From 9007e0fa70ca38b6b1ad78f6fd082d03d32d886c Mon Sep 17 00:00:00 2001 From: Mateo Wang <277851410+mateo-berri@users.noreply.github.com> Date: Wed, 28 Oct 2026 13:42:22 -0700 Subject: [PATCH 01/41] fix(files)!: scope provider and model file lists to the caller's managed files under require_managed_files (#47682) * test(files): cover owner-scoped file listing under require_managed_files * fix(files): list only the caller's managed files on every list route under require_managed_files * fix(files): narrow owner-scoped managed file lists to the named provider or model and page with after_id and before_id * fix(files): narrow scoped lists the way uploads route, and refuse lists that cannot be scoped A model name now matches every deployment that serves it, wildcards and aliases included, and a provider matches deployments by the provider uploads resolve, so anthropic/* and unprefixed Claude models count. The OpenAI and Azure lists also return the caller's pass-through uploads. An unknown provider and different after and after_id cursors answer 400, and a list with no matching deployment skips the database read. * fix(files): match stored files by model name and keep flag-off lists as on main Under require_managed_files, a narrowed list now matches a file's model mappings by deployment id or model name, so files LiteLLM stores in its own database show on their model and provider lists. The Anthropic list defaults to 20 rows when no limit is sent, after_id, before_id and the cursor 400s apply only under the flag, and an admin keeps the provider list when the managed files hook is not loaded. List errors are returned as values and mapped to HTTP errors in one place. * fix(files): keep admin provider lists and index the owner-scoped walk Under require_managed_files, an admin keeps the provider list as on main, and only a non-admin list reads the caller's own and team managed files. A stored file now also shows on the provider its model name routes to through a wildcard, with exact model names first, and a list routed to a model takes the default page size of the provider serving it. A new index on created_by and created_at keeps each chunk of the walk an index read, and the model check runs before the file object is parsed. * fix(files): filter provider and model lists in the query A non-admin list scoped to a provider or model now asks Postgres only for the caller's rows that map to one of that scope's deployments or models, or that live in file storage, instead of reading every owned row and filtering them in Python. A caller with many files on other providers no longer waits for the whole set to be read. Team members' files are matched in the same query. * test(files): type the fake raw query in the files list tests * refactor(proxy): list managed files through one typed database handle * fix(files): list alias and routing-group storage uploads on their provider --- .../proxy/hooks/managed_files.py | 151 ++++- .../migration.sql | 2 + .../litellm_proxy_extras/schema.prisma | 1 + litellm/constants.py | 1 + litellm/llms/base_llm/files/transformation.py | 23 +- .../openai_files_endpoints/files_endpoints.py | 255 +++++++- .../managed_id_rewriter.py | 6 + litellm/proxy/schema.prisma | 1 + schema.prisma | 1 + .../security/test_managed_file_listing.py | 263 ++++++++ .../proxy/test_managed_files_hook.py | 17 + .../test_files_endpoint.py | 569 +++++++++++++++++- ui/litellm-dashboard/src/lib/http/schema.d.ts | 6 + 13 files changed, 1245 insertions(+), 51 deletions(-) create mode 100644 litellm-proxy-extras/litellm_proxy_extras/migrations/20261021000000_managed_file_created_by_created_at_index/migration.sql create mode 100644 tests/integration/security/test_managed_file_listing.py diff --git a/enterprise/litellm_enterprise/proxy/hooks/managed_files.py b/enterprise/litellm_enterprise/proxy/hooks/managed_files.py index 7a07350e2d9..416444664a3 100644 --- a/enterprise/litellm_enterprise/proxy/hooks/managed_files.py +++ b/enterprise/litellm_enterprise/proxy/hooks/managed_files.py @@ -28,7 +28,7 @@ import httpx from fastapi import HTTPException from openai import APIConnectionError, AsyncOpenAI from openai.types.file_deleted import FileDeleted -from pydantic import BaseModel, JsonValue, TypeAdapter, ValidationError +from pydantic import BaseModel, ConfigDict, JsonValue, TypeAdapter, ValidationError from typing_extensions import ReadOnly, Unpack import litellm @@ -53,7 +53,7 @@ from litellm.litellm_core_utils.prompt_templates.common_utils import ( extract_file_metadata, ) from litellm.llms.base_llm.files.storage_backend_factory import get_storage_backend -from litellm.llms.base_llm.files.transformation import BaseFileEndpoints +from litellm.llms.base_llm.files.transformation import BaseFileEndpoints, FileListModelScope from litellm.llms.base_llm.managed_resources.isolation import ( build_list_page, build_owner_filter, @@ -309,6 +309,7 @@ def _is_xai_batch_output( class _ManagedFileRow(Protocol): unified_file_id: str file_object: OpenAIFileObject + model_mappings: Mapping[str, str] flat_model_file_ids: Sequence[str] storage_backend: Optional[str] storage_url: Optional[str] @@ -362,6 +363,8 @@ class _ManagedResourceDatabase(Protocol): @property def litellm_managedfiletable(self) -> _ManagedFileTableActions: ... + async def query_raw(self, query: str, *args: object) -> object: ... + @property def litellm_managedobjecttable(self) -> _ManagedObjectTableActions: ... @@ -385,10 +388,83 @@ class _RouterFileCallKwargs(TypedDict, total=False): custom_llm_provider: ReadOnly[str | None] +def _managed_database(prisma_client: _ManagedResourcePrismaClient) -> _ManagedResourceDatabase: + return prisma_client.db + + def _managed_file_table(prisma_client: _ManagedResourcePrismaClient) -> _ManagedFileTableActions: return prisma_client.db.litellm_managedfiletable +class _ListedFileRow(Protocol): + @property + def unified_file_id(self) -> str: ... + + @property + def file_object(self) -> object: ... + + @property + def model_mappings(self) -> Mapping[str, object]: ... + + @property + def storage_backend(self) -> str | None: ... + + +class _ScopedFileRow(BaseModel): + model_config = ConfigDict(frozen=True) + + unified_file_id: str + file_object: JsonValue = None + model_mappings: Mapping[str, JsonValue] + storage_backend: str | None = None + + +_SCOPED_FILE_ROWS: Final = TypeAdapter(tuple[_ScopedFileRow, ...]) + + +def _owner_condition(owner_filter: Mapping[str, object]) -> tuple[str, tuple[str, ...]]: + match owner_filter: + case {"OR": [{"created_by": str() as user_id}, {"team_id": str() as team_id}]}: + return '("created_by" = $1 OR "team_id" = $2)', (user_id, team_id) + case {"created_by": str() as user_id}: + return '"created_by" = $1', (user_id,) + case {"team_id": str() as team_id}: + return '"team_id" = $1', (team_id,) + case _: + return ("FALSE", ()) if owner_filter else ("TRUE", ()) + + +async def _scoped_file_rows( + database: _ManagedResourceDatabase, + owner_filter: Mapping[str, object], + model_mapping_keys: frozenset[str], + cursor_id: str | None, + direction: Literal["asc", "desc"], + take: int, +) -> tuple[_ScopedFileRow, ...]: + owner_condition, owner_params = _owner_condition(owner_filter) + keys_param: Final = len(owner_params) + 1 + comparison: Final = "<" if direction == "desc" else ">" + cursor_condition: Final = ( + f'AND ("created_at", "unified_file_id") {comparison} (SELECT "created_at", "unified_file_id" ' + f'FROM "LiteLLM_ManagedFileTable" WHERE "unified_file_id" = ${keys_param + 1}) ' + if cursor_id + else "" + ) + order: Final = direction.upper() + rows: Final = await database.query_raw( + 'SELECT "unified_file_id", "file_object", "model_mappings", "storage_backend" FROM "LiteLLM_ManagedFileTable" ' + f'WHERE {owner_condition} AND ("model_mappings" ?| ${keys_param}::text[] OR "storage_backend" IS NOT NULL) ' + f'{cursor_condition}ORDER BY "created_at" {order}, "unified_file_id" {order} ' + f"LIMIT ${keys_param + (2 if cursor_id else 1)}", + *owner_params, + sorted(model_mapping_keys), + *((cursor_id,) if cursor_id else ()), + take, + ) + return _SCOPED_FILE_ROWS.validate_python(rows) + + def _iter_provider_file_id_pairs( rows: Sequence[_ManagedFileRow], requested_provider_file_ids: frozenset[str], @@ -2265,14 +2341,23 @@ class _PROXY_LiteLLMManagedFiles(CustomLogger, BaseFileEndpoints): user_api_key_dict: UserAPIKeyAuth, limit: Optional[int] = None, after: Optional[str] = None, + model_scope: Optional[FileListModelScope] = None, + before: Optional[str] = None, **data: Dict, ) -> FileListPage: """List the managed files the caller owns, newest first. Pagination is keyset based on ``unified_file_id`` so a key that owns every file on the proxy still reads one bounded page at a time. - ``purpose`` is applied after parsing, because the managed file table - keeps it inside the ``file_object`` blob instead of a column, and rows + ``after`` pages toward older files and ``before`` toward newer ones, + each page still ordered newest first. + ``model_scope`` is applied in the query, so only rows whose + ``model_mappings`` name one of its deployment ids or model names, and + storage-backed rows, leave the database; a storage-backed row then + stays only when its model name is one the router sends to the listed + provider. ``purpose`` is applied after parsing, because the managed + file table keeps it inside the ``file_object`` blob instead of a + column. Rows whose blob will not parse drop out there too, so a chunk of rows can yield fewer matches than the page holds. Successive chunks are read until the page is full or the caller's rows run out, which keeps @@ -2281,9 +2366,8 @@ class _PROXY_LiteLLMManagedFiles(CustomLogger, BaseFileEndpoints): scan has to continue past it, the chunk widens to ``FILE_LIST_CONTINUATION_CHUNK_SIZE``, so the walk costs one query per that many rows instead of one per page. That bound is per query, not - per request: the work is still linear in the rows the caller owns, and - a filter matching nothing reads every one of them, with no index - covering either the owner filter or the sort. + per request: a ``purpose`` filter matching nothing still reads every + row that passes the query. """ validate_file_list_limit(limit) validate_file_list_purpose(purpose) @@ -2292,36 +2376,60 @@ class _PROXY_LiteLLMManagedFiles(CustomLogger, BaseFileEndpoints): if owner_filter is None: return FileListPage.model_validate(build_list_page([])) - if after: - cursor_row = await _managed_file_table(self.prisma_client).find_first( - where={**owner_filter, "unified_file_id": after} + if after and before: + raise ProxyException( + message="Pass either 'after' (or 'after_id') or 'before_id' when listing files, not both.", + type="invalid_request_error", + param="before_id", + code=400, + openai_code="invalid_value", + ) + cursor_param: Final = "before_id" if before else "after" + cursor: Final = before or after + database: Final = _managed_database(self.prisma_client) + if cursor: + cursor_row = await database.litellm_managedfiletable.find_first( + where={**owner_filter, "unified_file_id": cursor} ) if cursor_row is None: raise ProxyException( - message=f"Invalid 'after' cursor: no file found with id '{after}'.", + message=f"Invalid '{cursor_param}' cursor: no file found with id '{cursor}'.", type="invalid_request_error", - param="after", + param=cursor_param, code=400, openai_code="invalid_value", ) + if model_scope is not None and not model_scope.model_mapping_keys: + return FileListPage.model_validate(build_list_page([])) + page_size: Final = min(limit or MAX_FILE_LIST_LIMIT, MAX_FILE_LIST_LIMIT) matches: Final[List[OpenAIFileObject]] = [] - cursor_id = after + cursor_id = cursor chunk_size = page_size + 1 + direction: Final[Literal["asc", "desc"]] = "asc" if before else "desc" while len(matches) <= page_size: cursor_args: _CursorPageArgs = {"cursor": {"unified_file_id": cursor_id}, "skip": 1} if cursor_id else {} - chunk = await _managed_file_table(self.prisma_client).find_many( - where=owner_filter, - take=chunk_size, - order=[{"created_at": "desc"}, {"unified_file_id": "desc"}], - **cursor_args, + chunk: Sequence[_ListedFileRow] = ( + await _scoped_file_rows( + database, owner_filter, model_scope.model_mapping_keys, cursor_id, direction, chunk_size + ) + if model_scope is not None + else await database.litellm_managedfiletable.find_many( + where=owner_filter, + take=chunk_size, + order=[{"created_at": direction}, {"unified_file_id": direction}], + **cursor_args, + ) ) matches.extend( _public_file_object(parsed_file_object, row.unified_file_id) for row in chunk - if (parsed_file_object := _parse_managed_file_object(row.file_object, row.unified_file_id)) is not None + if ( + model_scope is None or model_scope.holds(row.model_mappings, stored=row.storage_backend is not None) + ) + and (parsed_file_object := _parse_managed_file_object(row.file_object, row.unified_file_id)) is not None and (purpose is None or parsed_file_object.purpose == purpose) ) if len(chunk) < chunk_size: @@ -2329,7 +2437,10 @@ class _PROXY_LiteLLMManagedFiles(CustomLogger, BaseFileEndpoints): cursor_id = chunk[-1].unified_file_id chunk_size = max(chunk_size, FILE_LIST_CONTINUATION_CHUNK_SIZE) - return FileListPage.model_validate(build_list_page(matches[:page_size], has_more=len(matches) > page_size)) + page: Final = matches[:page_size] + return FileListPage.model_validate( + build_list_page(page[::-1] if before else page, has_more=len(matches) > page_size) + ) def _is_batch_polling_enabled(self) -> bool: """ diff --git a/litellm-proxy-extras/litellm_proxy_extras/migrations/20261021000000_managed_file_created_by_created_at_index/migration.sql b/litellm-proxy-extras/litellm_proxy_extras/migrations/20261021000000_managed_file_created_by_created_at_index/migration.sql new file mode 100644 index 00000000000..8bbcb67bddf --- /dev/null +++ b/litellm-proxy-extras/litellm_proxy_extras/migrations/20261021000000_managed_file_created_by_created_at_index/migration.sql @@ -0,0 +1,2 @@ +CREATE INDEX CONCURRENTLY IF NOT EXISTS "LiteLLM_ManagedFileTable_created_by_created_at_idx" +ON "LiteLLM_ManagedFileTable" ("created_by", "created_at" DESC); diff --git a/litellm-proxy-extras/litellm_proxy_extras/schema.prisma b/litellm-proxy-extras/litellm_proxy_extras/schema.prisma index 83a0e6f9e05..4187f76c97c 100644 --- a/litellm-proxy-extras/litellm_proxy_extras/schema.prisma +++ b/litellm-proxy-extras/litellm_proxy_extras/schema.prisma @@ -1163,6 +1163,7 @@ model LiteLLM_ManagedFileTable { @@index([unified_file_id]) @@index([flat_model_file_ids], type: Gin) @@index([team_id, created_at(sort: Desc)]) + @@index([created_by, created_at(sort: Desc)]) } model LiteLLM_ManagedObjectTable { // for batches or finetuning jobs which use the diff --git a/litellm/constants.py b/litellm/constants.py index 2b083b03766..1d5c6404bfd 100644 --- a/litellm/constants.py +++ b/litellm/constants.py @@ -129,6 +129,7 @@ MAX_S3_OBJECT_DOWNLOAD_FILENAME_BYTES: Final = 1024 S3_LOG_PROMPTS_ONLY_ENV_VAR: Final = "S3_LOG_PROMPTS_ONLY" S3_PARTITION_GRANULARITY_ENV_VAR: Final = "S3_PARTITION_GRANULARITY" MAX_FILE_LIST_LIMIT: Final = 10000 +ANTHROPIC_DEFAULT_FILE_LIST_LIMIT: Final = 20 DEFAULT_SQS_FLUSH_INTERVAL_SECONDS: Final = get_env_int("DEFAULT_SQS_FLUSH_INTERVAL_SECONDS", 10) DEFAULT_NUM_WORKERS_LITELLM_PROXY: Final = get_env_int("DEFAULT_NUM_WORKERS_LITELLM_PROXY", 1) GRACEFUL_SHUTDOWN_TEARDOWN_HEADROOM_SECONDS: Final = 10 diff --git a/litellm/llms/base_llm/files/transformation.py b/litellm/llms/base_llm/files/transformation.py index 4346e214dc8..491fb1919b6 100644 --- a/litellm/llms/base_llm/files/transformation.py +++ b/litellm/llms/base_llm/files/transformation.py @@ -1,6 +1,7 @@ from abc import ABC, abstractmethod -from collections.abc import AsyncGenerator, Iterator, Mapping -from typing import TYPE_CHECKING, Any, Union +from collections.abc import AsyncGenerator, Callable, Iterable, Iterator, Mapping +from dataclasses import dataclass +from typing import TYPE_CHECKING, Any, Final, Union import httpx from openai.types.file_deleted import FileDeleted @@ -38,6 +39,22 @@ else: UserAPIKeyAuth = Any +def _routes_nowhere(_model_name: str) -> bool: + return False + + +@dataclass(frozen=True, slots=True) +class FileListModelScope: + model_mapping_keys: frozenset[str] + routes_stored_model_name: Callable[[str], bool] = _routes_nowhere + + def holds(self, model_mapping_keys: Iterable[str], stored: bool) -> bool: + mapped_keys: Final = frozenset(model_mapping_keys) + return not self.model_mapping_keys.isdisjoint(mapped_keys) or ( + stored and any(map(self.routes_stored_model_name, mapped_keys)) + ) + + class BaseFileUploadStream(ABC): """Re-iterable request body that yields an upload's bytes lazily. @@ -274,6 +291,8 @@ class BaseFileEndpoints(ABC): user_api_key_dict: UserAPIKeyAuth, limit: int | None = None, after: str | None = None, + model_scope: FileListModelScope | None = None, + before: str | None = None, **data: dict, ) -> FileListPage: pass diff --git a/litellm/proxy/openai_files_endpoints/files_endpoints.py b/litellm/proxy/openai_files_endpoints/files_endpoints.py index d2ceefffe47..5d63c3db222 100644 --- a/litellm/proxy/openai_files_endpoints/files_endpoints.py +++ b/litellm/proxy/openai_files_endpoints/files_endpoints.py @@ -6,9 +6,12 @@ ###################################################################### import asyncio +import functools +import itertools import traceback -from collections.abc import Mapping, Sequence -from typing import Any, BinaryIO, Final, TypedDict, cast, get_args +from collections.abc import Iterable, Iterator, Mapping, Sequence +from dataclasses import dataclass +from typing import Any, BinaryIO, Final, NoReturn, TypedDict, cast, get_args import httpx from fastapi import ( @@ -28,12 +31,13 @@ from typing_extensions import ReadOnly import litellm from litellm import CreateFileRequest, get_secret_str from litellm._logging import verbose_proxy_logger +from litellm.constants import ANTHROPIC_DEFAULT_FILE_LIST_LIMIT from litellm.litellm_core_utils.cloud_storage_security import ( is_managed_cloud_storage_uri, ) from litellm.litellm_core_utils.core_helpers import get_or_create_metadata_bucket from litellm.llms.base_llm.files.litellm_db_storage_backend import LITELLM_DB_STORAGE_BACKEND_NAME -from litellm.llms.base_llm.files.transformation import BaseFileEndpoints +from litellm.llms.base_llm.files.transformation import BaseFileEndpoints, FileListModelScope from litellm.llms.base_llm.managed_resources.isolation import build_list_page from litellm.proxy._types import * from litellm.proxy.auth.user_api_key_auth import user_api_key_auth @@ -112,6 +116,9 @@ from litellm.types.llms.openai import ( OpenAIFileObject, OpenAIFilesPurpose, ) +from litellm.types.proxy.auth.user_api_key_auth import user_api_key_has_admin_view +from litellm.types.router import Deployment +from litellm.types.utils import LlmProviders router: Final = APIRouter() @@ -1673,6 +1680,185 @@ def _as_file_list_page(response: object) -> object: return FileListPage.model_validate(page) +async def _list_owned_managed_files( + managed_files_obj: BaseFileEndpoints, + user_api_key_dict: UserAPIKeyAuth, + purpose: str | None, + limit: int | None, + after: str | None, + before: str | None, + model_scope: FileListModelScope | None, +) -> FileListPage: + return await managed_files_obj.afile_list( + purpose=purpose, + litellm_parent_otel_span=user_api_key_dict.parent_otel_span, + user_api_key_dict=user_api_key_dict, + limit=limit, + after=after, + before=before, + model_scope=model_scope, + ) + + +@dataclass(frozen=True, slots=True) +class _UnscopedFileList: + status_code: int + detail: str + + +@dataclass(frozen=True, slots=True) +class _ListedFiles: + model_scope: FileListModelScope | None + custom_llm_provider: str | None + + +def _raise_unscoped_file_list(failure: _UnscopedFileList) -> NoReturn: + raise HTTPException(status_code=failure.status_code, detail=failure.detail) + + +def _single_target_model_name(target_model_names: str) -> str | _UnscopedFileList: + target_model_names_list: Final = target_model_names.split(",") + if len(target_model_names_list) != 1: + return _UnscopedFileList( + status_code=400, + detail="target_model_names on list files must be a list of one model name. Example: ['gpt-4o']", + ) + return target_model_names_list[0] + + +async def _requested_custom_llm_provider(request: Request, provider: str | None) -> str | None: + return ( + provider + or get_custom_llm_provider_from_request_headers(request=request) + or get_custom_llm_provider_from_request_query(request=request) + or await get_custom_llm_provider_from_request_body(request=request) + ) + + +def _default_file_list_limit(custom_llm_provider: str | None) -> int | None: + return ANTHROPIC_DEFAULT_FILE_LIST_LIMIT if custom_llm_provider == LlmProviders.ANTHROPIC.value else None + + +def _deployment_model_mapping_keys(deployment: Deployment) -> frozenset[str]: + return frozenset(key for key in (deployment.model_name, deployment.model_info.id) if key is not None) + + +def _model_mapping_keys(deployments: Iterable[Deployment]) -> frozenset[str]: + return frozenset(itertools.chain.from_iterable(map(_deployment_model_mapping_keys, deployments))) + + +def _deployments_serving_model(llm_router: Router, model: str) -> tuple[Deployment, ...]: + deployment: Final = llm_router.get_deployment(model_id=model) if llm_router.has_model_id(model) else None + if deployment is not None: + return (deployment,) + return tuple(Deployment.model_validate(listed) for listed in llm_router.get_model_list(model_name=model) or ()) + + +def _resolved_deployment_provider(deployment: Deployment) -> str | None: + try: + return litellm.get_llm_provider( + model=deployment.litellm_params.model, + custom_llm_provider=deployment.litellm_params.custom_llm_provider, + )[1] + except litellm.BadRequestError: + return None + + +def _deployments_on_provider(llm_router: Router | None, custom_llm_provider: str) -> Iterator[Deployment]: + if llm_router is None: + return iter(()) + return ( + deployment + for deployment_id in llm_router.get_model_ids() + if (deployment := llm_router.get_deployment(model_id=deployment_id)) is not None + and _resolved_deployment_provider(deployment) == custom_llm_provider + ) + + +def _routed_model_name_on_provider( + llm_router: Router, custom_llm_provider: str, provider_model_names: frozenset[str], model_name: str +) -> bool: + if model_name in llm_router.model_names: + return model_name in provider_model_names + return llm_router.pattern_router.routes_to_provider(model_name, custom_llm_provider) + + +def _stored_model_name_routes_to_provider( + llm_router: Router, custom_llm_provider: str, provider_model_names: frozenset[str], model_name: str +) -> bool: + routing_group: Final = llm_router.get_routing_group(model_name) + routed_model_names: Final = ( + tuple(routing_group.models) if routing_group is not None else (llm_router.routable_model_group(model_name),) + ) + return any( + _routed_model_name_on_provider(llm_router, custom_llm_provider, provider_model_names, routed_model_name) + for routed_model_name in routed_model_names + ) + + +def _files_on_provider(llm_router: Router | None, custom_llm_provider: str) -> _ListedFiles | _UnscopedFileList: + from litellm.proxy.pass_through_endpoints.managed_id_rewriter import passthrough_file_model_id + + if custom_llm_provider not in litellm.provider_list: + return _UnscopedFileList(status_code=400, detail=f"Unknown provider '{custom_llm_provider}' for listing files.") + passthrough_model_id: Final = passthrough_file_model_id(custom_llm_provider) + deployments: Final = tuple(_deployments_on_provider(llm_router, custom_llm_provider)) + model_mapping_keys: Final = _model_mapping_keys(deployments) | frozenset( + (passthrough_model_id,) if passthrough_model_id is not None else () + ) + if llm_router is None: + return _ListedFiles(FileListModelScope(model_mapping_keys), custom_llm_provider) + return _ListedFiles( + FileListModelScope( + model_mapping_keys, + routes_stored_model_name=functools.partial( + _stored_model_name_routes_to_provider, + llm_router, + custom_llm_provider, + frozenset(deployment.model_name for deployment in deployments), + ), + ), + custom_llm_provider, + ) + + +def _files_serving_model(llm_router: Router, model: str) -> _ListedFiles: + deployments: Final = _deployments_serving_model(llm_router, model) + providers: Final = frozenset(map(_resolved_deployment_provider, deployments)) + return _ListedFiles( + FileListModelScope(_model_mapping_keys(deployments)), next(iter(providers)) if len(providers) == 1 else None + ) + + +async def _listed_files( + llm_router: Router | None, + user_api_key_dict: UserAPIKeyAuth, + custom_llm_provider: str | None, + target_model_names: str | None, + routed_model: str | None, +) -> _ListedFiles | _UnscopedFileList: + if routed_model is not None and llm_router is not None: + return _files_serving_model(llm_router, routed_model) + if target_model_names: + target_model_name: Final = _single_target_model_name(target_model_names) + if isinstance(target_model_name, _UnscopedFileList): + return target_model_name + if llm_router is None: + return _UnscopedFileList( + status_code=500, detail="LLM Router not initialized. Ensure models added to proxy." + ) + await get_authorized_credentials_for_model( + llm_router=llm_router, + model_id=target_model_name, + user_api_key_dict=user_api_key_dict, + operation_context="file list", + ) + return _files_serving_model(llm_router, target_model_name) + if custom_llm_provider is None: + return _ListedFiles(None, None) + return _files_on_provider(llm_router, custom_llm_provider) + + @router.get( "/{provider}/v1/files", dependencies=[Depends(user_api_key_auth)], @@ -1697,6 +1883,8 @@ async def list_files( purpose: str | None = None, limit: int | None = None, after: str | None = None, + after_id: str | None = None, + before_id: str | None = None, ): """ Returns information about a specific file. that can be used across - Assistants API, Batch API @@ -1739,6 +1927,10 @@ async def list_files( ) response: Any | None = None + owned_files_hook: Final = proxy_logging_obj.get_proxy_hook("managed_files") + list_owned_files_only: Final = litellm.require_managed_files is True and not user_api_key_has_admin_view( + user_api_key_dict + ) # Check for model-based credential routing (no file_id encoding check for list) should_route, model_used, _, credentials = await handle_model_based_routing( @@ -1750,7 +1942,36 @@ async def list_files( check_file_id_encoding=False, ) - if should_route and credentials is not None: + if list_owned_files_only: + if after and after_id and after != after_id: + raise HTTPException( + status_code=400, + detail="Pass the list cursor as either 'after' or 'after_id', not both with different ids.", + ) + if not isinstance(owned_files_hook, BaseFileEndpoints): + raise HTTPException( + status_code=500, + detail="Managed file listing is unavailable, so require_managed_files cannot scope the list to the caller.", + ) + listed_files: Final = await _listed_files( + llm_router=llm_router, + user_api_key_dict=user_api_key_dict, + custom_llm_provider=await _requested_custom_llm_provider(request=request, provider=provider), + target_model_names=target_model_names, + routed_model=model_used if should_route else None, + ) + if isinstance(listed_files, _UnscopedFileList): + _raise_unscoped_file_list(listed_files) + response = await _list_owned_managed_files( # rebind-ok: each dispatch branch binds response + managed_files_obj=owned_files_hook, + user_api_key_dict=user_api_key_dict, + purpose=purpose, + limit=limit if limit is not None else _default_file_list_limit(listed_files.custom_llm_provider), + after=after or after_id, + before=before_id, + model_scope=listed_files.model_scope, + ) + elif should_route and credentials is not None: # Use model-based routing with credentials from config prepare_data_with_credentials(data=data, credentials=credentials, include_internal_credentials=True) response = await litellm.afile_list( @@ -1762,12 +1983,9 @@ async def list_files( verbose_proxy_logger.debug("Listed files using model: %s", model_used) elif target_model_names and isinstance(target_model_names, str): - target_model_names_list: Final = target_model_names.split(",") - if len(target_model_names_list) != 1: - raise HTTPException( - status_code=400, - detail="target_model_names on list files must be a list of one model name. Example: ['gpt-4o']", - ) + target_model_name: Final = _single_target_model_name(target_model_names) + if isinstance(target_model_name, _UnscopedFileList): + _raise_unscoped_file_list(target_model_name) if llm_router is None: raise HTTPException( status_code=500, @@ -1775,7 +1993,7 @@ async def list_files( ) credentials = await get_authorized_credentials_for_model( llm_router=llm_router, - model_id=target_model_names_list[0], + model_id=target_model_name, user_api_key_dict=user_api_key_dict, operation_context="file list", ) @@ -1786,20 +2004,17 @@ async def list_files( **data, ) else: - custom_llm_provider: Final = ( - provider - or get_custom_llm_provider_from_request_headers(request=request) - or get_custom_llm_provider_from_request_query(request=request) - or await get_custom_llm_provider_from_request_body(request=request) - ) + custom_llm_provider: Final = await _requested_custom_llm_provider(request=request, provider=provider) managed_files_obj: Final = proxy_logging_obj.get_proxy_hook("managed_files") if custom_llm_provider is None and isinstance(managed_files_obj, BaseFileEndpoints): - response = await managed_files_obj.afile_list( - purpose=purpose, - litellm_parent_otel_span=user_api_key_dict.parent_otel_span, + response = await _list_owned_managed_files( # rebind-ok: each dispatch branch binds response + managed_files_obj=managed_files_obj, user_api_key_dict=user_api_key_dict, + purpose=purpose, limit=limit, after=after, + before=None, + model_scope=None, ) else: resolved_custom_llm_provider: Final = custom_llm_provider or "openai" diff --git a/litellm/proxy/pass_through_endpoints/managed_id_rewriter.py b/litellm/proxy/pass_through_endpoints/managed_id_rewriter.py index a5a04fb68d2..9790c150ae1 100644 --- a/litellm/proxy/pass_through_endpoints/managed_id_rewriter.py +++ b/litellm/proxy/pass_through_endpoints/managed_id_rewriter.py @@ -242,6 +242,12 @@ def _passthrough_sentinel_model_id(provider: str) -> str: return f"_passthrough_{provider}" +def passthrough_file_model_id(provider: str) -> str | None: + if (provider, "/v1/files") not in _LIST_ROUTE_TABLE: + return None + return _passthrough_sentinel_model_id(provider) + + # Key under which the provider marker is stored in a file row's model_mappings. # Its value lands in flat_model_file_ids (built from model_mappings.values()), # giving the file table a DB-queryable provider scope it otherwise lacks. diff --git a/litellm/proxy/schema.prisma b/litellm/proxy/schema.prisma index 83a0e6f9e05..4187f76c97c 100644 --- a/litellm/proxy/schema.prisma +++ b/litellm/proxy/schema.prisma @@ -1163,6 +1163,7 @@ model LiteLLM_ManagedFileTable { @@index([unified_file_id]) @@index([flat_model_file_ids], type: Gin) @@index([team_id, created_at(sort: Desc)]) + @@index([created_by, created_at(sort: Desc)]) } model LiteLLM_ManagedObjectTable { // for batches or finetuning jobs which use the diff --git a/schema.prisma b/schema.prisma index 83a0e6f9e05..4187f76c97c 100644 --- a/schema.prisma +++ b/schema.prisma @@ -1163,6 +1163,7 @@ model LiteLLM_ManagedFileTable { @@index([unified_file_id]) @@index([flat_model_file_ids], type: Gin) @@index([team_id, created_at(sort: Desc)]) + @@index([created_by, created_at(sort: Desc)]) } model LiteLLM_ManagedObjectTable { // for batches or finetuning jobs which use the diff --git a/tests/integration/security/test_managed_file_listing.py b/tests/integration/security/test_managed_file_listing.py new file mode 100644 index 00000000000..8c736b2b2bf --- /dev/null +++ b/tests/integration/security/test_managed_file_listing.py @@ -0,0 +1,263 @@ +import json +import uuid +from collections.abc import Callable, Iterator, Mapping +from contextlib import contextmanager +from dataclasses import dataclass, field +from itertools import count, islice +from pathlib import Path +from types import MappingProxyType +from typing import Final +from urllib.parse import parse_qs, urlsplit + +import pytest +import yaml +from integration._support.client import Gateway, gateway_from_environment +from integration._support.process import owned_proxy +from integration._support.wire import Reply, Request, Wire, wire_server +from anthropic import Anthropic +from openai import OpenAI +from pydantic import BaseModel, ConfigDict, JsonValue + +pytestmark: Final = pytest.mark.shares_worker_state + +UNMANAGED_FILE: Final = "file-uploaded-straight-to-the-provider" +UPLOAD: Final = b"isolation check\n" + + +@dataclass(frozen=True, slots=True) +class Deployment: + provider: str + model: str + litellm_model: str + api_key: str + api_path: str + + +OPENAI: Final = Deployment("openai", "gpt-4.1-managed-listing", "openai/gpt-4.1", "sk-managed-listing", "/v1") +ANTHROPIC: Final = Deployment( + "anthropic", "claude-managed-listing", "anthropic/claude-haiku-5-5", "sk-ant-managed-listing", "" +) +DEPLOYMENTS: Final = (pytest.param(OPENAI, id="openai"), pytest.param(ANTHROPIC, id="anthropic")) + + +@dataclass(frozen=True, slots=True) +class ListCall: + path: str + params: Mapping[str, str] = field(default_factory=lambda: MappingProxyType({})) + headers: Mapping[str, str] = field(default_factory=lambda: MappingProxyType({})) + + +LIST_FORMS: Final = ( + pytest.param(lambda d: ListCall(f"/{d.provider}/v1/files"), id="provider-path"), + pytest.param(lambda d: ListCall("/v1/files", params={"provider": d.provider}), id="provider-query"), + pytest.param(lambda d: ListCall("/v1/files", headers={"custom-llm-provider": d.provider}), id="provider-header"), + pytest.param(lambda d: ListCall("/v1/files", params={"target_model_names": d.model}), id="target-model-names"), + pytest.param(lambda d: ListCall("/v1/files", headers={"x-litellm-model": d.model}), id="model-header"), +) + + +class FilePage(BaseModel): + model_config = ConfigDict(extra="ignore", frozen=True) + + data: list[dict[str, JsonValue]] + has_more: bool + + @property + def ids(self) -> list[str]: + return [str(entry["id"]) for entry in self.data] + + +def _openai_file(file_id: str) -> dict[str, JsonValue]: + return { + "id": file_id, + "object": "file", + "bytes": len(UPLOAD), + "created_at": 1700000000, + "filename": "notes.txt", + "purpose": "user_data", + "status": "processed", + } + + +def _anthropic_file(file_id: str) -> dict[str, JsonValue]: + return { + "id": file_id, + "type": "file", + "filename": "notes.txt", + "mime_type": "text/plain", + "size_bytes": len(UPLOAD), + "created_at": "2026-01-01T00:00:00Z", + "downloadable": False, + } + + +def _reply(body: Mapping[str, JsonValue]) -> Reply: + return Reply(body=json.dumps(body).encode()) + + +def _list_reply(provider: str, file_ids: tuple[str, ...]) -> Reply: + if provider == ANTHROPIC.provider: + return _reply( + { + "data": [_anthropic_file(file_id) for file_id in file_ids], + "has_more": False, + "first_id": file_ids[0], + "last_id": file_ids[-1], + } + ) + return _reply({"object": "list", "data": [_openai_file(file_id) for file_id in file_ids]}) + + +def _responder(provider_files: Mapping[str, tuple[str, ...]]) -> Callable[[Request], Reply]: + uploads: Final = {provider: count() for provider in provider_files} + + def respond(request: Request) -> Reply: + target: Final = urlsplit(request.target) + path: Final = target.path + provider: Final = ANTHROPIC.provider if "x-api-key" in request.headers else OPENAI.provider + files: Final = provider_files[provider] + if (request.method, path) == ("GET", "/v1/models"): + return _reply({"object": "list", "data": []}) + if (request.method, path) == ("POST", "/v1/files"): + uploaded: Final = files[next(uploads[provider])] + return _reply(_anthropic_file(uploaded) if provider == ANTHROPIC.provider else _openai_file(uploaded)) + if (request.method, path) == ("GET", "/v1/files"): + limit: Final = int(parse_qs(target.query).get("limit", ["10000"])[0]) + return _list_reply(provider, (UNMANAGED_FILE, *reversed(files))[:limit]) + return Reply(status=404, body=json.dumps({"error": {"message": f"unscripted {request.target}"}}).encode()) + + return respond + + +def _config(wire: Wire) -> dict[str, JsonValue]: + return { + "model_list": [ + { + "model_name": deployment.model, + "litellm_params": { + "model": deployment.litellm_model, + "api_key": deployment.api_key, + "api_base": f"{wire.url}{deployment.api_path}", + }, + } + for deployment in (OPENAI, ANTHROPIC) + ], + "litellm_settings": {"require_managed_files": True}, + "general_settings": {"master_key": "os.environ/LITELLM_MASTER_KEY", "database_url": "os.environ/DATABASE_URL"}, + } + + +@dataclass(frozen=True, slots=True) +class Rig: + proxy: Gateway + owner: str + intruder: str + files_by_provider: Mapping[str, tuple[str, ...]] + every_file: tuple[str, ...] + + @property + def base_url(self) -> str: + return str(self.proxy.client.base_url).rstrip("/") + + def list(self, call: ListCall, key: str | None = None) -> FilePage: + response: Final = self.proxy.request("GET", call.path, key=key, params=call.params, headers=call.headers) + assert response.status_code == 200, response.text + return FilePage.model_validate_json(response.text) + + +def _upload(proxy: Gateway, key: str, deployments: tuple[Deployment, ...]) -> str: + with OpenAI(base_url=f"{str(proxy.client.base_url).rstrip('/')}/openai/v1", api_key=key, max_retries=0) as client: + uploaded: Final = client.files.with_raw_response.create( + file=("notes.txt", UPLOAD, "text/plain"), + purpose="user_data", + extra_body={"target_model_names": ",".join(deployment.model for deployment in deployments)}, + ) + assert uploaded.status_code == 200, uploaded.text + return uploaded.parse().id + + +@contextmanager +def _rig(directory: Path) -> Iterator[Rig]: + build: Final = uuid.uuid4().hex + provider_files: Final = { + OPENAI.provider: (f"file-older-{build}", f"file-newer-{build}"), + ANTHROPIC.provider: (f"file_older_{build}", f"file_newer_{build}"), + } + with gateway_from_environment() as shared, wire_server(_responder(provider_files)) as wire: + config: Final = directory / "config.yaml" + config.write_text(yaml.safe_dump(_config(wire))) + with owned_proxy(shared, directory, {}, config=config) as proxy, proxy.scenario() as scenario: + owner: Final = scenario.key(user_id=scenario.user()) + intruder: Final = scenario.key(user_id=scenario.user()) + on_both: Final = _upload(proxy, owner, (OPENAI, ANTHROPIC)) + on_openai: Final = _upload(proxy, owner, (OPENAI,)) + on_anthropic: Final = _upload(proxy, owner, (ANTHROPIC,)) + yield Rig( + proxy, + owner, + intruder, + MappingProxyType( + {OPENAI.provider: (on_openai, on_both), ANTHROPIC.provider: (on_anthropic, on_both)} + ), + (on_anthropic, on_openai, on_both), + ) + + +@pytest.fixture(scope="module") +def rig(tmp_path_factory: pytest.TempPathFactory) -> Iterator[Rig]: + with _rig(tmp_path_factory.mktemp("managed-file-listing")) as built: + yield built + + +@pytest.mark.parametrize("deployment", DEPLOYMENTS) +@pytest.mark.parametrize("form", LIST_FORMS) +def test_another_key_lists_none_of_the_files_on_the_shared_key( + rig: Rig, deployment: Deployment, form: Callable[[Deployment], ListCall] +) -> None: + page: Final = rig.list(form(deployment), key=rig.intruder) + assert (page.ids, page.has_more) == ([], False) + + +@pytest.mark.parametrize("deployment", DEPLOYMENTS) +@pytest.mark.parametrize("form", LIST_FORMS) +def test_owner_lists_their_own_managed_files_on_that_provider_or_model_newest_first( + rig: Rig, deployment: Deployment, form: Callable[[Deployment], ListCall] +) -> None: + page: Final = rig.list(form(deployment), key=rig.owner) + assert (page.ids, page.has_more) == (list(rig.files_by_provider[deployment.provider]), False) + + +def test_owner_lists_every_managed_file_when_no_provider_or_model_is_named(rig: Rig) -> None: + page: Final = rig.list(ListCall("/v1/files"), key=rig.owner) + assert (page.ids, page.has_more) == (list(rig.every_file), False) + + +@pytest.mark.parametrize("cursor_param", ["after", "after_id"]) +def test_owner_pages_through_a_provider_list_with_limit_and_a_cursor(rig: Rig, cursor_param: str) -> None: + first: Final = rig.list(ListCall("/openai/v1/files", params={"limit": "1"}), key=rig.owner) + second: Final = rig.list( + ListCall("/openai/v1/files", params={"limit": "1", cursor_param: first.ids[0]}), key=rig.owner + ) + expected: Final = rig.files_by_provider[OPENAI.provider] + assert [(page.ids, page.has_more) for page in (first, second)] == [ + ([expected[0]], True), + ([expected[1]], False), + ] + + +def test_anthropic_sdk_auto_pages_the_owners_files_to_the_end(rig: Rig) -> None: + with Anthropic(base_url=f"{rig.base_url}/anthropic", api_key=rig.owner, max_retries=0) as client: + listed: Final = [file.id for file in islice(client.beta.files.list(limit=1), 5)] + assert listed == list(rig.files_by_provider[ANTHROPIC.provider]) + + +def test_anthropic_sdk_pages_backwards_to_the_newest_file_and_stops(rig: Rig) -> None: + newest_first: Final = rig.files_by_provider[ANTHROPIC.provider] + with Anthropic(base_url=f"{rig.base_url}/anthropic", api_key=rig.owner, max_retries=0) as client: + listed: Final = [file.id for file in islice(client.beta.files.list(limit=1, before_id=newest_first[-1]), 5)] + assert listed == list(reversed(newest_first[:-1])) + + +def test_master_key_still_lists_the_files_on_the_shared_provider_key(rig: Rig) -> None: + listed: Final = rig.list(ListCall(f"/{ANTHROPIC.provider}/v1/files")).ids + assert UNMANAGED_FILE in listed diff --git a/tests/unit/enterprise/proxy/test_managed_files_hook.py b/tests/unit/enterprise/proxy/test_managed_files_hook.py index 01f20689e89..f9f4e17c760 100644 --- a/tests/unit/enterprise/proxy/test_managed_files_hook.py +++ b/tests/unit/enterprise/proxy/test_managed_files_hook.py @@ -15,6 +15,7 @@ import pytest from typing import Final, Optional from unittest.mock import AsyncMock, MagicMock, patch +from litellm.llms.base_llm.files.transformation import FileListModelScope from litellm.proxy._types import LitellmUserRoles, ProxyException, UserAPIKeyAuth from litellm.proxy.openai_files_endpoints.common_utils import BATCH_CREATE_HIDDEN_PARAM from litellm.types.llms.openai import FileListPage, OpenAIFileObject @@ -649,6 +650,22 @@ async def test_afile_list_denies_a_caller_with_no_identity_at_all(): assert table.find_many_calls == [] +@pytest.mark.asyncio +async def test_afile_list_reads_no_rows_when_no_deployment_can_hold_a_file() -> None: + managed_files, table = _make_managed_files_over_rows([_make_managed_file_row("unified-mine")]) + + response: Final = await managed_files.afile_list( + purpose=None, + litellm_parent_otel_span=None, + user_api_key_dict=_make_user_api_key_dict(), + model_scope=FileListModelScope(frozenset()), + ) + + assert response.data == [] + assert response.has_more is False + assert table.find_many_calls == [] + + @pytest.mark.asyncio async def test_afile_list_filters_by_purpose(): managed_files, _ = _make_managed_files_over_rows( diff --git a/tests/unit/proxy/openai_files_endpoint/test_files_endpoint.py b/tests/unit/proxy/openai_files_endpoint/test_files_endpoint.py index d47db69257b..ff63e37bcd8 100644 --- a/tests/unit/proxy/openai_files_endpoint/test_files_endpoint.py +++ b/tests/unit/proxy/openai_files_endpoint/test_files_endpoint.py @@ -1,3 +1,4 @@ +import itertools import json from typing import Final, List, Optional from urllib.parse import quote @@ -12,6 +13,7 @@ from pytest_mock import MockerFixture import litellm from litellm import Router +from litellm.constants import ANTHROPIC_DEFAULT_FILE_LIST_LIMIT from litellm.files.types import FileContentStreamingResult from litellm.proxy._types import LiteLLM_UserTableFiltered, LitellmUserRoles, UserAPIKeyAuth from litellm.proxy.hooks import get_proxy_hook @@ -4969,6 +4971,8 @@ def _managed_file_row( stored_file_id: Optional[str] = None, created_by: str = "test-user", team_id: Optional[str] = None, + model_mappings: Optional[dict[str, str]] = None, + storage_backend: Optional[str] = None, ) -> MagicMock: file_object = OpenAIFileObject( id=stored_file_id or unified_file_id, @@ -4984,6 +4988,8 @@ def _managed_file_row( file_object=file_object.model_dump(), created_by=created_by, team_id=team_id, + model_mappings=model_mappings or {}, + storage_backend=storage_backend, ) @@ -5010,16 +5016,49 @@ class _ManagedFileTableOverRows: return next(iter(self._owned_rows(where)), None) async def find_many(self, where, take=None, order=None, cursor=None, skip=0): - rows = self._owned_rows(where) - if cursor is not None: - start = next( - index - for index, row in enumerate(rows) - if row.unified_file_id == cursor["unified_file_id"] - ) - rows = rows[start + skip :] + owned_rows: Final = self._owned_rows(where) + oldest_first: Final = bool(order) and order[0].get("created_at") == "asc" + ordered_rows: Final = owned_rows[::-1] if oldest_first else owned_rows + start: Final = ( + 0 + if cursor is None + else skip + + next(index for index, row in enumerate(ordered_rows) if row.unified_file_id == cursor["unified_file_id"]) + ) + rows: Final = ordered_rows[start:] return rows if take is None else rows[:take] + async def query_raw(self, query: str, *params: object) -> list[dict[str, object]]: + keys_index: Final = next(index for index, param in enumerate(params) if isinstance(param, list)) + owner_ids: Final = frozenset(params[:keys_index]) + model_mapping_keys: Final = frozenset(params[keys_index]) + cursor_id: Final = params[keys_index + 1] if len(params) == keys_index + 3 else None + ordered_rows: Final = [ + row + for row in (self.rows[::-1] if '"created_at" ASC' in query else self.rows) + if not owner_ids or {row.created_by, row.team_id} & owner_ids + ] + start: Final = ( + 0 + if cursor_id is None + else 1 + next(index for index, row in enumerate(ordered_rows) if row.unified_file_id == cursor_id) + ) + listed_rows: Final = [ + row + for row in ordered_rows[start:] + if not model_mapping_keys.isdisjoint(row.model_mappings) + or ('"storage_backend" IS NOT NULL' in query and row.storage_backend is not None) + ] + return [ + { + "unified_file_id": row.unified_file_id, + "file_object": row.file_object, + "model_mappings": row.model_mappings, + "storage_backend": row.storage_backend, + } + for row in listed_rows[: params[-1]] + ] + def _setup_unscoped_list_files_route_over_real_hook( mocker, monkeypatch, llm_router: Router, rows @@ -5041,7 +5080,9 @@ def _setup_unscoped_list_files_route_over_real_hook( managed_files = PROXY_LiteLLMManagedFiles( internal_usage_cache=MagicMock(), prisma_client=MagicMock() ) - managed_files.prisma_client.db.litellm_managedfiletable = _ManagedFileTableOverRows(rows) + managed_file_table = _ManagedFileTableOverRows(rows) + managed_files.prisma_client.db.litellm_managedfiletable = managed_file_table + managed_files.prisma_client.db.query_raw = managed_file_table.query_raw proxy_logging_obj = setup_proxy_logging_object(monkeypatch, llm_router) proxy_logging_obj.proxy_hook_mapping["managed_files"] = managed_files @@ -6263,3 +6304,513 @@ def test_create_file_passthrough_fails_closed_when_guardrails_would_scan_the_bat assert error["param"] == "passthrough" assert "guardrails" in error["message"] assert forwarded_calls == [] + + +def _callers_files_on_every_deployment() -> list[MagicMock]: + return [ + _managed_file_row("unified-gpt-newer", model_mappings={"gpt-3.5-turbo-id": "file-gpt-newer"}), + _managed_file_row("unified-azure", model_mappings={"azure-gpt-3-5-turbo-id": "file-azure"}), + _managed_file_row("unified-gemini", model_mappings={"gemini-2.0-flash-id": "file-gemini"}), + _managed_file_row( + "unified-theirs", created_by="other-user", model_mappings={"gpt-3.5-turbo-id": "file-theirs"} + ), + _managed_file_row("unified-gpt-older", model_mappings={"gpt-3.5-turbo-id": "file-gpt-older"}), + ] + + +def _list_files_as_test_user(path: str, params: dict[str, str], headers: dict[str, str]) -> httpx.Response: + import litellm.proxy.proxy_server as ps + + try: + return client.get(path, params=params, headers={"Authorization": "Bearer test-key", **headers}) + finally: + app.dependency_overrides.pop(ps.user_api_key_auth, None) + + +@pytest.mark.parametrize( + ("path", "params", "headers", "expected_ids"), + [ + ("/openai/v1/files", {}, {}, ["unified-gpt-newer", "unified-gpt-older"]), + ("/v1/files", {"provider": "azure"}, {}, ["unified-azure"]), + ("/v1/files", {}, {"custom-llm-provider": "gemini"}, ["unified-gemini"]), + ("/v1/files", {"target_model_names": "gemini-2.0-flash"}, {}, ["unified-gemini"]), + ("/v1/files", {}, {"x-litellm-model": "gpt-3.5-turbo"}, ["unified-gpt-newer", "unified-gpt-older"]), + ("/v1/files", {}, {}, ["unified-gpt-newer", "unified-azure", "unified-gemini", "unified-gpt-older"]), + ], + ids=["provider-path", "provider-query", "provider-header", "target-model-names", "model-header", "no-provider"], +) +def test_require_managed_files_lists_the_callers_files_on_the_named_provider_or_model( + mocker: MockerFixture, + monkeypatch: pytest.MonkeyPatch, + llm_router: Router, + path: str, + params: dict[str, str], + headers: dict[str, str], + expected_ids: list[str], +) -> None: + _, provider_list = _setup_unscoped_list_files_route_over_real_hook( + mocker, monkeypatch, llm_router, _callers_files_on_every_deployment() + ) + monkeypatch.setattr("litellm.require_managed_files", True) + + response: Final = _list_files_as_test_user(path, params, headers) + + assert response.status_code == 200, response.text + assert [file["id"] for file in response.json()["data"]] == expected_ids + provider_list.assert_not_awaited() + + +def test_require_managed_files_lists_a_teammates_files_on_the_provider_and_no_other_teams( + mocker: MockerFixture, monkeypatch: pytest.MonkeyPatch, llm_router: Router +) -> None: + import litellm.proxy.proxy_server as ps + from litellm.proxy._types import LitellmUserRoles + + _setup_unscoped_list_files_route_over_real_hook( + mocker, + monkeypatch, + llm_router, + [ + _managed_file_row( + "unified-teammate", created_by="teammate", team_id="team-a", model_mappings={"gpt-3.5-turbo-id": "f-1"} + ), + _managed_file_row("unified-own", model_mappings={"gpt-3.5-turbo-id": "f-2"}), + _managed_file_row( + "unified-other-team", created_by="outsider", team_id="team-b", model_mappings={"gpt-3.5-turbo-id": "f-3"} + ), + ], + ) + app.dependency_overrides[ps.user_api_key_auth] = lambda: UserAPIKeyAuth( + api_key="test-key", user_role=LitellmUserRoles.INTERNAL_USER, user_id="test-user", team_id="team-a" + ) + monkeypatch.setattr("litellm.require_managed_files", True) + + response: Final = _list_files_as_test_user("/openai/v1/files", {}, {}) + + assert response.status_code == 200, response.text + assert [file["id"] for file in response.json()["data"]] == ["unified-teammate", "unified-own"] + + +def _router_with_wildcard_and_unprefixed_anthropic_deployments() -> Router: + unprefixed_anthropic_model: Final = next(model for model in sorted(litellm.anthropic_models) if "/" not in model) + return Router( + model_list=[ + { + "model_name": "gpt-3.5-turbo", + "litellm_params": {"model": "openai/gpt-3.5-turbo", "api_key": "openai_api_key"}, + "model_info": {"id": "gpt-3.5-turbo-id"}, + }, + { + "model_name": "anthropic/*", + "litellm_params": {"model": "anthropic/*", "api_key": "anthropic_api_key"}, + "model_info": {"id": "anthropic-wildcard-id"}, + }, + { + "model_name": "claude-unprefixed", + "litellm_params": {"model": unprefixed_anthropic_model, "api_key": "anthropic_api_key"}, + "model_info": {"id": "claude-unprefixed-id"}, + }, + ], + model_group_alias={"claude-alias": "claude-unprefixed"}, + routing_groups=[ + {"group_name": "claude-group", "models": ["claude-unprefixed"], "routing_strategy": "simple-shuffle"} + ], + ) + + +def _callers_files_on_wildcard_unprefixed_and_passthrough_deployments() -> list[MagicMock]: + return [ + _managed_file_row("unified-wildcard", model_mappings={"anthropic-wildcard-id": "file-wildcard"}), + _managed_file_row("unified-unprefixed", model_mappings={"claude-unprefixed-id": "file-unprefixed"}), + _managed_file_row("unified-gpt", model_mappings={"gpt-3.5-turbo-id": "file-gpt"}), + _managed_file_row( + "unified-passthrough", + model_mappings={ + "_passthrough_openai": "file-passthrough", + "_passthrough_provider_marker": "_passthrough_provider:openai", + }, + ), + _managed_file_row( + "unified-storage", + model_mappings={"claude-unprefixed": "litellm_db://unified-storage"}, + storage_backend="litellm_db", + ), + _managed_file_row( + "unified-wildcard-storage", + model_mappings={"anthropic/claude-haiku-5-5": "litellm_db://unified-wildcard-storage"}, + storage_backend="litellm_db", + ), + _managed_file_row( + "unified-alias-storage", + model_mappings={"claude-alias": "s3://bucket/unified-alias-storage"}, + storage_backend="s3", + ), + _managed_file_row( + "unified-group-storage", + model_mappings={"claude-group": "s3://bucket/unified-group-storage"}, + storage_backend="s3", + ), + ] + + +@pytest.mark.parametrize( + ("path", "params", "headers", "expected_ids"), + [ + ( + "/v1/files", + {"target_model_names": "anthropic/claude-haiku-5-5"}, + {}, + ["unified-wildcard", "unified-wildcard-storage"], + ), + ( + "/v1/files", + {}, + {"x-litellm-model": "anthropic/claude-haiku-5-5"}, + ["unified-wildcard", "unified-wildcard-storage"], + ), + ( + "/anthropic/v1/files", + {}, + {}, + [ + "unified-wildcard", + "unified-unprefixed", + "unified-storage", + "unified-wildcard-storage", + "unified-alias-storage", + "unified-group-storage", + ], + ), + ("/openai/v1/files", {}, {}, ["unified-gpt", "unified-passthrough"]), + ("/v1/files", {"target_model_names": "claude-unprefixed"}, {}, ["unified-unprefixed", "unified-storage"]), + ], + ids=[ + "wildcard-target-model-names", + "wildcard-model-header", + "unprefixed-provider", + "passthrough-provider", + "storage-backed-model", + ], +) +def test_require_managed_files_lists_the_callers_files_wherever_the_router_put_them( + mocker: MockerFixture, + monkeypatch: pytest.MonkeyPatch, + path: str, + params: dict[str, str], + headers: dict[str, str], + expected_ids: list[str], +) -> None: + _, provider_list = _setup_unscoped_list_files_route_over_real_hook( + mocker, + monkeypatch, + _router_with_wildcard_and_unprefixed_anthropic_deployments(), + _callers_files_on_wildcard_unprefixed_and_passthrough_deployments(), + ) + monkeypatch.setattr("litellm.require_managed_files", True) + + response: Final = _list_files_as_test_user(path, params, headers) + + assert response.status_code == 200, response.text + assert [file["id"] for file in response.json()["data"]] == expected_ids + provider_list.assert_not_awaited() + + +def _router_with_a_catch_all_openai_deployment() -> Router: + unprefixed_anthropic_model: Final = next(model for model in sorted(litellm.anthropic_models) if "/" not in model) + return Router( + model_list=[ + { + "model_name": "gpt-3.5-turbo", + "litellm_params": {"model": "openai/gpt-3.5-turbo", "api_key": "openai_api_key"}, + "model_info": {"id": "gpt-3.5-turbo-id"}, + }, + { + "model_name": "*", + "litellm_params": {"model": "openai/*", "api_key": "openai_api_key"}, + "model_info": {"id": "catch-all-id"}, + }, + { + "model_name": "claude-unprefixed", + "litellm_params": {"model": unprefixed_anthropic_model, "api_key": "anthropic_api_key"}, + "model_info": {"id": "claude-unprefixed-id"}, + }, + ] + ) + + +@pytest.mark.parametrize( + ("path", "expected_ids"), + [ + ("/openai/v1/files", ["unified-gpt", "unified-catch-all-storage"]), + ("/anthropic/v1/files", ["unified-unprefixed", "unified-storage"]), + ], + ids=["catch-all-provider", "explicit-provider"], +) +def test_require_managed_files_keeps_a_catch_all_route_from_claiming_files_on_other_providers( + mocker: MockerFixture, + monkeypatch: pytest.MonkeyPatch, + path: str, + expected_ids: list[str], +) -> None: + _, provider_list = _setup_unscoped_list_files_route_over_real_hook( + mocker, + monkeypatch, + _router_with_a_catch_all_openai_deployment(), + [ + _managed_file_row("unified-unprefixed", model_mappings={"claude-unprefixed-id": "file-unprefixed"}), + _managed_file_row("unified-gpt", model_mappings={"gpt-3.5-turbo-id": "file-gpt"}), + _managed_file_row( + "unified-storage", + model_mappings={"claude-unprefixed": "litellm_db://unified-storage"}, + storage_backend="litellm_db", + ), + _managed_file_row( + "unified-catch-all-storage", + model_mappings={"model-only-the-catch-all-serves": "litellm_db://unified-catch-all-storage"}, + storage_backend="litellm_db", + ), + ], + ) + monkeypatch.setattr("litellm.require_managed_files", True) + + response: Final = _list_files_as_test_user(path, {}, {}) + + assert response.status_code == 200, response.text + assert [file["id"] for file in response.json()["data"]] == expected_ids + provider_list.assert_not_awaited() + + +@pytest.mark.parametrize( + ("path", "params"), + [ + ("/antropic/v1/files", {}), + ("/openai/v1/files", {"after": "unified-gpt", "after_id": "unified-wildcard"}), + ("/v1/files", {"target_model_names": "gpt-3.5-turbo,claude-unprefixed"}), + ], + ids=["unknown-provider", "conflicting-after-cursors", "two-target-model-names"], +) +def test_require_managed_files_rejects_a_list_it_cannot_scope( + mocker: MockerFixture, monkeypatch: pytest.MonkeyPatch, path: str, params: dict[str, str] +) -> None: + _, provider_list = _setup_unscoped_list_files_route_over_real_hook( + mocker, + monkeypatch, + _router_with_wildcard_and_unprefixed_anthropic_deployments(), + _callers_files_on_wildcard_unprefixed_and_passthrough_deployments(), + ) + monkeypatch.setattr("litellm.require_managed_files", True) + + response: Final = _list_files_as_test_user(path, params, {}) + + assert response.status_code == 400, response.text + provider_list.assert_not_awaited() + + +@pytest.mark.parametrize("cursor_param", ["after", "after_id"]) +def test_require_managed_files_pages_a_provider_list_with_either_cursor_param( + mocker: MockerFixture, monkeypatch: pytest.MonkeyPatch, llm_router: Router, cursor_param: str +) -> None: + import litellm.proxy.proxy_server as ps + + _setup_unscoped_list_files_route_over_real_hook( + mocker, monkeypatch, llm_router, _callers_files_on_every_deployment() + ) + monkeypatch.setattr("litellm.require_managed_files", True) + + try: + first_page: Final = client.get( + "/openai/v1/files", params={"limit": "1"}, headers={"Authorization": "Bearer test-key"} + ) + second_page: Final = client.get( + "/openai/v1/files", + params={"limit": "1", cursor_param: first_page.json()["last_id"]}, + headers={"Authorization": "Bearer test-key"}, + ) + finally: + app.dependency_overrides.pop(ps.user_api_key_auth, None) + + assert (first_page.status_code, second_page.status_code) == (200, 200), (first_page.text, second_page.text) + assert [file["id"] for file in first_page.json()["data"]] == ["unified-gpt-newer"] + assert first_page.json()["has_more"] is True + assert [file["id"] for file in second_page.json()["data"]] == ["unified-gpt-older"] + assert second_page.json()["has_more"] is False + + +def test_require_managed_files_pages_toward_newer_files_with_before_id( + mocker: MockerFixture, monkeypatch: pytest.MonkeyPatch, llm_router: Router +) -> None: + import litellm.proxy.proxy_server as ps + + _setup_unscoped_list_files_route_over_real_hook( + mocker, monkeypatch, llm_router, _callers_files_on_every_deployment() + ) + monkeypatch.setattr("litellm.require_managed_files", True) + + try: + page_before_oldest: Final = client.get( + "/v1/files", + params={"limit": "2", "before_id": "unified-gpt-older"}, + headers={"Authorization": "Bearer test-key"}, + ) + newest_page: Final = client.get( + "/v1/files", + params={"limit": "2", "before_id": page_before_oldest.json()["first_id"]}, + headers={"Authorization": "Bearer test-key"}, + ) + finally: + app.dependency_overrides.pop(ps.user_api_key_auth, None) + + assert (page_before_oldest.status_code, newest_page.status_code) == (200, 200), ( + page_before_oldest.text, + newest_page.text, + ) + assert [file["id"] for file in page_before_oldest.json()["data"]] == ["unified-azure", "unified-gemini"] + assert page_before_oldest.json()["has_more"] is True + assert [file["id"] for file in newest_page.json()["data"]] == ["unified-gpt-newer"] + assert newest_page.json()["has_more"] is False + + +def test_managed_file_list_rejects_both_cursor_directions( + mocker: MockerFixture, monkeypatch: pytest.MonkeyPatch, llm_router: Router +) -> None: + _setup_unscoped_list_files_route_over_real_hook( + mocker, monkeypatch, llm_router, _callers_files_on_every_deployment() + ) + monkeypatch.setattr("litellm.require_managed_files", True) + + response: Final = _list_files_as_test_user( + "/v1/files", {"after": "unified-gpt-newer", "before_id": "unified-gpt-older"}, {} + ) + + assert response.status_code == 400, response.text + assert "not both" in response.json()["error"]["message"] + + +@pytest.mark.parametrize( + ("params", "headers"), + [({"target_model_names": "gemini-2.0-flash"}, {}), ({}, {"x-litellm-model": "gemini-2.0-flash"})], + ids=["target-model-names", "model-header"], +) +def test_require_managed_files_list_rejects_a_model_the_key_cannot_use( + mocker: MockerFixture, + monkeypatch: pytest.MonkeyPatch, + llm_router: Router, + params: dict[str, str], + headers: dict[str, str], +) -> None: + import litellm.proxy.proxy_server as ps + + _setup_unscoped_list_files_route_over_real_hook( + mocker, monkeypatch, llm_router, _callers_files_on_every_deployment() + ) + app.dependency_overrides[ps.user_api_key_auth] = lambda: UserAPIKeyAuth( + api_key="test-key", + user_role=LitellmUserRoles.INTERNAL_USER, + user_id="test-user", + models=["gpt-3.5-turbo"], + ) + monkeypatch.setattr("litellm.require_managed_files", True) + + response: Final = _list_files_as_test_user("/v1/files", params, headers) + + assert response.status_code == 403, response.text + assert "unified-gemini" not in response.text + + +def test_require_managed_files_list_fails_closed_without_the_managed_files_hook( + mocker: MockerFixture, monkeypatch: pytest.MonkeyPatch, llm_router: Router +) -> None: + proxy_logging_obj, provider_list = _setup_unscoped_list_files_route_over_real_hook( + mocker, monkeypatch, llm_router, _callers_files_on_every_deployment() + ) + proxy_logging_obj.proxy_hook_mapping.pop("managed_files") + provider_list.return_value = [_managed_file("file-on-the-shared-key")] + monkeypatch.setattr("litellm.require_managed_files", True) + + response: Final = _list_files_as_test_user("/openai/v1/files", {}, {}) + + assert response.status_code == 500, response.text + assert "file-on-the-shared-key" not in response.text + assert "require_managed_files cannot scope the list to the caller" in response.json()["error"]["message"] + provider_list.assert_not_awaited() + + +@pytest.mark.parametrize("hook_loaded", [True, False], ids=["hook-loaded", "no-hook"]) +def test_require_managed_files_keeps_the_provider_list_for_an_admin( + mocker: MockerFixture, + monkeypatch: pytest.MonkeyPatch, + llm_router: Router, + hook_loaded: bool, +) -> None: + import litellm.proxy.proxy_server as ps + + proxy_logging_obj, provider_list = _setup_unscoped_list_files_route_over_real_hook( + mocker, monkeypatch, llm_router, _callers_files_on_every_deployment() + ) + if not hook_loaded: + proxy_logging_obj.proxy_hook_mapping.pop("managed_files") + provider_list.return_value = [_managed_file("file-on-the-shared-key")] + app.dependency_overrides[ps.user_api_key_auth] = lambda: UserAPIKeyAuth( + api_key="test-key", user_role=LitellmUserRoles.PROXY_ADMIN, user_id="admin-user" + ) + monkeypatch.setattr("litellm.require_managed_files", True) + + response: Final = _list_files_as_test_user("/anthropic/v1/files", {}, {}) + + assert response.status_code == 200, response.text + assert [file["id"] for file in response.json()["data"]] == ["file-on-the-shared-key"] + + +def test_managed_file_list_without_require_managed_files_ignores_cursor_params_openai_does_not_define( + mocker: MockerFixture, monkeypatch: pytest.MonkeyPatch, llm_router: Router +) -> None: + _setup_unscoped_list_files_route_over_real_hook( + mocker, monkeypatch, llm_router, _callers_files_on_every_deployment() + ) + monkeypatch.setattr("litellm.require_managed_files", False) + + response: Final = _list_files_as_test_user( + "/v1/files", {"after": "unified-gpt-newer", "after_id": "unified-gemini", "before_id": "file-raw"}, {} + ) + + assert response.status_code == 200, response.text + assert [file["id"] for file in response.json()["data"]] == ["unified-azure", "unified-gemini", "unified-gpt-older"] + + +@pytest.mark.parametrize( + ("path", "headers", "expected_count", "has_more"), + [ + ("/anthropic/v1/files", {}, ANTHROPIC_DEFAULT_FILE_LIST_LIMIT, True), + ("/v1/files", {"x-litellm-model": "claude-unprefixed"}, ANTHROPIC_DEFAULT_FILE_LIST_LIMIT, True), + ("/anthropic/v1/files", {"x-litellm-model": "gpt-3.5-turbo"}, ANTHROPIC_DEFAULT_FILE_LIST_LIMIT + 1, False), + ("/v1/files", {}, 2 * (ANTHROPIC_DEFAULT_FILE_LIST_LIMIT + 1), False), + ], + ids=["anthropic-provider", "anthropic-model-header", "openai-model-header", "every-provider"], +) +def test_require_managed_files_pages_an_anthropic_list_at_the_anthropic_default_size( + mocker: MockerFixture, + monkeypatch: pytest.MonkeyPatch, + path: str, + headers: dict[str, str], + expected_count: int, + has_more: bool, +) -> None: + _setup_unscoped_list_files_route_over_real_hook( + mocker, + monkeypatch, + _router_with_wildcard_and_unprefixed_anthropic_deployments(), + [ + _managed_file_row( + f"unified-{deployment_id}-{index:03d}", model_mappings={deployment_id: f"file-{index:03d}"} + ) + for deployment_id, index in itertools.product( + ("claude-unprefixed-id", "gpt-3.5-turbo-id"), range(ANTHROPIC_DEFAULT_FILE_LIST_LIMIT + 1) + ) + ], + ) + monkeypatch.setattr("litellm.require_managed_files", True) + + response: Final = _list_files_as_test_user(path, {}, headers) + + assert response.status_code == 200, response.text + assert (len(response.json()["data"]), response.json()["has_more"]) == (expected_count, has_more) diff --git a/ui/litellm-dashboard/src/lib/http/schema.d.ts b/ui/litellm-dashboard/src/lib/http/schema.d.ts index 45f9630983f..70406105193 100644 --- a/ui/litellm-dashboard/src/lib/http/schema.d.ts +++ b/ui/litellm-dashboard/src/lib/http/schema.d.ts @@ -59857,6 +59857,8 @@ export interface operations { purpose?: string | null; limit?: number | null; after?: string | null; + after_id?: string | null; + before_id?: string | null; }; header?: never; path?: never; @@ -79053,6 +79055,8 @@ export interface operations { purpose?: string | null; limit?: number | null; after?: string | null; + after_id?: string | null; + before_id?: string | null; }; header?: never; path?: never; @@ -86388,6 +86392,8 @@ export interface operations { purpose?: string | null; limit?: number | null; after?: string | null; + after_id?: string | null; + before_id?: string | null; }; header?: never; path: { From cf7c35a1174eb9864cf35c1513c2d6d50b5df26a Mon Sep 17 00:00:00 2001 From: moe-berri Date: Wed, 28 Oct 2026 13:48:37 -0700 Subject: [PATCH 02/41] fix(anthropic): map model_context_window_exceeded stop reason to length (#48874) * test(anthropic): cover model_context_window_exceeded finish reason on chat and responses Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * fix(anthropic): map model_context_window_exceeded stop reason to length Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * fix(rust): mirror model_context_window_exceeded finish reason mapping Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> --------- Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- .../crates/core-utils/src/core_helpers.rs | 2 + litellm/litellm_core_utils/core_helpers.py | 1 + ...test_anthropic_context_window_stop_wire.py | 122 ++++++++++++++++++ .../litellm_core_utils/test_core_helpers.py | 1 + 4 files changed, 126 insertions(+) create mode 100644 tests/integration/providers/test_anthropic_context_window_stop_wire.py diff --git a/litellm-rust/crates/core-utils/src/core_helpers.rs b/litellm-rust/crates/core-utils/src/core_helpers.rs index ada1c3ceb1a..58ee88d301c 100644 --- a/litellm-rust/crates/core-utils/src/core_helpers.rs +++ b/litellm-rust/crates/core-utils/src/core_helpers.rs @@ -13,6 +13,7 @@ const FINISH_REASONS: &[(&str, &str)] = &[ ("max_tokens", "length"), ("refusal", "content_filter"), ("compaction", "length"), + ("model_context_window_exceeded", "length"), ("guardrail_intervened", "content_filter"), ("content_filtered", "content_filter"), ("content_filter", "content_filter"), @@ -74,6 +75,7 @@ mod tests { #[case::end_turn("end_turn", "stop")] #[case::stop_sequence("stop_sequence", "stop")] #[case::max_tokens("max_tokens", "length")] + #[case::model_context_window_exceeded("model_context_window_exceeded", "length")] #[case::refusal("refusal", "content_filter")] #[case::guardrail_intervened("guardrail_intervened", "content_filter")] #[case::content_filtered("content_filtered", "content_filter")] diff --git a/litellm/litellm_core_utils/core_helpers.py b/litellm/litellm_core_utils/core_helpers.py index 00c6ce97e78..55025508e40 100644 --- a/litellm/litellm_core_utils/core_helpers.py +++ b/litellm/litellm_core_utils/core_helpers.py @@ -266,6 +266,7 @@ _FINISH_REASON_MAP: Final[dict[str, OpenAIChatCompletionFinishReason]] = { "tool_use": "tool_calls", "refusal": "content_filter", "compaction": "length", + "model_context_window_exceeded": "length", # Cohere "COMPLETE": "stop", "ERROR_TOXIC": "content_filter", diff --git a/tests/integration/providers/test_anthropic_context_window_stop_wire.py b/tests/integration/providers/test_anthropic_context_window_stop_wire.py new file mode 100644 index 00000000000..38e3e7dffcd --- /dev/null +++ b/tests/integration/providers/test_anthropic_context_window_stop_wire.py @@ -0,0 +1,122 @@ +import json +import uuid +from collections.abc import Mapping +from typing import Final + +import pytest +from integration._support.client import Gateway +from integration._support.wire import Reply, Request, wire_server +from pydantic import JsonValue + +MODEL: Final = "claude-haiku-4-5" + + +def _sse(event: Mapping[str, JsonValue]) -> bytes: + return f"event: {event['type']}\ndata: {json.dumps(event)}\n\n".encode() + + +def _message(identity: str, stop_reason: str | None) -> dict[str, JsonValue]: + return { + "id": "msg_" + identity, + "type": "message", + "role": "assistant", + "model": MODEL, + "content": [{"type": "text", "text": "alpha bravo"}], + "stop_reason": stop_reason, + "stop_sequence": None, + "usage": {"input_tokens": 199878, "output_tokens": 12799}, + } + + +def _stream(identity: str, stop_reason: str) -> tuple[bytes, ...]: + return ( + _sse({"type": "message_start", "message": {**_message(identity, None), "content": []}}), + _sse({"type": "content_block_start", "index": 0, "content_block": {"type": "text", "text": ""}}), + _sse({"type": "content_block_delta", "index": 0, "delta": {"type": "text_delta", "text": "alpha bravo"}}), + _sse({"type": "content_block_stop", "index": 0}), + _sse( + { + "type": "message_delta", + "delta": {"stop_reason": stop_reason, "stop_sequence": None}, + "usage": {"output_tokens": 12799}, + } + ), + _sse({"type": "message_stop"}), + ) + + +def _anthropic(identity: str, stop_reason: str): + def respond(request: Request) -> Reply: + assert request.method == "POST" and request.target == "/v1/messages", request.target + if json.loads(request.body).get("stream"): + return Reply(content_type="text/event-stream", chunks=_stream(identity, stop_reason)) + return Reply(body=json.dumps(_message(identity, stop_reason)).encode()) + + return respond + + +def _events(gateway: Gateway, path: str, body: Mapping[str, JsonValue]) -> tuple[dict[str, JsonValue], ...]: + with gateway.client.stream( + "POST", path, json={**body, "stream": True}, headers={"Authorization": f"Bearer {gateway.key}"} + ) as response: + assert response.status_code == 200, response.read().decode() + return tuple( + json.loads(line.removeprefix("data: ")) + for line in response.iter_lines() + if line.startswith("data: ") and line != "data: [DONE]" + ) + + +def _chat_finish_reason(gateway: Gateway, model: str, *, stream: bool) -> JsonValue: + body: Final = {"model": model, "max_tokens": 64000, "messages": [{"role": "user", "content": "copy the list"}]} + if stream: + reasons: Final = tuple( + choice["finish_reason"] + for chunk in _events(gateway, "/v1/chat/completions", body) + for choice in chunk["choices"] # comprehension-ok: chunk -> choices flatten + if choice.get("finish_reason") + ) + assert len(reasons) == 1, reasons + return reasons[0] + response: Final = gateway.request("POST", "/v1/chat/completions", body) + assert response.status_code == 200, response.text + return response.json()["choices"][0]["finish_reason"] + + +def _responses_status(gateway: Gateway, model: str, *, stream: bool) -> JsonValue: + body: Final = {"model": model, "max_output_tokens": 64000, "input": "copy the list"} + if stream: + terminal: Final = tuple( + event["response"] + for event in _events(gateway, "/v1/responses", body) + if event["type"] in ("response.completed", "response.incomplete", "response.failed") + ) + assert len(terminal) == 1, terminal + return terminal[0]["status"] + response: Final = gateway.request("POST", "/v1/responses", body) + assert response.status_code == 200, response.text + return response.json()["status"] + + +@pytest.mark.parametrize("stream", [False, True], ids=["json", "stream"]) +@pytest.mark.parametrize( + ("stop_reason", "finish_reason"), + [("model_context_window_exceeded", "length"), ("max_tokens", "length"), ("end_turn", "stop")], +) +def test_chat_maps_anthropic_stop_reason(gateway: Gateway, stop_reason: str, finish_reason: str, stream: bool) -> None: + identity: Final = uuid.uuid4().hex + with wire_server(_anthropic(identity, stop_reason)) as wire, gateway.scenario() as scenario: + model: Final = scenario.model(model=f"anthropic/{MODEL}", api_base=wire.url, api_key="scripted-anthropic-key") + assert _chat_finish_reason(gateway, model, stream=stream) == finish_reason + + +@pytest.mark.parametrize("stream", [False, True], ids=["json", "stream"]) +@pytest.mark.parametrize( + ("stop_reason", "status"), + [("model_context_window_exceeded", "incomplete"), ("max_tokens", "incomplete"), ("end_turn", "completed")], +) +def test_responses_maps_anthropic_stop_reason(gateway: Gateway, stop_reason: str, status: str, stream: bool) -> None: + identity: Final = uuid.uuid4().hex + with wire_server(_anthropic(identity, stop_reason)) as wire, gateway.scenario() as scenario: + model: Final = scenario.model(model=f"anthropic/{MODEL}", api_base=wire.url, api_key="scripted-anthropic-key") + assert _responses_status(gateway, model, stream=stream) == status diff --git a/tests/unit/litellm_core_utils/test_core_helpers.py b/tests/unit/litellm_core_utils/test_core_helpers.py index d66c37ba8c2..66a8e6bbda6 100644 --- a/tests/unit/litellm_core_utils/test_core_helpers.py +++ b/tests/unit/litellm_core_utils/test_core_helpers.py @@ -256,6 +256,7 @@ class TestMapFinishReasonAnthropic: ("max_tokens", "length"), ("tool_use", "tool_calls"), ("compaction", "length"), + ("model_context_window_exceeded", "length"), ("content_filtered", "content_filter"), ], ) From 4516a251c940cc6ff7977b460a26a0582a0a18ff Mon Sep 17 00:00:00 2001 From: "devin-ai-integration[bot]" <158243242+devin-ai-integration[bot]@users.noreply.github.com> Date: Wed, 28 Oct 2026 14:39:25 -0700 Subject: [PATCH 03/41] feat(accounting): add accounting contracts, PricingSnapshot and ledger schema (LIT-10193) (#48862) * feat(accounting): add typed contracts and spend log adapters * refactor(accounting): reuse spend identity constants and register tests * refactor(accounting): remove unused spend log conversion API * feat(accounting): add PricingSnapshot, delivery enums and projection dimensions Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * feat(db): add LiteLLM_AccountingEvent and LiteLLM_AccountingDelivery ledger tables Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * feat(accounting): export JSON schema and semantic fixtures for the Rust cost crate Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * refactor(accounting): drop the schema and fixture export, the Python package is the contract Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * test(accounting): make tests/unit/accounting a package Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * fix(accounting): validate pricing snapshots and ledger rows * fix(otel): register the accounting ledger tables as span tables Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * fix(accounting): address review findings on dimensions, token subsets, multipliers and agent target Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> --------- Co-authored-by: Joshua Valluru <326636767+joshua-berri@users.noreply.github.com> Co-authored-by: joshua Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- .github/workflows/test-unit.yml | 1 + .../migration.sql | 47 ++ .../litellm_proxy_extras/schema.prisma | 70 +++ litellm/accounting/__init__.py | 62 +++ litellm/accounting/measurements.py | 13 + litellm/accounting/protocols.py | 85 +++ litellm/accounting/types.py | 428 ++++++++++++++++ litellm/integrations/otel/model/spans.py | 2 + litellm/proxy/schema.prisma | 70 +++ schema.prisma | 70 +++ .../database/test_accounting_ledger.py | 128 +++++ tests/unit/accounting/__init__.py | 0 tests/unit/accounting/test_measurements.py | 235 +++++++++ tests/unit/accounting/test_types.py | 484 ++++++++++++++++++ 14 files changed, 1695 insertions(+) create mode 100644 litellm-proxy-extras/litellm_proxy_extras/migrations/20261027000000_add_accounting_ledger_tables/migration.sql create mode 100644 litellm/accounting/__init__.py create mode 100644 litellm/accounting/measurements.py create mode 100644 litellm/accounting/protocols.py create mode 100644 litellm/accounting/types.py create mode 100644 tests/integration/database/test_accounting_ledger.py create mode 100644 tests/unit/accounting/__init__.py create mode 100644 tests/unit/accounting/test_measurements.py create mode 100644 tests/unit/accounting/test_types.py diff --git a/.github/workflows/test-unit.yml b/.github/workflows/test-unit.yml index e3980bd8f3b..2ddf21bca17 100644 --- a/.github/workflows/test-unit.yml +++ b/.github/workflows/test-unit.yml @@ -137,6 +137,7 @@ jobs: include: - shard: core-utils test-path: |- + tests/unit/accounting tests/unit/batch_completion tests/unit/decisions tests/unit/extraction diff --git a/litellm-proxy-extras/litellm_proxy_extras/migrations/20261027000000_add_accounting_ledger_tables/migration.sql b/litellm-proxy-extras/litellm_proxy_extras/migrations/20261027000000_add_accounting_ledger_tables/migration.sql new file mode 100644 index 00000000000..c0ad8692d36 --- /dev/null +++ b/litellm-proxy-extras/litellm_proxy_extras/migrations/20261027000000_add_accounting_ledger_tables/migration.sql @@ -0,0 +1,47 @@ +SET LOCAL lock_timeout = '10s'; + +CREATE TYPE "AccountingEventKind" AS ENUM ('inbound_usage', 'outbound_usage', 'cost'); + +CREATE TYPE "AccountingDeliveryState" AS ENUM ('pending', 'applied', 'parked'); + +CREATE TYPE "AccountingDeliveryTarget" AS ENUM ('LiteLLM_VerificationToken', 'LiteLLM_UserTable', 'LiteLLM_TeamTable', 'LiteLLM_TeamMembership', 'LiteLLM_OrganizationTable', 'LiteLLM_OrganizationMembership', 'LiteLLM_EndUserTable', 'LiteLLM_AgentsTable', 'LiteLLM_TagTable', 'LiteLLM_ProjectTable', 'LiteLLM_ModelAccessGroupBudgetTable', 'LiteLLM_BudgetTable', 'LiteLLM_BudgetWindowSpend', 'LiteLLM_DailyUserSpend', 'LiteLLM_DailyTeamSpend', 'LiteLLM_DailyOrganizationSpend', 'LiteLLM_DailyEndUserSpend', 'LiteLLM_DailyAgentSpend', 'LiteLLM_DailyTagSpend', 'LiteLLM_DailyModelUsage', 'LiteLLM_DailyToolSpend', 'LiteLLM_AutoRouterDailySpend'); + +-- CreateTable +CREATE TABLE IF NOT EXISTS "LiteLLM_AccountingEvent" ( + "event_id" TEXT NOT NULL, + "kind" "AccountingEventKind" NOT NULL, + "source_event_id" TEXT, + "schema_version" INTEGER NOT NULL, + "payload_hash" TEXT NOT NULL, + "occurred_at" TIMESTAMP(3) NOT NULL, + "recorded_at" TIMESTAMP(3) NOT NULL DEFAULT CURRENT_TIMESTAMP, + "payload" JSONB NOT NULL, + + CONSTRAINT "LiteLLM_AccountingEvent_schema_version_check" CHECK ("schema_version" > 0), + CONSTRAINT "LiteLLM_AccountingEvent_pkey" PRIMARY KEY ("event_id") +); + +-- CreateTable +CREATE TABLE IF NOT EXISTS "LiteLLM_AccountingDelivery" ( + "event_id" TEXT NOT NULL, + "target" "AccountingDeliveryTarget" NOT NULL, + "version" INTEGER NOT NULL, + "state" "AccountingDeliveryState" NOT NULL DEFAULT 'pending', + "attempts" INTEGER NOT NULL DEFAULT 0, + "next_attempt_at" TIMESTAMP(3) NOT NULL DEFAULT CURRENT_TIMESTAMP, + "applied_at" TIMESTAMP(3), + "last_error" TEXT, + + CONSTRAINT "LiteLLM_AccountingDelivery_version_check" CHECK ("version" > 0), + CONSTRAINT "LiteLLM_AccountingDelivery_attempts_check" CHECK ("attempts" >= 0), + CONSTRAINT "LiteLLM_AccountingDelivery_pkey" PRIMARY KEY ("event_id","target","version") +); + +-- CreateIndex +CREATE INDEX IF NOT EXISTS "LiteLLM_AccountingEvent_recorded_at_idx" ON "LiteLLM_AccountingEvent"("recorded_at"); + +-- CreateIndex +CREATE INDEX IF NOT EXISTS "LiteLLM_AccountingEvent_kind_source_event_id_idx" ON "LiteLLM_AccountingEvent"("kind", "source_event_id"); + +-- CreateIndex +CREATE INDEX IF NOT EXISTS "LiteLLM_AccountingDelivery_target_state_next_attempt_at_idx" ON "LiteLLM_AccountingDelivery"("target", "state", "next_attempt_at"); diff --git a/litellm-proxy-extras/litellm_proxy_extras/schema.prisma b/litellm-proxy-extras/litellm_proxy_extras/schema.prisma index 4187f76c97c..2965174114b 100644 --- a/litellm-proxy-extras/litellm_proxy_extras/schema.prisma +++ b/litellm-proxy-extras/litellm_proxy_extras/schema.prisma @@ -2061,3 +2061,73 @@ model LiteLLM_TelemetryReport { @@index([window_end]) @@index([created_at]) } + +enum AccountingEventKind { + inbound_usage + outbound_usage + cost +} + +enum AccountingDeliveryState { + pending + applied + parked +} + +enum AccountingDeliveryTarget { + LiteLLM_VerificationToken + LiteLLM_UserTable + LiteLLM_TeamTable + LiteLLM_TeamMembership + LiteLLM_OrganizationTable + LiteLLM_OrganizationMembership + LiteLLM_EndUserTable + LiteLLM_AgentsTable + LiteLLM_TagTable + LiteLLM_ProjectTable + LiteLLM_ModelAccessGroupBudgetTable + LiteLLM_BudgetTable + LiteLLM_BudgetWindowSpend + LiteLLM_DailyUserSpend + LiteLLM_DailyTeamSpend + LiteLLM_DailyOrganizationSpend + LiteLLM_DailyEndUserSpend + LiteLLM_DailyAgentSpend + LiteLLM_DailyTagSpend + LiteLLM_DailyModelUsage + LiteLLM_DailyToolSpend + LiteLLM_AutoRouterDailySpend +} + +// Durable accounting ledger. One row per usage or cost event; payload holds the +// private event envelope versioned by schema_version. Usage rows retain request-scoped +// pricing context; cost rows include resolved rates and the finalized pricing snapshot. +model LiteLLM_AccountingEvent { + event_id String @id + kind AccountingEventKind + source_event_id String? + schema_version Int + payload_hash String + occurred_at DateTime + recorded_at DateTime @default(now()) + payload Json + + @@index([recorded_at]) + @@index([kind, source_event_id]) +} + +// One row per (event, target table, version): pending work for the projector and, +// once applied in the same transaction as the target write, the receipt that stops a retry. +model LiteLLM_AccountingDelivery { + event_id String + target AccountingDeliveryTarget + version Int + state AccountingDeliveryState @default(pending) + attempts Int @default(0) + next_attempt_at DateTime @default(now()) + applied_at DateTime? + last_error String? + + @@id([event_id, target, version]) + @@index([target, state, next_attempt_at]) +} diff --git a/litellm/accounting/__init__.py b/litellm/accounting/__init__.py new file mode 100644 index 00000000000..d4e56f61279 --- /dev/null +++ b/litellm/accounting/__init__.py @@ -0,0 +1,62 @@ +"""Public accounting contracts. Importing these values does not enable accounting.""" + +from litellm.accounting.measurements import measurements_for +from litellm.accounting.protocols import BudgetEnforcer, CostAggregator, CostCompiler, UsageMeter +from litellm.accounting.types import ( + AdmissionDecision, + AggregateQuery, + AggregateRow, + CostEvent, + DeliveryState, + DeliveryTarget, + Dimension, + DimensionFilter, + DimensionName, + EventKind, + Granularity, + InboundUsageEvent, + Measurement, + Metric, + MetricName, + Multiplier, + NamedCost, + OutboundUsageEvent, + Outcome, + PricedQuantity, + PricingSnapshot, + Rate, + Unpriced, + validate_aggregate_rows, +) + +__all__ = [ + "AdmissionDecision", + "AggregateQuery", + "AggregateRow", + "BudgetEnforcer", + "CostAggregator", + "CostCompiler", + "CostEvent", + "DeliveryState", + "DeliveryTarget", + "Dimension", + "DimensionFilter", + "DimensionName", + "EventKind", + "Granularity", + "InboundUsageEvent", + "Measurement", + "Metric", + "MetricName", + "Multiplier", + "NamedCost", + "OutboundUsageEvent", + "Outcome", + "PricedQuantity", + "PricingSnapshot", + "Rate", + "Unpriced", + "UsageMeter", + "measurements_for", + "validate_aggregate_rows", +] diff --git a/litellm/accounting/measurements.py b/litellm/accounting/measurements.py new file mode 100644 index 00000000000..83011a280da --- /dev/null +++ b/litellm/accounting/measurements.py @@ -0,0 +1,13 @@ +from typing import Final + +from litellm.accounting.types import CostEvent, InboundUsageEvent, Measurement, OutboundUsageEvent + + +def measurements_for(event: InboundUsageEvent | OutboundUsageEvent | CostEvent) -> tuple[Measurement, ...]: + """Return the complete contribution set for one event, ready for one apply call.""" + metrics: Final = event.charges if isinstance(event, CostEvent) else event.metrics + occurred_at: Final = event.occurred_at if isinstance(event, CostEvent) else event.started_at + return tuple( + Measurement(source_event_id=event.event_id, occurred_at=occurred_at, dimensions=event.dimensions, metric=metric) + for metric in metrics or () + ) diff --git a/litellm/accounting/protocols.py b/litellm/accounting/protocols.py new file mode 100644 index 00000000000..8955f56e7a0 --- /dev/null +++ b/litellm/accounting/protocols.py @@ -0,0 +1,85 @@ +"""Contracts only. Implementations own persistence, pricing, projections and live budgets. + +The event row's ``payload`` is a private envelope versioned by the row's ``schema_version``. +Usage envelopes retain the public event and request-scoped pricing context: quantities with units, +model/provider, requested and billed tier/region when known, and per-request/deployment overrides. +These inputs must survive request teardown. Cost-map rates are resolved at compilation, not capture. +Cost envelopes retain the public CostEvent, including its required finalized pricing snapshot. +T5/T6 own envelope serialization; this public API does not introduce a second persisted schema. +""" + +from typing import Protocol + +from litellm.accounting.types import ( + AdmissionDecision, + AggregateQuery, + AggregateRow, + CostEvent, + Dimension, + InboundUsageEvent, + Measurement, + Metric, + OutboundUsageEvent, + Unpriced, +) + + +class UsageMeter(Protocol): + async def record(self, event: InboundUsageEvent | OutboundUsageEvent) -> None: + """Acknowledge commit, no-op identical IDs and reject conflicting content. + + Persist the event and its request-scoped pricing context before acknowledging. + HTTP acknowledgement is a separate policy. Unknown usage remains None. + """ + ... + + +class CostCompiler(Protocol): + async def compile(self, event: OutboundUsageEvent) -> CostEvent | Unpriced: + """Reuse the committed charge or finalize one charge and its delivery intent. + + Resolve captured request-scoped context by event identity and cost-map rates + at compilation. Persist the resolved inputs in the required snapshot. Replay + reuses the committed result and never consumes free allowance again. + Missing pricing is Unpriced, never an inferred zero. + """ + ... + + +class CostAggregator(Protocol): + async def apply(self, measurements: tuple[Measurement, ...]) -> None: + """Apply one source event's complete contribution set per target/version. + + Each target commits all contributions and its receipt atomically. A receipt + for the first metric must not suppress the rest. Usage need not wait for cost. + """ + ... + + async def query(self, query: AggregateQuery) -> tuple[AggregateRow, ...]: + """Return disjoint [start, end) rows per group using existing daily projections. + + Filters combine with AND, values within a filter with IN. DAY emits daily + buckets; None emits whole-interval totals. Reject unsupported metric/scope + combinations. Different groups may overlap, including non-additive tags. + """ + ... + + +class BudgetEnforcer(Protocol): + async def reserve( + self, reservation_id: str, dimensions: tuple[Dimension, ...], estimated: tuple[Metric, ...] + ) -> AdmissionDecision: + """Apply existing admission policies and estimates; duplicate IDs add no hold.""" + ... + + async def settle(self, reservation_id: str | None, cost_event: CostEvent) -> None: + """Settle the committed charge with atomic deduplication and hold mutation. + + Preserve scope/window attribution and retry unfinished settlement after + restart. SQL projection receipts do not provide live-counter atomicity. + """ + ... + + async def release(self, reservation_id: str) -> None: + """Idempotently release unused exposure at the request boundary, preserving charges.""" + ... diff --git a/litellm/accounting/types.py b/litellm/accounting/types.py new file mode 100644 index 00000000000..d0be29bf5b9 --- /dev/null +++ b/litellm/accounting/types.py @@ -0,0 +1,428 @@ +"""Accounting values use UTC instants, decimal strings on the wire and USD charges.""" + +from collections.abc import Mapping +from datetime import datetime, timezone +from decimal import Decimal +from enum import Enum +from typing import Annotated, Final, Literal, TypeAlias + +from pydantic import AfterValidator, AwareDatetime, BaseModel, ConfigDict, Field, model_validator +from typing_extensions import Self + +from litellm.types.utils import CallTypes + +_CONFIG: Final = ConfigDict(frozen=True, extra="forbid", validate_default=True) +Identifier: TypeAlias = Annotated[str, Field(strict=True, min_length=1, pattern=r"\S")] +Quantity: TypeAlias = Annotated[Decimal, Field(strict=True, ge=0, allow_inf_nan=False)] + + +def _utc(value: datetime) -> datetime: + return value.astimezone(timezone.utc) + + +Instant: TypeAlias = Annotated[AwareDatetime, AfterValidator(_utc)] + + +class MetricName(str, Enum): + """Reasoning tokens are included in completion tokens; above-1hr writes are included in cache creation tokens.""" + + REQUESTS = "requests" + SUCCESSFUL_REQUESTS = "successful_requests" + FAILED_REQUESTS = "failed_requests" + PROMPT_TOKENS = "prompt_tokens" + COMPLETION_TOKENS = "completion_tokens" + REASONING_TOKENS = "reasoning_tokens" + TOTAL_TOKENS = "total_tokens" + CACHE_READ_INPUT_TOKENS = "cache_read_input_tokens" + CACHE_CREATION_INPUT_TOKENS = "cache_creation_input_tokens" + CACHE_CREATION_INPUT_TOKENS_ABOVE_1HR = "cache_creation_input_tokens_above_1hr" + SPEND = "spend" + + +class DimensionName(str, Enum): + KEY = "key" + USER = "user" + TEAM = "team" + ORGANIZATION = "organization" + END_USER = "end_user" + MODEL = "model" + MODEL_GROUP = "model_group" + DEPLOYMENT = "deployment" + PROVIDER = "provider" + TOOL = "tool" + AGENT = "agent" + PROJECT = "project" + MODEL_ACCESS_GROUP = "model_access_group" + API_BASE = "api_base" + SERVICE_TIER = "service_tier" + TAG = "tag" + + +class Outcome(str, Enum): + SUCCEEDED = "succeeded" + FAILED = "failed" + CANCELLED = "cancelled" + + +class Granularity(str, Enum): + DAY = "day" + + +class EventKind(str, Enum): + INBOUND_USAGE = "inbound_usage" + OUTBOUND_USAGE = "outbound_usage" + COST = "cost" + + +class DeliveryState(str, Enum): + PENDING = "pending" + APPLIED = "applied" + PARKED = "parked" + + +class DeliveryTarget(str, Enum): + """One member per existing table the projector writes; each is applied and receipted independently.""" + + KEY_SPEND = "LiteLLM_VerificationToken" + USER_SPEND = "LiteLLM_UserTable" + TEAM_SPEND = "LiteLLM_TeamTable" + TEAM_MEMBER_SPEND = "LiteLLM_TeamMembership" + ORGANIZATION_SPEND = "LiteLLM_OrganizationTable" + ORGANIZATION_MEMBER_SPEND = "LiteLLM_OrganizationMembership" + END_USER_SPEND = "LiteLLM_EndUserTable" + AGENT_SPEND = "LiteLLM_AgentsTable" + TAG_SPEND = "LiteLLM_TagTable" + PROJECT_SPEND = "LiteLLM_ProjectTable" + MODEL_ACCESS_GROUP_SPEND = "LiteLLM_ModelAccessGroupBudgetTable" + BUDGET_SPEND = "LiteLLM_BudgetTable" + BUDGET_WINDOW_SPEND = "LiteLLM_BudgetWindowSpend" + DAILY_USER_SPEND = "LiteLLM_DailyUserSpend" + DAILY_TEAM_SPEND = "LiteLLM_DailyTeamSpend" + DAILY_ORGANIZATION_SPEND = "LiteLLM_DailyOrganizationSpend" + DAILY_END_USER_SPEND = "LiteLLM_DailyEndUserSpend" + DAILY_AGENT_SPEND = "LiteLLM_DailyAgentSpend" + DAILY_TAG_SPEND = "LiteLLM_DailyTagSpend" + DAILY_MODEL_USAGE = "LiteLLM_DailyModelUsage" + DAILY_TOOL_SPEND = "LiteLLM_DailyToolSpend" + AUTOROUTER_DAILY_SPEND = "LiteLLM_AutoRouterDailySpend" + + +_SUBSET_PAIRS: Final = ( + (MetricName.REASONING_TOKENS, MetricName.COMPLETION_TOKENS), + (MetricName.CACHE_CREATION_INPUT_TOKENS_ABOVE_1HR, MetricName.CACHE_CREATION_INPUT_TOKENS), +) + + +def _check_subsets(values: Mapping[str, Decimal]) -> None: + for part, whole in _SUBSET_PAIRS: + if part.value in values and whole.value in values and values[part.value] > values[whole.value]: + raise ValueError(f"{part.value} cannot exceed {whole.value}") + + +class Metric(BaseModel): + model_config = _CONFIG + + name: MetricName + value: Quantity + unit: Literal["count", "USD"] = "count" + + @model_validator(mode="after") + def validate_unit(self) -> Self: + if (self.name is MetricName.SPEND) != (self.unit == "USD"): + raise ValueError("SPEND requires USD; usage metrics require count") + if self.unit == "count" and self.value != self.value.to_integral_value(): + raise ValueError("Count metrics require integral values") + return self + + +class Dimension(BaseModel): + model_config = _CONFIG + + name: DimensionName + value: Identifier + + +_REPEATABLE_DIMENSIONS: Final = frozenset({DimensionName.TAG, DimensionName.MODEL_ACCESS_GROUP}) + + +def _dimensions(values: tuple[Dimension, ...]) -> tuple[Dimension, ...]: + if len(frozenset(values)) != len(values): + raise ValueError("Duplicate dimension values") + single_names: Final = tuple(value.name for value in values if value.name not in _REPEATABLE_DIMENSIONS) + if len(frozenset(single_names)) != len(single_names): + raise ValueError("Only tag and model access group dimensions may have multiple values") + return values + + +Dimensions: TypeAlias = Annotated[tuple[Dimension, ...], AfterValidator(_dimensions)] +_CALLER_METRICS: Final = frozenset((MetricName.REQUESTS, MetricName.SUCCESSFUL_REQUESTS, MetricName.FAILED_REQUESTS)) + + +def _metrics(values: tuple[Metric, ...]) -> tuple[Metric, ...]: + if len(frozenset(value.name for value in values)) != len(values): + raise ValueError("Duplicate metric names") + return values + + +Metrics: TypeAlias = Annotated[tuple[Metric, ...], AfterValidator(_metrics)] + + +def _interval(start: datetime, end: datetime, *, allow_empty: bool = False) -> None: + if end < start or (end == start and not allow_empty): + raise ValueError("End must follow start") + + +class InboundUsageEvent(BaseModel): + model_config = _CONFIG + + event_id: Identifier + request_id: Identifier + call_type: CallTypes + started_at: Instant + ended_at: Instant + outcome: Outcome + dimensions: Dimensions + metrics: Metrics + + @model_validator(mode="after") + def validate_caller_activity(self) -> Self: + _interval(self.started_at, self.ended_at, allow_empty=True) + if any(metric.name not in _CALLER_METRICS for metric in self.metrics): + raise ValueError("Inbound events own only caller request metrics") + if not any(metric.name is MetricName.REQUESTS and metric.value == 1 for metric in self.metrics): + raise ValueError("Inbound events count exactly one caller request") + for metric in self.metrics: + if metric.name is MetricName.SUCCESSFUL_REQUESTS and metric.value != int(self.outcome is Outcome.SUCCEEDED): + raise ValueError("Successful request count must match outcome") + if metric.name is MetricName.FAILED_REQUESTS and metric.value != int(self.outcome is not Outcome.SUCCEEDED): + raise ValueError("Failed request count must match outcome") + return self + + +class OutboundUsageEvent(BaseModel): + model_config = _CONFIG + + event_id: Identifier + attempt_id: Identifier + request_id: Identifier | None + retry_of_attempt_id: Identifier | None + call_type: CallTypes + started_at: Instant + ended_at: Instant + outcome: Outcome + dimensions: Dimensions + metrics: Metrics | None + + @model_validator(mode="after") + def validate_attempt_usage(self) -> Self: + _interval(self.started_at, self.ended_at, allow_empty=True) + if self.retry_of_attempt_id == self.attempt_id: + raise ValueError("An attempt cannot retry itself") + if self.metrics is not None and any( + metric.name in _CALLER_METRICS or metric.name is MetricName.SPEND for metric in self.metrics + ): + raise ValueError("Outbound events own only attempt usage metrics") + if self.metrics is not None: + _check_subsets({metric.name.value: metric.value for metric in self.metrics}) + return self + + +PricingUnit: TypeAlias = Literal["token", "second", "character", "image", "request"] + + +class Rate(BaseModel): + model_config = _CONFIG + + measure: Identifier + unit: PricingUnit + usd_per_unit: Quantity + + +class PricedQuantity(BaseModel): + """A pricing input in its billing unit. Only seconds may be fractional.""" + + model_config = _CONFIG + + measure: Identifier + unit: PricingUnit + value: Quantity + + @model_validator(mode="after") + def validate_discrete_quantity(self) -> Self: + if self.unit != "second" and self.value != self.value.to_integral_value(): + raise ValueError("Discrete priced quantities require integral values") + return self + + +class Multiplier(BaseModel): + """A regional or tier factor applied on top of the priced key, e.g. data_residency=eu or vertex_location=us-east5.""" + + model_config = _CONFIG + + name: Identifier + value: Identifier + factor: Annotated[Decimal, Field(strict=True, gt=0, allow_inf_nan=False)] + + +class NamedCost(BaseModel): + """A free-form extra charge by name, e.g. azure_model_router_flat_cost.""" + + model_config = _CONFIG + + name: Identifier + amount: Quantity + + +class PricingSnapshot(BaseModel): + """Retained inputs and result of authoritative compilation, reproducible after cost-map rates change. + + Request-scoped inputs and overrides come from the usage row's versioned payload. Cost-map rates are + resolved at compilation, not frozen at capture. Retries reuse the committed charge and snapshot. + Component names match ``CostBreakdown``: reasoning_cost is part of output_cost, not an additional charge. + Component sums and float tolerances belong to reconciliation, not this structural contract. + """ + + model_config = _CONFIG + + schema_version: Literal[1] = 1 + calculator_version: Identifier + model: Identifier + provider: Identifier | None + pricing_source: Literal["cost_map", "deployment", "per_request", "upstream_header"] + priced_model_key: Identifier + multipliers: tuple[Multiplier, ...] = () + rates: tuple[Rate, ...] = () + usage: tuple[PricedQuantity, ...] = () + input_cost: Quantity | None = None + output_cost: Quantity | None = None + cache_read_cost: Quantity | None = None + cache_creation_cost: Quantity | None = None + reasoning_cost: Quantity | None = None + tool_usage_cost: Quantity | None = None + guardrail_cost: Quantity | None = None + additional_costs: tuple[NamedCost, ...] = () + original_cost: Quantity | None = None + discount_percent: Quantity | None = None + discount_amount: Quantity | None = None + margin_percent: Quantity | None = None + margin_fixed_amount: Quantity | None = None + margin_total_amount: Quantity | None = None + upstream_reported_cost: Quantity | None = None + total_cost: Quantity + + @model_validator(mode="after") + def validate_snapshot(self) -> Self: + rate_keys: Final = frozenset((rate.measure, rate.unit) for rate in self.rates) + quantity_keys: Final = frozenset((quantity.measure, quantity.unit) for quantity in self.usage) + if len(rate_keys) != len(self.rates): + raise ValueError("Duplicate rate measures and units") + if len(quantity_keys) != len(self.usage): + raise ValueError("Duplicate priced quantity measures and units") + if not rate_keys.issubset(quantity_keys): + raise ValueError("Every rate requires one priced quantity with the same measure and unit") + if len(frozenset(cost.name for cost in self.additional_costs)) != len(self.additional_costs): + raise ValueError("Duplicate additional cost names") + if len(frozenset(multiplier.name for multiplier in self.multipliers)) != len(self.multipliers): + raise ValueError("Duplicate multiplier names") + _check_subsets({quantity.measure: quantity.value for quantity in self.usage if quantity.unit == "token"}) + return self + + +class CostEvent(BaseModel): + model_config = _CONFIG + + event_id: Identifier + usage_event_id: Identifier + occurred_at: Instant + dimensions: Dimensions + charges: Metrics + snapshot: PricingSnapshot + + @model_validator(mode="after") + def validate_charge(self) -> Self: + if len(self.charges) != 1 or self.charges[0].name is not MetricName.SPEND: + raise ValueError("A cost event requires one authoritative USD spend metric") + if self.event_id == self.usage_event_id: + raise ValueError("Cost and usage events require distinct identities") + if self.snapshot.total_cost != self.charges[0].value: + raise ValueError("Snapshot total must equal the authoritative spend") + return self + + +class Unpriced(BaseModel): + model_config = _CONFIG + + usage_event_id: Identifier + reason: Identifier + + +class Measurement(BaseModel): + model_config = _CONFIG + + source_event_id: Identifier + occurred_at: Instant + dimensions: Dimensions + metric: Metric + + +class DimensionFilter(BaseModel): + model_config = _CONFIG + + name: DimensionName + values: Annotated[tuple[Identifier, ...], Field(min_length=1)] + + +class AggregateQuery(BaseModel): + model_config = _CONFIG + + metric: MetricName + filters: tuple[DimensionFilter, ...] + group_by: tuple[DimensionName, ...] + start: Instant + end: Instant + granularity: Granularity | None + + @model_validator(mode="after") + def validate_query(self) -> Self: + _interval(self.start, self.end) + if len(frozenset(item.name for item in self.filters)) != len(self.filters): + raise ValueError("Use one IN filter per dimension; filters combine with AND") + if len(frozenset(self.group_by)) != len(self.group_by): + raise ValueError("Duplicate group-by dimensions") + if any( + (instant.hour, instant.minute, instant.second, instant.microsecond) != (0, 0, 0, 0) + for instant in (self.start, self.end) + ): + raise ValueError("Initial reporting queries require UTC day boundaries") + return self + + +class AggregateRow(BaseModel): + model_config = _CONFIG + + start: Instant + end: Instant + dimensions: Dimensions + value: Quantity + + @model_validator(mode="after") + def validate_bounds(self) -> Self: + _interval(self.start, self.end) + return self + + +def validate_aggregate_rows(rows: tuple[AggregateRow, ...]) -> tuple[AggregateRow, ...]: + """Validate [start, end) buckets per group; overlapping tag groups are not additive.""" + groups: Final = frozenset(frozenset(row.dimensions) for row in rows) + for group in groups: + ordered = sorted((row for row in rows if frozenset(row.dimensions) == group), key=lambda row: row.start) + if any(left.end > right.start for left, right in zip(ordered, ordered[1:])): + raise ValueError("Aggregate buckets overlap within a dimension group") + return rows + + +class AdmissionDecision(BaseModel): + model_config = _CONFIG + + allowed: bool + reason: Identifier | None = None diff --git a/litellm/integrations/otel/model/spans.py b/litellm/integrations/otel/model/spans.py index d8be607150c..e423b6698e6 100644 --- a/litellm/integrations/otel/model/spans.py +++ b/litellm/integrations/otel/model/spans.py @@ -338,6 +338,8 @@ _PRISMA_MODELS: Final[frozenset[str]] = frozenset( "LiteLLM_ToolTable", "LiteLLM_TelemetryReport", "LiteLLM_AccessGroupTable", + "LiteLLM_AccountingDelivery", + "LiteLLM_AccountingEvent", "LiteLLM_ClaudeCodePluginTable", "LiteLLM_MemoryTable", "LiteLLM_AdaptiveRouterState", diff --git a/litellm/proxy/schema.prisma b/litellm/proxy/schema.prisma index 4187f76c97c..2965174114b 100644 --- a/litellm/proxy/schema.prisma +++ b/litellm/proxy/schema.prisma @@ -2061,3 +2061,73 @@ model LiteLLM_TelemetryReport { @@index([window_end]) @@index([created_at]) } + +enum AccountingEventKind { + inbound_usage + outbound_usage + cost +} + +enum AccountingDeliveryState { + pending + applied + parked +} + +enum AccountingDeliveryTarget { + LiteLLM_VerificationToken + LiteLLM_UserTable + LiteLLM_TeamTable + LiteLLM_TeamMembership + LiteLLM_OrganizationTable + LiteLLM_OrganizationMembership + LiteLLM_EndUserTable + LiteLLM_AgentsTable + LiteLLM_TagTable + LiteLLM_ProjectTable + LiteLLM_ModelAccessGroupBudgetTable + LiteLLM_BudgetTable + LiteLLM_BudgetWindowSpend + LiteLLM_DailyUserSpend + LiteLLM_DailyTeamSpend + LiteLLM_DailyOrganizationSpend + LiteLLM_DailyEndUserSpend + LiteLLM_DailyAgentSpend + LiteLLM_DailyTagSpend + LiteLLM_DailyModelUsage + LiteLLM_DailyToolSpend + LiteLLM_AutoRouterDailySpend +} + +// Durable accounting ledger. One row per usage or cost event; payload holds the +// private event envelope versioned by schema_version. Usage rows retain request-scoped +// pricing context; cost rows include resolved rates and the finalized pricing snapshot. +model LiteLLM_AccountingEvent { + event_id String @id + kind AccountingEventKind + source_event_id String? + schema_version Int + payload_hash String + occurred_at DateTime + recorded_at DateTime @default(now()) + payload Json + + @@index([recorded_at]) + @@index([kind, source_event_id]) +} + +// One row per (event, target table, version): pending work for the projector and, +// once applied in the same transaction as the target write, the receipt that stops a retry. +model LiteLLM_AccountingDelivery { + event_id String + target AccountingDeliveryTarget + version Int + state AccountingDeliveryState @default(pending) + attempts Int @default(0) + next_attempt_at DateTime @default(now()) + applied_at DateTime? + last_error String? + + @@id([event_id, target, version]) + @@index([target, state, next_attempt_at]) +} diff --git a/schema.prisma b/schema.prisma index 4187f76c97c..2965174114b 100644 --- a/schema.prisma +++ b/schema.prisma @@ -2061,3 +2061,73 @@ model LiteLLM_TelemetryReport { @@index([window_end]) @@index([created_at]) } + +enum AccountingEventKind { + inbound_usage + outbound_usage + cost +} + +enum AccountingDeliveryState { + pending + applied + parked +} + +enum AccountingDeliveryTarget { + LiteLLM_VerificationToken + LiteLLM_UserTable + LiteLLM_TeamTable + LiteLLM_TeamMembership + LiteLLM_OrganizationTable + LiteLLM_OrganizationMembership + LiteLLM_EndUserTable + LiteLLM_AgentsTable + LiteLLM_TagTable + LiteLLM_ProjectTable + LiteLLM_ModelAccessGroupBudgetTable + LiteLLM_BudgetTable + LiteLLM_BudgetWindowSpend + LiteLLM_DailyUserSpend + LiteLLM_DailyTeamSpend + LiteLLM_DailyOrganizationSpend + LiteLLM_DailyEndUserSpend + LiteLLM_DailyAgentSpend + LiteLLM_DailyTagSpend + LiteLLM_DailyModelUsage + LiteLLM_DailyToolSpend + LiteLLM_AutoRouterDailySpend +} + +// Durable accounting ledger. One row per usage or cost event; payload holds the +// private event envelope versioned by schema_version. Usage rows retain request-scoped +// pricing context; cost rows include resolved rates and the finalized pricing snapshot. +model LiteLLM_AccountingEvent { + event_id String @id + kind AccountingEventKind + source_event_id String? + schema_version Int + payload_hash String + occurred_at DateTime + recorded_at DateTime @default(now()) + payload Json + + @@index([recorded_at]) + @@index([kind, source_event_id]) +} + +// One row per (event, target table, version): pending work for the projector and, +// once applied in the same transaction as the target write, the receipt that stops a retry. +model LiteLLM_AccountingDelivery { + event_id String + target AccountingDeliveryTarget + version Int + state AccountingDeliveryState @default(pending) + attempts Int @default(0) + next_attempt_at DateTime @default(now()) + applied_at DateTime? + last_error String? + + @@id([event_id, target, version]) + @@index([target, state, next_attempt_at]) +} diff --git a/tests/integration/database/test_accounting_ledger.py b/tests/integration/database/test_accounting_ledger.py new file mode 100644 index 00000000000..089ed872dee --- /dev/null +++ b/tests/integration/database/test_accounting_ledger.py @@ -0,0 +1,128 @@ +import os +import subprocess +import sys +from collections.abc import Iterator +from pathlib import Path +from typing import Final + +import psycopg +import pytest +from psycopg import sql + +from litellm.accounting import DeliveryState, DeliveryTarget, EventKind +from tests.integration._support.database import scratch_database + +_SCHEMA: Final = Path(__file__).resolve().parents[3] / "litellm-proxy-extras/litellm_proxy_extras/schema.prisma" +_EVENT_INSERT: Final = ( + 'INSERT INTO "LiteLLM_AccountingEvent" ' + "(event_id, kind, schema_version, payload_hash, occurred_at, payload) " + "VALUES ('event', %s, %s, 'hash', '2026-01-01', '{}') RETURNING kind::text, schema_version" +) +_DELIVERY_INSERT: Final = ( + 'INSERT INTO "LiteLLM_AccountingDelivery" (event_id, target, version, state, attempts) ' + "VALUES ('event', %s, %s, %s, %s) RETURNING target::text, version, state::text, attempts" +) + + +@pytest.fixture(scope="module") +def ledger_database() -> Iterator[str]: + with scratch_database("accounting") as database_url: + result: Final = subprocess.run( + [sys.executable, "-I", "-m", "prisma", "migrate", "deploy", "--schema", str(_SCHEMA)], + env={**os.environ, "DATABASE_URL": database_url}, + capture_output=True, + text=True, + timeout=300, + ) + assert result.returncode == 0, result.stdout + result.stderr + yield database_url + + +@pytest.fixture +def ledger(ledger_database: str) -> Iterator[psycopg.Connection[tuple[object, ...]]]: + with psycopg.connect(ledger_database) as connection: + try: + yield connection + finally: + connection.rollback() + + +@pytest.mark.parametrize("kind", tuple(EventKind)) +def test_ledger_accepts_every_public_event_kind( + ledger: psycopg.Connection[tuple[object, ...]], kind: EventKind +) -> None: + row: Final = ledger.execute(_EVENT_INSERT, (kind.value, 1)).fetchone() + assert row == (kind.value, 1) + + +@pytest.mark.parametrize("target", tuple(DeliveryTarget)) +@pytest.mark.parametrize("state", tuple(DeliveryState)) +def test_delivery_accepts_every_public_target_and_state( + ledger: psycopg.Connection[tuple[object, ...]], target: DeliveryTarget, state: DeliveryState +) -> None: + row: Final = ledger.execute(_DELIVERY_INSERT, (target.value, 1, state.value, 0)).fetchone() + assert row == (target.value, 1, state.value, 0) + + +@pytest.mark.parametrize( + ("enum_name", "values"), + [ + ("AccountingEventKind", tuple(value.value for value in EventKind)), + ("AccountingDeliveryState", tuple(value.value for value in DeliveryState)), + ("AccountingDeliveryTarget", tuple(value.value for value in DeliveryTarget)), + ], +) +def test_migrated_enums_match_the_public_contract( + ledger: psycopg.Connection[tuple[object, ...]], enum_name: str, values: tuple[str, ...] +) -> None: + rows: Final = ledger.execute( + "SELECT e.enumlabel FROM pg_enum e JOIN pg_type t ON t.oid = e.enumtypid WHERE t.typname = %s", + (enum_name,), + ).fetchall() + assert frozenset(row[0] for row in rows) == frozenset(values) + + +@pytest.mark.parametrize("version", [0, -1]) +def test_event_rejects_nonpositive_schema_versions( + ledger: psycopg.Connection[tuple[object, ...]], version: int +) -> None: + with pytest.raises(psycopg.errors.CheckViolation, match="AccountingEvent_schema_version_check"): + ledger.execute(_EVENT_INSERT, (EventKind.OUTBOUND_USAGE.value, version)) + + +@pytest.mark.parametrize( + ("version", "attempts", "constraint"), + [ + (0, 0, "AccountingDelivery_version_check"), + (-1, 0, "AccountingDelivery_version_check"), + (1, -1, "AccountingDelivery_attempts_check"), + ], +) +def test_delivery_rejects_invalid_versions_and_attempts( + ledger: psycopg.Connection[tuple[object, ...]], version: int, attempts: int, constraint: str +) -> None: + with pytest.raises(psycopg.errors.CheckViolation, match=constraint): + ledger.execute(_DELIVERY_INSERT, (DeliveryTarget.KEY_SPEND.value, version, "pending", attempts)) + + +@pytest.mark.parametrize("column", ["state", "target"]) +def test_delivery_rejects_unknown_enum_values(ledger: psycopg.Connection[tuple[object, ...]], column: str) -> None: + ledger.execute(_DELIVERY_INSERT, (DeliveryTarget.KEY_SPEND.value, 1, "pending", 0)) + with pytest.raises(psycopg.errors.InvalidTextRepresentation, match="invalid input value for enum"): + ledger.execute( + sql.SQL('UPDATE "LiteLLM_AccountingDelivery" SET {} = %s').format(sql.Identifier(column)), ("invalid",) + ) + + +def test_event_rejects_unknown_kind(ledger: psycopg.Connection[tuple[object, ...]]) -> None: + with pytest.raises(psycopg.errors.InvalidTextRepresentation, match="invalid input value for enum"): + ledger.execute(_EVENT_INSERT, ("invalid", 1)) + + +def test_new_delivery_defaults_to_pending_with_no_attempts(ledger: psycopg.Connection[tuple[object, ...]]) -> None: + row: Final = ledger.execute( + 'INSERT INTO "LiteLLM_AccountingDelivery" (event_id, target, version) ' + "VALUES ('event', %s, 1) RETURNING state::text, attempts", + (DeliveryTarget.KEY_SPEND.value,), + ).fetchone() + assert row == ("pending", 0) diff --git a/tests/unit/accounting/__init__.py b/tests/unit/accounting/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/unit/accounting/test_measurements.py b/tests/unit/accounting/test_measurements.py new file mode 100644 index 00000000000..b65c3076c5d --- /dev/null +++ b/tests/unit/accounting/test_measurements.py @@ -0,0 +1,235 @@ +from datetime import datetime, timezone +from decimal import Decimal +from typing import Final + +import pytest + +from litellm.accounting import ( + CostEvent, + Dimension, + DimensionName, + InboundUsageEvent, + Measurement, + Metric, + MetricName, + Outcome, + OutboundUsageEvent, + PricingSnapshot, + measurements_for, +) +from litellm.types.utils import CallTypes + +_START: Final = datetime(2026, 1, 1, 23, 59, 59, tzinfo=timezone.utc) +_END: Final = datetime(2026, 1, 2, tzinfo=timezone.utc) +_DIMENSIONS: Final = ( + Dimension(name=DimensionName.TEAM, value="team"), + Dimension(name=DimensionName.TAG, value="alpha"), + Dimension(name=DimensionName.TAG, value="beta"), +) + + +def _snapshot(amount: Decimal) -> PricingSnapshot: + return PricingSnapshot( + calculator_version="test", + model="test-model", + provider=None, + pricing_source="upstream_header", + priced_model_key="test-model", + upstream_reported_cost=amount, + total_cost=amount, + ) + + +@pytest.mark.parametrize( + ("outcome", "outcome_metric"), + [ + (Outcome.SUCCEEDED, MetricName.SUCCESSFUL_REQUESTS), + (Outcome.FAILED, MetricName.FAILED_REQUESTS), + (Outcome.CANCELLED, MetricName.FAILED_REQUESTS), + ], +) +def test_caller_measurements_keep_both_request_and_outcome_counts(outcome, outcome_metric): + event = InboundUsageEvent( + event_id="caller-event", + request_id="caller", + call_type=CallTypes.acompletion, + started_at=_START, + ended_at=_END, + outcome=outcome, + dimensions=_DIMENSIONS, + metrics=(Metric(name=MetricName.REQUESTS, value=Decimal(1)), Metric(name=outcome_metric, value=Decimal(1))), + ) + assert measurements_for(event) == ( + Measurement( + source_event_id="caller-event", + occurred_at=_START, + dimensions=_DIMENSIONS, + metric=Metric(name=MetricName.REQUESTS, value=Decimal(1)), + ), + Measurement( + source_event_id="caller-event", + occurred_at=_START, + dimensions=_DIMENSIONS, + metric=Metric(name=outcome_metric, value=Decimal(1)), + ), + ) + + +def test_usage_measurements_keep_every_metric_in_the_start_time_bucket(): + event = OutboundUsageEvent( + event_id="usage-event", + attempt_id="attempt", + request_id="caller", + retry_of_attempt_id=None, + call_type=CallTypes.acompletion, + started_at=_START, + ended_at=_END, + outcome=Outcome.SUCCEEDED, + dimensions=_DIMENSIONS, + metrics=( + Metric(name=MetricName.PROMPT_TOKENS, value=Decimal(10)), + Metric(name=MetricName.COMPLETION_TOKENS, value=Decimal(2)), + Metric(name=MetricName.TOTAL_TOKENS, value=Decimal(12)), + Metric(name=MetricName.CACHE_READ_INPUT_TOKENS, value=Decimal(3)), + Metric(name=MetricName.CACHE_CREATION_INPUT_TOKENS, value=Decimal(1)), + ), + ) + assert measurements_for(event) == ( + Measurement( + source_event_id="usage-event", + occurred_at=_START, + dimensions=_DIMENSIONS, + metric=Metric(name=MetricName.PROMPT_TOKENS, value=Decimal(10)), + ), + Measurement( + source_event_id="usage-event", + occurred_at=_START, + dimensions=_DIMENSIONS, + metric=Metric(name=MetricName.COMPLETION_TOKENS, value=Decimal(2)), + ), + Measurement( + source_event_id="usage-event", + occurred_at=_START, + dimensions=_DIMENSIONS, + metric=Metric(name=MetricName.TOTAL_TOKENS, value=Decimal(12)), + ), + Measurement( + source_event_id="usage-event", + occurred_at=_START, + dimensions=_DIMENSIONS, + metric=Metric(name=MetricName.CACHE_READ_INPUT_TOKENS, value=Decimal(3)), + ), + Measurement( + source_event_id="usage-event", + occurred_at=_START, + dimensions=_DIMENSIONS, + metric=Metric(name=MetricName.CACHE_CREATION_INPUT_TOKENS, value=Decimal(1)), + ), + ) + + +@pytest.mark.parametrize("metrics", [None, ()]) +def test_unknown_or_empty_usage_does_not_fabricate_zero_measurements(metrics): + event = OutboundUsageEvent( + event_id="usage-event", + attempt_id="attempt", + request_id=None, + retry_of_attempt_id=None, + call_type=CallTypes.acompletion, + started_at=_START, + ended_at=_END, + outcome=Outcome.SUCCEEDED, + dimensions=(), + metrics=metrics, + ) + assert measurements_for(event) == () + + +@pytest.mark.parametrize("amount", [Decimal(0), Decimal("0.1234567890123456789")]) +def test_cost_measurements_preserve_authoritative_amount_identity_and_occurrence(amount): + event = CostEvent( + event_id="cost-event", + usage_event_id="usage-event", + occurred_at=_END, + dimensions=_DIMENSIONS, + charges=(Metric(name=MetricName.SPEND, value=amount, unit="USD"),), + snapshot=_snapshot(amount), + ) + assert measurements_for(event) == ( + Measurement( + source_event_id="cost-event", + occurred_at=_END, + dimensions=_DIMENSIONS, + metric=Metric(name=MetricName.SPEND, value=amount, unit="USD"), + ), + ) + + +def test_billed_retry_and_success_contribute_one_caller_request_and_both_charges(): + caller = InboundUsageEvent( + event_id="caller-event", + request_id="caller", + call_type=CallTypes.acompletion, + started_at=_START, + ended_at=_END, + outcome=Outcome.SUCCEEDED, + dimensions=_DIMENSIONS, + metrics=(Metric(name=MetricName.REQUESTS, value=Decimal(1)),), + ) + retry = OutboundUsageEvent( + event_id="retry-usage", + attempt_id="retry", + request_id="caller", + retry_of_attempt_id=None, + call_type=CallTypes.acompletion, + started_at=_START, + ended_at=_END, + outcome=Outcome.FAILED, + dimensions=_DIMENSIONS, + metrics=(Metric(name=MetricName.TOTAL_TOKENS, value=Decimal(4)),), + ) + success = OutboundUsageEvent( + event_id="success-usage", + attempt_id="success", + request_id="caller", + retry_of_attempt_id="retry", + call_type=CallTypes.acompletion, + started_at=_START, + ended_at=_END, + outcome=Outcome.SUCCEEDED, + dimensions=_DIMENSIONS, + metrics=(Metric(name=MetricName.TOTAL_TOKENS, value=Decimal(12)),), + ) + retry_cost = CostEvent( + event_id="retry-cost", + usage_event_id="retry-usage", + occurred_at=_START, + dimensions=_DIMENSIONS, + charges=(Metric(name=MetricName.SPEND, value=Decimal("0.025"), unit="USD"),), + snapshot=_snapshot(Decimal("0.025")), + ) + success_cost = CostEvent( + event_id="success-cost", + usage_event_id="success-usage", + occurred_at=_START, + dimensions=_DIMENSIONS, + charges=(Metric(name=MetricName.SPEND, value=Decimal("0.125"), unit="USD"),), + snapshot=_snapshot(Decimal("0.125")), + ) + measurements = ( + *measurements_for(caller), + *measurements_for(retry), + *measurements_for(success), + *measurements_for(retry_cost), + *measurements_for(success_cost), + ) + assert sum(m.metric.value for m in measurements if m.metric.name is MetricName.REQUESTS) == Decimal(1) + assert sum(m.metric.value for m in measurements if m.metric.name is MetricName.TOTAL_TOKENS) == Decimal(16) + assert sum(m.metric.value for m in measurements if m.metric.name is MetricName.SPEND) == Decimal("0.150") + assert {m.source_event_id for m in measurements} == { + "caller-event", + "retry-usage", + "success-usage", + "retry-cost", + "success-cost", + } diff --git a/tests/unit/accounting/test_types.py b/tests/unit/accounting/test_types.py new file mode 100644 index 00000000000..f645ee6c2cc --- /dev/null +++ b/tests/unit/accounting/test_types.py @@ -0,0 +1,484 @@ +import re +from collections.abc import Mapping +from datetime import datetime, timedelta, timezone +from decimal import Decimal +from enum import Enum +from pathlib import Path +from typing import Final + +import pytest +from pydantic import ValidationError + +from litellm.accounting import ( + AdmissionDecision, + AggregateQuery, + AggregateRow, + CostEvent, + DeliveryState, + DeliveryTarget, + Dimension, + DimensionFilter, + DimensionName, + EventKind, + Granularity, + InboundUsageEvent, + Metric, + MetricName, + Multiplier, + Outcome, + OutboundUsageEvent, + PricedQuantity, + PricingSnapshot, + Rate, + validate_aggregate_rows, +) +from litellm.accounting.types import PricingUnit +from litellm.types.utils import CallTypes + +_START: Final = datetime(2026, 1, 1, tzinfo=timezone.utc) +_END: Final = datetime(2026, 1, 2, tzinfo=timezone.utc) +_REQUEST: Final = Metric(name=MetricName.REQUESTS, value=Decimal(1)) +_TOKEN: Final = Metric(name=MetricName.PROMPT_TOKENS, value=Decimal(10)) +_SPEND: Final = Metric(name=MetricName.SPEND, value=Decimal("0.1234567890123456789"), unit="USD") +_TEAM: Final = Dimension(name=DimensionName.TEAM, value="team") + + +def _inbound(metrics=(_REQUEST,), outcome=Outcome.SUCCEEDED): + return InboundUsageEvent( + event_id="inbound:request", + request_id="request", + call_type=CallTypes.acompletion, + started_at=_START, + ended_at=_START, + outcome=outcome, + dimensions=(_TEAM,), + metrics=metrics, + ) + + +def _outbound(metrics=(_TOKEN,), **changes): + return OutboundUsageEvent.model_validate( + { + "event_id": "outbound:attempt", + "attempt_id": "attempt", + "request_id": "request", + "retry_of_attempt_id": None, + "call_type": CallTypes.acompletion, + "started_at": _START, + "ended_at": _END, + "outcome": Outcome.SUCCEEDED, + "dimensions": (_TEAM,), + "metrics": metrics, + **changes, + } + ) + + +def _query(**changes): + return AggregateQuery.model_validate( + { + "metric": MetricName.SPEND, + "filters": (DimensionFilter(name=DimensionName.TEAM, values=("one", "two")),), + "group_by": (DimensionName.MODEL,), + "start": _START, + "end": _END, + "granularity": Granularity.DAY, + **changes, + } + ) + + +def test_wire_values_preserve_decimal_precision_and_normalize_aware_times(): + offset = timezone(timedelta(hours=-8)) + event = CostEvent( + event_id="cost:attempt", + usage_event_id="outbound:attempt", + occurred_at=datetime(2025, 12, 31, 16, tzinfo=offset), + dimensions=(_TEAM,), + charges=(_SPEND,), + snapshot=_snapshot(), + ) + wire = event.model_dump(mode="json") + assert wire["occurred_at"] == "2026-01-01T00:00:00Z" + assert wire["charges"] == [{"name": "spend", "value": "0.1234567890123456789", "unit": "USD"}] + assert event.occurred_at == _START + assert CostEvent.model_validate_json(event.model_dump_json()).charges[0].value == _SPEND.value + + +@pytest.mark.parametrize( + ("name", "value", "unit", "message"), + [ + (MetricName.SPEND, Decimal(1), "count", "SPEND requires USD"), + (MetricName.REQUESTS, Decimal(1), "USD", "SPEND requires USD"), + (MetricName.PROMPT_TOKENS, Decimal("1.5"), "count", "integral"), + (MetricName.SPEND, Decimal("NaN"), "USD", "finite"), + (MetricName.SPEND, Decimal("Infinity"), "USD", "finite"), + (MetricName.SPEND, Decimal(-1), "USD", "greater than or equal"), + (MetricName.SPEND, 0.1, "USD", "instance of Decimal"), + ], +) +def test_metrics_reject_wrong_units_nonfinite_values_and_binary_float_inputs(name, value, unit, message): + with pytest.raises(ValidationError, match=message): + Metric(name=name, value=value, unit=unit) + + +@pytest.mark.parametrize("metrics", [(_TOKEN,), (_SPEND,), ()]) +def test_inbound_activity_cannot_own_usage_or_cost_or_omit_request_count(metrics): + with pytest.raises(ValidationError, match="Inbound events"): + _inbound(metrics) + + +@pytest.mark.parametrize( + ("outcome", "metric"), + [ + (Outcome.FAILED, Metric(name=MetricName.SUCCESSFUL_REQUESTS, value=Decimal(1))), + (Outcome.SUCCEEDED, Metric(name=MetricName.FAILED_REQUESTS, value=Decimal(1))), + ], +) +def test_inbound_outcome_counts_cannot_contradict_outcome(outcome, metric): + with pytest.raises(ValidationError, match="count must match outcome"): + _inbound((_REQUEST, metric), outcome) + + +@pytest.mark.parametrize("metrics", [(_REQUEST,), (_SPEND,)]) +def test_outbound_attempts_cannot_own_caller_counts_or_charges(metrics): + with pytest.raises(ValidationError, match="Outbound events own only"): + _outbound(metrics) + + +@pytest.mark.parametrize( + ("changes", "message"), + [ + ({"started_at": datetime(2026, 1, 1)}, "timezone"), + ({"ended_at": _START - timedelta(seconds=1)}, "End must follow start"), + ({"retry_of_attempt_id": "attempt"}, "cannot retry itself"), + ({"dimensions": (_TEAM, _TEAM)}, "Duplicate dimension values"), + ({"dimensions": (_TEAM, Dimension(name=DimensionName.TEAM, value="other"))}, "Only tag and model access group"), + ( + { + "metrics": ( + Metric(name=MetricName.COMPLETION_TOKENS, value=Decimal(5)), + Metric(name=MetricName.REASONING_TOKENS, value=Decimal(6)), + ) + }, + "reasoning_tokens cannot exceed completion_tokens", + ), + ( + { + "metrics": ( + Metric(name=MetricName.CACHE_CREATION_INPUT_TOKENS, value=Decimal(5)), + Metric(name=MetricName.CACHE_CREATION_INPUT_TOKENS_ABOVE_1HR, value=Decimal(6)), + ) + }, + "cache_creation_input_tokens_above_1hr cannot exceed cache_creation_input_tokens", + ), + ({"event_id": " "}, "pattern"), + ({"metrics": (_TOKEN, _TOKEN)}, "Duplicate metric names"), + ], +) +def test_outbound_events_reject_invalid_identity_time_and_duplicate_attribution(changes, message): + with pytest.raises(ValidationError, match=message): + _outbound(**changes) + + +def test_unknown_and_known_zero_usage_are_distinct_and_values_are_immutable(): + unknown = _outbound(None) + zero = _outbound((Metric(name=MetricName.PROMPT_TOKENS, value=Decimal(0)),)) + assert unknown.metrics is None + assert zero.metrics is not None and zero.metrics[0].value == 0 + assert _inbound().started_at == _inbound().ended_at + with pytest.raises(ValidationError, match="frozen"): + zero.attempt_id = "different" + with pytest.raises(ValidationError, match="frozen"): + zero.metrics[0].value = Decimal(10) + + +@pytest.mark.parametrize("charges", [(), (_TOKEN,), (_SPEND, _SPEND)]) +def test_cost_events_require_one_authoritative_spend(charges): + with pytest.raises(ValidationError, match=r"authoritative USD spend|Duplicate metric"): + CostEvent( + event_id="cost", + usage_event_id="usage", + occurred_at=_START, + dimensions=(), + charges=charges, + snapshot=_snapshot(), + ) + + +def test_cost_event_cannot_reuse_its_usage_identity(): + with pytest.raises(ValidationError, match="distinct identities"): + CostEvent( + event_id="usage", + usage_event_id="usage", + occurred_at=_START, + dimensions=(), + charges=(_SPEND,), + snapshot=_snapshot(), + ) + + +@pytest.mark.parametrize("granularity", [Granularity.DAY, None]) +def test_daily_and_whole_interval_queries_preserve_filter_and_group_semantics(granularity): + query = _query(granularity=granularity) + assert query.granularity is granularity + assert query.filters[0].values == ("one", "two") + assert query.group_by == (DimensionName.MODEL,) + assert query.start == _START and query.end == _END + + +@pytest.mark.parametrize( + ("changes", "message"), + [ + ({"end": _START}, "End must follow start"), + ({"start": _START + timedelta(hours=1)}, "UTC day boundaries"), + ({"end": _END + timedelta(microseconds=1)}, "UTC day boundaries"), + ({"granularity": "hour"}, "day"), + ({"group_by": (DimensionName.TEAM, DimensionName.TEAM)}, "Duplicate group-by"), + ({"filters": (DimensionFilter(name=DimensionName.TEAM, values=("one",)),) * 2}, "one IN filter"), + ({"filters": ({"name": "team", "values": ()},)}, "at least 1"), + ], +) +def test_queries_reject_unsupported_or_ambiguous_ranges_and_filters(changes, message): + with pytest.raises(ValidationError, match=message): + _query(**changes) + + +def test_rows_allow_adjacent_buckets_and_overlapping_distinct_groups(): + first = AggregateRow(start=_START, end=_END, dimensions=(_TEAM,), value=Decimal(1)) + second = AggregateRow(start=_END, end=_END + timedelta(days=1), dimensions=(_TEAM,), value=Decimal(2)) + other = AggregateRow(start=_START, end=_END, dimensions=(), value=Decimal(3)) + rows = (second, other, first) + assert validate_aggregate_rows(rows) is rows + assert validate_aggregate_rows(()) == () + + +def test_rows_reject_overlap_within_a_group_regardless_of_dimension_order(): + model = Dimension(name=DimensionName.MODEL, value="model") + first = AggregateRow(start=_START, end=_END, dimensions=(_TEAM, model), value=Decimal(1)) + overlapping = AggregateRow(start=_START, end=_END, dimensions=(model, _TEAM), value=Decimal(2)) + with pytest.raises(ValueError, match="overlap within a dimension group"): + validate_aggregate_rows((first, overlapping)) + + +def test_rows_reject_empty_intervals(): + with pytest.raises(ValidationError, match="End must follow start"): + AggregateRow(start=_START, end=_START, dimensions=(), value=Decimal(0)) + + +def test_admission_decision_preserves_a_denial_reason(): + decision = AdmissionDecision(allowed=False, reason="Team budget exhausted") + assert not decision.allowed + assert decision.reason == "Team budget exhausted" + + +def _snapshot(**changes: object) -> PricingSnapshot: + return PricingSnapshot.model_validate( + { + "calculator_version": "1.80.0", + "model": "gpt-4o", + "provider": "openai", + "pricing_source": "cost_map", + "priced_model_key": "openai/gpt-4o", + "multipliers": (Multiplier(name="data_residency", value="eu", factor=Decimal("1.1")),), + "rates": (Rate(measure="input", unit="token", usd_per_unit=Decimal("0.0000025")),), + "usage": (PricedQuantity(measure="input", unit="token", value=_TOKEN.value),), + "input_cost": Decimal("0.000025"), + "total_cost": _SPEND.value, + **changes, + } + ) + + +def test_cost_event_carries_a_snapshot_whose_total_matches_the_spend(): + event = CostEvent( + event_id="cost:attempt", + usage_event_id="outbound:attempt", + occurred_at=_START, + dimensions=(_TEAM,), + charges=(_SPEND,), + snapshot=_snapshot(), + ) + assert event.snapshot is not None + assert event.snapshot.total_cost == _SPEND.value + assert event.snapshot.multipliers[0].factor == Decimal("1.1") + with pytest.raises(ValidationError, match="Snapshot total must equal the authoritative spend"): + CostEvent( + event_id="cost:attempt", + usage_event_id="outbound:attempt", + occurred_at=_START, + dimensions=(_TEAM,), + charges=(_SPEND,), + snapshot=_snapshot(total_cost=Decimal("0.5")), + ) + + +@pytest.mark.parametrize( + ("changes", "message"), + ( + ( + { + "rates": ( + Rate(measure="input", unit="token", usd_per_unit=Decimal(1)), + Rate(measure="input", unit="token", usd_per_unit=Decimal(2)), + ) + }, + "Duplicate rate measures and units", + ), + ({"usage": (_SPEND,)}, "PricedQuantity"), + ({"usage": (_REQUEST,)}, "PricedQuantity"), + ({"pricing_source": "guess"}, "pricing_source"), + ({"multipliers": ({"name": "x", "value": "y", "factor": Decimal(0)},)}, "factor"), + ({"schema_version": 2}, "schema_version"), + ( + {"additional_costs": ({"name": "fee", "amount": Decimal(1)}, {"name": "fee", "amount": Decimal(2)})}, + "Duplicate additional cost names", + ), + ), +) +def test_snapshots_reject_ambiguous_rates_wrong_usage_and_unknown_sources(changes: Mapping[str, object], message: str): + with pytest.raises(ValidationError, match=message): + _snapshot(**changes) + + +@pytest.mark.parametrize( + ("unit", "value"), + [("token", "10"), ("second", "1.25"), ("character", "25"), ("image", "2"), ("request", "1")], +) +def test_snapshot_preserves_matching_rates_and_quantities_for_each_billing_unit(unit: PricingUnit, value: str) -> None: + quantity: Final = PricedQuantity(measure="input", unit=unit, value=Decimal(value)) + rate: Final = Rate(measure="input", unit=unit, usd_per_unit=Decimal("0.003")) + amount: Final = quantity.value * rate.usd_per_unit + snapshot: Final = _snapshot(rates=(rate,), usage=(quantity,), total_cost=amount) + restored: Final = PricingSnapshot.model_validate_json(snapshot.model_dump_json()) + assert restored == snapshot + assert restored.usage[0].value * restored.rates[0].usd_per_unit == amount + + +@pytest.mark.parametrize( + ("usage", "message"), + [ + ((), "Every rate requires"), + (({"measure": "output", "unit": "token", "value": Decimal(10)},), "Every rate requires"), + (({"measure": "input", "unit": "second", "value": Decimal(10)},), "Every rate requires"), + ( + ({"measure": "input", "unit": "token", "value": Decimal(10)},) * 2, + "Duplicate priced quantity", + ), + ], +) +def test_snapshot_rejects_missing_mismatched_or_duplicate_priced_quantities( + usage: tuple[Mapping[str, object], ...], message: str +) -> None: + with pytest.raises(ValidationError, match=message): + _snapshot(usage=usage) + + +@pytest.mark.parametrize("unit", ["token", "character", "image", "request"]) +def test_priced_counts_remain_integral(unit: PricingUnit) -> None: + with pytest.raises(ValidationError, match="integral"): + PricedQuantity(measure="input", unit=unit, value=Decimal("1.25")) + + +@pytest.mark.parametrize( + ("value", "message"), + [ + (Decimal(-1), "greater than or equal"), + (Decimal("NaN"), "finite"), + (Decimal("Infinity"), "finite"), + (1.25, "instance of Decimal"), + ], +) +def test_pricing_values_reject_invalid_amounts(value: Decimal | float, message: str) -> None: + with pytest.raises(ValidationError, match=message): + PricedQuantity.model_validate({"measure": "input", "unit": "second", "value": value}) + with pytest.raises(ValidationError, match=message): + Rate.model_validate({"measure": "input", "unit": "second", "usd_per_unit": value}) + + +@pytest.mark.parametrize("snapshot_fields", [{}, {"snapshot": None}]) +def test_cost_event_cannot_omit_its_derivation(snapshot_fields: Mapping[str, object]) -> None: + with pytest.raises(ValidationError, match="snapshot"): + CostEvent.model_validate( + { + "event_id": "cost", + "usage_event_id": "usage", + "occurred_at": _START, + "dimensions": (), + "charges": (_SPEND,), + **snapshot_fields, + } + ) + + +def test_upstream_charge_snapshot_preserves_reported_cost_without_inventing_rates() -> None: + snapshot: Final = _snapshot( + pricing_source="upstream_header", rates=(), usage=(), upstream_reported_cost=_SPEND.value + ) + restored: Final = PricingSnapshot.model_validate_json(snapshot.model_dump_json()) + assert restored.upstream_reported_cost == restored.total_cost == _SPEND.value + assert restored.rates == restored.usage == () + + +def test_subset_token_counts_round_trip_without_changing_totals() -> None: + metrics: Final = ( + Metric(name=MetricName.COMPLETION_TOKENS, value=Decimal(10)), + Metric(name=MetricName.REASONING_TOKENS, value=Decimal(3)), + Metric(name=MetricName.CACHE_CREATION_INPUT_TOKENS, value=Decimal(8)), + Metric(name=MetricName.CACHE_CREATION_INPUT_TOKENS_ABOVE_1HR, value=Decimal(2)), + ) + event: Final = _outbound(metrics=metrics) + assert OutboundUsageEvent.model_validate_json(event.model_dump_json()).metrics == metrics + + +@pytest.mark.parametrize( + "schema_path", + [ + "schema.prisma", + "litellm/proxy/schema.prisma", + "litellm-proxy-extras/litellm_proxy_extras/schema.prisma", + ], +) +@pytest.mark.parametrize( + ("enum_name", "python_enum"), + [ + ("AccountingEventKind", EventKind), + ("AccountingDeliveryState", DeliveryState), + ("AccountingDeliveryTarget", DeliveryTarget), + ], +) +def test_prisma_accepts_exactly_the_public_accounting_enum_values( + schema_path: str, enum_name: str, python_enum: type[Enum] +) -> None: + schema: Final = (Path(__file__).resolve().parents[3] / schema_path).read_text() + declaration: Final = re.search(r"enum " + enum_name + r"\s*\{([^}]+)\}", schema) + assert declaration is not None, f"Missing persisted enum {enum_name}" + assert frozenset(declaration[1].split()) == frozenset(member.value for member in python_enum) + + +def test_one_event_carries_every_matched_model_access_group(): + groups = ( + Dimension(name=DimensionName.MODEL_ACCESS_GROUP, value="gpt-group"), + Dimension(name=DimensionName.MODEL_ACCESS_GROUP, value="claude-group"), + ) + assert _outbound(dimensions=(_TEAM, *groups)).dimensions == (_TEAM, *groups) + with pytest.raises(ValidationError, match="Duplicate dimension values"): + _outbound(dimensions=(groups[0], groups[0])) + + +def test_snapshots_reject_duplicate_multipliers_and_contradictory_token_counts(): + multiplier = Multiplier(name="data_residency", value="eu", factor=Decimal("1.1")) + with pytest.raises(ValidationError, match="Duplicate multiplier names"): + _snapshot(multipliers=(multiplier, multiplier)) + with pytest.raises(ValidationError, match="reasoning_tokens cannot exceed completion_tokens"): + _snapshot( + usage=( + PricedQuantity(measure="input", unit="token", value=_TOKEN.value), + PricedQuantity(measure="completion_tokens", unit="token", value=Decimal(5)), + PricedQuantity(measure="reasoning_tokens", unit="token", value=Decimal(6)), + ) + ) + + +def test_agent_live_spend_is_a_delivery_target(): + assert DeliveryTarget.AGENT_SPEND.value == "LiteLLM_AgentsTable" From 66ddbaab8624703e0868a6366e994fce73b5cc90 Mon Sep 17 00:00:00 2001 From: Kerry Lu Date: Wed, 28 Oct 2026 14:48:40 -0700 Subject: [PATCH 04/41] fix(ci): spend logs guard requires the approval to cover the head's spend-log lines (#48904) * fix(ci): spend logs guard requires the approval to cover the head's spend-log lines An approving review survives later pushes (dismiss_stale_reviews_on_push is off on main), so a PR approved once could add more LiteLLM_SpendLogs schema or migration changes and the guard kept passing. The guard now reads the review's commit_id and passes only when the approval is on the head, or the head adds no spend-log line the approved commit did not already carry. * fix(ci): keep approved_by untouched, add latest_approval beside it for the stale-approval check --------- Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- ci_cd/spend_logs_guard.py | 47 +++++++++++++++++++++++--- tests/unit/test_spend_logs_guard.py | 52 ++++++++++++++++++++++++++++- 2 files changed, 93 insertions(+), 6 deletions(-) diff --git a/ci_cd/spend_logs_guard.py b/ci_cd/spend_logs_guard.py index a04999c47c0..64e33c4fb70 100644 --- a/ci_cd/spend_logs_guard.py +++ b/ci_cd/spend_logs_guard.py @@ -4,7 +4,9 @@ needs an approving review from the spend-log owner before it can merge. CODEOWNERS can only match file paths, and every schema change edits the same schema.prisma copies, so a path rule would put every migration on the owner. This gate reads the diff instead: when no added line in the watched files mentions the table it passes at once, otherwise it passes only once -the owner's latest review on the pull request is an approval. The workflow re-runs on every review. +the owner's latest review on the pull request is an approval that still covers every spend-log line +in the head (approvals survive pushes, so a later commit may not add spend-log lines the owner never +saw). The workflow re-runs on every review. """ from __future__ import annotations @@ -13,7 +15,7 @@ import argparse import re import subprocess import sys -from collections.abc import Sequence +from collections.abc import Callable, Sequence from pathlib import Path from typing import Final @@ -37,6 +39,7 @@ class Review(BaseModel): model_config = ConfigDict(frozen=True, extra="ignore") state: str user: Reviewer + commit_id: str | None = None _REVIEWS: Final = TypeAdapter(list[Review]) @@ -75,6 +78,24 @@ def approved_by(reviews: Sequence[Review], login: str) -> bool: return bool(states) and states[-1] == "APPROVED" +def latest_approval(reviews: Sequence[Review], login: str) -> Review | None: + """`login`'s most recent APPROVED review on the pull request, the one GitHub keeps in force after pushes.""" + own: Final = tuple(review for review in reviews if review.user.login == login and review.state == "APPROVED") + return own[-1] if own else None + + +def approval_covers( + approval: Review, head: str, hits: Sequence[str], hits_at: Callable[[str], Sequence[str] | None] +) -> bool: + """True when the approval is on the head, or the head adds no spend-log line the approved commit did not.""" + if approval.commit_id is None: + return False + if approval.commit_id == head: + return True + reviewed: Final = hits_at(approval.commit_id) + return reviewed is not None and set(hits) <= set(reviewed) + + def added_diff(base: str, head: str) -> str: result: Final = subprocess.run( ("git", "diff", "--unified=0", "--no-color", f"{base}..{head}", "--", *WATCHED, MIGRATIONS_PREFIX), @@ -85,6 +106,15 @@ def added_diff(base: str, head: str) -> str: return result.stdout +def fetched(commit: str) -> bool: + result: Final = subprocess.run(("git", "fetch", "--no-tags", "--depth=1", "origin", commit), check=False) + return result.returncode == 0 + + +def reviewed_hits(base: str, commit: str) -> tuple[str, ...] | None: + return spend_log_hits(added_diff(base, commit)) if fetched(commit) else None + + class Arguments(BaseModel): model_config = ConfigDict(frozen=True) base: str @@ -104,14 +134,21 @@ def main(argv: Sequence[str]) -> int: if not hits: sys.stdout.write("No added line in schema.prisma or a migration names LiteLLM_SpendLogs; nothing to review.\n") return 0 - if approved_by(_REVIEWS.validate_json(args.reviews.read_bytes()), args.approver): + reviews: Final = _REVIEWS.validate_json(args.reviews.read_bytes()) + approval: Final = latest_approval(reviews, args.approver) if approved_by(reviews, args.approver) else None + if approval is not None and approval_covers( + approval, args.head, hits, lambda commit: reviewed_hits(args.base, commit) + ): sys.stdout.write(f"LiteLLM_SpendLogs change approved by {args.approver}: {', '.join(hits)}\n") return 0 listed: Final = "".join(f" {hit}\n" for hit in hits) - sys.stdout.write( + reason: Final = ( f"This pull request changes LiteLLM_SpendLogs and needs an approving review from @{args.approver}:\n" - f"{listed}The check re-runs when a review is submitted.\n" + if approval is None + else f"LiteLLM_SpendLogs lines changed after @{args.approver} approved {approval.commit_id}, " + f"a new approving review is needed:\n" ) + sys.stdout.write(f"{reason}{listed}The check re-runs when a review is submitted.\n") return 1 diff --git a/tests/unit/test_spend_logs_guard.py b/tests/unit/test_spend_logs_guard.py index 9a033045dd0..6a3355e1c52 100644 --- a/tests/unit/test_spend_logs_guard.py +++ b/tests/unit/test_spend_logs_guard.py @@ -42,6 +42,10 @@ class TestSpendLogHits: assert guard.spend_log_hits(diff) == ("litellm/proxy/schema.prisma: a LiteLLM_SpendLogs",) +def _review(login: str, state: str, commit_id: str | None = "head") -> "guard.Review": + return guard.Review(state=state, user=guard.Reviewer(login=login), commit_id=commit_id) + + class TestApprovedBy: def test_the_latest_review_of_the_approver_decides(self): reviews: Final = tuple( @@ -61,6 +65,42 @@ class TestApprovedBy: assert guard.approved_by((), "kerry-berri") is False +class TestLatestApproval: + def test_the_approvers_last_approved_review_is_returned_with_its_commit(self): + reviews: Final = ( + _review("kerry-berri", "APPROVED", "old"), + _review("someone-else", "APPROVED", "head"), + _review("kerry-berri", "APPROVED", "new"), + _review("kerry-berri", "COMMENTED", "head"), + ) + assert guard.latest_approval(reviews, "kerry-berri") is reviews[2] + assert guard.latest_approval(reviews[:2], "kerry-berri") is reviews[0] + assert guard.latest_approval(reviews[1:2], "kerry-berri") is None + assert guard.latest_approval((), "kerry-berri") is None + + +class TestApprovalCovers: + _HITS: Final = ("schema.prisma: a LiteLLM_SpendLogs", "schema.prisma: b LiteLLM_SpendLogs") + + def test_an_approval_on_the_head_covers_it_without_reading_the_approved_commit(self): + def unreadable(commit: str) -> None: + raise AssertionError(f"read {commit}") + + assert guard.approval_covers(_review("kerry-berri", "APPROVED", "head"), "head", self._HITS, unreadable) is True + + def test_an_older_approval_covers_the_head_only_when_it_saw_every_spend_log_line(self): + approval: Final = _review("kerry-berri", "APPROVED", "old") + assert guard.approval_covers(approval, "head", self._HITS, lambda commit: self._HITS) is True + assert guard.approval_covers(approval, "head", self._HITS[:1], lambda commit: self._HITS) is True + assert guard.approval_covers(approval, "head", self._HITS, lambda commit: self._HITS[:1]) is False + + def test_an_approval_whose_commit_cannot_be_read_does_not_cover_the_head(self): + gone: Final = _review("kerry-berri", "APPROVED", "gone") + assert guard.approval_covers(gone, "head", self._HITS, lambda commit: None) is False + without_commit: Final = _review("kerry-berri", "APPROVED", None) + assert guard.approval_covers(without_commit, "head", self._HITS, lambda commit: self._HITS) is False + + class TestExitCode: def test_a_pull_request_without_spend_log_lines_passes_without_reviews(self, tmp_path, monkeypatch, capsys): monkeypatch.setattr(guard, "added_diff", lambda base, head: _diff("schema.prisma", " foo String", 1)) @@ -75,5 +115,15 @@ class TestExitCode: reviews.write_text('[{"user": {"login": "someone-else"}, "state": "APPROVED"}]') assert guard.main(["--base", "a", "--head", "b", "--approver", "kerry-berri", "--reviews", str(reviews)]) == 1 assert "schema.prisma: x LiteLLM_SpendLogs" in capsys.readouterr().out - reviews.write_text('[{"user": {"login": "kerry-berri"}, "state": "APPROVED"}]') + reviews.write_text('[{"user": {"login": "kerry-berri"}, "state": "APPROVED", "commit_id": "b"}]') assert guard.main(["--base", "a", "--head", "b", "--approver", "kerry-berri", "--reviews", str(reviews)]) == 0 + + def test_an_approval_before_a_new_spend_log_line_was_pushed_fails(self, tmp_path, monkeypatch, capsys): + monkeypatch.setattr(guard, "added_diff", lambda base, head: _diff("schema.prisma", " x LiteLLM_SpendLogs", 7)) + monkeypatch.setattr(guard, "reviewed_hits", lambda base, commit: () if commit == "old" else None, raising=False) + reviews: Final = tmp_path / "r.json" + reviews.write_text('[{"user": {"login": "kerry-berri"}, "state": "APPROVED", "commit_id": "old"}]') + assert guard.main(["--base", "a", "--head", "b", "--approver", "kerry-berri", "--reviews", str(reviews)]) == 1 + assert "changed after @kerry-berri approved old" in capsys.readouterr().out + reviews.write_text('[{"user": {"login": "kerry-berri"}, "state": "APPROVED", "commit_id": "gone"}]') + assert guard.main(["--base", "a", "--head", "b", "--approver", "kerry-berri", "--reviews", str(reviews)]) == 1 From c6ca22ebc58baa4398728bdcbba5b5156105ee8b Mon Sep 17 00:00:00 2001 From: yucheng-berri Date: Wed, 28 Oct 2026 14:49:48 -0700 Subject: [PATCH 05/41] fix(guardrails)!: make hide-secrets honor skip_tool_message_in_guardrail (#48689) * fix(guardrails): make hide-secrets honor skip_tool_message_in_guardrail With the flag on, per guardrail or in litellm_settings, hide-secrets leaves role: tool messages and Anthropic tool_result blocks as sent, as the unified guardrails do. Tool call arguments are still masked. * test(guardrails): type the skip-tool walker test and assert the whole request --- .../enterprise_callbacks/secret_detection.py | 8 ++- litellm/proxy/guardrails/_content_utils.py | 35 ++++++---- .../test_hide_secrets_tool_traffic.py | 68 +++++++++++++++++++ .../proxy/guardrails/test_content_utils.py | 29 ++++++++ 4 files changed, 126 insertions(+), 14 deletions(-) diff --git a/enterprise/litellm_enterprise/enterprise_callbacks/secret_detection.py b/enterprise/litellm_enterprise/enterprise_callbacks/secret_detection.py index 65e4d50587f..c190e680347 100644 --- a/enterprise/litellm_enterprise/enterprise_callbacks/secret_detection.py +++ b/enterprise/litellm_enterprise/enterprise_callbacks/secret_detection.py @@ -32,6 +32,7 @@ from litellm.integrations.custom_guardrail import ( CustomGuardrail, log_guardrail_information, ) +from litellm.llms.base_llm.guardrail_translation.utils import effective_skip_tool_message_for_guardrail from litellm.proxy._types import UserAPIKeyAuth from litellm.proxy.guardrails._content_utils import iter_tool_json_fields, map_tool_json_text, walk_user_text from litellm.types.guardrails import GuardrailEventHooks @@ -753,7 +754,12 @@ class _ENTERPRISE_SecretDetection(CustomGuardrail): _masked_entity_count.set({}) # Covers multimodal list content + Responses-API input. - inspected = walk_user_text(data, self.redact_text, self.redact_tool_payload) + self._redact_prompt(data) + inspected = walk_user_text( + data, + self.redact_text, + self.redact_tool_payload, + skip_tool_message=effective_skip_tool_message_for_guardrail(self), + ) + self._redact_prompt(data) if inspected == 0: # Image-only, empty-text, and unsupported payloads inspected diff --git a/litellm/proxy/guardrails/_content_utils.py b/litellm/proxy/guardrails/_content_utils.py index b4830b7c771..9bea22acd71 100644 --- a/litellm/proxy/guardrails/_content_utils.py +++ b/litellm/proxy/guardrails/_content_utils.py @@ -305,6 +305,11 @@ def iter_message_text(data: Mapping[str, object]) -> Iterator[str]: yield from _iter_text_parts_in_content(message.get("content")) +def _is_tool_message(message: object) -> bool: + fields: Final = _text_object(message) + return fields is not None and str(fields.get("role") or "").lower() == "tool" + + def _call_item(item: object) -> tuple[str, object] | None: """Return ``(field, value)`` for a Responses tool-call input item, else None.""" fields: Final = _text_object(item) @@ -316,11 +321,14 @@ def walk_user_text( data: dict[str, object], visit: Callable[[str], str], visit_tool: Callable[[JsonValue], JsonValue] | None = None, + *, + skip_tool_message: bool = False, ) -> int: """Rewrite every text fragment in place via ``visit``. Mutates ``data["messages"]`` and ``data["input"]`` (including tool payloads), ``data["system"]`` and - ``data["instructions"]``. Returns the number + ``data["instructions"]``. ``skip_tool_message`` leaves ``role: tool`` messages and ``tool_result`` + blocks as sent. Returns the number of fragments visited so callers can short-circuit when nothing was inspected. """ @@ -353,6 +361,8 @@ def walk_user_text( def _rewrite_part(part: Mapping[str, object], native: bool) -> Mapping[str, object]: nonlocal visited + if skip_tool_message and part.get("type") == "tool_result": + return part text: Final = _part_text(part) rewritten: Final = {**part, **({"text": visit(text)} if text is not None else {})} if text is not None: @@ -384,18 +394,17 @@ def walk_user_text( return content messages: Final = data.get("messages") - if isinstance(messages, list): - for message in messages: - if not isinstance(message, dict): - continue - if "content" in message: - message["content"] = _rewrite_content(message["content"]) - calls = message.get("tool_calls") - if isinstance(calls, list): - for call in calls: - function = call.get("function") if isinstance(call, dict) else None - if isinstance(function, dict) and "arguments" in function: - function["arguments"] = _rewrite_arguments(function["arguments"]) + for message in _object_list(messages) or (): + if (skip_tool_message and _is_tool_message(message)) or not isinstance(message, dict): + continue + if "content" in message: + message["content"] = _rewrite_content(message["content"]) + calls = message.get("tool_calls") + if isinstance(calls, list): + for call in calls: + function = call.get("function") if isinstance(call, dict) else None + if isinstance(function, dict) and "arguments" in function: + function["arguments"] = _rewrite_arguments(function["arguments"]) if "system" in data: data["system"] = _rewrite_content(data["system"]) # rebind-ok: preserve the native hook's in-place contract diff --git a/tests/integration/security/test_hide_secrets_tool_traffic.py b/tests/integration/security/test_hide_secrets_tool_traffic.py index 42fdc0f5a44..60b49520b1f 100644 --- a/tests/integration/security/test_hide_secrets_tool_traffic.py +++ b/tests/integration/security/test_hide_secrets_tool_traffic.py @@ -413,3 +413,71 @@ def test_resent_tool_history_is_scanned_once(single_worker: Rig, shape: str) -> second: Final = timed(lambda: single_worker.proxy.request("POST", path, following)) assert first.value.status_code == 200 and second.value.status_code == 200, (first.value.text, second.value.text) assert_within_factor_of_control(second.seconds, control=first.seconds, factor=0.5) + + +@pytest.mark.parametrize("placement", ("guardrail", "global")) +def test_skip_tool_message_in_guardrail_forwards_tool_results(tmp_path: Path, placement: str) -> None: + def configure(configuration: dict[str, JsonValue]) -> None: + guardrails: Final = configuration["guardrails"] + settings: Final = configuration["litellm_settings"] + assert isinstance(guardrails, list) and isinstance(guardrails[0], dict) and isinstance(settings, dict) + params: Final = guardrails[0]["litellm_params"] + assert isinstance(params, dict) + (params if placement == "guardrail" else settings)["skip_tool_message_in_guardrail"] = True + + with tool_rig(tmp_path, configure=configure) as rig: + rig.wire.drain() + replies: Final = ( + rig.proxy.request( + "POST", + "/v1/messages", + { + "model": "claude", + "max_tokens": 32, + "cache": {"no-cache": True}, + "messages": [ + {"role": "user", "content": TEXT}, + { + "role": "assistant", + "content": [ + {"type": "tool_use", "id": "tool_1", "name": "read_file", "input": {"env": TEXT}} + ], + }, + { + "role": "user", + "content": [{"type": "tool_result", "tool_use_id": "tool_1", "content": TEXT}], + }, + ], + }, + ), + rig.proxy.request( + "POST", + "/v1/chat/completions", + { + "model": "gpt", + "cache": {"no-cache": True}, + "messages": [ + {"role": "user", "content": TEXT}, + { + "role": "assistant", + "content": None, + "tool_calls": [ + { + "id": "call_1", + "type": "function", + "function": {"name": "read_file", "arguments": json.dumps({"env": TEXT})}, + } + ], + }, + {"role": "tool", "tool_call_id": "call_1", "content": TEXT}, + ], + }, + ), + ) + assert all(reply.status_code == 200 for reply in replies), [reply.text for reply in replies] + sent: Final = rig.wire.drain() + assert len(sent) == 2 + for request in sent: + messages = json.loads(request.body)["messages"] + assert SECRET not in json.dumps(messages[:2]) + assert SECRET in json.dumps(messages[2]) diff --git a/tests/unit/proxy/guardrails/test_content_utils.py b/tests/unit/proxy/guardrails/test_content_utils.py index a668d5f609d..21d96a3ea97 100644 --- a/tests/unit/proxy/guardrails/test_content_utils.py +++ b/tests/unit/proxy/guardrails/test_content_utils.py @@ -2,6 +2,7 @@ import json from functools import reduce +from typing import Final import pytest from pydantic import JsonValue @@ -924,6 +925,34 @@ def test_native_writer_visits_each_tool_and_system_slot(data): assert "payload" not in json.dumps(data) +def test_skip_tool_message_leaves_tool_results_and_masks_the_rest() -> None: + data: Final[dict[str, object]] = { + "messages": [ + {"role": "user", "content": "secret"}, + { + "role": "assistant", + "content": [{"type": "tool_use", "id": "tool", "name": "read", "input": {"env": "secret"}}], + "tool_calls": [{"id": "call", "function": {"name": "read", "arguments": '{"env": "secret"}'}}], + }, + {"role": "user", "content": [{"type": "tool_result", "tool_use_id": "tool", "content": "secret"}]}, + {"role": "tool", "tool_call_id": "call", "content": "secret"}, + ] + } + assert walk_user_text(data, lambda text: text.replace("secret", "[MASKED]"), skip_tool_message=True) == 3 + assert data == { + "messages": [ + {"role": "user", "content": "[MASKED]"}, + { + "role": "assistant", + "content": [{"type": "tool_use", "id": "tool", "name": "read", "input": {"env": "[MASKED]"}}], + "tool_calls": [{"id": "call", "function": {"name": "read", "arguments": '{"env": "[MASKED]"}'}}], + }, + {"role": "user", "content": [{"type": "tool_result", "tool_use_id": "tool", "content": "secret"}]}, + {"role": "tool", "tool_call_id": "call", "content": "secret"}, + ] + } + + @pytest.mark.parametrize( "part", ( From 5215c2d3f77cb7cad6db861f323a4e892f5c16bf Mon Sep 17 00:00:00 2001 From: Kerry Lu Date: Wed, 28 Oct 2026 14:59:23 -0700 Subject: [PATCH 06/41] fix(pricing): add databricks priority pay per token prices (#48891) * fix(pricing): add databricks priority pay per token prices * fix(pricing): bill gemini 3.6 flash cache writes at its databricks input rate * fix(pricing): drop image token priority rate the cost map schema does not carry --------- Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- ...odel_prices_and_context_window_backup.json | 233 +++++++++++++++--- model_prices_and_context_window.json | 233 +++++++++++++++--- .../test_databricks_cost_calculator.py | 26 ++ 3 files changed, 428 insertions(+), 64 deletions(-) diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index 3e0e86012a4..1f2bd7d7837 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -21589,7 +21589,13 @@ ], "supports_vision": true, "supports_audio_input": true, - "supports_video_input": true + "supports_video_input": true, + "input_cost_per_token_priority": 5.6252e-07, + "output_cost_per_token_priority": 3.37498e-06, + "cache_read_input_token_cost_priority": 5.628e-08, + "cache_creation_input_token_cost_priority": 5.6252e-07, + "input_cost_per_audio_token_priority": 1.12497e-06, + "cache_read_input_audio_token_cost_priority": 1.1249e-07 }, "databricks/databricks-gemini-3-1-flash-image": { "cache_creation_input_token_cost": 6.2503e-07, @@ -21676,7 +21682,14 @@ "supports_reasoning": true, "supports_tool_choice": true, "supports_video_input": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 4.50002e-06, + "output_cost_per_token_priority": 2.699998e-05, + "cache_read_input_token_cost_priority": 4.5003e-07, + "cache_creation_input_token_cost_priority": 4.50002e-06, + "input_cost_per_token_above_200k_tokens_priority": 8.99997e-06, + "output_cost_per_token_above_200k_tokens_priority": 4.049997e-05, + "cache_read_input_token_cost_above_200k_tokens_priority": 8.9999e-07 }, "databricks/databricks-gemini-3-flash": { "cache_creation_input_token_cost": 6.2503e-07, @@ -21710,7 +21723,13 @@ ], "supports_vision": true, "supports_audio_input": true, - "supports_video_input": true + "supports_video_input": true, + "input_cost_per_token_priority": 1.12497e-06, + "output_cost_per_token_priority": 6.75003e-06, + "cache_read_input_token_cost_priority": 1.1249e-07, + "cache_creation_input_token_cost_priority": 1.12497e-06, + "input_cost_per_audio_token_priority": 2.25001e-06, + "cache_read_input_audio_token_cost_priority": 2.2498e-07 }, "databricks/databricks-gemini-3-pro": { "cache_creation_input_token_cost": 2.49998e-06, @@ -21757,7 +21776,11 @@ "supports_reasoning": true, "supports_sampling_params": false, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 1.35002e-06, + "output_cost_per_token_priority": 6.75003e-06, + "cache_read_input_token_cost_priority": 1.3503e-07, + "cache_creation_input_token_cost_priority": 1.35002e-06 }, "databricks/databricks-gemini-3-7-flash": { "cache_creation_input_token_cost": 7.4998e-07, @@ -21785,10 +21808,14 @@ "supports_reasoning": true, "supports_sampling_params": false, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 1.35002e-06, + "output_cost_per_token_priority": 6.75003e-06, + "cache_read_input_token_cost_priority": 1.3503e-07, + "cache_creation_input_token_cost_priority": 1.35002e-06 }, "databricks/databricks-gemini-3-6-flash": { - "cache_creation_input_token_cost": 1.87502e-06, + "cache_creation_input_token_cost": 7.4998e-07, "cache_read_input_token_cost": 7.497e-08, "input_cost_per_token": 7.4998e-07, "input_dbu_cost_per_token": 1.0714e-05, @@ -21816,7 +21843,11 @@ "supports_reasoning": true, "supports_sampling_params": false, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 1.35002e-06, + "output_cost_per_token_priority": 6.75003e-06, + "cache_read_input_token_cost_priority": 1.3503e-07, + "cache_creation_input_token_cost_priority": 1.35002e-06 }, "databricks/databricks-gemini-3-5-flash": { "cache_creation_input_token_cost": 1.87502e-06, @@ -21846,7 +21877,11 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 3.37498e-06, + "output_cost_per_token_priority": 2.025002e-05, + "cache_read_input_token_cost_priority": 3.3747e-07, + "cache_creation_input_token_cost_priority": 3.37498e-06 }, "databricks/databricks-gemini-3-5-flash-lite": { "cache_creation_input_token_cost": 3.7499e-07, @@ -21878,7 +21913,11 @@ "supports_response_schema": true, "supports_sampling_params": false, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 6.7501e-07, + "output_cost_per_token_priority": 5.62499e-06, + "cache_read_input_token_cost_priority": 6.748e-08, + "cache_creation_input_token_cost_priority": 6.7501e-07 }, "databricks/databricks-gemma-3-12b": { "cache_creation_input_token_cost": 1.5001e-07, @@ -21921,7 +21960,11 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": false + "supports_vision": false, + "input_cost_per_token_priority": 2.45e-06, + "output_cost_per_token_priority": 7.7e-06, + "cache_read_input_token_cost_priority": 4.55e-07, + "cache_creation_input_token_cost_priority": 2.45e-06 }, "databricks/databricks-glm-5-3": { "cache_creation_input_token_cost": 1.4e-06, @@ -21951,7 +21994,11 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": false, - "thinking_always_on": true + "thinking_always_on": true, + "input_cost_per_token_priority": 2.45e-06, + "output_cost_per_token_priority": 7.7e-06, + "cache_read_input_token_cost_priority": 4.55e-07, + "cache_creation_input_token_cost_priority": 2.45e-06 }, "databricks/databricks-glm-5-3-flash": { "cache_creation_input_token_cost": 1.5001e-07, @@ -21980,7 +22027,11 @@ "supports_reasoning": true, "supports_tool_choice": true, "supports_vision": true, - "thinking_always_on": true + "thinking_always_on": true, + "input_cost_per_token_priority": 2.625e-07, + "output_cost_per_token_priority": 8.75e-07, + "cache_read_input_token_cost_priority": 5.25e-08, + "cache_creation_input_token_cost_priority": 2.625e-07 }, "databricks/databricks-gpt-5": { "cache_creation_input_token_cost": 1.24999e-06, @@ -21999,7 +22050,11 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, "supports_prompt_caching": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "input_cost_per_token_priority": 2.49998e-06, + "output_cost_per_token_priority": 1.999998e-05, + "cache_read_input_token_cost_priority": 2.4997e-07, + "cache_creation_input_token_cost_priority": 2.49998e-06 }, "databricks/databricks-gpt-5-1": { "cache_creation_input_token_cost": 1.24999e-06, @@ -22018,7 +22073,11 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, "supports_prompt_caching": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "input_cost_per_token_priority": 2.49998e-06, + "output_cost_per_token_priority": 1.999998e-05, + "cache_read_input_token_cost_priority": 2.4997e-07, + "cache_creation_input_token_cost_priority": 2.49998e-06 }, "databricks/databricks-gpt-5-2": { "cache_creation_input_token_cost": 1.75e-06, @@ -22037,7 +22096,11 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, "supports_prompt_caching": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "input_cost_per_token_priority": 3.5e-06, + "output_cost_per_token_priority": 2.8e-05, + "cache_read_input_token_cost_priority": 3.5e-07, + "cache_creation_input_token_cost_priority": 3.5e-06 }, "databricks/databricks-gpt-5-3-codex": { "cache_creation_input_token_cost": 1.75e-06, @@ -22056,7 +22119,11 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, "supports_prompt_caching": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "input_cost_per_token_priority": 3.5e-06, + "output_cost_per_token_priority": 2.8e-05, + "cache_read_input_token_cost_priority": 3.5e-07, + "cache_creation_input_token_cost_priority": 3.5e-06 }, "databricks/databricks-gpt-5-4": { "cache_creation_input_token_cost": 2.49998e-06, @@ -22087,7 +22154,14 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 5.00003e-06, + "output_cost_per_token_priority": 2.999997e-05, + "cache_read_input_token_cost_priority": 5.0001e-07, + "cache_creation_input_token_cost_priority": 5.00003e-06, + "input_cost_per_token_above_272k_tokens_priority": 5.00003e-06, + "output_cost_per_token_above_272k_tokens_priority": 2.999997e-05, + "cache_read_input_token_cost_above_272k_tokens_priority": 5.0001e-07 }, "databricks/databricks-gpt-5-4-mini": { "cache_creation_input_token_cost": 7.4998e-07, @@ -22115,7 +22189,11 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 1.50003e-06, + "output_cost_per_token_priority": 8.99997e-06, + "cache_read_input_token_cost_priority": 1.5001e-07, + "cache_creation_input_token_cost_priority": 1.50003e-06 }, "databricks/databricks-gpt-5-4-nano": { "cache_creation_input_token_cost": 1.9999e-07, @@ -22175,7 +22253,15 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 8.00002e-06, + "output_cost_per_token_priority": 4.000003e-05, + "cache_read_input_token_cost_priority": 8.0003e-07, + "cache_creation_input_token_cost_priority": 9.99999e-06, + "input_cost_per_token_above_272k_tokens_priority": 1.599997e-05, + "output_cost_per_token_above_272k_tokens_priority": 6.000001e-05, + "cache_read_input_token_cost_above_272k_tokens_priority": 1.59999e-06, + "cache_creation_input_token_cost_above_272k_tokens_priority": 1.999998e-05 }, "databricks/databricks-gpt-5-6-terra": { "cache_creation_input_token_cost": 2.49998e-06, @@ -22207,7 +22293,15 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 4.00001e-06, + "output_cost_per_token_priority": 2.399999e-05, + "cache_read_input_token_cost_priority": 3.9998e-07, + "cache_creation_input_token_cost_priority": 5.00003e-06, + "input_cost_per_token_above_272k_tokens_priority": 8.00002e-06, + "output_cost_per_token_above_272k_tokens_priority": 3.600002e-05, + "cache_read_input_token_cost_above_272k_tokens_priority": 8.0003e-07, + "cache_creation_input_token_cost_above_272k_tokens_priority": 9.99999e-06 }, "databricks/databricks-gpt-5-6-luna": { "cache_creation_input_token_cost": 2.4997e-07, @@ -22239,7 +22333,15 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 3.9998e-07, + "output_cost_per_token_priority": 2.40002e-06, + "cache_read_input_token_cost_priority": 3.997e-08, + "cache_creation_input_token_cost_priority": 5.0001e-07, + "input_cost_per_token_above_272k_tokens_priority": 8.0003e-07, + "output_cost_per_token_above_272k_tokens_priority": 3.60003e-06, + "cache_read_input_token_cost_above_272k_tokens_priority": 8.001e-08, + "cache_creation_input_token_cost_above_272k_tokens_priority": 1.00002e-06 }, "databricks/databricks-gpt-5-5": { "cache_creation_input_token_cost": 5.00003e-06, @@ -22270,7 +22372,14 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 1.249997e-05, + "output_cost_per_token_priority": 7.50001e-05, + "cache_read_input_token_cost_priority": 1.24999e-06, + "cache_creation_input_token_cost_priority": 1.249997e-05, + "input_cost_per_token_above_272k_tokens_priority": 1.249997e-05, + "output_cost_per_token_above_272k_tokens_priority": 7.50001e-05, + "cache_read_input_token_cost_above_272k_tokens_priority": 1.24999e-06 }, "databricks/databricks-gpt-5-5-pro": { "cache_creation_input_token_cost": 2.999997e-05, @@ -22319,7 +22428,11 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, "supports_prompt_caching": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "input_cost_per_token_priority": 4.5003e-07, + "output_cost_per_token_priority": 3.60003e-06, + "cache_read_input_token_cost_priority": 4.501e-08, + "cache_creation_input_token_cost_priority": 4.5003e-07 }, "databricks/databricks-gpt-5-nano": { "cache_creation_input_token_cost": 4.998e-08, @@ -22469,7 +22582,11 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 5.25e-06, + "output_cost_per_token_priority": 2.625e-05, + "cache_read_input_token_cost_priority": 5.25e-07, + "cache_creation_input_token_cost_priority": 5.25e-06 }, "databricks/system.ai.kimi-k3": { "litellm_provider": "databricks", @@ -22551,7 +22668,11 @@ "supports_reasoning": true, "supports_tool_choice": true, "supports_vision": false, - "thinking_always_on": true + "thinking_always_on": true, + "input_cost_per_token_priority": 3.85e-07, + "output_cost_per_token_priority": 3.85e-06, + "cache_read_input_token_cost_priority": 3.85e-07, + "cache_creation_input_token_cost_priority": 3.85e-07 }, "databricks/databricks-qwen3-next-80b-a3b-instruct": { "cache_creation_input_token_cost": 1.5001e-07, @@ -77774,7 +77895,11 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 5.25e-07, + "output_cost_per_token_priority": 2.1e-06, + "cache_read_input_token_cost_priority": 5.25e-08, + "cache_creation_input_token_cost_priority": 5.25e-07 }, "databricks/databricks-deepseek-v4-1-flash-1017": { "cache_creation_input_token_cost": 3.0002e-07, @@ -77897,7 +78022,11 @@ "supports_reasoning": true, "supports_tool_choice": true, "supports_video_input": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 4.50002e-06, + "output_cost_per_token_priority": 2.250003e-05, + "cache_read_input_token_cost_priority": 2.2498e-07, + "cache_creation_input_token_cost_priority": 4.50002e-06 }, "databricks/databricks-gpt-5-4-pro": { "cache_creation_input_token_cost": 2.999997e-05, @@ -77959,7 +78088,15 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 4.00001e-06, + "output_cost_per_token_priority": 1.99997e-05, + "cache_read_input_token_cost_priority": 1.9999e-07, + "cache_creation_input_token_cost_priority": 5.00003e-06, + "input_cost_per_token_above_272k_tokens_priority": 8.00002e-06, + "output_cost_per_token_above_272k_tokens_priority": 2.99999e-05, + "cache_read_input_token_cost_above_272k_tokens_priority": 3.9998e-07, + "cache_creation_input_token_cost_above_272k_tokens_priority": 9.99999e-06 }, "databricks/databricks-gpt-6-1-terra": { "cache_creation_input_token_cost": 1.24999e-06, @@ -77991,7 +78128,15 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 1.99997e-06, + "output_cost_per_token_priority": 9.99999e-06, + "cache_read_input_token_cost_priority": 1.0003e-07, + "cache_creation_input_token_cost_priority": 2.49998e-06, + "input_cost_per_token_above_272k_tokens_priority": 4.00001e-06, + "output_cost_per_token_above_272k_tokens_priority": 1.500002e-05, + "cache_read_input_token_cost_above_272k_tokens_priority": 1.9999e-07, + "cache_creation_input_token_cost_above_272k_tokens_priority": 5.00003e-06 }, "databricks/databricks-gpt-6-astra": { "cache_creation_input_token_cost": 1.249997e-05, @@ -78023,7 +78168,15 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 1.999998e-05, + "output_cost_per_token_priority": 9.99999e-05, + "cache_read_input_token_cost_priority": 1.99997e-06, + "cache_creation_input_token_cost_priority": 2.500001e-05, + "input_cost_per_token_above_272k_tokens_priority": 4.000003e-05, + "output_cost_per_token_above_272k_tokens_priority": 0.0001500002, + "cache_read_input_token_cost_above_272k_tokens_priority": 4.00001e-06, + "cache_creation_input_token_cost_above_272k_tokens_priority": 5.000002e-05 }, "databricks/databricks-gpt-6-luna": { "cache_creation_input_token_cost": 1.2502e-07, @@ -78055,7 +78208,15 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 1.9999e-07, + "output_cost_per_token_priority": 1.0003e-06, + "cache_read_input_token_cost_priority": 2.002e-08, + "cache_creation_input_token_cost_priority": 2.4997e-07, + "input_cost_per_token_above_272k_tokens_priority": 3.9998e-07, + "output_cost_per_token_above_272k_tokens_priority": 1.5001e-06, + "cache_read_input_token_cost_above_272k_tokens_priority": 3.997e-08, + "cache_creation_input_token_cost_above_272k_tokens_priority": 5.0001e-07 }, "databricks/databricks-gpt-6-sol": { "cache_creation_input_token_cost": 2.49998e-06, @@ -78087,7 +78248,15 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 4.00001e-06, + "output_cost_per_token_priority": 1.99997e-05, + "cache_read_input_token_cost_priority": 3.9998e-07, + "cache_creation_input_token_cost_priority": 5.00003e-06, + "input_cost_per_token_above_272k_tokens_priority": 8.00002e-06, + "output_cost_per_token_above_272k_tokens_priority": 2.99999e-05, + "cache_read_input_token_cost_above_272k_tokens_priority": 8.0003e-07, + "cache_creation_input_token_cost_above_272k_tokens_priority": 9.99999e-06 }, "databricks/databricks-gpt-image-1": { "cache_creation_input_token_cost": 5.00003e-06, diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index 3e0e86012a4..1f2bd7d7837 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -21589,7 +21589,13 @@ ], "supports_vision": true, "supports_audio_input": true, - "supports_video_input": true + "supports_video_input": true, + "input_cost_per_token_priority": 5.6252e-07, + "output_cost_per_token_priority": 3.37498e-06, + "cache_read_input_token_cost_priority": 5.628e-08, + "cache_creation_input_token_cost_priority": 5.6252e-07, + "input_cost_per_audio_token_priority": 1.12497e-06, + "cache_read_input_audio_token_cost_priority": 1.1249e-07 }, "databricks/databricks-gemini-3-1-flash-image": { "cache_creation_input_token_cost": 6.2503e-07, @@ -21676,7 +21682,14 @@ "supports_reasoning": true, "supports_tool_choice": true, "supports_video_input": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 4.50002e-06, + "output_cost_per_token_priority": 2.699998e-05, + "cache_read_input_token_cost_priority": 4.5003e-07, + "cache_creation_input_token_cost_priority": 4.50002e-06, + "input_cost_per_token_above_200k_tokens_priority": 8.99997e-06, + "output_cost_per_token_above_200k_tokens_priority": 4.049997e-05, + "cache_read_input_token_cost_above_200k_tokens_priority": 8.9999e-07 }, "databricks/databricks-gemini-3-flash": { "cache_creation_input_token_cost": 6.2503e-07, @@ -21710,7 +21723,13 @@ ], "supports_vision": true, "supports_audio_input": true, - "supports_video_input": true + "supports_video_input": true, + "input_cost_per_token_priority": 1.12497e-06, + "output_cost_per_token_priority": 6.75003e-06, + "cache_read_input_token_cost_priority": 1.1249e-07, + "cache_creation_input_token_cost_priority": 1.12497e-06, + "input_cost_per_audio_token_priority": 2.25001e-06, + "cache_read_input_audio_token_cost_priority": 2.2498e-07 }, "databricks/databricks-gemini-3-pro": { "cache_creation_input_token_cost": 2.49998e-06, @@ -21757,7 +21776,11 @@ "supports_reasoning": true, "supports_sampling_params": false, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 1.35002e-06, + "output_cost_per_token_priority": 6.75003e-06, + "cache_read_input_token_cost_priority": 1.3503e-07, + "cache_creation_input_token_cost_priority": 1.35002e-06 }, "databricks/databricks-gemini-3-7-flash": { "cache_creation_input_token_cost": 7.4998e-07, @@ -21785,10 +21808,14 @@ "supports_reasoning": true, "supports_sampling_params": false, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 1.35002e-06, + "output_cost_per_token_priority": 6.75003e-06, + "cache_read_input_token_cost_priority": 1.3503e-07, + "cache_creation_input_token_cost_priority": 1.35002e-06 }, "databricks/databricks-gemini-3-6-flash": { - "cache_creation_input_token_cost": 1.87502e-06, + "cache_creation_input_token_cost": 7.4998e-07, "cache_read_input_token_cost": 7.497e-08, "input_cost_per_token": 7.4998e-07, "input_dbu_cost_per_token": 1.0714e-05, @@ -21816,7 +21843,11 @@ "supports_reasoning": true, "supports_sampling_params": false, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 1.35002e-06, + "output_cost_per_token_priority": 6.75003e-06, + "cache_read_input_token_cost_priority": 1.3503e-07, + "cache_creation_input_token_cost_priority": 1.35002e-06 }, "databricks/databricks-gemini-3-5-flash": { "cache_creation_input_token_cost": 1.87502e-06, @@ -21846,7 +21877,11 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 3.37498e-06, + "output_cost_per_token_priority": 2.025002e-05, + "cache_read_input_token_cost_priority": 3.3747e-07, + "cache_creation_input_token_cost_priority": 3.37498e-06 }, "databricks/databricks-gemini-3-5-flash-lite": { "cache_creation_input_token_cost": 3.7499e-07, @@ -21878,7 +21913,11 @@ "supports_response_schema": true, "supports_sampling_params": false, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 6.7501e-07, + "output_cost_per_token_priority": 5.62499e-06, + "cache_read_input_token_cost_priority": 6.748e-08, + "cache_creation_input_token_cost_priority": 6.7501e-07 }, "databricks/databricks-gemma-3-12b": { "cache_creation_input_token_cost": 1.5001e-07, @@ -21921,7 +21960,11 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": false + "supports_vision": false, + "input_cost_per_token_priority": 2.45e-06, + "output_cost_per_token_priority": 7.7e-06, + "cache_read_input_token_cost_priority": 4.55e-07, + "cache_creation_input_token_cost_priority": 2.45e-06 }, "databricks/databricks-glm-5-3": { "cache_creation_input_token_cost": 1.4e-06, @@ -21951,7 +21994,11 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": false, - "thinking_always_on": true + "thinking_always_on": true, + "input_cost_per_token_priority": 2.45e-06, + "output_cost_per_token_priority": 7.7e-06, + "cache_read_input_token_cost_priority": 4.55e-07, + "cache_creation_input_token_cost_priority": 2.45e-06 }, "databricks/databricks-glm-5-3-flash": { "cache_creation_input_token_cost": 1.5001e-07, @@ -21980,7 +22027,11 @@ "supports_reasoning": true, "supports_tool_choice": true, "supports_vision": true, - "thinking_always_on": true + "thinking_always_on": true, + "input_cost_per_token_priority": 2.625e-07, + "output_cost_per_token_priority": 8.75e-07, + "cache_read_input_token_cost_priority": 5.25e-08, + "cache_creation_input_token_cost_priority": 2.625e-07 }, "databricks/databricks-gpt-5": { "cache_creation_input_token_cost": 1.24999e-06, @@ -21999,7 +22050,11 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, "supports_prompt_caching": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "input_cost_per_token_priority": 2.49998e-06, + "output_cost_per_token_priority": 1.999998e-05, + "cache_read_input_token_cost_priority": 2.4997e-07, + "cache_creation_input_token_cost_priority": 2.49998e-06 }, "databricks/databricks-gpt-5-1": { "cache_creation_input_token_cost": 1.24999e-06, @@ -22018,7 +22073,11 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, "supports_prompt_caching": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "input_cost_per_token_priority": 2.49998e-06, + "output_cost_per_token_priority": 1.999998e-05, + "cache_read_input_token_cost_priority": 2.4997e-07, + "cache_creation_input_token_cost_priority": 2.49998e-06 }, "databricks/databricks-gpt-5-2": { "cache_creation_input_token_cost": 1.75e-06, @@ -22037,7 +22096,11 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, "supports_prompt_caching": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "input_cost_per_token_priority": 3.5e-06, + "output_cost_per_token_priority": 2.8e-05, + "cache_read_input_token_cost_priority": 3.5e-07, + "cache_creation_input_token_cost_priority": 3.5e-06 }, "databricks/databricks-gpt-5-3-codex": { "cache_creation_input_token_cost": 1.75e-06, @@ -22056,7 +22119,11 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, "supports_prompt_caching": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "input_cost_per_token_priority": 3.5e-06, + "output_cost_per_token_priority": 2.8e-05, + "cache_read_input_token_cost_priority": 3.5e-07, + "cache_creation_input_token_cost_priority": 3.5e-06 }, "databricks/databricks-gpt-5-4": { "cache_creation_input_token_cost": 2.49998e-06, @@ -22087,7 +22154,14 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 5.00003e-06, + "output_cost_per_token_priority": 2.999997e-05, + "cache_read_input_token_cost_priority": 5.0001e-07, + "cache_creation_input_token_cost_priority": 5.00003e-06, + "input_cost_per_token_above_272k_tokens_priority": 5.00003e-06, + "output_cost_per_token_above_272k_tokens_priority": 2.999997e-05, + "cache_read_input_token_cost_above_272k_tokens_priority": 5.0001e-07 }, "databricks/databricks-gpt-5-4-mini": { "cache_creation_input_token_cost": 7.4998e-07, @@ -22115,7 +22189,11 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 1.50003e-06, + "output_cost_per_token_priority": 8.99997e-06, + "cache_read_input_token_cost_priority": 1.5001e-07, + "cache_creation_input_token_cost_priority": 1.50003e-06 }, "databricks/databricks-gpt-5-4-nano": { "cache_creation_input_token_cost": 1.9999e-07, @@ -22175,7 +22253,15 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 8.00002e-06, + "output_cost_per_token_priority": 4.000003e-05, + "cache_read_input_token_cost_priority": 8.0003e-07, + "cache_creation_input_token_cost_priority": 9.99999e-06, + "input_cost_per_token_above_272k_tokens_priority": 1.599997e-05, + "output_cost_per_token_above_272k_tokens_priority": 6.000001e-05, + "cache_read_input_token_cost_above_272k_tokens_priority": 1.59999e-06, + "cache_creation_input_token_cost_above_272k_tokens_priority": 1.999998e-05 }, "databricks/databricks-gpt-5-6-terra": { "cache_creation_input_token_cost": 2.49998e-06, @@ -22207,7 +22293,15 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 4.00001e-06, + "output_cost_per_token_priority": 2.399999e-05, + "cache_read_input_token_cost_priority": 3.9998e-07, + "cache_creation_input_token_cost_priority": 5.00003e-06, + "input_cost_per_token_above_272k_tokens_priority": 8.00002e-06, + "output_cost_per_token_above_272k_tokens_priority": 3.600002e-05, + "cache_read_input_token_cost_above_272k_tokens_priority": 8.0003e-07, + "cache_creation_input_token_cost_above_272k_tokens_priority": 9.99999e-06 }, "databricks/databricks-gpt-5-6-luna": { "cache_creation_input_token_cost": 2.4997e-07, @@ -22239,7 +22333,15 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 3.9998e-07, + "output_cost_per_token_priority": 2.40002e-06, + "cache_read_input_token_cost_priority": 3.997e-08, + "cache_creation_input_token_cost_priority": 5.0001e-07, + "input_cost_per_token_above_272k_tokens_priority": 8.0003e-07, + "output_cost_per_token_above_272k_tokens_priority": 3.60003e-06, + "cache_read_input_token_cost_above_272k_tokens_priority": 8.001e-08, + "cache_creation_input_token_cost_above_272k_tokens_priority": 1.00002e-06 }, "databricks/databricks-gpt-5-5": { "cache_creation_input_token_cost": 5.00003e-06, @@ -22270,7 +22372,14 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 1.249997e-05, + "output_cost_per_token_priority": 7.50001e-05, + "cache_read_input_token_cost_priority": 1.24999e-06, + "cache_creation_input_token_cost_priority": 1.249997e-05, + "input_cost_per_token_above_272k_tokens_priority": 1.249997e-05, + "output_cost_per_token_above_272k_tokens_priority": 7.50001e-05, + "cache_read_input_token_cost_above_272k_tokens_priority": 1.24999e-06 }, "databricks/databricks-gpt-5-5-pro": { "cache_creation_input_token_cost": 2.999997e-05, @@ -22319,7 +22428,11 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, "supports_prompt_caching": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "input_cost_per_token_priority": 4.5003e-07, + "output_cost_per_token_priority": 3.60003e-06, + "cache_read_input_token_cost_priority": 4.501e-08, + "cache_creation_input_token_cost_priority": 4.5003e-07 }, "databricks/databricks-gpt-5-nano": { "cache_creation_input_token_cost": 4.998e-08, @@ -22469,7 +22582,11 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 5.25e-06, + "output_cost_per_token_priority": 2.625e-05, + "cache_read_input_token_cost_priority": 5.25e-07, + "cache_creation_input_token_cost_priority": 5.25e-06 }, "databricks/system.ai.kimi-k3": { "litellm_provider": "databricks", @@ -22551,7 +22668,11 @@ "supports_reasoning": true, "supports_tool_choice": true, "supports_vision": false, - "thinking_always_on": true + "thinking_always_on": true, + "input_cost_per_token_priority": 3.85e-07, + "output_cost_per_token_priority": 3.85e-06, + "cache_read_input_token_cost_priority": 3.85e-07, + "cache_creation_input_token_cost_priority": 3.85e-07 }, "databricks/databricks-qwen3-next-80b-a3b-instruct": { "cache_creation_input_token_cost": 1.5001e-07, @@ -77774,7 +77895,11 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 5.25e-07, + "output_cost_per_token_priority": 2.1e-06, + "cache_read_input_token_cost_priority": 5.25e-08, + "cache_creation_input_token_cost_priority": 5.25e-07 }, "databricks/databricks-deepseek-v4-1-flash-1017": { "cache_creation_input_token_cost": 3.0002e-07, @@ -77897,7 +78022,11 @@ "supports_reasoning": true, "supports_tool_choice": true, "supports_video_input": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 4.50002e-06, + "output_cost_per_token_priority": 2.250003e-05, + "cache_read_input_token_cost_priority": 2.2498e-07, + "cache_creation_input_token_cost_priority": 4.50002e-06 }, "databricks/databricks-gpt-5-4-pro": { "cache_creation_input_token_cost": 2.999997e-05, @@ -77959,7 +78088,15 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 4.00001e-06, + "output_cost_per_token_priority": 1.99997e-05, + "cache_read_input_token_cost_priority": 1.9999e-07, + "cache_creation_input_token_cost_priority": 5.00003e-06, + "input_cost_per_token_above_272k_tokens_priority": 8.00002e-06, + "output_cost_per_token_above_272k_tokens_priority": 2.99999e-05, + "cache_read_input_token_cost_above_272k_tokens_priority": 3.9998e-07, + "cache_creation_input_token_cost_above_272k_tokens_priority": 9.99999e-06 }, "databricks/databricks-gpt-6-1-terra": { "cache_creation_input_token_cost": 1.24999e-06, @@ -77991,7 +78128,15 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 1.99997e-06, + "output_cost_per_token_priority": 9.99999e-06, + "cache_read_input_token_cost_priority": 1.0003e-07, + "cache_creation_input_token_cost_priority": 2.49998e-06, + "input_cost_per_token_above_272k_tokens_priority": 4.00001e-06, + "output_cost_per_token_above_272k_tokens_priority": 1.500002e-05, + "cache_read_input_token_cost_above_272k_tokens_priority": 1.9999e-07, + "cache_creation_input_token_cost_above_272k_tokens_priority": 5.00003e-06 }, "databricks/databricks-gpt-6-astra": { "cache_creation_input_token_cost": 1.249997e-05, @@ -78023,7 +78168,15 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 1.999998e-05, + "output_cost_per_token_priority": 9.99999e-05, + "cache_read_input_token_cost_priority": 1.99997e-06, + "cache_creation_input_token_cost_priority": 2.500001e-05, + "input_cost_per_token_above_272k_tokens_priority": 4.000003e-05, + "output_cost_per_token_above_272k_tokens_priority": 0.0001500002, + "cache_read_input_token_cost_above_272k_tokens_priority": 4.00001e-06, + "cache_creation_input_token_cost_above_272k_tokens_priority": 5.000002e-05 }, "databricks/databricks-gpt-6-luna": { "cache_creation_input_token_cost": 1.2502e-07, @@ -78055,7 +78208,15 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 1.9999e-07, + "output_cost_per_token_priority": 1.0003e-06, + "cache_read_input_token_cost_priority": 2.002e-08, + "cache_creation_input_token_cost_priority": 2.4997e-07, + "input_cost_per_token_above_272k_tokens_priority": 3.9998e-07, + "output_cost_per_token_above_272k_tokens_priority": 1.5001e-06, + "cache_read_input_token_cost_above_272k_tokens_priority": 3.997e-08, + "cache_creation_input_token_cost_above_272k_tokens_priority": 5.0001e-07 }, "databricks/databricks-gpt-6-sol": { "cache_creation_input_token_cost": 2.49998e-06, @@ -78087,7 +78248,15 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "input_cost_per_token_priority": 4.00001e-06, + "output_cost_per_token_priority": 1.99997e-05, + "cache_read_input_token_cost_priority": 3.9998e-07, + "cache_creation_input_token_cost_priority": 5.00003e-06, + "input_cost_per_token_above_272k_tokens_priority": 8.00002e-06, + "output_cost_per_token_above_272k_tokens_priority": 2.99999e-05, + "cache_read_input_token_cost_above_272k_tokens_priority": 8.0003e-07, + "cache_creation_input_token_cost_above_272k_tokens_priority": 9.99999e-06 }, "databricks/databricks-gpt-image-1": { "cache_creation_input_token_cost": 5.00003e-06, diff --git a/tests/unit/llms/databricks/test_databricks_cost_calculator.py b/tests/unit/llms/databricks/test_databricks_cost_calculator.py index 34d3806d6eb..c35aa6e4cf0 100644 --- a/tests/unit/llms/databricks/test_databricks_cost_calculator.py +++ b/tests/unit/llms/databricks/test_databricks_cost_calculator.py @@ -253,3 +253,29 @@ def test_cost_per_token_bills_the_served_priority_tier( prompt_cost, completion_cost = cost_per_token(model="databricks/dbrx-tiered-test", usage=usage) assert prompt_cost == pytest.approx(30 * 0.001) assert completion_cost == pytest.approx(40 * 0.002) + + +RATES_WITHOUT_A_PRIORITY_FIELD: Final = frozenset({"input_cost_per_image_token"}) + + +def _is_standard_rate_key(key: str) -> bool: + return ( + ("_cost_per_" in key or key.endswith("_token_cost")) + and "dbu" not in key + and key not in RATES_WITHOUT_A_PRIORITY_FIELD + and not key.endswith(("_priority", "_flex", "_batches")) + ) + + +def test_every_databricks_priority_row_prices_every_standard_rate_at_the_priority_tier() -> None: + cost_map: Final = json.loads(MAIN_PRICES.read_text()) + priority_rows: Final = { + key: row + for key, row in cost_map.items() + if key.startswith("databricks/") and "input_cost_per_token_priority" in row + } + + assert priority_rows + for key, row in priority_rows.items(): + missing = sorted(f for f in row if _is_standard_rate_key(f) and f"{f}_priority" not in row) + assert not missing, f"{key} bills these at the standard rate on priority requests: {missing}" From b8fb50f70e82f5eed8a8d56cdb5b6b1ca4e1e7b7 Mon Sep 17 00:00:00 2001 From: Kerry Lu Date: Wed, 28 Oct 2026 15:03:40 -0700 Subject: [PATCH 07/41] fix(ci): keep the spend logs guard approval through a later COMMENTED review (LIT-10242) (#48901) Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- ci_cd/spend_logs_guard.py | 11 ++++++++--- tests/unit/test_spend_logs_guard.py | 23 +++++++++++++++++++++++ 2 files changed, 31 insertions(+), 3 deletions(-) diff --git a/ci_cd/spend_logs_guard.py b/ci_cd/spend_logs_guard.py index 64e33c4fb70..fac9d4ba5d2 100644 --- a/ci_cd/spend_logs_guard.py +++ b/ci_cd/spend_logs_guard.py @@ -27,7 +27,8 @@ WATCHED: Final = ( "litellm-proxy-extras/litellm_proxy_extras/schema.prisma", ) MIGRATIONS_PREFIX: Final = "litellm-proxy-extras/litellm_proxy_extras/migrations/" -TABLE: Final = re.compile(r"(? str: def approved_by(reviews: Sequence[Review], login: str) -> bool: - """True when `login`'s most recent review on the pull request is an approval.""" - states: Final = tuple(review.state for review in reviews if review.user.login == login) + """True when `login`'s most recent review that changes GitHub's approval state is an approval. + + COMMENTED and PENDING reviews leave an earlier approval in force on GitHub, so they are skipped here too. + """ + own: Final = tuple(review.state for review in reviews if review.user.login == login) + states: Final = tuple(state for state in own if state in DECIDING_STATES) return bool(states) and states[-1] == "APPROVED" diff --git a/tests/unit/test_spend_logs_guard.py b/tests/unit/test_spend_logs_guard.py index 6a3355e1c52..9ac2c923632 100644 --- a/tests/unit/test_spend_logs_guard.py +++ b/tests/unit/test_spend_logs_guard.py @@ -3,6 +3,8 @@ import sys from pathlib import Path from typing import Final +import pytest + _SCRIPT: Final = Path(__file__).resolve().parents[2] / "ci_cd" / "spend_logs_guard.py" _SPEC: Final = importlib.util.spec_from_file_location("spend_logs_guard", _SCRIPT) assert _SPEC is not None and _SPEC.loader is not None @@ -35,6 +37,13 @@ class TestSpendLogHits: elsewhere: Final = _diff("litellm/proxy/utils.py", 'table = "LiteLLM_SpendLogs"', 5) assert guard.spend_log_hits(removed + other + elsewhere) == () + def test_a_table_whose_name_only_starts_with_the_watched_one_is_not_a_hit(self): + archive: Final = _diff("schema.prisma", "model LiteLLM_SpendLogsArchive {", 900) + index: Final = _diff("schema.prisma", " @@index([team_id], map: \"LiteLLM_SpendLogs_team_id_idx\")", 760) + assert guard.spend_log_hits(archive + index) == ( + 'schema.prisma: @@index([team_id], map: "LiteLLM_SpendLogs_team_id_idx")', + ) + def test_two_files_in_one_diff_are_attributed_to_their_own_path(self): diff: Final = _diff("litellm/proxy/schema.prisma", " a LiteLLM_SpendLogs", 1) + _diff( "schema.prisma", " b String", 1 @@ -59,6 +68,20 @@ class TestApprovedBy: assert guard.approved_by(reviews, "kerry-berri") is True assert guard.approved_by(reviews[::-1], "kerry-berri") is False + @pytest.mark.parametrize( + ("states", "approved"), + ( + (("APPROVED", "COMMENTED"), True), + (("APPROVED", "PENDING"), True), + (("CHANGES_REQUESTED", "COMMENTED"), False), + (("APPROVED", "DISMISSED"), False), + (("COMMENTED",), False), + ), + ) + def test_a_comment_or_pending_review_does_not_change_the_answer(self, states, approved): + reviews: Final = tuple(guard.Review(state=state, user=guard.Reviewer(login="kerry-berri")) for state in states) + assert guard.approved_by(reviews, "kerry-berri") is approved + def test_no_review_from_the_approver_is_not_an_approval(self): other: Final = guard.Review(state="APPROVED", user=guard.Reviewer(login="someone-else")) assert guard.approved_by((other,), "kerry-berri") is False From e21de0e418c076be488eca83cf97d2ba3c1998cd Mon Sep 17 00:00:00 2001 From: Kerry Lu Date: Wed, 28 Oct 2026 15:11:26 -0700 Subject: [PATCH 08/41] fix(logging): bill each SDK call its own native Bedrock guardrail trace when calls share one metadata dict (LIT-10239) (#48923) Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- litellm/litellm_core_utils/litellm_logging.py | 2 +- ...ck_native_guardrail_shared_metadata_sdk.py | 67 +++++++++++++++++++ .../test_litellm_logging.py | 32 +++++++++ 3 files changed, 100 insertions(+), 1 deletion(-) create mode 100644 tests/integration/sdk/test_bedrock_native_guardrail_shared_metadata_sdk.py diff --git a/litellm/litellm_core_utils/litellm_logging.py b/litellm/litellm_core_utils/litellm_logging.py index 3b8157b2249..80f1f8f5eca 100644 --- a/litellm/litellm_core_utils/litellm_logging.py +++ b/litellm/litellm_core_utils/litellm_logging.py @@ -2758,7 +2758,7 @@ class Logging(LiteLLMLoggingBaseClass): if entry is None or not is_str_keyed_dict(litellm_params): return metadata: Final[object] = litellm_params.get("metadata") - bucket: Final = metadata if is_str_keyed_dict(metadata) else {} + bucket: Final = {**metadata} if is_str_keyed_dict(metadata) else {} litellm_params["metadata"] = bucket with_guardrail_entry(bucket, entry) diff --git a/tests/integration/sdk/test_bedrock_native_guardrail_shared_metadata_sdk.py b/tests/integration/sdk/test_bedrock_native_guardrail_shared_metadata_sdk.py new file mode 100644 index 00000000000..d0235310aaa --- /dev/null +++ b/tests/integration/sdk/test_bedrock_native_guardrail_shared_metadata_sdk.py @@ -0,0 +1,67 @@ +"""An SDK caller that reuses one ``metadata=`` dict across ``litellm.completion`` calls to a Bedrock deployment whose +``guardrailConfig`` has ``trace: enabled`` is billed, on each call, for that call's guardrail trace only, and the +caller's dict is left as the caller wrote it (LIT-10239)""" + +import json +from collections.abc import Mapping +from typing import Final + +import pytest +from integration._support.client import eventually +from integration._support.wire import Reply, Request, wire_server +from pydantic import JsonValue + +import litellm + +_MODEL: Final = "bedrock/anthropic.claude-sonnet-4-5-20250929-v1:0" +_PROMPT_TOKENS: Final = 5 +_COMPLETION_TOKENS: Final = 1 +_USAGE: Final[Mapping[str, int]] = {"contentPolicyUnits": 1, "topicPolicyUnits": 1} +_GUARDRAIL_IDS: Final = ("guardrail-one", "guardrail-two") + + +def _bedrock(request: Request) -> Reply: + """Bedrock's Converse runtime: a one-token answer and the trace of the guardrail the request named.""" + guardrail_id: Final = json.loads(request.body)["guardrailConfig"]["guardrailIdentifier"] + body: Final[dict[str, JsonValue]] = { + "output": {"message": {"role": "assistant", "content": [{"text": "ok"}]}}, + "stopReason": "end_turn", + "usage": {"inputTokens": _PROMPT_TOKENS, "outputTokens": _COMPLETION_TOKENS, "totalTokens": 6}, + "metrics": {"latencyMs": 3}, + "trace": {"guardrail": {"inputAssessment": {guardrail_id: {"invocationMetrics": {"usage": dict(_USAGE)}}}}}, + } + return Reply(body=json.dumps(body).encode()) + + +def _guardrail_cost() -> float: + rates: Final = litellm.model_cost["bedrock/guardrails"]["guardrail_cost_per_unit"] + return sum(units * rates[counter] for counter, units in _USAGE.items()) + + +def _llm_cost() -> float: + prices: Final = litellm.model_cost[_MODEL.removeprefix("bedrock/")] + return _PROMPT_TOKENS * prices["input_cost_per_token"] + _COMPLETION_TOKENS * prices["output_cost_per_token"] + + +def test_each_call_sharing_one_metadata_dict_is_billed_its_own_trace_only() -> None: + billed: list[float] = [] # mutable-ok: the success callback appends what each call was billed + litellm.success_callback.append( + lambda kwargs, response, start, end: billed.append(kwargs["standard_logging_object"]["response_cost"]) + ) + shared_metadata: Final = {"tag": "x"} + with wire_server(_bedrock) as wire: + for guardrail_id in _GUARDRAIL_IDS: + litellm.completion( + model=_MODEL, + messages=[{"role": "user", "content": "hello"}], + api_base=wire.url, + aws_access_key_id="AKIASCRIPTED", + aws_secret_access_key="scripted", + aws_region_name="us-east-1", + guardrailConfig={"guardrailIdentifier": guardrail_id, "guardrailVersion": "1", "trace": "enabled"}, + metadata=shared_metadata, + ) + costs: Final = eventually(lambda: tuple(billed), lambda seen: len(seen) == len(_GUARDRAIL_IDS), seconds=30) + + assert costs == pytest.approx((_llm_cost() + _guardrail_cost(),) * len(_GUARDRAIL_IDS)) + assert shared_metadata == {"tag": "x"} diff --git a/tests/unit/litellm_core_utils/test_litellm_logging.py b/tests/unit/litellm_core_utils/test_litellm_logging.py index e786753e1a8..299088a6c4f 100644 --- a/tests/unit/litellm_core_utils/test_litellm_logging.py +++ b/tests/unit/litellm_core_utils/test_litellm_logging.py @@ -13775,3 +13775,35 @@ def test_record_native_bedrock_guardrail_bills_nothing_for_a_response_served_fro logging_obj.record_native_bedrock_guardrail({"trace": trace}) assert "standard_logging_guardrail_information" not in logging_obj.model_call_details["litellm_params"]["metadata"] + + +def test_record_native_bedrock_guardrail_bills_each_call_its_own_trace_when_calls_share_one_metadata_dict() -> None: + import datetime + + from litellm.litellm_core_utils.litellm_logging import Logging + from litellm.litellm_core_utils.llm_cost_calc.guardrail_cost import guardrail_information_cost + + shared_metadata = {"tag": "x"} + traces = tuple( + {"guardrail": {"inputAssessment": {"gid": {"invocationMetrics": {"usage": {"contentPolicyUnits": units}}}}}} + for units in (1, 2) + ) + costs = [] + for call_index, trace in enumerate(traces): + logging_obj = Logging( + model="bedrock/anthropic.claude-sonnet-4-5-20250929-v1:0", + messages=[{"role": "user", "content": "hello"}], + stream=False, + call_type="completion", + start_time=datetime.datetime.now(), + litellm_call_id=f"call-shared-metadata-{call_index}", + function_id="fn", + ) + logging_obj.model_call_details["litellm_params"] = {"metadata": shared_metadata} + logging_obj.record_native_bedrock_guardrail({"trace": trace}) + recorded = logging_obj.model_call_details["litellm_params"]["metadata"]["standard_logging_guardrail_information"] + costs.append(guardrail_information_cost(recorded)) + + (first_cost, second_cost) = costs + assert second_cost == pytest.approx(2 * first_cost) + assert shared_metadata == {"tag": "x"} From 4a70d80aef0c600e4c493a77ab0a69d001d3d3bc Mon Sep 17 00:00:00 2001 From: Kerry Lu Date: Wed, 28 Oct 2026 15:23:52 -0700 Subject: [PATCH 09/41] fix(aiohttp_transport): keep the owned session open while a response stream is still being read (LIT-10237) (#48920) * fix(aiohttp_transport): keep the owned session open while a response stream is still being read (LIT-10237) * test(aiohttp_transport): annotate the stream regression test locals with Final and drop its docstring --------- Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- .../llms/custom_httpx/aiohttp_transport.py | 23 +++++++- .../custom_httpx/test_aiohttp_transport.py | 55 +++++++++++++++++++ 2 files changed, 77 insertions(+), 1 deletion(-) diff --git a/litellm/llms/custom_httpx/aiohttp_transport.py b/litellm/llms/custom_httpx/aiohttp_transport.py index 0e4d97fd770..9297e76624d 100644 --- a/litellm/llms/custom_httpx/aiohttp_transport.py +++ b/litellm/llms/custom_httpx/aiohttp_transport.py @@ -6,6 +6,7 @@ import ssl import sys import typing import urllib.request +import weakref from collections.abc import Callable, Generator from typing import ClassVar, Final @@ -109,6 +110,24 @@ def map_aiohttp_exceptions() -> Generator[None, None, None]: raise mapped_exc(message) from exc +def _release_transport_anchor(_transport: object) -> None: + """Release a transport anchored to a response stream. See ``_anchor_transport_to``.""" + + +def _anchor_transport_to(stream: "AiohttpResponseStream", transport: object) -> None: + """Keep the transport alive for as long as the stream can still read. + + The transport finalizer closes the session it owns, and nothing on the path from a + live response back to that session references the transport, so a client dropped + mid-stream would close the session under the body. ``weakref.finalize`` holds the + transport in its own registry rather than on the stream, which keeps the transport + out of the response's reference cycle (an exception carrying the response, for one) + so it is finalized by refcount once the anchor drops and can still schedule an async + close, instead of being collected in the same cyclic pass as the session it closes. + """ + weakref.finalize(stream, _release_transport_anchor, transport) + + class AiohttpResponseStream(httpx.AsyncByteStream): CHUNK_SIZE = 1024 * 16 @@ -463,10 +482,12 @@ class LiteLLMAiohttpTransport(AiohttpTransport): # Re-raise if it's a different RuntimeError raise + stream: Final = AiohttpResponseStream(response) + _anchor_transport_to(stream, self) return httpx.Response( status_code=response.status, headers=response.headers, - stream=AiohttpResponseStream(response), + stream=stream, request=request, ) diff --git a/tests/unit/llms/custom_httpx/test_aiohttp_transport.py b/tests/unit/llms/custom_httpx/test_aiohttp_transport.py index c7d5518ef5b..aed808223bd 100644 --- a/tests/unit/llms/custom_httpx/test_aiohttp_transport.py +++ b/tests/unit/llms/custom_httpx/test_aiohttp_transport.py @@ -5,6 +5,7 @@ import socket import sys import struct import threading +import weakref from typing import Final import aiohttp @@ -1383,3 +1384,57 @@ def test_a_transport_collected_while_its_loop_runs_elsewhere_closes_its_session_ loop.call_soon_threadsafe(loop.stop) thread.join(timeout=10) loop.close() + + +@pytest.mark.asyncio +async def test_a_stream_keeps_reading_after_the_client_that_opened_it_is_collected(): + from aiohttp import web + + release: Final = asyncio.Event() + body: Final = tuple(f"chunk{i}\n".encode() for i in range(4)) + + async def handler(request: web.Request) -> web.StreamResponse: + response: Final = web.StreamResponse() + await response.prepare(request) + await response.write(body[0]) + await release.wait() + for chunk in body[1:]: + await response.write(chunk) + await response.write_eof() + return response + + app: Final = web.Application() + app.router.add_get("/stream", handler) + runner: Final = web.AppRunner(app) + await runner.setup() + sock: Final = socket.socket() + sock.bind(("127.0.0.1", 0)) + port: Final[int] = sock.getsockname()[1] + await web.SockSite(runner, sock).start() + + transport: Final = LiteLLMAiohttpTransport(client=aiohttp.ClientSession) + client: Final = httpx.AsyncClient(transport=transport) + transport_ref: Final = weakref.ref(transport) + try: + response: Final = await client.send(client.build_request("GET", f"http://127.0.0.1:{port}/stream"), stream=True) + session: Final = transport.client + assert isinstance(session, aiohttp.ClientSession) + chunks: Final = response.aiter_bytes() + first: Final = await chunks.__anext__() + + del client, transport + gc.collect() + await asyncio.gather(*LiteLLMAiohttpTransport._background_close_tasks) + release.set() + + rest: Final = [chunk async for chunk in chunks] + assert first + b"".join(rest) == b"".join(body), (first, rest) + assert not session.closed, "the session was closed under a stream that was still being read" + + del response, chunks + gc.collect() + await asyncio.gather(*LiteLLMAiohttpTransport._background_close_tasks) + assert transport_ref() is None, "the finished stream kept the transport alive" + assert session.closed, "the owned session stayed open after its last stream finished" + finally: + await runner.cleanup() From 2791822f921fab9469b86ed37d8a2deadad0fc58 Mon Sep 17 00:00:00 2001 From: Kerry Lu Date: Wed, 28 Oct 2026 15:28:35 -0700 Subject: [PATCH 10/41] ci(spend-logs-guard): run only on pull_request_target, re-run by hand after approving (#48941) Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- .github/workflows/spend-logs-guard.yml | 4 ---- ci_cd/spend_logs_guard.py | 4 ++-- 2 files changed, 2 insertions(+), 6 deletions(-) diff --git a/.github/workflows/spend-logs-guard.yml b/.github/workflows/spend-logs-guard.yml index 354a1b7c947..29e9f236202 100644 --- a/.github/workflows/spend-logs-guard.yml +++ b/.github/workflows/spend-logs-guard.yml @@ -4,10 +4,6 @@ on: # zizmor: ignore[dangerous-triggers] runs the base branch's code only; the P pull_request_target: branches: - main - pull_request_review: - types: - - submitted - - dismissed permissions: contents: read diff --git a/ci_cd/spend_logs_guard.py b/ci_cd/spend_logs_guard.py index fac9d4ba5d2..a4d9a2e0a0d 100644 --- a/ci_cd/spend_logs_guard.py +++ b/ci_cd/spend_logs_guard.py @@ -6,7 +6,7 @@ so a path rule would put every migration on the owner. This gate reads the diff added line in the watched files mentions the table it passes at once, otherwise it passes only once the owner's latest review on the pull request is an approval that still covers every spend-log line in the head (approvals survive pushes, so a later commit may not add spend-log lines the owner never -saw). The workflow re-runs on every review. +saw). Reviews do not trigger the workflow, so the check is re-run by hand after approving. """ from __future__ import annotations @@ -153,7 +153,7 @@ def main(argv: Sequence[str]) -> int: else f"LiteLLM_SpendLogs lines changed after @{args.approver} approved {approval.commit_id}, " f"a new approving review is needed:\n" ) - sys.stdout.write(f"{reason}{listed}The check re-runs when a review is submitted.\n") + sys.stdout.write(f"{reason}{listed}After approving, re-run this job.\n") return 1 From 84c4c5aa5c530610863afe104b32dc50ebca37da Mon Sep 17 00:00:00 2001 From: Kerry Lu Date: Wed, 28 Oct 2026 15:32:27 -0700 Subject: [PATCH 11/41] fix(router): cool down failing deployments when fallbacks use string targets (LIT-10238) (#48913) Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- litellm/router_utils/cooldown_handlers.py | 32 ++++++++++++++----- .../sdk/test_router_cooldown_fallback_sdk.py | 28 ++++++++++++++++ .../router_utils/test_cooldown_handlers.py | 32 +++++++++++++++++++ 3 files changed, 84 insertions(+), 8 deletions(-) diff --git a/litellm/router_utils/cooldown_handlers.py b/litellm/router_utils/cooldown_handlers.py index 128bca6cca0..a1b1e327802 100644 --- a/litellm/router_utils/cooldown_handlers.py +++ b/litellm/router_utils/cooldown_handlers.py @@ -13,7 +13,7 @@ from datetime import datetime from types import MappingProxyType from typing import TYPE_CHECKING, Any, Final -from pydantic import TypeAdapter +from pydantic import TypeAdapter, ValidationError import litellm from litellm._internal_context import service_target @@ -417,17 +417,35 @@ def _should_run_cooldown_logic( return True -_FALLBACK_ENTRIES_ADAPTER: Final = TypeAdapter(tuple[Mapping[str, list[str | Mapping[str, object]]] | str, ...]) +_FALLBACK_ENTRIES_ADAPTER: Final = TypeAdapter(tuple[object, ...]) +_FALLBACK_ENTRY_ADAPTER: Final = TypeAdapter(Mapping[str, list[str | Mapping[str, object]] | str]) def _fallback_target_name(target: str | Mapping[str, object]) -> str: return target if isinstance(target, str) else str(target.get("model", "")) +def _fallback_target_names( + targets: Sequence[str | Mapping[str, object]] | str, +) -> list[str]: # mutable-ok: mirrors get_fallback_model_group's fallbacks shape + return [targets] if isinstance(targets, str) else [_fallback_target_name(target) for target in targets] + + def _fallback_chain_names( - entry: Mapping[str, Sequence[str | Mapping[str, object]]], -) -> Mapping[str, list[str]]: # mutable-ok: mirrors get_fallback_model_group's fallbacks shape - return {key: [_fallback_target_name(target) for target in targets] for key, targets in entry.items()} + entry: object, +) -> Mapping[str, list[str]] | None: # mutable-ok: mirrors get_fallback_model_group's fallbacks shape + try: + validated: Final = _FALLBACK_ENTRY_ADAPTER.validate_python(entry) + except ValidationError: + return None + return {key: _fallback_target_names(targets) for key, targets in validated.items()} + + +def _fallback_entries(configured: object) -> tuple[object, ...]: + try: + return _FALLBACK_ENTRIES_ADAPTER.validate_python(configured) + except ValidationError: + return () def _has_fallback_group(litellm_router_instance: LitellmRouter, requested_model_group: str | None) -> bool: @@ -437,9 +455,7 @@ def _has_fallback_group(litellm_router_instance: LitellmRouter, requested_model_ if requested_model_group is None or not configured: return False chains: Final = tuple( - _fallback_chain_names(entry) - for entry in _FALLBACK_ENTRIES_ADAPTER.validate_python(configured) - if isinstance(entry, Mapping) + chain for chain in map(_fallback_chain_names, _fallback_entries(configured)) if chain is not None ) fallback_group, _ = get_fallback_model_group(fallbacks=chains, model_group=requested_model_group) return bool(fallback_group) diff --git a/tests/integration/sdk/test_router_cooldown_fallback_sdk.py b/tests/integration/sdk/test_router_cooldown_fallback_sdk.py index 747adeee41d..e7508a2670e 100644 --- a/tests/integration/sdk/test_router_cooldown_fallback_sdk.py +++ b/tests/integration/sdk/test_router_cooldown_fallback_sdk.py @@ -12,6 +12,7 @@ from pydantic import JsonValue, TypeAdapter import litellm from litellm import Router +from litellm.router_utils.cooldown_handlers import async_get_cooldown_deployments from litellm.types.router import DeploymentTypedDict _MODEL: Final = "gpt-5.4" @@ -118,3 +119,30 @@ async def test_cooldown_skips_only_the_failed_duplicate_model_deployment() -> No "/primary/chat/completions", "/backup/chat/completions", ) + + +@pytest.mark.parametrize( + "router_kwargs", + [{"fallbacks": [{"primary": "backup"}]}, {"default_fallbacks": "backup"}], + ids=["string_fallback_target", "string_default_fallbacks"], +) +@pytest.mark.asyncio +async def test_string_valued_fallbacks_still_cool_down_the_failing_primary(router_kwargs: dict[str, object]) -> None: + with wire_server(_peer) as wire: + router: Final = Router( + model_list=[ + _deployment("primary", f"{wire.url}/primary", "primary-deployment"), + _deployment("backup", f"{wire.url}/backup", "backup-deployment"), + ], + allowed_fails=0, + num_retries=0, + **router_kwargs, + ) + with pytest.raises(litellm.AuthenticationError): + await router.acompletion( + model="primary", + messages=[{"role": "user", "content": f"cooldown {uuid.uuid4().hex}"}], + ) + cooled: Final = await async_get_cooldown_deployments(litellm_router_instance=router, parent_otel_span=None) + + assert tuple(cooled) == ("primary-deployment",) diff --git a/tests/unit/router_utils/test_cooldown_handlers.py b/tests/unit/router_utils/test_cooldown_handlers.py index 0d39016ea79..24459a5a250 100644 --- a/tests/unit/router_utils/test_cooldown_handlers.py +++ b/tests/unit/router_utils/test_cooldown_handlers.py @@ -25,6 +25,7 @@ from litellm.router_utils.cooldown_callbacks import router_cooldown_event_callba from litellm.router_utils.cooldown_handlers import ( _get_deployment_cooldown_policy, _has_explicit_allowed_fails_policy_for_exception, + _has_fallback_group, _increment_allowed_fails, _is_cooldown_required, _resolve_allowed_fails_from_policy, @@ -3149,6 +3150,8 @@ BENCHED: Final = (litellm.Timeout, litellm.Timeout, RouterRateLimitError, Router {}, BENCHED, ), + ({"allowed_fails": 1, "fallbacks": [{"test-model": "backup-model"}]}, {}, BENCHED), + ({"allowed_fails": 1, "default_fallbacks": "backup-model"}, {}, BENCHED), ], ids=[ "allowed_fails", @@ -3157,6 +3160,8 @@ BENCHED: Final = (litellm.Timeout, litellm.Timeout, RouterRateLimitError, Router "policy_names_timeout", "allowed_fails_with_a_fallback_group", "allowed_fails_with_a_dict_fallback_target", + "allowed_fails_with_a_string_fallback_target", + "allowed_fails_with_a_string_default_fallbacks", ], ) @pytest.mark.asyncio @@ -3180,6 +3185,33 @@ async def test_router_allowed_fails_benches_a_lone_deployment_only_when_the_poli assert len(route.calls) == outcomes.count(litellm.Timeout) +@pytest.mark.parametrize( + ("fallbacks", "expected"), + [ + ([{"test-model": "backup-model"}], True), + ([{"*": "backup-model"}], True), + ([{"test-model": ["backup-model"]}], True), + ([{"test-model": 5}, {"*": "backup-model"}], True), + ([{"test-model": 5}], False), + ([{"other-model": ["backup-model"]}], False), + ], + ids=[ + "string_target", + "string_catch_all_from_default_fallbacks", + "list_target", + "unparseable_entry_does_not_hide_the_catch_all", + "unparseable_entry_alone", + "another_groups_chain", + ], +) +def test_has_fallback_group_reads_string_targets_and_skips_unparseable_entries( + fallbacks: Sequence[Mapping[str, object]], expected: bool +) -> None: + router: Final = Router(model_list=[TIMED_OUT_DEPLOYMENT], fallbacks=list(fallbacks)) + + assert _has_fallback_group(router, "test-model") is expected + + async def _group_call_outcome(router: Router) -> str: try: response: Final = await router.acompletion( From e157c5e520acbd83ed75bb26ec3633c4b2cc09b1 Mon Sep 17 00:00:00 2001 From: Kerry Lu Date: Wed, 28 Oct 2026 15:34:15 -0700 Subject: [PATCH 12/41] fix(migration_data_checks): size btree entries by the bytes Postgres stores (LIT-10244) (#48911) * fix(migration_data_checks): size btree entries by the bytes Postgres stores (LIT-10244) * test(migration_data_checks): flag exactly the compressed keys Postgres rejects at the entry limit (LIT-10244) --------- Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- .../migration_data_checks.py | 11 +++- .../migrations/test_request_log_indexes.py | 64 +++++++++++++++++++ .../test_migration_data_checks.py | 15 +++++ 3 files changed, 87 insertions(+), 3 deletions(-) diff --git a/litellm-proxy-extras/litellm_proxy_extras/migration_data_checks.py b/litellm-proxy-extras/litellm_proxy_extras/migration_data_checks.py index 7b01e64dd73..e22639daaef 100644 --- a/litellm-proxy-extras/litellm_proxy_extras/migration_data_checks.py +++ b/litellm-proxy-extras/litellm_proxy_extras/migration_data_checks.py @@ -343,11 +343,16 @@ def btree_max_entry_bytes(block_size: int) -> int: def _value_bytes(column: ColumnType, identifier: "sql.Identifier") -> "sql.Composed": + """The bytes a value takes in the index entry: its length, or its stored size when + Postgres compressed it, since `index_form_tuple` keeps the compressed bytes.""" from psycopg import sql - if column.type_name in _OCTET_LENGTH_TYPES: - return sql.SQL("octet_length({})").format(identifier) - return sql.SQL("octet_length({}::text)").format(identifier) + length: Final = ( + sql.SQL("octet_length({})").format(identifier) + if column.type_name in _OCTET_LENGTH_TYPES + else sql.SQL("octet_length({}::text)").format(identifier) + ) + return sql.SQL("least({}, pg_column_size({}))").format(length, identifier) def _aligned(offset: "sql.Composable", align: str) -> "sql.Composed": diff --git a/tests/integration/database/migrations/test_request_log_indexes.py b/tests/integration/database/migrations/test_request_log_indexes.py index 1de2090b487..f558c355c84 100644 --- a/tests/integration/database/migrations/test_request_log_indexes.py +++ b/tests/integration/database/migrations/test_request_log_indexes.py @@ -1,3 +1,4 @@ +import hashlib import os import queue import secrets @@ -973,6 +974,7 @@ def test_the_migration_job_refuses_before_the_first_migration_when_spend_log_key _insert_spend_log_with_key(conn, "legacy-1", _legacy_bearer_token()) _insert_spend_log_with_key(conn, "legacy-2", _legacy_bearer_token()) _insert_spend_log_with_key(conn, "hashed-1", "a" * 64) + _insert_spend_log_with_key(conn, "repetitive-1", "z" * 3000) ledger_before: Final = _ledger(scratch_database) with pytest.raises(RuntimeError) as refused: @@ -1109,6 +1111,68 @@ def test_the_entry_size_check_draws_the_line_exactly_where_postgres_does( assert finding.detail == f"with an index entry over 2704 bytes, largest 2712 bytes ({described_as})" +@requires_db +def test_a_value_postgres_compresses_under_the_limit_is_not_counted_while_random_text_of_that_length_is( + scratch_database: str, +) -> None: + """Postgres compresses a long value before it checks the entry size, so 3000 repeated + bytes index fine while 3000 random hex bytes do not. The check must draw the same line.""" + create_index: Final = sql.SQL('CREATE INDEX "compress_probe_idx" ON "compress_probe" ("k")') + with psycopg.connect(scratch_database, autocommit=True) as conn: + conn.execute('CREATE TABLE "compress_probe" ("k" TEXT)') + conn.execute('INSERT INTO "compress_probe" ("k") VALUES (%s), (%s)', ("a" * 3000, secrets.token_hex(1500))) + findings: Final = migration_data_checks.find_blockers( + conn, "public", migration_data_checks.checks_for_statement(create_index.as_string(conn), "probe") + ) + with pytest.raises(psycopg.errors.ProgramLimitExceeded, match="index row size 3016 exceeds"): + conn.execute(create_index) + conn.execute('DELETE FROM "compress_probe" WHERE "k" <> %s', ("a" * 3000,)) + conn.execute(create_index) + stored: Final = conn.execute('SELECT octet_length("k"), pg_column_size("k") FROM "compress_probe"').fetchone() + + assert stored is not None and stored[0] == 3000 and stored[1] < 2704, stored + (finding,) = findings + assert isinstance(finding, migration_data_checks.BlockedChange) + assert finding.rows == 1, "only the random key is too long for the index once Postgres compresses the other" + assert finding.detail == "with an index entry over 2704 bytes, largest 3016 bytes (k up to 3000 bytes)" + + +def _postgres_rejects_an_index_on(conn: "psycopg.Connection[tuple[object, ...]]", key: str) -> bool: + conn.execute('DROP TABLE IF EXISTS "one_key"') + conn.execute('CREATE TABLE "one_key" ("k" TEXT)') + conn.execute('INSERT INTO "one_key" ("k") VALUES (%s)', (key,)) + try: + conn.execute('CREATE INDEX "one_key_idx" ON "one_key" ("k")') + except psycopg.errors.ProgramLimitExceeded: + return True + return False + + +@requires_db +def test_the_check_flags_exactly_the_compressed_keys_postgres_rejects_at_the_entry_limit(scratch_database: str) -> None: + """Keys that compress to a stored size just under and just over the entry limit go through the + check and through Postgres one by one. The check must flag a key exactly when Postgres rejects it.""" + noise: Final = "".join(hashlib.sha256(str(i).encode()).hexdigest() for i in range(40)) + keys: Final = tuple(noise[:n] + "a" * 1500 for n in range(2400, 2500)) + create_index: Final = sql.SQL('CREATE INDEX "stored_probe_idx" ON "stored_probe" ("k")') + with psycopg.connect(scratch_database, autocommit=True) as conn: + conn.execute('CREATE TABLE "stored_probe" ("k" TEXT)') + conn.execute('INSERT INTO "stored_probe" ("k") SELECT unnest(%s::text[])', (list(keys),)) + compressed: Final = conn.execute( + 'SELECT count(*) FROM "stored_probe" WHERE pg_column_size("k") < octet_length("k") AND octet_length("k") > 2704' + ).fetchone() + findings: Final = migration_data_checks.find_blockers( + conn, "public", migration_data_checks.checks_for_statement(create_index.as_string(conn), "probe") + ) + rejected: Final = tuple(key for key in keys if _postgres_rejects_an_index_on(conn, key)) + + assert compressed == (len(keys),), "every probe key must be longer than the limit yet stored compressed" + assert 0 < len(rejected) < len(keys), f"the probe must straddle the limit, Postgres rejected {len(rejected)}" + (finding,) = findings + assert isinstance(finding, migration_data_checks.BlockedChange) + assert finding.rows == len(rejected), f"the check must flag the {len(rejected)} keys Postgres rejects and no other" + + @requires_db def test_a_scan_that_cannot_finish_in_time_is_reported_as_unverified_rather_than_passed(scratch_database: str) -> None: with psycopg.connect(scratch_database, autocommit=True) as conn: diff --git a/tests/unit/litellm_proxy_extras/test_migration_data_checks.py b/tests/unit/litellm_proxy_extras/test_migration_data_checks.py index f3b787beda1..d4490927e29 100644 --- a/tests/unit/litellm_proxy_extras/test_migration_data_checks.py +++ b/tests/unit/litellm_proxy_extras/test_migration_data_checks.py @@ -7,7 +7,9 @@ from litellm_proxy_extras.migration_data_checks import ( MIGRATION_DATA_CHECK_TIMEOUT_ENV_VAR, BlockedChange, Check, + ColumnType, CouldNotVerify, + _count_query, btree_max_entry_bytes, checks_for_index, checks_for_migration, @@ -113,6 +115,19 @@ def test_the_btree_entry_limit_on_the_default_page_is_the_number_in_postgres_err assert btree_max_entry_bytes(8192) == 2704 +def test_the_entry_size_query_sizes_a_text_column_by_its_stored_bytes_when_postgres_compressed_it() -> None: + """Postgres compresses a long value before it checks the entry size, so a repetitive + 3000-byte key indexes fine. The query has to size it as stored, never by its length alone.""" + (check,) = checks_for_index( + "LiteLLM_SpendLogs", "LiteLLM_SpendLogs_api_key_startTime_idx", ("api_key", "startTime") + ) + types: Final = (ColumnType("text", -1, "i", -1), ColumnType("timestamp", 8, "d", 3)) + query: Final = _count_query(check, "public", types, 2704).as_string(None) + stored: Final = 'least(octet_length("api_key"), pg_column_size("api_key"))' + assert stored in query + assert 'octet_length("api_key")' not in query.replace(stored, ""), query + + def test_the_report_names_the_rows_the_statement_and_the_fix_and_says_nothing_was_applied() -> None: (check,) = checks_for_index( "LiteLLM_SpendLogs", "LiteLLM_SpendLogs_api_key_startTime_idx", ("api_key", "startTime") From 0b57b9976a609ab74ba5c5cf21f75079527527ff Mon Sep 17 00:00:00 2001 From: mrinal-berri Date: Wed, 28 Oct 2026 15:35:08 -0700 Subject: [PATCH 13/41] feat(daily-activity): per-guardrail spend breakdown in user, team and organization daily activity (#48702) * feat(daily-activity): read guardrail spend per key from the daily guardrail usage table Adds DailyActivityRepository.guardrail_spend, which sums LiteLLM_DailyGuardrailUsageUnits.cost by date, guardrail and key. A guardrail row counts only when its key has a spend row on that day under the same scope filters as the rest of the daily activity response, so user, key and date filters apply to guardrail cost too * feat(daily-activity): add a per-guardrail spend breakdown to /user/daily/activity/aggregated Each day now has breakdown.guardrails[] with metrics.spend and an api_key_breakdown carrying each key's spend and metadata. Guardrail entries only report spend, since that is the one metric stored per key. The breakdown is on the user endpoint only, is empty when a model filter is set because guardrail cost is not stored per model, and lists only keys within api_key_limit while the guardrail total keeps every key. spend and budgets are unchanged Resolves LIT-9785 * feat(daily-activity): add the guardrail breakdown to team and organization activity, with guardrail names /team/daily/activity/aggregated and /organization/daily/activity/aggregated now return breakdown.guardrails the same way the user endpoint does, since each key belongs to one team and one organization. Each guardrail entry also carries metadata.guardrail_name, read from LiteLLM_GuardrailsTable for guardrails created through the UI or API and falling back to the breakdown key for config.yaml guardrails, whose key is already their name * fix(daily-activity): keep daily activity up when the guardrail breakdown query fails, and treat an empty model filter as none A database error from the guardrail spend query now logs a warning and returns the response without guardrails, where it used to fail the whole aggregated response with a 500. An empty model= no longer drops the guardrail breakdown, matching how the rest of the endpoint already ignores it. The daily activity golden files and fixtures gain the empty guardrails key every breakdown now carries * refactor(daily-activity): fold guardrail breakdown in a worker thread and drop docstrings Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * feat(daily-activity): report untracked guardrail units next to spend Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> --------- Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- litellm/proxy/_lazy_openapi_snapshot.json | 75 ++++++++ .../common_daily_activity.py | 68 ++++++- .../daily_activity_routes.py | 1 + .../daily_activity_scopes.py | 4 + .../repositories/daily_activity_repository.py | 6 + litellm/repositories/daily_activity_sql.py | 19 ++ .../common_daily_activity.py | 29 +++ litellm/types/repositories/daily_activity.py | 10 + .../spend/fixtures/daily_activity_team.json | 2 +- .../spend/fixtures/daily_activity_user.json | 2 +- .../daily_activity_team_aggregated.json | 2 +- .../golden/daily_activity_team_paginated.json | 2 +- .../daily_activity_user_aggregated.json | 2 +- .../golden/daily_activity_user_paginated.json | 2 +- ...st_daily_activity_aggregated_breakdowns.py | 180 ++++++++++++++++++ .../test_daily_activity_routes.py | 1 + ui/litellm-dashboard/src/lib/http/schema.d.ts | 36 ++++ 17 files changed, 432 insertions(+), 9 deletions(-) diff --git a/litellm/proxy/_lazy_openapi_snapshot.json b/litellm/proxy/_lazy_openapi_snapshot.json index b10b5ae98b5..39adb99a0d0 100644 --- a/litellm/proxy/_lazy_openapi_snapshot.json +++ b/litellm/proxy/_lazy_openapi_snapshot.json @@ -3487,6 +3487,13 @@ "title": "Entities", "type": "object" }, + "guardrails": { + "additionalProperties": { + "$ref": "#/components/schemas/GuardrailMetric" + }, + "title": "Guardrails", + "type": "object" + }, "mcp_servers": { "additionalProperties": { "$ref": "#/components/schemas/MetricWithMetadata" @@ -3802,6 +3809,74 @@ "title": "ExportType", "type": "string" }, + "GuardrailKeyMetric": { + "properties": { + "metadata": { + "$ref": "#/components/schemas/KeyMetadata" + }, + "metrics": { + "$ref": "#/components/schemas/GuardrailSpendMetrics" + } + }, + "required": [ + "metrics", + "metadata" + ], + "title": "GuardrailKeyMetric", + "type": "object" + }, + "GuardrailMetadata": { + "properties": { + "guardrail_name": { + "title": "Guardrail Name", + "type": "string" + } + }, + "required": [ + "guardrail_name" + ], + "title": "GuardrailMetadata", + "type": "object" + }, + "GuardrailMetric": { + "properties": { + "api_key_breakdown": { + "additionalProperties": { + "$ref": "#/components/schemas/GuardrailKeyMetric" + }, + "title": "Api Key Breakdown", + "type": "object" + }, + "metadata": { + "$ref": "#/components/schemas/GuardrailMetadata" + }, + "metrics": { + "$ref": "#/components/schemas/GuardrailSpendMetrics" + } + }, + "required": [ + "metrics", + "metadata" + ], + "title": "GuardrailMetric", + "type": "object" + }, + "GuardrailSpendMetrics": { + "properties": { + "spend": { + "default": 0.0, + "title": "Spend", + "type": "number" + }, + "untracked_units": { + "default": 0, + "title": "Untracked Units", + "type": "integer" + } + }, + "title": "GuardrailSpendMetrics", + "type": "object" + }, "HTTPAuthSecurityScheme": { "description": "Defines a security scheme using HTTP authentication.", "properties": { diff --git a/litellm/proxy/management_endpoints/common_daily_activity.py b/litellm/proxy/management_endpoints/common_daily_activity.py index f948a49441a..66f386b792d 100644 --- a/litellm/proxy/management_endpoints/common_daily_activity.py +++ b/litellm/proxy/management_endpoints/common_daily_activity.py @@ -3,6 +3,7 @@ from collections.abc import Awaitable, Callable, Mapping, Sequence from collections.abc import Set as AbstractSet from dataclasses import dataclass, replace from datetime import date, datetime, timedelta +from itertools import groupby from types import MappingProxyType from typing import Final, Literal, NoReturn, Protocol @@ -28,6 +29,10 @@ from litellm.types.proxy.management_endpoints.common_daily_activity import ( DailySpendData, DailySpendMetadata, GroupedData, + GuardrailKeyMetric, + GuardrailMetadata, + GuardrailMetric, + GuardrailSpendMetrics, KeyMetadata, KeyMetricWithMetadata, MetricWithMetadata, @@ -39,6 +44,7 @@ from litellm.types.repositories.daily_activity import ( DailyActivityTable, EntityRollupRow, GroupingSetsRow, + GuardrailSpendRow, KeyMetadataRow, RollupMetricsRow, SpendLogsWindow, @@ -1013,17 +1019,68 @@ def _fold_entity_rollups_sync( ) +def _guardrail_metric(day: DailySpendData, rows: Sequence[GuardrailSpendRow]) -> GuardrailMetric: + return GuardrailMetric( + metrics=GuardrailSpendMetrics( + spend=sum(row.spend for row in rows), + untracked_units=sum(row.untracked_units for row in rows), + ), + metadata=GuardrailMetadata(guardrail_name=rows[0].guardrail_name), + api_key_breakdown={ + row.api_key: GuardrailKeyMetric( + metrics=GuardrailSpendMetrics(spend=row.spend, untracked_units=row.untracked_units), + metadata=day.breakdown.api_keys[row.api_key].metadata, + ) + for row in rows + if row.api_key in day.breakdown.api_keys + }, + ) + + +def _day_with_guardrails(day: DailySpendData, rows: Sequence[GuardrailSpendRow]) -> DailySpendData: + by_guardrail: Final = groupby(sorted(rows, key=lambda row: row.guardrail_id), key=lambda row: row.guardrail_id) + guardrails: Final = {guardrail_id: _guardrail_metric(day, tuple(group)) for guardrail_id, group in by_guardrail} + return day.model_copy(update={"breakdown": day.breakdown.model_copy(update={"guardrails": guardrails})}) + + +def _with_guardrail_breakdown( + results: Sequence[DailySpendData], rows: Sequence[GuardrailSpendRow] +) -> tuple[DailySpendData, ...]: + rows_by_date: Final = MappingProxyType( + {day: tuple(group) for day, group in groupby(sorted(rows, key=lambda row: row.date), key=lambda row: row.date)} + ) + return tuple(_day_with_guardrails(day, rows_by_date.get(day.date.strftime("%Y-%m-%d"), ())) for day in results) + + +async def _guardrail_spend_or_empty( + repository: DailyActivityRepository, scope: DailyActivityScope +) -> tuple[GuardrailSpendRow, ...]: + from prisma.errors import PrismaError + + try: + return await repository.guardrail_spend(scope) + except PrismaError as exc: + verbose_proxy_logger.warning("Returning daily activity without its guardrail breakdown: %s", exc) + return () + + async def get_daily_activity_aggregated( repository: DailyActivityRepository, scope: DailyActivityScope, *, entity_metadata_field: Mapping[str, dict[str, object]] | None = None, include_entity_breakdown: bool = False, + include_guardrail_breakdown: bool = False, api_key_limit: int = constants.USAGE_TOP_API_KEYS_DEFAULT, ) -> SpendAnalyticsPaginatedResponse: try: - aggregated_rows: Final = await repository.aggregated( - scope, include_entity_breakdown=include_entity_breakdown, api_key_limit=api_key_limit + aggregated_rows, guardrail_rows = await asyncio.gather( + repository.aggregated( + scope, include_entity_breakdown=include_entity_breakdown, api_key_limit=api_key_limit + ), + _guardrail_spend_or_empty(repository, scope) + if include_guardrail_breakdown and not scope.model + else asyncio.sleep(0, result=()), ) records: Final = aggregated_rows.grouping_rows aggregated: Final = await _aggregate_grouping_sets_records( @@ -1058,8 +1115,13 @@ async def get_daily_activity_aggregated( api_key_metadata=entity_key_metadata, entity_metadata_field=entity_metadata_field, ) + results: Final = ( + list(await asyncio.to_thread(_with_guardrail_breakdown, aggregated["results"], guardrail_rows)) + if guardrail_rows + else aggregated["results"] + ) return SpendAnalyticsPaginatedResponse( - results=aggregated["results"], + results=results, metadata=DailySpendMetadata( total_spend=aggregated["totals"].spend, total_flat_cost=aggregated["totals"].flat_cost, diff --git a/litellm/proxy/management_endpoints/daily_activity_routes.py b/litellm/proxy/management_endpoints/daily_activity_routes.py index 842873129e4..79e515944ce 100644 --- a/litellm/proxy/management_endpoints/daily_activity_routes.py +++ b/litellm/proxy/management_endpoints/daily_activity_routes.py @@ -300,6 +300,7 @@ def _register_aggregated_route(router: APIRouter, resolver: EntityScopeResolver, resolved.scope, entity_metadata_field=resolved.entity_metadata, include_entity_breakdown=resolver.include_entity_breakdown, + include_guardrail_breakdown=resolver.include_guardrail_breakdown, api_key_limit=api_key_limit, ) if resolver.entity != "team": diff --git a/litellm/proxy/management_endpoints/daily_activity_scopes.py b/litellm/proxy/management_endpoints/daily_activity_scopes.py index bc373388a4e..77246f5dda6 100644 --- a/litellm/proxy/management_endpoints/daily_activity_scopes.py +++ b/litellm/proxy/management_endpoints/daily_activity_scopes.py @@ -56,6 +56,7 @@ class EntityScopeResolver: resolve: EntityScopeResolve include_entity_breakdown: bool operation_names: OperationNames + include_guardrail_breakdown: bool = False def _query_ids(value: str | None) -> tuple[str, ...] | None: @@ -364,6 +365,7 @@ USER_RESOLVER = EntityScopeResolver( query=_user_query, resolve=_resolve_user, include_entity_breakdown=False, + include_guardrail_breakdown=True, operation_names=MappingProxyType( { "aggregated": "get_user_daily_activity_aggregated", @@ -385,6 +387,7 @@ TEAM_RESOLVER = EntityScopeResolver( query=_team_query, resolve=_resolve_team, include_entity_breakdown=True, + include_guardrail_breakdown=True, operation_names=MappingProxyType( { "aggregated": "get_team_daily_activity_aggregated", @@ -423,6 +426,7 @@ ORGANIZATION_RESOLVER = EntityScopeResolver( query=_organization_query, resolve=_resolve_organization, include_entity_breakdown=True, + include_guardrail_breakdown=True, operation_names=MappingProxyType( { "aggregated": "get_organization_daily_activity_aggregated", diff --git a/litellm/repositories/daily_activity_repository.py b/litellm/repositories/daily_activity_repository.py index 10c603fe56f..4ca23e4e89e 100644 --- a/litellm/repositories/daily_activity_repository.py +++ b/litellm/repositories/daily_activity_repository.py @@ -19,6 +19,7 @@ from litellm.repositories.daily_activity_sql import ( build_cache_leakage_keys_sql, build_entity_rollup_sql, build_export_sql, + build_guardrail_spend_sql, build_key_page_sql, build_key_search_sql, build_model_top_keys_sql, @@ -36,6 +37,7 @@ from litellm.types.repositories.daily_activity import ( ExportRow, ExportType, GroupingSetsRow, + GuardrailSpendRow, KeyMetadataRow, KeyPage, KeySpendRow, @@ -100,6 +102,7 @@ class DailyActivityDatabase(Protocol): _GROUPING_ADAPTER: Final = TypeAdapter(tuple[GroupingSetsRow, ...]) _ENTITY_ADAPTER: Final = TypeAdapter(tuple[EntityRollupRow, ...]) +_GUARDRAIL_SPEND_ADAPTER: Final = TypeAdapter(tuple[GuardrailSpendRow, ...]) _KEY_SPEND_ADAPTER: Final = TypeAdapter(tuple[KeySpendRow, ...]) _KEY_PAGE_TOTAL_ADAPTER: Final[TypeAdapter[int]] = TypeAdapter(int) _USER_SPEND_ADAPTER: Final = TypeAdapter(tuple[UserSpendRow, ...]) @@ -191,6 +194,9 @@ class DailyActivityRepository: distinct_api_keys=distinct_api_keys, ) + async def guardrail_spend(self, scope: DailyActivityScope) -> tuple[GuardrailSpendRow, ...]: + return _GUARDRAIL_SPEND_ADAPTER.validate_python(await self._query(build_guardrail_spend_sql(scope))) + async def search_keys(self, scope: DailyActivityScope, *, search: str, limit: int) -> tuple[str, ...]: if not 1 <= limit <= constants.USAGE_KEY_SEARCH_MAX: raise ValueError(f"limit must be between 1 and {constants.USAGE_KEY_SEARCH_MAX}") diff --git a/litellm/repositories/daily_activity_sql.py b/litellm/repositories/daily_activity_sql.py index 3d27e7d937e..7ec9fa67a67 100644 --- a/litellm/repositories/daily_activity_sql.py +++ b/litellm/repositories/daily_activity_sql.py @@ -261,6 +261,25 @@ def build_entity_rollup_sql(scope: DailyActivityScope, *, api_key_limit: int) -> return SqlQuery(sql=sql, params=(*where_params, PTU_SENTINEL_API_KEY, api_key_limit)) +def build_guardrail_spend_sql(scope: DailyActivityScope) -> SqlQuery: + pg_table: Final = PRISMA_TO_PG_TABLE[scope.table] + where_clause, where_params = build_where_clause(scope) + sql: Final = f""" + SELECT g.date, g.guardrail_id, COALESCE(t.guardrail_name, g.guardrail_id) AS guardrail_name, g.api_key, + COALESCE(SUM(g.cost), 0)::float AS spend, + COALESCE(SUM(CASE WHEN g.cost IS NULL THEN g.units ELSE g.untracked_units END), 0)::bigint AS untracked_units + FROM "LiteLLM_DailyGuardrailUsageUnits" g + LEFT JOIN "LiteLLM_GuardrailsTable" t ON t.guardrail_id = g.guardrail_id + WHERE g.date >= $1 AND g.date <= $2 + AND EXISTS ( + SELECT 1 FROM "{pg_table}" + WHERE {where_clause} AND "{pg_table}".date = g.date AND "{pg_table}".api_key = g.api_key + ) + GROUP BY g.date, g.guardrail_id, t.guardrail_name, g.api_key + """ + return SqlQuery(sql=sql, params=where_params) + + def _key_spend_select() -> str: return """ COALESCE(SUM(spend), 0)::float AS spend, diff --git a/litellm/types/proxy/management_endpoints/common_daily_activity.py b/litellm/types/proxy/management_endpoints/common_daily_activity.py index 213c911cfdc..3f1a7980b27 100644 --- a/litellm/types/proxy/management_endpoints/common_daily_activity.py +++ b/litellm/types/proxy/management_endpoints/common_daily_activity.py @@ -65,6 +65,34 @@ class MetricWithMetadata(MetricBase): api_key_breakdown: dict[str, KeyMetricWithMetadata] = Field(default_factory=dict) # api_key -> {metrics, metadata} +class GuardrailSpendMetrics(LiteLLMBaseModel): + model_config = ConfigDict(frozen=True) + + spend: float = 0.0 + untracked_units: int = 0 + + +class GuardrailKeyMetric(LiteLLMBaseModel): + model_config = ConfigDict(frozen=True) + + metrics: GuardrailSpendMetrics + metadata: KeyMetadata + + +class GuardrailMetadata(LiteLLMBaseModel): + model_config = ConfigDict(frozen=True) + + guardrail_name: str + + +class GuardrailMetric(LiteLLMBaseModel): + model_config = ConfigDict(frozen=True) + + metrics: GuardrailSpendMetrics + metadata: GuardrailMetadata + api_key_breakdown: Mapping[str, GuardrailKeyMetric] = Field(default_factory=dict) + + class BreakdownMetrics(LiteLLMBaseModel): """Breakdown of spend by different dimensions""" @@ -75,6 +103,7 @@ class BreakdownMetrics(LiteLLMBaseModel): endpoints: dict[str, MetricWithMetadata] = Field(default_factory=dict) # endpoint -> {metrics, metadata} api_keys: dict[str, KeyMetricWithMetadata] = Field(default_factory=dict) # api_key -> {metrics, metadata} entities: dict[str, MetricWithMetadata] = Field(default_factory=dict) # entity -> {metrics, metadata} + guardrails: Mapping[str, GuardrailMetric] = Field(default_factory=dict) # guardrail_id -> spend class DailySpendData(LiteLLMBaseModel): diff --git a/litellm/types/repositories/daily_activity.py b/litellm/types/repositories/daily_activity.py index 4111b83a891..b013ce43113 100644 --- a/litellm/types/repositories/daily_activity.py +++ b/litellm/types/repositories/daily_activity.py @@ -168,6 +168,16 @@ class EntityRollupRow(RollupMetricsRow): distinct_api_keys: int | None +@dataclass(frozen=True, slots=True) +class GuardrailSpendRow: + date: str + guardrail_id: str + guardrail_name: str + api_key: str + spend: float + untracked_units: int + + @dataclass(frozen=True, slots=True) class AggregatedRows: grouping_rows: tuple[GroupingSetsRow, ...] diff --git a/tests/integration/spend/fixtures/daily_activity_team.json b/tests/integration/spend/fixtures/daily_activity_team.json index 62377c07432..8f4956dd7fe 100644 --- a/tests/integration/spend/fixtures/daily_activity_team.json +++ b/tests/integration/spend/fixtures/daily_activity_team.json @@ -1 +1 @@ -{"results":[{"date":"2026-06-01","metrics":{"spend":1273.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1014,"completion_tokens":10,"cache_read_input_tokens":6,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1024,"successful_requests":5,"failed_requests":0,"api_requests":5,"total_response_time_ms":0,"timed_requests":0},"breakdown":{"mcp_servers":{},"models":{"model-ptu":{"metrics":{"spend":1000.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":0,"completion_tokens":0,"cache_read_input_tokens":0,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":0,"successful_requests":0,"failed_requests":0,"api_requests":0,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{}},"model-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-cache","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}}}},"model-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"deleted-target","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":false}}}},"model-popular":{"metrics":{"spend":270.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":9,"completion_tokens":6,"cache_read_input_tokens":3,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":15,"successful_requests":3,"failed_requests":0,"api_requests":3,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-a":{"metrics":{"spend":100.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":2,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":4,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-a","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-b":{"metrics":{"spend":90.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":3,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":5,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-b","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-c":{"metrics":{"spend":80.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":4,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":6,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-c","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}}}}},"model_groups":{"model-ptu":{"metrics":{"spend":1000.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":0,"completion_tokens":0,"cache_read_input_tokens":0,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":0,"successful_requests":0,"failed_requests":0,"api_requests":0,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{}},"model-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-cache","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}}}},"model-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"deleted-target","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":false}}}},"model-popular":{"metrics":{"spend":270.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":9,"completion_tokens":6,"cache_read_input_tokens":3,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":15,"successful_requests":3,"failed_requests":0,"api_requests":3,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-c":{"metrics":{"spend":80.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":4,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":6,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-c","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-a":{"metrics":{"spend":100.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":2,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":4,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-a","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-b":{"metrics":{"spend":90.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":3,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":5,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-b","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}}}}},"providers":{"provider-a":{"metrics":{"spend":1273.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1014,"completion_tokens":10,"cache_read_input_tokens":6,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1024,"successful_requests":5,"failed_requests":0,"api_requests":5,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"deleted-target","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":false}},"key-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-cache","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-b":{"metrics":{"spend":90.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":3,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":5,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-b","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-c":{"metrics":{"spend":80.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":4,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":6,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-c","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-a":{"metrics":{"spend":100.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":2,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":4,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-a","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}}}}},"endpoints":{"/v1/chat/completions":{"metrics":{"spend":1273.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1014,"completion_tokens":10,"cache_read_input_tokens":6,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1024,"successful_requests":5,"failed_requests":0,"api_requests":5,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-c":{"metrics":{"spend":80.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":4,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":6,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-c","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"deleted-target","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":false}},"key-b":{"metrics":{"spend":90.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":3,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":5,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-b","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-a":{"metrics":{"spend":100.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":2,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":4,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-a","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-cache","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}}}}},"api_keys":{"key-a":{"metrics":{"spend":100.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":2,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":4,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-a","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-b":{"metrics":{"spend":90.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":3,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":5,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-b","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-c":{"metrics":{"spend":80.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":4,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":6,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-c","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-cache","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"deleted-target","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":false}}},"entities":{"team-1":{"metrics":{"spend":1273.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1014,"completion_tokens":10,"cache_read_input_tokens":6,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1024,"successful_requests":5,"failed_requests":0,"api_requests":5,"total_response_time_ms":0,"timed_requests":0},"metadata":{"team_alias":"Usage Team"},"api_key_breakdown":{"key-a":{"metrics":{"spend":100.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":2,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":4,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-a","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-b":{"metrics":{"spend":90.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":3,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":5,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-b","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-c":{"metrics":{"spend":80.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":4,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":6,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-c","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-cache","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"deleted-target","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":false}}}}}}}],"metadata":{"total_spend":1273.0,"total_flat_cost":0.0,"total_ptu_hours":0.0,"total_prompt_tokens":1014,"total_completion_tokens":10,"total_tokens":1024,"total_api_requests":5,"total_successful_requests":5,"total_failed_requests":0,"total_cache_read_input_tokens":6,"total_cache_creation_input_tokens":0,"total_compression_saved_tokens":0,"total_compression_savings_spend":0.0,"total_prompt_caching_savings_spend":0.0,"total_gateway_injected_caching_savings_spend":0.0,"total_autorouter_savings_spend":0.0,"total_response_time_ms":0,"total_timed_requests":0,"page":1,"total_pages":1,"has_more":false,"api_key_limit":100,"total_api_keys":5,"entity_total_api_keys":{"team-1":5}}} +{"results":[{"date":"2026-06-01","metrics":{"spend":1273.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1014,"completion_tokens":10,"cache_read_input_tokens":6,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1024,"successful_requests":5,"failed_requests":0,"api_requests":5,"total_response_time_ms":0,"timed_requests":0},"breakdown":{"mcp_servers":{},"models":{"model-ptu":{"metrics":{"spend":1000.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":0,"completion_tokens":0,"cache_read_input_tokens":0,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":0,"successful_requests":0,"failed_requests":0,"api_requests":0,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{}},"model-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-cache","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}}}},"model-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"deleted-target","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":false}}}},"model-popular":{"metrics":{"spend":270.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":9,"completion_tokens":6,"cache_read_input_tokens":3,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":15,"successful_requests":3,"failed_requests":0,"api_requests":3,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-a":{"metrics":{"spend":100.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":2,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":4,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-a","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-b":{"metrics":{"spend":90.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":3,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":5,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-b","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-c":{"metrics":{"spend":80.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":4,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":6,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-c","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}}}}},"model_groups":{"model-ptu":{"metrics":{"spend":1000.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":0,"completion_tokens":0,"cache_read_input_tokens":0,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":0,"successful_requests":0,"failed_requests":0,"api_requests":0,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{}},"model-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-cache","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}}}},"model-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"deleted-target","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":false}}}},"model-popular":{"metrics":{"spend":270.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":9,"completion_tokens":6,"cache_read_input_tokens":3,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":15,"successful_requests":3,"failed_requests":0,"api_requests":3,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-c":{"metrics":{"spend":80.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":4,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":6,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-c","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-a":{"metrics":{"spend":100.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":2,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":4,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-a","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-b":{"metrics":{"spend":90.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":3,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":5,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-b","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}}}}},"providers":{"provider-a":{"metrics":{"spend":1273.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1014,"completion_tokens":10,"cache_read_input_tokens":6,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1024,"successful_requests":5,"failed_requests":0,"api_requests":5,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"deleted-target","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":false}},"key-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-cache","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-b":{"metrics":{"spend":90.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":3,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":5,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-b","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-c":{"metrics":{"spend":80.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":4,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":6,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-c","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-a":{"metrics":{"spend":100.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":2,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":4,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-a","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}}}}},"endpoints":{"/v1/chat/completions":{"metrics":{"spend":1273.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1014,"completion_tokens":10,"cache_read_input_tokens":6,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1024,"successful_requests":5,"failed_requests":0,"api_requests":5,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-c":{"metrics":{"spend":80.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":4,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":6,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-c","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"deleted-target","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":false}},"key-b":{"metrics":{"spend":90.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":3,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":5,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-b","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-a":{"metrics":{"spend":100.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":2,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":4,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-a","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-cache","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}}}}},"api_keys":{"key-a":{"metrics":{"spend":100.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":2,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":4,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-a","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-b":{"metrics":{"spend":90.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":3,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":5,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-b","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-c":{"metrics":{"spend":80.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":4,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":6,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-c","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-cache","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"deleted-target","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":false}}},"entities":{"team-1":{"metrics":{"spend":1273.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1014,"completion_tokens":10,"cache_read_input_tokens":6,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1024,"successful_requests":5,"failed_requests":0,"api_requests":5,"total_response_time_ms":0,"timed_requests":0},"metadata":{"team_alias":"Usage Team"},"api_key_breakdown":{"key-a":{"metrics":{"spend":100.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":2,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":4,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-a","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-b":{"metrics":{"spend":90.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":3,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":5,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-b","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-c":{"metrics":{"spend":80.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":4,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":6,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-c","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-cache","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"deleted-target","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":false}}}}},"guardrails":{}}}],"metadata":{"total_spend":1273.0,"total_flat_cost":0.0,"total_ptu_hours":0.0,"total_prompt_tokens":1014,"total_completion_tokens":10,"total_tokens":1024,"total_api_requests":5,"total_successful_requests":5,"total_failed_requests":0,"total_cache_read_input_tokens":6,"total_cache_creation_input_tokens":0,"total_compression_saved_tokens":0,"total_compression_savings_spend":0.0,"total_prompt_caching_savings_spend":0.0,"total_gateway_injected_caching_savings_spend":0.0,"total_autorouter_savings_spend":0.0,"total_response_time_ms":0,"total_timed_requests":0,"page":1,"total_pages":1,"has_more":false,"api_key_limit":100,"total_api_keys":5,"entity_total_api_keys":{"team-1":5}}} diff --git a/tests/integration/spend/fixtures/daily_activity_user.json b/tests/integration/spend/fixtures/daily_activity_user.json index c4302ee7045..ae05e661071 100644 --- a/tests/integration/spend/fixtures/daily_activity_user.json +++ b/tests/integration/spend/fixtures/daily_activity_user.json @@ -1 +1 @@ -{"results":[{"date":"2026-06-01","metrics":{"spend":1273.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1014,"completion_tokens":10,"cache_read_input_tokens":6,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1024,"successful_requests":5,"failed_requests":0,"api_requests":5,"total_response_time_ms":0,"timed_requests":0},"breakdown":{"mcp_servers":{},"models":{"model-ptu":{"metrics":{"spend":1000.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":0,"completion_tokens":0,"cache_read_input_tokens":0,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":0,"successful_requests":0,"failed_requests":0,"api_requests":0,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{}},"model-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-cache","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}}}},"model-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"deleted-target","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":false}}}},"model-popular":{"metrics":{"spend":270.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":9,"completion_tokens":6,"cache_read_input_tokens":3,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":15,"successful_requests":3,"failed_requests":0,"api_requests":3,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-a":{"metrics":{"spend":100.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":2,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":4,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-a","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-b":{"metrics":{"spend":90.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":3,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":5,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-b","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-c":{"metrics":{"spend":80.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":4,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":6,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-c","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}}}}},"model_groups":{"model-ptu":{"metrics":{"spend":1000.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":0,"completion_tokens":0,"cache_read_input_tokens":0,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":0,"successful_requests":0,"failed_requests":0,"api_requests":0,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{}},"model-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-cache","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}}}},"model-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"deleted-target","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":false}}}},"model-popular":{"metrics":{"spend":270.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":9,"completion_tokens":6,"cache_read_input_tokens":3,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":15,"successful_requests":3,"failed_requests":0,"api_requests":3,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-c":{"metrics":{"spend":80.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":4,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":6,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-c","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-a":{"metrics":{"spend":100.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":2,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":4,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-a","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-b":{"metrics":{"spend":90.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":3,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":5,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-b","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}}}}},"providers":{"provider-a":{"metrics":{"spend":1273.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1014,"completion_tokens":10,"cache_read_input_tokens":6,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1024,"successful_requests":5,"failed_requests":0,"api_requests":5,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"deleted-target","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":false}},"key-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-cache","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-b":{"metrics":{"spend":90.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":3,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":5,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-b","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-c":{"metrics":{"spend":80.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":4,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":6,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-c","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-a":{"metrics":{"spend":100.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":2,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":4,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-a","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}}}}},"endpoints":{"/v1/chat/completions":{"metrics":{"spend":1273.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1014,"completion_tokens":10,"cache_read_input_tokens":6,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1024,"successful_requests":5,"failed_requests":0,"api_requests":5,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-c":{"metrics":{"spend":80.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":4,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":6,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-c","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"deleted-target","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":false}},"key-b":{"metrics":{"spend":90.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":3,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":5,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-b","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-a":{"metrics":{"spend":100.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":2,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":4,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-a","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-cache","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}}}}},"api_keys":{"key-a":{"metrics":{"spend":100.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":2,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":4,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-a","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-b":{"metrics":{"spend":90.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":3,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":5,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-b","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-c":{"metrics":{"spend":80.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":4,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":6,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-c","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-cache","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"deleted-target","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":false}}},"entities":{"user-1":{"metrics":{"spend":1273.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1014,"completion_tokens":10,"cache_read_input_tokens":6,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1024,"successful_requests":5,"failed_requests":0,"api_requests":5,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-a":{"metrics":{"spend":100.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":2,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":4,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-a","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-b":{"metrics":{"spend":90.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":3,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":5,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-b","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-c":{"metrics":{"spend":80.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":4,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":6,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-c","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-cache","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"deleted-target","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":false}}}}}}}],"metadata":{"total_spend":1273.0,"total_flat_cost":0.0,"total_ptu_hours":0.0,"total_prompt_tokens":1014,"total_completion_tokens":10,"total_tokens":1024,"total_api_requests":5,"total_successful_requests":5,"total_failed_requests":0,"total_cache_read_input_tokens":6,"total_cache_creation_input_tokens":0,"total_compression_saved_tokens":0,"total_compression_savings_spend":0.0,"total_prompt_caching_savings_spend":0.0,"total_gateway_injected_caching_savings_spend":0.0,"total_autorouter_savings_spend":0.0,"total_response_time_ms":0,"total_timed_requests":0,"page":1,"total_pages":1,"has_more":false,"api_key_limit":100,"total_api_keys":5,"entity_total_api_keys":{"user-1":5}}} +{"results":[{"date":"2026-06-01","metrics":{"spend":1273.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1014,"completion_tokens":10,"cache_read_input_tokens":6,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1024,"successful_requests":5,"failed_requests":0,"api_requests":5,"total_response_time_ms":0,"timed_requests":0},"breakdown":{"mcp_servers":{},"models":{"model-ptu":{"metrics":{"spend":1000.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":0,"completion_tokens":0,"cache_read_input_tokens":0,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":0,"successful_requests":0,"failed_requests":0,"api_requests":0,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{}},"model-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-cache","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}}}},"model-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"deleted-target","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":false}}}},"model-popular":{"metrics":{"spend":270.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":9,"completion_tokens":6,"cache_read_input_tokens":3,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":15,"successful_requests":3,"failed_requests":0,"api_requests":3,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-a":{"metrics":{"spend":100.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":2,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":4,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-a","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-b":{"metrics":{"spend":90.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":3,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":5,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-b","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-c":{"metrics":{"spend":80.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":4,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":6,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-c","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}}}}},"model_groups":{"model-ptu":{"metrics":{"spend":1000.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":0,"completion_tokens":0,"cache_read_input_tokens":0,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":0,"successful_requests":0,"failed_requests":0,"api_requests":0,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{}},"model-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-cache","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}}}},"model-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"deleted-target","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":false}}}},"model-popular":{"metrics":{"spend":270.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":9,"completion_tokens":6,"cache_read_input_tokens":3,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":15,"successful_requests":3,"failed_requests":0,"api_requests":3,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-c":{"metrics":{"spend":80.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":4,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":6,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-c","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-a":{"metrics":{"spend":100.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":2,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":4,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-a","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-b":{"metrics":{"spend":90.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":3,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":5,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-b","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}}}}},"providers":{"provider-a":{"metrics":{"spend":1273.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1014,"completion_tokens":10,"cache_read_input_tokens":6,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1024,"successful_requests":5,"failed_requests":0,"api_requests":5,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"deleted-target","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":false}},"key-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-cache","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-b":{"metrics":{"spend":90.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":3,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":5,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-b","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-c":{"metrics":{"spend":80.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":4,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":6,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-c","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-a":{"metrics":{"spend":100.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":2,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":4,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-a","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}}}}},"endpoints":{"/v1/chat/completions":{"metrics":{"spend":1273.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1014,"completion_tokens":10,"cache_read_input_tokens":6,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1024,"successful_requests":5,"failed_requests":0,"api_requests":5,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-c":{"metrics":{"spend":80.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":4,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":6,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-c","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"deleted-target","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":false}},"key-b":{"metrics":{"spend":90.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":3,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":5,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-b","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-a":{"metrics":{"spend":100.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":2,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":4,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-a","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-cache","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}}}}},"api_keys":{"key-a":{"metrics":{"spend":100.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":2,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":4,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-a","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-b":{"metrics":{"spend":90.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":3,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":5,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-b","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-c":{"metrics":{"spend":80.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":4,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":6,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-c","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-cache","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"deleted-target","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":false}}},"entities":{"user-1":{"metrics":{"spend":1273.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1014,"completion_tokens":10,"cache_read_input_tokens":6,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1024,"successful_requests":5,"failed_requests":0,"api_requests":5,"total_response_time_ms":0,"timed_requests":0},"metadata":{},"api_key_breakdown":{"key-a":{"metrics":{"spend":100.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":2,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":4,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-a","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-b":{"metrics":{"spend":90.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":3,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":5,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-b","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-c":{"metrics":{"spend":80.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":4,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":6,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-c","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-cache":{"metrics":{"spend":2.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":1000,"completion_tokens":2,"cache_read_input_tokens":1,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":1002,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"alias-cache","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":true}},"key-target":{"metrics":{"spend":1.0,"flat_cost":0.0,"ptu_hours":0.0,"prompt_tokens":5,"completion_tokens":2,"cache_read_input_tokens":2,"cache_creation_input_tokens":0,"compression_saved_tokens":0,"compression_savings_spend":0.0,"prompt_caching_savings_spend":0.0,"gateway_injected_caching_savings_spend":0.0,"autorouter_savings_spend":0.0,"total_tokens":7,"successful_requests":1,"failed_requests":0,"api_requests":1,"total_response_time_ms":0,"timed_requests":0},"metadata":{"key_alias":"deleted-target","team_id":"team-1","user_id":"user-1","user_email":"user@example.com","key_exists":false}}}}},"guardrails":{}}}],"metadata":{"total_spend":1273.0,"total_flat_cost":0.0,"total_ptu_hours":0.0,"total_prompt_tokens":1014,"total_completion_tokens":10,"total_tokens":1024,"total_api_requests":5,"total_successful_requests":5,"total_failed_requests":0,"total_cache_read_input_tokens":6,"total_cache_creation_input_tokens":0,"total_compression_saved_tokens":0,"total_compression_savings_spend":0.0,"total_prompt_caching_savings_spend":0.0,"total_gateway_injected_caching_savings_spend":0.0,"total_autorouter_savings_spend":0.0,"total_response_time_ms":0,"total_timed_requests":0,"page":1,"total_pages":1,"has_more":false,"api_key_limit":100,"total_api_keys":5,"entity_total_api_keys":{"user-1":5}}} diff --git a/tests/integration/spend/golden/daily_activity_team_aggregated.json b/tests/integration/spend/golden/daily_activity_team_aggregated.json index 2956adda17a..91a5d33391b 100644 --- a/tests/integration/spend/golden/daily_activity_team_aggregated.json +++ b/tests/integration/spend/golden/daily_activity_team_aggregated.json @@ -26,7 +26,7 @@ }, "results": [ { - "breakdown": { + "breakdown": {"guardrails": {}, "api_keys": { "key-a": { "metadata": { diff --git a/tests/integration/spend/golden/daily_activity_team_paginated.json b/tests/integration/spend/golden/daily_activity_team_paginated.json index ca7239f8fa9..4d91b0da209 100644 --- a/tests/integration/spend/golden/daily_activity_team_paginated.json +++ b/tests/integration/spend/golden/daily_activity_team_paginated.json @@ -26,7 +26,7 @@ }, "results": [ { - "breakdown": { + "breakdown": {"guardrails": {}, "api_keys": { "key-a": { "metadata": { diff --git a/tests/integration/spend/golden/daily_activity_user_aggregated.json b/tests/integration/spend/golden/daily_activity_user_aggregated.json index 261ac22b2cb..185d333965a 100644 --- a/tests/integration/spend/golden/daily_activity_user_aggregated.json +++ b/tests/integration/spend/golden/daily_activity_user_aggregated.json @@ -26,7 +26,7 @@ }, "results": [ { - "breakdown": { + "breakdown": {"guardrails": {}, "api_keys": { "key-a": { "metadata": { diff --git a/tests/integration/spend/golden/daily_activity_user_paginated.json b/tests/integration/spend/golden/daily_activity_user_paginated.json index f1fb7ba50a4..00389289ebe 100644 --- a/tests/integration/spend/golden/daily_activity_user_paginated.json +++ b/tests/integration/spend/golden/daily_activity_user_paginated.json @@ -26,7 +26,7 @@ }, "results": [ { - "breakdown": { + "breakdown": {"guardrails": {}, "api_keys": { "key-a": { "metadata": { diff --git a/tests/integration/spend/test_daily_activity_aggregated_breakdowns.py b/tests/integration/spend/test_daily_activity_aggregated_breakdowns.py index 33baf9d0e2e..1c315a6e694 100644 --- a/tests/integration/spend/test_daily_activity_aggregated_breakdowns.py +++ b/tests/integration/spend/test_daily_activity_aggregated_breakdowns.py @@ -243,3 +243,183 @@ async def test_get_daily_activity_aggregated_model_group_rollups_fall_back_to_mo assert object_value(object_value(models["gpt-5"])["metrics"])["spend"] == 10.0 finally: _clean(day) + + +def _seed_guardrail_units(day: str, rows: Sequence[tuple[str, str, str, float | None, int, int]]) -> None: + for guardrail_id, api_key, usage_unit, cost, units, untracked_units in rows: + write_rows( + 'INSERT INTO "LiteLLM_DailyGuardrailUsageUnits" (guardrail_id, date, team_id, api_key, usage_unit,' + " units, cost, untracked_units, updated_at) VALUES (%s, %s, '', %s, %s, %s, %s, %s, now())", + ( + guardrail_id, + day, + api_key, + usage_unit, + str(units), + None if cost is None else str(cost), + str(untracked_units), + ), + ) + + +def _guardrail_spend(breakdown: dict[str, JsonValue]) -> dict[str, tuple[float, dict[str, float]]]: + return { + guardrail_id: ( + float(object_value(object_value(entry)["metrics"])["spend"]), + { + api_key: float(object_value(object_value(key_entry)["metrics"])["spend"]) + for api_key, key_entry in object_value(object_value(entry)["api_key_breakdown"]).items() + }, + ) + for guardrail_id, entry in object_value(breakdown["guardrails"]).items() + } + + +def _day_breakdown(gateway: Gateway, day: str, **params: str) -> dict[str, JsonValue]: + results: Final = _RESULTS.validate_python(_activity(gateway, day, **params)["results"]) + assert len(results) == 1 + return object_value(object_value(results[0])["breakdown"]) + + +@pytest.mark.asyncio +async def test_get_daily_activity_aggregated_breaks_down_guardrail_spend(gateway: Gateway) -> None: + day: Final = _unique_day() + _seed( + day, + [ + (_row_id(), "user-a", day, "key-a", "gpt-5", "", "bedrock", None, "/v1/chat/completions", 10, 1.0, 1, 1), + (_row_id(), "user-b", day, "key-b", "gpt-5", "", "bedrock", None, "/v1/chat/completions", 10, 2.0, 1, 1), + ], + ) + _seed_guardrail_units( + day, + [ + ("pii-filter", "key-a", "contentPolicyUnits", 0.10, 5, 2), + ("pii-filter", "key-a", "sensitiveInformationPolicyUnits", 0.05, 1, 0), + ("pii-filter", "key-b", "contentPolicyUnits", 0.20, 1, 0), + ("pii-filter", "key-b", "legacyUnits", None, 7, 0), + ("topic-filter", "key-a", "topicPolicyUnits", 0.01, 1, 0), + ("topic-filter", "key-b", "topicPolicyUnits", None, 1, 0), + ("pii-filter", "orphan-key", "contentPolicyUnits", 9.0, 1, 0), + ], + ) + try: + everyone: Final = _day_breakdown(gateway, day) + assert _guardrail_spend(everyone) == { + "pii-filter": (pytest.approx(0.35), {"key-a": pytest.approx(0.15), "key-b": pytest.approx(0.20)}), + "topic-filter": (pytest.approx(0.01), {"key-a": pytest.approx(0.01), "key-b": 0.0}), + } + api_keys: Final = object_value(everyone["api_keys"]) + pii_filter: Final = object_value(object_value(everyone["guardrails"])["pii-filter"]) + assert object_value(pii_filter["metrics"]) == {"spend": pytest.approx(0.35), "untracked_units": 9} + pii_api_keys: Final = object_value(pii_filter["api_key_breakdown"]) + assert object_value(object_value(pii_api_keys["key-a"])["metrics"]) == { + "spend": pytest.approx(0.15), + "untracked_units": 2, + } + assert object_value(object_value(pii_api_keys["key-b"])["metrics"]) == { + "spend": pytest.approx(0.20), + "untracked_units": 7, + } + + assert object_value(everyone["guardrails"])["topic-filter"] == { + "metrics": {"spend": pytest.approx(0.01), "untracked_units": 1}, + "metadata": {"guardrail_name": "topic-filter"}, + "api_key_breakdown": { + "key-a": { + "metrics": {"spend": pytest.approx(0.01), "untracked_units": 0}, + "metadata": object_value(api_keys["key-a"])["metadata"], + }, + "key-b": { + "metrics": {"spend": 0.0, "untracked_units": 1}, + "metadata": object_value(api_keys["key-b"])["metadata"], + }, + }, + } + assert object_value(object_value(object_value(everyone["models"])["gpt-5"])["metrics"])["spend"] == 3.0 + + assert _guardrail_spend(_day_breakdown(gateway, day, user_id="user-a")) == { + "pii-filter": (pytest.approx(0.15), {"key-a": pytest.approx(0.15)}), + "topic-filter": (pytest.approx(0.01), {"key-a": pytest.approx(0.01)}), + } + assert _guardrail_spend(_day_breakdown(gateway, day, api_key_limit="1")) == { + "pii-filter": (pytest.approx(0.35), {"key-b": pytest.approx(0.20)}), + "topic-filter": (pytest.approx(0.01), {"key-b": 0.0}), + } + limited_pii_filter: Final = object_value( + object_value(_day_breakdown(gateway, day, api_key_limit="1")["guardrails"])["pii-filter"] + ) + assert object_value(limited_pii_filter["metrics"])["untracked_units"] == 9 + assert set(object_value(limited_pii_filter["api_key_breakdown"])) == {"key-b"} + assert _guardrail_spend(_day_breakdown(gateway, day, model="gpt-5")) == {} + assert _guardrail_spend(_day_breakdown(gateway, day, model="")) == _guardrail_spend(everyone) + finally: + _clean(day) + write_rows('DELETE FROM "LiteLLM_DailyGuardrailUsageUnits" WHERE date = %s', (day,)) + + +def _entity_breakdown(gateway: Gateway, url: str, day: str, **params: str) -> dict[str, JsonValue]: + response: Final = gateway.request("GET", url, params={"start_date": day, "end_date": day, **params}) + assert response.status_code == 200, response.text + results: Final = _RESULTS.validate_python(object_value(response.json())["results"]) + assert len(results) == 1 + return object_value(object_value(results[0])["breakdown"]) + + +@pytest.mark.asyncio +@pytest.mark.parametrize( + ("url", "table", "entity_column", "filter_param"), + [ + ("/team/daily/activity/aggregated", "LiteLLM_DailyTeamSpend", "team_id", "team_ids"), + ( + "/organization/daily/activity/aggregated", + "LiteLLM_DailyOrganizationSpend", + "organization_id", + "organization_ids", + ), + ], +) +async def test_team_and_organization_activity_break_down_guardrail_spend_by_name( + gateway: Gateway, url: str, table: str, entity_column: str, filter_param: str +) -> None: + day: Final = _unique_day() + db_guardrail_id: Final = str(uuid.uuid4()) + db_guardrail_name: Final = f"pii-filter-{db_guardrail_id[:8]}" + for entity_id, api_key, spend in (("entity-a", "key-a", 1.0), ("entity-b", "key-b", 2.0)): + write_rows( + f'INSERT INTO "{table}" (id, {entity_column}, date, api_key, model, spend, api_requests,' + " successful_requests, updated_at) VALUES (%s, %s, %s, %s, 'gpt-5', %s, 1, 1, now())", + (_row_id(), entity_id, day, api_key, str(spend)), + ) + write_rows( + 'INSERT INTO "LiteLLM_GuardrailsTable" (guardrail_id, guardrail_name, litellm_params, updated_at)' + " VALUES (%s, %s, '{}', now())", + (db_guardrail_id, db_guardrail_name), + ) + _seed_guardrail_units( + day, + [ + (db_guardrail_id, "key-a", "contentPolicyUnits", 0.10, 1, 0), + (db_guardrail_id, "key-b", "contentPolicyUnits", 0.20, 1, 0), + ("config-topic-filter", "key-a", "topicPolicyUnits", 0.05, 1, 0), + (db_guardrail_id, "orphan-key", "contentPolicyUnits", 9.0, 1, 0), + ], + ) + try: + everyone: Final = _entity_breakdown(gateway, url, day) + assert _guardrail_spend(everyone) == { + db_guardrail_id: (pytest.approx(0.30), {"key-a": pytest.approx(0.10), "key-b": pytest.approx(0.20)}), + "config-topic-filter": (pytest.approx(0.05), {"key-a": pytest.approx(0.05)}), + } + guardrails: Final = object_value(everyone["guardrails"]) + assert object_value(guardrails[db_guardrail_id])["metadata"] == {"guardrail_name": db_guardrail_name} + assert object_value(guardrails["config-topic-filter"])["metadata"] == {"guardrail_name": "config-topic-filter"} + + assert _guardrail_spend(_entity_breakdown(gateway, url, day, **{filter_param: "entity-a"})) == { + db_guardrail_id: (pytest.approx(0.10), {"key-a": pytest.approx(0.10)}), + "config-topic-filter": (pytest.approx(0.05), {"key-a": pytest.approx(0.05)}), + } + finally: + write_rows(f'DELETE FROM "{table}" WHERE date = %s', (day,)) + write_rows('DELETE FROM "LiteLLM_DailyGuardrailUsageUnits" WHERE date = %s', (day,)) + write_rows('DELETE FROM "LiteLLM_GuardrailsTable" WHERE guardrail_id = %s', (db_guardrail_id,)) diff --git a/tests/unit/proxy/management_endpoints/test_daily_activity_routes.py b/tests/unit/proxy/management_endpoints/test_daily_activity_routes.py index 8fbead8c01c..2a273034a97 100644 --- a/tests/unit/proxy/management_endpoints/test_daily_activity_routes.py +++ b/tests/unit/proxy/management_endpoints/test_daily_activity_routes.py @@ -248,6 +248,7 @@ class _FakeRepository: def __init__(self, rows: tuple[_Activity, ...]) -> None: self._rows: Final = rows self.aggregated = AsyncMock(side_effect=self._aggregated) + self.guardrail_spend = AsyncMock(return_value=()) self.key_page = AsyncMock(side_effect=self._key_page) self.key_page_call: tuple[DailyActivityScope, int, int] | None = None self.search_keys = AsyncMock(side_effect=self._search_keys) diff --git a/ui/litellm-dashboard/src/lib/http/schema.d.ts b/ui/litellm-dashboard/src/lib/http/schema.d.ts index 70406105193..72cc8cdc94e 100644 --- a/ui/litellm-dashboard/src/lib/http/schema.d.ts +++ b/ui/litellm-dashboard/src/lib/http/schema.d.ts @@ -27721,6 +27721,10 @@ export interface components { entities?: { [key: string]: components["schemas"]["MetricWithMetadata"]; }; + /** Guardrails */ + guardrails?: { + [key: string]: components["schemas"]["GuardrailMetric"]; + }; /** Mcp Servers */ mcp_servers?: { [key: string]: components["schemas"]["MetricWithMetadata"]; @@ -33236,6 +33240,38 @@ export interface components { /** Updated At */ updated_at?: string | null; }; + /** GuardrailKeyMetric */ + GuardrailKeyMetric: { + metadata: components["schemas"]["KeyMetadata"]; + metrics: components["schemas"]["GuardrailSpendMetrics"]; + }; + /** GuardrailMetadata */ + GuardrailMetadata: { + /** Guardrail Name */ + guardrail_name: string; + }; + /** GuardrailMetric */ + GuardrailMetric: { + /** Api Key Breakdown */ + api_key_breakdown?: { + [key: string]: components["schemas"]["GuardrailKeyMetric"]; + }; + metadata: components["schemas"]["GuardrailMetadata"]; + metrics: components["schemas"]["GuardrailSpendMetrics"]; + }; + /** GuardrailSpendMetrics */ + GuardrailSpendMetrics: { + /** + * Spend + * @default 0 + */ + spend: number; + /** + * Untracked Units + * @default 0 + */ + untracked_units: number; + }; /** GuardrailSubmissionItem */ GuardrailSubmissionItem: { /** Created At */ From cf1006a3687de717f247e846cefedf823f2d8203 Mon Sep 17 00:00:00 2001 From: mrinal-berri Date: Wed, 28 Oct 2026 15:35:33 -0700 Subject: [PATCH 14/41] feat(ui): search and multi-select tools in the MCP toolset editor (#48601) * feat(ui): add per-server MCP tool catalog query hook One query per server keyed by server_id, so the toolset editor can load tools for many servers in parallel and a failing server surfaces as an error on its own row instead of an empty list. (LIT-10012) * feat(ui): add toolset tool search and bulk selection helpers Pure functions for the toolset editor: filter a server's tools by a search query, select/deselect many tools without duplicates or dropping other selections, and compute tri-state selection. (LIT-10012) * feat(ui): search tools and bulk select in the toolset editor The toolset editor search now matches tool names and descriptions as well as server names. Typing fetches every server's tools in parallel through the per-server query hook, auto-expands matching servers and hides servers that loaded with no match. Tool rows become checkboxes, each server header gets a tri-state select-all over its visible tools, and a bar offers select/deselect all matching tools. Bulk actions only touch visible tools, so selections on other servers are kept on save. (LIT-10012) * feat(ui): clear all and flag unconfirmed tools in the toolset panel Adds a Clear all action to the Your Toolset panel and labels selected tools whose server is gone, unreachable, or no longer lists the tool. Those tools stay selected and are still saved, so an outage never silently drops part of a toolset. (LIT-10012) * test(ui): cover toolset editor tool search and bulk selection Integration tests that drive the real toolset modal against stubbed networking and assert the saved tools payload: cross-server search by tool name and description, select/deselect all matching, server select-all, single-row toggles, Clear all, and that tools on unreachable or deleted servers survive a save. (LIT-10012) * fix(ui): pointer cursor on the toolset server select-all checkbox Base UI renders the checkbox as a span, which the global button cursor rule does not cover, so the server header select-all showed the default arrow. Use a pointer, and not-allowed while it is disabled. (LIT-10012) * fix(ui): stop toolset editor background fan-out and keep tools on failed refetch The toolset modal stays mounted when closed, so a finished search kept every server's tool query enabled and refocusing the window refetched all of them once stale. Fetch only while the editor is open, require both the live and debounced search before fanning out, and disable refetch on window focus for these queries. A failed refetch now keeps the tools that already loaded instead of flagging the server unreachable, and tools without a name are dropped. Trim comments that restated the code. (LIT-10012) * fix(ui): clear toolset search on close and only keep non-empty tool lists on failed refetch Reopening the editor after a search rendered once with the old query, which fetched every stale server; clear the search whenever the editor closes. A failed refetch now keeps the previous list only when it has tools, so a server that loaded none and then fails reports the error. (LIT-10012) * fix(ui): reset toolset editor state on close and avoid mutating selection sets Closing the editor now also clears expanded servers, so reopening it cannot refetch them for one render. Build the selection and expanded-server sets without in-place mutation, drop a comment that restated the code, and use screen queries in the new tests to stay within the prefer-screen-queries lint budget. (LIT-10012) --- ui/litellm-dashboard/eslint-suppressions.json | 3 - .../mcpServers/useMCPServerTools.test.ts | 118 ++++ .../hooks/mcpServers/useMCPServerTools.ts | 70 +++ .../MCPToolsetsTab.integration.test.tsx | 219 +++++++- .../_components/MCPToolsetsTab.tsx | 503 ++++++++++++------ .../_components/toolsetSelection.test.ts | 90 ++++ .../_components/toolsetSelection.ts | 63 +++ 7 files changed, 890 insertions(+), 176 deletions(-) create mode 100644 ui/litellm-dashboard/src/app/(dashboard)/hooks/mcpServers/useMCPServerTools.test.ts create mode 100644 ui/litellm-dashboard/src/app/(dashboard)/hooks/mcpServers/useMCPServerTools.ts create mode 100644 ui/litellm-dashboard/src/app/(dashboard)/mcp-servers/_components/toolsetSelection.test.ts create mode 100644 ui/litellm-dashboard/src/app/(dashboard)/mcp-servers/_components/toolsetSelection.ts diff --git a/ui/litellm-dashboard/eslint-suppressions.json b/ui/litellm-dashboard/eslint-suppressions.json index e67c1105d86..9fce93782b2 100644 --- a/ui/litellm-dashboard/eslint-suppressions.json +++ b/ui/litellm-dashboard/eslint-suppressions.json @@ -498,9 +498,6 @@ } }, "src/app/(dashboard)/mcp-servers/_components/MCPToolsetsTab.tsx": { - "no-nested-ternary": { - "count": 1 - }, "react-hooks/set-state-in-effect": { "count": 1 } diff --git a/ui/litellm-dashboard/src/app/(dashboard)/hooks/mcpServers/useMCPServerTools.test.ts b/ui/litellm-dashboard/src/app/(dashboard)/hooks/mcpServers/useMCPServerTools.test.ts new file mode 100644 index 00000000000..3ef8e88b2c9 --- /dev/null +++ b/ui/litellm-dashboard/src/app/(dashboard)/hooks/mcpServers/useMCPServerTools.test.ts @@ -0,0 +1,118 @@ +/* @vitest-environment jsdom */ +import React from "react"; +import { act, renderHook, waitFor } from "@testing-library/react"; +import { describe, it, expect, vi, beforeEach } from "vitest"; +import { QueryClient, QueryClientProvider } from "@tanstack/react-query"; +import { mcpServerToolKeys, useMCPServerTools } from "./useMCPServerTools"; +import * as networking from "@/components/networking"; + +vi.mock("@/components/networking", () => ({ + listMCPTools: vi.fn(), +})); + +vi.mock("@/app/(dashboard)/hooks/useAuthorized", () => ({ + default: vi.fn(() => ({ accessToken: "test-token-123" })), +})); + +const wrapper = ({ children }: { children: React.ReactNode }) => { + const queryClient = new QueryClient({ defaultOptions: { queries: { retry: false, gcTime: 0 } } }); + return React.createElement(QueryClientProvider, { client: queryClient }, children); +}; + +const toolsFor = (serverId: string) => ({ + tools: [{ name: `${serverId}_tool`, description: `from ${serverId}` }], + error: null, +}); + +describe("useMCPServerTools", () => { + beforeEach(() => { + vi.clearAllMocks(); + }); + + it("fetches only the enabled servers and leaves the rest idle", async () => { + vi.mocked(networking.listMCPTools).mockImplementation(async (_token, serverId) => toolsFor(serverId)); + + const { result } = renderHook(() => useMCPServerTools(["a", "b"], new Set(["a"])), { wrapper }); + + await waitFor(() => expect(result.current.get("a")?.tools).toEqual([{ name: "a_tool", description: "from a" }])); + expect(result.current.get("b")).toEqual({ tools: [], status: "idle" }); + expect(networking.listMCPTools).toHaveBeenCalledTimes(1); + expect(networking.listMCPTools).toHaveBeenCalledWith("test-token-123", "a"); + }); + + it("isolates a failing server from the others", async () => { + vi.mocked(networking.listMCPTools).mockImplementation(async (_token, serverId) => + serverId === "bad" ? { tools: [], error: "http_500", message: "upstream down" } : toolsFor(serverId), + ); + + const { result } = renderHook(() => useMCPServerTools(["good", "bad"], new Set(["good", "bad"])), { wrapper }); + + await waitFor(() => expect(result.current.get("bad")).toEqual({ tools: [], status: "error" })); + expect(result.current.get("good")).toEqual({ + tools: [{ name: "good_tool", description: "from good" }], + status: "success", + }); + }); + + it("reports a server that is being fetched as loading", () => { + vi.mocked(networking.listMCPTools).mockReturnValue(new Promise(() => {})); + + const { result } = renderHook(() => useMCPServerTools(["a"], new Set(["a"])), { wrapper }); + + expect(result.current.get("a")).toEqual({ tools: [], status: "loading" }); + }); + + it("normalizes a bare array response, tools without descriptions, and drops tools without a name", async () => { + vi.mocked(networking.listMCPTools).mockResolvedValue([ + { tool_name: "legacy" }, + "plain", + { description: "nameless" }, + ]); + + const { result } = renderHook(() => useMCPServerTools(["a"], new Set(["a"])), { wrapper }); + + await waitFor(() => + expect(result.current.get("a")?.tools).toEqual([ + { name: "legacy", description: "" }, + { name: "plain", description: "" }, + ]), + ); + }); + + it("reports an error when a refetch fails for a server that had loaded no tools", async () => { + const queryClient = new QueryClient({ defaultOptions: { queries: { retry: false } } }); + const sharedWrapper = ({ children }: { children: React.ReactNode }) => + React.createElement(QueryClientProvider, { client: queryClient }, children); + vi.mocked(networking.listMCPTools) + .mockResolvedValueOnce({ tools: [], error: null }) + .mockResolvedValueOnce({ tools: [], error: "http_502", message: "upstream down" }); + + const { result } = renderHook(() => useMCPServerTools(["a"], new Set(["a"])), { wrapper: sharedWrapper }); + await waitFor(() => expect(result.current.get("a")?.status).toBe("success")); + + await act(() => queryClient.refetchQueries({ queryKey: mcpServerToolKeys.detail("a") })); + + await waitFor(() => expect(result.current.get("a")).toEqual({ tools: [], status: "error" })); + }); + + it("keeps showing the loaded tools when a later refetch fails", async () => { + const queryClient = new QueryClient({ defaultOptions: { queries: { retry: false } } }); + const sharedWrapper = ({ children }: { children: React.ReactNode }) => + React.createElement(QueryClientProvider, { client: queryClient }, children); + vi.mocked(networking.listMCPTools) + .mockResolvedValueOnce(toolsFor("a")) + .mockResolvedValueOnce({ tools: [], error: "http_502", message: "upstream down" }); + + const { result } = renderHook(() => useMCPServerTools(["a"], new Set(["a"])), { wrapper: sharedWrapper }); + await waitFor(() => expect(result.current.get("a")?.status).toBe("success")); + + await act(() => queryClient.refetchQueries({ queryKey: mcpServerToolKeys.detail("a") })); + + await waitFor(() => expect(queryClient.getQueryState(mcpServerToolKeys.detail("a"))?.status).toBe("error")); + expect(networking.listMCPTools).toHaveBeenCalledTimes(2); + expect(result.current.get("a")).toEqual({ + tools: [{ name: "a_tool", description: "from a" }], + status: "success", + }); + }); +}); diff --git a/ui/litellm-dashboard/src/app/(dashboard)/hooks/mcpServers/useMCPServerTools.ts b/ui/litellm-dashboard/src/app/(dashboard)/hooks/mcpServers/useMCPServerTools.ts new file mode 100644 index 00000000000..bc4a6696a6e --- /dev/null +++ b/ui/litellm-dashboard/src/app/(dashboard)/hooks/mcpServers/useMCPServerTools.ts @@ -0,0 +1,70 @@ +import { useQueries, type QueryObserverResult } from "@tanstack/react-query"; +import { createQueryKeys } from "../common/queryKeysFactory"; +import { listMCPTools } from "@/components/networking"; +import useAuthorized from "../useAuthorized"; + +export const mcpServerToolKeys = createQueryKeys("mcpServerTools"); + +export interface MCPServerToolEntry { + name: string; + description: string; +} + +export type MCPServerToolsStatus = "idle" | "loading" | "error" | "success"; + +export interface MCPServerToolsState { + tools: MCPServerToolEntry[]; + status: MCPServerToolsStatus; +} + +type RawTool = string | { name?: string; tool_name?: string; description?: string }; + +const toEntry = (tool: RawTool): MCPServerToolEntry => + typeof tool === "string" + ? { name: tool, description: "" } + : { name: tool.name ?? tool.tool_name ?? "", description: tool.description ?? "" }; + +// listMCPTools never throws: it reports failures as `{ tools: [], error }`. +// Throw here so a broken server surfaces as a query error instead of an empty tool list. +export const fetchServerTools = async (accessToken: string, serverId: string): Promise => { + const result = await listMCPTools(accessToken, serverId); + if (!Array.isArray(result) && result?.error) { + throw new Error(result.message || result.error); + } + const toolList: RawTool[] = Array.isArray(result) ? result : result?.tools ?? []; + return toolList.map(toEntry).filter((tool) => tool.name !== ""); +}; + +const toStatus = ({ + status, + fetchStatus, + data, +}: Pick, "status" | "fetchStatus" | "data">): MCPServerToolsStatus => { + if (status === "error" && data !== undefined && data.length > 0) return "success"; + if (status !== "pending") return status; + return fetchStatus === "idle" ? "idle" : "loading"; +}; + +export const useMCPServerTools = ( + serverIds: readonly string[], + enabledServerIds: ReadonlySet, +): ReadonlyMap => { + const { accessToken } = useAuthorized(); + return useQueries({ + queries: serverIds.map((serverId) => ({ + queryKey: mcpServerToolKeys.detail(serverId), + queryFn: () => fetchServerTools(accessToken!, serverId), + enabled: !!accessToken && enabledServerIds.has(serverId), + staleTime: 5 * 60 * 1000, + retry: false, + refetchOnWindowFocus: false, + })), + combine: (results) => + new Map( + results.map((result, idx): [string, MCPServerToolsState] => [ + serverIds[idx], + { tools: result.data ?? [], status: toStatus(result) }, + ]), + ), + }); +}; diff --git a/ui/litellm-dashboard/src/app/(dashboard)/mcp-servers/_components/MCPToolsetsTab.integration.test.tsx b/ui/litellm-dashboard/src/app/(dashboard)/mcp-servers/_components/MCPToolsetsTab.integration.test.tsx index 81462c6bf9b..ba9f2bfa839 100644 --- a/ui/litellm-dashboard/src/app/(dashboard)/mcp-servers/_components/MCPToolsetsTab.integration.test.tsx +++ b/ui/litellm-dashboard/src/app/(dashboard)/mcp-servers/_components/MCPToolsetsTab.integration.test.tsx @@ -1,5 +1,5 @@ import React from "react"; -import { fireEvent, render, screen, waitFor, within } from "@testing-library/react"; +import { act, fireEvent, render, screen, waitFor, within } from "@testing-library/react"; import userEvent, { PointerEventsCheckLevel } from "@testing-library/user-event"; import { describe, it, expect, vi, beforeEach } from "vitest"; import { QueryClient, QueryClientProvider } from "@tanstack/react-query"; @@ -7,6 +7,7 @@ import { MCPToolsetsTab } from "./MCPToolsetsTab"; import * as networking from "@/components/networking"; import { useMCPToolsets } from "@/app/(dashboard)/hooks/mcpServers/useMCPToolsets"; import { useMCPServers } from "@/app/(dashboard)/hooks/mcpServers/useMCPServers"; +import { mcpServerToolKeys } from "@/app/(dashboard)/hooks/mcpServers/useMCPServerTools"; import { MCPToolset } from "@/components/mcp_tools/types"; vi.mock("@/components/networking", () => ({ @@ -20,6 +21,7 @@ vi.mock("@/components/networking", () => ({ vi.mock("@/app/(dashboard)/hooks/mcpServers/useMCPToolsets", () => ({ useMCPToolsets: vi.fn() })); vi.mock("@/app/(dashboard)/hooks/mcpServers/useMCPServers", () => ({ useMCPServers: vi.fn() })); +vi.mock("@/app/(dashboard)/hooks/useAuthorized", () => ({ default: () => ({ accessToken: "sk-test" }) })); const setup = () => userEvent.setup({ pointerEventsCheck: PointerEventsCheckLevel.Never }); @@ -197,6 +199,221 @@ describe("MCPToolsetsTab create/edit toolset form", () => { }); }); +describe("MCPToolsetsTab tool search and bulk selection", () => { + const servers = [ + { server_id: "gh", alias: "github", server_name: "GitHub" }, + { server_id: "lin", alias: "linear", server_name: "Linear" }, + { server_id: "broken", alias: "flaky", server_name: "Flaky" }, + ]; + const catalogs: Record = { + gh: [ + { name: "create_issue", description: "Open a GitHub issue" }, + { name: "list_pulls", description: "List pull requests" }, + ], + lin: [ + { name: "create_ticket", description: "Create a Linear issue" }, + { name: "list_projects", description: "List projects" }, + ], + }; + + const renderWithServers = (toolsets: MCPToolset[] = []) => { + vi.mocked(useMCPToolsets).mockReturnValue({ + data: toolsets, + isLoading: false, + } as unknown as ReturnType); + vi.mocked(useMCPServers).mockReturnValue({ data: servers, isSuccess: true } as unknown as ReturnType< + typeof useMCPServers + >); + const queryClient = new QueryClient({ defaultOptions: { queries: { retry: false, gcTime: 0 } } }); + render( + + + , + ); + return queryClient; + }; + + const editToolset = (tools: MCPToolset["tools"]) => + ({ toolset_id: "ts-1", toolset_name: "triage", description: "", tools }) as MCPToolset; + + const expectedUpdate = (tools: MCPToolset["tools"]) => ({ + toolset_id: "ts-1", + toolset_name: "triage", + description: "", + tools, + }); + + beforeEach(() => { + vi.clearAllMocks(); + vi.mocked(networking.listMCPTools).mockImplementation(async (_token, serverId) => + serverId in catalogs + ? { tools: catalogs[serverId], error: null } + : { tools: [], error: "http_502", message: "upstream down" }, + ); + vi.mocked(networking.createMCPToolset).mockResolvedValue( + {} as Awaited>, + ); + vi.mocked(networking.updateMCPToolset).mockResolvedValue( + {} as Awaited>, + ); + }); + + it("searches tool names and descriptions across collapsed servers and selects every match", async () => { + const user = setup(); + renderWithServers(); + + await openCreate(user); + fireEvent.change(screen.getByPlaceholderText("e.g. github-linear-tools"), { target: { value: "issues" } }); + await user.type(screen.getByRole("textbox", { name: "Search tools or servers" }), "issue"); + + expect(await screen.findByRole("checkbox", { name: /^create_ticket/ })).toBeInTheDocument(); + expect(screen.getByRole("checkbox", { name: /^create_issue/ })).toBeInTheDocument(); + expect(screen.queryByRole("checkbox", { name: /^list_pulls/ })).not.toBeInTheDocument(); + expect(await screen.findByText("2 matching tools")).toBeInTheDocument(); + + await user.click(screen.getByRole("button", { name: "Select all matching" })); + await user.click(screen.getByRole("button", { name: "Create Toolset" })); + + await waitFor(() => { + expect(networking.createMCPToolset).toHaveBeenCalledWith("sk-test", { + toolset_name: "issues", + description: "", + tools: [ + { server_id: "gh", tool_name: "create_issue" }, + { server_id: "lin", tool_name: "create_ticket" }, + ], + }); + }); + }); + + it("deselecting all matches keeps selected tools the search hides", async () => { + const user = setup(); + renderWithServers([ + editToolset([ + { server_id: "gh", tool_name: "list_pulls" }, + { server_id: "gh", tool_name: "create_issue" }, + { server_id: "lin", tool_name: "create_ticket" }, + ]), + ]); + + await openEditFor(user); + await user.type(screen.getByRole("textbox", { name: "Search tools or servers" }), "create"); + expect(await screen.findByRole("checkbox", { name: /^create_ticket/ })).toBeChecked(); + + await user.click(screen.getByRole("button", { name: "Deselect all matching" })); + await user.click(screen.getByRole("button", { name: "Save Changes" })); + + await waitFor(() => { + expect(networking.updateMCPToolset).toHaveBeenCalledWith( + "sk-test", + expectedUpdate([{ server_id: "gh", tool_name: "list_pulls" }]), + ); + }); + }); + + it("selects all of one server's tools from its header and toggles single rows, fetching only that server", async () => { + const user = setup(); + renderWithServers(); + + await openCreate(user); + fireEvent.change(screen.getByPlaceholderText("e.g. github-linear-tools"), { target: { value: "gh-only" } }); + await user.click(screen.getByRole("button", { name: /^github/ })); + expect(await screen.findByRole("checkbox", { name: /^list_pulls/ })).toBeInTheDocument(); + + await user.click(screen.getByRole("checkbox", { name: "Select all tools on github" })); + expect(screen.getByRole("checkbox", { name: /^list_pulls/ })).toBeChecked(); + // Clicking anywhere on a tool row toggles it exactly once. + await user.click(screen.getByText("List pull requests")); + expect(screen.getByRole("checkbox", { name: /^list_pulls/ })).not.toBeChecked(); + await user.click(screen.getByRole("button", { name: "Create Toolset" })); + + await waitFor(() => { + expect(networking.createMCPToolset).toHaveBeenCalledWith("sk-test", { + toolset_name: "gh-only", + description: "", + tools: [{ server_id: "gh", tool_name: "create_issue" }], + }); + }); + expect(networking.listMCPTools).toHaveBeenCalledTimes(1); + }); + + it("keeps and flags tools whose server is unreachable or gone when saving", async () => { + const user = setup(); + const tools = [ + { server_id: "broken", tool_name: "flaky_tool" }, + { server_id: "deleted", tool_name: "old_tool" }, + { server_id: "gh", tool_name: "create_issue" }, + ]; + renderWithServers([editToolset(tools)]); + + await openEditFor(user); + await user.type(screen.getByRole("textbox", { name: "Search tools or servers" }), "create"); + + expect(await screen.findByText("Couldn't load tools for this server.")).toBeInTheDocument(); + expect(await screen.findByText("server unreachable")).toBeInTheDocument(); + expect(screen.getByText("server not found")).toBeInTheDocument(); + + await user.click(screen.getByRole("button", { name: "Save Changes" })); + + await waitFor(() => { + expect(networking.updateMCPToolset).toHaveBeenCalledWith("sk-test", expectedUpdate(tools)); + }); + }); + + it("stops fetching server tools once the editor is closed after a search", async () => { + const user = setup(); + const queryClient = renderWithServers(); + + await openCreate(user); + await user.type(screen.getByRole("textbox", { name: "Search tools or servers" }), "issue"); + expect(await screen.findByRole("checkbox", { name: /^create_ticket/ })).toBeInTheDocument(); + await user.click(screen.getByRole("button", { name: "Cancel" })); + await waitFor(() => expect(screen.queryByRole("dialog")).not.toBeInTheDocument()); + const callsWhileOpen = vi.mocked(networking.listMCPTools).mock.calls.length; + + await act(() => queryClient.invalidateQueries({ queryKey: mcpServerToolKeys.all })); + await new Promise((resolve) => setTimeout(resolve, 50)); + expect(networking.listMCPTools).toHaveBeenCalledTimes(callsWhileOpen); + + await openCreate(user); + expect(screen.getByRole("textbox", { name: "Search tools or servers" })).toHaveValue(""); + await new Promise((resolve) => setTimeout(resolve, 50)); + expect(networking.listMCPTools).toHaveBeenCalledTimes(callsWhileOpen); + }); + + it("does not refetch a previously expanded server when the editor is reopened", async () => { + const user = setup(); + const queryClient = renderWithServers(); + + await openCreate(user); + await user.click(screen.getByRole("button", { name: /^github/ })); + expect(await screen.findByRole("checkbox", { name: /^list_pulls/ })).toBeInTheDocument(); + await user.click(screen.getByRole("button", { name: "Cancel" })); + await waitFor(() => expect(screen.queryByRole("dialog")).not.toBeInTheDocument()); + await act(() => queryClient.invalidateQueries({ queryKey: mcpServerToolKeys.all })); + + await openCreate(user); + await new Promise((resolve) => setTimeout(resolve, 50)); + + expect(networking.listMCPTools).toHaveBeenCalledTimes(1); + expect(screen.queryByRole("checkbox", { name: /^list_pulls/ })).not.toBeInTheDocument(); + }); + + it("clears the whole toolset with Clear all", async () => { + const user = setup(); + renderWithServers([editToolset([{ server_id: "gh", tool_name: "create_issue" }])]); + + await openEditFor(user); + await user.click(screen.getByRole("button", { name: "Clear all" })); + expect(screen.getByText("No tools added yet")).toBeInTheDocument(); + await user.click(screen.getByRole("button", { name: "Save Changes" })); + + await waitFor(() => { + expect(networking.updateMCPToolset).toHaveBeenCalledWith("sk-test", expectedUpdate([])); + }); + }); +}); + describe("MCPToolsetsTab usage guide", () => { it("shows the toolset URL on the gateway host", () => { renderTab(); diff --git a/ui/litellm-dashboard/src/app/(dashboard)/mcp-servers/_components/MCPToolsetsTab.tsx b/ui/litellm-dashboard/src/app/(dashboard)/mcp-servers/_components/MCPToolsetsTab.tsx index ff0e8f7b6a9..dc1477b79d3 100644 --- a/ui/litellm-dashboard/src/app/(dashboard)/mcp-servers/_components/MCPToolsetsTab.tsx +++ b/ui/litellm-dashboard/src/app/(dashboard)/mcp-servers/_components/MCPToolsetsTab.tsx @@ -1,28 +1,39 @@ -import React, { useState, useCallback } from "react"; +import React, { useState } from "react"; import { z } from "zod"; import { SortingState } from "@tanstack/react-table"; import { Inbox, Plus, X } from "lucide-react"; import { useMCPToolsets } from "@/app/(dashboard)/hooks/mcpServers/useMCPToolsets"; import { useMCPServers } from "@/app/(dashboard)/hooks/mcpServers/useMCPServers"; import { useQueryClient } from "@tanstack/react-query"; -import { DataTable } from "@/components/shared/DataTable"; +import { useDebouncedValue } from "@tanstack/react-pacer/debouncer"; import { - createMCPToolset, - updateMCPToolset, - deleteMCPToolset, - listMCPTools, - getGatewayBaseUrl, -} from "@/components/networking"; -import { MCPToolset, MCPToolsetTool } from "@/components/mcp_tools/types"; + useMCPServerTools, + type MCPServerToolEntry, + type MCPServerToolsState, +} from "@/app/(dashboard)/hooks/mcpServers/useMCPServerTools"; +import { DataTable } from "@/components/shared/DataTable"; +import { createMCPToolset, updateMCPToolset, deleteMCPToolset, getGatewayBaseUrl } from "@/components/networking"; +import { MCPServer, MCPToolset, MCPToolsetTool } from "@/components/mcp_tools/types"; import { FieldGroup } from "@/components/ui/field"; import { FormField } from "@/components/shared/form/FormField"; import { Button } from "@/components/ui/button"; +import { Checkbox } from "@/components/ui/checkbox"; import { Input } from "@/components/ui/input"; import { InputGroup, InputGroupAddon, InputGroupButton, InputGroupInput } from "@/components/ui/input-group"; import { UiLoadingSpinner } from "@/components/ui/ui-loading-spinner"; import { useZodForm } from "@/lib/forms/useZodForm"; import { toast } from "@/lib/toast"; import { displayToolName, getMCPToolsetTableColumns } from "./MCPToolsetTableColumns"; +import { + deselectTools, + filterServerTools, + isToolSelected, + matchesServerName, + selectTools, + selectionState, + toggleTool, + type SelectionState, +} from "./toolsetSelection"; import { Dialog, DialogContent, DialogFooter, DialogHeader, DialogTitle } from "@/components/ui/dialog"; interface MCPToolsetsTabProps { @@ -37,110 +48,191 @@ const toolsetSchema = z.object({ type ToolsetFormValues = z.infer; -interface MCPToolListProps { +const SEARCH_DEBOUNCE_WAIT_MS = 300; + +const IDLE_SERVER_TOOLS: MCPServerToolsState = { tools: [], status: "idle" }; + +const NO_SERVERS: ReadonlySet = new Set(); + +// The modal stays mounted while closed, so fetch only while it is open; otherwise a finished +// search keeps every server's query live and refetches all of them in the background. +const serversToFetch = ( + open: boolean, + searching: boolean, + serverIds: readonly string[], + expandedServerIds: ReadonlySet, +): ReadonlySet => { + if (!open) return NO_SERVERS; + return searching ? new Set(serverIds) : expandedServerIds; +}; + +const checkboxStateProps = (state: SelectionState) => ({ + checked: state === "all", + indeterminate: state === "some", +}); + +interface ServerView { + server: MCPServer; + serverName: string; + state: MCPServerToolsState; + visibleTools: readonly MCPServerToolEntry[]; +} + +const isPending = (state: MCPServerToolsState) => state.status === "idle" || state.status === "loading"; + +// While searching, a server stays listed until its tools load and turn out not to match. +const isHiddenBySearch = ( + search: string, + serverNames: readonly string[], + state: MCPServerToolsState, + visibleTools: readonly MCPServerToolEntry[], +) => { + if (!search.trim() || state.status !== "success") return false; + return visibleTools.length === 0 && !matchesServerName(search, serverNames); +}; + +const buildServerViews = ( + servers: readonly MCPServer[], + toolsByServer: ReadonlyMap, + search: string, +): ServerView[] => + servers.flatMap((server) => { + const state = toolsByServer.get(server.server_id) ?? IDLE_SERVER_TOOLS; + const names = [server.alias ?? "", server.server_name ?? ""]; + const visibleTools = filterServerTools(search, names, state.tools); + if (isHiddenBySearch(search, names, state, visibleTools)) return []; + return [{ server, serverName: server.alias || server.server_name || server.server_id, state, visibleTools }]; + }); + +const summarizeSearch = (serverViews: readonly ServerView[]) => ({ + matchingTools: serverViews.flatMap(({ server, visibleTools }) => + visibleTools.map((tool) => ({ server_id: server.server_id, tool_name: tool.name })), + ), + pendingServerCount: serverViews.filter(({ state }) => isPending(state)).length, +}); + +// Why a selected tool cannot be confirmed against its server. The tool stays selected and is +// still saved; the note only keeps it from looking lost. +const selectedToolIssue = ( + tool: MCPToolsetTool, + serverNameById: ReadonlyMap, + toolsByServer: ReadonlyMap, +): string | undefined => { + if (!serverNameById.has(tool.server_id)) return "server not found"; + const state = toolsByServer.get(tool.server_id); + if (state?.status === "error") return "server unreachable"; + if (state?.status === "success" && !state.tools.some((t) => t.name === tool.tool_name)) return "tool not found"; + return undefined; +}; + +interface ServerToolGroupProps { serverId: string; serverName: string; - accessToken: string | null; + state: MCPServerToolsState; + visibleTools: readonly MCPServerToolEntry[]; + expanded: boolean; + searchActive: boolean; selectedTools: MCPToolsetTool[]; - onToggle: (tool: MCPToolsetTool) => void; + onToggleExpanded: () => void; + onToggleTool: (tool: MCPToolsetTool) => void; + onSetToolsSelected: (tools: MCPToolsetTool[], selected: boolean) => void; } -interface ToolEntry { - name: string; - description?: string; -} +function ServerToolGroup({ + serverId, + serverName, + state, + visibleTools, + expanded, + searchActive, + selectedTools, + onToggleExpanded, + onToggleTool, + onSetToolsSelected, +}: ServerToolGroupProps) { + const selectedCount = selectedTools.filter((t) => t.server_id === serverId).length; + const visibleRefs = visibleTools.map((tool) => ({ server_id: serverId, tool_name: tool.name })); + const groupState = selectionState(selectedTools, visibleRefs); -function MCPToolList({ serverId, serverName, accessToken, selectedTools, onToggle }: MCPToolListProps) { - const [tools, setTools] = useState([]); - const [loading, setLoading] = useState(false); - const [expanded, setExpanded] = useState(false); - - const selectedSet = new Set(selectedTools.filter((t) => t.server_id === serverId).map((t) => t.tool_name)); - - const fetchTools = useCallback(async () => { - if (!accessToken || tools.length > 0) return; - setLoading(true); - try { - const result = await listMCPTools(accessToken, serverId); - const toolList = Array.isArray(result) ? result : result?.tools ?? []; - setTools(toolList.map((t: any) => ({ name: t.name ?? t.tool_name ?? t, description: t.description ?? "" }))); - } catch { - setTools([]); - } finally { - setLoading(false); + const renderBody = () => { + if (isPending(state)) { + return ( +
+ +
+ ); } - }, [accessToken, serverId, tools.length]); - - const handleToggle = () => { - if (!expanded) fetchTools(); - setExpanded(!expanded); + if (state.status === "error") { + return

Couldn't load tools for this server.

; + } + if (visibleTools.length === 0) { + return ( +

+ {searchActive ? "No matching tools on this server." : "No tools found for this server."} +

+ ); + } + return ( +
+ {visibleRefs.map((ref, idx) => { + const tool = visibleTools[idx]; + const selected = isToolSelected(selectedTools, ref); + return ( + + ); + })} +
+ ); }; return (
- - {expanded && ( -
- {loading ? ( -
- -
- ) : tools.length === 0 ? ( -

No tools found for this server.

- ) : ( -
- {tools.map((tool) => { - const selected = selectedSet.has(tool.name); - return ( - - ); - })} -
- )} -
- )} +
+ onSetToolsSelected(visibleRefs, checked)} + /> + +
+ {expanded &&
{renderBody()}
}
); } @@ -149,21 +241,29 @@ interface CreateToolsetModalProps { open: boolean; onClose: () => void; onSave: (name: string, description: string | undefined, tools: MCPToolsetTool[]) => Promise; - accessToken: string | null; initialToolset?: MCPToolset; } -function CreateToolsetModal({ open, onClose, onSave, accessToken, initialToolset }: CreateToolsetModalProps) { - const form = useZodForm(toolsetSchema, { - defaultValues: { - toolset_name: initialToolset?.toolset_name || "", - description: initialToolset?.description || "", - }, - }); +const toolsetFormDefaults = (toolset?: MCPToolset): ToolsetFormValues => ({ + toolset_name: toolset?.toolset_name || "", + description: toolset?.description || "", +}); + +function CreateToolsetModal({ open, onClose, onSave, initialToolset }: CreateToolsetModalProps) { + const form = useZodForm(toolsetSchema, { defaultValues: toolsetFormDefaults(initialToolset) }); const [selectedTools, setSelectedTools] = useState(initialToolset?.tools || []); const [saving, setSaving] = useState(false); - const [serverSearch, setServerSearch] = useState(""); - const { data: mcpServers = [] } = useMCPServers(); + const [search, setSearch] = useState(""); + const [debouncedSearch] = useDebouncedValue(search, { wait: SEARCH_DEBOUNCE_WAIT_MS }); + const [expandedServerIds, setExpandedServerIds] = useState>(() => new Set()); + const { data: mcpServers = [], isSuccess: serversLoaded } = useMCPServers(); + const serverIds = React.useMemo(() => mcpServers.map((s) => s.server_id), [mcpServers]); + const searching = search.trim() !== "" && debouncedSearch.trim() !== ""; + const fetchServerIds = React.useMemo( + () => serversToFetch(open, searching, serverIds, expandedServerIds), + [open, searching, serverIds, expandedServerIds], + ); + const toolsByServer = useMCPServerTools(serverIds, fetchServerIds); const serverPrefixById = React.useMemo( () => new Map(mcpServers.map((s) => [s.server_id, s.alias || s.server_name || s.server_id])), [mcpServers], @@ -171,41 +271,54 @@ function CreateToolsetModal({ open, onClose, onSave, accessToken, initialToolset React.useEffect(() => { if (open) { - form.reset({ - toolset_name: initialToolset?.toolset_name || "", - description: initialToolset?.description || "", - }); + form.reset(toolsetFormDefaults(initialToolset)); setSelectedTools(initialToolset?.tools || []); - setServerSearch(""); + setSearch(""); + setExpandedServerIds(new Set()); } }, [open, initialToolset, form]); const handleToggleTool = (tool: MCPToolsetTool) => { - setSelectedTools((prev) => { - const exists = prev.some((t) => t.server_id === tool.server_id && t.tool_name === tool.tool_name); - return exists - ? prev.filter((t) => !(t.server_id === tool.server_id && t.tool_name === tool.tool_name)) - : [...prev, tool]; - }); + setSelectedTools((prev) => toggleTool(prev, tool)); + }; + + const handleSetToolsSelected = (tools: MCPToolsetTool[], selected: boolean) => { + setSelectedTools((prev) => (selected ? selectTools(prev, tools) : deselectTools(prev, tools))); + }; + + const handleToggleExpanded = (serverId: string) => { + setExpandedServerIds((prev) => + prev.has(serverId) ? new Set([...prev].filter((id) => id !== serverId)) : new Set([...prev, serverId]), + ); + }; + + // The create modal stays mounted, so state left from the last session would fetch servers + // on the first render after reopening, before the open effect resets it. + const handleClose = () => { + setSearch(""); + setExpandedServerIds(new Set()); + onClose(); }; const handleSubmit = async (values: ToolsetFormValues) => { setSaving(true); try { await onSave(values.toolset_name, values.description, selectedTools); - onClose(); + handleClose(); } finally { setSaving(false); } }; - const filteredServers = mcpServers.filter((s) => { - const q = serverSearch.toLowerCase(); - return !q || (s.alias || "").toLowerCase().includes(q) || (s.server_name || "").toLowerCase().includes(q); - }); + const searchActive = search.trim() !== ""; + const serverViews = buildServerViews(mcpServers, toolsByServer, search); + const { matchingTools, pendingServerCount } = searchActive + ? summarizeSearch(serverViews) + : { matchingTools: [], pendingServerCount: 0 }; + const matchingState = selectionState(selectedTools, matchingTools); return ( - !open && onClose()}> + !open && handleClose()}> {initialToolset ? "Edit Toolset" : "New Toolset"} @@ -229,32 +342,65 @@ function CreateToolsetModal({ open, onClose, onSave, accessToken, initialToolset setServerSearch(e.target.value)} + placeholder="Search tools or servers..." + aria-label="Search tools or servers" + value={search} + onChange={(e) => setSearch(e.target.value)} /> - {serverSearch && ( + {search && ( - setServerSearch("")}> + setSearch("")}> )} + {searchActive && ( +
+

+ {matchingTools.length} matching {matchingTools.length === 1 ? "tool" : "tools"} + {pendingServerCount > 0 && + ` · loading ${pendingServerCount} ${pendingServerCount === 1 ? "server" : "servers"}…`} +

+
+ + +
+
+ )}
- {filteredServers.length === 0 ? ( + {serverViews.length === 0 ? (

- {mcpServers.length === 0 ? "No MCP servers configured" : "No servers match your search"} + {mcpServers.length === 0 ? "No MCP servers configured" : "No tools or servers match your search"}

) : ( - filteredServers.map((server) => ( - ( + handleToggleExpanded(server.server_id)} + onToggleTool={handleToggleTool} + onSetToolsSelected={handleSetToolsSelected} /> )) )} @@ -266,41 +412,60 @@ function CreateToolsetModal({ open, onClose, onSave, accessToken, initialToolset {/* Right panel: Your Toolset */}
-

- Your Toolset{" "} - ({selectedTools.length} tools) -

+
+

+ Your Toolset{" "} + ({selectedTools.length} tools) +

+ +
{selectedTools.length === 0 ? (

No tools added yet

) : ( - selectedTools.map((tool, idx) => ( -
- - ✕ - - - )) + + ); + }) )}
-
diff --git a/ui/litellm-dashboard/src/components/shared/SearchSelect.integration.test.tsx b/ui/litellm-dashboard/src/components/shared/SearchSelect.integration.test.tsx index 9f320049b09..ecb8dd161f3 100644 --- a/ui/litellm-dashboard/src/components/shared/SearchSelect.integration.test.tsx +++ b/ui/litellm-dashboard/src/components/shared/SearchSelect.integration.test.tsx @@ -63,6 +63,24 @@ describe("SearchSelect", () => { expect(onValueChange).toHaveBeenLastCalledWith("team-3"); }); + it("offers the typed text as a value only when the caller accepts it", async () => { + const user = userEvent.setup(); + const onValueChange = vi.fn(); + const customValue = { hint: "Custom value", accepts: (typed: string) => typed.endsWith("*") }; + const { rerender } = render(); + const input = screen.getByRole("combobox"); + await user.click(input); + fireEvent.change(input, { target: { value: "team-*" } }); + expect(screen.queryByRole("option", { name: /team-\*/ })).not.toBeInTheDocument(); + + rerender(); + fireEvent.change(input, { target: { value: "team-4" } }); + expect(screen.queryByRole("option", { name: /team-4/ })).not.toBeInTheDocument(); + fireEvent.change(input, { target: { value: "team-*" } }); + await user.click(await screen.findByRole("option", { name: /team-\*/ })); + expect(onValueChange).toHaveBeenLastCalledWith("team-*"); + }); + it("filters the options client-side as you type", async () => { const user = userEvent.setup(); render(); diff --git a/ui/litellm-dashboard/src/components/shared/SearchSelect.tsx b/ui/litellm-dashboard/src/components/shared/SearchSelect.tsx index 3649114826d..e139720a2ed 100644 --- a/ui/litellm-dashboard/src/components/shared/SearchSelect.tsx +++ b/ui/litellm-dashboard/src/components/shared/SearchSelect.tsx @@ -1,6 +1,6 @@ "use client"; -import React from "react"; +import React, { useState } from "react"; import { Combobox, ComboboxContent, @@ -19,6 +19,11 @@ export interface SearchSelectOption { icon?: React.ReactNode; } +interface CustomValue { + hint: string; + accepts: (typed: string) => boolean; +} + interface SearchSelectProps { options: SearchSelectOption[]; value?: string | null; @@ -29,6 +34,8 @@ interface SearchSelectProps { className?: string; inputId?: string; allowClear?: boolean; + customValue?: CustomValue; + reservedValues?: readonly string[]; "aria-label"?: string; id?: string; } @@ -49,13 +56,22 @@ export function SearchSelect({ className, inputId, allowClear = true, + customValue, + reservedValues = [], "aria-label": ariaLabel, id, }: SearchSelectProps) { + const [query, setQuery] = useState(""); const selected = value == null || value === "" ? null : options.find((option) => option.value === value) ?? { label: value, value }; - const items = + const known = selected !== null && !options.some((option) => option.value === selected.value) ? [selected, ...options] : options; + const typed = query.trim(); + const isNewValue = typed !== "" && !reservedValues.includes(typed) && !known.some((option) => option.value === typed); + const items = + customValue !== undefined && isNewValue && customValue.accepts(typed) + ? [...known, { label: typed, value: typed, sublabel: customValue.hint }] + : known; return ( a.value === b.value} itemToStringLabel={(item: SearchSelectOption) => item.label} filter={matchesQuery} + onInputValueChange={setQuery} disabled={disabled} > x. - * - model_max_budget: Optional[dict] - Model-specific max budget for user. [Docs](https://docs.litellm.ai/docs/proxy/users#add-model-specific-budgets-to-keys) + * - model_max_budget: Optional[dict] - Model-specific max budget for user, keyed by model name, `*` pattern, or model access group. [Docs](https://docs.litellm.ai/docs/proxy/users#add-model-specific-budgets-to-keys) * - budget_fallbacks: Optional[Dict[str, List[str]]] - Per-model fallback chain tried in order when that model's own `model_max_budget` is exceeded, e.g. {"gpt-4o": ["gpt-4o-mini"]}. * - model_rpm_limit: Optional[float] - Model-specific rpm limit for user. [Docs](https://docs.litellm.ai/docs/proxy/users#add-model-specific-limits-to-keys) * - mcp_rpm_limit: Optional[dict] - Per-MCP-server rpm limit, keyed by MCP server name {"github": 100, "slack": 200}. Enforced for keys and teams only; values set on a user are stored but not enforced per user. @@ -18917,7 +18917,7 @@ export interface paths { * - permissions: Optional[dict] - [Not Implemented Yet] User-specific permissions, eg. turning off pii masking. * - metadata: Optional[dict] - Metadata for user, store information for user. Example metadata = {"team": "core-infra", "app": "app2", "email": "ishaan@berri.ai" } * - max_parallel_requests: Optional[int] - Rate limit a user based on the number of parallel requests. Raises 429 error, if user's parallel requests > x. - * - model_max_budget: Optional[dict] - Model-specific max budget for user. [Docs](https://docs.litellm.ai/docs/proxy/users#add-model-specific-budgets-to-keys) + * - model_max_budget: Optional[dict] - Model-specific max budget for user, keyed by model name, `*` pattern, or model access group. [Docs](https://docs.litellm.ai/docs/proxy/users#add-model-specific-budgets-to-keys) * - budget_fallbacks: Optional[Dict[str, List[str]]] - Per-model fallback chain tried in order when that model's own `model_max_budget` is exceeded, e.g. {"gpt-4o": ["gpt-4o-mini"]}. * - model_rpm_limit: Optional[float] - Model-specific rpm limit for user. [Docs](https://docs.litellm.ai/docs/proxy/users#add-model-specific-limits-to-keys) * - mcp_rpm_limit: Optional[dict] - Per-MCP-server rpm limit, keyed by MCP server name {"github": 100, "slack": 200}. Enforced for keys and teams only; values set on a user are stored but not enforced per user. From 41c4c3788e0e39c972f6bf4ed71451e61cbb24f8 Mon Sep 17 00:00:00 2001 From: tessellark <349031000+tessellark@users.noreply.github.com> Date: Wed, 28 Oct 2026 15:42:51 -0700 Subject: [PATCH 17/41] test(routing): audit #46601 (#48910) * test(routing): audit #46601 Adds integration cells for the falsy budget 400, the empty and none-only declarations under drop_params, Vertex AI Claude, Databricks Claude, and the Bedrock native output retry * test(routing): expect the Anthropic default max_tokens on Bedrock Invoke chat Reads the default from litellm instead of pinning 4096, which #48502 raised to 128000 * test(routing): assert whole sent bodies in the Vertex AI and Bedrock native output cells --------- Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com> --- .../_reasoning_effort_levels_support.py | 5 + .../test_reasoning_effort_levels_wire.py | 320 +++++++++++++++++- 2 files changed, 316 insertions(+), 9 deletions(-) diff --git a/tests/integration/routing/_reasoning_effort_levels_support.py b/tests/integration/routing/_reasoning_effort_levels_support.py index 8d43695e419..d295f6e6d56 100644 --- a/tests/integration/routing/_reasoning_effort_levels_support.py +++ b/tests/integration/routing/_reasoning_effort_levels_support.py @@ -32,6 +32,7 @@ GEMINI: Final = "gemini-3-pro-preview" ZAI_GLM: Final = "glm-5" ECHOED_KEY_MODEL: Final = f"{KIMI}-echoed-key" BEDROCK_CLAUDE: Final = f"us.anthropic.{CLAUDE}" +DATABRICKS_CLAUDE: Final = f"databricks-{CLAUDE}" AZURE_API_VERSION: Final = "2025-04-01-preview" SYNTHETIC_KEY: Final = "synthetic-provider-key" _MARKER: Final = re.compile(r"marker-([0-9a-f]{32})") @@ -215,6 +216,10 @@ def backend(kind: str, wire_url: str) -> Backend: f"bedrock/invoke/{BEDROCK_CLAUDE}", MappingProxyType({"api_base": wire_url, "api_key": SYNTHETIC_KEY, "aws_region_name": "us-east-1"}), ) + case "databricks_claude": + return Backend( + f"databricks/{DATABRICKS_CLAUDE}", MappingProxyType({"api_base": wire_url, "api_key": SYNTHETIC_KEY}) + ) case _: raise AssertionError(f"unknown backend kind {kind}") diff --git a/tests/integration/routing/test_reasoning_effort_levels_wire.py b/tests/integration/routing/test_reasoning_effort_levels_wire.py index 3d88aad547c..d15867ac572 100644 --- a/tests/integration/routing/test_reasoning_effort_levels_wire.py +++ b/tests/integration/routing/test_reasoning_effort_levels_wire.py @@ -1,4 +1,5 @@ import asyncio +import json from collections.abc import Callable, Mapping from dataclasses import dataclass from types import MappingProxyType @@ -8,9 +9,11 @@ import anthropic import httpx import openai import pytest +from integration._support.anthropic_sse import message_json from integration._support.client import Gateway, eventually, list_value, object_value, string_value from integration._support.openai_wire import openai_error -from integration._support.wire import Request, wire_server +from integration._support.vertex import service_account_json +from integration._support.wire import Reply, Request, wire_server from integration.routing._reasoning_effort_levels_support import ( AZURE_API_VERSION, CLAUDE, @@ -34,6 +37,7 @@ from integration.routing._reasoning_effort_levels_support import ( marker_of, messages_request, new_marker, + path_of, prompt, provider_reply, responses_request, @@ -41,6 +45,8 @@ from integration.routing._reasoning_effort_levels_support import ( ) from pydantic import JsonValue +from litellm.constants import DEFAULT_ANTHROPIC_CHAT_MAX_TOKENS + CHAT: Final = "/v1/chat/completions" MESSAGES: Final = "/v1/messages" RESPONSES: Final = "/v1/responses" @@ -49,6 +55,7 @@ _ADAPTIVE: Final[dict[str, JsonValue]] = {"type": "adaptive"} _STRING_BUDGET: Final[dict[str, JsonValue]] = {"type": "enabled", "budget_tokens": "2048"} _XHIGH_BUDGET: Final[dict[str, JsonValue]] = {"type": "enabled", "budget_tokens": 9000} _BOOL_BUDGET: Final[dict[str, JsonValue]] = {"type": "enabled", "budget_tokens": True} +_FALSE_BUDGET: Final[dict[str, JsonValue]] = {"type": "enabled", "budget_tokens": False} _NULL_BUDGET: Final[dict[str, JsonValue]] = {"type": "enabled", "budget_tokens": None} _LIST_BUDGET: Final[dict[str, JsonValue]] = {"type": "enabled", "budget_tokens": [2048]} _EMPTY_BUDGET: Final[dict[str, JsonValue]] = {"type": "enabled", "budget_tokens": ""} @@ -67,6 +74,8 @@ _KIMI_INFO: Final[Mapping[str, JsonValue]] = MappingProxyType({"reasoning_effort _CLAUDE_INFO: Final[Mapping[str, JsonValue]] = MappingProxyType({"reasoning_effort_levels": list(CLAUDE_LEVELS)}) _GPT_INFO: Final[Mapping[str, JsonValue]] = MappingProxyType({"reasoning_effort_levels": list(GPT_LEVELS)}) _UNDECLARED: Final[Mapping[str, JsonValue]] = MappingProxyType({}) +_EMPTY_DECLARATION: Final[Mapping[str, JsonValue]] = MappingProxyType({"reasoning_effort_levels": []}) +_NONE_ONLY_DECLARATION: Final[Mapping[str, JsonValue]] = MappingProxyType({"reasoning_effort_levels": ["none"]}) _ECHOED_MAP_KEY: Final[Mapping[str, JsonValue]] = MappingProxyType( {"key": "azure/chat-latest", "litellm_provider": "azure"} ) @@ -134,7 +143,7 @@ def _invoke_chat(effort: str, thinking: Mapping[str, JsonValue], betas: tuple[st "messages": [{"role": "user", "content": [{"type": "text", "text": prompt(marker)}]}], "thinking": dict(thinking), "output_config": {"effort": effort}, - "max_tokens": 4096, + "max_tokens": DEFAULT_ANTHROPIC_CHAT_MAX_TOKENS, "anthropic_version": "bedrock-2023-05-31", "anthropic_beta": list(betas), } @@ -144,6 +153,14 @@ def _with_output_config(expected: Expected, output_config: JsonValue) -> Expecte return lambda marker: {**expected(marker), "output_config": output_config} +def _with_fields(expected: Expected, fields: Mapping[str, JsonValue]) -> Expected: + return lambda marker: {**expected(marker), **fields} + + +def _without(expected: Expected, field: str) -> Expected: + return lambda marker: {key: value for key, value in expected(marker).items() if key != field} + + def _converse_chat(effort: str, thinking: Mapping[str, JsonValue]) -> Expected: return lambda marker: { "messages": [{"role": "user", "content": [{"text": prompt(marker)}]}], @@ -320,6 +337,7 @@ _OPENAI_RESPONSES: Final = "/v1/responses" _AZURE_RESPONSES: Final = f"/openai/responses?api-version={AZURE_API_VERSION}" _UNSUPPORTED: Final = "does not support parameters" _GEMINI: Final = f"/models/{GEMINI}:generateContent" +_DATABRICKS: Final = "/chat/completions" CELLS: Final[Mapping[str, Cell]] = MappingProxyType( { @@ -1750,6 +1768,139 @@ CELLS: Final[Mapping[str, Cell]] = MappingProxyType( _anthropic_chat_without_effort(_ADAPTIVE), _ANTHROPIC, ), + "chat-anthropic-false-budget-400": Cell( + CHAT, "anthropic", _CLAUDE_INFO, {"thinking": _FALSE_BUDGET}, None, _ANTHROPIC, 400, _NON_NUMERIC_BUDGET + ), + "chat-anthropic-undeclared-null-budget-400": Cell( + CHAT, "anthropic", _UNDECLARED, {"thinking": _NULL_BUDGET}, None, _ANTHROPIC, 400, _NON_NUMERIC_BUDGET + ), + "chat-anthropic-undeclared-false-budget-drop-params-sends-adaptive-without-effort": Cell( + CHAT, + "anthropic", + _UNDECLARED, + {"thinking": _FALSE_BUDGET, "drop_params": True}, + _anthropic_chat_without_effort(_ADAPTIVE), + _ANTHROPIC, + ), + "chat-anthropic-undeclared-missing-budget-sends-low": Cell( + CHAT, + "anthropic", + _UNDECLARED, + {"thinking": _MISSING_BUDGET}, + _anthropic_chat("low", _ADAPTIVE), + _ANTHROPIC, + ), + "messages-anthropic-undeclared-false-budget-400": Cell( + MESSAGES, "anthropic", _UNDECLARED, {"thinking": _FALSE_BUDGET}, None, _ANTHROPIC, 400, _NON_NUMERIC_BUDGET + ), + "messages-anthropic-empty-string-budget-400": Cell( + MESSAGES, "anthropic", _CLAUDE_INFO, {"thinking": _EMPTY_BUDGET}, None, _ANTHROPIC, 400, _NON_NUMERIC_BUDGET + ), + "messages-anthropic-undeclared-null-budget-drop-params-sends-adaptive-without-effort": Cell( + MESSAGES, + "anthropic", + _UNDECLARED, + {"thinking": _NULL_BUDGET, "drop_params": True}, + _native_messages(None, _ADAPTIVE), + _ANTHROPIC, + ), + "responses-anthropic-false-budget-400": Cell( + RESPONSES, + "anthropic", + _CLAUDE_INFO, + {"thinking": _FALSE_BUDGET}, + None, + _ANTHROPIC, + 400, + _NON_NUMERIC_BUDGET, + ), + "chat-converse-undeclared-false-budget-400": Cell( + CHAT, + "bedrock_converse", + _UNDECLARED, + {"thinking": _FALSE_BUDGET}, + None, + _CONVERSE, + 400, + _NON_NUMERIC_BUDGET, + ), + "messages-invoke-undeclared-null-budget-400": Cell( + MESSAGES, "bedrock_invoke", _UNDECLARED, {"thinking": _NULL_BUDGET}, None, _INVOKE, 400, _NON_NUMERIC_BUDGET + ), + "chat-databricks-claude-undeclared-false-budget-400": Cell( + CHAT, + "databricks_claude", + _UNDECLARED, + {"thinking": _FALSE_BUDGET}, + None, + _DATABRICKS, + 400, + _NON_NUMERIC_BUDGET, + ), + "chat-anthropic-empty-declaration-budget-drop-params-sends-adaptive-without-effort": Cell( + CHAT, + "anthropic", + _EMPTY_DECLARATION, + {"thinking": _BUDGET, "drop_params": True}, + _anthropic_chat_without_effort(_ADAPTIVE), + _ANTHROPIC, + ), + "chat-anthropic-none-only-declaration-budget-drop-params-sends-adaptive-without-effort": Cell( + CHAT, + "anthropic", + _NONE_ONLY_DECLARATION, + {"thinking": _BUDGET, "drop_params": True}, + _anthropic_chat_without_effort(_ADAPTIVE), + _ANTHROPIC, + ), + "chat-anthropic-empty-declaration-budget-keeps-its-effort": Cell( + CHAT, + "anthropic", + _EMPTY_DECLARATION, + {"thinking": _BUDGET}, + _anthropic_chat("medium", _ADAPTIVE), + _ANTHROPIC, + ), + "chat-converse-empty-declaration-budget-drop-params-sends-adaptive-without-effort": Cell( + CHAT, + "bedrock_converse", + _EMPTY_DECLARATION, + {"thinking": _BUDGET, "drop_params": True}, + _converse_chat_without_effort(_ADAPTIVE), + _CONVERSE, + ), + "messages-anthropic-empty-declaration-budget-drop-params-sends-adaptive-without-effort": Cell( + MESSAGES, + "anthropic", + _EMPTY_DECLARATION, + {"thinking": _BUDGET, "drop_params": True}, + _native_messages(None, _ADAPTIVE), + _ANTHROPIC, + ), + "messages-anthropic-none-only-declaration-budget-drop-params-sends-adaptive-without-effort": Cell( + MESSAGES, + "anthropic", + _NONE_ONLY_DECLARATION, + {"thinking": _BUDGET, "drop_params": True}, + _native_messages(None, _ADAPTIVE), + _ANTHROPIC, + ), + "messages-anthropic-empty-declaration-budget-keeps-its-effort": Cell( + MESSAGES, + "anthropic", + _EMPTY_DECLARATION, + {"thinking": _BUDGET}, + _native_messages({"effort": "medium"}, _ADAPTIVE), + _ANTHROPIC, + ), + "messages-converse-empty-declaration-budget-drop-params-sends-adaptive-without-effort": Cell( + MESSAGES, + "bedrock_converse", + _EMPTY_DECLARATION, + {"thinking": _BUDGET, "drop_params": True}, + _converse_messages(None), + _CONVERSE, + ), } ) @@ -1815,20 +1966,65 @@ def _bedrock_native_schema(body: Mapping[str, JsonValue], *, invoke: bool) -> Js return JSON_OBJECT.validate_json(string_value(json_schema["schema"])) +_CONVERSE_NATIVE_OUTPUT: Final[dict[str, JsonValue]] = { + "textFormat": { + "type": "json_schema", + "structure": {"jsonSchema": {"name": "answer", "schema": json.dumps(_NATIVE_OUTPUT_SCHEMA)}}, + } +} +_INVOKE_NATIVE_OUTPUT: Final[dict[str, JsonValue]] = { + "effort": "high", + "format": {"type": "json_schema", "schema": _NATIVE_OUTPUT_SCHEMA}, +} +_JSON_TOOL: Final = "json_tool_call" +_CONVERSE_TOOL_SCHEMA: Final[dict[str, JsonValue]] = { + key: value for key, value in _NATIVE_OUTPUT_SCHEMA.items() if key != "additionalProperties" +} +_CONVERSE_WITH_NATIVE_OUTPUT: Final = _with_fields( + _converse_chat("high", _ADAPTIVE), {"outputConfig": _CONVERSE_NATIVE_OUTPUT} +) +_INVOKE_WITH_NATIVE_OUTPUT: Final = _with_output_config( + _invoke_chat("high", _ADAPTIVE, ("structured-outputs-2025-11-13",)), _INVOKE_NATIVE_OUTPUT +) +_CONVERSE_WITH_JSON_TOOL: Final = _with_fields( + _converse_chat("high", _ADAPTIVE), + { + "toolConfig": { + "tools": [ + { + "toolSpec": { + "name": _JSON_TOOL, + "description": _JSON_TOOL, + "inputSchema": {"json": _CONVERSE_TOOL_SCHEMA}, + } + } + ] + } + }, +) +_INVOKE_WITH_JSON_TOOL: Final = _with_fields( + _without(_invoke_chat("high", _ADAPTIVE, ()), "anthropic_beta"), + {"tools": [{"name": _JSON_TOOL, "input_schema": _NATIVE_OUTPUT_SCHEMA}]}, +) + + @pytest.mark.parametrize( - ("kind", "target", "fields"), + ("kind", "target", "fields", "expected"), [ - pytest.param("bedrock_converse", _CONVERSE, {"thinking": _BUDGET}, id="converse-budget"), + pytest.param( + "bedrock_converse", _CONVERSE, {"thinking": _BUDGET}, _CONVERSE_WITH_NATIVE_OUTPUT, id="converse-budget" + ), pytest.param( "bedrock_invoke", _INVOKE, {"thinking": _BUDGET, "output_config": {"effort": "medium"}}, + _INVOKE_WITH_NATIVE_OUTPUT, id="invoke-output-config-beside-budget", ), ], ) def test_a_deployment_opted_into_bedrock_native_output_keeps_it_beside_its_declared_level( - gateway: Gateway, kind: str, target: str, fields: Mapping[str, JsonValue] + gateway: Gateway, kind: str, target: str, fields: Mapping[str, JsonValue], expected: Expected ) -> None: marker: Final = new_marker() response_format: Final[dict[str, JsonValue]] = { @@ -1844,10 +2040,116 @@ def test_a_deployment_opted_into_bedrock_native_output_keeps_it_beside_its_decla assert response.status_code == 200, response.text[:2000] assert answer(marker) in response.text, response.text[:2000] assert [(request.method, request.target) for request in received] == [("POST", target)] - body: Final = body_of(received[0]) - assert _bedrock_native_schema(body, invoke=target == _INVOKE) == _NATIVE_OUTPUT_SCHEMA, body - assert {"toolConfig", "tools"}.isdisjoint(body), body - assert sent_effort(received[0]) == "high", body + assert body_of(received[0]) == expected(marker) + + +_NATIVE_OUTPUT_REJECTION: Final = ( + "The model returned the following errors: output_config.format: Extra inputs are not permitted" +) + + +def _rejecting_native_output(request: Request) -> Reply: + native: Final = _bedrock_native_schema(body_of(request), invoke=path_of(request).endswith("/invoke")) + if native is None: + return provider_reply(request) + return Reply(status=400, body=json.dumps({"message": _NATIVE_OUTPUT_REJECTION}).encode()) + + +@pytest.mark.parametrize( + ("kind", "target", "rejected", "retried"), + [ + pytest.param( + "bedrock_converse", _CONVERSE, _CONVERSE_WITH_NATIVE_OUTPUT, _CONVERSE_WITH_JSON_TOOL, id="converse" + ), + pytest.param("bedrock_invoke", _INVOKE, _INVOKE_WITH_NATIVE_OUTPUT, _INVOKE_WITH_JSON_TOOL, id="invoke"), + ], +) +def test_the_retry_after_bedrock_rejects_native_output_keeps_the_declared_level( + gateway: Gateway, kind: str, target: str, rejected: Expected, retried: Expected +) -> None: + marker: Final = new_marker() + response_format: Final[dict[str, JsonValue]] = { + "type": "json_schema", + "json_schema": {"name": "answer", "schema": _NATIVE_OUTPUT_SCHEMA}, + } + with wire_server(_rejecting_native_output) as wire, gateway.scenario() as scenario: + model: Final = deployment(scenario, kind, wire, _CLAUDE_INFO_WITH_NATIVE_OUTPUT) + response: Final = gateway.request( + "POST", CHAT, chat_request(model, marker, response_format=response_format, thinking=_BUDGET) + ) + received: Final = wire.drain() + assert response.status_code == 200, response.text[:2000] + assert [(request.method, request.target) for request in received] == [("POST", target), ("POST", target)] + assert [body_of(request) for request in received] == [rejected(marker), retried(marker)] + + +_VERTEX_PROJECT: Final = "scripted-project" +_VERTEX_LOCATION: Final = "us-east5" +_VERTEX_TOKEN: Final = "/_oauth/token" +_VERTEX_RAW_PREDICT: Final = ( + f"/v1/projects/{_VERTEX_PROJECT}/locations/{_VERTEX_LOCATION}/publishers/anthropic/models/{CLAUDE}:rawPredict" +) +_VERTEX_ACCESS_TOKEN: Final = json.dumps({"access_token": "scripted", "token_type": "Bearer", "expires_in": 3600}) + + +def _vertex(expected: Expected) -> Expected: + return _with_fields(_without(expected, "model"), {"anthropic_version": "vertex-2023-10-16"}) + + +def _vertex_reply(request: Request) -> Reply: + if request.target == _VERTEX_TOKEN: + return Reply(body=_VERTEX_ACCESS_TOKEN.encode()) + marker: Final = marker_of(request) + return Reply(body=message_json(f"msg_{marker}", CLAUDE, answer(marker))) + + +@pytest.mark.parametrize( + ("route", "fields", "expected", "status"), + [ + pytest.param( + CHAT, + {"reasoning_effort": "medium"}, + _vertex(_anthropic_chat("high", {"type": "adaptive", "display": "summarized"})), + 200, + id="chat-medium-to-high", + ), + pytest.param( + MESSAGES, + {"thinking": _BUDGET}, + _vertex(_native_messages({"effort": "high"}, _ADAPTIVE)), + 200, + id="messages-budget-to-high", + ), + pytest.param(CHAT, {"thinking": _FALSE_BUDGET}, None, 400, id="chat-false-budget-400"), + ], +) +def test_a_vertex_ai_claude_deployment_gets_one_of_its_declared_levels( + gateway: Gateway, route: str, fields: dict[str, JsonValue], expected: Expected | None, status: int +) -> None: + marker: Final = new_marker() + with wire_server(_vertex_reply) as wire, gateway.scenario() as scenario: + model: Final = scenario.model( + model=f"vertex_ai/{CLAUDE}", + api_base=wire.url, + api_key=None, + vertex_project=_VERTEX_PROJECT, + vertex_location=_VERTEX_LOCATION, + vertex_credentials=service_account_json(_VERTEX_PROJECT, wire.url), + model_info=dict(_CLAUDE_INFO), + ) + body: Final = ( + chat_request(model, marker, **fields) if route == CHAT else messages_request(model, marker, **fields) + ) + response: Final = gateway.request("POST", route, body) + received: Final = tuple(request for request in wire.drain() if request.target != _VERTEX_TOKEN) + assert response.status_code == status, response.text[:2000] + if expected is None: + assert _NON_NUMERIC_BUDGET in response.text, response.text[:2000] + assert received == (), [request.body[:300] for request in received] + return + assert answer(marker) in response.text, response.text[:2000] + assert [(request.method, request.target) for request in received] == [("POST", _VERTEX_RAW_PREDICT)] + assert body_of(received[0]) == expected(marker) def _single(received: tuple[Request, ...], marker: str) -> Request: From ddb8a14ee97e3abb6c251edfd3a4ba53467688db Mon Sep 17 00:00:00 2001 From: Mateo Wang <277851410+mateo-berri@users.noreply.github.com> Date: Wed, 28 Oct 2026 15:44:43 -0700 Subject: [PATCH 18/41] feat(skills): add a litellm-setup agent skill (#47731) * feat(skills): add a litellm-setup agent skill * fix(skills): run the quickstart for Docker with Postgres in litellm-setup Use scripts/quickstart.sh for the Docker plus Postgres path instead of a hand-rolled Postgres container, keep uv with the user's own Postgres, load .env as data instead of sourcing it, pin the uv Python range, chmod an existing .env, list bedrock_converse and vertex_ai rows in the cost map lookup, ask before editing the shell profile, and stop the health wait when the proxy exits * fix(skills): guide empty Postgres, docker sudo, and key fixes in litellm-setup Ask for a new empty database on the uv path and DIRECT_URL behind a pooler, add the baseline error to troubleshooting, and drop inherited database settings in the loader line. Check python3 and lsof, explain Docker's permission denied, give quickstart commands by the folder's path, and fix provider keys by recreating the container or deleting the quickstart model. Drop the structure-only test. * fix(skills): keep shell profiles untouched when litellm-setup installs uv The uv installer edits shell profiles unless UV_NO_MODIFY_PATH is set, so the skill now sets it and puts ~/.local/bin on PATH for later commands. It also says litellm --setup runs in a separate terminal and how to retire the shared master-key UI login * fix(skills): offer litellm --setup only on the uv path The Docker paths install no host litellm CLI, and the old wording also passed the skill's own step number on to the user * fix(skills): ask before using an env provider key, print model names config.yaml accepts, probe the port with curl * fix(skills): state the env key default in the question, chmod the wizard config, keep uv's bin dir on PATH * fix(skills): keep litellm-setup on its own .env and install when the shell has others * fix(skills): read the setup wizard's config without printing keys and hand off the exact restart command * fix(skills): hide the database-access dev skill from end-user skill installs * fix(skills): give the litellm --setup path a restart command that keeps its config * docs(skills): point litellm --setup users to the master key in litellm_config.yaml * fix(skills): hide every non-end-user skill, use Azure deployment names, keep the restart block out of the user's shell * test(skills): find non-end-user skills in-process instead of through git * docs(skills): price Azure deployments with base_model and put uv on PATH in the start block --- .agents/skills/database-access/SKILL.md | 2 + .../.agents/skills/rust-error/SKILL.md | 2 + .../.agents/skills/rust-gateway/SKILL.md | 2 + .../.agents/skills/rust-string-enums/SKILL.md | 2 + .../.agents/skills/rust-tracing/SKILL.md | 2 + skills/litellm-setup/SKILL.md | 205 ++++++++++++++++++ .../slack-gif-creator/SKILL.md | 2 + .../test-delete-skill/SKILL.md | 2 + .../test-skill-litellm/SKILL.md | 2 + tests/unit/test_repo_agent_skills.py | 68 ++++++ .../.agents/skills/url-state/SKILL.md | 2 + 11 files changed, 291 insertions(+) create mode 100644 skills/litellm-setup/SKILL.md create mode 100644 tests/unit/test_repo_agent_skills.py diff --git a/.agents/skills/database-access/SKILL.md b/.agents/skills/database-access/SKILL.md index 796dfafe4f6..13fb7b7e286 100644 --- a/.agents/skills/database-access/SKILL.md +++ b/.agents/skills/database-access/SKILL.md @@ -1,6 +1,8 @@ --- name: database-access description: Add database queries, extract existing Prisma calls into repositories, or explicitly port queries to Rust and SQLx without expanding migration scope +metadata: + internal: true --- # Database access diff --git a/litellm-rust/.agents/skills/rust-error/SKILL.md b/litellm-rust/.agents/skills/rust-error/SKILL.md index fd5d6222287..74da1ecbcf4 100644 --- a/litellm-rust/.agents/skills/rust-error/SKILL.md +++ b/litellm-rust/.agents/skills/rust-error/SKILL.md @@ -1,6 +1,8 @@ --- name: rust-error description: Define, wrap, convert, classify, report or test Rust error types in litellm-rust with thiserror. Use when adding an error type or variant, mapping one crate's error into another, turning an error into a gateway response or Python exception, deciding retry or fault from an error, or replacing call sites that build errors from strings +metadata: + internal: true --- # Rust errors diff --git a/litellm-rust/.agents/skills/rust-gateway/SKILL.md b/litellm-rust/.agents/skills/rust-gateway/SKILL.md index 66b93ee128d..4fee0b54a68 100644 --- a/litellm-rust/.agents/skills/rust-gateway/SKILL.md +++ b/litellm-rust/.agents/skills/rust-gateway/SKILL.md @@ -1,6 +1,8 @@ --- name: rust-gateway description: Add or change an HTTP route, extractor, response envelope, stream or router mount in the litellm-rust gateway crates (crates/gateway and crates/gateway-*). Use when adding an endpoint, parsing a request body, choosing an error envelope for an API surface, relaying SSE, or writing a router-level test +metadata: + internal: true --- # Rust gateway diff --git a/litellm-rust/.agents/skills/rust-string-enums/SKILL.md b/litellm-rust/.agents/skills/rust-string-enums/SKILL.md index 5fc067790a0..352f6427dd9 100644 --- a/litellm-rust/.agents/skills/rust-string-enums/SKILL.md +++ b/litellm-rust/.agents/skills/rust-string-enums/SKILL.md @@ -1,6 +1,8 @@ --- name: rust-string-enums description: Define or refactor Rust string-valued enums and their Serde adapters in litellm-rust, using Strum and serde_with while preserving parsing, wire values, and schemas +metadata: + internal: true --- # Rust string enums diff --git a/litellm-rust/.agents/skills/rust-tracing/SKILL.md b/litellm-rust/.agents/skills/rust-tracing/SKILL.md index c1c7f43c456..a8da3f3bb14 100644 --- a/litellm-rust/.agents/skills/rust-tracing/SKILL.md +++ b/litellm-rust/.agents/skills/rust-tracing/SKILL.md @@ -1,6 +1,8 @@ --- name: rust-tracing description: Add or change Rust diagnostic tracing in litellm-rust, including route spans, subscriber layers, and Python logger delivery +metadata: + internal: true --- # Rust tracing diff --git a/skills/litellm-setup/SKILL.md b/skills/litellm-setup/SKILL.md new file mode 100644 index 00000000000..7c824751d6f --- /dev/null +++ b/skills/litellm-setup/SKILL.md @@ -0,0 +1,205 @@ +--- +name: litellm-setup +description: Set up a self-hosted LiteLLM proxy, the OpenAI-compatible LLM gateway, on this machine, from install through provider keys, config.yaml, a master key, optional Postgres and Admin UI, and a verified chat completion. Use when the user asks to set up LiteLLM, install the LiteLLM proxy, self-host an LLM gateway, or put one endpoint in front of OpenAI, Anthropic, Gemini, Azure, Bedrock, Ollama, or other providers. Not for using the LiteLLM Python SDK inside an app +--- + +# Set up a LiteLLM proxy + +Take the user from nothing to a running LiteLLM proxy that has answered one chat completion. Run every command yourself unless a step says the user does it + +Never print a secret. Provider keys, the master key, the salt key, the database URL, and virtual keys live only in `.env`, and `config.yaml` reads them through `os.environ/NAME`. Check that a value is there with `grep -c '^NAME=.' .env`, never with `cat .env` or `echo "$NAME"`. Load `.env` into a shell with this line, which exports each line as it is, so a `&` or `$` in a value stays literal, and first drops any database settings the shell already had and sets `LITELLM_MODE=PRODUCTION`, which stops `litellm` from also loading a `.env` it finds above its install folder, such as `~/.env`, so only this `.env` picks the database. The steps that need it start with it: + +```bash +unset DATABASE_URL DIRECT_URL DATABASE_HOST; export LITELLM_MODE=PRODUCTION; while IFS= read -r line || [ -n "$line" ]; do case "$line" in [A-Za-z_]*=*) export "$line" ;; esac; done < .env +``` + +## 1. Ask once + +Ask everything in one message, with defaults the user can accept by saying "go": + +- Providers and models. Default is OpenAI with its newest flagship model, using the `OPENAI_API_KEY` in your environment when `[ -n "$OPENAI_API_KEY" ] && echo set` says it is there. Run that check before asking, and say in the question that the default uses that variable, naming it but never its value, since a key in your environment can belong to another account +- Whether they want virtual keys, teams, spend tracking, or the Admin UI. Those need Postgres, plain routing does not. Default is no +- uv (default) or Docker. With Postgres, Docker runs the repo's quickstart, which brings its own database, and uv needs a Postgres server they already have, with a new, empty database for LiteLLM, so someone with their own Postgres server picks uv +- Folder and port. Defaults are `./litellm-proxy` (`./litellm-gateway` for the quickstart) and 4000 + +Then follow the steps for their answers. Plain routing skips steps 3 and 6. The quickstart is steps 2, 3, 8, and 9. uv with their own Postgres skips step 3 + +In the same message, say that someone who picks uv and would rather answer prompts can run `litellm --setup` in a separate terminal window once LiteLLM is installed, not through Claude Code. Never run it yourself, since it needs a real terminal and without one prints "Setup cancelled." and writes nothing. Give them the command as `LITELLM_MODE=PRODUCTION /litellm --setup`, with `` replaced by the folder `uv tool dir --bin` prints, to run from the folder they want the config in. It writes `litellm_config.yaml` there with the provider keys and master key in plain text and offers to start the proxy, which then listens on every network interface and stops with Ctrl+C in that window. When the user is done, run this in that folder. It makes the file readable only by them, keeps it out of git (the wizard does neither), and prints only the model names. Never print that file or run `grep`, `sed`, or `cat` on it: + +```bash +chmod 600 litellm_config.yaml; grep -qx 'litellm_config.yaml' .gitignore 2>/dev/null || echo 'litellm_config.yaml' >> .gitignore +"$(uv tool dir)/litellm/bin/python" -I -c 'import yaml; print(*(m["model_name"] for m in yaml.safe_load(open("litellm_config.yaml"))["model_list"]), sep="\n")' +``` + +Then skip to step 8 and run it against their proxy, starting that shell command with this line instead of the loader line, which reads the master key without printing it: + +```bash +export LITELLM_MASTER_KEY="$("$(uv tool dir)/litellm/bin/python" -I -c 'import yaml; print(yaml.safe_load(open("litellm_config.yaml"))["general_settings"]["master_key"])')" +``` + +## 2. Check prerequisites + +```bash +uv --version; docker info --format '{{.ServerVersion}}'; docker compose version; curl --version | head -1; openssl version; command -v python3 +``` + +On native Windows, ask the user to run Claude Code inside WSL, since every command here is POSIX shell. Every path needs curl, openssl, and python3, so ask the user to install any that is missing. The uv path needs uv, which downloads a supported Python (LiteLLM supports 3.10 to 3.14) when the `--python` range in step 4 rules out the one on the machine. If uv is missing, offer to install it, and only after the user says yes run `curl -LsSf https://astral.sh/uv/install.sh | env UV_NO_MODIFY_PATH=1 sh`, which leaves their shell profile alone (https://docs.astral.sh/uv/getting-started/installation/). It puts uv in `~/.local/bin`, so start every later command that runs `uv` or `litellm` with `export PATH="$HOME/.local/bin:$PATH";`. The Docker path needs a running Docker daemon, and the quickstart also needs Docker Compose v2. If `docker info` says `permission denied`, the user's account cannot use Docker without sudo. Offer the uv path, or ask them to add themselves to the `docker` group the way https://docs.docker.com/engine/install/linux-postinstall/ shows and start a new login session before you check again + +Check the port is free with `curl -s -o /dev/null --max-time 2 http://127.0.0.1:4000; echo $?`, which prints `7` (could not connect) only when nothing listens there. Don't use `lsof` for this, since it misses ports that root or Docker hold. If it is taken, pick another port and use it everywhere below, because a proxy asked for a busy 4000 silently moves to a random port + +## 3. Postgres with Docker: run the quickstart + +The repo's quickstart starts the gateway, Postgres, and the Admin UI in Docker, listening on this machine only, and puts a `litellm-gateway` command in that folder that stops, starts, updates, or removes them. It asks nothing when it has no terminal. It writes the master key, salt key, and database password to its own `.env` and never prints them. Download it, then run it with the folder and port from step 1. It returns once the gateway is ready: + +```bash +curl -fsSL https://raw.githubusercontent.com/BerriAI/litellm/main/scripts/quickstart.sh -o quickstart.sh +LITELLM_DIR="$PWD/litellm-gateway" LITELLM_PORT=4000 sh quickstart.sh --yes +cd litellm-gateway +``` + +Models live in the database here, so add each one through the API instead of a `config.yaml`, naming it the way step 5 says. Pipe the request body in, so the provider key never shows up in output or on a command line. Unless the user agreed to the key in your environment, as step 5 says, ask them to add the line `OPENAI_API_KEY=` to this `.env` themselves. Run the loader line first: + +```bash +python3 -c 'import json, os, sys; json.dump({"model_name": "gpt-6.1-sol", "litellm_params": {"model": "openai/gpt-6.1-sol", "api_key": os.environ["OPENAI_API_KEY"]}}, sys.stdout)' \ + | curl -sS -o /dev/null -w '%{http_code}\n' http://127.0.0.1:4000/model/new \ + -H "Authorization: Bearer $LITELLM_MASTER_KEY" -H "Content-Type: application/json" -d @- +``` + +Run that in one shell command that starts with the loader line, since the master key is in `.env`. `200` means the model was added. Then go to step 8 + +## 4. Install + +uv: + +```bash +uv tool install --python '>=3.10,<3.15' 'litellm[proxy]' +export PATH="$(uv tool dir --bin):$PATH"; litellm --version +``` + +With their own Postgres, install `'litellm[proxy,extra_proxy]'` instead, with the same `--python` range, which adds the Prisma client the proxy uses for the database. Running `uv tool install` again with the new extras replaces the old install. Start every later command that runs `litellm` with `export PATH="$(uv tool dir --bin):$PATH";`, so this install runs even when the shell has no `litellm` or an older one from a venv or `pip install litellm`. `uv tool update-shell` puts it on the PATH of new shells by editing the user's shell profile, so run it only after the user says yes + +Docker: + +```bash +docker pull docker.litellm.ai/berriai/litellm:main-stable +``` + +## 5. Write `.env` and `config.yaml` + +Run every command from here on in the folder from step 1. Create `.env` readable only by the user, keep it out of git, and add a master key unless it already has one. The proxy refuses to start without one, and `sk-` is the format of every key it issues: + +```bash +mkdir -p litellm-proxy && cd litellm-proxy +touch .env && chmod 600 .env; grep -qx '.env' .gitignore 2>/dev/null || echo '.env' >> .gitignore +grep -q '^LITELLM_MASTER_KEY=.' .env || printf 'LITELLM_MASTER_KEY=sk-%s\n' "$(openssl rand -hex 32)" >> .env +``` + +Copy a provider key that is already in your environment only after the user agreed to that variable, in their step 1 answer or when you ask now, and copy it without printing it, as in `printf 'OPENAI_API_KEY=%s\n' "$OPENAI_API_KEY" >> .env`. Otherwise ask the user to add the line `OPENAI_API_KEY=` to `.env` themselves and tell you when it is there. If they paste a key into the chat anyway, write it to `.env` and never repeat it. Keep one `NAME=value` per line with no quotes, so `docker --env-file` and the loader line read the same value + +Confirm each model's current LiteLLM name in the cost map instead of trusting memory, since model names go stale fast. This prints the chat models of the provider named last (`openai`, `anthropic`, `gemini`, `azure`, `bedrock`, `vertex_ai`, and so on) the way `config.yaml` takes them, with the provider in front. It matches by prefix, so `bedrock` also lists the newer Bedrock models and `vertex_ai` lists every Vertex family. For Ollama, use a model from `ollama list` instead: + +```bash +curl -fsSL https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json | python3 -c ' +import json, sys +for name, row in json.load(sys.stdin).items(): + provider = str(row.get("litellm_provider", "")) if isinstance(row, dict) else "" + if provider.startswith(sys.argv[1]) and row.get("mode") == "chat": + prefix = "bedrock" if provider == "bedrock_converse" else "vertex_ai" if provider.startswith("vertex_ai") else provider + print(name if name.startswith(prefix + "/") else prefix + "/" + name) +' openai +``` + +In `config.yaml`, `model` is a name from that list, as in `openai/gpt-6.1-sol` or `bedrock/us.anthropic.`, and `model_name` is what clients will send. If the folder already has a `config.yaml`, show the user its `model_name` lines and ask before replacing it. This example uses the newest GPT model in the cost map when this skill was written, so replace it with what the user picked: + +```yaml +model_list: + - model_name: gpt-6.1-sol + litellm_params: + model: openai/gpt-6.1-sol + api_key: os.environ/OPENAI_API_KEY + +general_settings: + master_key: os.environ/LITELLM_MASTER_KEY +``` + +Add one entry per model. On Azure, ask the user for each deployment's name and the model it runs. Use `azure/` as `model`, since Azure routes by deployment and a name from the lookup works only when the deployment has that same name, and add `model_info:` with `base_model: azure/` to the entry, since LiteLLM prices each call by that name. Providers that need more than a key, like Azure or Bedrock, list their `litellm_params` on their page under https://docs.litellm.ai/docs/providers. Point every secret at `.env` through `os.environ/NAME` the same way + +## 6. Their own Postgres, on the uv path + +Ask the user to create a new, empty database for LiteLLM on their server (`CREATE DATABASE litellm;`) and add `DATABASE_URL=postgresql://...` for it to `.env` themselves. The proxy creates its tables there on first start and stops on a database that already holds other tables. If that URL goes through a transaction pooler such as PgBouncer, they also add `DIRECT_URL` for the same database without the pooler, which the migrations use. Then add a salt key. It encrypts the provider credentials the proxy stores in the database and falls back to the master key when unset, so set it once and never change it: + +```bash +grep -q '^LITELLM_SALT_KEY=.' .env || printf 'LITELLM_SALT_KEY=sk-%s\n' "$(openssl rand -hex 32)" >> .env +``` + +Generate the Prisma client once inside the tool's environment. A pip or uv install doesn't ship it generated (Docker images do): + +```bash +tool="$(uv tool dir)/litellm" +schema=$("$tool/bin/python" -I -c 'import importlib.util, os; print(os.path.join(importlib.util.find_spec("litellm").submodule_search_locations[0], "proxy", "schema.prisma"))') +PATH="$tool/bin:$PATH" prisma generate --schema "$schema" +``` + +## 7. Start the proxy + +Bind to `127.0.0.1` unless the user wants other machines to reach it + +uv, in one shell command that starts with the loader line, since the installed `litellm` does not read `.env` by itself: + +```bash +export PATH="$HOME/.local/bin:$PATH"; export PATH="$(uv tool dir --bin):$PATH" +nohup litellm --config config.yaml --host 127.0.0.1 --port 4000 > litellm.log 2>&1 & +echo $! > litellm.pid +``` + +Docker: + +```bash +docker run -d --name litellm --env-file .env -v "$PWD/config.yaml:/app/config.yaml:ro" -p 127.0.0.1:4000:4000 \ + docker.litellm.ai/berriai/litellm:main-stable --config /app/config.yaml --port 4000 +``` + +Wait on the health route, stopping early if the proxy exits. The first start with Postgres applies the database migrations and takes longer. Define `alive` for the path you took, `alive() { kill -0 "$(cat litellm.pid)" 2>/dev/null; }` for uv or `alive() { [ "$(docker inspect -f '{{.State.Running}}' litellm 2>/dev/null)" = true ]; }` for Docker, in the same shell command as the loop: + +```bash +for i in $(seq 1 180); do curl -fsS http://127.0.0.1:4000/health/liveliness >/dev/null 2>&1 && break; alive || break; sleep 1; done +curl -sS http://127.0.0.1:4000/health/readiness +``` + +Readiness reports `"db":"connected"` with Postgres and `"db":"Not connected"` without it. If the proxy never comes up, read the end of `litellm.log` (or `docker logs litellm`) and check the troubleshooting list + +## 8. Verify with a chat completion + +Use a `model_name` from `config.yaml` (or step 3), in one shell command that starts with the loader line: + +```bash +curl -sS http://127.0.0.1:4000/v1/chat/completions \ + -H "Authorization: Bearer $LITELLM_MASTER_KEY" -H "Content-Type: application/json" \ + -d '{"model": "gpt-6.1-sol", "messages": [{"role": "user", "content": "Say hello in five words."}]}' +``` + +Show the user the reply in `choices[0].message.content`. An error here means the setup is not done, so fix it before handing off + +## 9. Hand off + +Tell the user, without printing any secret: + +- The base URL is `http://127.0.0.1:4000`. Any OpenAI-compatible client works with it as the base URL and the master key or a virtual key as the API key +- The master key is `LITELLM_MASTER_KEY` in the folder's `.env`, or `general_settings.master_key` in `litellm_config.yaml` when they used `litellm --setup` +- To stop it, `kill "$(cat litellm.pid)"` (uv), `docker stop litellm` (Docker), `/litellm-gateway stop` (quickstart), or Ctrl+C in its window (`litellm --setup`). To start it again, `docker start litellm`, `/litellm-gateway start`, on uv the whole step 7 command as you ran it, loader line and PATH export included, wrapped in `( ... )` so their shell keeps its own settings, shown as a code block and never paraphrased, since a new shell has neither, or for `litellm --setup` `LITELLM_MODE=PRODUCTION /litellm --config litellm_config.yaml --host 127.0.0.1 --port ` in a terminal window in that folder, with `` and `` filled in, since running the wizard again writes a new config with a new master key. `update` and `uninstall` update or remove a quickstart install. Give the quickstart commands with the folder's full path, since a bare `litellm-gateway` is on PATH only when `~/.local/bin` is, and can belong to an earlier install +- With Postgres, the Admin UI is at `http://127.0.0.1:4000/ui` (user `admin`, password the master key), and `POST /key/generate` with the master key creates virtual keys with their own budgets and model lists. Without Postgres, both need step 3 or 6 first +- The Admin UI warns that the master-key login is a shared secret, which is fine while only this machine reaches the proxy. Before anyone else can reach it, invite an admin with their own password on the Internal Users page, then set `general_settings.disable_env_credential_login: true` in the proxy config +- To manage the running proxy from a coding agent (keys, teams, users, models, MCP servers), install `npx skills add BerriAI/litellm-skills -g` + +## Troubleshooting + +- `refused to start: no master key is set` means the loader line did not run in the shell command that started `litellm`, or `--env-file .env` is missing from `docker run` +- `a database URL is set but the prisma CLI is neither on PATH nor importable` means the install lacks `extra_proxy`, so rerun step 4 with it, or, on plain routing, that the command that started `litellm` did not start with the loader line +- `The Client hasn't been generated yet` or `Unable to find Prisma binaries` means the Prisma generate command in step 6 has not run +- `Incorrect API key provided` or another provider `AuthenticationError` means a provider key is wrong. Fix it in `.env` and restart, since the proxy reads the environment only at startup. On uv, stop it and rerun step 7. On Docker, `docker rm -f litellm` and rerun step 7, since `docker restart` keeps the environment the container was created with. On the quickstart the key is saved with the model, so delete the model and add it again: `GET /model/info` lists each model's `model_info.id`, and `POST /model/delete` with `{"id": ""}` deletes it +- `Cannot automatically baseline this database` means `DATABASE_URL` points at a database that already holds tables LiteLLM did not create. Point it at a new, empty database +- `LLM Provider NOT provided` means a `model` in `config.yaml` (or step 3) lacks its provider in front. Use the name as the step 5 lookup prints it +- `Invalid model name passed in model=...` means the request's `model` is not a `model_name` in `config.yaml` or step 3. `GET /v1/models` lists the ones the key can use +- `No connected db.` on a proxy without Postgres means the bearer token is not the master key +- `Conflict. The container name "/litellm" is already in use` means an earlier run left a container. `docker start litellm` reuses it, or with the user's OK `docker rm -f litellm` and rerun step 7 +- Inside Docker, `localhost` in `config.yaml` (an Ollama `api_base`, say) is the container itself. Use `host.docker.internal`, and on Linux also pass `--add-host=host.docker.internal:host-gateway` diff --git a/tests/unit/skills/test_skills_data/slack-gif-creator/SKILL.md b/tests/unit/skills/test_skills_data/slack-gif-creator/SKILL.md index 3cae971b731..d2cdfd17bde 100644 --- a/tests/unit/skills/test_skills_data/slack-gif-creator/SKILL.md +++ b/tests/unit/skills/test_skills_data/slack-gif-creator/SKILL.md @@ -1,6 +1,8 @@ --- name: slack-gif-creator description: Knowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." +metadata: + internal: true --- # Slack GIF Creator diff --git a/tests/unit/skills/test_skills_data/test-delete-skill/SKILL.md b/tests/unit/skills/test_skills_data/test-delete-skill/SKILL.md index ba3fda9ab7b..ef1fd1a7b24 100644 --- a/tests/unit/skills/test_skills_data/test-delete-skill/SKILL.md +++ b/tests/unit/skills/test_skills_data/test-delete-skill/SKILL.md @@ -1,6 +1,8 @@ --- name: test-delete-skill description: A test skill created specifically for deletion testing +metadata: + internal: true --- # Test Delete Skill diff --git a/tests/unit/skills/test_skills_data/test-skill-litellm/SKILL.md b/tests/unit/skills/test_skills_data/test-skill-litellm/SKILL.md index ec83b580f0e..b3d54ccdf3c 100644 --- a/tests/unit/skills/test_skills_data/test-skill-litellm/SKILL.md +++ b/tests/unit/skills/test_skills_data/test-skill-litellm/SKILL.md @@ -1,6 +1,8 @@ --- name: test-skill-litellm description: A test skill created by LiteLLM automated tests +metadata: + internal: true --- # Test Skill diff --git a/tests/unit/test_repo_agent_skills.py b/tests/unit/test_repo_agent_skills.py new file mode 100644 index 00000000000..591241f0cf9 --- /dev/null +++ b/tests/unit/test_repo_agent_skills.py @@ -0,0 +1,68 @@ +from collections.abc import Iterator +from fnmatch import fnmatch +from pathlib import Path +from typing import Final + +import pytest +import yaml + +_REPO_ROOT: Final = Path(__file__).resolve().parents[2] +_SKILLS_DIR: Final = _REPO_ROOT / "skills" +_SKILL_DIRECTORIES: Final = tuple(sorted(skill_file.parent for skill_file in _SKILLS_DIR.glob("*/SKILL.md"))) +_FULL_DEPTH_MAX_DEPTH: Final = 5 +_FULL_DEPTH_SKIPPED_NAMES: Final = frozenset(("node_modules", ".git", "dist", "build", "__pycache__")) +_GITIGNORED_NAME_PATTERNS: Final = tuple( + line.strip().rstrip("/") + for line in (_REPO_ROOT / ".gitignore").read_text(encoding="utf-8").splitlines() + if line.strip() and not line.startswith(("#", "!")) and "/" not in line.strip().rstrip("/") +) + + +def _absent_from_a_clone_or_skipped_by_the_cli(name: str) -> bool: + return name in _FULL_DEPTH_SKIPPED_NAMES or any(fnmatch(name, pattern) for pattern in _GITIGNORED_NAME_PATTERNS) + + +def _full_depth_skill_directories(directory: Path, depth: int) -> Iterator[Path]: + if (directory / "SKILL.md").is_file(): + yield directory + if depth == _FULL_DEPTH_MAX_DEPTH: + return + for child in sorted(directory.iterdir()): + if child.is_dir() and not _absent_from_a_clone_or_skipped_by_the_cli(child.name): + yield from _full_depth_skill_directories(child, depth + 1) + + +_NON_END_USER_SKILL_DIRECTORIES: Final = tuple( + directory for directory in _full_depth_skill_directories(_REPO_ROOT, 0) if directory.parent != _SKILLS_DIR +) + + +def _frontmatter(skill_directory: Path) -> object: + opening, frontmatter, _body = (skill_directory / "SKILL.md").read_text(encoding="utf-8").split("---\n", 2) + assert opening == "", f"{skill_directory.name}/SKILL.md must open with a --- frontmatter block" + return yaml.safe_load(frontmatter) + + +@pytest.mark.parametrize("skill_directory", _SKILL_DIRECTORIES, ids=lambda path: path.name) +def test_skill_frontmatter_has_string_name_matching_its_directory_and_a_description(skill_directory: Path) -> None: + frontmatter: Final = _frontmatter(skill_directory) + assert isinstance(frontmatter, dict), frontmatter + assert frontmatter.get("name") == skill_directory.name, frontmatter + description: Final = frontmatter.get("description") + assert isinstance(description, str) and description.strip(), frontmatter + + +@pytest.mark.parametrize( + "skill_directory", + _NON_END_USER_SKILL_DIRECTORIES, + ids=lambda path: str(path.relative_to(_REPO_ROOT)), +) +def test_skill_outside_skills_folder_is_hidden_from_end_user_skill_installs(skill_directory: Path) -> None: + frontmatter: Final = _frontmatter(skill_directory) + assert isinstance(frontmatter, dict), frontmatter + metadata: Final = frontmatter.get("metadata") + assert isinstance(metadata, dict) and metadata.get("internal") is True, ( + f"{skill_directory} needs `metadata: {{internal: true}}`, or `npx skills add BerriAI/litellm` installs it " + "for end users, from a root dot-folder always and from anywhere with --full-depth (skills CLI 1.7.4, " + "checked 2026-10-28)" + ) diff --git a/ui/litellm-dashboard/.agents/skills/url-state/SKILL.md b/ui/litellm-dashboard/.agents/skills/url-state/SKILL.md index 986118d81a7..8bf8fadd19e 100644 --- a/ui/litellm-dashboard/.agents/skills/url-state/SKILL.md +++ b/ui/litellm-dashboard/.agents/skills/url-state/SKILL.md @@ -1,6 +1,8 @@ --- name: url-state description: Read or write dashboard URL query state (tabs, filters, pagination, deep links, demo flags) with nuqs instead of raw history or URLSearchParams +metadata: + internal: true --- # URL state From 752b5c053a50f1cb56d1d099f2660a4459dafc1c Mon Sep 17 00:00:00 2001 From: Kerry Lu Date: Wed, 28 Oct 2026 15:52:15 -0700 Subject: [PATCH 19/41] fix(ci): flag any schema.prisma change inside model LiteLLM_SpendLogs in the spend logs guard (LIT-10240) (#48905) Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- ci_cd/spend_logs_guard.py | 107 ++++++++++++++---- tests/unit/test_spend_logs_guard.py | 163 +++++++++++++++++++++++----- 2 files changed, 220 insertions(+), 50 deletions(-) diff --git a/ci_cd/spend_logs_guard.py b/ci_cd/spend_logs_guard.py index a4d9a2e0a0d..5e6d38fba57 100644 --- a/ci_cd/spend_logs_guard.py +++ b/ci_cd/spend_logs_guard.py @@ -2,11 +2,12 @@ needs an approving review from the spend-log owner before it can merge. CODEOWNERS can only match file paths, and every schema change edits the same schema.prisma copies, -so a path rule would put every migration on the owner. This gate reads the diff instead: when no -added line in the watched files mentions the table it passes at once, otherwise it passes only once -the owner's latest review on the pull request is an approval that still covers every spend-log line -in the head (approvals survive pushes, so a later commit may not add spend-log lines the owner never -saw). Reviews do not trigger the workflow, so the check is re-run by hand after approving. +so a path rule would put every migration on the owner. This gate reads the diff instead: a schema.prisma +hunk counts when it adds or removes a line inside the `model LiteLLM_SpendLogs { ... }` block, a migration +hunk counts when an added line names the table. Without such a hunk it passes at once, otherwise it passes +only once the owner's latest review on the pull request is an approval that still covers every spend-log line +in the head (approvals survive pushes, so a later commit may not add spend-log lines the owner never saw). +Reviews do not trigger the workflow, so the check is re-run by hand after approving. """ from __future__ import annotations @@ -15,8 +16,10 @@ import argparse import re import subprocess import sys -from collections.abc import Callable, Sequence +from collections.abc import Callable, Mapping, Sequence +from dataclasses import dataclass from pathlib import Path +from types import MappingProxyType from typing import Final from pydantic import BaseModel, ConfigDict, TypeAdapter @@ -29,6 +32,8 @@ WATCHED: Final = ( MIGRATIONS_PREFIX: Final = "litellm-proxy-extras/litellm_proxy_extras/migrations/" TABLE: Final = re.compile(r"(? bool: return path in WATCHED or (path.startswith(MIGRATIONS_PREFIX) and path.endswith("/migration.sql")) -def _added_lines(section: str) -> tuple[str, ...]: - body: Final = section.split("\n@@", 1) - return tuple(line[1:] for line in body[1].splitlines() if line.startswith("+")) if len(body) == 2 else () +def model_block(schema: str) -> range: + """1-based line range of `model LiteLLM_SpendLogs { ... }` including both braces, empty when the schema has none.""" + lines: Final = schema.splitlines() + starts: Final = tuple(number for number, line in enumerate(lines, 1) if MODEL_START.match(line)) + if not starts: + return range(0) + ends: Final = tuple(number for number, line in enumerate(lines, 1) if number > starts[0] and line.startswith("}")) + return range(starts[0], (ends[0] if ends else len(lines)) + 1) -def spend_log_hits(diff: str) -> tuple[str, ...]: - """`path: added line` for every added line in a watched file that names the table.""" +def _hunks(section: str) -> tuple[Hunk, ...]: + headers: Final = tuple(HUNK.finditer(section)) + ends: Final = tuple(header.start() for header in headers[1:]) + (len(section),) + return tuple(_hunk(header, section[header.end() : end]) for header, end in zip(headers, ends)) + + +def _hunk(header: re.Match[str], body: str) -> Hunk: + lines: Final = body.splitlines() + return Hunk( + removed=tuple(enumerate((line[1:] for line in lines if line.startswith("-")), int(header.group(1)))), + added=tuple(enumerate((line[1:] for line in lines if line.startswith("+")), int(header.group(2)))), + ) + + +def spend_log_hits(diff: str, blocks: Mapping[str, Block]) -> tuple[str, ...]: + """`path: +line` or `path: -line` for every changed line that touches the table. + + A schema.prisma line counts when its number falls inside that copy's `blocks` entry (head for added lines, + merge base for removed ones), a migration line when it is added and names the table.""" sections: Final = re.split(r"^diff --git [^\n]*\n", diff, flags=re.MULTILINE) - return tuple(f"{path}: {line.strip()}" for path, line in _watched_added_lines(sections) if TABLE.search(line)) - - -def _watched_added_lines(sections: Sequence[str]) -> tuple[tuple[str, str], ...]: - pairs: Final = tuple((_path_of(section), section) for section in sections) + paths: Final = tuple((_path_of(section), section) for section in sections) return tuple( - (path, line) for path, section in pairs if watched(path) for line in _added_lines(section) - ) # comprehension-ok: path then its lines + hit for path, section in paths if watched(path) for hit in _section_hits(path, _hunks(section), blocks) + ) # comprehension-ok: path then its hits + + +def _section_hits(path: str, hunks: Sequence[Hunk], blocks: Mapping[str, Block]) -> tuple[str, ...]: + if path not in blocks: + return tuple(f"{path}: +{line.strip()}" for hunk in hunks for _, line in hunk.added if TABLE.search(line)) + block: Final = blocks[path] + removed: Final = tuple( + f"{path}: -{line.strip()}" for hunk in hunks for number, line in hunk.removed if number in block.base + ) # comprehension-ok: hunk then its lines + added: Final = tuple( + f"{path}: +{line.strip()}" for hunk in hunks for number, line in hunk.added if number in block.head + ) # comprehension-ok: hunk then its lines + return removed + added def _path_of(section: str) -> str: @@ -111,13 +161,25 @@ def added_diff(base: str, head: str) -> str: return result.stdout +def file_at(revision: str, path: str) -> str: + """Contents of `path` at `revision`, empty when the file does not exist there.""" + result: Final = subprocess.run(("git", "show", f"{revision}:{path}"), check=False, capture_output=True, text=True) + return result.stdout if result.returncode == 0 else "" + + +def schema_blocks(base: str, head: str) -> Mapping[str, Block]: + return MappingProxyType( + {path: Block(base=model_block(file_at(base, path)), head=model_block(file_at(head, path))) for path in WATCHED} + ) + + def fetched(commit: str) -> bool: result: Final = subprocess.run(("git", "fetch", "--no-tags", "--depth=1", "origin", commit), check=False) return result.returncode == 0 def reviewed_hits(base: str, commit: str) -> tuple[str, ...] | None: - return spend_log_hits(added_diff(base, commit)) if fetched(commit) else None + return spend_log_hits(added_diff(base, commit), schema_blocks(base, commit)) if fetched(commit) else None class Arguments(BaseModel): @@ -135,9 +197,12 @@ def main(argv: Sequence[str]) -> int: parser.add_argument("--approver", required=True, help="GitHub login whose approval unlocks the merge") parser.add_argument("--reviews", required=True, help="JSON file with the pull request's reviews") args: Final = Arguments.model_validate(vars(parser.parse_args(argv))) - hits: Final = spend_log_hits(added_diff(args.base, args.head)) + hits: Final = spend_log_hits(added_diff(args.base, args.head), schema_blocks(args.base, args.head)) if not hits: - sys.stdout.write("No added line in schema.prisma or a migration names LiteLLM_SpendLogs; nothing to review.\n") + sys.stdout.write( + "No change inside model LiteLLM_SpendLogs in schema.prisma and no added migration line names it; " + "nothing to review.\n" + ) return 0 reviews: Final = _REVIEWS.validate_json(args.reviews.read_bytes()) approval: Final = latest_approval(reviews, args.approver) if approved_by(reviews, args.approver) else None diff --git a/tests/unit/test_spend_logs_guard.py b/tests/unit/test_spend_logs_guard.py index 9ac2c923632..eb54099cf8b 100644 --- a/tests/unit/test_spend_logs_guard.py +++ b/tests/unit/test_spend_logs_guard.py @@ -1,3 +1,5 @@ +from __future__ import annotations + import importlib.util import sys from pathlib import Path @@ -12,46 +14,144 @@ guard: Final = importlib.util.module_from_spec(_SPEC) sys.modules["spend_logs_guard"] = guard _SPEC.loader.exec_module(guard) +SCHEMA: Final = ( + "generator client {\n" + ' provider = "prisma-client-py"\n' + "}\n" + "\n" + "model LiteLLM_TeamTable {\n" + " team_id String @id\n" + "}\n" + "\n" + "model LiteLLM_SpendLogs {\n" + " request_id String @id\n" + " api_key String\n" + ' @@index([api_key], map: "LiteLLM_SpendLogs_api_key_idx")\n' + "}\n" + "\n" + "model LiteLLM_UserTable {\n" + " user_id String @id\n" + "}\n" +) +_NO_BLOCKS: Final = {} -def _diff(path: str, added: str, start: int = 10) -> str: - return f"diff --git a/{path} b/{path}\n--- a/{path}\n+++ b/{path}\n@@ -{start},0 +{start},1 @@\n+{added}\n" + +def _hunk(old_start: int, old_count: int, new_start: int, new_count: int, *lines: str) -> str: + body: Final = "".join(f"{line}\n" for line in lines) + return f"@@ -{old_start},{old_count} +{new_start},{new_count} @@\n{body}" + + +def _diff(path: str, *hunks: str) -> str: + return f"diff --git a/{path} b/{path}\n--- a/{path}\n+++ b/{path}\n" + "".join(hunks) + + +def _added(path: str, added: str, start: int = 10) -> str: + return _diff(path, _hunk(start, 0, start, 1, f"+{added}")) + + +def _blocks(path: str, base: str = SCHEMA, head: str = SCHEMA) -> dict[str, guard.Block]: + return {path: guard.Block(base=guard.model_block(base), head=guard.model_block(head))} + + +def _with_line(schema: str, after: int, line: str) -> str: + lines: Final = schema.splitlines(keepends=True) + return "".join(lines[:after]) + f"{line}\n" + "".join(lines[after:]) + + +class TestModelBlock: + def test_the_block_runs_from_the_model_line_through_its_closing_brace(self): + assert guard.model_block(SCHEMA) == range(9, 14) + + def test_a_schema_without_the_model_has_an_empty_block(self): + assert guard.model_block(SCHEMA.replace("LiteLLM_SpendLogs", "LiteLLM_ErrorLogs")) == range(0) class TestSpendLogHits: - def test_an_added_schema_line_naming_the_table_is_a_hit(self): + def test_a_field_added_inside_the_block_is_a_hit_without_naming_the_table(self): + head: Final = _with_line(SCHEMA, 9, " gauntlet_probe String?") + diff: Final = _added("schema.prisma", " gauntlet_probe String?") + assert guard.spend_log_hits(diff, _blocks("schema.prisma", head=head)) == ( + "schema.prisma: +gauntlet_probe String?", + ) + + def test_a_field_removed_from_the_block_is_a_hit(self): + head: Final = SCHEMA.replace(" api_key String\n", "") + diff: Final = _diff("schema.prisma", _hunk(11, 1, 10, 0, "- api_key String")) + assert guard.spend_log_hits(diff, _blocks("schema.prisma", head=head)) == ("schema.prisma: -api_key String",) + + def test_the_whole_model_appearing_or_disappearing_is_a_hit(self): + without: Final = SCHEMA.replace( + SCHEMA[SCHEMA.index("model LiteLLM_SpendLogs") : SCHEMA.index("model LiteLLM_UserTable")], "" + ) + body: Final = tuple(SCHEMA.splitlines()[8:14]) + added: Final = _diff("schema.prisma", _hunk(8, 0, 9, 6, *(f"+{line}" for line in body))) + removed: Final = _diff("schema.prisma", _hunk(9, 6, 8, 0, *(f"-{line}" for line in body))) + assert len(guard.spend_log_hits(added, _blocks("schema.prisma", base=without))) == 5 + assert len(guard.spend_log_hits(removed, _blocks("schema.prisma", head=without))) == 5 + + def test_a_line_naming_the_table_in_another_model_is_not_a_hit(self): + head: Final = _with_line(SCHEMA, 6, " spend_logs LiteLLM_SpendLogs[]") diff: Final = _diff( - "schema.prisma", ' @@index([api_key, startTime], map: "LiteLLM_SpendLogs_api_key_startTime_idx")', 740 - ) - assert guard.spend_log_hits(diff) == ( - 'schema.prisma: @@index([api_key, startTime], map: "LiteLLM_SpendLogs_api_key_startTime_idx")', + "schema.prisma", + _hunk(6, 0, 7, 1, "+ spend_logs LiteLLM_SpendLogs[]"), + _hunk(16, 1, 17, 0, "- user_id String @id"), ) + assert guard.spend_log_hits(diff, _blocks("schema.prisma", head=head)) == () - def test_an_added_migration_line_is_a_hit_in_any_case(self): + def test_each_hunk_is_placed_by_its_own_header(self): + head: Final = _with_line(_with_line(SCHEMA, 2, ' output = "./client"'), 10, " new_col String?") + diff: Final = _diff( + "schema.prisma", + _hunk(2, 0, 3, 1, '+ output = "./client"'), + _hunk(9, 0, 11, 1, "+ new_col String?"), + ) + assert guard.spend_log_hits(diff, _blocks("schema.prisma", head=head)) == ("schema.prisma: +new_col String?",) + + def test_an_added_migration_line_is_a_hit_in_any_case_and_a_removed_one_is_not(self): path: Final = "litellm-proxy-extras/litellm_proxy_extras/migrations/20260101000000_x/migration.sql" - hits: Final = guard.spend_log_hits(_diff(path, 'CREATE INDEX "i" ON "litellm_spendlogs"("api_key");', 3)) - assert hits == (f'{path}: CREATE INDEX "i" ON "litellm_spendlogs"("api_key");',) + diff: Final = _diff( + path, + _hunk( + 3, + 1, + 3, + 1, + '-DROP INDEX "LiteLLM_SpendLogs_old_idx";', + '+CREATE INDEX "i" ON "litellm_spendlogs"("api_key");', + ), + ) + assert guard.spend_log_hits(diff, _NO_BLOCKS) == ( + f'{path}: +CREATE INDEX "i" ON "litellm_spendlogs"("api_key");', + ) - def test_a_removed_line_or_another_table_or_an_unwatched_file_is_not_a_hit(self): - removed: Final = "--- a/schema.prisma\n+++ b/schema.prisma\n@@ -1,1 +1,0 @@\n-model LiteLLM_SpendLogs {\n" - other: Final = _diff("schema.prisma", " team_id String // LiteLLM_TeamTable, not MyLiteLLM_SpendLogs", 5) - elsewhere: Final = _diff("litellm/proxy/utils.py", 'table = "LiteLLM_SpendLogs"', 5) - assert guard.spend_log_hits(removed + other + elsewhere) == () + def test_another_table_in_a_migration_or_an_unwatched_file_is_not_a_hit(self): + path: Final = "litellm-proxy-extras/litellm_proxy_extras/migrations/20260101000000_x/migration.sql" + other: Final = _added( + path, 'ALTER TABLE "LiteLLM_TeamTable" ADD COLUMN "x" TEXT; -- not MyLiteLLM_SpendLogs', 5 + ) + elsewhere: Final = _added("litellm/proxy/utils.py", 'table = "LiteLLM_SpendLogs"', 5) + assert guard.spend_log_hits(other + elsewhere, _NO_BLOCKS) == () def test_a_table_whose_name_only_starts_with_the_watched_one_is_not_a_hit(self): - archive: Final = _diff("schema.prisma", "model LiteLLM_SpendLogsArchive {", 900) - index: Final = _diff("schema.prisma", " @@index([team_id], map: \"LiteLLM_SpendLogs_team_id_idx\")", 760) - assert guard.spend_log_hits(archive + index) == ( - 'schema.prisma: @@index([team_id], map: "LiteLLM_SpendLogs_team_id_idx")', + index: Final = ' @@index([team_id], map: "LiteLLM_SpendLogs_team_id_idx")' + head: Final = _with_line(_with_line(SCHEMA, 18, "model LiteLLM_SpendLogsArchive {"), 12, index) + diff: Final = _diff( + "schema.prisma", + _hunk(12, 0, 13, 1, f"+{index}"), + _hunk(18, 0, 20, 1, "+model LiteLLM_SpendLogsArchive {"), + ) + assert guard.spend_log_hits(diff, _blocks("schema.prisma", head=head)) == ( + 'schema.prisma: +@@index([team_id], map: "LiteLLM_SpendLogs_team_id_idx")', ) def test_two_files_in_one_diff_are_attributed_to_their_own_path(self): - diff: Final = _diff("litellm/proxy/schema.prisma", " a LiteLLM_SpendLogs", 1) + _diff( - "schema.prisma", " b String", 1 - ) - assert guard.spend_log_hits(diff) == ("litellm/proxy/schema.prisma: a LiteLLM_SpendLogs",) + head: Final = _with_line(SCHEMA, 9, " a String") + diff: Final = _added("litellm/proxy/schema.prisma", " a String") + _added("schema.prisma", " b String", 2) + blocks: Final = _blocks("litellm/proxy/schema.prisma", head=head) | _blocks("schema.prisma") + assert guard.spend_log_hits(diff, blocks) == ("litellm/proxy/schema.prisma: +a String",) -def _review(login: str, state: str, commit_id: str | None = "head") -> "guard.Review": +def _review(login: str, state: str, commit_id: str | None = "head") -> guard.Review: return guard.Review(state=state, user=guard.Reviewer(login=login), commit_id=commit_id) @@ -126,24 +226,29 @@ class TestApprovalCovers: class TestExitCode: def test_a_pull_request_without_spend_log_lines_passes_without_reviews(self, tmp_path, monkeypatch, capsys): - monkeypatch.setattr(guard, "added_diff", lambda base, head: _diff("schema.prisma", " foo String", 1)) + monkeypatch.setattr(guard, "added_diff", lambda base, head: _added("schema.prisma", " foo String", 6)) + monkeypatch.setattr(guard, "schema_blocks", lambda base, head: _blocks("schema.prisma")) reviews: Final = tmp_path / "r.json" reviews.write_text("[]") assert guard.main(["--base", "a", "--head", "b", "--approver", "kerry-berri", "--reviews", str(reviews)]) == 0 assert "nothing to review" in capsys.readouterr().out def test_a_spend_log_change_fails_until_the_approver_approves(self, tmp_path, monkeypatch, capsys): - monkeypatch.setattr(guard, "added_diff", lambda base, head: _diff("schema.prisma", " x LiteLLM_SpendLogs", 7)) + head: Final = _with_line(SCHEMA, 9, " x String?") + monkeypatch.setattr(guard, "added_diff", lambda base_sha, head_sha: _added("schema.prisma", " x String?")) + monkeypatch.setattr(guard, "schema_blocks", lambda base_sha, head_sha: _blocks("schema.prisma", head=head)) reviews: Final = tmp_path / "r.json" reviews.write_text('[{"user": {"login": "someone-else"}, "state": "APPROVED"}]') assert guard.main(["--base", "a", "--head", "b", "--approver", "kerry-berri", "--reviews", str(reviews)]) == 1 - assert "schema.prisma: x LiteLLM_SpendLogs" in capsys.readouterr().out + assert "schema.prisma: +x String?" in capsys.readouterr().out reviews.write_text('[{"user": {"login": "kerry-berri"}, "state": "APPROVED", "commit_id": "b"}]') assert guard.main(["--base", "a", "--head", "b", "--approver", "kerry-berri", "--reviews", str(reviews)]) == 0 def test_an_approval_before_a_new_spend_log_line_was_pushed_fails(self, tmp_path, monkeypatch, capsys): - monkeypatch.setattr(guard, "added_diff", lambda base, head: _diff("schema.prisma", " x LiteLLM_SpendLogs", 7)) - monkeypatch.setattr(guard, "reviewed_hits", lambda base, commit: () if commit == "old" else None, raising=False) + head: Final = _with_line(SCHEMA, 9, " x String?") + monkeypatch.setattr(guard, "added_diff", lambda base_sha, head_sha: _added("schema.prisma", " x String?")) + monkeypatch.setattr(guard, "schema_blocks", lambda base_sha, head_sha: _blocks("schema.prisma", head=head)) + monkeypatch.setattr(guard, "reviewed_hits", lambda base_sha, commit: () if commit == "old" else None) reviews: Final = tmp_path / "r.json" reviews.write_text('[{"user": {"login": "kerry-berri"}, "state": "APPROVED", "commit_id": "old"}]') assert guard.main(["--base", "a", "--head", "b", "--approver", "kerry-berri", "--reviews", str(reviews)]) == 1 From 0f786fa1c7b99ee860643597675ed7daf0d289d0 Mon Sep 17 00:00:00 2001 From: Kerry Lu Date: Wed, 28 Oct 2026 15:52:44 -0700 Subject: [PATCH 20/41] feat(router): add deployment rpd and tpd daily limits to enforce_model_rate_limits (#48875) * feat(router): add deployment rpd and tpd daily limits to enforce_model_rate_limits Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * fix(router): route rate limit cache increments through typed helpers * chore(ui): regenerate schema.d.ts for deployment rpd and tpd fields * test(types): add rpd and tpd to the owned litellm_params inventory * fix(router): run daily rpd/tpd checks after the minute checks and increment rpm and rpd in one cache trip * fix(router): flatten the async rpm reject branch for the ruff strict gate * fix(router): hand rpm and rpd back on any rate limit reject A request the rpm, rpd or tpd check rejects no longer keeps its request counts, so retries against a deployment whose rpd is spent cannot fill its rpm and block the in-group fallover with the minute 429. * test(router): a minute-rejected request no longer keeps its rpm count --------- Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- litellm/caching/in_memory_cache.py | 4 +- .../litellm_core_utils/get_litellm_params.py | 2 + .../pre_call_checks/model_rate_limit_check.py | 228 +++++++++++--- litellm/types/litellm_params.py | 2 + litellm/types/router.py | 4 + ..._enforce_model_rate_limits_daily_limits.py | 96 ++++++ tests/unit/caching/test_in_memory_cache.py | 25 +- .../test_enforce_model_rate_limits.py | 286 +++++++++++++++++- .../test_router/test_io_token_rate_limits.py | 2 +- tests/unit/types/test_litellm_params.py | 2 + ui/litellm-dashboard/src/lib/http/schema.d.ts | 8 + 11 files changed, 589 insertions(+), 70 deletions(-) create mode 100644 tests/integration/routing/test_enforce_model_rate_limits_daily_limits.py diff --git a/litellm/caching/in_memory_cache.py b/litellm/caching/in_memory_cache.py index 1cc3e106aa0..6b40c48c4ef 100644 --- a/litellm/caching/in_memory_cache.py +++ b/litellm/caching/in_memory_cache.py @@ -274,7 +274,9 @@ class InMemoryCache(BaseCache): ) -> list[float] | None: results: Final = [] for increment in increment_list: - result = await self.async_increment(increment["key"], increment["increment_value"], **kwargs) + result = await self.async_increment( + increment["key"], increment["increment_value"], ttl=increment["ttl"], **kwargs + ) results.append(result) return results diff --git a/litellm/litellm_core_utils/get_litellm_params.py b/litellm/litellm_core_utils/get_litellm_params.py index ba964d200ec..8178b7e98b8 100644 --- a/litellm/litellm_core_utils/get_litellm_params.py +++ b/litellm/litellm_core_utils/get_litellm_params.py @@ -75,6 +75,8 @@ OPTIONAL_KWARGS_KEYS: Final = ( "rpm", "itpm", "otpm", + "tpd", + "rpd", "use_xai_oauth", "chatgpt_auth_file", "github_copilot_auth_type", diff --git a/litellm/router_utils/pre_call_checks/model_rate_limit_check.py b/litellm/router_utils/pre_call_checks/model_rate_limit_check.py index ea3fdd601dc..9eebcf39515 100644 --- a/litellm/router_utils/pre_call_checks/model_rate_limit_check.py +++ b/litellm/router_utils/pre_call_checks/model_rate_limit_check.py @@ -5,15 +5,19 @@ When enabled via router_settings.optional_pre_call_checks: ["enforce_model_rate_ - tpm/rpm: combined TPM + optional RPM (legacy) - itpm/otpm: separate input/output tokens per minute +- tpd/rpd: tokens and requests per UTC calendar day When a deployment sets both itpm/otpm and tpm/rpm, both are enforced. A warning is logged the first time such a deployment is seen. """ import contextlib -from typing import TYPE_CHECKING, Any, Final +from collections.abc import Callable +from datetime import datetime, timedelta +from typing import TYPE_CHECKING, Any, Final, Literal import httpx +from pydantic import BaseModel, ConfigDict import litellm from litellm._internal_context import with_service_target @@ -33,6 +37,7 @@ from litellm.router_utils.pre_call_checks.io_token_rate_limit_check import ( io_token_refund_failure, ) from litellm.router_utils.routing_read_batch import ROUTER_USAGE_TARGET +from litellm.types.caching import RedisPipelineIncrementOperation from litellm.types.router import RouterErrors from litellm.types.utils import StandardLoggingPayload from litellm.utils import get_utc_datetime @@ -47,6 +52,62 @@ else: class RoutingArgs: ttl: int = 60 # 1min (RPM/TPM expire key) + daily_ttl: int = 86400 + + +DAY_FORMAT: Final = "%Y-%m-%d" + + +def seconds_until_utc_midnight(now: datetime) -> int: + next_midnight: Final = (now + timedelta(days=1)).replace(hour=0, minute=0, second=0, microsecond=0) + return max(1, int((next_midnight - now).total_seconds())) + + +def _first_set(*values: int | None) -> int | None: + return next((value for value in values if value is not None), None) + + +class _DailyLimitFields(BaseModel): + model_config = ConfigDict(frozen=True, extra="ignore", from_attributes=True) + tpd: int | None = None + rpd: int | None = None + + +class _DailyLimitParams(_DailyLimitFields): + model: str | None = None + + +class _DailyLimitModelInfo(_DailyLimitFields): + id: str | None = None + + +class _DailyLimitDeployment(_DailyLimitFields): + model_name: str = "" + litellm_params: _DailyLimitParams = _DailyLimitParams() + model_info: _DailyLimitModelInfo = _DailyLimitModelInfo() + + def limits(self) -> tuple[int | None, int | None]: + return ( + _first_set(self.tpd, self.litellm_params.tpd, self.model_info.tpd), + _first_set(self.rpd, self.litellm_params.rpd, self.model_info.rpd), + ) + + def cache_keys(self, current_day: str) -> tuple[str, str]: + prefix: Final = f"{self.model_info.id}:{self.litellm_params.model}" + return f"{prefix}:tpd:{current_day}", f"{prefix}:rpd:{current_day}" + + def limit_error(self, kind: Literal["tpd", "rpd"], limit: int, usage: int, now: datetime) -> litellm.RateLimitError: + return litellm.RateLimitError( + message=f"Model rate limit exceeded. {kind.upper()} limit={limit}, current usage={usage}", + llm_provider="", + model=self.litellm_params.model, + response=httpx.Response( + status_code=429, + content=f"{RouterErrors.user_defined_ratelimit_error.value} {kind} limit={limit}. current usage={usage}. id={self.model_info.id}, model_group={self.model_name}", + headers={"retry-after": str(seconds_until_utc_midnight(now))}, + request=httpx.Request(method="model_rate_limit_check", url="https://github.com/BerriAI/litellm"), + ), + ) class ModelRateLimitingCheck(CustomLogger): @@ -60,12 +121,60 @@ class ModelRateLimitingCheck(CustomLogger): this check actively enforces those limits across ALL routing strategies. """ - def __init__(self, dual_cache: DualCache): + def __init__(self, dual_cache: DualCache, clock: Callable[[], datetime] = get_utc_datetime): self.dual_cache = dual_cache + self.clock = clock # model_ids already warned about itpm/otpm + tpm/rpm on the same deployment, # so the warning is logged once per deployment rather than per request. self._io_token_conflict_warned_ids: set[str] = set() + def _increment(self, key: str, value: int, ttl: int) -> int: + return self.dual_cache.increment_cache(key=key, value=value, ttl=ttl) + + async def _async_increment( + self, key: str, value: int, ttl: int, parent_otel_span: Span | None = None + ) -> int | None: + count: Final = await self.dual_cache.async_increment_cache( + key=key, value=value, ttl=ttl, parent_otel_span=parent_otel_span + ) + return None if count is None else int(count) + + @staticmethod + def _request_count_operations( + rpm_key: str | None, rpd_key: str | None, increment_value: int + ) -> tuple[RedisPipelineIncrementOperation, ...]: + return tuple( + RedisPipelineIncrementOperation(key=key, increment_value=increment_value, ttl=ttl) + for key, ttl in ((rpm_key, RoutingArgs.ttl), (rpd_key, RoutingArgs.daily_ttl)) + if key is not None + ) + + async def _async_refund_request_counts( + self, rpm_key: str | None, rpd_key: str | None, parent_otel_span: Span | None + ) -> None: + operations: Final = self._request_count_operations(rpm_key, rpd_key, -1) + if operations: + await self.dual_cache.async_increment_cache_pipeline( + increment_list=list(operations), parent_otel_span=parent_otel_span + ) + + async def _async_increment_request_counts( + self, rpm_key: str | None, rpd_key: str | None, parent_otel_span: Span | None + ) -> tuple[int | None, int | None]: + operations: Final = self._request_count_operations(rpm_key, rpd_key, 1) + if not operations: + return None, None + counts: Final = await self.dual_cache.async_increment_cache_pipeline( + increment_list=list(operations), parent_otel_span=parent_otel_span + ) + if counts is None or len(counts) != len(operations): + return None, None + count_by_key: Final = {op["key"]: int(count) for op, count in zip(operations, counts)} + return ( + None if rpm_key is None else count_by_key.get(rpm_key), + None if rpd_key is None else count_by_key.get(rpd_key), + ) + def _warn_io_token_and_tpm_rpm_coexist_once(self, deployment: dict) -> None: tpm_limit, rpm_limit = self._get_deployment_limits(deployment) if tpm_limit is None and rpm_limit is None: @@ -129,6 +238,21 @@ class ModelRateLimitingCheck(CustomLogger): return tpm, rpm + def _check_daily_limits(self, deployment: _DailyLimitDeployment, now: datetime) -> None: + tpd_limit, rpd_limit = deployment.limits() + if tpd_limit is None and rpd_limit is None: + return + tpd_key, rpd_key = deployment.cache_keys(now.strftime(DAY_FORMAT)) + if tpd_limit is not None: + current_tpd: Final = self._get_current_tpm(tpd_key, tpd_limit) + if current_tpd is not None and current_tpd >= tpd_limit: + raise deployment.limit_error("tpd", tpd_limit, current_tpd, now) + if rpd_limit is not None: + current_rpd: Final = self._increment(rpd_key, 1, RoutingArgs.daily_ttl) + if current_rpd > rpd_limit: + self._increment(rpd_key, -1, RoutingArgs.daily_ttl) + raise deployment.limit_error("rpd", rpd_limit, current_rpd, now) + def _get_cache_keys(self, deployment: dict, current_minute: str) -> tuple[str, str]: """Get the cache keys for TPM and RPM tracking.""" model_id: Final = deployment.get("model_info", {}).get("id") @@ -178,13 +302,10 @@ class ModelRateLimitingCheck(CustomLogger): ) io_reservation_made = True + dt: Final = self.clock() + daily: Final = _DailyLimitDeployment.model_validate(deployment) tpm_limit, rpm_limit = self._get_deployment_limits(deployment) - # If no limits are set, allow the request - if tpm_limit is None and rpm_limit is None: - return deployment - - dt: Final = get_utc_datetime() current_minute: Final = dt.strftime("%H-%M") tpm_key, rpm_key = self._get_cache_keys(deployment, current_minute) @@ -213,8 +334,9 @@ class ModelRateLimitingCheck(CustomLogger): # Check RPM limit (atomic increment-first to avoid race conditions) if rpm_limit is not None: - current_rpm: Final = self.dual_cache.increment_cache(key=rpm_key, value=1, ttl=RoutingArgs.ttl) + current_rpm: Final = self._increment(rpm_key, 1, RoutingArgs.ttl) if current_rpm is not None and current_rpm > rpm_limit: + self._increment(rpm_key, -1, RoutingArgs.ttl) raise litellm.RateLimitError( message=f"Model rate limit exceeded. RPM limit={rpm_limit}, current usage={current_rpm}", llm_provider="", @@ -230,6 +352,12 @@ class ModelRateLimitingCheck(CustomLogger): ), ) + try: + self._check_daily_limits(daily, dt) + except litellm.RateLimitError: + if rpm_limit is not None: + self._increment(rpm_key, -1, RoutingArgs.ttl) + raise return deployment except litellm.RateLimitError: @@ -259,15 +387,14 @@ class ModelRateLimitingCheck(CustomLogger): ) io_reservation_made = True + dt: Final = self.clock() + daily: Final = _DailyLimitDeployment.model_validate(deployment) + tpd_limit, rpd_limit = daily.limits() tpm_limit, rpm_limit = self._get_deployment_limits(deployment) - # If no limits are set, allow the request - if tpm_limit is None and rpm_limit is None: - return deployment - - dt: Final = get_utc_datetime() current_minute: Final = dt.strftime("%H-%M") tpm_key, rpm_key = self._get_cache_keys(deployment, current_minute) + tpd_key, rpd_key = daily.cache_keys(dt.strftime(DAY_FORMAT)) model_id: Final = deployment.get("model_info", {}).get("id") model_name: Final = deployment.get("litellm_params", {}).get("model") @@ -293,30 +420,37 @@ class ModelRateLimitingCheck(CustomLogger): num_retries=0, # Don't retry - return 429 immediately ) - # Check RPM limit (atomic increment-first to avoid race conditions) - if rpm_limit is not None: - current_rpm: Final = await self.dual_cache.async_increment_cache( - key=rpm_key, - value=1, - ttl=RoutingArgs.ttl, - parent_otel_span=parent_otel_span, - ) - if current_rpm is not None and current_rpm > rpm_limit: - raise litellm.RateLimitError( - message=f"Model rate limit exceeded. RPM limit={rpm_limit}, current usage={current_rpm}", - llm_provider="", - model=model_name, - response=httpx.Response( - status_code=429, - content=f"{RouterErrors.user_defined_ratelimit_error.value} rpm limit={rpm_limit}. current usage={current_rpm}. id={model_id}, model_group={model_group}", - headers={"retry-after": str(60)}, - request=httpx.Request( - method="model_rate_limit_check", - url="https://github.com/BerriAI/litellm", - ), + if tpd_limit is not None: + current_tpd: Final = await self._async_get_current_tpm(tpd_key, tpd_limit, parent_otel_span) + if current_tpd is not None and current_tpd >= tpd_limit: + raise daily.limit_error("tpd", tpd_limit, current_tpd, dt) + + charged_rpm_key: Final = rpm_key if rpm_limit is not None else None + charged_rpd_key: Final = rpd_key if rpd_limit is not None else None + current_rpm, current_rpd = await self._async_increment_request_counts( + charged_rpm_key, charged_rpd_key, parent_otel_span + ) + if rpm_limit is not None and current_rpm is not None and current_rpm > rpm_limit: + await self._async_refund_request_counts(charged_rpm_key, charged_rpd_key, parent_otel_span) + raise litellm.RateLimitError( + message=f"Model rate limit exceeded. RPM limit={rpm_limit}, current usage={current_rpm}", + llm_provider="", + model=model_name, + response=httpx.Response( + status_code=429, + content=f"{RouterErrors.user_defined_ratelimit_error.value} rpm limit={rpm_limit}. current usage={current_rpm}. id={model_id}, model_group={model_group}", + headers={"retry-after": str(60)}, + request=httpx.Request( + method="model_rate_limit_check", + url="https://github.com/BerriAI/litellm", ), - num_retries=0, # Don't retry - return 429 immediately - ) + ), + num_retries=0, # Don't retry - return 429 immediately + ) + + if rpd_limit is not None and current_rpd is not None and current_rpd > rpd_limit: + await self._async_refund_request_counts(charged_rpm_key, charged_rpd_key, parent_otel_span) + raise daily.limit_error("rpd", rpd_limit, current_rpd, dt) return deployment @@ -373,17 +507,16 @@ class ModelRateLimitingCheck(CustomLogger): if not model or not total_tokens: return - dt: Final = get_utc_datetime() + dt: Final = self.clock() current_minute: Final = dt.strftime("%H-%M") tpm_key: Final = f"{model_id}:{model}:tpm:{current_minute}" + tpd_key: Final = f"{model_id}:{model}:tpd:{dt.strftime(DAY_FORMAT)}" verbose_router_logger.debug("[TPM TRACKING] Incrementing %s by %s", tpm_key, total_tokens) - await self.dual_cache.async_increment_cache( - key=tpm_key, - value=total_tokens, - ttl=RoutingArgs.ttl, - ) + tokens: Final = int(total_tokens) + await self._async_increment(tpm_key, tokens, RoutingArgs.ttl) + await self._async_increment(tpd_key, tokens, RoutingArgs.daily_ttl) except Exception as e: verbose_router_logger.debug("Error in ModelRateLimitingCheck.async_log_success_event: %s", e) @@ -434,15 +567,14 @@ class ModelRateLimitingCheck(CustomLogger): if not model or not total_tokens: return - dt: Final = get_utc_datetime() + dt: Final = self.clock() current_minute: Final = dt.strftime("%H-%M") tpm_key: Final = f"{model_id}:{model}:tpm:{current_minute}" + tpd_key: Final = f"{model_id}:{model}:tpd:{dt.strftime(DAY_FORMAT)}" - self.dual_cache.increment_cache( - key=tpm_key, - value=total_tokens, - ttl=RoutingArgs.ttl, - ) + tokens: Final = int(total_tokens) + self._increment(tpm_key, tokens, RoutingArgs.ttl) + self._increment(tpd_key, tokens, RoutingArgs.daily_ttl) except Exception as e: verbose_router_logger.debug("Error in ModelRateLimitingCheck.log_success_event: %s", e) diff --git a/litellm/types/litellm_params.py b/litellm/types/litellm_params.py index 717b74e5741..9196b7a226f 100644 --- a/litellm/types/litellm_params.py +++ b/litellm/types/litellm_params.py @@ -189,6 +189,8 @@ class DeploymentOptions: tpm: int | None = None itpm: int | None = None otpm: int | None = None + rpd: int | None = None + tpd: int | None = None kubernetes_pod_discovery: bool | None = None default_api_key_rpm_limit: int | None = None default_api_key_tpm_limit: int | None = None diff --git a/litellm/types/router.py b/litellm/types/router.py index ba9b82e2a57..f8de2128a80 100644 --- a/litellm/types/router.py +++ b/litellm/types/router.py @@ -533,6 +533,8 @@ class GenericLiteLLMParams(CredentialLiteLLMParams, CustomPricingLiteLLMParams): rpm: int | None = None itpm: int | None = None otpm: int | None = None + tpd: int | None = None + rpd: int | None = None timeout: float | str | httpx.Timeout | SDKTimeout | None = None # if str, pass in as os.environ/ # read timeout of a live stream, a faked stream takes the longer of this and timeout. If str, pass in as os.environ/ stream_timeout: float | str | None = None @@ -738,6 +740,8 @@ class LiteLLMParamsTypedDict(TypedDict, total=False): rpm: int | None itpm: int | None otpm: int | None + tpd: ReadOnly[int | None] + rpd: ReadOnly[int | None] order: int | None weight: int | None max_parallel_requests: int | None diff --git a/tests/integration/routing/test_enforce_model_rate_limits_daily_limits.py b/tests/integration/routing/test_enforce_model_rate_limits_daily_limits.py new file mode 100644 index 00000000000..c670171a149 --- /dev/null +++ b/tests/integration/routing/test_enforce_model_rate_limits_daily_limits.py @@ -0,0 +1,96 @@ +from __future__ import annotations + +import uuid +from collections.abc import Iterator +from dataclasses import dataclass +from pathlib import Path +from typing import Final + +import httpx +import pytest +from integration._support.client import gateway_from_environment +from integration._support.openai_wire import chat_reply +from integration._support.process import OwnedProxy, graceful_stop_seconds, owned_proxy_process +from integration._support.wire import Reply, Request, Wire, wire_server +from integration.providers._cache_control_marks_support import owned_config +from integration.routing.test_payment_required_mapping import OPENAI_MODEL, is_model_info_probe, model_list_reply +from pydantic import JsonValue + +pytestmark: Final = pytest.mark.timeout(2 * graceful_stop_seconds() + 120) + +GROUP: Final = "daily-limited-group" +LIMITED_ID: Final = "free-tier-two-a-day" +UNLIMITED_ID: Final = "paid-tier" +RPD: Final = 2 +EXTRA_REQUESTS: Final = 2 + + +@dataclass(frozen=True, slots=True) +class Rig: + owned: OwnedProxy + limited: Wire + unlimited: Wire + + +def _reply(request: Request) -> Reply: + if is_model_info_probe(request): + return model_list_reply() + return chat_reply("chatcmpl-daily", OPENAI_MODEL, "served", stream=False) + + +def _deployment(model_id: str, wire: Wire, **params: JsonValue) -> dict[str, JsonValue]: + return { + "model_name": GROUP, + "litellm_params": { + "model": OPENAI_MODEL, + "api_base": wire.url + "/v1", + "api_key": "synthetic-openai-key", + **params, + }, + "model_info": {"id": model_id}, + } + + +def _config(directory: Path, limited: Wire, unlimited: Wire) -> Path: + return owned_config( + directory, + [_deployment(LIMITED_ID, limited, rpd=RPD, order=1), _deployment(UNLIMITED_ID, unlimited, order=2)], + litellm_settings={"cache": False}, + router_settings={"optional_pre_call_checks": ["enforce_model_rate_limits"], "num_retries": 1}, + ) + + +@pytest.fixture(scope="module") +def rig(tmp_path_factory: pytest.TempPathFactory) -> Iterator[Rig]: + directory: Final = tmp_path_factory.mktemp("daily-limits") + with ( + gateway_from_environment() as gateway, + wire_server(_reply) as limited, + wire_server(_reply) as unlimited, + owned_proxy_process(gateway, directory, {}, config=_config(directory, limited, unlimited)) as owned, + ): + yield Rig(owned, limited, unlimited) + + +def _chat(rig: Rig) -> httpx.Response: + return rig.owned.gateway.request( + "POST", + "/v1/chat/completions", + {"model": GROUP, "messages": [{"role": "user", "content": f"daily {uuid.uuid4().hex}"}]}, + ) + + +def _forwarded(wire: Wire) -> int: + return sum(1 for request in wire.drain() if not is_model_info_probe(request)) + + +def test_requests_past_the_deployment_rpd_are_served_by_the_other_deployment_in_the_group(rig: Rig) -> None: + responses: Final = tuple(_chat(rig) for _ in range(RPD + EXTRA_REQUESTS)) + + assert [response.status_code for response in responses] == [200] * (RPD + EXTRA_REQUESTS), [ + response.text for response in responses + ] + assert [response.headers["x-litellm-model-id"] for response in responses] == [LIMITED_ID] * RPD + [ + UNLIMITED_ID + ] * EXTRA_REQUESTS + assert (_forwarded(rig.limited), _forwarded(rig.unlimited)) == (RPD, EXTRA_REQUESTS) diff --git a/tests/unit/caching/test_in_memory_cache.py b/tests/unit/caching/test_in_memory_cache.py index 75232476ecd..901832937b5 100644 --- a/tests/unit/caching/test_in_memory_cache.py +++ b/tests/unit/caching/test_in_memory_cache.py @@ -235,13 +235,9 @@ def test_in_memory_cache_eviction_order(): # Add items with different TTLs now = time.time() - in_memory_cache.set_cache( - key="early_expire", value="value_1", ttl=100 - ) # expires in 100 seconds + in_memory_cache.set_cache(key="early_expire", value="value_1", ttl=100) # expires in 100 seconds time.sleep(0.01) - in_memory_cache.set_cache( - key="late_expire", value="value_2", ttl=200 - ) # expires in 200 seconds + in_memory_cache.set_cache(key="late_expire", value="value_2", ttl=200) # expires in 200 seconds # Verify TTL order early_ttl = in_memory_cache.ttl_dict["early_expire"] @@ -379,3 +375,20 @@ def test_in_memory_cache_injected_clock_controls_expiry_and_eviction() -> None: assert cache.get_cache("second") is None assert cache.get_cache("third") == "replacement" assert cache.get_cache("fourth") == "new" + + +@pytest.mark.asyncio +async def test_async_increment_pipeline_keeps_each_key_for_its_own_ttl() -> None: + now: Final = {"t": 1_000.0} + cache: Final = InMemoryCache(default_ttl=600, clock=lambda: now["t"]) + + await cache.async_increment_pipeline( + increment_list=[ + {"key": "minute", "increment_value": 1, "ttl": 60}, + {"key": "day", "increment_value": 1, "ttl": 86400}, + ] + ) + now["t"] += 601 + + assert cache.get_cache("minute") is None + assert cache.get_cache("day") == 1 diff --git a/tests/unit/test_router/test_enforce_model_rate_limits.py b/tests/unit/test_router/test_enforce_model_rate_limits.py index 7577064b7f9..2a260fdf0b8 100644 --- a/tests/unit/test_router/test_enforce_model_rate_limits.py +++ b/tests/unit/test_router/test_enforce_model_rate_limits.py @@ -6,7 +6,7 @@ regardless of the routing strategy being used. """ import asyncio -from datetime import timedelta +from datetime import datetime, timedelta, timezone from unittest.mock import AsyncMock, MagicMock import pytest @@ -222,11 +222,9 @@ class TestModelRateLimitingCheck: check.log_success_event(kwargs, None, None, None) - # Verify increment_cache was called - mock_cache.increment_cache.assert_called_once() - _, kwarg_params = mock_cache.increment_cache.call_args - assert "test-id:gpt-4:tpm:" in kwarg_params["key"] - assert kwarg_params["value"] == 50 + keys = tuple(call.kwargs["key"] for call in mock_cache.increment_cache.call_args_list) + assert any(key.startswith("test-id:gpt-4:tpm:") for key in keys), keys + assert all(call.kwargs["value"] == 50 for call in mock_cache.increment_cache.call_args_list) class TestModelRateLimitingCheckAsync: @@ -253,7 +251,7 @@ class TestModelRateLimitingCheckAsync: """Test that RateLimitError is raised when RPM limit is exceeded (async).""" mock_cache = MagicMock() mock_cache.async_get_cache = AsyncMock(return_value=None) - mock_cache.async_increment_cache = AsyncMock(return_value=11) # Over limit + mock_cache.async_increment_cache_pipeline = AsyncMock(return_value=[11.0]) # Over limit check = ModelRateLimitingCheck(dual_cache=mock_cache) @@ -274,7 +272,7 @@ class TestModelRateLimitingCheckAsync: """Test that requests are allowed when under the limit (async).""" mock_cache = MagicMock() mock_cache.async_get_cache = AsyncMock(return_value=None) - mock_cache.async_increment_cache = AsyncMock(return_value=6) + mock_cache.async_increment_cache_pipeline = AsyncMock(return_value=[6.0]) check = ModelRateLimitingCheck(dual_cache=mock_cache) @@ -339,7 +337,7 @@ class TestModelRateLimitingCheckAsync: ): redis_cache = MagicMock() redis_cache.async_get_cache = AsyncMock(side_effect=RedisCircuitBreakerOpenError()) - redis_cache.async_increment = AsyncMock(return_value=2) + redis_cache.async_increment_pipeline = AsyncMock(return_value=[2.0]) check = ModelRateLimitingCheck(dual_cache=_dual_cache_with_local_tpm(5, redis_cache)) with pytest.raises(litellm.RateLimitError) as exc_info: @@ -375,11 +373,9 @@ class TestModelRateLimitingCheckAsync: await check.async_log_success_event(kwargs, None, None, None) - # Verify async_increment_cache was called - mock_cache.async_increment_cache.assert_called_once() - _, kwarg_params = mock_cache.async_increment_cache.call_args - assert "test-id:gpt-4:tpm:" in kwarg_params["key"] - assert kwarg_params["value"] == 50 + keys = tuple(call.kwargs["key"] for call in mock_cache.async_increment_cache.call_args_list) + assert any(key.startswith("test-id:gpt-4:tpm:") for key in keys), keys + assert all(call.kwargs["value"] == 50 for call in mock_cache.async_increment_cache.call_args_list) class TestRouterWithEnforceModelRateLimits: @@ -466,3 +462,265 @@ class TestModelRateLimitConcurrency: assert len(successes) == 2, f"Expected 2 successes, got {len(successes)}" assert len(failures) == 2, f"Expected 2 rate limit errors, got {len(failures)}" + + +DAY_ONE = datetime(2026, 10, 28, 23, 59, 30, tzinfo=timezone.utc) +RPD_DEPLOYMENT = { + "litellm_params": {"model": "gpt-4.1-mini", "rpd": 2}, + "model_info": {"id": "daily-test-id"}, + "model_name": "daily-model", +} +RPM_AND_RPD_DEPLOYMENT = {**RPD_DEPLOYMENT, "litellm_params": {"model": "gpt-4.1-mini", "rpm": 1, "rpd": 2}} + + +class Clock: + def __init__(self, now: datetime): + self.now = now + + def __call__(self) -> datetime: + return self.now + + +def _pre_call(check: ModelRateLimitingCheck, deployment: dict[str, object]) -> object: + return check.pre_call_check(deployment) + + +async def _async_pre_call(check: ModelRateLimitingCheck, deployment: dict[str, object]) -> object: + return await check.async_pre_call_check(deployment) + + +async def _async_log_success(check: ModelRateLimitingCheck, kwargs: dict[str, object]) -> None: + await check.async_log_success_event(kwargs, None, None, None) + + +class TestDailyLimits: + def test_pre_call_check_rejects_the_request_over_rpd(self): + check = ModelRateLimitingCheck(dual_cache=DualCache(), clock=Clock(DAY_ONE)) + + assert _pre_call(check, RPD_DEPLOYMENT) == RPD_DEPLOYMENT + assert _pre_call(check, RPD_DEPLOYMENT) == RPD_DEPLOYMENT + with pytest.raises(litellm.RateLimitError) as exc_info: + _pre_call(check, RPD_DEPLOYMENT) + + assert "RPD limit=2, current usage=3" in str(exc_info.value) + assert "rpd limit=2. current usage=3. id=daily-test-id, model_group=daily-model" in exc_info.value.response.text + + @pytest.mark.asyncio + async def test_async_pre_call_check_rejects_the_request_over_rpd_and_lets_the_router_retry(self): + check = ModelRateLimitingCheck(dual_cache=DualCache(), clock=Clock(DAY_ONE)) + + assert await _async_pre_call(check, RPD_DEPLOYMENT) == RPD_DEPLOYMENT + assert await _async_pre_call(check, RPD_DEPLOYMENT) == RPD_DEPLOYMENT + with pytest.raises(litellm.RateLimitError) as exc_info: + await _async_pre_call(check, RPD_DEPLOYMENT) + + assert "RPD limit=2" in str(exc_info.value) + assert exc_info.value.num_retries is None + + @pytest.mark.asyncio + async def test_async_pre_call_check_rejects_once_the_day_tokens_reach_tpd(self): + deployment = {**RPD_DEPLOYMENT, "litellm_params": {"model": "gpt-4.1-mini", "tpd": 100}} + check = ModelRateLimitingCheck(dual_cache=DualCache(), clock=Clock(DAY_ONE)) + kwargs = { + "standard_logging_object": { + "model_id": "daily-test-id", + "total_tokens": 60, + "hidden_params": {"litellm_model_name": "gpt-4.1-mini"}, + } + } + + await _async_log_success(check, kwargs) + assert await _async_pre_call(check, deployment) == deployment + await _async_log_success(check, kwargs) + with pytest.raises(litellm.RateLimitError) as exc_info: + await _async_pre_call(check, deployment) + + assert "TPD limit=100, current usage=120" in str(exc_info.value) + + def test_pre_call_check_minute_rejected_request_does_not_spend_rpd(self): + clock = Clock(DAY_ONE) + check = ModelRateLimitingCheck(dual_cache=DualCache(), clock=clock) + + assert _pre_call(check, RPM_AND_RPD_DEPLOYMENT) == RPM_AND_RPD_DEPLOYMENT + with pytest.raises(litellm.RateLimitError, match="RPM limit=1"): + _pre_call(check, RPM_AND_RPD_DEPLOYMENT) + clock.now = DAY_ONE - timedelta(minutes=1) + + assert _pre_call(check, RPM_AND_RPD_DEPLOYMENT) == RPM_AND_RPD_DEPLOYMENT + clock.now = DAY_ONE - timedelta(minutes=2) + with pytest.raises(litellm.RateLimitError, match="RPD limit=2, current usage=3"): + _pre_call(check, RPM_AND_RPD_DEPLOYMENT) + + @pytest.mark.asyncio + async def test_async_pre_call_check_minute_rejected_request_does_not_spend_rpd(self): + clock = Clock(DAY_ONE) + check = ModelRateLimitingCheck(dual_cache=DualCache(), clock=clock) + + assert await _async_pre_call(check, RPM_AND_RPD_DEPLOYMENT) == RPM_AND_RPD_DEPLOYMENT + with pytest.raises(litellm.RateLimitError, match="RPM limit=1"): + await _async_pre_call(check, RPM_AND_RPD_DEPLOYMENT) + clock.now = DAY_ONE - timedelta(minutes=1) + + assert await _async_pre_call(check, RPM_AND_RPD_DEPLOYMENT) == RPM_AND_RPD_DEPLOYMENT + clock.now = DAY_ONE - timedelta(minutes=2) + with pytest.raises(litellm.RateLimitError, match="RPD limit=2, current usage=3"): + await _async_pre_call(check, RPM_AND_RPD_DEPLOYMENT) + + def test_pre_call_check_rejected_requests_hand_back_rpm_and_rpd(self): + clock = Clock(DAY_ONE) + check = ModelRateLimitingCheck(dual_cache=DualCache(), clock=clock) + + assert _pre_call(check, RPM_AND_RPD_DEPLOYMENT) == RPM_AND_RPD_DEPLOYMENT + with pytest.raises(litellm.RateLimitError, match="RPM limit=1, current usage=2"): + _pre_call(check, RPM_AND_RPD_DEPLOYMENT) + with pytest.raises(litellm.RateLimitError, match="RPM limit=1, current usage=2"): + _pre_call(check, RPM_AND_RPD_DEPLOYMENT) + clock.now = DAY_ONE - timedelta(minutes=1) + assert _pre_call(check, RPM_AND_RPD_DEPLOYMENT) == RPM_AND_RPD_DEPLOYMENT + + clock.now = DAY_ONE - timedelta(minutes=2) + with pytest.raises(litellm.RateLimitError, match="RPD limit=2, current usage=3"): + _pre_call(check, RPM_AND_RPD_DEPLOYMENT) + with pytest.raises(litellm.RateLimitError, match="RPD limit=2, current usage=3"): + _pre_call(check, RPM_AND_RPD_DEPLOYMENT) + + @pytest.mark.asyncio + async def test_async_pre_call_check_rejected_requests_hand_back_rpm_and_rpd(self): + clock = Clock(DAY_ONE) + check = ModelRateLimitingCheck(dual_cache=DualCache(), clock=clock) + + assert await _async_pre_call(check, RPM_AND_RPD_DEPLOYMENT) == RPM_AND_RPD_DEPLOYMENT + with pytest.raises(litellm.RateLimitError, match="RPM limit=1, current usage=2"): + await _async_pre_call(check, RPM_AND_RPD_DEPLOYMENT) + with pytest.raises(litellm.RateLimitError, match="RPM limit=1, current usage=2"): + await _async_pre_call(check, RPM_AND_RPD_DEPLOYMENT) + clock.now = DAY_ONE - timedelta(minutes=1) + assert await _async_pre_call(check, RPM_AND_RPD_DEPLOYMENT) == RPM_AND_RPD_DEPLOYMENT + + clock.now = DAY_ONE - timedelta(minutes=2) + with pytest.raises(litellm.RateLimitError, match="RPD limit=2, current usage=3"): + await _async_pre_call(check, RPM_AND_RPD_DEPLOYMENT) + with pytest.raises(litellm.RateLimitError, match="RPD limit=2, current usage=3"): + await _async_pre_call(check, RPM_AND_RPD_DEPLOYMENT) + + @pytest.mark.asyncio + async def test_async_pre_call_check_refunds_rpm_and_rpd_in_one_cache_trip(self): + mock_cache = MagicMock() + mock_cache.async_get_cache = AsyncMock(return_value=None) + mock_cache.async_increment_cache_pipeline = AsyncMock(side_effect=[[1.0, 3.0], [0.0, 2.0]]) + check = ModelRateLimitingCheck(dual_cache=mock_cache, clock=Clock(DAY_ONE)) + + with pytest.raises(litellm.RateLimitError, match="RPD limit=2, current usage=3"): + await _async_pre_call(check, RPM_AND_RPD_DEPLOYMENT) + + assert mock_cache.async_increment_cache_pipeline.await_count == 2 + refund = mock_cache.async_increment_cache_pipeline.call_args_list[1].kwargs["increment_list"] + assert [(op["key"], op["increment_value"], op["ttl"]) for op in refund] == [ + ("daily-test-id:gpt-4.1-mini:rpm:23-59", -1, 60), + ("daily-test-id:gpt-4.1-mini:rpd:2026-10-28", -1, 86400), + ] + + @pytest.mark.asyncio + async def test_async_pre_call_check_increments_rpm_and_rpd_in_one_cache_trip(self): + mock_cache = MagicMock() + mock_cache.async_get_cache = AsyncMock(return_value=None) + mock_cache.async_increment_cache_pipeline = AsyncMock(return_value=[1.0, 1.0]) + check = ModelRateLimitingCheck(dual_cache=mock_cache, clock=Clock(DAY_ONE)) + + assert await _async_pre_call(check, RPM_AND_RPD_DEPLOYMENT) == RPM_AND_RPD_DEPLOYMENT + + assert mock_cache.async_increment_cache_pipeline.await_count == 1 + operations = mock_cache.async_increment_cache_pipeline.call_args.kwargs["increment_list"] + assert [(op["key"], op["ttl"]) for op in operations] == [ + ("daily-test-id:gpt-4.1-mini:rpm:23-59", 60), + ("daily-test-id:gpt-4.1-mini:rpd:2026-10-28", 86400), + ] + assert not mock_cache.async_increment_cache.called + + def test_pre_call_check_reads_the_shared_tpd_count_from_redis(self): + redis_cache = MagicMock() + redis_cache.get_cache.return_value = 100 + deployment = {**RPD_DEPLOYMENT, "litellm_params": {"model": "gpt-4.1-mini", "tpd": 100}} + check = ModelRateLimitingCheck(dual_cache=DualCache(redis_cache=redis_cache), clock=Clock(DAY_ONE)) + + with pytest.raises(litellm.RateLimitError) as exc_info: + _pre_call(check, deployment) + + assert "TPD limit=100, current usage=100" in str(exc_info.value) + assert redis_cache.get_cache.call_args.kwargs["key"] == "daily-test-id:gpt-4.1-mini:tpd:2026-10-28" + + def test_rpd_count_starts_fresh_after_utc_midnight(self): + clock = Clock(DAY_ONE) + check = ModelRateLimitingCheck(dual_cache=DualCache(), clock=clock) + _pre_call(check, RPD_DEPLOYMENT) + _pre_call(check, RPD_DEPLOYMENT) + with pytest.raises(litellm.RateLimitError): + _pre_call(check, RPD_DEPLOYMENT) + + clock.now = DAY_ONE + timedelta(seconds=30) + + assert _pre_call(check, RPD_DEPLOYMENT) == RPD_DEPLOYMENT + assert _pre_call(check, RPD_DEPLOYMENT) == RPD_DEPLOYMENT + with pytest.raises(litellm.RateLimitError) as exc_info: + _pre_call(check, RPD_DEPLOYMENT) + assert "current usage=3" in str(exc_info.value) + + @pytest.mark.parametrize( + ("now", "seconds_to_midnight"), + [ + (DAY_ONE, 30), + (datetime(2026, 10, 28, 0, 0, 0, tzinfo=timezone.utc), 86400), + (datetime(2026, 10, 28, 12, 30, 15, 500000, tzinfo=timezone.utc), 41384), + ], + ) + def test_daily_429_retry_after_is_seconds_until_utc_midnight(self, now: datetime, seconds_to_midnight: int): + deployment = {**RPD_DEPLOYMENT, "litellm_params": {"model": "gpt-4.1-mini", "rpd": 0}} + check = ModelRateLimitingCheck(dual_cache=DualCache(), clock=Clock(now)) + + with pytest.raises(litellm.RateLimitError) as exc_info: + _pre_call(check, deployment) + + assert exc_info.value.response.headers["retry-after"] == str(seconds_to_midnight) + + @pytest.mark.asyncio + async def test_router_serves_the_request_from_the_other_deployment_once_rpd_is_spent(self): + router = Router( + model_list=[ + { + "model_name": "daily-group", + "litellm_params": { + "model": "openai/gpt-4.1-mini", + "api_key": "k", + "mock_response": "a", + "rpd": 1, + "order": 1, + }, + "model_info": {"id": "spent-first"}, + }, + { + "model_name": "daily-group", + "litellm_params": { + "model": "openai/gpt-4.1-mini", + "api_key": "k", + "mock_response": "b", + "order": 2, + }, + "model_info": {"id": "always-free"}, + }, + ], + num_retries=1, + ) + check = ModelRateLimitingCheck(dual_cache=router.cache, clock=Clock(DAY_ONE)) + router.optional_callbacks = [check] + litellm.logging_callback_manager.add_litellm_callback(check) + try: + served = [ + (await router.acompletion(model="daily-group", messages=[{"role": "user", "content": "hi"}])) + .choices[0] + .message.content + for _ in range(4) + ] + finally: + router.discard() + + assert served == ["a", "b", "b", "b"] diff --git a/tests/unit/test_router/test_io_token_rate_limits.py b/tests/unit/test_router/test_io_token_rate_limits.py index a5a68271111..0ae22534278 100644 --- a/tests/unit/test_router/test_io_token_rate_limits.py +++ b/tests/unit/test_router/test_io_token_rate_limits.py @@ -859,7 +859,7 @@ class TestModelRateLimitingCheckIOTokens: with pytest.raises(litellm.RateLimitError): await check.async_pre_call_check(deployment) - assert await dual_cache.async_get_cache(key=rpm_key) == 6 + assert await dual_cache.async_get_cache(key=rpm_key) == 5, "the rejected request hands its rpm count back" assert (await dual_cache.async_get_cache(key=itpm_key) or 0) == 0 assert ITPM_RESERVED_KEY not in request_kwargs["metadata"] assert any("both limit types are enforced" in record.message for record in caplog.records) diff --git a/tests/unit/types/test_litellm_params.py b/tests/unit/types/test_litellm_params.py index 740342f54d9..cc1ea645012 100644 --- a/tests/unit/types/test_litellm_params.py +++ b/tests/unit/types/test_litellm_params.py @@ -154,6 +154,8 @@ OPTION_NAMES: Final = ( "model_info", "rpm", "tpm", + "rpd", + "tpd", "itpm", "otpm", "default_api_key_rpm_limit", diff --git a/ui/litellm-dashboard/src/lib/http/schema.d.ts b/ui/litellm-dashboard/src/lib/http/schema.d.ts index a09672d7d8f..a49f1099121 100644 --- a/ui/litellm-dashboard/src/lib/http/schema.d.ts +++ b/ui/litellm-dashboard/src/lib/http/schema.d.ts @@ -35647,6 +35647,8 @@ export interface components { regional_processing_uplift_multiplier_us?: number | null; /** Responses Api Version */ responses_api_version?: string | null; + /** Rpd */ + rpd?: number | null; /** Rpm */ rpm?: number | null; /** S3 Access Key Id */ @@ -35699,6 +35701,8 @@ export interface components { token_exchange_profile?: string | null; /** Token Exchange Scope */ token_exchange_scope?: string | null; + /** Tpd */ + tpd?: number | null; /** Tpm */ tpm?: number | null; /** Use Chat Completions Api */ @@ -51039,6 +51043,8 @@ export interface components { regional_processing_uplift_multiplier_us?: number | null; /** Responses Api Version */ responses_api_version?: string | null; + /** Rpd */ + rpd?: number | null; /** Rpm */ rpm?: number | null; /** S3 Access Key Id */ @@ -51091,6 +51097,8 @@ export interface components { token_exchange_profile?: string | null; /** Token Exchange Scope */ token_exchange_scope?: string | null; + /** Tpd */ + tpd?: number | null; /** Tpm */ tpm?: number | null; /** Use Chat Completions Api */ From 8e8f6dc66ed693f21f60eddce1939fd179621832 Mon Sep 17 00:00:00 2001 From: tessellark <349031000+tessellark@users.noreply.github.com> Date: Wed, 28 Oct 2026 15:53:00 -0700 Subject: [PATCH 21/41] feat(tencent): add Seedream image generation and editing through TokenHub (#48735) * feat(tencent): add Seedream image generation and editing through TokenHub * feat(tencent): price Seedream image generation and edits from TokenHub usage * test(tencent): cover Seedream image generation, edits, and pricing through TokenHub * fix(tencent): keep non-Seedream image models on the OpenAI-compatible route Only seedream- models use TokenHub's wand route, so Hy-Image and other tencent image models keep working as on main. The response now carries TokenHub's image size and format, Pro's flat fallback bills its default 2K size, and the new tests are typed. * fix(tencent): bill other Tencent image models as before and forward user and TokenHub params on edits * fix(tencent): keep dropped edit params out and parse form-string watermark and seed * fix(tencent): send extra_body edit fields to TokenHub as given * fix(tencent): keep the routed model on Seedream edits and bill a deployment's per-token price * fix(tencent): accept string drop params on edits, rebuild nested form fields, and honor a per-image deployment price An image edit with a non-list additional_drop_params answered 500 for every provider after this branch validated it as a list, so a non-list value now drops nothing again, as chat treats it. Seedream edits sent through the OpenAI SDK arrive as multipart form fields, so TokenHub-only booleans and integers are parsed and bracketed keys like optimize_prompt_options[mode] are rebuilt into objects. A deployment that sets only output_cost_per_image is billed that price, and the TokenHub URL uses the shared api_base helpers. --------- Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com> --- litellm/images/main.py | 57 ++++- litellm/images/utils.py | 40 ++- .../litellm_core_utils/llm_cost_calc/utils.py | 83 +++---- .../flux_lora_depth_transformation.py | 5 +- .../llms/fal_ai/image_edit/transformation.py | 44 +--- litellm/llms/tencent/cost_calculator.py | 47 +++- litellm/llms/tencent/image_edit/__init__.py | 0 .../llms/tencent/image_edit/transformation.py | 147 +++++++++++ .../llms/tencent/image_generation/__init__.py | 0 .../image_generation/transformation.py | 200 +++++++++++++++ ...odel_prices_and_context_window_backup.json | 33 +++ .../provider_endpoints_support_backup.json | 4 +- litellm/utils.py | 16 ++ model_prices_and_context_window.json | 33 +++ provider_endpoints_support.json | 4 +- tests/unit/images/test_main.py | 18 ++ .../unit/llms/tencent/image_edit/__init__.py | 0 .../test_tencent_image_edit_transformation.py | 232 ++++++++++++++++++ .../llms/tencent/image_generation/__init__.py | 0 ...tencent_image_generation_transformation.py | 215 ++++++++++++++++ .../tencent/test_tencent_cost_calculator.py | 128 ++++++++++ 21 files changed, 1204 insertions(+), 102 deletions(-) create mode 100644 litellm/llms/tencent/image_edit/__init__.py create mode 100644 litellm/llms/tencent/image_edit/transformation.py create mode 100644 litellm/llms/tencent/image_generation/__init__.py create mode 100644 litellm/llms/tencent/image_generation/transformation.py create mode 100644 tests/unit/llms/tencent/image_edit/__init__.py create mode 100644 tests/unit/llms/tencent/image_edit/test_tencent_image_edit_transformation.py create mode 100644 tests/unit/llms/tencent/image_generation/__init__.py create mode 100644 tests/unit/llms/tencent/image_generation/test_tencent_image_generation_transformation.py create mode 100644 tests/unit/llms/tencent/test_tencent_cost_calculator.py diff --git a/litellm/images/main.py b/litellm/images/main.py index 84156919774..d0618e5e804 100644 --- a/litellm/images/main.py +++ b/litellm/images/main.py @@ -1,9 +1,9 @@ import asyncio import contextvars import importlib -from collections.abc import Coroutine, Mapping +from collections.abc import Collection, Coroutine, Mapping from functools import partial -from typing import TYPE_CHECKING, Final, Literal, cast, overload +from typing import TYPE_CHECKING, Final, Literal, cast, get_type_hints, overload from pydantic import TypeAdapter, ValidationError @@ -61,6 +61,7 @@ from litellm.types.utils import ( ProviderRoutingHints, ) from litellm.utils import ( + EXTRA_BODY_ROUTING_KEYS, ImageResponse, ProviderConfigManager, get_llm_provider, @@ -466,7 +467,7 @@ def image_generation( litellm.LlmProviders.QWENCLOUD, litellm.LlmProviders.QWEN_AI_PLATFORM, litellm.LlmProviders.EDENAI, - ): + ) or (custom_llm_provider == litellm.LlmProviders.TENCENT and image_generation_config is not None): if image_generation_config is None: raise ValueError(f"image generation config is not supported for {custom_llm_provider}") @@ -807,6 +808,32 @@ def image_variation( return response +_IMAGE_EDIT_OPENAI_PARAMS: Final = frozenset(get_type_hints(ImageEditOptionalRequestParams)) +_DROP_PARAM_KWARGS: Final = frozenset({"drop_params", "additional_drop_params"}) +_ADDITIONAL_DROP_PARAMS: Final = TypeAdapter[tuple[str, ...]](tuple[str, ...]) + + +def _listed_drop_params(additional_drop_params: object) -> tuple[str, ...]: + try: + return _ADDITIONAL_DROP_PARAMS.validate_python(additional_drop_params) + except ValidationError: + return () + + +def _tokenhub_passthrough_edit_params( + non_default_params: Mapping[str, object], + extra_body: Mapping[str, object] | None, + mapped_params: Collection[str], + additional_drop_params: Collection[str] | None, +) -> Mapping[str, object]: + dropped: Final = frozenset(additional_drop_params or ()) + unmapped_excluded: Final = _DROP_PARAM_KWARGS | _IMAGE_EDIT_OPENAI_PARAMS | frozenset(mapped_params) | dropped + return { + **{key: value for key, value in non_default_params.items() if key not in unmapped_excluded}, + **{key: value for key, value in (extra_body or {}).items() if key not in dropped | EXTRA_BODY_ROUTING_KEYS}, + } + + @client def image_edit( image: FileTypes | list[FileTypes] | None = None, @@ -935,22 +962,24 @@ def image_edit( raise ValueError(f"image edit is not supported for {custom_llm_provider}") local_vars.update(kwargs) + provider_supported_params: Final[frozenset[str]] = frozenset( + image_edit_provider_config.get_supported_openai_params(model) + ) # Get ImageEditOptionalRequestParams with only valid parameters image_edit_optional_params: Final[ImageEditOptionalRequestParams] = ( _get_ImageEditRequestUtils().get_requested_image_edit_optional_param( local_vars, - provider_supported_params=frozenset( - image_edit_provider_config.get_supported_openai_params(model) - ).intersection(non_default_params), + provider_supported_params=provider_supported_params.intersection(non_default_params), ) ) # Get optional parameters for the responses API + additional_drop_params: Final = _listed_drop_params(kwargs.get("additional_drop_params")) image_edit_request_params: Final[dict] = _get_ImageEditRequestUtils().get_optional_params_image_edit( model=model, image_edit_provider_config=image_edit_provider_config, image_edit_optional_params=image_edit_optional_params, drop_params=kwargs.get("drop_params"), - additional_drop_params=kwargs.get("additional_drop_params"), + additional_drop_params=additional_drop_params, ) if image_edit_provider_config.use_multipart_form_data() and ( @@ -1039,7 +1068,19 @@ def image_edit( image=images, prompt=prompt, image_edit_provider_config=image_edit_provider_config, - image_edit_optional_request_params=image_edit_request_params, + image_edit_optional_request_params={ + **image_edit_request_params, + **( + _tokenhub_passthrough_edit_params( + non_default_params, + extra_body if isinstance(extra_body, dict) else None, + provider_supported_params, + additional_drop_params, + ) + if custom_llm_provider == litellm.LlmProviders.TENCENT.value + else {} + ), + }, custom_llm_provider=custom_llm_provider, litellm_params=litellm_params, logging_obj=litellm_logging_obj, diff --git a/litellm/images/utils.py b/litellm/images/utils.py index 24454954714..4e0cde24a68 100644 --- a/litellm/images/utils.py +++ b/litellm/images/utils.py @@ -1,6 +1,9 @@ +import base64 +import os from collections.abc import Collection, Mapping from io import BufferedReader, BytesIO -from typing import Any, Final, cast, get_type_hints +from pathlib import Path +from typing import Any, Final, Protocol, cast, get_type_hints, runtime_checkable import litellm from litellm.litellm_core_utils.token_counter import get_image_type @@ -16,7 +19,7 @@ class ImageEditRequestUtils: image_edit_provider_config: BaseImageEditConfig, image_edit_optional_params: ImageEditOptionalRequestParams, drop_params: bool | None = None, - additional_drop_params: list[str] | None = None, + additional_drop_params: Collection[str] | None = None, ) -> dict: """ Get optional parameters for the image edit API. @@ -145,3 +148,36 @@ class ImageEditRequestUtils: except Exception: # If anything goes wrong, default to PNG return FILE_MIME_TYPES[FileType.PNG] + + +@runtime_checkable +class _SeekableBinaryReader(Protocol): + def tell(self) -> int: ... + + def seek(self, offset: int) -> int: ... + + def read(self) -> bytes: ... + + +def _read_image_bytes(image: object) -> bytes: + if isinstance(image, bytes): + return image + if isinstance(image, tuple): + return _read_image_bytes(image[1]) + if isinstance(image, os.PathLike): + return Path(image).read_bytes() + if isinstance(image, _SeekableBinaryReader): + position: Final = image.tell() + image.seek(0) + data: Final = image.read() + image.seek(position) + return data + raise ValueError(f"Unsupported image type for image edit: {type(image).__name__}") + + +def image_to_data_url(image: object) -> str: + if isinstance(image, str): + return image + image_bytes: Final = _read_image_bytes(image) + mime_type: Final = ImageEditRequestUtils.get_image_content_type(image_bytes) + return f"data:{mime_type};base64,{base64.b64encode(image_bytes).decode('utf-8')}" diff --git a/litellm/litellm_core_utils/llm_cost_calc/utils.py b/litellm/litellm_core_utils/llm_cost_calc/utils.py index 3b49f239afe..4b0b875af6d 100644 --- a/litellm/litellm_core_utils/llm_cost_calc/utils.py +++ b/litellm/litellm_core_utils/llm_cost_calc/utils.py @@ -6,7 +6,7 @@ from collections.abc import Mapping, Sequence from dataclasses import dataclass, field, replace from datetime import datetime, timezone, tzinfo from types import MappingProxyType -from typing import Final, Literal, TypedDict, cast +from typing import Final, Literal, Protocol, TypedDict, cast from zoneinfo import ZoneInfo, ZoneInfoNotFoundError from typing_extensions import ReadOnly @@ -2085,6 +2085,41 @@ def calculate_image_response_web_search_cost( ) +class _ModelPricedImageCostCalculator(Protocol): + def __call__(self, model: str, image_response: ImageResponse, model_info: ModelInfo | None = None) -> float: ... + + +def _model_priced_image_cost_calculator( + custom_llm_provider: str | None, model: str +) -> _ModelPricedImageCostCalculator | None: + match custom_llm_provider: + case litellm.LlmProviders.RECRAFT.value: + from litellm.llms.recraft.cost_calculator import cost_calculator as recraft_image_cost_calculator + + return recraft_image_cost_calculator + case litellm.LlmProviders.AIML.value: + from litellm.llms.aiml.image_generation.cost_calculator import cost_calculator as aiml_image_cost_calculator + + return aiml_image_cost_calculator + case litellm.LlmProviders.COMETAPI.value: + from litellm.llms.cometapi.image_generation.cost_calculator import ( + cost_calculator as cometapi_image_cost_calculator, + ) + + return cometapi_image_cost_calculator + case litellm.LlmProviders.RUNWAYML.value: + from litellm.llms.runwayml.cost_calculator import cost_calculator as runwayml_image_cost_calculator + + return runwayml_image_cost_calculator + case litellm.LlmProviders.TENCENT.value: + from litellm.llms.tencent.cost_calculator import image_cost_calculator as tencent_image_cost_calculator + from litellm.llms.tencent.image_generation.transformation import is_tokenhub_image_model + + return tencent_image_cost_calculator if is_tokenhub_image_model(model) else None + case _: + return None + + class CostCalculatorUtils: @staticmethod def call_type_has_image_response(call_type: str) -> bool: @@ -2131,9 +2166,6 @@ class CostCalculatorUtils: from litellm.llms.gemini.image_generation.cost_calculator import ( cost_calculator as gemini_image_cost_calculator, ) - from litellm.llms.recraft.cost_calculator import ( - cost_calculator as recraft_image_cost_calculator, - ) from litellm.llms.vertex_ai.image_generation.cost_calculator import ( cost_calculator as vertex_ai_image_cost_calculator, ) @@ -2146,7 +2178,10 @@ class CostCalculatorUtils: ) resolved_n: Final = n if n is not None else (len(completion_response.data) if completion_response.data else 0) pricing: Final = deployment_pricing(model_info) + model_priced_calculator: Final = _model_priced_image_cost_calculator(custom_llm_provider, model) + if model_priced_calculator is not None: + return model_priced_calculator(model=model, image_response=completion_response, model_info=pricing) if custom_llm_provider == litellm.LlmProviders.VERTEX_AI.value: if isinstance(completion_response, ImageResponse): return vertex_ai_image_cost_calculator( @@ -2164,36 +2199,6 @@ class CostCalculatorUtils: optional_params=optional_params, ) raise TypeError("completion_response must be of type ImageResponse for bedrock image cost calculation") - elif custom_llm_provider == litellm.LlmProviders.RECRAFT.value: - from litellm.llms.recraft.cost_calculator import ( - cost_calculator as recraft_image_cost_calculator, - ) - - return recraft_image_cost_calculator( - model=model, - image_response=completion_response, - model_info=pricing, - ) - elif custom_llm_provider == litellm.LlmProviders.AIML.value: - from litellm.llms.aiml.image_generation.cost_calculator import ( - cost_calculator as aiml_image_cost_calculator, - ) - - return aiml_image_cost_calculator( - model=model, - image_response=completion_response, - model_info=pricing, - ) - elif custom_llm_provider == litellm.LlmProviders.COMETAPI.value: - from litellm.llms.cometapi.image_generation.cost_calculator import ( - cost_calculator as cometapi_image_cost_calculator, - ) - - return cometapi_image_cost_calculator( - model=model, - image_response=completion_response, - model_info=pricing, - ) elif custom_llm_provider == litellm.LlmProviders.GEMINI.value: if call_type in ( CallTypes.image_edit.value, @@ -2237,16 +2242,6 @@ class CostCalculatorUtils: optional_params=optional_params, model_info=pricing, ) - elif custom_llm_provider == litellm.LlmProviders.RUNWAYML.value: - from litellm.llms.runwayml.cost_calculator import ( - cost_calculator as runwayml_image_cost_calculator, - ) - - return runwayml_image_cost_calculator( - model=model, - image_response=completion_response, - model_info=pricing, - ) elif ( custom_llm_provider == litellm.LlmProviders.OPENAI.value or custom_llm_provider == litellm.LlmProviders.AZURE.value diff --git a/litellm/llms/fal_ai/image_edit/flux_lora_depth_transformation.py b/litellm/llms/fal_ai/image_edit/flux_lora_depth_transformation.py index 344d9d3c75b..6d0891295bd 100644 --- a/litellm/llms/fal_ai/image_edit/flux_lora_depth_transformation.py +++ b/litellm/llms/fal_ai/image_edit/flux_lora_depth_transformation.py @@ -4,13 +4,14 @@ from typing import Final from httpx._types import RequestFiles +from litellm.images.utils import image_to_data_url from litellm.litellm_core_utils.url_utils import join_api_base from litellm.secret_managers.main import get_secret_str from litellm.types.images.main import ImageEditOptionalRequestParams from litellm.types.router import GenericLiteLLMParams from litellm.types.utils import FileTypes -from .transformation import DEFAULT_BASE_URL, FalAIImageEditConfig, to_data_url +from .transformation import DEFAULT_BASE_URL, FalAIImageEditConfig FLUX_LORA_DEPTH_ENDPOINT: Final[str] = "fal-ai/flux-lora-depth" SUPPORTED_OPENAI_PARAMS: Final[tuple[str, ...]] = ("n", "size") @@ -68,7 +69,7 @@ class FalAIFluxLoraDepthEditConfig(FalAIImageEditConfig): ) request_body: Final[dict[str, object]] = { # mutable-ok: base class contract returns a dict "prompt": prompt, - "image_url": to_data_url(next(iter(images))), + "image_url": image_to_data_url(next(iter(images))), **provider_params, } return request_body, () diff --git a/litellm/llms/fal_ai/image_edit/transformation.py b/litellm/llms/fal_ai/image_edit/transformation.py index ea4e8cac5bf..5cbf635fd47 100644 --- a/litellm/llms/fal_ai/image_edit/transformation.py +++ b/litellm/llms/fal_ai/image_edit/transformation.py @@ -1,15 +1,12 @@ -import base64 -import os from collections.abc import Mapping -from pathlib import Path from types import MappingProxyType -from typing import TYPE_CHECKING, Final, Protocol, runtime_checkable +from typing import TYPE_CHECKING, Final import httpx from httpx._types import RequestFiles from pydantic import ConfigDict, TypeAdapter -from litellm.images.utils import ImageEditRequestUtils +from litellm.images.utils import image_to_data_url from litellm.llms.base_llm.image_edit.transformation import BaseImageEditConfig from litellm.llms.fal_ai.image_generation.gpt_image_2_transformation import ( map_gpt_image_quality, @@ -39,39 +36,6 @@ PARAM_TRANSLATION: Final[Mapping[str, str]] = MappingProxyType( ) -@runtime_checkable -class _SeekableBinaryReader(Protocol): - def tell(self) -> int: ... - - def seek(self, offset: int) -> int: ... - - def read(self) -> bytes: ... - - -def _read_image_bytes(image: object) -> bytes: - if isinstance(image, bytes): - return image - if isinstance(image, tuple): - return _read_image_bytes(image[1]) - if isinstance(image, os.PathLike): - return Path(image).read_bytes() - if isinstance(image, _SeekableBinaryReader): - position: Final = image.tell() - image.seek(0) - data: Final = image.read() - image.seek(position) - return data - raise ValueError(f"Unsupported image type for Fal AI image edit: {type(image).__name__}") - - -def to_data_url(image: object) -> str: - if isinstance(image, str): - return image - image_bytes: Final = _read_image_bytes(image) - mime_type: Final = ImageEditRequestUtils.get_image_content_type(image_bytes) - return f"data:{mime_type};base64,{base64.b64encode(image_bytes).decode('utf-8')}" - - def _first(value: object) -> object: return value[0] if isinstance(value, list) and value else value @@ -146,14 +110,14 @@ class FalAIImageEditConfig(BaseImageEditConfig): raise ValueError("Fal AI image edit requires at least one input image") mask: Final = _first(image_edit_optional_request_params.get("mask")) mask_field: Final[Mapping[str, str]] = ( - MappingProxyType({"mask_url": to_data_url(mask)}) if mask is not None else MappingProxyType({}) + MappingProxyType({"mask_url": image_to_data_url(mask)}) if mask is not None else MappingProxyType({}) ) provider_params: Final[Mapping[str, object]] = MappingProxyType( {key: value for key, value in image_edit_optional_request_params.items() if key != "mask"} ) request_body: Final[dict[str, object]] = { # mutable-ok: base class contract returns a dict "prompt": prompt, - "image_urls": tuple(to_data_url(img) for img in images), + "image_urls": tuple(image_to_data_url(img) for img in images), **mask_field, **provider_params, } diff --git a/litellm/llms/tencent/cost_calculator.py b/litellm/llms/tencent/cost_calculator.py index d9aebdc3284..5d2abe3acf4 100644 --- a/litellm/llms/tencent/cost_calculator.py +++ b/litellm/llms/tencent/cost_calculator.py @@ -1,6 +1,49 @@ -from litellm.litellm_core_utils.llm_cost_calc.utils import generic_cost_per_token -from litellm.types.utils import Usage +from typing import Final + +from litellm.litellm_core_utils.llm_cost_calc.utils import ( + calculate_image_response_cost_from_usage, + flat_image_cost, + generic_cost_per_token, + resolve_image_model_info, +) +from litellm.types.utils import ImageResponse, ModelInfo, Usage def cost_per_token(model: str, usage: Usage) -> tuple[float, float]: return generic_cost_per_token(model=model, usage=usage, custom_llm_provider="tencent") + + +def _deployment_image_token_pricing(model_info: ModelInfo | None) -> ModelInfo | None: + if model_info is None or model_info.get("output_cost_per_image_token") is not None: + return model_info + output_cost_per_token: Final = model_info.get("output_cost_per_token") + if output_cost_per_token is None: + return model_info + priced: Final[ModelInfo] = {**model_info, "output_cost_per_image_token": output_cost_per_token} + return priced + + +def _deployment_prices_per_image_only(model_info: ModelInfo | None) -> bool: + return ( + model_info is not None + and model_info.get("output_cost_per_image") is not None + and model_info.get("output_cost_per_image_token") is None + and model_info.get("output_cost_per_token") is None + ) + + +def image_cost_calculator(model: str, image_response: ImageResponse, model_info: ModelInfo | None = None) -> float: + resolved_model_info: Final = resolve_image_model_info( + model=model, custom_llm_provider="tencent", model_info=_deployment_image_token_pricing(model_info) + ) + if _deployment_prices_per_image_only(model_info): + return flat_image_cost(resolved_model_info, image_response) + token_based_cost: Final = calculate_image_response_cost_from_usage( + model=model, + image_response=image_response, + custom_llm_provider="tencent", + model_info=resolved_model_info, + ) + if token_based_cost is not None: + return token_based_cost + return flat_image_cost(resolved_model_info, image_response) diff --git a/litellm/llms/tencent/image_edit/__init__.py b/litellm/llms/tencent/image_edit/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/litellm/llms/tencent/image_edit/transformation.py b/litellm/llms/tencent/image_edit/transformation.py new file mode 100644 index 00000000000..3033d161ce7 --- /dev/null +++ b/litellm/llms/tencent/image_edit/transformation.py @@ -0,0 +1,147 @@ +from collections.abc import Mapping +from types import MappingProxyType +from typing import TYPE_CHECKING, Final, TypeAlias + +import httpx +from httpx._types import RequestFiles +from pydantic import TypeAdapter +from typing_extensions import override + +from litellm.images.utils import image_to_data_url +from litellm.llms.base_llm.chat.transformation import BaseLLMException +from litellm.llms.base_llm.image_edit.transformation import BaseImageEditConfig +from litellm.llms.tencent.image_generation.transformation import ( + TOKENHUB_IMAGE_PARAMS, + tokenhub_image_headers, + tokenhub_image_params, + tokenhub_image_response, + tokenhub_image_url, +) +from litellm.types.images.main import ImageEditOptionalRequestParams +from litellm.types.router import GenericLiteLLMParams +from litellm.types.utils import ImageResponse + +if TYPE_CHECKING: + from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj + +_IMAGES: Final = TypeAdapter(tuple[object, ...]) +_FORM_BOOLEANS: Final = MappingProxyType({"true": True, "false": False}) +_OPENAI_PARAMS: Final = frozenset(TOKENHUB_IMAGE_PARAMS) + +_FormEntries: TypeAlias = tuple[tuple[tuple[str, ...], object], ...] + + +def _form_scalar(value: object) -> object: + match value: + case str() if value.lower() in _FORM_BOOLEANS: + return _FORM_BOOLEANS[value.lower()] + case str() if value.removeprefix("-").isdigit(): + return int(value) + case _: + return value + + +def _form_path(key: str) -> tuple[str, ...]: + root, bracket, rest = key.partition("[") + if not bracket or not rest.endswith("]"): + return (key,) + return (root, *rest.removesuffix("]").split("][")) + + +def _nested_form_value(entries: _FormEntries) -> object: + match entries: + case (((), value),): + return _form_scalar(value) + case _: + return _nested_form_fields(tuple(entry for entry in entries if entry[0])) + + +def _nested_form_fields(entries: _FormEntries) -> Mapping[str, object]: + roots: Final = tuple(dict.fromkeys(path[0] for path, _ in entries)) + return { + root: _nested_form_value(tuple((path[1:], value) for path, value in entries if path[0] == root)) + for root in roots + } + + +def _tokenhub_form_params(params: Mapping[str, object]) -> Mapping[str, object]: + tokenhub_entries: Final = tuple( + (_form_path(key), value) for key, value in params.items() if key not in _OPENAI_PARAMS + ) + return { + **{key: value for key, value in params.items() if key in _OPENAI_PARAMS}, + **_nested_form_fields(tokenhub_entries), + } + + +class TencentImageEditConfig(BaseImageEditConfig): + @override + def get_supported_openai_params(self, model: str) -> list[str]: # mutable-ok: mirrors override contract + return list(TOKENHUB_IMAGE_PARAMS) + + @override + def map_openai_params( + self, + image_edit_optional_params: ImageEditOptionalRequestParams, + model: str, + drop_params: bool, + ) -> dict[str, object]: # mutable-ok: mirrors override contract + return dict(tokenhub_image_params(image_edit_optional_params, model, drop_params)) + + @override + def validate_environment( + self, + headers: Mapping[str, str], + model: str, + api_key: str | None = None, + litellm_params: Mapping[str, object] | None = None, + api_base: str | None = None, + ) -> dict[str, str]: # mutable-ok: mirrors override contract + return dict(tokenhub_image_headers(headers, api_key, model)) + + @override + def get_complete_url( + self, + model: str, + api_base: str | None, + litellm_params: Mapping[str, object], + ) -> str: + return tokenhub_image_url(api_base) + + @override + def use_multipart_form_data(self) -> bool: + return False + + @override + def transform_image_edit_request( + self, + model: str, + prompt: str | None, + image: object, + image_edit_optional_request_params: Mapping[str, object], + litellm_params: GenericLiteLLMParams, + headers: Mapping[str, str], + ) -> tuple[dict[str, object], RequestFiles]: # mutable-ok: mirrors override contract + images: Final = _IMAGES.validate_python(image) if isinstance(image, list) else (image,) + return { + "model": model, + "prompt": prompt, + "images": tuple(image_to_data_url(img) for img in images if img is not None), + **_tokenhub_form_params(image_edit_optional_request_params), + }, () + + @override + def transform_image_edit_response( + self, + model: str, + raw_response: httpx.Response, + logging_obj: "LiteLLMLoggingObj", + ) -> ImageResponse: + image_response: Final = tokenhub_image_response(raw_response) + if image_response is None: + raise BaseLLMException( + status_code=raw_response.status_code, + message=raw_response.text, + headers=raw_response.headers, + ) + return image_response diff --git a/litellm/llms/tencent/image_generation/__init__.py b/litellm/llms/tencent/image_generation/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/litellm/llms/tencent/image_generation/transformation.py b/litellm/llms/tencent/image_generation/transformation.py new file mode 100644 index 00000000000..fb4b102e8be --- /dev/null +++ b/litellm/llms/tencent/image_generation/transformation.py @@ -0,0 +1,200 @@ +from collections.abc import Mapping, Sequence +from typing import TYPE_CHECKING, Final + +import httpx +from pydantic import BaseModel, ConfigDict, TypeAdapter, ValidationError +from typing_extensions import override + +from litellm.exceptions import AuthenticationError, UnsupportedParamsError +from litellm.litellm_core_utils.url_utils import ensure_api_route, join_api_base +from litellm.llms.base_llm.chat.transformation import BaseLLMException +from litellm.llms.base_llm.image_generation.transformation import ( + BaseImageGenerationConfig, +) +from litellm.secret_managers.main import get_secret_str +from litellm.types.llms.openai import ( + AllMessageValues, + OpenAIImageGenerationOptionalParams, +) +from litellm.types.utils import ( + ImageObject, + ImageResponse, + ImageUsage, + ImageUsageInputTokensDetails, + LlmProviders, +) + +if TYPE_CHECKING: + from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj + +TOKENHUB_API_BASE: Final = "https://tokenhub-intl.tencentcloudmaas.com/v1" +TOKENHUB_IMAGE_PATH: Final = "/wand/si-image/generation" +TOKENHUB_IMAGE_MODEL_PREFIX: Final = "seedream-" +TOKENHUB_IMAGE_PARAMS: Final[tuple[OpenAIImageGenerationOptionalParams, ...]] = ( + "n", + "size", + "response_format", + "output_format", + "background", + "user", +) + +_JSON_OBJECT: Final = TypeAdapter(Mapping[str, object], config=ConfigDict(hide_input_in_errors=True)) + + +class _TokenHubImage(BaseModel): + model_config = ConfigDict(frozen=True) + + url: str | None = None + b64_json: str | None = None + size: str | None = None + output_format: str | None = None + + +class _TokenHubUsage(BaseModel): + model_config = ConfigDict(frozen=True) + + total_tokens: int + + +class _TokenHubImageResponse(BaseModel): + model_config = ConfigDict(frozen=True, hide_input_in_errors=True) + + created: int + data: tuple[_TokenHubImage, ...] + tokenhub_usage: _TokenHubUsage | None = None + + +def is_tokenhub_image_model(model: str) -> bool: + return model.removeprefix(f"{LlmProviders.TENCENT.value}/").startswith(TOKENHUB_IMAGE_MODEL_PREFIX) + + +def tokenhub_image_url(api_base: str | None) -> str: + base: Final = api_base or get_secret_str("TENCENT_API_BASE") or TOKENHUB_API_BASE + return join_api_base(ensure_api_route(base, "v1"), TOKENHUB_IMAGE_PATH) + + +def tokenhub_image_headers(headers: Mapping[str, str], api_key: str | None, model: str) -> Mapping[str, str]: + resolved_api_key: Final = api_key or get_secret_str("TENCENT_API_KEY") + if not resolved_api_key: + raise AuthenticationError( + message="TENCENT_API_KEY is not set. Set it in the environment or pass api_key.", + llm_provider="tencent", + model=model, + ) + return {**headers, "Authorization": f"Bearer {resolved_api_key}", "Content-Type": "application/json"} + + +def tokenhub_image_params(params: Mapping[str, object], model: str, drop_params: bool) -> Mapping[str, object]: + n: Final = params.get("n") + if n is not None and n != 1 and not drop_params: + raise UnsupportedParamsError( + status_code=400, + message=f"Tencent TokenHub returns one image per request, so n={n} is not supported for {model}. " + + "To send n=1 instead, set `litellm.drop_params = True`.", + ) + return { + key: value + for key, value in params.items() + if key != "n" and value is not None and not (key == "background" and value == "auto") + } + + +def tokenhub_image_response(raw_response: httpx.Response) -> ImageResponse | None: + try: + parsed: Final = _TokenHubImageResponse.model_validate_json(raw_response.content) + except ValidationError: + return None + total_tokens: Final = parsed.tokenhub_usage.total_tokens if parsed.tokenhub_usage is not None else 0 + first_image: Final = parsed.data[0] if parsed.data else _TokenHubImage() + return ImageResponse( + created=parsed.created, + data=[ImageObject(url=image.url, b64_json=image.b64_json) for image in parsed.data], + size=first_image.size, + output_format=first_image.output_format, + usage=ImageUsage( + input_tokens=0, + input_tokens_details=ImageUsageInputTokensDetails(image_tokens=0, text_tokens=0), + output_tokens=total_tokens, + total_tokens=total_tokens, + ), + ) + + +class TencentImageGenerationConfig(BaseImageGenerationConfig): + @override + def get_supported_openai_params( + self, model: str + ) -> list[OpenAIImageGenerationOptionalParams]: # mutable-ok: mirrors override contract + return list(TOKENHUB_IMAGE_PARAMS) + + @override + def map_openai_params( + self, + non_default_params: Mapping[str, object], + optional_params: Mapping[str, object], + model: str, + drop_params: bool, + ) -> dict[str, object]: # mutable-ok: mirrors override contract + return {**optional_params, **tokenhub_image_params(non_default_params, model, drop_params)} + + @override + def get_complete_url( + self, + api_base: str | None, + api_key: str | None, + model: str, + optional_params: Mapping[str, object], + litellm_params: Mapping[str, object], + stream: bool | None = None, + ) -> str: + return tokenhub_image_url(api_base) + + @override + def validate_environment( + self, + headers: Mapping[str, str], + model: str, + messages: Sequence[AllMessageValues], + optional_params: Mapping[str, object], + litellm_params: Mapping[str, object], + api_key: str | None = None, + api_base: str | None = None, + ) -> dict[str, str]: # mutable-ok: mirrors override contract + return dict(tokenhub_image_headers(headers, api_key, model)) + + @override + def transform_image_generation_request( + self, + model: str, + prompt: str, + optional_params: Mapping[str, object], + litellm_params: Mapping[str, object], + headers: Mapping[str, str], + ) -> dict[str, object]: # mutable-ok: mirrors override contract + extra_body: Final = _JSON_OBJECT.validate_python(optional_params.get("extra_body") or {}) + params: Final = {key: value for key, value in optional_params.items() if key != "extra_body"} + return {"model": model, "prompt": prompt, **params, **extra_body} + + @override + def transform_image_generation_response( + self, + model: str, + raw_response: httpx.Response, + model_response: ImageResponse, + logging_obj: "LiteLLMLoggingObj", + request_data: Mapping[str, object], + optional_params: Mapping[str, object], + litellm_params: Mapping[str, object], + encoding: object, + api_key: str | None = None, + json_mode: bool | None = None, + ) -> ImageResponse: + image_response: Final = tokenhub_image_response(raw_response) + if image_response is None: + raise BaseLLMException( + status_code=raw_response.status_code, + message=raw_response.text, + headers=raw_response.headers, + ) + return image_response diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index 1f2bd7d7837..dc043dc4d26 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -61986,6 +61986,39 @@ "supports_reasoning": true, "supports_vision": false }, + "tencent/seedream-image-v5.0-flash": { + "litellm_provider": "tencent", + "mode": "image_generation", + "output_cost_per_image": 0.018, + "output_cost_per_image_token": 1.6e-06, + "source": "https://cloud.tencent.com/document/product/1823/130055", + "supported_endpoints": [ + "/v1/images/generations", + "/v1/images/edits" + ] + }, + "tencent/seedream-image-v5.0-lite": { + "litellm_provider": "tencent", + "mode": "image_generation", + "output_cost_per_image": 0.035, + "output_cost_per_image_token": 1.6e-06, + "source": "https://www.tencentcloud.com/document/product/1300/78937", + "supported_endpoints": [ + "/v1/images/generations", + "/v1/images/edits" + ] + }, + "tencent/seedream-image-v5.0-pro": { + "litellm_provider": "tencent", + "mode": "image_generation", + "output_cost_per_image": 0.09, + "output_cost_per_image_token": 1.6e-06, + "source": "https://www.tencentcloud.com/document/product/1300/78937", + "supported_endpoints": [ + "/v1/images/generations", + "/v1/images/edits" + ] + }, "cognition/swe-1.6": { "input_cost_per_token": 5e-07, "output_cost_per_token": 2.5e-06, diff --git a/litellm/provider_endpoints_support_backup.json b/litellm/provider_endpoints_support_backup.json index d73ac19bae1..0a6c2bcb2a9 100644 --- a/litellm/provider_endpoints_support_backup.json +++ b/litellm/provider_endpoints_support_backup.json @@ -3793,8 +3793,8 @@ "responses": true, "text_completion": true, "embeddings": false, - "image_generations": false, - "image_edits": false, + "image_generations": true, + "image_edits": true, "audio_transcriptions": false, "audio_speech": false, "rerank": false, diff --git a/litellm/utils.py b/litellm/utils.py index a231644addf..1ef39059e1b 100644 --- a/litellm/utils.py +++ b/litellm/utils.py @@ -10418,6 +10418,13 @@ class ProviderConfigManager: return get_modelscope_image_generation_config(model) elif LlmProviders.EDENAI == provider: return litellm.EdenAIImageGenerationConfig() + elif LlmProviders.TENCENT == provider: + from litellm.llms.tencent.image_generation.transformation import ( + TencentImageGenerationConfig, + is_tokenhub_image_model, + ) + + return TencentImageGenerationConfig() if is_tokenhub_image_model(model) else None return None @staticmethod @@ -10606,6 +10613,15 @@ class ProviderConfigManager: ) return get_openrouter_image_edit_config(model) + elif LlmProviders.TENCENT == provider: + from litellm.llms.tencent.image_edit.transformation import ( + TencentImageEditConfig, + ) + from litellm.llms.tencent.image_generation.transformation import ( + is_tokenhub_image_model, + ) + + return TencentImageEditConfig() if is_tokenhub_image_model(model) else None return None @staticmethod diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index 1f2bd7d7837..dc043dc4d26 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -61986,6 +61986,39 @@ "supports_reasoning": true, "supports_vision": false }, + "tencent/seedream-image-v5.0-flash": { + "litellm_provider": "tencent", + "mode": "image_generation", + "output_cost_per_image": 0.018, + "output_cost_per_image_token": 1.6e-06, + "source": "https://cloud.tencent.com/document/product/1823/130055", + "supported_endpoints": [ + "/v1/images/generations", + "/v1/images/edits" + ] + }, + "tencent/seedream-image-v5.0-lite": { + "litellm_provider": "tencent", + "mode": "image_generation", + "output_cost_per_image": 0.035, + "output_cost_per_image_token": 1.6e-06, + "source": "https://www.tencentcloud.com/document/product/1300/78937", + "supported_endpoints": [ + "/v1/images/generations", + "/v1/images/edits" + ] + }, + "tencent/seedream-image-v5.0-pro": { + "litellm_provider": "tencent", + "mode": "image_generation", + "output_cost_per_image": 0.09, + "output_cost_per_image_token": 1.6e-06, + "source": "https://www.tencentcloud.com/document/product/1300/78937", + "supported_endpoints": [ + "/v1/images/generations", + "/v1/images/edits" + ] + }, "cognition/swe-1.6": { "input_cost_per_token": 5e-07, "output_cost_per_token": 2.5e-06, diff --git a/provider_endpoints_support.json b/provider_endpoints_support.json index d73ac19bae1..0a6c2bcb2a9 100644 --- a/provider_endpoints_support.json +++ b/provider_endpoints_support.json @@ -3793,8 +3793,8 @@ "responses": true, "text_completion": true, "embeddings": false, - "image_generations": false, - "image_edits": false, + "image_generations": true, + "image_edits": true, "audio_transcriptions": false, "audio_speech": false, "rerank": false, diff --git a/tests/unit/images/test_main.py b/tests/unit/images/test_main.py index 88d5ae1fc2e..29424e50079 100644 --- a/tests/unit/images/test_main.py +++ b/tests/unit/images/test_main.py @@ -357,6 +357,24 @@ def test_image_edit_prices_a_vertex_deployment_at_its_configured_location( assert cost_at("us-central1") == pytest.approx(0.044) +def test_image_edit_accepts_a_string_additional_drop_params(respx_mock: respx.MockRouter) -> None: + edits_route: Final = respx_mock.post("https://api.openai.com/v1/images/edits").mock( + return_value=httpx.Response(status_code=200, json={"created": 1792551838, "data": [{"b64_json": "aW1n"}]}) + ) + + response: Final = litellm.image_edit( + model="openai/gpt-image-2.5-sunburst", + image=b"\x89PNG\r\n\x1a\nfakepng", + prompt="make the circle blue", + api_key="sk-test", + quality="high", + additional_drop_params="quality", + ) + + assert edits_route.called + assert response.data[0].b64_json == "aW1n" + + class TestCustomLogger(CustomLogger): __test__ = False diff --git a/tests/unit/llms/tencent/image_edit/__init__.py b/tests/unit/llms/tencent/image_edit/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/unit/llms/tencent/image_edit/test_tencent_image_edit_transformation.py b/tests/unit/llms/tencent/image_edit/test_tencent_image_edit_transformation.py new file mode 100644 index 00000000000..506c43f6cba --- /dev/null +++ b/tests/unit/llms/tencent/image_edit/test_tencent_image_edit_transformation.py @@ -0,0 +1,232 @@ +import base64 +import io +import json +from typing import Final + +import httpx +import pytest + +import litellm +from litellm.llms.custom_httpx.http_handler import HTTPHandler + +PNG_BYTES: Final = b"\x89PNG\r\n\x1a\n" + b"\x00" * 32 +TOKENHUB_RESPONSE: Final[dict[str, object]] = { + "created": 1792551838, + "data": [{"output_format": "jpeg", "size": "2048x2048", "url": "https://cos.example/edited.jpg"}], + "model": "doubao-seedream-5-0-lite", + "tokenhub_usage": {"total_tokens": 21875}, + "request_id": "4f795d48-b227-4c35-af2c-1a0ba288b6b6", +} + + +def _client(captured: dict[str, object]) -> HTTPHandler: + def respond(request: httpx.Request) -> httpx.Response: + captured["url"] = str(request.url) + captured["content_type"] = request.headers.get("content-type") + captured["authorization"] = request.headers.get("authorization") + captured["body"] = json.loads(request.content) + return httpx.Response(200, json=TOKENHUB_RESPONSE) + + return HTTPHandler(client=httpx.Client(transport=httpx.MockTransport(respond))) + + +def test_edit_sends_uploaded_images_as_data_urls_in_a_json_body() -> None: + captured: Final[dict[str, object]] = {} + png_data_url: Final = f"data:image/png;base64,{base64.b64encode(PNG_BYTES).decode()}" + + response: Final = litellm.image_edit( + model="tencent/seedream-image-v5.0-lite", + image=[PNG_BYTES, io.BytesIO(PNG_BYTES), "https://ref.example/style.png"], + prompt="turn the square into a paper lantern", + size="2K", + api_key="sk-tokenhub", + client=_client(captured), + ) + + assert captured["url"] == "https://tokenhub-intl.tencentcloudmaas.com/v1/wand/si-image/generation" + assert captured["content_type"] == "application/json" + assert captured["authorization"] == "Bearer sk-tokenhub" + assert captured["body"] == { + "model": "seedream-image-v5.0-lite", + "prompt": "turn the square into a paper lantern", + "images": [png_data_url, png_data_url, "https://ref.example/style.png"], + "size": "2K", + } + assert response.data[0].url == "https://cos.example/edited.jpg" + assert response.usage.total_tokens == 21875 + + +def test_edit_rejects_a_mask_without_drop_params() -> None: + captured: Final[dict[str, object]] = {} + + with pytest.raises(litellm.UnsupportedParamsError): + litellm.image_edit( + model="tencent/seedream-image-v5.0-pro", + image=PNG_BYTES, + mask=PNG_BYTES, + prompt="turn the square into a paper lantern", + api_key="sk-tokenhub", + client=_client(captured), + ) + + assert captured == {} + + +def test_edit_drops_a_mask_under_drop_params() -> None: + captured: Final[dict[str, object]] = {} + + litellm.image_edit( + model="tencent/seedream-image-v5.0-pro", + image=PNG_BYTES, + mask=PNG_BYTES, + prompt="turn the square into a paper lantern", + api_key="sk-tokenhub", + client=_client(captured), + drop_params=True, + ) + + assert captured["body"] == { + "model": "seedream-image-v5.0-pro", + "prompt": "turn the square into a paper lantern", + "images": [f"data:image/png;base64,{base64.b64encode(PNG_BYTES).decode()}"], + } + + +def test_edit_is_not_offered_for_tencent_models_outside_seedream() -> None: + captured: Final[dict[str, object]] = {} + + with pytest.raises(Exception, match="image edit is not supported for tencent"): + litellm.image_edit( + model="tencent/hy-image-v3.5-preview", + image=PNG_BYTES, + prompt="turn the square into a paper lantern", + api_key="sk-tokenhub", + client=_client(captured), + ) + + assert captured == {} + + +@pytest.mark.parametrize( + ("form_params", "expected_body_params"), + [ + ({"watermark": "false", "seed": "7"}, {"watermark": False, "seed": 7}), + ({"watermark": "True", "seed": "-1"}, {"watermark": True, "seed": -1}), + ({"watermark": "maybe"}, {"watermark": "maybe"}), + ( + {"layer_decomposition": "true", "optimize_prompt_options[mode]": "fast"}, + {"layer_decomposition": True, "optimize_prompt_options": {"mode": "fast"}}, + ), + ( + {"sequential_image_generation_options[max_images]": "3", "user": "12345"}, + {"sequential_image_generation_options": {"max_images": 3}, "user": "12345"}, + ), + ], +) +def test_edit_parses_tokenhub_values_sent_as_form_strings( + form_params: dict[str, object], expected_body_params: dict[str, object] +) -> None: + captured: Final[dict[str, object]] = {} + + litellm.image_edit( + model="tencent/seedream-image-v5.0-flash", + image=PNG_BYTES, + prompt="turn the square into a paper lantern", + api_key="sk-tokenhub", + client=_client(captured), + **form_params, + ) + + assert captured["body"] == { + "model": "seedream-image-v5.0-flash", + "prompt": "turn the square into a paper lantern", + "images": [f"data:image/png;base64,{base64.b64encode(PNG_BYTES).decode()}"], + **expected_body_params, + } + + +@pytest.mark.parametrize( + ("params", "expected_body_params"), + [ + ({"background": "auto", "watermark": False}, {"watermark": False}), + ({"watermark": False, "seed": 7, "additional_drop_params": ["watermark"]}, {"seed": 7}), + ({"extra_body": {"watermark": False}, "additional_drop_params": ["watermark"]}, {}), + ], +) +def test_edit_keeps_dropped_params_out_of_the_tokenhub_body( + params: dict[str, object], expected_body_params: dict[str, object] +) -> None: + captured: Final[dict[str, object]] = {} + + litellm.image_edit( + model="tencent/seedream-image-v5.0-flash", + image=PNG_BYTES, + prompt="turn the square into a paper lantern", + api_key="sk-tokenhub", + client=_client(captured), + **params, + ) + + assert captured["body"] == { + "model": "seedream-image-v5.0-flash", + "prompt": "turn the square into a paper lantern", + "images": [f"data:image/png;base64,{base64.b64encode(PNG_BYTES).decode()}"], + **expected_body_params, + } + + +@pytest.mark.parametrize( + ("params", "expected_body_params"), + [ + ({"extra_body": {"response_format": "b64_json"}}, {"response_format": "b64_json"}), + ({"size": "1K", "extra_body": {"size": "2K"}}, {"size": "2K"}), + ({"extra_body": {"model": "seedream-image-v5.0-pro", "seed": 7}}, {"seed": 7}), + ({"extra_body": '{"seed": 7}'}, {}), + ], +) +def test_edit_sends_extra_body_fields_as_given( + params: dict[str, object], expected_body_params: dict[str, object] +) -> None: + captured: Final[dict[str, object]] = {} + + litellm.image_edit( + model="tencent/seedream-image-v5.0-flash", + image=PNG_BYTES, + prompt="turn the square into a paper lantern", + api_key="sk-tokenhub", + client=_client(captured), + **params, + ) + + assert captured["body"] == { + "model": "seedream-image-v5.0-flash", + "prompt": "turn the square into a paper lantern", + "images": [f"data:image/png;base64,{base64.b64encode(PNG_BYTES).decode()}"], + **expected_body_params, + } + + +def test_edit_forwards_tokenhub_only_params_and_extra_body() -> None: + captured: Final[dict[str, object]] = {} + + litellm.image_edit( + model="tencent/seedream-image-v5.0-flash", + image=PNG_BYTES, + prompt="turn the square into a paper lantern", + api_key="sk-tokenhub", + client=_client(captured), + user="end-user-42", + watermark=False, + seed=7, + extra_body={"sequential_image_generation": "disabled"}, + ) + + assert captured["body"] == { + "model": "seedream-image-v5.0-flash", + "prompt": "turn the square into a paper lantern", + "images": [f"data:image/png;base64,{base64.b64encode(PNG_BYTES).decode()}"], + "user": "end-user-42", + "watermark": False, + "seed": 7, + "sequential_image_generation": "disabled", + } diff --git a/tests/unit/llms/tencent/image_generation/__init__.py b/tests/unit/llms/tencent/image_generation/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/unit/llms/tencent/image_generation/test_tencent_image_generation_transformation.py b/tests/unit/llms/tencent/image_generation/test_tencent_image_generation_transformation.py new file mode 100644 index 00000000000..ab16382cdae --- /dev/null +++ b/tests/unit/llms/tencent/image_generation/test_tencent_image_generation_transformation.py @@ -0,0 +1,215 @@ +import json +from typing import Final + +import httpx +import openai +import pytest + +import litellm +from litellm.llms.custom_httpx.http_handler import HTTPHandler + +TOKENHUB_RESPONSE: Final[dict[str, object]] = { + "created": 1792551825, + "data": [{"output_format": "jpeg", "size": "1248x832", "url": "https://cos.example/lantern.jpg"}], + "model": "doubao-seedream-5-0-pro-260628", + "tokenhub_usage": {"total_tokens": 28125}, + "request_id": "c57b0374-b5fe-47db-b4b2-751315f920aa", +} + + +def _client(captured: dict[str, object], status_code: int = 200, body: dict[str, object] | None = None) -> HTTPHandler: + def respond(request: httpx.Request) -> httpx.Response: + captured["url"] = str(request.url) + captured["authorization"] = request.headers.get("authorization") + captured["body"] = json.loads(request.content) + return httpx.Response(status_code, json=TOKENHUB_RESPONSE if body is None else body) + + return HTTPHandler(client=httpx.Client(transport=httpx.MockTransport(respond))) + + +def test_generation_posts_openai_and_tokenhub_params_to_the_wand_route() -> None: + captured: Final[dict[str, object]] = {} + + response: Final = litellm.image_generation( + model="tencent/seedream-image-v5.0-pro", + prompt="a red paper lantern", + size="2K", + response_format="url", + api_key="sk-tokenhub", + client=_client(captured), + images=["https://ref.example/lantern.png"], + watermark=False, + ) + + assert captured["url"] == "https://tokenhub-intl.tencentcloudmaas.com/v1/wand/si-image/generation" + assert captured["authorization"] == "Bearer sk-tokenhub" + assert captured["body"] == { + "model": "seedream-image-v5.0-pro", + "prompt": "a red paper lantern", + "size": "2K", + "response_format": "url", + "images": ["https://ref.example/lantern.png"], + "watermark": False, + } + assert response.data[0].url == "https://cos.example/lantern.jpg" + assert response.size == "1248x832" + assert response.output_format == "jpeg" + assert response.created == 1792551825 + assert response.usage.total_tokens == 28125 + assert response.usage.output_tokens == 28125 + + +def test_generation_returns_base64_images() -> None: + captured: Final[dict[str, object]] = {} + body: Final[dict[str, object]] = { + **TOKENHUB_RESPONSE, + "data": [{"output_format": "png", "size": "1024x1024", "b64_json": "aW1n"}], + } + + response: Final = litellm.image_generation( + model="tencent/seedream-image-v5.0-flash", + prompt="a red paper lantern", + response_format="b64_json", + api_key="sk-tokenhub", + client=_client(captured, body=body), + ) + + assert response.data[0].b64_json == "aW1n" + assert response.data[0].url is None + + +@pytest.mark.parametrize( + ("api_base", "expected_url"), + [ + ("https://tokenhub.example", "https://tokenhub.example/v1/wand/si-image/generation"), + ("https://tokenhub.example/v1", "https://tokenhub.example/v1/wand/si-image/generation"), + ("https://tokenhub.example/v1/", "https://tokenhub.example/v1/wand/si-image/generation"), + ], +) +def test_generation_resolves_a_configured_api_base_to_the_wand_route(api_base: str, expected_url: str) -> None: + captured: Final[dict[str, object]] = {} + + litellm.image_generation( + model="tencent/seedream-image-v5.0-lite", + prompt="a red paper lantern", + api_key="sk-tokenhub", + api_base=api_base, + client=_client(captured), + ) + + assert captured["url"] == expected_url + + +@pytest.mark.parametrize( + ("params", "expected_body_params"), + [ + ({"n": 1}, {}), + ({"n": 3, "drop_params": True}, {}), + ({"background": "auto"}, {}), + ({"background": "transparent", "output_format": "png"}, {"background": "transparent", "output_format": "png"}), + ({"user": "end-user-42"}, {"user": "end-user-42"}), + ], +) +def test_generation_sends_only_values_tokenhub_accepts( + params: dict[str, object], expected_body_params: dict[str, object] +) -> None: + captured: Final[dict[str, object]] = {} + + litellm.image_generation( + model="tencent/seedream-image-v5.0-pro", + prompt="a red paper lantern", + api_key="sk-tokenhub", + client=_client(captured), + **params, + ) + + assert captured["body"] == { + "model": "seedream-image-v5.0-pro", + "prompt": "a red paper lantern", + **expected_body_params, + } + + +def test_generation_rejects_more_than_one_image_without_drop_params() -> None: + captured: Final[dict[str, object]] = {} + + with pytest.raises(litellm.UnsupportedParamsError) as exc_info: + litellm.image_generation( + model="tencent/seedream-image-v5.0-pro", + prompt="a red paper lantern", + n=2, + api_key="sk-tokenhub", + client=_client(captured), + ) + + assert exc_info.value.status_code == 400 + assert captured == {} + + +@pytest.mark.parametrize( + ("error", "expected_message"), + [ + ( + { + "error": { + "message": "The model or service ID 'seedream-image-v5.0-pro' does not exist.", + "code": "400004", + "type": "gateway_error", + } + }, + "does not exist", + ), + ({"Type": 2, "Code": 1001, "Msg": "Invalid param", "Desc": ""}, "Invalid param"), + ], +) +def test_generation_surfaces_the_tokenhub_error_message(error: dict[str, object], expected_message: str) -> None: + with pytest.raises(litellm.BadRequestError) as exc_info: + litellm.image_generation( + model="tencent/seedream-image-v5.0-pro", + prompt="a red paper lantern", + api_key="sk-tokenhub", + client=_client({}, status_code=400, body=error), + ) + + assert expected_message in str(exc_info.value) + + +def test_generation_without_a_tokenhub_key_is_an_authentication_error(monkeypatch: pytest.MonkeyPatch) -> None: + monkeypatch.delenv("TENCENT_API_KEY", raising=False) + captured: Final[dict[str, object]] = {} + + with pytest.raises(litellm.AuthenticationError) as exc_info: + litellm.image_generation( + model="tencent/seedream-image-v5.0-pro", + prompt="a red paper lantern", + client=_client(captured), + ) + + assert exc_info.value.status_code == 401 + assert "TENCENT_API_KEY" in str(exc_info.value) + assert captured == {} + + +def test_generation_keeps_other_tencent_image_models_on_the_openai_compatible_route() -> None: + captured: Final[dict[str, object]] = {} + + def respond(request: httpx.Request) -> httpx.Response: + captured["url"] = str(request.url) + captured["body"] = json.loads(request.content) + return httpx.Response(200, json={"created": 1792551825, "data": [{"url": "https://cos.example/hy.png"}]}) + + response: Final = litellm.image_generation( + model="tencent/hy-image-v3.5-preview", + prompt="a red paper lantern", + quality="high", + api_key="sk-tokenhub", + client=openai.OpenAI( + api_key="sk-tokenhub", + base_url="https://tokenhub-intl.tencentcloudmaas.com/v1", + http_client=httpx.Client(transport=httpx.MockTransport(respond)), + ), + ) + + assert captured["url"] == "https://tokenhub-intl.tencentcloudmaas.com/v1/images/generations" + assert captured["body"] == {"model": "hy-image-v3.5-preview", "prompt": "a red paper lantern", "quality": "high"} + assert response.data[0].url == "https://cos.example/hy.png" diff --git a/tests/unit/llms/tencent/test_tencent_cost_calculator.py b/tests/unit/llms/tencent/test_tencent_cost_calculator.py new file mode 100644 index 00000000000..7f98ee02810 --- /dev/null +++ b/tests/unit/llms/tencent/test_tencent_cost_calculator.py @@ -0,0 +1,128 @@ +from typing import Final + +import httpx +import openai +import pytest + +import litellm +from litellm.llms.custom_httpx.http_handler import HTTPHandler +from litellm.types.utils import ImageObject, ImageResponse, ImageUsage, ImageUsageInputTokensDetails + +SEEDREAM_MODELS: Final = ( + "tencent/seedream-image-v5.0-pro", + "tencent/seedream-image-v5.0-flash", + "tencent/seedream-image-v5.0-lite", +) + + +def _image_response(total_tokens: int, images: int = 1) -> ImageResponse: + return ImageResponse( + created=1792551825, + data=[ImageObject(url=f"https://cos.example/{index}.jpg") for index in range(images)], + usage=ImageUsage( + input_tokens=0, + input_tokens_details=ImageUsageInputTokensDetails(image_tokens=0, text_tokens=0), + output_tokens=total_tokens, + total_tokens=total_tokens, + ), + ) + + +@pytest.mark.parametrize("model", SEEDREAM_MODELS) +@pytest.mark.parametrize("call_type", ["image_generation", "image_edit"]) +def test_seedream_bills_tokenhub_usage_tokens_for_a_tier_size( + local_model_cost_map: None, model: str, call_type: str +) -> None: + token_rate: Final = litellm.model_cost[model]["output_cost_per_image_token"] + + cost: Final = litellm.completion_cost( + completion_response=_image_response(total_tokens=56250), + model=model, + call_type=call_type, + size="2K", + ) + + assert cost == pytest.approx(56250 * token_rate) + + +@pytest.mark.parametrize("model", SEEDREAM_MODELS) +def test_seedream_falls_back_to_the_per_image_price_without_usage(local_model_cost_map: None, model: str) -> None: + per_image_price: Final = litellm.model_cost[model]["output_cost_per_image"] + + cost: Final = litellm.completion_cost( + completion_response=_image_response(total_tokens=0, images=2), + model=model, + call_type="image_generation", + size="2K", + ) + + assert cost == pytest.approx(2 * per_image_price) + + +async def test_a_deployment_priced_tencent_image_model_outside_seedream_bills_its_deployment_price() -> None: + def respond(request: httpx.Request) -> httpx.Response: + return httpx.Response(200, json={"created": 1792551825, "data": [{"url": "https://cos.example/hy.png"}]}) + + router: Final = litellm.Router( + model_list=[ + { + "model_name": "hy-image", + "litellm_params": {"model": "tencent/hy-image-v3.5-preview", "api_key": "sk-tokenhub"}, + "model_info": {"input_cost_per_image": 0.04}, + } + ] + ) + + response: Final = await router.aimage_generation( + model="hy-image", + prompt="a red paper lantern", + client=openai.AsyncOpenAI( + api_key="sk-tokenhub", + base_url="https://tokenhub-intl.tencentcloudmaas.com/v1", + http_client=httpx.AsyncClient(transport=httpx.MockTransport(respond)), + ), + ) + + assert response._hidden_params["response_cost"] == pytest.approx(0.04) + + +@pytest.mark.parametrize( + ("deployment_price", "expected_cost"), + [ + ({"output_cost_per_token": 0.000001}, 11250 * 0.000001), + ({"output_cost_per_image": 0.05}, 0.05), + ], +) +def test_a_seedream_deployment_priced_in_the_ui_bills_its_own_rate( + deployment_price: dict[str, float], expected_cost: float +) -> None: + def respond(request: httpx.Request) -> httpx.Response: + return httpx.Response( + 200, + json={ + "created": 1792551825, + "data": [{"output_format": "jpeg", "size": "1024x1024", "url": "https://cos.example/0.jpg"}], + "tokenhub_usage": {"total_tokens": 11250}, + }, + ) + + router: Final = litellm.Router( + model_list=[ + { + "model_name": "seedream-flash", + "litellm_params": { + "model": "tencent/seedream-image-v5.0-flash", + "api_key": "sk-tokenhub", + **deployment_price, + }, + } + ] + ) + + response: Final = router.image_generation( + model="seedream-flash", + prompt="a red paper lantern", + client=HTTPHandler(client=httpx.Client(transport=httpx.MockTransport(respond))), + ) + + assert response._hidden_params["response_cost"] == pytest.approx(expected_cost) From 6063187fea3715087bddb7d23e7eaefaa5dea82d Mon Sep 17 00:00:00 2001 From: quietjules <349048734+quietjules@users.noreply.github.com> Date: Wed, 28 Oct 2026 15:53:06 -0700 Subject: [PATCH 22/41] test(spend): audit #47506 billed background cost reaching the creator's loggers (#48919) Adds integration cells for a different key reading the response, message redaction, a rotated, deleted, or team-deleted creator, and a creator whose Langfuse keys resolve to nothing Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com> --- .../test_background_response_billed_once.py | 183 ++++++++++++++++-- 1 file changed, 172 insertions(+), 11 deletions(-) diff --git a/tests/integration/spend/test_background_response_billed_once.py b/tests/integration/spend/test_background_response_billed_once.py index b207e46204a..5f948fac3ab 100644 --- a/tests/integration/spend/test_background_response_billed_once.py +++ b/tests/integration/spend/test_background_response_billed_once.py @@ -1,6 +1,7 @@ import asyncio import base64 import json +import math import uuid from collections.abc import Callable, Mapping from concurrent.futures import ThreadPoolExecutor @@ -10,9 +11,18 @@ from typing import Final import httpx import openai import pytest -from integration._support.client import Gateway, Scenario, eventually, held, list_value, object_value, string_value +from integration._support.client import ( + Gateway, + Scenario, + delete_key_if_present, + eventually, + held, + list_value, + object_value, + string_value, +) from integration._support.database import read_rows, write_rows -from integration._support.otlp_sink import decode_spans +from integration._support.otlp_sink import Span, decode_spans from integration._support.upstream import delete_scenario, register_scenario from integration._support.wire import Request, wire_server from integration.caching.response_cache_case import entry_key, redis_store @@ -859,7 +869,7 @@ LANGFUSE_COST: Final = "langfuse.observation.cost_details" _LANGFUSE_COST_DETAILS: Final = TypeAdapter(dict[str, float]) -def _langfuse_callback(host: str) -> dict[str, JsonValue]: +def _langfuse_callback(host: str, **callback_vars: JsonValue) -> dict[str, JsonValue]: return { "callback_name": "langfuse", "callback_type": "success", @@ -867,12 +877,17 @@ def _langfuse_callback(host: str) -> dict[str, JsonValue]: "langfuse_public_key": LANGFUSE_PUBLIC_KEY, "langfuse_secret_key": LANGFUSE_SECRET_KEY, "langfuse_host": host, + **callback_vars, }, } +def _logging_to(langfuse_url: str, **callback_vars: JsonValue) -> dict[str, JsonValue]: + return {"logging": [_langfuse_callback(langfuse_url, **callback_vars)]} + + def _key_with_its_own_logging(scenario: Scenario, langfuse_url: str) -> str: - return scenario.key(metadata={"logging": [_langfuse_callback(langfuse_url)]}) + return scenario.key(metadata=_logging_to(langfuse_url)) def _key_in_a_team_with_logging(scenario: Scenario, langfuse_url: str) -> str: @@ -892,17 +907,33 @@ def _trace_exports(langfuse: Recorder) -> tuple[Request, ...]: return tuple(request for request in langfuse.requests() if request.target.startswith(LANGFUSE_TRACES_PATH)) -def _exported_costs(langfuse: Recorder) -> tuple[float, ...]: - spans: Final = tuple( +def _exported_spans(langfuse: Recorder) -> tuple[Span, ...]: + return tuple( chain.from_iterable( decode_spans(export.body, export.headers.get("content-type", "")) for export in _trace_exports(langfuse) ) ) - return tuple( - _LANGFUSE_COST_DETAILS.validate_json(string_value(span["attributes"][LANGFUSE_COST]))["total"] - for span in spans - if LANGFUSE_COST in span["attributes"] - ) + + +def _span_cost(span: Span) -> float | None: + cost_details: Final = span["attributes"].get(LANGFUSE_COST) + return None if cost_details is None else _LANGFUSE_COST_DETAILS.validate_json(string_value(cost_details))["total"] + + +def _exported_costs(langfuse: Recorder) -> tuple[float, ...]: + return tuple(cost for cost in map(_span_cost, _exported_spans(langfuse)) if cost is not None) + + +def _priced_spans(langfuse: Recorder) -> tuple[Span, ...]: + return tuple(span for span in _exported_spans(langfuse) if (_span_cost(span) or 0.0) > 0) + + +def _billed_once(langfuse: Recorder, seconds: float) -> Span: + priced: Final = eventually(lambda: _priced_spans(langfuse), lambda spans: len(spans) > 0, seconds=seconds) + costs: Final = tuple(_span_cost(span) or 0.0 for span in priced) + assert len(costs) == 1, costs + assert math.isclose(costs[0], PRICE), costs + return priced[0] @pytest.mark.timeout(240) @@ -925,3 +956,133 @@ def test_the_billed_cost_reaches_the_langfuse_project_set_on_the_creating_key_or assert [cost for cost in exported if cost > 0] == [pytest.approx(PRICE)] credentials: Final = base64.b64encode(f"{LANGFUSE_PUBLIC_KEY}:{LANGFUSE_SECRET_KEY}".encode()).decode() assert {export.headers.get("authorization") for export in _trace_exports(langfuse)} == {f"Basic {credentials}"} + + +@pytest.mark.timeout(240) +def test_the_billed_cost_reaches_the_creating_keys_langfuse_project_and_never_the_readers(gateway: Gateway) -> None: + with ( + wire_server(langfuse_sink) as creator_wire, + wire_server(langfuse_sink) as reader_wire, + gateway.scenario() as scenario, + ): + creator_langfuse: Final = Recorder(creator_wire) + reader_langfuse: Final = Recorder(reader_wire) + team_id: Final = scenario.team() + user_id: Final = scenario.member(team_id) + creator: Final = scenario.key(team_id=team_id, user_id=user_id, metadata=_logging_to(creator_wire.url)) + reader: Final = scenario.key(team_id=team_id, user_id=user_id, metadata=_logging_to(reader_wire.url)) + upstream: Final = scripted_upstream(scenario, background_routes("completed")) + response_id: Final = create_background_response(scenario, creator, upstream.model) + reads: Final = tuple(read_response(gateway, reader, response_id) for _ in range(READS)) + assert reads in (settling_reads("completed", READS), free_reads("completed", READS)) + _billed_once(creator_langfuse, SPEND_LANDING_SECONDS) + eventually(lambda: _trace_exports(reader_langfuse), lambda exports: len(exports) > 0) + assert _priced_spans(reader_langfuse) == () + + +MESSAGE_LOGGING: Final[Mapping[str, Mapping[str, JsonValue]]] = { + "logged": {}, + "redacted": {"turn_off_message_logging": "true"}, +} + + +@pytest.mark.timeout(240) +@pytest.mark.parametrize("message_logging", MESSAGE_LOGGING.keys()) +def test_the_billed_observation_carries_the_response_text_only_when_the_creating_key_keeps_message_logging( + gateway: Gateway, message_logging: str +) -> None: + with wire_server(langfuse_sink) as wire, gateway.scenario() as scenario: + langfuse: Final = Recorder(wire) + key: Final = scenario.key(metadata=_logging_to(wire.url, **MESSAGE_LOGGING[message_logging])) + upstream: Final = scripted_upstream(scenario, background_routes("completed")) + response_id: Final = create_background_response(scenario, key, upstream.model) + reads: Final = tuple(read_response(gateway, key, response_id) for _ in range(READS)) + assert reads in (settling_reads("completed", READS), free_reads("completed", READS)) + billed: Final = _billed_once(langfuse, SPEND_LANDING_SECONDS) + assert ("pong" in json.dumps(billed["attributes"])) is (message_logging == "logged"), billed["attributes"] + + +def _generated_key(scenario: Scenario, **fields: JsonValue) -> str: + key: Final = string_value(scenario.gateway.post("/key/generate", fields)["key"]) + scenario.cleanups.callback(delete_key_if_present, scenario.gateway, key) + return key + + +def _delete_team_if_present(gateway: Gateway, team_id: str) -> None: + if gateway.rows('SELECT team_id FROM "LiteLLM_TeamTable" WHERE team_id = %s', (team_id,)): + gateway.post("/team/delete", {"team_ids": [team_id]}) + + +Retirement = Callable[[Scenario, str, str], tuple[str, Callable[[], object]]] + + +def _rotated_with_grace_then_relogged( + scenario: Scenario, billed_url: str, stale_url: str +) -> tuple[str, Callable[[], object]]: + key: Final = _generated_key(scenario, metadata=_logging_to(stale_url)) + + def rotate() -> None: + rotated: Final = string_value( + scenario.gateway.post("/key/regenerate", {"key": key, "grace_period": "1h"})["key"] + ) + scenario.cleanups.callback(delete_key_if_present, scenario.gateway, rotated) + scenario.gateway.post("/key/update", {"key": rotated, "metadata": _logging_to(billed_url)}) + + return key, rotate + + +def _deleted_key(scenario: Scenario, billed_url: str, _stale_url: str) -> tuple[str, Callable[[], object]]: + key: Final = _generated_key(scenario, metadata=_logging_to(billed_url)) + return key, lambda: scenario.gateway.post("/key/delete", {"keys": [key]}) + + +def _deleted_team(scenario: Scenario, billed_url: str, _stale_url: str) -> tuple[str, Callable[[], object]]: + team_id: Final = string_value( + scenario.gateway.post("/team/new", {"team_alias": f"integration-{uuid.uuid4().hex}"})["team_id"] + ) + scenario.cleanups.callback(_delete_team_if_present, scenario.gateway, team_id) + added: Final = scenario.gateway.request("POST", f"/team/{team_id}/callback", _langfuse_callback(billed_url)) + assert added.status_code == 200, added.text + key: Final = _generated_key(scenario, team_id=team_id) + return key, lambda: scenario.gateway.post("/team/delete", {"team_ids": [team_id]}) + + +RETIREMENTS: Final[Mapping[str, Retirement]] = { + "key_rotated_with_grace": _rotated_with_grace_then_relogged, + "key_deleted": _deleted_key, + "team_deleted": _deleted_team, +} + + +@pytest.mark.timeout(240) +@pytest.mark.parametrize("retired", RETIREMENTS.keys()) +def test_a_retired_creators_billed_cost_reaches_its_own_langfuse_project(gateway: Gateway, retired: str) -> None: + with ( + wire_server(langfuse_sink) as billed_wire, + wire_server(langfuse_sink) as stale_wire, + gateway.scenario() as scenario, + ): + billed_langfuse: Final = Recorder(billed_wire) + stale_langfuse: Final = Recorder(stale_wire) + key, retire = RETIREMENTS[retired](scenario, billed_wire.url, stale_wire.url) + upstream: Final = scripted_upstream(scenario, background_routes("queued")) + response_id: Final = create_background_response(scenario, key, upstream.model) + retire() + upstream.script(background_routes("completed")) + _billed_once(billed_langfuse, ONE_POLL_SECONDS + SPEND_LANDING_SECONDS) + assert [row["status"] for row in managed_rows(response_id)] == ["completed"] + assert _priced_spans(stale_langfuse) == () + + +@pytest.mark.timeout(240) +def test_a_creator_whose_langfuse_keys_resolve_to_nothing_is_still_billed_once(gateway: Gateway) -> None: + unset: Final = f"os.environ/INTEGRATION_UNSET_LANGFUSE_{uuid.uuid4().hex.upper()}" + with wire_server(langfuse_sink) as wire, gateway.scenario() as scenario: + key: Final = scenario.key(metadata=_logging_to(wire.url, langfuse_public_key=unset, langfuse_secret_key=unset)) + upstream: Final = scripted_upstream(scenario, background_routes("completed")) + response_id: Final = create_background_response(scenario, key, upstream.model) + reads: Final = tuple(read_response(gateway, key, response_id) for _ in range(READS)) + assert reads in (settling_reads("completed", READS), free_reads("completed", READS)) + assert_charged_exactly(lambda: key_spend(key), PRICE) + assert math.isclose(number(one_charged_row(upstream.model_id)["spend"]), PRICE) + assert [row["status"] for row in managed_rows(response_id)] == ["completed"] From 8bbce867d05b239593890118d31797c77060f8a4 Mon Sep 17 00:00:00 2001 From: mubashir1osmani Date: Wed, 28 Oct 2026 18:58:52 -0400 Subject: [PATCH 23/41] feat(proxy): accept request priority from the x-litellm-priority header (#47618) * feat(proxy): accept request priority from the x-litellm-priority header Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * fix(proxy): only apply x-litellm-priority on scheduler-queued routes Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * fix(proxy): apply x-litellm-priority only where the router schedules the request The header now maps onto priority only for chat and text completions that a router serves, so the Cursor route's responses call and proxies started without a config never forward it to the provider. /queue/chat/completions now honors the header too, with the header winning over a body priority. * test(proxy): use a real ProxyLogging in the priority header routing test * fix(proxy): ignore a malformed priority header under drop_params instead of dropping the body priority * fix(proxy): treat an empty priority header as unset and pass a valid one on as an integer --------- Co-authored-by: Mubashir Osmani Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com> --- litellm/proxy/common_request_processing.py | 12 +- litellm/proxy/litellm_pre_call_utils.py | 3 + litellm/proxy/proxy_server.py | 11 +- litellm/router.py | 15 +- litellm/router_utils/request_priority.py | 9 + .../router_code_coverage.py | 2 + .../proxy/proxy_server/test_routes_queue.py | 48 ++++++ .../proxy/test_common_request_processing.py | 163 ++++++++++++++++++ 8 files changed, 257 insertions(+), 6 deletions(-) diff --git a/litellm/proxy/common_request_processing.py b/litellm/proxy/common_request_processing.py index f5d6417fb79..70e305f46fb 100644 --- a/litellm/proxy/common_request_processing.py +++ b/litellm/proxy/common_request_processing.py @@ -266,6 +266,8 @@ from litellm.proxy.anthropic_endpoints.streaming_model_restamp import ( AnthropicStreamModelRestamper, ) from litellm.proxy.litellm_pre_call_utils import ( + ROUTER_SCHEDULED_ROUTE_TYPES, + SCHEDULER_PRIORITY_HEADER, LiteLLMProxyRequestSetup, add_litellm_data_to_request, get_metadata_variable_name, @@ -2478,14 +2480,20 @@ class ProxyBaseLLMRequestProcessing: self.data["model"] = alias_target aliased_model: Final = self.data.get("model") + if llm_router is not None and route_type in ROUTER_SCHEDULED_ROUTE_TYPES and isinstance(aliased_model, str): + self.data.update( + llm_router.priority_header_update( + model=aliased_model, kwargs=self.data, header=request.headers.get(SCHEDULER_PRIORITY_HEADER) + ) + ) litellm_call_id: Final = resolve_litellm_call_id(request.headers.get("x-litellm-call-id")) self.data["litellm_call_id"] = litellm_call_id bind_telemetry_call_id(litellm_call_id) - DDSpanTagger.tag_call_id(self.data.get("litellm_call_id")) + DDSpanTagger.tag_call_id(litellm_call_id) DDSpanTagger.tag_request( user_api_key_dict=user_api_key_dict, - requested_model=self.data.get("model"), + requested_model=aliased_model, ) ### AUTO STREAM USAGE TRACKING ### diff --git a/litellm/proxy/litellm_pre_call_utils.py b/litellm/proxy/litellm_pre_call_utils.py index 24a31723faa..601d741c189 100644 --- a/litellm/proxy/litellm_pre_call_utils.py +++ b/litellm/proxy/litellm_pre_call_utils.py @@ -118,6 +118,9 @@ _SESSION_ID_VALUE_RE: Final = re.compile(r"^[a-zA-Z0-9_\-]{8,}$") _SHA256_HEX_RE: Final = re.compile(r"^[0-9a-f]{64}$") +SCHEDULER_PRIORITY_HEADER: Final = "x-litellm-priority" +ROUTER_SCHEDULED_ROUTE_TYPES: Final = frozenset({"acompletion", "atext_completion"}) + # W3C Trace Context traceparent header: https://www.w3.org/TR/trace-context/ # e.g. "00-4bf92f3577b34da6a3ce929d0e0e4736-00f067aa0ba902b7-01" _TRACEPARENT_RE: Final = re.compile(r"^[0-9a-f]{2}-([0-9a-f]{32})-[0-9a-f]{16}-[0-9a-f]{2}$", re.IGNORECASE) diff --git a/litellm/proxy/proxy_server.py b/litellm/proxy/proxy_server.py index 64fb2791913..70e1c25a70a 100644 --- a/litellm/proxy/proxy_server.py +++ b/litellm/proxy/proxy_server.py @@ -666,7 +666,11 @@ from litellm.proxy.list_api.common import ( problem_response, request_validation_problem, ) -from litellm.proxy.litellm_pre_call_utils import LiteLLMProxyRequestSetup, add_litellm_data_to_request +from litellm.proxy.litellm_pre_call_utils import ( + SCHEDULER_PRIORITY_HEADER, + LiteLLMProxyRequestSetup, + add_litellm_data_to_request, +) from litellm.proxy.logging_endpoints.callback_logs_endpoints import ( rust_control_plane_router, ) @@ -18481,6 +18485,11 @@ async def async_queue_request( raise HTTPException(status_code=500, detail={"error": CommonProxyErrors.no_llm_router.value}) router: Final = llm_router + data.update( + router.priority_header_update( + model=data["model"], kwargs=data, header=request.headers.get(SCHEDULER_PRIORITY_HEADER) + ) + ) queued_priority, _ = router.split_request_priority(model=data["model"], kwargs=data) if "stream" in data and data["stream"] is True: # use generate_responses to stream responses diff --git a/litellm/router.py b/litellm/router.py index 3e6d82d9b07..5ca09499c3b 100644 --- a/litellm/router.py +++ b/litellm/router.py @@ -359,6 +359,7 @@ from litellm.router_utils.request_priority import ( InvalidPriority, parse_default_priority, priority_error, + priority_header_fields, request_drops_params, resolve_request_priority, unqueueable_priority_message, @@ -5411,14 +5412,22 @@ class Router: request_priority: Final = resolve_request_priority( requested=kwargs.get("priority"), default_priority=self.default_priority, - drops_params=lambda: request_drops_params( - kwargs, self.default_litellm_params, self._request_deployment_params(model=model, kwargs=kwargs) - ), + drops_params=lambda: self._drops_params(model=model, kwargs=kwargs), ) if isinstance(request_priority, InvalidPriority): raise priority_error(model=model, message=request_priority.message) return request_priority, {key: value for key, value in kwargs.items() if key != "priority"} + def priority_header_update( + self, model: str, kwargs: Mapping[str, object], header: str | None + ) -> Mapping[str, int | str]: + return priority_header_fields(header, drops_params=lambda: self._drops_params(model=model, kwargs=kwargs)) + + def _drops_params(self, model: str, kwargs: Mapping[str, object]) -> bool: + return request_drops_params( + kwargs, self.default_litellm_params, self._request_deployment_params(model=model, kwargs=kwargs) + ) + async def _schedule_factory( self, model: str, diff --git a/litellm/router_utils/request_priority.py b/litellm/router_utils/request_priority.py index 14608cc1bcf..61627d4b0cf 100644 --- a/litellm/router_utils/request_priority.py +++ b/litellm/router_utils/request_priority.py @@ -47,6 +47,15 @@ def resolve_request_priority( return priority +def priority_header_fields(header: str | None, drops_params: Callable[[], bool]) -> Mapping[str, int | str]: + if not header: + return {} + priority: Final = _as_priority(header) + if isinstance(priority, InvalidPriority): + return {} if drops_params() else {"priority": header} + return {"priority": priority} + + def parse_default_priority(value: object) -> int | None | InvalidPriority: if value is None: return None diff --git a/tests/code_coverage_tests/router_code_coverage.py b/tests/code_coverage_tests/router_code_coverage.py index eee02a2f6b6..4c9afd26dad 100644 --- a/tests/code_coverage_tests/router_code_coverage.py +++ b/tests/code_coverage_tests/router_code_coverage.py @@ -175,6 +175,8 @@ ignored_function_names = [ "_responses_continuation", # Tested through the Responses mid-stream fallback tests in tests/unit/llms/base_llm/responses/test_transformation.py "_responses_continuation_ceilings", # Tested through the Responses mid-stream fallback tests in tests/unit/llms/base_llm/responses/test_transformation.py "_team_deployments_named", # Tested through the team lookups in the concurrent delete tests in test_router.py + "priority_header_update", # Tested through the priority header tests in test_common_request_processing.py + "_drops_params", # Read by split_request_priority and priority_header_update, tested through both "_responses_event_announces_background_run", # Tested through the background Responses stream tests in test_router.py "_aresponses_background_stream_can_release", # Tested through the background Responses stream tests in test_router.py "_raise_released_stream_error", # Tested through the background Responses stream content-policy tests in test_router.py diff --git a/tests/unit/proxy/proxy_server/test_routes_queue.py b/tests/unit/proxy/proxy_server/test_routes_queue.py index a86314e3063..5cbf89f5e68 100644 --- a/tests/unit/proxy/proxy_server/test_routes_queue.py +++ b/tests/unit/proxy/proxy_server/test_routes_queue.py @@ -191,6 +191,54 @@ def test_queue_chat_completions_rejects_a_request_it_cannot_queue_as_an_invalid_ assert add_request.await_count == 0 +@pytest.mark.parametrize( + ("default_priority", "priority_header", "priority_fields", "expected_priority"), + [ + (5, "0", {"priority": 5}, 0), + (5, "0", {}, 0), + (None, "high", {"priority": 1, "drop_params": True}, 1), + (None, "", {"priority": 1}, 1), + ], + ids=("header-beats-body", "header-only", "drop-params-keeps-body", "empty-header-keeps-body"), +) +def test_queue_chat_completions_queues_at_the_priority_header( + client, + auth_as, + monkeypatch: pytest.MonkeyPatch, + default_priority: int | None, + priority_header: str, + priority_fields: dict[str, object], + expected_priority: int, +) -> None: + router: Final = _queue_router(monkeypatch, default_priority=default_priority) + payload: Final = {"model": "gpt-4", "messages": [{"role": "user", "content": "hi"}], **priority_fields} + with patch.object(router.scheduler, "add_request", wraps=router.scheduler.add_request) as add_request: + with auth_as(): + response: Final = client.post( + "/queue/chat/completions", json=payload, headers={"x-litellm-priority": priority_header} + ) + assert response.status_code == 200, response.text + assert [call.kwargs["request"].priority for call in add_request.await_args_list] == [expected_priority] + assert response.headers.get("x-litellm-priority") == str(expected_priority) + + +def test_queue_chat_completions_rejects_a_malformed_priority_header_like_a_malformed_body_value( + client, auth_as, monkeypatch: pytest.MonkeyPatch +) -> None: + router: Final = _queue_router(monkeypatch, default_priority=5) + payload: Final = {"model": "gpt-4", "messages": [{"role": "user", "content": "hi"}], "priority": 1} + with patch.object(router.scheduler, "add_request", wraps=router.scheduler.add_request) as add_request: + with auth_as(): + response: Final = client.post( + "/queue/chat/completions", json=payload, headers={"x-litellm-priority": "high"} + ) + assert response.status_code == 400, response.text + error: Final = response.json()["error"] + assert (error["type"], error["param"]) == ("invalid_request_error", "priority") + assert "priority must be an integer" in error["message"] + assert add_request.await_count == 0 + + def test_streamed_queue_responses_report_the_priority_the_router_queued_at( client, auth_as, monkeypatch: pytest.MonkeyPatch ) -> None: diff --git a/tests/unit/proxy/test_common_request_processing.py b/tests/unit/proxy/test_common_request_processing.py index 790790459c3..fdcbcbfcd05 100644 --- a/tests/unit/proxy/test_common_request_processing.py +++ b/tests/unit/proxy/test_common_request_processing.py @@ -35,6 +35,7 @@ from litellm.constants import ( ) from litellm.integrations.custom_logger import CustomLogger from litellm.integrations.opentelemetry import UserAPIKeyAuth +from litellm.proxy.common_utils.user_api_key_cache import UserApiKeyCache from litellm.proxy.common_request_processing import ( ProxyBaseLLMRequestProcessing, ProxyConfig, @@ -513,6 +514,168 @@ class TestProxyBaseLLMRequestProcessing: assert exc_info.value.param == "model" add_litellm_data_to_request.assert_not_awaited() + @pytest.mark.asyncio + @pytest.mark.parametrize( + ("path", "route_type", "has_router", "priority_header", "body_priority", "expected_priority"), + [ + ("/v1/chat/completions", "acompletion", True, "0", 5, 0), + ("/v1/completions", "atext_completion", True, "0", 5, 0), + ("/v1/chat/completions", "acompletion", True, "1", None, 1), + ("/v1/completions", "atext_completion", True, "1", None, 1), + ("/v1/chat/completions", "acompletion", True, None, 5, 5), + ("/v1/chat/completions", "acompletion", True, "", 5, 5), + ("/cursor/chat/completions", "aresponses", True, "0", None, None), + ("/v1/responses", "aresponses", True, "0", 5, 5), + ("/v1/embeddings", "aembedding", True, "0", None, None), + ("/v1/chat/completions", "acompletion", False, "0", None, None), + ("/v1/chat/completions", "acompletion", False, "0", 5, 5), + ], + ids=( + "chat-header-beats-body", + "text-completion-header-beats-body", + "chat-header-only", + "text-completion-header-only", + "chat-body-only", + "chat-empty-header-keeps-body", + "cursor-responses-route-ignores-header", + "responses-keeps-body", + "embeddings-ignores-header", + "routerless-ignores-header", + "routerless-keeps-body", + ), + ) + async def test_common_processing_pre_call_logic_applies_priority_header_only_where_the_router_schedules( + self, + path: str, + route_type: Literal["acompletion", "atext_completion", "aresponses", "aembedding"], + has_router: bool, + priority_header: str | None, + body_priority: int | None, + expected_priority: str | int | None, + ) -> None: + from litellm.proxy.proxy_server import ProxyConfig as RealProxyConfig + + header_entries: Final[tuple[tuple[bytes, bytes], ...]] = ( + () if priority_header is None else ((b"x-litellm-priority", priority_header.encode()),) + ) + request: Final = Request( + { + "type": "http", + "method": "POST", + "path": path, + "scheme": "http", + "server": ("localhost", 4000), + "client": ("127.0.0.1", 12345), + "query_string": b"", + "headers": ((b"content-type", b"application/json"), *header_entries), + } + ) + router: Final = ( + Router( + model_list=[ + { + "model_name": "qa-model", + "litellm_params": {"model": "openai/gpt-4o", "api_key": "sk-test", "mock_response": "hi"}, + } + ] + ) + if has_router + else None + ) + processing_obj: Final = ProxyBaseLLMRequestProcessing( + data={ + "model": "qa-model", + "messages": [{"role": "user", "content": "hi"}], + **({"priority": body_priority} if body_priority is not None else {}), + } + ) + returned_data, _ = await processing_obj.common_processing_pre_call_logic( + request=request, + general_settings={}, + user_api_key_dict=ProxyUserAPIKeyAuth(api_key="sk-test"), + proxy_logging_obj=ProxyLogging(user_api_key_cache=UserApiKeyCache()), + proxy_config=RealProxyConfig(), + route_type=route_type, + llm_router=router, + ) + + assert ("priority" in returned_data, returned_data.get("priority")) == ( + expected_priority is not None, + expected_priority, + ) + + @pytest.mark.asyncio + @pytest.mark.parametrize( + ("requested_model", "deployment_drops_params", "body_priority", "expected_priority"), + [ + ("qa-model", True, 5, 5), + ("qa-model", True, None, None), + ("key-alias", True, 5, 5), + ("qa-model", False, 5, "high"), + ], + ids=( + "drop-params-keeps-body", + "drop-params-without-body-leaves-default", + "drop-params-read-from-the-aliased-deployment", + "no-drop-params-passes-header-on-to-400", + ), + ) + async def test_common_processing_pre_call_logic_drops_only_a_malformed_priority_header_under_drop_params( + self, + requested_model: str, + deployment_drops_params: bool, + body_priority: int | None, + expected_priority: str | int | None, + ) -> None: + from litellm.proxy.proxy_server import ProxyConfig as RealProxyConfig + + request: Final = Request( + { + "type": "http", + "method": "POST", + "path": "/v1/chat/completions", + "scheme": "http", + "server": ("localhost", 4000), + "client": ("127.0.0.1", 12345), + "query_string": b"", + "headers": ((b"content-type", b"application/json"), (b"x-litellm-priority", b"high")), + } + ) + router: Final = Router( + model_list=[ + { + "model_name": "qa-model", + "litellm_params": { + "model": "openai/gpt-4o", + "api_key": "sk-test", + "mock_response": "hi", + "drop_params": deployment_drops_params, + }, + } + ] + ) + processing_obj: Final = ProxyBaseLLMRequestProcessing( + data={ + "model": requested_model, + "messages": [{"role": "user", "content": "hi"}], + **({"priority": body_priority} if body_priority is not None else {}), + } + ) + returned_data, _ = await processing_obj.common_processing_pre_call_logic( + request=request, + general_settings={}, + user_api_key_dict=ProxyUserAPIKeyAuth(api_key="sk-test", aliases={"key-alias": "qa-model"}), + proxy_logging_obj=ProxyLogging(user_api_key_cache=UserApiKeyCache()), + proxy_config=RealProxyConfig(), + route_type="acompletion", + llm_router=router, + ) + + assert ("priority" in returned_data, returned_data.get("priority")) == ( + expected_priority is not None, + expected_priority, + ) + @pytest.mark.asyncio @pytest.mark.parametrize("safe_memory_mode", [False, True]) @pytest.mark.parametrize( From 3f526a5017219f477fb24aa752511b9667ce51c0 Mon Sep 17 00:00:00 2001 From: agustin18 Date: Wed, 28 Oct 2026 20:03:03 -0300 Subject: [PATCH 24/41] fix(websearch): strip Claude Code instructional prefix in short-circuit search (#43458) * fix(websearch): strip Claude Code instructional prefix in short-circuit search Signed-off-by: agustin18 * fix(websearch): preserve quotes for unwrapped search queries Signed-off-by: agustin18 * test(websearch): replace mock with pure unit tests and cover empty queries Signed-off-by: agustin18 * style(websearch): fix ruff formatting blank line in handler Signed-off-by: agustin18 * refactor(websearch): enforce Final annotations, prevent rebinding, and retain e2e coverage Signed-off-by: agustin18 * test(websearch): assert server_tool_use block unconditionally Signed-off-by: agustin18 * fix(websearch): search only the query inside Claude Code's web search prompt Match Claude Code's exact "Perform a web search for the query:" prefix instead of a broad regex family, keep any quotes the user typed so exact-phrase searches still work, and test the search request at the HTTP boundary. --------- Signed-off-by: agustin18 Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com> --- .../websearch_interception/handler.py | 5 +- .../test_websearch_short_circuit.py | 64 +++++++++++++++++++ 2 files changed, 68 insertions(+), 1 deletion(-) diff --git a/litellm/integrations/websearch_interception/handler.py b/litellm/integrations/websearch_interception/handler.py index a38b8bed595..6e1ba3eea08 100644 --- a/litellm/integrations/websearch_interception/handler.py +++ b/litellm/integrations/websearch_interception/handler.py @@ -97,6 +97,8 @@ WEBSEARCH_NATIVE_BLOCKS_METADATA_KEY: Final = "websearch_native_blocks" _RESPONSE_CONTENT_FIELD: Final = "content" +_CLAUDE_CODE_WEB_SEARCH_PROMPT_PREFIX: Final = "Perform a web search for the query:" + _ResponseT: Final = TypeVar("_ResponseT") @@ -363,7 +365,8 @@ class WebSearchInterceptionLogger(CustomLogger): get_last_user_message, ) - query: Final = get_last_user_message(cast(list[AllMessageValues], messages)) + last_user_message: Final = get_last_user_message(cast(list[AllMessageValues], messages)) or "" + query: Final = last_user_message.removeprefix(_CLAUDE_CODE_WEB_SEARCH_PROMPT_PREFIX).strip() if not query: return None diff --git a/tests/unit/integrations/websearch_interception/test_websearch_short_circuit.py b/tests/unit/integrations/websearch_interception/test_websearch_short_circuit.py index fdc3771e903..414bcc11602 100644 --- a/tests/unit/integrations/websearch_interception/test_websearch_short_circuit.py +++ b/tests/unit/integrations/websearch_interception/test_websearch_short_circuit.py @@ -6,10 +6,13 @@ and executes the search directly without routing through the backend LLM. """ import json +from typing import Final from unittest.mock import AsyncMock, patch import pytest +import respx +import litellm from litellm.integrations.websearch_interception.handler import ( WebSearchInterceptionLogger, ) @@ -169,6 +172,27 @@ class TestTryShortCircuitSearch: assert result is None + @pytest.mark.asyncio + @pytest.mark.parametrize( + "empty_prompt", + [ + " ", + "Perform a web search for the query: ", + ], + ) + async def test_does_not_short_circuit_empty_or_whitespace_user_message(self, empty_prompt: str): + """User message that is whitespace or empty query → returns None without executing search.""" + logger = WebSearchInterceptionLogger(enabled_providers=["github_copilot"]) + + result = await logger.try_short_circuit_search( + model="github_copilot/claude-sonnet-4", + messages=[{"role": "user", "content": empty_prompt}], + tools=[{"type": "web_search_20250305", "name": "web_search", "max_uses": 8}], + custom_llm_provider="github_copilot", + ) + + assert result is None + @pytest.mark.asyncio async def test_search_failure_returns_error_text(self): """Search failure → response with error message, not exception""" @@ -226,6 +250,46 @@ class TestTryShortCircuitSearch: # Query extraction tests # --------------------------------------------------------------------------- +PERPLEXITY_SEARCH_URL: Final = "https://api.perplexity.ai/search" + + +@pytest.fixture +def perplexity_search(monkeypatch: pytest.MonkeyPatch, respx_mock: respx.MockRouter) -> respx.Route: + monkeypatch.setenv("PERPLEXITYAI_API_KEY", "test-perplexity-key") + monkeypatch.delenv("PERPLEXITY_API_BASE", raising=False) + monkeypatch.setattr(litellm, "disable_aiohttp_transport", True) + litellm.in_memory_llm_clients_cache.flush_cache() + return respx_mock.post(PERPLEXITY_SEARCH_URL).respond( + json={"results": [{"title": "Releases", "url": "https://example.com/releases", "snippet": "Release notes"}]} + ) + + +@pytest.mark.asyncio +@pytest.mark.parametrize( + ("user_message", "searched_query"), + [ + ("Perform a web search for the query: litellm latest release", "litellm latest release"), + ('Perform a web search for the query: "litellm proxy"', '"litellm proxy"'), + ], +) +async def test_short_circuit_searches_the_query_inside_claude_codes_prompt( + perplexity_search: respx.Route, user_message: str, searched_query: str +): + logger: Final = WebSearchInterceptionLogger(enabled_providers=["bedrock"]) + + result: Final = await logger.try_short_circuit_search( + model="bedrock/converse/us.anthropic.claude-sonnet-4-5-20250929-v1:0", + messages=[{"role": "user", "content": user_message}], + tools=[{"type": "web_search_20250305", "name": "web_search", "max_uses": 8}], + custom_llm_provider="bedrock", + ) + + assert perplexity_search.call_count == 1 + assert json.loads(perplexity_search.calls.last.request.content)["query"] == searched_query + assert result is not None + server_tool_use: Final = next(block for block in result["content"] if block["type"] == "server_tool_use") + assert server_tool_use["input"]["query"] == searched_query + # --------------------------------------------------------------------------- # Integration with entry point From d30c702c5cfe98d0f2c6832ce5ef5bb0ca8019de Mon Sep 17 00:00:00 2001 From: joshua-berri Date: Wed, 28 Oct 2026 16:11:51 -0700 Subject: [PATCH 25/41] feat(accounting): add durable event ledger writer (#48883) * feat(accounting): add durable event ledger writer * fix(accounting): validate delegated transaction managers locally * docs(accounting): clarify ledger integration contracts * fix(accounting): return ledger conflicts as values and keep cleanup on the kind index Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * fix(accounting): import assert_never from typing_extensions for Python 3.10 Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * test(accounting): annotate ledger integration test locals as Final Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * test(accounting): finish ledger test conventions --------- Co-authored-by: Joshua Valluru <326636767+joshua-berri@users.noreply.github.com> Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- litellm/proxy/db/accounting_ledger.py | 269 ++++++++++++ .../database/test_accounting_ledger.py | 412 +++++++++++++++++- tests/unit/proxy/db/test_accounting_ledger.py | 103 +++++ 3 files changed, 781 insertions(+), 3 deletions(-) create mode 100644 litellm/proxy/db/accounting_ledger.py create mode 100644 tests/unit/proxy/db/test_accounting_ledger.py diff --git a/litellm/proxy/db/accounting_ledger.py b/litellm/proxy/db/accounting_ledger.py new file mode 100644 index 00000000000..91f8307283e --- /dev/null +++ b/litellm/proxy/db/accounting_ledger.py @@ -0,0 +1,269 @@ +from __future__ import annotations + +import hashlib +import json +from collections.abc import AsyncGenerator, Callable +from contextlib import AbstractAsyncContextManager, asynccontextmanager +from dataclasses import dataclass +from datetime import datetime, timedelta, timezone +from typing import TYPE_CHECKING, Final, Literal, TypeAlias + +from pydantic import AwareDatetime, ConfigDict, JsonValue, TypeAdapter +from typing_extensions import assert_never + +from litellm.accounting import CostEvent, DeliveryTarget, EventKind, InboundUsageEvent, OutboundUsageEvent +from litellm.proxy.db.db_span import db_span +from litellm.proxy.db.routing_prisma_wrapper import writer_wrapper + +if TYPE_CHECKING: + from prisma import Prisma + + from litellm.proxy.utils import PrismaClient + +AccountingEvent: TypeAlias = InboundUsageEvent | OutboundUsageEvent | CostEvent +_JSON: Final = TypeAdapter(dict[str, JsonValue]) +_INSTANT: Final[TypeAdapter[datetime]] = TypeAdapter(AwareDatetime) +_STRINGS: Final = TypeAdapter(tuple[dict[str, str], ...]) +_TIMEOUT: Final = timedelta(seconds=10) + + +class AccountingConflictError(ValueError): + pass + + +_Transaction: TypeAlias = AbstractAsyncContextManager["Prisma"] + + +@dataclass(frozen=True, slots=True) +class _AcceptedEvent: + event_id: str + kind: EventKind + source_event_id: str | None + occurred_at: datetime + payload: str + payload_hash: str + targets: tuple[str, ...] + schema_version: int + version: int + + +def _json(value: object) -> str: + return json.dumps(value, sort_keys=True, separators=(",", ":"), allow_nan=False) + + +def _prepare( + event: AccountingEvent, + targets: tuple[DeliveryTarget, ...], + envelope: str, + schema_version: int, + version: int, +) -> _AcceptedEvent: + if type(schema_version) is not int or type(version) is not int or min(schema_version, version) < 1: + raise ValueError("Accounting versions must be positive integers") + validated: Final = type(event).model_validate_json(event.model_dump_json()) + payload: Final = _JSON.validate_json(envelope) + if payload.get("event") != validated.model_dump(mode="json"): + raise ValueError("Accounting envelope must contain the matching public event") + destinations: Final = tuple(sorted(frozenset(DeliveryTarget(target).value for target in targets))) + serialized: Final = _json(payload) + identity: Final = _json((schema_version, version, destinations, payload)) + kind: Final = ( + EventKind.COST + if isinstance(validated, CostEvent) + else EventKind.INBOUND_USAGE + if isinstance(validated, InboundUsageEvent) + else EventKind.OUTBOUND_USAGE + ) + return _AcceptedEvent( + event_id=validated.event_id, + kind=kind, + source_event_id=validated.usage_event_id if isinstance(validated, CostEvent) else None, + occurred_at=validated.occurred_at if isinstance(validated, CostEvent) else validated.ended_at, + payload=serialized, + payload_hash=hashlib.sha256(identity.encode()).hexdigest(), + targets=destinations, + schema_version=schema_version, + version=version, + ) + + +_INSERT_EVENT: Final = """ +INSERT INTO "LiteLLM_AccountingEvent" + (event_id, kind, source_event_id, schema_version, payload_hash, occurred_at, payload) +VALUES ($1, $2::"AccountingEventKind", $3, $4::int, $5, $6::timestamp, $7::jsonb) +ON CONFLICT (event_id) DO NOTHING RETURNING event_id +""" +_INSERT_DELIVERIES: Final = """ +INSERT INTO "LiteLLM_AccountingDelivery" (event_id, target, version) +SELECT $1, value::"AccountingDeliveryTarget", $3::int FROM jsonb_array_elements_text($2::jsonb) +""" +_LOCK_SOURCE: Final = """ +SELECT event_id FROM "LiteLLM_AccountingEvent" +WHERE event_id = $1 AND kind = 'outbound_usage' FOR UPDATE +""" +_EXISTING_COST: Final = """ +SELECT event_id FROM "LiteLLM_AccountingEvent" WHERE kind = 'cost' AND source_event_id = $1 +""" +_LOCK_EVENT: Final = """ +SELECT payload_hash FROM "LiteLLM_AccountingEvent" WHERE event_id = $1 FOR UPDATE +""" +_COMPLETED_FAMILY: Final = """ +root.source_event_id IS NULL AND root.kind IN ('inbound_usage', 'outbound_usage') +AND root.recorded_at < $1::timestamp +AND (root.kind = 'inbound_usage' OR EXISTS ( + SELECT 1 FROM "LiteLLM_AccountingEvent" AS cost WHERE cost.source_event_id = root.event_id AND cost.kind = 'cost' +)) +AND NOT EXISTS ( + SELECT 1 FROM "LiteLLM_AccountingEvent" AS member + WHERE (member.event_id = root.event_id OR (member.kind = 'cost' AND member.source_event_id = root.event_id)) + AND (member.recorded_at >= $1::timestamp + OR EXISTS (SELECT 1 FROM "LiteLLM_AccountingDelivery" AS delivery + WHERE delivery.event_id = member.event_id AND delivery.state != 'applied')) +) +""" +_RETENTION_CANDIDATES: Final = ( + 'SELECT root.event_id FROM "LiteLLM_AccountingEvent" AS root WHERE ' + + _COMPLETED_FAMILY + + " ORDER BY root.recorded_at, root.event_id LIMIT $2::int FOR UPDATE OF root SKIP LOCKED" +) +_ELIGIBLE_GROUP: Final = ( + 'SELECT root.event_id FROM "LiteLLM_AccountingEvent" AS root WHERE root.event_id = $2 AND ' + _COMPLETED_FAMILY +) +_DELETE_DELIVERIES: Final = """ +DELETE FROM "LiteLLM_AccountingDelivery" WHERE event_id IN ( + SELECT event_id FROM "LiteLLM_AccountingEvent" WHERE event_id = $1 OR (kind = 'cost' AND source_event_id = $1) +) +""" +_DELETE_EVENTS: Final = """ +DELETE FROM "LiteLLM_AccountingEvent" WHERE event_id = $1 OR (kind = 'cost' AND source_event_id = $1) +""" + + +_Conflict: TypeAlias = Literal["missing_source", "competing_cost", "identity_conflict"] + + +async def _accept(db: Prisma, record: _AcceptedEvent) -> Literal["inserted", "duplicate"] | _Conflict: + if record.source_event_id is not None: + source: Final = _STRINGS.validate_python(await db.query_raw(_LOCK_SOURCE, record.source_event_id)) + if not source: + return "missing_source" + costs: Final = _STRINGS.validate_python(tuple(await db.query_raw(_EXISTING_COST, record.source_event_id))) + if any(cost["event_id"] != record.event_id for cost in costs): + return "competing_cost" + inserted: Final = _STRINGS.validate_python( + await db.query_raw( + _INSERT_EVENT, + record.event_id, + record.kind.value, + record.source_event_id, + record.schema_version, + record.payload_hash, + record.occurred_at.astimezone(timezone.utc).isoformat(), + record.payload, + ) + ) + if not inserted: + existing: Final = _STRINGS.validate_python(tuple(await db.query_raw(_LOCK_EVENT, record.event_id))) + if not existing or existing[0]["payload_hash"] != record.payload_hash: + return "identity_conflict" + return "duplicate" + await db.execute_raw(_INSERT_DELIVERIES, record.event_id, _json(record.targets), record.version) + return "inserted" + + +async def _delete_completed(db: Prisma, event_id: str, cutoff: str) -> int: + if not _STRINGS.validate_python(await db.query_raw(_ELIGIBLE_GROUP, cutoff, event_id)): + return 0 + await db.execute_raw(_DELETE_DELIVERIES, event_id) + return await db.execute_raw(_DELETE_EVENTS, event_id) + + +class AccountingLedger: + def __init__(self, transaction: Callable[[], _Transaction]) -> None: + self._transaction: Final = transaction + + @classmethod + def for_client(cls, client: PrismaClient) -> AccountingLedger: + from prisma.client import TransactionManager + + managers: Final = TypeAdapter(TransactionManager, config=ConfigDict(arbitrary_types_allowed=True)) + + @asynccontextmanager + async def transaction() -> AsyncGenerator[Prisma]: + primary: Final = writer_wrapper(client.db) + manager: Final = managers.validate_python(primary.tx(timeout=_TIMEOUT)) + async with db_span("accounting_ledger", "LiteLLM_AccountingEvent"), manager as db: + await db.execute_raw("SET LOCAL statement_timeout = 5000") + await db.execute_raw("SET LOCAL lock_timeout = 1000") + yield db + + return cls(transaction) + + async def insert_event( + self, + event: AccountingEvent, + targets: tuple[DeliveryTarget, ...], + *, + envelope: str, + schema_version: int = 1, + version: int = 1, + ) -> Literal["inserted", "duplicate"]: + """Accept T5/T6's serialized envelope, containing its public event under `event`. + + The immutable acceptance hash includes the target set and delivery version. + Changing either for an accepted event conflicts. T7 must provide a separate + add-target or new-version workflow that preserves receipts and prevents + duplicate contributions; redelivery cannot silently create or reset work. + + The Prisma adapter opens one interactive transaction per event, including + two SET LOCAL statements and a source-row lock for costs. I/O is async, + but awaiting this method inline adds commit latency per event, potentially + several times per request. T5 owns bounded asynchronous capture and an + explicit batching strategy; looping over this method does not batch commits. + Success acknowledges commit, independently of HTTP response acknowledgement. + """ + record: Final = _prepare(event, targets, envelope, schema_version, version) + async with self._transaction() as db: + result: Final = await _accept(db, record) + match result: + case "inserted" | "duplicate": + return result + case "missing_source": + raise AccountingConflictError("Cost source must be a durably accepted outbound usage event") + case "competing_cost": + raise AccountingConflictError("Usage event already has an authoritative cost") + case "identity_conflict": + raise AccountingConflictError("Event identity conflicts with its accepted content or delivery set") + case _: + assert_never(result) + + async def purge_completed( + self, + *, + redelivery_before: datetime | None = None, + recovery_before: datetime | None = None, + limit: int = 100, + ) -> int: + """Delete at most `limit` completed families once both retention owners permit it. + + T8 must supply a coherent recovery watermark including outstanding holds. + Omitting either watermark disables deletion. Each usage row and its cost + remain together so a retained charge never loses its source or vice versa. + Applied receipts are deleted with the family, ending its retained dedup + protection. The caller must guarantee events recorded before + redelivery_before cannot be redelivered through supported paths. + T8's recovery_before must also protect checkpoints, + outstanding holds and unfinished settlements; age alone is insufficient. + """ + if type(limit) is not int or not 1 <= limit <= 1000: + raise ValueError("Accounting retention batch size must be between 1 and 1000") + if redelivery_before is None or recovery_before is None: + return 0 + cutoff: Final = min( + _INSTANT.validate_python(redelivery_before).astimezone(timezone.utc), + _INSTANT.validate_python(recovery_before).astimezone(timezone.utc), + ).isoformat() + async with self._transaction() as db: + roots: Final = _STRINGS.validate_python(tuple(await db.query_raw(_RETENTION_CANDIDATES, cutoff, limit))) + deleted: Final = tuple([await _delete_completed(db, root["event_id"], cutoff) for root in roots]) + return sum(deleted) diff --git a/tests/integration/database/test_accounting_ledger.py b/tests/integration/database/test_accounting_ledger.py index 089ed872dee..0c2ceb62eaf 100644 --- a/tests/integration/database/test_accounting_ledger.py +++ b/tests/integration/database/test_accounting_ledger.py @@ -1,16 +1,48 @@ +from __future__ import annotations + +import asyncio +import json import os import subprocess import sys -from collections.abc import Iterator +from collections.abc import AsyncGenerator, AsyncIterator, Iterator, Mapping, Sequence +from contextlib import asynccontextmanager +from datetime import datetime, timezone +from decimal import Decimal +from itertools import chain from pathlib import Path -from typing import Final +from typing import TYPE_CHECKING, Final, TypeAlias import psycopg import pytest +import pytest_asyncio from psycopg import sql +from pydantic import JsonValue -from litellm.accounting import DeliveryState, DeliveryTarget, EventKind +from litellm.accounting import ( + CostEvent, + DeliveryState, + DeliveryTarget, + EventKind, + InboundUsageEvent, + Metric, + MetricName, + OutboundUsageEvent, + Outcome, + PricingSnapshot, +) +from litellm.proxy.db import accounting_ledger +from litellm.proxy.db.accounting_ledger import AccountingConflictError, AccountingEvent, AccountingLedger +from litellm.types.utils import CallTypes from tests.integration._support.database import scratch_database +from tests.integration._support.prisma_clients import connected_proxy_prisma + +if TYPE_CHECKING: + from prisma import Prisma + + from litellm.proxy.utils import PrismaClient + +_Writer: TypeAlias = tuple["PrismaClient", AccountingLedger] _SCHEMA: Final = Path(__file__).resolve().parents[3] / "litellm-proxy-extras/litellm_proxy_extras/schema.prisma" _EVENT_INSERT: Final = ( @@ -126,3 +158,377 @@ def test_new_delivery_defaults_to_pending_with_no_attempts(ledger: psycopg.Conne (DeliveryTarget.KEY_SPEND.value,), ).fetchone() assert row == ("pending", 0) + + +_AT: Final = datetime(2026, 1, 1, tzinfo=timezone.utc) +_CUTOFF: Final = datetime(2026, 2, 1, tzinfo=timezone.utc) +_TARGETS: Final = (DeliveryTarget.USER_SPEND, DeliveryTarget.TEAM_SPEND) + + +def _usage(event_id: str = "usage", *, inbound: bool = False) -> InboundUsageEvent | OutboundUsageEvent: + fields: Final = dict( + event_id=event_id, + request_id="request", + call_type=CallTypes.completion, + started_at=_AT, + ended_at=_AT, + outcome=Outcome.SUCCEEDED, + dimensions=(), + ) + if inbound: + return InboundUsageEvent(**fields, metrics=(Metric(name=MetricName.REQUESTS, value=Decimal(1)),)) + return OutboundUsageEvent(**fields, attempt_id=event_id, retry_of_attempt_id=None, metrics=None) + + +def _cost(source: str = "usage", event_id: str = "cost", amount: str = "0.125") -> CostEvent: + return CostEvent( + event_id=event_id, + usage_event_id=source, + occurred_at=_AT, + dimensions=(), + charges=(Metric(name=MetricName.SPEND, value=Decimal(amount), unit="USD"),), + snapshot=PricingSnapshot( + calculator_version="fixture", + model="fixture", + provider=None, + pricing_source="cost_map", + priced_model_key="fixture", + total_cost=Decimal(amount), + ), + ) + + +def _envelope(event: AccountingEvent, *, context: Mapping[str, JsonValue] | None = None) -> str: + return json.dumps({"event": event.model_dump(mode="json"), "context": context or {}}) + + +async def _insert( + store: AccountingLedger, + event: AccountingEvent, + targets: tuple[DeliveryTarget, ...] = _TARGETS, + *, + context: Mapping[str, JsonValue] | None = None, +) -> str: + return await store.insert_event(event, targets, envelope=_envelope(event, context=context)) + + +def _rows(database_url: str, statement: str, parameters: tuple[object, ...] = ()) -> Sequence[tuple[object, ...]]: + with psycopg.connect(database_url) as connection: + cursor: Final = connection.execute(statement, parameters) + return cursor.fetchall() if cursor.description is not None else [] + + +def _complete_and_age(database_url: str) -> None: + with psycopg.connect(database_url) as connection: + connection.execute('UPDATE "LiteLLM_AccountingEvent" SET recorded_at = %s', (_AT,)) + connection.execute("UPDATE \"LiteLLM_AccountingDelivery\" SET state = 'applied', applied_at = %s", (_AT,)) + + +@pytest_asyncio.fixture(loop_scope="function") +async def writer(ledger_database: str, monkeypatch: pytest.MonkeyPatch) -> AsyncIterator[_Writer]: + monkeypatch.setenv("DATABASE_URL", ledger_database) + monkeypatch.delenv("DATABASE_URL_READ_REPLICA", raising=False) + _rows(ledger_database, 'TRUNCATE "LiteLLM_AccountingDelivery", "LiteLLM_AccountingEvent"') + async with connected_proxy_prisma(ledger_database) as client: + yield client, AccountingLedger.for_client(client) + + +@pytest.mark.asyncio +async def test_event_and_complete_delivery_set_commit_together(writer: _Writer, ledger_database: str) -> None: + _, store = writer + event: Final = _usage() + assert await _insert(store, event, context={"requested_region": "eu", "quantity": "1.25"}) == "inserted" + rows: Final = _rows(ledger_database, 'SELECT kind::text, source_event_id, payload FROM "LiteLLM_AccountingEvent"') + assert rows == [ + ("outbound_usage", None, json.loads(_envelope(event, context={"requested_region": "eu", "quantity": "1.25"}))) + ] + deliveries: Final = _rows( + ledger_database, 'SELECT target::text, state::text, attempts FROM "LiteLLM_AccountingDelivery"' + ) + assert set(deliveries) == {(target.value, "pending", 0) for target in _TARGETS} + + +@pytest.mark.asyncio +async def test_concurrent_duplicates_preserve_completed_deliveries(writer: _Writer, ledger_database: str) -> None: + _, store = writer + event: Final = _usage(inbound=True) + results: Final = await asyncio.gather(*(_insert(store, event) for _ in range(8))) + assert results.count("inserted") == 1 and results.count("duplicate") == 7 + _complete_and_age(ledger_database) + before: Final = _rows(ledger_database, 'SELECT * FROM "LiteLLM_AccountingDelivery" ORDER BY target') + reordered: Final = json.dumps(json.loads(_envelope(event)), sort_keys=True, indent=2) + assert await store.insert_event(event, tuple(reversed(_TARGETS)) + _TARGETS, envelope=reordered) == "duplicate" + assert _rows(ledger_database, 'SELECT * FROM "LiteLLM_AccountingDelivery" ORDER BY target') == before + assert _rows(ledger_database, 'SELECT count(*) FROM "LiteLLM_AccountingEvent"') == [(1,)] + + +@pytest.mark.asyncio +@pytest.mark.parametrize("change", ["payload", "context", "targets", "schema_version", "version"]) +async def test_conflicting_acceptance_cannot_overwrite_or_add_work( + writer: _Writer, ledger_database: str, change: str +) -> None: + _, store = writer + event: Final = _usage() + await _insert(store, event) + altered: Final = event.model_copy(update={"outcome": Outcome.FAILED}) if change == "payload" else event + targets: Final = (DeliveryTarget.KEY_SPEND,) if change == "targets" else _TARGETS + context: Final = {"override": "0.5"} if change == "context" else {} + options: Final = {change: 2} if change in ("schema_version", "version") else {} + with pytest.raises(AccountingConflictError, match="conflicts"): + await store.insert_event(altered, targets, envelope=_envelope(altered, context=context), **options) + assert _rows(ledger_database, 'SELECT payload FROM "LiteLLM_AccountingEvent"') == [(json.loads(_envelope(event)),)] + assert set(_rows(ledger_database, 'SELECT target::text FROM "LiteLLM_AccountingDelivery"')) == { + (t.value,) for t in _TARGETS + } + + +@pytest.mark.asyncio +async def test_competing_cost_identities_accept_only_one_authoritative_charge( + writer: _Writer, ledger_database: str +) -> None: + _, store = writer + await _insert(store, _usage()) + results: Final = await asyncio.gather( + _insert(store, _cost()), _insert(store, _cost(event_id="other-cost")), return_exceptions=True + ) + assert sum(result == "inserted" for result in results) == 1 + assert sum(isinstance(result, AccountingConflictError) for result in results) == 1 + accepted: Final = _rows( + ledger_database, "SELECT event_id, payload FROM \"LiteLLM_AccountingEvent\" WHERE kind = 'cost'" + ) + assert len(accepted) == 1 + event: Final = CostEvent.model_validate_json(json.dumps(accepted[0][1]["event"])) + assert await _insert(store, event) == "duplicate" + assert event.snapshot.total_cost == Decimal("0.125") + with pytest.raises(AccountingConflictError): + await _insert(store, _cost(event_id=event.event_id, amount="0.5")) + + +@pytest.mark.asyncio +@pytest.mark.parametrize("source", ["missing", "inbound", "cost"]) +async def test_cost_requires_an_existing_outbound_source(writer: _Writer, ledger_database: str, source: str) -> None: + _, store = writer + if source == "inbound": + await _insert(store, _usage("source", inbound=True)) + if source == "cost": + await _insert(store, _usage()) + await _insert(store, _cost(event_id="source")) + with pytest.raises(AccountingConflictError, match="outbound usage"): + await _insert(store, _cost(source="source", event_id="invalid")) + assert _rows(ledger_database, "SELECT count(*) FROM \"LiteLLM_AccountingEvent\" WHERE event_id = 'invalid'") == [ + (0,) + ] + + +@pytest.mark.asyncio +@pytest.mark.parametrize("failure", ["before_commit", "after_commit", "cancellation"]) +async def test_interruption_rolls_back_or_retry_recognizes_commit( + writer: _Writer, ledger_database: str, failure: str +) -> None: + client, store = writer + + @asynccontextmanager + async def interrupted() -> AsyncGenerator[Prisma]: + async with client.writer_db.tx() as tx: + yield tx + if failure == "before_commit": + raise RuntimeError("injected pre-commit failure") + if failure == "cancellation": + raise asyncio.CancelledError + raise RuntimeError("injected lost acknowledgement") + + error: Final = asyncio.CancelledError if failure == "cancellation" else RuntimeError + with pytest.raises(error): + await _insert(AccountingLedger(interrupted), _usage()) + expected: Final = 1 if failure == "after_commit" else 0 + assert _rows(ledger_database, 'SELECT count(*) FROM "LiteLLM_AccountingEvent"') == [(expected,)] + assert _rows(ledger_database, 'SELECT count(*) FROM "LiteLLM_AccountingDelivery"') == [(expected * len(_TARGETS),)] + assert await _insert(store, _usage()) == ("duplicate" if expected else "inserted") + + +@pytest.mark.asyncio +async def test_delivery_insert_failure_rolls_back_event(writer: _Writer, ledger_database: str) -> None: + from prisma.errors import RawQueryError + + _, store = writer + with psycopg.connect(ledger_database) as connection: + connection.execute( + 'ALTER TABLE "LiteLLM_AccountingDelivery" ADD CONSTRAINT reject_delivery CHECK (version < 1)' + ) + try: + with pytest.raises(RawQueryError, match="reject_delivery"): + await _insert(store, _usage()) + assert _rows(ledger_database, 'SELECT count(*) FROM "LiteLLM_AccountingEvent"') == [(0,)] + finally: + with psycopg.connect(ledger_database) as connection: + connection.execute('ALTER TABLE "LiteLLM_AccountingDelivery" DROP CONSTRAINT reject_delivery') + + +@pytest.mark.asyncio +@pytest.mark.parametrize("blocked", ["pending", "parked", "unpriced", "recent_cost", "recovery"]) +async def test_retention_preserves_incomplete_or_recovery_needed_families( + writer: _Writer, ledger_database: str, blocked: str +) -> None: + _, store = writer + await _insert(store, _usage()) + if blocked != "unpriced": + await _insert(store, _cost()) + _complete_and_age(ledger_database) + with psycopg.connect(ledger_database) as connection: + if blocked in ("pending", "parked"): + connection.execute( + "UPDATE \"LiteLLM_AccountingDelivery\" SET state = %s WHERE event_id = 'cost'", (blocked,) + ) + if blocked == "recent_cost": + connection.execute( + "UPDATE \"LiteLLM_AccountingEvent\" SET recorded_at = %s WHERE event_id = 'cost'", (_CUTOFF,) + ) + before: Final = _rows(ledger_database, 'SELECT event_id FROM "LiteLLM_AccountingEvent" ORDER BY event_id') + assert ( + await store.purge_completed( + redelivery_before=_CUTOFF, recovery_before=_AT if blocked == "recovery" else _CUTOFF + ) + == 0 + ) + assert _rows(ledger_database, 'SELECT event_id FROM "LiteLLM_AccountingEvent" ORDER BY event_id') == before + + +@pytest.mark.asyncio +async def test_retention_removes_completed_family_and_receipts_together(writer: _Writer, ledger_database: str) -> None: + _, store = writer + await _insert(store, _usage()) + await _insert(store, _cost()) + _complete_and_age(ledger_database) + assert await store.purge_completed(redelivery_before=_CUTOFF, recovery_before=_CUTOFF) == 2 + assert _rows(ledger_database, 'SELECT count(*) FROM "LiteLLM_AccountingEvent"') == [(0,)] + assert _rows(ledger_database, 'SELECT count(*) FROM "LiteLLM_AccountingDelivery"') == [(0,)] + + +@pytest.mark.asyncio +async def test_retention_respects_batch_bound_and_skips_locked_families(writer: _Writer, ledger_database: str) -> None: + client, store = writer + for name in ("a", "b", "c"): + await _insert(store, _usage(name, inbound=True)) + _complete_and_age(ledger_database) + async with client.writer_db.tx() as tx: + await tx.query_raw("SELECT event_id FROM \"LiteLLM_AccountingEvent\" WHERE event_id = 'a' FOR UPDATE") + assert await store.purge_completed(redelivery_before=_CUTOFF, recovery_before=_CUTOFF, limit=1) == 1 + assert _rows(ledger_database, 'SELECT event_id FROM "LiteLLM_AccountingEvent" ORDER BY event_id') == [ + ("a",), + ("c",), + ] + assert await store.purge_completed(redelivery_before=_CUTOFF, recovery_before=_CUTOFF) == 2 + + +@pytest.mark.asyncio +async def test_retention_makes_progress_past_pending_and_unpriced_events(writer: _Writer, ledger_database: str) -> None: + _, store = writer + await _insert(store, _usage("a", inbound=True)) + await _insert(store, _usage("b")) + await _insert(store, _usage("c", inbound=True)) + _complete_and_age(ledger_database) + _rows(ledger_database, "UPDATE \"LiteLLM_AccountingDelivery\" SET state = 'pending' WHERE event_id = 'a'") + assert await store.purge_completed(redelivery_before=_CUTOFF, recovery_before=_CUTOFF, limit=1) == 1 + assert _rows(ledger_database, 'SELECT event_id FROM "LiteLLM_AccountingEvent" ORDER BY event_id') == [ + ("a",), + ("b",), + ] + + +@pytest.mark.asyncio +async def test_zero_cost_and_empty_delivery_sets_are_durable(writer: _Writer, ledger_database: str) -> None: + _, store = writer + await _insert(store, _usage(), targets=()) + event: Final = _cost(amount="0") + assert await _insert(store, event, targets=()) == "inserted" + assert await _insert(store, event, targets=()) == "duplicate" + assert _rows(ledger_database, 'SELECT count(*) FROM "LiteLLM_AccountingDelivery"') == [(0,)] + _complete_and_age(ledger_database) + assert await store.purge_completed(redelivery_before=_CUTOFF, recovery_before=_CUTOFF) == 2 + + +@pytest.mark.asyncio +async def test_retention_cannot_delete_source_while_cost_acceptance_is_uncommitted( + writer: _Writer, ledger_database: str +) -> None: + client, store = writer + await _insert(store, _usage()) + _complete_and_age(ledger_database) + inserted: Final = asyncio.Event() + release: Final = asyncio.Event() + + @asynccontextmanager + async def paused_commit() -> AsyncGenerator[Prisma]: + async with client.writer_db.tx() as tx: + yield tx + inserted.set() + await release.wait() + + task: Final = asyncio.create_task(_insert(AccountingLedger(paused_commit), _cost())) + try: + await asyncio.wait_for(inserted.wait(), timeout=5) + assert await store.purge_completed(redelivery_before=_CUTOFF, recovery_before=_CUTOFF) == 0 + finally: + release.set() + await task + assert _rows(ledger_database, 'SELECT event_id FROM "LiteLLM_AccountingEvent" ORDER BY event_id') == [ + ("cost",), + ("usage",), + ] + + +@pytest.mark.asyncio +async def test_retention_failure_rolls_back_event_and_receipt_deletion(writer: _Writer, ledger_database: str) -> None: + client, store = writer + await _insert(store, _usage(inbound=True)) + _complete_and_age(ledger_database) + + @asynccontextmanager + async def interrupted() -> AsyncGenerator[Prisma]: + async with client.writer_db.tx() as tx: + yield tx + raise RuntimeError("injected cleanup failure") + + with pytest.raises(RuntimeError, match="cleanup failure"): + await AccountingLedger(interrupted).purge_completed(redelivery_before=_CUTOFF, recovery_before=_CUTOFF) + assert _rows(ledger_database, 'SELECT count(*) FROM "LiteLLM_AccountingEvent"') == [(1,)] + assert _rows(ledger_database, 'SELECT count(*) FROM "LiteLLM_AccountingDelivery"') == [(len(_TARGETS),)] + assert await store.purge_completed(redelivery_before=_CUTOFF, recovery_before=_CUTOFF) == 1 + + +@pytest.mark.asyncio +async def test_ledger_reads_and_writes_use_primary_when_replica_is_configured( + writer: _Writer, ledger_database: str, monkeypatch: pytest.MonkeyPatch +) -> None: + with scratch_database("replica") as replica_url: + monkeypatch.setenv("DATABASE_URL_READ_REPLICA", replica_url) + async with connected_proxy_prisma(ledger_database) as client: + store: Final = AccountingLedger.for_client(client) + await _insert(store, _usage()) + assert await _insert(store, _usage()) == "duplicate" + assert await _insert(store, _cost()) == "inserted" + _complete_and_age(ledger_database) + assert await store.purge_completed(redelivery_before=_CUTOFF, recovery_before=_CUTOFF) == 2 + assert _rows(replica_url, "SELECT to_regclass('public.\"LiteLLM_AccountingEvent\"')") == [(None,)] + + +def _index_conditions(plan: object) -> tuple[tuple[object, object], ...]: + if isinstance(plan, dict): + own: Final = ((plan.get("Index Name"), plan.get("Index Cond")),) if "Index Name" in plan else () + children: Final = (plan.get("Plan"), *plan.get("Plans", ())) + return own + tuple(chain.from_iterable(_index_conditions(child) for child in children if child is not None)) + return tuple(chain.from_iterable(_index_conditions(item) for item in plan)) if isinstance(plan, list) else () + + +@pytest.mark.parametrize("statement", [accounting_ledger._DELETE_EVENTS, accounting_ledger._DELETE_DELIVERIES]) +def test_family_cleanup_lookups_use_the_kind_source_index(ledger_database: str, statement: str) -> None: + explained: Final = "EXPLAIN (FORMAT JSON) " + statement.replace("$1", "%s") + with psycopg.connect(ledger_database) as connection: + connection.execute("SET enable_seqscan = off") + plan: Final = connection.execute(explained, ("usage",) * statement.count("$1")).fetchone() + assert plan is not None + conditions: Final = _index_conditions(plan[0]) + assert any( + name == "LiteLLM_AccountingEvent_kind_source_event_id_idx" and "source_event_id" in str(condition) + for name, condition in conditions + ), conditions diff --git a/tests/unit/proxy/db/test_accounting_ledger.py b/tests/unit/proxy/db/test_accounting_ledger.py new file mode 100644 index 00000000000..3916fe869e6 --- /dev/null +++ b/tests/unit/proxy/db/test_accounting_ledger.py @@ -0,0 +1,103 @@ +import json +from datetime import datetime, timezone +from decimal import Decimal +from typing import Final, Literal, NoReturn + +import pytest + +from litellm.accounting import InboundUsageEvent, Metric, MetricName, Outcome +from litellm.proxy.db.accounting_ledger import AccountingLedger +from litellm.types.utils import CallTypes + + +def _no_transaction() -> NoReturn: + raise AssertionError("Invalid accounting input must not open a database transaction") + + +@pytest.fixture +def event() -> InboundUsageEvent: + return InboundUsageEvent( + event_id="caller", + request_id="caller", + call_type=CallTypes.completion, + started_at=datetime(2026, 1, 1, tzinfo=timezone.utc), + ended_at=datetime(2026, 1, 1, tzinfo=timezone.utc), + outcome=Outcome.SUCCEEDED, + dimensions=(), + metrics=(Metric(name=MetricName.REQUESTS, value=Decimal(1)),), + ) + + +@pytest.mark.asyncio +@pytest.mark.parametrize("envelope", ["{}", "[]", '{"event":null}', '{"event":{}}', "not json"]) +async def test_invalid_envelopes_fail_before_database_access(event: InboundUsageEvent, envelope: str) -> None: + with pytest.raises(ValueError, match=r"Accounting envelope|validation error"): + await AccountingLedger(_no_transaction).insert_event(event, (), envelope=envelope) + + +@pytest.mark.asyncio +@pytest.mark.parametrize("version", [0, -1, True, 1.5]) +@pytest.mark.parametrize("field", ["version", "schema_version"]) +async def test_invalid_versions_fail_before_database_access( + event: InboundUsageEvent, version: int | float, field: Literal["version", "schema_version"] +) -> None: + with pytest.raises(ValueError, match="positive integers"): + await AccountingLedger(_no_transaction).insert_event( + event, + (), + envelope="{}", + **{field: version}, # pyright: ignore[reportArgumentType] # Exercise invalid runtime versions + ) + + +@pytest.mark.asyncio +async def test_constructed_invalid_events_are_revalidated(event: InboundUsageEvent) -> None: + invalid: Final = event.model_copy(update={"event_id": ""}) + with pytest.raises(ValueError, match="event_id"): + await AccountingLedger(_no_transaction).insert_event(invalid, (), envelope="{}") + + +@pytest.mark.asyncio +@pytest.mark.parametrize("limit", [0, -1, 1001, True, 1.5]) +async def test_retention_rejects_invalid_batch_bounds(limit: int | float) -> None: + with pytest.raises(ValueError, match="batch size"): + await AccountingLedger(_no_transaction).purge_completed( + limit=limit # pyright: ignore[reportArgumentType] # Exercise invalid runtime batch sizes + ) + + +@pytest.mark.asyncio +@pytest.mark.parametrize("missing", ["both", "redelivery", "recovery"]) +async def test_retention_requires_both_safety_watermarks(missing: Literal["both", "redelivery", "recovery"]) -> None: + cutoff: Final = datetime(2026, 1, 1, tzinfo=timezone.utc) + deleted: Final = await AccountingLedger(_no_transaction).purge_completed( + redelivery_before=cutoff if missing == "recovery" else None, + recovery_before=cutoff if missing == "redelivery" else None, + ) + assert deleted == 0 + + +@pytest.mark.asyncio +async def test_retention_rejects_ambiguous_naive_watermarks() -> None: + with pytest.raises(ValueError, match="timezone"): + await AccountingLedger(_no_transaction).purge_completed( + redelivery_before=datetime(2026, 1, 1), recovery_before=datetime(2026, 1, 1, tzinfo=timezone.utc) + ) + + +@pytest.mark.asyncio +async def test_nonfinite_context_is_rejected_before_database_access(event: InboundUsageEvent) -> None: + envelope: Final = json.dumps({"event": event.model_dump(mode="json"), "context": {"price": float("nan")}}) + with pytest.raises(ValueError, match="Out of range float"): + await AccountingLedger(_no_transaction).insert_event(event, (), envelope=envelope) + + +@pytest.mark.asyncio +async def test_unknown_delivery_target_is_rejected_before_database_access(event: InboundUsageEvent) -> None: + envelope: Final = json.dumps({"event": event.model_dump(mode="json")}) + with pytest.raises(ValueError, match="not a valid DeliveryTarget"): + await AccountingLedger(_no_transaction).insert_event( + event, + ("unknown",), # pyright: ignore[reportArgumentType] # Exercise an unknown runtime target + envelope=envelope, + ) From f89b9c61504801fcc99e40e37af532bb38bbfa30 Mon Sep 17 00:00:00 2001 From: rj-cairn <349020452+rj-cairn@users.noreply.github.com> Date: Wed, 28 Oct 2026 16:12:10 -0700 Subject: [PATCH 26/41] refactor(types): replace Any with proven types in 28 files (#48903) * refactor(types): replace Any with proven types in 52 files Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * fix(types): use typing_extensions.TypedDict for validated seams Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * fix(types): revert strict Anthropic response validation Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * fix(types): accept a null finish reason from Bedrock Cohere and TwelveLabs Pegasus * test(bedrock): annotate the Pegasus null finish reason test --------- Co-authored-by: mateo Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com> --- litellm/litellm_core_utils/litellm_logging.py | 4 +- litellm/litellm_core_utils/safe_json_dumps.py | 32 +++++++-------- ...mazon_twelvelabs_pegasus_transformation.py | 15 ++++++- .../base_invoke_transformation.py | 39 +++++++++++++++---- .../llms/bedrock/realtime/transformation.py | 3 +- .../llms/dashscope/embed/transformation.py | 30 +++++++++++--- litellm/llms/databricks/streaming_utils.py | 9 ++++- litellm/llms/gigachat/authenticator.py | 15 ++++++- litellm/llms/openai/realtime/handler.py | 7 +++- .../perplexity/embedding/transformation.py | 33 ++++++++++++---- .../vertex_gemma_models/transformation.py | 22 ++++++++--- .../volcengine/responses/transformation.py | 5 ++- litellm/proxy/client/models.py | 9 +++-- .../guardrail_hooks/openai/moderations.py | 2 +- litellm/proxy/litellm_pre_call_utils.py | 12 +++--- .../auto_router_endpoints.py | 5 ++- .../router_settings_endpoints.py | 4 +- .../management_helpers/team_roster_sync.py | 3 +- .../mcp_server/discoverable_endpoints.py | 3 +- .../pass_through_endpoints.py | 24 ++++++++---- .../spend_tracking/key_metadata_recovery.py | 5 ++- .../secret_managers/aws_secret_manager_v2.py | 23 ++++++++--- .../integrations/rag/bedrock_knowledgebase.py | 10 ++--- litellm/types/llms/bedrock_invoke_agents.py | 10 ++--- litellm/types/llms/gemini.py | 11 +++--- litellm/types/llms/langgraph.py | 11 +++--- litellm/types/mcp_server/tool_registry.py | 10 ++--- .../guardrails/guardrail_hooks/enkryptai.py | 10 ++--- .../test_base_invoke_transformation.py | 27 +++++++++++++ .../test_twelvelabs_pegasus_transformation.py | 28 +++++++++++++ 30 files changed, 308 insertions(+), 113 deletions(-) diff --git a/litellm/litellm_core_utils/litellm_logging.py b/litellm/litellm_core_utils/litellm_logging.py index 80f1f8f5eca..8b66c27b930 100644 --- a/litellm/litellm_core_utils/litellm_logging.py +++ b/litellm/litellm_core_utils/litellm_logging.py @@ -3481,7 +3481,7 @@ class Logging(LiteLLMLoggingBaseClass): start_time=start_time, end_time=end_time, print_verbose=print_verbose, - level=LogfireLevel.INFO.value, + level=LogfireLevel.INFO, ) if callback == "lunary" and lunaryLogger is not None: @@ -4442,7 +4442,7 @@ class Logging(LiteLLMLoggingBaseClass): response_obj=result, start_time=start_time, end_time=end_time, - level=LogfireLevel.ERROR.value, + level=LogfireLevel.ERROR, print_verbose=print_verbose, ) diff --git a/litellm/litellm_core_utils/safe_json_dumps.py b/litellm/litellm_core_utils/safe_json_dumps.py index 63242a580e7..24c401281ef 100644 --- a/litellm/litellm_core_utils/safe_json_dumps.py +++ b/litellm/litellm_core_utils/safe_json_dumps.py @@ -34,7 +34,7 @@ def safe_json_structure( def _transform(key: str | None, value: str) -> str: return value if value_transform is None else value_transform(key, value) - def _serialize(obj: object, seen: set[int], depth: int, key: str | None = None) -> Any: + def _serialize(obj: object, seen: set[int], depth: int, key: str | None = None) -> object: # Check for maximum depth. if depth > max_depth: return "MaxDepthExceeded" @@ -48,32 +48,32 @@ def safe_json_structure( if id(obj) in seen: return "CircularReference Detected" seen.add(id(obj)) - result: dict | list | tuple | set | str if isinstance(obj, dict): - result = {} - for k, v in obj.items(): - if isinstance(k, (str)): - clean_k = k.replace("\x00", "") if "\x00" in k else k - result[clean_k] = _serialize(v, seen, depth + 1, clean_k) + mapping_result: Final = { + (clean_k := k.replace("\x00", "") if "\x00" in k else k): _serialize(v, seen, depth + 1, clean_k) + for k, v in obj.items() + if isinstance(k, str) + } seen.remove(id(obj)) - return result + return mapping_result elif isinstance(obj, list): - result = [_serialize(item, seen, depth + 1, key) for item in obj] + list_result: Final = [_serialize(item, seen, depth + 1, key) for item in obj] seen.remove(id(obj)) - return result + return list_result elif isinstance(obj, tuple): - result = tuple(_serialize(item, seen, depth + 1, key) for item in obj) + tuple_result: Final = tuple(_serialize(item, seen, depth + 1, key) for item in obj) seen.remove(id(obj)) - return result + return tuple_result elif isinstance(obj, set): - result = sorted([_serialize(item, seen, depth + 1, key) for item in obj]) + sortable_values: Final[list[Any]] = [_serialize(item, seen, depth + 1, key) for item in obj] + set_result: Final = sorted(sortable_values) seen.remove(id(obj)) - return result + return set_result elif isinstance(obj, BaseModel): dumped: Final = obj.model_dump() - result = _serialize(dumped, seen, depth + 1, key) + model_result: Final = _serialize(dumped, seen, depth + 1, key) seen.remove(id(obj)) - return result + return model_result else: # Fall back to string conversion for non-serializable objects. try: diff --git a/litellm/llms/bedrock/chat/invoke_transformations/amazon_twelvelabs_pegasus_transformation.py b/litellm/llms/bedrock/chat/invoke_transformations/amazon_twelvelabs_pegasus_transformation.py index f7566ac4923..dc14ef005b5 100644 --- a/litellm/llms/bedrock/chat/invoke_transformations/amazon_twelvelabs_pegasus_transformation.py +++ b/litellm/llms/bedrock/chat/invoke_transformations/amazon_twelvelabs_pegasus_transformation.py @@ -10,6 +10,8 @@ import time from typing import TYPE_CHECKING, Any, Final import httpx +from pydantic import ConfigDict, TypeAdapter, with_config +from typing_extensions import NotRequired, ReadOnly, TypedDict import litellm from litellm._logging import verbose_logger @@ -34,6 +36,15 @@ else: LiteLLMLoggingObj = Any +@with_config(ConfigDict(extra="allow", strict=True)) +class _PegasusResponse(TypedDict): + message: ReadOnly[NotRequired[str | None]] + finishReason: ReadOnly[NotRequired[str | None]] + + +_PEGASUS_RESPONSE: Final = TypeAdapter(_PegasusResponse) + + class AmazonTwelveLabsPegasusConfig(AmazonInvokeConfig, BaseConfig): """ Handles transforming OpenAI-style requests into Bedrock InvokeModel requests for @@ -207,7 +218,7 @@ class AmazonTwelveLabsPegasusConfig(AmazonInvokeConfig, BaseConfig): ModelResponse with choices[0].message.content and finish_reason """ try: - completion_response: Final = raw_response.json() + completion_response: Final = _PEGASUS_RESPONSE.validate_python(raw_response.json()) except Exception as e: raise BedrockError( message=f"Error parsing response: {raw_response.text}, error: {e}", @@ -224,7 +235,7 @@ class AmazonTwelveLabsPegasusConfig(AmazonInvokeConfig, BaseConfig): message_content: Final = completion_response.get("message", "") # Extract finish reason and map to LiteLLM format - finish_reason_raw: Final = completion_response.get("finishReason", "stop") + finish_reason_raw: Final = completion_response.get("finishReason") or "stop" finish_reason: Final = map_finish_reason(finish_reason_raw) # Set the response content diff --git a/litellm/llms/bedrock/chat/invoke_transformations/base_invoke_transformation.py b/litellm/llms/bedrock/chat/invoke_transformations/base_invoke_transformation.py index 51b5a99272e..852179e8caa 100644 --- a/litellm/llms/bedrock/chat/invoke_transformations/base_invoke_transformation.py +++ b/litellm/llms/bedrock/chat/invoke_transformations/base_invoke_transformation.py @@ -1,11 +1,12 @@ import copy import json import time -from collections.abc import Mapping +from collections.abc import Mapping, Sequence from typing import TYPE_CHECKING, Any, Final, cast, get_args import httpx -from pydantic import TypeAdapter, ValidationError +from pydantic import ConfigDict, TypeAdapter, ValidationError, with_config +from typing_extensions import NotRequired, ReadOnly, TypedDict import litellm from litellm._logging import verbose_logger @@ -48,6 +49,27 @@ from litellm.llms.bedrock.base_aws_llm import BaseAWSLLM _GUARDRAIL_CONFIG_VALIDATOR: Final["TypeAdapter[GuardrailConfigBlock]"] = TypeAdapter(GuardrailConfigBlock) + +@with_config(ConfigDict(extra="allow", strict=True)) +class _CohereGeneration(TypedDict): + text: ReadOnly[str | None] + finish_reason: ReadOnly[str | None] + + +@with_config(ConfigDict(extra="allow", strict=True)) +class _CohereInvokeResponse(TypedDict): + text: NotRequired[ReadOnly[str | None]] + generations: NotRequired[ReadOnly[Sequence[_CohereGeneration]]] + + +@with_config(ConfigDict(extra="allow", strict=True)) +class _LlamaInvokeResponse(TypedDict): + generation: ReadOnly[str | None] + + +_COHERE_INVOKE_RESPONSE: Final = TypeAdapter(_CohereInvokeResponse) +_LLAMA_INVOKE_RESPONSE: Final = TypeAdapter(_LlamaInvokeResponse) + _GUARDRAIL_CONFIG_EXPECTED_FORMAT: Final = ( "{'guardrailIdentifier': str, 'guardrailVersion': str, 'trace': 'enabled'|'disabled'|'enabled_full'}" ) @@ -310,12 +332,13 @@ class AmazonInvokeConfig(BaseConfig, BaseAWSLLM): outputText: str | None = None try: if provider == "cohere": - if "text" in completion_response: - outputText = completion_response["text"] - elif "generations" in completion_response: - outputText = completion_response["generations"][0]["text"] + cohere_response: Final = _COHERE_INVOKE_RESPONSE.validate_python(completion_response) + if "text" in cohere_response: + outputText = cohere_response["text"] + elif "generations" in cohere_response: + outputText = cohere_response["generations"][0]["text"] model_response.choices[0].finish_reason = map_finish_reason( - completion_response["generations"][0]["finish_reason"] + cohere_response["generations"][0]["finish_reason"] or "stop" ) elif provider == "anthropic": return litellm.AmazonAnthropicClaudeConfig().transform_response( @@ -361,7 +384,7 @@ class AmazonInvokeConfig(BaseConfig, BaseAWSLLM): elif provider == "ai21": outputText = completion_response.get("completions")[0].get("data").get("text") elif provider == "meta" or provider == "llama" or provider == "deepseek_r1": - outputText = completion_response["generation"] + outputText = _LLAMA_INVOKE_RESPONSE.validate_python(completion_response)["generation"] elif provider == "mistral": outputText = litellm.AmazonMistralConfig.get_outputText(completion_response, model_response) else: # amazon titan diff --git a/litellm/llms/bedrock/realtime/transformation.py b/litellm/llms/bedrock/realtime/transformation.py index 9cb38822881..952d8858f78 100644 --- a/litellm/llms/bedrock/realtime/transformation.py +++ b/litellm/llms/bedrock/realtime/transformation.py @@ -53,6 +53,7 @@ from litellm.types.realtime import ( RealtimeResponseTypedDict, ) +_JSON_OBJECT_MESSAGE: Final = TypeAdapter(dict[str, JsonValue], config=ConfigDict(strict=True)) _JSON_VALUE: Final = TypeAdapter(object) _JSON_OBJECT: Final = TypeAdapter(Mapping[str, object]) @@ -856,7 +857,7 @@ class BedrockRealtimeConfig(BaseRealtimeConfig): List of Bedrock format messages (JSON strings) """ try: - json_message: Final[dict[str, JsonValue]] = json.loads(message) + json_message: Final = _JSON_OBJECT_MESSAGE.validate_python(json.loads(message)) except json.JSONDecodeError: verbose_logger.warning("Invalid JSON message: %s", message[:200]) return [] diff --git a/litellm/llms/dashscope/embed/transformation.py b/litellm/llms/dashscope/embed/transformation.py index d0c2b4dff0c..35132346261 100644 --- a/litellm/llms/dashscope/embed/transformation.py +++ b/litellm/llms/dashscope/embed/transformation.py @@ -14,6 +14,8 @@ Docs - https://help.aliyun.com/zh/model-studio/text-embedding-synchronous-api from typing import Final import httpx +from pydantic import ConfigDict, TypeAdapter, with_config +from typing_extensions import NotRequired, ReadOnly, TypedDict from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj from litellm.litellm_core_utils.url_utils import ensure_api_route @@ -28,6 +30,23 @@ from ..common_utils import DashScopeError DEFAULT_API_BASE: Final = "https://dashscope.aliyuncs.com/compatible-mode/v1" +@with_config(ConfigDict(extra="allow", strict=True)) +class _DashScopeUsage(TypedDict, total=False): + prompt_tokens: NotRequired[ReadOnly[int]] + total_tokens: NotRequired[ReadOnly[int]] + + +@with_config(ConfigDict(extra="allow", strict=True)) +class _DashScopeEmbeddingResponse(TypedDict, total=False): + data: NotRequired[ReadOnly[list[object]]] + model: NotRequired[ReadOnly[str | None]] + usage: NotRequired[ReadOnly[_DashScopeUsage | None]] + id: NotRequired[ReadOnly[object]] + + +_DASHSCOPE_EMBEDDING_RESPONSE: Final = TypeAdapter(_DashScopeEmbeddingResponse) + + class DashScopeEmbeddingConfig(BaseEmbeddingConfig): """ Reference: https://help.aliyun.com/zh/model-studio/text-embedding-synchronous-api @@ -152,11 +171,12 @@ class DashScopeEmbeddingConfig(BaseEmbeddingConfig): message=message, ) + parsed_response: Final = _DASHSCOPE_EMBEDDING_RESPONSE.validate_python(response_json) model_response.object = "list" - model_response.data = response_json.get("data", []) - model_response.model = response_json.get("model", model) + model_response.data = parsed_response.get("data", []) + model_response.model = parsed_response.get("model", model) - usage: Final = response_json.get("usage") or {} + usage: Final = parsed_response.get("usage") or {} prompt_tokens: Final = usage.get("prompt_tokens", 0) total_tokens: Final = usage.get("total_tokens", prompt_tokens) setattr( @@ -169,8 +189,8 @@ class DashScopeEmbeddingConfig(BaseEmbeddingConfig): ), ) - if "id" in response_json: - setattr(model_response, "id", response_json["id"]) + if "id" in parsed_response: + setattr(model_response, "id", parsed_response["id"]) return model_response diff --git a/litellm/llms/databricks/streaming_utils.py b/litellm/llms/databricks/streaming_utils.py index eca87f9dd5f..cc85012b803 100644 --- a/litellm/llms/databricks/streaming_utils.py +++ b/litellm/llms/databricks/streaming_utils.py @@ -1,6 +1,8 @@ import json from typing import Final +from pydantic import ConfigDict, TypeAdapter + import litellm from litellm import verbose_logger from litellm.types.llms.openai import ( @@ -23,6 +25,9 @@ def _usage_block(usage: Usage) -> ChatCompletionUsageBlock: ) +_JSON_CHUNK: Final = TypeAdapter(dict[str, object], config=ConfigDict(strict=True)) + + class ModelResponseIterator: def __init__(self, streaming_response, sync_stream: bool): self.streaming_response = streaming_response @@ -99,7 +104,7 @@ class ModelResponseIterator: self.response_iterator = self.streaming_response chunk: Final = self._next_sse_data() try: - json_chunk: Final = json.loads(chunk) + json_chunk: Final = _JSON_CHUNK.validate_python(json.loads(chunk)) return self.chunk_parser(chunk=json_chunk) except ValueError as e: verbose_logger.debug( @@ -126,7 +131,7 @@ class ModelResponseIterator: if chunk == "[DONE]": raise StopAsyncIteration try: - json_chunk: Final = json.loads(chunk) + json_chunk: Final = _JSON_CHUNK.validate_python(json.loads(chunk)) return self.chunk_parser(chunk=json_chunk) except ValueError as e: verbose_logger.debug( diff --git a/litellm/llms/gigachat/authenticator.py b/litellm/llms/gigachat/authenticator.py index d43e203d7a1..f37c6e88896 100644 --- a/litellm/llms/gigachat/authenticator.py +++ b/litellm/llms/gigachat/authenticator.py @@ -12,6 +12,8 @@ from types import MappingProxyType from typing import Final import httpx +from pydantic import ConfigDict, TypeAdapter, with_config +from typing_extensions import NotRequired, ReadOnly, TypedDict from litellm._logging import verbose_logger from litellm.caching.caching import InMemoryCache @@ -37,6 +39,17 @@ _token_cache: Final = InMemoryCache() _NO_LITELLM_PARAMS: Final[Mapping[str, object]] = MappingProxyType({}) +@with_config(ConfigDict(extra="allow", strict=True)) +class _GigaChatTokenResponse(TypedDict): + tok: NotRequired[ReadOnly[str | None]] + access_token: NotRequired[ReadOnly[str | None]] + exp: NotRequired[ReadOnly[int | float | str | None]] + expires_at: NotRequired[ReadOnly[int | float | str | None]] + + +_TOKEN_RESPONSE: Final = TypeAdapter(_GigaChatTokenResponse) + + class GigaChatAuthError(BaseLLMException): """GigaChat authentication error.""" @@ -230,7 +243,7 @@ async def _request_token_async( def _parse_token_response(response: httpx.Response) -> tuple[str, int]: """Parse OAuth token response.""" - data: Final = response.json() + data: Final = _TOKEN_RESPONSE.validate_python(response.json()) # GigaChat returns either 'tok'/'exp' or 'access_token'/'expires_at' access_token: Final = data.get("tok") or data.get("access_token") diff --git a/litellm/llms/openai/realtime/handler.py b/litellm/llms/openai/realtime/handler.py index 080fc3d24c9..b7e4cc6868f 100644 --- a/litellm/llms/openai/realtime/handler.py +++ b/litellm/llms/openai/realtime/handler.py @@ -5,7 +5,7 @@ This requires websockets, and is currently only supported on LiteLLM Proxy. """ import ssl -from typing import Any, Final, cast +from typing import TYPE_CHECKING, Final, cast from litellm._logging import redact_string, verbose_logger from litellm.constants import REALTIME_WEBSOCKET_MAX_MESSAGE_SIZE_BYTES @@ -23,6 +23,9 @@ from ....llms.custom_httpx.http_handler import get_shared_realtime_ssl_context from ..common_utils import is_openai_backed_api_base from ..openai import OpenAIChatCompletion +if TYPE_CHECKING: + from fastapi import WebSocket + class OpenAIRealtime(OpenAIChatCompletion): """ @@ -131,7 +134,7 @@ class OpenAIRealtime(OpenAIChatCompletion): async def async_realtime( self, model: str, - websocket: Any, + websocket: "WebSocket", logging_obj: LiteLLMLogging, api_base: str | None = None, api_key: str | None = None, diff --git a/litellm/llms/perplexity/embedding/transformation.py b/litellm/llms/perplexity/embedding/transformation.py index 7b4da933976..147d344795d 100644 --- a/litellm/llms/perplexity/embedding/transformation.py +++ b/litellm/llms/perplexity/embedding/transformation.py @@ -12,12 +12,14 @@ This module decodes them into float arrays for OpenAI-compatible responses. """ import base64 +import builtins import struct -from collections.abc import Mapping -from typing import Final +from collections.abc import Mapping, Sequence +from typing import Final, Literal import httpx -from pydantic import ConfigDict, TypeAdapter +from pydantic import ConfigDict, TypeAdapter, with_config +from typing_extensions import ReadOnly, TypedDict from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj from litellm.litellm_core_utils.url_utils import join_api_base @@ -31,6 +33,17 @@ _JSON_OBJECT: Final = TypeAdapter(Mapping[str, object], config=ConfigDict(hide_i _INT: Final = TypeAdapter(int) +@with_config(ConfigDict(extra="allow", strict=True)) +class _PerplexityEmbeddingResponse(TypedDict, total=False): + model: ReadOnly[str | None] + object: ReadOnly[Literal["list"]] + data: ReadOnly[Sequence[builtins.object] | Mapping[str, builtins.object] | Literal[""]] + usage: ReadOnly[builtins.object] + + +_EMBEDDING_RESPONSE_ADAPTER: Final = TypeAdapter(_PerplexityEmbeddingResponse) + + class PerplexityEmbeddingError(BaseLLMException): def __init__( self, @@ -153,14 +166,18 @@ class PerplexityEmbeddingConfig(BaseEmbeddingConfig): litellm_params: dict = {}, ) -> EmbeddingResponse: try: - raw_response_json: Final = raw_response.json() + raw_response_json: Final[object] = raw_response.json() except Exception: raise PerplexityEmbeddingError(message=raw_response.text, status_code=raw_response.status_code) - model_response.model = raw_response_json.get("model", model) - model_response.object = raw_response_json.get("object", "list") + response_data: Final[_PerplexityEmbeddingResponse] = _EMBEDDING_RESPONSE_ADAPTER.validate_python( + raw_response_json + ) - raw_data: Final = raw_response_json.get("data", []) + model_response.model = response_data.get("model", model) + model_response.object = response_data.get("object", "list") + + raw_data: Final = response_data.get("data", []) decoded_data: Final[list[dict[str, object]]] = [] for item in raw_data: decoded_item = dict(_JSON_OBJECT.validate_python(item)) @@ -168,7 +185,7 @@ class PerplexityEmbeddingConfig(BaseEmbeddingConfig): decoded_data.append(decoded_item) model_response.data = decoded_data - usage_data: Final = _JSON_OBJECT.validate_python(raw_response_json.get("usage", {})) + usage_data: Final = _JSON_OBJECT.validate_python(response_data.get("usage", {})) usage: Final = Usage( prompt_tokens=_INT.validate_python( usage_data.get("prompt_tokens", 0) or usage_data.get("total_tokens", 0) or 0 diff --git a/litellm/llms/vertex_ai/vertex_gemma_models/transformation.py b/litellm/llms/vertex_ai/vertex_gemma_models/transformation.py index 9fa86ec30f5..ec2fd16f93e 100644 --- a/litellm/llms/vertex_ai/vertex_gemma_models/transformation.py +++ b/litellm/llms/vertex_ai/vertex_gemma_models/transformation.py @@ -9,10 +9,11 @@ The actual message transformation reuses OpenAIGPTConfig since Gemma uses OpenAI """ from collections.abc import Callable -from typing import TYPE_CHECKING, Any, Final, cast +from typing import TYPE_CHECKING, Final, cast import httpx -from pydantic import ValidationError +from pydantic import ConfigDict, TypeAdapter, ValidationError +from typing_extensions import NotRequired, ReadOnly, TypeAliasType, TypedDict from litellm.llms.base_llm.chat.transformation import BaseLLMException from litellm.llms.custom_httpx.http_handler import ( @@ -32,6 +33,17 @@ if TYPE_CHECKING: from litellm.litellm_core_utils.tokenizer import Encoding as Tokenizer +class _PredictionsEnvelope(TypedDict): + predictions: NotRequired[ReadOnly[dict[str, object]]] + + +_PredictionsEnvelopeType = TypeAliasType("_PredictionsEnvelopeType", _PredictionsEnvelope) +_PREDICTIONS_ENVELOPE: Final[TypeAdapter[_PredictionsEnvelope]] = TypeAdapter( + _PredictionsEnvelopeType, + config=ConfigDict(strict=True, extra="allow"), +) + + def parse_vertex_gemma_container_error(predictions: object) -> VertexGemmaContainerError | None: try: return VertexGemmaContainerError.model_validate(predictions) @@ -142,7 +154,7 @@ class VertexGemmaConfig(OpenAIGPTConfig): def _unwrap_predictions_response( self, - response_json: dict[str, Any], + response_json: _PredictionsEnvelope, ) -> dict[str, object]: """ Unwrap the Vertex Gemma predictions format to OpenAI format. @@ -354,7 +366,7 @@ class VertexGemmaConfig(OpenAIGPTConfig): message=f"Request failed: {response.text}", ) - response_json: Final = response.json() + response_json: Final = _PREDICTIONS_ENVELOPE.validate_python(response.json()) # Unwrap predictions to get OpenAI-compatible response openai_response: Final = self._unwrap_predictions_response(response_json) @@ -449,7 +461,7 @@ class VertexGemmaConfig(OpenAIGPTConfig): message=f"Request failed: {response.text}", ) - response_json: Final = response.json() + response_json: Final = _PREDICTIONS_ENVELOPE.validate_python(response.json()) # Unwrap predictions to get OpenAI-compatible response openai_response: Final = self._unwrap_predictions_response(response_json) diff --git a/litellm/llms/volcengine/responses/transformation.py b/litellm/llms/volcengine/responses/transformation.py index 747ea21e4fb..4076ed06fc2 100644 --- a/litellm/llms/volcengine/responses/transformation.py +++ b/litellm/llms/volcengine/responses/transformation.py @@ -3,6 +3,7 @@ from types import UnionType from typing import TYPE_CHECKING, Any, Final, Literal, Protocol, Union, get_args, get_origin import httpx +from pydantic import ConfigDict, TypeAdapter from pydantic import fields as pyd_fields import litellm @@ -37,6 +38,8 @@ if TYPE_CHECKING: else: LiteLLMLoggingObj = Any +_RESPONSE_BODY: Final = TypeAdapter(dict[str, object], config=ConfigDict(strict=True)) + class _EventModelClass(Protocol): @property @@ -420,7 +423,7 @@ class VolcEngineResponsesAPIConfig(OpenAIResponsesAPIConfig): @staticmethod def _parsed_response_body(raw_response: httpx.Response) -> dict[str, object]: - return raw_response.json() + return _RESPONSE_BODY.validate_python(raw_response.json()) @staticmethod def _annotation_origin(annotation: object) -> object: diff --git a/litellm/proxy/client/models.py b/litellm/proxy/client/models.py index 1361a3e6364..859322f061d 100644 --- a/litellm/proxy/client/models.py +++ b/litellm/proxy/client/models.py @@ -3,9 +3,12 @@ from collections.abc import Mapping from typing import Any, Final import requests +from pydantic import ConfigDict, TypeAdapter from .exceptions import NotFoundError, UnauthorizedError +_JSON_OBJECT: Final = TypeAdapter(dict[str, object], config=ConfigDict(strict=True)) + class ModelsManagementClient: def __init__(self, base_url: str, api_key: str | None = None, timeout: int = 30): @@ -109,7 +112,7 @@ class ModelsManagementClient: try: response: Final = session.send(request.prepare(), timeout=self._timeout) response.raise_for_status() - return response.json() + return _JSON_OBJECT.validate_python(response.json()) except requests.exceptions.HTTPError as e: if e.response.status_code == 401: raise UnauthorizedError(e) @@ -145,7 +148,7 @@ class ModelsManagementClient: try: response: Final = session.send(request.prepare(), timeout=self._timeout) response.raise_for_status() - return response.json() + return _JSON_OBJECT.validate_python(response.json()) except requests.exceptions.HTTPError as e: if e.response.status_code == 401: raise UnauthorizedError(e) @@ -287,7 +290,7 @@ class ModelsManagementClient: try: response: Final = session.send(request.prepare(), timeout=self._timeout) response.raise_for_status() - return response.json() + return _JSON_OBJECT.validate_python(response.json()) except requests.exceptions.HTTPError as e: if e.response.status_code == 401: raise UnauthorizedError(e) diff --git a/litellm/proxy/guardrails/guardrail_hooks/openai/moderations.py b/litellm/proxy/guardrails/guardrail_hooks/openai/moderations.py index 98be35079c3..16b3c45f6f8 100644 --- a/litellm/proxy/guardrails/guardrail_hooks/openai/moderations.py +++ b/litellm/proxy/guardrails/guardrail_hooks/openai/moderations.py @@ -169,7 +169,7 @@ class OpenAIModerationGuardrail(OpenAIGuardrailBase, CustomGuardrail): from litellm.types.llms.openai import OpenAIModerationResponse - return OpenAIModerationResponse(**response.json()) + return OpenAIModerationResponse.model_validate(response.json()) def _check_moderation_result(self, moderation_response: "OpenAIModerationResponse") -> None: """ diff --git a/litellm/proxy/litellm_pre_call_utils.py b/litellm/proxy/litellm_pre_call_utils.py index 601d741c189..d4528bc85e8 100644 --- a/litellm/proxy/litellm_pre_call_utils.py +++ b/litellm/proxy/litellm_pre_call_utils.py @@ -691,7 +691,7 @@ def _metadata_string_tags(metadata_objects: Iterable[Mapping[str, object]]) -> t def _strip_untrusted_request_header_controls( - headers: Any, + headers: object, *, allow_client_message_redaction_opt_out: bool = False, ) -> None: @@ -1695,7 +1695,7 @@ class LiteLLMProxyRequestSetup: @staticmethod def add_internal_user_from_user_mapping( - general_settings: dict | None, + general_settings: Mapping[str, object] | None, user_api_key_dict: UserAPIKeyAuth, headers: dict, ) -> UserAPIKeyAuth: @@ -1714,7 +1714,7 @@ class LiteLLMProxyRequestSetup: return user_api_key_dict @staticmethod - def get_user_from_headers(headers: dict, general_settings: dict | None = None) -> str | None: + def get_user_from_headers(headers: dict, general_settings: Mapping[str, object] | None = None) -> str | None: """ Get the user from the specified header if `general_settings.user_header_name` is set. """ @@ -1735,7 +1735,9 @@ class LiteLLMProxyRequestSetup: return user @staticmethod - def get_openai_org_id_from_headers(headers: dict, general_settings: dict | None = None) -> str | None: + def get_openai_org_id_from_headers( + headers: dict, general_settings: Mapping[str, object] | None = None + ) -> str | None: """ Get the OpenAI Org ID from the headers. """ @@ -1826,7 +1828,7 @@ class LiteLLMProxyRequestSetup: headers: dict, request_data: Mapping[str, object], user_api_key_dict: UserAPIKeyAuth, - general_settings: dict[str, object] | None = None, + general_settings: Mapping[str, object] | None = None, ) -> LitellmDataForBackendLLMCall: """ - Adds user from headers diff --git a/litellm/proxy/management_endpoints/auto_router_endpoints.py b/litellm/proxy/management_endpoints/auto_router_endpoints.py index 7a843aecc0c..4f48f3c57b9 100644 --- a/litellm/proxy/management_endpoints/auto_router_endpoints.py +++ b/litellm/proxy/management_endpoints/auto_router_endpoints.py @@ -231,9 +231,12 @@ def _shadow_eval_attempts(prisma_client: "PrismaClient") -> _ShadowEvalAttemptTa return _ShadowEvalAttemptRepository(prisma_client).table +_RAW_ROWS: Final = TypeAdapter(Sequence[dict[str, object]], config=ConfigDict(strict=True)) + + async def _query_raw(prisma_client: "PrismaClient", query: str, *args: object) -> Sequence[Mapping[str, object]]: async with db_span("auto_router_report_query", sql_relation(query)): - return await prisma_client.db.query_raw(query, *args) + return _RAW_ROWS.validate_python(await prisma_client.db.query_raw(query, *args)) async def _execute_raw(prisma_client: "PrismaClient", query: str, *args: object) -> int: diff --git a/litellm/proxy/management_endpoints/router_settings_endpoints.py b/litellm/proxy/management_endpoints/router_settings_endpoints.py index 1f4ca442d20..41c4b5cfaf7 100644 --- a/litellm/proxy/management_endpoints/router_settings_endpoints.py +++ b/litellm/proxy/management_endpoints/router_settings_endpoints.py @@ -10,7 +10,7 @@ GET /router/fields - Get router settings field definitions without values (for U import inspect from collections.abc import Mapping from types import MappingProxyType -from typing import Any, Final, get_args +from typing import Final, get_args from fastapi import APIRouter, Depends from pydantic import Field @@ -32,7 +32,7 @@ router: Final = APIRouter() class RouterSettingsResponse(LiteLLMBaseModel): fields: list[RouterSettingsField] = Field(description="List of all configurable router settings with metadata") - current_values: dict[str, Any] = Field(description="Current values of router settings") + current_values: dict[str, object] = Field(description="Current values of router settings") routing_strategy_descriptions: dict[str, str] = Field(description="Descriptions for each routing strategy option") routing_group_strategies: tuple[str, ...] = Field( description="Strategies supported when constructing a routing group" diff --git a/litellm/proxy/management_helpers/team_roster_sync.py b/litellm/proxy/management_helpers/team_roster_sync.py index dc0ba2e47d4..08be455a55d 100644 --- a/litellm/proxy/management_helpers/team_roster_sync.py +++ b/litellm/proxy/management_helpers/team_roster_sync.py @@ -40,6 +40,7 @@ from litellm.proxy.management_helpers.access_group_team_sync import ( ) from litellm.proxy.utils import PrismaClient, ProxyLogging from litellm.repositories.chunked_in import delete_many_in, find_many_in, update_many_in +from litellm.repositories.team_repository import TeamRepository if TYPE_CHECKING: from prisma import Prisma @@ -144,7 +145,7 @@ def _changes_nothing(plan: RosterPlan) -> bool: async def _unchanged_roster(prisma_client: PrismaClient, team_id: str) -> RosterSync | TeamGone: - team_row: Final = await prisma_client.writer_db.litellm_teamtable.find_unique(where={"team_id": team_id}) + team_row: Final = await TeamRepository(prisma_client, use_writer=True).table.find_unique(where={"team_id": team_id}) if team_row is None: return TeamGone(team_id=team_id) team: Final = LiteLLM_TeamTable.model_validate(team_row.model_dump()) diff --git a/litellm/proxy/mcp_server/discoverable_endpoints.py b/litellm/proxy/mcp_server/discoverable_endpoints.py index eb12b15cdae..07a267d6caa 100644 --- a/litellm/proxy/mcp_server/discoverable_endpoints.py +++ b/litellm/proxy/mcp_server/discoverable_endpoints.py @@ -300,12 +300,13 @@ def decode_state_hash(encrypted_state: str) -> dict: if decrypted_json is None: raise ValueError("Failed to decrypt state parameter") - state_data: Final = json.loads(decrypted_json) + state_data: Final = _DECODED_STATE.validate_python(json.loads(decrypted_json)) return state_data _MANAGED_AUTH_CODE_PREFIX: Final = "llm_scope_" _DECODED_JSON: Final = TypeAdapter(object) +_DECODED_STATE: Final = TypeAdapter(dict[str, object], config=ConfigDict(strict=True)) class _ManagedAuthorizationCode(LiteLLMBaseModel): diff --git a/litellm/proxy/pass_through_endpoints/pass_through_endpoints.py b/litellm/proxy/pass_through_endpoints/pass_through_endpoints.py index 650dcae3239..e017eee3d27 100644 --- a/litellm/proxy/pass_through_endpoints/pass_through_endpoints.py +++ b/litellm/proxy/pass_through_endpoints/pass_through_endpoints.py @@ -30,6 +30,7 @@ from pydantic import TypeAdapter from starlette.datastructures import UploadFile as StarletteUploadFile from starlette.routing import BaseRoute, Route from starlette.websockets import WebSocketState +from typing_extensions import ReadOnly from websockets.asyncio.client import connect from websockets.exceptions import ( ConnectionClosedError, @@ -173,8 +174,17 @@ def _is_mergeable_client_metadata(key: str, value: object) -> bool: return not key.startswith("user_api_key_") and (key != "model_info" or isinstance(value, Mapping)) +class _RegisteredPassThroughRoute(TypedDict): + endpoint_id: ReadOnly[str] + path: ReadOnly[str | None] + type: ReadOnly[str] + methods: ReadOnly[str | bool | list[str] | tuple[str, ...] | Mapping[str, object]] + auth: ReadOnly[bool] + passthrough_params: ReadOnly[Mapping[str, object]] + + # Global registry to track registered pass-through routes and prevent memory leaks -_registered_pass_through_routes: Final[dict[str, dict[str, str | bool | list[str] | Mapping[str, object]]]] = {} +_registered_pass_through_routes: Final[dict[str, _RegisteredPassThroughRoute]] = {} def get_response_body(response: httpx.Response) -> dict | None: @@ -2239,7 +2249,7 @@ def create_pass_through_route( status_code=status.HTTP_405_METHOD_NOT_ALLOWED, detail=f"Method {request.method} is not allowed for pass-through endpoint {path}.", ) - target_params: Final = { + target_params: Final[dict[str, object]] = { "target": target, "custom_headers": custom_headers, "forward_headers": _forward_headers, @@ -2271,7 +2281,9 @@ def create_pass_through_route( # Ensure custom_headers is a dict. Botocore returns a HeadersDict # for SigV4-prepared requests, which is a Mapping but not a dict. - headers_dict: Final = dict(param_custom_headers) if isinstance(param_custom_headers, Mapping) else {} + headers_dict: Final[dict[str, object]] = ( + {**param_custom_headers} if isinstance(param_custom_headers, Mapping) else {} + ) # Ensure query_params and custom_body are dicts or None final_query_params: Final = query_params_data if isinstance(query_params_data, dict) else {} @@ -3104,9 +3116,7 @@ class SafeRouteAdder: return f"{path}/{{subpath:path}}" return None - def registered_route_methods( - route_info: Mapping[str, str | bool | Sequence[str] | Mapping[str, object]], - ) -> tuple[tuple[str, str], ...]: + def registered_route_methods(route_info: _RegisteredPassThroughRoute) -> tuple[tuple[str, str], ...]: starlette_path: Final = registered_route_path(route_info) methods: Final = route_info.get("methods") if starlette_path is None or not isinstance(methods, (list, tuple)): @@ -3462,7 +3472,7 @@ class InitPassThroughEndpointHelpers: return (comparison_route, *forwarded) @staticmethod - def get_registered_pass_through_route(route: str, method: str | None = None) -> dict[str, Any] | None: + def get_registered_pass_through_route(route: str, method: str | None = None) -> _RegisteredPassThroughRoute | None: """Get passthrough params for a given route and optionally filter by HTTP method""" comparison_route: Final = InitPassThroughEndpointHelpers._route_for_registry_lookup(route) for key in _registered_pass_through_routes: diff --git a/litellm/proxy/spend_tracking/key_metadata_recovery.py b/litellm/proxy/spend_tracking/key_metadata_recovery.py index 5442c0f3f81..8fd4fbfcb8b 100644 --- a/litellm/proxy/spend_tracking/key_metadata_recovery.py +++ b/litellm/proxy/spend_tracking/key_metadata_recovery.py @@ -6,7 +6,7 @@ from datetime import datetime, timedelta from types import MappingProxyType from typing import Final, Literal, TypeVar -from pydantic import TypeAdapter +from pydantic import ConfigDict, TypeAdapter from typing_extensions import ReadOnly, TypedDict from litellm._logging import verbose_proxy_logger @@ -155,6 +155,7 @@ class _DailyUserSpendOwnerRow(LiteLLMBaseModel): last_owner: str | None = None +_RAW_ROWS: Final = TypeAdapter(Sequence[dict[str, object]], config=ConfigDict(strict=True)) _TOKEN_DIGEST_ROWS: Final = TypeAdapter(tuple[_TokenDigestRow, ...]) _SPEND_LOG_DIGEST_ROWS: Final = TypeAdapter(tuple[_SpendLogDigestRow, ...]) _DAILY_USER_SPEND_OWNER_ROWS: Final = TypeAdapter(tuple[_DailyUserSpendOwnerRow, ...]) @@ -194,7 +195,7 @@ async def _rows_within_the_statement_timeout( await transaction.execute_raw(_SPEND_LOG_STATEMENT_TIMEOUT_SQL) for setting in planner_settings: await transaction.execute_raw(setting) - return await transaction.query_raw(sql, *params) + return _RAW_ROWS.validate_python(await transaction.query_raw(sql, *params)) async def _reverse_hash_key_metadata( diff --git a/litellm/secret_managers/aws_secret_manager_v2.py b/litellm/secret_managers/aws_secret_manager_v2.py index 98d12e497da..de8566e6fdc 100644 --- a/litellm/secret_managers/aws_secret_manager_v2.py +++ b/litellm/secret_managers/aws_secret_manager_v2.py @@ -21,6 +21,8 @@ from types import MappingProxyType from typing import TYPE_CHECKING, Final import httpx +from pydantic import ConfigDict, TypeAdapter, with_config +from typing_extensions import ReadOnly, TypedDict import litellm from litellm._logging import verbose_logger @@ -41,6 +43,15 @@ if TYPE_CHECKING: from botocore.awsrequest import HTTPHeaders +@with_config(ConfigDict(extra="allow", strict=True)) +class _SecretStringResponse(TypedDict): + SecretString: ReadOnly[str] + + +_SECRET_STRING_RESPONSE: Final = TypeAdapter(_SecretStringResponse) +_AWS_SECRET_MANAGER_RESPONSE: Final = TypeAdapter(dict[str, object], config=ConfigDict(strict=True)) + + class AWSSecretsManagerV2(BaseAWSLLM, BaseSecretManager): def __init__( self, @@ -154,7 +165,7 @@ class AWSSecretsManagerV2(BaseAWSLLM, BaseSecretManager): try: response: Final = await async_client.post(url=endpoint_url, headers=headers, data=body.decode("utf-8")) response.raise_for_status() - return response.json()["SecretString"] + return _SECRET_STRING_RESPONSE.validate_python(response.json())["SecretString"] except httpx.TimeoutException: raise ValueError("Timeout error occurred") except Exception as e: @@ -204,7 +215,7 @@ class AWSSecretsManagerV2(BaseAWSLLM, BaseSecretManager): try: response: Final = sync_client.post(url=endpoint_url, headers=headers, data=body.decode("utf-8")) - return response.json()["SecretString"] + return _SECRET_STRING_RESPONSE.validate_python(response.json())["SecretString"] except httpx.TimeoutException: raise ValueError("Timeout error occurred") except httpx.HTTPStatusError as e: @@ -424,7 +435,7 @@ class AWSSecretsManagerV2(BaseAWSLLM, BaseSecretManager): try: response: Final = await async_client.post(url=endpoint_url, headers=headers, data=body.decode("utf-8")) response.raise_for_status() - return response.json() + return _AWS_SECRET_MANAGER_RESPONSE.validate_python(response.json()) except httpx.HTTPStatusError as err: raise ValueError(f"HTTP error occurred: {err.response.text}") except httpx.TimeoutException: @@ -481,7 +492,7 @@ class AWSSecretsManagerV2(BaseAWSLLM, BaseSecretManager): try: response: Final = await async_client.post(url=endpoint_url, headers=headers, data=body.decode("utf-8")) response.raise_for_status() - return response.json() + return _AWS_SECRET_MANAGER_RESPONSE.validate_python(response.json()) except httpx.HTTPStatusError as err: raise ValueError(f"HTTP error occurred: {err.response.text}") except httpx.TimeoutException: @@ -533,7 +544,7 @@ class AWSSecretsManagerV2(BaseAWSLLM, BaseSecretManager): try: response: Final = await async_client.post(url=endpoint_url, headers=headers, data=body.decode("utf-8")) response.raise_for_status() - return response.json() + return _AWS_SECRET_MANAGER_RESPONSE.validate_python(response.json()) except httpx.HTTPStatusError as err: raise ValueError(f"HTTP error occurred: {err.response.text}") except httpx.TimeoutException: @@ -613,7 +624,7 @@ class AWSSecretsManagerV2(BaseAWSLLM, BaseSecretManager): try: response: Final = await async_client.post(url=endpoint_url, headers=headers, data=body.decode("utf-8")) response.raise_for_status() - return response.json() + return _AWS_SECRET_MANAGER_RESPONSE.validate_python(response.json()) except httpx.HTTPStatusError as err: raise ValueError(f"HTTP error occurred: {err.response.text}") except httpx.TimeoutException: diff --git a/litellm/types/integrations/rag/bedrock_knowledgebase.py b/litellm/types/integrations/rag/bedrock_knowledgebase.py index 7156d8101e1..7288eca03b4 100644 --- a/litellm/types/integrations/rag/bedrock_knowledgebase.py +++ b/litellm/types/integrations/rag/bedrock_knowledgebase.py @@ -1,4 +1,4 @@ -from typing import Any, Literal +from typing import Literal from typing_extensions import ReadOnly, TypedDict @@ -40,7 +40,7 @@ class BedrockKBRetrievalResult(TypedDict, total=False): content: BedrockKBContent | None location: BedrockKBLocation | None score: float | None - metadata: dict[str, Any] | None + metadata: dict[str, object] | None class BedrockKBResponse(TypedDict, total=False): @@ -87,21 +87,21 @@ class BedrockKBMetadataConfiguration(TypedDict, total=False): class BedrockKBModelConfiguration(TypedDict, total=False): """Model configuration for reranking.""" - additionalModelRequestFields: dict[str, Any] | None + additionalModelRequestFields: dict[str, object] | None modelArn: str | None class BedrockKBRerankingConfiguration(TypedDict, total=False): """Configuration for reranking in vector search.""" - bedrockRerankingConfiguration: dict[str, Any] | None # This could be further typed if needed + bedrockRerankingConfiguration: dict[str, object] | None # This could be further typed if needed type: str | None class BedrockKBVectorSearchConfiguration(TypedDict, total=False): """Configuration for vector search.""" - filter: dict[str, Any] | None + filter: dict[str, object] | None implicitFilterConfiguration: BedrockKBImplicitFilterConfiguration | None numberOfResults: int | None overrideSearchType: str | None diff --git a/litellm/types/llms/bedrock_invoke_agents.py b/litellm/types/llms/bedrock_invoke_agents.py index 62f3c18ca7a..b334c986d6b 100644 --- a/litellm/types/llms/bedrock_invoke_agents.py +++ b/litellm/types/llms/bedrock_invoke_agents.py @@ -4,7 +4,7 @@ Type definitions for AWS Bedrock Invoke Agent API responses. https://docs.aws.amazon.com/bedrock/latest/APIReference/API_agent-runtime_InvokeAgent.html """ -from typing import Any, Final +from typing import Final from typing_extensions import TypedDict @@ -39,7 +39,7 @@ class InvokeAgentModelInvocationInput(TypedDict, total=False): """Model invocation input details.""" foundationModel: str | None - inferenceConfiguration: dict[str, Any] | None + inferenceConfiguration: dict[str, object] | None text: str | None traceId: str | None type: str | None @@ -49,9 +49,9 @@ class InvokeAgentModelInvocationOutput(TypedDict, total=False): """Model invocation output details.""" metadata: InvokeAgentMetadata | None - parsedResponse: dict[str, Any] | None - rawResponse: dict[str, Any] | None - reasoningContent: dict[str, Any] | None + parsedResponse: dict[str, object] | None + rawResponse: dict[str, object] | None + reasoningContent: dict[str, object] | None traceId: str | None diff --git a/litellm/types/llms/gemini.py b/litellm/types/llms/gemini.py index c300c26bcd3..2ad81e64e05 100644 --- a/litellm/types/llms/gemini.py +++ b/litellm/types/llms/gemini.py @@ -1,5 +1,6 @@ +from collections.abc import Mapping from enum import Enum -from typing import Any, Literal +from typing import Literal from typing_extensions import Required, TypedDict @@ -235,7 +236,7 @@ class GeminiVideoGenerationInstance(TypedDict, total=False): """Instance data for Gemini video generation request""" prompt: Required[str] - image: dict[str, Any] + image: Mapping[str, object] class GeminiVideoGenerationParameters(LiteLLMBaseModel): @@ -263,7 +264,7 @@ class GeminiVideoGenerationParameters(LiteLLMBaseModel): negativePrompt: str | None = None """Text describing what not to include in the video.""" - lastFrame: Any | None = None + lastFrame: object | None = None """ The final image for interpolation video to transition. Should be used with the 'image' parameter. @@ -275,7 +276,7 @@ class GeminiVideoGenerationParameters(LiteLLMBaseModel): Only supported in Veo 3.1 (list of VideoGenerationReferenceImage objects). """ - video: Any | None = None + video: object | None = None """ Video to be used for video extension (Video object). Only supported in Veo 3.1 & Veo 3 Fast. @@ -354,5 +355,5 @@ class GeminiLongRunningOperationResponse(LiteLLMBaseModel): response: GeminiOperationResponse | None = None """Response object when operation is complete""" - error: dict[str, Any] | None = None + error: dict[str, object] | None = None """Error details if operation failed""" diff --git a/litellm/types/llms/langgraph.py b/litellm/types/llms/langgraph.py index 7e34329992d..83f7bb090d1 100644 --- a/litellm/types/llms/langgraph.py +++ b/litellm/types/llms/langgraph.py @@ -4,7 +4,8 @@ Type definitions for LangGraph API. LangGraph provides a streaming and non-streaming API for running agents. """ -from typing import Any, Literal +from collections.abc import Mapping +from typing import Literal from typing_extensions import TypedDict @@ -29,8 +30,8 @@ class LangGraphRequest(TypedDict, total=False): assistant_id: str input: LangGraphInput stream_mode: str | None - config: dict[str, Any] | None - metadata: dict[str, Any] | None + config: Mapping[str, object] | None + metadata: Mapping[str, object] | None # Response Types - Streaming @@ -38,7 +39,7 @@ class LangGraphStreamEvent(TypedDict, total=False): """Single event in a LangGraph stream response.""" event: str - data: Any + data: object # Response Types - Non-streaming @@ -55,7 +56,7 @@ class LangGraphResponse(TypedDict, total=False): """Non-streaming response structure from LangGraph.""" messages: list[LangGraphResponseMessage] - values: dict[str, Any] + values: dict[str, object] # Parsed response for internal use diff --git a/litellm/types/mcp_server/tool_registry.py b/litellm/types/mcp_server/tool_registry.py index 738d37c9503..080e27b6ed3 100644 --- a/litellm/types/mcp_server/tool_registry.py +++ b/litellm/types/mcp_server/tool_registry.py @@ -1,5 +1,5 @@ from collections.abc import Callable -from typing import Any, ClassVar +from typing import ClassVar from pydantic import ConfigDict, Field @@ -10,7 +10,7 @@ class MCPTool(LiteLLMBaseModel): model_config: ClassVar[ConfigDict] = ConfigDict(arbitrary_types_allowed=True) name: str description: str - input_schema: dict[str, Any] + input_schema: dict[str, object] handler: Callable server_id: str | None = Field(default=None, frozen=True) @@ -18,18 +18,18 @@ class MCPTool(LiteLLMBaseModel): class ToolSchema(LiteLLMBaseModel): name: str description: str - inputSchema: dict[str, Any] + inputSchema: dict[str, object] class ListToolsResponse(LiteLLMBaseModel): tools: list[ToolSchema] nextCursor: str | None = None - _meta: dict[str, Any] | None = None + _meta: dict[str, object] | None = None class CallToolRequest(LiteLLMBaseModel): method: str = "tools/call" - params: dict[str, Any] + params: dict[str, object] class ContentItem(LiteLLMBaseModel): diff --git a/litellm/types/proxy/guardrails/guardrail_hooks/enkryptai.py b/litellm/types/proxy/guardrails/guardrail_hooks/enkryptai.py index d1eba1dd4a9..f936c9d85cf 100644 --- a/litellm/types/proxy/guardrails/guardrail_hooks/enkryptai.py +++ b/litellm/types/proxy/guardrails/guardrail_hooks/enkryptai.py @@ -1,5 +1,3 @@ -from typing import Any - from pydantic import Field from typing_extensions import TypedDict @@ -20,7 +18,7 @@ class EnkryptAIPolicyViolationDetail(TypedDict, total=False): class EnkryptAIPIIDetail(TypedDict, total=False): """Details for PII detection.""" - pii: dict[str, Any] + pii: dict[str, object] class EnkryptAIToxicityDetail(TypedDict, total=False): @@ -85,8 +83,8 @@ class EnkryptAIResponseDetails(TypedDict, total=False): toxicity: EnkryptAIToxicityDetail | None keyword_detected: EnkryptAIKeywordDetail | None bias: EnkryptAIBiasDetail | None - prompt_injection: dict[str, Any] | None - jailbreak: dict[str, Any] | None + prompt_injection: dict[str, object] | None + jailbreak: dict[str, object] | None class EnkryptAIResponse(TypedDict, total=False): @@ -107,7 +105,7 @@ class EnkryptAIProcessedResult(TypedDict): | EnkryptAIToxicityDetail | EnkryptAIKeywordDetail | EnkryptAIBiasDetail - | dict[str, Any], + | dict[str, object], ] diff --git a/tests/unit/llms/bedrock/chat/invoke_transformations/test_base_invoke_transformation.py b/tests/unit/llms/bedrock/chat/invoke_transformations/test_base_invoke_transformation.py index 1890c68b30d..c36644ba536 100644 --- a/tests/unit/llms/bedrock/chat/invoke_transformations/test_base_invoke_transformation.py +++ b/tests/unit/llms/bedrock/chat/invoke_transformations/test_base_invoke_transformation.py @@ -242,6 +242,33 @@ def test_transform_response_hands_json_mode_to_nova(): assert json.loads(result.choices[0].message.content) == {"city": "Paris", "temperature": 21} +@pytest.mark.parametrize(("finish_reason", "expected"), [("MAX_TOKENS", "length"), (None, "stop")]) +def test_transform_response_accepts_null_cohere_finish_reason(finish_reason: str | None, expected: str) -> None: + from litellm.types.utils import ModelResponse + + raw_response: Final = httpx.Response( + 200, + json={"generations": [{"text": "Hello there", "finish_reason": finish_reason}]}, + request=httpx.Request("POST", "https://bedrock"), + ) + + result: Final = AmazonInvokeConfig().transform_response( + model="invoke/cohere.command-text-v14", + raw_response=raw_response, + model_response=ModelResponse(), + logging_obj=MagicMock(), + request_data={}, + messages=[{"role": "user", "content": "hi"}], + optional_params={}, + litellm_params={}, + encoding=None, + api_key=None, + ) + + assert result.choices[0].message.content == "Hello there" + assert result.choices[0].finish_reason == expected + + def _stream_invoke_completion_with_spied_client(**kwargs: object) -> tuple[MagicMock, MagicMock]: mock_response = MagicMock() mock_response.status_code = 200 diff --git a/tests/unit/llms/bedrock/chat/invoke_transformations/test_twelvelabs_pegasus_transformation.py b/tests/unit/llms/bedrock/chat/invoke_transformations/test_twelvelabs_pegasus_transformation.py index d90e9933b88..5b3cb2e0c92 100644 --- a/tests/unit/llms/bedrock/chat/invoke_transformations/test_twelvelabs_pegasus_transformation.py +++ b/tests/unit/llms/bedrock/chat/invoke_transformations/test_twelvelabs_pegasus_transformation.py @@ -1,6 +1,13 @@ +from typing import Final +from unittest.mock import Mock + +import httpx +import pytest + from litellm.llms.bedrock.chat.invoke_transformations.amazon_twelvelabs_pegasus_transformation import ( AmazonTwelveLabsPegasusConfig, ) +from litellm.types.utils import ModelResponse def _make_messages() -> list[dict]: @@ -84,3 +91,24 @@ def test_transform_request_includes_s3_media(): s3_location = request["mediaSource"]["s3Location"] assert s3_location["uri"] == "s3://test-bucket/video.mp4" assert s3_location["bucketOwner"] == "123456789012" + + +@pytest.mark.parametrize( + ("finish_reason", "expected"), + [("length", "length"), (None, "stop")], +) +def test_transform_response_accepts_null_finish_reason(finish_reason: str | None, expected: str) -> None: + result: Final = AmazonTwelveLabsPegasusConfig().transform_response( + model="twelvelabs.pegasus-1-2-v1:0", + raw_response=httpx.Response(200, json={"message": "A cat jumps.", "finishReason": finish_reason}), + model_response=ModelResponse(), + logging_obj=Mock(), + request_data={}, + messages=_make_messages(), + optional_params={}, + litellm_params={}, + encoding=None, + ) + + assert result.choices[0].message.content == "A cat jumps." + assert result.choices[0].finish_reason == expected From 9e878cfae2edb7a66ecb3d977b2f3320c5955e6d Mon Sep 17 00:00:00 2001 From: moe-berri Date: Wed, 28 Oct 2026 16:26:28 -0700 Subject: [PATCH 27/41] test(gemini): cover every web source for a grounded passage (#48861) * fix(gemini): preserve every grounding web source * test(gemini): cover all grounding sources per passage Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> --------- Co-authored-by: Zabir Raihan Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- .../test_gemini_grounding_annotations_wire.py | 65 +++++++++++++++++++ ...test_vertex_and_google_ai_studio_gemini.py | 33 ++++++++++ 2 files changed, 98 insertions(+) create mode 100644 tests/integration/providers/test_gemini_grounding_annotations_wire.py diff --git a/tests/integration/providers/test_gemini_grounding_annotations_wire.py b/tests/integration/providers/test_gemini_grounding_annotations_wire.py new file mode 100644 index 00000000000..11df7c214f3 --- /dev/null +++ b/tests/integration/providers/test_gemini_grounding_annotations_wire.py @@ -0,0 +1,65 @@ +import json +from typing import Final + +import pytest +from integration._support.client import Gateway +from integration._support.wire import Reply, Request, wire_server + + +def _grounded_reply(request: Request) -> Reply: + body: Final = { + "candidates": [ + { + "content": {"role": "model", "parts": [{"text": "The passage"}]}, + "finishReason": "STOP", + "groundingMetadata": { + "groundingChunks": [ + {"retrievedContext": {"uri": "gs://documents/source"}}, + {"web": {"uri": "https://example.com/first", "title": "First"}}, + {"web": {"uri": "https://example.com/second", "title": "Second"}}, + ], + "groundingSupports": [ + {"segment": {"startIndex": 4, "endIndex": 11}, "groundingChunkIndices": [0, 99, 2, 1]} + ], + }, + } + ], + "usageMetadata": {"promptTokenCount": 5, "candidatesTokenCount": 2, "totalTokenCount": 7}, + } + assert json.loads(request.body)["contents"] == [{"role": "user", "parts": [{"text": "Find sources"}]}] + if ":streamGenerateContent" in request.target: + return Reply(content_type="text/event-stream", chunks=(f"data: {json.dumps(body)}\n\n".encode(),)) + return Reply(body=json.dumps(body).encode()) + + +@pytest.mark.parametrize("stream", [False, True], ids=["non-streaming", "streaming"]) +def test_chat_grounding_annotations_include_every_web_source(gateway: Gateway, stream: bool) -> None: + with wire_server(_grounded_reply) as wire, gateway.scenario() as scenario: + model: Final = scenario.model(model="gemini/gemini-3.8-flash", api_base=wire.url, api_key="synthetic-key") + response: Final = gateway.request( + "POST", + "/v1/chat/completions", + {"model": model, "messages": [{"role": "user", "content": "Find sources"}], "stream": stream}, + ) + assert response.status_code == 200, response.text + payloads: Final = ( + [json.loads(line[6:]) for line in response.text.splitlines() if line.startswith("data: {")] + if stream + else [response.json()] + ) + messages: Final = tuple(payload["choices"][0]["delta" if stream else "message"] for payload in payloads) + annotations: Final = tuple(message["annotations"] for message in messages if message.get("annotations")) + assert annotations == ( + [ + { + "type": "url_citation", + "url_citation": { + "start_index": 4, + "end_index": 11, + "url": f"https://example.com/{name.lower()}", + "title": name, + }, + } + for name in ("Second", "First") + ], + ), response.text diff --git a/tests/unit/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py b/tests/unit/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py index 1546d1e9fe0..3e1bca18f08 100644 --- a/tests/unit/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py +++ b/tests/unit/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py @@ -3514,6 +3514,39 @@ def test_vertex_ai_annotation_conversion(): assert "Current time" in time_annotation["url_citation"]["title"] +@pytest.mark.parametrize( + "indices, expected_indices", + [([2, 0], [2, 0]), ([1, 99, -1, 2, 0], [2, 0]), ([0], [0]), ([], []), ([1, 99], [])], + ids=["multiple-sources", "non-web-first", "single-source", "empty", "no-web"], +) +def test_grounding_annotations_preserve_all_web_sources(indices: list[int], expected_indices: list[int]) -> None: + chunks: Final = [ + {"web": {"uri": "https://example.com/first", "title": "First"}}, + {"retrievedContext": {"uri": "gs://documents/source"}}, + {"web": {"uri": "https://example.com/second", "title": "Second"}}, + ] + metadata: Final = { + "groundingChunks": chunks, + "groundingSupports": [ + {"segment": {"startIndex": 4, "endIndex": 11}, "groundingChunkIndices": indices} + ], + } + annotations: Final = VertexGeminiConfig._convert_grounding_metadata_to_annotations([metadata], "The passage") + expected: Final = [ + { + "type": "url_citation", + "url_citation": { + "start_index": 4, + "end_index": 11, + "url": chunks[index]["web"]["uri"], + "title": chunks[index]["web"]["title"], + }, + } + for index in expected_indices + ] + assert json.dumps(annotations) == json.dumps(expected) + + def test_vertex_ai_annotation_empty_grounding_metadata(): """ Test handling of empty or missing grounding metadata. From 96fe78f96ecb6d0a5f92a6b7e68a3b770bb25213 Mon Sep 17 00:00:00 2001 From: Kerry Lu Date: Wed, 28 Oct 2026 16:32:45 -0700 Subject: [PATCH 28/41] fix(router): reject string fallback targets at Router init (LIT-10238) (#48943) * fix(router): reject string fallback targets at Router init (LIT-10238) * fix(router): keep direct target fallback entries when rejecting string targets * test(router): set string fallbacks after init in cooldown tolerance tests --------- Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- litellm/router.py | 15 +++++- .../router_utils/test_cooldown_handlers.py | 7 +-- tests/unit/test_router/test_router.py | 48 +++++++++++++++++++ 3 files changed, 63 insertions(+), 7 deletions(-) diff --git a/litellm/router.py b/litellm/router.py index 5ca09499c3b..408cb20945f 100644 --- a/litellm/router.py +++ b/litellm/router.py @@ -1370,7 +1370,7 @@ class Router: }, ] - router = Router(model_list=model_list, fallbacks=[{"azure-gpt-3.5-turbo": "openai-gpt-3.5-turbo"}]) + router = Router(model_list=model_list, fallbacks=[{"azure-gpt-3.5-turbo": ["openai-gpt-3.5-turbo"]}]) ``` """ @@ -1596,6 +1596,10 @@ class Router: if default_fallbacks is not None or litellm.default_fallbacks is not None: _fallbacks = default_fallbacks or litellm.default_fallbacks + if isinstance(_fallbacks, str): + raise TypeError( + f"default_fallbacks must be a list of model names, got '{_fallbacks}'. Use ['{_fallbacks}']" + ) if self.fallbacks is not None: self.fallbacks.append({"*": _fallbacks}) else: @@ -2839,7 +2843,7 @@ class Router: self._initialize_core_endpoints() self._initialize_specialized_endpoints() - def validate_fallbacks(self, fallback_param: list | None): + def validate_fallbacks(self, fallback_param: Sequence[Mapping[str, object]] | None): """ Validate the fallbacks parameter. """ @@ -2852,6 +2856,13 @@ class Router: raise ValueError( f"Dictionary '{fallback_dict}' must have exactly one key, but has {len(fallback_dict)} keys." ) + ((model_group, targets),) = fallback_dict.items() + if isinstance(targets, list) or check_non_standard_fallback_format(fallbacks=[fallback_dict]): + continue + raise TypeError( + f"Fallbacks for '{model_group}' must be a list of model names, got '{targets}'. " + f"Use {{'{model_group}': ['{targets}']}}" + ) def _add_encrypted_content_affinity_check(self, enable_global_affinity: bool) -> None: def _move_before_deployment_affinity( diff --git a/tests/unit/router_utils/test_cooldown_handlers.py b/tests/unit/router_utils/test_cooldown_handlers.py index 24459a5a250..2e1e3acaabb 100644 --- a/tests/unit/router_utils/test_cooldown_handlers.py +++ b/tests/unit/router_utils/test_cooldown_handlers.py @@ -3150,8 +3150,6 @@ BENCHED: Final = (litellm.Timeout, litellm.Timeout, RouterRateLimitError, Router {}, BENCHED, ), - ({"allowed_fails": 1, "fallbacks": [{"test-model": "backup-model"}]}, {}, BENCHED), - ({"allowed_fails": 1, "default_fallbacks": "backup-model"}, {}, BENCHED), ], ids=[ "allowed_fails", @@ -3160,8 +3158,6 @@ BENCHED: Final = (litellm.Timeout, litellm.Timeout, RouterRateLimitError, Router "policy_names_timeout", "allowed_fails_with_a_fallback_group", "allowed_fails_with_a_dict_fallback_target", - "allowed_fails_with_a_string_fallback_target", - "allowed_fails_with_a_string_default_fallbacks", ], ) @pytest.mark.asyncio @@ -3207,7 +3203,8 @@ async def test_router_allowed_fails_benches_a_lone_deployment_only_when_the_poli def test_has_fallback_group_reads_string_targets_and_skips_unparseable_entries( fallbacks: Sequence[Mapping[str, object]], expected: bool ) -> None: - router: Final = Router(model_list=[TIMED_OUT_DEPLOYMENT], fallbacks=list(fallbacks)) + router: Final = Router(model_list=[TIMED_OUT_DEPLOYMENT]) + router.fallbacks = list(fallbacks) assert _has_fallback_group(router, "test-model") is expected diff --git a/tests/unit/test_router/test_router.py b/tests/unit/test_router/test_router.py index 3c7a222d9a1..300c9cff6ff 100644 --- a/tests/unit/test_router/test_router.py +++ b/tests/unit/test_router/test_router.py @@ -31228,3 +31228,51 @@ async def test_anthropic_caller_rate_limit_does_not_block_other_credentials( finally: await GLOBAL_LOGGING_WORKER.flush() await client.close() + + +_FALLBACK_VALIDATION_MODEL_LIST: Final = ( + {"model_name": "gpt-4", "litellm_params": {"model": "openai/gpt-4", "api_key": "fake-key"}}, + {"model_name": "gpt-3.5", "litellm_params": {"model": "openai/gpt-3.5-turbo", "api_key": "fake-key"}}, +) + + +@pytest.mark.parametrize( + ("router_kwargs", "expected_message"), + [ + ({"fallbacks": [{"gpt-4": "gpt-3.5"}]}, "Fallbacks for 'gpt-4' must be a list of model names, got 'gpt-3.5'"), + ({"default_fallbacks": "gpt-3.5"}, "default_fallbacks must be a list of model names, got 'gpt-3.5'"), + ( + {"content_policy_fallbacks": [{"gpt-4": "gpt-3.5"}]}, + "Fallbacks for 'gpt-4' must be a list of model names, got 'gpt-3.5'", + ), + ], + ids=["string_fallback_target", "string_default_fallbacks", "string_content_policy_fallback_target"], +) +def test_router_rejects_a_string_fallback_target_at_init( + router_kwargs: Mapping[str, object], expected_message: str +) -> None: + with pytest.raises(TypeError, match="must be a list of model names") as excinfo: + Router(model_list=list(_FALLBACK_VALIDATION_MODEL_LIST), **router_kwargs) + + assert expected_message in str(excinfo.value) + + +@pytest.mark.parametrize( + ("router_kwargs", "expected_fallbacks"), + [ + ({"fallbacks": [{"gpt-4": ["gpt-3.5"]}]}, [{"gpt-4": ["gpt-3.5"]}]), + ({"default_fallbacks": ["gpt-3.5"]}, [{"*": ["gpt-3.5"]}]), + ( + {"fallbacks": [{"gpt-4": [{"model": "gpt-3.5", "api_key": "other-key"}]}]}, + [{"gpt-4": [{"model": "gpt-3.5", "api_key": "other-key"}]}], + ), + ({"fallbacks": [{"model": "gpt-3.5"}]}, [{"model": "gpt-3.5"}]), + ], + ids=["list_fallback_target", "list_default_fallbacks", "dict_override_fallback_target", "direct_target_entry"], +) +def test_router_keeps_supported_fallback_shapes( + router_kwargs: Mapping[str, object], expected_fallbacks: list[dict[str, object]] +) -> None: + router: Final = Router(model_list=list(_FALLBACK_VALIDATION_MODEL_LIST), **router_kwargs) + + assert router.fallbacks == expected_fallbacks From ab07b7207e37e52f1ac8126817d101c90f5d01a0 Mon Sep 17 00:00:00 2001 From: "devin-ai-integration[bot]" <158243242+devin-ai-integration[bot]@users.noreply.github.com> Date: Wed, 28 Oct 2026 16:59:57 -0700 Subject: [PATCH 29/41] fix(vertex-ai): add batch above 200k and image output batch prices for gemini models (#48945) * fix(vertex-ai): add batch and flex above 200k and image output tier prices for gemini models * fix(vertex-ai): drop flex above 200k and image priority fields the cost map schema rejects Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> --------- Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- ...odel_prices_and_context_window_backup.json | 28 ++++++++++++++++--- model_prices_and_context_window.json | 28 ++++++++++++++++--- 2 files changed, 48 insertions(+), 8 deletions(-) diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index dc043dc4d26..8719565a96f 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -28853,6 +28853,7 @@ "mode": "image_generation", "output_cost_per_image": 0.039, "output_cost_per_image_token": 3e-05, + "output_cost_per_image_token_batches": 1.5e-05, "output_cost_per_reasoning_token": 2.5e-06, "output_cost_per_token": 2.5e-06, "output_cost_per_token_batches": 1.25e-06, @@ -28910,6 +28911,7 @@ "mode": "image_generation", "output_cost_per_image": 0.134, "output_cost_per_image_token": 0.00012, + "output_cost_per_image_token_batches": 6e-05, "output_cost_per_token": 1.2e-05, "output_cost_per_token_above_200k_tokens": 1.8e-05, "output_cost_per_token_above_200k_tokens_priority": 3.24e-05, @@ -28963,12 +28965,13 @@ "mode": "image_generation", "output_cost_per_image": 0.134, "output_cost_per_image_token": 0.00012, + "output_cost_per_image_token_batches": 6e-05, "output_cost_per_token": 1.2e-05, "output_cost_per_token_priority": 2.16e-05, "output_cost_per_token_above_200k_tokens": 1.8e-05, "output_cost_per_token_above_200k_tokens_priority": 3.24e-05, "output_cost_per_token_batches": 6e-06, - "source": "https://ai.google.dev/gemini-api/docs/pricing", + "source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing", "supported_endpoints": [ "/v1/chat/completions", "/v1/images/generations", @@ -29014,6 +29017,7 @@ "mode": "image_generation", "output_cost_per_image": 0.0672, "output_cost_per_image_token": 6e-05, + "output_cost_per_image_token_batches": 3e-05, "output_cost_per_token": 3e-06, "output_cost_per_token_batches": 1.5e-06, "output_cost_per_token_flex": 1.5e-06, @@ -29059,9 +29063,10 @@ "mode": "image_generation", "output_cost_per_image": 0.0672, "output_cost_per_image_token": 6e-05, + "output_cost_per_image_token_batches": 3e-05, "output_cost_per_token": 3e-06, "output_cost_per_token_batches": 1.5e-06, - "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing#gemini-models", + "source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing", "supported_endpoints": [ "/v1/chat/completions", "/v1/images/generations", @@ -29106,6 +29111,7 @@ "mode": "image_generation", "output_cost_per_image": 0.0336, "output_cost_per_image_token": 3e-05, + "output_cost_per_image_token_batches": 1.5e-05, "output_cost_per_token": 1.5e-06, "output_cost_per_token_batches": 7.5e-07, "output_cost_per_token_flex": 7.5e-07, @@ -29672,10 +29678,12 @@ }, "search_free_requests": {"count": 10000, "period": "day", "shared_group": "gemini-2.5-pro"}, "google_maps_grounding_cost_per_query": 0.025, + "input_cost_per_token_above_200k_tokens_batches": 1.25e-06, "input_cost_per_token_above_200k_tokens_priority": 4.5e-06, "input_cost_per_token_batches": 6.25e-07, "input_cost_per_token_flex": 6.25e-07, "input_cost_per_token_priority": 2.25e-06, + "output_cost_per_token_above_200k_tokens_batches": 7.5e-06, "output_cost_per_token_above_200k_tokens_priority": 2.7e-05, "output_cost_per_token_batches": 5e-06, "output_cost_per_token_flex": 5e-06, @@ -29690,6 +29698,7 @@ "google_maps_grounding_cost_per_query": 0.014, "input_cost_per_token": 2e-06, "input_cost_per_token_above_200k_tokens": 4e-06, + "input_cost_per_token_above_200k_tokens_batches": 2e-06, "input_cost_per_token_above_200k_tokens_priority": 7.2e-06, "input_cost_per_token_batches": 1e-06, "input_cost_per_token_flex": 1e-06, @@ -29701,6 +29710,7 @@ "output_cost_per_image": 0.00012, "output_cost_per_token": 1.2e-05, "output_cost_per_token_above_200k_tokens": 1.8e-05, + "output_cost_per_token_above_200k_tokens_batches": 9e-06, "output_cost_per_token_above_200k_tokens_priority": 3.24e-05, "output_cost_per_token_batches": 6e-06, "output_cost_per_token_flex": 6e-06, @@ -30229,6 +30239,7 @@ "google_maps_grounding_cost_per_query": 0.014, "input_cost_per_token": 2e-06, "input_cost_per_token_above_200k_tokens": 4e-06, + "input_cost_per_token_above_200k_tokens_batches": 2e-06, "input_cost_per_token_above_200k_tokens_priority": 7.2e-06, "input_cost_per_token_batches": 1e-06, "input_cost_per_token_flex": 1e-06, @@ -30240,6 +30251,7 @@ "output_cost_per_image": 0.00012, "output_cost_per_token": 1.2e-05, "output_cost_per_token_above_200k_tokens": 1.8e-05, + "output_cost_per_token_above_200k_tokens_batches": 9e-06, "output_cost_per_token_above_200k_tokens_priority": 3.24e-05, "output_cost_per_token_batches": 6e-06, "output_cost_per_token_flex": 6e-06, @@ -51910,6 +51922,7 @@ "mode": "image_generation", "output_cost_per_image": 0.039, "output_cost_per_image_token": 3e-05, + "output_cost_per_image_token_batches": 1.5e-05, "output_cost_per_reasoning_token": 2.5e-06, "output_cost_per_token": 2.5e-06, "output_cost_per_token_batches": 1.25e-06, @@ -51968,6 +51981,7 @@ "mode": "image_generation", "output_cost_per_image": 0.134, "output_cost_per_image_token": 0.00012, + "output_cost_per_image_token_batches": 6e-05, "output_cost_per_token": 1.2e-05, "output_cost_per_token_above_200k_tokens": 1.8e-05, "output_cost_per_token_above_200k_tokens_priority": 3.24e-05, @@ -52002,13 +52016,14 @@ "mode": "image_generation", "output_cost_per_image": 0.134, "output_cost_per_image_token": 0.00012, + "output_cost_per_image_token_batches": 6e-05, "output_cost_per_token": 1.2e-05, "output_cost_per_token_priority": 2.16e-05, "output_cost_per_token_above_200k_tokens": 1.8e-05, "output_cost_per_token_above_200k_tokens_priority": 3.24e-05, "output_cost_per_token_batches": 6e-06, "supports_reasoning": false, - "source": "https://docs.cloud.google.com/vertex-ai/generative-ai/docs/models/gemini/3-pro-image", + "source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing", "supported_endpoints": [ "/v1/chat/completions", "/v1/images/generations", @@ -52034,6 +52049,7 @@ "mode": "image_generation", "output_cost_per_image": 0.0672, "output_cost_per_image_token": 6e-05, + "output_cost_per_image_token_batches": 3e-05, "output_cost_per_token": 3e-06, "output_cost_per_token_batches": 1.5e-06, "output_cost_per_token_flex": 1.5e-06, @@ -52061,10 +52077,11 @@ "mode": "image_generation", "output_cost_per_image": 0.0672, "output_cost_per_image_token": 6e-05, + "output_cost_per_image_token_batches": 3e-05, "output_cost_per_token": 3e-06, "output_cost_per_token_batches": 1.5e-06, "supports_reasoning": false, - "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing#gemini-models", + "source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing", "supported_endpoints": [ "/v1/chat/completions", "/v1/images/generations", @@ -52089,6 +52106,7 @@ "mode": "image_generation", "output_cost_per_image": 0.0336, "output_cost_per_image_token": 3e-05, + "output_cost_per_image_token_batches": 1.5e-05, "output_cost_per_token": 1.5e-06, "output_cost_per_token_batches": 7.5e-07, "output_cost_per_token_flex": 7.5e-07, @@ -74400,6 +74418,7 @@ "mode": "image_generation", "output_cost_per_image": 0.0336, "output_cost_per_image_token": 3e-05, + "output_cost_per_image_token_batches": 1.5e-05, "output_cost_per_token": 7.5e-06, "output_cost_per_token_batches": 3.75e-06, "output_cost_per_token_priority": 1.35e-05, @@ -74445,6 +74464,7 @@ "mode": "image_generation", "output_cost_per_image": 0.0336, "output_cost_per_image_token": 3e-05, + "output_cost_per_image_token_batches": 1.5e-05, "output_cost_per_token": 7.5e-06, "output_cost_per_token_batches": 3.75e-06, "output_cost_per_token_priority": 1.35e-05, diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index dc043dc4d26..8719565a96f 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -28853,6 +28853,7 @@ "mode": "image_generation", "output_cost_per_image": 0.039, "output_cost_per_image_token": 3e-05, + "output_cost_per_image_token_batches": 1.5e-05, "output_cost_per_reasoning_token": 2.5e-06, "output_cost_per_token": 2.5e-06, "output_cost_per_token_batches": 1.25e-06, @@ -28910,6 +28911,7 @@ "mode": "image_generation", "output_cost_per_image": 0.134, "output_cost_per_image_token": 0.00012, + "output_cost_per_image_token_batches": 6e-05, "output_cost_per_token": 1.2e-05, "output_cost_per_token_above_200k_tokens": 1.8e-05, "output_cost_per_token_above_200k_tokens_priority": 3.24e-05, @@ -28963,12 +28965,13 @@ "mode": "image_generation", "output_cost_per_image": 0.134, "output_cost_per_image_token": 0.00012, + "output_cost_per_image_token_batches": 6e-05, "output_cost_per_token": 1.2e-05, "output_cost_per_token_priority": 2.16e-05, "output_cost_per_token_above_200k_tokens": 1.8e-05, "output_cost_per_token_above_200k_tokens_priority": 3.24e-05, "output_cost_per_token_batches": 6e-06, - "source": "https://ai.google.dev/gemini-api/docs/pricing", + "source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing", "supported_endpoints": [ "/v1/chat/completions", "/v1/images/generations", @@ -29014,6 +29017,7 @@ "mode": "image_generation", "output_cost_per_image": 0.0672, "output_cost_per_image_token": 6e-05, + "output_cost_per_image_token_batches": 3e-05, "output_cost_per_token": 3e-06, "output_cost_per_token_batches": 1.5e-06, "output_cost_per_token_flex": 1.5e-06, @@ -29059,9 +29063,10 @@ "mode": "image_generation", "output_cost_per_image": 0.0672, "output_cost_per_image_token": 6e-05, + "output_cost_per_image_token_batches": 3e-05, "output_cost_per_token": 3e-06, "output_cost_per_token_batches": 1.5e-06, - "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing#gemini-models", + "source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing", "supported_endpoints": [ "/v1/chat/completions", "/v1/images/generations", @@ -29106,6 +29111,7 @@ "mode": "image_generation", "output_cost_per_image": 0.0336, "output_cost_per_image_token": 3e-05, + "output_cost_per_image_token_batches": 1.5e-05, "output_cost_per_token": 1.5e-06, "output_cost_per_token_batches": 7.5e-07, "output_cost_per_token_flex": 7.5e-07, @@ -29672,10 +29678,12 @@ }, "search_free_requests": {"count": 10000, "period": "day", "shared_group": "gemini-2.5-pro"}, "google_maps_grounding_cost_per_query": 0.025, + "input_cost_per_token_above_200k_tokens_batches": 1.25e-06, "input_cost_per_token_above_200k_tokens_priority": 4.5e-06, "input_cost_per_token_batches": 6.25e-07, "input_cost_per_token_flex": 6.25e-07, "input_cost_per_token_priority": 2.25e-06, + "output_cost_per_token_above_200k_tokens_batches": 7.5e-06, "output_cost_per_token_above_200k_tokens_priority": 2.7e-05, "output_cost_per_token_batches": 5e-06, "output_cost_per_token_flex": 5e-06, @@ -29690,6 +29698,7 @@ "google_maps_grounding_cost_per_query": 0.014, "input_cost_per_token": 2e-06, "input_cost_per_token_above_200k_tokens": 4e-06, + "input_cost_per_token_above_200k_tokens_batches": 2e-06, "input_cost_per_token_above_200k_tokens_priority": 7.2e-06, "input_cost_per_token_batches": 1e-06, "input_cost_per_token_flex": 1e-06, @@ -29701,6 +29710,7 @@ "output_cost_per_image": 0.00012, "output_cost_per_token": 1.2e-05, "output_cost_per_token_above_200k_tokens": 1.8e-05, + "output_cost_per_token_above_200k_tokens_batches": 9e-06, "output_cost_per_token_above_200k_tokens_priority": 3.24e-05, "output_cost_per_token_batches": 6e-06, "output_cost_per_token_flex": 6e-06, @@ -30229,6 +30239,7 @@ "google_maps_grounding_cost_per_query": 0.014, "input_cost_per_token": 2e-06, "input_cost_per_token_above_200k_tokens": 4e-06, + "input_cost_per_token_above_200k_tokens_batches": 2e-06, "input_cost_per_token_above_200k_tokens_priority": 7.2e-06, "input_cost_per_token_batches": 1e-06, "input_cost_per_token_flex": 1e-06, @@ -30240,6 +30251,7 @@ "output_cost_per_image": 0.00012, "output_cost_per_token": 1.2e-05, "output_cost_per_token_above_200k_tokens": 1.8e-05, + "output_cost_per_token_above_200k_tokens_batches": 9e-06, "output_cost_per_token_above_200k_tokens_priority": 3.24e-05, "output_cost_per_token_batches": 6e-06, "output_cost_per_token_flex": 6e-06, @@ -51910,6 +51922,7 @@ "mode": "image_generation", "output_cost_per_image": 0.039, "output_cost_per_image_token": 3e-05, + "output_cost_per_image_token_batches": 1.5e-05, "output_cost_per_reasoning_token": 2.5e-06, "output_cost_per_token": 2.5e-06, "output_cost_per_token_batches": 1.25e-06, @@ -51968,6 +51981,7 @@ "mode": "image_generation", "output_cost_per_image": 0.134, "output_cost_per_image_token": 0.00012, + "output_cost_per_image_token_batches": 6e-05, "output_cost_per_token": 1.2e-05, "output_cost_per_token_above_200k_tokens": 1.8e-05, "output_cost_per_token_above_200k_tokens_priority": 3.24e-05, @@ -52002,13 +52016,14 @@ "mode": "image_generation", "output_cost_per_image": 0.134, "output_cost_per_image_token": 0.00012, + "output_cost_per_image_token_batches": 6e-05, "output_cost_per_token": 1.2e-05, "output_cost_per_token_priority": 2.16e-05, "output_cost_per_token_above_200k_tokens": 1.8e-05, "output_cost_per_token_above_200k_tokens_priority": 3.24e-05, "output_cost_per_token_batches": 6e-06, "supports_reasoning": false, - "source": "https://docs.cloud.google.com/vertex-ai/generative-ai/docs/models/gemini/3-pro-image", + "source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing", "supported_endpoints": [ "/v1/chat/completions", "/v1/images/generations", @@ -52034,6 +52049,7 @@ "mode": "image_generation", "output_cost_per_image": 0.0672, "output_cost_per_image_token": 6e-05, + "output_cost_per_image_token_batches": 3e-05, "output_cost_per_token": 3e-06, "output_cost_per_token_batches": 1.5e-06, "output_cost_per_token_flex": 1.5e-06, @@ -52061,10 +52077,11 @@ "mode": "image_generation", "output_cost_per_image": 0.0672, "output_cost_per_image_token": 6e-05, + "output_cost_per_image_token_batches": 3e-05, "output_cost_per_token": 3e-06, "output_cost_per_token_batches": 1.5e-06, "supports_reasoning": false, - "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing#gemini-models", + "source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing", "supported_endpoints": [ "/v1/chat/completions", "/v1/images/generations", @@ -52089,6 +52106,7 @@ "mode": "image_generation", "output_cost_per_image": 0.0336, "output_cost_per_image_token": 3e-05, + "output_cost_per_image_token_batches": 1.5e-05, "output_cost_per_token": 1.5e-06, "output_cost_per_token_batches": 7.5e-07, "output_cost_per_token_flex": 7.5e-07, @@ -74400,6 +74418,7 @@ "mode": "image_generation", "output_cost_per_image": 0.0336, "output_cost_per_image_token": 3e-05, + "output_cost_per_image_token_batches": 1.5e-05, "output_cost_per_token": 7.5e-06, "output_cost_per_token_batches": 3.75e-06, "output_cost_per_token_priority": 1.35e-05, @@ -74445,6 +74464,7 @@ "mode": "image_generation", "output_cost_per_image": 0.0336, "output_cost_per_image_token": 3e-05, + "output_cost_per_image_token_batches": 1.5e-05, "output_cost_per_token": 7.5e-06, "output_cost_per_token_batches": 3.75e-06, "output_cost_per_token_priority": 1.35e-05, From d06b889d413237881e3dc27486e4345dd7aa18f9 Mon Sep 17 00:00:00 2001 From: rj-cairn <349020452+rj-cairn@users.noreply.github.com> Date: Wed, 28 Oct 2026 17:00:29 -0700 Subject: [PATCH 30/41] fix(streaming): keep merged think tags balanced and split SSE lines only on CR/LF (#48755) * fix(vertex_ai): split Gemini SSE lines on CR/LF only so U+2028 in text keeps the stream The Gemini and Vertex AI streaming path read the SSE body with httpx iter_lines()/aiter_lines(). Those follow str.splitlines(), which also breaks on U+0085, U+2028, U+2029, \x0b, \x0c and \x1c-\x1e. JSON leaves U+0085, U+2028 and U+2029 unescaped inside strings, so a data: line whose text holds one of them was cut in two. The tail is not a data: line and the head never parses, and the accumulate-and-retry fallback then carries the unparsed prefix forward, so every later event is appended to it and the stream ends with no more content and no usage. SSE lines end only at CRLF, LF or CR. The new iter_sse_lines/aiter_sse_lines split decoded text on exactly those, treating a CR at the end of one chunk and an LF at the start of the next as one break. The first request and every long-decode hop now read through them. Co-Authored-By: Claude Opus 5.5 * fix(streaming): keep merged think tags balanced and split shared SSE lines only on CR/LF Close an open block on the first content or tool-call delta and at stream end, reopen it when reasoning returns, and route the shared chat completions stream through the SSE line splitter so U+2028 and U+2029 inside JSON no longer drop chunks. * fix(streaming): exact-match the SSE [DONE] marker in BaseModelResponseIterator * fix(streaming): split every SSE reader on CR/LF only and keep the logged think block closed Switch the remaining httpx stream readers (Anthropic chat, Google GenAI, Interactions, Codestral, openai_like, Predibase, LangGraph, Vertex Agent Engine, A2A providers, Azure AI agents, sandboxes, the A2A endpoint, and others) from iter_lines/aiter_lines to iter_sse_lines/aiter_sse_lines, so a raw U+2028 or U+2029 inside a JSON string no longer splits an event. Keep the chunk that closes a think block in the logged and cached response when a stream ends without a finish_reason, close the block before a legacy function_call, and make sync Codestral FIM streaming send stream=true and use the text-completion-codestral stream parser. Test fakes that only mocked iter_lines now stream real httpx bytes and assert the content they feed in. * fix(streaming): split buffered SSE parsers only on CR and LF, shard the codestral tests Buffered SSE and JSONL parsers used str.splitlines(), which also breaks on U+2028, U+2029 and U+0085 that JSON allows raw inside strings. They now use split_sse_lines from sse_lines.py, which also gains a str.split fast path. Adds tests/unit/llms/codestral to the unit workflow shard. * fix(streaming): keep U+2028 in restamped Anthropic frames and fold CRLF before splitting SSE lines * test: type the new stream tests, drop a comment, and benchmark the SSE line splitter * test: type the iterator class parameters in the line-separator stream tests * fix(responses): stream tool call arguments opened in a text chunk * fix(responses): hold a tool call that shares a chunk with text until the text ends * test: type the a2a SSE handler helper and the sandbox fake responses --------- Co-authored-by: Tim Vink Co-authored-by: Claude Opus 5.5 Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com> Co-authored-by: michelligabriele --- .github/workflows/test-unit.yml | 1 + .../bedrock_agentcore/transformation.py | 5 +- .../providers/watsonx_orchestrate/handler.py | 5 +- .../transformation.py | 3 +- litellm/google_genai/streaming_iterator.py | 9 +- litellm/interactions/streaming_iterator.py | 5 +- .../litellm_core_utils/streaming_handler.py | 57 +++--- litellm/llms/anthropic/chat/handler.py | 5 +- .../anthropic/messages/stream_assembly.py | 3 +- litellm/llms/azure_ai/agents/handler.py | 3 +- litellm/llms/base_llm/base_model_iterator.py | 2 +- .../base_llm/guardrail_translation/utils.py | 3 +- .../llms/base_llm/sandbox/transformation.py | 3 +- litellm/llms/bedrock/files/transformation.py | 3 +- .../guardrail_translation/handler.py | 3 +- litellm/llms/bedrock/search/transformation.py | 5 +- .../llms/chatgpt/responses/transformation.py | 3 +- litellm/llms/codestral/completion/handler.py | 9 +- litellm/llms/custom_httpx/aiohttp_handler.py | 3 +- litellm/llms/custom_httpx/llm_http_handler.py | 5 +- litellm/llms/custom_httpx/sse_lines.py | 74 ++++++++ .../guardrail_translation/handler.py | 3 +- .../huggingface/embedding/transformation.py | 3 +- litellm/llms/langgraph/chat/sse_iterator.py | 5 +- litellm/llms/oobabooga/chat/oobabooga.py | 3 +- litellm/llms/openai_like/chat/handler.py | 9 +- litellm/llms/predibase/chat/handler.py | 5 +- .../vertex_ai/agent_engine/transformation.py | 5 +- .../vertex_and_google_ai_studio_gemini.py | 9 +- .../proxy/agent_endpoints/a2a_endpoints.py | 3 +- .../streaming_model_restamp.py | 4 +- .../anthropic_passthrough_logging_handler.py | 7 +- .../tinyfish_passthrough_logging_handler.py | 3 +- .../managed_id_rewriter.py | 3 +- .../response_polling/background_streaming.py | 3 +- .../streaming_iterator.py | 13 +- tests/benchmarks/test_sse_lines_benchmarks.py | 27 +++ .../test_bedrock_agentcore_a2a.py | 32 ++++ ...test_watsonx_orchestrate_transformation.py | 20 ++- ...responses_transformation_transformation.py | 18 ++ .../test_google_genai_streaming_iterator.py | 83 +++++---- .../test_interactions_streaming_iterator.py | 45 ++++- .../test_streaming_handler.py | 114 +++++++++++- .../chat/test_anthropic_chat_handler.py | 59 ++++++- .../messages/test_stream_assembly.py | 24 +++ .../unit/llms/azure_ai/agents/test_handler.py | 23 ++- .../azure_ai/agents/test_transformation.py | 20 +-- .../guardrail_translation/test_utils.py | 17 ++ .../test_bedrock_files_transformation.py | 14 ++ .../guardrail_translation/test_handler.py | 14 +- .../test_agentcore_search_transformation.py | 15 ++ .../test_chatgpt_responses_transformation.py | 31 ++++ .../test_cloudflare_transformation.py | 47 ++++- tests/unit/llms/codestral/__init__.py | 0 .../llms/codestral/completion/__init__.py | 0 .../test_codestral_completion_handler.py | 50 ++++++ .../llms/custom_httpx/test_aiohttp_handler.py | 37 ++++ .../unit/llms/custom_httpx/test_sse_lines.py | 64 +++++++ .../test_databricks_chat_transformation.py | 14 +- .../test_fireworks_ai_chat_transformation.py | 6 +- .../test_google_genai_guardrail_handler.py | 16 ++ .../chat/test_langgraph_sse_iterator.py | 32 ++++ .../llms/oobabooga/chat/test_oobabooga.py | 29 ++++ .../chat/test_openai_like_chat_handler.py | 37 +++- .../test_snowflake_chat_transformation.py | 30 +--- .../llms/test_predibase_transformation.py | 47 +++++ tests/unit/llms/triton/test_triton.py | 11 +- .../agent_engine/test_transformation.py | 47 ++++- ...test_vertex_and_google_ai_studio_gemini.py | 64 ++++++- .../llms/vertex_ai/test_http_status_201.py | 8 +- .../agent_endpoints/test_a2a_endpoints.py | 138 +++++++-------- .../test_streaming_model_restamp.py | 23 +++ ...t_anthropic_passthrough_logging_handler.py | 48 +++++ ...st_tinyfish_passthrough_logging_handler.py | 7 + .../test_managed_id_rewriter.py | 33 +++- .../test_response_polling_handler.py | 31 +++- .../test_streaming_iterator_transformation.py | 100 +++++++++++ tests/unit/sandbox/test_e2b_sandbox.py | 28 ++- .../unit/sandbox/test_opensandbox_sandbox.py | 13 +- tests/unit/test_groq_streaming_encoding.py | 164 ++++-------------- 80 files changed, 1554 insertions(+), 413 deletions(-) create mode 100644 litellm/llms/custom_httpx/sse_lines.py create mode 100644 tests/benchmarks/test_sse_lines_benchmarks.py create mode 100644 tests/unit/llms/codestral/__init__.py create mode 100644 tests/unit/llms/codestral/completion/__init__.py create mode 100644 tests/unit/llms/codestral/completion/test_codestral_completion_handler.py create mode 100644 tests/unit/llms/custom_httpx/test_sse_lines.py create mode 100644 tests/unit/llms/langgraph/chat/test_langgraph_sse_iterator.py diff --git a/.github/workflows/test-unit.yml b/.github/workflows/test-unit.yml index 2ddf21bca17..80d5018fe9a 100644 --- a/.github/workflows/test-unit.yml +++ b/.github/workflows/test-unit.yml @@ -245,6 +245,7 @@ jobs: tests/unit/llms/clarifai tests/unit/llms/claude_code tests/unit/llms/cloudflare + tests/unit/llms/codestral tests/unit/llms/codex tests/unit/llms/cohere tests/unit/llms/cometapi diff --git a/litellm/a2a_protocol/providers/bedrock_agentcore/transformation.py b/litellm/a2a_protocol/providers/bedrock_agentcore/transformation.py index 1cea0194715..d4400fbb4cc 100644 --- a/litellm/a2a_protocol/providers/bedrock_agentcore/transformation.py +++ b/litellm/a2a_protocol/providers/bedrock_agentcore/transformation.py @@ -19,6 +19,7 @@ from litellm.a2a_protocol.utils import ( ) from litellm.exceptions import BadRequestError from litellm.llms.bedrock.chat.agentcore.transformation import AmazonAgentCoreConfig +from litellm.llms.custom_httpx.sse_lines import aiter_sse_lines RUNTIME_SESSION_ID_MIN_LENGTH: Final = 33 RUNTIME_SESSION_ID_MAX_LENGTH: Final = 256 @@ -50,7 +51,7 @@ _RESERVED_PREFIX_HEADERS: Final[tuple[str, ...]] = ( class _SSELineSource(Protocol): """Minimal streaming-response surface used to read SSE lines.""" - def aiter_lines(self) -> AsyncIterator[str]: ... + def aiter_text(self) -> AsyncIterator[str]: ... def _filter_reserved_headers( @@ -244,7 +245,7 @@ class BedrockAgentCoreA2ATransformation: Yields: Parsed JSON dicts from SSE data lines """ - async for line in response.aiter_lines(): + async for line in aiter_sse_lines(response.aiter_text()): line = line.strip() if not line: continue diff --git a/litellm/a2a_protocol/providers/watsonx_orchestrate/handler.py b/litellm/a2a_protocol/providers/watsonx_orchestrate/handler.py index 0a2871d25cc..3937c47ea3c 100644 --- a/litellm/a2a_protocol/providers/watsonx_orchestrate/handler.py +++ b/litellm/a2a_protocol/providers/watsonx_orchestrate/handler.py @@ -20,6 +20,7 @@ from litellm.llms.custom_httpx.http_handler import ( AsyncHTTPHandler, get_async_httpx_client, ) +from litellm.llms.custom_httpx.sse_lines import aiter_sse_lines from litellm.types.llms.custom_http import httpxSpecialProvider _IBM_CLOUD_IAM_URL: Final = "https://iam.cloud.ibm.com/identity/token" @@ -74,7 +75,7 @@ class _WXORun(TypedDict, total=False): class _SSELineSource(Protocol): - def aiter_lines(self) -> AsyncIterator[str]: ... + def aiter_text(self) -> AsyncIterator[str]: ... class _WXOView(TypedDict, total=False): @@ -231,7 +232,7 @@ class WatsonxOrchestrateHandler: async def _accumulate_wxo_sse_text(response: _SSELineSource) -> str: source: Final[_WXOView] = {"sse_source": response} accumulated_text = "" - async for line in source["sse_source"].aiter_lines(): + async for line in aiter_sse_lines(source["sse_source"].aiter_text()): if not line.startswith("data:"): continue data_str = line[5:].strip() diff --git a/litellm/completion_extras/litellm_responses_transformation/transformation.py b/litellm/completion_extras/litellm_responses_transformation/transformation.py index 47a3a903d82..ee3639ac387 100644 --- a/litellm/completion_extras/litellm_responses_transformation/transformation.py +++ b/litellm/completion_extras/litellm_responses_transformation/transformation.py @@ -34,6 +34,7 @@ from litellm.llms.base_llm.base_model_iterator import BaseModelResponseIterator from litellm.llms.base_llm.bridges.completion_transformation import ( CompletionTransformationBridge, ) +from litellm.llms.custom_httpx.sse_lines import split_sse_lines from litellm.responses.litellm_completion_transformation.tool_search import is_anthropic_tool_search_chat_tool from litellm.responses.sse_output_recovery import ( parse_sse_json_chunk, @@ -1351,7 +1352,7 @@ class LiteLLMResponsesTransformationHandler(CompletionTransformationBridge): recovered_output_items: Final[dict[int, dict[str, object]]] = {} recovered_text_only_items: Final[dict[int, dict[str, object]]] = {} - for chunk in raw_sse.splitlines(): + for chunk in split_sse_lines(raw_sse): parsed_chunk = parse_sse_json_chunk(chunk) if parsed_chunk is None: continue diff --git a/litellm/google_genai/streaming_iterator.py b/litellm/google_genai/streaming_iterator.py index a22c47cb1e7..08ac7d4f9dc 100644 --- a/litellm/google_genai/streaming_iterator.py +++ b/litellm/google_genai/streaming_iterator.py @@ -4,6 +4,7 @@ from typing import TYPE_CHECKING, Any, Final import litellm from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj +from litellm.llms.custom_httpx.sse_lines import aiter_sse_lines, iter_sse_lines from litellm.proxy.pass_through_endpoints.success_handler import ( PassThroughEndpointLogging, ) @@ -131,9 +132,9 @@ class GoogleGenAIGenerateContentStreamingIterator(BaseGoogleGenAIGenerateContent self.model = model self.generate_content_provider_config = generate_content_provider_config self.litellm_metadata = litellm_metadata - # Gemini streamGenerateContent uses SSE line framing; iter_lines keeps + # Gemini streamGenerateContent uses SSE line framing; iter_sse_lines keeps # large inlineData payloads (e.g. image/jpeg) intact within one event. - self.stream_iterator = response.iter_lines() + self.stream_iterator = iter_sse_lines(response.iter_text()) def __iter__(self): return self @@ -182,9 +183,9 @@ class AsyncGoogleGenAIGenerateContentStreamingIterator(BaseGoogleGenAIGenerateCo self.model = model self.generate_content_provider_config = generate_content_provider_config self.litellm_metadata = litellm_metadata - # Gemini streamGenerateContent uses SSE line framing; aiter_lines keeps + # Gemini streamGenerateContent uses SSE line framing; aiter_sse_lines keeps # large inlineData payloads (e.g. image/jpeg) intact within one event. - self.stream_iterator = response.aiter_lines() + self.stream_iterator = aiter_sse_lines(response.aiter_text()) def __aiter__(self): return self diff --git a/litellm/interactions/streaming_iterator.py b/litellm/interactions/streaming_iterator.py index f4ad57bf7b4..c7faa3fc510 100644 --- a/litellm/interactions/streaming_iterator.py +++ b/litellm/interactions/streaming_iterator.py @@ -21,6 +21,7 @@ from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLogging from litellm.litellm_core_utils.llm_response_utils.get_api_base import get_api_base from litellm.litellm_core_utils.thread_pool_executor import executor from litellm.llms.base_llm.interactions.transformation import BaseInteractionsAPIConfig +from litellm.llms.custom_httpx.sse_lines import aiter_sse_lines, iter_sse_lines from litellm.types.interactions import ( InteractionsAPIStreamingResponse, ) @@ -145,7 +146,7 @@ class InteractionsAPIStreamingIterator(BaseInteractionsAPIStreamingIterator): litellm_metadata=litellm_metadata, custom_llm_provider=custom_llm_provider, ) - self.stream_iterator = response.aiter_lines() + self.stream_iterator = aiter_sse_lines(response.aiter_text()) def __aiter__(self): return self @@ -212,7 +213,7 @@ class SyncInteractionsAPIStreamingIterator(BaseInteractionsAPIStreamingIterator) litellm_metadata=litellm_metadata, custom_llm_provider=custom_llm_provider, ) - self.stream_iterator = response.iter_lines() + self.stream_iterator = iter_sse_lines(response.iter_text()) def __iter__(self): return self diff --git a/litellm/litellm_core_utils/streaming_handler.py b/litellm/litellm_core_utils/streaming_handler.py index 3fb25de43cc..2348d8e57d6 100644 --- a/litellm/litellm_core_utils/streaming_handler.py +++ b/litellm/litellm_core_utils/streaming_handler.py @@ -1263,7 +1263,7 @@ class CustomStreamWrapper: ) if _is_delta_empty: - model_response.choices[0].delta = Delta(content=None) # ensure empty delta chunk returned + model_response.choices[0].delta = Delta(content=self._close_open_think_block()) # get any function call arguments model_response.choices[0].finish_reason = map_finish_reason( finish_reason=self.received_finish_reason @@ -1292,21 +1292,26 @@ class CustomStreamWrapper: if self.merge_reasoning_content_in_choices is not True: return delta: Final = model_response.choices[0].delta - reasoning_content: Final = getattr(delta, "reasoning_content", None) + reasoning_content: Final = getattr(delta, "reasoning_content", None) or "" content: Final = delta.content or "" - if reasoning_content: - self._emitted_disqualifying_content = True - closes_thinking: Final = bool(content) and not self.sent_last_thinking_block - opening: Final = "" if self.sent_first_thinking_block else "" - delta.content = opening + reasoning_content + ("" + content if closes_thinking else "") - self.sent_first_thinking_block = True - self.sent_last_thinking_block = self.sent_last_thinking_block or closes_thinking - elif self.sent_first_thinking_block and not self.sent_last_thinking_block and content: - delta.content = "" + content - self.sent_last_thinking_block = True - + ends_thinking: Final = bool(content or delta.tool_calls or delta.function_call) + inside_think_block: Final = self.sent_first_thinking_block and not self.sent_last_thinking_block + opening: Final = "" if reasoning_content and not inside_think_block else "" + closing: Final = "" if ends_thinking and (reasoning_content or inside_think_block) else "" if hasattr(delta, "reasoning_content"): del delta.reasoning_content + if not reasoning_content and not closing: + return + delta.content = opening + reasoning_content + closing + content + self.sent_first_thinking_block = True + self.sent_last_thinking_block = bool(closing) + self._emitted_disqualifying_content = self._emitted_disqualifying_content or bool(reasoning_content) + + def _close_open_think_block(self) -> str | None: + if not self.sent_first_thinking_block or self.sent_last_thinking_block: + return None + self.sent_last_thinking_block = True + return "" def _dispatch_provider_chunk( self, @@ -1932,8 +1937,20 @@ class CustomStreamWrapper: model_response.choices[0].finish_reason == "stop" and self.tool_call ): # don't overwrite for other - potential error finish reasons model_response.choices[0].finish_reason = "tool_calls" + close_tag: Final = self._close_open_think_block() + if close_tag is not None: + model_response.choices[0].delta.content = close_tag return model_response + def _final_chunk_for_logging(self, processed_chunk: "ModelResponseStream") -> "ModelResponseStream | None": + if self.received_finish_reason is not None or self.intermittent_finish_reason is not None: + return processed_chunk + if processed_chunk.choices[0].delta.content is None: + return None + return processed_chunk.model_copy( + update={"choices": [processed_chunk.choices[0].model_copy(update={"finish_reason": None})]} + ) + def _record_usage_only_chunk(self, model_response: "ModelResponseStream") -> None: """ Keep provider usage-only chunks (e.g. OpenRouter's post-finish chunk, which carries a @@ -2173,10 +2190,9 @@ class CustomStreamWrapper: else: self.sent_last_chunk = True processed_chunk: Final = self.finish_reason_handler() - # The logged response is built from self.chunks; keep a finish_reason the provider sent on its - # last content chunk (stripped there), but never add the synthetic "stop" used when it sent none. - if self.received_finish_reason is not None or self.intermittent_finish_reason is not None: - self.chunks.append(processed_chunk) + final_chunk_for_logging: Final = self._final_chunk_for_logging(processed_chunk) + if final_chunk_for_logging is not None: + self.chunks.append(final_chunk_for_logging) if self.stream_options is None: self._set_final_chunk_usage(processed_chunk) ## LOGGING @@ -2406,10 +2422,9 @@ class CustomStreamWrapper: else: self.sent_last_chunk = True processed_chunk: Final = self.finish_reason_handler() - # The logged response is built from self.chunks; keep a finish_reason the provider sent on its - # last content chunk (stripped there), but never add the synthetic "stop" used when it sent none. - if self.received_finish_reason is not None or self.intermittent_finish_reason is not None: - self.chunks.append(processed_chunk) + final_chunk_for_logging: Final = self._final_chunk_for_logging(processed_chunk) + if final_chunk_for_logging is not None: + self.chunks.append(final_chunk_for_logging) if self.stream_options is None: self._set_final_chunk_usage(processed_chunk) # see sync __next__'s sibling branch: deliberately do NOT restore diff --git a/litellm/llms/anthropic/chat/handler.py b/litellm/llms/anthropic/chat/handler.py index 13ceea3e31b..8a82ed98991 100644 --- a/litellm/llms/anthropic/chat/handler.py +++ b/litellm/llms/anthropic/chat/handler.py @@ -38,6 +38,7 @@ from litellm.llms.custom_httpx.http_handler import ( get_async_httpx_client, get_httpx_client, ) +from litellm.llms.custom_httpx.sse_lines import aiter_sse_lines, iter_sse_lines from litellm.types.llms.anthropic import ( ContentBlockDelta, ContentBlockStart, @@ -233,7 +234,7 @@ async def make_call( raise AnthropicError(status_code=500, message=str(e)) completion_stream: Final = ModelResponseIterator( - streaming_response=response.aiter_lines(), + streaming_response=aiter_sse_lines(response.aiter_text()), sync_stream=False, json_mode=json_mode, speed=speed, @@ -304,7 +305,7 @@ def make_sync_call( ) completion_stream: Final = ModelResponseIterator( - streaming_response=response.iter_lines(), + streaming_response=iter_sse_lines(response.iter_text()), sync_stream=True, json_mode=json_mode, speed=speed, diff --git a/litellm/llms/anthropic/messages/stream_assembly.py b/litellm/llms/anthropic/messages/stream_assembly.py index 90bdad8f4bc..c103eb0f975 100644 --- a/litellm/llms/anthropic/messages/stream_assembly.py +++ b/litellm/llms/anthropic/messages/stream_assembly.py @@ -13,6 +13,7 @@ from litellm.llms.anthropic.chat.handler import ( ) from litellm.llms.anthropic.chat.transformation import AnthropicConfig from litellm.llms.anthropic.messages.native.utils import MESSAGE_DELTA_OPTIONAL_KEYS +from litellm.llms.custom_httpx.sse_lines import split_sse_lines from litellm.types.utils import ( Choices, Message, @@ -269,7 +270,7 @@ def build_complete_streaming_response_legacy( def extract_sse_data(event_str: str) -> dict | None: """Parse the JSON object from the ``data:`` line of an Anthropic SSE event.""" - for line in event_str.splitlines(): + for line in split_sse_lines(event_str): stripped = line.strip() if stripped.startswith("data:"): payload = stripped[len("data:") :].strip() diff --git a/litellm/llms/azure_ai/agents/handler.py b/litellm/llms/azure_ai/agents/handler.py index 94f1399f01b..423a09c5c92 100644 --- a/litellm/llms/azure_ai/agents/handler.py +++ b/litellm/llms/azure_ai/agents/handler.py @@ -35,6 +35,7 @@ from litellm.llms.azure_ai.agents.transformation import ( AzureAIAgentsConfig, AzureAIAgentsError, ) +from litellm.llms.custom_httpx.sse_lines import aiter_sse_lines from litellm.types.llms.openai import ( ChatCompletionAnnotation, ChatCompletionAnnotationURLCitation, @@ -632,7 +633,7 @@ class AzureAIAgentsHandler: current_event = None - async for line in response.aiter_lines(): + async for line in aiter_sse_lines(response.aiter_text()): line = line.strip() if line.startswith("event:"): diff --git a/litellm/llms/base_llm/base_model_iterator.py b/litellm/llms/base_llm/base_model_iterator.py index 032951be74a..3afc685b3dd 100644 --- a/litellm/llms/base_llm/base_model_iterator.py +++ b/litellm/llms/base_llm/base_model_iterator.py @@ -96,7 +96,7 @@ class BaseModelResponseIterator: released (and a backend like vLLM aborts generation) when the stream is abandoned before its natural end. - ``streaming_response`` is usually a bare ``aiter_lines()`` generator + ``streaming_response`` is usually a bare ``aiter_sse_lines()`` generator that holds no reference to the response, so the handler that owns the response attaches it here after construction.""" if self.http_response is not None: diff --git a/litellm/llms/base_llm/guardrail_translation/utils.py b/litellm/llms/base_llm/guardrail_translation/utils.py index 725146fe894..fb886e42780 100644 --- a/litellm/llms/base_llm/guardrail_translation/utils.py +++ b/litellm/llms/base_llm/guardrail_translation/utils.py @@ -6,6 +6,7 @@ from typing import Final, TypeVar, cast # noqa: TID251 # a rebuilt chat row ha from pydantic import BaseModel, ConfigDict, TypeAdapter, ValidationError +from litellm.llms.custom_httpx.sse_lines import split_sse_lines from litellm.types.llms.anthropic_messages.anthropic_response import AnthropicUsage from litellm.types.llms.openai import AllMessageValues, ResponseAPIUsage @@ -24,7 +25,7 @@ def _anthropic_stream_chunk_events(item: object) -> list[dict]: events: Final[list[dict]] = [] for block in chunk.split("\n\n"): - for line in block.splitlines(): + for line in split_sse_lines(block): stripped = line.strip() if not stripped.startswith("data:"): continue diff --git a/litellm/llms/base_llm/sandbox/transformation.py b/litellm/llms/base_llm/sandbox/transformation.py index fe0e3abb87e..9b9b4e0a947 100644 --- a/litellm/llms/base_llm/sandbox/transformation.py +++ b/litellm/llms/base_llm/sandbox/transformation.py @@ -12,6 +12,7 @@ from typing import Final import httpx from pydantic import Field, PrivateAttr +from litellm.llms.custom_httpx.sse_lines import aiter_sse_lines from litellm.types.llms.base import LiteLLMPydanticObjectBase SANDBOX_MAX_OUTPUT_BYTES: Final = 10 * 1024 * 1024 @@ -99,7 +100,7 @@ class BaseSandboxConfig: async def _read_capped_lines(self, response: httpx.Response) -> list[str]: lines: Final[list[str]] = [] total = 0 - async for line in response.aiter_lines(): + async for line in aiter_sse_lines(response.aiter_text()): total += len(line.encode("utf-8")) if total > SANDBOX_MAX_OUTPUT_BYTES: raise ValueError( diff --git a/litellm/llms/bedrock/files/transformation.py b/litellm/llms/bedrock/files/transformation.py index 8c4702bcf99..0624c2c65db 100644 --- a/litellm/llms/bedrock/files/transformation.py +++ b/litellm/llms/bedrock/files/transformation.py @@ -49,6 +49,7 @@ from litellm.llms.base_llm.files.transformation import ( BaseFilesConfig, LiteLLMLoggingObj, ) +from litellm.llms.custom_httpx.sse_lines import split_sse_lines from litellm.types.llms.base import LiteLLMBaseModel from litellm.types.llms.bedrock import AwsAuthParams, BedrockBatchRecordKind from litellm.types.llms.openai import ( @@ -400,7 +401,7 @@ def _batch_input_line(line_number: int, line: str) -> _BatchInputLine: def _non_blank_lines(file_content: str) -> Iterator[tuple[int, str]]: - numbered_lines: Final = enumerate(file_content.splitlines(), start=1) + numbered_lines: Final = enumerate(split_sse_lines(file_content), start=1) return ((line_number, line) for line_number, line in numbered_lines if line.strip()) diff --git a/litellm/llms/bedrock/passthrough/guardrail_translation/handler.py b/litellm/llms/bedrock/passthrough/guardrail_translation/handler.py index 1b01b0ec6d0..3a65a8928bf 100644 --- a/litellm/llms/bedrock/passthrough/guardrail_translation/handler.py +++ b/litellm/llms/bedrock/passthrough/guardrail_translation/handler.py @@ -45,6 +45,7 @@ from litellm.llms.bedrock.passthrough.guardrail_translation.structured_edits imp structured_edits, tool_edits, ) +from litellm.llms.custom_httpx.sse_lines import split_sse_lines from litellm.proxy.guardrails.auto_router_compression import is_compression_guardrail from litellm.types.utils import GenericGuardrailAPIInputs @@ -947,7 +948,7 @@ def _invoke_frames(items: Sequence[object]) -> tuple[InvokeStreamFrame, ...]: def _sse_events(sse: bytes) -> tuple[Mapping[str, object], ...]: return tuple( _JSON_OBJECT.validate_python(json.loads(line[len("data:") :])) - for line in sse.decode().splitlines() + for line in split_sse_lines(sse.decode()) if line.startswith("data:") ) diff --git a/litellm/llms/bedrock/search/transformation.py b/litellm/llms/bedrock/search/transformation.py index 565fc12aa1e..6f72cadad67 100644 --- a/litellm/llms/bedrock/search/transformation.py +++ b/litellm/llms/bedrock/search/transformation.py @@ -47,6 +47,7 @@ from litellm.llms.base_llm.search.transformation import ( ) from litellm.llms.bedrock.base_aws_llm import BaseAWSLLM from litellm.llms.bedrock.common_utils import BedrockError +from litellm.llms.custom_httpx.sse_lines import split_sse_lines from litellm.secret_managers.main import get_secret_str # AgentCore web-search rejects queries longer than 200 characters @@ -146,7 +147,9 @@ def _iter_sse_events(text: str) -> Iterator[Mapping[str, object]]: progress notifications before the JSON-RPC response. """ for chunk in _SSE_EVENT_SEPARATOR.split(text): - payload = "\n".join(line[len("data:") :].lstrip() for line in chunk.splitlines() if line.startswith("data:")) + payload = "\n".join( + line[len("data:") :].lstrip() for line in split_sse_lines(chunk) if line.startswith("data:") + ) if not payload: continue try: diff --git a/litellm/llms/chatgpt/responses/transformation.py b/litellm/llms/chatgpt/responses/transformation.py index 040b7b8c1d8..72b5cde21d1 100644 --- a/litellm/llms/chatgpt/responses/transformation.py +++ b/litellm/llms/chatgpt/responses/transformation.py @@ -11,6 +11,7 @@ from litellm.litellm_core_utils.llm_response_utils.convert_dict_to_response impo safe_convert_created_field, ) from litellm.litellm_core_utils.url_utils import join_api_base +from litellm.llms.custom_httpx.sse_lines import split_sse_lines from litellm.llms.openai.common_utils import OpenAIError from litellm.llms.openai.responses.transformation import OpenAIResponsesAPIConfig from litellm.responses.sse_output_recovery import ( @@ -199,7 +200,7 @@ class ChatGPTResponsesAPIConfig(OpenAIResponsesAPIConfig): error_message = None streamed_output_items: Final[dict[int, dict[str, object]]] = {} text_only_output_items: Final[dict[int, dict[str, object]]] = {} - for chunk in body_text.splitlines(): + for chunk in split_sse_lines(body_text): parsed_chunk = parse_sse_json_chunk(chunk) if parsed_chunk is None: continue diff --git a/litellm/llms/codestral/completion/handler.py b/litellm/llms/codestral/completion/handler.py index 6bebfe2d444..6b172e77295 100644 --- a/litellm/llms/codestral/completion/handler.py +++ b/litellm/llms/codestral/completion/handler.py @@ -20,6 +20,7 @@ from litellm.llms.custom_httpx.http_handler import ( AsyncHTTPHandler, get_async_httpx_client, ) +from litellm.llms.custom_httpx.sse_lines import aiter_sse_lines, iter_sse_lines from litellm.types.utils import TextChoices from litellm.utils import CustomStreamWrapper, TextCompletionResponse @@ -109,7 +110,7 @@ async def make_call( if response.status_code != 200: raise TextCompletionCodestralError(status_code=response.status_code, message=response.text) - completion_stream: Final = response.aiter_lines() + completion_stream: Final = aiter_sse_lines(response.aiter_text()) # LOGGING logging_obj.post_call( input=messages, @@ -339,13 +340,13 @@ class CodestralTextCompletion: response = litellm.module_level_client.post( completion_url, headers=headers, - data=json.dumps(data), + data=json.dumps({**data, "stream": True}), stream=stream, ) _response: Final = CustomStreamWrapper( - response.iter_lines(), + iter_sse_lines(response.iter_text()), model, - custom_llm_provider="codestral", + custom_llm_provider="text-completion-codestral", logging_obj=logging_obj, ) return _response diff --git a/litellm/llms/custom_httpx/aiohttp_handler.py b/litellm/llms/custom_httpx/aiohttp_handler.py index 447d7b685f7..4060d84f7f4 100644 --- a/litellm/llms/custom_httpx/aiohttp_handler.py +++ b/litellm/llms/custom_httpx/aiohttp_handler.py @@ -21,6 +21,7 @@ from litellm.llms.custom_httpx.http_handler import ( get_httpx_client, get_ssl_configuration, ) +from litellm.llms.custom_httpx.sse_lines import iter_sse_lines from litellm.types.llms.openai import FileTypes from litellm.types.utils import HttpHandlerRequestFields, ImageResponse, LlmProviders from litellm.utils import CustomStreamWrapper, ModelResponse, ProviderConfigManager @@ -466,7 +467,7 @@ class BaseLLMAIOHTTPHandler: ) else: completion_stream = provider_config.get_model_response_iterator( - streaming_response=response.iter_lines(), sync_stream=True + streaming_response=iter_sse_lines(response.iter_text()), sync_stream=True ) # LOGGING diff --git a/litellm/llms/custom_httpx/llm_http_handler.py b/litellm/llms/custom_httpx/llm_http_handler.py index afa12b58b29..e7da6586b97 100644 --- a/litellm/llms/custom_httpx/llm_http_handler.py +++ b/litellm/llms/custom_httpx/llm_http_handler.py @@ -118,6 +118,7 @@ from litellm.llms.custom_httpx.http_handler import ( get_httpx_client, header_value, ) +from litellm.llms.custom_httpx.sse_lines import aiter_sse_lines, iter_sse_lines from litellm.responses.streaming_iterator import ( BaseResponsesAPIStreamingIterator, MockResponsesAPIStreamingIterator, @@ -1086,7 +1087,7 @@ class BaseLLMHTTPHandler: completion_stream: object = MockResponseIterator(model_response=model_response, json_mode=json_mode) else: completion_stream = provider_config.get_model_response_iterator_for_request( - streaming_response=response.iter_lines(), + streaming_response=iter_sse_lines(response.iter_text()), sync_stream=True, json_mode=json_mode, optional_params=_OPTIONAL_PARAMS.validate_python(optional_params), @@ -1229,7 +1230,7 @@ class BaseLLMHTTPHandler: completion_stream: object = MockResponseIterator(model_response=model_response, json_mode=json_mode) else: completion_stream = provider_config.get_model_response_iterator_for_request( - streaming_response=response.aiter_lines(), + streaming_response=aiter_sse_lines(response.aiter_text()), sync_stream=False, json_mode=json_mode, optional_params=_OPTIONAL_PARAMS.validate_python(optional_params), diff --git a/litellm/llms/custom_httpx/sse_lines.py b/litellm/llms/custom_httpx/sse_lines.py new file mode 100644 index 00000000000..71ea7d621eb --- /dev/null +++ b/litellm/llms/custom_httpx/sse_lines.py @@ -0,0 +1,74 @@ +""" +Split a decoded Server-Sent Events body into lines the way the SSE spec does. + +httpx's `iter_lines()` / `aiter_lines()` follow `str.splitlines()`, which also breaks on U+0085, U+2028, +U+2029, \\x0b, \\x0c and \\x1c-\\x1e. JSON leaves U+0085, U+2028 and U+2029 unescaped inside strings, so a +`data:` line carrying one of them gets cut in half. SSE lines end only at CRLF, LF or CR +(https://html.spec.whatwg.org/multipage/server-sent-events.html#parsing-an-event-stream) +""" + +import io +import re +from collections.abc import AsyncIterable, AsyncIterator, Iterable, Iterator, Sequence +from typing import Final + +_SSE_LINE_BREAK: Final = re.compile(r"\r\n|\r|\n") +_SSE_LINE_WITH_BREAK: Final = re.compile(r"[^\r\n]*(?:\r\n|\r|\n)|[^\r\n]+") + + +def _split_lines(text: str) -> Sequence[str]: + if "\r" not in text: + return text.split("\n") + crlf_folded: Final = text.replace("\r\n", "\n") + if "\r" not in crlf_folded: + return crlf_folded.split("\n") + return _SSE_LINE_BREAK.split(text) + + +class _SSELineSplitter: + def __init__(self) -> None: + self._partial_line: io.StringIO = io.StringIO() + self._after_cr: bool = False + + def feed(self, text: str) -> tuple[str, ...]: + if not text: + return () + fresh: Final = text[1:] if self._after_cr and text.startswith("\n") else text + self._after_cr = fresh.endswith("\r") + *ended, tail = _split_lines(fresh) + if not ended: + _ = self._partial_line.write(tail) + return () + first: Final = self._partial_line.getvalue() + ended[0] + self._partial_line = io.StringIO() + _ = self._partial_line.write(tail) + return (first, *ended[1:]) + + def flush(self) -> tuple[str, ...]: + rest: Final = self._partial_line.getvalue() + self._partial_line = io.StringIO() + return (rest,) if rest else () + + +def iter_sse_lines(text_chunks: Iterable[str]) -> Iterator[str]: + splitter: Final = _SSELineSplitter() + for text in text_chunks: + yield from splitter.feed(text) + yield from splitter.flush() + + +def split_sse_lines(text: str) -> tuple[str, ...]: + return tuple(iter_sse_lines((text,))) + + +def split_sse_lines_keepends(text: str) -> tuple[str, ...]: + return tuple(_SSE_LINE_WITH_BREAK.findall(text)) + + +async def aiter_sse_lines(text_chunks: AsyncIterable[str]) -> AsyncIterator[str]: + splitter: Final = _SSELineSplitter() + async for text in text_chunks: + for line in splitter.feed(text): + yield line + for line in splitter.flush(): + yield line diff --git a/litellm/llms/gemini/google_genai/guardrail_translation/handler.py b/litellm/llms/gemini/google_genai/guardrail_translation/handler.py index 34007bc8fd4..b6b2aa6c28f 100644 --- a/litellm/llms/gemini/google_genai/guardrail_translation/handler.py +++ b/litellm/llms/gemini/google_genai/guardrail_translation/handler.py @@ -23,6 +23,7 @@ from litellm.llms.base_llm.guardrail_translation.base_translation import ( BaseTranslation, StreamTransformSink, ) +from litellm.llms.custom_httpx.sse_lines import split_sse_lines from litellm.types.utils import GenericGuardrailAPIInputs if TYPE_CHECKING: @@ -121,7 +122,7 @@ def _parse_json_dict_or_none(payload: str) -> Mapping[str, object] | None: def _sse_payload_texts(sse_text: str) -> tuple[str, ...]: return tuple( text - for line in sse_text.splitlines() + for line in split_sse_lines(sse_text) if line.startswith("data:") for payload in (line[len("data:") :].strip(),) if payload and payload != "[DONE]" diff --git a/litellm/llms/huggingface/embedding/transformation.py b/litellm/llms/huggingface/embedding/transformation.py index 1b7d6483739..6e6a117227e 100644 --- a/litellm/llms/huggingface/embedding/transformation.py +++ b/litellm/llms/huggingface/embedding/transformation.py @@ -17,6 +17,7 @@ from litellm.litellm_core_utils.prompt_templates.factory import ( ) from litellm.litellm_core_utils.streaming_handler import CustomStreamWrapper from litellm.llms.base_llm.chat.transformation import BaseConfig, BaseLLMException +from litellm.llms.custom_httpx.sse_lines import iter_sse_lines from litellm.secret_managers.main import get_secret_str from litellm.types.llms.openai import AllMessageValues from litellm.types.utils import Choices, Message, ModelResponse, Usage @@ -357,7 +358,7 @@ class HuggingFaceEmbeddingConfig(BaseConfig): api_key: str | None = None, ) -> list[dict[str, str]]: streamed_response: Final = CustomStreamWrapper( - completion_stream=response.iter_lines(), + completion_stream=iter_sse_lines(response.iter_text()), model=model, custom_llm_provider="huggingface", logging_obj=logging_obj, diff --git a/litellm/llms/langgraph/chat/sse_iterator.py b/litellm/llms/langgraph/chat/sse_iterator.py index 575820bceee..6fe355ad98c 100644 --- a/litellm/llms/langgraph/chat/sse_iterator.py +++ b/litellm/llms/langgraph/chat/sse_iterator.py @@ -12,6 +12,7 @@ import httpx from pydantic import TypeAdapter from litellm._logging import verbose_logger +from litellm.llms.custom_httpx.sse_lines import aiter_sse_lines, iter_sse_lines from litellm.types.utils import Delta, ModelResponseStream, StreamingChoices _DECODED_JSON: Final = TypeAdapter(object) @@ -36,12 +37,12 @@ class LangGraphSSEStreamIterator: def __iter__(self): """Initialize sync iteration.""" - self.line_iterator = self.response.iter_lines() + self.line_iterator = iter_sse_lines(self.response.iter_text()) return self def __aiter__(self): """Initialize async iteration.""" - self.async_line_iterator = self.response.aiter_lines() + self.async_line_iterator = aiter_sse_lines(self.response.aiter_text()) return self def _parse_sse_line(self, line: str) -> ModelResponseStream | None: diff --git a/litellm/llms/oobabooga/chat/oobabooga.py b/litellm/llms/oobabooga/chat/oobabooga.py index 8879c20c5eb..b4dc57ff5ed 100644 --- a/litellm/llms/oobabooga/chat/oobabooga.py +++ b/litellm/llms/oobabooga/chat/oobabooga.py @@ -5,6 +5,7 @@ from typing import TYPE_CHECKING, Final import litellm from litellm.litellm_core_utils.url_utils import join_api_base from litellm.llms.custom_httpx.http_handler import get_httpx_client +from litellm.llms.custom_httpx.sse_lines import iter_sse_lines from litellm.utils import EmbeddingResponse, ModelResponse, Usage from ..common_utils import OobaboogaError @@ -74,7 +75,7 @@ def completion( stream=optional_params["stream"] if "stream" in optional_params else False, ) if "stream" in optional_params and optional_params["stream"] is True: - return response.iter_lines() + return iter_sse_lines(response.iter_text()) else: return oobabooga_config.transform_response( model=model, diff --git a/litellm/llms/openai_like/chat/handler.py b/litellm/llms/openai_like/chat/handler.py index 2c74ef19e7d..a450947afdf 100644 --- a/litellm/llms/openai_like/chat/handler.py +++ b/litellm/llms/openai_like/chat/handler.py @@ -16,6 +16,7 @@ from litellm import LlmProviders from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj from litellm.llms.base_llm.base_model_iterator import MockResponseIterator from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler, HTTPHandler +from litellm.llms.custom_httpx.sse_lines import aiter_sse_lines, iter_sse_lines from litellm.llms.databricks.streaming_utils import ModelResponseIterator from litellm.llms.openai.chat.gpt_transformation import OpenAIGPTConfig from litellm.llms.openai.openai import OpenAIConfig @@ -65,7 +66,9 @@ async def make_call( model_response: Final = _fake_streamed_model_response(response.json()) completion_stream = MockResponseIterator(model_response=model_response) else: - completion_stream = ModelResponseIterator(streaming_response=response.aiter_lines(), sync_stream=False) + completion_stream = ModelResponseIterator( + streaming_response=aiter_sse_lines(response.aiter_text()), sync_stream=False + ) # LOGGING logging_obj.post_call( input=messages, @@ -103,7 +106,9 @@ def make_sync_call( model_response: Final = _fake_streamed_model_response(response.json()) completion_stream = MockResponseIterator(model_response=model_response) else: - completion_stream = ModelResponseIterator(streaming_response=response.iter_lines(), sync_stream=True) + completion_stream = ModelResponseIterator( + streaming_response=iter_sse_lines(response.iter_text()), sync_stream=True + ) # LOGGING logging_obj.post_call( diff --git a/litellm/llms/predibase/chat/handler.py b/litellm/llms/predibase/chat/handler.py index d0a61ea6e00..2ba4a63e08b 100644 --- a/litellm/llms/predibase/chat/handler.py +++ b/litellm/llms/predibase/chat/handler.py @@ -14,6 +14,7 @@ from litellm.llms.custom_httpx.http_handler import ( AsyncHTTPHandler, get_async_httpx_client, ) +from litellm.llms.custom_httpx.sse_lines import aiter_sse_lines, iter_sse_lines from litellm.utils import CustomStreamWrapper, ModelResponse from ..common_utils import PredibaseError @@ -34,7 +35,7 @@ async def make_call( if response.status_code != 200: raise PredibaseError(status_code=response.status_code, message=response.text) - completion_stream: Final = response.aiter_lines() + completion_stream: Final = aiter_sse_lines(response.aiter_text()) # LOGGING logging_obj.post_call( input=messages, @@ -163,7 +164,7 @@ class PredibaseChatCompletion: timeout=timeout, ) _response: Final = CustomStreamWrapper( - response.iter_lines(), + iter_sse_lines(response.iter_text()), model, custom_llm_provider="predibase", logging_obj=logging_obj, diff --git a/litellm/llms/vertex_ai/agent_engine/transformation.py b/litellm/llms/vertex_ai/agent_engine/transformation.py index 9f592398f74..0339533e1f5 100644 --- a/litellm/llms/vertex_ai/agent_engine/transformation.py +++ b/litellm/llms/vertex_ai/agent_engine/transformation.py @@ -22,6 +22,7 @@ from litellm.litellm_core_utils.prompt_templates.common_utils import ( convert_content_list_to_str, ) from litellm.llms.base_llm.chat.transformation import BaseConfig, BaseLLMException +from litellm.llms.custom_httpx.sse_lines import aiter_sse_lines, iter_sse_lines from litellm.llms.vertex_ai.agent_engine.sse_iterator import ( VertexAgentEngineResponseIterator, ) @@ -357,7 +358,7 @@ class VertexAgentEngineConfig(BaseConfig, VertexBase): ) -> VertexAgentEngineResponseIterator: """Return a streaming iterator for SSE responses.""" return VertexAgentEngineResponseIterator( - streaming_response=raw_response.iter_lines(), + streaming_response=iter_sse_lines(raw_response.iter_text()), sync_stream=True, ) @@ -465,7 +466,7 @@ class VertexAgentEngineConfig(BaseConfig, VertexBase): # Create iterator for SSE stream (async) completion_stream: Final = VertexAgentEngineResponseIterator( - streaming_response=response.aiter_lines(), + streaming_response=aiter_sse_lines(response.aiter_text()), sync_stream=False, ) diff --git a/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py b/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py index f27593dc94e..c506192f253 100644 --- a/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py +++ b/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py @@ -41,6 +41,7 @@ from litellm.llms.custom_httpx.http_handler import ( get_async_httpx_client, get_httpx_client, ) +from litellm.llms.custom_httpx.sse_lines import aiter_sse_lines, iter_sse_lines from litellm.types.llms.anthropic import AnthropicThinkingParam from litellm.types.llms.base import CachedTokensDetails from litellm.types.llms.gemini import BidiGenerateContentServerMessage @@ -3241,7 +3242,7 @@ async def make_call( raise completion_stream: Final = ModelResponseIterator( - streaming_response=response.aiter_lines(), + streaming_response=aiter_sse_lines(response.aiter_text()), sync_stream=False, logging_obj=logging_obj, response_headers=response.headers, @@ -3312,7 +3313,7 @@ def make_sync_call( raise completion_stream: Final = ModelResponseIterator( - streaming_response=response.iter_lines(), + streaming_response=iter_sse_lines(response.iter_text()), sync_stream=True, logging_obj=logging_obj, response_headers=response.headers, @@ -4190,7 +4191,7 @@ class ModelResponseIterator: log_failed_hop(e) return self._truncated_stream_end(token) self._start_hop(hop_response, token) - self.streaming_response = hop_response.aiter_lines() + self.streaming_response = aiter_sse_lines(hop_response.aiter_text()) self.async_response_iterator = self.streaming_response return None @@ -4204,7 +4205,7 @@ class ModelResponseIterator: log_failed_hop(e) return self._truncated_stream_end(token) self._start_hop(hop_response, token) - self.streaming_response = hop_response.iter_lines() + self.streaming_response = iter_sse_lines(hop_response.iter_text()) self.response_iterator = self.streaming_response return None diff --git a/litellm/proxy/agent_endpoints/a2a_endpoints.py b/litellm/proxy/agent_endpoints/a2a_endpoints.py index 4691c7b0bc5..219ce906340 100644 --- a/litellm/proxy/agent_endpoints/a2a_endpoints.py +++ b/litellm/proxy/agent_endpoints/a2a_endpoints.py @@ -26,6 +26,7 @@ from litellm._logging import verbose_proxy_logger from litellm.litellm_core_utils.core_helpers import as_str_mapping from litellm.litellm_core_utils.url_utils import SSRFError, validate_url from litellm.llms.a2a.common_utils import resolve_a2a_hop_auth_header +from litellm.llms.custom_httpx.sse_lines import aiter_sse_lines from litellm.proxy._types import UserAPIKeyAuth from litellm.proxy.a2a.version_convert import ( A2AVersion, @@ -320,7 +321,7 @@ async def _a2a_sse_event_source( "error": {"code": -32603, "message": resp.reason_phrase}, } return - async for line in resp.aiter_lines(): + async for line in aiter_sse_lines(resp.aiter_text()): stripped = line.strip() if not stripped.startswith("data:"): continue diff --git a/litellm/proxy/anthropic_endpoints/streaming_model_restamp.py b/litellm/proxy/anthropic_endpoints/streaming_model_restamp.py index f748a754fe0..32e0f6e8682 100644 --- a/litellm/proxy/anthropic_endpoints/streaming_model_restamp.py +++ b/litellm/proxy/anthropic_endpoints/streaming_model_restamp.py @@ -14,6 +14,8 @@ from typing import Final from pydantic import TypeAdapter, ValidationError +from litellm.llms.custom_httpx.sse_lines import split_sse_lines_keepends + _MESSAGE_START_EVENT: Final = "message_start" _MESSAGE_START_MARKER: Final = b"message_start" _SSE_DATA_FIELD: Final = "data:" @@ -52,7 +54,7 @@ def _restamped_data_line(line: str, requested_model: str) -> str | None: def _restamped_frame(frame: str, requested_model: str) -> str | None: - lines: Final = frame.splitlines(keepends=True) + lines: Final = split_sse_lines_keepends(frame) restamped: Final = tuple(_restamped_data_line(line, requested_model) for line in lines) if all(line is None for line in restamped): return None diff --git a/litellm/proxy/pass_through_endpoints/llm_provider_handlers/anthropic_passthrough_logging_handler.py b/litellm/proxy/pass_through_endpoints/llm_provider_handlers/anthropic_passthrough_logging_handler.py index 7c7eb51bc5d..5415314f29b 100644 --- a/litellm/proxy/pass_through_endpoints/llm_provider_handlers/anthropic_passthrough_logging_handler.py +++ b/litellm/proxy/pass_through_endpoints/llm_provider_handlers/anthropic_passthrough_logging_handler.py @@ -29,6 +29,7 @@ from litellm.llms.anthropic.messages.stream_assembly import ( split_sse_chunk_into_events, ) from litellm.llms.bedrock.common_utils import BEDROCK_INVOKE_TRACE_FIELD +from litellm.llms.custom_httpx.sse_lines import split_sse_lines from litellm.proxy._types import PassThroughEndpointLoggingTypedDict from litellm.proxy.auth.auth_utils import get_end_user_id_from_request_body from litellm.proxy.pass_through_endpoints.llm_provider_handlers.batch_attribution import ( @@ -61,7 +62,7 @@ _JSON_OBJECT: Final = TypeAdapter(Mapping[str, object], config=ConfigDict(hide_i def _event_payload(event: str) -> Mapping[str, object] | None: data_lines: Final = tuple( - line.strip()[len("data:") :].strip() for line in event.splitlines() if line.strip().startswith("data:") + line.strip()[len("data:") :].strip() for line in split_sse_lines(event) if line.strip().startswith("data:") ) if not data_lines or data_lines[0] in ("", "[DONE]"): return None @@ -362,7 +363,7 @@ class AnthropicPassthroughLoggingHandler: ) -> str | None: for raw in all_chunks: text = raw.decode("utf-8") if isinstance(raw, bytes) else raw - for line in text.splitlines(): + for line in split_sse_lines(text): if not line.startswith("data:"): continue try: @@ -402,7 +403,7 @@ class AnthropicPassthroughLoggingHandler: """ for raw in reversed(all_chunks): text = raw.decode("utf-8") if isinstance(raw, bytes) else raw - for line in reversed(text.splitlines()): + for line in reversed(split_sse_lines(text)): if not line.startswith("data:"): continue try: diff --git a/litellm/proxy/pass_through_endpoints/llm_provider_handlers/tinyfish_passthrough_logging_handler.py b/litellm/proxy/pass_through_endpoints/llm_provider_handlers/tinyfish_passthrough_logging_handler.py index 2afc1d17bc2..0c4a3f64852 100644 --- a/litellm/proxy/pass_through_endpoints/llm_provider_handlers/tinyfish_passthrough_logging_handler.py +++ b/litellm/proxy/pass_through_endpoints/llm_provider_handlers/tinyfish_passthrough_logging_handler.py @@ -21,6 +21,7 @@ from litellm.llms.custom_httpx.http_handler import ( AsyncHTTPHandler, get_async_httpx_client, ) +from litellm.llms.custom_httpx.sse_lines import split_sse_lines from litellm.proxy._types import PassThroughEndpointLoggingTypedDict from litellm.types.llms.custom_http import httpxSpecialProvider from litellm.types.passthrough_endpoints.tinyfish import ( @@ -81,7 +82,7 @@ def sse_poller_spawned(logging_obj: LiteLLMLoggingObj) -> bool: def run_id_from_sse_frames(frames: bytes) -> str | None: - return _run_id_from_sse_chunks(frames.decode("utf-8", errors="replace").splitlines()) + return _run_id_from_sse_chunks(split_sse_lines(frames.decode("utf-8", errors="replace"))) def resolve_tinyfish_agent_api_base() -> str: diff --git a/litellm/proxy/pass_through_endpoints/managed_id_rewriter.py b/litellm/proxy/pass_through_endpoints/managed_id_rewriter.py index 9790c150ae1..cae8ccc93ee 100644 --- a/litellm/proxy/pass_through_endpoints/managed_id_rewriter.py +++ b/litellm/proxy/pass_through_endpoints/managed_id_rewriter.py @@ -51,6 +51,7 @@ from litellm.llms.base_llm.managed_resources.isolation import ( can_access_resource, resolve_resource_owner_id, ) +from litellm.llms.custom_httpx.sse_lines import split_sse_lines from litellm.proxy._types import UserAPIKeyAuth from litellm.proxy.batches_endpoints.common_utils import validate_batch_list_limit from litellm.proxy.common_utils.sse_keepalive import split_complete_sse_frames @@ -830,7 +831,7 @@ _SSE_EVENT_ADAPTER: Final = TypeAdapter(Mapping[str, JsonValue]) def _first_streamed_response(frames: bytes) -> tuple[str, Mapping[str, JsonValue]] | None: - for line in frames.decode("utf-8", errors="replace").splitlines(): + for line in split_sse_lines(frames.decode("utf-8", errors="replace")): if not line.startswith(_SSE_DATA_PREFIX): continue try: diff --git a/litellm/proxy/response_polling/background_streaming.py b/litellm/proxy/response_polling/background_streaming.py index 5f6a7432afb..1eb76f8f558 100644 --- a/litellm/proxy/response_polling/background_streaming.py +++ b/litellm/proxy/response_polling/background_streaming.py @@ -18,6 +18,7 @@ from starlette.types import Message from typing_extensions import ReadOnly, TypedDict from litellm._logging import verbose_proxy_logger +from litellm.llms.custom_httpx.sse_lines import split_sse_lines from litellm.proxy.auth.user_api_key_auth import UserAPIKeyAuth from litellm.proxy.common_request_processing import ProxyBaseLLMRequestProcessing from litellm.proxy.response_polling.polling_handler import ResponsePollingHandler @@ -81,7 +82,7 @@ class _StreamsBody(Protocol): def _sse_frame_data(frame: str) -> str | None: - return next((line[6:].strip() for line in frame.splitlines() if line.startswith("data: ")), None) + return next((line[6:].strip() for line in split_sse_lines(frame) if line.startswith("data: ")), None) async def _never_receive() -> Message: diff --git a/litellm/responses/litellm_completion_transformation/streaming_iterator.py b/litellm/responses/litellm_completion_transformation/streaming_iterator.py index 219944de521..982158040c4 100644 --- a/litellm/responses/litellm_completion_transformation/streaming_iterator.py +++ b/litellm/responses/litellm_completion_transformation/streaming_iterator.py @@ -165,6 +165,7 @@ class LiteLLMCompletionStreamingIterator(ResponsesAPIStreamingIterator): self._upstream_exhausted: bool = False self._response_id_primed: bool = False self._pending_tool_events: list[BaseLiteLLMOpenAIResponseObject] = [] + self._tool_calls_held_behind_text: tuple[object, ...] = () self._tool_output_index_by_call_id: dict[str, int] = {} self._tool_args_by_call_id: dict[str, str] = {} self._tool_item_id_by_call_id: dict[str, str] = {} # mutable-ok: filled per call id as tool call events stream @@ -388,6 +389,15 @@ class LiteLLMCompletionStreamingIterator(ResponsesAPIStreamingIterator): return "" return self._tool_call_id_by_index.get(index, "") + def _hold_tool_calls_behind_text(self, tool_calls: object) -> None: + if isinstance(tool_calls, list): + self._tool_calls_held_behind_text = (*self._tool_calls_held_behind_text, *tool_calls) + + def _release_held_tool_calls(self) -> tuple[object, ...]: + held: Final = self._tool_calls_held_behind_text + self._tool_calls_held_behind_text = () + return held + def _queue_tool_call_delta_events(self, tool_calls: object) -> None: """ Convert chat-completions streaming `tool_calls` deltas into Responses API streaming events. @@ -1365,6 +1375,7 @@ class LiteLLMCompletionStreamingIterator(ResponsesAPIStreamingIterator): if delta_content: if not self.sent_message_item_added_event: self._queue_message_item_added_events() + self._hold_tool_calls_behind_text(getattr(chunk.choices[0].delta, "tool_calls", None)) self._sequence_number += 1 text_delta_event: Final = OutputTextDeltaEvent( type=ResponsesAPIStreamEvents.OUTPUT_TEXT_DELTA, @@ -1379,7 +1390,7 @@ class LiteLLMCompletionStreamingIterator(ResponsesAPIStreamingIterator): # Priority 3: Handle tool call deltas (if any) -> queue events and emit them # For each tool call delta, we emit events one at a time to match OpenAI's streaming behavior if chunk.choices and hasattr(chunk.choices[0].delta, "tool_calls") and chunk.choices[0].delta.tool_calls: - self._queue_tool_call_delta_events(chunk.choices[0].delta.tool_calls) + self._queue_tool_call_delta_events([*self._release_held_tool_calls(), *chunk.choices[0].delta.tool_calls]) # Return one pending tool event at a time if self._pending_tool_events: return self._pending_tool_events.pop(0) diff --git a/tests/benchmarks/test_sse_lines_benchmarks.py b/tests/benchmarks/test_sse_lines_benchmarks.py new file mode 100644 index 00000000000..1320fae49a6 --- /dev/null +++ b/tests/benchmarks/test_sse_lines_benchmarks.py @@ -0,0 +1,27 @@ +from collections.abc import Sequence +from typing import Final + +import pytest + +from litellm.llms.custom_httpx.sse_lines import iter_sse_lines + +EVENT: Final = 'data: {"choices":[{"delta":{"content":"hello world"}}]}' + + +def _chunks(line_break: str) -> Sequence[str]: + body: Final = "".join(f"{EVENT}{line_break}{line_break}" for _ in range(2000)) + return tuple(body[i : i + 512] for i in range(0, len(body), 512)) + + +LF_CHUNKS: Final = _chunks("\n") +CRLF_CHUNKS: Final = _chunks("\r\n") + + +@pytest.mark.benchmark +def test_iter_sse_lines_lf_stream() -> None: + assert sum(1 for _ in iter_sse_lines(LF_CHUNKS)) == 4000 + + +@pytest.mark.benchmark +def test_iter_sse_lines_crlf_stream() -> None: + assert sum(1 for _ in iter_sse_lines(CRLF_CHUNKS)) == 4000 diff --git a/tests/unit/a2a_protocol/providers/bedrock_agentcore/test_bedrock_agentcore_a2a.py b/tests/unit/a2a_protocol/providers/bedrock_agentcore/test_bedrock_agentcore_a2a.py index 09f09170241..cd3f7622fa2 100644 --- a/tests/unit/a2a_protocol/providers/bedrock_agentcore/test_bedrock_agentcore_a2a.py +++ b/tests/unit/a2a_protocol/providers/bedrock_agentcore/test_bedrock_agentcore_a2a.py @@ -10,6 +10,7 @@ Verifies that: """ import json +from typing import Final from unittest.mock import AsyncMock, MagicMock, patch import httpx @@ -560,6 +561,37 @@ class TestStreaming: assert sent_body["params"]["message"]["messageId"] == "msg-001" assert [event["result"]["kind"] for event in events] == ["task", "status-update"] + @pytest.mark.asyncio + @pytest.mark.usefixtures("httpx_transport") + async def test_streaming_keeps_unicode_line_separators_inside_an_sse_event(self) -> None: + from litellm.a2a_protocol.providers.bedrock_agentcore.config import ( + BedrockAgentCoreA2AConfig, + ) + + event: Final = { + "jsonrpc": "2.0", + "id": "req-001", + "result": {"kind": "artifact-update", "artifact": {"parts": [{"kind": "text", "text": "a\u2028b\u2029c"}]}}, + } + with respx.mock(assert_all_called=True) as router: + router.post(url__regex=r".*/invocations.*").mock( + return_value=httpx.Response( + 200, + headers={"content-type": "text/event-stream"}, + text=f"data: {json.dumps(event, ensure_ascii=False)}\n\n", + ) + ) + events: Final = [ + streamed + async for streamed in BedrockAgentCoreA2AConfig().handle_streaming( + request_id="req-001", + params=SAMPLE_PARAMS, + litellm_params=SAMPLE_LITELLM_PARAMS, + ) + ] + + assert events == [event] + class TestConfigManager: """Test that config manager routes 'bedrock' correctly.""" diff --git a/tests/unit/a2a_protocol/providers/watsonx_orchestrate/test_watsonx_orchestrate_transformation.py b/tests/unit/a2a_protocol/providers/watsonx_orchestrate/test_watsonx_orchestrate_transformation.py index a300560ae9d..ac3f0390f84 100644 --- a/tests/unit/a2a_protocol/providers/watsonx_orchestrate/test_watsonx_orchestrate_transformation.py +++ b/tests/unit/a2a_protocol/providers/watsonx_orchestrate/test_watsonx_orchestrate_transformation.py @@ -1,6 +1,7 @@ import asyncio import json import time +from typing import Final import httpx import pytest @@ -36,13 +37,8 @@ class _ShortTtlTokenClient: return _JsonResponse({"access_token": f"token-{self.calls}", "expires_in": 30}) -class _SSELines: - def __init__(self, lines): - self.lines = lines - - async def aiter_lines(self): - for line in self.lines: - yield line +def _sse_response(lines: list[str]) -> httpx.Response: + return httpx.Response(200, stream=httpx.ByteStream("".join(f"{line}\n" for line in lines).encode())) class _InvalidJsonStreamResponse: @@ -218,7 +214,7 @@ def test_cp4d_token_ttl_from_absolute_expiration(): @pytest.mark.asyncio async def test_accumulate_wxo_sse_text_ignores_non_dict_json_events(): - response = _SSELines( + response = _sse_response( [ "data: null", "data: true", @@ -230,6 +226,14 @@ async def test_accumulate_wxo_sse_text_ignores_non_dict_json_events(): ) +@pytest.mark.asyncio +async def test_accumulate_wxo_sse_text_keeps_unicode_line_separators_inside_an_event() -> None: + separated_text: Final = "before\u2028between\u2029after" + response: Final = _sse_response([f"data: {json.dumps({'results': separated_text}, ensure_ascii=False)}", ""]) + + assert await WatsonxOrchestrateHandler._accumulate_wxo_sse_text(response) == separated_text + + @pytest.mark.asyncio async def test_short_lived_tokens_are_not_served_from_cache(): client = _ShortTtlTokenClient() diff --git a/tests/unit/completion_extras/litellm_responses_transformation/test_completion_extras_litellm_responses_transformation_transformation.py b/tests/unit/completion_extras/litellm_responses_transformation/test_completion_extras_litellm_responses_transformation_transformation.py index 9aa494c1aab..46f1dbf2e7a 100644 --- a/tests/unit/completion_extras/litellm_responses_transformation/test_completion_extras_litellm_responses_transformation_transformation.py +++ b/tests/unit/completion_extras/litellm_responses_transformation/test_completion_extras_litellm_responses_transformation_transformation.py @@ -1102,6 +1102,24 @@ def test_recover_output_items_merges_text_only_items_at_distinct_indices(): assert recovered[1]["content"][0]["text"] == "From OUTPUT_TEXT_DONE only" +def test_recover_output_items_keeps_an_item_whose_text_holds_unicode_line_separators() -> None: + from litellm.completion_extras.litellm_responses_transformation.transformation import ( + LiteLLMResponsesTransformationHandler, + ) + + item: Final = { + "type": "message", + "id": "msg_item_0", + "role": "assistant", + "status": "completed", + "content": [{"type": "output_text", "text": "From\u2028OUTPUT_ITEM\u2029DONE", "annotations": []}], + } + event: Final = {"type": "response.output_item.done", "output_index": 0, "item": item} + raw_sse: Final = f"data: {json.dumps(event, ensure_ascii=False)}\ndata: [DONE]\n" + + assert LiteLLMResponsesTransformationHandler._recover_output_items_from_raw_sse(raw_sse) == [item] + + def test_transform_response_prefers_completed_output_from_raw_sse(): from litellm.completion_extras.litellm_responses_transformation.transformation import ( LiteLLMResponsesTransformationHandler, diff --git a/tests/unit/google_genai/test_google_genai_streaming_iterator.py b/tests/unit/google_genai/test_google_genai_streaming_iterator.py index 120c99b6e42..cbb9638b3d9 100644 --- a/tests/unit/google_genai/test_google_genai_streaming_iterator.py +++ b/tests/unit/google_genai/test_google_genai_streaming_iterator.py @@ -62,7 +62,9 @@ def _logging_obj( ], ) def test_streaming_logging_targets_the_provider_that_served_the_request( - iterator_cls: type, + iterator_cls: ( + type[AsyncGoogleGenAIGenerateContentStreamingIterator] | type[GoogleGenAIGenerateContentStreamingIterator] + ), custom_llm_provider: str, expected_endpoint_type: EndpointType, ): @@ -99,18 +101,15 @@ def _large_inline_data_event() -> str: return f"data: {json.dumps(payload)}" +def _sse_response(body: str) -> httpx.Response: + return httpx.Response(200, stream=httpx.ByteStream(body.encode())) + + @pytest.mark.asyncio async def test_async_streaming_iterator_yields_complete_sse_events(): """Large inlineData must not be split across byte-chunk boundaries.""" - mock_response = MagicMock() - - async def _aiter_lines(): - yield _large_inline_data_event() - - mock_response.aiter_lines = _aiter_lines - iterator = AsyncGoogleGenAIGenerateContentStreamingIterator( - response=mock_response, + response=_sse_response(_large_inline_data_event() + "\n\n"), model="gemini-3.1-flash-image-preview", logging_obj=_logging_obj(), generate_content_provider_config=MagicMock(), @@ -128,11 +127,8 @@ async def test_async_streaming_iterator_yields_complete_sse_events(): def test_sync_streaming_iterator_yields_complete_sse_events(): - mock_response = MagicMock() - mock_response.iter_lines.return_value = iter([_large_inline_data_event()]) - iterator = GoogleGenAIGenerateContentStreamingIterator( - response=mock_response, + response=_sse_response(_large_inline_data_event() + "\n\n"), model="gemini-3.1-flash-image-preview", logging_obj=_logging_obj(), generate_content_provider_config=MagicMock(), @@ -150,17 +146,8 @@ def test_sync_streaming_iterator_yields_complete_sse_events(): @pytest.mark.asyncio async def test_async_streaming_iterator_preserves_multi_field_sse_event(): - mock_response = MagicMock() - - async def _aiter_lines(): - yield "event: message" - yield 'data: {"text":"hi"}' - yield "" - - mock_response.aiter_lines = _aiter_lines - iterator = AsyncGoogleGenAIGenerateContentStreamingIterator( - response=mock_response, + response=_sse_response('event: message\ndata: {"text":"hi"}\n\n'), model="gemini-test", logging_obj=_logging_obj(), generate_content_provider_config=MagicMock(), @@ -174,16 +161,8 @@ async def test_async_streaming_iterator_preserves_multi_field_sse_event(): @pytest.mark.asyncio async def test_async_streaming_iterator_forwards_sse_comment_events(): - mock_response = MagicMock() - - async def _aiter_lines(): - yield ": keepalive" - yield "" - - mock_response.aiter_lines = _aiter_lines - iterator = AsyncGoogleGenAIGenerateContentStreamingIterator( - response=mock_response, + response=_sse_response(": keepalive\n\n"), model="gemini-test", logging_obj=_logging_obj(), generate_content_provider_config=MagicMock(), @@ -195,6 +174,46 @@ async def test_async_streaming_iterator_forwards_sse_comment_events(): assert chunk == b": keepalive\n\n" +_LINE_SEPARATED_TEXT: Final = "before\u2028between\u2029after" + + +async def _first_chunk( + iterator: GoogleGenAIGenerateContentStreamingIterator | AsyncGoogleGenAIGenerateContentStreamingIterator, +) -> bytes: + if isinstance(iterator, AsyncGoogleGenAIGenerateContentStreamingIterator): + return await iterator.__anext__() + return next(iterator) + + +@pytest.mark.asyncio +@pytest.mark.parametrize( + "iterator_cls", + [AsyncGoogleGenAIGenerateContentStreamingIterator, GoogleGenAIGenerateContentStreamingIterator], +) +async def test_streaming_iterator_keeps_unicode_line_separators_inside_the_event( + iterator_cls: ( + type[AsyncGoogleGenAIGenerateContentStreamingIterator] | type[GoogleGenAIGenerateContentStreamingIterator] + ), +) -> None: + payload: Final = {"candidates": [{"content": {"parts": [{"text": _LINE_SEPARATED_TEXT}]}}]} + event: Final = f"data: {json.dumps(payload, ensure_ascii=False)}\n\n" + iterator: Final = iterator_cls( + response=_sse_response(event), + model="gemini-3.8-flash", + logging_obj=_logging_obj(), + generate_content_provider_config=MagicMock(), + litellm_metadata={}, + custom_llm_provider="gemini", + ) + + chunk: Final = await _first_chunk(iterator) + + assert chunk == event.encode() + assert ( + json.loads(chunk[len(b"data: ") : -2])["candidates"][0]["content"]["parts"][0]["text"] == _LINE_SEPARATED_TEXT + ) + + def _gemini_event(text: str, finish_reason: str | None = None) -> str: candidate: Final = {"content": {"role": "model", "parts": [{"text": text}]}, "index": 0} payload: Final = { diff --git a/tests/unit/interactions/test_interactions_streaming_iterator.py b/tests/unit/interactions/test_interactions_streaming_iterator.py index afcd0d559d7..fb1fe92d637 100644 --- a/tests/unit/interactions/test_interactions_streaming_iterator.py +++ b/tests/unit/interactions/test_interactions_streaming_iterator.py @@ -16,7 +16,10 @@ import pytest import litellm from litellm.integrations.custom_logger import CustomLogger from litellm.interactions import streaming_iterator as interactions_streaming_iterator_module -from litellm.interactions.streaming_iterator import InteractionsAPIStreamingIterator +from litellm.interactions.streaming_iterator import ( + InteractionsAPIStreamingIterator, + SyncInteractionsAPIStreamingIterator, +) from litellm.litellm_core_utils import thread_pool_executor as thread_pool_executor_module from litellm.litellm_core_utils.litellm_logging import Logging as LitellmLogging from litellm.llms.gemini.interactions.transformation import GoogleAIStudioInteractionsConfig @@ -148,3 +151,43 @@ async def test_stream_logs_its_first_event_time_measured_from_the_call_start(): assert first_event_at is not None assert recorder.logged == [(call_start, first_event_at)] + + +_LINE_SEPARATED_TEXT: Final = "before\u2028between\u2029after" + + +async def _first_event( + iterator: InteractionsAPIStreamingIterator | SyncInteractionsAPIStreamingIterator, +) -> InteractionsAPIStreamingResponse: + if isinstance(iterator, InteractionsAPIStreamingIterator): + return await iterator.__anext__() + return next(iterator) + + +@pytest.mark.asyncio +@pytest.mark.parametrize("iterator_cls", [InteractionsAPIStreamingIterator, SyncInteractionsAPIStreamingIterator]) +async def test_stream_keeps_unicode_line_separators_inside_the_delta_text( + iterator_cls: type[InteractionsAPIStreamingIterator] | type[SyncInteractionsAPIStreamingIterator], +) -> None: + delta: Final = {"type": "text", "text": _LINE_SEPARATED_TEXT} + event: Final = {"event_type": "step.delta", "id": "interaction-1", "delta": delta} + iterator: Final = iterator_cls( + response=httpx.Response( + 200, stream=httpx.ByteStream(f"data: {json.dumps(event, ensure_ascii=False)}\n\n".encode()) + ), + model="gemini/gemini-3.1-pro-preview", + interactions_api_config=GoogleAIStudioInteractionsConfig(), + logging_obj=LitellmLogging( + model="gemini/gemini-3.1-pro-preview", + messages=[{"role": "user", "content": "hi"}], + stream=True, + call_type="ainteraction", + start_time=datetime(2026, 10, 9, 12, 0, 0), + litellm_call_id="interactions-separators", + function_id="interactions-separators", + ), + ) + + first: Final = await _first_event(iterator) + + assert first.delta == delta diff --git a/tests/unit/litellm_core_utils/test_streaming_handler.py b/tests/unit/litellm_core_utils/test_streaming_handler.py index a1b58521d7e..79978ecfe8e 100644 --- a/tests/unit/litellm_core_utils/test_streaming_handler.py +++ b/tests/unit/litellm_core_utils/test_streaming_handler.py @@ -506,6 +506,118 @@ def test_multi_chunk_reasoning_and_content( assert initialized_custom_stream_wrapper.sent_last_thinking_block is True +def _stream_chunk(delta: dict[str, object], finish_reason: str | None = None) -> ModelResponseStream: + return ModelResponseStream(choices=[{"index": 0, "delta": delta, "finish_reason": finish_reason}]) + + +def _content_through_first_call(deltas: Sequence[Delta]) -> str: + first_call: Final = next( + (index for index, delta in enumerate(deltas) if delta.tool_calls or delta.function_call), len(deltas) + ) + return "".join(delta.content or "" for delta in deltas[: first_call + 1]) + + +_GET_WEATHER_CALL: Final = { + "index": 0, + "id": "call_1", + "type": "function", + "function": {"name": "get_weather", "arguments": '{"city": "Paris"}'}, +} + + +@pytest.mark.asyncio +@pytest.mark.parametrize("is_async", [True, False]) +@pytest.mark.parametrize( + "chunks,expected_content,expected_tool_calls", + [ + ( + ( + _stream_chunk({"reasoning_content": "Check the weather"}), + _stream_chunk({"tool_calls": [_GET_WEATHER_CALL]}), + _stream_chunk({}, "tool_calls"), + ), + "Check the weather", + 1, + ), + ( + ( + _stream_chunk({"reasoning_content": "Pick a function"}), + _stream_chunk({"function_call": {"name": "get_weather", "arguments": "{}"}}), + _stream_chunk({}, "function_call"), + ), + "Pick a function", + 0, + ), + ( + ( + _stream_chunk({"reasoning_content": "Step one"}), + _stream_chunk({"reasoning_content": ", step two"}), + _stream_chunk({}, "length"), + ), + "Step one, step two", + 0, + ), + ( + ( + _stream_chunk({"reasoning_content": "a"}), + _stream_chunk({"content": "b"}), + _stream_chunk({"reasoning_content": "c"}), + _stream_chunk({"content": "d"}), + _stream_chunk({}, "stop"), + ), + "abcd", + 0, + ), + ( + (_stream_chunk({"reasoning_content": "Still thinking"}),), + "Still thinking", + 0, + ), + ( + (_stream_chunk({"content": "hi"}), _stream_chunk({}, "stop")), + "hi", + 0, + ), + ], + ids=[ + "reasoning_then_tool_call", + "reasoning_then_function_call", + "reasoning_cut_by_length", + "reasoning_after_answer", + "stream_ends_mid_reasoning", + "answer_without_reasoning", + ], +) +async def test_merged_reasoning_stream_keeps_think_tags_balanced( + logging_obj: Logging, + is_async: bool, + chunks: tuple[ModelResponseStream, ...], + expected_content: str, + expected_tool_calls: int, +) -> None: + response: Final = CustomStreamWrapper( + completion_stream=ModelResponseListIterator(model_responses=[chunk.model_copy(deep=True) for chunk in chunks]), + model="deepseek-reasoner", + logging_obj=logging_obj, + ) + response.merge_reasoning_content_in_choices = True + + deltas: Final = ( + tuple([chunk.choices[0].delta async for chunk in response]) + if is_async + else tuple(chunk.choices[0].delta for chunk in response) + ) + logged: Final = litellm.stream_chunk_builder(chunks=response.chunks) + + assert "".join(delta.content or "" for delta in deltas) == expected_content + assert _content_through_first_call(deltas) == expected_content + assert sum(len(delta.tool_calls or ()) for delta in deltas) == expected_tool_calls + assert logged.choices[0].message.content == (expected_content or None) + assert any(chunk.choices[0].finish_reason for chunk in response.chunks if chunk.choices) == any( + chunk.choices[0].finish_reason for chunk in chunks + ) + + def test_strip_sse_data_from_chunk(): """Test the static method that strips 'data: ' prefix from SSE chunks""" # Test with string inputs @@ -7849,7 +7961,7 @@ def gemini_mock_post_streaming(url, **kwargs): mock_response = MagicMock() mock_response.status_code = 200 mock_response.headers = {"Content-Type": "text/event-stream"} - mock_response.iter_lines = MagicMock(return_value=stream_response()) + mock_response.iter_text = MagicMock(return_value=stream_response()) return mock_response diff --git a/tests/unit/llms/anthropic/chat/test_anthropic_chat_handler.py b/tests/unit/llms/anthropic/chat/test_anthropic_chat_handler.py index 1f4a5aaa256..355d407001c 100644 --- a/tests/unit/llms/anthropic/chat/test_anthropic_chat_handler.py +++ b/tests/unit/llms/anthropic/chat/test_anthropic_chat_handler.py @@ -15,7 +15,7 @@ import litellm from litellm._uuid import uuid from litellm.constants import RESPONSE_FORMAT_TOOL_NAME from litellm.litellm_core_utils.prompt_templates.factory import anthropic_messages_pt -from litellm.llms.anthropic.chat.handler import ModelResponseIterator, _envelope_verdict, make_call +from litellm.llms.anthropic.chat.handler import ModelResponseIterator, _envelope_verdict, make_call, make_sync_call from litellm.llms.base_llm.base_utils import response_format_schema, unwrap_response_format_envelope from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler, HTTPHandler from litellm.types.llms.openai import ( @@ -27,24 +27,34 @@ from litellm.types.utils import ChatCompletionDeltaToolCall, ModelResponseStream from tests.unit.llms.anthropic.anthropic_docs_refusal import ANTHROPIC_DOCS_REFUSAL_MESSAGE +def _text_delta_stream(text: str) -> bytes: + events: Final = ( + ("content_block_start", {"type": "content_block_start", "index": 0, "content_block": {"type": "text", "text": ""}}), + ("content_block_delta", {"type": "content_block_delta", "index": 0, "delta": {"type": "text_delta", "text": text}}), + ) + return "".join( + f"event: {name}\ndata: {json.dumps(payload, ensure_ascii=False)}\n\n" for name, payload in events + ).encode() + + +def _streamed_text(chunks: list[object]) -> str: + return "".join(chunk.choices[0].delta.content or "" for chunk in chunks if isinstance(chunk, ModelResponseStream)) + + @pytest.mark.asyncio async def test_make_call_passes_logging_obj_to_client_post(): """make_call must pass logging_obj to client.post so track_llm_api_timing can set llm_api_duration_ms for litellm_overhead_time_ms.""" mock_client = AsyncMock() - mock_response = MagicMock() - mock_response.aiter_lines = MagicMock( - return_value=iter([b'data: {"type":"message_start"}\n', b'data: {"type":"message_delta"}\n']) - ) - mock_client.post.return_value = mock_response + mock_client.post.return_value = httpx.Response(200, stream=httpx.ByteStream(_text_delta_stream("Hi there"))) logging_obj = MagicMock() - await make_call( + completion_stream, _ = await make_call( client=mock_client, api_base="https://api.anthropic.com/v1/messages", headers={}, data="{}", - model="claude-3-5-haiku", + model="claude-haiku-5-5", messages=[{"role": "user", "content": "Hi"}], logging_obj=logging_obj, timeout=60.0, @@ -54,6 +64,39 @@ async def test_make_call_passes_logging_obj_to_client_post(): mock_client.post.assert_called_once() call_kwargs = mock_client.post.call_args[1] assert call_kwargs.get("logging_obj") is logging_obj + assert _streamed_text([chunk async for chunk in completion_stream]) == "Hi there" + + +_LINE_SEPARATED_TEXT: Final = "before\u2028between\u2029after" + + +async def _streamed_chunks(sync_stream: bool, text: str) -> list[object]: + transport: Final = httpx.MockTransport( + lambda request: httpx.Response(200, stream=httpx.ByteStream(_text_delta_stream(text))) + ) + call_args: Final = { + "api_base": "https://api.anthropic.com/v1/messages", + "headers": {}, + "data": "{}", + "model": "claude-haiku-5-5", + "messages": [{"role": "user", "content": "Hi"}], + "logging_obj": MagicMock(), + "timeout": 60.0, + "json_mode": False, + } + if sync_stream: + sync_iterator, _ = make_sync_call(client=HTTPHandler(client=httpx.Client(transport=transport)), **call_args) + return list(sync_iterator) + async_iterator, _ = await make_call(client=AsyncHTTPHandler(transport=transport), **call_args) + return [chunk async for chunk in async_iterator] + + +@pytest.mark.asyncio +@pytest.mark.parametrize("sync_stream", [True, False]) +async def test_streamed_text_keeps_unicode_line_separators(sync_stream: bool) -> None: + chunks: Final = await _streamed_chunks(sync_stream, _LINE_SEPARATED_TEXT) + + assert _streamed_text(chunks) == _LINE_SEPARATED_TEXT def test_anthropic_completion_does_not_send_deployment_default_limits(): diff --git a/tests/unit/llms/anthropic/messages/test_stream_assembly.py b/tests/unit/llms/anthropic/messages/test_stream_assembly.py index f7d6b8d4f8c..4993d2fa9bf 100644 --- a/tests/unit/llms/anthropic/messages/test_stream_assembly.py +++ b/tests/unit/llms/anthropic/messages/test_stream_assembly.py @@ -144,6 +144,30 @@ def test_anthropic_message_delta_fields_reads_the_fields_the_openai_shape_drops( assert dict(anthropic_message_delta_fields(GOOGLE_STREAM)) == {} +def test_anthropic_message_delta_fields_reads_a_message_delta_whose_json_holds_unicode_line_separators() -> None: + safeguard_results: Final = [ + { + "type": "dangerous_tool_use", + "status": { + "type": "available", + "tool_uses": { + "toolu_01": {"type": "evaluated", "outcome": "flagged", "explanation": "deletes\u2028the\u2029root"} + }, + }, + } + ] + message_delta: Final = { + "type": "message_delta", + "delta": {"stop_reason": "end_turn", "safeguard_results": safeguard_results}, + "usage": {"output_tokens": 2}, + } + raw_message_delta_frame: Final = f"event: message_delta\ndata: {json.dumps(message_delta, ensure_ascii=False)}\n\n" + + fields: Final = anthropic_message_delta_fields((*ANTHROPIC_STREAM[:5], raw_message_delta_frame.encode())) + + assert dict(fields) == {"safeguard_results": safeguard_results} + + @pytest.mark.parametrize( "chunks", [GOOGLE_STREAM, INVALID_UTF8, (ANTHROPIC_ERROR,), ()], diff --git a/tests/unit/llms/azure_ai/agents/test_handler.py b/tests/unit/llms/azure_ai/agents/test_handler.py index 71466bb172c..14b6094747f 100644 --- a/tests/unit/llms/azure_ai/agents/test_handler.py +++ b/tests/unit/llms/azure_ai/agents/test_handler.py @@ -173,7 +173,7 @@ async def test_handler_acompletion_with_thread_id_reuses_that_thread() -> None: def _sse(event: str, data: dict[str, object] | str) -> str: - return f"event: {event}\ndata: {data if isinstance(data, str) else json.dumps(data)}\n\n" + return f"event: {event}\ndata: {data if isinstance(data, str) else json.dumps(data, ensure_ascii=False)}\n\n" def _message_delta(text: str) -> dict[str, object]: @@ -222,3 +222,24 @@ async def test_acompletion_stream_yields_agent_message_deltas_then_stop(respx_mo "thread": {"messages": [{"role": "user", "content": "Hi Agent, what is 10 + 5?"}]}, } assert chunks == [("10 + 5", None), (" = 15", None), (None, "stop")] + + +@pytest.mark.asyncio +async def test_acompletion_stream_keeps_unicode_line_separators_inside_a_delta(respx_mock: respx.MockRouter) -> None: + separated_text: Final = "before\u2028between\u2029after" + _ = respx_mock.post(f"{API_BASE}/threads/runs").respond( + 200, + text=_sse("thread.message.delta", _message_delta(separated_text)) + _sse("done", "[DONE]"), + headers={"content-type": "text/event-stream"}, + ) + + response: Final = await litellm.acompletion( + model=f"azure_ai/agents/{AGENT_ID}", + messages=[{"role": "user", "content": "Say it with separators"}], + api_base=API_BASE, + api_key="azure-ad-token", + stream=True, + ) + chunks: Final = [(chunk.choices[0].delta.content, chunk.choices[0].finish_reason) async for chunk in response] + + assert chunks == [(separated_text, None), (None, "stop")] diff --git a/tests/unit/llms/azure_ai/agents/test_transformation.py b/tests/unit/llms/azure_ai/agents/test_transformation.py index 10676438858..413908c8313 100644 --- a/tests/unit/llms/azure_ai/agents/test_transformation.py +++ b/tests/unit/llms/azure_ai/agents/test_transformation.py @@ -1,7 +1,7 @@ import json from typing import Final -from unittest.mock import MagicMock +import httpx import pytest from litellm.llms.azure_ai.agents.transformation import AzureAIAgentsConfig @@ -428,15 +428,10 @@ async def test_azure_ai_agents_streaming_accumulates_annotations_from_multiple_t "data: [DONE]", ] - async def mock_aiter_lines(): - for line in sse_lines: - yield line - - mock_response = MagicMock() - mock_response.aiter_lines = MagicMock(return_value=mock_aiter_lines()) + sse_response = httpx.Response(200, stream=httpx.ByteStream("\n".join(sse_lines).encode())) chunks = [] - async for chunk in handler._process_sse_stream(mock_response, "azure_ai/agents/asst_123"): + async for chunk in handler._process_sse_stream(sse_response, "azure_ai/agents/asst_123"): chunks.append(chunk) final_chunk = chunks[-1] @@ -498,15 +493,10 @@ async def test_azure_ai_agents_streaming_annotations_from_completed_message(): "data: [DONE]", ] - async def mock_aiter_lines(): - for line in sse_lines: - yield line - - mock_response = MagicMock() - mock_response.aiter_lines = MagicMock(return_value=mock_aiter_lines()) + sse_response = httpx.Response(200, stream=httpx.ByteStream("\n".join(sse_lines).encode())) chunks = [] - async for chunk in handler._process_sse_stream(mock_response, "azure_ai/agents/asst_123"): + async for chunk in handler._process_sse_stream(sse_response, "azure_ai/agents/asst_123"): chunks.append(chunk) assert len(chunks) >= 1 diff --git a/tests/unit/llms/base_llm/guardrail_translation/test_utils.py b/tests/unit/llms/base_llm/guardrail_translation/test_utils.py index 65208a2bd3b..acd50616f90 100644 --- a/tests/unit/llms/base_llm/guardrail_translation/test_utils.py +++ b/tests/unit/llms/base_llm/guardrail_translation/test_utils.py @@ -1,9 +1,11 @@ +import json from typing import Final import pytest import litellm from litellm.llms.base_llm.guardrail_translation.utils import ( + blocked_response_usage, effective_skip_system_message_for_guardrail, effective_skip_tool_message_for_guardrail, ) @@ -32,3 +34,18 @@ def test_a_per_guardrail_skip_flag_wins_over_the_global_setting( assert effective_skip_system_message_for_guardrail(guardrail) is expected assert effective_skip_tool_message_for_guardrail(guardrail) is expected + + +def test_blocked_response_usage_reads_a_message_delta_whose_json_holds_unicode_line_separators() -> None: + message_start: Final = {"type": "message_start", "message": {"usage": {"input_tokens": 11, "output_tokens": 1}}} + message_delta: Final = { + "type": "message_delta", + "delta": {"stop_reason": "stop_sequence", "stop_sequence": "END\u2028OF\u2029TURN"}, + "usage": {"output_tokens": 7}, + } + stream: Final = "".join( + f"event: {event['type']}\ndata: {json.dumps(event, ensure_ascii=False)}\n\n" + for event in (message_start, message_delta) + ).encode() + + assert blocked_response_usage([stream]) == {"input_tokens": 11, "output_tokens": 7} diff --git a/tests/unit/llms/bedrock/files/test_bedrock_files_transformation.py b/tests/unit/llms/bedrock/files/test_bedrock_files_transformation.py index 7e05ef500f5..9d9e39df944 100644 --- a/tests/unit/llms/bedrock/files/test_bedrock_files_transformation.py +++ b/tests/unit/llms/bedrock/files/test_bedrock_files_transformation.py @@ -4458,6 +4458,20 @@ class TestMalformedBatchJsonl: _VALID_BATCH_RECORD: Final = json.loads(_VALID_BATCH_LINE) +def test_file_content_conversion_keeps_a_line_whose_json_holds_unicode_line_separators() -> None: + text: Final = "Hello\u2028from\u2029batch" + body: Final = {**_VALID_BATCH_RECORD["body"], "messages": [{"role": "user", "content": text}]} + record: Final = {**_VALID_BATCH_RECORD, "body": body} + file_content: Final = f"{json.dumps(record, ensure_ascii=False)}\n".encode() + + bedrock_jsonl, _ = BedrockJsonlFilesTransformation().transform_openai_file_content_to_bedrock_file_content( + file_content + ) + + (bedrock_record,) = (json.loads(line) for line in bedrock_jsonl.split("\n")) + assert bedrock_record["modelInput"]["messages"] == [{"role": "user", "content": [{"text": text}]}] + + def _batch_line(body: object) -> str: return json.dumps({**_VALID_BATCH_RECORD, "body": body}) diff --git a/tests/unit/llms/bedrock/passthrough/guardrail_translation/test_handler.py b/tests/unit/llms/bedrock/passthrough/guardrail_translation/test_handler.py index 5aaee1f0ee4..aec1672b335 100644 --- a/tests/unit/llms/bedrock/passthrough/guardrail_translation/test_handler.py +++ b/tests/unit/llms/bedrock/passthrough/guardrail_translation/test_handler.py @@ -18,7 +18,7 @@ import types from collections.abc import Callable, Mapping, Sequence from datetime import datetime from itertools import chain -from typing import ClassVar, Literal, TypeAlias +from typing import ClassVar, Final, Literal, TypeAlias from unittest.mock import AsyncMock, MagicMock, patch import httpx @@ -60,6 +60,7 @@ from litellm.llms.bedrock.passthrough.guardrail_translation.handler import ( BedrockPassthroughGuardrailHandler, _extract_converse_texts, _is_converse_endpoint, + _sse_events, _write_back_texts, ) @@ -3840,3 +3841,14 @@ class TestInvokeStreamGuardrailRelay: assert received_key is not None and received_key.texts == ("In Paris",) assert rewritten_key == received_key + + +def test_sse_events_keeps_a_data_line_whose_json_holds_unicode_line_separators() -> None: + event: Final = { + "type": "content_block_delta", + "index": 0, + "delta": {"type": "text_delta", "text": "Card\u2028on\u2029file"}, + } + sse: Final = f"event: content_block_delta\ndata: {json.dumps(event, ensure_ascii=False)}\n\n".encode() + + assert _sse_events(sse) == (event,) diff --git a/tests/unit/llms/bedrock/search/test_agentcore_search_transformation.py b/tests/unit/llms/bedrock/search/test_agentcore_search_transformation.py index 664dbf87fe9..01cca1fc270 100644 --- a/tests/unit/llms/bedrock/search/test_agentcore_search_transformation.py +++ b/tests/unit/llms/bedrock/search/test_agentcore_search_transformation.py @@ -8,6 +8,7 @@ the sharded CI (coverage collection runs against this tree). import json import os +from typing import Final import httpx import pytest @@ -682,3 +683,17 @@ def test_transform_search_response_ignores_text_blocks_without_result_objects(bl body = {"result": {"content": [{"type": "text", "text": block_text}]}} assert _transform_text(json.dumps(body)).results == [] + + +def test_transform_search_response_keeps_a_result_whose_sse_json_holds_unicode_line_separators() -> None: + snippet: Final = "Snippet\u2028for\u2029result" + results: Final = [{"title": "Result", "url": "https://example.com/1", "text": snippet}] + body: Final = { + "jsonrpc": "2.0", + "id": 1, + "result": {"content": [{"type": "text", "text": json.dumps(results, ensure_ascii=False)}]}, + } + + response: Final = _transform_text(f"event: message\ndata: {json.dumps(body, ensure_ascii=False)}\n\n") + + assert _as_tuples(response) == [("Result", "https://example.com/1", snippet, None, None)] diff --git a/tests/unit/llms/chatgpt/responses/test_chatgpt_responses_transformation.py b/tests/unit/llms/chatgpt/responses/test_chatgpt_responses_transformation.py index a6f7d613f28..635363e77ea 100644 --- a/tests/unit/llms/chatgpt/responses/test_chatgpt_responses_transformation.py +++ b/tests/unit/llms/chatgpt/responses/test_chatgpt_responses_transformation.py @@ -587,6 +587,37 @@ class TestChatGPTResponsesMultiAccount: assert headers_default["ChatGPT-Account-Id"] == "acct-default" +def test_chatgpt_non_stream_sse_keeps_an_output_item_whose_text_holds_unicode_line_separators() -> None: + text: Final = "Hello\u2028from\u2029stream" + output_item_done: Final = { + "type": "response.output_item.done", + "output_index": 0, + "item": {"type": "message", "role": "assistant", "content": [{"type": "output_text", "text": text}]}, + } + completed: Final = { + "type": "response.completed", + "response": { + "id": "resp_test", + "object": "response", + "created_at": 1700000000, + "status": "completed", + "model": "gpt-6.1-sol", + "output": [], + }, + } + sse_body: Final = "".join( + f"data: {json.dumps(event, ensure_ascii=False)}\n" for event in (output_item_done, completed) + ) + + parsed: Final = ChatGPTResponsesAPIConfig().transform_response_api_response( + model="chatgpt/gpt-6.1-sol", + raw_response=httpx.Response(200, headers={"content-type": "text/event-stream"}, text=sse_body), + logging_obj=MagicMock(), + ) + + assert parsed.output_text == text + + @pytest.mark.parametrize("call", ["responses", "completion"]) def test_prompt_cache_key_keeps_one_session_header_across_requests( call: str, tmp_path: Path, monkeypatch: pytest.MonkeyPatch diff --git a/tests/unit/llms/cloudflare/test_cloudflare_transformation.py b/tests/unit/llms/cloudflare/test_cloudflare_transformation.py index 3a45c83556c..ccbfc02f0e6 100644 --- a/tests/unit/llms/cloudflare/test_cloudflare_transformation.py +++ b/tests/unit/llms/cloudflare/test_cloudflare_transformation.py @@ -1,6 +1,6 @@ import asyncio import json -from collections.abc import Callable, Iterator +from collections.abc import AsyncIterator, Callable, Iterator from typing import Final from unittest.mock import MagicMock @@ -380,6 +380,35 @@ def test_completion_cloudflare_stream(sync_mode: bool, _cloudflare_httpx_transpo assert "language" in content.lower() +@pytest.mark.asyncio +@pytest.mark.parametrize("sync_mode", [True, False], ids=["sync", "async"]) +async def test_cloudflare_stream_keeps_unicode_line_separators_inside_content( + sync_mode: bool, _cloudflare_httpx_transport: None +) -> None: + streamed_text: Final = ("alpha\u2028", "beta\u2029", "gamma") + base: Final = {"id": "chatcmpl-cf", "object": "chat.completion.chunk", "created": 1, "model": "@cf/glm"} + body: Final = "".join( + f"data: {json.dumps({**base, 'choices': [{'index': 0, 'delta': {'content': text}}]}, ensure_ascii=False)}\n\n" + for text in streamed_text + ) + with respx.mock(assert_all_called=True) as api: + api.post(f"{FAKE_API_BASE}/chat/completions").respond( + 200, headers={"content-type": "text/event-stream"}, content=(body + "data: [DONE]\n\n").encode() + ) + request: Final = { + "model": "cloudflare/@cf/glm", + "messages": [{"role": "user", "content": "repeat"}], + "stream": True, + "api_base": FAKE_API_BASE, + "api_key": FAKE_API_KEY, + } + chunks: Final = ( + tuple(completion(**request)) if sync_mode else tuple([c async for c in await acompletion(**request)]) + ) + + assert "".join(c.choices[0].delta.content or "" for c in chunks) == "".join(streamed_text) + + def _make_mock_response(json_data: Dict[str, Any]) -> MagicMock: mock = MagicMock(spec=httpx.Response) mock.status_code = 200 @@ -397,13 +426,13 @@ def test_acompletion_cloudflare_stream(sync_mode, monkeypatch): if sync_mode: - def _iter_lines(): + def _iter_text() -> Iterator[str]: for chunk in raw_chunks: - yield f"data: {chunk}" - yield "data: [DONE]" + yield f"data: {chunk}\n\n" + yield "data: [DONE]\n\n" mock_resp = MagicMock() - mock_resp.iter_lines.return_value = _iter_lines() + mock_resp.iter_text.return_value = _iter_text() mock_resp.status_code = 200 mock_resp.headers = {"content-type": "text/event-stream"} @@ -420,13 +449,13 @@ def test_acompletion_cloudflare_stream(sync_mode, monkeypatch): mock_post.assert_called_once() else: - async def _aiter_lines(): + async def _aiter_text() -> AsyncIterator[str]: for chunk in raw_chunks: - yield f"data: {chunk}" - yield "data: [DONE]" + yield f"data: {chunk}\n\n" + yield "data: [DONE]\n\n" mock_resp = MagicMock() - mock_resp.aiter_lines.return_value = _aiter_lines() + mock_resp.aiter_text.return_value = _aiter_text() mock_resp.status_code = 200 mock_resp.headers = {"content-type": "text/event-stream"} diff --git a/tests/unit/llms/codestral/__init__.py b/tests/unit/llms/codestral/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/unit/llms/codestral/completion/__init__.py b/tests/unit/llms/codestral/completion/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/unit/llms/codestral/completion/test_codestral_completion_handler.py b/tests/unit/llms/codestral/completion/test_codestral_completion_handler.py new file mode 100644 index 00000000000..9ce50d8ceca --- /dev/null +++ b/tests/unit/llms/codestral/completion/test_codestral_completion_handler.py @@ -0,0 +1,50 @@ +import json +from typing import Final + +import httpx +import pytest +import respx + +import litellm + +_LINE_SEPARATED_TEXT: Final = "before\u2028between\u2029after" + + +def _fim_stream_body(text: str) -> bytes: + chunks: Final = ( + {"id": "c1", "object": "chat.completion.chunk", "created": 1, "model": "codestral-latest", + "choices": [{"index": 0, "delta": {"role": "assistant", "content": text}, "finish_reason": None}]}, + {"id": "c1", "object": "chat.completion.chunk", "created": 1, "model": "codestral-latest", + "choices": [{"index": 0, "delta": {"content": ""}, "finish_reason": "stop"}]}, + ) + events: Final = "".join(f"data: {json.dumps(chunk, ensure_ascii=False)}\n\n" for chunk in chunks) + return f"{events}data: [DONE]\n\n".encode() + + +async def _streamed_texts(sync_stream: bool) -> list[str | None]: + request: Final = { + "model": "text-completion-codestral/codestral-latest", + "prompt": "def f(", + "stream": True, + "api_key": "codestral-key", + "max_retries": 0, + } + if sync_stream: + return [chunk.choices[0].text for chunk in litellm.text_completion(**request)] + return [chunk.choices[0].text async for chunk in await litellm.atext_completion(**request)] + + +@pytest.mark.asyncio +@pytest.mark.parametrize("sync_stream", [True, False]) +async def test_fim_stream_keeps_unicode_line_separators_inside_a_delta( + sync_stream: bool, respx_mock: respx.MockRouter, monkeypatch: pytest.MonkeyPatch +) -> None: + monkeypatch.setattr(litellm, "disable_aiohttp_transport", True) + route: Final = respx_mock.post("https://codestral.mistral.ai/v1/fim/completions").mock( + return_value=httpx.Response(200, content=_fim_stream_body(_LINE_SEPARATED_TEXT)) + ) + + texts: Final = await _streamed_texts(sync_stream) + + assert texts == [_LINE_SEPARATED_TEXT, None] + assert json.loads(route.calls.last.request.content)["stream"] is True diff --git a/tests/unit/llms/custom_httpx/test_aiohttp_handler.py b/tests/unit/llms/custom_httpx/test_aiohttp_handler.py index 83c38b65beb..bb66a70da84 100644 --- a/tests/unit/llms/custom_httpx/test_aiohttp_handler.py +++ b/tests/unit/llms/custom_httpx/test_aiohttp_handler.py @@ -1,11 +1,17 @@ +import json +from typing import Final from unittest.mock import AsyncMock, Mock, patch import aiohttp +import httpx import pytest import litellm +from litellm.llms.aiohttp_openai.chat.transformation import AiohttpOpenAIChatConfig from litellm.llms.custom_httpx.aiohttp_handler import BaseLLMAIOHTTPHandler from litellm.llms.custom_httpx.aiohttp_transport import LiteLLMAiohttpTransport +from litellm.llms.custom_httpx.http_handler import HTTPHandler +from litellm.types.utils import ModelResponseStream class TestBaseLLMAIOHTTPHandler: @@ -445,3 +451,34 @@ class TestBaseLLMAIOHTTPHandler: # Should use transport, not connector mock_transport.get_valid_client_session.assert_called_once() assert result is mock_session_from_transport + + +def test_make_sync_call_stream_keeps_unicode_line_separators_inside_a_delta() -> None: + text: Final = "before\u2028between\u2029after" + chunk: Final = { + "id": "c1", + "object": "chat.completion.chunk", + "created": 1, + "model": "gpt-5.5", + "choices": [{"index": 0, "delta": {"content": text}, "finish_reason": None}], + } + body: Final = f"data: {json.dumps(chunk, ensure_ascii=False)}\n\ndata: [DONE]\n\n".encode() + transport: Final = httpx.MockTransport(lambda request: httpx.Response(200, stream=httpx.ByteStream(body))) + + completion_stream, _ = BaseLLMAIOHTTPHandler().make_sync_call( + provider_config=AiohttpOpenAIChatConfig(), + api_base="https://example.com/v1/chat/completions", + headers={}, + data={"model": "gpt-5.5", "messages": []}, + model="gpt-5.5", + messages=[], + logging_obj=Mock(), + litellm_params={}, + timeout=5.0, + client=HTTPHandler(client=httpx.Client(transport=transport)), + ) + + contents: Final = [ + chunk.choices[0].delta.content for chunk in completion_stream if isinstance(chunk, ModelResponseStream) + ] + assert contents == [text] diff --git a/tests/unit/llms/custom_httpx/test_sse_lines.py b/tests/unit/llms/custom_httpx/test_sse_lines.py new file mode 100644 index 00000000000..9bb03f6da98 --- /dev/null +++ b/tests/unit/llms/custom_httpx/test_sse_lines.py @@ -0,0 +1,64 @@ +from collections.abc import AsyncIterator, Sequence +from typing import Final + +import pytest + +from litellm.llms.custom_httpx.sse_lines import ( + aiter_sse_lines, + iter_sse_lines, + split_sse_lines, + split_sse_lines_keepends, +) + +CASES: Final = ( + pytest.param(("data: a\u2028b\r\n\r\n",), ("data: a\u2028b", ""), id="u2028-stays-inside-the-line"), + pytest.param( + ("data: \u0085\u2029\x0b\x0c\x1c\x1d\x1e\n",), + ("data: \u0085\u2029\x0b\x0c\x1c\x1d\x1e",), + id="no-splitlines-only-separator-breaks-a-line", + ), + pytest.param(("a\r\nb\nc\rd\n",), ("a", "b", "c", "d"), id="crlf-lf-and-cr-each-end-a-line"), + pytest.param(("a\r", "\nb\n"), ("a", "b"), id="crlf-split-across-chunks-is-one-break"), + pytest.param(("a\r", "\n\r\n"), ("a", ""), id="crlf-split-then-blank-line"), + pytest.param(("a\r", "", "\nb\n"), ("a", "b"), id="empty-chunk-between-cr-and-lf"), + pytest.param(("a\r", "\r"), ("a", ""), id="cr-then-cr-is-two-breaks"), + pytest.param(("a\r\r\nb\n",), ("a", "", "b"), id="cr-then-crlf-in-one-chunk-is-two-breaks"), + pytest.param(("a\r\nb\r\n\r\n",), ("a", "b", ""), id="crlf-only-chunk"), + pytest.param(("da", "ta: x", "y\n"), ("data: xy",), id="line-spanning-chunks-is-joined"), + pytest.param(("data: x",), ("data: x",), id="unterminated-last-line-is-kept"), + pytest.param(("",), (), id="empty-body"), +) + + +async def _async_chunks(chunks: Sequence[str]) -> AsyncIterator[str]: + for chunk in chunks: + yield chunk + + +@pytest.mark.parametrize(("chunks", "expected"), CASES) +def test_iter_sse_lines_splits_only_on_cr_lf_and_crlf(chunks: Sequence[str], expected: Sequence[str]) -> None: + assert tuple(iter_sse_lines(chunks)) == tuple(expected) + + +@pytest.mark.parametrize(("chunks", "expected"), CASES) +async def test_aiter_sse_lines_splits_only_on_cr_lf_and_crlf(chunks: Sequence[str], expected: Sequence[str]) -> None: + assert tuple([line async for line in aiter_sse_lines(_async_chunks(chunks))]) == tuple(expected) + + +@pytest.mark.parametrize(("chunks", "expected"), CASES) +def test_split_sse_lines_splits_a_whole_body_only_on_cr_lf_and_crlf( + chunks: Sequence[str], expected: Sequence[str] +) -> None: + assert split_sse_lines("".join(chunks)) == tuple(expected) + + +@pytest.mark.parametrize( + ("text", "expected"), + ( + pytest.param("a\r\nb\rc\nd", ("a\r\n", "b\r", "c\n", "d"), id="each-break-stays-on-its-line"), + pytest.param("data: a\u2028b\u2029c\n\n", ("data: a\u2028b\u2029c\n", "\n"), id="u2028-and-u2029-stay-inside"), + pytest.param("", (), id="empty-body"), + ), +) +def test_split_sse_lines_keepends_splits_only_on_cr_lf_and_crlf(text: str, expected: Sequence[str]) -> None: + assert split_sse_lines_keepends(text) == tuple(expected) diff --git a/tests/unit/llms/databricks/chat/test_databricks_chat_transformation.py b/tests/unit/llms/databricks/chat/test_databricks_chat_transformation.py index b3a224209e9..eb2e1d8e347 100644 --- a/tests/unit/llms/databricks/chat/test_databricks_chat_transformation.py +++ b/tests/unit/llms/databricks/chat/test_databricks_chat_transformation.py @@ -1,6 +1,6 @@ import copy import json -from collections.abc import Iterator +from collections.abc import AsyncIterator, Iterator from typing import Final from unittest.mock import MagicMock, patch @@ -2160,13 +2160,13 @@ def mock_chat_response_nonanthropic_prompt_caching() -> Dict[str, Any]: def mock_http_handler_chat_streaming_response() -> MagicMock: mock_stream_chunks = mock_chat_streaming_response_chunks() - def mock_iter_lines(): + def mock_iter_text() -> Iterator[str]: for chunk in mock_stream_chunks: for line in chunk.splitlines(): - yield line + yield f"{line}\n" mock_response = MagicMock() - mock_response.iter_lines.side_effect = mock_iter_lines + mock_response.iter_text.side_effect = mock_iter_text mock_response.status_code = 200 return mock_response @@ -2175,13 +2175,13 @@ def mock_http_handler_chat_streaming_response() -> MagicMock: def mock_http_handler_chat_async_streaming_response() -> MagicMock: mock_stream_chunks = mock_chat_streaming_response_chunks() - async def mock_iter_lines(): + async def mock_aiter_text() -> AsyncIterator[str]: for chunk in mock_stream_chunks: for line in chunk.splitlines(): - yield line + yield f"{line}\n" mock_response = MagicMock() - mock_response.aiter_lines.return_value = mock_iter_lines() + mock_response.aiter_text.return_value = mock_aiter_text() mock_response.status_code = 200 return mock_response diff --git a/tests/unit/llms/fireworks_ai/chat/test_fireworks_ai_chat_transformation.py b/tests/unit/llms/fireworks_ai/chat/test_fireworks_ai_chat_transformation.py index 04b791cf289..61e5545b8bf 100644 --- a/tests/unit/llms/fireworks_ai/chat/test_fireworks_ai_chat_transformation.py +++ b/tests/unit/llms/fireworks_ai/chat/test_fireworks_ai_chat_transformation.py @@ -1329,7 +1329,7 @@ def test_streaming_surfaces_fireworks_response_fields(): raw_response = MagicMock() raw_response.status_code = 200 raw_response.headers = {} - raw_response.iter_lines = lambda: iter(sse_lines) + raw_response.iter_text = lambda: iter(f"{line}\n" for line in sse_lines) client = HTTPHandler() with patch.object(client, "post", return_value=raw_response): @@ -1788,7 +1788,7 @@ def test_streaming_preserves_selected_model_for_private_accounting(): raw_response = MagicMock() raw_response.status_code = 200 raw_response.headers = {} - raw_response.iter_lines = lambda: iter(sse_lines) + raw_response.iter_text = lambda: iter(f"{line}\n" for line in sse_lines) client = HTTPHandler() with patch.object(client, "post", return_value=raw_response): @@ -2275,7 +2275,7 @@ def test_completion_forwards_litellm_user_id_as_user(call_kwargs: dict[str, obje def test_completion_forwards_litellm_user_id_when_streaming() -> None: client: Final = _fireworks_chat_client() - client.post.return_value.iter_lines = lambda: iter(()) + client.post.return_value.iter_text = lambda: iter(()) litellm.completion( model="fireworks_ai/accounts/fireworks/models/kimi-k3", messages=[{"role": "user", "content": "hi"}], diff --git a/tests/unit/llms/gemini/google_genai/guardrail_translation/test_google_genai_guardrail_handler.py b/tests/unit/llms/gemini/google_genai/guardrail_translation/test_google_genai_guardrail_handler.py index 4119ce99423..53ddb1135b4 100644 --- a/tests/unit/llms/gemini/google_genai/guardrail_translation/test_google_genai_guardrail_handler.py +++ b/tests/unit/llms/gemini/google_genai/guardrail_translation/test_google_genai_guardrail_handler.py @@ -4,6 +4,7 @@ Tests for the Google GenAI generateContent guardrail translation handler. import json from types import SimpleNamespace +from typing import Final from unittest.mock import AsyncMock, MagicMock import pytest @@ -190,6 +191,21 @@ async def test_streaming_raw_sse_chunks_accumulate_text_across_split_frames(): assert guardrail.apply_guardrail.call_args.kwargs["inputs"]["texts"] == ["harmful answer"] +@pytest.mark.asyncio +async def test_streaming_raw_sse_text_holding_unicode_line_separators_reaches_the_guardrail_whole() -> None: + guardrail: Final = _mock_guardrail(["clean"]) + text: Final = "harmful\u2028answer\u2029here" + payload: Final = {"candidates": [{"content": {"parts": [{"text": text}]}}]} + chunk: Final = f"data: {json.dumps(payload, ensure_ascii=False)}\r\n\r\n" + + await GoogleGenAIGenerateContentHandler().process_output_streaming_response( + responses_so_far=[chunk], + guardrail_to_apply=guardrail, + ) + + assert guardrail.apply_guardrail.call_args.kwargs["inputs"]["texts"] == [text] + + @pytest.mark.asyncio async def test_streaming_blocking_guardrail_exception_propagates(): handler = GoogleGenAIGenerateContentHandler() diff --git a/tests/unit/llms/langgraph/chat/test_langgraph_sse_iterator.py b/tests/unit/llms/langgraph/chat/test_langgraph_sse_iterator.py new file mode 100644 index 00000000000..06485362162 --- /dev/null +++ b/tests/unit/llms/langgraph/chat/test_langgraph_sse_iterator.py @@ -0,0 +1,32 @@ +import json +from typing import Final + +import httpx +import pytest + +from litellm.llms.langgraph.chat.sse_iterator import LangGraphSSEStreamIterator + +_LINE_SEPARATED_TEXT: Final = "before\u2028between\u2029after" + + +def _messages_event_stream(text: str) -> httpx.Response: + event: Final = ["messages", [[{"type": "AIMessageChunk", "content": text}, {}]]] + return httpx.Response(200, stream=httpx.ByteStream(f"data: {json.dumps(event, ensure_ascii=False)}\n\n".encode())) + + +async def _streamed_contents(iterator: LangGraphSSEStreamIterator, sync_stream: bool) -> list[str | None]: + if sync_stream: + return [chunk.choices[0].delta.content for chunk in iterator] + return [chunk.choices[0].delta.content async for chunk in iterator] + + +@pytest.mark.asyncio +@pytest.mark.parametrize("sync_stream", [True, False]) +async def test_stream_keeps_unicode_line_separators_inside_a_message_chunk(sync_stream: bool) -> None: + iterator: Final = LangGraphSSEStreamIterator( + response=_messages_event_stream(_LINE_SEPARATED_TEXT), model="langgraph/agent" + ) + + contents: Final = await _streamed_contents(iterator, sync_stream) + + assert contents == [_LINE_SEPARATED_TEXT, None] diff --git a/tests/unit/llms/oobabooga/chat/test_oobabooga.py b/tests/unit/llms/oobabooga/chat/test_oobabooga.py index 395a4fb5715..332b6913890 100644 --- a/tests/unit/llms/oobabooga/chat/test_oobabooga.py +++ b/tests/unit/llms/oobabooga/chat/test_oobabooga.py @@ -1,7 +1,12 @@ +import json +from typing import Final from unittest.mock import MagicMock, patch +import httpx +import respx import litellm +from litellm.llms.oobabooga.chat import oobabooga MOCK_COMPLETION_RESPONSE = { "choices": [{"message": {"role": "assistant", "content": "hi there"}}], @@ -50,3 +55,27 @@ def test_url_valued_model_still_targets_that_url(): mock_post.assert_called_once() called_url = mock_post.call_args[0][0] assert called_url == "https://sdk-user.example/v1/chat/completions" + + +def test_stream_keeps_unicode_line_separators_inside_a_data_line(respx_mock: respx.MockRouter) -> None: + text: Final = "before\u2028between\u2029after" + chunk: Final = {"choices": [{"index": 0, "delta": {"content": text}, "finish_reason": None}]} + data_line: Final = f"data: {json.dumps(chunk, ensure_ascii=False)}" + respx_mock.post("https://ooba.example/v1/chat/completions").mock( + return_value=httpx.Response(200, content=f"{data_line}\n\ndata: [DONE]\n\n".encode()) + ) + + lines: Final = oobabooga.completion( + model="my-model", + messages=[{"role": "user", "content": "hello"}], + api_base="https://ooba.example", + model_response=litellm.ModelResponse(), + print_verbose=print, + encoding=None, + api_key=None, + logging_obj=MagicMock(), + optional_params={"stream": True}, + litellm_params={}, + ) + + assert [line for line in lines if line] == [data_line, "data: [DONE]"] diff --git a/tests/unit/llms/openai_like/chat/test_openai_like_chat_handler.py b/tests/unit/llms/openai_like/chat/test_openai_like_chat_handler.py index 75b60b2dd6c..3df1efe311b 100644 --- a/tests/unit/llms/openai_like/chat/test_openai_like_chat_handler.py +++ b/tests/unit/llms/openai_like/chat/test_openai_like_chat_handler.py @@ -4,9 +4,10 @@ from typing import Final from unittest.mock import Mock import httpx +import pytest -from litellm.llms.custom_httpx.http_handler import HTTPHandler -from litellm.llms.openai_like.chat.handler import make_sync_call +from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler, HTTPHandler +from litellm.llms.openai_like.chat.handler import make_call, make_sync_call _FAKE_STREAMED_REPLY: Final = { "id": "chatcmpl-fake-stream", @@ -73,3 +74,35 @@ def test_fake_stream_keeps_reasoning_every_tool_call_and_token_detail(): assert [call.function.name for call in delta.tool_calls] == ["get_weather", "get_time"] assert chunks[0].usage.prompt_tokens_details.cached_tokens == 64 assert chunks[0].usage.completion_tokens_details.reasoning_tokens == 1920 + + +_LINE_SEPARATED_TEXT: Final = "before\u2028between\u2029after" + + +def _streamed_reply(request: httpx.Request) -> httpx.Response: + chunk: Final = {"id": "chatcmpl-separators", "choices": [{"index": 0, "delta": {"content": _LINE_SEPARATED_TEXT}}]} + body: Final = f"data: {json.dumps(chunk, ensure_ascii=False)}\n\ndata: [DONE]\n\n" + return httpx.Response(200, stream=httpx.ByteStream(body.encode())) + + +async def _streamed_chunks(sync_stream: bool) -> list[dict[str, object]]: + call_args: Final = { + "api_base": "https://openai-like.example/v1/chat/completions", + "headers": {}, + "data": json.dumps({"model": "openai/gpt-oss-120b", "messages": [], "stream": True}), + "model": "openai/gpt-oss-120b", + "messages": [], + "logging_obj": Mock(), + } + transport: Final = httpx.MockTransport(_streamed_reply) + if sync_stream: + return list(make_sync_call(client=HTTPHandler(client=httpx.Client(transport=transport)), **call_args)) + return [chunk async for chunk in await make_call(client=AsyncHTTPHandler(transport=transport), **call_args)] + + +@pytest.mark.asyncio +@pytest.mark.parametrize("sync_stream", [True, False]) +async def test_stream_keeps_unicode_line_separators_inside_the_delta(sync_stream: bool) -> None: + chunks: Final = await _streamed_chunks(sync_stream) + + assert "".join(str(chunk["text"]) for chunk in chunks) == _LINE_SEPARATED_TEXT diff --git a/tests/unit/llms/snowflake/chat/test_snowflake_chat_transformation.py b/tests/unit/llms/snowflake/chat/test_snowflake_chat_transformation.py index f9d12b7eeea..36c319d9524 100644 --- a/tests/unit/llms/snowflake/chat/test_snowflake_chat_transformation.py +++ b/tests/unit/llms/snowflake/chat/test_snowflake_chat_transformation.py @@ -902,20 +902,14 @@ class TestSnowflakeChatCompletion: @pytest.mark.parametrize("sync_mode", [True, False]) def test_chat_completion_snowflake_stream(self, sync_mode): - raw_chunks = _streaming_chunks() + sse_body = "".join(f"data: {chunk}\n\n" for chunk in _streaming_chunks()) + "data: [DONE]\n\n" + mock_resp = httpx.Response( + 200, + headers={"content-type": "text/event-stream"}, + stream=httpx.ByteStream(sse_body.encode()), + ) if sync_mode: - - def _iter_lines(): - for chunk in raw_chunks: - yield f"data: {chunk}" - yield "data: [DONE]" - - mock_resp = MagicMock() - mock_resp.iter_lines.return_value = _iter_lines() - mock_resp.status_code = 200 - mock_resp.headers = {"content-type": "text/event-stream"} - with patch.object(HTTPHandler, "post", return_value=mock_resp) as mock_post: response = completion( model="snowflake/mistral-7b", @@ -930,16 +924,6 @@ class TestSnowflakeChatCompletion: mock_post.assert_called_once() else: - async def _aiter_lines(): - for chunk in raw_chunks: - yield f"data: {chunk}" - yield "data: [DONE]" - - mock_resp = MagicMock() - mock_resp.aiter_lines.return_value = _aiter_lines() - mock_resp.status_code = 200 - mock_resp.headers = {"content-type": "text/event-stream"} - async def _run(): with patch.object( AsyncHTTPHandler, @@ -964,8 +948,8 @@ class TestSnowflakeChatCompletion: chunks_received = asyncio.run(_run()) - assert len(chunks_received) > 0 content = "".join(c.choices[0].delta.content for c in chunks_received if c.choices[0].delta.content) + assert content == "The sky is blue" async def test_snowflake_claude_async_completion_inlines_remote_images_off_the_event_loop(async_only_image_fetch): diff --git a/tests/unit/llms/test_predibase_transformation.py b/tests/unit/llms/test_predibase_transformation.py index a02840ed4ef..7afbd05b221 100644 --- a/tests/unit/llms/test_predibase_transformation.py +++ b/tests/unit/llms/test_predibase_transformation.py @@ -1,3 +1,4 @@ +import json from typing import Final from unittest.mock import AsyncMock, Mock @@ -639,3 +640,49 @@ def test_predibase_completion_passes_existing_config_to_async_completion(monkeyp assert result == "async-result" assert captured["async_kwargs"]["predibase_config"] is captured["config_instance"] + + +_LINE_SEPARATED_TEXT: Final = "before\u2028between\u2029after" + + +def _generate_stream_body(text: str) -> bytes: + events: Final = ( + {"token": {"text": text}}, + {"token": {"text": ""}, "generated_text": text, "details": {"finish_reason": "eos_token"}}, + ) + return "".join(f"data:{json.dumps(event, ensure_ascii=False)}\n\n" for event in events).encode() + + +async def _streamed_contents(sync_stream: bool) -> list[str | None]: + request: Final = { + "model": "predibase/llama-3-8b-instruct", + "messages": [{"role": "user", "content": "hi"}], + "stream": True, + "predibase_tenant_id": "tenant-1", + "api_key": "pb-key", + "max_retries": 0, + } + if sync_stream: + return [chunk.choices[0].delta.content for chunk in litellm.completion(**request)] + return [chunk.choices[0].delta.content async for chunk in await litellm.acompletion(**request)] + + +@pytest.mark.asyncio +@pytest.mark.parametrize("sync_stream", [True, False]) +async def test_predibase_stream_keeps_unicode_line_separators_inside_a_token( + sync_stream: bool, respx_mock: respx.MockRouter, monkeypatch: pytest.MonkeyPatch +) -> None: + monkeypatch.setattr(litellm, "disable_hf_tokenizer_download", True) + monkeypatch.setattr(litellm, "disable_aiohttp_transport", True) + monkeypatch.setattr( + litellm, + "custom_prompt_dict", + {"llama-3-8b-instruct": {"roles": {}, "initial_prompt_value": "", "final_prompt_value": ""}}, + ) + respx_mock.post( + "https://serving.app.predibase.com/tenant-1/deployments/v2/llms/llama-3-8b-instruct/generate_stream" + ).mock(return_value=httpx.Response(200, content=_generate_stream_body(_LINE_SEPARATED_TEXT))) + + contents: Final = await _streamed_contents(sync_stream) + + assert contents == [_LINE_SEPARATED_TEXT, None] diff --git a/tests/unit/llms/triton/test_triton.py b/tests/unit/llms/triton/test_triton.py index 7c0de603a3e..238c30816b2 100644 --- a/tests/unit/llms/triton/test_triton.py +++ b/tests/unit/llms/triton/test_triton.py @@ -1,5 +1,7 @@ import json import traceback +from collections.abc import Iterator +from typing import Final from unittest.mock import MagicMock, patch import httpx @@ -210,8 +212,8 @@ def test_completion_triton_generate_api(stream, monkeypatch, respx_mock): mock_response = MagicMock() if stream: - def mock_iter_lines(): - mock_output = "".join( + def mock_iter_text() -> Iterator[str]: + mock_output: Final = "".join( [ 'data: {"model_name":"ensemble","model_version":"1","sequence_end":false,"sequence_id":0,"sequence_start":false,"text_output":"' + t @@ -219,10 +221,9 @@ def test_completion_triton_generate_api(stream, monkeypatch, respx_mock): for t in ["I", " am", " an", " AI", " assistant"] ] ) - for out in mock_output.split("\n"): - yield out + yield mock_output - mock_response.iter_lines = mock_iter_lines + mock_response.iter_text = mock_iter_text else: def return_val(): diff --git a/tests/unit/llms/vertex_ai/agent_engine/test_transformation.py b/tests/unit/llms/vertex_ai/agent_engine/test_transformation.py index 06aaca44e8a..20333fd0465 100644 --- a/tests/unit/llms/vertex_ai/agent_engine/test_transformation.py +++ b/tests/unit/llms/vertex_ai/agent_engine/test_transformation.py @@ -14,7 +14,7 @@ import pytest from litellm.litellm_core_utils.litellm_logging import Logging -from litellm.llms.custom_httpx.http_handler import get_async_httpx_client +from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler, HTTPHandler, get_async_httpx_client from litellm.llms.vertex_ai.agent_engine.sse_iterator import ( VertexAgentEngineResponseIterator, ) @@ -170,3 +170,48 @@ async def test_async_stream_wrapper_without_a_client_posts_through_the_cached_ve assert type(wrapper) is CustomStreamWrapper assert [str(request.url) for request in sent_requests] == [api_base] assert json.loads(sent_requests[0].content) == {"class_method": "stream_query"} + + +_LINE_SEPARATED_TEXT: Final = "before\u2028between\u2029after" + + +def _separated_text_reply(request: httpx.Request) -> httpx.Response: + chunk: Final = {"content": {"parts": [{"text": _LINE_SEPARATED_TEXT}], "role": "model"}, "finish_reason": "STOP"} + return httpx.Response(200, stream=httpx.ByteStream(f"data: {json.dumps(chunk, ensure_ascii=False)}\n\n".encode())) + + +async def _streamed_text(sync_stream: bool) -> str: + messages: Final = [{"role": "user", "content": "hi"}] + call_args: Final = { + "model": "agent_engine/123", + "custom_llm_provider": "vertex_ai", + "logging_obj": Logging( + model="agent_engine/123", + messages=messages, + stream=True, + call_type="completion" if sync_stream else "acompletion", + start_time=datetime(2026, 1, 1), + litellm_call_id="call-separators", + function_id="fn-separators", + ), + "api_base": "https://us-central1-aiplatform.googleapis.com/v1/reasoningEngines/123:streamQuery", + "headers": {}, + "data": {"class_method": "stream_query"}, + "messages": messages, + "litellm_params": {}, + } + transport: Final = httpx.MockTransport(_separated_text_reply) + if sync_stream: + sync_wrapper: Final = VertexAgentEngineConfig().get_sync_custom_stream_wrapper( + client=HTTPHandler(client=httpx.Client(transport=transport)), **call_args + ) + return "".join(chunk.choices[0].delta.content or "" for chunk in sync_wrapper) + async_wrapper: Final = await VertexAgentEngineConfig().get_async_custom_stream_wrapper( + client=AsyncHTTPHandler(transport=transport), **call_args + ) + return "".join([chunk.choices[0].delta.content or "" async for chunk in async_wrapper]) + + +@pytest.mark.parametrize("sync_stream", [True, False]) +async def test_stream_keeps_unicode_line_separators_inside_the_text(sync_stream: bool) -> None: + assert await _streamed_text(sync_stream) == _LINE_SEPARATED_TEXT diff --git a/tests/unit/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py b/tests/unit/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py index 3e1bca18f08..67c78b62151 100644 --- a/tests/unit/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py +++ b/tests/unit/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py @@ -3,8 +3,9 @@ import json import logging import re import threading -from collections.abc import Awaitable, Callable, Iterator, Mapping +from collections.abc import AsyncIterator, Awaitable, Callable, Iterator, Mapping from copy import deepcopy +from itertools import pairwise from typing import Final, List, Literal, cast, get_args from unittest.mock import MagicMock, Mock, patch @@ -34,7 +35,7 @@ from litellm.llms.vertex_ai.gemini.vertex_and_google_ai_studio_gemini import ( ) from litellm.types.llms.base import CachedTokensDetails from litellm.types.llms.vertex_ai import GeminiFinishReason, RequestBody, UsageMetadata -from litellm.types.utils import ChoiceLogprobs, ModelInfo, Usage +from litellm.types.utils import ChoiceLogprobs, ModelInfo, ModelResponseStream, Usage from litellm.utils import _invalidate_model_cost_lowercase_map, CustomStreamWrapper from litellm.litellm_core_utils.logging_worker import GLOBAL_LOGGING_WORKER from litellm.integrations.custom_logger import CustomLogger @@ -10125,7 +10126,7 @@ def test_vertex_ai_streaming_response_id(): mock_response = MagicMock() mock_response.status_code = 200 - mock_response.iter_lines = MagicMock(return_value=stream_response()) + mock_response.iter_text = MagicMock(return_value=stream_response()) return mock_response logging_obj = MagicMock() @@ -10690,6 +10691,63 @@ async def test_gemini_sends_a_boolean_for_a_gateway_default_written_as_text_or_n assert sent["includeThoughts"] is expected +class _BodyInChunks(httpx.SyncByteStream, httpx.AsyncByteStream): + def __init__(self, chunks: tuple[bytes, ...]) -> None: + self._chunks: Final = chunks + + def __iter__(self) -> Iterator[bytes]: + yield from self._chunks + + async def __aiter__(self) -> AsyncIterator[bytes]: + for chunk in self._chunks: + yield chunk + + +def _split_at(body: bytes, cuts: tuple[int, ...]) -> tuple[bytes, ...]: + return tuple(body[start:end] for start, end in pairwise((0, *sorted(cuts), len(body)))) + + +async def _gemini_stream(is_async: bool, chunks: tuple[bytes, ...]) -> list[ModelResponseStream]: + def handler(request: httpx.Request) -> httpx.Response: + return httpx.Response(200, stream=_BodyInChunks(chunks), headers={"content-type": "text/event-stream"}) + + kwargs: Final = { + "model": "gemini/gemini-3.8-flash", + "messages": [{"role": "user", "content": "hi"}], + "stream": True, + "stream_options": {"include_usage": True}, + "api_key": "fake-gemini-key", + } + if is_async: + stream: Final = await litellm.acompletion( + **kwargs, client=AsyncHTTPHandler(transport=httpx.MockTransport(handler)) + ) + return [chunk async for chunk in stream] + return list( + litellm.completion(**kwargs, client=HTTPHandler(client=httpx.Client(transport=httpx.MockTransport(handler)))) + ) + + +@pytest.mark.parametrize("is_async", [True, False]) +async def test_gemini_stream_keeps_text_and_usage_when_a_part_contains_a_unicode_line_separator(is_async: bool) -> None: + events: Final = ( + {"candidates": [{"content": {"role": "model", "parts": [{"text": "line one\u2028line two"}]}}]}, + { + "candidates": [{"content": {"role": "model", "parts": [{"text": " and three"}]}, "finishReason": "STOP"}], + "usageMetadata": {"promptTokenCount": 10, "candidatesTokenCount": 8, "totalTokenCount": 18}, + }, + ) + body: Final = "".join(f"data: {json.dumps(event, ensure_ascii=False)}\r\n\r\n" for event in events).encode() + cut_inside_separator_and_crlf: Final = (body.index(b"\xe2\x80\xa8") + 1, body.index(b"\r") + 1) + + received: Final = await _gemini_stream(is_async, _split_at(body, cut_inside_separator_and_crlf)) + + usage: Final = next(chunk.usage for chunk in reversed(received) if getattr(chunk, "usage", None)) + text: Final = "".join(c.choices[0].delta.content or "" for c in received if c.choices) + assert text == "line one\u2028line two and three" + assert (usage.prompt_tokens, usage.completion_tokens) == (10, 8) + + @pytest.mark.parametrize("chunk_indices", [[0, 1], [2, 1, 0]]) def test_grounding_support_preserves_every_web_source(chunk_indices: list[int]) -> None: metadata: Final = { diff --git a/tests/unit/llms/vertex_ai/test_http_status_201.py b/tests/unit/llms/vertex_ai/test_http_status_201.py index 34502d5cebe..0ed469acfa7 100644 --- a/tests/unit/llms/vertex_ai/test_http_status_201.py +++ b/tests/unit/llms/vertex_ai/test_http_status_201.py @@ -39,8 +39,8 @@ class TestVertexAIHTTPStatus201(unittest.TestCase): # Create a mock response with status code 201 mock_response = MagicMock() mock_response.status_code = 201 - mock_response.aiter_lines = MagicMock() - mock_response.aiter_lines.return_value = ["test response"] + mock_response.aiter_text = MagicMock() + mock_response.aiter_text.return_value = ["test response"] # Setup mock client mock_client = MagicMock() @@ -76,8 +76,8 @@ class TestVertexAIHTTPStatus201(unittest.TestCase): # Create a mock response with status code 201 mock_response = MagicMock() mock_response.status_code = 201 - mock_response.iter_lines = MagicMock() - mock_response.iter_lines.return_value = ["test response"] + mock_response.iter_text = MagicMock() + mock_response.iter_text.return_value = ["test response"] mock_post.return_value = mock_response # Call the make_sync_call function diff --git a/tests/unit/proxy/agent_endpoints/test_a2a_endpoints.py b/tests/unit/proxy/agent_endpoints/test_a2a_endpoints.py index a7d7ac5627c..7abe6ad7799 100644 --- a/tests/unit/proxy/agent_endpoints/test_a2a_endpoints.py +++ b/tests/unit/proxy/agent_endpoints/test_a2a_endpoints.py @@ -4,10 +4,11 @@ Mock tests for A2A endpoints. Tests that invoke_agent_a2a properly integrates with add_litellm_data_to_request. """ +import asyncio import json import socket import sys -from collections.abc import Awaitable, Callable, Mapping +from collections.abc import AsyncIterator, Awaitable, Callable, Mapping from contextlib import AbstractContextManager, ExitStack from dataclasses import dataclass from typing import Final @@ -24,6 +25,21 @@ from litellm.types.agents import AgentCaller AddLiteLLMData = Callable[..., Awaitable[dict[str, object]]] +class _SSELineStream(httpx.AsyncByteStream): + def __init__(self, lines: tuple[str, ...], first_line_delay: float) -> None: + self._lines = lines + self._first_line_delay = first_line_delay + + async def __aiter__(self) -> AsyncIterator[bytes]: + await asyncio.sleep(self._first_line_delay) + for line in self._lines: + yield f"{line}\n".encode() + + +def _sse_upstream_response(*lines: str, first_line_delay: float = 0.0) -> httpx.Response: + return httpx.Response(200, stream=_SSELineStream(lines, first_line_delay)) + + @dataclass(frozen=True, slots=True) class CapturedAgentCall: request_id: object @@ -941,13 +957,7 @@ async def test_task_methods_extract_litellm_params_before_forwarding(method: str mock_http_response.json.return_value = upstream_response mock_http_response.is_success = True - async def fake_aiter_lines(): - yield 'data: {"jsonrpc":"2.0","id":"req-1","result":{"taskId":"task-1"}}' - - mock_resp = AsyncMock() - mock_resp.is_success = True - mock_resp.aiter_lines = fake_aiter_lines - mock_resp.aclose = AsyncMock() + mock_resp = _sse_upstream_response('data: {"jsonrpc":"2.0","id":"req-1","result":{"taskId":"task-1"}}') mock_async_client = MagicMock() mock_async_client.build_request = MagicMock(return_value=MagicMock()) @@ -982,8 +992,8 @@ async def test_task_methods_extract_litellm_params_before_forwarding(method: str user_api_key_dict=user_api_key_dict, ) if method == "tasks/resubscribe": - async for _ in response.body_iterator: - pass + streamed = "".join([chunk async for chunk in response.body_iterator]) + assert "task-1" in streamed if method == "tasks/resubscribe": forwarded_body = mock_async_client.build_request.call_args.kwargs["json"] @@ -1007,14 +1017,7 @@ async def test_subscribe_to_task_returns_sse_stream(): 'data: {"jsonrpc":"2.0","id":"req-1","result":{"taskId":"task-1","status":{"state":"completed"}}}', ] - async def fake_aiter_lines(): - for line in sse_lines: - yield line - - mock_resp = AsyncMock() - mock_resp.is_success = True - mock_resp.aiter_lines = fake_aiter_lines - mock_resp.aclose = AsyncMock() + mock_resp = _sse_upstream_response(*sse_lines) mock_async_client = MagicMock() mock_async_client.build_request = MagicMock(return_value=MagicMock()) @@ -1064,13 +1067,9 @@ async def test_subscribe_to_task_calls_pre_call_hook(): mock_request = _make_request_mock("tasks/resubscribe", {"id": "task-1"}) user_api_key_dict = UserAPIKeyAuth(api_key="sk-test", user_id="u1", team_id="t1") - async def fake_aiter_lines(): - yield 'data: {"jsonrpc":"2.0","id":"req-1","result":{"taskId":"task-1","status":{"state":"completed"}}}' - - mock_resp = AsyncMock() - mock_resp.is_success = True - mock_resp.aiter_lines = fake_aiter_lines - mock_resp.aclose = AsyncMock() + mock_resp = _sse_upstream_response( + 'data: {"jsonrpc":"2.0","id":"req-1","result":{"taskId":"task-1","status":{"state":"completed"}}}' + ) mock_async_client = MagicMock() mock_async_client.build_request = MagicMock(return_value=MagicMock()) @@ -1117,9 +1116,9 @@ async def test_subscribe_to_task_calls_pre_call_hook(): ) assert response.media_type == "text/event-stream" - async for _ in response.body_iterator: - pass + streamed = "".join([chunk async for chunk in response.body_iterator]) + assert "completed" in streamed mock_proxy_logging.pre_call_hook.assert_awaited_once() call_kwargs = mock_proxy_logging.pre_call_hook.await_args.kwargs assert call_kwargs.get("call_type") == "asend_message" @@ -1149,16 +1148,10 @@ async def test_subscribe_to_task_runs_post_call_streaming_guardrail(): mock_request = _make_request_mock("tasks/resubscribe", {"id": "task-1"}) user_api_key_dict = UserAPIKeyAuth(api_key="sk-test", user_id="u1", team_id="t1") - async def fake_aiter_lines(): - yield ( - 'data: {"jsonrpc":"2.0","id":"req-1","result":' - '{"kind":"message","parts":[{"kind":"text","text":"resubscribe-secret"}]}}' - ) - - mock_resp = AsyncMock() - mock_resp.is_success = True - mock_resp.aiter_lines = fake_aiter_lines - mock_resp.aclose = AsyncMock() + mock_resp = _sse_upstream_response( + 'data: {"jsonrpc":"2.0","id":"req-1","result":' + '{"kind":"message","parts":[{"kind":"text","text":"resubscribe-secret"}]}}' + ) mock_async_client = MagicMock() mock_async_client.build_request = MagicMock(return_value=MagicMock()) @@ -1600,14 +1593,7 @@ async def test_pascal_method_names_normalize_to_wire_format(pascal_method: str, mock_http_response.is_success = True mock_http_response.raise_for_status = MagicMock() - async def _empty_aiter_lines(): - return - yield # make it an async generator - - mock_sse_resp = AsyncMock() - mock_sse_resp.is_success = True - mock_sse_resp.aiter_lines = _empty_aiter_lines - mock_sse_resp.aclose = AsyncMock() + mock_sse_resp = _sse_upstream_response() mock_async_client = MagicMock() mock_async_client.build_request = MagicMock(return_value=MagicMock()) @@ -2349,13 +2335,7 @@ async def test_task_methods_forward_trace_header(method: str): mock_http_response.json.return_value = upstream_response mock_http_response.is_success = True - async def fake_aiter_lines(): - yield 'data: {"jsonrpc":"2.0","id":"req-1","result":{"taskId":"task-1"}}' - - mock_resp = AsyncMock() - mock_resp.is_success = True - mock_resp.aiter_lines = fake_aiter_lines - mock_resp.aclose = AsyncMock() + mock_resp = _sse_upstream_response('data: {"jsonrpc":"2.0","id":"req-1","result":{"taskId":"task-1"}}') mock_async_client = MagicMock() mock_async_client.build_request = MagicMock(return_value=MagicMock()) @@ -2390,8 +2370,8 @@ async def test_task_methods_forward_trace_header(method: str): user_api_key_dict=user_api_key_dict, ) if method == "tasks/resubscribe": - async for _ in response.body_iterator: - pass + streamed = "".join([chunk async for chunk in response.body_iterator]) + assert "task-1" in streamed if method == "tasks/resubscribe": forwarded_headers = mock_async_client.build_request.call_args.kwargs["headers"] @@ -2646,17 +2626,12 @@ def test_served_version_falls_back_to_header_when_unconfigured(): assert _served_version(_agent(None), _request_with_a2a_header(None)) == "0.3" -def _sse_agent_handler(lines): - mock_resp = AsyncMock() - mock_resp.is_success = True - mock_resp.aiter_lines = lines - mock_resp.aclose = AsyncMock() - - mock_async_client = MagicMock() +def _sse_agent_handler(upstream_response: httpx.Response) -> MagicMock: + mock_async_client: Final = MagicMock() mock_async_client.build_request = MagicMock(return_value=MagicMock()) - mock_async_client.send = AsyncMock(return_value=mock_resp) + mock_async_client.send = AsyncMock(return_value=upstream_response) - mock_handler = MagicMock() + mock_handler: Final = MagicMock() mock_handler.client = mock_async_client return mock_handler @@ -2685,13 +2660,13 @@ async def test_forward_jsonrpc_sse_pings_while_the_upstream_agent_is_still_silen monkeypatch.setattr(litellm, "sse_keepalive_ping_interval_seconds", 0.05) - async def _slow_lines(): - await asyncio.sleep(0.3) - yield 'data: {"jsonrpc": "2.0", "id": "req-1", "result": {"kind": "task"}}' + slow_upstream = _sse_upstream_response( + 'data: {"jsonrpc": "2.0", "id": "req-1", "result": {"kind": "task"}}', first_line_delay=0.3 + ) with patch( "litellm.llms.custom_httpx.http_handler.get_async_httpx_client", - return_value=_sse_agent_handler(_slow_lines), + return_value=_sse_agent_handler(slow_upstream), ): response = await _resubscribe_response() assert response.headers["x-accel-buffering"] == "no" @@ -2713,12 +2688,11 @@ async def test_forward_jsonrpc_sse_is_untouched_while_keepalives_are_unconfigure monkeypatch.setattr(litellm, "sse_keepalive_ping_interval_seconds", None) - async def _lines(): - yield 'data: {"jsonrpc": "2.0", "id": "req-1", "result": {"kind": "task"}}' - with patch( "litellm.llms.custom_httpx.http_handler.get_async_httpx_client", - return_value=_sse_agent_handler(_lines), + return_value=_sse_agent_handler( + _sse_upstream_response('data: {"jsonrpc": "2.0", "id": "req-1", "result": {"kind": "task"}}') + ), ): response = await _resubscribe_response() assert "x-accel-buffering" not in response.headers @@ -2728,6 +2702,28 @@ async def test_forward_jsonrpc_sse_is_untouched_while_keepalives_are_unconfigure assert json.loads(chunks[-1].removeprefix("data: "))["result"]["kind"] == "task" +@pytest.mark.asyncio +async def test_forward_jsonrpc_sse_keeps_unicode_line_separators_inside_an_event(monkeypatch: pytest.MonkeyPatch) -> None: + monkeypatch.setattr(litellm, "sse_keepalive_ping_interval_seconds", None) + separated_text: Final = "before\u2028between\u2029after" + event: Final = { + "jsonrpc": "2.0", + "id": "req-1", + "result": {"kind": "message", "parts": [{"kind": "text", "text": separated_text}]}, + } + + with patch( + "litellm.llms.custom_httpx.http_handler.get_async_httpx_client", + return_value=_sse_agent_handler(_sse_upstream_response(f"data: {json.dumps(event, ensure_ascii=False)}")), + ): + response: Final = await _resubscribe_response() + chunks: Final = [chunk async for chunk in response.body_iterator] + + assert [json.loads(chunk.removeprefix("data: "))["result"]["parts"][0]["text"] for chunk in chunks] == [ + separated_text + ] + + async def _stream_message_response(): from litellm.proxy.agent_endpoints.a2a_endpoints import _handle_stream_message diff --git a/tests/unit/proxy/anthropic_endpoints/test_streaming_model_restamp.py b/tests/unit/proxy/anthropic_endpoints/test_streaming_model_restamp.py index 85bb8304b95..5911fcdc073 100644 --- a/tests/unit/proxy/anthropic_endpoints/test_streaming_model_restamp.py +++ b/tests/unit/proxy/anthropic_endpoints/test_streaming_model_restamp.py @@ -4,6 +4,7 @@ Tests for restamping the public model on Anthropic Messages streaming chunks. import json from collections.abc import AsyncIterator, Sequence +from typing import Final import pytest @@ -46,6 +47,28 @@ def test_restamps_sse_bytes_frame(): assert b"event: message_start" in restamped +def test_restamps_sse_bytes_frame_holding_a_line_separator() -> None: + payload: Final = { + "type": "message_start", + "message": { + "id": "msg_1", + "type": "message", + "role": "assistant", + "model": "claude-haiku-4-5-20251001", + "content": [{"type": "text", "text": "alpha\u2028beta"}], + }, + } + frame: Final = f"event: message_start\ndata: {json.dumps(payload, ensure_ascii=False)}\n\n".encode() + + restamped: Final = restamp_anthropic_stream_chunk_model(frame, "claude-auto-1") + + assert isinstance(restamped, bytes) + data_line: Final = next(line for line in restamped.decode().split("\n") if line.startswith("data:")) + message: Final = json.loads(data_line[len("data:") :])["message"] + assert message["model"] == "claude-auto-1" + assert message["content"] == [{"type": "text", "text": "alpha\u2028beta"}] + + def test_restamps_event_dict(): chunk = {"type": "message_start", "message": {"id": "msg_1", "model": "claude-sonnet-4-6"}} diff --git a/tests/unit/proxy/pass_through_endpoints/llm_provider_handlers/test_anthropic_passthrough_logging_handler.py b/tests/unit/proxy/pass_through_endpoints/llm_provider_handlers/test_anthropic_passthrough_logging_handler.py index e09a179eadf..47b6af9aeb3 100644 --- a/tests/unit/proxy/pass_through_endpoints/llm_provider_handlers/test_anthropic_passthrough_logging_handler.py +++ b/tests/unit/proxy/pass_through_endpoints/llm_provider_handlers/test_anthropic_passthrough_logging_handler.py @@ -2503,6 +2503,33 @@ def test_stream_was_interrupted_skips_data_lines_that_are_not_json_objects( assert AnthropicPassthroughLoggingHandler._stream_was_interrupted(all_chunks) is interrupted + +def test_stream_was_interrupted_reads_a_message_delta_whose_json_holds_unicode_line_separators() -> None: + message_delta: Final = { + "type": "message_delta", + "delta": {"stop_reason": "stop_sequence", "stop_sequence": "END\u2028OF\u2029TURN"}, + "usage": {"output_tokens": 3}, + } + all_chunks: Final = [ + 'data: {"type": "content_block_stop", "index": 0}', + f"data: {json.dumps(message_delta, ensure_ascii=False)}", + ] + + assert AnthropicPassthroughLoggingHandler._stream_was_interrupted(all_chunks) is False + + +def test_extract_model_reads_a_message_start_whose_json_holds_unicode_line_separators() -> None: + message_start: Final = { + "type": "message_start", + "message": { + "model": "claude-sonnet-5", + "content": [{"type": "text", "text": "Prefill\u2028line\u2029two"}], + }, + } + chunk: Final = f"event: message_start\ndata: {json.dumps(message_start, ensure_ascii=False)}\n\n" + + assert AnthropicPassthroughLoggingHandler._extract_model_from_anthropic_chunks([chunk]) == "claude-sonnet-5" + @pytest.mark.asyncio async def test_anthropic_passthrough_handler(mock_httpx_response, mock_response, mock_logging_obj): """ @@ -3004,3 +3031,24 @@ def test_anthropic_passthrough_stream_bills_the_bedrock_guardrail_trace_of_the_m (entry,) = _recorded_guardrail_entries(mock_logging_obj) assert entry["guardrail_usage"] == {"topicPolicyUnits": 3, "contentPolicyUnits": 3} assert entry["guardrail_cost"] == pytest.approx(6 * 0.00015) + + +def test_anthropic_passthrough_stream_bills_a_bedrock_guardrail_trace_whose_json_holds_unicode_line_separators( + monkeypatch: pytest.MonkeyPatch, mock_logging_obj: LiteLLMLoggingObj, all_chunks: list[str] +) -> None: + monkeypatch.setattr(litellm, "model_cost", _BEDROCK_TRACE_PRICING) + mock_logging_obj.model_call_details["litellm_params"] = {"metadata": {}} + trace: Final = {"guardrail": {**_BEDROCK_INVOKE_TRACE["guardrail"], "actionReason": "Blocked\u2028by\u2029policy"}} + stop_event: Final = json.dumps({"type": "message_stop", "amazon-bedrock-trace": trace}, ensure_ascii=False) + AnthropicPassthroughLoggingHandler.handle_logging_anthropic_collected_chunks( + litellm_logging_obj=mock_logging_obj, + passthrough_success_handler_obj=MagicMock(), + url_route="/v1/messages", + request_body={"model": "claude", "messages": []}, + endpoint_type=MagicMock(), + start_time=datetime(2025, 1, 1), + all_chunks=[*all_chunks, f"event: message_stop\ndata: {stop_event}\n\n"], + end_time=datetime(2025, 1, 1), + ) + (entry,) = _recorded_guardrail_entries(mock_logging_obj) + assert entry["guardrail_usage"] == {"topicPolicyUnits": 3, "contentPolicyUnits": 3} diff --git a/tests/unit/proxy/pass_through_endpoints/llm_provider_handlers/test_tinyfish_passthrough_logging_handler.py b/tests/unit/proxy/pass_through_endpoints/llm_provider_handlers/test_tinyfish_passthrough_logging_handler.py index 1b83c5140ca..b0f84f2ecbe 100644 --- a/tests/unit/proxy/pass_through_endpoints/llm_provider_handlers/test_tinyfish_passthrough_logging_handler.py +++ b/tests/unit/proxy/pass_through_endpoints/llm_provider_handlers/test_tinyfish_passthrough_logging_handler.py @@ -1,6 +1,7 @@ import asyncio import json from datetime import datetime +from typing import Final from unittest.mock import AsyncMock, MagicMock import httpx @@ -267,6 +268,12 @@ class TestRunIdFromSseFrames: frames = b'data: {"run_id": "run-7", "event": "INITIALIZED"}\n\ndata: {"run_id": "run-7", "event": "ACTION"}\n\n' assert run_id_from_sse_frames(frames) == "run-7" + def test_finds_run_id_in_a_frame_whose_json_holds_unicode_line_separators(self) -> None: + event: Final = {"run_id": "run-7", "event": "ACTION", "message": "Clicked\u2028Submit\u2029button"} + frames: Final = f"data: {json.dumps(event, ensure_ascii=False)}\n\n".encode() + + assert run_id_from_sse_frames(frames) == "run-7" + def test_skips_frames_without_run_id(self): frames = b': keepalive\n\ndata: not-json\n\ndata: {"event": "HEARTBEAT"}\n\n' assert run_id_from_sse_frames(frames) is None diff --git a/tests/unit/proxy/pass_through_endpoints/test_managed_id_rewriter.py b/tests/unit/proxy/pass_through_endpoints/test_managed_id_rewriter.py index 49bc2e9c29e..2496aa47d22 100644 --- a/tests/unit/proxy/pass_through_endpoints/test_managed_id_rewriter.py +++ b/tests/unit/proxy/pass_through_endpoints/test_managed_id_rewriter.py @@ -30,7 +30,7 @@ from litellm.llms.base_llm.managed_resources.utils import( resolve_passthrough_managed_id_provider, ) from tests._vcr_conftest_common import install_live_call_probe, record_vcr_outcome -from typing import Any +from typing import Any, Final def _user() -> UserAPIKeyAuth: @@ -254,6 +254,37 @@ async def test_streamed_response_with_cr_only_frame_delimiters_is_still_owned_an assert output == _response_stream_bytes(managed_id).replace(b"\n", b"\r") +def _instructed_response_stream_bytes(raw_id: str, instructions: str) -> bytes: + events: Final = ( + {"type": "response.created", "response": {"id": raw_id, "status": "in_progress", "instructions": instructions}}, + {"type": "response.completed", "response": {"id": raw_id, "status": "completed", "instructions": instructions}}, + ) + return b"".join( + f"event: {event['type']}\ndata: {json.dumps(event, ensure_ascii=False)}\n\n".encode() for event in events + ) + + +@pytest.mark.asyncio +async def test_streamed_response_whose_json_holds_unicode_line_separators_is_still_owned_and_rewritten() -> None: + instructions: Final = "Answer\u2028in\u2029verse" + pc: Final = _prisma_client() + + output: Final = await _collect( + rewrite_streamed_response_ids( + stream=_chunks(_instructed_response_stream_bytes(RAW_RESPONSE_ID, instructions), 4096), + provider="openai", + method="POST", + route="/openai_passthrough/v1/responses", + user_api_key_dict=_user(), + prisma_client=pc, + ) + ) + + pc.db.litellm_managedobjecttable.upsert.assert_awaited_once() + managed_id: Final = pc.db.litellm_managedobjecttable.upsert.await_args.kwargs["data"]["create"]["unified_object_id"] + assert output == _instructed_response_stream_bytes(managed_id, instructions) + + @pytest.mark.asyncio async def test_streamed_bytes_untouched_on_routes_without_a_response_id(): pc = _prisma_client() diff --git a/tests/unit/proxy/response_polling/test_response_polling_handler.py b/tests/unit/proxy/response_polling/test_response_polling_handler.py index 63e38ce533a..42acad28488 100644 --- a/tests/unit/proxy/response_polling/test_response_polling_handler.py +++ b/tests/unit/proxy/response_polling/test_response_polling_handler.py @@ -17,7 +17,7 @@ following the OpenAI Response API format. import asyncio import json from datetime import datetime, timezone -from typing import Any, Dict, Optional +from typing import Any, Dict, Final, Optional from unittest.mock import AsyncMock, Mock, patch import pytest @@ -1400,7 +1400,7 @@ def _make_sse_stream(events: list) -> Mock: async def _body_iterator(): for event in events: - yield f"data: {json.dumps(event)}" + yield f"data: {json.dumps(event, ensure_ascii=False)}" yield "data: [DONE]" mock_response = Mock() @@ -1639,6 +1639,33 @@ class TestBackgroundStreamingTerminalEvents: assert final_call.kwargs["status"] == "completed" assert final_call.kwargs["usage"] == {"input_tokens": 10, "output_tokens": 50} + @pytest.mark.asyncio + async def test_response_completed_whose_json_holds_unicode_line_separators_keeps_its_output(self) -> None: + from litellm.proxy.response_polling.background_streaming import ( + background_streaming_task, + ) + + item: Final = { + "id": "item_1", + "type": "message", + "content": [{"type": "output_text", "text": "Line\u2028one\u2029two"}], + } + events: Final = [ + {"type": "response.completed", "response": {"id": "resp_123", "status": "completed", "output": [item]}} + ] + handler: Final = AsyncMock(spec=ResponsePollingHandler) + + with patch( + "litellm.proxy.response_polling.background_streaming.ProxyBaseLLMRequestProcessing" + ) as MockProcessor: + MockProcessor.return_value.base_process_llm_request = AsyncMock(return_value=_make_sse_stream(events)) + await background_streaming_task(**_make_background_streaming_kwargs("poll_u2028", handler)) + + calls: Final = handler.update_state.call_args_list + outputs: Final = [call.kwargs["output"] for call in calls if "output" in call.kwargs] + assert outputs[-1] == [item] + assert calls[-1].kwargs["status"] == "completed" + @pytest.mark.asyncio async def test_fallback_status_derived_from_event_type_when_status_field_missing( self, diff --git a/tests/unit/responses/litellm_completion_transformation/test_streaming_iterator_transformation.py b/tests/unit/responses/litellm_completion_transformation/test_streaming_iterator_transformation.py index ed90acbd2fa..0b971bc1696 100644 --- a/tests/unit/responses/litellm_completion_transformation/test_streaming_iterator_transformation.py +++ b/tests/unit/responses/litellm_completion_transformation/test_streaming_iterator_transformation.py @@ -1288,6 +1288,106 @@ async def test_reasoning_item_closes_before_a_tool_call_from_the_same_chunk_open ] +def _text_opening_a_tool_call_chunk(content: str, arguments: str = "") -> ModelResponseStream: + return ModelResponseStream( + id=CHAT_COMPLETION_ID, + created=1748575031, + model="claude-haiku-5-5", + object="chat.completion.chunk", + choices=[ + StreamingChoices( + index=0, + delta=Delta( + role="assistant", + content=content, + tool_calls=[ + { + "id": "call_weather", + "type": "function", + "function": {"name": "get_weather", "arguments": arguments}, + "index": 0, + } + ], + ), + finish_reason=None, + ) + ], + ) + + +def _tool_arguments_chunk(arguments: str, finish_reason: str | None = None) -> ModelResponseStream: + return ModelResponseStream( + id=CHAT_COMPLETION_ID, + created=1748575031, + model="claude-haiku-5-5", + object="chat.completion.chunk", + choices=[ + StreamingChoices( + index=0, + delta=Delta( + content=None, + tool_calls=[{"type": "function", "function": {"arguments": arguments}, "index": 0}], + ), + finish_reason=finish_reason, + ) + ], + ) + + +@pytest.mark.parametrize("sync_mode", [True, False]) +@pytest.mark.asyncio +async def test_tool_call_opened_in_a_text_chunk_streams_each_argument_delta(sync_mode: bool) -> None: + argument_pieces: Final = ('{"', "city", '"', ': "', "Paris", '"', "}") + iterator: Final = _build_iterator( + [ + _text_opening_a_tool_call_chunk(""), + *(_tool_arguments_chunk(piece) for piece in argument_pieces), + _chunk("", finish_reason="tool_calls"), + ] + ) + + events: Final = await _collect_events(iterator, sync_mode) + + assert _event_text(events, ResponsesAPIStreamEvents.OUTPUT_TEXT_DELTA) == "" + assert [ + event.delta + for event in events + if getattr(event, "type", None) == ResponsesAPIStreamEvents.FUNCTION_CALL_ARGUMENTS_DELTA + ] == list(argument_pieces) + + +@pytest.mark.parametrize("sync_mode", [True, False]) +@pytest.mark.asyncio +async def test_tool_call_sharing_a_chunk_with_text_opens_after_the_whole_message(sync_mode: bool) -> None: + iterator: Final = _build_iterator( + [ + _text_opening_a_tool_call_chunk("Checking ", arguments='{"city": "Paris"}'), + _chunk("the "), + _chunk("weather now."), + _chunk("", finish_reason="tool_calls"), + ] + ) + + events: Final = await _collect_events(iterator, sync_mode) + + event_types: Final = [getattr(event, "type", None) for event in events] + function_call_added_at: Final = next( + position + for position, event in enumerate(events) + if event_types[position] == ResponsesAPIStreamEvents.OUTPUT_ITEM_ADDED and event.item.type == "function_call" + ) + streamed: Final = [ + event + for event in events + if getattr(event, "type", None) + in (ResponsesAPIStreamEvents.OUTPUT_TEXT_DELTA, ResponsesAPIStreamEvents.FUNCTION_CALL_ARGUMENTS_DELTA) + ] + assert ResponsesAPIStreamEvents.OUTPUT_TEXT_DELTA not in event_types[function_call_added_at:] + assert [event.sequence_number for event in streamed] == sorted(event.sequence_number for event in streamed) + assert _event_text(events, ResponsesAPIStreamEvents.OUTPUT_TEXT_DELTA) == "Checking the weather now." + assert _event_text(events, ResponsesAPIStreamEvents.FUNCTION_CALL_ARGUMENTS_DELTA) == '{"city": "Paris"}' + + @pytest.mark.parametrize("sync_mode", [True, False]) @pytest.mark.asyncio async def test_tool_only_stream_puts_the_function_call_at_output_index_zero(sync_mode: bool): diff --git a/tests/unit/sandbox/test_e2b_sandbox.py b/tests/unit/sandbox/test_e2b_sandbox.py index e01b9120416..e58392ca266 100644 --- a/tests/unit/sandbox/test_e2b_sandbox.py +++ b/tests/unit/sandbox/test_e2b_sandbox.py @@ -7,6 +7,7 @@ integration tests live in tests/integration/sandbox/test_e2b_sandbox.py. """ import json +from typing import Final import httpx import pytest @@ -19,19 +20,18 @@ from litellm.llms.e2b.sandbox.transformation import ( ) +def _line_stream(lines: list[str]) -> httpx.Response: + return httpx.Response(200, stream=httpx.ByteStream("".join(f"{line}\n" for line in lines).encode())) + + class FakeResponse: - def __init__(self, *, json_data=None, lines=None, status_code=200): + def __init__(self, *, json_data: object = None, status_code: int = 200) -> None: self._json = json_data - self._lines = lines or [] self.status_code = status_code def json(self): return self._json - async def aiter_lines(self): - for line in self._lines: - yield line - class FakeHTTPClient: """Records outbound requests and returns canned responses keyed by URL.""" @@ -61,7 +61,7 @@ class FakeHTTPClient: if url.endswith("/execute"): if self.execute_raises is not None: raise self.execute_raises - return FakeResponse(lines=self.execute_lines) + return _line_stream(self.execute_lines) raise AssertionError(f"unexpected POST {url}") async def delete(self, url, headers=None, **kwargs): @@ -249,6 +249,20 @@ def test_parse_lines_skips_non_json_lines(): assert result.error is None +@pytest.mark.asyncio +async def test_run_code_keeps_unicode_line_separators_inside_stdout() -> None: + separated_text: Final = "before\u2028between\u2029after\n" + client: Final = FakeHTTPClient( + execute_lines=[json.dumps({"type": "stdout", "text": separated_text, "timestamp": 1}, ensure_ascii=False)] + ) + handle: Final = ContainerHandle(id="sbx_sep", provider="e2b", domain="e2b.app") + handle._hidden_params = {"envd_access_token": "tok"} + + result: Final = await E2BSandboxConfig().arun_code(container=handle, code="print(text)", client=client) + + assert result.stdout == separated_text + + @pytest.mark.asyncio async def test_run_code_aborts_on_output_over_cap(): big_line = "x" * (MAX_OUTPUT_BYTES + 1) diff --git a/tests/unit/sandbox/test_opensandbox_sandbox.py b/tests/unit/sandbox/test_opensandbox_sandbox.py index 2928dea100e..36ae848786c 100644 --- a/tests/unit/sandbox/test_opensandbox_sandbox.py +++ b/tests/unit/sandbox/test_opensandbox_sandbox.py @@ -27,10 +27,13 @@ def sse(data): return f"data: {json.dumps(data)}" +def _line_stream(lines: list[str]) -> httpx.Response: + return httpx.Response(200, stream=httpx.ByteStream("".join(f"{line}\n" for line in lines).encode())) + + class FakeResponse: - def __init__(self, *, json_data=None, lines=None, status_code=200): + def __init__(self, *, json_data: object = None, status_code: int = 200) -> None: self._json = json_data - self._lines = lines or [] self.status_code = status_code def json(self): @@ -40,10 +43,6 @@ class FakeResponse: if self.status_code >= 400: raise http_status_error(self.status_code) - async def aiter_lines(self): - for line in self._lines: - yield line - class FakeHTTPClient: def __init__( @@ -93,7 +92,7 @@ class FakeHTTPClient: if url.endswith("/code"): if self.execute_raises is not None: raise self.execute_raises - return FakeResponse(lines=self.execute_lines) + return _line_stream(self.execute_lines) raise AssertionError(f"unexpected POST {url}") async def get(self, url, headers=None, params=None, **kwargs): diff --git a/tests/unit/test_groq_streaming_encoding.py b/tests/unit/test_groq_streaming_encoding.py index ffe87fb9890..37f05272f67 100644 --- a/tests/unit/test_groq_streaming_encoding.py +++ b/tests/unit/test_groq_streaming_encoding.py @@ -6,148 +6,58 @@ UTF-8 encoded content in streaming responses, specifically fixing the ASCII encoding error described in issue #12660. """ -import asyncio -from unittest.mock import AsyncMock, Mock +import json +from typing import Final +from unittest.mock import Mock +import httpx import pytest +from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler, HTTPHandler from litellm.llms.openai_like.chat.handler import make_call, make_sync_call - -class MockResponse: - """Mock httpx response for testing UTF-8 handling.""" - - def __init__(self, test_content: str): - self.test_content = test_content - self.status_code = 200 - - def iter_text(self, encoding="utf-8"): - """Mock iter_text that yields content with the specified encoding.""" - yield self.test_content - - async def aiter_text(self, encoding="utf-8"): - """Mock aiter_text that yields content with the specified encoding.""" - yield self.test_content - - def iter_lines(self): - """Mock iter_lines method for synchronous streaming.""" - yield self.test_content - - async def aiter_lines(self): - """Mock aiter_lines method for asynchronous streaming.""" - yield self.test_content - - def json(self): - return {"choices": [{"delta": {"content": "test"}}]} +_UNICODE_TEXTS: Final = ( + "The symbol µ represents micro", + "Testing © character", + "Testing ™ character", + "Testing € character", + "Testing 北京 character", + "Testing 🚀 character", + "Testing Ñoño character", +) -class MockSyncClient: - """Mock synchronous HTTP client for testing.""" - - def __init__(self, response_content: str): - self.response_content = response_content - - def post(self, *args, **kwargs): - return MockResponse(self.response_content) +def _utf8_stream_transport(text: str) -> httpx.MockTransport: + chunk: Final = {"choices": [{"index": 0, "delta": {"content": text}}]} + body: Final = f"data: {json.dumps(chunk, ensure_ascii=False)}\n\n".encode() + return httpx.MockTransport(lambda request: httpx.Response(200, stream=httpx.ByteStream(body))) -class MockAsyncClient: - """Mock asynchronous HTTP client for testing.""" - - def __init__(self, response_content: str): - self.response_content = response_content - - async def post(self, *args, **kwargs): - return MockResponse(self.response_content) +def _call_args() -> dict[str, object]: + return { + "api_base": "https://test.com/v1/chat/completions", + "headers": {"Authorization": "Bearer test"}, + "data": '{"model": "test", "messages": []}', + "model": "test-model", + "messages": [], + "logging_obj": Mock(), + } -def test_utf8_streaming_sync(): - """Test that synchronous streaming handles UTF-8 characters correctly.""" - # Content with the µ character that was causing issues - test_content = ( - 'data: {"choices":[{"delta":{"content":"The symbol µ represents micro"}}]}\n\n' +@pytest.mark.parametrize("text", _UNICODE_TEXTS) +def test_utf8_streaming_sync(text: str) -> None: + completion_stream: Final = make_sync_call( + client=HTTPHandler(client=httpx.Client(transport=_utf8_stream_transport(text))), **_call_args() ) - mock_client = MockSyncClient(test_content) - mock_logging = Mock() - - # This should not raise an ASCII encoding error - completion_stream = make_sync_call( - client=mock_client, - api_base="https://test.com/v1/chat/completions", - headers={"Authorization": "Bearer test"}, - data='{"model": "test", "messages": []}', - model="test-model", - messages=[], - logging_obj=mock_logging, - ) - - # Verify we can iterate through the stream without encoding errors - assert completion_stream is not None + assert "".join(chunk["text"] for chunk in completion_stream) == text @pytest.mark.asyncio -async def test_utf8_streaming_async(): - """Test that asynchronous streaming handles UTF-8 characters correctly.""" - # Content with the µ character that was causing issues - test_content = ( - 'data: {"choices":[{"delta":{"content":"The symbol µ represents micro"}}]}\n\n' +@pytest.mark.parametrize("text", _UNICODE_TEXTS) +async def test_utf8_streaming_async(text: str) -> None: + completion_stream: Final = await make_call( + client=AsyncHTTPHandler(transport=_utf8_stream_transport(text)), **_call_args() ) - mock_client = MockAsyncClient(test_content) - mock_logging = Mock() - - # This should not raise an ASCII encoding error - completion_stream = await make_call( - client=mock_client, - api_base="https://test.com/v1/chat/completions", - headers={"Authorization": "Bearer test"}, - data='{"model": "test", "messages": []}', - model="test-model", - messages=[], - logging_obj=mock_logging, - ) - - # Verify we can iterate through the stream without encoding errors - assert completion_stream is not None - - -def test_various_unicode_characters(): - """Test streaming with various Unicode characters that could cause issues.""" - unicode_test_cases = [ - "µ", # Micro symbol (the original issue) - "©", # Copyright symbol - "™", # Trademark symbol - "€", # Euro symbol - "北京", # Chinese characters - "🚀", # Emoji - "Ñoño", # Spanish characters with tildes - ] - - for unicode_char in unicode_test_cases: - test_content = f'data: {{"choices":[{{"delta":{{"content":"Testing {unicode_char} character"}}}}]}}\n\n' - - mock_client = MockSyncClient(test_content) - mock_logging = Mock() - - # This should not raise an ASCII encoding error for any Unicode character - completion_stream = make_sync_call( - client=mock_client, - api_base="https://test.com/v1/chat/completions", - headers={"Authorization": "Bearer test"}, - data='{"model": "test", "messages": []}', - model="test-model", - messages=[], - logging_obj=mock_logging, - ) - - assert ( - completion_stream is not None - ), f"Failed to handle Unicode character: {unicode_char}" - - -if __name__ == "__main__": - test_utf8_streaming_sync() - asyncio.run(test_utf8_streaming_async()) - test_various_unicode_characters() - print("All UTF-8 streaming tests passed!") + assert "".join([chunk["text"] async for chunk in completion_stream]) == text From 7650f766b006077442abe758ee4b22b5c436ad18 Mon Sep 17 00:00:00 2001 From: Nate Armstrong Date: Wed, 28 Oct 2026 17:01:52 -0700 Subject: [PATCH 31/41] refactor(router): fence the router surface and document the router-core boundary (#48839) Add rustdoc overviews for a new litellm-router-core crate and for litellm-model-catalog, a module docstring on litellm/router.py, and a shrink-only allowlist of the Router members and router_utils modules used outside the router, checked in CI by router_surface_check.py. Move the router-free helpers out of litellm/router_utils: batch_utils to litellm_core_utils, and the auto-router naming contract, tuning baseline and baseline request next to the auto-router plugin. No behaviour change. --- .github/workflows/test-linting.yml | 3 + litellm-rust/Cargo.lock | 4 + litellm-rust/crates/model-catalog/src/lib.rs | 33 +++ litellm-rust/crates/router-core/Cargo.toml | 9 + litellm-rust/crates/router-core/src/lib.rs | 63 +++++ .../batch_utils.py | 0 .../prompt_templates/common_utils.py | 2 +- .../llms/anthropic/prompt_cache_prediction.py | 2 +- litellm/proxy/common_request_processing.py | 2 +- .../guardrails/auto_router_compression.py | 2 +- litellm/proxy/health_check.py | 2 +- .../proxy/hooks/autorouter_baseline_cache.py | 2 +- .../proxy/hooks/model_max_budget_limiter.py | 2 +- litellm/proxy/litellm_pre_call_utils.py | 2 +- .../auto_router_endpoints.py | 12 +- .../model_management_endpoints.py | 31 +-- .../auto_router_availability.py | 4 +- .../auto_router_permissions.py | 5 +- litellm/proxy/proxy_server.py | 14 +- litellm/router.py | 61 +++-- .../auto_router}/auto_router_model_naming.py | 0 .../auto_router_tuning_baseline.py | 0 .../auto_router}/baseline_request.py | 0 litellm/router_strategy/budget_limiter.py | 2 +- litellm/router_strategy/least_busy.py | 2 +- litellm/router_strategy/lowest_cost.py | 2 +- litellm/router_strategy/lowest_latency.py | 2 +- litellm/router_strategy/lowest_tpm_rpm.py | 2 +- litellm/router_strategy/lowest_tpm_rpm_v2.py | 2 +- .../router_utils/fallback_event_handlers.py | 2 +- .../router_code_coverage.py | 4 +- .../router_surface_allowlist.json | 211 ++++++++++++++++ .../router_surface_check.py | 235 ++++++++++++++++++ .../unbounded_in_baseline.txt | 2 +- .../test_batch_utils.py | 13 +- .../test_auto_router_endpoints.py | 4 +- .../test_model_management_endpoints.py | 49 ++-- .../test_auto_router_availability.py | 7 +- .../unit/proxy/proxy_server/test_lifecycle.py | 12 +- .../test_auto_router_model_naming.py | 8 +- .../test_auto_router_tuning_baseline.py | 4 +- .../test_baseline_request.py | 2 +- .../router_strategy/test_complexity_router.py | 36 ++- tests/unit/router_strategy/test_llm_v2.py | 4 +- tests/unit/test_private_usage_aliases.py | 2 +- .../test_router/test_router_helper_utils.py | 15 +- .../test_router/test_router_surface_check.py | 58 +++++ 47 files changed, 797 insertions(+), 138 deletions(-) create mode 100644 litellm-rust/crates/router-core/Cargo.toml create mode 100644 litellm-rust/crates/router-core/src/lib.rs rename litellm/{router_utils => litellm_core_utils}/batch_utils.py (100%) rename litellm/{router_utils => router_strategy/auto_router}/auto_router_model_naming.py (100%) rename litellm/{router_utils => router_strategy/auto_router}/auto_router_tuning_baseline.py (100%) rename litellm/{router_utils => router_strategy/auto_router}/baseline_request.py (100%) create mode 100644 tests/code_coverage_tests/router_surface_allowlist.json create mode 100644 tests/code_coverage_tests/router_surface_check.py rename tests/unit/{router_utils => litellm_core_utils}/test_batch_utils.py (97%) rename tests/unit/{router_utils => router_strategy}/test_auto_router_model_naming.py (99%) rename tests/unit/{router_utils => router_strategy}/test_auto_router_tuning_baseline.py (99%) rename tests/unit/{router_utils => router_strategy}/test_baseline_request.py (95%) create mode 100644 tests/unit/test_router/test_router_surface_check.py diff --git a/.github/workflows/test-linting.yml b/.github/workflows/test-linting.yml index b9f8d7e7dfb..89ca60ad9d1 100644 --- a/.github/workflows/test-linting.yml +++ b/.github/workflows/test-linting.yml @@ -561,6 +561,9 @@ jobs: - name: router_code_coverage run: uv run --no-sync python ./tests/code_coverage_tests/router_code_coverage.py + - name: router_surface_check + run: uv run --no-sync python ./tests/code_coverage_tests/router_surface_check.py + - name: test_chat_completion_imports run: uv run --no-sync python ./tests/code_coverage_tests/test_chat_completion_imports.py diff --git a/litellm-rust/Cargo.lock b/litellm-rust/Cargo.lock index 9c45c5c4b1c..10e590b5006 100644 --- a/litellm-rust/Cargo.lock +++ b/litellm-rust/Cargo.lock @@ -4393,6 +4393,10 @@ dependencies = [ "rstest", ] +[[package]] +name = "litellm-router-core" +version = "0.1.0" + [[package]] name = "litellm-router-types" version = "0.1.0" diff --git a/litellm-rust/crates/model-catalog/src/lib.rs b/litellm-rust/crates/model-catalog/src/lib.rs index 066c2c83c6b..d42c05e6fe7 100644 --- a/litellm-rust/crates/model-catalog/src/lib.rs +++ b/litellm-rust/crates/model-catalog/src/lib.rs @@ -1,3 +1,36 @@ +//! The model catalog: per-model pricing, context limits and capabilities. +//! +//! This is the Rust form of `model_prices_and_context_window.json`, the map Python reads as +//! `litellm.model_cost` and through `litellm.get_model_info`. It answers "what does this model +//! cost and what can it do"; it knows nothing about configured deployments. The router's view of +//! deployments (groups, aliases, wildcard patterns, team mappings) is the router registry, defined +//! in `litellm-router-core`, which reads this catalog through its dynamic config. +//! +//! # Structure +//! - `Catalog`: the parsed map. `Catalog::parse` takes the JSON body and a `Provenance` (where +//! the body came from: the bundled copy in the repo or the remote copy, with revision and etag), +//! and keeps every entry in file order with its raw fields. +//! - `ModelEntry`: one model. `ModelEntry::fields` is the raw JSON object, so unknown fields are +//! never lost; `ModelEntry::info` is the typed mirror `ModelInfo` (pricing in `pricing`, +//! modalities and modes in `capabilities`). +//! - Aliases: an entry may list `aliases`; each alias resolves to its canonical key. Problems +//! found while indexing them (bad lists, collisions) are kept as `AliasIssue`s instead of +//! failing the parse. +//! - `FallbackGeneralizations` / `FallbackRule`: regex rules that map unknown model ids onto known +//! families. They are data in the file, not entries. +//! - Validation: `Catalog::validate` checks a catalog against `IntegrityLimits` (a remote copy that +//! lost most of its models is refused), and `validate_registry` checks every entry against the +//! checked-in contract. +//! +//! # Lookups +//! `Catalog::lookup` resolves a key exactly first (entry or alias), then case-insensitively, and +//! returns a `ModelMatch` with both the key that matched and the canonical key of the entry. Provider +//! prefixes are not stripped here; callers decide which keys to try. +//! +//! # Readers +//! Cost tracking, the router (through `RouterDynamicConfig`) and the gateway. Nothing in this crate +//! performs I/O; loading and refreshing the file is the caller's job. + mod capabilities; mod catalog; mod error; diff --git a/litellm-rust/crates/router-core/Cargo.toml b/litellm-rust/crates/router-core/Cargo.toml new file mode 100644 index 00000000000..af97bd52c57 --- /dev/null +++ b/litellm-rust/crates/router-core/Cargo.toml @@ -0,0 +1,9 @@ +[package] +name = "litellm-router-core" +version = "0.1.0" +edition.workspace = true +license.workspace = true +repository.workspace = true +description = "The router core behind litellm.Router: registry, settings, health, selection and the call loop" + +[dependencies] diff --git a/litellm-rust/crates/router-core/src/lib.rs b/litellm-rust/crates/router-core/src/lib.rs new file mode 100644 index 00000000000..05d95d02c84 --- /dev/null +++ b/litellm-rust/crates/router-core/src/lib.rs @@ -0,0 +1,63 @@ +//! Routing for LiteLLM: decide which deployment serves a request, and drive retries, fallbacks +//! and cooldowns around the attempts. +//! +//! This crate defines the boundary `litellm.Router` is moving behind. Today the Python router in +//! `litellm/router.py` implements all of it; this crate is where the Rust core will implement the +//! same interface. Until then it holds this overview and nothing else, so the interface is written +//! down in code before either side changes. +//! +//! # The public surface +//! Code outside the router uses `litellm.Router` through two kinds of members and nothing else: +//! - The endpoint methods (`acompletion`, `aembedding`, `aresponses`, `aanthropic_messages`, the +//! files, batches, assistants and vector store wrappers, 37 in all). Each is a thin wrapper over +//! `call(op, request)`. +//! - Eight entry points: `registry()`, `settings()`, `health()`, `select()`, `apply()`, +//! `observe()`, `register_plugin()` and `retire()`. +//! +//! Every other member that outside code uses today has a named replacement among those. The old +//! names stay available for about a month through a `DeprecatedRouterMembers` mixin, which +//! implements them over the entry points and warns on use. A CI check +//! (`tests/code_coverage_tests/router_surface_check.py`) keeps an allowlist of the members outside +//! code touches; the list may only shrink. +//! +//! # Structure +//! - `Registry`: the configured deployments, indexed by id, model group, alias, wildcard pattern +//! and team. Immutable and versioned. Seven queries: `deployment(id)`, `deployments(filter)`, +//! `names(filter)`, `resolve(name, scope)`, `group(name)`, `access_groups()` and +//! `credentials(id)`. "Registry" is the router's view of what is configured; per-model pricing +//! and capabilities are the model catalog (`litellm-model-catalog`), which the router reads +//! through its dynamic config. +//! - `RouterConfig` / `RouterDynamicConfig`: everything the router needs to know, passed in. The +//! router never reads process globals. Static config (model list, strategy, retry, timeout and +//! cooldown policy, Redis settings, plugins) is fixed for the router's life. Dynamic config +//! (pricing, model aliases, fallbacks, retry counts, callbacks) is a versioned snapshot read +//! once per call. +//! - `Settings`: one typed, frozen view of the configuration a caller may read. +//! - `Health`: cooldowns, blocked deployments and usage, kept in shared cache keys so Python and +//! Rust workers see the same state. `report(id, outcome)` records outcomes the router did not +//! see itself (a health check, an interceptor that marks a deployment failed). +//! - `select(model, request)`: resolve a name, filter eligible deployments, pick one with the +//! configured strategy, without calling it. +//! - `apply(changes)`: one `ChangeSet` (upserts, deletes, settings), applied atomically, producing +//! a new registry version. +//! - `call(op, request)`: select, run the attempt, and retry or fall back until it succeeds or the +//! policy is exhausted. +//! +//! # Execution +//! Each attempt resolves an `Executor` after a deployment is picked. A Rust-native operation runs +//! its `inference-*` route machine in-process; anything else goes through `Host::invoke`. A +//! started attempt is never replayed on a different executor. +//! +//! # Hosts +//! `Host` is what the embedding process provides: `invoke` for non-native operations, `select` +//! for strategies and sub-routers not yet ported, `allow_fallback` for access and budget checks, +//! `sleep`, `observe` for routing events, and `cache`, a DualCache-style store that holds all +//! routing state under the same keys, encodings and TTLs the Python router uses. `PythonHost` +//! serves them from the existing Python code when the router runs inside Python; `NativeHost` +//! serves them in the Rust server, where an operation without a Rust implementation makes the +//! config fail at build time, never partway through a request. +//! +//! # Plugins +//! Routing strategies, pre-call checks and sub-routers (auto, complexity, quality, adaptive) +//! register through `register_plugin` and see a `RoutingContext` (registry, settings, health and +//! the current request), never the router itself. diff --git a/litellm/router_utils/batch_utils.py b/litellm/litellm_core_utils/batch_utils.py similarity index 100% rename from litellm/router_utils/batch_utils.py rename to litellm/litellm_core_utils/batch_utils.py diff --git a/litellm/litellm_core_utils/prompt_templates/common_utils.py b/litellm/litellm_core_utils/prompt_templates/common_utils.py index cf6bdc69556..101f069c0f8 100644 --- a/litellm/litellm_core_utils/prompt_templates/common_utils.py +++ b/litellm/litellm_core_utils/prompt_templates/common_utils.py @@ -21,8 +21,8 @@ from pydantic import TypeAdapter import litellm from litellm import verbose_logger +from litellm.litellm_core_utils.batch_utils import InMemoryFile from litellm.litellm_core_utils.core_helpers import as_str_mapping -from litellm.router_utils.batch_utils import InMemoryFile from litellm.types.llms.openai import ( AllMessageValues, ChatCompletionAssistantMessage, diff --git a/litellm/llms/anthropic/prompt_cache_prediction.py b/litellm/llms/anthropic/prompt_cache_prediction.py index 7a4c707860d..17c178d169e 100644 --- a/litellm/llms/anthropic/prompt_cache_prediction.py +++ b/litellm/llms/anthropic/prompt_cache_prediction.py @@ -29,7 +29,7 @@ from litellm.llms.anthropic.messages.native.transformation import ( AnthropicMessagesConfig, ) from litellm.llms.anthropic.messages.native.utils import AnthropicMessagesRequestUtils, prepare_native_messages -from litellm.router_utils.baseline_request import ( +from litellm.router_strategy.auto_router.baseline_request import ( BASELINE_PARAMETERS, capture_baseline_parameters, ) diff --git a/litellm/proxy/common_request_processing.py b/litellm/proxy/common_request_processing.py index 70e305f46fb..9bb451495c8 100644 --- a/litellm/proxy/common_request_processing.py +++ b/litellm/proxy/common_request_processing.py @@ -139,8 +139,8 @@ from litellm.proxy.utils import ( # noqa: F401 # legacy module exports ) from litellm.responses.streaming_iterator import partial_responses_incomplete_event_on_disconnect, spawn_logging_task from litellm.router import Router +from litellm.router_strategy.auto_router.auto_router_model_naming import AUTO_ROUTER_MODEL_PREFIX from litellm.router_utils.add_retry_fallback_headers import get_hidden_params_dict -from litellm.router_utils.auto_router_model_naming import AUTO_ROUTER_MODEL_PREFIX from litellm.router_utils.common_utils import resolve_model_group_alias, resolve_served_model from litellm.types.extraction import ExtractionResponse from litellm.types.guardrails import GuardrailEventHooks diff --git a/litellm/proxy/guardrails/auto_router_compression.py b/litellm/proxy/guardrails/auto_router_compression.py index 1267fb15db8..3071bcbb1dc 100644 --- a/litellm/proxy/guardrails/auto_router_compression.py +++ b/litellm/proxy/guardrails/auto_router_compression.py @@ -14,7 +14,7 @@ from typing import TYPE_CHECKING, Final from litellm._logging import verbose_proxy_logger from litellm.litellm_core_utils.core_helpers import get_or_create_metadata_bucket -from litellm.router_utils.auto_router_model_naming import AUTO_ROUTER_MODEL_PREFIX +from litellm.router_strategy.auto_router.auto_router_model_naming import AUTO_ROUTER_MODEL_PREFIX from litellm.types.utils import GenericGuardrailAPIInputs if TYPE_CHECKING: diff --git a/litellm/proxy/health_check.py b/litellm/proxy/health_check.py index 0d2f8dc9a0c..be32405154c 100644 --- a/litellm/proxy/health_check.py +++ b/litellm/proxy/health_check.py @@ -27,7 +27,7 @@ from litellm.constants import ( HEALTH_CHECK_TIMEOUT_SECONDS, ) from litellm.litellm_core_utils.health_check_helpers import native_health_check_mode -from litellm.router_utils.auto_router_model_naming import ( +from litellm.router_strategy.auto_router.auto_router_model_naming import ( StrategyRouterDependency, classify_strategy_router_model, strategy_router_dependencies, diff --git a/litellm/proxy/hooks/autorouter_baseline_cache.py b/litellm/proxy/hooks/autorouter_baseline_cache.py index c4232609124..1147ce7ec12 100644 --- a/litellm/proxy/hooks/autorouter_baseline_cache.py +++ b/litellm/proxy/hooks/autorouter_baseline_cache.py @@ -37,8 +37,8 @@ from litellm.proxy.spend_tracking.savings import ( _effective_model_info, # pyright: ignore[reportPrivateUsage] # existing deployment-price owner _proxy_llm_router, # pyright: ignore[reportPrivateUsage] # existing optional proxy-router owner ) +from litellm.router_strategy.auto_router.baseline_request import baseline_request from litellm.router_strategy.complexity_router.context_compaction import compaction_applied -from litellm.router_utils.baseline_request import baseline_request from litellm.types.llms.base import LiteLLMBaseModel from litellm.types.router import BaselineRouteStamp from litellm.types.utils import CallTypes, ModelInfo, Usage diff --git a/litellm/proxy/hooks/model_max_budget_limiter.py b/litellm/proxy/hooks/model_max_budget_limiter.py index 11a47b2d424..1c4a61cfad8 100644 --- a/litellm/proxy/hooks/model_max_budget_limiter.py +++ b/litellm/proxy/hooks/model_max_budget_limiter.py @@ -19,6 +19,7 @@ from litellm._internal_context import with_service_target from litellm._logging import verbose_proxy_logger from litellm.caching.caching import DualCache from litellm.integrations.custom_logger import Span +from litellm.litellm_core_utils.batch_utils import batch_charged_once_marker_key, batch_id_to_charge_once from litellm.litellm_core_utils.duration_parser import duration_in_seconds from litellm.litellm_core_utils.litellm_logging import StandardLoggingPayloadSetup from litellm.llms.bedrock.common_utils import get_bedrock_base_model, get_bedrock_cross_region_inference_regions @@ -30,7 +31,6 @@ from litellm.proxy.hooks.model_budget_shared_windows import ( SharedWindows, ) from litellm.router_strategy.budget_limiter import RouterBudgetLimiting, budget_window_seconds -from litellm.router_utils.batch_utils import batch_charged_once_marker_key, batch_id_to_charge_once from litellm.types.llms.openai import AllMessageValues from litellm.types.utils import BudgetConfig, LlmProviders, StandardLoggingPayload diff --git a/litellm/proxy/litellm_pre_call_utils.py b/litellm/proxy/litellm_pre_call_utils.py index d4528bc85e8..2c05f724374 100644 --- a/litellm/proxy/litellm_pre_call_utils.py +++ b/litellm/proxy/litellm_pre_call_utils.py @@ -3290,7 +3290,7 @@ def _per_user_credential_names_for_groups( GITHUB_COPILOT_AUTH_TYPE_KEY, GITHUB_COPILOT_PER_USER_AUTH_TYPE, ) - from litellm.router_utils.auto_router_model_naming import classify_strategy_router_model + from litellm.router_strategy.auto_router.auto_router_model_naming import classify_strategy_router_model names: Final[list[str]] = [] # mutable-ok: accumulates one name per per-user deployment for group in model_groups: diff --git a/litellm/proxy/management_endpoints/auto_router_endpoints.py b/litellm/proxy/management_endpoints/auto_router_endpoints.py index 4f48f3c57b9..3dceab6b7f7 100644 --- a/litellm/proxy/management_endpoints/auto_router_endpoints.py +++ b/litellm/proxy/management_endpoints/auto_router_endpoints.py @@ -60,16 +60,16 @@ from litellm.repositories.table_repositories import PrismaTableRepository from litellm.repositories.team_repository import TeamRepository from litellm.repositories.user_repository import UserRepository from litellm.repositories.verification_token_repository import VerificationTokenRepository +from litellm.router_strategy.auto_router.auto_router_model_naming import ( + StrategyRouterDependencyRole, + classify_strategy_router_model, + strategy_router_dependencies, +) from litellm.router_strategy.complexity_router import ComplexityRouter from litellm.router_strategy.complexity_router.request_models import ( AutoRouterRoutingTestRequest, RequestComplexityRouterConfig, ) -from litellm.router_utils.auto_router_model_naming import ( - StrategyRouterDependencyRole, - classify_strategy_router_model, - strategy_router_dependencies, -) from litellm.types.llms.base import LiteLLMBaseModel from litellm.types.management_endpoints.auto_router_endpoints import ( SHADOW_EVAL_TURN_VALVE, @@ -411,7 +411,7 @@ async def validate_complexity_router_config( auto_router_environment_reference_error(data.complexity_router_config, user_api_key_dict) ) - from litellm.router_utils.auto_router_model_naming import ( + from litellm.router_strategy.auto_router.auto_router_model_naming import ( validate_complexity_router_config_write, ) diff --git a/litellm/proxy/management_endpoints/model_management_endpoints.py b/litellm/proxy/management_endpoints/model_management_endpoints.py index 45006b14df6..8cd3559ba86 100644 --- a/litellm/proxy/management_endpoints/model_management_endpoints.py +++ b/litellm/proxy/management_endpoints/model_management_endpoints.py @@ -131,6 +131,23 @@ from litellm.repositories.project_repository import ProjectRepository from litellm.repositories.table_repositories import ModelTableRepository from litellm.repositories.team_repository import TeamRepository from litellm.router import Router +from litellm.router_strategy.auto_router.auto_router_model_naming import ( + GATED_AUTO_ROUTER_CAPABILITIES, + STRATEGY_ROUTER_PARAM_FIELDS, + GatedAutoRouterCapability, + capability_limit_violation, + carries_complexity_router_settings, + count_capability_routers, + gated_capabilities_of, + is_complexity_router_model, + validate_complexity_router_config_placement, + validate_complexity_router_config_write, + validate_strategy_router_model_write, +) +from litellm.router_strategy.auto_router.auto_router_tuning_baseline import ( + is_mutable_tuned_candidate, + tuning_quota_violation, +) from litellm.router_strategy.complexity_router import ( DEFAULT_CLASSIFIER_CONTEXT_WINDOW_SIZE, ClassificationRubric, @@ -144,20 +161,6 @@ from litellm.router_strategy.complexity_router import ( normalize_classification_prompt, ) from litellm.router_strategy.complexity_router.config import resolve_complexity_router_config_write -from litellm.router_utils.auto_router_model_naming import ( - GATED_AUTO_ROUTER_CAPABILITIES, - STRATEGY_ROUTER_PARAM_FIELDS, - GatedAutoRouterCapability, - capability_limit_violation, - carries_complexity_router_settings, - count_capability_routers, - gated_capabilities_of, - is_complexity_router_model, - validate_complexity_router_config_placement, - validate_complexity_router_config_write, - validate_strategy_router_model_write, -) -from litellm.router_utils.auto_router_tuning_baseline import is_mutable_tuned_candidate, tuning_quota_violation from litellm.types.llms.base import LiteLLMBaseModel from litellm.types.llms.bedrock import AwsSessionTag from litellm.types.proxy.management_endpoints.model_management_endpoints import ( diff --git a/litellm/proxy/management_helpers/auto_router_availability.py b/litellm/proxy/management_helpers/auto_router_availability.py index 9b511ce0817..ce7859aff2d 100644 --- a/litellm/proxy/management_helpers/auto_router_availability.py +++ b/litellm/proxy/management_helpers/auto_router_availability.py @@ -7,14 +7,14 @@ from typing import Final from pydantic import Json, TypeAdapter, ValidationError from litellm.proxy.common_utils.encrypt_decrypt_utils import decrypt_value_helper -from litellm.router_utils.auto_router_model_naming import ( +from litellm.router_strategy.auto_router.auto_router_model_naming import ( GATED_AUTO_ROUTER_CAPABILITIES, capability_limit_violation, classify_strategy_router_model, count_capability_routers, gated_capabilities_of, ) -from litellm.router_utils.auto_router_tuning_baseline import ( +from litellm.router_strategy.auto_router.auto_router_tuning_baseline import ( is_mutable_tuned_candidate, mutable_tuned_identities, tuning_quota_violation, diff --git a/litellm/proxy/management_helpers/auto_router_permissions.py b/litellm/proxy/management_helpers/auto_router_permissions.py index 995446db7a6..8597f0d3667 100644 --- a/litellm/proxy/management_helpers/auto_router_permissions.py +++ b/litellm/proxy/management_helpers/auto_router_permissions.py @@ -34,13 +34,16 @@ from litellm.repositories.prisma_protocols import DatabaseClient from litellm.repositories.project_repository import ProjectRepository from litellm.repositories.table_repositories import TeamMembershipRepository from litellm.router import Router +from litellm.router_strategy.auto_router.auto_router_model_naming import ( + classify_strategy_router_model, + strategy_router_dependencies, +) from litellm.router_strategy.complexity_router.config import ( ComplexityRouterConfigWrite, normalize_classifier_config_aliases, resolve_complexity_router_config_write, ) from litellm.router_strategy.complexity_router.request_models import RequestComplexityRouterConfig -from litellm.router_utils.auto_router_model_naming import classify_strategy_router_model, strategy_router_dependencies from litellm.types.llms.base import LiteLLMBaseModel from litellm.types.router import Deployment, updateDeployment diff --git a/litellm/proxy/proxy_server.py b/litellm/proxy/proxy_server.py index 70e1c25a70a..b263412ce0f 100644 --- a/litellm/proxy/proxy_server.py +++ b/litellm/proxy/proxy_server.py @@ -168,12 +168,7 @@ from litellm.proxy.common_utils.token_count_settings import ( ) from litellm.proxy.management_helpers.auto_router_availability import AutoRouterCatalogEntry, build_auto_router_catalog from litellm.proxy.management_helpers.model_group_custom_fields import model_group_custom_fields -from litellm.router_utils.access_windows import access_windows_config_error -from litellm.router_utils.add_retry_fallback_headers import ( - get_fallback_errors_from_headers, - get_hidden_params_dict, -) -from litellm.router_utils.auto_router_model_naming import ( +from litellm.router_strategy.auto_router.auto_router_model_naming import ( GATED_AUTO_ROUTER_CAPABILITIES, STRATEGY_ROUTER_PARAM_FIELDS, capability_limit_violation, @@ -181,12 +176,17 @@ from litellm.router_utils.auto_router_model_naming import ( count_capability_routers, validate_complexity_router_config_placement, ) -from litellm.router_utils.auto_router_tuning_baseline import ( +from litellm.router_strategy.auto_router.auto_router_tuning_baseline import ( TUNING_BASELINE_PARAM_NAME, mutable_tuned_identities, snapshot_tuning_baselines, tuning_limit_violation, ) +from litellm.router_utils.access_windows import access_windows_config_error +from litellm.router_utils.add_retry_fallback_headers import ( + get_fallback_errors_from_headers, + get_hidden_params_dict, +) from litellm.router_utils.common_utils import resolve_model_group_alias from litellm.router_utils.routing_groups import parse_routing_groups from litellm.types.caching import RedisPipelineIncrementOperation diff --git a/litellm/router.py b/litellm/router.py index 408cb20945f..fc888495e99 100644 --- a/litellm/router.py +++ b/litellm/router.py @@ -7,6 +7,29 @@ # # Thank you ! We ❤️ you! - Krrish & Ishaan +""" +litellm.Router: load balancing, retries, fallbacks and cooldowns across deployments. + +Public surface. The router is moving behind a narrow interface; code outside +the router should end up using a Router through two kinds of members only: + +- endpoint methods (acompletion, aembedding, aresponses, aanthropic_messages, + the files, batches, assistants and vector store wrappers, ...), each a thin + wrapper around one call; +- entry points (being added) that replace direct access to internals: registry() for + deployment lookups, settings() for configuration reads, health() for + cooldowns, blocks and usage, select() to pick a deployment without calling + it, apply() for runtime changes, observe() for routing events, + register_plugin() for strategies and sub-routers, and retire() to release + the router. + +Everything else on the class is internal. The members outside code uses today +are listed in tests/code_coverage_tests/router_surface_allowlist.json, and a CI +check (router_surface_check.py next to it) fails on any new outside use, so the +list only shrinks while callers move to the entry points. The same interface +is documented for the Rust core in litellm-rust/crates/router-core. +""" + import asyncio import contextlib import copy @@ -85,6 +108,15 @@ from litellm.integrations.custom_logger import CustomLogger from litellm.integrations.otel.routing import routing_decision_attributes from litellm.integrations.otel.runtime import phase_attributes, phase_event, phase_span from litellm.litellm_core_utils.asyncify import run_async_function +from litellm.litellm_core_utils.batch_utils import ( + batch_create_routing, + deployment_supported_endpoints, + get_router_metadata_variable_name, + is_batch_retrieve_call_type, + replace_model_in_jsonl, + should_replace_model_in_jsonl, + without_deployments_that_cannot_batch, +) from litellm.litellm_core_utils.completion_timeout import CompletionTimeout from litellm.litellm_core_utils.core_helpers import ( coerce_token_limit, @@ -148,6 +180,15 @@ from litellm.llms.openai_like.model_info import ( get_openai_compatible_model_info, ) from litellm.responses.bridge_check import keep_registered_mode +from litellm.router_strategy.auto_router.auto_router_model_naming import ( + AUTO_ROUTER_MODEL_PREFIX, + STRATEGY_ROUTER_PARAM_FIELDS, + GatedAutoRouterCapability, + capability_limit_violation, + claimed_capabilities, + classify_strategy_router_model, + count_capability_routers, +) from litellm.router_strategy.base_routing_strategy import BaseRoutingStrategy from litellm.router_strategy.budget_limiter import RouterBudgetLimiting from litellm.router_strategy.complexity_router.context_compaction import ( @@ -187,24 +228,6 @@ from litellm.router_utils.add_retry_fallback_headers import ( replace_complexity_router_headers, response_total_token_count, ) -from litellm.router_utils.auto_router_model_naming import ( - AUTO_ROUTER_MODEL_PREFIX, - STRATEGY_ROUTER_PARAM_FIELDS, - GatedAutoRouterCapability, - capability_limit_violation, - claimed_capabilities, - classify_strategy_router_model, - count_capability_routers, -) -from litellm.router_utils.batch_utils import ( - batch_create_routing, - deployment_supported_endpoints, - get_router_metadata_variable_name, - is_batch_retrieve_call_type, - replace_model_in_jsonl, - should_replace_model_in_jsonl, - without_deployments_that_cannot_batch, -) from litellm.router_utils.client_initalization_utils import InitalizeCachedClient, MaxParallelRequestsLimit from litellm.router_utils.clientside_credential_handler import ( get_dynamic_litellm_params, @@ -16271,7 +16294,7 @@ class Router: to the deployment that actually served the request. Every attempt therefore writes or clears, never just writes. """ - from litellm.router_utils.baseline_request import capture_baseline_parameters + from litellm.router_strategy.auto_router.baseline_request import capture_baseline_parameters from litellm.types.router import BaselineRouteStamp phase_attributes(routing_decision_attributes(routing_decision)) diff --git a/litellm/router_utils/auto_router_model_naming.py b/litellm/router_strategy/auto_router/auto_router_model_naming.py similarity index 100% rename from litellm/router_utils/auto_router_model_naming.py rename to litellm/router_strategy/auto_router/auto_router_model_naming.py diff --git a/litellm/router_utils/auto_router_tuning_baseline.py b/litellm/router_strategy/auto_router/auto_router_tuning_baseline.py similarity index 100% rename from litellm/router_utils/auto_router_tuning_baseline.py rename to litellm/router_strategy/auto_router/auto_router_tuning_baseline.py diff --git a/litellm/router_utils/baseline_request.py b/litellm/router_strategy/auto_router/baseline_request.py similarity index 100% rename from litellm/router_utils/baseline_request.py rename to litellm/router_strategy/auto_router/baseline_request.py diff --git a/litellm/router_strategy/budget_limiter.py b/litellm/router_strategy/budget_limiter.py index eab232eae18..1d502dd8d95 100644 --- a/litellm/router_strategy/budget_limiter.py +++ b/litellm/router_strategy/budget_limiter.py @@ -36,12 +36,12 @@ from litellm._logging import verbose_router_logger from litellm.caching.caching import DualCache from litellm.caching.redis_cache import RedisCache, RedisPipelineIncrementOperation, log_redis_failure from litellm.integrations.custom_logger import CustomLogger, Span +from litellm.litellm_core_utils.batch_utils import batch_charged_once_marker_key, batch_id_to_charge_once from litellm.litellm_core_utils.core_helpers import ( get_metadata_variable_name_from_kwargs, ) from litellm.litellm_core_utils.duration_parser import duration_in_seconds_from from litellm.router_strategy.tag_based_routing import get_tags_from_request_kwargs -from litellm.router_utils.batch_utils import batch_charged_once_marker_key, batch_id_to_charge_once from litellm.router_utils.cooldown_callbacks import ( get_prometheus_logger_from_callbacks, ) diff --git a/litellm/router_strategy/least_busy.py b/litellm/router_strategy/least_busy.py index 6f3e0936641..8e0c3d6a7e7 100644 --- a/litellm/router_strategy/least_busy.py +++ b/litellm/router_strategy/least_busy.py @@ -10,7 +10,7 @@ from litellm._logging import verbose_router_logger from litellm.caching.caching import DualCache from litellm.caching.redis_cache import log_redis_failure from litellm.integrations.custom_logger import CustomLogger -from litellm.router_utils.batch_utils import is_batch_retrieve_call_type +from litellm.litellm_core_utils.batch_utils import is_batch_retrieve_call_type IN_FLIGHT_COUNT_TTL_SECONDS: Final = 60 * 60 diff --git a/litellm/router_strategy/lowest_cost.py b/litellm/router_strategy/lowest_cost.py index 5894da9cb4f..17d03fd4ff8 100644 --- a/litellm/router_strategy/lowest_cost.py +++ b/litellm/router_strategy/lowest_cost.py @@ -10,7 +10,7 @@ from litellm._internal_context import with_service_target from litellm._logging import verbose_router_logger from litellm.caching.caching import DualCache from litellm.integrations.custom_logger import CustomLogger -from litellm.router_utils.batch_utils import is_batch_retrieve_call_type +from litellm.litellm_core_utils.batch_utils import is_batch_retrieve_call_type from litellm.router_utils.deployment_pricing import deployment_pricing_models, str_keyed diff --git a/litellm/router_strategy/lowest_latency.py b/litellm/router_strategy/lowest_latency.py index 4e994d3fdc6..c352934db6c 100644 --- a/litellm/router_strategy/lowest_latency.py +++ b/litellm/router_strategy/lowest_latency.py @@ -13,9 +13,9 @@ from litellm import ModelResponse, token_counter, verbose_logger from litellm._internal_context import with_service_target from litellm.caching.caching import DualCache from litellm.integrations.custom_logger import CustomLogger +from litellm.litellm_core_utils.batch_utils import is_batch_retrieve_call_type from litellm.litellm_core_utils.core_helpers import get_parent_otel_span_from_kwargs, safe_divide_seconds from litellm.responses.utils import ResponseAPILoggingUtils -from litellm.router_utils.batch_utils import is_batch_retrieve_call_type from litellm.types.llms.openai import ResponsesAPIResponse from litellm.types.utils import LiteLLMPydanticObjectBase diff --git a/litellm/router_strategy/lowest_tpm_rpm.py b/litellm/router_strategy/lowest_tpm_rpm.py index 2d373e0c266..9636d7f009d 100644 --- a/litellm/router_strategy/lowest_tpm_rpm.py +++ b/litellm/router_strategy/lowest_tpm_rpm.py @@ -9,7 +9,7 @@ from litellm._internal_context import with_service_target from litellm._logging import verbose_router_logger from litellm.caching.caching import DualCache from litellm.integrations.custom_logger import CustomLogger -from litellm.router_utils.batch_utils import is_batch_retrieve_call_type +from litellm.litellm_core_utils.batch_utils import is_batch_retrieve_call_type from litellm.types.utils import LiteLLMPydanticObjectBase from litellm.utils import print_verbose diff --git a/litellm/router_strategy/lowest_tpm_rpm_v2.py b/litellm/router_strategy/lowest_tpm_rpm_v2.py index 527ea830d55..0719b30d36d 100644 --- a/litellm/router_strategy/lowest_tpm_rpm_v2.py +++ b/litellm/router_strategy/lowest_tpm_rpm_v2.py @@ -15,8 +15,8 @@ from litellm._internal_context import with_service_target from litellm._logging import verbose_logger, verbose_router_logger from litellm.caching.caching import DualCache from litellm.integrations.custom_logger import CustomLogger +from litellm.litellm_core_utils.batch_utils import is_batch_retrieve_call_type from litellm.litellm_core_utils.core_helpers import get_parent_otel_span_from_kwargs -from litellm.router_utils.batch_utils import is_batch_retrieve_call_type from litellm.types.router import RouterErrors from litellm.types.utils import LiteLLMPydanticObjectBase, StandardLoggingPayload from litellm.utils import get_utc_datetime, print_verbose diff --git a/litellm/router_utils/fallback_event_handlers.py b/litellm/router_utils/fallback_event_handlers.py index 0a77ad5b9ea..6d1de1f9f6c 100644 --- a/litellm/router_utils/fallback_event_handlers.py +++ b/litellm/router_utils/fallback_event_handlers.py @@ -13,6 +13,7 @@ from pydantic import TypeAdapter import litellm from litellm._logging import verbose_router_logger from litellm.integrations.custom_logger import CustomLogger +from litellm.litellm_core_utils.batch_utils import get_router_metadata_variable_name from litellm.litellm_core_utils.core_helpers import get_metadata_variable_name_from_kwargs, safe_deep_copy from litellm.litellm_core_utils.credential_accessor import reject_api_key_secret_reference from litellm.litellm_core_utils.get_llm_provider_logic import inferred_provider @@ -22,7 +23,6 @@ from litellm.router_utils.add_retry_fallback_headers import ( add_fallback_headers_to_response, get_fallback_error_info, ) -from litellm.router_utils.batch_utils import get_router_metadata_variable_name from litellm.router_utils.cooldown_handlers import ( _first_present, # pyright: ignore[reportPrivateUsage] - shared internal helper, used across router_utils cast_exception_status_to_int, diff --git a/tests/code_coverage_tests/router_code_coverage.py b/tests/code_coverage_tests/router_code_coverage.py index 4c9afd26dad..9fb3633d53c 100644 --- a/tests/code_coverage_tests/router_code_coverage.py +++ b/tests/code_coverage_tests/router_code_coverage.py @@ -187,13 +187,13 @@ ignored_function_names = [ def main(): router_file = [ "./litellm/router.py", - "./litellm/router_utils/batch_utils.py", + "./litellm/litellm_core_utils/batch_utils.py", "./litellm/router_utils/pattern_match_deployments.py", ] # router_file = [ # "../../litellm/router.py", # "../../litellm/router_utils/pattern_match_deployments.py", - # "../../litellm/router_utils/batch_utils.py", + # "../../litellm/litellm_core_utils/batch_utils.py", # ] ## LOCAL TESTING tests_dir = "./tests/" # Update this path if your tests directory is located elsewhere # tests_dir = "../../tests/" # LOCAL TESTING diff --git a/tests/code_coverage_tests/router_surface_allowlist.json b/tests/code_coverage_tests/router_surface_allowlist.json new file mode 100644 index 00000000000..428fd9b4011 --- /dev/null +++ b/tests/code_coverage_tests/router_surface_allowlist.json @@ -0,0 +1,211 @@ +{ + "_doc": "Router members and litellm.router_utils modules used outside the router. Shrink-only: see tests/code_coverage_tests/router_surface_check.py.", + "members_outside": [ + "_get_client", + "_zero_cost_cache", + "a_add_message", + "aadapter_completion", + "aanthropic_messages", + "abatch_completion", + "abatch_completion_fastest_response", + "acancel_batch", + "acancel_fine_tuning_job", + "acompletion", + "acreate_assistants", + "acreate_batch", + "acreate_file", + "acreate_fine_tuning_job", + "acreate_thread", + "adaptive_routers", + "adecisions", + "adelete_assistant", + "aembedding", + "afile_content", + "afile_delete", + "aget_assistants", + "aget_messages", + "aget_responses", + "aget_thread", + "alist_batches", + "alist_fine_tuning_jobs", + "allm_passthrough_route", + "amoderation", + "aocr", + "are_all_deployments_blocked", + "arefresh_model_info", + "aretrieve_batch", + "aretrieve_fine_tuning_job", + "arm_routing_read_prefetch", + "arun_thread", + "async_get_available_deployment", + "async_get_fully_unhealthy_model_names", + "async_get_healthy_deployments", + "async_pre_routing_hook", + "auto_routers", + "avector_store_create", + "avector_store_search", + "background_health_check_model_groups", + "bind_to_request_logging", + "cache", + "cache_responses", + "cached_deployment_model_info", + "cached_model_group_info", + "completion", + "complexity_routers", + "config_deployments", + "content_policy_fallbacks", + "context_window_fallbacks", + "cooldown_time", + "default_deployment", + "default_litellm_params", + "delete_deployment", + "deployment_latency_map", + "deployment_names", + "deployment_usable_by_team", + "deployments_for_request", + "deployments_named", + "discard", + "embedding", + "enable_health_check_routing", + "fallbacks", + "generate_model_id", + "get_all_deployments", + "get_available_deployment_for_pass_through", + "get_available_guardrail", + "get_candidate_model_ids_for_route", + "get_configured_display_name", + "get_configured_mode", + "get_configured_service_tiers", + "get_configured_token_limits", + "get_credential_deployment", + "get_deployment", + "get_deployment_by_model_group_name", + "get_deployment_credentials", + "get_deployment_credentials_with_provider", + "get_deployment_model_info", + "get_discovered_model_info", + "get_fully_blocked_model_names", + "get_model_access_groups", + "get_model_from_alias", + "get_model_group_info", + "get_model_group_usage", + "get_model_ids", + "get_model_info", + "get_model_list", + "get_model_list_from_model_alias", + "get_model_list_from_routing_groups", + "get_model_list_of_routed_group", + "get_model_listing_info", + "get_model_names", + "get_remaining_model_group_usage", + "get_routable_deployments", + "get_routable_upstream_model", + "get_router_deployment_budget_limiter", + "get_routing_group", + "get_settings", + "get_valid_args", + "guardrail_list", + "has_model_id", + "health_check_ignore_transient_errors", + "health_state_cache", + "is_deployment_blocked", + "is_model_access_group_for_wildcard_route", + "is_recognized_model", + "kubernetes_pod_discovery", + "map_team_model", + "model_group_alias", + "model_list", + "model_name_to_deployment_indices", + "model_names", + "pattern_router", + "priority_header_update", + "prompt_caching_pin_cache", + "provider_budget_config", + "quality_routers", + "resolve_model_name_from_model_id", + "resolve_served_model_name", + "resolved_litellm_models", + "routable_model_group", + "router_general_settings", + "routing_groups", + "routing_plugins", + "schedule_acompletion", + "search_tools", + "serve_through_request_logging", + "should_include_deployment", + "split_request_priority", + "team_model_to_deployment_indices", + "team_pattern_routers", + "team_ptu_ceiling", + "team_public_model_names", + "team_public_model_names_of", + "update_redis_cache", + "update_settings", + "upsert_deployment", + "wildcard_route_deployments", + "wildcard_routed_models" + ], + "members_router_family": [ + "_count_pre_call_check_tokens", + "_pre_call_checks", + "_update_kwargs_before_fallbacks", + "acompletion", + "adecisions", + "aembedding", + "allowed_fails", + "allowed_fails_policy", + "aresponses", + "async_function_with_fallbacks", + "async_get_healthy_deployments", + "cache", + "cached_deployment_model_info", + "cooldown_cache", + "cooldown_time", + "default_max_parallel_requests", + "deployments_named", + "disable_cooldowns", + "embedding", + "enable_pre_call_checks", + "enable_tag_filtering", + "explicit_allowed_fails", + "fallback_access_check", + "fallback_budget_check", + "fallbacks", + "get_all_deployments", + "get_allowed_fails_from_policy", + "get_deployment", + "get_model_group", + "get_model_ids", + "get_model_info", + "get_model_list", + "get_router_model_info", + "log_retry", + "model_list", + "provider_default_deployment_ids", + "routing_group_has_alternatives", + "routing_strategy", + "search_tools", + "set_custom_routing_strategy", + "slack_alerting_logger", + "tag_filtering_match_any", + "tag_routing_prefix", + "team_model_has_alternatives" + ], + "router_utils_imports_outside": [ + "litellm.router_utils.access_windows", + "litellm.router_utils.add_retry_fallback_headers", + "litellm.router_utils.clientside_credential_handler", + "litellm.router_utils.common_utils", + "litellm.router_utils.cooldown_handlers", + "litellm.router_utils.deployment_pricing", + "litellm.router_utils.fallback_event_handlers", + "litellm.router_utils.get_retry_from_policy", + "litellm.router_utils.health_state_cache", + "litellm.router_utils.pattern_match_deployments", + "litellm.router_utils.ptu_shares", + "litellm.router_utils.reasoning_effort_capability", + "litellm.router_utils.router_callbacks.track_deployment_metrics", + "litellm.router_utils.routing_groups", + "litellm.router_utils.search_api_router" + ] +} diff --git a/tests/code_coverage_tests/router_surface_check.py b/tests/code_coverage_tests/router_surface_check.py new file mode 100644 index 00000000000..0b35bac5240 --- /dev/null +++ b/tests/code_coverage_tests/router_surface_check.py @@ -0,0 +1,235 @@ +""" +Fence for the Router's public surface. + +Code outside the router may use only the Router members listed in +router_surface_allowlist.json, and may import only the litellm.router_utils +modules listed there. The lists hold what is used today and may only shrink: + +- a member or module used outside the router that is not listed fails the + check (use an existing entry point instead, or discuss widening the surface); +- a listed member or module that nothing outside the router uses any more + also fails, so the entry is removed in the same change. + +"Outside" means every non-test file under litellm/ and enterprise/ except +litellm/router.py and the router's own modules. Uses from the router family +(litellm/router_utils, litellm/router_strategy) are listed separately, because +those modules become private parts of the router core in a later step. + +Member uses are found the same way the boundary survey found them: attribute +access on a receiver whose name looks like a router (llm_router, +litellm_router_instance, router, ...) or on the Router class, plus +getattr/setattr/hasattr with a constant name on such a receiver. Members +include the container file endpoints Router.__init__ registers by name. + +Run: python tests/code_coverage_tests/router_surface_check.py [--write] +--write regenerates the allowlist from the current tree (for shrinking it). +""" + +import ast +import json +import os +import re +import sys +from typing import Dict, List, Set + +ROOT = os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "..")) +ALLOWLIST = os.path.join(os.path.dirname(__file__), "router_surface_allowlist.json") + +ROUTER_FILES = { + "litellm/router.py", +} +ROUTER_CLASS_FILES = ["litellm/router.py"] +ROUTER_CLASS_NAMES = {"Router"} + +SCAN_ROOTS = ["litellm", "enterprise"] +SKIP_DIRS = { + ".git", + "node_modules", + ".venv", + "__pycache__", + "litellm-rust", + "ui", + "docs", + "cookbook", + "tests", + "test", +} + +RECEIVER = re.compile( + r"(^|[._])(llm_router|router|litellm_router|litellm_router_instance|router_instance" + r"|_router|proxy_router|llm_router_instance|router_obj|mock_router|Router)$" +) +# Router.__init__ registers these with setattr(self, name, ...), so they never appear as self.. +CONTAINER_FILE_ENDPOINTS = "litellm/containers/endpoints.json" +DYNAMIC_FUNCS = {"getattr", "setattr", "hasattr"} + + +def _area(rel: str) -> str: + if rel.startswith(("litellm/router_utils/", "litellm/router_strategy/")): + return "router_family" + return "outside" + + +def _is_test(rel: str) -> bool: + name = os.path.basename(rel) + return "/tests/" in rel or name.startswith("test_") or name.endswith("_test.py") + + +def router_members() -> Set[str]: + members: Set[str] = set() + for rel in ROUTER_CLASS_FILES: + with open(os.path.join(ROOT, rel), encoding="utf-8") as f: + tree = ast.parse(f.read()) + for cls in tree.body: + if not isinstance(cls, ast.ClassDef) or cls.name not in ROUTER_CLASS_NAMES: + continue + for node in cls.body: + if isinstance(node, (ast.FunctionDef, ast.AsyncFunctionDef)): + members.add(node.name) + elif isinstance(node, ast.AnnAssign) and isinstance(node.target, ast.Name): + members.add(node.target.id) + elif isinstance(node, ast.Assign): + members.update(t.id for t in node.targets if isinstance(t, ast.Name)) + for node in ast.walk(cls): + if ( + isinstance(node, ast.Attribute) + and isinstance(node.value, ast.Name) + and node.value.id == "self" + and isinstance(node.ctx, ast.Store) + ): + members.add(node.attr) + with open(os.path.join(ROOT, CONTAINER_FILE_ENDPOINTS), encoding="utf-8") as f: + for endpoint in json.load(f)["endpoints"]: + members.update((endpoint["name"], endpoint["async_name"])) + return {m for m in members if not (m.startswith("__") and m.endswith("__"))} + + +def _python_files(): + for top in SCAN_ROOTS: + base = os.path.join(ROOT, top) + for dirpath, dirnames, filenames in os.walk(base): + dirnames[:] = [d for d in dirnames if d not in SKIP_DIRS] + for fn in filenames: + if fn.endswith(".py"): + path = os.path.join(dirpath, fn) + yield os.path.relpath(path, ROOT).replace(os.sep, "/"), path + + +def scan() -> Dict[str, Dict[str, Set[str]]]: + """Return {"members": {area: {member: files}}, "router_utils_imports": {module: files}}.""" + members = router_members() + uses: Dict[str, Dict[str, Set[str]]] = {"outside": {}, "router_family": {}} + imports: Dict[str, Set[str]] = {} + for rel, path in _python_files(): + if rel in ROUTER_FILES or _is_test(rel): + continue + with open(path, encoding="utf-8") as f: + src = f.read() + has_router = "router" in src.lower() + if not has_router: + continue + try: + tree = ast.parse(src) + except SyntaxError: + continue + area = _area(rel) + for name in member_uses(tree, members): + uses[area].setdefault(name, set()).add(rel) + if area == "outside": + for node in ast.walk(tree): + for mod in _router_utils_modules(node): + imports.setdefault(mod, set()).add(rel) + return {"members": uses, "router_utils_imports": imports} + + +def member_uses(tree: ast.AST, members: Set[str]) -> Set[str]: + """The Router members a module uses, on an instance or on the Router class itself.""" + used: Set[str] = set() + for node in ast.walk(tree): + if isinstance(node, ast.Attribute) and node.attr in members: + if RECEIVER.search(ast.unparse(node.value)): + used.add(node.attr) + elif ( + isinstance(node, ast.Call) + and isinstance(node.func, ast.Name) + and node.func.id in DYNAMIC_FUNCS + and len(node.args) >= 2 + and isinstance(node.args[1], ast.Constant) + and node.args[1].value in members + and RECEIVER.search(ast.unparse(node.args[0])) + ): + used.add(node.args[1].value) + return used + + +def _router_utils_modules(node: ast.AST) -> List[str]: + if isinstance(node, ast.ImportFrom) and node.module: + if node.module == "litellm.router_utils": + return ["litellm.router_utils." + a.name for a in node.names] + if node.module.startswith("litellm.router_utils."): + return [node.module] + if isinstance(node, ast.Import): + return [a.name for a in node.names if a.name.startswith("litellm.router_utils.")] + return [] + + +def build_allowlist(found) -> dict: + return { + "_doc": ( + "Router members and litellm.router_utils modules used outside the router. " + "Shrink-only: see tests/code_coverage_tests/router_surface_check.py." + ), + "members_outside": sorted(found["members"]["outside"]), + "members_router_family": sorted(found["members"]["router_family"]), + "router_utils_imports_outside": sorted(found["router_utils_imports"]), + } + + +def compare(allow: dict, found) -> List[str]: + errors: List[str] = [] + sections = [ + ("members_outside", found["members"]["outside"], "Router member"), + ("members_router_family", found["members"]["router_family"], "Router member (router family)"), + ("router_utils_imports_outside", found["router_utils_imports"], "litellm.router_utils import"), + ] + for key, used, what in sections: + listed = set(allow.get(key, [])) + for name in sorted(set(used) - listed): + files = ", ".join(sorted(used[name])) + errors.append( + f"new {what} outside the router: {name!r} in {files}. The router surface may only " + f"shrink: use an entry point instead." + ) + for name in sorted(listed - set(used)): + errors.append( + f"{what} {name!r} is listed in {os.path.basename(ALLOWLIST)} [{key}] but nothing uses it " + f"any more: remove it from the list." + ) + return errors + + +def main(argv: List[str]) -> int: + found = scan() + if "--write" in argv: + with open(ALLOWLIST, "w", encoding="utf-8") as f: + json.dump(build_allowlist(found), f, indent=2) + f.write("\n") + print(f"wrote {ALLOWLIST}") + return 0 + with open(ALLOWLIST, encoding="utf-8") as f: + allow = json.load(f) + errors = compare(allow, found) + for e in errors: + print(e) + if errors: + return 1 + print( + f"router surface ok: {len(allow['members_outside'])} members outside, " + f"{len(allow['members_router_family'])} from the router family, " + f"{len(allow['router_utils_imports_outside'])} router_utils modules imported outside" + ) + return 0 + + +if __name__ == "__main__": + sys.exit(main(sys.argv[1:])) diff --git a/tests/code_coverage_tests/unbounded_in_baseline.txt b/tests/code_coverage_tests/unbounded_in_baseline.txt index 1da964681a9..d8c6e2241f4 100644 --- a/tests/code_coverage_tests/unbounded_in_baseline.txt +++ b/tests/code_coverage_tests/unbounded_in_baseline.txt @@ -129,4 +129,4 @@ litellm/proxy/utils.py prefetch_config_params prisma param_name.in `param_names` litellm/repositories/daily_activity_repository.py DailyActivityRepository.daily_rows prisma [scope.entity_id_field].in `list(scope.entity_ids)` 0 litellm/repositories/daily_activity_repository.py DailyActivityRepository.daily_rows prisma api_key.in `list(scope.api_keys)` 0 litellm/repositories/daily_activity_repository.py DailyActivityRepository.daily_rows prisma not.in `list(scope.exclude_entity_ids)` 0 -litellm/router_utils/auto_router_model_naming.py raw-sql classifier_type.IN `IN ({_LLM_CLASSIFIER_TYPES_SQL})` 0 +litellm/router_strategy/auto_router/auto_router_model_naming.py raw-sql classifier_type.IN `IN ({_LLM_CLASSIFIER_TYPES_SQL})` 0 diff --git a/tests/unit/router_utils/test_batch_utils.py b/tests/unit/litellm_core_utils/test_batch_utils.py similarity index 97% rename from tests/unit/router_utils/test_batch_utils.py rename to tests/unit/litellm_core_utils/test_batch_utils.py index 79fe86a37c0..cd37d635c43 100644 --- a/tests/unit/router_utils/test_batch_utils.py +++ b/tests/unit/litellm_core_utils/test_batch_utils.py @@ -7,7 +7,7 @@ from typing import Dict, Final, List import pytest import litellm -from litellm.router_utils.batch_utils import ( +from litellm.litellm_core_utils.batch_utils import ( InMemoryFile, get_router_metadata_variable_name, parse_jsonl_with_embedded_newlines, @@ -116,7 +116,7 @@ def test_router_metadata_variable_name(): def test_non_json_input(): """Test that replace_model_in_jsonl returns original content for non-JSON input""" - from litellm.router_utils.batch_utils import replace_model_in_jsonl + from litellm.litellm_core_utils.batch_utils import replace_model_in_jsonl # Test with non-JSON string non_json_str = "This is not a JSON string" @@ -138,7 +138,7 @@ def test_non_json_input(): def test_should_replace_model_in_jsonl(): """Test that should_replace_model_in_jsonl returns the correct value""" - from litellm.router_utils.batch_utils import should_replace_model_in_jsonl + from litellm.litellm_core_utils.batch_utils import should_replace_model_in_jsonl assert should_replace_model_in_jsonl(purpose="batch") is True assert should_replace_model_in_jsonl(purpose="batch", passthrough=True) is False @@ -340,7 +340,7 @@ def test_replace_model_in_jsonl_with_embedded_newlines(): def test_is_batch_retrieve_call_type_matches_only_batch_retrieves(): - from litellm.router_utils.batch_utils import is_batch_retrieve_call_type + from litellm.litellm_core_utils.batch_utils import is_batch_retrieve_call_type from litellm.types.utils import CallTypes assert is_batch_retrieve_call_type(CallTypes.aretrieve_batch.value) is True @@ -389,7 +389,8 @@ def setup_and_teardown(): ) def test_batch_charge_identity_requires_a_cost_bearing_retrieval(call_type: str, cost: float, eligible: bool) -> None: from openai.types import Batch - from litellm.router_utils.batch_utils import batch_id_to_charge_once + + from litellm.litellm_core_utils.batch_utils import batch_id_to_charge_once batch: Final = Batch( id="batch-budget", completion_window="24h", created_at=1, endpoint="/v1/chat/completions", @@ -401,7 +402,7 @@ def test_batch_charge_identity_requires_a_cost_bearing_retrieval(call_type: str, def test_batch_charge_markers_keep_counters_and_batches_independent() -> None: from litellm.caching.in_memory_cache import InMemoryCache - from litellm.router_utils.batch_utils import batch_charged_once_marker_key + from litellm.litellm_core_utils.batch_utils import batch_charged_once_marker_key cache: Final = InMemoryCache() cache.set_cache(batch_charged_once_marker_key("provider-spend", "first-batch"), 1) diff --git a/tests/unit/proxy/management_endpoints/test_auto_router_endpoints.py b/tests/unit/proxy/management_endpoints/test_auto_router_endpoints.py index 748b291d5ff..c13e3d5812b 100644 --- a/tests/unit/proxy/management_endpoints/test_auto_router_endpoints.py +++ b/tests/unit/proxy/management_endpoints/test_auto_router_endpoints.py @@ -170,7 +170,7 @@ async def test_preview_and_serving_share_default_model_resolution( explicit_default: str | None, expected: str, ): - from litellm.router_utils.auto_router_model_naming import validate_complexity_router_config_write + from litellm.router_strategy.auto_router.auto_router_model_naming import validate_complexity_router_config_write from litellm.types.management_endpoints.auto_router_endpoints import ComplexityRouterConfigValidationRequest config: Final = { @@ -1240,7 +1240,7 @@ class TestAutoRouterBenchmarks: def test_the_listed_kinds_match_the_router_types_traffic_can_record(self): from typing import get_args, get_type_hints - from litellm.router_utils.auto_router_model_naming import StrategyRouterKind + from litellm.router_strategy.auto_router.auto_router_model_naming import StrategyRouterKind from litellm.types.utils import StandardLoggingRoutingDecision readonly_router_type: Final = get_type_hints(StandardLoggingRoutingDecision, include_extras=True)["router_type"] diff --git a/tests/unit/proxy/management_endpoints/test_model_management_endpoints.py b/tests/unit/proxy/management_endpoints/test_model_management_endpoints.py index fc6749aed0f..437d948f8f5 100644 --- a/tests/unit/proxy/management_endpoints/test_model_management_endpoints.py +++ b/tests/unit/proxy/management_endpoints/test_model_management_endpoints.py @@ -1,6 +1,6 @@ -import inspect import asyncio import contextlib +import inspect import json import socket from collections.abc import Iterator, Mapping @@ -1461,10 +1461,10 @@ class TestDeleteModelClearsRouterRegistry: add_deployment never restores, so an unguarded pop would make it permanently unroutable (the same cross-tenant DoS clear_cache was hardened against). """ + from litellm.proxy.management_endpoints.model_management_endpoints import ModelInfoDelete from litellm.proxy.management_endpoints.model_management_endpoints import ( delete_model as delete_model_endpoint, ) - from litellm.proxy.management_endpoints.model_management_endpoints import ModelInfoDelete model_id = "regular-del-1" admin_user = UserAPIKeyAuth(user_id="admin", user_role=LitellmUserRoles.PROXY_ADMIN) @@ -2739,11 +2739,11 @@ class TestAddAndDeleteModelLifecycle: - Delete same model again → raises (model not found) """ from litellm.proxy.management_endpoints.model_management_endpoints import ( + ModelInfoDelete, add_new_model, - delete_model as delete_model_endpoint, ) from litellm.proxy.management_endpoints.model_management_endpoints import ( - ModelInfoDelete, + delete_model as delete_model_endpoint, ) model_id = "lifecycle-test-model-123" @@ -2833,6 +2833,8 @@ class TestDeleteTeamBYOKModelGhost: async def test_delete_strips_public_name_and_refreshes_cache(self): from litellm.proxy.management_endpoints.model_management_endpoints import ( ModelInfoDelete, + ) + from litellm.proxy.management_endpoints.model_management_endpoints import ( delete_model as delete_model_endpoint, ) @@ -2916,6 +2918,8 @@ class TestDeleteTeamBYOKModelGhost: alias cleanup (delete_team_model_alias), preserving legacy behavior.""" from litellm.proxy.management_endpoints.model_management_endpoints import ( ModelInfoDelete, + ) + from litellm.proxy.management_endpoints.model_management_endpoints import ( delete_model as delete_model_endpoint, ) @@ -2984,6 +2988,8 @@ class TestDeleteTeamBYOKModelGhost: team.models when one replica is deleted but a sibling still backs it.""" from litellm.proxy.management_endpoints.model_management_endpoints import ( ModelInfoDelete, + ) + from litellm.proxy.management_endpoints.model_management_endpoints import ( delete_model as delete_model_endpoint, ) @@ -3065,6 +3071,8 @@ class TestDeleteTeamBYOKModelGhost: deployment still serves it.""" from litellm.proxy.management_endpoints.model_management_endpoints import ( ModelInfoDelete, + ) + from litellm.proxy.management_endpoints.model_management_endpoints import ( delete_model as delete_model_endpoint, ) @@ -3145,6 +3153,8 @@ class TestDeleteTeamBYOKModelGhost: the alias would break routing that works.""" from litellm.proxy.management_endpoints.model_management_endpoints import ( ModelInfoDelete, + ) + from litellm.proxy.management_endpoints.model_management_endpoints import ( delete_model as delete_model_endpoint, ) @@ -3252,6 +3262,8 @@ class TestDeleteModelTeamAuth: async def test_proxy_admin_can_delete_model_when_team_deleted(self): from litellm.proxy.management_endpoints.model_management_endpoints import ( ModelInfoDelete, + ) + from litellm.proxy.management_endpoints.model_management_endpoints import ( delete_model as delete_model_endpoint, ) @@ -3287,6 +3299,8 @@ class TestDeleteModelTeamAuth: """A missing team must never let a non-admin delete the orphan (no fail-open).""" from litellm.proxy.management_endpoints.model_management_endpoints import ( ModelInfoDelete, + ) + from litellm.proxy.management_endpoints.model_management_endpoints import ( delete_model as delete_model_endpoint, ) from litellm.proxy.proxy_server import ProxyException @@ -3322,6 +3336,8 @@ class TestDeleteModelTeamAuth: """The auth check must not add a redundant team query on the live-team path.""" from litellm.proxy.management_endpoints.model_management_endpoints import ( ModelInfoDelete, + ) + from litellm.proxy.management_endpoints.model_management_endpoints import ( delete_model as delete_model_endpoint, ) from litellm.proxy.proxy_server import ProxyException @@ -4392,7 +4408,6 @@ class TestModelInfoCostMapEchoFilter: def test_echoed_cost_map_metadata_is_not_persisted(self): import litellm - from litellm.proxy.management_endpoints.model_management_endpoints import update_db_model from litellm.types.router import Deployment, LiteLLM_Params, ModelInfo @@ -4418,7 +4433,6 @@ class TestModelInfoCostMapEchoFilter: def test_an_edited_value_survives_the_echo_filter(self): import litellm - from litellm.proxy.management_endpoints.model_management_endpoints import update_db_model from litellm.types.router import Deployment, LiteLLM_Params, ModelInfo @@ -4483,7 +4497,6 @@ class TestModelInfoCostMapEchoFilter: def test_metadata_without_a_cost_map_key_is_persisted(self): import litellm - from litellm.proxy.management_endpoints.model_management_endpoints import update_db_model from litellm.types.router import Deployment, LiteLLM_Params, ModelInfo from litellm.types.utils import echoed_cost_map_fields @@ -4513,7 +4526,6 @@ class TestModelInfoCostMapEchoFilter: def test_a_stored_mode_survives_an_echoed_save(self): import litellm - from litellm.proxy.management_endpoints.model_management_endpoints import update_db_model from litellm.types.router import Deployment, LiteLLM_Params, ModelInfo @@ -4536,7 +4548,6 @@ class TestModelInfoCostMapEchoFilter: def test_resetting_an_override_to_the_cost_map_value_removes_it(self): import litellm - from litellm.proxy.management_endpoints.model_management_endpoints import update_db_model from litellm.types.router import Deployment, LiteLLM_Params, ModelInfo @@ -4560,7 +4571,6 @@ class TestModelInfoCostMapEchoFilter: def test_reset_is_recognised_after_the_router_registered_the_override(self, monkeypatch: pytest.MonkeyPatch): import litellm - from litellm.proxy.management_endpoints.model_management_endpoints import update_db_model from litellm.types.router import Deployment, LiteLLM_Params, ModelInfo @@ -4586,7 +4596,6 @@ class TestModelInfoCostMapEchoFilter: from types import MappingProxyType import litellm - from litellm.proxy.management_endpoints.model_management_endpoints import update_db_model from litellm.types.router import Deployment, LiteLLM_Params, ModelInfo @@ -4611,7 +4620,6 @@ class TestModelInfoCostMapEchoFilter: def test_echo_is_compared_against_the_deployments_lookup_not_the_key(self): import litellm - from litellm.proxy.management_endpoints.model_management_endpoints import update_db_model from litellm.types.router import Deployment, LiteLLM_Params, ModelInfo @@ -4653,7 +4661,6 @@ class TestModelInfoCostMapEchoFilter: def test_base_model_wins_over_litellm_params_model_for_the_lookup(self): import litellm - from litellm.proxy.management_endpoints.model_management_endpoints import update_db_model from litellm.types.router import Deployment, LiteLLM_Params, ModelInfo @@ -4681,7 +4688,6 @@ class TestModelInfoCostMapEchoFilter: def test_encrypted_stored_model_is_decrypted_for_the_lookup(self, monkeypatch): import litellm - from litellm.proxy.management_endpoints.model_management_endpoints import update_db_model from litellm.types.router import Deployment, LiteLLM_Params, ModelInfo @@ -6431,7 +6437,7 @@ class TestStrategyRouterWriteValidation: AUTO_ROUTER_CAPABILITY_SLOT_LOCK_KEY, _auto_router_capability_slot, ) - from litellm.router_utils.auto_router_model_naming import gated_capabilities_of + from litellm.router_strategy.auto_router.auto_router_model_naming import gated_capabilities_of capabilities: Final = gated_capabilities_of(effective_params) @@ -6568,7 +6574,7 @@ class TestStrategyRouterWriteValidation: from fastapi import HTTPException from litellm.proxy.management_endpoints.model_management_endpoints import _auto_router_capability_slot - from litellm.router_utils.auto_router_tuning_baseline import snapshot_tuning_baselines + from litellm.router_strategy.auto_router.auto_router_tuning_baseline import snapshot_tuning_baselines configs = { "_TUNED_A": self._TUNED_A, @@ -6624,7 +6630,7 @@ class TestStrategyRouterWriteValidation: from litellm.proxy import proxy_server from litellm.proxy.management_endpoints.model_management_endpoints import _auto_router_capability_slot - from litellm.router_utils.auto_router_tuning_baseline import snapshot_tuning_baselines + from litellm.router_strategy.auto_router.auto_router_tuning_baseline import snapshot_tuning_baselines fake = self._FakeDb([]) fake.tx_obj.litellm_proxymodeltable.find_many = AsyncMock( @@ -6669,7 +6675,7 @@ class TestStrategyRouterWriteValidation: row id before the slot is entered, so a second tuned router is refused before its DB write.""" from litellm.proxy._types import ProxyException from litellm.proxy.management_endpoints.model_management_endpoints import add_new_model - from litellm.router_utils.auto_router_tuning_baseline import snapshot_tuning_baselines + from litellm.router_strategy.auto_router.auto_router_tuning_baseline import snapshot_tuning_baselines admin = UserAPIKeyAuth(user_id="admin", user_role=LitellmUserRoles.PROXY_ADMIN) baselines = snapshot_tuning_baselines([self._db_router_row("a", self._TUNED_A)]) @@ -8980,7 +8986,10 @@ class TestTeamMemberAutoRouterWrites: from fastapi import HTTPException from litellm.proxy.management_endpoints.model_management_endpoints import _auto_router_capability_slot - from litellm.proxy.management_helpers.auto_router_permissions import MemberAutoRouterWrite, validate_member_auto_router_config + from litellm.proxy.management_helpers.auto_router_permissions import ( + MemberAutoRouterWrite, + validate_member_auto_router_config, + ) row: Final = self._row() database: Final = self._database(self._team(), row) @@ -10677,6 +10686,8 @@ class TestFederationGateScopesToWhatTheWriteTouches: and the team that owns the model must be able to take it off their page.""" from litellm.proxy.management_endpoints.model_management_endpoints import ( ModelInfoDelete, + ) + from litellm.proxy.management_endpoints.model_management_endpoints import ( delete_model as delete_model_endpoint, ) diff --git a/tests/unit/proxy/management_helpers/test_auto_router_availability.py b/tests/unit/proxy/management_helpers/test_auto_router_availability.py index a88dcd7f66d..0eaae60b4ea 100644 --- a/tests/unit/proxy/management_helpers/test_auto_router_availability.py +++ b/tests/unit/proxy/management_helpers/test_auto_router_availability.py @@ -1,16 +1,15 @@ from collections.abc import Mapping -from typing import Final from types import SimpleNamespace - -from litellm.proxy.common_utils.encrypt_decrypt_utils import encrypt_value_helper +from typing import Final import pytest +from litellm.proxy.common_utils.encrypt_decrypt_utils import encrypt_value_helper from litellm.proxy.management_helpers.auto_router_availability import ( auto_router_availability, build_auto_router_catalog, ) -from litellm.router_utils.auto_router_tuning_baseline import snapshot_tuning_baselines +from litellm.router_strategy.auto_router.auto_router_tuning_baseline import snapshot_tuning_baselines def deployment( diff --git a/tests/unit/proxy/proxy_server/test_lifecycle.py b/tests/unit/proxy/proxy_server/test_lifecycle.py index 11a44a0ec16..8b1d7c1222a 100644 --- a/tests/unit/proxy/proxy_server/test_lifecycle.py +++ b/tests/unit/proxy/proxy_server/test_lifecycle.py @@ -33,9 +33,9 @@ from pydantic import BaseModel from typing_extensions import TypedDict import litellm.proxy.proxy_server as ps +from litellm.proxy.management_helpers.audit_logs import track_audit_task from litellm.proxy.proxy_server import ( ProxyStartupEvent, - initialize_shared_aiohttp_session, _resolve_pydantic_type, _resolve_typed_dict_type, cleanup_router_config_variables, @@ -43,6 +43,7 @@ from litellm.proxy.proxy_server import ( get_litellm_model_info, initialize, initialize_from_worker_config, + initialize_shared_aiohttp_session, load_from_azure_key_vault, proxy_shutdown_event, proxy_startup_event, @@ -50,8 +51,6 @@ from litellm.proxy.proxy_server import ( save_worker_config, ) -from litellm.proxy.management_helpers.audit_logs import track_audit_task - from .conftest import normalize # --------------------------------------------------------------------------- @@ -1165,7 +1164,7 @@ def test_proxy_startup_event_warns_for_global_budget_without_database(): @pytest.mark.asyncio async def test_tuning_baseline_v3_is_created_alongside_the_legacy_row(): - from litellm.router_utils.auto_router_tuning_baseline import DEFAULT_TUNING_FINGERPRINT + from litellm.router_strategy.auto_router.auto_router_tuning_baseline import DEFAULT_TUNING_FINGERPRINT prisma_client = MagicMock() prisma_client.db.litellm_config.find_unique = AsyncMock(return_value=None) @@ -1186,7 +1185,10 @@ async def test_tuning_baseline_v3_is_created_alongside_the_legacy_row(): @pytest.mark.asyncio async def test_scorer_baseline_upgrade_preserves_existing_routers_and_is_not_refreshed_on_restart(): - from litellm.router_utils.auto_router_tuning_baseline import mutable_tuned_identities, snapshot_tuning_baselines + from litellm.router_strategy.auto_router.auto_router_tuning_baseline import ( + mutable_tuned_identities, + snapshot_tuning_baselines, + ) deployments = [ { diff --git a/tests/unit/router_utils/test_auto_router_model_naming.py b/tests/unit/router_strategy/test_auto_router_model_naming.py similarity index 99% rename from tests/unit/router_utils/test_auto_router_model_naming.py rename to tests/unit/router_strategy/test_auto_router_model_naming.py index 1cfa3aa5c91..44fcfdaf5be 100644 --- a/tests/unit/router_utils/test_auto_router_model_naming.py +++ b/tests/unit/router_strategy/test_auto_router_model_naming.py @@ -3,9 +3,7 @@ from typing import Final import pytest -from litellm.router_strategy.complexity_router.fuse_presets import get_fuse_presets -from litellm.router_strategy.complexity_router.jev_classifier import DEFAULT_JEV_INSTRUCTIONS -from litellm.router_utils.auto_router_model_naming import ( +from litellm.router_strategy.auto_router.auto_router_model_naming import ( GATED_AUTO_ROUTER_CAPABILITIES, capability_limit_violation, carries_complexity_router_settings, @@ -18,6 +16,8 @@ from litellm.router_utils.auto_router_model_naming import ( validate_complexity_router_config_write, validate_strategy_router_model_write, ) +from litellm.router_strategy.complexity_router.fuse_presets import get_fuse_presets +from litellm.router_strategy.complexity_router.jev_classifier import DEFAULT_JEV_INSTRUCTIONS COMPLEXITY_FIELDS = frozenset({"complexity_router_config"}) SEMANTIC_FIELDS = frozenset({"auto_router_config", "auto_router_default_model", "auto_router_embedding_model"}) @@ -596,7 +596,7 @@ def test_custom_classifier_prompt_capability(config: Mapping[str, object], expec ], ) def test_is_complexity_router_model(model: str | None, expected: bool) -> None: - from litellm.router_utils.auto_router_model_naming import is_complexity_router_model + from litellm.router_strategy.auto_router.auto_router_model_naming import is_complexity_router_model assert is_complexity_router_model(model) is expected diff --git a/tests/unit/router_utils/test_auto_router_tuning_baseline.py b/tests/unit/router_strategy/test_auto_router_tuning_baseline.py similarity index 99% rename from tests/unit/router_utils/test_auto_router_tuning_baseline.py rename to tests/unit/router_strategy/test_auto_router_tuning_baseline.py index d0d98ce7042..4ee219a7784 100644 --- a/tests/unit/router_utils/test_auto_router_tuning_baseline.py +++ b/tests/unit/router_strategy/test_auto_router_tuning_baseline.py @@ -7,8 +7,7 @@ from typing import Final import pytest -from litellm.router_strategy.complexity_router.config import ComplexityRouterConfig -from litellm.router_utils.auto_router_tuning_baseline import ( +from litellm.router_strategy.auto_router.auto_router_tuning_baseline import ( DEFAULT_TUNING_FINGERPRINT, HEURISTIC_V1_TUNING_FIELDS, heuristic_v1_router_fingerprint, @@ -19,6 +18,7 @@ from litellm.router_utils.auto_router_tuning_baseline import ( tuning_limit_violation, tuning_quota_violation, ) +from litellm.router_strategy.complexity_router.config import ComplexityRouterConfig _TIERS = {"SIMPLE": "cheap", "MEDIUM": "mid", "COMPLEX": "strong", "REASONING": "top"} _ALT_TIERS = {**_TIERS, "COMPLEX": "other-strong"} diff --git a/tests/unit/router_utils/test_baseline_request.py b/tests/unit/router_strategy/test_baseline_request.py similarity index 95% rename from tests/unit/router_utils/test_baseline_request.py rename to tests/unit/router_strategy/test_baseline_request.py index 37d21597752..eb785a41d63 100644 --- a/tests/unit/router_utils/test_baseline_request.py +++ b/tests/unit/router_strategy/test_baseline_request.py @@ -1,6 +1,6 @@ from typing import Final -from litellm.router_utils.baseline_request import baseline_request, capture_baseline_parameters +from litellm.router_strategy.auto_router.baseline_request import baseline_request, capture_baseline_parameters def test_baseline_snapshot_owns_nested_caller_settings_and_overrides_routed_settings() -> None: diff --git a/tests/unit/router_strategy/test_complexity_router.py b/tests/unit/router_strategy/test_complexity_router.py index ca39c7f5c8f..018f93779a6 100644 --- a/tests/unit/router_strategy/test_complexity_router.py +++ b/tests/unit/router_strategy/test_complexity_router.py @@ -24,14 +24,6 @@ from pydantic import TypeAdapter, ValidationError import litellm from litellm import Router -from litellm.integrations.custom_logger import CustomLogger -from litellm.types.utils import StandardLoggingRoutingDecision -from litellm.router_utils.auto_router_model_naming import ( - CUSTOMIZATION_CAPABILITY, - GATED_AUTO_ROUTER_CAPABILITIES, - HEURISTIC_V2_CAPABILITY, - count_capability_routers, -) from litellm._logging import verbose_router_logger from litellm.caching.dual_cache import DualCache from litellm.caching.in_memory_cache import InMemoryCache @@ -42,7 +34,19 @@ from litellm.constants import ( RETURN_RAW_MODEL_NAME_METADATA_KEY, SESSION_ID_GENERATED_METADATA_KEY, ) +from litellm.integrations.custom_logger import CustomLogger +from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler from litellm.router import as_output_cap, stamped_client_ceiling +from litellm.router_strategy.auto_router.auto_router_model_naming import ( + CUSTOMIZATION_CAPABILITY, + GATED_AUTO_ROUTER_CAPABILITIES, + HEURISTIC_V2_CAPABILITY, + count_capability_routers, +) +from litellm.router_strategy.complexity_router.capability_classifier import ( + CAPABILITY_CLASSIFIER_SYSTEM_PROMPT, + CapabilityClassifierVerdict, +) from litellm.router_strategy.complexity_router.complexity_router import ( _CLASSIFICATION_CURRENT_MESSAGE_ONLY, _CLASSIFICATION_WITH_CONVERSATION, @@ -58,18 +62,14 @@ from litellm.router_strategy.complexity_router.complexity_router import ( classification_system_prompt, custom_tier_classification_prompt, ) -from litellm.router_strategy.complexity_router.capability_classifier import ( - CAPABILITY_CLASSIFIER_SYSTEM_PROMPT, - CapabilityClassifierVerdict, -) from litellm.router_strategy.complexity_router.config import ( - CapabilityCalibrationConfig, - CapabilityClassifierConfig, DEFAULT_CLASSIFICATION_RUBRIC, DEFAULT_CLASSIFIER_CONTEXT_WINDOW_SIZE, DEFAULT_COMPLEXITY_CONFIG, DEFAULT_TECHNICAL_KEYWORDS, TIER_SEVERITY_ORDER, + CapabilityCalibrationConfig, + CapabilityClassifierConfig, ClassificationRubric, ClassifierLLMConfig, ComplexityRouterConfig, @@ -83,10 +83,12 @@ from litellm.router_strategy.complexity_router.jev_classifier import ( JevUsage, ) from litellm.router_strategy.complexity_router.llm_v2 import LLM_V2_PROMPT_VERSION +from litellm.router_strategy.complexity_router.request_models import RequestComplexityRouterConfig from litellm.router_strategy.complexity_router.tier_predictor import ( TierGlobalStatistic, TrainedTierArtifact, ) +from litellm.types.llms.openai import ResponsesAPIResponse from litellm.types.router import ( Deployment, LiteLLM_Params, @@ -94,10 +96,7 @@ from litellm.types.router import ( RouterErrors, TaggedPreRoutingStrategy, ) -from litellm.types.llms.openai import ResponsesAPIResponse -from litellm.router_strategy.complexity_router.request_models import RequestComplexityRouterConfig -from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler - +from litellm.types.utils import StandardLoggingRoutingDecision requires_semantic_router = pytest.mark.skipif( sys.version_info >= (3, 14), reason="The semantic-router extra excludes Python 3.14" @@ -16299,7 +16298,6 @@ class TestTierHealthFailover: tests can prove the probe passes a kwargs copy and forwards the prompt arguments. """ import litellm as litellm_module - from litellm.types.router import RouterRateLimitError probed_kwargs = [] diff --git a/tests/unit/router_strategy/test_llm_v2.py b/tests/unit/router_strategy/test_llm_v2.py index 3fd2e8808e7..8c1e9e48661 100644 --- a/tests/unit/router_strategy/test_llm_v2.py +++ b/tests/unit/router_strategy/test_llm_v2.py @@ -4,11 +4,12 @@ from typing import Final from unittest.mock import AsyncMock, MagicMock import pytest -import litellm from pydantic import ValidationError +import litellm from litellm import ModelResponse, Router from litellm.caching.dual_cache import DualCache +from litellm.router_strategy.auto_router.auto_router_model_naming import strategy_router_dependencies from litellm.router_strategy.complexity_router.complexity_router import ComplexityRouter from litellm.router_strategy.complexity_router.config import ComplexityRouterConfig, ComplexityTier from litellm.router_strategy.complexity_router.fuse_presets import get_fuse_presets @@ -21,7 +22,6 @@ from litellm.router_strategy.complexity_router.llm_v2 import ( LLMV2Verdict, llm_v2_response_format, ) -from litellm.router_utils.auto_router_model_naming import strategy_router_dependencies from litellm.types.llms.openai import ResponsesAPIResponse diff --git a/tests/unit/test_private_usage_aliases.py b/tests/unit/test_private_usage_aliases.py index 1d9c2e87100..d0c5c8247d3 100644 --- a/tests/unit/test_private_usage_aliases.py +++ b/tests/unit/test_private_usage_aliases.py @@ -1969,7 +1969,7 @@ ALIAS_CASES: Final = ( ("litellm.router_utils.add_retry_fallback_headers", "HiddenParamsAsyncIteratorWrapper", "_inner", "inner", True), ("litellm.router_utils.add_retry_fallback_headers", "", "_HiddenParamsHost", "HiddenParamsHost", False), ( - "litellm.router_utils.batch_utils", + "litellm.litellm_core_utils.batch_utils", "", "_get_router_metadata_variable_name", "get_router_metadata_variable_name", diff --git a/tests/unit/test_router/test_router_helper_utils.py b/tests/unit/test_router/test_router_helper_utils.py index af8e7f53462..513136ae150 100644 --- a/tests/unit/test_router/test_router_helper_utils.py +++ b/tests/unit/test_router/test_router_helper_utils.py @@ -4,20 +4,21 @@ import gc import json import os import weakref -from datetime import datetime, timezone from collections.abc import Iterator, Mapping +from datetime import datetime, timezone from types import MethodType from typing import Final from unittest.mock import AsyncMock, MagicMock, patch -import litellm import pytest + +import litellm from litellm import Router from litellm.caching.dual_cache import DualCache from litellm.caching.in_memory_cache import InMemoryCache from litellm.constants import DEFAULT_AUTO_ROUTER_MAX_INPUT_CHARS, ROUTER_USAGE_COUNTED_TOKENS_METADATA_KEY -from litellm.integrations.SlackAlerting.slack_alerting import SlackAlerting from litellm.integrations.custom_logger import CustomLogger +from litellm.integrations.SlackAlerting.slack_alerting import SlackAlerting from litellm.litellm_core_utils.streaming_handler import CustomStreamWrapper from litellm.router_strategy.adaptive_router.hooks import AdaptiveRouterPostCallHook from litellm.types.litellm_params import RoutingStrategyName @@ -801,10 +802,11 @@ async def test_deployment_callback_on_failure(model_list): def test_deployment_callback_respects_cooldown_time(model_list): """Ensure per-model cooldown_time is honored even when exception headers are present.""" - import httpx import time from unittest.mock import patch + import httpx + router = Router(model_list=model_list) class FakeException(Exception): @@ -1811,8 +1813,8 @@ def test_get_allowed_fails_from_policy(model_list, exception_type, exception_nam def test_initialize_alerting(model_list): """Test if the 'initialize_alerting' function is working correctly""" - from litellm.types.router import AlertingConfig from litellm.integrations.SlackAlerting.slack_alerting import SlackAlerting + from litellm.types.router import AlertingConfig router = Router( model_list=model_list, alerting_config=AlertingConfig(webhook_url="test") @@ -2245,6 +2247,7 @@ def test_deployments_by_pattern(model_list): ) def test_pattern_match_deployment_set_model_name(user_request_model, model_name, litellm_model, expected_model): from re import Match + from litellm.router_utils.pattern_match_deployments import PatternMatchRouter pattern_router = PatternMatchRouter() @@ -2702,7 +2705,7 @@ def test_handle_clientside_credential_metadata_loading(model_list, function_name ) def test_handle_clientside_credential_metadata_variable_name(model_list, function_name, metadata_key): """Test that _handle_clientside_credential uses the correct metadata variable name based on function name""" - from litellm.router_utils.batch_utils import get_router_metadata_variable_name + from litellm.litellm_core_utils.batch_utils import get_router_metadata_variable_name router = Router(model_list=model_list) diff --git a/tests/unit/test_router/test_router_surface_check.py b/tests/unit/test_router/test_router_surface_check.py new file mode 100644 index 00000000000..a97fd029380 --- /dev/null +++ b/tests/unit/test_router/test_router_surface_check.py @@ -0,0 +1,58 @@ +""" +The router surface fence (tests/code_coverage_tests/router_surface_check.py) must see every way +outside code reaches a Router member, or new dependencies slip past the shrink-only allowlist. +""" + +import ast +import importlib.util +import pathlib + +import pytest + +_CHECK = pathlib.Path(__file__).parents[2] / "code_coverage_tests" / "router_surface_check.py" + + +@pytest.fixture(scope="module") +def surface_check(): + spec = importlib.util.spec_from_file_location("router_surface_check", _CHECK) + assert spec is not None and spec.loader is not None + module = importlib.util.module_from_spec(spec) + spec.loader.exec_module(module) + return module + + +def _uses(surface_check, src: str) -> set: + return surface_check.member_uses(ast.parse(src), surface_check.router_members()) + + +def test_instance_and_dynamic_access_are_uses(surface_check): + assert _uses(surface_check, "llm_router.get_model_list()\ngetattr(router, 'model_list')") == { + "get_model_list", + "model_list", + } + + +def test_class_level_calls_are_uses(surface_check): + assert _uses(surface_check, "litellm.Router.get_valid_args()") == {"get_valid_args"} + assert _uses(surface_check, "from litellm import Router\nRouter.get_valid_args()") == {"get_valid_args"} + + +def test_container_file_endpoints_registered_by_name_are_members(surface_check): + members = surface_check.router_members() + assert {"aupload_container_file", "upload_container_file", "alist_container_files"} <= members + assert _uses(surface_check, "await llm_router.aupload_container_file(file=f)") == {"aupload_container_file"} + + +def test_unlisted_class_level_use_fails_the_fence(surface_check): + found = { + "members": {"outside": {"get_valid_args": {"litellm/new_module.py"}}, "router_family": {}}, + "router_utils_imports": {}, + } + allow = {"members_outside": [], "members_router_family": [], "router_utils_imports_outside": []} + errors = surface_check.compare(allow, found) + assert len(errors) == 1 + assert "'get_valid_args' in litellm/new_module.py" in errors[0] + + +def test_unrelated_receivers_are_not_uses(surface_check): + assert _uses(surface_check, "client.get_model_list()\nRouterConfig.model_list") == set() From 49ccce7813725cc87b64d1b0f1b097dca5c681af Mon Sep 17 00:00:00 2001 From: paveln83 <349044532+paveln83@users.noreply.github.com> Date: Wed, 28 Oct 2026 17:12:07 -0700 Subject: [PATCH 32/41] test(extraction): audit #47473 (#48898) * test(extraction): audit #47473 * test(extraction): check over-budget extraction never reaches Cloudflare --------- Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com> --- .../management/test_model_health_check.py | 37 ++ ...test_cloudflare_clairic_extraction_wire.py | 418 ++++++++++++++++++ 2 files changed, 455 insertions(+) create mode 100644 tests/integration/providers/test_cloudflare_clairic_extraction_wire.py diff --git a/tests/integration/management/test_model_health_check.py b/tests/integration/management/test_model_health_check.py index 57a05daca94..992c1e8e875 100644 --- a/tests/integration/management/test_model_health_check.py +++ b/tests/integration/management/test_model_health_check.py @@ -26,6 +26,20 @@ _DECISIONS_PROBE_REPLY: Final = JsonResponse( "usage": {"input_tokens": 10, "output_tokens": 1}, }, ) +_EXTRACTION_PROBE_REPLY: Final = JsonResponse( + content_type="application/json", + body={ + "result": { + "model": "clairic", + "answers": {"reachable": {"type": "text", "value": "health", "confidence": 0.9}}, + "truncated": False, + "usage": {"input_tokens": 10, "output_tokens": 0}, + }, + "success": True, + "errors": [], + "messages": [], + }, +) _CONFIGURED_PROBE_REPLY: Final = JsonResponse( content_type="application/json", body={ @@ -287,6 +301,29 @@ def test_decisions_mode_health_check_resolves_the_mode_from_the_cost_map_and_sen ] +def test_extraction_mode_health_check_resolves_the_mode_from_the_cost_map_and_sends_the_extraction_probe( + gateway: Gateway, +) -> None: + with gateway.scenario() as scenario: + handle: Final = register_scenario(f"health-extraction-{uuid.uuid4().hex[:12]}", _EXTRACTION_PROBE_REPLY) + scenario.cleanups.callback(delete_scenario, handle) + model: Final = scenario.model( + model="cloudflare/@cf/cloudflare/clairic", api_base=handle.api_base(), api_key="synthetic-cloudflare-key" + ) + report: Final = _health_report(gateway, model) + assert (report["healthy_count"], report["unhealthy_count"]) == (1, 0), report + assert _probes_sent_to(gateway, handle) == [ + ( + f"/{handle.scenario_id}/ai/run/@cf/cloudflare/clairic", + { + "model": "clairic", + "state": os.environ.get("DEFAULT_HEALTH_CHECK_PROMPT", "test from litellm"), + "fields": {"reachable": {"type": "text", "instructions": "Any word from the text"}}, + }, + ) + ] + + @pytest.mark.parametrize("mode", ("decisions", "evaluation"), ids=("decisions", "legacy-evaluation")) def test_decisions_mode_health_check_sends_the_configured_state_and_questions(gateway: Gateway, mode: str) -> None: with gateway.scenario() as scenario: diff --git a/tests/integration/providers/test_cloudflare_clairic_extraction_wire.py b/tests/integration/providers/test_cloudflare_clairic_extraction_wire.py new file mode 100644 index 00000000000..7b2a48c9676 --- /dev/null +++ b/tests/integration/providers/test_cloudflare_clairic_extraction_wire.py @@ -0,0 +1,418 @@ +import json +import math +import re +import uuid +from collections.abc import Iterator, Mapping +from dataclasses import dataclass +from pathlib import Path +from typing import Final, Literal + +import httpx +import pytest +import yaml +from integration._support.client import JSON_OBJECT, Gateway, eventually, object_value +from integration._support.database import read_rows +from integration._support.process import OwnedProxy, WorkerHealthcheck, graceful_stop_seconds, owned_proxy_process +from integration._support.wire import Reply, Request, Wire, wire_server +from pydantic import JsonValue, TypeAdapter + +Route = Literal["/v1/extraction", "/extraction"] + +_API_KEY: Final = "synthetic-cloudflare-key" +_CLAIRIC: Final = "cloudflare/@cf/cloudflare/clairic" +_CLAIRIC_PATH: Final = "/ai/run/@cf/cloudflare/clairic" +_INPUT_TOKENS: Final = 412 +_EMAIL: Final = "billing@acmetools.example" +_SCENARIO: Final = re.compile(r"scenario-([a-z-]+)") +_JSON_ROWS: Final = TypeAdapter(list[dict[str, JsonValue]]) +_FIELDS: Final[dict[str, JsonValue]] = { + "vendor": {"type": "text", "instructions": "Who sent the invoice?"}, + "total": {"type": "number", "unit": "USD"}, +} +_SPEND_QUERY: Final = ( + 'SELECT spend, status, call_type, model_group, prompt_tokens, completion_tokens FROM "LiteLLM_SpendLogs" ' + "WHERE request_id = %s" +) +_OWNED_TIMEOUT: Final = int(2 * graceful_stop_seconds() + 120) +_REQUEST_GUARDRAIL: Final = "extraction-mask-email-in" +_RESPONSE_GUARDRAIL: Final = "extraction-mask-email-out" +_OWNED_MODEL: Final = "invoice-reader" + + +def _state(marker: str, scenario: str = "plain") -> str: + return f"Invoice from Acme Tools, total $174. scenario-{scenario} marker-{marker}" + + +def _vendor_span(state: str) -> list[JsonValue]: + start: Final = state.index("Acme Tools") + return [start, start + len("Acme Tools")] + + +def _email_span(state: str) -> list[JsonValue]: + start: Final = state.index(_EMAIL) + return [start, start + len(_EMAIL)] + + +def _answers(state: str) -> dict[str, JsonValue]: + return { + "vendor": {"type": "text", "value": "Acme Tools", "confidence": 0.97, "span": _vendor_span(state)}, + "total": {"type": "number", "value": 174, "confidence": 0.95}, + } + + +def _envelope(result: Mapping[str, JsonValue]) -> Reply: + return Reply(body=json.dumps({"result": result, "success": True, "errors": [], "messages": []}).encode()) + + +def _clairic_reply(request: Request) -> Reply: + state: Final = JSON_OBJECT.validate_json(request.body)["state"] + assert isinstance(state, str), request.body + return _envelope( + { + "model": "clairic", + "answers": _answers(state), + "truncated": False, + "usage": {"input_tokens": _INPUT_TOKENS, "output_tokens": 0}, + } + ) + + +def _expected_response(model: str, state: str) -> dict[str, JsonValue]: + return { + "model": model, + "answers": _answers(state), + "truncated": False, + "usage": {"input_tokens": _INPUT_TOKENS, "output_tokens": 0}, + } + + +def _upstream_body(request: Request) -> dict[str, JsonValue]: + assert (request.method, request.target) == ("POST", _CLAIRIC_PATH), (request.method, request.target) + assert request.headers["authorization"] == f"Bearer {_API_KEY}", request.headers + return JSON_OBJECT.validate_json(request.body) + + +def _body(model: str, marker: str, **extra: JsonValue) -> dict[str, JsonValue]: + return {"model": model, "state": _state(marker), "fields": _FIELDS, **extra} + + +def _extract( + gateway: Gateway, + model: str, + marker: str, + *, + route: Route = "/v1/extraction", + key: str | None = None, + **extra: JsonValue, +) -> httpx.Response: + return gateway.request("POST", route, _body(model, marker, **extra), key=key) + + +def _number(value: JsonValue) -> float: + assert isinstance(value, (int, float)) and not isinstance(value, bool), value + return float(value) + + +def _expected_spend() -> float: + cost_map: Final = JSON_OBJECT.validate_json(Path("model_prices_and_context_window.json").read_bytes()) + priced: Final = cost_map.get(_CLAIRIC) + assert priced is not None, f"{_CLAIRIC} has no cost map entry" + return _INPUT_TOKENS * _number(object_value(priced)["input_cost_per_token"]) + + +def _spend_row(call_id: str) -> dict[str, JsonValue]: + rows: Final = eventually(lambda: read_rows(_SPEND_QUERY, (call_id,)), lambda found: len(found) == 1, seconds=70) + return rows[0] + + +def _assert_billed(gateway: Gateway, response: httpx.Response, model: str) -> None: + expected: Final = _expected_spend() + assert math.isclose(float(response.headers["x-litellm-response-cost"]), expected, rel_tol=1e-9), response.headers + call_id: Final = response.headers["x-litellm-call-id"] + row: Final = _spend_row(call_id) + assert (row["status"], row["call_type"], row["model_group"], row["prompt_tokens"], row["completion_tokens"]) == ( + "success", + "aextraction", + model, + _INPUT_TOKENS, + 0, + ), row + assert math.isclose(_number(row["spend"]), expected, rel_tol=1e-9), row + listed: Final = gateway.request("GET", "/spend/logs", params={"request_id": call_id}) + assert listed.status_code == 200, listed.text + (reported,) = _JSON_ROWS.validate_json(listed.content) + assert (reported["request_id"], reported["model_group"]) == (call_id, model), reported + assert math.isclose(_number(reported["spend"]), expected, rel_tol=1e-9), reported + + +@pytest.mark.parametrize("route", ("/v1/extraction", "/extraction")) +def test_extraction_reaches_clairic_on_both_routes_and_is_billed(gateway: Gateway, route: Route) -> None: + marker: Final = uuid.uuid4().hex + with wire_server(_clairic_reply) as wire, gateway.scenario() as scenario: + model: Final = scenario.model(model=_CLAIRIC, api_base=wire.url, api_key=_API_KEY) + response: Final = _extract(gateway, model, marker, route=route, metadata={"tags": [f"audit-{marker}"]}) + assert response.status_code == 200, response.text + assert response.json() == _expected_response(model, _state(marker)) + (request,) = wire.drain() + assert _upstream_body(request) == {"model": "clairic", "state": _state(marker), "fields": _FIELDS} + _assert_billed(gateway, response, model) + + +_INVALID_BODIES: Final[tuple[tuple[str, dict[str, JsonValue]], ...]] = ( + ("unknown_type", {"fields": {"vendor": {"type": "paragraph"}}}), + ("list_without_items", {"fields": {"items": {"type": "list"}}}), + ("remote_image", {"images": ["https://images.example/receipt.png"]}), +) + + +@pytest.mark.parametrize(("label", "change"), _INVALID_BODIES, ids=[label for label, _ in _INVALID_BODIES]) +def test_invalid_extraction_bodies_are_refused_before_any_upstream_call( + gateway: Gateway, label: str, change: dict[str, JsonValue] +) -> None: + marker: Final = uuid.uuid4().hex + with wire_server(_clairic_reply) as wire, gateway.scenario() as scenario: + model: Final = scenario.model(model=_CLAIRIC, api_base=wire.url, api_key=_API_KEY) + response: Final = gateway.request("POST", "/v1/extraction", {**_body(model, marker), **change}) + assert response.status_code == 400, (label, response.text) + assert "Invalid Extraction request" in response.text, (label, response.text) + assert wire.drain() == (), label + + +@pytest.mark.parametrize(("param", "value"), (("temperature", 0.2), ("stream", True)), ids=("temperature", "stream")) +def test_unsupported_top_level_params_are_refused_unless_the_deployment_drops_params( + gateway: Gateway, param: str, value: JsonValue +) -> None: + marker: Final = uuid.uuid4().hex + with wire_server(_clairic_reply) as wire, gateway.scenario() as scenario: + strict: Final = scenario.model(model=_CLAIRIC, api_base=wire.url, api_key=_API_KEY) + dropping: Final = scenario.model(model=_CLAIRIC, api_base=wire.url, api_key=_API_KEY, drop_params=True) + refused: Final = _extract(gateway, strict, marker, **{param: value}) + assert refused.status_code == 400, refused.text + assert param in refused.text, refused.text + assert wire.drain() == () + dropped: Final = _extract(gateway, dropping, marker, **{param: value}) + assert dropped.status_code == 200, dropped.text + assert dropped.headers["content-type"].startswith("application/json"), dropped.headers + assert dropped.json() == _expected_response(dropping, _state(marker)) + (request,) = wire.drain() + assert _upstream_body(request) == {"model": "clairic", "state": _state(marker), "fields": _FIELDS} + _assert_billed(gateway, dropped, dropping) + + +def test_key_checks_on_extraction_match_chat(gateway: Gateway) -> None: + marker: Final = uuid.uuid4().hex + with wire_server(_clairic_reply) as wire, gateway.scenario() as scenario: + model: Final = scenario.model(model=_CLAIRIC, api_base=wire.url, api_key=_API_KEY) + anonymous: Final = gateway.client.post("/v1/extraction", json=_body(model, marker)) + assert anonymous.status_code == 401, anonymous.text + restricted: Final = scenario.key(models=[f"other-{uuid.uuid4().hex}"]) + refused: Final = _extract(gateway, model, marker, key=restricted) + chat_refused: Final = gateway.request( + "POST", + "/v1/chat/completions", + {"model": model, "messages": [{"role": "user", "content": marker}]}, + key=restricted, + ) + assert 400 <= refused.status_code < 500, refused.text + assert refused.status_code == chat_refused.status_code, (refused.text, chat_refused.text) + assert wire.drain() == () + allowed: Final = scenario.key(models=[model], max_budget=1e-06) + first: Final = _extract(gateway, model, marker, key=allowed) + assert first.status_code == 200, first.text + assert len(wire.drain()) == 1 + blocked, upstream_calls = eventually( + lambda: (_extract(gateway, model, uuid.uuid4().hex, key=allowed), len(wire.drain())), + lambda answer: answer[0].status_code != 200, + seconds=70, + ) + chat_blocked: Final = gateway.request( + "POST", + "/v1/chat/completions", + {"model": model, "messages": [{"role": "user", "content": marker}]}, + key=allowed, + ) + assert 400 <= blocked.status_code < 500, blocked.text + assert blocked.status_code == chat_blocked.status_code, (blocked.text, chat_blocked.text) + assert upstream_calls == 0 + assert wire.drain() == () + + +def test_a_small_key_tpm_limit_serves_extraction_and_counts_its_input_tokens(gateway: Gateway) -> None: + with wire_server(_clairic_reply) as wire, gateway.scenario() as scenario: + model: Final = scenario.model(model=_CLAIRIC, api_base=wire.url, api_key=_API_KEY) + key: Final = scenario.key(models=[model], tpm_limit=100) + first: Final = _extract(gateway, model, uuid.uuid4().hex, key=key) + assert first.status_code == 200, first.text + assert len(wire.drain()) == 1 + limited, upstream_calls = eventually( + lambda: (_extract(gateway, model, uuid.uuid4().hex, key=key), len(wire.drain())), + lambda answer: answer[0].status_code != 200, + seconds=70, + ) + assert limited.status_code == 429, limited.text + assert upstream_calls == 0 + + +def test_cloudflare_errors_keep_their_status_and_log_an_unbilled_failure(gateway: Gateway) -> None: + marker: Final = uuid.uuid4().hex + refusal: Final = json.dumps( + { + "result": None, + "success": False, + "errors": [{"code": 5006, "message": f"min above max {marker}"}], + "messages": [], + } + ).encode() + with wire_server(lambda _: Reply(status=422, body=refusal)) as wire, gateway.scenario() as scenario: + model: Final = scenario.model(model=_CLAIRIC, api_base=wire.url, api_key=_API_KEY) + response: Final = _extract(gateway, model, marker) + assert response.status_code == 422, response.text + assert f"min above max {marker}" in response.text, response.text + assert len(wire.drain()) >= 1 + row: Final = _spend_row(response.headers["x-litellm-call-id"]) + assert (row["status"], row["call_type"], row["model_group"], _number(row["spend"])) == ( + "failure", + "aextraction", + model, + 0.0, + ), row + + +def _owned_reply(request: Request) -> Reply: + body: Final = JSON_OBJECT.validate_json(request.body) + state: Final = body["state"] + assert isinstance(state, str), request.body + scenario: Final = _SCENARIO.search(state) + assert scenario is not None, state + if scenario.group(1) != "mask-out": + return _clairic_reply(request) + return _envelope( + { + "model": "clairic", + "answers": { + "vendor": {"type": "text", "value": "Acme Tools", "confidence": 0.97, "span": _vendor_span(state)}, + "contact_email": {"type": "text", "value": _EMAIL, "confidence": 0.9, "span": _email_span(state)}, + }, + "truncated": False, + "state_text": state, + "usage": {"input_tokens": _INPUT_TOKENS, "output_tokens": 0}, + } + ) + + +def _owned_config(upstream_url: str) -> dict[str, JsonValue]: + base: Final = JSON_OBJECT.validate_python(yaml.safe_load(Path("tests/integration/proxy_config.yaml").read_text())) + email_filter: Final[list[JsonValue]] = [{"pattern_type": "prebuilt", "pattern_name": "email", "action": "MASK"}] + return { + **base, + "guardrails": [ + { + "guardrail_name": _REQUEST_GUARDRAIL, + "litellm_params": {"guardrail": "litellm_content_filter", "mode": "pre_call", "patterns": email_filter}, + }, + { + "guardrail_name": _RESPONSE_GUARDRAIL, + "litellm_params": { + "guardrail": "litellm_content_filter", + "mode": "post_call", + "patterns": email_filter, + }, + }, + ], + "model_list": [ + { + "model_name": _OWNED_MODEL, + "litellm_params": {"model": _CLAIRIC, "api_base": upstream_url, "api_key": _API_KEY}, + } + ], + } + + +@dataclass(frozen=True, slots=True) +class Rig: + proxy: OwnedProxy + upstream: Wire + + +@pytest.fixture(scope="module") +def rig(session_gateway: Gateway, tmp_path_factory: pytest.TempPathFactory) -> Iterator[Rig]: + with wire_server(_owned_reply) as upstream: + directory: Final = tmp_path_factory.mktemp("clairic-extraction-owned") + config_path: Final = directory / "clairic-extraction.yaml" + config_path.write_text(yaml.safe_dump(_owned_config(upstream.url))) + with owned_proxy_process( + session_gateway, + directory, + {}, + config=config_path, + workers=2, + extra_arguments=("--max_tokens", "256", "--temperature", "0.2"), + healthcheck=WorkerHealthcheck.LENIENT, + ) as proxy: + yield Rig(proxy, upstream) + + +def _owned_extract(rig: Rig, state: str, **extra: JsonValue) -> httpx.Response: + return rig.proxy.gateway.request( + "POST", "/v1/extraction", {"model": _OWNED_MODEL, "state": state, "fields": _FIELDS, **extra} + ) + + +def _applied_guardrails(response: httpx.Response) -> frozenset[str]: + return frozenset(name.strip() for name in response.headers.get("x-litellm-applied-guardrails", "").split(",")) + + +@pytest.mark.timeout(_OWNED_TIMEOUT) +def test_cli_generation_defaults_never_reach_an_extraction_call(rig: Rig) -> None: + state: Final = _state(uuid.uuid4().hex) + response: Final = _owned_extract(rig, state) + assert response.status_code == 200, response.text + assert response.json() == _expected_response(_OWNED_MODEL, state) + (request,) = rig.upstream.drain() + assert _upstream_body(request) == {"model": "clairic", "state": state, "fields": _FIELDS} + + +@pytest.mark.timeout(_OWNED_TIMEOUT) +def test_a_request_guardrail_masks_the_state_before_clairic_and_drops_spans_into_it(rig: Rig) -> None: + state: Final = f"{_state(uuid.uuid4().hex, 'mask-in')} Questions go to {_EMAIL}." + response: Final = _owned_extract(rig, state, guardrails=[_REQUEST_GUARDRAIL]) + assert response.status_code == 200, response.text + assert _REQUEST_GUARDRAIL in _applied_guardrails(response), response.headers + (request,) = rig.upstream.drain() + sent: Final = _upstream_body(request) + sent_state: Final = sent["state"] + assert isinstance(sent_state, str) and _EMAIL not in sent_state, sent + assert sent_state.startswith(state.removesuffix(f"{_EMAIL}.")) and "REDACTED" in sent_state, sent + assert sent == {"model": "clairic", "state": sent_state, "fields": _FIELDS} + answers: Final = object_value(response.json()["answers"]) + assert object_value(answers["vendor"]) == {"type": "text", "value": "Acme Tools", "confidence": 0.97}, answers + + +@pytest.mark.timeout(_OWNED_TIMEOUT) +def test_a_response_guardrail_masks_text_answers_and_state_text(rig: Rig) -> None: + state: Final = f"{_state(uuid.uuid4().hex, 'mask-out')} Questions go to {_EMAIL}." + fields: Final[dict[str, JsonValue]] = {**_FIELDS, "contact_email": {"type": "text"}} + response: Final = rig.proxy.gateway.request( + "POST", + "/v1/extraction", + { + "model": _OWNED_MODEL, + "state": state, + "fields": fields, + "guardrails": [_RESPONSE_GUARDRAIL], + "options": {"return_state_text": True}, + }, + ) + assert response.status_code == 200, response.text + assert _RESPONSE_GUARDRAIL in _applied_guardrails(response), response.headers + (request,) = rig.upstream.drain() + assert _upstream_body(request)["state"] == state + body: Final = response.json() + state_text: Final = body["state_text"] + assert isinstance(state_text, str) and _EMAIL not in state_text and "REDACTED" in state_text, body + answers: Final = object_value(body["answers"]) + contact: Final = object_value(answers["contact_email"]) + assert "span" not in contact, answers + masked: Final = contact["value"] + assert isinstance(masked, str) and _EMAIL not in masked and "REDACTED" in masked, answers + assert object_value(answers["vendor"]) == {"type": "text", "value": "Acme Tools", "confidence": 0.97}, answers From 93cfc5265eb4df6e8aa463913c684c08728b5561 Mon Sep 17 00:00:00 2001 From: yuneng-jiang Date: Wed, 28 Oct 2026 17:44:48 -0700 Subject: [PATCH 33/41] ci: run zizmor with --no-ignores and move the spend logs and cost map guards to pull_request (#48947) * ci(zizmor): run zizmor 1.31.1 without ignores and own the workflow zizmor-action has no input for --no-ignores, so the workflow now installs the zizmor 1.31.1 release binary directly, pinned by SHA-256 like grype in image-scan.yml, and runs it with --no-ignores and --no-config. Inline `zizmor: ignore` comments and config-file ignores no longer hide findings. Behavior otherwise matches the old action defaults (regular persona, online audits with the job token, medium and above, GitHub annotations) Adds a CODEOWNERS entry for .github/workflows/zizmor.yml * ci: run spend-logs-guard and cost-map-guard on pull_request Both guards ran on pull_request_target behind a zizmor ignore comment. They now run on pull_request and check out the base commit by sha, so the guard scripts still come from the base branch while the PR's head is only fetched and diffed. The zizmor ignore comments are gone --- .github/CODEOWNERS | 1 + .github/workflows/cost-map-guard.yml | 5 +++-- .github/workflows/spend-logs-guard.yml | 6 +++--- .github/workflows/zizmor.yml | 17 +++++++++++------ 4 files changed, 18 insertions(+), 11 deletions(-) diff --git a/.github/CODEOWNERS b/.github/CODEOWNERS index f6c06f7d83f..158110cbbab 100644 --- a/.github/CODEOWNERS +++ b/.github/CODEOWNERS @@ -2,3 +2,4 @@ /litellm/model_prices_and_context_window_backup.json @mateo-berri @ryan-crabbe-berri @kerry-berri @joshua-berri /helm/ @yassin-berriai /terraform/ @yassin-berriai +/.github/workflows/zizmor.yml @yuneng-berri diff --git a/.github/workflows/cost-map-guard.yml b/.github/workflows/cost-map-guard.yml index f9172b9551a..05b22017cbb 100644 --- a/.github/workflows/cost-map-guard.yml +++ b/.github/workflows/cost-map-guard.yml @@ -1,7 +1,7 @@ name: Cost map guard -on: # zizmor: ignore[dangerous-triggers] runs the base branch's code only; the PR's cost map files are read as data and never executed - pull_request_target: +on: + pull_request: branches: - main - "litellm_**" @@ -19,6 +19,7 @@ jobs: steps: - uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955 # v4.3.0 with: + ref: ${{ github.event.pull_request.base.sha }} persist-credentials: false - name: Fetch the pull request head and its merge base id: revisions diff --git a/.github/workflows/spend-logs-guard.yml b/.github/workflows/spend-logs-guard.yml index 29e9f236202..cc3bda9b880 100644 --- a/.github/workflows/spend-logs-guard.yml +++ b/.github/workflows/spend-logs-guard.yml @@ -1,7 +1,7 @@ name: Spend logs guard -on: # zizmor: ignore[dangerous-triggers] runs the base branch's code only; the PR's diff is read as data and never executed - pull_request_target: +on: + pull_request: branches: - main @@ -19,7 +19,7 @@ jobs: steps: - uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955 # v4.3.0 with: - ref: ${{ github.event.pull_request.base.ref }} + ref: ${{ github.event.pull_request.base.sha }} persist-credentials: false - name: Fetch the pull request head, its merge base and its reviews id: revisions diff --git a/.github/workflows/zizmor.yml b/.github/workflows/zizmor.yml index 73f9efb8df9..eba23e795f3 100644 --- a/.github/workflows/zizmor.yml +++ b/.github/workflows/zizmor.yml @@ -27,10 +27,15 @@ jobs: with: persist-credentials: false + - name: Install zizmor + run: | + curl -fsSL --retry 3 -o "$RUNNER_TEMP/zizmor.tar.gz" \ + https://github.com/zizmorcore/zizmor/releases/download/v1.31.1/zizmor-x86_64-unknown-linux-gnu.tar.gz + echo "d185f726bdaf92df78273217dfdd24ebcd6b0ef48cf3ce776ff5762b6e138c99 $RUNNER_TEMP/zizmor.tar.gz" | sha256sum -c - + tar xzf "$RUNNER_TEMP/zizmor.tar.gz" -C "$RUNNER_TEMP" zizmor + - name: Run zizmor - uses: zizmorcore/zizmor-action@5f14fd08f7cf1cb1609c1e344975f152c7ee938d # v0.5.6 - with: - version: "1.24.1" - min-severity: medium - advanced-security: false - annotations: true + env: + GH_TOKEN: ${{ github.token }} + run: | + "$RUNNER_TEMP/zizmor" --no-ignores --no-config --min-severity=medium --format=github . From e1580b7b837b680d0a0770cfa5f6c88938bc6050 Mon Sep 17 00:00:00 2001 From: Kerry Lu Date: Wed, 28 Oct 2026 17:48:09 -0700 Subject: [PATCH 34/41] refactor(images): remove dall-e-2 and dall-e-3 code paths (#48703) OpenAI shut down dall-e-2 and dall-e-3 on 2026-05-12. Drop their image generation and edit configs, the Azure dall-e-2 create-and-poll flow, and the silent dall-e-2 fallback when no model is passed. image_generation without a model or provider now fails locally with "model is required". image_edit and image_variation leave model out of the provider request instead of sending dall-e-2, so providers that ignore model (Topaz) keep working and OpenAI returns its own 400. --- litellm/__init__.py | 1 - litellm/constants.py | 1 - litellm/cost_calculator.py | 11 +- litellm/images/main.py | 32 ++- .../get_llm_provider_logic.py | 1 - litellm/llms/azure/azure.py | 196 +----------------- .../llms/azure/image_generation/__init__.py | 10 +- .../dall_e_2_transformation.py | 7 - .../dall_e_3_transformation.py | 7 - .../azure_ai/image_generation/__init__.py | 10 +- .../dall_e_2_transformation.py | 7 - .../dall_e_3_transformation.py | 7 - litellm/llms/custom_httpx/aiohttp_handler.py | 5 +- litellm/llms/openai/image_edit/__init__.py | 26 --- .../image_edit/dalle2_transformation.py | 98 --------- .../llms/openai/image_edit/transformation.py | 2 +- .../llms/openai/image_generation/__init__.py | 17 -- .../dall_e_2_transformation.py | 81 -------- .../dall_e_3_transformation.py | 81 -------- .../llms/openai/image_variations/handler.py | 2 +- litellm/utils.py | 11 +- tests/unit/images/test_image_edit.py | 18 ++ tests/unit/images/test_main.py | 23 +- .../test_azure_image_generation_init.py | 18 -- ...t_openai_image_generation_extra_headers.py | 2 +- .../llms/openai/test_openai_common_utils.py | 38 ++-- 26 files changed, 89 insertions(+), 623 deletions(-) delete mode 100644 litellm/llms/azure/image_generation/dall_e_2_transformation.py delete mode 100644 litellm/llms/azure/image_generation/dall_e_3_transformation.py delete mode 100644 litellm/llms/azure_ai/image_generation/dall_e_2_transformation.py delete mode 100644 litellm/llms/azure_ai/image_generation/dall_e_3_transformation.py delete mode 100644 litellm/llms/openai/image_edit/dalle2_transformation.py delete mode 100644 litellm/llms/openai/image_generation/dall_e_2_transformation.py delete mode 100644 litellm/llms/openai/image_generation/dall_e_3_transformation.py diff --git a/litellm/__init__.py b/litellm/__init__.py index 8dfe5adca1c..c649a48211a 100644 --- a/litellm/__init__.py +++ b/litellm/__init__.py @@ -1336,7 +1336,6 @@ all_embedding_models = ( ) ####### IMAGE GENERATION MODELS ################### -openai_image_generation_models = ["dall-e-2", "dall-e-3"] ####### VIDEO GENERATION MODELS ################### openai_video_generation_models = ["sora-2"] diff --git a/litellm/constants.py b/litellm/constants.py index 1d5c6404bfd..25770d960bb 100644 --- a/litellm/constants.py +++ b/litellm/constants.py @@ -764,7 +764,6 @@ ANTHROPIC_WEB_SEARCH_TOOL_MAX_USES: Final = { # Used for web search interception across providers LITELLM_WEB_SEARCH_TOOL_NAME: Final = "litellm_web_search" -DEFAULT_IMAGE_ENDPOINT_MODEL: Final = "dall-e-2" DEFAULT_VIDEO_ENDPOINT_MODEL: Final = "sora-2" DEFAULT_GOOGLE_VIDEO_DURATION_SECONDS: Final = get_env_int("DEFAULT_GOOGLE_VIDEO_DURATION_SECONDS", 8) diff --git a/litellm/cost_calculator.py b/litellm/cost_calculator.py index 86b22069a39..087d5b5b718 100644 --- a/litellm/cost_calculator.py +++ b/litellm/cost_calculator.py @@ -1592,14 +1592,6 @@ def completion_cost( for part in ws_tier_parts ) - if ( - (call_type == "aimage_generation" or call_type == "image_generation") - and model is not None - and isinstance(model, str) - and len(model) == 0 - and custom_llm_provider == "azure" - ): - model = "dall-e-2" # for dall-e-2, azure expects an empty model name # Handle Inputs to completion_cost prompt_tokens = 0 prompt_characters: int | None = None @@ -1674,7 +1666,8 @@ def completion_cost( raw_candidates if custom_pricing is True else _priced_candidates_before_unpriced_rules(raw_candidates) ) - for idx, (model, candidate_pricing_provider) in enumerate(potential_model_names): + for idx, (candidate_model, candidate_pricing_provider) in enumerate(potential_model_names): + model = candidate_model try: if verbose_logger.isEnabledFor(logging.DEBUG): verbose_logger.debug("selected model name for cost calculation: %s", model) diff --git a/litellm/images/main.py b/litellm/images/main.py index d0618e5e804..6e77cbb7046 100644 --- a/litellm/images/main.py +++ b/litellm/images/main.py @@ -16,7 +16,6 @@ import litellm # client is imported from litellm as it's a decorator from litellm import client -from litellm.constants import DEFAULT_IMAGE_ENDPOINT_MODEL from litellm.constants import request_timeout as DEFAULT_REQUEST_TIMEOUT from litellm.exceptions import LiteLLMUnknownProvider from litellm.litellm_core_utils.get_llm_provider_logic import ( @@ -299,17 +298,14 @@ def image_generation( if custom_llm_provider == "chatgpt" or (model is not None and model.startswith("chatgpt/")) else None ) - dynamic_api_key: str | None = None - if model is not None or custom_llm_provider is not None: - model, custom_llm_provider, dynamic_api_key, api_base = get_llm_provider( - model=model, - custom_llm_provider=custom_llm_provider, - api_base=api_base, - litellm_params=chatgpt_params, - ) - else: - model = "dall-e-2" - custom_llm_provider = "openai" # default to dall-e-2 on openai + if model is None and custom_llm_provider is None: + raise litellm.BadRequestError(message="model is required", model="", llm_provider="") + model, custom_llm_provider, dynamic_api_key, api_base = get_llm_provider( + model=model, + custom_llm_provider=custom_llm_provider, + api_base=api_base, + litellm_params=chatgpt_params, + ) model_response.hidden_params["model"] = model openai_params: Final = [ "user", @@ -712,7 +708,7 @@ async def aimage_variation(*args, **kwargs) -> ImageResponse: @client def image_variation( image: FileTypes, - model: str = "dall-e-2", # set to dall-e-2 by default - like OpenAI. + model: str | None = None, n: int = 1, response_format: Literal["url", "b64_json"] = "url", size: str | None = None, @@ -728,7 +724,7 @@ def image_variation( litellm_params: Final = get_litellm_params(**kwargs) # get the custom llm provider model, custom_llm_provider, dynamic_api_key, api_base = get_llm_provider( - model=model, + model=model or "", custom_llm_provider=litellm_params.get("custom_llm_provider", None), api_base=litellm_params.get("api_base", None), api_key=litellm_params.get("api_key", None), @@ -747,7 +743,7 @@ def image_variation( response: ImageResponse | None = None provider_config: Final = ProviderConfigManager.get_provider_model_info( - model=model or "", # openai defaults to dall-e-2 + model=model, provider=llm_provider, ) @@ -772,7 +768,7 @@ def image_variation( model_response=model_response, api_key=api_key, api_base=api_base, - model=model, + model=model or None, image=image, timeout=litellm_params.get("timeout", None), custom_llm_provider=custom_llm_provider, @@ -790,7 +786,7 @@ def image_variation( model_response=model_response, api_key=api_key, api_base=api_base, - model=model, + model=model or None, image=image, timeout=litellm_params.get("timeout", None) or DEFAULT_REQUEST_TIMEOUT, custom_llm_provider=custom_llm_provider, @@ -901,7 +897,7 @@ def image_edit( # get llm provider logic litellm_params: Final = GenericLiteLLMParams(**kwargs) model, custom_llm_provider, _, _ = get_llm_provider( - model=model or DEFAULT_IMAGE_ENDPOINT_MODEL, + model=model or "", custom_llm_provider=custom_llm_provider, ) diff --git a/litellm/litellm_core_utils/get_llm_provider_logic.py b/litellm/litellm_core_utils/get_llm_provider_logic.py index 19c1d5c414c..d9933087334 100644 --- a/litellm/litellm_core_utils/get_llm_provider_logic.py +++ b/litellm/litellm_core_utils/get_llm_provider_logic.py @@ -437,7 +437,6 @@ def get_llm_provider( model in litellm.open_ai_chat_completion_models or "ft:gpt-3.5-turbo" in model or "ft:gpt-4" in model # catches ft:gpt-4-0613, ft:gpt-4o - or model in litellm.openai_image_generation_models or model.startswith("gpt-image") or model in litellm.openai_video_generation_models ): diff --git a/litellm/llms/azure/azure.py b/litellm/llms/azure/azure.py index 02f1ba3472a..5a181faad11 100644 --- a/litellm/llms/azure/azure.py +++ b/litellm/llms/azure/azure.py @@ -13,13 +13,12 @@ from openai import ( BadRequestError, OpenAI, ) -from pydantic import BaseModel, TypeAdapter +from pydantic import BaseModel import litellm -from litellm.constants import AZURE_OPERATION_POLLING_TIMEOUT, DEFAULT_MAX_RETRIES +from litellm.constants import DEFAULT_MAX_RETRIES from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj from litellm.litellm_core_utils.logging_utils import speech_request_body, track_llm_api_timing -from litellm.litellm_core_utils.url_utils import SSRFError, assert_same_origin from litellm.llms.custom_httpx.http_handler import ( AsyncHTTPHandler, HTTPHandler, @@ -34,7 +33,6 @@ from litellm.types.utils import ( from litellm.utils import ( CustomStreamWrapper, convert_to_model_response_object, - modify_url, ) from ...types.llms.openai import HttpxBinaryResponseContent @@ -59,8 +57,6 @@ from .image_generation import ( ) from .image_generation.http_utils import azure_deployment_image_generation_json_body -_DECODED_JSON: Final = TypeAdapter(object) - class AzureOpenAIAssistantsAPIConfig: """ @@ -860,11 +856,6 @@ class AzureChatCompletion(BaseAzureLLM, BaseLLM): headers: dict, deployment_name: str | None = None, ) -> httpx.Response: - """ - Implemented for azure dall-e-2 image gen calls - - Alternative to needing a custom transport implementation - """ if client is None: _params: Final = {} if timeout is not None: @@ -881,95 +872,6 @@ class AzureChatCompletion(BaseAzureLLM, BaseLLM): else: async_handler = client - if ( - "images/generations" in api_base - and api_version - in [ # dall-e-3 starts from `2023-12-01-preview` so we should be able to avoid conflict - "2023-06-01-preview", - "2023-07-01-preview", - "2023-08-01-preview", - "2023-09-01-preview", - "2023-10-01-preview", - ] - ): # CREATE + POLL for azure dall-e-2 calls - api_base = modify_url(original_url=api_base, new_path="/openai/images/generations:submit") - - data.pop( - "model", None - ) # REMOVE 'model' from dall-e-2 arg https://learn.microsoft.com/en-us/azure/ai-services/openai/reference#request-a-generated-image-dall-e-2-preview - response = await async_handler.post( - url=api_base, - data=json.dumps(data), - headers=headers, - ) - if "operation-location" in response.headers: - operation_location_url: Final = response.headers["operation-location"] - else: - raise AzureOpenAIError(status_code=500, message=response.text) - # Reject polling URLs that don't share an origin with ``api_base``. - # Without this an upstream-controlled or attacker-controlled - # value would receive the operator's Azure API key in the - # request headers below. VERIA-51. - try: - assert_same_origin(operation_location_url, api_base) - except SSRFError as ssrf_err: - raise AzureOpenAIError( - status_code=502, - message=f"Rejected polling URL: {ssrf_err}", - ) - response = await async_handler.get( - url=operation_location_url, - headers=headers, - ) - - await response.aread() - - timeout_secs: Final[int] = AZURE_OPERATION_POLLING_TIMEOUT - start_time: Final = time.time() - if "status" not in response.json(): - # Don't reflect the raw response body — when the polling - # URL points at an internal JSON API (cloud metadata - # service etc.) reflecting it here turns Blind SSRF into - # Full-Read SSRF. VERIA-51. - raise AzureOpenAIError( - status_code=502, - message="Polling response missing 'status' field", - ) - while response.json()["status"] not in ["succeeded", "failed"]: - if time.time() - start_time > timeout_secs: - raise AzureOpenAIError(status_code=408, message="Operation polling timed out.") - - await asyncio.sleep(int(response.headers.get("retry-after") or 10)) - response = await async_handler.get( - url=operation_location_url, - headers=headers, - ) - await response.aread() - - if response.json()["status"] == "failed": - error_data: Final = response.json() - # Preserve Azure error details (e.g. content_policy_violation, - # inner_error, content_filter_results) as structured body so - # exception_type() can route them correctly. - _error_body: Final = error_data.get("error", error_data) - _error_msg: Final = ( - _error_body.get("message", "Image generation failed") - if isinstance(_error_body, dict) - else json.dumps(error_data) - ) - raise AzureOpenAIError( - status_code=400, - message=_error_msg, - body=error_data, - ) - - result: Final = _DECODED_JSON.validate_python(response.json()["result"]) - return httpx.Response( - status_code=200, - headers=response.headers, - content=json.dumps(result).encode("utf-8"), - request=httpx.Request(method="POST", url="https://api.openai.com/v1"), - ) request_json: Final = azure_deployment_image_generation_json_body(api_base, data, deployment_name) return await async_handler.post( url=api_base, @@ -988,11 +890,6 @@ class AzureChatCompletion(BaseAzureLLM, BaseLLM): headers: dict, deployment_name: str | None = None, ) -> httpx.Response: - """ - Implemented for azure dall-e-2 image gen calls - - Alternative to needing a custom transport implementation - """ if client is None: _params: Final = {} if timeout is not None: @@ -1006,87 +903,6 @@ class AzureChatCompletion(BaseAzureLLM, BaseLLM): else: sync_handler = client - if ( - "images/generations" in api_base - and api_version - in [ # dall-e-3 starts from `2023-12-01-preview` so we should be able to avoid conflict - "2023-06-01-preview", - "2023-07-01-preview", - "2023-08-01-preview", - "2023-09-01-preview", - "2023-10-01-preview", - ] - ): # CREATE + POLL for azure dall-e-2 calls - api_base = modify_url(original_url=api_base, new_path="/openai/images/generations:submit") - - data.pop( - "model", None - ) # REMOVE 'model' from dall-e-2 arg https://learn.microsoft.com/en-us/azure/ai-services/openai/reference#request-a-generated-image-dall-e-2-preview - response = sync_handler.post( - url=api_base, - data=json.dumps(data), - headers=headers, - ) - if "operation-location" in response.headers: - operation_location_url: Final = response.headers["operation-location"] - else: - raise AzureOpenAIError(status_code=500, message=response.text) - try: - assert_same_origin(operation_location_url, api_base) - except SSRFError as ssrf_err: - raise AzureOpenAIError( - status_code=502, - message=f"Rejected polling URL: {ssrf_err}", - ) - response = sync_handler.get( - url=operation_location_url, - headers=headers, - ) - - response.read() - - timeout_secs: Final[int] = AZURE_OPERATION_POLLING_TIMEOUT - start_time: Final = time.time() - if "status" not in response.json(): - raise AzureOpenAIError( - status_code=502, - message="Polling response missing 'status' field", - ) - while response.json()["status"] not in ["succeeded", "failed"]: - if time.time() - start_time > timeout_secs: - raise AzureOpenAIError(status_code=408, message="Operation polling timed out.") - - time.sleep(int(response.headers.get("retry-after") or 10)) - response = sync_handler.get( - url=operation_location_url, - headers=headers, - ) - response.read() - - if response.json()["status"] == "failed": - error_data: Final = response.json() - # Preserve Azure error details (e.g. content_policy_violation, - # inner_error, content_filter_results) as structured body so - # exception_type() can route them correctly. - _error_body: Final = error_data.get("error", error_data) - _error_msg: Final = ( - _error_body.get("message", "Image generation failed") - if isinstance(_error_body, dict) - else json.dumps(error_data) - ) - raise AzureOpenAIError( - status_code=400, - message=_error_msg, - body=error_data, - ) - - result: Final = _DECODED_JSON.validate_python(response.json()["result"]) - return httpx.Response( - status_code=200, - headers=response.headers, - content=json.dumps(result).encode("utf-8"), - request=httpx.Request(method="POST", url="https://api.openai.com/v1"), - ) request_json: Final = azure_deployment_image_generation_json_body(api_base, data, deployment_name) return sync_handler.post( url=api_base, @@ -1193,10 +1009,10 @@ class AzureChatCompletion(BaseAzureLLM, BaseLLM): deployment_name=model, ) - provider_config: Final = get_azure_image_generation_config(data.get("model", "dall-e-2")) + provider_config: Final = get_azure_image_generation_config(data.get("model", "")) if provider_config is not None: return provider_config.transform_image_generation_response( - model=data.get("model", "dall-e-2"), + model=data.get("model", ""), raw_response=httpx_response, model_response=model_response or ImageResponse(), logging_obj=logging_obj, @@ -1329,10 +1145,10 @@ class AzureChatCompletion(BaseAzureLLM, BaseLLM): headers=request_headers, deployment_name=model, ) - provider_config: Final = get_azure_image_generation_config(data.get("model", "dall-e-2")) + provider_config: Final = get_azure_image_generation_config(data.get("model", "")) if isinstance(provider_config, AzureFoundryMAIImageGenerationConfig): return provider_config.transform_image_generation_response( - model=data.get("model", "dall-e-2"), + model=data.get("model", ""), raw_response=httpx_response, model_response=model_response or ImageResponse(), logging_obj=logging_obj, diff --git a/litellm/llms/azure/image_generation/__init__.py b/litellm/llms/azure/image_generation/__init__.py index a2a905f2287..7e9b5f5d99c 100644 --- a/litellm/llms/azure/image_generation/__init__.py +++ b/litellm/llms/azure/image_generation/__init__.py @@ -4,14 +4,10 @@ from litellm.llms.base_llm.image_generation.transformation import ( BaseImageGenerationConfig, ) -from .dall_e_2_transformation import AzureDallE2ImageGenerationConfig -from .dall_e_3_transformation import AzureDallE3ImageGenerationConfig from .gpt_transformation import AzureGPTImageGenerationConfig from .http_utils import azure_deployment_image_generation_json_body __all__ = [ - "AzureDallE2ImageGenerationConfig", - "AzureDallE3ImageGenerationConfig", "AzureGPTImageGenerationConfig", "azure_deployment_image_generation_json_body", ] @@ -21,11 +17,7 @@ def get_azure_image_generation_config(model: str) -> BaseImageGenerationConfig: model = model.lower() model = model.replace("-", "") model = model.replace("_", "") - if model == "" or "dalle2" in model: # empty model is dall-e-2 - return AzureDallE2ImageGenerationConfig() - elif "dalle3" in model: - return AzureDallE3ImageGenerationConfig() - elif AzureFoundryMAIImageGenerationConfig.is_mai_model(model): + if AzureFoundryMAIImageGenerationConfig.is_mai_model(model): return AzureFoundryMAIImageGenerationConfig() else: verbose_logger.debug( diff --git a/litellm/llms/azure/image_generation/dall_e_2_transformation.py b/litellm/llms/azure/image_generation/dall_e_2_transformation.py deleted file mode 100644 index 788753e4ea9..00000000000 --- a/litellm/llms/azure/image_generation/dall_e_2_transformation.py +++ /dev/null @@ -1,7 +0,0 @@ -from litellm.llms.openai.image_generation import DallE2ImageGenerationConfig - - -class AzureDallE2ImageGenerationConfig(DallE2ImageGenerationConfig): - """ - Azure dall-e-2 image generation config - """ diff --git a/litellm/llms/azure/image_generation/dall_e_3_transformation.py b/litellm/llms/azure/image_generation/dall_e_3_transformation.py deleted file mode 100644 index e6974f9317e..00000000000 --- a/litellm/llms/azure/image_generation/dall_e_3_transformation.py +++ /dev/null @@ -1,7 +0,0 @@ -from litellm.llms.openai.image_generation import DallE3ImageGenerationConfig - - -class AzureDallE3ImageGenerationConfig(DallE3ImageGenerationConfig): - """ - Azure dall-e-3 image generation config - """ diff --git a/litellm/llms/azure_ai/image_generation/__init__.py b/litellm/llms/azure_ai/image_generation/__init__.py index 742922ad295..bd3c11d36f9 100644 --- a/litellm/llms/azure_ai/image_generation/__init__.py +++ b/litellm/llms/azure_ai/image_generation/__init__.py @@ -3,15 +3,11 @@ from litellm.llms.base_llm.image_generation.transformation import ( BaseImageGenerationConfig, ) -from .dall_e_2_transformation import AzureFoundryDallE2ImageGenerationConfig -from .dall_e_3_transformation import AzureFoundryDallE3ImageGenerationConfig from .flux_transformation import AzureFoundryFluxImageGenerationConfig from .gpt_transformation import AzureFoundryGPTImageGenerationConfig from .mai_transformation import AzureFoundryMAIImageGenerationConfig __all__ = [ - "AzureFoundryDallE2ImageGenerationConfig", - "AzureFoundryDallE3ImageGenerationConfig", "AzureFoundryFluxImageGenerationConfig", "AzureFoundryGPTImageGenerationConfig", "AzureFoundryMAIImageGenerationConfig", @@ -22,11 +18,7 @@ def get_azure_ai_image_generation_config(model: str) -> BaseImageGenerationConfi model = model.lower() model = model.replace("-", "") model = model.replace("_", "") - if model == "" or "dalle2" in model: # empty model is dall-e-2 - return AzureFoundryDallE2ImageGenerationConfig() - elif "dalle3" in model: - return AzureFoundryDallE3ImageGenerationConfig() - elif AzureFoundryMAIImageGenerationConfig.is_mai_model(model): + if AzureFoundryMAIImageGenerationConfig.is_mai_model(model): return AzureFoundryMAIImageGenerationConfig() elif "flux" in model: return AzureFoundryFluxImageGenerationConfig() diff --git a/litellm/llms/azure_ai/image_generation/dall_e_2_transformation.py b/litellm/llms/azure_ai/image_generation/dall_e_2_transformation.py deleted file mode 100644 index 2c8a5426628..00000000000 --- a/litellm/llms/azure_ai/image_generation/dall_e_2_transformation.py +++ /dev/null @@ -1,7 +0,0 @@ -from litellm.llms.openai.image_generation import DallE2ImageGenerationConfig - - -class AzureFoundryDallE2ImageGenerationConfig(DallE2ImageGenerationConfig): - """ - Azure dall-e-2 image generation config - """ diff --git a/litellm/llms/azure_ai/image_generation/dall_e_3_transformation.py b/litellm/llms/azure_ai/image_generation/dall_e_3_transformation.py deleted file mode 100644 index 3ceb9b405da..00000000000 --- a/litellm/llms/azure_ai/image_generation/dall_e_3_transformation.py +++ /dev/null @@ -1,7 +0,0 @@ -from litellm.llms.openai.image_generation import DallE3ImageGenerationConfig - - -class AzureFoundryDallE3ImageGenerationConfig(DallE3ImageGenerationConfig): - """ - Azure dall-e-3 image generation config - """ diff --git a/litellm/llms/custom_httpx/aiohttp_handler.py b/litellm/llms/custom_httpx/aiohttp_handler.py index 4060d84f7f4..e6544968d24 100644 --- a/litellm/llms/custom_httpx/aiohttp_handler.py +++ b/litellm/llms/custom_httpx/aiohttp_handler.py @@ -561,11 +561,8 @@ class BaseLLMAIOHTTPHandler: organization: str | None = None, headers: dict | None = None, ) -> ImageResponse: - if model is None: - raise ValueError("model is required for non-openai image variations") - provider_config: Final = ProviderConfigManager.get_provider_image_variation_config( - model=model, # openai defaults to dall-e-2 + model=model or "", provider=LlmProviders(custom_llm_provider), ) diff --git a/litellm/llms/openai/image_edit/__init__.py b/litellm/llms/openai/image_edit/__init__.py index 63a41b65e8a..261010071c9 100644 --- a/litellm/llms/openai/image_edit/__init__.py +++ b/litellm/llms/openai/image_edit/__init__.py @@ -1,31 +1,5 @@ -from typing import Final - -from litellm.llms.base_llm.image_edit.transformation import BaseImageEditConfig - -from .dalle2_transformation import DallE2ImageEditConfig from .transformation import OpenAIImageEditConfig __all__ = [ - "DallE2ImageEditConfig", "OpenAIImageEditConfig", - "get_openai_image_edit_config", ] - - -def get_openai_image_edit_config(model: str) -> BaseImageEditConfig: - """ - Get the appropriate OpenAI image edit config based on the model. - - Args: - model: The model name (e.g., "dall-e-2", "gpt-image-1") - - Returns: - The appropriate config instance for the model - """ - model_normalized: Final = model.lower().replace("-", "").replace("_", "") - - if model_normalized == "dalle2": - return DallE2ImageEditConfig() - else: - # Default to standard OpenAI config for gpt-image-1 and other models - return OpenAIImageEditConfig() diff --git a/litellm/llms/openai/image_edit/dalle2_transformation.py b/litellm/llms/openai/image_edit/dalle2_transformation.py deleted file mode 100644 index f349959b14f..00000000000 --- a/litellm/llms/openai/image_edit/dalle2_transformation.py +++ /dev/null @@ -1,98 +0,0 @@ -from io import BufferedReader -from typing import TYPE_CHECKING, Any, Final, cast - -from httpx._types import RequestFiles - -import litellm -from litellm.images.utils import ImageEditRequestUtils -from litellm.types.images.main import ImageEditRequestParams -from litellm.types.llms.openai import FileTypes -from litellm.types.router import GenericLiteLLMParams - -from .transformation import OpenAIImageEditConfig - -if TYPE_CHECKING: - from litellm.litellm_core_utils.litellm_logging import Logging as _LiteLLMLoggingObj - - LiteLLMLoggingObj = _LiteLLMLoggingObj -else: - LiteLLMLoggingObj = Any - - -class DallE2ImageEditConfig(OpenAIImageEditConfig): - """ - DALL-E-2 specific configuration for image edit API. - - DALL-E-2 only supports editing a single image (not an array). - Uses "image" field name instead of "image[]". - """ - - def transform_image_edit_request( - self, - model: str, - prompt: str | None, - image: FileTypes | None, - image_edit_optional_request_params: dict, - litellm_params: GenericLiteLLMParams, - headers: dict, - ) -> tuple[dict, RequestFiles]: - """ - Transform image edit request for DALL-E-2. - - DALL-E-2 only accepts a single image with field name "image" (not "image[]"). - """ - request_params: Final = { - "model": model, - **image_edit_optional_request_params, - } - if image is not None: - request_params["image"] = image - if prompt is not None: - request_params["prompt"] = prompt - - request: Final = ImageEditRequestParams(**request_params) - request_dict: Final = cast(dict, request) - - ######################################################### - # Separate images and masks as `files` and send other parameters as `data` - ######################################################### - _image_list: Final = request_dict.get("image") - _mask = request_dict.get("mask") - data_without_files: Final = {k: v for k, v in request_dict.items() if k not in ["image", "mask"]} - files_list: Final[list[tuple[str, Any]]] = [] - - # Handle image parameter - DALL-E-2 only supports single image - if _image_list is not None: - image_list: Final = [_image_list] if not isinstance(_image_list, list) else _image_list - - # Validate only one image is provided - if len(image_list) > 1: - raise litellm.BadRequestError( - message="DALL-E-2 only supports editing a single image. Please provide one image.", - model=model, - llm_provider="openai", - ) - - # Use "image" field name (singular) for DALL-E-2 - for _image in image_list: - if _image is not None: - self._add_image_to_files( - files_list=files_list, - image=_image, - field_name="image", - ) - - # Handle mask parameter if provided - if _mask is not None: - # Handle case where mask can be a list (extract first mask) - if isinstance(_mask, list): - _mask = _mask[0] if _mask else None - - if _mask is not None: - mask_content_type: Final[str] = ImageEditRequestUtils.get_image_content_type(_mask) - if isinstance(_mask, BufferedReader): - files_list.append(("mask", (_mask.name, _mask, mask_content_type))) - else: - files_list.append(("mask", ("mask.png", _mask, mask_content_type))) - - return data_without_files, files_list diff --git a/litellm/llms/openai/image_edit/transformation.py b/litellm/llms/openai/image_edit/transformation.py index 9def833052b..0bba58096ef 100644 --- a/litellm/llms/openai/image_edit/transformation.py +++ b/litellm/llms/openai/image_edit/transformation.py @@ -96,7 +96,7 @@ class OpenAIImageEditConfig(BaseImageEditConfig): """ # Build request params, only including non-None values request_params: Final = { - "model": model, + **({"model": model} if model else {}), **image_edit_optional_request_params, } if image is not None: diff --git a/litellm/llms/openai/image_generation/__init__.py b/litellm/llms/openai/image_generation/__init__.py index e20c80f20bb..4cffbe81a0e 100644 --- a/litellm/llms/openai/image_generation/__init__.py +++ b/litellm/llms/openai/image_generation/__init__.py @@ -1,9 +1,3 @@ -from litellm.llms.base_llm.image_generation.transformation import ( - BaseImageGenerationConfig, -) - -from .dall_e_2_transformation import DallE2ImageGenerationConfig -from .dall_e_3_transformation import DallE3ImageGenerationConfig from .gpt_transformation import GPTImageGenerationConfig from .guardrail_translation import ( OpenAIImageGenerationHandler, @@ -11,18 +5,7 @@ from .guardrail_translation import ( ) __all__ = [ - "DallE2ImageGenerationConfig", - "DallE3ImageGenerationConfig", "GPTImageGenerationConfig", "OpenAIImageGenerationHandler", "guardrail_translation_mappings", ] - - -def get_openai_image_generation_config(model: str) -> BaseImageGenerationConfig: - if model.startswith("dall-e-2") or model == "": # empty model is dall-e-2 - return DallE2ImageGenerationConfig() - elif model.startswith("dall-e-3"): - return DallE3ImageGenerationConfig() - else: - return GPTImageGenerationConfig() diff --git a/litellm/llms/openai/image_generation/dall_e_2_transformation.py b/litellm/llms/openai/image_generation/dall_e_2_transformation.py deleted file mode 100644 index ffc6f1d5fe9..00000000000 --- a/litellm/llms/openai/image_generation/dall_e_2_transformation.py +++ /dev/null @@ -1,81 +0,0 @@ -from typing import TYPE_CHECKING, Final - -import httpx - -from litellm.llms.base_llm.image_generation.transformation import ( - BaseImageGenerationConfig, -) -from litellm.types.llms.openai import OpenAIImageGenerationOptionalParams -from litellm.types.utils import ImageResponse -from litellm.utils import convert_to_model_response_object - -if TYPE_CHECKING: - from litellm.litellm_core_utils.logging import Logging as LiteLLMLoggingObj - - from litellm.litellm_core_utils.tokenizer import Encoding as Tokenizer - - -class DallE2ImageGenerationConfig(BaseImageGenerationConfig): - """ - OpenAI dall-e-2 image generation config - """ - - def get_supported_openai_params(self, model: str) -> list[OpenAIImageGenerationOptionalParams]: - return ["n", "response_format", "quality", "size", "user"] - - def map_openai_params( - self, - non_default_params: dict, - optional_params: dict, - model: str, - drop_params: bool, - ) -> dict: - supported_params: Final = self.get_supported_openai_params(model) - for k in non_default_params: - if k not in optional_params: - if k in supported_params: - optional_params[k] = non_default_params[k] - elif drop_params: - pass - else: - raise ValueError( - f"Parameter {k} is not supported for model {model}. Supported parameters are {supported_params}. Set drop_params=True to drop unsupported parameters." - ) - - return optional_params - - def transform_image_generation_response( - self, - model: str, - raw_response: httpx.Response, - model_response: ImageResponse, - logging_obj: "LiteLLMLoggingObj", - request_data: dict, - optional_params: dict, - litellm_params: dict, - encoding: "Tokenizer | None", - api_key: str | None = None, - json_mode: bool | None = None, - ) -> ImageResponse: - response: Final = raw_response.json() - - stringified_response: Final = response - ## LOGGING - logging_obj.post_call( - input=request_data.get("prompt", ""), - api_key=api_key, - additional_args={"complete_input_dict": request_data}, - original_response=stringified_response, - ) - image_response: Final[ImageResponse] = convert_to_model_response_object( - response_object=stringified_response, - model_response_object=model_response, - response_type="image_generation", - ) - - # set optional params - image_response.size = optional_params.get("size", "1024x1024") # default is always 1024x1024 - image_response.quality = optional_params.get("quality", "standard") # always standard for dall-e-2 - image_response.output_format = optional_params.get("output_format", "png") # always png for dall-e-2 - - return image_response diff --git a/litellm/llms/openai/image_generation/dall_e_3_transformation.py b/litellm/llms/openai/image_generation/dall_e_3_transformation.py deleted file mode 100644 index 90b7eaedf2f..00000000000 --- a/litellm/llms/openai/image_generation/dall_e_3_transformation.py +++ /dev/null @@ -1,81 +0,0 @@ -from typing import TYPE_CHECKING, Final - -import httpx - -from litellm.llms.base_llm.image_generation.transformation import ( - BaseImageGenerationConfig, -) -from litellm.types.llms.openai import OpenAIImageGenerationOptionalParams -from litellm.types.utils import ImageResponse -from litellm.utils import convert_to_model_response_object - -if TYPE_CHECKING: - from litellm.litellm_core_utils.logging import Logging as LiteLLMLoggingObj - - from litellm.litellm_core_utils.tokenizer import Encoding as Tokenizer - - -class DallE3ImageGenerationConfig(BaseImageGenerationConfig): - """ - OpenAI dall-e-3 image generation config - """ - - def get_supported_openai_params(self, model: str) -> list[OpenAIImageGenerationOptionalParams]: - return ["n", "response_format", "quality", "size", "user", "style"] - - def map_openai_params( - self, - non_default_params: dict, - optional_params: dict, - model: str, - drop_params: bool, - ) -> dict: - supported_params: Final = self.get_supported_openai_params(model) - for k in non_default_params: - if k not in optional_params: - if k in supported_params: - optional_params[k] = non_default_params[k] - elif drop_params: - pass - else: - raise ValueError( - f"Parameter {k} is not supported for model {model}. Supported parameters are {supported_params}. Set drop_params=True to drop unsupported parameters." - ) - - return optional_params - - def transform_image_generation_response( - self, - model: str, - raw_response: httpx.Response, - model_response: ImageResponse, - logging_obj: "LiteLLMLoggingObj", - request_data: dict, - optional_params: dict, - litellm_params: dict, - encoding: "Tokenizer | None", - api_key: str | None = None, - json_mode: bool | None = None, - ) -> ImageResponse: - response: Final = raw_response.json() - - stringified_response: Final = response - ## LOGGING - logging_obj.post_call( - input=request_data.get("prompt", ""), - api_key=api_key, - additional_args={"complete_input_dict": request_data}, - original_response=stringified_response, - ) - image_response: Final[ImageResponse] = convert_to_model_response_object( - response_object=stringified_response, - model_response_object=model_response, - response_type="image_generation", - ) - - # set optional params - image_response.size = optional_params.get("size", "1024x1024") # default is always 1024x1024 - image_response.quality = optional_params.get("quality", "hd") # always hd for dall-e-3 - image_response.output_format = optional_params.get("output_format", "png") # always png for dall-e-3 - - return image_response diff --git a/litellm/llms/openai/image_variations/handler.py b/litellm/llms/openai/image_variations/handler.py index 89dcdc40e02..01f01d3de1d 100644 --- a/litellm/llms/openai/image_variations/handler.py +++ b/litellm/llms/openai/image_variations/handler.py @@ -151,7 +151,7 @@ class OpenAIImageVariationsHandler: ) -> ImageResponse: try: provider_config: Final = ProviderConfigManager.get_provider_image_variation_config( - model=model or "", # openai defaults to dall-e-2 + model=model or "", provider=LlmProviders.OPENAI, ) diff --git a/litellm/utils.py b/litellm/utils.py index 1ef39059e1b..be1e11dd39f 100644 --- a/litellm/utils.py +++ b/litellm/utils.py @@ -7664,7 +7664,6 @@ def validate_environment( model in litellm.open_ai_chat_completion_models or model in litellm.open_ai_text_completion_models or model in litellm.open_ai_embedding_models - or model in litellm.openai_image_generation_models or model.startswith("gpt-image") ): if "OPENAI_API_KEY" in os.environ: @@ -10311,11 +10310,9 @@ class ProviderConfigManager: provider: LlmProviders, ) -> BaseImageGenerationConfig | None: if LlmProviders.OPENAI == provider: - from litellm.llms.openai.image_generation import ( - get_openai_image_generation_config, - ) + from litellm.llms.openai.image_generation import GPTImageGenerationConfig - return get_openai_image_generation_config(model) + return GPTImageGenerationConfig() elif LlmProviders.AZURE == provider: from litellm.llms.azure.image_generation import ( get_azure_image_generation_config, @@ -10546,9 +10543,9 @@ class ProviderConfigManager: provider: LlmProviders, ) -> BaseImageEditConfig | None: if LlmProviders.OPENAI == provider: - from litellm.llms.openai.image_edit import get_openai_image_edit_config + from litellm.llms.openai.image_edit import OpenAIImageEditConfig - return get_openai_image_edit_config(model=model) + return OpenAIImageEditConfig() elif LlmProviders.HOSTED_VLLM == provider: from litellm.llms.hosted_vllm.image_edit import get_hosted_vllm_image_edit_config diff --git a/tests/unit/images/test_image_edit.py b/tests/unit/images/test_image_edit.py index fb7a2869b2a..4f858050e3e 100644 --- a/tests/unit/images/test_image_edit.py +++ b/tests/unit/images/test_image_edit.py @@ -105,6 +105,24 @@ def test_openai_image_edit_sync_accepts_bytesio_images(respx_mock: respx.MockRou assert _multipart_image_parts(route.calls[0].request) == (_FIRST_IMAGE,) +def test_openai_image_edit_without_a_model_leaves_model_out_of_the_request( + respx_mock: respx.MockRouter, httpx_transport: None +) -> None: + route: Final = respx_mock.post("https://api.openai.com/v1/images/edits").mock( + return_value=httpx.Response(200, json=_EDIT_RESPONSE) + ) + + litellm.image_edit( + prompt="combine the reference images", + image=io.BytesIO(_FIRST_IMAGE), + custom_llm_provider="openai", + api_key="fake-key", + ) + + assert route.call_count == 1 + assert b'name="model"' not in route.calls[0].request.read() + + @pytest.mark.asyncio async def test_openai_image_edit_accepts_bytesio_images(respx_mock: respx.MockRouter, httpx_transport: None) -> None: route: Final = respx_mock.post("https://api.openai.com/v1/images/edits").mock( diff --git a/tests/unit/images/test_main.py b/tests/unit/images/test_main.py index 29424e50079..61b2f28b727 100644 --- a/tests/unit/images/test_main.py +++ b/tests/unit/images/test_main.py @@ -300,7 +300,7 @@ async def test_aimage_generation_ignores_a_non_string_base_url_like_the_sync_pat ) response: Final = await litellm.aimage_generation( - model="dall-e-3", + model="gpt-image-2.5-sunburst", prompt="a red circle", api_base=api_base, base_url=httpx.URL(api_base), @@ -688,3 +688,24 @@ async def test_chatgpt_images_forward_client_login_and_workspace( assert "cookie" not in request.headers assert "x-litellm-api-key" not in request.headers assert json.loads(request.content) == {"model": "gpt-image-2", "prompt": "A blue circle"} + + +def test_image_generation_without_a_model_or_provider_fails_before_calling_any_provider( + respx_mock: respx.MockRouter, +) -> None: + with pytest.raises(litellm.BadRequestError, match="model is required"): + litellm.image_generation(prompt="a red circle", api_key="fake_openai_api_key") + + assert not respx_mock.calls + + +def test_topaz_image_variation_works_without_a_model(respx_mock: respx.MockRouter) -> None: + route: Final = respx_mock.post("https://api.topazlabs.com/image/v1/enhance").mock( + return_value=httpx.Response(200, content=b"enhanced-image", headers={"content-type": "image/png"}) + ) + + response: Final = litellm.image_variation(image=b"source-image", custom_llm_provider="topaz", api_key="fake-key") + + assert route.call_count == 1 + assert response.data is not None + assert response.data[0].b64_json == base64.b64encode(b"enhanced-image").decode() diff --git a/tests/unit/llms/azure/image_generation/test_azure_image_generation_init.py b/tests/unit/llms/azure/image_generation/test_azure_image_generation_init.py index eabd5c8427d..83e01fce642 100644 --- a/tests/unit/llms/azure/image_generation/test_azure_image_generation_init.py +++ b/tests/unit/llms/azure/image_generation/test_azure_image_generation_init.py @@ -20,27 +20,9 @@ from litellm.llms.azure.image_generation.http_utils import ( azure_deployment_image_generation_json_body, ) from litellm.llms.custom_httpx.http_handler import HTTPHandler -from litellm.llms.azure.image_generation import ( - AzureDallE3ImageGenerationConfig, - get_azure_image_generation_config, -) from litellm.utils import get_optional_params_image_gen -@pytest.mark.parametrize( - "received_model, expected_config", - [ - ("dall-e-3", AzureDallE3ImageGenerationConfig), - ("dalle-3", AzureDallE3ImageGenerationConfig), - ("openai_dall_e_3", AzureDallE3ImageGenerationConfig), - ], -) -def test_azure_image_generation_config(received_model, expected_config): - assert isinstance( - get_azure_image_generation_config(received_model), expected_config - ) - - def test_azure_deployment_image_generation_json_body(): """Deployment-scoped Azure image URL must not send ``model`` in JSON.""" api = ( diff --git a/tests/unit/llms/openai/image_generation/test_openai_image_generation_extra_headers.py b/tests/unit/llms/openai/image_generation/test_openai_image_generation_extra_headers.py index 11df07f5fea..e103412e0c3 100644 --- a/tests/unit/llms/openai/image_generation/test_openai_image_generation_extra_headers.py +++ b/tests/unit/llms/openai/image_generation/test_openai_image_generation_extra_headers.py @@ -265,7 +265,7 @@ class TestImageGenerationEntryPointHeaders: test_headers = {"cf-aig-authorization": "Bearer my-secret"} await litellm.aimage_generation( - model="dall-e-3", + model="gpt-image-2.5-sunburst", prompt="A white cat", extra_headers=test_headers, client=mock_openai_client, diff --git a/tests/unit/llms/openai/test_openai_common_utils.py b/tests/unit/llms/openai/test_openai_common_utils.py index 831df30c98a..8dc92e23fdf 100644 --- a/tests/unit/llms/openai/test_openai_common_utils.py +++ b/tests/unit/llms/openai/test_openai_common_utils.py @@ -46,12 +46,12 @@ API_FUNCTION_PARAMS = [ ( "image_generation", False, - {"model": "dall-e-3", "prompt": "A beautiful sunset over mountains"}, + {"model": "gpt-image-2.5-sunburst", "prompt": "A beautiful sunset over mountains"}, ), ( "image_generation", True, - {"model": "dall-e-3", "prompt": "A beautiful sunset over mountains"}, + {"model": "gpt-image-2.5-sunburst", "prompt": "A beautiful sunset over mountains"}, ), ( "speech", @@ -84,11 +84,7 @@ async def test_openai_client_reuse(function_name, is_async, args): """ # Determine which client class to mock based on whether the test is async - client_path = ( - "litellm.llms.openai.openai.AsyncOpenAI" - if is_async - else "litellm.llms.openai.openai.OpenAI" - ) + client_path = "litellm.llms.openai.openai.AsyncOpenAI" if is_async else "litellm.llms.openai.openai.OpenAI" # Create the appropriate patches with ( @@ -98,9 +94,7 @@ async def test_openai_client_reuse(function_name, is_async, args): ): # Setup the mock to return None first time (cache miss) then a client for subsequent calls mock_client = MagicMock() - mock_get_cache.side_effect = [None] + [ - mock_client - ] * 9 # First call returns None, rest return the mock client + mock_get_cache.side_effect = [None] + [mock_client] * 9 # First call returns None, rest return the mock client # Make 10 API calls for _ in range(10): @@ -118,9 +112,9 @@ async def test_openai_client_reuse(function_name, is_async, args): pass # Verify client was created only once - assert ( - mock_client_class.call_count == 1 - ), f"{'Async' if is_async else ''}OpenAI client should be created only once" + assert mock_client_class.call_count == 1, ( + f"{'Async' if is_async else ''}OpenAI client should be created only once" + ) # Verify the client was cached assert mock_set_cache.call_count == 1, "Client should be cached once" @@ -144,12 +138,8 @@ def test_precomputed_init_params_match_inspect_signature(): _OPENAI_INIT_PARAMS, ) - expected_openai = tuple( - p for p in inspect.signature(OpenAI.__init__).parameters if p != "self" - ) - expected_azure = tuple( - p for p in inspect.signature(AzureOpenAI.__init__).parameters if p != "self" - ) + expected_openai = tuple(p for p in inspect.signature(OpenAI.__init__).parameters if p != "self") + expected_azure = tuple(p for p in inspect.signature(AzureOpenAI.__init__).parameters if p != "self") assert _OPENAI_INIT_PARAMS == expected_openai assert _AZURE_OPENAI_INIT_PARAMS == expected_azure @@ -500,7 +490,10 @@ async def test_sdk_api_factories_keep_httpx_transport_and_request_timeouts(api: moved: Final = mock.get(host="sdk-default.example", path="/v1/moved-models").mock( return_value=httpx.Response( 200, - json={"object": "list", "data": [{"id": "sdk-compat", "object": "model", "created": 0, "owned_by": "t"}]}, + json={ + "object": "list", + "data": [{"id": "sdk-compat", "object": "model", "created": 0, "owned_by": "t"}], + }, ) ) sdk_client: Final = _sdk_api_client(api, is_async, timeout) @@ -542,7 +535,10 @@ async def test_azure_gateway_and_image_variations_use_the_callers_async_session( assert azure_client is not None assert azure_client._client is session response: Final = await aimage_variation( - image=BytesIO(b"image-bytes"), api_key="transport-only", api_base="https://sdk-default.example/v1" + image=BytesIO(b"image-bytes"), + custom_llm_provider="openai", + api_key="transport-only", + api_base="https://sdk-default.example/v1", ) assert response.data[0].url == "https://example.com/image.png" assert not session.is_closed From 77cad6e9032369fab248870d8ca9cfa94367484a Mon Sep 17 00:00:00 2001 From: zoveki <349028329+zoveki@users.noreply.github.com> Date: Wed, 28 Oct 2026 18:05:03 -0700 Subject: [PATCH 35/41] test(proxy): audit #46997 (#48918) * test(proxy): audit #46997 Add request-path cells for the global proxy spend counter: a key with no user spends toward litellm max_budget, and two proxies sharing Redis add every request to one counter that matches the proxy budget row * test(proxy): wait for the recorded budget spend and shared counter before checking them --------- Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com> --- .../spend/test_global_proxy_spend_counter.py | 159 +++++++++++++++++- 1 file changed, 158 insertions(+), 1 deletion(-) diff --git a/tests/integration/spend/test_global_proxy_spend_counter.py b/tests/integration/spend/test_global_proxy_spend_counter.py index 0185e48ffbd..e818487cf53 100644 --- a/tests/integration/spend/test_global_proxy_spend_counter.py +++ b/tests/integration/spend/test_global_proxy_spend_counter.py @@ -1,16 +1,113 @@ +import json import os import uuid +from itertools import product +from pathlib import Path from typing import Final +import httpx import pytest +import yaml +from integration._support.client import Gateway, eventually, object_value +from integration._support.process import graceful_stop_seconds, owned_proxy +from integration._support.redis_process import owned_redis +from pydantic import JsonValue, TypeAdapter +from redis import Redis import litellm from litellm.caching.in_memory_cache import InMemoryCache from litellm.caching.redis_cache import RedisCache -from litellm.constants import GLOBAL_PROXY_SPEND_CACHE_KEY +from litellm.constants import GLOBAL_PROXY_SPEND_CACHE_KEY, LITELLM_PROXY_BUDGET_NAME from litellm.proxy.common_utils.user_api_key_cache import UserApiKeyCache from litellm.proxy.proxy_server import update_cache +BUDGET_MODEL: Final = "proxy-budget-model" +CONFIG: Final = TypeAdapter(dict[str, JsonValue]) +SECTION: Final = TypeAdapter(dict[str, JsonValue]) +OWNED_PROXY_SECONDS: Final = 2 * graceful_stop_seconds() + 120 + + +def _proxy_budget_config(gateway: Gateway, directory: Path, max_budget: float) -> Path: + loaded: Final = CONFIG.validate_python( + yaml.safe_load((Path(__file__).resolve().parents[1] / "proxy_config.yaml").read_text()) + ) + model: Final[JsonValue] = { + "model_name": BUDGET_MODEL, + "litellm_params": { + "model": "openai/gpt-4o-mini", + "api_key": "integration-provider-key", + "api_base": f"{gateway.upstream_url}/v1", + "input_cost_per_token": 0.001, + "output_cost_per_token": 0.002, + }, + } + path: Final = directory / "proxy_budget_config.yaml" + path.write_text( + yaml.safe_dump( + { + **loaded, + "model_list": [model], + "general_settings": { + **SECTION.validate_python(loaded["general_settings"]), + "user_api_key_cache_ttl": 600, + }, + "litellm_settings": { + **SECTION.validate_python(loaded["litellm_settings"]), + "max_budget": max_budget, + "budget_duration": "30d", + }, + } + ) + ) + return path + + +def _chat(proxy: Gateway, key: str, marker: str) -> httpx.Response: + return proxy.request( + "POST", + "/v1/chat/completions", + {"model": BUDGET_MODEL, "max_tokens": 20, "messages": [{"role": "user", "content": marker}]}, + key=key, + ) + + +def _served_id(proxy: Gateway, key: str) -> str: + response: Final = _chat(proxy, key, f"proxy budget {uuid.uuid4().hex}") + assert response.status_code == 200, response.text + identity: Final = response.json()["id"] + assert isinstance(identity, str), response.text + return identity + + +def _logged_spend(proxy: Gateway, request_ids: tuple[str, ...]) -> tuple[float, ...]: + rows: Final = eventually( + lambda: proxy.rows( + 'SELECT spend FROM "LiteLLM_SpendLogs" WHERE request_id = ANY(%s::text[])', + ("{" + ",".join(request_ids) + "}",), + ), + lambda found: len(found) == len(request_ids), + seconds=70, + ) + return tuple(float(str(row["spend"])) for row in rows) + + +def _proxy_budget_spend(proxy: Gateway) -> float | None: + rows: Final = proxy.rows('SELECT spend FROM "LiteLLM_UserTable" WHERE user_id = %s', (LITELLM_PROXY_BUDGET_NAME,)) + return float(str(rows[0]["spend"])) if rows else None + + +def _counter_key(redis: Redis) -> str: + found: Final = tuple(str(key, "utf-8") for key in redis.scan_iter(match=f"*{GLOBAL_PROXY_SPEND_CACHE_KEY}")) + assert len(found) == 1, found + return found[0] + + +def _upstream_saw(proxy: Gateway, marker: str) -> bool: + with httpx.Client(base_url=proxy.upstream_url, timeout=5, trust_env=False) as upstream: + response: Final = upstream.get("/__observations") + response.raise_for_status() + return marker in json.dumps(object_value(response.json())["requests"]) + async def _add_request_cost(cache: UserApiKeyCache, response_cost: float) -> None: await update_cache( @@ -56,3 +153,63 @@ async def test_global_proxy_spend_adds_every_instance_cost_and_stays_cleared_aft assert await redis_cache.async_get_cache(key=GLOBAL_PROXY_SPEND_CACHE_KEY) is None finally: await redis_cache.async_delete_cache(key=GLOBAL_PROXY_SPEND_CACHE_KEY) + + +@pytest.mark.timeout(OWNED_PROXY_SECONDS) +def test_a_key_with_no_user_spends_toward_the_proxy_max_budget(gateway: Gateway, tmp_path: Path) -> None: + with ( + owned_proxy(gateway, tmp_path, {}, config=_proxy_budget_config(gateway, tmp_path, max_budget=0.01)) as owned, + owned.scenario() as scenario, + ): + key: Final = scenario.key(models=[BUDGET_MODEL]) + (spent,) = _logged_spend(owned, (_served_id(owned, key),)) + assert spent > 0.01, spent + eventually( + lambda: _proxy_budget_spend(owned), + lambda recorded: recorded is not None and recorded > 0.01, + seconds=70, + ) + + marker: Final = f"over the proxy budget {uuid.uuid4().hex}" + refused: Final = _chat(owned, key, marker) + + assert refused.status_code != 200, refused.text + assert object_value(refused.json()["error"])["type"] == "budget_exceeded", refused.text + assert not _upstream_saw(owned, marker), refused.text + + +@pytest.mark.timeout(2 * OWNED_PROXY_SECONDS) +def test_two_proxies_add_every_request_to_one_proxy_spend_counter(gateway: Gateway, tmp_path: Path) -> None: + config: Final = _proxy_budget_config(gateway, tmp_path, max_budget=1000.0) + with owned_redis(tmp_path / "redis") as cache, Redis(host=cache.host, port=cache.port) as redis: + shared_redis: Final = {"REDIS_HOST": cache.host, "REDIS_PORT": str(cache.port)} + with owned_proxy(gateway, tmp_path / "first", shared_redis, config=config) as first: + assert first.database_url is not None + with ( + owned_proxy( + gateway, tmp_path / "second", {**shared_redis, "DATABASE_URL": first.database_url}, config=config + ) as second, + first.scenario() as scenario, + ): + user_key: Final = scenario.key(user_id=scenario.user(), models=[BUDGET_MODEL]) + userless_key: Final = scenario.key(models=[BUDGET_MODEL]) + counter: Final = _counter_key(redis) + ttl_before: Final = redis.ttl(counter) + + request_ids: Final = tuple( + _served_id(proxy, key) for key, proxy in product((user_key, userless_key), (first, second)) + ) + total: Final = sum(_logged_spend(first, request_ids)) + persisted: Final = eventually( + lambda: _proxy_budget_spend(first), + lambda spend: spend is not None and spend == pytest.approx(total), + seconds=70, + ) + + assert persisted == pytest.approx(total) + eventually( + lambda: redis.get(counter), + lambda shared: shared is not None and float(str(shared, "utf-8")) == pytest.approx(total), + seconds=70, + ) + assert 0 < redis.ttl(counter) <= ttl_before From 07d314beb598ae4cb5d263fda3e1ecc531633112 Mon Sep 17 00:00:00 2001 From: tavren27 <348969968+tavren27@users.noreply.github.com> Date: Wed, 28 Oct 2026 18:05:08 -0700 Subject: [PATCH 36/41] test(responses): audit #47445 (#48902) * test(responses): audit #47445 Integration cells for billing every round of a streamed MCP auto-execute on one spend row, on a native Responses upstream and the chat bridge, across tool rounds, a rejected or broken follow-up, a post-call guardrail block, a client hang-up, the response cache, and the next bridged turn * test(responses): check a broken follow-up bills its own nonzero usage --------- Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com> --- .../integration/mcp/test_mcp_llm_endpoints.py | 365 +++++++++++++++++- 1 file changed, 351 insertions(+), 14 deletions(-) diff --git a/tests/integration/mcp/test_mcp_llm_endpoints.py b/tests/integration/mcp/test_mcp_llm_endpoints.py index 394221708bf..f751d1f8f4c 100644 --- a/tests/integration/mcp/test_mcp_llm_endpoints.py +++ b/tests/integration/mcp/test_mcp_llm_endpoints.py @@ -6,7 +6,7 @@ from concurrent.futures import ThreadPoolExecutor from contextlib import contextmanager from dataclasses import dataclass from hashlib import sha256 -from itertools import chain +from itertools import chain, count from pathlib import Path from typing import Final, Literal, TypeVar @@ -50,6 +50,7 @@ def _llm_gateway(tmp_path_factory: pytest.TempPathFactory) -> Iterator[tuple[Gat directory: Final = tmp_path_factory.mktemp("mcp-llm-gateway") stock: Final = JSON_OBJECT.validate_python(yaml.safe_load(Path("tests/integration/proxy_config.yaml").read_text())) config: Final = directory / "stored-responses.yaml" + (directory / f"{STREAM_BLOCK_MODULE}.py").write_text(STREAM_BLOCK_SOURCE) config.write_text( yaml.safe_dump( { @@ -57,8 +58,17 @@ def _llm_gateway(tmp_path_factory: pytest.TempPathFactory) -> Iterator[tuple[Gat "general_settings": {**object_value(stock["general_settings"]), "store_prompts_in_spend_logs": True}, "litellm_settings": { **object_value(stock["litellm_settings"]), - "cost_margin_config": {"hosted_vllm": {"fixed_amount": FIXED_MARGIN}}, + "cost_margin_config": { + "hosted_vllm": {"fixed_amount": FIXED_MARGIN}, + "deepseek": {"fixed_amount": FIXED_MARGIN}, + }, }, + "guardrails": [ + { + "guardrail_name": STREAM_BLOCK, + "litellm_params": {"guardrail": f"{STREAM_BLOCK_MODULE}.BlockGuard", "mode": "post_call"}, + } + ], } ) ) @@ -79,7 +89,7 @@ def gateway(_llm_gateway: tuple[Gateway, Mapping[str, str]]) -> Iterator[Gateway yield private -Surface = Literal["chat", "responses", "messages", "messages_bridge"] +Surface = Literal["chat", "responses", "messages", "messages_bridge", "responses_bridge"] SURFACES: Final[tuple[Surface, ...]] = ("chat", "responses", "messages", "messages_bridge") ADD: Final = {"a": 2, "b": 3} ANSWER: Final = "the sum is 5" @@ -88,6 +98,19 @@ AUTO: Final = {**GATEWAY_REF, "require_approval": "never"} OUTAGE: Final = "bridge-outage" DIRECT: Final = "answer-without-tools" FIXED_MARGIN: Final = 0.25 +STREAM_BLOCK: Final = "mcp-stream-block" +STREAM_BLOCK_MODULE: Final = "mcp_stream_block" +STREAM_BLOCK_DETAIL: Final = "blocked by output policy" +STREAM_BLOCK_SOURCE: Final = ( + "from fastapi import HTTPException\n\n" + "from litellm.integrations.custom_guardrail import CustomGuardrail\n\n\n" + "class BlockGuard(CustomGuardrail):\n" + " async def async_post_call_streaming_iterator_hook(self, user_api_key_dict, response, request_data):\n" + " async for _ in response:\n" + " pass\n" + f' raise HTTPException(status_code=400, detail="{STREAM_BLOCK_DETAIL}")\n' + " yield\n" +) JSON_OBJECT: Final = TypeAdapter(dict[str, JsonValue]) JSON_VALUE: Final[TypeAdapter[JsonValue]] = TypeAdapter(JsonValue) @@ -128,6 +151,8 @@ class Billing: first: tuple[int, int] = (10, 5) follow_up: tuple[int, int] = (10, 5) follow_up_fails: bool = False + follow_up_breaks: bool = False + tool_rounds: int = 1 EVEN_ROUNDS: Final = Billing() @@ -138,6 +163,10 @@ def _fixed_turn(tool: str) -> Callable[[Mapping[str, JsonValue]], Turn]: return lambda _: Turn(tool, json.dumps(ADD), ANSWER) +def _slow_turn(tool: str, seconds: float) -> Callable[[Mapping[str, JsonValue]], Turn]: + return lambda _: Turn(tool, json.dumps({"seconds": seconds}), ANSWER) + + def _echoing_turn(body: Mapping[str, JsonValue]) -> Turn: names: Final = _tool_names(body) return Turn(names[0] if names else "", json.dumps({"query": _prompt(body)}), _tool_result_text(body) or "") @@ -183,6 +212,50 @@ def _item_events(index: int, item: Mapping[str, JsonValue]) -> tuple[dict[str, J ) +def _event_stream(events: Sequence[Mapping[str, JsonValue]]) -> Reply: + return Reply( + content_type="text/event-stream", + chunks=tuple( + f"event: {event['type']}\ndata: {json.dumps({**event, 'sequence_number': number})}\n\n".encode() + for number, event in enumerate(events) + ), + ) + + +def _broken_follow_up(target: str, identity: str, tokens: tuple[int, int]) -> Reply: + outage: Final = {"message": FOLLOW_UP_OUTAGE, "type": "server_error", "code": None} + if target.endswith("/chat/completions"): + delta: Final = {"role": "assistant", "content": "the sum"} + started: Final = { + "id": f"chatcmpl-{identity}", + "object": "chat.completion.chunk", + "created": 1, + "model": "gpt-4o-mini", + "choices": [{"index": 0, "delta": delta, "finish_reason": None}], + } + return Reply( + content_type="text/event-stream", + chunks=tuple(f"data: {json.dumps(frame)}\n\n".encode() for frame in (started, {"error": outage})), + ) + input_tokens, output_tokens = tokens + failed: Final[dict[str, JsonValue]] = { + "id": f"resp_{identity}", + "object": "response", + "created_at": 1, + "status": "failed", + "model": "gpt-4o-mini", + "output": [], + "error": {"code": "server_error", "message": FOLLOW_UP_OUTAGE}, + "usage": {"input_tokens": input_tokens, "output_tokens": output_tokens, "total_tokens": sum(tokens)}, + } + return _event_stream( + ( + {"type": "response.created", "response": {**failed, "status": "in_progress", "error": None, "usage": None}}, + {"type": "response.failed", "response": failed}, + ) + ) + + def _responses_stream(response: Mapping[str, JsonValue], items: Sequence[Mapping[str, JsonValue]]) -> Reply: item_events: Final = tuple(chain.from_iterable(_item_events(index, item) for index, item in enumerate(items))) events: Final = ( @@ -191,13 +264,7 @@ def _responses_stream(response: Mapping[str, JsonValue], items: Sequence[Mapping *item_events, {"type": "response.completed", "response": response}, ) - return Reply( - content_type="text/event-stream", - chunks=tuple( - f"event: {event['type']}\ndata: {json.dumps({**event, 'sequence_number': number})}\n\n".encode() - for number, event in enumerate(events) - ), - ) + return _event_stream(events) def _block_ids(message: JsonValue, kind: str, field: str) -> frozenset[str]: @@ -252,6 +319,8 @@ def _anthropic_tool_result_rule(body: Mapping[str, JsonValue]) -> Reply | None: def _model_double( plan: Callable[[Mapping[str, JsonValue]], Turn], billing: Billing = EVEN_ROUNDS, parallel: int = 1 ) -> Callable[[Request], Reply]: + served: Final = count() + def respond(request: Request) -> Reply: if request.method == "GET" and request.target.endswith("/models"): return _json({"object": "list", "data": []}) @@ -260,11 +329,12 @@ def _model_double( outage: Final = {"error": {"message": "scripted provider outage", "type": "server_error", "code": None}} return Reply(status=500, body=json.dumps(outage).encode()) turn: Final = plan(body) + call_number: Final = next(served) answered: Final = _has_tool_result(body) if answered and billing.follow_up_fails: failure: Final = {"error": {"message": FOLLOW_UP_OUTAGE, "type": "server_error", "code": None}} return Reply(status=500, body=json.dumps(failure).encode()) - done: Final = answered or DIRECT in _prompt(body) + done: Final = DIRECT in _prompt(body) or (answered and call_number >= billing.tool_rounds) calls: Final = range(1, parallel + 1) prompt_tokens, completion_tokens = billing.follow_up if answered else billing.first identity: Final = uuid.uuid4().hex[:12] @@ -273,9 +343,15 @@ def _model_double( "completion_tokens": completion_tokens, "total_tokens": prompt_tokens + completion_tokens, } + if answered and billing.follow_up_breaks: + return _broken_follow_up(request.target, identity, (prompt_tokens, completion_tokens)) if request.target.endswith("/chat/completions"): requested: Final = tuple( - {"id": f"call_{call}", "type": "function", "function": {"name": turn.tool, "arguments": turn.arguments}} + { + "id": f"call_{identity}_{call}", + "type": "function", + "function": {"name": turn.tool, "arguments": turn.arguments}, + } for call in calls ) message: Final = ( @@ -460,6 +536,7 @@ def _upstream_model(surface: Surface) -> str: "responses": "openai/responses/gpt-4o-mini", "messages": "anthropic/claude-sonnet-4-5", "messages_bridge": "hosted_vllm/gpt-4o-mini", + "responses_bridge": "deepseek/gpt-4o-mini", }[surface] @@ -475,11 +552,15 @@ def _rig( billing: Billing = EVEN_ROUNDS, prices: Mapping[str, JsonValue] | None = None, parallel: int = 1, + slow_tool_seconds: float | None = None, ) -> Iterator[Rig]: alias: Final = "llm" + uuid.uuid4().hex[:8] + plan: Final = ( + _fixed_turn(f"{alias}-add") if slow_tool_seconds is None else _slow_turn(f"{alias}-slow", slow_tool_seconds) + ) with ( - mcp_peer() as peer, - wire_server(_model_double(_fixed_turn(f"{alias}-add"), billing, parallel)) as wire, + mcp_peer(rich=slow_tool_seconds is not None) as peer, + wire_server(_model_double(plan, billing, parallel)) as wire, gateway.scenario() as scenario, ): server_id: Final = register_mcp(scenario, peer, alias) @@ -1048,6 +1129,262 @@ def test_bridged_next_turn_replays_the_mcp_turn_answer(gateway: Gateway) -> None assert ANSWER in _assistant_texts(opening), opening +StreamedUpstream = Literal["messages_bridge", "responses_bridge"] +STREAMED_UPSTREAMS: Final[tuple[StreamedUpstream, ...]] = ("messages_bridge", "responses_bridge") +STREAMED_IDS: Final = ("native", "bridged") +SLOW_TOOL_SECONDS: Final = 5.0 + + +def _events(response: httpx.Response) -> tuple[dict[str, JsonValue], ...]: + return tuple( + JSON_OBJECT.validate_json(line.removeprefix("data: ")) + for line in response.text.splitlines() + if line.startswith("data: ") and line != "data: [DONE]" + ) + + +def _completed(response: httpx.Response) -> dict[str, JsonValue]: + finished: Final = tuple( + object_value(event["response"]) for event in _events(response) if event["type"] == "response.completed" + ) + assert len(finished) == 1, response.text + return finished[0] + + +def _model_rows(call_id: str, settled: Callable[[list[dict[str, JsonValue]]], bool]) -> list[dict[str, JsonValue]]: + return eventually( + lambda: read_rows( + "SELECT request_id, status, spend, prompt_tokens, completion_tokens " + 'FROM "LiteLLM_SpendLogs" WHERE litellm_call_id=%s AND call_type NOT IN (%s, %s)', + (call_id, "call_mcp_tool", "list_mcp_tools"), + ), + settled, + seconds=70, + return_last_on_timeout=True, + ) + + +def _billed(rows: Sequence[Mapping[str, JsonValue]]) -> float: + return sum(float(str(row["spend"])) for row in rows) + + +def _bills_at_least(expected: float) -> Callable[[list[dict[str, JsonValue]]], bool]: + return lambda rows: bool(rows) and _billed(rows) >= expected - 1e-9 + + +def _statuses(rows: Sequence[Mapping[str, JsonValue]]) -> list[str]: + return sorted(str(row["status"]) for row in rows) + + +def _logs_a_failure_billing(expected: float) -> Callable[[list[dict[str, JsonValue]]], bool]: + return lambda rows: "failure" in _statuses(rows) and _billed(rows) >= expected - 1e-9 + + +def _limited_key(rig: Rig) -> str: + return rig.scenario.key(object_permission={"mcp_servers": [rig.server_id]}, max_parallel_requests=1) + + +def _answers_directly(rig: Rig, key: str) -> int: + return rig.send(key, [AUTO], prompt=f"{DIRECT} {rig.alias}").status_code + + +STREAMED_ROUNDS: Final = ( + pytest.param("messages_bridge", 1, id="native"), + pytest.param("responses_bridge", 1, id="bridged"), + pytest.param("messages_bridge", 2, id="native-two-tool-rounds"), + pytest.param("responses_bridge", 2, id="bridged-two-tool-rounds"), + pytest.param("messages_bridge", 0, id="native-no-tool-call"), + pytest.param("responses_bridge", 0, id="bridged-no-tool-call"), +) + + +@pytest.mark.parametrize(("upstream", "tool_rounds"), STREAMED_ROUNDS) +def test_streaming_responses_auto_execute_bills_every_round_once_on_one_success_row( + gateway: Gateway, upstream: StreamedUpstream, tool_rounds: int +) -> None: + rounds: Final = (FIRST_ROUND, *(FOLLOW_UP_ROUND,) * tool_rounds) + billing: Final = Billing(FIRST_ROUND, FOLLOW_UP_ROUND, tool_rounds=max(tool_rounds, 1)) + with _rig(gateway, "responses", upstream=upstream, billing=billing, prices=PRICES) as rig: + prompt: Final = None if tool_rounds else f"{DIRECT} {rig.alias}" + response: Final = rig.send(_granted_key(rig), [AUTO], prompt=prompt, stream=True) + assert response.status_code == 200, response.text + completed: Final = _completed(response) + assert rig.final_text(completed) == ANSWER, response.text + assert len(rig.upstream_tools()) == len(rounds) + assert len(_peer_add_calls(rig.peer)) == tool_rounds + expected: Final = _price(*rounds) + FIXED_MARGIN + rows: Final = _model_rows(response.headers["x-litellm-call-id"], _bills_at_least(expected)) + assert [row["status"] for row in rows] == ["success"], rows + row: Final = rows[0] + assert (row["prompt_tokens"], row["completion_tokens"]) == ( + sum(prompt_tokens for prompt_tokens, _ in rounds), + sum(completion_tokens for _, completion_tokens in rounds), + ), row + usage: Final = object_value(completed["usage"]) + assert (usage["input_tokens"], usage["output_tokens"]) == (row["prompt_tokens"], row["completion_tokens"]) + assert float(str(row["spend"])) == pytest.approx(expected), row + + +@pytest.mark.parametrize("upstream", STREAMED_UPSTREAMS, ids=STREAMED_IDS) +def test_streaming_responses_rejected_follow_up_bills_round_one_once( + gateway: Gateway, upstream: StreamedUpstream +) -> None: + billing: Final = Billing(FIRST_ROUND, FOLLOW_UP_ROUND, follow_up_fails=True) + with _rig(gateway, "responses", upstream=upstream, billing=billing, prices=PRICES) as rig: + response: Final = rig.send(_granted_key(rig), [AUTO], stream=True) + assert FOLLOW_UP_OUTAGE in response.text, response.text + assert ANSWER not in response.text, response.text + assert len(rig.upstream_tools()) == 2 + expected: Final = _price(FIRST_ROUND) + FIXED_MARGIN + rows: Final = _model_rows(response.headers["x-litellm-call-id"], _bills_at_least(expected)) + assert _billed(rows) == pytest.approx(expected), rows + + +BROKEN_FOLLOW_UPS: Final = ( + pytest.param("messages_bridge", FOLLOW_UP_ROUND, id="native"), + pytest.param("responses_bridge", None, id="bridged"), +) + + +@pytest.mark.parametrize(("upstream", "reported_usage"), BROKEN_FOLLOW_UPS) +def test_streaming_responses_follow_up_broken_mid_stream_bills_round_one_once_and_itself_on_a_failure_row( + gateway: Gateway, upstream: StreamedUpstream, reported_usage: tuple[int, int] | None +) -> None: + billing: Final = Billing(FIRST_ROUND, FOLLOW_UP_ROUND, follow_up_breaks=True) + with _rig(gateway, "responses", upstream=upstream, billing=billing, prices=PRICES) as rig: + response: Final = rig.send(_granted_key(rig), [AUTO], stream=True) + assert ANSWER not in response.text, response.text + assert len(rig.upstream_tools()) == 2 + round_one: Final = _price(FIRST_ROUND) + FIXED_MARGIN + rows: Final = _model_rows(response.headers["x-litellm-call-id"], _logs_a_failure_billing(round_one)) + assert _statuses(rows) == ["failure", "success"], rows + (success,) = (row for row in rows if row["status"] == "success") + (failure,) = (row for row in rows if row["status"] == "failure") + assert (success["prompt_tokens"], success["completion_tokens"]) == FIRST_ROUND, rows + assert float(str(success["spend"])) == pytest.approx(round_one), rows + failed_round: Final = (int(str(failure["prompt_tokens"])), int(str(failure["completion_tokens"]))) + assert min(failed_round) > 0, rows + assert reported_usage in (None, failed_round), rows + assert float(str(failure["spend"])) == pytest.approx(_price(failed_round)), rows + + +@pytest.mark.parametrize("upstream", STREAMED_UPSTREAMS, ids=STREAMED_IDS) +def test_streaming_responses_blocked_by_a_post_call_guardrail_bills_every_round_once( + gateway: Gateway, upstream: StreamedUpstream +) -> None: + billing: Final = Billing(FIRST_ROUND, FOLLOW_UP_ROUND) + with _rig(gateway, "responses", upstream=upstream, billing=billing, prices=PRICES) as rig: + key: Final = _granted_key(rig) + one_round: Final = rig.send(key, [], prompt=f"{DIRECT} {rig.alias}", stream=True, guardrails=[STREAM_BLOCK]) + assert STREAM_BLOCK_DETAIL in one_round.text, one_round.text + one_round_rows: Final = _model_rows( + one_round.headers["x-litellm-call-id"], _logs_a_failure_billing(_price(FIRST_ROUND) + FIXED_MARGIN) + ) + rig.wire.drain() + response: Final = rig.send(key, [AUTO], stream=True, guardrails=[STREAM_BLOCK]) + assert STREAM_BLOCK_DETAIL in response.text, response.text + assert len(rig.upstream_tools()) == 2 and len(_peer_add_calls(rig.peer)) == 1 + expected: Final = _price(FIRST_ROUND, FOLLOW_UP_ROUND) + FIXED_MARGIN + rows: Final = _model_rows(response.headers["x-litellm-call-id"], _logs_a_failure_billing(expected)) + assert _statuses(rows) == _statuses(one_round_rows), (rows, one_round_rows) + assert _billed(rows) == pytest.approx(expected), rows + + +@pytest.mark.parametrize( + ("surface", "stream"), + (("responses", True), ("responses", False), ("chat", False)), + ids=("responses-stream", "responses", "chat"), +) +def test_auto_executed_tool_call_runs_inside_the_requests_one_parallel_slot( + gateway: Gateway, surface: Surface, stream: bool +) -> None: + with _rig(gateway, surface, slow_tool_seconds=SLOW_TOOL_SECONDS) as rig: + key: Final = _limited_key(rig) + with ThreadPoolExecutor(max_workers=1) as pool: + running: Final = pool.submit(rig.send, key, [AUTO], None, stream=stream) + eventually(lambda: _peer_add_calls(rig.peer), lambda found: len(found) == 1, seconds=30) + during: Final = _answers_directly(rig, key) + served: Final = running.result(timeout=90) + assert during == 429 + assert served.status_code == 200 and ANSWER in served.text, served.text + eventually(lambda: _answers_directly(rig, key), lambda status: status == 200, seconds=30) + + +@pytest.mark.parametrize("upstream", STREAMED_UPSTREAMS, ids=STREAMED_IDS) +def test_streaming_responses_hung_up_during_the_tool_call_bills_each_served_round_once_and_frees_the_slot( + gateway: Gateway, upstream: StreamedUpstream +) -> None: + billing: Final = Billing(FIRST_ROUND, FOLLOW_UP_ROUND) + with _rig( + gateway, + "responses", + upstream=upstream, + billing=billing, + prices=PRICES, + slow_tool_seconds=SLOW_TOOL_SECONDS, + ) as rig: + key: Final = _limited_key(rig) + with rig.gateway.client.stream( + "POST", + "/v1/responses", + headers={"Authorization": f"Bearer {key}"}, + json={"model": rig.model, "input": f"add {rig.alias}", "tools": [AUTO], "stream": True}, + timeout=90, + ) as hung_up: + assert hung_up.status_code == 200 + call_id: Final = hung_up.headers["x-litellm-call-id"] + eventually(lambda: _peer_add_calls(rig.peer), lambda found: len(found) == 1, seconds=30) + eventually(lambda: _answers_directly(rig, key), lambda status: status == 200, seconds=60) + posted: Final = tuple( + JSON_OBJECT.validate_json(request.body) for request in rig.wire.drain() if request.method == "POST" + ) + served: Final = tuple( + FOLLOW_UP_ROUND if _has_tool_result(body) else FIRST_ROUND for body in posted if DIRECT not in _prompt(body) + ) + expected: Final = _price(*served) + FIXED_MARGIN + rows: Final = _model_rows(call_id, _bills_at_least(expected)) + assert served and rows, (served, rows) + assert _billed(rows) == pytest.approx(expected), (served, rows) + + +@pytest.mark.parametrize("upstream", STREAMED_UPSTREAMS, ids=STREAMED_IDS) +def test_streaming_responses_repeated_from_the_response_cache_bills_nothing( + gateway: Gateway, upstream: StreamedUpstream +) -> None: + billing: Final = Billing(FIRST_ROUND, FOLLOW_UP_ROUND) + with _rig(gateway, "responses", upstream=upstream, billing=billing, prices=PRICES) as rig: + key: Final = _granted_key(rig) + paid: Final = rig.send(key, [AUTO], stream=True) + assert paid.status_code == 200 and rig.final_text(_completed(paid)) == ANSWER, paid.text + paid_price: Final = _price(FIRST_ROUND, FOLLOW_UP_ROUND) + FIXED_MARGIN + _model_rows(paid.headers["x-litellm-call-id"], _bills_at_least(paid_price)) + assert len(rig.upstream_tools()) == 2 + cached: Final = rig.send(key, [AUTO], stream=True) + assert cached.status_code == 200 and rig.final_text(_completed(cached)) == ANSWER, cached.text + assert rig.upstream_tools() == () + rows: Final = _model_rows( + cached.headers["x-litellm-call-id"], lambda found: "success" in {row["status"] for row in found} + ) + assert "success" in {row["status"] for row in rows} and _billed(rows) == 0.0, rows + + +def test_bridged_next_turn_after_a_streamed_mcp_turn_replays_its_answer(gateway: Gateway) -> None: + billing: Final = Billing(FIRST_ROUND, FOLLOW_UP_ROUND) + with _rig(gateway, "responses", upstream="responses_bridge", billing=billing, prices=PRICES) as rig: + key: Final = _granted_key(rig) + first: Final = rig.send(key, [AUTO], stream=True) + assert first.status_code == 200, first.text + first_id: Final = string_value(_completed(first)["id"]) + _model_rows(first.headers["x-litellm-call-id"], _bills_at_least(_price(FIRST_ROUND, FOLLOW_UP_ROUND))) + rig.wire.drain() + second: Final = rig.send(key, [AUTO], previous_response_id=first_id, cache={"no-cache": True}) + assert second.status_code == 200, second.text + opening: Final = next( + JSON_OBJECT.validate_json(request.body) for request in rig.wire.drain() if request.method == "POST" + ) + assert ANSWER in _assistant_texts(opening), opening + + Bridge = Literal["chat", "responses", "messages"] BRIDGES: Final[tuple[Bridge, ...]] = ("chat", "responses") Client = Literal["sync", "async"] From 3b9a59348f564d7746cb05a9e16341f42321a219 Mon Sep 17 00:00:00 2001 From: zoveki <349028329+zoveki@users.noreply.github.com> Date: Wed, 28 Oct 2026 18:06:58 -0700 Subject: [PATCH 37/41] fix(vertex_ai)!: send the caller's anthropic-version in the Vertex Claude /v1/messages body, not the header (#48814) * fix(vertex_ai): drop the caller's anthropic-version header on Claude /v1/messages Vertex reads the API version from the body's anthropic_version and answers 400 when a forwarded anthropic-version header names a different one, which anthropic SDK 2.x always does. Strip the header on the Python and LITELLM_RUST Vertex messages routes. * fix(vertex_ai): carry the caller's anthropic-version into the Claude /v1/messages body Vertex reads the API version from the body alone, so the caller's anthropic-version header now becomes the body's anthropic_version instead of a pinned vertex-2023-10-16. The 2023-01-01 and 2023-06-01 first-party versions keep vertex-2023-10-16, and the header itself is still never sent. Same change on the Python proxy and the Rust gateway. * fix(vertex_ai): follow drop_params for an anthropic-version Vertex does not accept Only versions Vertex accepts in the body (vertex-2023-10-16, 2026-10-21) are carried into anthropic_version. 2023-06-01, 2023-01-01, a blank header, or no header keep vertex-2023-10-16. Any other value drops to vertex-2023-10-16 under drop_params and answers a 400 naming drop_params without it, on the Python proxy and the Rust gateway. * refactor(vertex_ai): return the unsupported anthropic-version error as a value and keep Rust helpers and tests in place * fix(rust gateway): resolve os.environ drop_params and accept the same flag words as Python The gateway read drop_params: os.environ/NAME and words like yes as off, so the drop_params opt-out from the new anthropic-version 400 did not work there * docs(rust router): name Router::from_config in the crate README --------- Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com> --- litellm-rust/Cargo.lock | 1 + .../crates/core-utils/src/serde_compat.rs | 41 ++++- litellm-rust/crates/gateway/src/lib.rs | 8 +- .../crates/inference-messages/src/handler.rs | 88 ++++++++- .../src/base_llm/messages/transformation.rs | 9 + .../llms/src/vertex_ai/messages/AGENTS.md | 3 +- .../src/vertex_ai/messages/transformation.rs | 173 ++++++++++++++++-- litellm-rust/crates/router/Cargo.toml | 1 + litellm-rust/crates/router/README.md | 2 +- litellm-rust/crates/router/src/lib.rs | 88 ++++++++- litellm-rust/crates/router/tests/router.rs | 8 +- .../anthropic/messages/transformation.py | 59 +++++- ...artner_models_anthropic_messages_config.py | 83 ++++++++- 13 files changed, 522 insertions(+), 42 deletions(-) diff --git a/litellm-rust/Cargo.lock b/litellm-rust/Cargo.lock index 10e590b5006..66a345fc438 100644 --- a/litellm-rust/Cargo.lock +++ b/litellm-rust/Cargo.lock @@ -4387,6 +4387,7 @@ name = "litellm-router" version = "0.1.0" dependencies = [ "litellm-config", + "litellm-core-utils", "litellm-inference", "litellm-inference-messages", "litellm-router-types", diff --git a/litellm-rust/crates/core-utils/src/serde_compat.rs b/litellm-rust/crates/core-utils/src/serde_compat.rs index 3e4d82d3a3e..6f40a4af60a 100644 --- a/litellm-rust/crates/core-utils/src/serde_compat.rs +++ b/litellm-rust/crates/core-utils/src/serde_compat.rs @@ -1,13 +1,25 @@ -pub fn parse_str_bool(value: &str) -> Option { - let token = value.trim_matches(|character: char| { +fn python_strip(value: &str) -> &str { + value.trim_matches(|character: char| { character.is_whitespace() || matches!(character, '\u{1c}'..='\u{1f}') - }); + }) +} + +pub fn parse_str_bool(value: &str) -> Option { + let token = python_strip(value); if token.eq_ignore_ascii_case("true") { return Some(true); } token.eq_ignore_ascii_case("false").then_some(false) } +pub fn parse_pydantic_bool(value: &str) -> Option { + match python_strip(value).to_ascii_lowercase().as_str() { + "1" | "on" | "t" | "true" | "y" | "yes" => Some(true), + "0" | "off" | "f" | "false" | "n" | "no" => Some(false), + _ => None, + } +} + /// `redis-py` string Booleans: only `1`, `true`, and `yes` (case-insensitive) are true. pub fn parse_redis_bool(value: &str) -> bool { value == "1" || value.eq_ignore_ascii_case("true") || value.eq_ignore_ascii_case("yes") @@ -34,4 +46,27 @@ mod tests { ) { assert_eq!(parse_str_bool(input), expected, "{input:?}"); } + + #[rstest] + #[case::yes(" YES ", Some(true))] + #[case::on("On", Some(true))] + #[case::one("1", Some(true))] + #[case::t("t", Some(true))] + #[case::y("y", Some(true))] + #[case::control_whitespace_true("\u{1c}TRUE\u{1f}", Some(true))] + #[case::no("No", Some(false))] + #[case::off("OFF", Some(false))] + #[case::zero("0", Some(false))] + #[case::f("F", Some(false))] + #[case::n("n", Some(false))] + #[case::false_word("false", Some(false))] + #[case::two("2", None)] + #[case::empty("", None)] + #[case::unknown("maybe", None)] + fn boolean_words_follow_pydantic_lax_bool_parsing( + #[case] input: &str, + #[case] expected: Option, + ) { + assert_eq!(parse_pydantic_bool(input), expected, "{input:?}"); + } } diff --git a/litellm-rust/crates/gateway/src/lib.rs b/litellm-rust/crates/gateway/src/lib.rs index b8ff8e46033..12a9bf9110a 100644 --- a/litellm-rust/crates/gateway/src/lib.rs +++ b/litellm-rust/crates/gateway/src/lib.rs @@ -37,6 +37,7 @@ pub fn build_inference(config: &Config) -> Result, Error> { .map(|value| value.expose().to_owned()) .or_else(|| std::env::var(name).ok()) }; + let router = ModelRouter::from_config(config, &lookup); let settings = &config.litellm_settings; let http_settings = HttpSettings::from_layers([ HttpSettingsLayer::from_environment(&lookup), @@ -64,12 +65,7 @@ pub fn build_inference(config: &Config) -> Result, Error> { Arc::new(EnvironmentSecrets::python_compatible(client)), )); let resources = CoreResources::new(pool); - Ok(Arc::new(Gateway::new( - resources, - http, - secrets, - ModelRouter::from_model_list(&config.model_list), - )?)) + Ok(Arc::new(Gateway::new(resources, http, secrets, router)?)) } pub async fn build_redis(config: &Config) -> Result { diff --git a/litellm-rust/crates/inference-messages/src/handler.rs b/litellm-rust/crates/inference-messages/src/handler.rs index fa802d20587..9d0bf09197e 100644 --- a/litellm-rust/crates/inference-messages/src/handler.rs +++ b/litellm-rust/crates/inference-messages/src/handler.rs @@ -141,9 +141,13 @@ async fn provider_call( &env_lookup, )?; - let body = config.transform_anthropic_messages_request( - trimmed, - &MessagesTransformContext::new(shaping.capabilities, shaping.settings.drop_params), + let body = config.request_body( + config.transform_anthropic_messages_request( + trimmed, + &MessagesTransformContext::new(shaping.capabilities, shaping.settings.drop_params), + )?, + &validated.headers, + shaping.settings.drop_params, )?; let environment = ValidatedEnvironment { headers: config.request_headers( @@ -514,6 +518,84 @@ mod tests { ); } + fn vertex_call_requesting( + shaping: MessagesShaping, + requested: &str, + drop_params: bool, + ) -> MessagesCall { + MessagesCall { + body: body( + json!({"model": "claude-sonnet-4-5", "messages": [{"role": "user", "content": "hi"}], "max_tokens": 16}), + ), + api_key: None, + api_base: Some("https://vertex.test".into()), + custom_llm_provider: Some("vertex_ai".into()), + litellm_params: Default::default(), + extra_headers: Some(Map::from_iter([ + ("authorization".to_string(), json!("Bearer vertex-token")), + ("anthropic-version".to_string(), json!(requested)), + ])), + provider_specific_header: None, + timeout: None, + shaping: MessagesShaping { + settings: MessagesSettings { + drop_params, + ..shaping.settings + }, + ..shaping + }, + } + } + + fn vertex_project(name: &str) -> Option { + (name == "VERTEXAI_PROJECT").then(|| "version-project".to_string()) + } + + #[rstest] + #[case::sdk_2("2026-10-21", false, "2026-10-21")] + #[case::sdk_1("2023-06-01", false, "vertex-2023-10-16")] + #[case::unknown_under_drop_params("latest", true, "vertex-2023-10-16")] + #[tokio::test] + async fn vertex_gets_the_requested_version_in_the_body_alone( + shaping: MessagesShaping, + #[case] requested: &str, + #[case] drop_params: bool, + #[case] body_version: &str, + ) { + let call = provider_call_with_secrets( + vertex_call_requesting(shaping, requested, drop_params), + &vertex_project, + ) + .await + .unwrap(); + let version_headers: Vec<&str> = call + .wire + .headers + .iter() + .filter(|(name, _)| name.eq_ignore_ascii_case("anthropic-version")) + .map(|(_, value)| value.as_str()) + .collect(); + assert_eq!( + (version_headers, &call.wire.body["anthropic_version"]), + (Vec::<&str>::new(), &json!(body_version)) + ); + } + + #[rstest] + #[tokio::test] + async fn vertex_refuses_an_unsupported_version_without_drop_params(shaping: MessagesShaping) { + assert!(matches!( + provider_call_with_secrets( + vertex_call_requesting(shaping, "latest", false), + &vertex_project + ) + .await, + Err(Error::Llm(litellm_llms::Error::InvalidRequest( + litellm_llms::ErrorDetail::UnsupportedParameter { .. } + ))) + )); + } + #[rstest] #[tokio::test] async fn missing_credentials_are_reported_before_the_request_is_transformed( diff --git a/litellm-rust/crates/llms/src/base_llm/messages/transformation.rs b/litellm-rust/crates/llms/src/base_llm/messages/transformation.rs index 8a88af40d6b..744ce89afdb 100644 --- a/litellm-rust/crates/llms/src/base_llm/messages/transformation.rs +++ b/litellm-rust/crates/llms/src/base_llm/messages/transformation.rs @@ -77,6 +77,15 @@ pub trait BaseMessagesConfig: Sync { headers } + fn request_body( + &self, + request: MessagesRequest, + _headers: &Headers, + _drop_params: bool, + ) -> Result { + Ok(request) + } + /// The JSON that goes on the wire, for a host whose body differs from the typed request: /// Python's configs `pop("model")` when the model is addressed by the URL. fn wire_body(&self, body: Value) -> Value { diff --git a/litellm-rust/crates/llms/src/vertex_ai/messages/AGENTS.md b/litellm-rust/crates/llms/src/vertex_ai/messages/AGENTS.md index b5d637d6bc7..d95da65ac9d 100644 --- a/litellm-rust/crates/llms/src/vertex_ai/messages/AGENTS.md +++ b/litellm-rust/crates/llms/src/vertex_ai/messages/AGENTS.md @@ -1,6 +1,7 @@ # rules -- Claude on Vertex AI: reuses `anthropic/messages` shaping, moves the model into the publisher URL, puts the Vertex `anthropic_version` in the body, and authenticates with a Google access token +- Claude on Vertex AI: reuses `anthropic/messages` shaping, moves the model into the publisher URL, and authenticates with a Google access token +- Vertex reads the API version only from the body's `anthropic_version` and answers 400 when an `anthropic-version` header names a different one, so the caller's header never reaches Vertex. A version Vertex accepts in the body is carried there, `2023-06-01`, `2023-01-01`, or no header get `vertex-2023-10-16`, and any other version drops under `drop_params` and is refused without it - Applies Vertex's beta policy and strips `output_config.effort` for models that do not accept it # references diff --git a/litellm-rust/crates/llms/src/vertex_ai/messages/transformation.rs b/litellm-rust/crates/llms/src/vertex_ai/messages/transformation.rs index 2e77dbc1ca8..46e5a68dcb8 100644 --- a/litellm-rust/crates/llms/src/vertex_ai/messages/transformation.rs +++ b/litellm-rust/crates/llms/src/vertex_ai/messages/transformation.rs @@ -3,10 +3,10 @@ use litellm_auth_gcp::{ VertexConfig, get_vertex_ai_location, get_vertex_ai_project, get_vertex_ai_project_from_credentials, secret_names, }; -use litellm_http::request::has_header; +use litellm_http::request::{has_header, header_value, without_headers}; use litellm_llms_types::{ formats::messages::{MessagesOptionalParams, MessagesRequest, MessagesTool, OutputConfig}, - providers::anthropic::{AnthropicBeta, BetaProvider, BetaSet}, + providers::anthropic::{AnthropicBeta, BetaProvider, BetaSet, VERSION_HEADER}, recognized::Recognized, }; use litellm_router_types::LitellmParams; @@ -36,6 +36,8 @@ use litellm_llms_types::providers::vertex_ai::{ }; const WEB_SEARCH_TOOL_PREFIX: &str = "web_search"; +const VERSIONS_SERVED_AS_VERTEX_DEFAULT: [&str; 3] = ["", "2023-01-01", "2023-06-01"]; +const VERSIONS_VERTEX_ACCEPTS_IN_BODY: [&str; 2] = [ANTHROPIC_VERSION, "2026-10-21"]; /// Claude on Vertex AI, Python's `VertexAIPartnerModelsAnthropicMessagesConfig`: the Anthropic /// payload addressed to a project and location, authenticated with a Google access token. @@ -90,12 +92,6 @@ impl BaseMessagesConfig for VertexAiPartnerModelsAnthropicMessagesConfig { .params .output_config .and_then(|config| sanitize_output_config(config, accepts_effort)), - extra: request - .params - .extra - .into_iter() - .chain([("anthropic_version".to_string(), json!(ANTHROPIC_VERSION))]) - .collect(), ..request.params }, ..request @@ -133,7 +129,35 @@ impl BaseMessagesConfig for VertexAiPartnerModelsAnthropicMessagesConfig { } fn request_headers(&self, headers: Headers, request: &MessagesRequest) -> Headers { - merge_beta_headers(headers, vertex_feature_betas(request)) + merge_beta_headers( + without_headers(headers, &[VERSION_HEADER]), + vertex_feature_betas(request), + ) + } + + fn request_body( + &self, + request: MessagesRequest, + headers: &Headers, + drop_params: bool, + ) -> Result { + let version = vertex_version( + &request.model, + header_value(headers, VERSION_HEADER), + drop_params, + )?; + Ok(MessagesRequest { + params: MessagesOptionalParams { + extra: request + .params + .extra + .into_iter() + .chain([("anthropic_version".to_string(), json!(version))]) + .collect(), + ..request.params + }, + ..request + }) } /// Vertex addresses the model in the URL and rejects it in the body. @@ -225,6 +249,26 @@ fn vertex_feature_betas(request: &MessagesRequest) -> BetaSet { ) } +fn vertex_version<'a>( + model: &str, + requested: Option<&'a str>, + drop_params: bool, +) -> Result<&'a str, Error> { + match requested.map(str::trim).unwrap_or_default() { + version if VERSIONS_SERVED_AS_VERTEX_DEFAULT.contains(&version) => Ok(ANTHROPIC_VERSION), + version if VERSIONS_VERTEX_ACCEPTS_IN_BODY.contains(&version) => Ok(version), + _ if drop_params => Ok(ANTHROPIC_VERSION), + version => Err(Error::InvalidRequest( + crate::ErrorDetail::UnsupportedParameter { + model: model.into(), + param: VERSION_HEADER.into(), + value: version.into(), + hint: String::new(), + }, + )), + } +} + #[cfg(test)] mod tests { use rstest::rstest; @@ -471,6 +515,28 @@ mod tests { ); } + #[rstest] + #[case::sdk_spelling("anthropic-version")] + #[case::any_case("Anthropic-Version")] + fn a_forwarded_anthropic_version_never_reaches_vertex(#[case] name: &str) { + let request: MessagesRequest = serde_json::from_value(json!({ + "model": "claude-sonnet-4-5", + "max_tokens": 16, + "messages": [{"role": "user", "content": "hi"}], + })) + .unwrap(); + assert_eq!( + VERTEX_ANTHROPIC_MESSAGES_CONFIG.request_headers( + vec![ + (name.to_string(), "2026-10-21".to_string()), + ("x-request-id".to_string(), "req-1".to_string()), + ], + &request, + ), + vec![("x-request-id".to_string(), "req-1".to_string())] + ); + } + fn transformed(value: Value, context: &MessagesTransformContext) -> Value { let request: MessagesRequest = serde_json::from_value(value).unwrap(); serde_json::to_value( @@ -482,16 +548,87 @@ mod tests { } #[rstest] - fn the_body_carries_the_vertex_version_and_the_wire_drops_the_model() { - let body = transformed( - json!({ - "model": "claude-sonnet-4-5@20250929", - "max_tokens": 16, - "messages": [{"role": "user", "content": "hi"}] - }), - &MessagesTransformContext::default(), + #[case::sdk_2(&[("anthropic-version", "2026-10-21")], false, "2026-10-21")] + #[case::any_case(&[("Anthropic-Version", "2026-10-21")], false, "2026-10-21")] + #[case::vertex_name(&[("anthropic-version", "vertex-2023-10-16")], false, "vertex-2023-10-16")] + #[case::sdk_1(&[("anthropic-version", "2023-06-01")], false, "vertex-2023-10-16")] + #[case::first_version(&[("anthropic-version", "2023-01-01")], false, "vertex-2023-10-16")] + #[case::blank(&[("anthropic-version", " ")], false, "vertex-2023-10-16")] + #[case::no_header(&[], false, "vertex-2023-10-16")] + #[case::unknown_under_drop_params(&[("anthropic-version", "latest")], true, "vertex-2023-10-16")] + fn the_requested_version_goes_in_the_body( + #[case] forwarded: &[(&str, &str)], + #[case] drop_params: bool, + #[case] body_version: &str, + ) { + let request: MessagesRequest = serde_json::from_value(json!({ + "model": "claude-sonnet-4-5", + "max_tokens": 16, + "messages": [{"role": "user", "content": "hi"}], + })) + .unwrap(); + let headers: Headers = forwarded + .iter() + .map(|(name, value)| (name.to_string(), value.to_string())) + .collect(); + let body = serde_json::to_value( + VERTEX_ANTHROPIC_MESSAGES_CONFIG + .request_body(request, &headers, drop_params) + .unwrap(), + ) + .unwrap(); + assert_eq!(body["anthropic_version"], json!(body_version)); + } + + #[rstest] + #[case::unreleased("2025-01-01")] + #[case::alias("latest")] + fn an_unsupported_version_is_refused_without_drop_params(#[case] requested: &str) { + let request: MessagesRequest = serde_json::from_value(json!({ + "model": "claude-sonnet-4-5", + "max_tokens": 16, + "messages": [{"role": "user", "content": "hi"}], + })) + .unwrap(); + let headers: Headers = vec![("anthropic-version".to_string(), requested.to_string())]; + let Error::InvalidRequest(detail) = VERTEX_ANTHROPIC_MESSAGES_CONFIG + .request_body(request, &headers, false) + .unwrap_err() + else { + panic!("expected an invalid request"); + }; + assert_eq!( + detail.to_string(), + format!( + "claude-sonnet-4-5 does not support anthropic-version={requested}. \ + To drop unsupported params, set `litellm.drop_params = True`." + ) ); - assert_eq!(body["anthropic_version"], json!("vertex-2023-10-16")); + } + + #[rstest] + fn the_body_carries_the_vertex_version_and_the_wire_drops_the_model() { + let request: MessagesRequest = serde_json::from_value(json!({ + "model": "claude-sonnet-4-5@20250929", + "max_tokens": 16, + "messages": [{"role": "user", "content": "hi"}] + })) + .unwrap(); + let body = serde_json::to_value( + VERTEX_ANTHROPIC_MESSAGES_CONFIG + .request_body( + VERTEX_ANTHROPIC_MESSAGES_CONFIG + .transform_anthropic_messages_request( + request, + &MessagesTransformContext::default(), + ) + .unwrap(), + &Vec::new(), + false, + ) + .unwrap(), + ) + .unwrap(); assert_eq!(body["model"], json!("claude-sonnet-4-5@20250929")); let wire = VERTEX_ANTHROPIC_MESSAGES_CONFIG.wire_body(body); assert_eq!(wire.get("model"), None); diff --git a/litellm-rust/crates/router/Cargo.toml b/litellm-rust/crates/router/Cargo.toml index a356bedcf16..988d1b68805 100644 --- a/litellm-rust/crates/router/Cargo.toml +++ b/litellm-rust/crates/router/Cargo.toml @@ -7,6 +7,7 @@ repository.workspace = true [dependencies] litellm-config.workspace = true +litellm-core-utils.workspace = true litellm-inference.workspace = true litellm-inference-messages.workspace = true litellm-router-types.workspace = true diff --git a/litellm-rust/crates/router/README.md b/litellm-rust/crates/router/README.md index 972d37d14be..a734e634104 100644 --- a/litellm-rust/crates/router/README.md +++ b/litellm-rust/crates/router/README.md @@ -1,4 +1,4 @@ -`litellm-router` scaffolds the model-list setup and deployment lookup portion of Python's `litellm.Router`. `Router::from_model_list(&config.model_list)` maps configured public names to provider deployments. Programmatic callers can collect `(String, Deployment)` entries into a `Router` +`litellm-router` scaffolds the model-list setup and deployment lookup portion of Python's `litellm.Router`. `Router::from_config(&config, &env)` maps configured public names to provider deployments and reads `drop_params` from `litellm_settings` and each deployment, resolving an `os.environ/NAME` value through `env`. Programmatic callers can collect `(String, Deployment)` entries into a `Router` Lookup is exact and returns `None` for an unknown name. This extraction preserves the gateway's existing behavior: the last entry wins when public names repeat. Multiple deployments per model group, routing strategies, retries, cooldowns, and fallbacks are not implemented yet diff --git a/litellm-rust/crates/router/src/lib.rs b/litellm-rust/crates/router/src/lib.rs index fb48213e8a7..b75c1a14446 100644 --- a/litellm-rust/crates/router/src/lib.rs +++ b/litellm-rust/crates/router/src/lib.rs @@ -2,16 +2,33 @@ mod deployment; use std::collections::HashMap; -use litellm_config::Model; +use litellm_config::{Config, Spelled}; +use litellm_core_utils::{serde_compat::parse_pydantic_bool, settings::Lookup}; +use litellm_inference_messages::{MessagesSettings, MessagesShaping}; pub use deployment::Deployment; #[derive(Clone, Debug, Default)] pub struct Router(HashMap); +fn is_on(flag: Option<&Spelled>, env: &impl Lookup) -> bool { + match flag { + Some(Spelled::Value(on)) => *on, + Some(Spelled::Text(text)) => match text.strip_prefix("os.environ/") { + Some(name) => env + .get(name) + .is_some_and(|value| parse_pydantic_bool(&value) == Some(true)), + None => parse_pydantic_bool(text) == Some(true), + }, + None => false, + } +} + impl Router { - pub fn from_model_list(model_list: &[Model]) -> Self { - model_list + pub fn from_config(config: &Config, env: &impl Lookup) -> Self { + let drops_params_globally = is_on(config.litellm_settings.drop_params.as_ref(), env); + config + .model_list .iter() .map(|model| { ( @@ -26,6 +43,14 @@ impl Router { api_base: model.litellm_params.api_base.clone(), custom_llm_provider: model.litellm_params.custom_llm_provider.clone(), litellm_params: model.litellm_params.clone(), + shaping: MessagesShaping { + settings: MessagesSettings { + drop_params: drops_params_globally + || is_on(model.litellm_params.drop_params.as_ref(), env), + ..MessagesSettings::default() + }, + ..MessagesShaping::default() + }, ..Deployment::default() }, ) @@ -43,3 +68,60 @@ impl FromIterator<(String, Deployment)> for Router { Self(entries.into_iter().collect()) } } + +#[cfg(test)] +mod tests { + use litellm_config::Config; + use rstest::rstest; + + use super::Router; + + #[rstest] + #[case::unset("", "", false)] + #[case::deployment_flag("", "\n drop_params: true", true)] + #[case::deployment_word("", "\n drop_params: \" True \"", true)] + #[case::deployment_off("", "\n drop_params: false", false)] + #[case::deployment_unknown_word("", "\n drop_params: maybe", false)] + #[case::deployment_yes("", "\n drop_params: yes", true)] + #[case::deployment_no("", "\n drop_params: \"no\"", false)] + #[case::deployment_env_reference("", "\n drop_params: os.environ/DROP_PARAMS_FLAG", true)] + #[case::deployment_unset_env_reference( + "", + "\n drop_params: os.environ/MISSING_FLAG", + false + )] + #[case::global_env_reference( + "litellm_settings:\n drop_params: os.environ/DROP_PARAMS_FLAG\n", + "", + true + )] + #[case::global_flag("litellm_settings:\n drop_params: true\n", "", true)] + #[case::global_word("litellm_settings:\n drop_params: \"true\"\n", "", true)] + #[case::global_off("litellm_settings:\n drop_params: false\n", "", false)] + #[case::global_over_deployment_off( + "litellm_settings:\n drop_params: true\n", + "\n drop_params: false", + true + )] + fn configured_drop_params_reach_the_deployment( + #[case] settings: &str, + #[case] deployment: &str, + #[case] expected: bool, + ) { + let config = Config::from_yaml(&format!( + "{settings}model_list:\n - model_name: public-model\n litellm_params:\n model: provider/model{deployment}" + )) + .unwrap(); + + assert_eq!( + Router::from_config(&config, &|name: &str| (name == "DROP_PARAMS_FLAG") + .then(|| " Yes ".to_owned())) + .get("public-model") + .unwrap() + .shaping + .settings + .drop_params, + expected + ); + } +} diff --git a/litellm-rust/crates/router/tests/router.rs b/litellm-rust/crates/router/tests/router.rs index bbaeb806b96..03e78ab9d11 100644 --- a/litellm-rust/crates/router/tests/router.rs +++ b/litellm-rust/crates/router/tests/router.rs @@ -16,7 +16,7 @@ fn configuration_preserves_deployment_parameters(#[case] parameters: &str) { "model_list:\n - model_name: public-model\n litellm_params:\n model: provider/model\n {parameters}" )) .unwrap(); - let router = Router::from_model_list(&config.model_list); + let router = Router::from_config(&config, &|_: &str| None); let deployment = router.get(&config.model_list[0].model_name).unwrap(); let params = &config.model_list[0].litellm_params; @@ -44,7 +44,7 @@ fn configuration_types_the_provider_params_of_a_deployment() { azure_ad_token: ignored-by-every-group", ) .unwrap(); - let router = Router::from_model_list(&config.model_list); + let router = Router::from_config(&config, &|_: &str| None); let params = &router.get("public-model").unwrap().litellm_params; assert_eq!( @@ -74,7 +74,7 @@ fn lookup_uses_public_names(#[case] name: &str, #[case] expected: Option<&str>) model: provider/b", ) .unwrap(); - let router = Router::from_model_list(&config.model_list); + let router = Router::from_config(&config, &|_: &str| None); assert_eq!(router.get(name).map(|entry| entry.model.as_str()), expected); } @@ -84,7 +84,7 @@ fn empty_configuration_has_no_deployment() { let config = Config::from_yaml("model_list: []").unwrap(); assert!( - Router::from_model_list(&config.model_list) + Router::from_config(&config, &|_: &str| None) .get("") .is_none() ); diff --git a/litellm/llms/vertex_ai/vertex_ai_partner_models/anthropic/messages/transformation.py b/litellm/llms/vertex_ai/vertex_ai_partner_models/anthropic/messages/transformation.py index 6d8eb228786..1f0ee728cd1 100644 --- a/litellm/llms/vertex_ai/vertex_ai_partner_models/anthropic/messages/transformation.py +++ b/litellm/llms/vertex_ai/vertex_ai_partner_models/anthropic/messages/transformation.py @@ -1,6 +1,8 @@ from collections.abc import Mapping, Sequence from typing import Final +from litellm.exceptions import UnsupportedParamsError +from litellm.litellm_core_utils.core_helpers import drops_unsupported_params from litellm.llms.anthropic.common_utils import AnthropicModelInfo from litellm.llms.anthropic.messages.native.transformation import ( AnthropicMessagesConfig, @@ -17,6 +19,41 @@ from litellm.types.router import GenericLiteLLMParams from ....vertex_llm_base import VertexBase from ..output_params_utils import sanitize_vertex_anthropic_output_params +_VERSION_HEADER_VERTEX_READS_FROM_BODY: Final = "anthropic-version" +_VERTEX_DEFAULT_VERSION: Final = "vertex-2023-10-16" +_VERSIONS_SERVED_AS_VERTEX_DEFAULT: Final = frozenset({"", "2023-01-01", "2023-06-01"}) +_VERSIONS_VERTEX_ACCEPTS_IN_BODY: Final = frozenset({_VERTEX_DEFAULT_VERSION, "2026-10-21"}) + + +def _requested_version(headers: Mapping[str, object]) -> str: + requested: Final = next( + (value for name, value in headers.items() if name.lower() == _VERSION_HEADER_VERTEX_READS_FROM_BODY), + "", + ) + return requested.strip() if isinstance(requested, str) else "" + + +def _vertex_version(headers: Mapping[str, object], model: str, drop_params: bool) -> str | UnsupportedParamsError: + requested: Final = _requested_version(headers) + if requested in _VERSIONS_SERVED_AS_VERTEX_DEFAULT: + return _VERTEX_DEFAULT_VERSION + if requested in _VERSIONS_VERTEX_ACCEPTS_IN_BODY: + return requested + if drop_params: + return _VERTEX_DEFAULT_VERSION + return UnsupportedParamsError( + message=( + f"{model} does not support {_VERSION_HEADER_VERTEX_READS_FROM_BODY}={requested}. " + "To drop unsupported params, set `litellm.drop_params = True`." + ), + llm_provider="vertex_ai", + model=model, + ) + + +def _without_version_header(headers: Mapping[str, str]) -> dict[str, str]: # mutable-ok: the caller sends these headers + return {name: value for name, value in headers.items() if name.lower() != _VERSION_HEADER_VERTEX_READS_FROM_BODY} + class VertexAIPartnerModelsAnthropicMessagesConfig(AnthropicMessagesConfig, VertexBase): @property @@ -198,7 +235,7 @@ class VertexAIPartnerModelsAnthropicMessagesConfig(AnthropicMessagesConfig, Vert messages: list[dict], anthropic_messages_optional_request_params: dict, litellm_params: GenericLiteLLMParams, - headers: dict, + headers: dict[str, object], # mutable-ok: mirrors the provider transform contract ) -> dict: anthropic_messages_request: Final = super().transform_anthropic_messages_request( model=model, @@ -212,10 +249,28 @@ class VertexAIPartnerModelsAnthropicMessagesConfig(AnthropicMessagesConfig, Vert self._remove_scope_from_cache_control(anthropic_messages_request) - anthropic_messages_request["anthropic_version"] = "vertex-2023-10-16" + version: Final = _vertex_version( + headers, model=model, drop_params=drops_unsupported_params(litellm_params.drop_params) + ) + if isinstance(version, UnsupportedParamsError): + raise version + anthropic_messages_request["anthropic_version"] = version anthropic_messages_request.pop("model", None) # do not pass model in request body to vertex ai sanitize_vertex_anthropic_output_params(anthropic_messages_request, model) return anthropic_messages_request + + def sign_request( + self, + headers: Mapping[str, str], + optional_params: Mapping[str, object], + request_data: Mapping[str, object], + api_base: str, + api_key: str | None = None, + model: str | None = None, + stream: bool | None = None, + fake_stream: bool | None = None, + ) -> tuple[dict[str, str], bytes | None]: # mutable-ok: the caller sends these headers + return _without_version_header(headers), None diff --git a/tests/unit/llms/vertex_ai/vertex_ai_partner_models/anthropic/test_vertex_ai_partner_models_anthropic_messages_config.py b/tests/unit/llms/vertex_ai/vertex_ai_partner_models/anthropic/test_vertex_ai_partner_models_anthropic_messages_config.py index 6c10d524531..5916ce9f7a1 100644 --- a/tests/unit/llms/vertex_ai/vertex_ai_partner_models/anthropic/test_vertex_ai_partner_models_anthropic_messages_config.py +++ b/tests/unit/llms/vertex_ai/vertex_ai_partner_models/anthropic/test_vertex_ai_partner_models_anthropic_messages_config.py @@ -6,12 +6,14 @@ from datetime import datetime, timedelta from queue import SimpleQueue from threading import get_ident from typing import Final -from unittest.mock import MagicMock, patch +from unittest.mock import AsyncMock, MagicMock, patch import pytest import requests import responses +import litellm +from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler from litellm.llms.vertex_ai.vertex_ai_partner_models.anthropic.messages.transformation import ( VertexAIPartnerModelsAnthropicMessagesConfig, ) @@ -94,6 +96,85 @@ async def test_messages_auth_load_and_refresh_leave_the_event_loop(stream: bool, assert headers == {"Authorization": "Bearer incoming", "anthropic-beta": "retained-beta", "x-request-id": "req-1"} +def _vertex_client() -> MagicMock: + vertex_response: Final = MagicMock() + vertex_response.json.return_value = { + "id": "msg_1", + "type": "message", + "role": "assistant", + "content": [{"type": "text", "text": "hi"}], + "model": "claude-sonnet-5", + "stop_reason": "end_turn", + "stop_sequence": None, + "usage": {"input_tokens": 1, "output_tokens": 1}, + } + client: Final = MagicMock(spec=AsyncHTTPHandler) + client.post = AsyncMock(return_value=vertex_response) + return client + + +async def _send_to_vertex(client: MagicMock, caller_headers: Mapping[str, str], drop_params: bool) -> None: + with responses.RequestsMock(assert_all_requests_are_fired=False) as oauth: + oauth.add(responses.POST, "https://oauth2.googleapis.com/token", json={"access_token": "vertex-token", "expires_in": 3600}) + await litellm.anthropic.messages.acreate( + model="vertex_ai/claude-sonnet-5", + messages=[{"role": "user", "content": "hi"}], + max_tokens=16, + vertex_ai_project="version-project", + vertex_ai_location="global", + vertex_credentials=json.dumps( + { + "type": "authorized_user", + "client_id": "version-header-client", + "client_secret": "synthetic-secret", + "refresh_token": "synthetic-refresh", + } + ), + client=client, + drop_params=drop_params, + provider_specific_header={ + "custom_llm_provider": "vertex_ai", + "extra_headers": {**caller_headers, "x-request-id": "req-1"}, + }, + ) + + +@pytest.mark.parametrize( + ("caller_headers", "drop_params", "body_version"), + ( + ({"anthropic-version": "2026-10-21"}, False, "2026-10-21"), + ({"Anthropic-Version": "2026-10-21"}, False, "2026-10-21"), + ({"anthropic-version": "vertex-2023-10-16"}, False, "vertex-2023-10-16"), + ({"anthropic-version": "2023-06-01"}, False, "vertex-2023-10-16"), + ({"anthropic-version": "2023-01-01"}, False, "vertex-2023-10-16"), + ({"anthropic-version": " "}, False, "vertex-2023-10-16"), + ({}, False, "vertex-2023-10-16"), + ({"anthropic-version": "latest"}, True, "vertex-2023-10-16"), + ), +) +@pytest.mark.asyncio +async def test_caller_anthropic_version_reaches_vertex_only_in_the_body( + caller_headers: dict[str, str], drop_params: bool, body_version: str +) -> None: + client: Final = _vertex_client() + await _send_to_vertex(client, caller_headers, drop_params) + + sent: Final = client.post.call_args.kwargs + sent_header_names: Final = {name.lower() for name in sent["headers"]} + assert "anthropic-version" not in sent_header_names + assert sent["headers"]["x-request-id"] == "req-1" + assert json.loads(sent["data"])["anthropic_version"] == body_version + + +@pytest.mark.parametrize("requested", ("2025-01-01", "latest")) +@pytest.mark.asyncio +async def test_unsupported_anthropic_version_is_refused_without_drop_params(requested: str) -> None: + client: Final = _vertex_client() + with pytest.raises(litellm.BadRequestError, match=f"anthropic-version={requested}. To drop unsupported params"): + await _send_to_vertex(client, {"anthropic-version": requested}, drop_params=False) + client.post.assert_not_awaited() + + def test_validate_environment_uses_vertex_ai_location(): config = VertexAIPartnerModelsAnthropicMessagesConfig() headers = {} From 57c1df8c33b01bbee21b6496649428124635301b Mon Sep 17 00:00:00 2001 From: yujonglee Date: Wed, 28 Oct 2026 18:24:40 -0700 Subject: [PATCH 38/41] refactor(model-catalog): parse, lower and validate the catalog into typed prices (#48853) Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- .github/workflows/test-rust.yml | 2 + .github/workflows/test-unit.yml | 2 +- litellm-rust/Cargo.lock | 34 +- litellm-rust/Cargo.toml | 4 + litellm-rust/crates/cost/Cargo.toml | 15 +- litellm-rust/crates/cost/README.md | 12 +- litellm-rust/crates/cost/benches/calculate.rs | 66 -- litellm-rust/crates/cost/examples/charge.rs | 40 - litellm-rust/crates/cost/src/call.rs | 42 + litellm-rust/crates/cost/src/error.rs | 47 +- litellm-rust/crates/cost/src/lib.rs | 388 +------- litellm-rust/crates/cost/src/money.rs | 257 ++++++ litellm-rust/crates/cost/src/price.rs | 505 +++++++++++ litellm-rust/crates/cost/src/price/meter.rs | 197 ++++ litellm-rust/crates/cost/src/priced.rs | 53 ++ litellm-rust/crates/cost/src/schedule.rs | 241 +++++ litellm-rust/crates/cost/src/usage.rs | 104 +++ litellm-rust/crates/cost/tests/calculation.rs | 535 ----------- .../cost/tests/generate_python_reference.py | 96 -- .../crates/cost/tests/python_reference.tsv | 9 - .../crates/model-catalog-refresh/Cargo.toml | 19 + .../crates/model-catalog-refresh/src/error.rs | 105 +++ .../crates/model-catalog-refresh/src/lib.rs | 9 + .../model-catalog-refresh/src/policy.rs | 20 + .../model-catalog-refresh/src/source.rs | 17 + .../crates/model-catalog-refresh/src/step.rs | 66 ++ litellm-rust/crates/model-catalog/AGENTS.md | 28 +- litellm-rust/crates/model-catalog/Cargo.toml | 10 +- .../crates/model-catalog/src/bundled.rs | 3 + .../crates/model-catalog/src/capabilities.rs | 67 -- .../crates/model-catalog/src/catalog.rs | 485 ++++++---- .../crates/model-catalog/src/error.rs | 391 +++++++- .../crates/model-catalog/src/fallback.rs | 9 +- .../crates/model-catalog/src/layers.rs | 32 + litellm-rust/crates/model-catalog/src/lib.rs | 46 +- .../crates/model-catalog/src/lower.rs | 855 ++++++++++++++++++ .../crates/model-catalog/src/model.rs | 22 + .../crates/model-catalog/src/model_info.rs | 734 ++++----------- .../crates/model-catalog/src/pricing.rs | 97 -- .../crates/model-catalog/src/register.rs | 26 + .../crates/model-catalog/src/resolve/image.rs | 18 + .../crates/model-catalog/src/resolve/mod.rs | 46 + .../crates/model-catalog/src/schema.rs | 197 ++-- .../crates/model-catalog/src/store/AGENTS.md | 35 + .../crates/model-catalog/src/store/mod.rs | 460 ++++++++++ .../model-catalog/src/store/name_index.rs | 35 + .../crates/model-catalog/src/validation.rs | 252 ------ .../crates/model-catalog/src/vocabulary.rs | 126 +++ .../model-catalog/src/wire/attributes.rs | 344 +++++++ .../crates/model-catalog/src/wire/blocks.rs | 182 ++++ .../crates/model-catalog/src/wire/entry.rs | 193 ++++ .../crates/model-catalog/src/wire/flat_key.rs | 359 ++++++++ .../crates/model-catalog/src/wire/lax.rs | 128 +++ .../crates/model-catalog/src/wire/mod.rs | 10 + .../crates/model-catalog/tests/catalog.rs | 632 +++++++------ .../tests/generate_price_lookup_reference.py | 91 ++ .../model-catalog/tests/price_lookup.rs | 89 ++ .../tests/price_lookup_reference.json | 89 ++ .../tests/registry_validation.rs | 93 -- .../crates/model-catalog/tests/schema.rs | 73 +- 60 files changed, 6345 insertions(+), 2797 deletions(-) delete mode 100644 litellm-rust/crates/cost/benches/calculate.rs delete mode 100644 litellm-rust/crates/cost/examples/charge.rs create mode 100644 litellm-rust/crates/cost/src/call.rs create mode 100644 litellm-rust/crates/cost/src/money.rs create mode 100644 litellm-rust/crates/cost/src/price.rs create mode 100644 litellm-rust/crates/cost/src/price/meter.rs create mode 100644 litellm-rust/crates/cost/src/priced.rs create mode 100644 litellm-rust/crates/cost/src/schedule.rs create mode 100644 litellm-rust/crates/cost/src/usage.rs delete mode 100644 litellm-rust/crates/cost/tests/calculation.rs delete mode 100644 litellm-rust/crates/cost/tests/generate_python_reference.py delete mode 100644 litellm-rust/crates/cost/tests/python_reference.tsv create mode 100644 litellm-rust/crates/model-catalog-refresh/Cargo.toml create mode 100644 litellm-rust/crates/model-catalog-refresh/src/error.rs create mode 100644 litellm-rust/crates/model-catalog-refresh/src/lib.rs create mode 100644 litellm-rust/crates/model-catalog-refresh/src/policy.rs create mode 100644 litellm-rust/crates/model-catalog-refresh/src/source.rs create mode 100644 litellm-rust/crates/model-catalog-refresh/src/step.rs create mode 100644 litellm-rust/crates/model-catalog/src/bundled.rs delete mode 100644 litellm-rust/crates/model-catalog/src/capabilities.rs create mode 100644 litellm-rust/crates/model-catalog/src/layers.rs create mode 100644 litellm-rust/crates/model-catalog/src/lower.rs create mode 100644 litellm-rust/crates/model-catalog/src/model.rs delete mode 100644 litellm-rust/crates/model-catalog/src/pricing.rs create mode 100644 litellm-rust/crates/model-catalog/src/register.rs create mode 100644 litellm-rust/crates/model-catalog/src/resolve/image.rs create mode 100644 litellm-rust/crates/model-catalog/src/resolve/mod.rs create mode 100644 litellm-rust/crates/model-catalog/src/store/AGENTS.md create mode 100644 litellm-rust/crates/model-catalog/src/store/mod.rs create mode 100644 litellm-rust/crates/model-catalog/src/store/name_index.rs delete mode 100644 litellm-rust/crates/model-catalog/src/validation.rs create mode 100644 litellm-rust/crates/model-catalog/src/vocabulary.rs create mode 100644 litellm-rust/crates/model-catalog/src/wire/attributes.rs create mode 100644 litellm-rust/crates/model-catalog/src/wire/blocks.rs create mode 100644 litellm-rust/crates/model-catalog/src/wire/entry.rs create mode 100644 litellm-rust/crates/model-catalog/src/wire/flat_key.rs create mode 100644 litellm-rust/crates/model-catalog/src/wire/lax.rs create mode 100644 litellm-rust/crates/model-catalog/src/wire/mod.rs create mode 100644 litellm-rust/crates/model-catalog/tests/generate_price_lookup_reference.py create mode 100644 litellm-rust/crates/model-catalog/tests/price_lookup.rs create mode 100644 litellm-rust/crates/model-catalog/tests/price_lookup_reference.json delete mode 100644 litellm-rust/crates/model-catalog/tests/registry_validation.rs diff --git a/.github/workflows/test-rust.yml b/.github/workflows/test-rust.yml index 636a5031e5f..04af770b24f 100644 --- a/.github/workflows/test-rust.yml +++ b/.github/workflows/test-rust.yml @@ -4,6 +4,7 @@ on: push: paths: - "litellm-rust/**" + - "model_prices_and_context_window.json" - "litellm/rust_bridge/**" - "scripts/generate_trace_types.py" - "scripts/trace_codegen/**" @@ -38,6 +39,7 @@ on: - "litellm_**" paths: - "litellm-rust/**" + - "model_prices_and_context_window.json" - "litellm/rust_bridge/**" - "scripts/generate_trace_types.py" - "scripts/trace_codegen/**" diff --git a/.github/workflows/test-unit.yml b/.github/workflows/test-unit.yml index 80d5018fe9a..4768eeb633e 100644 --- a/.github/workflows/test-unit.yml +++ b/.github/workflows/test-unit.yml @@ -42,7 +42,7 @@ jobs: id: rust-bridge-key if: steps.changes.outputs.decision != 'skip' env: - RUST_BRIDGE_CACHE_KEY: ${{ runner.os }}-rust-bridge-editable-${{ hashFiles('litellm-rust/Cargo.lock', 'litellm-rust/Cargo.toml', 'litellm-rust/crates/**', 'pyproject.toml') }} + RUST_BRIDGE_CACHE_KEY: ${{ runner.os }}-rust-bridge-editable-${{ hashFiles('litellm-rust/Cargo.lock', 'litellm-rust/Cargo.toml', 'litellm-rust/crates/**', 'model_prices_and_context_window.json', 'pyproject.toml') }} run: echo "key=$RUST_BRIDGE_CACHE_KEY" >> "$GITHUB_OUTPUT" - name: Set up Python diff --git a/litellm-rust/Cargo.lock b/litellm-rust/Cargo.lock index 66a345fc438..f7d7e5f4dab 100644 --- a/litellm-rust/Cargo.lock +++ b/litellm-rust/Cargo.lock @@ -3150,6 +3150,7 @@ dependencies = [ "defmt", "jiff-core", "jiff-static", + "jiff-tzdb", "log", "portable-atomic", "portable-atomic-util", @@ -3178,6 +3179,12 @@ dependencies = [ "syn 2.0.119", ] +[[package]] +name = "jiff-tzdb" +version = "0.1.9" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "fa8377070c6bae868759445e5a77f66d84f0b72f3a054bfb00e6d038b8282da7" + [[package]] name = "jni" version = "0.22.4" @@ -3729,9 +3736,12 @@ dependencies = [ name = "litellm-cost" version = "0.1.0" dependencies = [ - "criterion", - "proptest", + "jiff", "rstest", + "schemars 1.2.2", + "serde", + "serde_json", + "strum", "thiserror 2.0.19", ] @@ -4277,17 +4287,35 @@ name = "litellm-model-catalog" version = "0.1.0" dependencies = [ "indexmap 2.14.0", + "jiff", "jsonschema", + "litellm-cost", "litellm-error", - "litellm-llms-types", + "macro_rules_attribute", "rstest", "schemars 1.2.2", "serde", "serde_json", + "serde_with", + "strum", "thiserror 2.0.19", "time", ] +[[package]] +name = "litellm-model-catalog-refresh" +version = "0.1.0" +dependencies = [ + "litellm-error", + "litellm-http", + "litellm-model-catalog", + "reqwest", + "rstest", + "thiserror 2.0.19", + "tokio", + "url", +] + [[package]] name = "litellm-python-bridge" version = "0.1.0" diff --git a/litellm-rust/Cargo.toml b/litellm-rust/Cargo.toml index b7b22cfd32c..49a33e08456 100644 --- a/litellm-rust/Cargo.toml +++ b/litellm-rust/Cargo.toml @@ -73,6 +73,9 @@ litellm-python-compat = { path = "crates/python-compat" } litellm-circuit-breaker = { path = "crates/circuit-breaker" } litellm-redis-client = { path = "crates/redis-client" } litellm-redis-testing = { path = "crates/redis-testing" } +litellm-cost = { path = "crates/cost" } +litellm-model-catalog = { path = "crates/model-catalog" } +litellm-model-catalog-refresh = { path = "crates/model-catalog-refresh" } askama = { version = "0.16.1", default-features = false, features = ["derive", "std"] } tracing = "0.1" @@ -109,6 +112,7 @@ serde = { version = "1.0", features = ["derive"] } serde_json = { version = "1.0", features = ["float_roundtrip"] } serde_with = { version = "=3.21.0", default-features = false, features = ["std", "macros"] } sha2 = "0.10" +sha1 = "0.10" syn = { version = "2", default-features = false } sqlx = { version = "0.9.0", default-features = false, features = ["json", "macros", "postgres", "runtime-tokio", "chrono", "tls-rustls-ring-native-roots"] } chrono = { version = "0.4.45", default-features = false, features = ["std"] } diff --git a/litellm-rust/crates/cost/Cargo.toml b/litellm-rust/crates/cost/Cargo.toml index 6921e6151b5..858287c7860 100644 --- a/litellm-rust/crates/cost/Cargo.toml +++ b/litellm-rust/crates/cost/Cargo.toml @@ -5,14 +5,17 @@ edition.workspace = true license.workspace = true repository.workspace = true +[features] +serde = ["dep:serde"] +schema = ["serde", "dep:schemars"] + [dependencies] +jiff = { version = "0.2.37", default-features = false, features = ["std"] } +schemars = { workspace = true, optional = true } +serde = { workspace = true, optional = true } thiserror.workspace = true +strum.workspace = true [dev-dependencies] -criterion.workspace = true -proptest.workspace = true rstest.workspace = true - -[[bench]] -name = "calculate" -harness = false +serde_json.workspace = true diff --git a/litellm-rust/crates/cost/README.md b/litellm-rust/crates/cost/README.md index ee4722b6438..f8fc9403917 100644 --- a/litellm-rust/crates/cost/README.md +++ b/litellm-rust/crates/cost/README.md @@ -1,13 +1,3 @@ # litellm-cost -This crate calculates text token charges from rates and usage supplied by its caller. It is standalone and has no Python bridge or proxy integration - -Call `compile(&pricing)` once for an immutable plan, then `plan.calculate(&request)` for each supported request. `calculate(&pricing, &request)` compiles on each call. A successful result exposes pre-multiplier component costs, selected rates, the multiplier, and derived `input()`, `output()`, and `total()` values - -The caller states whether `prompt_tokens` includes cache tokens. Threshold selection uses total input tokens for either convention and selects one rate for the whole request. Thresholds are sorted when compiled, and duplicate thresholds or tier overrides fail deterministically. `Fast` selects priority rates; unknown tiers use standard rates - -`Rate::Missing`, `Rate::Null`, and `Rate::Value(0.0)` remain distinct. Missing cache rates fall back to the selected input rate, and an absent one-hour write rate falls back to the selected write rate. Missing input or output rates return typed errors, including for zero usage. Python's sparse-entry behavior remains outside this native contract - -The supported off-peak shape is one non-wrapping UTC daily window. The caller supplies the applicable regional multiplier after provider-specific selection. Negative or non-finite rates, ambiguous rules, inconsistent cache counts, incomplete write splits, invalid windows and overflow return errors. Callers must decline unsupported inputs before native execution if their public contract accepts those shapes - -This crate does not select models, read catalogs, fetch provider prices, normalize multimodal usage, process provider-reported costs, or calculate non-token charges. It does not change proxy behavior. The reference fixture was generated by `tests/generate_python_reference.py` against the Python implementation at the commit recorded in `tests/python_reference.tsv`, using synthetic rates and fixed usage +This crate holds the price data model that `litellm-model-catalog` lowers each catalog entry into, plus the usage, call context and priced outcome types that `calculate` will compute over. There is no `calculate` until the token pricing PR lands. `UnitPrice` is USD per unit and `Usd` is an amount, and only the operations that make sense between them exist, so a price is never added to an amount. `Prices::band` and `Prices::rate` resolve a token rate the way Python's `_get_token_base_cost` does diff --git a/litellm-rust/crates/cost/benches/calculate.rs b/litellm-rust/crates/cost/benches/calculate.rs deleted file mode 100644 index 2561b606e6c..00000000000 --- a/litellm-rust/crates/cost/benches/calculate.rs +++ /dev/null @@ -1,66 +0,0 @@ -use criterion::{Criterion, criterion_group, criterion_main}; -use litellm_cost::{ - Pricing, PromptConvention, Rate, Rates, Request, ServiceTier, ThresholdPolicy, ThresholdRates, - Usage, calculate, compile, -}; -use std::hint::black_box; - -fn bench(c: &mut Criterion) { - let pricing = Pricing { - standard: Rates { - input: Rate::Value(0.000002), - output: Rate::Value(0.000008), - cache_read: Rate::Value(0.0000005), - cache_write: Rate::Missing, - cache_write_1h: Rate::Missing, - }, - tiers: &[], - thresholds: &[], - off_peak: None, - }; - let request = Request { - usage: Usage { - prompt_tokens: 1000, - completion_tokens: 200, - cache_read_tokens: 250, - cache_write_tokens: 0, - cache_write_5m_tokens: None, - cache_write_1h_tokens: None, - prompt_convention: PromptConvention::IncludesCache, - }, - service_tier: ServiceTier::Standard, - threshold_policy: ThresholdPolicy::Exclusive, - region_multiplier: None, - billed_at_utc_minute: None, - }; - let plan = compile(&pricing).unwrap(); - c.bench_function("native_compiled_calculation", |b| { - b.iter(|| black_box(plan.calculate(black_box(&request)).unwrap())) - }); - c.bench_function("native_full_wrapper", |b| { - b.iter(|| black_box(calculate(black_box(&pricing), black_box(&request)).unwrap())) - }); - c.bench_function("native_rate_compilation", |b| { - b.iter(|| black_box(compile(black_box(&pricing)).unwrap())) - }); - let threshold = ThresholdRates { - above_prompt_tokens: 1000, - standard: Rates { - input: Rate::Value(0.000004), - output: Rate::Value(0.000016), - ..Rates::EMPTY - }, - tiers: &[], - }; - let threshold_pricing = Pricing { - thresholds: &[threshold], - ..pricing - }; - let threshold_plan = compile(&threshold_pricing).unwrap(); - c.bench_function("native_threshold_boundary", |b| { - b.iter(|| black_box(threshold_plan.calculate(black_box(&request)).unwrap())) - }); -} - -criterion_group!(benches, bench); -criterion_main!(benches); diff --git a/litellm-rust/crates/cost/examples/charge.rs b/litellm-rust/crates/cost/examples/charge.rs deleted file mode 100644 index a3a83febd74..00000000000 --- a/litellm-rust/crates/cost/examples/charge.rs +++ /dev/null @@ -1,40 +0,0 @@ -use litellm_cost::{ - Pricing, PromptConvention, Rate, Rates, Request, ServiceTier, ThresholdPolicy, Usage, compile, -}; - -fn main() { - let pricing = Pricing { - standard: Rates { - input: Rate::Value(2.0), - output: Rate::Value(4.0), - cache_read: Rate::Value(0.5), - cache_write: Rate::Value(3.0), - cache_write_1h: Rate::Missing, - }, - tiers: &[], - thresholds: &[], - off_peak: None, - }; - let request = Request { - usage: Usage { - prompt_tokens: 100, - completion_tokens: 20, - cache_read_tokens: 25, - cache_write_tokens: 10, - cache_write_5m_tokens: None, - cache_write_1h_tokens: None, - prompt_convention: PromptConvention::IncludesCache, - }, - service_tier: ServiceTier::Standard, - threshold_policy: ThresholdPolicy::Exclusive, - region_multiplier: None, - billed_at_utc_minute: None, - }; - let cost = compile(&pricing).unwrap().calculate(&request).unwrap(); - println!( - "input={} output={} total={}", - cost.input(), - cost.output(), - cost.total() - ); -} diff --git a/litellm-rust/crates/cost/src/call.rs b/litellm-rust/crates/cost/src/call.rs new file mode 100644 index 00000000000..3f44ec03e9b --- /dev/null +++ b/litellm-rust/crates/cost/src/call.rs @@ -0,0 +1,42 @@ +use crate::price::{Channel, Prices, Residency, ServiceTier}; + +/// Facts about the call, never about the row or the provider +#[derive(Clone, Debug, PartialEq)] +pub struct CallContext { + /// Requested service tier, `"fast"` maps to `Priority` as on main + pub tier: ServiceTier, + /// Online, batch or decisions + pub channel: Channel, + /// OpenAI data residency, picks the residency uplift + pub residency: Option, + /// Vertex call sent to a regional endpoint, picks the regional uplift + pub non_global_endpoint: bool, + /// Anthropic `speed: fast`, picks the speed uplift + pub speed: Option, + /// Anthropic and Bedrock `inference_geo`, picks the geo uplift + pub inference_geo: Option, + /// When the call was billed, for off-peak windows + pub at: jiff::Timestamp, + /// Per-call custom pricing, lowered like a row and applied field by field over it + pub overlay: Option, +} + +/// Anthropic `speed` +#[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)] +pub enum Speed { + Fast, +} + +/// Anthropic and Bedrock `inference_geo`, lowercased as on main +#[derive(Clone, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)] +pub struct InferenceGeo(Box); + +impl InferenceGeo { + pub fn new(geo: &str) -> Self { + Self(geo.to_lowercase().into()) + } + + pub fn as_str(&self) -> &str { + &self.0 + } +} diff --git a/litellm-rust/crates/cost/src/error.rs b/litellm-rust/crates/cost/src/error.rs index 1753080617f..7ef5eb87597 100644 --- a/litellm-rust/crates/cost/src/error.rs +++ b/litellm-rust/crates/cost/src/error.rs @@ -1,27 +1,24 @@ -#[derive(Clone, Copy, Debug, Eq, PartialEq, thiserror::Error)] +use thiserror::Error; + +#[derive(Clone, Copy, Debug, PartialEq, Eq, Error, strum::VariantArray)] pub enum Error { - #[error("model has no input rate")] - MissingInputRate, - #[error("model has no output rate")] - MissingOutputRate, - #[error("rate must be finite and nonnegative")] - InvalidRate, - #[error("region multiplier must be finite and positive")] - InvalidRegionMultiplier, - #[error("off-peak pricing needs a billing minute within the day")] - InvalidBillingTime, - #[error("off-peak window must start before it ends within the day")] - InvalidOffPeakWindow, - #[error("cached tokens exceed prompt tokens")] - CacheExceedsPrompt, - #[error("cache write details do not add up to the cache write tokens")] - InvalidCacheWriteDetails, - #[error("token count or cost overflows")] - TokenCountOverflow, - #[error("service tier is repeated or cannot carry its own rates")] - DuplicateTier, - #[error("prompt token threshold is repeated")] - DuplicateThreshold, - #[error("service tier is repeated or cannot carry its own rates within a threshold")] - DuplicateThresholdTier, + #[error("expected a UTC window HH:MM-HH:MM")] + InvalidUtcWindow, + #[error("expected an ISO weekday number 1-7 or an English day name")] + InvalidWeekday, +} + +#[cfg(test)] +mod tests { + use rstest::rstest; + + use super::Error; + use crate::schedule::{UtcWindow, Weekday}; + + #[rstest] + #[case::window("not-a-window".parse::().unwrap_err(), Error::InvalidUtcWindow)] + #[case::weekday("not-a-weekday".parse::().unwrap_err(), Error::InvalidWeekday)] + fn schedule_parsers_share_the_crate_error(#[case] error: Error, #[case] expected: Error) { + assert_eq!(error, expected); + } } diff --git a/litellm-rust/crates/cost/src/lib.rs b/litellm-rust/crates/cost/src/lib.rs index 24c69afd9d4..192dfa361d6 100644 --- a/litellm-rust/crates/cost/src/lib.rs +++ b/litellm-rust/crates/cost/src/lib.rs @@ -1,383 +1,9 @@ +pub mod call; mod error; +pub mod money; +pub mod price; +pub mod priced; +pub mod schedule; +pub mod usage; -pub use error::Error; - -#[derive(Clone, Copy, Debug, PartialEq)] -pub enum Rate { - Missing, - Null, - Value(f64), -} - -impl Rate { - fn value(self) -> Option { - match self { - Self::Value(value) => Some(value), - Self::Missing | Self::Null => None, - } - } - - fn or(self, fallback: Self) -> Self { - if self.value().is_some() { - self - } else { - fallback - } - } -} - -#[derive(Clone, Copy, Debug, PartialEq)] -pub struct Rates { - pub input: Rate, - pub output: Rate, - pub cache_read: Rate, - pub cache_write: Rate, - pub cache_write_1h: Rate, -} - -impl Rates { - pub const EMPTY: Self = Self { - input: Rate::Missing, - output: Rate::Missing, - cache_read: Rate::Missing, - cache_write: Rate::Missing, - cache_write_1h: Rate::Missing, - }; - - fn overlay(self, base: Self) -> Self { - Self { - input: self.input.or(base.input), - output: self.output.or(base.output), - cache_read: self.cache_read.or(base.cache_read), - cache_write: self.cache_write.or(base.cache_write), - cache_write_1h: self.cache_write_1h.or(base.cache_write_1h), - } - } -} - -#[derive(Clone, Copy, Debug, Eq, PartialEq)] -pub enum ServiceTier { - Standard, - Flex, - Priority, - Fast, - Ultrafast, - Unknown, -} - -#[derive(Clone, Copy, Debug, Eq, PartialEq)] -pub enum ThresholdPolicy { - Exclusive, - Inclusive, -} - -#[derive(Clone, Copy, Debug, Eq, PartialEq)] -pub enum PromptConvention { - IncludesCache, - ExcludesCache, -} - -#[derive(Clone, Copy, Debug, PartialEq)] -pub struct Usage { - pub prompt_tokens: u64, - pub completion_tokens: u64, - pub cache_read_tokens: u64, - pub cache_write_tokens: u64, - pub cache_write_5m_tokens: Option, - pub cache_write_1h_tokens: Option, - pub prompt_convention: PromptConvention, -} - -#[derive(Clone, Copy, Debug, PartialEq)] -pub struct TierRates { - pub tier: ServiceTier, - pub rates: Rates, -} - -#[derive(Clone, Copy, Debug, PartialEq)] -pub struct ThresholdRates<'a> { - pub above_prompt_tokens: u64, - pub standard: Rates, - pub tiers: &'a [TierRates], -} - -#[derive(Clone, Copy, Debug, PartialEq)] -pub struct OffPeakRates { - pub start_utc_minute: u16, - pub end_utc_minute: u16, - pub rates: Rates, -} - -#[derive(Clone, Copy, Debug, PartialEq)] -pub struct Pricing<'a> { - pub standard: Rates, - pub tiers: &'a [TierRates], - pub thresholds: &'a [ThresholdRates<'a>], - pub off_peak: Option, -} - -#[derive(Clone, Copy, Debug, PartialEq)] -pub struct Request { - pub usage: Usage, - pub service_tier: ServiceTier, - pub threshold_policy: ThresholdPolicy, - pub region_multiplier: Option, - pub billed_at_utc_minute: Option, -} - -#[derive(Clone, Copy, Debug, PartialEq)] -pub struct Cost { - pub uncached_input: f64, - pub cache_read: f64, - pub cache_write_5m: f64, - pub cache_write_1h: f64, - pub output: f64, - pub multiplier: f64, - pub rates: EffectiveRates, -} - -impl Cost { - pub fn input(self) -> f64 { - (self.uncached_input + self.cache_read + self.cache_write_5m + self.cache_write_1h) - * self.multiplier - } - - pub fn output(self) -> f64 { - self.output * self.multiplier - } - - pub fn total(self) -> f64 { - self.input() + self.output() - } -} - -#[derive(Clone, Copy, Debug, PartialEq)] -pub struct EffectiveRates { - pub input: f64, - pub output: f64, - pub cache_read: f64, - pub cache_write_5m: f64, - pub cache_write_1h: f64, -} - -fn selected_tier(tier: ServiceTier) -> ServiceTier { - if tier == ServiceTier::Fast { - ServiceTier::Priority - } else { - tier - } -} - -#[derive(Clone, Debug)] -struct CompiledThreshold { - above_prompt_tokens: u64, - standard: Rates, - tiers: Vec, -} - -#[derive(Clone, Debug)] -pub struct PricingPlan { - standard: Rates, - tiers: Vec, - thresholds: Vec, - off_peak: Option, -} - -fn valid_rates(rates: Rates) -> bool { - [ - rates.input, - rates.output, - rates.cache_read, - rates.cache_write, - rates.cache_write_1h, - ] - .into_iter() - .all(|rate| { - rate.value() - .is_none_or(|value| value.is_finite() && value >= 0.0) - }) -} - -fn validate_tiers(tiers: &[TierRates], duplicate: Error) -> Result<(), Error> { - if tiers.iter().any(|entry| !valid_rates(entry.rates)) { - return Err(Error::InvalidRate); - } - if tiers.iter().enumerate().any(|(index, entry)| { - matches!( - entry.tier, - ServiceTier::Standard | ServiceTier::Unknown | ServiceTier::Fast - ) || tiers[..index] - .iter() - .any(|previous| previous.tier == entry.tier) - }) { - return Err(duplicate); - } - Ok(()) -} - -pub fn compile(pricing: &Pricing<'_>) -> Result { - if !valid_rates(pricing.standard) { - return Err(Error::InvalidRate); - } - validate_tiers(pricing.tiers, Error::DuplicateTier)?; - if let Some(window) = pricing.off_peak { - if window.start_utc_minute >= 1440 - || window.end_utc_minute > 1440 - || window.start_utc_minute >= window.end_utc_minute - { - return Err(Error::InvalidOffPeakWindow); - } - if !valid_rates(window.rates) { - return Err(Error::InvalidRate); - } - } - let mut thresholds: Vec<_> = pricing - .thresholds - .iter() - .map(|entry| { - if !valid_rates(entry.standard) { - return Err(Error::InvalidRate); - } - validate_tiers(entry.tiers, Error::DuplicateThresholdTier)?; - Ok(CompiledThreshold { - above_prompt_tokens: entry.above_prompt_tokens, - standard: entry.standard, - tiers: entry.tiers.to_vec(), - }) - }) - .collect::>()?; - thresholds.sort_unstable_by_key(|entry| entry.above_prompt_tokens); - if thresholds - .windows(2) - .any(|pair| pair[0].above_prompt_tokens == pair[1].above_prompt_tokens) - { - return Err(Error::DuplicateThreshold); - } - Ok(PricingPlan { - standard: pricing.standard, - tiers: pricing.tiers.to_vec(), - thresholds, - off_peak: pricing.off_peak, - }) -} - -impl PricingPlan { - fn resolve_rates(&self, request: &Request, threshold_tokens: u64) -> Result { - let tier = selected_tier(request.service_tier); - let base = self - .tiers - .iter() - .find(|entry| tier != ServiceTier::Standard && entry.tier == tier) - .map_or(self.standard, |entry| entry.rates.overlay(self.standard)); - let threshold = self.thresholds.iter().rev().find(|entry| { - threshold_tokens > entry.above_prompt_tokens - || (request.threshold_policy == ThresholdPolicy::Inclusive - && threshold_tokens == entry.above_prompt_tokens) - }); - let selected = threshold.map_or(base, |entry| { - let standard = entry.standard.overlay(base); - entry - .tiers - .iter() - .find(|specific| tier != ServiceTier::Standard && specific.tier == tier) - .map_or(standard, |specific| specific.rates.overlay(standard)) - }); - match self.off_peak { - None => Ok(selected), - Some(window) => { - if window.start_utc_minute >= 1440 - || window.end_utc_minute > 1440 - || window.start_utc_minute >= window.end_utc_minute - { - return Err(Error::InvalidOffPeakWindow); - } - let minute = request - .billed_at_utc_minute - .ok_or(Error::InvalidBillingTime)?; - if minute >= 1440 { - return Err(Error::InvalidBillingTime); - } - if (window.start_utc_minute..window.end_utc_minute).contains(&minute) { - Ok(window.rates.overlay(selected)) - } else { - Ok(selected) - } - } - } - } - - fn checked_rate(rate: Rate, missing: Error) -> Result { - let value = rate.value().ok_or(missing)?; - if !value.is_finite() || value < 0.0 { - return Err(Error::InvalidRate); - } - Ok(value) - } - - pub fn calculate(&self, request: &Request) -> Result { - let usage = request.usage; - let cached = usage - .cache_read_tokens - .checked_add(usage.cache_write_tokens) - .ok_or(Error::TokenCountOverflow)?; - let (regular, threshold_tokens) = match usage.prompt_convention { - PromptConvention::IncludesCache => ( - usage - .prompt_tokens - .checked_sub(cached) - .ok_or(Error::CacheExceedsPrompt)?, - usage.prompt_tokens, - ), - PromptConvention::ExcludesCache => ( - usage.prompt_tokens, - usage - .prompt_tokens - .checked_add(cached) - .ok_or(Error::TokenCountOverflow)?, - ), - }; - let writes = match (usage.cache_write_5m_tokens, usage.cache_write_1h_tokens) { - (None, None) => (usage.cache_write_tokens, 0), - (Some(five), Some(one)) if five.checked_add(one) == Some(usage.cache_write_tokens) => { - (five, one) - } - _ => return Err(Error::InvalidCacheWriteDetails), - }; - let rates = self.resolve_rates(request, threshold_tokens)?; - let input = Self::checked_rate(rates.input, Error::MissingInputRate)?; - let output = Self::checked_rate(rates.output, Error::MissingOutputRate)?; - let read = Self::checked_rate(rates.cache_read.or(rates.input), Error::MissingInputRate)?; - let write = Self::checked_rate(rates.cache_write.or(rates.input), Error::MissingInputRate)?; - let write_1h = Self::checked_rate( - rates.cache_write_1h.or(rates.cache_write).or(rates.input), - Error::MissingInputRate, - )?; - let multiplier = request.region_multiplier.unwrap_or(1.0); - if !multiplier.is_finite() || multiplier <= 0.0 { - return Err(Error::InvalidRegionMultiplier); - } - let cost = Cost { - uncached_input: regular as f64 * input, - cache_read: usage.cache_read_tokens as f64 * read, - cache_write_5m: writes.0 as f64 * write, - cache_write_1h: writes.1 as f64 * write_1h, - output: usage.completion_tokens as f64 * output, - multiplier, - rates: EffectiveRates { - input, - output, - cache_read: read, - cache_write_5m: write, - cache_write_1h: write_1h, - }, - }; - if !cost.total().is_finite() { - return Err(Error::TokenCountOverflow); - } - Ok(cost) - } -} - -pub fn calculate(pricing: &Pricing<'_>, request: &Request) -> Result { - compile(pricing)?.calculate(request) -} +pub use error::*; diff --git a/litellm-rust/crates/cost/src/money.rs b/litellm-rust/crates/cost/src/money.rs new file mode 100644 index 00000000000..0140fff52d2 --- /dev/null +++ b/litellm-rust/crates/cost/src/money.rs @@ -0,0 +1,257 @@ +use std::iter::Sum; +use std::ops::{Add, Mul}; + +use crate::priced::Quantity; + +#[derive(Clone, Copy, Debug, PartialEq, PartialOrd)] +pub struct UnitPrice(f64); + +#[derive(Clone, Copy, Debug, Default, PartialEq, PartialOrd)] +pub struct Usd(f64); + +#[derive(Clone, Copy, Debug, PartialEq, PartialOrd)] +pub struct Multiplier(f64); + +impl UnitPrice { + pub fn new(usd_per_unit: f64) -> Option { + (usd_per_unit.is_finite() && usd_per_unit >= 0.0).then_some(Self(usd_per_unit)) + } + pub fn get(self) -> f64 { + self.0 + } +} + +impl Usd { + pub const ZERO: Self = Self(0.0); + + pub fn new(amount: f64) -> Option { + (amount.is_finite() && amount >= 0.0).then_some(Self(amount)) + } + pub fn get(self) -> f64 { + self.0 + } +} + +impl Multiplier { + pub fn new(factor: f64) -> Option { + (factor.is_finite() && factor >= 1.0).then_some(Self(factor)) + } + pub fn get(self) -> f64 { + self.0 + } +} + +impl Mul for UnitPrice { + type Output = Usd; + + fn mul(self, quantity: Quantity) -> Usd { + match quantity { + Quantity::Count(units) => Usd(self.0 * units as f64), + Quantity::Duration(duration) => Usd(self.0 * duration.as_secs_f64()), + } + } +} + +impl Mul for UnitPrice { + type Output = UnitPrice; + + fn mul(self, factor: Multiplier) -> UnitPrice { + UnitPrice(self.0 * factor.0) + } +} + +impl Mul for Usd { + type Output = Usd; + + fn mul(self, factor: Multiplier) -> Usd { + Usd(self.0 * factor.0) + } +} + +impl Add for Usd { + type Output = Usd; + + fn add(self, other: Usd) -> Usd { + Usd(self.0 + other.0) + } +} + +impl Sum for Usd { + fn sum>(amounts: I) -> Usd { + amounts.fold(Usd::ZERO, Add::add) + } +} + +#[cfg(feature = "serde")] +mod serde_impls { + use std::fmt; + + use serde::de::{self, Deserialize, Deserializer, Visitor}; + + use super::{Multiplier, UnitPrice}; + + #[derive(Clone, Copy)] + struct Number(&'static str); + + impl Visitor<'_> for Number { + type Value = f64; + + fn expecting(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result { + formatter.write_str(self.0) + } + + fn visit_i64(self, value: i64) -> Result { + Ok(value as f64) + } + + fn visit_u64(self, value: u64) -> Result { + Ok(value as f64) + } + + fn visit_f64(self, value: f64) -> Result { + Ok(value) + } + + fn visit_str(self, value: &str) -> Result { + value + .trim() + .parse() + .map_err(|_| E::invalid_value(de::Unexpected::Str(value), &self)) + } + } + + fn checked<'de, D: Deserializer<'de>, T>( + deserializer: D, + expected: Number, + new: fn(f64) -> Option, + ) -> Result { + let number = deserializer.deserialize_any(expected)?; + new(number) + .ok_or_else(|| de::Error::invalid_value(de::Unexpected::Float(number), &expected)) + } + + impl<'de> Deserialize<'de> for UnitPrice { + fn deserialize>(deserializer: D) -> Result { + checked(deserializer, Number("a non-negative USD amount"), Self::new) + } + } + + impl<'de> Deserialize<'de> for Multiplier { + fn deserialize>(deserializer: D) -> Result { + checked( + deserializer, + Number("a multiplier of at least 1"), + Self::new, + ) + } + } +} + +#[cfg(feature = "schema")] +mod schema { + use std::borrow::Cow; + + use schemars::{JsonSchema, Schema, SchemaGenerator, json_schema}; + + use super::{Multiplier, UnitPrice}; + + impl JsonSchema for UnitPrice { + fn inline_schema() -> bool { + true + } + fn schema_name() -> Cow<'static, str> { + "UnitPrice".into() + } + fn json_schema(_: &mut SchemaGenerator) -> Schema { + json_schema!({"type": "number", "minimum": 0}) + } + } + + impl JsonSchema for Multiplier { + fn inline_schema() -> bool { + true + } + fn schema_name() -> Cow<'static, str> { + "Multiplier".into() + } + fn json_schema(_: &mut SchemaGenerator) -> Schema { + json_schema!({"type": "number", "minimum": 1}) + } + } +} + +#[cfg(test)] +mod tests { + use rstest::rstest; + + use super::*; + + #[rstest] + #[case::price(0.5)] + #[case::free(0.0)] + fn unit_price_keeps_finite_non_negative_values(#[case] value: f64) { + assert_eq!(UnitPrice::new(value).map(UnitPrice::get), Some(value)); + } + + #[rstest] + #[case::negative(-1e-9)] + #[case::nan(f64::NAN)] + #[case::infinite(f64::INFINITY)] + fn unit_price_rejects_what_no_price_can_be(#[case] value: f64) { + assert_eq!(UnitPrice::new(value), None); + } + + #[rstest] + #[case::uplift(1.1, true)] + #[case::neutral(1.0, true)] + #[case::discount(0.5, false)] + #[case::nan(f64::NAN, false)] + fn multiplier_is_never_a_discount(#[case] value: f64, #[case] accepted: bool) { + assert_eq!(Multiplier::new(value).is_some(), accepted); + } + + #[rstest] + fn amounts_follow_units_times_price_then_uplift() { + let price = UnitPrice::new(0.25).unwrap(); + let uplift = Multiplier::new(2.0).unwrap(); + + let total: Usd = [ + price * Quantity::Count(4), + (price * uplift) * Quantity::Count(2), + ] + .into_iter() + .sum(); + + assert_eq!(total.get(), 2.0); + assert_eq!((total * uplift).get(), 4.0); + } + + #[rstest] + fn a_duration_is_billed_in_fractional_seconds() { + let per_second = UnitPrice::new(0.01).unwrap(); + + let amount = per_second * Quantity::Duration(std::time::Duration::from_millis(2_500)); + + assert!((amount.get() - 0.025).abs() < 1e-12); + } + + #[cfg(feature = "serde")] + #[rstest] + #[case::number(serde_json::json!(1.5e-6), Some(1.5e-6))] + #[case::integer(serde_json::json!(2), Some(2.0))] + #[case::numeric_string(serde_json::json!(" 0.000002 "), Some(2e-6))] + #[case::negative(serde_json::json!(-1), None)] + #[case::word(serde_json::json!("free"), None)] + #[case::boolean(serde_json::json!(true), None)] + fn unit_price_reads_numbers_and_numeric_strings( + #[case] written: serde_json::Value, + #[case] expected: Option, + ) { + use serde::Deserialize; + + assert_eq!( + UnitPrice::deserialize(&written).ok().map(UnitPrice::get), + expected + ); + } +} diff --git a/litellm-rust/crates/cost/src/price.rs b/litellm-rust/crates/cost/src/price.rs new file mode 100644 index 00000000000..a90bedacb0f --- /dev/null +++ b/litellm-rust/crates/cost/src/price.rs @@ -0,0 +1,505 @@ +mod meter; + +use std::collections::BTreeMap; +use std::num::NonZeroU64; + +pub use meter::*; + +use crate::call::{InferenceGeo, Speed}; +use crate::money::{Multiplier, UnitPrice}; +use crate::schedule::Schedule; + +#[derive(Clone, Copy, Debug, Default, PartialEq, Eq, PartialOrd, Ord, Hash)] +pub enum ServiceTier { + #[default] + Standard, + Priority, + Flex, + Ultrafast, + Balanced, +} + +#[derive(Clone, Copy, Debug, Default, PartialEq, Eq, PartialOrd, Ord, Hash)] +pub enum Band { + #[default] + Base, + OverPromptTokens(NonZeroU64), +} + +#[derive(Clone, Copy, Debug, Default, PartialEq, Eq, PartialOrd, Ord, Hash)] +pub struct When { + pub tier: ServiceTier, + pub band: Band, +} + +#[derive(Clone, Copy, Debug, Default, PartialEq, Eq, PartialOrd, Ord, Hash)] +pub enum Channel { + #[default] + Online, + Batch, + Decisions, +} + +#[derive(Clone, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)] +pub struct Price { + pub meter: Meter, + pub channel: Channel, + pub when: When, +} + +#[derive(Clone, Debug, Default, PartialEq)] +pub struct Prices { + rates: BTreeMap, + off_peak: Option, + billing: Billing, +} + +/// Row attributes pricing reads, lowered by model-catalog from the row and its provider +#[derive(Clone, Copy, Debug, Default, PartialEq, Eq)] +pub struct Billing { + /// Unit the row bills text in. `Character` only when the row prices characters and not tokens + pub text_unit: TextUnit, + /// Whether a band's floor is in the band. xAI rows bill the band from the floor itself + pub band_edge: BandEdge, + /// Rate a cache read borrows when the row has no cache-read price. Fireworks rows borrow half the input rate + pub cache_read_fallback: CacheReadFallback, + /// Whether a row with no token price is priced by wall clock, from its `mode` + pub wall_clock: WallClock, + /// From `web_search_billing_unit`, whether web search and Maps grounding bill per query or once per prompt + pub web_search: WebSearchUnit, + /// Whether `tiered_pricing` bills each meter in its own tier or the whole call at the tier the input picks. + /// Dashscope rows bill the whole call (inferred, 10 checks no other row has that shape) + pub tier_scope: TierScope, +} + +#[derive(Clone, Copy, Debug, Default, PartialEq, Eq)] +pub enum TextUnit { + #[default] + Token, + Character, +} + +#[derive(Clone, Copy, Debug, Default, PartialEq, Eq)] +pub enum BandEdge { + #[default] + Exclusive, + Inclusive, +} + +#[derive(Clone, Copy, Debug, Default, PartialEq, Eq)] +pub enum CacheReadFallback { + #[default] + InputRate, + HalfInputRate, +} + +/// `ResponseTime` for `chat`, `completion`, `embedding`, `responses` or no mode, `AudioFile` for +/// `audio_transcription`, which prices the input audio when it has one, `Never` for every other mode +#[derive(Clone, Copy, Debug, Default, PartialEq, Eq)] +pub enum WallClock { + Never, + #[default] + ResponseTime, + AudioFile, +} + +#[derive(Clone, Copy, Debug, Default, PartialEq, Eq)] +pub enum WebSearchUnit { + PerQuery, + #[default] + PerPrompt, +} + +#[derive(Clone, Copy, Debug, Default, PartialEq, Eq)] +pub enum TierScope { + #[default] + PerMeter, + WholeCall, +} + +#[derive(Clone, Debug, PartialEq)] +pub struct OffPeak { + schedule: Schedule, + rates: OffPeakRates, +} + +#[derive(Clone, Copy, Debug, Default, PartialEq)] +pub struct OffPeakRates { + pub input: Option, + pub output: Option, + pub cache_read: Option, + pub explicit_cache_read: Option, + pub cache_write: Option, + pub reasoning: Option, +} + +#[derive(Clone, Debug, Default, PartialEq)] +pub struct Adjustments(BTreeMap); + +/// Which call fact picks which multiplier in `Adjustments`, lowered by model-catalog from the row +#[derive(Clone, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)] +pub enum Uplift { + /// OpenAI data residency + Residency(Residency), + /// Vertex non-global endpoint, from `regional_endpoint_uplift_multiplier` + NonGlobalEndpoint, + /// Anthropic `speed: fast`, from `provider_specific_entry.fast` + Speed(Speed), + /// Anthropic and Bedrock geo, from `provider_specific_entry.`, lowercased as on main + InferenceGeo(InferenceGeo), +} + +#[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)] +pub enum Residency { + Eu, + Us, +} + +impl Band { + pub fn over_prompt_tokens(tokens: u64) -> Self { + NonZeroU64::new(tokens).map_or(Self::Base, Self::OverPromptTokens) + } +} + +impl Price { + pub fn online(meter: Meter, when: When) -> Self { + Self { + meter, + channel: Channel::Online, + when, + } + } +} + +impl Prices { + pub fn new( + rates: BTreeMap, + off_peak: Option, + billing: Billing, + ) -> Self { + Self { + rates, + off_peak, + billing, + } + } + + pub fn get(&self, price: &Price) -> Option { + self.rates.get(price).copied() + } + + pub fn iter(&self) -> impl Iterator { + self.rates.iter().map(|(price, rate)| (price, *rate)) + } + + pub fn off_peak(&self) -> Option<&OffPeak> { + self.off_peak.as_ref() + } + + pub fn billing(&self) -> Billing { + self.billing + } + + /// The online prompt band a request falls in. Every token meter's `_above_` prices count, + /// and the highest band the prompt passes wins. `billing.band_edge` says whether the floor is in the band. + pub fn band(&self, tier: ServiceTier, prompt_tokens: u64) -> Band { + let inclusive = self.billing.band_edge == BandEdge::Inclusive; + self.rates + .keys() + .filter(|price| { + price.channel == Channel::Online + && price.meter.sets_prompt_band() + && (price.when.tier == ServiceTier::Standard || price.when.tier == tier) + }) + .filter_map(|price| match price.when.band { + Band::OverPromptTokens(floor) => Some(floor.get()), + Band::Base => None, + }) + .filter(|floor| prompt_tokens > *floor || (inclusive && prompt_tokens == *floor)) + .max() + .map_or(Band::Base, Band::over_prompt_tokens) + } + + /// The online rate for `meter`, tried at `(tier, band)`, `(standard, band)`, `(tier, base)` + /// and `(standard, base)` in that order. A missing cache price falls back to the input price + /// resolved the same way. + pub fn rate(&self, meter: &Meter, tier: ServiceTier, band: Band) -> Option { + [ + When { tier, band }, + When { + tier: ServiceTier::Standard, + band, + }, + When { + tier, + band: Band::Base, + }, + When::default(), + ] + .into_iter() + .find_map(|when| self.get(&Price::online(meter.clone(), when))) + .or_else(|| self.rate(&meter.fallback()?, tier, band)) + } +} + +impl OffPeak { + pub fn new(schedule: Schedule, rates: OffPeakRates) -> Option { + (rates != OffPeakRates::default()).then_some(Self { schedule, rates }) + } + + pub fn schedule(&self) -> &Schedule { + &self.schedule + } + + pub fn rates(&self) -> &OffPeakRates { + &self.rates + } +} + +impl Adjustments { + pub fn new(uplifts: impl IntoIterator) -> Self { + Self(uplifts.into_iter().collect()) + } + + pub fn get(&self, uplift: &Uplift) -> Option { + self.0.get(uplift).copied() + } +} + +#[cfg(test)] +mod tests { + use rstest::{fixture, rstest}; + + use super::*; + + const INPUT: Meter = Meter::TEXT_INPUT; + const OUTPUT: Meter = Meter::Output(OutputUnit::Token(TokenKind::Text)); + const CACHE_READ: Meter = Meter::CacheRead { + kind: TokenKind::Text, + mode: CacheReadMode::Implicit, + }; + const CACHE_WRITE_HOUR: Meter = Meter::CacheWrite { + kind: TokenKind::Text, + ttl: CacheTtl::OneHour, + }; + + fn over(tokens: u64) -> Band { + Band::over_prompt_tokens(tokens) + } + + fn prices(rates: &[(Meter, Channel, ServiceTier, Band, f64)]) -> Prices { + priced_with(rates, Billing::default()) + } + + fn priced_with(rates: &[(Meter, Channel, ServiceTier, Band, f64)], billing: Billing) -> Prices { + Prices::new( + rates + .iter() + .map(|(meter, channel, tier, band, rate)| { + ( + Price { + meter: meter.clone(), + channel: *channel, + when: When { + tier: *tier, + band: *band, + }, + }, + UnitPrice::new(*rate).unwrap(), + ) + }) + .collect(), + None, + billing, + ) + } + + #[fixture] + fn priority_with_a_standard_band() -> Prices { + use Channel::Online; + use ServiceTier::{Priority, Standard}; + prices(&[ + (INPUT, Online, Standard, Band::Base, 1.0), + (INPUT, Online, Priority, Band::Base, 2.0), + (INPUT, Online, Standard, over(200_000), 10.0), + (OUTPUT, Online, Standard, Band::Base, 3.0), + (OUTPUT, Online, Priority, over(200_000), 30.0), + ]) + } + + #[rstest] + #[case::below(1_000, false, Band::Base)] + #[case::at_floor_exclusive(200_000, false, Band::Base)] + #[case::at_floor_inclusive(200_000, true, over(200_000))] + #[case::above(200_001, false, over(200_000))] + fn the_band_starts_strictly_above_its_floor_unless_inclusive( + #[case] prompt_tokens: u64, + #[case] inclusive: bool, + #[case] expected: Band, + ) { + let band_edge = if inclusive { + BandEdge::Inclusive + } else { + BandEdge::Exclusive + }; + let prices = priced_with( + &[( + INPUT, + Channel::Online, + ServiceTier::Standard, + over(200_000), + 10.0, + )], + Billing { + band_edge, + ..Billing::default() + }, + ); + assert_eq!(prices.band(ServiceTier::Standard, prompt_tokens), expected); + } + + #[rstest] + fn a_band_only_another_tier_prices_does_not_apply() { + let prices = prices(&[( + OUTPUT, + Channel::Online, + ServiceTier::Flex, + over(100_000), + 1.0, + )]); + assert_eq!(prices.band(ServiceTier::Standard, 300_000), Band::Base); + assert_eq!(prices.band(ServiceTier::Flex, 300_000), over(100_000)); + } + + #[rstest] + fn batch_bands_never_pick_the_online_band() { + let prices = prices(&[( + INPUT, + Channel::Batch, + ServiceTier::Standard, + over(100_000), + 1.0, + )]); + assert_eq!(prices.band(ServiceTier::Standard, 300_000), Band::Base); + } + + #[rstest] + fn the_highest_band_passed_wins() { + let prices = prices(&[ + ( + INPUT, + Channel::Online, + ServiceTier::Standard, + over(100_000), + 1.0, + ), + ( + OUTPUT, + Channel::Online, + ServiceTier::Standard, + over(200_000), + 1.0, + ), + ]); + assert_eq!(prices.band(ServiceTier::Standard, 300_000), over(200_000)); + } + + #[rstest] + #[case::tier_base(INPUT, ServiceTier::Priority, Band::Base, Some(2.0))] + #[case::standard_band_beats_tier_base(INPUT, ServiceTier::Priority, over(200_000), Some(10.0))] + #[case::tier_band(OUTPUT, ServiceTier::Priority, over(200_000), Some(30.0))] + #[case::standard_base_when_nothing_else(OUTPUT, ServiceTier::Flex, over(200_000), Some(3.0))] + #[case::cache_read_falls_back_to_the_resolved_input( + CACHE_READ, + ServiceTier::Priority, + over(200_000), + Some(10.0) + )] + #[case::one_hour_write_falls_back_through_write_to_input( + CACHE_WRITE_HOUR, + ServiceTier::Priority, + Band::Base, + Some(2.0) + )] + #[case::twelve_hour_write_falls_back_through_write_to_input( + Meter::CacheWrite { kind: TokenKind::Text, ttl: CacheTtl::TwelveHours }, + ServiceTier::Standard, + Band::Base, + Some(1.0) + )] + #[case::no_fallback_for_other_meters( + Meter::Input(InputUnit::Token(TokenKind::Audio)), + ServiceTier::Standard, + Band::Base, + None + )] + fn rate_walks_tier_then_band( + priority_with_a_standard_band: Prices, + #[case] meter: Meter, + #[case] tier: ServiceTier, + #[case] band: Band, + #[case] expected: Option, + ) { + assert_eq!( + priority_with_a_standard_band + .rate(&meter, tier, band) + .map(UnitPrice::get), + expected + ); + } + + #[rstest] + fn an_explicit_cache_price_wins_over_the_input_fallback() { + let prices = prices(&[ + ( + INPUT, + Channel::Online, + ServiceTier::Standard, + Band::Base, + 1.0, + ), + ( + CACHE_READ, + Channel::Online, + ServiceTier::Standard, + Band::Base, + 0.0, + ), + ]); + assert_eq!( + prices + .rate(&CACHE_READ, ServiceTier::Standard, Band::Base) + .map(UnitPrice::get), + Some(0.0) + ); + } + + #[rstest] + fn off_peak_needs_a_price() { + let schedule = Schedule { + windows: Vec::new(), + weekday_timezone: None, + }; + assert_eq!( + OffPeak::new(schedule.clone(), OffPeakRates::default()), + None + ); + assert!( + OffPeak::new( + schedule, + OffPeakRates { + output: UnitPrice::new(1.0), + ..OffPeakRates::default() + } + ) + .is_some() + ); + } + + #[rstest] + #[case::zero(0, Band::Base)] + #[case::positive(272_000, Band::OverPromptTokens(NonZeroU64::new(272_000).unwrap()))] + fn a_zero_floor_is_the_base_band(#[case] tokens: u64, #[case] expected: Band) { + assert_eq!(Band::over_prompt_tokens(tokens), expected); + } +} diff --git a/litellm-rust/crates/cost/src/price/meter.rs b/litellm-rust/crates/cost/src/price/meter.rs new file mode 100644 index 00000000000..6e68981d011 --- /dev/null +++ b/litellm-rust/crates/cost/src/price/meter.rs @@ -0,0 +1,197 @@ +use std::num::NonZeroU64; + +#[derive(Clone, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)] +pub enum Meter { + Input(InputUnit), + Output(OutputUnit), + CacheRead { + kind: TokenKind, + mode: CacheReadMode, + }, + CacheWrite { + kind: TokenKind, + ttl: CacheTtl, + }, + Feature(Feature), + WallClockSecond, + WebSearch(SearchContextSize), + Guardrail(GuardrailUnit), +} + +#[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)] +pub enum TokenKind { + Text, + Audio, + Image, + Video, +} + +#[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)] +pub enum InputUnit { + Token(TokenKind), + Character, + Image, + Pixel, + Query { over_results: Option }, + Request, + Second, + TextMessage, + AudioSecond, + VideoSecond { over_seconds: Option }, +} + +#[derive(Clone, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)] +pub enum OutputUnit { + Token(TokenKind), + ReasoningToken, + Character, + Image(Option), + Pixel, + Second(Option), + VideoSecond, +} + +#[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)] +pub enum CacheReadMode { + Implicit, + Explicit, +} + +#[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)] +pub enum CacheTtl { + Default, + OneHour, + TwelveHours, +} + +#[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)] +pub enum Feature { + AnnotationPage, + CitationToken, + CodeInterpreterSession, + GoogleMapsGroundingQuery, + OcrCredit, + OcrPage, + /// `browser_open_cost_per_request`, Groq `browser.open` + BrowserOpen, + /// `cost_per_additional_result`, Parallel AI results past the ones the query price includes + AdditionalResult, +} + +#[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)] +pub enum SearchContextSize { + Low, + Medium, + High, +} + +#[derive(Clone, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)] +pub struct GuardrailUnit(Box); + +#[derive(Clone, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)] +pub struct ImageSize(Box); + +#[derive(Clone, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)] +pub struct Resolution(Box); + +impl Meter { + pub(crate) const TEXT_INPUT: Self = Self::Input(InputUnit::Token(TokenKind::Text)); + + pub(crate) fn sets_prompt_band(&self) -> bool { + matches!( + self, + Self::Input(InputUnit::Token(TokenKind::Text)) + | Self::Output(OutputUnit::Token(TokenKind::Text)) + | Self::CacheRead { + kind: TokenKind::Text, + mode: CacheReadMode::Implicit, + } + | Self::CacheWrite { + kind: TokenKind::Text, + .. + } + ) + } + + pub(crate) fn fallback(&self) -> Option { + match self { + Self::CacheRead { + kind: TokenKind::Text, + mode: CacheReadMode::Implicit, + } + | Self::CacheWrite { + kind: TokenKind::Text, + ttl: CacheTtl::Default, + } => Some(Self::TEXT_INPUT), + Self::CacheWrite { + kind: TokenKind::Text, + ttl: CacheTtl::OneHour | CacheTtl::TwelveHours, + } => Some(Self::CacheWrite { + kind: TokenKind::Text, + ttl: CacheTtl::Default, + }), + _ => None, + } + } +} + +impl GuardrailUnit { + pub fn new(counter: &str) -> Self { + Self(counter.into()) + } + pub fn as_str(&self) -> &str { + &self.0 + } +} + +impl ImageSize { + pub fn new(label: &str) -> Option { + let number = label.strip_suffix('K').unwrap_or(label); + let (whole, fraction) = number.split_once('.').unwrap_or((number, "0")); + (is_digits(whole) && is_digits(fraction)).then(|| Self(label.into())) + } + pub fn as_str(&self) -> &str { + &self.0 + } +} + +impl Resolution { + pub fn new(label: &str) -> Option { + let digits = label.strip_suffix(['p', 'k'])?; + is_digits(digits).then(|| Self(label.into())) + } + pub fn as_str(&self) -> &str { + &self.0 + } +} + +fn is_digits(value: &str) -> bool { + !value.is_empty() && value.bytes().all(|byte| byte.is_ascii_digit()) +} + +#[cfg(test)] +mod tests { + use rstest::rstest; + + use super::*; + + #[rstest] + #[case::kilo("1K", true)] + #[case::half_kilo("0.5K", true)] + #[case::pixels("1024", true)] + #[case::word("big", false)] + #[case::bare_suffix("K", false)] + #[case::trailing_dot("1.K", false)] + fn image_size_labels_are_numbers_with_an_optional_k(#[case] label: &str, #[case] valid: bool) { + assert_eq!(ImageSize::new(label).is_some(), valid); + } + + #[rstest] + #[case::progressive("720p", true)] + #[case::kilo("4k", true)] + #[case::bare("720", false)] + #[case::word("hd", false)] + fn resolution_labels_are_numbers_with_p_or_k(#[case] label: &str, #[case] valid: bool) { + assert_eq!(Resolution::new(label).is_some(), valid); + } +} diff --git a/litellm-rust/crates/cost/src/priced.rs b/litellm-rust/crates/cost/src/priced.rs new file mode 100644 index 00000000000..535560c1b9a --- /dev/null +++ b/litellm-rust/crates/cost/src/priced.rs @@ -0,0 +1,53 @@ +use std::time::Duration; + +use crate::money::{UnitPrice, Usd}; +use crate::price::Meter; + +#[derive(Clone, Debug, PartialEq)] +pub enum Outcome { + Priced(Priced), + /// No meter with usage had a price + Unpriced { + missing: Vec, + }, +} + +#[derive(Clone, Debug, PartialEq)] +pub struct Priced { + /// One line per billed meter, fallbacks included + pub lines: Vec, + /// Meters with usage and no price and no fallback + pub missing: Vec, + /// Sum of `lines` + pub total: Usd, +} + +#[derive(Clone, Debug, PartialEq)] +pub struct Line { + /// The meter this line bills + pub meter: Meter, + /// How much of the meter the call used + pub quantity: Quantity, + /// Rate after off-peak and before uplifts + pub unit_price: UnitPrice, + /// `unit_price * quantity` times every uplift that applies + pub amount: Usd, + /// The meter whose rate was borrowed, when this meter had none + pub fallback_from: Option, + /// Which half of main's `(prompt, completion)` pair the line belongs to + pub side: Side, +} + +/// A billed amount of one meter +#[derive(Clone, Copy, Debug, PartialEq, Eq)] +pub enum Quantity { + Count(u64), + Duration(Duration), +} + +/// Meters main bills as one total (video, Replicate, Amazon Transcribe) take a fixed side, chosen in 09 +#[derive(Clone, Copy, Debug, PartialEq, Eq)] +pub enum Side { + Prompt, + Completion, +} diff --git a/litellm-rust/crates/cost/src/schedule.rs b/litellm-rust/crates/cost/src/schedule.rs new file mode 100644 index 00000000000..adebc044cca --- /dev/null +++ b/litellm-rust/crates/cost/src/schedule.rs @@ -0,0 +1,241 @@ +use std::str::FromStr; + +use crate::Error; + +#[derive(Clone, Debug, PartialEq, Eq)] +pub struct Schedule { + pub windows: Vec, + pub weekday_timezone: Option, +} + +#[derive(Clone, Debug, PartialEq, Eq)] +pub struct Window { + pub hours: Vec, + pub weekdays: Option>, +} + +#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)] +pub struct UtcWindow { + pub start_minute: u16, + pub end_minute: u16, +} + +#[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)] +pub enum Weekday { + Monday, + Tuesday, + Wednesday, + Thursday, + Friday, + Saturday, + Sunday, +} + +impl FromStr for UtcWindow { + type Err = Error; + + fn from_str(window: &str) -> Result { + let (start, end) = window.split_once('-').ok_or(Error::InvalidUtcWindow)?; + Ok(Self { + start_minute: minute_of_day(start).ok_or(Error::InvalidUtcWindow)?, + end_minute: minute_of_day(end).ok_or(Error::InvalidUtcWindow)?, + }) + } +} + +fn minute_of_day(clock: &str) -> Option { + let (hour, minute) = clock.split_once(':')?; + if hour.len() != 2 || minute.len() != 2 { + return None; + } + let hour: u16 = hour.parse().ok().filter(|hour| *hour < 24)?; + let minute: u16 = minute.parse().ok().filter(|minute| *minute < 60)?; + Some(hour * 60 + minute) +} + +impl Weekday { + pub fn from_iso(number: u8) -> Option { + match number { + 1 => Some(Self::Monday), + 2 => Some(Self::Tuesday), + 3 => Some(Self::Wednesday), + 4 => Some(Self::Thursday), + 5 => Some(Self::Friday), + 6 => Some(Self::Saturday), + 7 => Some(Self::Sunday), + _ => None, + } + } +} + +impl FromStr for Weekday { + type Err = Error; + + fn from_str(name: &str) -> Result { + match name.trim().to_ascii_lowercase().as_str() { + "mon" | "monday" => Ok(Self::Monday), + "tue" | "tues" | "tuesday" => Ok(Self::Tuesday), + "wed" | "wednesday" => Ok(Self::Wednesday), + "thu" | "thur" | "thurs" | "thursday" => Ok(Self::Thursday), + "fri" | "friday" => Ok(Self::Friday), + "sat" | "saturday" => Ok(Self::Saturday), + "sun" | "sunday" => Ok(Self::Sunday), + _ => Err(Error::InvalidWeekday), + } + } +} + +#[cfg(feature = "serde")] +mod serde_impls { + use std::fmt; + + use serde::de::{self, Deserialize, Deserializer, Visitor}; + + use super::{UtcWindow, Weekday}; + + impl<'de> Deserialize<'de> for UtcWindow { + fn deserialize>(deserializer: D) -> Result { + let window = String::deserialize(deserializer)?; + window.parse().map_err(de::Error::custom) + } + } + + struct WrittenWeekday; + + impl Visitor<'_> for WrittenWeekday { + type Value = Weekday; + + fn expecting(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result { + formatter.write_str("an ISO weekday number 1-7 or an English day name") + } + + fn visit_u64(self, number: u64) -> Result { + u8::try_from(number) + .ok() + .and_then(Weekday::from_iso) + .ok_or_else(|| E::invalid_value(de::Unexpected::Unsigned(number), &self)) + } + + fn visit_i64(self, number: i64) -> Result { + u64::try_from(number) + .map_err(|_| E::invalid_value(de::Unexpected::Signed(number), &self)) + .and_then(|number| self.visit_u64(number)) + } + + fn visit_str(self, name: &str) -> Result { + name.parse() + .map_err(|_| E::invalid_value(de::Unexpected::Str(name), &self)) + } + } + + impl<'de> Deserialize<'de> for Weekday { + fn deserialize>(deserializer: D) -> Result { + deserializer.deserialize_any(WrittenWeekday) + } + } +} + +#[cfg(feature = "schema")] +mod schema { + use std::borrow::Cow; + + use schemars::{JsonSchema, Schema, SchemaGenerator, json_schema}; + + use super::{UtcWindow, Weekday}; + + impl JsonSchema for UtcWindow { + fn inline_schema() -> bool { + true + } + fn schema_name() -> Cow<'static, str> { + "UtcWindow".into() + } + fn json_schema(_: &mut SchemaGenerator) -> Schema { + json_schema!({ + "type": "string", + "pattern": r"^([01]\d|2[0-3]):[0-5]\d-([01]\d|2[0-3]):[0-5]\d$" + }) + } + } + + impl JsonSchema for Weekday { + fn inline_schema() -> bool { + true + } + fn schema_name() -> Cow<'static, str> { + "Weekday".into() + } + fn json_schema(_: &mut SchemaGenerator) -> Schema { + json_schema!({ + "anyOf": [ + {"type": "integer", "minimum": 1, "maximum": 7}, + { + "type": "string", + "pattern": r"(?i)^\s*(mon|monday|tue|tues|tuesday|wed|wednesday|thu|thur|thurs|thursday|fri|friday|sat|saturday|sun|sunday)\s*$" + } + ] + }) + } + } +} + +#[cfg(test)] +mod tests { + use rstest::rstest; + + use super::*; + + #[rstest] + #[case::same_day("10:00-18:30", Some((600, 1110)))] + #[case::wraps_midnight("22:00-02:00", Some((1320, 120)))] + #[case::whole_day("00:00-00:00", Some((0, 0)))] + #[case::hour_out_of_range("25:00-01:00", None)] + #[case::minute_out_of_range("10:60-11:00", None)] + #[case::one_digit_hour("9:00-10:00", None)] + #[case::no_separator("10:00", None)] + fn utc_window_reads_hh_mm_ranges(#[case] text: &str, #[case] expected: Option<(u16, u16)>) { + assert_eq!( + text.parse::() + .ok() + .map(|window| (window.start_minute, window.end_minute)), + expected + ); + } + + #[rstest] + #[case::short("thurs", Some(Weekday::Thursday))] + #[case::upper("SATURDAY", Some(Weekday::Saturday))] + #[case::padded(" mon ", Some(Weekday::Monday))] + #[case::unknown("someday", None)] + #[case::empty("", None)] + fn weekday_names_match_python_after_trimming( + #[case] name: &str, + #[case] expected: Option, + ) { + assert_eq!(name.parse().ok(), expected); + } + + #[rstest] + #[case::monday(1, Some(Weekday::Monday))] + #[case::sunday(7, Some(Weekday::Sunday))] + #[case::zero(0, None)] + #[case::eight(8, None)] + fn weekday_numbers_are_iso(#[case] number: u8, #[case] expected: Option) { + assert_eq!(Weekday::from_iso(number), expected); + } + + #[cfg(feature = "serde")] + #[rstest] + #[case::number(serde_json::json!(6), Some(Weekday::Saturday))] + #[case::name(serde_json::json!("Sun"), Some(Weekday::Sunday))] + #[case::negative(serde_json::json!(-1), None)] + #[case::boolean(serde_json::json!(true), None)] + fn weekday_reads_numbers_and_names( + #[case] written: serde_json::Value, + #[case] expected: Option, + ) { + use serde::Deserialize; + + assert_eq!(Weekday::deserialize(&written).ok(), expected); + } +} diff --git a/litellm-rust/crates/cost/src/usage.rs b/litellm-rust/crates/cost/src/usage.rs new file mode 100644 index 00000000000..8eed83ac87b --- /dev/null +++ b/litellm-rust/crates/cost/src/usage.rs @@ -0,0 +1,104 @@ +use std::collections::BTreeMap; +use std::time::Duration; + +use crate::price::{CacheReadMode, CacheTtl, ImageSize, Resolution, SearchContextSize, TokenKind}; + +/// What one call consumed. Python fills it from the provider response, `calculate` never reads the catalog for it +#[derive(Clone, Debug, Default, PartialEq, Eq)] +pub struct Usage { + /// Prompt size as the provider reported it. Only picks `_above_Nk_tokens` bands, never billed. + /// Filled even when the call bills characters, `0` when the provider reports none + pub prompt_tokens: u64, + /// Billed token quantities by meter + pub tokens: Tokens, + /// Billed character quantities. Used only when the row's `billing.text_unit` is `Character` + pub characters: Characters, + /// Billed time per second meter, never a bare number of seconds or milliseconds + pub durations: BTreeMap, + /// Images sent and returned + pub images: Images, + /// Search queries, web searches and grounding requests + pub search: Search, + /// Realtime text messages, xAI bills them per message + pub text_messages: u64, +} + +#[derive(Clone, Debug, Default, PartialEq, Eq)] +pub struct Tokens { + /// Uncached prompt tokens by modality + pub input: BTreeMap, + /// Prompt tokens served from cache, by modality and implicit or explicit cache + pub cache_read: BTreeMap<(TokenKind, CacheReadMode), u64>, + /// Prompt tokens written to cache, by modality and cache lifetime + pub cache_write: BTreeMap<(TokenKind, CacheTtl), u64>, + /// Completion tokens by modality, reasoning excluded + pub output: BTreeMap, + /// Reasoning tokens, billed at the reasoning rate or the text output rate + pub reasoning: u64, + /// Perplexity citation tokens, prompt side + pub citation: u64, +} + +#[derive(Clone, Copy, Debug, Default, PartialEq, Eq)] +pub struct Characters { + /// Characters of the formatted prompt + pub prompt: u64, + /// Characters of the response text + pub completion: u64, +} + +#[derive(Clone, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)] +pub enum DurationMeter { + /// Response time of the call, for rows priced by wall clock + WallClock, + /// Length of the input audio, a transcription file or a realtime transcription + InputAudio, + /// Realtime voice session time + LiveVoice, + /// Length of generated video, already multiplied by the video count. `resolution` picks `output_cost_per_second_` + OutputVideo { resolution: Option }, +} + +#[derive(Clone, Debug, Default, PartialEq, Eq)] +pub struct Images { + /// Images sent in the request + pub input: u64, + /// Pixels sent in the request, for rows priced per input pixel + pub input_pixels: u64, + /// One entry per returned image, since price can depend on each image's size + pub output: Vec, +} + +#[derive(Clone, Debug, Default, PartialEq, Eq)] +pub struct ImageOut { + /// Size the request asked for, picks a size keyed price like fal's `output_cost_per_image_` + pub size: Option, + /// Pixel width the response reported, for per-pixel and nearest-size pricing + pub width: Option, + /// Pixel height the response reported, for per-pixel and nearest-size pricing + pub height: Option, +} + +#[derive(Clone, Debug, Default, PartialEq, Eq)] +pub struct Search { + /// Search API queries, for `input_cost_per_query` rows + pub queries: u64, + /// Results the query asked for, picks the `max_results_range` tier + pub max_results: Option, + /// Web searches a model ran as a tool + pub web_search: Option, + /// Google Maps grounding requests + pub maps_grounding: u64, + /// Groq `browser.open` calls + pub browser_opens: u64, + /// Results past the ones the query price includes. Python counts them, since the included 10 is a request default + pub additional_results: u64, +} + +#[derive(Clone, Copy, Debug, PartialEq, Eq)] +pub struct WebSearch { + /// Web search requests the model made + pub requests: u64, + /// Context size the request asked for, picks `search_context_size_` + pub context: SearchContextSize, +} diff --git a/litellm-rust/crates/cost/tests/calculation.rs b/litellm-rust/crates/cost/tests/calculation.rs deleted file mode 100644 index af1a73d8c61..00000000000 --- a/litellm-rust/crates/cost/tests/calculation.rs +++ /dev/null @@ -1,535 +0,0 @@ -use litellm_cost::{ - Error, OffPeakRates, Pricing, PromptConvention, Rate, Rates, Request, ServiceTier, - ThresholdPolicy, ThresholdRates, TierRates, Usage, calculate, compile, -}; -use rstest::rstest; - -fn rates(input: Rate, output: Rate) -> Rates { - Rates { - input, - output, - ..Rates::EMPTY - } -} - -fn request() -> Request { - Request { - usage: Usage { - prompt_tokens: 100, - completion_tokens: 20, - cache_read_tokens: 25, - cache_write_tokens: 10, - cache_write_5m_tokens: None, - cache_write_1h_tokens: None, - prompt_convention: PromptConvention::IncludesCache, - }, - service_tier: ServiceTier::Standard, - threshold_policy: ThresholdPolicy::Exclusive, - region_multiplier: None, - billed_at_utc_minute: None, - } -} - -fn pricing(standard: Rates) -> Pricing<'static> { - Pricing { - standard, - tiers: &[], - thresholds: &[], - off_peak: None, - } -} - -#[test] -fn breakdown_and_total_agree() { - let standard = Rates { - cache_read: Rate::Value(0.5), - cache_write: Rate::Value(3.0), - ..rates(Rate::Value(2.0), Rate::Value(4.0)) - }; - let result = calculate(&pricing(standard), &request()).unwrap(); - assert_eq!(result.uncached_input, 65.0 * 2.0); - assert_eq!(result.cache_read, 25.0 * 0.5); - assert_eq!(result.cache_write_5m, 10.0 * 3.0); - assert_eq!(result.output(), 20.0 * 4.0); - assert_eq!(result.total(), result.input() + result.output()); - assert_eq!(result.rates.cache_read, 0.5); -} - -#[rstest] -#[case::missing(Rate::Missing, Rate::Missing, 200.0)] -#[case::null(Rate::Null, Rate::Missing, 200.0)] -#[case::zero(Rate::Value(0.0), Rate::Value(0.0), 130.0)] -fn absent_null_and_zero_cache_rates_are_distinct( - #[case] cache_read: Rate, - #[case] cache_write: Rate, - #[case] expected_input: f64, -) { - let base = rates(Rate::Value(2.0), Rate::Value(4.0)); - let standard = Rates { - cache_read, - cache_write, - ..base - }; - assert_eq!( - calculate(&pricing(standard), &request()).unwrap().input(), - expected_input - ); -} - -#[test] -fn equivalent_prompt_conventions_select_the_same_threshold() { - let threshold = ThresholdRates { - above_prompt_tokens: 90, - standard: rates(Rate::Value(5.0), Rate::Value(8.0)), - tiers: &[], - }; - let specification = Pricing { - standard: rates(Rate::Value(2.0), Rate::Value(4.0)), - tiers: &[], - thresholds: &[threshold], - off_peak: None, - }; - let included = request(); - let excluded = Request { - usage: Usage { - prompt_tokens: 65, - prompt_convention: PromptConvention::ExcludesCache, - ..included.usage - }, - ..included - }; - let plan = compile(&specification).unwrap(); - assert_eq!(plan.calculate(&included), plan.calculate(&excluded)); - assert_eq!(plan.calculate(&included).unwrap().rates.input, 5.0); -} - -#[test] -fn split_writes_and_invalid_accounting() { - let standard = Rates { - cache_read: Rate::Value(0.5), - cache_write: Rate::Value(3.0), - cache_write_1h: Rate::Value(5.0), - ..rates(Rate::Value(2.0), Rate::Value(4.0)) - }; - let base = request(); - let split = Request { - usage: Usage { - cache_write_5m_tokens: Some(4), - cache_write_1h_tokens: Some(6), - ..base.usage - }, - ..base - }; - let result = calculate(&pricing(standard), &split).unwrap(); - assert_eq!(result.cache_write_5m, 12.0); - assert_eq!(result.cache_write_1h, 30.0); - let overlapping = Request { - usage: Usage { - prompt_tokens: 30, - ..split.usage - }, - ..split - }; - assert_eq!( - calculate(&pricing(standard), &overlapping), - Err(Error::CacheExceedsPrompt) - ); - let incomplete = Request { - usage: Usage { - cache_write_1h_tokens: None, - ..split.usage - }, - ..split - }; - assert_eq!( - calculate(&pricing(standard), &incomplete), - Err(Error::InvalidCacheWriteDetails) - ); -} - -#[test] -fn threshold_tiers_and_boundaries() { - let priority = TierRates { - tier: ServiceTier::Priority, - rates: rates(Rate::Value(3.0), Rate::Missing), - }; - let threshold = ThresholdRates { - above_prompt_tokens: 100, - standard: rates(Rate::Value(5.0), Rate::Value(8.0)), - tiers: &[ - TierRates { - tier: ServiceTier::Priority, - rates: rates(Rate::Value(7.0), Rate::Missing), - }, - TierRates { - tier: ServiceTier::Flex, - rates: rates(Rate::Value(6.0), Rate::Missing), - }, - ], - }; - let specification = Pricing { - standard: rates(Rate::Value(2.0), Rate::Value(4.0)), - tiers: &[priority], - thresholds: &[threshold], - off_peak: None, - }; - let base = request(); - let no_cache = Request { - usage: Usage { - cache_read_tokens: 0, - cache_write_tokens: 0, - ..base.usage - }, - ..base - }; - let fast = Request { - service_tier: ServiceTier::Fast, - ..no_cache - }; - let inclusive = Request { - threshold_policy: ThresholdPolicy::Inclusive, - ..fast - }; - let flex = Request { - service_tier: ServiceTier::Flex, - ..inclusive - }; - assert_eq!(calculate(&specification, &no_cache).unwrap().input(), 200.0); - assert_eq!(calculate(&specification, &fast).unwrap().input(), 300.0); - assert_eq!( - calculate(&specification, &inclusive).unwrap().input(), - 700.0 - ); - assert_eq!(calculate(&specification, &flex).unwrap().input(), 600.0); -} - -#[rstest] -#[case::ultrafast_above_threshold(ServiceTier::Ultrafast, 300_000, 9_301_000.0, 37_000.0)] -#[case::ultrafast_at_threshold(ServiceTier::Ultrafast, 272_000, 544_500.0, 5_000.0)] -#[case::standard_above_threshold(ServiceTier::Standard, 300_000, 3_300_600.0, 13_000.0)] -#[case::priority_above_threshold(ServiceTier::Priority, 300_000, 5_701_000.0, 23_000.0)] -fn tiered_long_context_rates_are_selected_by_service_tier( - #[case] service_tier: ServiceTier, - #[case] prompt_tokens: u64, - #[case] expected_input: f64, - #[case] expected_output: f64, -) { - let standard = Rates { - cache_read: Rate::Value(3.0), - ..rates(Rate::Value(1.0), Rate::Value(2.0)) - }; - let tiers = [ - TierRates { - tier: ServiceTier::Priority, - rates: Rates { - cache_read: Rate::Value(5.0), - ..rates(Rate::Value(3.0), Rate::Value(4.0)) - }, - }, - TierRates { - tier: ServiceTier::Ultrafast, - rates: Rates { - cache_read: Rate::Value(7.0), - ..rates(Rate::Value(2.0), Rate::Value(5.0)) - }, - }, - ]; - let threshold_tiers = [ - TierRates { - tier: ServiceTier::Priority, - rates: Rates { - cache_read: Rate::Value(29.0), - ..rates(Rate::Value(19.0), Rate::Value(23.0)) - }, - }, - TierRates { - tier: ServiceTier::Ultrafast, - rates: Rates { - cache_read: Rate::Value(41.0), - ..rates(Rate::Value(31.0), Rate::Value(37.0)) - }, - }, - ]; - let thresholds = [ThresholdRates { - above_prompt_tokens: 272_000, - standard: Rates { - cache_read: Rate::Value(17.0), - ..rates(Rate::Value(11.0), Rate::Value(13.0)) - }, - tiers: &threshold_tiers, - }]; - let pricing = Pricing { - standard, - tiers: &tiers, - thresholds: &thresholds, - off_peak: None, - }; - let base = request(); - let long_context_request = Request { - usage: Usage { - prompt_tokens, - completion_tokens: 1_000, - cache_read_tokens: 100, - cache_write_tokens: 0, - ..base.usage - }, - service_tier, - ..base - }; - let cost = calculate(&pricing, &long_context_request).unwrap(); - - assert_eq!(cost.input(), expected_input); - assert_eq!(cost.output(), expected_output); -} - -#[test] -fn compile_rejects_ambiguous_rates() { - let duplicate = ThresholdRates { - above_prompt_tokens: 100, - standard: Rates::EMPTY, - tiers: &[], - }; - let specification = Pricing { - standard: rates(Rate::Value(1.0), Rate::Value(1.0)), - tiers: &[], - thresholds: &[duplicate, duplicate], - off_peak: None, - }; - assert_eq!( - compile(&specification).err(), - Some(Error::DuplicateThreshold) - ); - let invalid = pricing(rates(Rate::Value(f64::NAN), Rate::Value(1.0))); - assert_eq!(compile(&invalid).err(), Some(Error::InvalidRate)); -} - -#[test] -fn off_peak_is_one_non_wrapping_utc_window() { - let specification = Pricing { - standard: rates(Rate::Value(2.0), Rate::Value(4.0)), - tiers: &[], - thresholds: &[], - off_peak: Some(OffPeakRates { - start_utc_minute: 60, - end_utc_minute: 120, - rates: rates(Rate::Value(1.0), Rate::Value(2.0)), - }), - }; - let base = request(); - let start = Request { - billed_at_utc_minute: Some(60), - ..base - }; - let end = Request { - billed_at_utc_minute: Some(120), - ..base - }; - assert_eq!( - calculate(&specification, &base), - Err(Error::InvalidBillingTime) - ); - assert_eq!(calculate(&specification, &start).unwrap().input(), 100.0); - assert_eq!(calculate(&specification, &end).unwrap().input(), 200.0); -} - -#[test] -fn missing_rates_and_free_rates_remain_distinct() { - let base = request(); - let empty = Request { - usage: Usage { - prompt_tokens: 0, - completion_tokens: 0, - cache_read_tokens: 0, - cache_write_tokens: 0, - ..base.usage - }, - ..base - }; - assert_eq!( - calculate(&pricing(Rates::EMPTY), &empty), - Err(Error::MissingInputRate) - ); - assert_eq!( - calculate(&pricing(rates(Rate::Value(0.0), Rate::Missing)), &empty), - Err(Error::MissingOutputRate) - ); - assert_eq!( - calculate(&pricing(rates(Rate::Value(0.0), Rate::Value(0.0))), &empty) - .unwrap() - .total(), - 0.0 - ); -} - -#[test] -fn matches_executed_python_reference_cases() { - for row in include_str!("python_reference.tsv") - .lines() - .filter(|line| !line.starts_with('#')) - { - let fields: Vec<_> = row.split('\t').collect(); - let count = |index: usize| fields[index].parse::().unwrap(); - let number = |index: usize| fields[index].parse::().unwrap(); - let optional_rate = |index: usize| { - if fields[index].is_empty() { - Rate::Missing - } else { - Rate::Value(number(index)) - } - }; - let threshold = ThresholdRates { - above_prompt_tokens: if fields[9].is_empty() { 0 } else { count(9) }, - standard: rates(optional_rate(10), optional_rate(11)), - tiers: &[], - }; - let thresholds = if fields[9].is_empty() { - &[][..] - } else { - std::slice::from_ref(&threshold) - }; - let specification = Pricing { - standard: Rates { - cache_read: optional_rate(7), - cache_write: optional_rate(8), - ..rates(Rate::Value(number(5)), Rate::Value(number(6))) - }, - tiers: &[], - thresholds, - off_peak: None, - }; - let base = request(); - let input = Request { - usage: Usage { - prompt_tokens: count(1), - completion_tokens: count(2), - cache_read_tokens: count(3), - cache_write_tokens: count(4), - ..base.usage - }, - ..base - }; - let actual = calculate(&specification, &input).unwrap(); - assert_eq!(actual.input(), number(12), "{}", fields[0]); - assert_eq!(actual.output(), number(13), "{}", fields[0]); - } -} - -proptest::proptest! { - #[test] - fn equivalent_usage_conventions_and_breakdown_agree( - regular in 0_u64..1000, - read in 0_u64..1000, - write in 0_u64..1000, - output in 0_u64..1000, - ) { - let threshold = ThresholdRates { - above_prompt_tokens: 1000, - standard: rates(Rate::Value(5.0), Rate::Value(8.0)), - tiers: &[], - }; - let specification = Pricing { - standard: rates(Rate::Value(2.0), Rate::Value(4.0)), - tiers: &[], - thresholds: &[threshold], - off_peak: None, - }; - let base = request(); - let included = Request { - usage: Usage { - prompt_tokens: regular + read + write, - completion_tokens: output, - cache_read_tokens: read, - cache_write_tokens: write, - ..base.usage - }, - ..base - }; - let excluded = Request { - usage: Usage { - prompt_tokens: regular, - prompt_convention: PromptConvention::ExcludesCache, - ..included.usage - }, - ..included - }; - let plan = compile(&specification).unwrap(); - let left = plan.calculate(&included).unwrap(); - let right = plan.calculate(&excluded).unwrap(); - proptest::prop_assert_eq!(left, right); - proptest::prop_assert_eq!(left.total(), left.input() + left.output()); - } -} - -#[test] -fn regional_multiplier_applies_after_input_components_are_summed() { - let standard = Rates { - cache_read: Rate::Value(0.5), - cache_write: Rate::Value(3.0), - ..rates(Rate::Value(2.0), Rate::Value(4.0)) - }; - let base = request(); - let regional = Request { - region_multiplier: Some(1.1), - ..base - }; - let result = calculate(&pricing(standard), ®ional).unwrap(); - assert_eq!(result.input(), (65.0 * 2.0 + 25.0 * 0.5 + 10.0 * 3.0) * 1.1); - assert_eq!(result.output(), 20.0 * 4.0 * 1.1); - let invalid = Request { - region_multiplier: Some(f64::NAN), - ..base - }; - assert_eq!( - calculate(&pricing(standard), &invalid), - Err(Error::InvalidRegionMultiplier) - ); -} - -#[test] -fn compilation_sorts_thresholds_and_rejects_duplicate_tiers() { - let high = ThresholdRates { - above_prompt_tokens: 200, - standard: rates(Rate::Value(7.0), Rate::Missing), - tiers: &[], - }; - let low = ThresholdRates { - above_prompt_tokens: 100, - standard: rates(Rate::Value(5.0), Rate::Missing), - tiers: &[], - }; - let specification = Pricing { - standard: rates(Rate::Value(2.0), Rate::Value(4.0)), - tiers: &[], - thresholds: &[high, low], - off_peak: None, - }; - let base = request(); - let above_both = Request { - usage: Usage { - prompt_tokens: 201, - cache_read_tokens: 0, - cache_write_tokens: 0, - ..base.usage - }, - ..base - }; - assert_eq!( - compile(&specification) - .unwrap() - .calculate(&above_both) - .unwrap() - .rates - .input, - 7.0 - ); - let duplicate = TierRates { - tier: ServiceTier::Flex, - rates: Rates::EMPTY, - }; - let invalid = Pricing { - tiers: &[duplicate, duplicate], - thresholds: &[], - ..specification - }; - assert_eq!(compile(&invalid).err(), Some(Error::DuplicateTier)); -} diff --git a/litellm-rust/crates/cost/tests/generate_python_reference.py b/litellm-rust/crates/cost/tests/generate_python_reference.py deleted file mode 100644 index 4a179555650..00000000000 --- a/litellm-rust/crates/cost/tests/generate_python_reference.py +++ /dev/null @@ -1,96 +0,0 @@ -import subprocess -from dataclasses import dataclass -from pathlib import Path - -from litellm.litellm_core_utils.llm_cost_calc.utils import generic_cost_per_token -from litellm.types.utils import Usage - - -@dataclass(frozen=True, slots=True) -class Case: - name: str - prompt: int - completion: int - cache_read: int - cache_write: int - input_rate: float - output_rate: float - cache_read_rate: float | None = None - cache_write_rate: float | None = None - threshold: int | None = None - threshold_input_rate: float | None = None - threshold_output_rate: float | None = None - - -CASES = ( - Case("ordinary", 100, 20, 0, 0, 2.0, 4.0), - Case("cache_fallback", 100, 20, 25, 10, 2.0, 4.0), - Case("cache_specific", 100, 20, 25, 10, 2.0, 4.0, 0.5, 3.0), - Case("free_cache", 100, 20, 25, 10, 2.0, 4.0, 0.0, 0.0), - Case("threshold_below", 99, 20, 0, 0, 2.0, 4.0, threshold=100, threshold_input_rate=5.0, threshold_output_rate=8.0), - Case("threshold_at", 100, 20, 0, 0, 2.0, 4.0, threshold=100, threshold_input_rate=5.0, threshold_output_rate=8.0), - Case( - "threshold_above", 101, 20, 0, 0, 2.0, 4.0, threshold=100, threshold_input_rate=5.0, threshold_output_rate=8.0 - ), - Case( - "cache_threshold_above", - 101, - 20, - 25, - 10, - 2.0, - 4.0, - threshold=100, - threshold_input_rate=5.0, - threshold_output_rate=8.0, - ), -) - - -def reference(case: Case) -> tuple[float, float]: - info = {"input_cost_per_token": case.input_rate, "output_cost_per_token": case.output_rate} - if case.cache_read_rate is not None: - info["cache_read_input_token_cost"] = case.cache_read_rate - if case.cache_write_rate is not None: - info["cache_creation_input_token_cost"] = case.cache_write_rate - if case.threshold is not None: - info[f"input_cost_per_token_above_{case.threshold}_tokens"] = case.threshold_input_rate - info[f"output_cost_per_token_above_{case.threshold}_tokens"] = case.threshold_output_rate - details = {"cached_tokens": case.cache_read, "cache_write_tokens": case.cache_write} - usage = Usage(prompt_tokens=case.prompt, completion_tokens=case.completion, prompt_tokens_details=details) - return generic_cost_per_token( - model="synthetic", - usage=usage, - custom_llm_provider="openai", - model_info=info, - ) - - -def main() -> None: - revision = subprocess.check_output(("git", "rev-parse", "HEAD"), text=True).strip() - rows = ("# Python reference commit: " + revision,) + tuple( - "\t".join( - str(value) if value is not None else "" - for value in ( - case.name, - case.prompt, - case.completion, - case.cache_read, - case.cache_write, - case.input_rate, - case.output_rate, - case.cache_read_rate, - case.cache_write_rate, - case.threshold, - case.threshold_input_rate, - case.threshold_output_rate, - *reference(case), - ) - ) - for case in CASES - ) - Path(__file__).with_name("python_reference.tsv").write_text("\n".join(rows) + "\n") - - -if __name__ == "__main__": - main() diff --git a/litellm-rust/crates/cost/tests/python_reference.tsv b/litellm-rust/crates/cost/tests/python_reference.tsv deleted file mode 100644 index 5abb1329a61..00000000000 --- a/litellm-rust/crates/cost/tests/python_reference.tsv +++ /dev/null @@ -1,9 +0,0 @@ -# Python reference commit: dc4be2fd987c993aefcf16444e34f960c12c8627 -ordinary 100 20 0 0 2.0 4.0 200.0 80.0 -cache_fallback 100 20 25 10 2.0 4.0 200.0 80.0 -cache_specific 100 20 25 10 2.0 4.0 0.5 3.0 172.5 80.0 -free_cache 100 20 25 10 2.0 4.0 0.0 0.0 130.0 80.0 -threshold_below 99 20 0 0 2.0 4.0 100 5.0 8.0 198.0 80.0 -threshold_at 100 20 0 0 2.0 4.0 100 5.0 8.0 200.0 80.0 -threshold_above 101 20 0 0 2.0 4.0 100 5.0 8.0 505.0 160.0 -cache_threshold_above 101 20 25 10 2.0 4.0 100 5.0 8.0 505.0 160.0 diff --git a/litellm-rust/crates/model-catalog-refresh/Cargo.toml b/litellm-rust/crates/model-catalog-refresh/Cargo.toml new file mode 100644 index 00000000000..27219042c6c --- /dev/null +++ b/litellm-rust/crates/model-catalog-refresh/Cargo.toml @@ -0,0 +1,19 @@ +[package] +name = "litellm-model-catalog-refresh" +version = "0.1.0" +edition.workspace = true +license.workspace = true +repository.workspace = true + +[dependencies] +litellm-http.workspace = true +litellm-model-catalog.workspace = true +thiserror.workspace = true +url.workspace = true + +[dev-dependencies] +litellm-error.workspace = true +litellm-http = { workspace = true, features = ["test-support"] } +reqwest.workspace = true +rstest.workspace = true +tokio.workspace = true diff --git a/litellm-rust/crates/model-catalog-refresh/src/error.rs b/litellm-rust/crates/model-catalog-refresh/src/error.rs new file mode 100644 index 00000000000..656f32370cb --- /dev/null +++ b/litellm-rust/crates/model-catalog-refresh/src/error.rs @@ -0,0 +1,105 @@ +use std::io; + +use thiserror::Error; + +#[derive(Debug, Error)] +#[error(transparent)] +pub struct Error(Box); + +#[derive(Debug, Error)] +enum Repr { + #[error(transparent)] + Http(#[from] litellm_http::Error), + #[error("could not read the catalog")] + Read(#[source] io::Error), + #[error("invalid catalog URL")] + InvalidUrl(#[source] url::ParseError), + #[error("unsupported catalog URL scheme {scheme}")] + UnsupportedScheme { scheme: String }, +} + +impl Error { + pub fn unsupported_scheme(scheme: String) -> Self { + Self(Box::new(Repr::UnsupportedScheme { scheme })) + } +} + +impl From for Error { + fn from(cause: litellm_http::Error) -> Self { + Self(Box::new(Repr::Http(cause))) + } +} + +impl From for Error { + fn from(cause: io::Error) -> Self { + Self(Box::new(Repr::Read(cause))) + } +} + +impl From for Error { + fn from(cause: url::ParseError) -> Self { + Self(Box::new(Repr::InvalidUrl(cause))) + } +} + +#[cfg(test)] +mod tests { + use litellm_error::{Report, find}; + use rstest::rstest; + + use super::*; + use crate::{DoneReason, Reply, Step}; + + #[rstest] + fn exhausted_transport_keeps_its_typed_cause() { + let reply = Reply::Transport { + error: io::Error::from(io::ErrorKind::ConnectionReset).into(), + }; + let Reply::Transport { error } = reply else { + panic!("transport failure") + }; + let step = Step::Done { + reason: DoneReason::Exhausted(error), + }; + let Step::Done { + reason: DoneReason::Exhausted(error), + } = step + else { + panic!("exhausted attempts") + }; + let cause = find::(&error).expect("typed I/O cause"); + assert_eq!(cause.kind(), io::ErrorKind::ConnectionReset); + assert_eq!( + Report::new(&error) + .to_string() + .matches(&cause.to_string()) + .count(), + 1 + ); + assert!(!error.to_string().contains(&cause.to_string())); + } + + #[rstest] + fn invalid_source_keeps_its_typed_cause() { + let error = Error::from(url::Url::parse("http://[invalid").unwrap_err()); + assert!(find::(&error).is_some()); + } + + #[rstest] + #[tokio::test] + async fn http_failures_keep_the_cause_without_the_url() { + let cause = litellm_http::Client::for_test() + .get("http://localhost:invalid/private?api_key=secret") + .send() + .await + .unwrap_err(); + let error = Error::from(litellm_http::Error::from(cause)); + assert!(find::(&error).is_some()); + let report = Report::new(&error).to_string(); + assert!(!report.contains("secret"), "{report}"); + assert!(!report.contains("private"), "{report}"); + let debug = format!("{error:?}"); + assert!(!debug.contains("secret"), "{debug}"); + assert!(!debug.contains("private"), "{debug}"); + } +} diff --git a/litellm-rust/crates/model-catalog-refresh/src/lib.rs b/litellm-rust/crates/model-catalog-refresh/src/lib.rs new file mode 100644 index 00000000000..006c4887fad --- /dev/null +++ b/litellm-rust/crates/model-catalog-refresh/src/lib.rs @@ -0,0 +1,9 @@ +mod error; +mod policy; +mod source; +mod step; + +pub use error::Error; +pub use policy::*; +pub use source::*; +pub use step::*; diff --git a/litellm-rust/crates/model-catalog-refresh/src/policy.rs b/litellm-rust/crates/model-catalog-refresh/src/policy.rs new file mode 100644 index 00000000000..bc32e40408e --- /dev/null +++ b/litellm-rust/crates/model-catalog-refresh/src/policy.rs @@ -0,0 +1,20 @@ +use std::time::Duration; + +use litellm_model_catalog::IntegrityLimits; + +/// Main's defaults, 3 attempts, 30 s max wait, 5 s timeout, backoff base 2.0, jitter 0.0 to 1.0 s +#[derive(Clone, Copy, Debug, PartialEq)] +pub struct RetryPolicy { + pub max_attempts: u32, + pub max_wait: Duration, + pub timeout: Duration, + pub backoff_base_seconds: f64, + pub jitter_min_seconds: f64, + pub jitter_max_seconds: f64, + /// Main's 50 model floor and 0.5 shrink ratio + pub limits: IntegrityLimits, +} + +/// Main's default `LITELLM_MODEL_COST_MAP_URL` +pub const DEFAULT_REGISTRY_URL: &str = + "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json"; diff --git a/litellm-rust/crates/model-catalog-refresh/src/source.rs b/litellm-rust/crates/model-catalog-refresh/src/source.rs new file mode 100644 index 00000000000..713e0386941 --- /dev/null +++ b/litellm-rust/crates/model-catalog-refresh/src/source.rs @@ -0,0 +1,17 @@ +use std::path::PathBuf; + +use url::Url; + +/// Where a fetched catalog comes from +#[derive(Clone, Debug, PartialEq, Eq)] +pub enum Source { + Http(Url), + File(PathBuf), + S3(S3Object), +} + +#[derive(Clone, Debug, PartialEq, Eq)] +pub struct S3Object { + pub bucket: String, + pub key: String, +} diff --git a/litellm-rust/crates/model-catalog-refresh/src/step.rs b/litellm-rust/crates/model-catalog-refresh/src/step.rs new file mode 100644 index 00000000000..822d1807916 --- /dev/null +++ b/litellm-rust/crates/model-catalog-refresh/src/step.rs @@ -0,0 +1,66 @@ +use std::time::Duration; + +use litellm_model_catalog::Catalog; + +use crate::Error; +use crate::source::Source; + +/// One attempt. Where to fetch from is the `Source`, so the request carries no URL +#[derive(Clone, Debug, PartialEq, Eq)] +pub struct FetchRequest { + pub if_none_match: Option, + pub timeout: Duration, +} + +#[derive(Debug)] +pub enum Reply { + /// An HTTP, S3 or file answer. The body is read only on 200 + Response { + status: u16, + etag: Option, + retry_after: Option, + body: Option>, + }, + /// No answer reached us + Transport { error: Error }, +} + +/// What the fetch state machine asks its driver to do next +#[derive(Debug)] +pub enum Step { + Fetch(FetchRequest), + Sleep(Duration), + /// A parsed document to install, with its rejected row errors + Adopt { + catalog: Box, + installed: Installed, + errors: Vec, + }, + /// 304, a final status, or retries exhausted + Done { + reason: DoneReason, + }, +} + +#[derive(Debug)] +pub enum DoneReason { + NotModified, + Status(u16), + Exhausted(Error), +} + +/// What is installed now, the input to the next revalidation +#[derive(Clone, Debug, PartialEq, Eq)] +pub struct Installed { + pub origin: Origin, + pub etag: Option, + /// `git_blob_id` of the installed bytes + pub revision: String, + pub model_count: usize, +} + +#[derive(Clone, Debug, PartialEq, Eq)] +pub enum Origin { + Bundled, + Remote { source: Source }, +} diff --git a/litellm-rust/crates/model-catalog/AGENTS.md b/litellm-rust/crates/model-catalog/AGENTS.md index 0a380788c19..fc4b28d4ef0 100644 --- a/litellm-rust/crates/model-catalog/AGENTS.md +++ b/litellm-rust/crates/model-catalog/AGENTS.md @@ -1,5 +1,27 @@ -## Validation +## Layout -For `model_prices_and_context_window.json` validation, we should eventually: +`wire/` reads one entry as written. `EntryWire::read` makes one pass over the entry's keys and puts every non-null key in exactly one place, which is an attribute, a flat price parsed by `FlatKey::parse`, a price block, a `supports_*` capability or `unknown`. A known key with a bad value becomes an `Error` that names the key, and the rest of the entry is still read -- Stop skipping this crate's tests +`lower.rs` is the only code that knows both the wire types and `litellm-cost`. It lowers an `EntryWire` into `Prices`, `Adjustments` and `Model` and checks the rules that span keys, such as one price per meter and condition, contiguous tiers, no suffix inside a block, and off-peak needing hours and prices. Lowering only goes one way. Nothing turns `Prices` back into JSON, since the raw fields are kept + +`catalog.rs` reads the document, indexes aliases and runs the checks across entries. A bad entry keeps its raw fields and its errors are listed in `Loaded.errors`. Deciding what to do with them is the caller's job. `Catalog::written` gives back every top-level key and value as the document wrote them, in document order. `Catalog::served` gives back every model and alias as Python serves it, without `sample_spec` and `fallback_generalizations` + +`Billing` on `Prices` holds the row attributes pricing reads (text unit, band edge, cache-read fallback, wall clock, web search unit, tier scope). `lower.rs` derives it from the row and its `litellm_provider`, so `litellm-cost` holds no provider list + +`store/` is the catalog's in-memory store and the only place that indexes model names. Its rules are in [src/store/AGENTS.md](src/store/AGENTS.md) + +`layers.rs`, `resolve/`, `model_info.rs` and `register.rs` hold types only. The functions over them land with their first caller + +Catalog errors use one opaque `Error` with typed causes and JSON Pointer accessors (`path`, `key`, `field`). `Error::rejects` alone decides whether an error fails the document, rejects its row or neither, and `ErrorKind` is the stable name hosts report it by + +## Prices + +The price vocabulary lives in `litellm-cost`, and only this crate knows JSON key names. Every price is keyed by `Meter`, `Channel` and `When`. A variation only one meter has stays in the meter (cache TTL, image size, resolution, `_above_Ns_interval`, `max_results_range`). Service tier and prompt band make up `When`. Batch and decisions are a `Channel` because Python bills them with their own rules. Off-peak is an overlay on `Prices`, and regional uplifts are `Adjustments`, never prices + +A new `N`, image size or resolution in an existing key shape parses with no Rust change. A new unit, token kind, feature or tier needs a variant in `litellm-cost` plus its spelling in `wire/flat_key.rs` and `FLAT_KEY_PATTERN` in `schema.rs` + +`Prices::band` and `Prices::rate` are pinned to Python's `_get_token_base_cost` by `tests/price_lookup_reference.json`. After a billing change in Python, regenerate it from the repo root with `PYTHONPATH=. python litellm-rust/crates/model-catalog/tests/generate_price_lookup_reference.py` + +## Tests + +`BUNDLED_REGISTRY` embeds the root `model_prices_and_context_window.json`, so the tests over it fail a JSON change with a bad value or a key no type reads diff --git a/litellm-rust/crates/model-catalog/Cargo.toml b/litellm-rust/crates/model-catalog/Cargo.toml index 2c2866d7090..45d16515dcd 100644 --- a/litellm-rust/crates/model-catalog/Cargo.toml +++ b/litellm-rust/crates/model-catalog/Cargo.toml @@ -6,15 +6,19 @@ license.workspace = true repository.workspace = true [features] -schema = ["dep:schemars", "litellm-llms-types/schema"] +schema = ["dep:schemars", "litellm-cost/schema"] [dependencies] -litellm-llms-types.workspace = true +litellm-cost = { workspace = true, features = ["serde"] } indexmap = { version = "2.14.0", features = ["serde"] } +jiff = { version = "0.2.37", default-features = false, features = ["std", "tzdb-bundle-always"] } +macro_rules_attribute.workspace = true schemars = { workspace = true, optional = true } serde.workspace = true -serde_json.workspace = true +serde_json = { workspace = true, features = ["preserve_order"] } +serde_with.workspace = true +strum.workspace = true thiserror.workspace = true time.workspace = true diff --git a/litellm-rust/crates/model-catalog/src/bundled.rs b/litellm-rust/crates/model-catalog/src/bundled.rs new file mode 100644 index 00000000000..d71a2d57ebb --- /dev/null +++ b/litellm-rust/crates/model-catalog/src/bundled.rs @@ -0,0 +1,3 @@ +/// The registry document this build serves when no other one is loaded. +pub const BUNDLED_REGISTRY: &[u8] = + include_bytes!("../../../../model_prices_and_context_window.json"); diff --git a/litellm-rust/crates/model-catalog/src/capabilities.rs b/litellm-rust/crates/model-catalog/src/capabilities.rs deleted file mode 100644 index 30724766aeb..00000000000 --- a/litellm-rust/crates/model-catalog/src/capabilities.rs +++ /dev/null @@ -1,67 +0,0 @@ -use serde::{Deserialize, Serialize}; - -/// Primary API surface / task type of the model. -#[derive(Clone, Copy, Debug, Deserialize, Eq, PartialEq, Serialize)] -#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))] -#[serde(rename_all = "snake_case")] -pub enum Mode { - AudioSpeech, - AudioTranscription, - Chat, - Completion, - Decisions, - Embedding, - Evaluation, - Guardrail, - ImageEdit, - ImageGeneration, - Moderation, - Ocr, - Realtime, - Rerank, - Responses, - Search, - VectorStore, - VideoGeneration, -} - -/// Gemini audio generation API the model is served through. -#[derive(Clone, Copy, Debug, Deserialize, Eq, PartialEq, Serialize)] -#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))] -#[serde(rename_all = "snake_case")] -pub enum VertexAiAudioApi { - LyriaPredict, - LyriaInteractions, -} - -/// Audio container format the model can return. -#[derive(Clone, Copy, Debug, Deserialize, Eq, PartialEq, Serialize)] -#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))] -#[serde(rename_all = "snake_case")] -pub enum AudioFormat { - Mp3, - Wav, -} - -/// Input modality the model accepts. -#[derive(Clone, Copy, Debug, Deserialize, Eq, PartialEq, Serialize)] -#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))] -#[serde(rename_all = "snake_case")] -pub enum InputModality { - Text, - Image, - Audio, - Video, -} - -/// Output modality the model can produce. -#[derive(Clone, Copy, Debug, Deserialize, Eq, PartialEq, Serialize)] -#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))] -#[serde(rename_all = "snake_case")] -pub enum OutputModality { - Text, - Image, - Audio, - Video, - Code, -} diff --git a/litellm-rust/crates/model-catalog/src/catalog.rs b/litellm-rust/crates/model-catalog/src/catalog.rs index 37412ef1e14..ac5f4ffcf25 100644 --- a/litellm-rust/crates/model-catalog/src/catalog.rs +++ b/litellm-rust/crates/model-catalog/src/catalog.rs @@ -1,54 +1,29 @@ -use crate::error::Error; -use crate::fallback::{FallbackGeneralizations, FallbackRule}; -use crate::model_info::ModelInfo; +use std::borrow::Cow; +use std::collections::HashSet; + use indexmap::IndexMap; use serde::Deserialize; -use serde_json::{Map, Value}; -use std::collections::HashMap; +use serde_json::Value; -#[derive(Clone, Debug, Default, PartialEq, Eq)] -pub struct Provenance { - pub source: Option, - pub revision: Option, - pub etag: Option, -} +use crate::error::{Error, Loaded, Rejects, Repr}; +use crate::fallback::{FallbackGeneralizations, FallbackRule}; +use crate::model::Model; +use crate::store::{Rows, Store}; +use crate::wire::{EntryWire, Fields}; + +const SAMPLE_SPEC: &str = "sample_spec"; +const FALLBACK_GENERALIZATIONS: &str = "fallback_generalizations"; #[derive(Clone, Copy, Debug, PartialEq)] pub struct IntegrityLimits { - pub reference_model_count: usize, - pub min_model_count: usize, - pub min_reference_ratio: f64, -} - -#[derive(Clone, Debug, PartialEq, Eq, thiserror::Error)] -pub enum AliasIssue { - #[error("model {model:?} has aliases that are not an array")] - InvalidList { model: String }, - #[error("model {model:?} has an alias that is not a string")] - InvalidName { model: String }, - #[error("alias {alias:?} of model {model:?} is already a model name")] - CanonicalCollision { model: String, alias: String }, - #[error("alias {alias:?} of model {model:?} is already an alias of another model")] - AliasCollision { model: String, alias: String }, + min_model_count: usize, + min_reference_ratio: f64, } #[derive(Clone, Debug)] pub struct ModelEntry { - fields: Map, - info: ModelInfo, -} - -impl ModelEntry { - pub fn field(&self, name: &str) -> Option<&Value> { - self.fields.get(name) - } - pub fn fields(&self) -> &Map { - &self.fields - } - /// The entry deserialized into the typed mirror of the catalog schema. - pub fn info(&self) -> &ModelInfo { - &self.info - } + fields: Fields, + model: Option, } #[derive(Clone, Copy, Debug)] @@ -58,142 +33,312 @@ pub struct ModelMatch<'a> { pub entry: &'a ModelEntry, } -#[derive(Debug)] +#[derive(Clone, Debug)] +struct Section { + index: usize, + key: String, + value: Value, +} + +/// One fetched key kept off the served map because its row failed validation +#[derive(Clone, Debug)] +pub struct Rejection<'a> { + pub key: String, + pub errors: Vec<&'a Error>, + /// The bundled row for the key is served instead. `false` means the key is left out + pub replaced_by_bundled: bool, +} + +#[derive(Clone, Copy, Debug, serde::Serialize)] +#[serde(untagged)] +pub enum Written<'a> { + Entry(&'a Fields), + Section(&'a Value), +} + +#[derive(Clone, Debug)] pub struct Catalog { + store: Store, + sections: Vec
, + fallback_generalizations: Option, +} + +/// Models and aliases, an alias served as its owner's row +#[derive(Clone, Debug)] +struct CatalogRows { entries: IndexMap, aliases: IndexMap, - lowercase_keys: HashMap, - sample_spec: Option, - fallback_generalizations: Option, - provenance: Provenance, - alias_issues: Vec, +} + +impl CatalogRows { + fn owner<'a>(&'a self, key: &'a str) -> &'a str { + self.aliases.get(key).map_or(key, String::as_str) + } + + fn fields(&self, key: &str) -> Option<&Fields> { + Some(self.entries.get(self.owner(key))?.fields()) + } +} + +/// Model names match ignoring case, and keys are grouped by their raw `litellm_provider` +impl Rows for CatalogRows { + type Row = Fields; + type Stamp = (); + + fn fold(name: &str) -> String { + name.to_lowercase() + } + + fn stamp(&self) -> Self::Stamp {} + + fn contains(&self, key: &str) -> bool { + self.entries.contains_key(key) || self.aliases.contains_key(key) + } + + fn row(&self, key: &str) -> Option> { + self.fields(key).map(Cow::Borrowed) + } + + fn scan(&self, visit: &mut dyn FnMut(&str, Option<&str>)) { + let keys = self.entries.keys().chain(self.aliases.keys()); + for key in keys { + let provider = self + .fields(key) + .and_then(|fields| fields.get("litellm_provider")?.as_str()); + visit(key, provider); + } + } +} + +impl IntegrityLimits { + pub fn new(min_model_count: usize, min_reference_ratio: f64) -> Option { + (0.0..=1.0).contains(&min_reference_ratio).then_some(Self { + min_model_count, + min_reference_ratio, + }) + } +} + +impl ModelEntry { + pub fn field(&self, name: &str) -> Option<&Value> { + self.fields.get(name) + } + + pub fn fields(&self) -> &Fields { + &self.fields + } + + pub fn model(&self) -> Option<&Model> { + self.model.as_ref() + } } impl Catalog { - pub fn parse(body: &[u8], provenance: Provenance) -> Result { - let root: IndexMap = serde_json::from_slice(body)?; - if root.is_empty() { - return Err(Error::Empty); - } + pub fn parse(body: &[u8]) -> Result, Error> { + let document = serde_json::from_slice::(body) + .map_err(|error| Error::new("", Repr::Json(error)))?; - let mut entries = IndexMap::with_capacity(root.len()); + let mut problems = Vec::new(); + let mut sections = Vec::new(); + let mut entries = IndexMap::with_capacity(document.len()); let mut alias_lists = Vec::new(); - let mut alias_issues = Vec::new(); - let mut sample_spec = None; - let mut fallback_generalizations = None; - for (name, value) in root { - match name.as_str() { - "sample_spec" => { - sample_spec = Some(value); + for (index, (name, value)) in document.into_iter().enumerate() { + let fields = match value { + Value::Object(fields) if !is_section(&name) => fields, + value => { + if !is_section(&name) { + problems.push(Error::from_key(&name, Repr::NotAnObject)); + } + sections.push(Section { + index, + key: name, + value, + }); continue; } - "fallback_generalizations" => { - fallback_generalizations = - Some(serde_json::from_value::(value)?); - continue; - } - _ => {} - } - let Value::Object(mut fields) = value else { - return Err(Error::EntryNotObject { model: name }); }; - let info = ModelInfo::deserialize(&fields)?; - if let Some(aliases) = fields.remove("aliases") { - match aliases { - Value::Array(names) => alias_lists.push((name.clone(), names)), - _ => alias_issues.push(AliasIssue::InvalidList { - model: name.clone(), - }), - } + let (model, aliases, row_problems) = read_row(&fields); + problems.extend( + row_problems + .into_iter() + .map(|problem| problem.prepend(&name)), + ); + if let Some(aliases) = aliases { + alias_lists.push((name.clone(), aliases)); } - entries.insert(name, ModelEntry { fields, info }); + entries.insert(name, ModelEntry { fields, model }); + } + if entries.is_empty() { + return Err(Error::new("", Repr::Empty)); } - let mut aliases = IndexMap::new(); + let mut aliases: IndexMap = IndexMap::new(); for (model, names) in alias_lists { - for name in names { - let Value::String(alias) = name else { - alias_issues.push(AliasIssue::InvalidName { - model: model.clone(), - }); - continue; - }; + for alias in names { if entries.contains_key(&alias) { - alias_issues.push(AliasIssue::CanonicalCollision { - model: model.clone(), - alias, - }); - } else if aliases.contains_key(&alias) { - alias_issues.push(AliasIssue::AliasCollision { - model: model.clone(), - alias, - }); + problems.push(Error::from_key(&model, Repr::AliasShadowsModel { alias })); + } else if let Some(taken_by) = aliases.get(&alias) { + problems.push(Error::from_key( + &model, + Repr::AliasTaken { + alias, + taken_by: taken_by.clone(), + }, + )); } else { aliases.insert(alias, model.clone()); } } } - let lowercase_keys = entries - .keys() - .chain(aliases.keys()) - .map(|key| (key.to_lowercase(), key.clone())) - .collect(); - Ok(Self { - entries, - aliases, - lowercase_keys, - sample_spec, - fallback_generalizations, - provenance, - alias_issues, + let fallback_generalizations = sections + .iter() + .find(|section| section.key == FALLBACK_GENERALIZATIONS) + .and_then(|section| { + FallbackGeneralizations::deserialize(§ion.value) + .map_err(|error| { + problems.push(Error::from_key(§ion.key, Repr::Invalid(error))) + }) + .ok() + }); + Ok(Loaded { + value: Self { + store: Store::new(CatalogRows { entries, aliases }), + sections, + fallback_generalizations, + }, + errors: problems, }) } - pub fn validate(&self, limits: IntegrityLimits) -> Result<(), Error> { - if !limits.min_reference_ratio.is_finite() - || !(0.0..=1.0).contains(&limits.min_reference_ratio) + pub fn row_errors(model: &str, fields: &Fields) -> Vec { + let (_, _, errors) = read_row(fields); + errors + .into_iter() + .map(|error| error.prepend(model)) + .collect() + } + + pub fn written(&self) -> impl Iterator)> { + let mut sections = self.sections.iter().peekable(); + let mut entries = self.rows().entries.iter(); + (0..self.rows().entries.len() + self.sections.len()).map_while(move |index| match sections + .next_if(|section| section.index == index) { - return Err(Error::InvalidRatio); - } - let actual = self.entries.len(); + Some(section) => Some((section.key.as_str(), Written::Section(§ion.value))), + None => entries + .next() + .map(|(name, entry)| (name.as_str(), Written::Entry(&entry.fields))), + }) + } + + pub fn served(&self) -> impl Iterator { + let rows = self.rows(); + rows.entries + .iter() + .map(|(name, entry)| (name.as_str(), entry.fields())) + .chain( + rows.aliases + .iter() + .map(|(alias, owner)| (alias.as_str(), rows.entries[owner].fields())), + ) + } + + pub fn check_size(&self, reference: usize, limits: IntegrityLimits) -> Result<(), Error> { + let actual = self.model_count(); if actual < limits.min_model_count { - return Err(Error::BelowMinimum { - actual, - minimum: limits.min_model_count, - }); + return Err(Error::new( + "", + Repr::BelowMinimum { + actual, + minimum: limits.min_model_count, + }, + )); } - if limits.reference_model_count > 0 - && (actual as f64) < (limits.reference_model_count as f64) * limits.min_reference_ratio - { - return Err(Error::Shrunk { - actual, - reference: limits.reference_model_count, - ratio: limits.min_reference_ratio, - }); + if (actual as f64) < (reference as f64) * limits.min_reference_ratio { + return Err(Error::new( + "", + Repr::Shrunk { + actual, + reference, + ratio: limits.min_reference_ratio, + }, + )); } Ok(()) } - pub fn lookup(&self, key: &str) -> Option> { - let matched_key = if self.entries.contains_key(key) || self.aliases.contains_key(key) { - key - } else { - self.lowercase_keys.get(&key.to_lowercase())?.as_str() - }; - let canonical_key = self - .aliases - .get(matched_key) - .map(String::as_str) - .unwrap_or(matched_key); - let (canonical_key, entry) = self.entries.get_key_value(canonical_key)?; - let matched_key = self + pub fn without_rejected(self, errors: &[Error]) -> (Self, Vec<&Error>) { + let row_names: HashSet = self + .rows() .entries - .get_key_value(matched_key) + .keys() + .cloned() + .chain( + self.sections + .iter() + .filter(|section| !is_section(§ion.key)) + .map(|section| section.key.clone()), + ) + .collect(); + let rejected: HashSet = errors + .iter() + .filter(|error| error.rejects() == Rejects::Row) + .filter_map(|error| crate::error::pointer_root(error.path())) + .filter(|root| row_names.contains(root)) + .collect(); + let removed_errors = errors + .iter() + .filter(|error| { + crate::error::pointer_root(error.path()) + .is_some_and(|root| rejected.contains(&root)) + }) + .collect(); + let removed_positions: HashSet = self + .written() + .enumerate() + .filter(|(_, (key, _))| rejected.contains(*key)) + .map(|(index, _)| index) + .collect(); + let mut catalog = self; + catalog.store.edit(|rows| { + rows.entries.retain(|name, _| !rejected.contains(name)); + rows.aliases.retain(|_, owner| !rejected.contains(owner)); + }); + catalog + .sections + .retain(|section| is_section(§ion.key) || !rejected.contains(§ion.key)); + for section in &mut catalog.sections { + let removed_before = removed_positions + .iter() + .filter(|index| **index < section.index) + .count(); + section.index -= removed_before; + } + (catalog, removed_errors) + } + + pub fn unrecognized(&self) -> impl Iterator { + self.rows().entries.iter().flat_map(|(name, entry)| { + entry + .model + .iter() + .flat_map(|model| &model.unrecognized) + .map(move |key| (name.as_str(), key.as_str())) + }) + } + + pub fn lookup(&self, key: &str) -> Option> { + let rows = self.rows(); + let found = self.store.find_key(key)?; + let (canonical_key, entry) = rows.entries.get_key_value(rows.owner(&found))?; + let matched_key = rows + .entries + .get_key_value(found.as_str()) .map(|(key, _)| key.as_str()) .or_else(|| { - self.aliases - .get_key_value(matched_key) + rows.aliases + .get_key_value(found.as_str()) .map(|(key, _)| key.as_str()) })?; Some(ModelMatch { @@ -204,30 +349,58 @@ impl Catalog { } pub fn model_count(&self) -> usize { - self.entries.len() + self.rows().entries.len() } + pub fn model_names(&self) -> impl Iterator { - self.entries.keys().map(String::as_str) - } - pub fn alias_count(&self) -> usize { - self.aliases.len() + self.rows().entries.keys().map(String::as_str) } + pub fn aliases(&self) -> &IndexMap { - &self.aliases + &self.rows().aliases } - pub fn alias_issues(&self) -> &[AliasIssue] { - &self.alias_issues + + fn rows(&self) -> &CatalogRows { + self.store.rows() } + pub fn sample_spec(&self) -> Option<&Value> { - self.sample_spec.as_ref() - } - pub fn fallback_generalizations(&self) -> Option<&FallbackGeneralizations> { - self.fallback_generalizations.as_ref() + self.sections + .iter() + .find(|section| section.key == SAMPLE_SPEC) + .map(|section| §ion.value) } + pub fn fallback_rules(&self) -> Option<&[FallbackRule]> { Some(self.fallback_generalizations.as_ref()?.rules.as_slice()) } - pub fn provenance(&self) -> &Provenance { - &self.provenance +} + +fn read_row(fields: &Fields) -> (Option, Option>, Vec) { + let wire = EntryWire::read(fields); + let aliases = wire.attributes.aliases.clone(); + let timezone_error = unknown_timezone(&wire); + match Model::try_from(wire) { + Ok(model) => (Some(model), aliases, timezone_error.into_iter().collect()), + Err(mut errors) => { + errors.extend(timezone_error); + (None, aliases, errors) + } } } + +fn is_section(key: &str) -> bool { + key == SAMPLE_SPEC || key == FALLBACK_GENERALIZATIONS +} + +fn unknown_timezone(wire: &EntryWire) -> Option { + let zone = wire.off_peak.as_ref()?.weekday_timezone.as_deref()?.trim(); + (!zone.is_empty() && jiff::tz::TimeZone::get(zone).is_err()).then(|| { + Error::from_path( + "off_peak_pricing.weekday_timezone", + Repr::UnknownTimezone { + zone: zone.to_owned(), + }, + ) + }) +} diff --git a/litellm-rust/crates/model-catalog/src/error.rs b/litellm-rust/crates/model-catalog/src/error.rs index 00902858be4..425ae47687f 100644 --- a/litellm-rust/crates/model-catalog/src/error.rs +++ b/litellm-rust/crates/model-catalog/src/error.rs @@ -1,21 +1,392 @@ +use std::num::ParseIntError; + +use strum::{EnumDiscriminants, IntoStaticStr, VariantArray}; use thiserror::Error; #[derive(Debug, Error)] -pub enum Error { - #[error("invalid catalog JSON")] - Json(#[from] serde_json::Error), - #[error("catalog is empty")] +#[error(transparent)] +pub struct Error(Box); + +#[derive(Debug, Error)] +enum Context { + #[error(transparent)] + Bare(Repr), + #[error("{}", display_path(path))] + At { + path: String, + #[source] + cause: Repr, + }, +} + +#[derive(Debug, Error)] +pub(crate) enum DateCause { + #[error("invalid date component")] + Number(#[from] ParseIntError), + #[error("calendar component out of range")] + Range(#[from] time::error::ComponentRange), +} + +#[derive(Debug, Error, EnumDiscriminants)] +#[strum_discriminants(name(ErrorKind), derive(IntoStaticStr, VariantArray), vis(pub))] +pub(crate) enum Repr { + #[error("invalid JSON")] + #[strum_discriminants(strum(serialize = "json"))] + Json(#[source] serde_json::Error), + #[error("has no models")] + #[strum_discriminants(strum(serialize = "empty"))] Empty, - #[error("model {model:?} must be an object")] - EntryNotObject { model: String }, - #[error("catalog has {actual} models, below minimum {minimum}")] + #[error("has {actual} models, below minimum {minimum}")] + #[strum_discriminants(strum(serialize = "below_minimum"))] BelowMinimum { actual: usize, minimum: usize }, - #[error("catalog has {actual} models, below {ratio} of reference count {reference}")] + #[error("has {actual} models, below {ratio} of reference count {reference}")] + #[strum_discriminants(strum(serialize = "shrunk"))] Shrunk { actual: usize, reference: usize, ratio: f64, }, - #[error("minimum reference ratio must be finite and between zero and one")] - InvalidRatio, + #[error("must be a JSON object")] + #[strum_discriminants(strum(serialize = "not_an_object"))] + NotAnObject, + #[error("invalid value")] + #[strum_discriminants(strum(serialize = "invalid"))] + Invalid(#[source] serde_json::Error), + #[error("not a field this block reads")] + #[strum_discriminants(strum(serialize = "unknown_block_key"))] + UnknownBlockKey, + #[error("required")] + #[strum_discriminants(strum(serialize = "missing_provider"))] + MissingProvider, + #[error("{zone:?} is not an IANA time zone")] + #[strum_discriminants(strum(serialize = "unknown_timezone"))] + UnknownTimezone { zone: String }, + #[error("alias {alias:?} is already a model name")] + #[strum_discriminants(strum(serialize = "alias_shadows_model"))] + AliasShadowsModel { alias: String }, + #[error("alias {alias:?} already belongs to {taken_by}")] + #[strum_discriminants(strum(serialize = "alias_taken"))] + AliasTaken { alias: String, taken_by: String }, + #[error("{first} and {second} price the same meter under the same conditions")] + #[strum_discriminants(strum(serialize = "duplicate_price"))] + DuplicatePrice { first: String, second: String }, + #[error("a price inside {block} takes no band or tier suffix")] + #[strum_discriminants(strum(serialize = "suffix_inside_block"))] + SuffixInsideBlock { block: &'static str }, + #[error("a tier needs exactly one of range and max_results_range")] + #[strum_discriminants(strum(serialize = "tier_range"))] + TierRange, + #[error("the range ends below its start")] + #[strum_discriminants(strum(serialize = "tier_reversed"))] + TierReversed, + #[error("starts at {start}, but the tiers before it end at {expected}")] + #[strum_discriminants(strum(serialize = "tier_gap"))] + TierGap { start: u64, expected: u64 }, + #[error("tiered_pricing mixes token ranges and result ranges")] + #[strum_discriminants(strum(serialize = "mixed_tier_basis"))] + MixedTierBasis, + #[error("a tier by result count only prices input_cost_per_query")] + #[strum_discriminants(strum(serialize = "not_priced_by_results"))] + NotPricedByResults, + #[error("tiered_pricing and prompt token bands are set in two places")] + #[strum_discriminants(strum(serialize = "bands_from_two_sources"))] + BandsFromTwoSources, + #[error("has no off-peak rate")] + #[strum_discriminants(strum(serialize = "not_an_off_peak_rate"))] + NotAnOffPeakRate, + #[error("off_peak_pricing has no prices")] + #[strum_discriminants(strum(serialize = "no_off_peak_prices"))] + NoOffPeakPrices, + #[error("off_peak_pricing needs hours_utc or windows")] + #[strum_discriminants(strum(serialize = "no_off_peak_hours"))] + NoOffPeakHours, + #[error("must list at least one entry")] + #[strum_discriminants(strum(serialize = "empty_list"))] + EmptyList, + #[error("expected a calendar date YYYY-MM-DD")] + #[strum_discriminants(strum(serialize = "invalid_date"))] + InvalidDate { + #[source] + cause: Option, + }, +} + +/// What a problem takes out of the served map +#[derive(Clone, Copy, Debug, PartialEq, Eq, VariantArray)] +pub enum Rejects { + Document, + Row, + Nothing, +} + +impl Error { + pub fn kind(&self) -> ErrorKind { + ErrorKind::from(self.repr()) + } + + pub fn rejects(&self) -> Rejects { + match self.repr() { + Repr::Json(_) | Repr::Empty | Repr::BelowMinimum { .. } | Repr::Shrunk { .. } => { + Rejects::Document + } + Repr::NotAnObject + | Repr::Invalid(_) + | Repr::UnknownBlockKey + | Repr::UnknownTimezone { .. } + | Repr::AliasShadowsModel { .. } + | Repr::AliasTaken { .. } + | Repr::DuplicatePrice { .. } + | Repr::SuffixInsideBlock { .. } + | Repr::TierRange + | Repr::TierReversed + | Repr::TierGap { .. } + | Repr::MixedTierBasis + | Repr::NotPricedByResults + | Repr::BandsFromTwoSources + | Repr::NotAnOffPeakRate + | Repr::NoOffPeakPrices + | Repr::NoOffPeakHours + | Repr::EmptyList + | Repr::InvalidDate { .. } => Rejects::Row, + Repr::MissingProvider => Rejects::Nothing, + } + } +} + +#[derive(Debug)] +pub struct Loaded { + pub value: T, + pub errors: Vec, +} + +impl From for Error { + fn from(cause: Repr) -> Self { + Self(Box::new(Context::Bare(cause))) + } +} + +impl From for Error { + fn from(cause: DateCause) -> Self { + Repr::InvalidDate { cause: Some(cause) }.into() + } +} + +impl Error { + fn repr(&self) -> &Repr { + match &*self.0 { + Context::Bare(cause) | Context::At { cause, .. } => cause, + } + } + + pub(crate) fn new(path: impl Into, cause: Repr) -> Self { + Self(Box::new(Context::At { + path: path.into(), + cause, + })) + } + + pub(crate) fn from_key(key: &str, cause: Repr) -> Self { + Self::new(pointer_for_key(key), cause) + } + + pub(crate) fn from_path(path: &str, cause: Repr) -> Self { + Self::new(path_to_pointer(path), cause) + } + + pub fn path(&self) -> &str { + match &*self.0 { + Context::Bare(_) => "", + Context::At { path, .. } => path, + } + } + + pub fn key(&self) -> Option { + pointer_root(self.path()) + } + + pub fn field(&self) -> Option { + let segments: Vec<_> = self + .path() + .split('/') + .skip(2) + .map(unescape_pointer_segment) + .collect(); + (!segments.is_empty()).then(|| segments.join("/")) + } + + pub(crate) fn prepend(self, key: &str) -> Self { + let (path, cause) = match *self.0 { + Context::Bare(cause) => (String::new(), cause), + Context::At { path, cause } => (path, cause), + }; + Self::new(format!("/{}{path}", escape_pointer_segment(key)), cause) + } +} + +pub(crate) fn pointer_root(path: &str) -> Option { + path.strip_prefix('/') + .and_then(|path| path.split('/').next()) + .map(unescape_pointer_segment) +} + +fn escape_pointer_segment(segment: &str) -> String { + segment.replace('~', "~0").replace('/', "~1") +} + +pub(crate) fn pointer_for_key(key: &str) -> String { + format!("/{}", escape_pointer_segment(key)) +} + +pub(crate) fn pointer_child(parent: &str, key: &str) -> String { + format!("{parent}/{}", escape_pointer_segment(key)) +} + +fn unescape_pointer_segment(segment: &str) -> String { + segment.replace("~1", "/").replace("~0", "~") +} + +fn path_to_pointer(key: &str) -> String { + let mut pointer = String::new(); + let mut segment = String::new(); + let mut chars = key.chars(); + let mut push_segment = |segment: &mut String| { + if !segment.is_empty() { + pointer.push('/'); + pointer.push_str(&escape_pointer_segment(segment)); + segment.clear(); + } + }; + while let Some(character) = chars.next() { + match character { + '.' => push_segment(&mut segment), + '[' => { + push_segment(&mut segment); + let mut index = String::new(); + for character in chars.by_ref() { + if character == ']' { + break; + } + index.push(character); + } + push_segment(&mut index); + } + character => segment.push(character), + } + } + push_segment(&mut segment); + pointer +} + +fn display_path(path: &str) -> String { + if path.is_empty() { + return "catalog".to_owned(); + } + path.split('/') + .skip(1) + .map(unescape_pointer_segment) + .enumerate() + .fold(String::new(), |mut display, (index, segment)| { + if index == 0 { + display.push_str(&segment); + } else if segment.bytes().all(|byte| byte.is_ascii_digit()) { + display.push('['); + display.push_str(&segment); + display.push(']'); + } else { + display.push('.'); + display.push_str(&segment); + } + display + }) +} + +#[cfg(test)] +mod tests { + use litellm_error::{Report, find}; + use rstest::rstest; + use serde_json::Value; + + use super::*; + use crate::{Catalog, wire::IsoDate}; + + #[rstest] + #[case::json(Repr::Json(serde_json::from_str::("{").unwrap_err()), Rejects::Document)] + #[case::empty(Repr::Empty, Rejects::Document)] + #[case::below_minimum(Repr::BelowMinimum { actual: 1, minimum: 2 }, Rejects::Document)] + #[case::shrunk(Repr::Shrunk { actual: 1, reference: 3, ratio: 0.5 }, Rejects::Document)] + #[case::not_an_object(Repr::NotAnObject, Rejects::Row)] + #[case::invalid(Repr::Invalid(serde_json::from_str::("x").unwrap_err()), Rejects::Row)] + #[case::unknown_block_key(Repr::UnknownBlockKey, Rejects::Row)] + #[case::missing_provider(Repr::MissingProvider, Rejects::Nothing)] + #[case::unknown_timezone(Repr::UnknownTimezone { zone: "invalid".into() }, Rejects::Row)] + #[case::alias_shadows_model(Repr::AliasShadowsModel { alias: "model".into() }, Rejects::Row)] + #[case::alias_taken(Repr::AliasTaken { alias: "alias".into(), taken_by: "owner".into() }, Rejects::Row)] + #[case::duplicate_price(Repr::DuplicatePrice { first: "a".into(), second: "b".into() }, Rejects::Row)] + #[case::suffix_inside_block(Repr::SuffixInsideBlock { block: "tiered_pricing" }, Rejects::Row)] + #[case::tier_range(Repr::TierRange, Rejects::Row)] + #[case::tier_reversed(Repr::TierReversed, Rejects::Row)] + #[case::tier_gap(Repr::TierGap { start: 1, expected: 0 }, Rejects::Row)] + #[case::mixed_tier_basis(Repr::MixedTierBasis, Rejects::Row)] + #[case::not_priced_by_results(Repr::NotPricedByResults, Rejects::Row)] + #[case::bands_from_two_sources(Repr::BandsFromTwoSources, Rejects::Row)] + #[case::not_an_off_peak_rate(Repr::NotAnOffPeakRate, Rejects::Row)] + #[case::no_off_peak_prices(Repr::NoOffPeakPrices, Rejects::Row)] + #[case::no_off_peak_hours(Repr::NoOffPeakHours, Rejects::Row)] + #[case::empty_list(Repr::EmptyList, Rejects::Row)] + #[case::invalid_date(Repr::InvalidDate { cause: None }, Rejects::Row)] + fn policy_classifies_each_error(#[case] error: Repr, #[case] expected: Rejects) { + assert_eq!(Error::from(error).rejects(), expected); + } + + #[rstest] + #[case::root("", "x", "/x")] + #[case::empty_model("/", "x", "//x")] + #[case::empty_parent_segment("/m/", "x", "/m//x")] + #[case::escaped("/m~1n", "a~/b", "/m~1n/a~0~1b")] + fn pointers_preserve_segments(#[case] parent: &str, #[case] key: &str, #[case] expected: &str) { + assert_eq!(pointer_child(parent, key), expected); + } + + #[rstest] + #[case::malformed(br#"{"#)] + #[case::bad_price(br#"{"m":{"litellm_provider":"test","input_cost_per_token":"free"}}"#)] + fn json_causes_are_typed_and_rendered_once(#[case] body: &[u8]) { + let error = match Catalog::parse(body) { + Ok(loaded) => loaded.errors.into_iter().next().expect("invalid price"), + Err(error) => error, + }; + let cause = find::(&error) + .expect("typed JSON cause") + .to_string(); + let report = Report::new(&error).to_string(); + assert_eq!(report.matches(&cause).count(), 1, "{report}"); + assert!(!error.to_string().contains(&cause)); + } + + #[rstest] + #[case::year("xxxx-01-01")] + #[case::month("2026-xx-01")] + #[case::day("2026-01-xx")] + fn date_number_errors_keep_the_cause(#[case] value: &str) { + let error = value.parse::().unwrap_err(); + let cause = find::(&error) + .expect("typed integer cause") + .to_string(); + assert_eq!(error.rejects(), Rejects::Row); + assert_eq!(Report::new(&error).to_string().matches(&cause).count(), 1); + assert!(!error.to_string().contains(&cause)); + } + + #[rstest] + #[case::month("2026-13-01")] + #[case::day("2026-02-30")] + fn date_range_errors_keep_the_cause(#[case] value: &str) { + let error = value.parse::().unwrap_err(); + let cause = find::(&error) + .expect("typed calendar cause") + .to_string(); + assert_eq!(error.rejects(), Rejects::Row); + assert_eq!(Report::new(&error).to_string().matches(&cause).count(), 1); + assert!(!error.to_string().contains(&cause)); + } } diff --git a/litellm-rust/crates/model-catalog/src/fallback.rs b/litellm-rust/crates/model-catalog/src/fallback.rs index 62291a84929..f5a0cadb6b9 100644 --- a/litellm-rust/crates/model-catalog/src/fallback.rs +++ b/litellm-rust/crates/model-catalog/src/fallback.rs @@ -1,10 +1,9 @@ -use serde::{Deserialize, Serialize}; use serde_json::Value; use std::collections::BTreeMap; /// One regex rule generalizing unknown model ids to known families. -#[derive(Clone, Debug, Deserialize, PartialEq, Serialize)] -#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))] +#[macro_rules_attribute::apply(crate::wire_type)] +#[derive(Eq, Hash)] pub struct FallbackRule { pub name: String, pub pattern: String, @@ -15,8 +14,8 @@ pub struct FallbackRule { } /// Regex rules that generalize unknown model ids to known families; not a model entry. -#[derive(Clone, Debug, Deserialize, PartialEq, Serialize)] -#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))] +#[macro_rules_attribute::apply(crate::wire_type)] +#[derive(Eq, Hash)] #[serde(deny_unknown_fields)] pub struct FallbackGeneralizations { pub rules: Vec, diff --git a/litellm-rust/crates/model-catalog/src/layers.rs b/litellm-rust/crates/model-catalog/src/layers.rs new file mode 100644 index 00000000000..881f0f6af9b --- /dev/null +++ b/litellm-rust/crates/model-catalog/src/layers.rs @@ -0,0 +1,32 @@ +use std::sync::Arc; + +use indexmap::IndexMap; +use serde_json::Value; + +use crate::catalog::Catalog; +use crate::register::Registrations; +use crate::wire::Fields; + +/// One field of a fetched row over the layer below, RFC 7386. An absent field inherits, JSON `null` is `Remove` +#[derive(Clone, Debug, PartialEq)] +pub enum Patch { + Set(Value), + Remove, +} + +/// The four layers served rows are composed from, lowest first +#[derive(Clone, Debug)] +#[expect( + dead_code, + reason = "02 and 07 add the functions that read these layers" +)] +pub struct Layers { + /// 0, the wheel's embedded map + bundled: Arc, + /// 1, the last installed fetch, rejected rows left out + fetched: Option>, + /// 2, `register(persist=False)`, cleared by every install and refilled by the router replay + transient: IndexMap, + /// 3, `register(persist=True)`, applied over every install + persisted: Registrations, +} diff --git a/litellm-rust/crates/model-catalog/src/lib.rs b/litellm-rust/crates/model-catalog/src/lib.rs index d42c05e6fe7..fe419dde054 100644 --- a/litellm-rust/crates/model-catalog/src/lib.rs +++ b/litellm-rust/crates/model-catalog/src/lib.rs @@ -7,20 +7,18 @@ //! in `litellm-router-core`, which reads this catalog through its dynamic config. //! //! # Structure -//! - `Catalog`: the parsed map. `Catalog::parse` takes the JSON body and a `Provenance` (where -//! the body came from: the bundled copy in the repo or the remote copy, with revision and etag), -//! and keeps every entry in file order with its raw fields. +//! - `Catalog`: the parsed map. `Catalog::parse` takes the JSON body and returns `Loaded`, +//! keeping every entry in file order with its raw fields and collecting row and alias errors. //! - `ModelEntry`: one model. `ModelEntry::fields` is the raw JSON object, so unknown fields are -//! never lost; `ModelEntry::info` is the typed mirror `ModelInfo` (pricing in `pricing`, -//! modalities and modes in `capabilities`). +//! never lost; `ModelEntry::model` is the optional typed `Model` with attributes, prices, +//! adjustments and capabilities. Invalid rows retain their raw fields without a typed model. //! - Aliases: an entry may list `aliases`; each alias resolves to its canonical key. Problems -//! found while indexing them (bad lists, collisions) are kept as `AliasIssue`s instead of -//! failing the parse. +//! found while indexing them (bad lists, collisions) are kept in `Loaded::errors` instead of +//! failing the document. //! - `FallbackGeneralizations` / `FallbackRule`: regex rules that map unknown model ids onto known //! families. They are data in the file, not entries. -//! - Validation: `Catalog::validate` checks a catalog against `IntegrityLimits` (a remote copy that -//! lost most of its models is refused), and `validate_registry` checks every entry against the -//! checked-in contract. +//! - Validation: `Catalog::check_size` checks a catalog against `IntegrityLimits` (a remote copy +//! that lost most of its models is refused), and `Catalog::row_errors` validates one row. //! //! # Lookups //! `Catalog::lookup` resolves a key exactly first (entry or alias), then case-insensitively, and @@ -31,21 +29,37 @@ //! Cost tracking, the router (through `RouterDynamicConfig`) and the gateway. Nothing in this crate //! performs I/O; loading and refreshing the file is the caller's job. -mod capabilities; +macro_rules_attribute::attribute_alias! { + #[apply(wire_type)] = + #[derive(Clone, Debug, PartialEq, serde::Serialize, serde::Deserialize)] + #[cfg_attr(feature = "schema", derive(schemars::JsonSchema))]; +} + +mod bundled; mod catalog; mod error; mod fallback; +mod layers; +mod lower; +mod model; mod model_info; -mod pricing; -mod validation; +mod register; +mod resolve; +mod store; +mod vocabulary; +pub mod wire; -pub use capabilities::*; +pub use bundled::*; pub use catalog::*; pub use error::*; pub use fallback::*; +pub use layers::*; +pub use model::*; pub use model_info::*; -pub use pricing::*; -pub use validation::*; +pub use register::*; +pub use resolve::*; +pub use store::*; +pub use vocabulary::*; #[cfg(feature = "schema")] mod schema; diff --git a/litellm-rust/crates/model-catalog/src/lower.rs b/litellm-rust/crates/model-catalog/src/lower.rs new file mode 100644 index 00000000000..4c3fc38a01a --- /dev/null +++ b/litellm-rust/crates/model-catalog/src/lower.rs @@ -0,0 +1,855 @@ +use std::collections::BTreeMap; +use std::collections::btree_map::Entry; +use std::num::NonZeroU64; + +use litellm_cost::money::UnitPrice; +use litellm_cost::price::{ + Adjustments, Band, BandEdge, Billing, CacheReadFallback, CacheReadMode, CacheTtl, + GuardrailUnit, InputUnit, Meter, OffPeak, OffPeakRates, OutputUnit, Price, Prices, Residency, + SearchContextSize, TextUnit, TierScope, TokenKind, Uplift, WallClock, WebSearchUnit, When, +}; +use litellm_cost::schedule::{Schedule, Window}; + +use crate::error::{Error, Repr, pointer_child, pointer_for_key}; +use crate::model::Model; +use crate::vocabulary::Mode; +use crate::wire::{EntryWire, FlatPrice, OffPeakBlock, TierRow, WebSearchBillingUnit}; + +impl TryFrom<&EntryWire> for Prices { + type Error = Vec; + + fn try_from(wire: &EntryWire) -> Result { + let mut problems = Vec::new(); + let sourced = flat_prices(&wire.prices) + .chain(tier_prices(wire, &mut problems)) + .chain(search_context_prices(wire)) + .chain(guardrail_prices(wire)); + let rates = dedupe(sourced, &mut problems); + let off_peak = wire + .off_peak + .as_ref() + .and_then(|block| off_peak(block, &mut problems)); + if problems.is_empty() { + let billing = billing(wire, &rates); + Ok(Prices::new(rates, off_peak, billing)) + } else { + Err(problems) + } + } +} + +fn billing(wire: &EntryWire, rates: &BTreeMap) -> Billing { + let provider = wire.attributes.litellm_provider.as_deref(); + let prices_meter = |priced: fn(&Meter) -> bool| rates.keys().any(|price| priced(&price.meter)); + let prices_tokens = prices_meter(|meter| { + matches!( + meter, + Meter::Input(InputUnit::Token(_)) | Meter::Output(OutputUnit::Token(_)) + ) + }); + let prices_characters = prices_meter(|meter| { + matches!( + meter, + Meter::Input(InputUnit::Character) | Meter::Output(OutputUnit::Character) + ) + }); + let prices_cache_read = prices_meter(|meter| { + matches!( + meter, + Meter::CacheRead { + kind: TokenKind::Text, + .. + } + ) + }); + Billing { + text_unit: if prices_characters && !prices_tokens { + TextUnit::Character + } else { + TextUnit::Token + }, + band_edge: if provider == Some("xai") { + BandEdge::Inclusive + } else { + BandEdge::Exclusive + }, + cache_read_fallback: if provider == Some("fireworks_ai") && !prices_cache_read { + CacheReadFallback::HalfInputRate + } else { + CacheReadFallback::InputRate + }, + wall_clock: match wire.attributes.mode { + None | Some(Mode::Chat | Mode::Completion | Mode::Embedding | Mode::Responses) => { + WallClock::ResponseTime + } + Some(Mode::AudioTranscription) => WallClock::AudioFile, + Some(_) => WallClock::Never, + }, + web_search: match wire.attributes.web_search_billing_unit { + Some(WebSearchBillingUnit::PerQuery) => WebSearchUnit::PerQuery, + Some(WebSearchBillingUnit::PerPrompt) | None => WebSearchUnit::PerPrompt, + }, + tier_scope: if provider == Some("dashscope") && wire.tier_rows.is_some() { + TierScope::WholeCall + } else { + TierScope::PerMeter + }, + } +} + +impl From<&EntryWire> for Adjustments { + fn from(wire: &EntryWire) -> Self { + Adjustments::new( + [ + (Uplift::NonGlobalEndpoint, wire.regional_endpoint_uplift), + (Uplift::Residency(Residency::Eu), wire.residency_uplift_eu), + (Uplift::Residency(Residency::Us), wire.residency_uplift_us), + ] + .into_iter() + .filter_map(|(uplift, factor)| Some((uplift, factor?))), + ) + } +} + +impl TryFrom for Model { + type Error = Vec; + + fn try_from(wire: EntryWire) -> Result { + let prices = Prices::try_from(&wire); + let adjustments = Adjustments::from(&wire); + let EntryWire { + attributes, + capabilities, + unknown, + problems, + .. + } = wire; + let missing_provider = attributes.litellm_provider.is_none().then_some(Error::new( + pointer_for_key("litellm_provider"), + Repr::MissingProvider, + )); + match prices { + Ok(prices) if problems.is_empty() && missing_provider.is_none() => Ok(Model { + attributes, + prices, + adjustments, + capabilities, + unrecognized: unknown.into_iter().map(|(key, _)| key).collect(), + }), + prices => Err(problems + .into_iter() + .chain(missing_provider) + .chain(prices.err().into_iter().flatten()) + .collect()), + } + } +} + +struct Sourced { + price: Price, + rate: UnitPrice, + source: String, + path: String, +} + +fn flat_prices(prices: &[FlatPrice]) -> impl Iterator + '_ { + prices.iter().map(|flat| { + let (channel, when) = flat.flat.channel_and_when(); + Sourced { + price: Price { + meter: flat.flat.meter.clone(), + channel, + when, + }, + rate: flat.price, + source: flat.key.clone(), + path: flat.path.clone(), + } + }) +} + +fn dedupe( + sourced: impl Iterator, + problems: &mut Vec, +) -> BTreeMap { + let mut seen: BTreeMap = BTreeMap::new(); + for Sourced { + price, + rate, + source, + path, + } in sourced + { + match seen.entry(price) { + Entry::Occupied(first) if first.get().0 == rate => {} + Entry::Occupied(first) => problems.push(Error::new( + path, + Repr::DuplicatePrice { + first: first.get().1.clone(), + second: source.clone(), + }, + )), + Entry::Vacant(slot) => { + slot.insert((rate, source, path)); + } + } + } + seen.into_iter() + .map(|(price, (rate, _, _))| (price, rate)) + .collect() +} + +#[derive(Clone, Copy, PartialEq, Eq)] +enum TierBasis { + PromptTokens, + Results, +} + +fn tier_basis(row: &TierRow) -> Result<(TierBasis, [u64; 2]), Error> { + match (row.range, row.max_results_range) { + (Some(range), None) => Ok((TierBasis::PromptTokens, range)), + (None, Some(range)) => Ok((TierBasis::Results, range)), + _ => Err(Error::new(row.path.clone(), Repr::TierRange)), + } +} + +fn tier_prices(wire: &EntryWire, problems: &mut Vec) -> Vec { + let Some(rows) = wire.tier_rows.as_deref().filter(|rows| !rows.is_empty()) else { + return Vec::new(); + }; + let based: Vec<_> = rows + .iter() + .filter_map(|row| match tier_basis(row) { + Ok((basis, range)) => Some((basis, range, row)), + Err(problem) => { + problems.push(problem); + None + } + }) + .collect(); + let Some(&(basis, ..)) = based.first() else { + return Vec::new(); + }; + if based.iter().any(|(other, ..)| *other != basis) { + problems.push(Error::from_key("tiered_pricing", Repr::MixedTierBasis)); + return Vec::new(); + } + if basis == TierBasis::PromptTokens { + problems.extend( + wire.prices + .iter() + .filter(|flat| flat.flat.band != Band::Base) + .map(|flat| Error::new(flat.path.clone(), Repr::BandsFromTwoSources)), + ); + } + let mut ordered = based; + ordered.sort_by_key(|(_, [start, _], _)| *start); + problems.extend(ladder_problems(basis, &ordered)); + ordered + .iter() + .flat_map(|(basis, [start, _], row)| { + row.prices.iter().map(move |flat| (*basis, *start, flat)) + }) + .filter_map( + |(basis, start, flat)| match tier_price(basis, start, flat) { + Ok(sourced) => Some(sourced), + Err(problem) => { + problems.push(problem); + None + } + }, + ) + .collect() +} + +fn ladder_problems(basis: TierBasis, ordered: &[(TierBasis, [u64; 2], &TierRow)]) -> Vec { + let step = match basis { + TierBasis::PromptTokens => 0, + TierBasis::Results => 1, + }; + let reversed = ordered + .iter() + .filter(|(_, [start, end], _)| end < start) + .map(|(.., row)| Error::new(row.path.clone(), Repr::TierReversed)); + let first = ordered + .first() + .filter(|(_, [start, _], _)| *start > step) + .map(|(_, [start, _], row)| { + Error::new( + row.path.clone(), + Repr::TierGap { + start: *start, + expected: 0, + }, + ) + }); + let gaps = ordered.windows(2).filter_map(|pair| { + let (_, [_, end], _) = pair[0]; + let (_, [start, _], row) = pair[1]; + (end.checked_add(step) != Some(start)).then(|| { + Error::new( + row.path.clone(), + Repr::TierGap { + start, + expected: end, + }, + ) + }) + }); + reversed.chain(first).chain(gaps).collect() +} + +fn tier_price(basis: TierBasis, start: u64, flat: &FlatPrice) -> Result { + if flat.flat.has_suffix() { + return Err(Error::new( + flat.path.clone(), + Repr::SuffixInsideBlock { + block: "tiered_pricing", + }, + )); + } + let (meter, band) = match (basis, &flat.flat.meter) { + (TierBasis::PromptTokens, meter) => (meter.clone(), Band::over_prompt_tokens(start)), + (TierBasis::Results, Meter::Input(InputUnit::Query { over_results: None })) => ( + Meter::Input(InputUnit::Query { + over_results: NonZeroU64::new(start.saturating_sub(1)), + }), + Band::Base, + ), + (TierBasis::Results, _) => { + return Err(Error::new(flat.path.clone(), Repr::NotPricedByResults)); + } + }; + Ok(Sourced { + price: Price::online( + meter, + When { + band, + ..When::default() + }, + ), + rate: flat.price, + source: flat.key.clone(), + path: flat.path.clone(), + }) +} + +fn search_context_prices(wire: &EntryWire) -> impl Iterator + '_ { + wire.search_context.iter().flat_map(|costs| { + [ + (SearchContextSize::Low, "low", costs.search_context_size_low), + ( + SearchContextSize::Medium, + "medium", + costs.search_context_size_medium, + ), + ( + SearchContextSize::High, + "high", + costs.search_context_size_high, + ), + ] + .into_iter() + .filter_map(|(size, name, rate)| { + Some(Sourced { + price: Price::online(Meter::WebSearch(size), When::default()), + rate: rate?, + source: format!("search_context_cost_per_query.search_context_size_{name}"), + path: pointer_child( + &pointer_for_key("search_context_cost_per_query"), + &format!("search_context_size_{name}"), + ), + }) + }) + }) +} + +fn guardrail_prices(wire: &EntryWire) -> impl Iterator + '_ { + wire.guardrail.iter().flatten().map(|(unit, rate)| Sourced { + price: Price::online(Meter::Guardrail(GuardrailUnit::new(unit)), When::default()), + rate: *rate, + source: format!("guardrail_cost_per_unit.{unit}"), + path: pointer_child(&pointer_for_key("guardrail_cost_per_unit"), unit), + }) +} + +fn off_peak(block: &OffPeakBlock, problems: &mut Vec) -> Option { + let rates = off_peak_rates(&block.prices, problems); + let schedule = schedule(block, problems)?; + let off_peak = OffPeak::new(schedule, rates); + if off_peak.is_none() { + problems.push(Error::from_key("off_peak_pricing", Repr::NoOffPeakPrices)); + } + off_peak +} + +fn off_peak_rates(prices: &[FlatPrice], problems: &mut Vec) -> OffPeakRates { + let mut rates = OffPeakRates::default(); + for flat in prices { + if flat.flat.has_suffix() { + problems.push(Error::new( + flat.path.clone(), + Repr::SuffixInsideBlock { + block: "off_peak_pricing", + }, + )); + continue; + } + let slot = match flat.flat.meter { + Meter::Input(InputUnit::Token(TokenKind::Text)) => &mut rates.input, + Meter::Output(OutputUnit::Token(TokenKind::Text)) => &mut rates.output, + Meter::Output(OutputUnit::ReasoningToken) => &mut rates.reasoning, + Meter::CacheRead { + kind: TokenKind::Text, + mode: CacheReadMode::Implicit, + } => &mut rates.cache_read, + Meter::CacheRead { + kind: TokenKind::Text, + mode: CacheReadMode::Explicit, + } => &mut rates.explicit_cache_read, + Meter::CacheWrite { + kind: TokenKind::Text, + ttl: CacheTtl::Default, + } => &mut rates.cache_write, + _ => { + problems.push(Error::new(flat.path.clone(), Repr::NotAnOffPeakRate)); + continue; + } + }; + *slot = Some(flat.price); + } + rates +} + +fn schedule(block: &OffPeakBlock, problems: &mut Vec) -> Option { + let empty = |key: String| Error::from_path(&key, Repr::EmptyList); + let every_day = block + .hours_utc + .iter() + .map(|hours| ("off_peak_pricing.hours_utc".to_owned(), hours, None)); + let windows = block + .windows + .iter() + .flatten() + .enumerate() + .map(|(index, window)| { + ( + format!("off_peak_pricing.windows[{index}].hours_utc"), + &window.hours_utc, + Some((index, window.weekdays.as_ref())), + ) + }); + let windows: Vec = every_day + .chain(windows) + .map(|(key, hours, weekdays)| { + if hours.windows().is_empty() { + problems.push(empty(key)); + } + if let Some((index, Some(days))) = weekdays + && days.is_empty() + { + problems.push(empty(format!("off_peak_pricing.windows[{index}].weekdays"))); + } + Window { + hours: hours.windows().to_vec(), + weekdays: weekdays.and_then(|(_, days)| days.cloned()), + } + }) + .collect(); + if block.windows.as_ref().is_some_and(Vec::is_empty) { + problems.push(empty("off_peak_pricing.windows".to_owned())); + } + if block.hours_utc.is_none() && block.windows.is_none() { + problems.push(Error::from_key("off_peak_pricing", Repr::NoOffPeakHours)); + return None; + } + Some(Schedule { + windows, + weekday_timezone: block.weekday_timezone.clone(), + }) +} + +#[cfg(test)] +mod tests { + use litellm_cost::money::Multiplier; + use litellm_cost::price::{Channel, ServiceTier}; + use litellm_cost::schedule::{UtcWindow, Weekday}; + use rstest::rstest; + use serde_json::{Value, json}; + + use super::*; + + const TEXT_INPUT: Meter = Meter::Input(InputUnit::Token(TokenKind::Text)); + + fn wire(entry: Value) -> EntryWire { + let Value::Object(fields) = entry else { + panic!("entry must be an object") + }; + let wire = EntryWire::read(&fields); + assert!(wire.problems.is_empty(), "{:?}", wire.problems); + wire + } + + fn prices(entry: Value) -> Result> { + Prices::try_from(&wire(entry)) + } + + fn online(meter: Meter, tier: ServiceTier, band: Band) -> Price { + Price::online(meter, When { tier, band }) + } + + #[rstest] + #[case::flat_key(json!({"input_cost_per_token": 1e-6}), online(TEXT_INPUT, ServiceTier::Standard, Band::Base))] + #[case::band_and_tier( + json!({"input_cost_per_token_above_272k_tokens_priority": 1e-6}), + online(TEXT_INPUT, ServiceTier::Priority, Band::over_prompt_tokens(272_000)) + )] + #[case::batch( + json!({"input_cost_per_token_batches": 1e-6}), + Price { meter: TEXT_INPUT, channel: Channel::Batch, when: When::default() } + )] + #[case::first_token_tier_is_the_base_band( + json!({"tiered_pricing": [{"range": [0.0, 256000.0], "input_cost_per_token": 1e-6}]}), + online(TEXT_INPUT, ServiceTier::Standard, Band::Base) + )] + #[case::result_tier_stays_in_its_meter( + json!({"tiered_pricing": [ + {"max_results_range": [0, 25], "input_cost_per_query": 2e-6}, + {"max_results_range": [26, 100], "input_cost_per_query": 1e-6} + ]}), + online( + Meter::Input(InputUnit::Query { over_results: NonZeroU64::new(25) }), + ServiceTier::Standard, + Band::Base + ) + )] + #[case::search_context( + json!({"search_context_cost_per_query": {"search_context_size_low": 1e-6}}), + online(Meter::WebSearch(SearchContextSize::Low), ServiceTier::Standard, Band::Base) + )] + #[case::character_band_is_a_prompt_token_band( + json!({"input_cost_per_character_above_128k_tokens": 1e-6}), + online(Meter::Input(InputUnit::Character), ServiceTier::Standard, Band::over_prompt_tokens(128_000)) + )] + #[case::guardrail( + json!({"guardrail_cost_per_unit": {"contentPolicyUnits": 1e-6}}), + online(Meter::Guardrail(GuardrailUnit::new("contentPolicyUnits")), ServiceTier::Standard, Band::Base) + )] + fn every_spelling_lowers_to_one_price(#[case] entry: Value, #[case] expected: Price) { + assert_eq!(prices(entry).unwrap().get(&expected), UnitPrice::new(1e-6)); + } + + #[rstest] + fn two_spellings_that_agree_are_one_price() { + let prices = prices(json!({ + "input_cost_per_token": 1e-6, + "tiered_pricing": [ + {"range": [0, 10], "input_cost_per_token": 1e-6}, + {"range": [10, 20], "input_cost_per_token": 2e-6} + ] + })) + .unwrap(); + assert_eq!( + prices.get(&Price::online(TEXT_INPUT, When::default())), + UnitPrice::new(1e-6) + ); + } + + #[rstest] + fn token_tiers_lower_to_prompt_bands_at_each_start() { + let prices = prices(json!({"tiered_pricing": [ + {"range": [0, 256000], "input_cost_per_token": 1e-6}, + {"range": [256000, 1000000], "input_cost_per_token": 2e-6} + ]})) + .unwrap(); + assert_eq!( + prices.band(ServiceTier::Standard, 256_001), + Band::over_prompt_tokens(256_000) + ); + assert_eq!( + prices + .rate( + &TEXT_INPUT, + ServiceTier::Standard, + Band::over_prompt_tokens(256_000) + ) + .map(UnitPrice::get), + Some(2e-6) + ); + } + + #[rstest] + #[case::two_spellings_of_one_price( + json!({ + "input_cost_per_token": 1e-6, + "tiered_pricing": [{"range": [0, 10], "input_cost_per_token": 2e-6}] + }), + "/tiered_pricing/0/input_cost_per_token", + "duplicate_price" + )] + #[case::zero_band_is_the_base_price( + json!({"input_cost_per_token": 1e-6, "input_cost_per_token_above_0k_tokens": 2e-6}), + "/input_cost_per_token_above_0k_tokens", + "duplicate_price" + )] + #[case::tier_without_range( + json!({"tiered_pricing": [{"input_cost_per_token": 1e-6}]}), + "/tiered_pricing/0", + "tier_range" + )] + #[case::tier_with_suffix( + json!({"tiered_pricing": [{"range": [0, 10], "input_cost_per_token_priority": 1e-6}]}), + "/tiered_pricing/0/input_cost_per_token_priority", + "suffix_inside_block" + )] + #[case::reversed_range( + json!({"tiered_pricing": [{"range": [0, 10], "input_cost_per_token": 1e-6}, {"range": [10, 5], "input_cost_per_token": 1e-6}]}), + "/tiered_pricing/1", + "tier_reversed" + )] + #[case::gap_between_tiers( + json!({"tiered_pricing": [ + {"range": [0, 10], "input_cost_per_token": 1e-6}, + {"range": [20, 30], "input_cost_per_token": 2e-6} + ]}), + "/tiered_pricing/1", + "tier_gap" + )] + #[case::ladder_not_from_zero( + json!({"tiered_pricing": [{"range": [5, 10], "input_cost_per_token": 1e-6}]}), + "/tiered_pricing/0", + "tier_gap" + )] + #[case::result_ladder_overlaps( + json!({"tiered_pricing": [ + {"max_results_range": [1, 10], "input_cost_per_query": 1e-6}, + {"max_results_range": [10, 20], "input_cost_per_query": 2e-6} + ]}), + "/tiered_pricing/1", + "tier_gap" + )] + #[case::mixed_basis( + json!({"tiered_pricing": [ + {"range": [0, 10], "input_cost_per_token": 1e-6}, + {"max_results_range": [11, 20], "input_cost_per_query": 2e-6} + ]}), + "/tiered_pricing", + "mixed_tier_basis" + )] + #[case::result_tier_prices_a_token( + json!({"tiered_pricing": [{"max_results_range": [0, 10], "input_cost_per_token": 1e-6}]}), + "/tiered_pricing/0/input_cost_per_token", + "not_priced_by_results" + )] + #[case::two_band_sources( + json!({ + "output_cost_per_token_above_200k_tokens": 1e-6, + "tiered_pricing": [{"range": [0, 10], "input_cost_per_token": 1e-6}] + }), + "/output_cost_per_token_above_200k_tokens", + "bands_from_two_sources" + )] + #[case::off_peak_without_hours( + json!({"off_peak_pricing": {"input_cost_per_token": 1e-6}}), + "/off_peak_pricing", + "no_off_peak_hours" + )] + #[case::off_peak_without_prices( + json!({"off_peak_pricing": {"hours_utc": "00:00-01:00"}}), + "/off_peak_pricing", + "no_off_peak_prices" + )] + #[case::off_peak_empty_windows( + json!({"off_peak_pricing": {"windows": [], "input_cost_per_token": 1e-6}}), + "/off_peak_pricing/windows", + "empty_list" + )] + #[case::off_peak_empty_hours( + json!({"off_peak_pricing": {"hours_utc": [], "input_cost_per_token": 1e-6}}), + "/off_peak_pricing/hours_utc", + "empty_list" + )] + #[case::off_peak_empty_weekdays( + json!({"off_peak_pricing": { + "windows": [{"hours_utc": "00:00-01:00", "weekdays": []}], + "input_cost_per_token": 1e-6 + }}), + "/off_peak_pricing/windows/0/weekdays", + "empty_list" + )] + #[case::off_peak_image_price( + json!({"off_peak_pricing": {"hours_utc": "00:00-01:00", "output_cost_per_image": 1}}), + "/off_peak_pricing/output_cost_per_image", + "not_an_off_peak_rate" + )] + #[case::off_peak_band( + json!({"off_peak_pricing": {"hours_utc": "00:00-01:00", "input_cost_per_token_above_200k_tokens": 1}}), + "/off_peak_pricing/input_cost_per_token_above_200k_tokens", + "suffix_inside_block" + )] + fn rules_across_keys_name_where_each_price_came_from( + #[case] entry: Value, + #[case] expected_path: &str, + #[case] expected_variant: &str, + ) { + let problems = prices(entry).unwrap_err(); + assert!( + problems + .iter() + .any(|problem| problem.path() == expected_path + && <&str>::from(problem.kind()) == expected_variant), + "{problems:?}" + ); + } + + #[rstest] + fn off_peak_hours_and_windows_form_one_schedule_with_its_rates() { + let prices = prices(json!({"off_peak_pricing": { + "hours_utc": ["00:00-01:00"], + "windows": [{"hours_utc": "10:00-12:00", "weekdays": [6, " sun "]}], + "weekday_timezone": "Asia/Shanghai", + "output_cost_per_token": 1e-6, + "explicit_cache_read_input_token_cost": 2e-6 + }})) + .unwrap(); + let off_peak = prices.off_peak().unwrap(); + let window = |text: &str| text.parse::().unwrap(); + + assert_eq!( + off_peak.schedule(), + &Schedule { + windows: vec![ + Window { + hours: vec![window("00:00-01:00")], + weekdays: None, + }, + Window { + hours: vec![window("10:00-12:00")], + weekdays: Some(vec![Weekday::Saturday, Weekday::Sunday]), + }, + ], + weekday_timezone: Some("Asia/Shanghai".into()), + } + ); + assert_eq!( + off_peak.rates(), + &OffPeakRates { + output: UnitPrice::new(1e-6), + explicit_cache_read: UnitPrice::new(2e-6), + ..OffPeakRates::default() + } + ); + assert_eq!(prices.iter().count(), 0); + } + + #[rstest] + fn uplifts_are_adjustments_not_prices() { + let wire = wire(json!({ + "regional_endpoint_uplift_multiplier": 1.1, + "regional_processing_uplift_multiplier_eu": "1.2" + })); + let adjustments = Adjustments::from(&wire); + + assert_eq!(Prices::try_from(&wire).unwrap().iter().count(), 0); + assert_eq!( + adjustments + .get(&Uplift::NonGlobalEndpoint) + .map(Multiplier::get), + Some(1.1) + ); + assert_eq!( + adjustments + .get(&Uplift::Residency(Residency::Eu)) + .map(Multiplier::get), + Some(1.2) + ); + assert_eq!(adjustments.get(&Uplift::Residency(Residency::Us)), None); + } + + #[rstest] + fn a_model_reports_field_provider_and_price_problems_together() { + let Value::Object(fields) = json!({ + "mode": "chatt", + "tiered_pricing": [{"input_cost_per_token": 1e-6}] + }) else { + unreachable!() + }; + let problems: Vec = Model::from_fields(&fields) + .unwrap_err() + .iter() + .map(|problem| litellm_error::Report::new(problem).to_string()) + .collect(); + + assert_eq!(problems.len(), 3, "{problems:?}"); + assert!(problems[0].starts_with("mode: invalid value: unknown variant `chatt`")); + assert_eq!(problems[1], "litellm_provider: required"); + assert_eq!( + problems[2], + "tiered_pricing[0]: a tier needs exactly one of range and max_results_range" + ); + } + + #[rstest] + fn a_null_value_is_absent() { + let Value::Object(fields) = json!({ + "litellm_provider": "openai", + "input_cost_per_token": null, + "mode": null + }) else { + unreachable!() + }; + let model = Model::from_fields(&fields).unwrap(); + + assert_eq!(model.prices.iter().count(), 0); + assert_eq!(model.attributes.mode, None); + assert!(model.unrecognized.is_empty()); + } + + #[rstest] + #[case::tokens_win_over_characters( + json!({"litellm_provider": "vertex_ai-language-models", "input_cost_per_character": 1e-6, "input_cost_per_token": 1e-6}), + Billing::default() + )] + #[case::characters_only( + json!({"litellm_provider": "vertex_ai-language-models", "input_cost_per_character": 1e-6}), + Billing { text_unit: TextUnit::Character, ..Billing::default() } + )] + #[case::xai_band_is_inclusive( + json!({"litellm_provider": "xai", "input_cost_per_token": 1e-6}), + Billing { band_edge: BandEdge::Inclusive, ..Billing::default() } + )] + #[case::fireworks_without_cache_read_borrows_half( + json!({"litellm_provider": "fireworks_ai", "input_cost_per_token": 1e-6}), + Billing { cache_read_fallback: CacheReadFallback::HalfInputRate, ..Billing::default() } + )] + #[case::fireworks_with_cache_read_keeps_it( + json!({"litellm_provider": "fireworks_ai", "input_cost_per_token": 1e-6, "cache_read_input_token_cost": 1e-7}), + Billing::default() + )] + #[case::transcription_prices_the_audio_file( + json!({"litellm_provider": "openai", "mode": "audio_transcription", "input_cost_per_second": 1e-4}), + Billing { wall_clock: WallClock::AudioFile, ..Billing::default() } + )] + #[case::provisioned_chat_prices_response_time( + json!({"litellm_provider": "bedrock", "mode": "chat", "cost_per_second": 1e-2}), + Billing::default() + )] + #[case::image_generation_is_never_wall_clock( + json!({"litellm_provider": "openai", "mode": "image_generation"}), + Billing { wall_clock: WallClock::Never, ..Billing::default() } + )] + #[case::gemini_per_query_search( + json!({"litellm_provider": "gemini", "web_search_billing_unit": "per_query"}), + Billing { web_search: WebSearchUnit::PerQuery, ..Billing::default() } + )] + #[case::dashscope_tiers_bill_the_whole_call( + json!({"litellm_provider": "dashscope", "tiered_pricing": [{"range": [0, 32000], "input_cost_per_token": 1e-6}]}), + Billing { tier_scope: TierScope::WholeCall, ..Billing::default() } + )] + fn billing_is_lowered_from_the_row_and_its_provider( + #[case] entry: Value, + #[case] expected: Billing, + ) { + assert_eq!(prices(entry).unwrap().billing(), expected); + } +} diff --git a/litellm-rust/crates/model-catalog/src/model.rs b/litellm-rust/crates/model-catalog/src/model.rs new file mode 100644 index 00000000000..c1acd1c14cb --- /dev/null +++ b/litellm-rust/crates/model-catalog/src/model.rs @@ -0,0 +1,22 @@ +use std::collections::BTreeMap; + +use litellm_cost::price::{Adjustments, Prices}; + +use crate::error::Error; +use crate::wire::{Attributes, EntryWire, Fields}; + +/// One catalog entry read into types. +#[derive(Clone, Debug, PartialEq)] +pub struct Model { + pub attributes: Attributes, + pub prices: Prices, + pub adjustments: Adjustments, + pub capabilities: BTreeMap, + pub unrecognized: Vec, +} + +impl Model { + pub fn from_fields(fields: &Fields) -> Result> { + Self::try_from(EntryWire::read(fields)) + } +} diff --git a/litellm-rust/crates/model-catalog/src/model_info.rs b/litellm-rust/crates/model-catalog/src/model_info.rs index ebee2c64140..8f7549c1c6d 100644 --- a/litellm-rust/crates/model-catalog/src/model_info.rs +++ b/litellm-rust/crates/model-catalog/src/model_info.rs @@ -1,582 +1,194 @@ -use crate::capabilities::{AudioFormat, InputModality, Mode, OutputModality, VertexAiAudioApi}; -use crate::pricing::{OffPeakPricing, SearchContextCostPerQuery, TieredRate, WebSearchBillingUnit}; -use litellm_llms_types::formats::chat_completions::ReasoningEffort; -use serde::{Deserialize, Serialize}; +use indexmap::IndexMap; use serde_json::Value; -use std::collections::BTreeMap; -/// Typed mirror of one catalog model entry. -#[derive(Clone, Debug, Deserialize, PartialEq, Serialize)] -#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))] -pub struct ModelInfo { - #[serde(skip_serializing_if = "Option::is_none")] - pub annotation_cost_per_page: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub annotation_cost_per_page_batches: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub audio_transcription_config: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub bedrock_converse_supports_strict_tools: Option, - /// Highest reasoning effort the Bedrock output_config accepts for this model. - #[serde(skip_serializing_if = "Option::is_none")] - pub bedrock_output_config_effort_ceiling: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_creation_input_audio_token_cost: Option, - /// USD per token written to the provider's prompt cache. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_creation_input_token_cost: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_creation_input_token_cost_above_32k_tokens: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_creation_input_token_cost_above_100k_tokens: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_creation_input_token_cost_above_100k_tokens_batches: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_creation_input_token_cost_above_128k_tokens: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_creation_input_token_cost_above_1hr: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_creation_input_token_cost_above_1hr_above_100k_tokens: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_creation_input_token_cost_above_1hr_above_200k_tokens: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_creation_input_token_cost_above_200k_tokens: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_creation_input_token_cost_above_200k_tokens_batches: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_creation_input_token_cost_above_256k_tokens: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_creation_input_token_cost_above_272k_tokens: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_creation_input_token_cost_above_272k_tokens_batches: Option, - /// Flex service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_creation_input_token_cost_above_272k_tokens_flex: Option, - /// Priority service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_creation_input_token_cost_above_272k_tokens_priority: Option, - /// Ultrafast service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_creation_input_token_cost_above_272k_tokens_ultrafast: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_creation_input_token_cost_batches: Option, - /// Flex service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_creation_input_token_cost_flex: Option, - /// Priority service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_creation_input_token_cost_priority: Option, - /// Ultrafast service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_creation_input_token_cost_ultrafast: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_read_input_audio_token_cost: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_read_input_image_token_cost: Option, - /// USD per prompt token served from the provider's prompt cache. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_read_input_token_cost: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_read_input_token_cost_above_32k_tokens: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_read_input_token_cost_above_100k_tokens: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_read_input_token_cost_above_100k_tokens_batches: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_read_input_token_cost_above_128k_tokens: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_read_input_token_cost_above_200k_tokens: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_read_input_token_cost_above_200k_tokens_batches: Option, - /// Priority service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_read_input_token_cost_above_200k_tokens_priority: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_read_input_token_cost_above_256k_tokens: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_read_input_token_cost_above_272k_tokens: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_read_input_token_cost_above_272k_tokens_batches: Option, - /// Flex service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_read_input_token_cost_above_272k_tokens_flex: Option, - /// Priority service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_read_input_token_cost_above_272k_tokens_priority: Option, - /// Ultrafast service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_read_input_token_cost_above_272k_tokens_ultrafast: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_read_input_token_cost_above_512k_tokens: Option, - /// Balanced service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_read_input_token_cost_balanced: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_read_input_token_cost_batches: Option, - /// Flex service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_read_input_token_cost_flex: Option, - /// Priority service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_read_input_token_cost_priority: Option, - /// Ultrafast service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_read_input_token_cost_ultrafast: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub citation_cost_per_token: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub code_interpreter_cost_per_session: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub comment: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub computer_use_input_cost_per_1k_tokens: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub computer_use_output_cost_per_1k_tokens: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub cost_per_second: Option, - /// Reasoning effort the provider applies when the request omits reasoning_effort. Gates whether a non-default temperature or the top_p/logprobs sampling params are accepted, which hold only when the effort resolves to 'none'. - #[serde(skip_serializing_if = "Option::is_none")] - pub default_reasoning_effort: Option, - /// Date the provider deprecates the model, YYYY-MM-DD. - #[serde(skip_serializing_if = "Option::is_none")] - pub deprecation_date: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub file_search_cost_per_1k_calls: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub file_search_cost_per_gb_per_day: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub gemini_audio_only_live: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub gemini_native_audio: Option, - /// USD per Grounding with Google Maps request; billed per query or per prompt per web_search_billing_unit. - #[serde(skip_serializing_if = "Option::is_none")] - pub google_maps_grounding_cost_per_query: Option, - /// USD cost per billable guardrail unit, keyed by the provider's usage counter name (e.g. Bedrock's contentPolicyUnits). - #[serde(skip_serializing_if = "Option::is_none")] - pub guardrail_cost_per_unit: Option>, - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_audio_per_second: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_audio_per_second_above_128k_tokens: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_audio_token: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_audio_token_batches: Option, - /// Priority service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_audio_token_priority: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_character: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_character_above_128k_tokens: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_image: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_image_above_128k_tokens: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_image_token: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_image_token_batches: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_pixel: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_query: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_request: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_second: Option, - /// USD per prompt token. - #[serde(skip_serializing_if = "Option::is_none")] +/// The fields main's `ModelInfoBase(...)` copies (`litellm/utils.py` at c7c895db45), one typed field each, +/// read from the raw row without lowering it, so building one never fails +#[derive(Clone, Debug, Default, PartialEq)] +pub struct ModelInfoFields { + pub key: String, + pub max_tokens: Option, + pub max_input_tokens: Option, + pub max_output_tokens: Option, pub input_cost_per_token: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_token_above_32k_tokens: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_token_above_100k_tokens: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_token_above_100k_tokens_batches: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_token_above_128k_tokens: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_token_above_200k_tokens: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_token_above_200k_tokens_batches: Option, - /// Priority service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_token_above_200k_tokens_priority: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_token_above_256k_tokens: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_token_above_272k_tokens: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_token_above_272k_tokens_batches: Option, - /// Flex service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_token_above_272k_tokens_flex: Option, - /// Priority service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_token_above_272k_tokens_priority: Option, - /// Ultrafast service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_token_above_272k_tokens_ultrafast: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_token_above_512k_tokens: Option, - /// Balanced service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_token_balanced: Option, - /// USD per prompt token via the provider's batch API. - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_token_batches: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_token_cache_hit: Option, - /// Flex service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] pub input_cost_per_token_flex: Option, - /// Priority service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] pub input_cost_per_token_priority: Option, - /// Ultrafast service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] + pub input_cost_per_token_balanced: Option, pub input_cost_per_token_ultrafast: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_video_per_second: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_video_per_second_above_128k_tokens: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_video_per_second_above_15s_interval: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_video_per_second_above_8s_interval: Option, - #[serde(skip_serializing_if = "Option::is_none")] + pub cache_creation_input_token_cost: Option, + pub cache_creation_input_token_cost_above_200k_tokens: Option, + pub cache_creation_input_token_cost_above_272k_tokens: Option, + pub cache_creation_input_token_cost_above_272k_tokens_priority: Option, + pub cache_creation_input_token_cost_above_272k_tokens_flex: Option, + pub cache_creation_input_token_cost_above_272k_tokens_ultrafast: Option, + pub cache_creation_input_token_cost_flex: Option, + pub cache_creation_input_token_cost_priority: Option, + pub cache_creation_input_token_cost_ultrafast: Option, + pub cache_read_input_token_cost: Option, + pub cache_read_input_audio_token_cost: Option, + pub cache_read_input_audio_token_cost_flex: Option, + pub cache_read_input_audio_token_cost_priority: Option, + pub cache_read_input_audio_token_cost_batches: Option, + pub explicit_cache_read_input_token_cost: Option, + pub prompt_cache_min_tokens: Option, + pub cache_read_input_token_cost_above_200k_tokens: Option, + pub cache_read_input_token_cost_above_200k_tokens_priority: Option, + pub cache_read_input_token_cost_above_272k_tokens: Option, + pub cache_read_input_token_cost_above_272k_tokens_priority: Option, + pub cache_read_input_token_cost_above_272k_tokens_flex: Option, + pub cache_read_input_token_cost_above_272k_tokens_ultrafast: Option, + pub cache_read_input_token_cost_above_512k_tokens: Option, + pub cache_read_input_token_cost_flex: Option, + pub cache_read_input_token_cost_priority: Option, + pub cache_read_input_token_cost_balanced: Option, + pub cache_read_input_token_cost_ultrafast: Option, + pub cache_read_input_token_cost_batches: Option, + pub cache_read_input_token_cost_above_200k_tokens_batches: Option, + pub cache_read_input_token_cost_above_272k_tokens_batches: Option, + pub cache_creation_input_token_cost_batches: Option, + pub cache_creation_input_token_cost_above_200k_tokens_batches: Option, + pub cache_creation_input_token_cost_above_272k_tokens_batches: Option, + pub cache_creation_input_token_cost_above_1hr: Option, + pub cache_creation_input_token_cost_above_12hr: Option, + pub cache_creation_input_token_cost_above_12hr_flex: Option, + pub cache_creation_input_token_cost_above_12hr_priority: Option, + pub cache_creation_input_token_cost_above_12hr_above_272k_tokens: Option, + pub cache_creation_input_token_cost_above_12hr_above_272k_tokens_flex: Option, + pub cache_creation_input_token_cost_above_12hr_above_272k_tokens_priority: Option, + pub cache_creation_input_token_cost_above_12hr_batches: Option, + pub cache_creation_input_token_cost_above_12hr_ultrafast: Option, + pub cache_creation_input_token_cost_above_12hr_above_272k_tokens_batches: Option, + pub cache_creation_input_token_cost_above_12hr_above_272k_tokens_ultrafast: Option, + pub off_peak_pricing: Option, + pub input_cost_per_character: Option, + pub input_cost_per_token_above_128k_tokens: Option, + pub input_cost_per_token_above_200k_tokens: Option, + pub input_cost_per_token_above_200k_tokens_priority: Option, + pub input_cost_per_token_above_272k_tokens: Option, + pub input_cost_per_token_above_272k_tokens_priority: Option, + pub input_cost_per_token_above_272k_tokens_flex: Option, + pub input_cost_per_token_above_272k_tokens_ultrafast: Option, + pub input_cost_per_token_above_512k_tokens: Option, + pub input_cost_per_query: Option, + pub cost_per_second: Option, + pub input_cost_per_second: Option, + pub input_cost_per_text_message: Option, + pub input_cost_per_audio_token: Option, + pub input_cost_per_audio_token_flex: Option, + pub input_cost_per_audio_token_priority: Option, + pub input_cost_per_image_token: Option, pub input_cost_per_video_token: Option, - #[serde(skip_serializing_if = "Option::is_none")] + pub input_cost_per_audio_token_batches: Option, + pub input_cost_per_image_token_batches: Option, + pub input_cost_per_image: Option, + pub input_cost_per_audio_per_second: Option, + pub input_cost_per_video_per_second: Option, + pub input_cost_per_token_batches: Option, pub input_cost_per_video_token_batches: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub input_dbu_cost_per_token: Option, - /// LiteLLM provider slug; one of https://docs.litellm.ai/docs/providers. - #[serde(skip_serializing_if = "Option::is_none")] - pub litellm_provider: Option, - /// Maximum prompt/context tokens the model accepts. - #[serde(skip_serializing_if = "Option::is_none")] - pub max_input_tokens: Option, - /// Maximum tokens the model can generate in one response. - #[serde(skip_serializing_if = "Option::is_none")] - pub max_output_tokens: Option, - /// Legacy field: max output tokens if the provider specifies it, else max input tokens. - #[serde(skip_serializing_if = "Option::is_none")] - pub max_tokens: Option, - /// Free-form notes about the entry (e.g. pricing derivation). - #[serde(skip_serializing_if = "Option::is_none")] - pub metadata: Option>, - /// Primary API surface / task type of the model. - #[serde(skip_serializing_if = "Option::is_none")] - pub mode: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub ocr_cost_per_credit: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub ocr_cost_per_page: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub ocr_cost_per_page_batches: Option, - /// Rates that replace the same-named base fields while the request falls inside the stated UTC windows. - #[serde(skip_serializing_if = "Option::is_none")] - pub off_peak_pricing: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_audio_token: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_character: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_character_above_128k_tokens: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_image: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_image_1024: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_image_1536: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_image_512: Option, - #[serde( - rename = "output_cost_per_image_0.5K", - skip_serializing_if = "Option::is_none" - )] - pub output_cost_per_image_0_5k: Option, - #[serde( - rename = "output_cost_per_image_1K", - skip_serializing_if = "Option::is_none" - )] - pub output_cost_per_image_1k: Option, - #[serde( - rename = "output_cost_per_image_2K", - skip_serializing_if = "Option::is_none" - )] - pub output_cost_per_image_2k: Option, - #[serde( - rename = "output_cost_per_image_4K", - skip_serializing_if = "Option::is_none" - )] - pub output_cost_per_image_4k: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_image_token: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_pixel: Option, - /// USD per reasoning/thinking token, when billed separately. - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_reasoning_token: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_second: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_second_1080p: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_second_2k: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_second_480p: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_second_4k: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_second_720p: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_second_768p: Option, - /// USD per generated token. - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_token: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_token_above_32k_tokens: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_token_above_100k_tokens: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_token_above_100k_tokens_batches: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_token_above_128k_tokens: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_token_above_200k_tokens: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_token_above_200k_tokens_batches: Option, - /// Priority service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_token_above_200k_tokens_priority: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_token_above_256k_tokens: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_token_above_272k_tokens: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_token_above_272k_tokens_batches: Option, - /// Flex service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_token_above_272k_tokens_flex: Option, - /// Priority service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_token_above_272k_tokens_priority: Option, - /// Ultrafast service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_token_above_272k_tokens_ultrafast: Option, - /// Rate applied once the prompt exceeds the token threshold in the field name. - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_token_above_512k_tokens: Option, - /// Balanced service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_token_balanced: Option, - /// USD per generated token via the provider's batch API. - #[serde(skip_serializing_if = "Option::is_none")] + pub input_cost_per_token_above_200k_tokens_batches: Option, + pub input_cost_per_token_above_272k_tokens_batches: Option, pub output_cost_per_token_batches: Option, - /// Flex service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] + pub output_cost_per_token_above_200k_tokens_batches: Option, + pub output_cost_per_token_above_272k_tokens_batches: Option, + pub output_cost_per_token: Option, pub output_cost_per_token_flex: Option, - /// Priority service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] pub output_cost_per_token_priority: Option, - /// Ultrafast service-tier rate for the same-named base field. - #[serde(skip_serializing_if = "Option::is_none")] + pub output_cost_per_token_balanced: Option, pub output_cost_per_token_ultrafast: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_video_per_second: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_video_token: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub output_dbu_cost_per_token: Option, - /// Embedding dimension for embedding models. - #[serde(skip_serializing_if = "Option::is_none")] - pub output_vector_size: Option, - /// Smallest prefix the provider will actually cache; absent means the provider default applies. - #[serde(skip_serializing_if = "Option::is_none")] - pub prompt_cache_min_tokens: Option, - /// Provider-internal routing hints (e.g. bedrock_invocation_schema). - #[serde(skip_serializing_if = "Option::is_none")] - pub provider_specific_entry: Option>, - /// Exact reasoning_effort levels this deployment accepts; wins over supports_* flags. - #[serde(skip_serializing_if = "Option::is_none")] - pub reasoning_effort_levels: Option>, - /// Multiplier applied to all token costs when served from a non-global Vertex AI endpoint (e.g. 1.10 = +10%). - #[serde(skip_serializing_if = "Option::is_none")] - pub regional_endpoint_uplift_multiplier: Option, - /// Multiplier applied to all token costs for EU data residency (e.g. 1.10 = +10%). - #[serde(skip_serializing_if = "Option::is_none")] pub regional_processing_uplift_multiplier_eu: Option, - /// Multiplier applied to all token costs for US data residency (e.g. 1.10 = +10%). - #[serde(skip_serializing_if = "Option::is_none")] pub regional_processing_uplift_multiplier_us: Option, - /// Provider default requests-per-minute limit. - #[serde(skip_serializing_if = "Option::is_none")] - pub rpm: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub rules: Option>, - /// USD cost per web search query, keyed by search context size. - #[serde(skip_serializing_if = "Option::is_none")] - pub search_context_cost_per_query: Option, - /// URL of the provider pricing/model page this entry was taken from. - #[serde(skip_serializing_if = "Option::is_none")] - pub source: Option, - /// Audio container formats the model can return. - #[serde(skip_serializing_if = "Option::is_none")] - pub supported_audio_formats: Option>, - /// OpenAI-style API routes this model can be called through, e.g. /v1/chat/completions. - #[serde(skip_serializing_if = "Option::is_none")] + pub regional_endpoint_uplift_multiplier: Option, + pub output_cost_per_audio_token: Option, + pub output_cost_per_character: Option, + pub output_cost_per_reasoning_token: Option, + pub output_cost_per_reasoning_token_flex: Option, + pub output_cost_per_reasoning_token_priority: Option, + pub output_cost_per_token_above_128k_tokens: Option, + pub output_cost_per_character_above_128k_tokens: Option, + pub output_cost_per_token_above_200k_tokens: Option, + pub output_cost_per_token_above_200k_tokens_priority: Option, + pub output_cost_per_token_above_272k_tokens: Option, + pub output_cost_per_token_above_272k_tokens_priority: Option, + pub output_cost_per_token_above_272k_tokens_flex: Option, + pub output_cost_per_token_above_272k_tokens_ultrafast: Option, + pub output_cost_per_token_above_512k_tokens: Option, + pub output_cost_per_second: Option, + pub output_cost_per_second_1080p: Option, + pub output_cost_per_second_480p: Option, + pub output_cost_per_second_720p: Option, + pub output_cost_per_second_768p: Option, + pub output_cost_per_second_2k: Option, + pub output_cost_per_second_4k: Option, + pub output_cost_per_video_per_second: Option, + pub output_cost_per_image: Option, + pub output_cost_per_pixel: Option, + pub output_cost_per_image_token: Option, + pub output_cost_per_image_token_batches: Option, + pub output_cost_per_video_token: Option, + pub output_vector_size: Option, + pub citation_cost_per_token: Option, + pub tiered_pricing: Option, + pub litellm_provider: Option, + pub mode: Option, pub supported_endpoints: Option>, - /// Input modalities the model accepts. - #[serde(skip_serializing_if = "Option::is_none")] - pub supported_modalities: Option>, - /// Output modalities the model can produce. - #[serde(skip_serializing_if = "Option::is_none")] - pub supported_output_modalities: Option>, - /// Cloud regions the model is available in ('global' or region ids). - #[serde(skip_serializing_if = "Option::is_none")] - pub supported_regions: Option>, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_adaptive_thinking: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_anthropic_compaction: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_anthropic_thinking_payload: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_assistant_prefill: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_audio_input: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_audio_output: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_bedrock_runtime_chat_completions_response_format: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_bedrock_runtime_chat_completions_tools_with_reasoning: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_computer_use: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_embedding_image_input: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_fast_mode: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_forced_tool_use: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_function_calling: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_image_input: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_image_size: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_legacy_thinking: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_low_reasoning_effort: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_max_reasoning_effort: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_mid_conversation_system: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_minimal_reasoning_effort: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_multimodal: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_native_streaming: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_native_structured_output: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_none_reasoning_effort: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_nova_canvas_image_edit: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_output_config: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_parallel_function_calling: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_parallel_tool_use_config: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_pdf_input: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_prompt_cache_breakpoint: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_prompt_caching: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_reasoning: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_response_schema: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_sampling_params: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_speed: Option, - #[serde(skip_serializing_if = "Option::is_none")] pub supports_system_messages: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_thinking_cache_preservation: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_tool_choice: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_tool_search: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_url_context: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_video_input: Option, - #[serde(skip_serializing_if = "Option::is_none")] + pub supports_response_schema: Option, pub supports_vision: Option, - #[serde(skip_serializing_if = "Option::is_none")] + pub supports_function_calling: Option, + pub supports_parallel_function_calling: Option, + pub supports_tool_choice: Option, + pub supports_assistant_prefill: Option, + pub supports_prompt_caching: Option, + pub supports_prompt_cache_breakpoint: Option, + pub supports_thinking_cache_preservation: Option, + pub supports_audio_input: Option, + pub supports_video_input: Option, + pub supports_box_questions: Option, + pub supports_audio_output: Option, + pub supports_pdf_input: Option, + pub supports_embedding_image_input: Option, + pub supports_native_streaming: Option, + pub supports_native_structured_output: Option, + pub supports_bedrock_native_structured_output: Option, pub supports_web_search: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub supports_xhigh_reasoning_effort: Option, - #[serde(skip_serializing_if = "Option::is_none")] + pub supports_url_context: Option, + pub supports_reasoning: Option, + pub supports_adaptive_thinking: Option, + pub supports_legacy_thinking: Option, pub thinking_always_on: Option, - /// Context-length or result-count tiered rates; each tier's costs apply within its range. - #[serde(skip_serializing_if = "Option::is_none")] - pub tiered_pricing: Option>, - /// Provider default tokens-per-minute limit. - #[serde(skip_serializing_if = "Option::is_none")] - pub tpm: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub use_openai_responses_path: Option, - #[serde(skip_serializing_if = "Option::is_none")] + pub supports_tool_search: Option, + pub supports_mid_conversation_system: Option, + pub supports_anthropic_thinking_payload: Option, + pub supports_anthropic_compaction: Option, + pub supports_none_reasoning_effort: Option, + pub supports_minimal_reasoning_effort: Option, + pub supports_low_reasoning_effort: Option, + pub supports_xhigh_reasoning_effort: Option, + pub supports_max_reasoning_effort: Option, + pub reasoning_effort_levels: Option, + pub default_reasoning_effort: Option, + pub bedrock_output_config_effort_ceiling: Option, + pub bedrock_converse_supports_strict_tools: Option, + pub supports_regex_lookaround: Option, + pub supports_computer_use: Option, + pub search_context_cost_per_query: Option, + pub web_search_billing_unit: Option, + pub search_free_requests: Option, + pub google_maps_grounding_cost_per_query: Option, + pub tpm: Option, + pub rpm: Option, + pub ocr_cost_per_page: Option, + pub ocr_cost_per_page_batches: Option, + pub ocr_cost_per_credit: Option, + pub annotation_cost_per_page: Option, + pub annotation_cost_per_page_batches: Option, + pub provider_specific_entry: Option, pub uses_embed_content: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub vector_store_cost_per_gb_per_day: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub vertex_ai_audio_api: Option, - /// Whether web search is billed per query or per prompt. - #[serde(skip_serializing_if = "Option::is_none")] - pub web_search_billing_unit: Option, + pub supports_image_size: Option, + pub supported_audio_formats: Option, + pub token_count_framing: Option, + pub token_count_model: Option, + pub token_count_tokenizer: Option, + pub vertex_ai_audio_api: Option, + /// Every key matching `ABOVE_THRESHOLD_COST_KEY_PATTERN`, copied as written + pub above_threshold_costs: IndexMap, } diff --git a/litellm-rust/crates/model-catalog/src/pricing.rs b/litellm-rust/crates/model-catalog/src/pricing.rs deleted file mode 100644 index b8ed652451c..00000000000 --- a/litellm-rust/crates/model-catalog/src/pricing.rs +++ /dev/null @@ -1,97 +0,0 @@ -use serde::{Deserialize, Serialize}; - -/// Whether web search is billed per query or per prompt. -#[derive(Clone, Copy, Debug, Deserialize, Eq, PartialEq, Serialize)] -#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))] -#[serde(rename_all = "snake_case")] -pub enum WebSearchBillingUnit { - PerQuery, - PerPrompt, -} - -/// UTC "HH:MM-HH:MM" window, or a list of them; a window may wrap past midnight. -#[derive(Clone, Debug, Deserialize, PartialEq, Serialize)] -#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))] -#[serde(untagged)] -pub enum UtcHours { - Single(String), - Multiple(Vec), -} - -/// ISO-8601 weekday number (1 = Monday .. 7 = Sunday) or English day name. -#[derive(Clone, Debug, Deserialize, PartialEq, Serialize)] -#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))] -#[serde(untagged)] -pub enum Weekday { - Number(u8), - Name(String), -} - -/// One off-peak window entry inside `windows`. -#[derive(Clone, Debug, Deserialize, PartialEq, Serialize)] -#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))] -#[serde(deny_unknown_fields)] -pub struct OffPeakWindow { - pub hours_utc: UtcHours, - #[serde(skip_serializing_if = "Option::is_none")] - pub weekdays: Option>, -} - -/// Rates that replace the same-named base fields inside the stated UTC windows. -#[derive(Clone, Debug, Default, Deserialize, PartialEq, Serialize)] -#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))] -#[serde(deny_unknown_fields)] -pub struct OffPeakPricing { - #[serde(skip_serializing_if = "Option::is_none")] - pub hours_utc: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub windows: Option>, - #[serde(skip_serializing_if = "Option::is_none")] - pub weekday_timezone: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_token: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_token: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_reasoning_token: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_read_input_token_cost: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_creation_input_token_cost: Option, -} - -/// USD cost per web search query, keyed by search context size. -#[derive(Clone, Debug, Default, Deserialize, PartialEq, Serialize)] -#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))] -#[serde(deny_unknown_fields)] -pub struct SearchContextCostPerQuery { - #[serde(skip_serializing_if = "Option::is_none")] - pub search_context_size_low: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub search_context_size_medium: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub search_context_size_high: Option, -} - -/// One tier of a context-length or result-count tiered rate. -#[derive(Clone, Debug, Default, Deserialize, PartialEq, Serialize)] -#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))] -#[serde(deny_unknown_fields)] -pub struct TieredRate { - #[serde(skip_serializing_if = "Option::is_none")] - pub range: Option<[f64; 2]>, - #[serde(skip_serializing_if = "Option::is_none")] - pub max_results_range: Option<[f64; 2]>, - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_token: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_token: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub output_cost_per_reasoning_token: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_read_input_token_cost: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub cache_creation_input_token_cost: Option, - #[serde(skip_serializing_if = "Option::is_none")] - pub input_cost_per_query: Option, -} diff --git a/litellm-rust/crates/model-catalog/src/register.rs b/litellm-rust/crates/model-catalog/src/register.rs new file mode 100644 index 00000000000..ed34082ae57 --- /dev/null +++ b/litellm-rust/crates/model-catalog/src/register.rs @@ -0,0 +1,26 @@ +use std::collections::{BTreeSet, HashMap}; + +use indexmap::IndexMap; + +use crate::wire::Fields; + +/// Where one `register` call writes and which row it inherits from +#[derive(Clone, Debug, PartialEq, Eq)] +pub struct RegistrationTarget { + /// Always the caller's key, never a case variant of it + pub write: String, + /// The first candidate that is an exact key whose row passes `provider_matches` + pub inherit_from: Option, +} + +/// Python's inheritance candidates in main's order, `get_model_info`'s then `_resolve_builtin_model_cost_entry`'s +#[derive(Clone, Debug, PartialEq, Eq)] +pub struct RegistrationCandidates(pub Vec); + +/// Layer 3, and the fields `register` last wrote, which litellm owns again +#[derive(Clone, Debug, Default)] +#[expect(dead_code, reason = "07 adds the functions that read registrations")] +pub struct Registrations { + persisted: IndexMap, + registered_fields: HashMap>, +} diff --git a/litellm-rust/crates/model-catalog/src/resolve/image.rs b/litellm-rust/crates/model-catalog/src/resolve/image.rs new file mode 100644 index 00000000000..d24e170b32c --- /dev/null +++ b/litellm-rust/crates/model-catalog/src/resolve/image.rs @@ -0,0 +1,18 @@ +/// What an image call asks for. `resolve_image` builds main's eight key spellings from it +#[derive(Clone, Debug, PartialEq, Eq)] +pub struct ImageCandidates { + pub model: String, + pub provider: Option, + /// Width and height, `1024 x 1024` when the request names none, as on main + pub size: Option<(u32, u32)>, + pub quality: Option, +} + +/// A size and quality keyed row, `{quality}/{W}-x-{H}/{model}`, parsed once per index build for fal's nearest match +#[derive(Clone, Debug, PartialEq, Eq)] +pub struct ImageKey { + pub quality: Option, + pub width: u32, + pub height: u32, + pub model: String, +} diff --git a/litellm-rust/crates/model-catalog/src/resolve/mod.rs b/litellm-rust/crates/model-catalog/src/resolve/mod.rs new file mode 100644 index 00000000000..fbaead622aa --- /dev/null +++ b/litellm-rust/crates/model-catalog/src/resolve/mod.rs @@ -0,0 +1,46 @@ +mod image; + +pub use image::*; + +/// The seven names main's `get_model_info` tries, built by Python, walked in this order by Rust +#[derive(Clone, Debug, Default, PartialEq, Eq)] +pub struct Candidates { + pub combined_model_name: Option, + pub model: String, + pub region_free_combined_model_name: Option, + pub split_model: Option, + pub combined_stripped_model_name: Option, + pub stripped_model_name: Option, + pub provider_prefixed_model_name: Option, +} + +#[derive(Clone, Copy, Debug, PartialEq, Eq)] +pub enum CandidateSlot { + CombinedModelName, + Model, + RegionFreeCombinedModelName, + SplitModel, + CombinedStrippedModelName, + StrippedModelName, + ProviderPrefixedModelName, +} + +/// Which step matched, so a test can name the step that changed +#[derive(Clone, Copy, Debug, PartialEq, Eq)] +pub struct Via { + pub slot: CandidateSlot, + /// Matched through the case-insensitive index rather than an exact key + pub case_insensitive: bool, +} + +#[derive(Clone, Debug, PartialEq, Eq)] +pub enum Resolution { + Found { + key: String, + via: Via, + }, + /// Every candidate tried, in order + Missing { + tried: Vec, + }, +} diff --git a/litellm-rust/crates/model-catalog/src/schema.rs b/litellm-rust/crates/model-catalog/src/schema.rs index 7de988b3ff3..66f9933a3e3 100644 --- a/litellm-rust/crates/model-catalog/src/schema.rs +++ b/litellm-rust/crates/model-catalog/src/schema.rs @@ -1,9 +1,80 @@ -use schemars::Schema; +use std::collections::BTreeMap; + +use litellm_cost::money::{Multiplier, UnitPrice}; +use schemars::{JsonSchema, Schema}; use serde_json::{Map, Value, json}; +use crate::fallback::FallbackGeneralizations; +use crate::wire::{Attributes, OffPeakWindow, SearchContextCostPerQuery, UtcHours}; + +#[derive(JsonSchema)] +#[schemars(rename = "ModelEntry")] +struct ModelEntry { + #[serde(flatten)] + _attributes: Attributes, + #[serde(flatten)] + _blocks: PriceBlocks, +} + +#[derive(JsonSchema)] +#[allow(dead_code)] +struct PriceBlocks { + /// USD cost per billable guardrail unit, keyed by the provider's usage counter name (e.g. Bedrock's contentPolicyUnits). + guardrail_cost_per_unit: Option>, + /// Rates that replace the same-named base fields while the request falls inside the stated UTC windows. + off_peak_pricing: Option, + /// Multiplier applied to all token costs when served from a non-global Vertex AI endpoint (e.g. 1.10 = +10%). + regional_endpoint_uplift_multiplier: Option, + /// Multiplier applied to all token costs for EU data residency (e.g. 1.10 = +10%). + regional_processing_uplift_multiplier_eu: Option, + /// Multiplier applied to all token costs for US data residency (e.g. 1.10 = +10%). + regional_processing_uplift_multiplier_us: Option, + /// USD cost per web search query, keyed by search context size. + search_context_cost_per_query: Option, + /// Context-length or result-count tiered rates; each tier's costs apply within its range. + tiered_pricing: Option>, +} + +/// Rates that replace the same-named base fields inside the stated UTC windows. +#[derive(JsonSchema)] +#[allow(dead_code)] +struct OffPeakPricing { + hours_utc: Option, + windows: Option>, + weekday_timezone: Option, +} + +/// One tier of a context-length or result-count tiered rate. +#[derive(JsonSchema)] +#[allow(dead_code)] +struct TieredRate { + range: Option<[u64; 2]>, + max_results_range: Option<[u64; 2]>, +} + +#[derive(JsonSchema)] +#[allow(dead_code)] +struct RegistryDocument { + sample_spec: Option>, + fallback_generalizations: Option, +} + +pub(crate) const FLAT_KEY_PATTERN: &str = concat!( + r"^(", + r"input_cost_per_((audio_|image_|video_)?token|character|image|pixel|query|request|second|text_message|audio_per_second|video_per_second(_above_\d+s_interval)?)", + r"|output_cost_per_((audio_|image_|video_|reasoning_)?token|character|image(_\d+(\.\d+)?K?)?|pixel|second(_\d+[pk])?|video_per_second)", + r"|cache_read_input_(audio_|image_|video_)?token_cost", + r"|explicit_cache_read_input_token_cost", + r"|cache_creation_input_(audio_|image_|video_)?token_cost(_above_1hr|_above_12hr)?", + r"|annotation_cost_per_page|citation_cost_per_token|code_interpreter_cost_per_session", + r"|google_maps_grounding_cost_per_query|ocr_cost_per_credit|ocr_cost_per_page|cost_per_second", + r"|browser_open_cost_per_request|cost_per_additional_result", + r")(_above_\d+k?_tokens)?(_(priority|flex|ultrafast|balanced|batches|decisions))?$", +); + /// JSON Schema for one model entry, including registry validation constraints. pub fn model_entry_json_schema() -> Schema { - let mut schema = serde_json::to_value(schemars::schema_for!(crate::ModelInfo)) + let mut schema = serde_json::to_value(schemars::schema_for!(ModelEntry)) .expect("derived model schema serializes"); remove_nullable_optional_fields(&mut schema); decorate_model_entry(&mut schema); @@ -22,24 +93,21 @@ pub fn registry_json_schema() -> Schema { entry.as_object_mut().unwrap().remove("$schema"); definitions.insert("modelEntry".into(), entry); - let mut fallback = serde_json::to_value(schemars::schema_for!(crate::FallbackGeneralizations)) - .expect("derived fallback schema serializes"); - remove_nullable_optional_fields(&mut fallback); - definitions.extend(take_definitions(&mut fallback)); - fallback.as_object_mut().unwrap().remove("$schema"); - - let root = json!({ - "$schema": "https://json-schema.org/draft/2020-12/schema", - "title": "LiteLLM model prices and context window registry", - "type": "object", - "properties": { - "sample_spec": {"type": "object"}, - "fallback_generalizations": fallback - }, - "additionalProperties": {"$ref": "#/$defs/modelEntry"}, - "$defs": definitions - }); - Schema::from(root.as_object().unwrap().clone()) + let mut registry = serde_json::to_value(schemars::schema_for!(RegistryDocument)) + .expect("derived registry schema serializes"); + remove_nullable_optional_fields(&mut registry); + definitions.extend(take_definitions(&mut registry)); + let root = registry.as_object_mut().unwrap(); + root.insert( + "title".into(), + "LiteLLM model prices and context window registry".into(), + ); + root.insert( + "additionalProperties".into(), + json!({"$ref": "#/$defs/modelEntry"}), + ); + root.insert("$defs".into(), Value::Object(definitions)); + Schema::from(root.clone()) } fn take_definitions(schema: &mut Value) -> Map { @@ -80,42 +148,21 @@ fn remove_nullable_optional_fields(value: &mut Value) { } fn decorate_model_entry(schema: &mut Value) { + let rates = json!({FLAT_KEY_PATTERN: {"type": "number", "minimum": 0}}); let object = schema.as_object_mut().unwrap(); object.insert("required".into(), json!(["litellm_provider"])); object.insert("additionalProperties".into(), Value::Bool(true)); - let properties = object - .get_mut("properties") - .unwrap() - .as_object_mut() - .unwrap(); - properties.insert( - "aliases".into(), - json!({"type": "array", "items": {"type": "string"}}), + object.insert( + "patternProperties".into(), + json!({ + FLAT_KEY_PATTERN: {"type": "number", "minimum": 0}, + "^supports_": {"type": "boolean"}, + }), ); - properties.get_mut("deprecation_date").unwrap()["format"] = json!("date"); - properties.get_mut("deprecation_date").unwrap()["pattern"] = - json!(r"^\d{4}-(0[1-9]|1[0-2])-(0[1-9]|[12]\d|3[01])$"); - - properties.iter_mut().for_each(|(name, property)| { - if name.contains("cost") { - property["minimum"] = json!(0); - } else if name.contains("uplift_multiplier") { - property["minimum"] = json!(1); - } - }); - properties.get_mut("guardrail_cost_per_unit").unwrap()["additionalProperties"]["minimum"] = - json!(0); - let definitions = object.get_mut("$defs").unwrap().as_object_mut().unwrap(); - for definition in ["OffPeakPricing", "TieredRate", "SearchContextCostPerQuery"] { - let properties = definitions[definition]["properties"] - .as_object_mut() - .unwrap(); - properties.iter_mut().for_each(|(name, property)| { - if name.contains("cost") || definition == "SearchContextCostPerQuery" { - property["minimum"] = json!(0); - } - }); + for definition in ["OffPeakPricing", "TieredRate"] { + definitions[definition]["patternProperties"] = rates.clone(); + definitions[definition]["additionalProperties"] = Value::Bool(false); } definitions["OffPeakPricing"]["anyOf"] = json!([ {"required": ["hours_utc"]}, @@ -123,15 +170,43 @@ fn decorate_model_entry(schema: &mut Value) { ]); definitions["OffPeakPricing"]["properties"]["windows"]["minItems"] = json!(1); definitions["OffPeakWindow"]["properties"]["weekdays"]["minItems"] = json!(1); - definitions["TieredRate"]["properties"]["range"]["items"]["minimum"] = json!(0); - definitions["TieredRate"]["properties"]["max_results_range"]["items"]["minimum"] = json!(0); - definitions["Weekday"]["anyOf"][0]["minimum"] = json!(1); - definitions["Weekday"]["anyOf"][0]["maximum"] = json!(7); - definitions["Weekday"]["anyOf"][1]["pattern"] = json!( - r"(?i)^(mon|monday|tue|tues|tuesday|wed|wednesday|thu|thur|thurs|thursday|fri|friday|sat|saturday|sun|sunday)$" - ); - let window_pattern = json!(r"^([01]\d|2[0-3]):[0-5]\d-([01]\d|2[0-3]):[0-5]\d$"); - definitions["UtcHours"]["anyOf"][0]["pattern"] = window_pattern.clone(); - definitions["UtcHours"]["anyOf"][1]["items"]["pattern"] = window_pattern; definitions["UtcHours"]["anyOf"][1]["minItems"] = json!(1); } + +#[cfg(test)] +mod tests { + use rstest::rstest; + + use super::FLAT_KEY_PATTERN; + use crate::wire::FlatKey; + + #[rstest] + #[case::input_token("input_cost_per_token")] + #[case::cache_write_one_hour_band( + "cache_creation_input_token_cost_above_1hr_above_200k_tokens" + )] + #[case::image_label("output_cost_per_image_0.5K")] + #[case::resolution_tier("output_cost_per_second_4k_flex")] + #[case::feature_batch("annotation_cost_per_page_batches")] + #[case::video_interval("input_cost_per_video_per_second_above_8s_interval")] + #[case::explicit_cache_read("explicit_cache_read_input_token_cost")] + #[case::wall_clock("cost_per_second")] + #[case::unused_meter("cost_per_token")] + #[case::other_currency("input_dbu_cost_per_token")] + #[case::typo("input_cost_per_tokne")] + #[case::read_has_no_ttl("cache_read_input_token_cost_above_1hr")] + #[case::interval_on_a_token_meter("input_cost_per_token_above_8s_interval")] + fn flat_key_pattern_agrees_with_the_parser(#[case] key: &str) { + let schema = serde_json::json!({ + "type": "object", + "patternProperties": {FLAT_KEY_PATTERN: {}}, + "additionalProperties": false, + }); + let validator = jsonschema::validator_for(&schema).unwrap(); + + assert_eq!( + validator.is_valid(&serde_json::json!({key: 1})), + FlatKey::parse(key).is_some() + ); + } +} diff --git a/litellm-rust/crates/model-catalog/src/store/AGENTS.md b/litellm-rust/crates/model-catalog/src/store/AGENTS.md new file mode 100644 index 00000000000..f1dacd2e217 --- /dev/null +++ b/litellm-rust/crates/model-catalog/src/store/AGENTS.md @@ -0,0 +1,35 @@ +# Store + +`Store` is an in-memory store, a small database built for exactly three queries. It knows nothing about the catalog. A source picks its own `Rows::Row` type, how names fold for matching, and the group each key is listed under. The catalog's sources fold to lowercase and group by raw `litellm_provider` + +| Query | Method | Real caller | +|---|---|---| +| row by exact key | `Store::get` | reading `litellm.model_cost["gpt-4o"]` | +| actual key for a name that folds to the same form | `Store::find_key` | `get_model_info("GPT-4o")` finding `gpt-4o` | +| keys in a group | `Store::keys_in` | the provider sets, `litellm.open_ai_chat_completion_models` | + +## Rules + +- nothing in `store/` imports from the rest of the crate. Wire types, case rules and field names such as `litellm_provider` belong to the source that implements `Rows` +- the index in `name_index.rs` is private to the store. Never build a folded-name map or a provider grouping anywhere else in the workspace. A new lookup pattern is a new `Store` method backed by `NameIndex` +- the index holds keys, never rows. Rows are always read from `Rows`, so a price edit is seen on the next read and never rebuilds the index +- `find_key` and `keys_in` never return a key the rows lack. A hit on a removed key rebuilds the index once and asks again +- `keys_in` matches the group exactly as the source wrote it. Provider aliases such as `vertex_ai` over `vertex_ai-language-models` are Python's job +- when two keys fold to the same name the later key in row order wins, like main's `_model_cost_lowercase_map` + +## Stamps + +`Rows::stamp` is how the store learns that the index may be stale. The store keeps the stamp it built the index at and rebuilds when the source returns a different one. A source must change its stamp whenever a key or a key's group changes, since those are the two things the index holds + +| Source | Stamp | Blind spot | +|---|---|---| +| `Catalog`'s `CatalogRows` | `()`, every change goes through `Store::edit`, which drops the index | none | +| python-bridge's bound `litellm.model_cost` | the dict object and its `len` | a delete plus an add between two reads, or a `litellm_provider` rewrite on an existing key | + +A source with a blind spot calls `Store::invalidate` when it knows keys changed. Until then a key added in the blind spot is not found, and a key moved to another provider stays listed under the old one. Removed keys are still never returned + +## Tests + +Tests live in `mod.rs`'s `tests` module and drive the store through `ModelCost`, a fake `litellm.model_cost` with real bundled keys whose row is just its `litellm_provider`. Each Python write is a method (`set`, `del`, `rebind`) and its stamp is what python-bridge can see. Add a scenario as a named test that reads like the Python user code it stands for + +The two property tests compare every answer with a linear scan of the same dict. With a stamp that sees every key change, answers must match exactly. With the dict stamp, every answer must be sound (present, same folded name), and `invalidate` must make them match again. A store change that fails either is a bug in the store, not in the test diff --git a/litellm-rust/crates/model-catalog/src/store/mod.rs b/litellm-rust/crates/model-catalog/src/store/mod.rs new file mode 100644 index 00000000000..2220526f8b6 --- /dev/null +++ b/litellm-rust/crates/model-catalog/src/store/mod.rs @@ -0,0 +1,460 @@ +mod name_index; + +use std::borrow::Cow; +use std::fmt; +use std::sync::{PoisonError, RwLock}; + +use name_index::NameIndex; + +/// Where a `Store`'s rows live. The store rebuilds its index whenever `stamp` changes, so a source must change its +/// stamp whenever a key or a key's group changes, or call `Store::invalidate` when it cannot +pub trait Rows { + type Row: ToOwned + ?Sized; + type Stamp: PartialEq; + + /// The name two spellings of one key share, which `Store::find_key` matches on + fn fold(name: &str) -> String; + fn stamp(&self) -> Self::Stamp; + fn contains(&self, key: &str) -> bool; + fn row(&self, key: &str) -> Option>; + /// Every key in row order with the group it is listed under + fn scan(&self, visit: &mut dyn FnMut(&str, Option<&str>)); +} + +/// An in-memory store for three queries (row by exact key, actual key for a name by its folded form, keys in a group) +/// that owns the only index behind them +pub struct Store { + rows: R, + index: RwLock>>, +} + +struct Built { + stamp: S, + names: NameIndex, +} + +impl Store { + pub fn new(rows: R) -> Self { + Self { + rows, + index: RwLock::new(None), + } + } + + pub fn rows(&self) -> &R { + &self.rows + } + + /// Changes the rows in place. The index is rebuilt on the next query + pub fn edit(&mut self, edit: impl FnOnce(&mut R) -> T) -> T { + let edited = edit(&mut self.rows); + *self.index.get_mut().unwrap_or_else(PoisonError::into_inner) = None; + edited + } + + pub fn get(&self, key: &str) -> Option> { + self.rows.row(key) + } + + /// The key itself when present, otherwise the actual key that folds to the same name. Never a key the rows lack + pub fn find_key(&self, name: &str) -> Option { + if self.rows.contains(name) { + return Some(name.to_owned()); + } + let folded = R::fold(name); + let hit = self.query(|names| names.key_for(&folded).map(str::to_owned))?; + if self.rows.contains(&hit) { + return Some(hit); + } + self.invalidate(); + self.query(|names| names.key_for(&folded).map(str::to_owned)) + .filter(|key| self.rows.contains(key)) + } + + /// Keys whose group is exactly `group`, in row order. Never a key the rows lack + pub fn keys_in(&self, group: &str) -> Vec { + self.query(|names| names.keys_in(group).to_vec()) + .into_iter() + .filter(|key| self.rows.contains(key)) + .collect() + } + + /// Drops the index, for a source whose stamp cannot see every key change + pub fn invalidate(&self) { + *self.index.write().unwrap_or_else(PoisonError::into_inner) = None; + } + + fn query(&self, read: impl Fn(&NameIndex) -> T) -> T { + let stamp = self.rows.stamp(); + if let Some(built) = self + .index + .read() + .unwrap_or_else(PoisonError::into_inner) + .as_ref() + .filter(|built| built.stamp == stamp) + { + return read(&built.names); + } + let names = NameIndex::scan(&self.rows); + let answer = read(&names); + *self.index.write().unwrap_or_else(PoisonError::into_inner) = Some(Built { stamp, names }); + answer + } +} + +impl Clone for Store { + fn clone(&self) -> Self { + Self::new(self.rows.clone()) + } +} + +impl fmt::Debug for Store { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("Store") + .field("rows", &self.rows) + .finish_non_exhaustive() + } +} + +#[cfg(test)] +mod tests { + use std::cell::{Cell, RefCell}; + + use indexmap::IndexMap; + use rstest::{fixture, rstest}; + + use super::*; + + const BUNDLED: &[(&str, Option<&str>)] = &[ + ("gpt-4o", Some("openai")), + ("o3-pro", Some("openai")), + ("azure/gpt-4o", Some("azure")), + ("claude-sonnet-4-5", Some("anthropic")), + ("gemini-2.5-pro", Some("vertex_ai-language-models")), + ( + "fireworks_ai/accounts/fireworks/models/llama-v3p1-8b-instruct", + Some("fireworks_ai"), + ), + ( + "anyscale/meta-llama/Meta-Llama-3-8B-Instruct", + Some("anyscale"), + ), + ("sample_free_row", None), + ]; + + /// `litellm.model_cost` as Python sees it. The stamp is what python-bridge can observe for free, the dict's + /// identity and its length, so a write that keeps both hides a key change from the store + struct ModelCost { + dict: RefCell>>, + identity: Cell, + index_changes: Cell, + honest_stamp: bool, + scans: Cell, + } + + impl ModelCost { + fn new(rows: &[(&str, Option<&str>)], honest_stamp: bool) -> Self { + Self { + dict: RefCell::new( + rows.iter() + .map(|(key, provider)| ((*key).to_owned(), provider.map(str::to_owned))) + .collect(), + ), + identity: Cell::new(0), + index_changes: Cell::new(0), + honest_stamp, + scans: Cell::new(0), + } + } + + /// `model_cost[key] = {"litellm_provider": provider, ...}` + fn set(&self, key: &str, provider: Option<&str>) { + let previous = self + .dict + .borrow_mut() + .insert(key.to_owned(), provider.map(str::to_owned)); + if previous.as_ref().map(Option::as_deref) != Some(provider) { + self.index_changes.set(self.index_changes.get() + 1); + } + } + + /// `del model_cost[key]` + fn del(&self, key: &str) { + if self.dict.borrow_mut().shift_remove(key).is_some() { + self.index_changes.set(self.index_changes.get() + 1); + } + } + + /// `litellm.model_cost = {...}`, a new dict object + fn rebind(&self, rows: &[(&str, Option<&str>)]) { + *self.dict.borrow_mut() = rows + .iter() + .map(|(key, provider)| ((*key).to_owned(), provider.map(str::to_owned))) + .collect(); + self.identity.set(self.identity.get() + 1); + self.index_changes.set(self.index_changes.get() + 1); + } + } + + /// A row is its `litellm_provider`, which is also the group the store lists its key under + impl Rows for &ModelCost { + type Row = Option; + type Stamp = (u64, u64); + + fn fold(name: &str) -> String { + name.to_lowercase() + } + + fn stamp(&self) -> Self::Stamp { + if self.honest_stamp { + (self.identity.get(), self.index_changes.get()) + } else { + (self.identity.get(), self.dict.borrow().len() as u64) + } + } + + fn contains(&self, key: &str) -> bool { + self.dict.borrow().contains_key(key) + } + + fn row(&self, key: &str) -> Option>> { + Some(Cow::Owned(self.dict.borrow().get(key)?.clone())) + } + + fn scan(&self, visit: &mut dyn FnMut(&str, Option<&str>)) { + self.scans.set(self.scans.get() + 1); + for (key, provider) in self.dict.borrow().iter() { + visit(key, provider.as_deref()); + } + } + } + + #[fixture] + fn model_cost() -> ModelCost { + ModelCost::new(BUNDLED, false) + } + + /// Rows only `Store::edit` can change, so the stamp never moves + #[derive(Debug)] + struct Owned(IndexMap>); + + impl Rows for Owned { + type Row = Option; + type Stamp = (); + + fn fold(name: &str) -> String { + name.to_lowercase() + } + + fn stamp(&self) -> Self::Stamp {} + + fn contains(&self, key: &str) -> bool { + self.0.contains_key(key) + } + + fn row(&self, key: &str) -> Option>> { + self.0.get(key).map(Cow::Borrowed) + } + + fn scan(&self, visit: &mut dyn FnMut(&str, Option<&str>)) { + for (key, provider) in &self.0 { + visit(key, provider.as_deref()); + } + } + } + + fn owned(rows: &[(&str, Option<&str>)]) -> Store { + Store::new(Owned( + rows.iter() + .map(|(key, provider)| ((*key).to_owned(), provider.map(str::to_owned))) + .collect(), + )) + } + + #[rstest] + #[case::bundled_key_as_written("gpt-4o", Some("gpt-4o"))] + #[case::user_typed_upper_case("GPT-4o", Some("gpt-4o"))] + #[case::provider_prefixed("Azure/GPT-4o", Some("azure/gpt-4o"))] + #[case::fireworks_path_in_another_case( + "fireworks_ai/accounts/fireworks/models/Llama-v3p1-8B-Instruct", + Some("fireworks_ai/accounts/fireworks/models/llama-v3p1-8b-instruct") + )] + #[case::mixed_case_bundled_key( + "anyscale/meta-llama/meta-llama-3-8b-instruct", + Some("anyscale/meta-llama/Meta-Llama-3-8B-Instruct") + )] + #[case::unmapped_model("my-finetune", None)] + #[case::prefix_is_not_a_match("gpt-4", None)] + fn find_key_maps_a_requested_model_to_its_key( + model_cost: ModelCost, + #[case] model: &str, + #[case] expected: Option<&str>, + ) { + assert_eq!(Store::new(&model_cost).find_key(model).as_deref(), expected); + } + + #[rstest] + #[case::the_bundled_spelling("gpt-4o", "gpt-4o")] + #[case::the_registered_spelling("GPT-4o", "GPT-4o")] + #[case::another_spelling_takes_the_later_key("Gpt-4O", "GPT-4o")] + fn a_registered_case_variant_is_its_own_row( + model_cost: ModelCost, + #[case] model: &str, + #[case] expected: &str, + ) { + model_cost.set("GPT-4o", Some("openai")); + assert_eq!( + Store::new(&model_cost).find_key(model).as_deref(), + Some(expected) + ); + } + + #[rstest] + #[case::openai_in_map_order("openai", &["gpt-4o", "o3-pro"])] + #[case::azure("azure", &["azure/gpt-4o"])] + #[case::raw_vertex_family("vertex_ai-language-models", &["gemini-2.5-pro"])] + #[case::provider_alias_is_python_s_job("vertex_ai", &[])] + #[case::matched_as_written("OpenAI", &[])] + #[case::unknown_provider("acme", &[])] + fn keys_in_lists_the_keys_a_raw_provider_names( + model_cost: ModelCost, + #[case] provider: &str, + #[case] expected: &[&str], + ) { + assert_eq!(Store::new(&model_cost).keys_in(provider), expected); + } + + #[rstest] + fn a_model_written_straight_into_the_dict_is_found_in_any_case(model_cost: ModelCost) { + let store = Store::new(&model_cost); + assert_eq!(store.find_key("mycorp/custom-model"), None); + + model_cost.set("MyCorp/Custom-Model", Some("openai")); + + assert_eq!( + store.find_key("mycorp/custom-model").as_deref(), + Some("MyCorp/Custom-Model") + ); + assert_eq!( + store.keys_in("openai"), + ["gpt-4o", "o3-pro", "MyCorp/Custom-Model"] + ); + } + + #[rstest] + fn a_price_edit_reuses_the_index(model_cost: ModelCost) { + let store = Store::new(&model_cost); + assert_eq!(store.find_key("GPT-4o").as_deref(), Some("gpt-4o")); + + model_cost.set("gpt-4o", Some("openai")); + model_cost.set("o3-pro", Some("openai")); + model_cost.set("gpt-4o", Some("openai")); + + assert_eq!(store.find_key("O3-PRO").as_deref(), Some("o3-pro")); + assert_eq!(store.keys_in("openai"), ["gpt-4o", "o3-pro"]); + assert_eq!(model_cost.scans.get(), 1); + } + + #[rstest] + fn a_reload_that_rebinds_the_dict_is_seen_even_at_the_same_length(model_cost: ModelCost) { + let store = Store::new(&model_cost); + assert_eq!(store.find_key("GPT-4o").as_deref(), Some("gpt-4o")); + + let mut reloaded = BUNDLED.to_vec(); + reloaded[0] = ("gpt-4o-2024-11-20", Some("openai")); + model_cost.rebind(&reloaded); + + assert_eq!(store.find_key("GPT-4o"), None); + assert_eq!( + store.find_key("GPT-4o-2024-11-20").as_deref(), + Some("gpt-4o-2024-11-20") + ); + } + + #[rstest] + fn a_deleted_model_is_never_returned(model_cost: ModelCost) { + let store = Store::new(&model_cost); + assert_eq!(store.find_key("O3-PRO").as_deref(), Some("o3-pro")); + + model_cost.del("o3-pro"); + + assert_eq!(store.find_key("O3-PRO"), None); + assert_eq!(store.keys_in("openai"), ["gpt-4o"]); + } + + #[rstest] + fn a_respelled_key_at_the_same_length_is_found_after_one_rebuild(model_cost: ModelCost) { + let store = Store::new(&model_cost); + assert_eq!(store.find_key("o3-PRO").as_deref(), Some("o3-pro")); + + model_cost.del("o3-pro"); + model_cost.set("O3-Pro", Some("openai")); + + assert_eq!(store.find_key("o3-PRO").as_deref(), Some("O3-Pro")); + assert_eq!(model_cost.scans.get(), 2); + } + + #[rstest] + fn a_swap_at_the_same_length_hides_the_new_key_until_invalidate(model_cost: ModelCost) { + let store = Store::new(&model_cost); + assert_eq!(store.find_key("O3-PRO").as_deref(), Some("o3-pro")); + + model_cost.del("o3-pro"); + model_cost.set("MyCorp/Custom-Model", Some("openai")); + + assert_eq!(store.find_key("mycorp/custom-model"), None); + assert_eq!(store.keys_in("openai"), ["gpt-4o"]); + + store.invalidate(); + + assert_eq!( + store.find_key("mycorp/custom-model").as_deref(), + Some("MyCorp/Custom-Model") + ); + assert_eq!(store.keys_in("openai"), ["gpt-4o", "MyCorp/Custom-Model"]); + } + + #[rstest] + fn a_provider_rewrite_on_an_existing_key_needs_invalidate(model_cost: ModelCost) { + let store = Store::new(&model_cost); + assert_eq!(store.keys_in("openai"), ["gpt-4o", "o3-pro"]); + + model_cost.set("o3-pro", Some("azure")); + + assert_eq!(store.keys_in("openai"), ["gpt-4o", "o3-pro"]); + + store.invalidate(); + + assert_eq!(store.keys_in("openai"), ["gpt-4o"]); + assert_eq!(store.keys_in("azure"), ["o3-pro", "azure/gpt-4o"]); + } + + #[rstest] + fn an_edit_on_owned_rows_is_seen_by_the_next_query() { + let mut store = owned(BUNDLED); + assert_eq!(store.find_key("GPT-4o").as_deref(), Some("gpt-4o")); + + store.edit(|rows| { + rows.0.shift_remove("gpt-4o"); + rows.0 + .insert("MyCorp/Custom-Model".to_owned(), Some("openai".to_owned())); + }); + + assert_eq!(store.find_key("GPT-4o"), None); + assert_eq!( + store.find_key("mycorp/custom-model").as_deref(), + Some("MyCorp/Custom-Model") + ); + assert_eq!(store.keys_in("openai"), ["o3-pro", "MyCorp/Custom-Model"]); + } + + #[rstest] + fn get_reads_the_row_by_exact_key_only() { + let store = owned(BUNDLED); + assert_eq!( + store.get("azure/gpt-4o").as_deref(), + Some(&Some("azure".to_owned())) + ); + assert_eq!(store.get("Azure/GPT-4o"), None); + } +} diff --git a/litellm-rust/crates/model-catalog/src/store/name_index.rs b/litellm-rust/crates/model-catalog/src/store/name_index.rs new file mode 100644 index 00000000000..86f05cb31af --- /dev/null +++ b/litellm-rust/crates/model-catalog/src/store/name_index.rs @@ -0,0 +1,35 @@ +use std::collections::HashMap; + +use super::Rows; + +/// Keys only, never rows, so a value write never makes it stale. A later key wins when two keys fold to one name +#[derive(Debug, Default)] +pub(super) struct NameIndex { + by_folded: HashMap, + by_group: HashMap>, +} + +impl NameIndex { + pub(super) fn scan(rows: &R) -> Self { + let mut index = Self::default(); + rows.scan(&mut |key, group| { + index.by_folded.insert(R::fold(key), key.to_owned()); + if let Some(group) = group { + index + .by_group + .entry(group.to_owned()) + .or_default() + .push(key.to_owned()); + } + }); + index + } + + pub(super) fn key_for(&self, folded: &str) -> Option<&str> { + self.by_folded.get(folded).map(String::as_str) + } + + pub(super) fn keys_in(&self, group: &str) -> &[String] { + self.by_group.get(group).map_or(&[], Vec::as_slice) + } +} diff --git a/litellm-rust/crates/model-catalog/src/validation.rs b/litellm-rust/crates/model-catalog/src/validation.rs deleted file mode 100644 index 6dd4174148c..00000000000 --- a/litellm-rust/crates/model-catalog/src/validation.rs +++ /dev/null @@ -1,252 +0,0 @@ -use std::collections::BTreeSet; - -use serde_json::{Map, Value}; -use thiserror::Error; - -use crate::{AliasIssue, Catalog, ModelInfo, UtcHours, Weekday}; - -#[derive(Debug, Error)] -pub enum RegistryValidationError { - #[error("invalid registry entry {model:?}")] - Entry { - model: String, - #[source] - reason: EntryViolation, - }, - #[error(transparent)] - Alias(AliasIssue), -} - -#[derive(Debug, Error)] -pub enum EntryViolation { - #[error("entry must be an object")] - NotObject, - #[error("aliases must be an array")] - AliasesNotArray, - #[error("aliases must contain strings")] - AliasNotString, - #[error("entry does not match the model schema")] - Schema(#[source] serde_json::Error), - #[error("litellm_provider is required")] - MissingProvider, - #[error("invalid deprecation_date")] - DeprecationDate(#[source] time::error::Parse), - #[error("off_peak_pricing needs hours or windows")] - OffPeakWithoutHours, - #[error("off_peak_pricing.hours_utc is invalid")] - InvalidHours, - #[error("off_peak_pricing.windows is empty")] - EmptyWindows, - #[error("off_peak_pricing.weekdays is invalid")] - InvalidWeekdays, - #[error("typed entry could not be serialized")] - Serialize(#[source] serde_json::Error), - #[error("entry has an unknown field, null, or changed value: {}", fields.join(", "))] - ChangedFields { fields: Vec }, - #[error("{field} must be nonnegative")] - Negative { field: String }, - #[error("{field} must be at least one")] - UpliftBelowOne { field: String }, -} - -pub fn validate_model_entry(model: &str, value: &Value) -> Result<(), RegistryValidationError> { - validate_entry_inner(value).map_err(|reason| RegistryValidationError::Entry { - model: model.to_owned(), - reason, - }) -} - -pub fn validate_registry(catalog: &Catalog) -> Result<(), RegistryValidationError> { - if let Some(issue) = catalog.alias_issues().first() { - return Err(RegistryValidationError::Alias(issue.clone())); - } - catalog.model_names().try_for_each(|name| { - let entry = catalog.lookup(name).expect("catalog name must resolve"); - validate_model_entry(name, &Value::Object(entry.entry.fields().clone())) - }) -} - -fn json_eq(left: &Value, right: &Value) -> bool { - match (left, right) { - (Value::Number(left), Value::Number(right)) => left.as_f64() == right.as_f64(), - (Value::Array(left), Value::Array(right)) => { - left.len() == right.len() && left.iter().zip(right).all(|(a, b)| json_eq(a, b)) - } - (Value::Object(left), Value::Object(right)) => { - left.len() == right.len() - && left - .iter() - .all(|(key, value)| right.get(key).is_some_and(|other| json_eq(value, other))) - } - _ => left == right, - } -} - -fn keys(value: &Map) -> BTreeSet { - value.keys().cloned().collect() -} - -fn symmetric_difference(left: &BTreeSet, right: &BTreeSet) -> Vec { - left.symmetric_difference(right).cloned().collect() -} - -fn validate_entry_inner(value: &Value) -> Result<(), EntryViolation> { - let object = value.as_object().ok_or(EntryViolation::NotObject)?; - if let Some(aliases) = object.get("aliases") { - let names = aliases.as_array().ok_or(EntryViolation::AliasesNotArray)?; - if names.iter().any(|name| !name.is_string()) { - return Err(EntryViolation::AliasNotString); - } - } - let info: ModelInfo = serde_json::from_value(value.clone()).map_err(EntryViolation::Schema)?; - if info.litellm_provider.is_none() { - return Err(EntryViolation::MissingProvider); - } - validate_dates_and_windows(&info)?; - let serialized = serde_json::to_value(info).map_err(EntryViolation::Serialize)?; - let mut expected = object.clone(); - expected.remove("aliases"); - if !json_eq(&Value::Object(expected.clone()), &serialized) { - let actual = serialized - .as_object() - .expect("ModelInfo serializes as an object"); - return Err(EntryViolation::ChangedFields { - fields: symmetric_difference(&keys(&expected), &keys(actual)), - }); - } - check_prices("", value) -} - -fn validate_dates_and_windows(info: &ModelInfo) -> Result<(), EntryViolation> { - if let Some(date) = &info.deprecation_date { - let format = time::format_description::parse_borrowed::<2>("[year]-[month]-[day]").unwrap(); - time::Date::parse(date, &format).map_err(EntryViolation::DeprecationDate)?; - } - let Some(pricing) = &info.off_peak_pricing else { - return Ok(()); - }; - if pricing.hours_utc.is_none() && pricing.windows.is_none() { - return Err(EntryViolation::OffPeakWithoutHours); - } - if let Some(hours) = &pricing.hours_utc { - validate_hours(hours)?; - } - if let Some(windows) = &pricing.windows { - if windows.is_empty() { - return Err(EntryViolation::EmptyWindows); - } - windows.iter().try_for_each(|window| { - validate_hours(&window.hours_utc)?; - if let Some(days) = &window.weekdays - && (days.is_empty() || days.iter().any(|day| !valid_weekday(day))) - { - return Err(EntryViolation::InvalidWeekdays); - } - Ok(()) - })?; - } - Ok(()) -} - -fn validate_hours(hours: &UtcHours) -> Result<(), EntryViolation> { - let values = match hours { - UtcHours::Single(value) => std::slice::from_ref(value), - UtcHours::Multiple(values) => values.as_slice(), - }; - if values.is_empty() || values.iter().any(|value| !valid_utc_window(value)) { - return Err(EntryViolation::InvalidHours); - } - Ok(()) -} - -fn valid_utc_window(value: &str) -> bool { - let Some((start, end)) = value.split_once('-') else { - return false; - }; - [start, end].into_iter().all(|clock| { - let Some((hour, minute)) = clock.split_once(':') else { - return false; - }; - hour.len() == 2 - && minute.len() == 2 - && hour.parse::().is_ok_and(|hour| hour < 24) - && minute.parse::().is_ok_and(|minute| minute < 60) - }) -} - -fn valid_weekday(day: &Weekday) -> bool { - match day { - Weekday::Number(number) => (1..=7).contains(number), - Weekday::Name(name) => matches!( - name.to_ascii_lowercase().as_str(), - "mon" - | "monday" - | "tue" - | "tues" - | "tuesday" - | "wed" - | "wednesday" - | "thu" - | "thur" - | "thurs" - | "thursday" - | "fri" - | "friday" - | "sat" - | "saturday" - | "sun" - | "sunday" - ), - } -} - -fn check_prices(path: &str, value: &Value) -> Result<(), EntryViolation> { - let Some(object) = value.as_object() else { - return Ok(()); - }; - object.iter().try_for_each(|(key, field)| { - let field_path = format!("{path}.{key}"); - if (key.contains("cost") - || path.ends_with(".guardrail_cost_per_unit") - || path.ends_with(".search_context_cost_per_query")) - && let Some(number) = field.as_f64() - && number < 0.0 - { - return Err(EntryViolation::Negative { - field: relative(&field_path), - }); - } - if key.contains("uplift_multiplier") - && let Some(number) = field.as_f64() - && number < 1.0 - { - return Err(EntryViolation::UpliftBelowOne { - field: relative(&field_path), - }); - } - if matches!(key.as_str(), "range" | "max_results_range") - && field.as_array().is_some_and(|values| { - values - .iter() - .any(|value| value.as_f64().is_some_and(|n| n < 0.0)) - }) - { - return Err(EntryViolation::Negative { - field: relative(&field_path), - }); - } - if matches!(key.as_str(), "metadata" | "provider_specific_entry") { - return Ok(()); - } - match field.as_array() { - Some(items) => items.iter().enumerate().try_for_each(|(index, item)| { - check_prices(&format!("{field_path}[{index}]"), item) - }), - None => check_prices(&field_path, field), - } - }) -} - -fn relative(field_path: &str) -> String { - field_path.trim_start_matches('.').to_owned() -} diff --git a/litellm-rust/crates/model-catalog/src/vocabulary.rs b/litellm-rust/crates/model-catalog/src/vocabulary.rs new file mode 100644 index 00000000000..2b7ac8bd1ab --- /dev/null +++ b/litellm-rust/crates/model-catalog/src/vocabulary.rs @@ -0,0 +1,126 @@ +/// Primary API surface / task type of the model. +#[macro_rules_attribute::apply(crate::wire_type)] +#[derive(Copy, Eq, Hash)] +#[serde(rename_all = "snake_case")] +pub enum Mode { + AudioSpeech, + AudioTranscription, + Chat, + Completion, + Decisions, + Embedding, + Evaluation, + Extraction, + Guardrail, + ImageEdit, + ImageGeneration, + Moderation, + Ocr, + Realtime, + Rerank, + Responses, + Search, + VectorStore, + VideoGeneration, +} + +/// Gemini audio generation API the model is served through. +#[macro_rules_attribute::apply(crate::wire_type)] +#[derive(Copy, Eq, Hash)] +#[serde(rename_all = "snake_case")] +pub enum VertexAiAudioApi { + LyriaPredict, + LyriaInteractions, +} + +/// Audio container format the model can return. +#[macro_rules_attribute::apply(crate::wire_type)] +#[derive(Copy, Eq, Hash)] +#[serde(rename_all = "snake_case")] +pub enum AudioFormat { + Mp3, + Wav, +} + +/// Input modality the model accepts. +#[macro_rules_attribute::apply(crate::wire_type)] +#[derive(Copy, Eq, Hash)] +#[serde(rename_all = "snake_case")] +pub enum InputModality { + Text, + Image, + Audio, + Video, +} + +/// Output modality the model can produce. +#[macro_rules_attribute::apply(crate::wire_type)] +#[derive(Copy, Eq, Hash)] +#[serde(rename_all = "snake_case")] +pub enum OutputModality { + Text, + Image, + Audio, + Video, + Code, +} + +/// API route a model can be called through. +#[macro_rules_attribute::apply(crate::wire_type)] +#[derive(Copy, Eq, Hash)] +pub enum Endpoint { + #[serde(rename = "/v1/chat/completions")] + ChatCompletions, + #[serde(rename = "/v1/completions")] + Completions, + #[serde(rename = "/v1/responses")] + Responses, + #[serde(rename = "/v1/messages")] + Messages, + #[serde(rename = "/v1/batch")] + Batch, + #[serde(rename = "/v1/embeddings")] + Embeddings, + #[serde(rename = "/v1/images/generations")] + ImageGenerations, + #[serde(rename = "/v1/images/edits")] + ImageEdits, + #[serde(rename = "/v1/audio/speech")] + AudioSpeech, + #[serde(rename = "/v1/audio/transcriptions")] + AudioTranscriptions, + #[serde(rename = "/v1/realtime")] + Realtime, + #[serde(rename = "/v1/realtime/transcription_sessions")] + RealtimeTranscriptionSessions, + #[serde(rename = "/v1/videos")] + Videos, + #[serde(rename = "/v1/ocr")] + Ocr, + #[serde(rename = "/v1/listen")] + Listen, + #[serde(rename = "/v1/systemone")] + SystemOne, + #[serde(rename = "/v1/decisions")] + Decisions, + #[serde(rename = "/v1beta/interactions")] + Interactions, + #[serde(rename = "/v1beta/models")] + Models, + #[serde(rename = "/vertex_ai/live")] + VertexAiLive, +} + +/// Reasoning effort level a model accepts or applies. +#[macro_rules_attribute::apply(crate::wire_type)] +#[derive(Copy, Eq, Hash)] +#[serde(rename_all = "snake_case")] +pub enum EffortLevel { + None, + Minimal, + Low, + Medium, + High, + Xhigh, + Max, +} diff --git a/litellm-rust/crates/model-catalog/src/wire/attributes.rs b/litellm-rust/crates/model-catalog/src/wire/attributes.rs new file mode 100644 index 00000000000..93c10a81524 --- /dev/null +++ b/litellm-rust/crates/model-catalog/src/wire/attributes.rs @@ -0,0 +1,344 @@ +use std::collections::BTreeMap; +use std::fmt; +use std::str::FromStr; + +use litellm_cost::money::UnitPrice; +use serde::de::{self, Deserializer}; +use serde::{Deserialize, Serialize}; +use serde_json::Value; + +use crate::error::{DateCause, Error, Repr}; +use crate::vocabulary::{ + AudioFormat, EffortLevel, Endpoint, InputModality, Mode, OutputModality, VertexAiAudioApi, +}; +use crate::wire::lax::Count; + +macro_rules! attributes { + ($($(#[$doc:meta])* $name:ident: $ty:ty $(as $via:ty)?),* $(,)?) => { + /// The named, non-price fields of a catalog entry. + #[derive(Clone, Debug, Default, PartialEq)] + #[cfg_attr(feature = "schema", derive(schemars::JsonSchema))] + pub struct Attributes { + $($(#[$doc])* pub $name: Option<$ty>,)* + } + + impl Attributes { + pub(crate) fn read( + &mut self, + key: &str, + value: &Value, + ) -> Option> { + match key { + $(stringify!($name) => Some( + attributes!(@read value, $ty $(, $via)?).map(|read| self.$name = Some(read)), + ),)* + _ => None, + } + } + } + }; + (@read $value:ident, $ty:ty) => { + <$ty as Deserialize>::deserialize($value) + }; + (@read $value:ident, $ty:ty, $via:ty) => { + serde_with::de::DeserializeAsWrap::<$ty, $via>::deserialize($value) + .map(serde_with::de::DeserializeAsWrap::into_inner) + }; +} + +attributes! { + /// Other names this entry is served under. + aliases: Vec, + audio_transcription_config: String, + bedrock_converse_supports_strict_tools: bool, + /// Highest reasoning effort the Bedrock output_config accepts for this model. + bedrock_output_config_effort_ceiling: EffortLevel, + comment: String, + /// Reasoning effort the provider applies when the request omits reasoning_effort. Gates whether a non-default temperature or the top_p/logprobs sampling params are accepted, which hold only when the effort resolves to 'none'. + default_reasoning_effort: EffortLevel, + /// Date the provider deprecates the model, YYYY-MM-DD. + deprecation_date: IsoDate, + gemini_audio_only_live: bool, + gemini_native_audio: bool, + input_cost_per_token_cache_hit: UnitPrice, + /// Databricks units, not USD. + input_dbu_cost_per_token: Dbu, + /// LiteLLM provider slug; one of https://docs.litellm.ai/docs/providers. + litellm_provider: String, + /// Maximum prompt/context tokens the model accepts. + max_input_tokens: u64 as Count, + /// Maximum tokens the model can generate in one response. + max_output_tokens: u64 as Count, + /// Legacy field: max output tokens if the provider specifies it, else max input tokens. + max_tokens: u64 as Count, + /// Free-form notes about the entry (e.g. pricing derivation). + metadata: BTreeMap, + /// Primary API surface / task type of the model. + mode: Mode, + /// Databricks units, not USD. + output_dbu_cost_per_token: Dbu, + /// Embedding dimension for embedding models. + output_vector_size: u64 as Count, + /// Smallest prefix the provider will actually cache; absent means the provider default applies. + prompt_cache_min_tokens: u64 as Count, + /// Provider-internal routing hints (e.g. bedrock_invocation_schema). + provider_specific_entry: BTreeMap, + reasoning_contexts: Vec, + /// Exact reasoning_effort levels this deployment accepts; wins over supports_* flags. + reasoning_effort_levels: Vec, + reasoning_modes: Vec, + /// Provider default requests-per-minute limit. + rpm: u64 as Count, + /// Grounded search requests the provider includes for free before search is billed. + search_free_requests: SearchFreeRequests, + /// URL of the provider pricing/model page this entry was taken from. + source: String, + /// Audio container formats the model can return. + supported_audio_formats: Vec, + /// OpenAI-style API routes this model can be called through, e.g. /v1/chat/completions. + supported_endpoints: Vec, + /// Input modalities the model accepts. + supported_modalities: Vec, + /// Output modalities the model can produce. + supported_output_modalities: Vec, + /// Cloud regions the model is available in ('global' or region ids). + supported_regions: Vec, + thinking_always_on: bool, + token_count_framing: u64 as Count, + token_count_model: TokenCountModel, + token_count_tokenizer: TokenCountTokenizer, + /// Provider default tokens-per-minute limit. + tpm: u64 as Count, + use_openai_responses_path: bool, + uses_embed_content: bool, + vertex_ai_audio_api: VertexAiAudioApi, + /// Whether web search is billed per query or per prompt. + web_search_billing_unit: WebSearchBillingUnit, +} + +/// The model whose provider token counter counts tokens for this entry. +#[macro_rules_attribute::apply(crate::wire_type)] +#[derive(Eq, Hash)] +#[serde(deny_unknown_fields)] +pub struct TokenCountModel { + pub model: String, + #[serde(skip_serializing_if = "Option::is_none")] + pub litellm_provider: Option, + #[serde(skip_serializing_if = "Option::is_none")] + pub region: Option, +} + +/// Free grounded search requests per period, shared across the models of one group. +#[macro_rules_attribute::apply(crate::wire_type)] +#[derive(Eq, Hash)] +#[serde(deny_unknown_fields)] +pub struct SearchFreeRequests { + pub count: u64, + pub period: FreeRequestPeriod, + pub shared_group: String, +} + +#[macro_rules_attribute::apply(crate::wire_type)] +#[derive(Copy, Eq, Hash)] +#[serde(rename_all = "snake_case")] +pub enum FreeRequestPeriod { + Day, + Month, +} + +/// Local tokenizer that counts tokens for this entry. +#[macro_rules_attribute::apply(crate::wire_type)] +#[derive(Copy, Eq, Hash)] +#[serde(rename_all = "snake_case")] +pub enum TokenCountTokenizer { + KimiK3, +} + +/// Whether web search is billed per query or per prompt. +#[macro_rules_attribute::apply(crate::wire_type)] +#[derive(Copy, Eq, Hash)] +#[serde(rename_all = "snake_case")] +pub enum WebSearchBillingUnit { + PerQuery, + PerPrompt, +} + +/// A non-negative amount of Databricks units. +#[derive(Clone, Copy, Debug, PartialEq, PartialOrd, Serialize)] +#[serde(transparent)] +pub struct Dbu(f64); + +impl Dbu { + pub fn new(units: f64) -> Option { + (units.is_finite() && units >= 0.0).then_some(Self(units)) + } + pub fn get(self) -> f64 { + self.0 + } +} + +impl<'de> Deserialize<'de> for Dbu { + fn deserialize>(deserializer: D) -> Result { + let units = f64::deserialize(deserializer)?; + Self::new(units).ok_or_else(|| { + de::Error::invalid_value( + de::Unexpected::Float(units), + &"a non-negative number of Databricks units", + ) + }) + } +} + +/// A calendar date written YYYY-MM-DD. +#[derive( + Clone, + Copy, + Debug, + PartialEq, + Eq, + PartialOrd, + Ord, + Hash, + serde_with::DeserializeFromStr, + serde_with::SerializeDisplay, +)] +pub struct IsoDate(pub time::Date); + +impl FromStr for IsoDate { + type Err = Error; + + fn from_str(date: &str) -> Result { + let mut parts = date.split('-'); + let (Some(year), Some(month), Some(day), None) = + (parts.next(), parts.next(), parts.next(), parts.next()) + else { + return Err(Repr::InvalidDate { cause: None }.into()); + }; + if year.len() != 4 || month.len() != 2 || day.len() != 2 { + return Err(Repr::InvalidDate { cause: None }.into()); + } + let number = |part: &str| part.parse::().map_err(DateCause::from); + let month = time::Month::try_from(number(month)?).map_err(DateCause::from)?; + time::Date::from_calendar_date(year.parse().map_err(DateCause::from)?, month, number(day)?) + .map(Self) + .map_err(DateCause::from) + .map_err(Error::from) + } +} + +impl fmt::Display for IsoDate { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + write!( + f, + "{:04}-{:02}-{:02}", + self.0.year(), + u8::from(self.0.month()), + self.0.day() + ) + } +} + +#[cfg(feature = "schema")] +mod schema { + use std::borrow::Cow; + + use schemars::{JsonSchema, Schema, SchemaGenerator, json_schema}; + + use super::{Dbu, IsoDate}; + + impl JsonSchema for IsoDate { + fn inline_schema() -> bool { + true + } + fn schema_name() -> Cow<'static, str> { + "IsoDate".into() + } + fn json_schema(_: &mut SchemaGenerator) -> Schema { + json_schema!({ + "type": "string", + "format": "date", + "pattern": r"^\d{4}-(0[1-9]|1[0-2])-(0[1-9]|[12]\d|3[01])$" + }) + } + } + + impl JsonSchema for Dbu { + fn inline_schema() -> bool { + true + } + fn schema_name() -> Cow<'static, str> { + "Dbu".into() + } + fn json_schema(_: &mut SchemaGenerator) -> Schema { + json_schema!({"type": "number", "minimum": 0}) + } + } +} + +#[cfg(test)] +mod tests { + use rstest::rstest; + use serde_json::json; + + use super::*; + + #[rstest] + #[case::leap_day("2028-02-29", true)] + #[case::not_a_leap_year("2026-02-29", false)] + #[case::impossible_day("2026-02-31", false)] + #[case::one_digit_month("2026-2-01", false)] + #[case::extra_part("2026-02-01-01", false)] + fn iso_dates_are_real_calendar_dates(#[case] text: &str, #[case] valid: bool) { + assert_eq!(text.parse::().is_ok(), valid); + } + + #[rstest] + #[case::count_written_as_text("max_tokens", json!("8192"), true)] + #[case::negative_count("max_tokens", json!(-1), false)] + #[case::known_mode("mode", json!("chat"), true)] + #[case::unknown_mode("mode", json!("chatt"), false)] + #[case::negative_dbu("input_dbu_cost_per_token", json!(-1), false)] + #[case::reasoning_contexts("reasoning_contexts", json!(["reasoning", "non_reasoning"]), true)] + #[case::reasoning_modes("reasoning_modes", json!(["none", "adaptive"]), true)] + #[case::reasoning_context_not_a_list("reasoning_contexts", json!("reasoning"), false)] + #[case::reasoning_mode_not_a_string("reasoning_modes", json!([1]), false)] + fn each_attribute_reads_its_own_type( + #[case] key: &str, + #[case] value: Value, + #[case] valid: bool, + ) { + assert_eq!( + Attributes::default() + .read(key, &value) + .map(|read| read.is_ok()), + Some(valid) + ); + } + + #[rstest] + fn keys_that_are_not_attributes_are_left_to_the_caller() { + assert!( + Attributes::default() + .read("input_cost_per_token", &json!(1)) + .is_none() + ); + } + + #[rstest] + fn reasoning_metadata_is_retained() { + let mut attributes = Attributes::default(); + attributes + .read("reasoning_contexts", &json!(["reasoning", "non_reasoning"])) + .unwrap() + .unwrap(); + attributes + .read("reasoning_modes", &json!(["none", "adaptive"])) + .unwrap() + .unwrap(); + assert_eq!( + attributes.reasoning_contexts.unwrap(), + ["reasoning", "non_reasoning"] + ); + assert_eq!(attributes.reasoning_modes.unwrap(), ["none", "adaptive"]); + } +} diff --git a/litellm-rust/crates/model-catalog/src/wire/blocks.rs b/litellm-rust/crates/model-catalog/src/wire/blocks.rs new file mode 100644 index 00000000000..3a355686b7c --- /dev/null +++ b/litellm-rust/crates/model-catalog/src/wire/blocks.rs @@ -0,0 +1,182 @@ +use litellm_cost::money::UnitPrice; +use litellm_cost::schedule::{UtcWindow, Weekday}; +use serde::Deserialize; +use serde::de::DeserializeOwned; +use serde_json::{Map, Value}; +use serde_with::de::DeserializeAsWrap; + +use crate::error::{Error, Repr, pointer_child, pointer_for_key}; +use crate::wire::flat_key::FlatKey; +use crate::wire::lax::Count; + +/// A flat price key with its spelling, kept so problems can name it. +#[derive(Clone, Debug, PartialEq)] +pub struct FlatPrice { + pub key: String, + pub path: String, + pub flat: FlatKey, + pub price: UnitPrice, +} + +#[derive(Clone, Debug, Default, PartialEq)] +pub struct TierRow { + pub path: String, + pub range: Option<[u64; 2]>, + pub max_results_range: Option<[u64; 2]>, + pub prices: Vec, +} + +#[derive(Clone, Debug, Default, PartialEq)] +pub struct OffPeakBlock { + pub hours_utc: Option, + pub windows: Option>, + pub weekday_timezone: Option, + pub prices: Vec, +} + +/// One UTC window or a list of them. +#[derive(Clone, Debug, PartialEq, Eq, Deserialize)] +#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))] +#[serde(untagged)] +pub enum UtcHours { + Single(UtcWindow), + Multiple(Vec), +} + +/// One off-peak window entry inside `windows`. +#[derive(Clone, Debug, PartialEq, Eq, Deserialize)] +#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))] +#[serde(deny_unknown_fields)] +pub struct OffPeakWindow { + pub hours_utc: UtcHours, + pub weekdays: Option>, +} + +/// USD cost per web search query, keyed by search context size. +#[derive(Clone, Debug, Default, PartialEq, Deserialize)] +#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))] +#[serde(deny_unknown_fields)] +pub struct SearchContextCostPerQuery { + pub search_context_size_low: Option, + pub search_context_size_medium: Option, + pub search_context_size_high: Option, +} + +impl UtcHours { + pub fn windows(&self) -> &[UtcWindow] { + match self { + Self::Single(window) => std::slice::from_ref(window), + Self::Multiple(windows) => windows, + } + } +} + +pub(crate) struct Reader<'a> { + pub(crate) problems: &'a mut Vec, +} + +impl Reader<'_> { + pub(crate) fn value(&mut self, path: String, value: &Value) -> Option { + T::deserialize(value) + .map_err(|error| self.problems.push(Error::new(path, Repr::Invalid(error)))) + .ok() + } + + pub(crate) fn price( + &mut self, + key: String, + path: String, + flat: FlatKey, + value: &Value, + ) -> Option { + let price = self.value(path.clone(), value)?; + Some(FlatPrice { + key, + path, + flat, + price, + }) + } + + fn object<'v>(&mut self, path: &str, value: &'v Value) -> Option<&'v Map> { + match value { + Value::Object(fields) => Some(fields), + other => self + .value::>(path.to_owned(), other) + .and(None), + } + } + + fn block_price( + &mut self, + name: &str, + key: String, + path: String, + value: &Value, + ) -> Option { + let Some(flat) = FlatKey::parse(name) else { + self.problems.push(Error::new(path, Repr::UnknownBlockKey)); + return None; + }; + self.price(key, path, flat, value) + } + + pub(crate) fn tier_rows(&mut self, value: &Value) -> Option> { + let path = pointer_for_key("tiered_pricing"); + let rows: Vec = self.value(path.clone(), value)?; + Some( + rows.iter() + .enumerate() + .filter_map(|(index, row)| self.tier_row(path.clone(), index, row)) + .collect(), + ) + } + + fn tier_row(&mut self, parent: String, index: usize, value: &Value) -> Option { + let path = format!("{parent}/{index}"); + let display_path = format!("tiered_pricing[{index}]"); + let fields = self.object(&path, value)?; + let mut row = TierRow { + path: path.clone(), + ..TierRow::default() + }; + for (name, value) in fields.iter().filter(|(_, value)| !value.is_null()) { + let key = format!("{display_path}.{name}"); + let child_path = pointer_child(&path, name); + match name.as_str() { + "range" => row.range = self.bounds(child_path, value), + "max_results_range" => row.max_results_range = self.bounds(child_path, value), + _ => row + .prices + .extend(self.block_price(name, key, child_path, value)), + } + } + Some(row) + } + + fn bounds(&mut self, path: String, value: &Value) -> Option<[u64; 2]> { + DeserializeAsWrap::<[u64; 2], [Count; 2]>::deserialize(value) + .map(DeserializeAsWrap::into_inner) + .map_err(|error| self.problems.push(Error::new(path, Repr::Invalid(error)))) + .ok() + } + + pub(crate) fn off_peak(&mut self, value: &Value) -> Option { + let path = pointer_for_key("off_peak_pricing"); + let fields = self.object(&path, value)?; + let mut block = OffPeakBlock::default(); + for (name, value) in fields.iter().filter(|(_, value)| !value.is_null()) { + let key = format!("off_peak_pricing.{name}"); + let child_path = pointer_child(&path, name); + match name.as_str() { + "hours_utc" => block.hours_utc = self.value(child_path, value), + "windows" => block.windows = self.value(child_path, value), + "weekday_timezone" => block.weekday_timezone = self.value(child_path, value), + _ => block + .prices + .extend(self.block_price(name, key, child_path, value)), + } + } + Some(block) + } +} diff --git a/litellm-rust/crates/model-catalog/src/wire/entry.rs b/litellm-rust/crates/model-catalog/src/wire/entry.rs new file mode 100644 index 00000000000..9c155c6496a --- /dev/null +++ b/litellm-rust/crates/model-catalog/src/wire/entry.rs @@ -0,0 +1,193 @@ +use std::collections::BTreeMap; + +use litellm_cost::money::{Multiplier, UnitPrice}; +use serde_json::{Map, Value}; + +use crate::error::{Error, Repr, pointer_for_key}; +use crate::wire::attributes::Attributes; +use crate::wire::blocks::{FlatPrice, OffPeakBlock, Reader, SearchContextCostPerQuery, TierRow}; +use crate::wire::flat_key::FlatKey; + +pub type Fields = Map; + +/// One catalog entry as written. Every non-null key lands in exactly one field, and a known key +/// with a bad value is kept out and listed in `problems`. +#[derive(Debug, Default)] +pub struct EntryWire { + pub attributes: Attributes, + pub prices: Vec, + pub capabilities: BTreeMap, + pub tier_rows: Option>, + pub off_peak: Option, + pub search_context: Option, + pub guardrail: Option>, + pub regional_endpoint_uplift: Option, + pub residency_uplift_eu: Option, + pub residency_uplift_us: Option, + pub unknown: Fields, + pub problems: Vec, +} + +impl EntryWire { + pub fn read(fields: &Fields) -> Self { + let mut problems = Vec::new(); + let mut wire = Self::default(); + let mut reader = Reader { + problems: &mut problems, + }; + for (key, value) in fields.iter().filter(|(_, value)| !value.is_null()) { + wire.read_key(&mut reader, key, value); + } + Self { problems, ..wire } + } + + fn read_key(&mut self, reader: &mut Reader<'_>, key: &str, value: &Value) { + let path = || pointer_for_key(key); + match key { + "tiered_pricing" => self.tier_rows = reader.tier_rows(value), + "off_peak_pricing" => self.off_peak = reader.off_peak(value), + "search_context_cost_per_query" => self.search_context = reader.value(path(), value), + "guardrail_cost_per_unit" => self.guardrail = reader.value(path(), value), + "regional_endpoint_uplift_multiplier" => { + self.regional_endpoint_uplift = reader.value(path(), value) + } + "regional_processing_uplift_multiplier_eu" => { + self.residency_uplift_eu = reader.value(path(), value) + } + "regional_processing_uplift_multiplier_us" => { + self.residency_uplift_us = reader.value(path(), value) + } + _ => self.read_open_key(reader, key, value), + } + } + + fn read_open_key(&mut self, reader: &mut Reader<'_>, key: &str, value: &Value) { + if let Some(read) = self.attributes.read(key, value) { + if let Err(error) = read { + reader + .problems + .push(Error::from_key(key, Repr::Invalid(error))); + } + return; + } + if let Some(flat) = FlatKey::parse(key) { + self.prices + .extend(reader.price(key.to_owned(), pointer_for_key(key), flat, value)); + return; + } + if key.starts_with("supports_") { + if let Some(flag) = reader.value(pointer_for_key(key), value) { + self.capabilities.insert(key.to_owned(), flag); + } + return; + } + self.unknown.insert(key.to_owned(), value.clone()); + } +} + +#[cfg(test)] +mod tests { + use rstest::rstest; + use serde_json::json; + + use super::*; + + fn read(entry: Value) -> EntryWire { + let Value::Object(fields) = entry else { + panic!("entry must be an object") + }; + EntryWire::read(&fields) + } + + fn problem_keys(wire: &EntryWire) -> Vec { + wire.problems + .iter() + .map(|problem| problem.path().to_owned()) + .collect() + } + + #[rstest] + fn every_key_lands_in_exactly_one_place() { + let wire = read(json!({ + "litellm_provider": "openai", + "max_tokens": "8192", + "input_cost_per_token": "0.000001", + "tiered_pricing": [{"range": [0, 10], "output_cost_per_token": 2e-6}], + "supports_vision": true, + "id": "deployment-1", + "mode": null + })); + + assert_eq!(wire.attributes.litellm_provider.as_deref(), Some("openai")); + assert_eq!(wire.attributes.max_tokens, Some(8192)); + assert_eq!(wire.attributes.mode, None); + assert_eq!( + wire.prices + .iter() + .map(|price| price.key.as_str()) + .collect::>(), + ["input_cost_per_token"] + ); + assert_eq!(wire.tier_rows.unwrap()[0].prices.len(), 1); + assert_eq!( + wire.capabilities, + BTreeMap::from([("supports_vision".to_owned(), true)]) + ); + assert_eq!(wire.unknown.keys().collect::>(), ["id"]); + assert!(wire.problems.is_empty()); + } + + #[rstest] + fn every_bad_value_is_reported_by_its_key_at_once() { + let wire = read(json!({ + "mode": "chatt", + "max_tokens": -1, + "input_cost_per_token": "x", + "output_cost_per_image_0.5K": "x", + "supports_vision": "yes", + "regional_endpoint_uplift_multiplier": 0.5, + "tiered_pricing": [ + {"range": [0, 10], "input_cost_per_token": 1e-6}, + {"range": [10.5, 20], "typo_cost": 1, "output_cost_per_token": -1} + ], + "off_peak_pricing": {"hours_utc": "25:00-01:00", "input_cost_per_token": 1e-6} + })); + + assert_eq!( + problem_keys(&wire) + .into_iter() + .collect::>(), + [ + "/input_cost_per_token", + "/max_tokens", + "/mode", + "/off_peak_pricing/hours_utc", + "/output_cost_per_image_0.5K", + "/regional_endpoint_uplift_multiplier", + "/supports_vision", + "/tiered_pricing/1/output_cost_per_token", + "/tiered_pricing/1/range", + "/tiered_pricing/1/typo_cost", + ] + .map(str::to_owned) + .into_iter() + .collect() + ); + } + + #[rstest] + #[case::tiers_not_a_list(json!({"tiered_pricing": {"range": [0, 1]}}), "/tiered_pricing")] + #[case::row_not_an_object(json!({"tiered_pricing": [3]}), "/tiered_pricing/0")] + #[case::off_peak_not_an_object(json!({"off_peak_pricing": true}), "/off_peak_pricing")] + #[case::unknown_search_size( + json!({"search_context_cost_per_query": {"search_context_size_huge": 1}}), + "/search_context_cost_per_query" + )] + #[case::negative_guardrail( + json!({"guardrail_cost_per_unit": {"contentPolicyUnits": -1}}), + "/guardrail_cost_per_unit" + )] + fn a_bad_block_is_reported_at_the_block(#[case] entry: Value, #[case] expected: &str) { + assert_eq!(problem_keys(&read(entry)), [expected]); + } +} diff --git a/litellm-rust/crates/model-catalog/src/wire/flat_key.rs b/litellm-rust/crates/model-catalog/src/wire/flat_key.rs new file mode 100644 index 00000000000..906d917b487 --- /dev/null +++ b/litellm-rust/crates/model-catalog/src/wire/flat_key.rs @@ -0,0 +1,359 @@ +use std::num::NonZeroU64; + +use litellm_cost::price::{ + Band, CacheReadMode, CacheTtl, Channel, Feature, ImageSize, InputUnit, Meter, OutputUnit, + Resolution, ServiceTier, TokenKind, When, +}; + +/// A flat price key as the grammar reads it: `{meter}[_above_{N}k_tokens][_{tier}]`. +#[derive(Clone, Debug, PartialEq, Eq)] +pub struct FlatKey { + pub meter: Meter, + pub band: Band, + pub tier: Option, +} + +#[derive(Clone, Copy, Debug, PartialEq, Eq)] +pub enum TierSuffix { + Service(ServiceTier), + Batches, + Decisions, +} + +impl FlatKey { + pub fn parse(key: &str) -> Option { + let (rest, tier) = split_tier(key); + let (base, suffix) = split_above(rest)?; + let (meter, band) = match suffix { + Above::None => (meter(base)?, Band::Base), + Above::PromptTokens(tokens) => (meter(base)?, Band::over_prompt_tokens(tokens)), + Above::Seconds(seconds) => match meter(base)? { + Meter::Input(InputUnit::VideoSecond { over_seconds: None }) => ( + Meter::Input(InputUnit::VideoSecond { + over_seconds: NonZeroU64::new(seconds), + }), + Band::Base, + ), + _ => return None, + }, + }; + Some(Self { meter, band, tier }) + } + + pub fn has_suffix(&self) -> bool { + self.band != Band::Base || self.tier.is_some() + } + + pub fn channel_and_when(&self) -> (Channel, When) { + let (channel, tier) = match self.tier { + None => (Channel::Online, ServiceTier::Standard), + Some(TierSuffix::Service(tier)) => (Channel::Online, tier), + Some(TierSuffix::Batches) => (Channel::Batch, ServiceTier::Standard), + Some(TierSuffix::Decisions) => (Channel::Decisions, ServiceTier::Standard), + }; + ( + channel, + When { + tier, + band: self.band, + }, + ) + } +} + +enum Above { + None, + PromptTokens(u64), + Seconds(u64), +} + +fn split_tier(key: &str) -> (&str, Option) { + let Some((head, tail)) = key.rsplit_once('_') else { + return (key, None); + }; + let tier = match tail { + "priority" => TierSuffix::Service(ServiceTier::Priority), + "flex" => TierSuffix::Service(ServiceTier::Flex), + "ultrafast" => TierSuffix::Service(ServiceTier::Ultrafast), + "balanced" => TierSuffix::Service(ServiceTier::Balanced), + "batches" => TierSuffix::Batches, + "decisions" => TierSuffix::Decisions, + _ => return (key, None), + }; + (head, Some(tier)) +} + +fn split_above(key: &str) -> Option<(&str, Above)> { + let Some((head, tail)) = key.rsplit_once("_above_") else { + return Some((key, Above::None)); + }; + if tail == "1hr" || tail == "12hr" { + return Some((key, Above::None)); + } + if let Some(thousands) = tail.strip_suffix("k_tokens") { + return Some(( + head, + Above::PromptTokens(number(thousands)?.checked_mul(1000)?), + )); + } + if let Some(tokens) = tail.strip_suffix("_tokens") { + return Some((head, Above::PromptTokens(number(tokens)?))); + } + let seconds = tail.strip_suffix("s_interval")?; + Some((head, Above::Seconds(number(seconds)?))) +} + +fn number(digits: &str) -> Option { + (!digits.is_empty() && digits.bytes().all(|byte| byte.is_ascii_digit())) + .then(|| digits.parse().ok()) + .flatten() +} + +fn meter(base: &str) -> Option { + let fixed = match base { + "cost_per_second" => Some(Meter::WallClockSecond), + "explicit_cache_read_input_token_cost" => Some(Meter::CacheRead { + kind: TokenKind::Text, + mode: CacheReadMode::Explicit, + }), + "annotation_cost_per_page" => Some(Meter::Feature(Feature::AnnotationPage)), + "citation_cost_per_token" => Some(Meter::Feature(Feature::CitationToken)), + "code_interpreter_cost_per_session" => { + Some(Meter::Feature(Feature::CodeInterpreterSession)) + } + "google_maps_grounding_cost_per_query" => { + Some(Meter::Feature(Feature::GoogleMapsGroundingQuery)) + } + "ocr_cost_per_credit" => Some(Meter::Feature(Feature::OcrCredit)), + "ocr_cost_per_page" => Some(Meter::Feature(Feature::OcrPage)), + "browser_open_cost_per_request" => Some(Meter::Feature(Feature::BrowserOpen)), + "cost_per_additional_result" => Some(Meter::Feature(Feature::AdditionalResult)), + _ => None, + }; + if fixed.is_some() { + return fixed; + } + if let Some(unit) = base.strip_prefix("input_cost_per_") { + return input_unit(unit).map(Meter::Input); + } + if let Some(unit) = base.strip_prefix("output_cost_per_") { + return output_unit(unit).map(Meter::Output); + } + if let Some(kind) = base + .strip_prefix("cache_read_input_") + .and_then(|rest| rest.strip_suffix("_cost")) + { + return Some(Meter::CacheRead { + kind: token_kind(kind)?, + mode: CacheReadMode::Implicit, + }); + } + let rest = base.strip_prefix("cache_creation_input_")?; + let (kind, ttl) = if let Some(kind) = rest.strip_suffix("_cost_above_1hr") { + (kind, CacheTtl::OneHour) + } else if let Some(kind) = rest.strip_suffix("_cost_above_12hr") { + (kind, CacheTtl::TwelveHours) + } else { + (rest.strip_suffix("_cost")?, CacheTtl::Default) + }; + Some(Meter::CacheWrite { + kind: token_kind(kind)?, + ttl, + }) +} + +fn token_kind(unit: &str) -> Option { + match unit { + "token" => Some(TokenKind::Text), + "audio_token" => Some(TokenKind::Audio), + "image_token" => Some(TokenKind::Image), + "video_token" => Some(TokenKind::Video), + _ => None, + } +} + +fn input_unit(unit: &str) -> Option { + if let Some(kind) = token_kind(unit) { + return Some(InputUnit::Token(kind)); + } + match unit { + "character" => Some(InputUnit::Character), + "image" => Some(InputUnit::Image), + "pixel" => Some(InputUnit::Pixel), + "query" => Some(InputUnit::Query { over_results: None }), + "request" => Some(InputUnit::Request), + "second" => Some(InputUnit::Second), + "text_message" => Some(InputUnit::TextMessage), + "audio_per_second" => Some(InputUnit::AudioSecond), + "video_per_second" => Some(InputUnit::VideoSecond { over_seconds: None }), + _ => None, + } +} + +fn output_unit(unit: &str) -> Option { + if let Some(kind) = token_kind(unit) { + return Some(OutputUnit::Token(kind)); + } + match unit.split_once('_') { + Some(("image", label)) => { + return ImageSize::new(label).map(|size| OutputUnit::Image(Some(size))); + } + Some(("second", label)) => { + return Resolution::new(label).map(|resolution| OutputUnit::Second(Some(resolution))); + } + _ => {} + } + match unit { + "reasoning_token" => Some(OutputUnit::ReasoningToken), + "character" => Some(OutputUnit::Character), + "image" => Some(OutputUnit::Image(None)), + "pixel" => Some(OutputUnit::Pixel), + "second" => Some(OutputUnit::Second(None)), + "video_per_second" => Some(OutputUnit::VideoSecond), + _ => None, + } +} + +#[cfg(test)] +mod tests { + use rstest::rstest; + + use super::*; + + const TEXT_INPUT: Meter = Meter::Input(InputUnit::Token(TokenKind::Text)); + const TEXT_OUTPUT: Meter = Meter::Output(OutputUnit::Token(TokenKind::Text)); + + fn key(meter: Meter, band: Band, tier: Option) -> FlatKey { + FlatKey { meter, band, tier } + } + + #[rstest] + #[case::input_token("input_cost_per_token", key(TEXT_INPUT, Band::Base, None))] + #[case::band_and_tier( + "output_cost_per_token_above_272k_tokens_priority", + key( + TEXT_OUTPUT, + Band::over_prompt_tokens(272_000), + Some(TierSuffix::Service(ServiceTier::Priority)) + ) + )] + #[case::band_without_k( + "input_cost_per_token_above_128_tokens", + key(TEXT_INPUT, Band::over_prompt_tokens(128), None) + )] + #[case::zero_band_is_base( + "input_cost_per_token_above_0k_tokens", + key(TEXT_INPUT, Band::Base, None) + )] + #[case::batch( + "input_cost_per_token_batches", + key(TEXT_INPUT, Band::Base, Some(TierSuffix::Batches)) + )] + #[case::reasoning( + "output_cost_per_reasoning_token", + key(Meter::Output(OutputUnit::ReasoningToken), Band::Base, None) + )] + #[case::cache_read_audio( + "cache_read_input_audio_token_cost", + key(Meter::CacheRead { kind: TokenKind::Audio, mode: CacheReadMode::Implicit }, Band::Base, None) + )] + #[case::explicit_cache_read( + "explicit_cache_read_input_token_cost", + key(Meter::CacheRead { kind: TokenKind::Text, mode: CacheReadMode::Explicit }, Band::Base, None) + )] + #[case::cache_write_one_hour_band( + "cache_creation_input_token_cost_above_1hr_above_200k_tokens", + key( + Meter::CacheWrite { kind: TokenKind::Text, ttl: CacheTtl::OneHour }, + Band::over_prompt_tokens(200_000), + None + ) + )] + #[case::cache_write_twelve_hours_priority( + "cache_creation_input_token_cost_above_12hr_above_272k_tokens_priority", + key( + Meter::CacheWrite { kind: TokenKind::Text, ttl: CacheTtl::TwelveHours }, + Band::over_prompt_tokens(272_000), + Some(TierSuffix::Service(ServiceTier::Priority)) + ) + )] + #[case::image_label( + "output_cost_per_image_0.5K", + key( + Meter::Output(OutputUnit::Image(ImageSize::new("0.5K"))), + Band::Base, + None + ) + )] + #[case::image_token_is_a_token( + "output_cost_per_image_token", + key(Meter::Output(OutputUnit::Token(TokenKind::Image)), Band::Base, None) + )] + #[case::resolution( + "output_cost_per_second_720p", + key( + Meter::Output(OutputUnit::Second(Resolution::new("720p"))), + Band::Base, + None + ) + )] + #[case::video_interval( + "input_cost_per_video_per_second_above_8s_interval", + key(Meter::Input(InputUnit::VideoSecond { over_seconds: NonZeroU64::new(8) }), Band::Base, None) + )] + #[case::feature_batch( + "ocr_cost_per_page_batches", + key( + Meter::Feature(Feature::OcrPage), + Band::Base, + Some(TierSuffix::Batches) + ) + )] + #[case::wall_clock("cost_per_second", key(Meter::WallClockSecond, Band::Base, None))] + #[case::browser_open( + "browser_open_cost_per_request", + key(Meter::Feature(Feature::BrowserOpen), Band::Base, None) + )] + #[case::additional_result( + "cost_per_additional_result", + key(Meter::Feature(Feature::AdditionalResult), Band::Base, None) + )] + fn parses_every_key_shape(#[case] text: &str, #[case] expected: FlatKey) { + assert_eq!(FlatKey::parse(text), Some(expected)); + } + + #[rstest] + #[case::typo("input_cost_per_tokne")] + #[case::unused_meter("cost_per_token")] + #[case::other_currency("input_dbu_cost_per_token")] + #[case::cache_hit_alias("input_cost_per_token_cache_hit")] + #[case::non_numeric_band("input_cost_per_token_above_xk_tokens")] + #[case::interval_on_a_token_meter("input_cost_per_token_above_8s_interval")] + #[case::read_has_no_ttl("cache_read_input_token_cost_above_1hr")] + #[case::cache_unit_must_be_token("cache_read_input_character_cost")] + #[case::bad_image_size("output_cost_per_image_big")] + #[case::bad_resolution("output_cost_per_second_hd")] + #[case::unknown_feature("telepathy_cost_per_page")] + #[case::empty("")] + fn rejects_keys_outside_the_grammar(#[case] text: &str) { + assert_eq!(FlatKey::parse(text), None); + } + + #[rstest] + #[case::plain("input_cost_per_token", (Channel::Online, When::default()))] + #[case::service_tier( + "input_cost_per_token_flex", + (Channel::Online, When { tier: ServiceTier::Flex, band: Band::Base }) + )] + #[case::batch( + "input_cost_per_token_above_200k_tokens_batches", + (Channel::Batch, When { tier: ServiceTier::Standard, band: Band::over_prompt_tokens(200_000) }) + )] + #[case::decisions("input_cost_per_token_decisions", (Channel::Decisions, When::default()))] + fn batch_and_decisions_pick_a_channel_not_a_tier( + #[case] text: &str, + #[case] expected: (Channel, When), + ) { + assert_eq!(FlatKey::parse(text).unwrap().channel_and_when(), expected); + } +} diff --git a/litellm-rust/crates/model-catalog/src/wire/lax.rs b/litellm-rust/crates/model-catalog/src/wire/lax.rs new file mode 100644 index 00000000000..d514a6a82c1 --- /dev/null +++ b/litellm-rust/crates/model-catalog/src/wire/lax.rs @@ -0,0 +1,128 @@ +use std::fmt; + +use serde::Deserializer; +use serde::de::{self, Visitor}; +use serde_with::DeserializeAs; + +pub(crate) struct LaxI64; + +pub(crate) struct Count; + +impl<'de> DeserializeAs<'de, i64> for LaxI64 { + fn deserialize_as>(deserializer: D) -> Result { + deserializer.deserialize_any(Self) + } +} + +impl Visitor<'_> for LaxI64 { + type Value = i64; + + fn expecting(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result { + formatter.write_str("an integer in the i64 range") + } + + fn visit_i64(self, value: i64) -> Result { + Ok(value) + } + + fn visit_u64(self, value: u64) -> Result { + i64::try_from(value).map_err(E::custom) + } + + fn visit_f64(self, value: f64) -> Result { + integral_float(value).ok_or_else(|| E::custom("expected an integer in the i64 range")) + } + + fn visit_str(self, value: &str) -> Result { + integer_string(value.trim()) + .ok_or_else(|| E::custom("expected an integer in the i64 range")) + } + + fn visit_bool(self, value: bool) -> Result { + Ok(i64::from(value)) + } +} + +impl<'de> DeserializeAs<'de, u64> for Count { + fn deserialize_as>(deserializer: D) -> Result { + let count = LaxI64::deserialize_as(deserializer)?; + u64::try_from(count).map_err(|_| { + de::Error::invalid_value(de::Unexpected::Signed(count), &"a count of zero or more") + }) + } +} + +fn integer_string(value: &str) -> Option { + let integer = match value.split_once('.') { + Some((integer, fraction)) => { + if fraction.is_empty() || !fraction.bytes().all(|byte| byte == b'0') { + return None; + } + integer + } + None => value, + }; + if integer.starts_with('_') || integer.ends_with('_') || integer.contains("__") { + return None; + } + let digits = integer.strip_prefix(['+', '-']).unwrap_or(integer); + if digits.is_empty() + || digits.starts_with('_') + || !digits + .bytes() + .all(|byte| byte.is_ascii_digit() || byte == b'_') + { + return None; + } + integer.replace('_', "").parse().ok() +} + +fn integral_float(value: f64) -> Option { + (value.is_finite() + && value.fract() == 0.0 + && value >= i64::MIN as f64 + && value < -(i64::MIN as f64)) + .then_some(value as i64) +} + +#[cfg(test)] +mod tests { + use rstest::rstest; + use serde_json::{Value, json}; + use serde_with::de::DeserializeAsWrap; + + use super::*; + + fn count(value: Value) -> Option { + DeserializeAsWrap::::deserialize(value) + .ok() + .map(DeserializeAsWrap::into_inner) + } + + use serde::Deserialize; + + #[rstest] + #[case::integer(json!(128000), Some(128_000))] + #[case::whole_float(json!(128000.0), Some(128_000))] + #[case::string(json!(" 8192 "), Some(8192))] + #[case::separated(json!("1_000"), Some(1000))] + #[case::trailing_zeros(json!("2.000"), Some(2))] + #[case::boolean(json!(true), Some(1))] + #[case::maximum(json!(i64::MAX.to_string()), Some(i64::MAX as u64))] + #[case::negative(json!(-5), None)] + #[case::fraction(json!(2.5), None)] + #[case::precise_fraction(json!("1.0000000000000001"), None)] + #[case::exponent(json!("1e3"), None)] + #[case::missing_fraction(json!("2."), None)] + #[case::repeated_separator(json!("2__0"), None)] + #[case::above_i64(json!(u64::MAX), None)] + #[case::float_above_i64(json!(9_223_372_036_854_775_808.0), None)] + #[case::word(json!("8k"), None)] + #[case::object(json!({}), None)] + fn counts_accept_whole_numbers_however_they_are_written( + #[case] written: Value, + #[case] expected: Option, + ) { + assert_eq!(count(written), expected); + } +} diff --git a/litellm-rust/crates/model-catalog/src/wire/mod.rs b/litellm-rust/crates/model-catalog/src/wire/mod.rs new file mode 100644 index 00000000000..e8f23b7bd0d --- /dev/null +++ b/litellm-rust/crates/model-catalog/src/wire/mod.rs @@ -0,0 +1,10 @@ +mod attributes; +mod blocks; +mod entry; +mod flat_key; +mod lax; + +pub use attributes::*; +pub use blocks::*; +pub use entry::*; +pub use flat_key::*; diff --git a/litellm-rust/crates/model-catalog/tests/catalog.rs b/litellm-rust/crates/model-catalog/tests/catalog.rs index d384dfea9a8..30c070cfb52 100644 --- a/litellm-rust/crates/model-catalog/tests/catalog.rs +++ b/litellm-rust/crates/model-catalog/tests/catalog.rs @@ -1,303 +1,405 @@ -use std::path::{Path, PathBuf}; +use std::collections::BTreeSet; -use litellm_model_catalog::{AliasIssue, Catalog, Error, IntegrityLimits, Mode, Provenance}; +use litellm_model_catalog::{ + BUNDLED_REGISTRY, Catalog, Error, ErrorKind, IntegrityLimits, Mode, Rejects, Written, +}; use rstest::{fixture, rstest}; -use serde_json::json; +use serde_json::{Map, Value, json}; const ALPHA_FIXTURE: &[u8] = br#"{ - "sample_spec":{"explanation":"example"}, + "sample_spec":{"explanation":"example","input_cost_per_token":"not checked"}, "fallback_generalizations":{"rules":[{"name":"family","pattern":"^new-","model_info":{"mode":"chat"}}]}, "Alpha":{"litellm_provider":"test","aliases":["short"],"price":0,"enabled":false, "optional":null,"unknown":{"nested":[1,{"x":true}]}} }"#; -#[fixture] -fn repo_root() -> PathBuf { - Path::new(env!("CARGO_MANIFEST_DIR")).join("../../..") -} - -#[fixture] -fn current_catalog(repo_root: PathBuf) -> Catalog { - let body = std::fs::read(repo_root.join("model_prices_and_context_window.json")).unwrap(); - Catalog::parse(&body, Provenance::default()).unwrap() -} - -#[fixture] -fn backup_catalog(repo_root: PathBuf) -> Catalog { - let body = std::fs::read(repo_root.join("litellm/model_prices_and_context_window_backup.json")) - .unwrap(); - Catalog::parse(&body, Provenance::default()).unwrap() -} - #[fixture] fn fixture_catalog() -> Catalog { - Catalog::parse( - ALPHA_FIXTURE, - Provenance { - source: Some("fixture".into()), - revision: Some("rev".into()), - etag: None, - }, - ) - .unwrap() + Catalog::parse(ALPHA_FIXTURE).unwrap().value +} + +fn loaded(document: Value) -> litellm_model_catalog::Loaded { + Catalog::parse(document.to_string().as_bytes()).unwrap() +} + +fn catalog(document: Value) -> Catalog { + loaded(document).value +} + +fn written(catalog: &Catalog) -> String { + let document: Map = catalog + .written() + .map(|(key, value)| { + let value = match value { + Written::Entry(fields) => serde_json::to_value(fields).unwrap(), + Written::Section(value) => value.clone(), + }; + (key.to_owned(), value) + }) + .collect(); + serde_json::to_string(&document).unwrap() +} + +fn report(error: &Error) -> String { + litellm_error::Report::new(error).to_string() +} + +fn error_text(document: Value) -> Vec { + loaded(document).errors.iter().map(report).collect() } #[rstest] #[case::decisions("decisions", Mode::Decisions)] #[case::legacy_evaluation("evaluation", Mode::Evaluation)] fn parses_decision_modes(#[case] mode: &str, #[case] expected: Mode) { - let source = json!({"Test":{"litellm_provider":"test","mode":mode}}).to_string(); - let catalog = Catalog::parse(source.as_bytes(), Provenance::default()).unwrap(); - + let catalog = catalog(json!({"Test": {"litellm_provider": "test", "mode": mode}})); assert_eq!( - catalog.lookup("Test").unwrap().entry.info().mode, + catalog + .lookup("Test") + .unwrap() + .entry + .model() + .unwrap() + .attributes + .mode, Some(expected) ); - assert_eq!(serde_json::to_value(expected).unwrap(), json!(mode)); } #[rstest] -#[ignore] -fn preserves_fields_and_metadata(fixture_catalog: Catalog) { - let catalog = fixture_catalog; - let entry = catalog.lookup("SHORT").unwrap(); - assert_eq!(entry.canonical_key, "Alpha"); - assert_eq!(entry.matched_key, "short"); - assert_eq!(entry.entry.field("price"), Some(&json!(0))); - assert_eq!(entry.entry.field("enabled"), Some(&json!(false))); - assert_eq!(entry.entry.field("optional"), Some(&json!(null))); - assert_eq!(entry.entry.field("missing"), None); +fn unknown_keys_are_kept_raw_and_listed_as_unrecognized(fixture_catalog: Catalog) { + let entry = fixture_catalog.lookup("Alpha").unwrap().entry; assert_eq!( - entry.entry.field("unknown"), + entry.field("unknown"), Some(&json!({"nested":[1,{"x":true}]})) ); - assert_eq!(entry.entry.field("aliases"), None); - assert_eq!(entry.entry.info().litellm_provider.as_deref(), Some("test")); assert_eq!( - catalog.sample_spec(), - Some(&json!({"explanation":"example"})) - ); - assert_eq!(catalog.fallback_rules().unwrap().len(), 1); - assert_eq!(catalog.provenance().revision.as_deref(), Some("rev")); - assert_eq!(catalog.model_count(), 1); -} - -#[rstest] -#[ignore] -fn snapshot_does_not_borrow_source() { - let mut source = ALPHA_FIXTURE.to_vec(); - let catalog = Catalog::parse(&source, Provenance::default()).unwrap(); - source.fill(b' '); - - let entry = catalog.lookup("short").unwrap(); - assert_eq!(entry.canonical_key, "Alpha"); - assert_eq!(entry.entry.field("price"), Some(&json!(0))); -} - -#[rstest] -#[case("Shared", "First")] -#[case("Second", "Second")] -#[case("shared", "Second")] -#[case("FIRST", "First")] -#[case("sHaReD", "Second")] -#[ignore] -fn alias_collisions_and_case_fallback_follow_entry_order( - #[case] lookup: &str, - #[case] expected: &str, -) { - let catalog = Catalog::parse( - br#"{ - "First":{"aliases":["Shared","Second","first"],"value":1}, - "Second":{"aliases":["Shared","sHaReD"],"value":2}, - "SHARED":{"value":3} - }"#, - Provenance::default(), - ) - .unwrap(); - assert_eq!(catalog.lookup(lookup).unwrap().canonical_key, expected); - assert_eq!(catalog.alias_count(), 3); - assert!( - catalog - .alias_issues() - .contains(&AliasIssue::CanonicalCollision { - model: "First".into(), - alias: "Second".into(), - }) - ); - assert!( - catalog - .alias_issues() - .contains(&AliasIssue::AliasCollision { - model: "Second".into(), - alias: "Shared".into(), - }) + fixture_catalog.unrecognized().collect::>(), + BTreeSet::from([ + ("Alpha", "enabled"), + ("Alpha", "price"), + ("Alpha", "unknown") + ]) ); } -#[test] -#[ignore] -fn json_entry_order_controls_alias_ownership_and_case_fallback() { - let forward = Catalog::parse( - br#"{ - "Alpha":{"aliases":["shared"]}, - "Beta":{"aliases":["shared"]}, - "Foo":{}, - "fOO":{} - }"#, - Provenance::default(), - ) - .unwrap(); - let reversed = Catalog::parse( - br#"{ - "fOO":{}, - "Foo":{}, - "Beta":{"aliases":["shared"]}, - "Alpha":{"aliases":["shared"]} - }"#, - Provenance::default(), - ) - .unwrap(); - - assert_eq!(forward.lookup("shared").unwrap().canonical_key, "Alpha"); - assert_eq!(reversed.lookup("shared").unwrap().canonical_key, "Beta"); - assert_eq!(forward.lookup("foo").unwrap().canonical_key, "fOO"); - assert_eq!(reversed.lookup("foo").unwrap().canonical_key, "Foo"); -} - -#[derive(Debug)] -enum ValidationOutcome { - Ok, - Shrunk, - BelowMinimum, - InvalidRatio, -} - #[rstest] -#[case( - IntegrityLimits { - reference_model_count: 2, - min_model_count: 1, - min_reference_ratio: 0.5, - }, - ValidationOutcome::Ok -)] -#[case( - IntegrityLimits { - reference_model_count: 3, - min_model_count: 1, - min_reference_ratio: 0.5, - }, - ValidationOutcome::Shrunk -)] -#[case( - IntegrityLimits { - reference_model_count: 0, - min_model_count: 2, - min_reference_ratio: 0.5, - }, - ValidationOutcome::BelowMinimum -)] -#[case( - IntegrityLimits { - reference_model_count: 0, - min_model_count: 0, - min_reference_ratio: f64::NAN, - }, - ValidationOutcome::InvalidRatio -)] -#[ignore] -fn integrity_uses_canonical_count_and_strict_shrink_boundary( - #[case] limits: IntegrityLimits, - #[case] expected: ValidationOutcome, -) { - let catalog = Catalog::parse( - br#"{"sample_spec":{},"fallback_generalizations":{"rules":[]},"a":{"aliases":["b","c"]}}"#, - Provenance::default(), - ) - .unwrap(); - let actual = catalog.validate(limits); - match expected { - ValidationOutcome::Ok => assert!(actual.is_ok()), - ValidationOutcome::Shrunk => { - assert!(matches!(actual, Err(Error::Shrunk { actual: 1, .. }))) - } - ValidationOutcome::BelowMinimum => { - assert!(matches!(actual, Err(Error::BelowMinimum { actual: 1, .. }))) - } - ValidationOutcome::InvalidRatio => assert!(matches!(actual, Err(Error::InvalidRatio))), - } -} - -#[derive(Debug)] -enum MalformedOutcome { - Empty, - Json, - EntryNotObject, -} - -#[rstest] -#[case::empty(b"{}", MalformedOutcome::Empty)] -#[case::invalid_json(b"{", MalformedOutcome::Json)] -#[case::entry_not_object(br#"{"a":1}"#, MalformedOutcome::EntryNotObject)] -#[case::fallback_rules_missing( - br#"{"fallback_generalizations":{},"a":{}}"#, - MalformedOutcome::Json -)] -#[ignore] -fn malformed_input_and_aliases_have_typed_outcomes( - #[case] body: &[u8], - #[case] expected: MalformedOutcome, -) { - let actual = Catalog::parse(body, Provenance::default()); - match expected { - MalformedOutcome::Empty => assert!(matches!(actual, Err(Error::Empty))), - MalformedOutcome::Json => assert!(matches!(actual, Err(Error::Json(_)))), - MalformedOutcome::EntryNotObject => { - assert!(matches!(actual, Err(Error::EntryNotObject { .. }))) - } - } -} - -#[rstest] -#[ignore] -fn invalid_aliases_are_reported_not_fatal() { - let catalog = Catalog::parse( - br#"{"a":{"aliases":"bad"},"b":{"aliases":[9,"ok"]}}"#, - Provenance::default(), - ) - .unwrap(); +fn sections_aliases_and_raw_fields_are_read_in_their_document_views(fixture_catalog: Catalog) { + let found = fixture_catalog.lookup("SHORT").unwrap(); + assert_eq!(found.canonical_key, "Alpha"); + assert_eq!(found.matched_key, "short"); + assert_eq!(found.entry.field("optional"), Some(&json!(null))); + assert_eq!(found.entry.field("aliases"), Some(&json!(["short"]))); assert_eq!( - catalog.alias_issues(), - &[ - AliasIssue::InvalidList { model: "a".into() }, - AliasIssue::InvalidName { model: "b".into() }, - ] + fixture_catalog.sample_spec(), + Some(&json!({"explanation":"example","input_cost_per_token":"not checked"})) + ); + assert_eq!(fixture_catalog.fallback_rules().unwrap()[0].name, "family"); + assert_eq!(fixture_catalog.model_count(), 1); + assert_eq!( + fixture_catalog + .served() + .map(|(key, _)| key) + .collect::>(), + ["Alpha", "short"] ); - assert_eq!(catalog.lookup("ok").unwrap().canonical_key, "b"); - assert!(catalog.lookup("missing").is_none()); } #[rstest] -#[ignore] -fn parses_current_and_packaged_catalogs_against_independent_baseline( - current_catalog: Catalog, - backup_catalog: Catalog, -) { - assert!(current_catalog.model_count() > 0); - assert!(backup_catalog.model_count() > 0); - assert!(current_catalog.sample_spec().is_some()); - assert!(backup_catalog.sample_spec().is_some()); - // Snapshot from 2026-09-23; the backup file mirrors the current file and cannot detect shrinkage. - const REFERENCE_MODEL_COUNT: usize = 4303; - current_catalog - .validate(IntegrityLimits { - reference_model_count: REFERENCE_MODEL_COUNT, - min_model_count: 50, - min_reference_ratio: 0.9, - }) - .unwrap(); - assert!(current_catalog.model_names().all(|name| { - let entry = current_catalog.lookup(name).unwrap().entry; - entry.info().litellm_provider.is_some() == entry.field("litellm_provider").is_some() +fn a_bad_entry_is_reported_by_model_and_key_and_kept_raw() { + let parsed = loaded(json!({ + "Good": {"litellm_provider": "test"}, + "Bad": {"litellm_provider": "test", "mode": "chatt", "input_cost_per_token": "free"}, + "Worse": {"mode": "chat"} })); + let catalog = parsed.value; + let problems: Vec = parsed.errors.iter().map(report).collect(); + assert_eq!(problems.len(), 3, "{problems:?}"); + assert!(problems[0].starts_with("Bad.mode: invalid value: unknown variant `chatt`")); + assert!(problems[1].starts_with("Bad.input_cost_per_token: invalid value")); + assert_eq!(problems[2], "Worse.litellm_provider: required"); + assert!(catalog.lookup("Good").unwrap().entry.model().is_some()); + assert!(catalog.lookup("Bad").unwrap().entry.model().is_none()); + assert_eq!(catalog.model_count(), 3); + assert!( + parsed + .errors + .iter() + .find(|error| error.rejects() == Rejects::Document) + .is_none() + ); +} + +#[rstest] +#[case::alias_names_a_model( + json!({"First": {"litellm_provider": "t", "aliases": ["Second"]}, "Second": {"litellm_provider": "t"}}), + r#"First: alias "Second" is already a model name"# +)] +#[case::alias_claimed_twice( + json!({"First": {"litellm_provider": "t", "aliases": ["shared"]}, "Second": {"litellm_provider": "t", "aliases": ["shared"]}}), + r#"Second: alias "shared" already belongs to First"# +)] +#[case::aliases_not_a_list( + json!({"a": {"litellm_provider": "t", "aliases": "bad"}}), + "a.aliases: invalid value: invalid type: string \"bad\", expected a sequence" +)] +#[case::entry_not_object( + json!({"a": 1, "valid": {"litellm_provider": "test"}}), + "a: must be a JSON object" +)] +#[case::fallback_rules_missing( + json!({"fallback_generalizations": {}, "a": {"litellm_provider": "t"}}), + "fallback_generalizations: invalid value: missing field `rules`" +)] +#[case::unknown_timezone( + json!({"a": {"litellm_provider": "t", "off_peak_pricing": { + "windows": [{"hours_utc": "00:00-01:00"}], + "weekday_timezone": "Not/AZone", + "input_cost_per_token": 1e-6 + }}}), + r#"a.off_peak_pricing.weekday_timezone: "Not/AZone" is not an IANA time zone"# +)] +fn document_level_problems_are_reported(#[case] document: Value, #[case] expected: &str) { + assert_eq!(error_text(document), [expected]); +} + +#[rstest] +fn escaped_row_paths_and_row_validation_share_the_parse_rules() { + let document = json!({ + "open/router~v1": {"litellm_provider": "test", "input_cost_per_token": -1} + }); + let parsed = loaded(document); + let fields = parsed + .value + .lookup("open/router~v1") + .unwrap() + .entry + .fields(); + let row_errors = Catalog::row_errors("open/router~v1", fields); + assert_eq!( + row_errors + .iter() + .map(|error| (error.path(), report(error))) + .collect::>(), + parsed + .errors + .iter() + .map(|error| (error.path(), report(error))) + .collect::>() + ); + assert_eq!( + row_errors[0].path(), + "/open~1router~0v1/input_cost_per_token" + ); + assert!(row_errors[0].kind() == ErrorKind::Invalid); +} + +#[rstest] +fn malformed_fallback_is_collected_while_models_are_read() { + let parsed = loaded(json!({ + "fallback_generalizations": {"rules": "bad"}, + "a": {"litellm_provider": "test"} + })); + assert_eq!(parsed.value.model_count(), 1); + assert_eq!(parsed.errors.len(), 1); + assert_eq!(parsed.errors[0].path(), "/fallback_generalizations"); +} + +#[rstest] +fn rejected_rows_and_their_aliases_are_removed_without_disturbing_written_order() { + let body = r#"{"before":{"litellm_provider":"test","aliases":["before-alias"]},"bad":{"litellm_provider":"test","aliases":["bad-alias"],"mode":"chatt"},"not-object":4,"sample_spec":{"keep":true},"after":{"litellm_provider":"test"}}"#; + let parsed = Catalog::parse(body.as_bytes()).unwrap(); + let (catalog, removed) = parsed.value.without_rejected(&parsed.errors); + + assert_eq!(removed.len(), 2); + assert_eq!( + catalog.lookup("before-alias").unwrap().canonical_key, + "before" + ); + assert_eq!( + catalog.lookup("BEFORE-ALIAS").unwrap().matched_key, + "before-alias" + ); + assert!(catalog.lookup("bad").is_none()); + assert!(catalog.lookup("BAD").is_none()); + assert!(catalog.lookup("Bad-Alias").is_none()); + assert_eq!(catalog.model_count(), 2); + assert_eq!( + written(&catalog), + r#"{"before":{"litellm_provider":"test","aliases":["before-alias"]},"sample_spec":{"keep":true},"after":{"litellm_provider":"test"}}"# + ); +} + +#[rstest] +fn invalid_empty_key_rows_are_removed() { + let parsed = loaded(json!({ + "": {"litellm_provider": "test", "mode": "chatt"}, + "good": {"litellm_provider": "test"} + })); + assert_eq!(parsed.errors[0].path(), "//mode"); + + let (catalog, removed) = parsed.value.without_rejected(&parsed.errors); + assert_eq!(catalog.model_count(), 1); + assert!(catalog.lookup("").is_none()); + assert!(catalog.lookup("good").is_some()); + assert_eq!(removed.len(), 1); +} + +#[rstest] +fn timezone_errors_reject_providerless_rows_without_rejecting_other_providerless_rows() { + let parsed = loaded(json!({ + "good": {"mode": "chat"}, + "bad": {"off_peak_pricing": { + "windows": [{"hours_utc": "00:00-01:00"}], + "weekday_timezone": "Not/AZone", + "input_cost_per_token": 1e-6 + }} + })); + assert_eq!(parsed.errors.len(), 3); + assert!( + parsed + .errors + .iter() + .any(|error| error.kind() == ErrorKind::UnknownTimezone) + ); + + let (catalog, removed) = parsed.value.without_rejected(&parsed.errors); + assert_eq!(catalog.model_count(), 1); + assert!(catalog.lookup("good").is_some()); + assert!(catalog.lookup("bad").is_none()); + assert_eq!(removed.len(), 2); +} + +#[rstest] +fn bundled_per_row_errors_match_document_errors() { + let parsed = Catalog::parse(BUNDLED_REGISTRY).unwrap(); + let row_errors: Vec<_> = parsed + .value + .written() + .flat_map(|(key, written)| match written { + Written::Entry(fields) => Catalog::row_errors(key, fields), + Written::Section(_) => Vec::new(), + }) + .map(|error| (error.path().to_owned(), report(&error))) + .collect(); + let document_errors: Vec<_> = parsed + .errors + .iter() + .map(|error| (error.path().to_owned(), report(error))) + .collect(); + + assert_eq!(row_errors, document_errors); +} + +#[rstest] +fn three_hundred_rows_filter_one_row_with_two_invalid_prices() { + let valid_rows: Map = (0..299) + .map(|index| { + ( + format!("valid-{index:03}"), + json!({"litellm_provider": "test"}), + ) + }) + .collect(); + let mut document = valid_rows; + document.insert( + "bad".to_owned(), + json!({ + "litellm_provider": "test", + "input_cost_per_token": -1, + "output_cost_per_token": -1 + }), + ); + let parsed = Catalog::parse(serde_json::to_vec(&document).unwrap().as_slice()).unwrap(); + assert_eq!(parsed.value.model_count(), 300); + assert_eq!(parsed.errors.len(), 2); + + let (catalog, removed) = parsed.value.without_rejected(&parsed.errors); + assert_eq!(catalog.model_count(), 299); + assert_eq!(removed.len(), 2); +} + +#[rstest] +fn integrity_limits_count_canonical_models_not_aliases() { + let fetched = catalog(json!({"a": {"litellm_provider": "t", "aliases": ["b", "c"]}})); + assert!( + fetched + .check_size(2, IntegrityLimits::new(1, 0.5).unwrap()) + .is_ok() + ); + assert_eq!( + fetched + .check_size(3, IntegrityLimits::new(1, 0.5).unwrap()) + .unwrap_err() + .kind(), + ErrorKind::Shrunk + ); + assert_eq!( + fetched + .check_size(2, IntegrityLimits::new(2, 0.5).unwrap()) + .unwrap_err() + .kind(), + ErrorKind::BelowMinimum + ); +} + +#[rstest] +#[case::empty(b"{}", "empty")] +#[case::only_sections(br#"{"sample_spec":{}}"#, "empty")] +#[case::invalid_json(b"{", "json")] +#[case::array_root_is_json_error(b"[]", "json")] +fn malformed_documents_are_rejected_whole(#[case] body: &[u8], #[case] kind: &str) { + let error = Catalog::parse(body).unwrap_err(); + assert_eq!(error.rejects(), Rejects::Document); + assert_eq!(<&str>::from(error.kind()), kind); +} + +#[rstest] +fn written_is_the_document_in_order_including_sections_and_bad_entries() { + let body = concat!( + r#"{"b":{"litellm_provider":"t","z":1,"a":null},"sample_spec":{"x":"y"},"n":4,"#, + r#""fallback_generalizations":{"rules":"bad"},"a":{"mode":"chatt","price":1.0}}"# + ); + let catalog = Catalog::parse(body.as_bytes()).unwrap().value; + assert_eq!(written(&catalog), body); + assert_eq!(catalog.model_count(), 2); +} + +#[rstest] +fn the_bundled_registry_parses_with_no_problems_and_every_key_understood() { + let loaded = Catalog::parse(BUNDLED_REGISTRY).unwrap(); + let catalog = loaded.value; + assert!(loaded.errors.is_empty(), "{:#?}", loaded.errors); + let unrecognized: Vec<_> = catalog.unrecognized().collect(); + assert!(unrecognized.is_empty(), "{unrecognized:#?}"); + assert!( + catalog + .fallback_rules() + .is_some_and(|rules| !rules.is_empty()) + ); + const REFERENCE_MODEL_COUNT: usize = 4303; + assert!(catalog.model_count() as f64 >= REFERENCE_MODEL_COUNT as f64 * 0.9); +} + +#[rstest] +fn the_bundled_registry_is_written_back_as_it_was_read() { + let document: Value = serde_json::from_slice(BUNDLED_REGISTRY).unwrap(); + assert_eq!( + written(&Catalog::parse(BUNDLED_REGISTRY).unwrap().value), + serde_json::to_string(&document).unwrap() + ); +} + +#[rstest] +fn a_bad_value_keeps_its_json_error_as_the_source_and_names_its_key_and_field() { + let parsed = + loaded(json!({"open/router": {"litellm_provider": "t", "input_cost_per_token": "free"}})); + let error = &parsed.errors[0]; + let cause = litellm_error::find::(error) + .expect("the JSON error is in the chain") + .to_string(); + assert_eq!(report(error).matches(&cause).count(), 1); + assert!(!error.to_string().contains(&cause)); + assert_eq!(error.key().as_deref(), Some("open/router")); + assert_eq!(error.field().as_deref(), Some("input_cost_per_token")); } diff --git a/litellm-rust/crates/model-catalog/tests/generate_price_lookup_reference.py b/litellm-rust/crates/model-catalog/tests/generate_price_lookup_reference.py new file mode 100644 index 00000000000..e49b635ff4a --- /dev/null +++ b/litellm-rust/crates/model-catalog/tests/generate_price_lookup_reference.py @@ -0,0 +1,91 @@ +import itertools +import json +import re +import subprocess +from pathlib import Path +from typing import Final + +from litellm.litellm_core_utils.llm_cost_calc.utils import _get_token_base_cost +from litellm.types.utils import Usage + +REPO_ROOT: Final = Path(__file__).resolve().parents[4] +TOKEN_RATE_KEY: Final = re.compile( + r"^(input_cost_per_token|output_cost_per_token|cache_read_input_token_cost" + r"|cache_creation_input_token_cost(_above_1hr)?)" + r"(_above_(\d+)(k?)_tokens)?(_(priority|flex|ultrafast|balanced|batches|decisions))?$" +) +SERVICE_TIERS: Final = ("priority", "flex", "ultrafast", "balanced") +LARGEST_PROMPT: Final = 10_000_000 +INCLUSIVE_PROVIDERS: Final = frozenset({"xai"}) + + +def token_rates(entry: dict[str, object]) -> dict[str, object]: + return { + key: value + for key, value in entry.items() + if value is not None and (key == "tiered_pricing" or TOKEN_RATE_KEY.match(key)) + } + + +def band_floors(info: dict[str, object]) -> frozenset[int]: + matches: Final = (TOKEN_RATE_KEY.match(key) for key in info) + flat: Final = frozenset( + int(match.group(4)) * (1000 if match.group(5) else 1) + for match in matches + if match is not None and match.group(4) is not None + ) + rows: Final = info.get("tiered_pricing") + tiers: Final = ( + frozenset(int(row["range"][0]) for row in rows if isinstance(row, dict) and "range" in row) + if isinstance(rows, list) + else frozenset() + ) + return flat | tiers + + +def tiers_in(info: dict[str, object]) -> tuple[str | None, ...]: + return (None, *(tier for tier in SERVICE_TIERS if any(key.endswith(f"_{tier}") for key in info))) + + +def prompts(info: dict[str, object]) -> tuple[int, ...]: + floors: Final = band_floors(info) + return tuple(sorted(({1, LARGEST_PROMPT} | floors | {floor + 1 for floor in floors}) - {0})) + + +def check(info: dict[str, object], tier: str | None, prompt: int, inclusive: bool) -> list[object]: + usage: Final = Usage(prompt_tokens=prompt, completion_tokens=1, total_tokens=prompt + 1) + rates: Final = _get_token_base_cost(info, usage, service_tier=tier, threshold_is_inclusive=inclusive) + one_hour: Final = rates.cache_creation_rates_by_ttl.get("1hr", rates.cache_creation_rate) + return [tier, prompt, [rates.input_rate, rates.output_rate, rates.cache_creation_rate, one_hour, rates.cache_read_rate]] + + +def shape(info: dict[str, object], inclusive: bool) -> str: + rows: Final = info.get("tiered_pricing") + row_keys: Final = [sorted(row) for row in rows if isinstance(row, dict)] if isinstance(rows, list) else [] + return json.dumps([sorted(info), row_keys, inclusive]) + + +def checks(info: dict[str, object], inclusive: bool) -> list[list[object]]: + return [check(info, tier, prompt, inclusive) for tier, prompt in itertools.product(tiers_in(info), prompts(info))] + + +def main() -> None: + registry: Final = json.loads((REPO_ROOT / "model_prices_and_context_window.json").read_text()) + entries: Final = tuple( + (token_rates(entry), entry.get("litellm_provider") in INCLUSIVE_PROVIDERS) + for name, entry in registry.items() + if name not in ("sample_spec", "fallback_generalizations") and isinstance(entry, dict) + ) + by_shape: Final = {shape(info, inclusive): (info, inclusive) for info, inclusive in reversed(entries) if info} + models: Final = tuple( + json.dumps({"info": info, "inclusive": inclusive, "checks": checks(info, inclusive)}) + for info, inclusive in by_shape.values() + ) + revision: Final = subprocess.check_output(("git", "rev-parse", "HEAD"), text=True, cwd=REPO_ROOT).strip() + Path(__file__).with_name("price_lookup_reference.json").write_text( + f'{{"python_reference_commit": "{revision}", "models": [\n' + ",\n".join(models) + "\n]}\n" + ) + + +if __name__ == "__main__": + main() diff --git a/litellm-rust/crates/model-catalog/tests/price_lookup.rs b/litellm-rust/crates/model-catalog/tests/price_lookup.rs new file mode 100644 index 00000000000..cc16e9f23e9 --- /dev/null +++ b/litellm-rust/crates/model-catalog/tests/price_lookup.rs @@ -0,0 +1,89 @@ +use litellm_cost::money::UnitPrice; +use litellm_cost::price::{ + CacheReadMode, CacheTtl, InputUnit, Meter, OutputUnit, Prices, ServiceTier, TokenKind, +}; +use litellm_model_catalog::wire::{EntryWire, Fields}; +use rstest::rstest; +use serde::Deserialize; + +#[derive(Deserialize)] +struct Reference { + models: Vec, +} + +#[derive(Deserialize)] +struct ReferenceModel { + info: Fields, + inclusive: bool, + checks: Vec<(Option, u64, [f64; 5])>, +} + +const METERS: [Meter; 5] = [ + Meter::Input(InputUnit::Token(TokenKind::Text)), + Meter::Output(OutputUnit::Token(TokenKind::Text)), + Meter::CacheWrite { + kind: TokenKind::Text, + ttl: CacheTtl::Default, + }, + Meter::CacheWrite { + kind: TokenKind::Text, + ttl: CacheTtl::OneHour, + }, + Meter::CacheRead { + kind: TokenKind::Text, + mode: CacheReadMode::Implicit, + }, +]; + +fn service_tier(name: Option<&str>) -> ServiceTier { + match name { + None => ServiceTier::Standard, + Some("priority") => ServiceTier::Priority, + Some("flex") => ServiceTier::Flex, + Some("ultrafast") => ServiceTier::Ultrafast, + Some("balanced") => ServiceTier::Balanced, + Some(other) => panic!("the reference uses an unknown service tier {other}"), + } +} + +fn mismatch( + prices: &Prices, + (tier_name, prompt, python): &(Option, u64, [f64; 5]), +) -> Option { + let tier = service_tier(tier_name.as_deref()); + let band = prices.band(tier, *prompt); + let rust = METERS + .each_ref() + .map(|meter| prices.rate(meter, tier, band).map_or(0.0, UnitPrice::get)); + (rust != *python) + .then(|| format!("tier={tier_name:?} prompt={prompt}: rust {rust:?}, python {python:?}")) +} + +#[rstest] +fn band_and_rate_match_python_token_base_cost() { + let reference: Reference = + serde_json::from_str(include_str!("price_lookup_reference.json")).unwrap(); + + let mismatches: Vec = reference + .models + .iter() + .flat_map(|model| { + let provider = if model.inclusive { "xai" } else { "openai" }; + let fields: Fields = model + .info + .clone() + .into_iter() + .chain([("litellm_provider".to_owned(), provider.into())]) + .collect(); + let prices = Prices::try_from(&EntryWire::read(&fields)).unwrap(); + model + .checks + .iter() + .filter_map(move |check| mismatch(&prices, check)) + .map(move |found| format!("{:?} {found}", model.info)) + }) + .collect(); + + assert!(reference.models.len() > 50); + assert!(mismatches.is_empty(), "{mismatches:#?}"); +} diff --git a/litellm-rust/crates/model-catalog/tests/price_lookup_reference.json b/litellm-rust/crates/model-catalog/tests/price_lookup_reference.json new file mode 100644 index 00000000000..4b87b696723 --- /dev/null +++ b/litellm-rust/crates/model-catalog/tests/price_lookup_reference.json @@ -0,0 +1,89 @@ +{"python_reference_commit": "c7c895db456eda349a7be9661b4b21f578f9b589", "models": [ +{"info": {"cache_creation_input_token_cost": 1.25e-05, "cache_creation_input_token_cost_above_272k_tokens": 2.5e-05, "cache_read_input_token_cost": 1e-06, "cache_read_input_token_cost_above_272k_tokens": 2e-06, "input_cost_per_token": 1e-05, "input_cost_per_token_above_272k_tokens": 2e-05, "output_cost_per_token": 5e-05, "output_cost_per_token_above_272k_tokens": 7.5e-05}, "inclusive": false, "checks": [[null, 1, [1e-05, 5e-05, 1.25e-05, 1.25e-05, 1e-06]], [null, 272000, [1e-05, 5e-05, 1.25e-05, 1.25e-05, 1e-06]], [null, 272001, [2e-05, 7.5e-05, 2.5e-05, 2.5e-05, 2e-06]], [null, 10000000, [2e-05, 7.5e-05, 2.5e-05, 2.5e-05, 2e-06]]]}, +{"info": {"cache_read_input_token_cost": 1e-07, "input_cost_per_token": 1e-06, "output_cost_per_token": 3e-06}, "inclusive": false, "checks": [[null, 1, [1e-06, 3e-06, 1e-06, 1e-06, 1e-07]], [null, 10000000, [1e-06, 3e-06, 1e-06, 1e-06, 1e-07]]]}, +{"info": {"input_cost_per_token": 1.25e-05, "output_cost_per_token": 1.25e-05}, "inclusive": false, "checks": [[null, 1, [1.25e-05, 1.25e-05, 1.25e-05, 1.25e-05, 1.25e-05]], [null, 10000000, [1.25e-05, 1.25e-05, 1.25e-05, 1.25e-05, 1.25e-05]]]}, +{"info": {"cache_creation_input_token_cost": 2.5e-06, "cache_creation_input_token_cost_above_272k_tokens": 5e-06, "cache_creation_input_token_cost_above_272k_tokens_flex": 2.5e-06, "cache_creation_input_token_cost_above_272k_tokens_priority": 1e-05, "cache_creation_input_token_cost_flex": 1.25e-06, "cache_creation_input_token_cost_priority": 5e-06, "cache_read_input_token_cost": 2e-07, "cache_read_input_token_cost_above_272k_tokens": 4e-07, "cache_read_input_token_cost_above_272k_tokens_flex": 2e-07, "cache_read_input_token_cost_above_272k_tokens_priority": 8e-07, "cache_read_input_token_cost_batches": 1e-07, "cache_read_input_token_cost_above_272k_tokens_batches": 2e-07, "cache_creation_input_token_cost_batches": 1.25e-06, "cache_creation_input_token_cost_above_272k_tokens_batches": 2.5e-06, "cache_read_input_token_cost_flex": 1e-07, "cache_read_input_token_cost_priority": 4e-07, "input_cost_per_token": 2e-06, "input_cost_per_token_above_272k_tokens": 4e-06, "input_cost_per_token_above_272k_tokens_flex": 2e-06, "input_cost_per_token_above_272k_tokens_priority": 8e-06, "input_cost_per_token_batches": 1e-06, "input_cost_per_token_above_272k_tokens_batches": 2e-06, "input_cost_per_token_flex": 1e-06, "input_cost_per_token_priority": 4e-06, "output_cost_per_token": 1e-05, "output_cost_per_token_above_272k_tokens": 1.5e-05, "output_cost_per_token_above_272k_tokens_flex": 7.5e-06, "output_cost_per_token_above_272k_tokens_priority": 3e-05, "output_cost_per_token_batches": 5e-06, "output_cost_per_token_above_272k_tokens_batches": 7.5e-06, "output_cost_per_token_flex": 5e-06, "output_cost_per_token_priority": 2e-05}, "inclusive": false, "checks": [[null, 1, [2e-06, 1e-05, 2.5e-06, 2.5e-06, 2e-07]], [null, 272000, [2e-06, 1e-05, 2.5e-06, 2.5e-06, 2e-07]], [null, 272001, [4e-06, 1.5e-05, 5e-06, 5e-06, 4e-07]], [null, 10000000, [4e-06, 1.5e-05, 5e-06, 5e-06, 4e-07]], ["priority", 1, [4e-06, 2e-05, 5e-06, 5e-06, 4e-07]], ["priority", 272000, [4e-06, 2e-05, 5e-06, 5e-06, 4e-07]], ["priority", 272001, [8e-06, 3e-05, 1e-05, 1e-05, 8e-07]], ["priority", 10000000, [8e-06, 3e-05, 1e-05, 1e-05, 8e-07]], ["flex", 1, [1e-06, 5e-06, 1.25e-06, 1.25e-06, 1e-07]], ["flex", 272000, [1e-06, 5e-06, 1.25e-06, 1.25e-06, 1e-07]], ["flex", 272001, [2e-06, 7.5e-06, 2.5e-06, 2.5e-06, 2e-07]], ["flex", 10000000, [2e-06, 7.5e-06, 2.5e-06, 2.5e-06, 2e-07]]]}, +{"info": {"cache_creation_input_token_cost": 1.25e-06, "cache_creation_input_token_cost_above_1hr": 2e-06, "cache_creation_input_token_cost_batches": 6.25e-07, "cache_read_input_token_cost": 1e-07, "cache_read_input_token_cost_batches": 5e-08, "input_cost_per_token": 1e-06, "input_cost_per_token_batches": 5e-07, "output_cost_per_token": 5e-06, "output_cost_per_token_batches": 2.5e-06}, "inclusive": false, "checks": [[null, 1, [1e-06, 5e-06, 1.25e-06, 2e-06, 1e-07]], [null, 10000000, [1e-06, 5e-06, 1.25e-06, 2e-06, 1e-07]]]}, +{"info": {"input_cost_per_token_above_100k_tokens_batches": 2.5e-07, "output_cost_per_token_above_100k_tokens_batches": 1.25e-06, "cache_creation_input_token_cost_above_100k_tokens_batches": 3.125e-07, "cache_read_input_token_cost_above_100k_tokens_batches": 2.5e-08, "cache_creation_input_token_cost": 1.25e-07, "cache_creation_input_token_cost_above_1hr": 2e-07, "cache_creation_input_token_cost_batches": 6.25e-08, "cache_read_input_token_cost": 1e-08, "cache_read_input_token_cost_batches": 5e-09, "input_cost_per_token": 1e-07, "input_cost_per_token_batches": 5e-08, "output_cost_per_token": 5e-07, "output_cost_per_token_batches": 2.5e-07, "input_cost_per_token_above_100k_tokens": 5e-07, "output_cost_per_token_above_100k_tokens": 2.5e-06, "cache_creation_input_token_cost_above_100k_tokens": 6.25e-07, "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1e-06, "cache_read_input_token_cost_above_100k_tokens": 5e-08}, "inclusive": false, "checks": [[null, 1, [1e-07, 5e-07, 1.25e-07, 2e-07, 1e-08]], [null, 100000, [1e-07, 5e-07, 1.25e-07, 2e-07, 1e-08]], [null, 100001, [5e-07, 2.5e-06, 6.25e-07, 1e-06, 5e-08]], [null, 10000000, [5e-07, 2.5e-06, 6.25e-07, 1e-06, 5e-08]]]}, +{"info": {"input_cost_per_token": 1.0000000000000001e-07, "output_cost_per_token": 5e-07, "cache_read_input_token_cost": 1e-08, "cache_creation_input_token_cost": 1.25e-07, "tiered_pricing": [{"range": [0, 272000], "input_cost_per_token": 1.0000000000000001e-07, "output_cost_per_token": 5e-07, "cache_read_input_token_cost": 1e-08, "cache_creation_input_token_cost": 1.25e-07}, {"range": [272000, 1050000], "input_cost_per_token": 2.0000000000000002e-07, "output_cost_per_token": 7.5e-07, "cache_read_input_token_cost": 2e-08, "cache_creation_input_token_cost": 2.5e-07}]}, "inclusive": false, "checks": [[null, 1, [1.0000000000000001e-07, 5e-07, 1.25e-07, 1.25e-07, 1e-08]], [null, 272000, [1.0000000000000001e-07, 5e-07, 1.25e-07, 1.25e-07, 1e-08]], [null, 272001, [2.0000000000000002e-07, 7.5e-07, 2.5e-07, 2.5e-07, 2e-08]], [null, 10000000, [2.0000000000000002e-07, 7.5e-07, 2.5e-07, 2.5e-07, 2e-08]]]}, +{"info": {"cache_creation_input_token_cost": 1e-06, "cache_read_input_token_cost": 8e-08, "input_cost_per_token": 8e-07, "output_cost_per_token": 4e-06}, "inclusive": false, "checks": [[null, 1, [8e-07, 4e-06, 1e-06, 1e-06, 8e-08]], [null, 10000000, [8e-07, 4e-06, 1e-06, 1e-06, 8e-08]]]}, +{"info": {"input_cost_per_token": 2e-06, "output_cost_per_token": 6e-06, "cache_read_input_token_cost": 5e-07, "tiered_pricing": [{"range": [0, 200000], "input_cost_per_token": 2e-06, "output_cost_per_token": 6e-06, "cache_read_input_token_cost": 5e-07}, {"range": [200000, 500000], "input_cost_per_token": 4e-06, "output_cost_per_token": 1.2e-05, "cache_read_input_token_cost": 1e-06}]}, "inclusive": false, "checks": [[null, 1, [2e-06, 6e-06, 2e-06, 2e-06, 5e-07]], [null, 200000, [2e-06, 6e-06, 2e-06, 2e-06, 5e-07]], [null, 200001, [4e-06, 1.2e-05, 4e-06, 4e-06, 1e-06]], [null, 10000000, [4e-06, 1.2e-05, 4e-06, 4e-06, 1e-06]]]}, +{"info": {"cache_read_input_token_cost": 7.5e-08, "cache_read_input_token_cost_priority": 1.5e-07, "input_cost_per_token": 7.5e-07, "input_cost_per_token_priority": 1.5e-06, "output_cost_per_token": 4.5e-06, "output_cost_per_token_priority": 9e-06}, "inclusive": false, "checks": [[null, 1, [7.5e-07, 4.5e-06, 7.5e-07, 7.5e-07, 7.5e-08]], [null, 10000000, [7.5e-07, 4.5e-06, 7.5e-07, 7.5e-07, 7.5e-08]], ["priority", 1, [1.5e-06, 9e-06, 1.5e-06, 1.5e-06, 1.5e-07]], ["priority", 10000000, [1.5e-06, 9e-06, 1.5e-06, 1.5e-06, 1.5e-07]]]}, +{"info": {"cache_read_input_token_cost": 2e-07, "cache_read_input_token_cost_above_200k_tokens": 4e-07, "input_cost_per_token": 1.25e-06, "input_cost_per_token_above_200k_tokens": 2.5e-06, "output_cost_per_token": 2.5e-06, "output_cost_per_token_above_200k_tokens": 5e-06}, "inclusive": false, "checks": [[null, 1, [1.25e-06, 2.5e-06, 1.25e-06, 1.25e-06, 2e-07]], [null, 200000, [1.25e-06, 2.5e-06, 1.25e-06, 1.25e-06, 2e-07]], [null, 200001, [2.5e-06, 5e-06, 2.5e-06, 2.5e-06, 4e-07]], [null, 10000000, [2.5e-06, 5e-06, 2.5e-06, 2.5e-06, 4e-07]]]}, +{"info": {"cache_creation_input_token_cost": 1.25e-06, "cache_creation_input_token_cost_above_1hr": 2e-06, "cache_read_input_token_cost": 1e-07, "input_cost_per_token": 1e-06, "output_cost_per_token": 5e-06}, "inclusive": false, "checks": [[null, 1, [1e-06, 5e-06, 1.25e-06, 2e-06, 1e-07]], [null, 10000000, [1e-06, 5e-06, 1.25e-06, 2e-06, 1e-07]]]}, +{"info": {"cache_read_input_token_cost": 8.25e-08, "cache_read_input_token_cost_flex": 4.125e-08, "cache_read_input_token_cost_priority": 1.44375e-07, "input_cost_per_token": 3.3e-07, "input_cost_per_token_batches": 1.65e-07, "input_cost_per_token_flex": 1.65e-07, "input_cost_per_token_priority": 5.775e-07, "output_cost_per_token": 2.75e-06, "output_cost_per_token_batches": 1.375e-06, "output_cost_per_token_flex": 1.375e-06, "output_cost_per_token_priority": 4.8125e-06}, "inclusive": false, "checks": [[null, 1, [3.3e-07, 2.75e-06, 3.3e-07, 3.3e-07, 8.25e-08]], [null, 10000000, [3.3e-07, 2.75e-06, 3.3e-07, 3.3e-07, 8.25e-08]], ["priority", 1, [5.775e-07, 4.8125e-06, 5.775e-07, 5.775e-07, 1.44375e-07]], ["priority", 10000000, [5.775e-07, 4.8125e-06, 5.775e-07, 5.775e-07, 1.44375e-07]], ["flex", 1, [1.65e-07, 1.375e-06, 1.65e-07, 1.65e-07, 4.125e-08]], ["flex", 10000000, [1.65e-07, 1.375e-06, 1.65e-07, 1.65e-07, 4.125e-08]]]}, +{"info": {"cache_read_input_token_cost": 7.5e-08, "cache_read_input_token_cost_batches": 3.75e-08, "cache_read_input_token_cost_flex": 3.75e-08, "cache_read_input_token_cost_priority": 1.5e-07, "input_cost_per_token": 7.5e-07, "output_cost_per_token": 4.5e-06, "input_cost_per_token_batches": 3.75e-07, "input_cost_per_token_flex": 3.75e-07, "input_cost_per_token_priority": 1.5e-06, "output_cost_per_token_batches": 2.25e-06, "output_cost_per_token_flex": 2.25e-06, "output_cost_per_token_priority": 9e-06}, "inclusive": false, "checks": [[null, 1, [7.5e-07, 4.5e-06, 7.5e-07, 7.5e-07, 7.5e-08]], [null, 10000000, [7.5e-07, 4.5e-06, 7.5e-07, 7.5e-07, 7.5e-08]], ["priority", 1, [1.5e-06, 9e-06, 1.5e-06, 1.5e-06, 1.5e-07]], ["priority", 10000000, [1.5e-06, 9e-06, 1.5e-06, 1.5e-06, 1.5e-07]], ["flex", 1, [3.75e-07, 2.25e-06, 3.75e-07, 3.75e-07, 3.75e-08]], ["flex", 10000000, [3.75e-07, 2.25e-06, 3.75e-07, 3.75e-07, 3.75e-08]]]}, +{"info": {"cache_read_input_token_cost": 5.5e-09, "input_cost_per_token": 5.5e-08, "input_cost_per_token_batches": 2.75e-08, "output_cost_per_token": 4.4e-07, "output_cost_per_token_batches": 2.2e-07, "cache_read_input_token_cost_batches": 2.75e-09}, "inclusive": false, "checks": [[null, 1, [5.5e-08, 4.4e-07, 5.5e-08, 5.5e-08, 5.5e-09]], [null, 10000000, [5.5e-08, 4.4e-07, 5.5e-08, 5.5e-08, 5.5e-09]]]}, +{"info": {"output_cost_per_token": 0.0}, "inclusive": false, "checks": [[null, 1, [0.0, 0.0, 0.0, 0.0, 0.0]], [null, 10000000, [0.0, 0.0, 0.0, 0.0, 0.0]]]}, +{"info": {"input_cost_per_token_above_100k_tokens": 5e-07, "output_cost_per_token_above_100k_tokens": 2.5e-06, "cache_creation_input_token_cost_above_100k_tokens": 6.25e-07, "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1e-06, "cache_read_input_token_cost_above_100k_tokens": 5e-08, "cache_creation_input_token_cost": 1.25e-07, "cache_creation_input_token_cost_above_1hr": 2e-07, "cache_creation_input_token_cost_batches": 6.25e-08, "cache_read_input_token_cost": 1e-08, "cache_read_input_token_cost_batches": 5e-09, "input_cost_per_token": 1e-07, "input_cost_per_token_batches": 5e-08, "output_cost_per_token": 5e-07, "output_cost_per_token_batches": 2.5e-07}, "inclusive": false, "checks": [[null, 1, [1e-07, 5e-07, 1.25e-07, 2e-07, 1e-08]], [null, 100000, [1e-07, 5e-07, 1.25e-07, 2e-07, 1e-08]], [null, 100001, [5e-07, 2.5e-06, 6.25e-07, 1e-06, 5e-08]], [null, 10000000, [5e-07, 2.5e-06, 6.25e-07, 1e-06, 5e-08]]]}, +{"info": {"input_cost_per_token_above_100k_tokens": 5.5e-07, "output_cost_per_token_above_100k_tokens": 2.75e-06, "cache_creation_input_token_cost_above_100k_tokens": 6.875e-07, "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1.1e-06, "cache_read_input_token_cost_above_100k_tokens": 5.5e-08, "cache_creation_input_token_cost": 1.375e-07, "cache_creation_input_token_cost_above_1hr": 2.2e-07, "cache_read_input_token_cost": 1.1e-08, "input_cost_per_token": 1.1e-07, "output_cost_per_token": 5.5e-07}, "inclusive": false, "checks": [[null, 1, [1.1e-07, 5.5e-07, 1.375e-07, 2.2e-07, 1.1e-08]], [null, 100000, [1.1e-07, 5.5e-07, 1.375e-07, 2.2e-07, 1.1e-08]], [null, 100001, [5.5e-07, 2.75e-06, 6.875e-07, 1.1e-06, 5.5e-08]], [null, 10000000, [5.5e-07, 2.75e-06, 6.875e-07, 1.1e-06, 5.5e-08]]]}, +{"info": {"cache_creation_input_token_cost": 4.125e-06, "cache_creation_input_token_cost_flex": 2.0625e-06, "cache_creation_input_token_cost_priority": 7.21875e-06, "cache_read_input_token_cost": 3.3e-07, "cache_read_input_token_cost_flex": 1.65e-07, "cache_read_input_token_cost_priority": 5.775e-07, "input_cost_per_token": 3.3e-06, "input_cost_per_token_flex": 1.65e-06, "input_cost_per_token_priority": 5.775e-06, "output_cost_per_token": 1.65e-05, "output_cost_per_token_flex": 8.25e-06, "output_cost_per_token_priority": 2.8875e-05}, "inclusive": false, "checks": [[null, 1, [3.3e-06, 1.65e-05, 4.125e-06, 4.125e-06, 3.3e-07]], [null, 10000000, [3.3e-06, 1.65e-05, 4.125e-06, 4.125e-06, 3.3e-07]], ["priority", 1, [5.775e-06, 2.8875e-05, 7.21875e-06, 7.21875e-06, 5.775e-07]], ["priority", 10000000, [5.775e-06, 2.8875e-05, 7.21875e-06, 7.21875e-06, 5.775e-07]], ["flex", 1, [1.65e-06, 8.25e-06, 2.0625e-06, 2.0625e-06, 1.65e-07]], ["flex", 10000000, [1.65e-06, 8.25e-06, 2.0625e-06, 2.0625e-06, 1.65e-07]]]}, +{"info": {"cache_read_input_token_cost": 1.375e-07, "cache_read_input_token_cost_priority": 2.75e-07, "input_cost_per_token": 1.375e-06, "input_cost_per_token_batches": 6.875e-07, "input_cost_per_token_priority": 2.75e-06, "output_cost_per_token": 1.1e-05, "output_cost_per_token_batches": 5.5e-06, "output_cost_per_token_priority": 2.2e-05, "cache_read_input_token_cost_batches": 6.875e-08}, "inclusive": false, "checks": [[null, 1, [1.375e-06, 1.1e-05, 1.375e-06, 1.375e-06, 1.375e-07]], [null, 10000000, [1.375e-06, 1.1e-05, 1.375e-06, 1.375e-06, 1.375e-07]], ["priority", 1, [2.75e-06, 2.2e-05, 2.75e-06, 2.75e-06, 2.75e-07]], ["priority", 10000000, [2.75e-06, 2.2e-05, 2.75e-06, 2.75e-06, 2.75e-07]]]}, +{"info": {"input_cost_per_token": 5e-06, "input_cost_per_token_batches": 2.5e-06, "output_cost_per_token": 1.5e-05, "output_cost_per_token_batches": 7.5e-06}, "inclusive": false, "checks": [[null, 1, [5e-06, 1.5e-05, 5e-06, 5e-06, 5e-06]], [null, 10000000, [5e-06, 1.5e-05, 5e-06, 5e-06, 5e-06]]]}, +{"info": {"cache_creation_input_token_cost": 2.1e-06, "cache_creation_input_token_cost_batches": 1.05e-06, "cache_creation_input_token_cost_flex": 1.05e-06, "cache_creation_input_token_cost_priority": 3.675e-06, "cache_read_input_token_cost": 3.12e-07, "cache_read_input_token_cost_batches": 1.56e-07, "cache_read_input_token_cost_flex": 1.56e-07, "cache_read_input_token_cost_priority": 5.46e-07, "input_cost_per_token": 1.68e-06, "input_cost_per_token_batches": 8.4e-07, "input_cost_per_token_flex": 8.4e-07, "input_cost_per_token_priority": 2.94e-06, "output_cost_per_token": 5.28e-06, "output_cost_per_token_batches": 2.64e-06, "output_cost_per_token_flex": 2.64e-06, "output_cost_per_token_priority": 9.24e-06}, "inclusive": false, "checks": [[null, 1, [1.68e-06, 5.28e-06, 2.1e-06, 2.1e-06, 3.12e-07]], [null, 10000000, [1.68e-06, 5.28e-06, 2.1e-06, 2.1e-06, 3.12e-07]], ["priority", 1, [2.94e-06, 9.24e-06, 3.675e-06, 3.675e-06, 5.46e-07]], ["priority", 10000000, [2.94e-06, 9.24e-06, 3.675e-06, 3.675e-06, 5.46e-07]], ["flex", 1, [8.4e-07, 2.64e-06, 1.05e-06, 1.05e-06, 1.56e-07]], ["flex", 10000000, [8.4e-07, 2.64e-06, 1.05e-06, 1.05e-06, 1.56e-07]]]}, +{"info": {"input_cost_per_token": 1.1e-05, "input_cost_per_token_above_272k_tokens": 2.2e-05, "cache_creation_input_token_cost": 1.375e-05, "cache_creation_input_token_cost_above_272k_tokens": 2.75e-05, "cache_creation_input_token_cost_above_272k_tokens_ultrafast": 0.000165, "cache_creation_input_token_cost_ultrafast": 8.25e-05, "cache_read_input_token_cost": 1.1e-06, "cache_read_input_token_cost_above_272k_tokens": 2.2e-06, "cache_read_input_token_cost_above_272k_tokens_ultrafast": 1.32e-05, "cache_read_input_token_cost_ultrafast": 6.6e-06, "input_cost_per_token_above_272k_tokens_ultrafast": 0.000132, "input_cost_per_token_ultrafast": 6.6e-05, "output_cost_per_token": 5.5e-05, "output_cost_per_token_above_272k_tokens": 8.25e-05, "output_cost_per_token_above_272k_tokens_ultrafast": 0.000495, "output_cost_per_token_ultrafast": 0.00033}, "inclusive": false, "checks": [[null, 1, [1.1e-05, 5.5e-05, 1.375e-05, 1.375e-05, 1.1e-06]], [null, 272000, [1.1e-05, 5.5e-05, 1.375e-05, 1.375e-05, 1.1e-06]], [null, 272001, [2.2e-05, 8.25e-05, 2.75e-05, 2.75e-05, 2.2e-06]], [null, 10000000, [2.2e-05, 8.25e-05, 2.75e-05, 2.75e-05, 2.2e-06]], ["ultrafast", 1, [6.6e-05, 0.00033, 8.25e-05, 8.25e-05, 6.6e-06]], ["ultrafast", 272000, [6.6e-05, 0.00033, 8.25e-05, 8.25e-05, 6.6e-06]], ["ultrafast", 272001, [0.000132, 0.000495, 0.000165, 0.000165, 1.32e-05]], ["ultrafast", 10000000, [0.000132, 0.000495, 0.000165, 0.000165, 1.32e-05]]]}, +{"info": {"cache_creation_input_token_cost": 1.25e-05, "cache_creation_input_token_cost_above_272k_tokens": 2.5e-05, "cache_creation_input_token_cost_above_272k_tokens_flex": 1.25e-05, "cache_creation_input_token_cost_above_272k_tokens_priority": 5e-05, "cache_creation_input_token_cost_flex": 6.25e-06, "cache_creation_input_token_cost_priority": 2.5e-05, "cache_read_input_token_cost": 1e-06, "cache_read_input_token_cost_above_272k_tokens": 2e-06, "cache_read_input_token_cost_above_272k_tokens_flex": 1e-06, "cache_read_input_token_cost_above_272k_tokens_priority": 4e-06, "cache_read_input_token_cost_batches": 5e-07, "cache_read_input_token_cost_above_272k_tokens_batches": 1e-06, "cache_creation_input_token_cost_batches": 6.25e-06, "cache_creation_input_token_cost_above_272k_tokens_batches": 1.25e-05, "cache_creation_input_token_cost_above_272k_tokens_ultrafast": 0.00015, "cache_creation_input_token_cost_ultrafast": 7.5e-05, "cache_read_input_token_cost_above_272k_tokens_ultrafast": 1.2e-05, "cache_read_input_token_cost_flex": 5e-07, "cache_read_input_token_cost_priority": 2e-06, "cache_read_input_token_cost_ultrafast": 6e-06, "input_cost_per_token": 1e-05, "input_cost_per_token_above_272k_tokens": 2e-05, "input_cost_per_token_above_272k_tokens_flex": 1e-05, "input_cost_per_token_above_272k_tokens_priority": 4e-05, "input_cost_per_token_batches": 5e-06, "input_cost_per_token_above_272k_tokens_batches": 1e-05, "input_cost_per_token_above_272k_tokens_ultrafast": 0.00012, "input_cost_per_token_flex": 5e-06, "input_cost_per_token_priority": 2e-05, "input_cost_per_token_ultrafast": 6e-05, "output_cost_per_token": 5e-05, "output_cost_per_token_above_272k_tokens": 7.5e-05, "output_cost_per_token_above_272k_tokens_flex": 3.75e-05, "output_cost_per_token_above_272k_tokens_priority": 0.00015, "output_cost_per_token_batches": 2.5e-05, "output_cost_per_token_above_272k_tokens_batches": 3.75e-05, "output_cost_per_token_above_272k_tokens_ultrafast": 0.00045, "output_cost_per_token_flex": 2.5e-05, "output_cost_per_token_priority": 0.0001, "output_cost_per_token_ultrafast": 0.0003}, "inclusive": false, "checks": [[null, 1, [1e-05, 5e-05, 1.25e-05, 1.25e-05, 1e-06]], [null, 272000, [1e-05, 5e-05, 1.25e-05, 1.25e-05, 1e-06]], [null, 272001, [2e-05, 7.5e-05, 2.5e-05, 2.5e-05, 2e-06]], [null, 10000000, [2e-05, 7.5e-05, 2.5e-05, 2.5e-05, 2e-06]], ["priority", 1, [2e-05, 0.0001, 2.5e-05, 2.5e-05, 2e-06]], ["priority", 272000, [2e-05, 0.0001, 2.5e-05, 2.5e-05, 2e-06]], ["priority", 272001, [4e-05, 0.00015, 5e-05, 5e-05, 4e-06]], ["priority", 10000000, [4e-05, 0.00015, 5e-05, 5e-05, 4e-06]], ["flex", 1, [5e-06, 2.5e-05, 6.25e-06, 6.25e-06, 5e-07]], ["flex", 272000, [5e-06, 2.5e-05, 6.25e-06, 6.25e-06, 5e-07]], ["flex", 272001, [1e-05, 3.75e-05, 1.25e-05, 1.25e-05, 1e-06]], ["flex", 10000000, [1e-05, 3.75e-05, 1.25e-05, 1.25e-05, 1e-06]], ["ultrafast", 1, [6e-05, 0.0003, 7.5e-05, 7.5e-05, 6e-06]], ["ultrafast", 272000, [6e-05, 0.0003, 7.5e-05, 7.5e-05, 6e-06]], ["ultrafast", 272001, [0.00012, 0.00045, 0.00015, 0.00015, 1.2e-05]], ["ultrafast", 10000000, [0.00012, 0.00045, 0.00015, 0.00015, 1.2e-05]]]}, +{"info": {"input_cost_per_token": 2.2e-06, "output_cost_per_token": 6.6e-06, "cache_read_input_token_cost": 5.5e-07, "cache_read_input_token_cost_flex": 2.75e-07, "cache_read_input_token_cost_priority": 9.625e-07, "input_cost_per_token_flex": 1.1e-06, "input_cost_per_token_priority": 3.85e-06, "output_cost_per_token_flex": 3.3e-06, "output_cost_per_token_priority": 1.155e-05}, "inclusive": false, "checks": [[null, 1, [2.2e-06, 6.6e-06, 2.2e-06, 2.2e-06, 5.5e-07]], [null, 10000000, [2.2e-06, 6.6e-06, 2.2e-06, 2.2e-06, 5.5e-07]], ["priority", 1, [3.85e-06, 1.155e-05, 3.85e-06, 3.85e-06, 9.625e-07]], ["priority", 10000000, [3.85e-06, 1.155e-05, 3.85e-06, 3.85e-06, 9.625e-07]], ["flex", 1, [1.1e-06, 3.3e-06, 1.1e-06, 1.1e-06, 2.75e-07]], ["flex", 10000000, [1.1e-06, 3.3e-06, 1.1e-06, 1.1e-06, 2.75e-07]]]}, +{"info": {"cache_read_input_token_cost": 3e-06, "cache_read_input_token_cost_above_272k_tokens": 6e-06, "input_cost_per_token": 3e-05, "input_cost_per_token_above_272k_tokens": 6e-05, "output_cost_per_token": 0.00018, "output_cost_per_token_above_272k_tokens": 0.00027}, "inclusive": false, "checks": [[null, 1, [3e-05, 0.00018, 3e-05, 3e-05, 3e-06]], [null, 272000, [3e-05, 0.00018, 3e-05, 3e-05, 3e-06]], [null, 272001, [6e-05, 0.00027, 6e-05, 6e-05, 6e-06]], [null, 10000000, [6e-05, 0.00027, 6e-05, 6e-05, 6e-06]]]}, +{"info": {"input_cost_per_token": 1.5e-07, "input_cost_per_token_batches": 1.2e-07, "output_cost_per_token": 0}, "inclusive": false, "checks": [[null, 1, [1.5e-07, 0.0, 1.5e-07, 1.5e-07, 1.5e-07]], [null, 10000000, [1.5e-07, 0.0, 1.5e-07, 1.5e-07, 1.5e-07]]]}, +{"info": {"cache_read_input_token_cost_priority": 6e-07, "input_cost_per_token_priority": 1.2e-06, "output_cost_per_token_priority": 1.2e-06}, "inclusive": false, "checks": [[null, 1, [0.0, 0.0, 0.0, 0.0, 0.0]], [null, 10000000, [0.0, 0.0, 0.0, 0.0, 0.0]], ["priority", 1, [1.2e-06, 1.2e-06, 1.2e-06, 1.2e-06, 6e-07]], ["priority", 10000000, [1.2e-06, 1.2e-06, 1.2e-06, 1.2e-06, 6e-07]]]}, +{"info": {"cache_read_input_token_cost": 2e-07, "input_cost_per_token": 1.25e-06, "output_cost_per_token": 2.5e-06, "input_cost_per_token_above_200k_tokens": 2.5e-06, "output_cost_per_token_above_200k_tokens": 5e-06, "cache_read_input_token_cost_above_200k_tokens": 4e-07}, "inclusive": true, "checks": [[null, 1, [1.25e-06, 2.5e-06, 1.25e-06, 1.25e-06, 2e-07]], [null, 200000, [2.5e-06, 5e-06, 2.5e-06, 2.5e-06, 4e-07]], [null, 200001, [2.5e-06, 5e-06, 2.5e-06, 2.5e-06, 4e-07]], [null, 10000000, [2.5e-06, 5e-06, 2.5e-06, 2.5e-06, 4e-07]]]}, +{"info": {"cache_read_input_token_cost": 2e-07, "cache_read_input_token_cost_batches": 1.6e-07, "input_cost_per_token": 1.25e-06, "input_cost_per_token_batches": 1e-06, "output_cost_per_token": 2.5e-06, "output_cost_per_token_batches": 2e-06, "input_cost_per_token_above_200k_tokens": 2.5e-06, "input_cost_per_token_above_200k_tokens_batches": 2e-06, "output_cost_per_token_above_200k_tokens": 5e-06, "output_cost_per_token_above_200k_tokens_batches": 4e-06, "cache_read_input_token_cost_above_200k_tokens": 4e-07, "cache_read_input_token_cost_above_200k_tokens_batches": 3.2e-07}, "inclusive": true, "checks": [[null, 1, [1.25e-06, 2.5e-06, 1.25e-06, 1.25e-06, 2e-07]], [null, 200000, [2.5e-06, 5e-06, 2.5e-06, 2.5e-06, 4e-07]], [null, 200001, [2.5e-06, 5e-06, 2.5e-06, 2.5e-06, 4e-07]], [null, 10000000, [2.5e-06, 5e-06, 2.5e-06, 2.5e-06, 4e-07]]]}, +{"info": {"input_cost_per_token": 5e-06}, "inclusive": false, "checks": [[null, 1, [5e-06, 0.0, 5e-06, 5e-06, 5e-06]], [null, 10000000, [5e-06, 0.0, 5e-06, 5e-06, 5e-06]]]}, +{"info": {"cache_read_input_token_cost": 1.5e-08, "input_cost_per_token": 6e-08, "input_cost_per_token_batches": 3e-08, "output_cost_per_token": 2.4e-07, "output_cost_per_token_batches": 1.2e-07}, "inclusive": false, "checks": [[null, 1, [6e-08, 2.4e-07, 6e-08, 6e-08, 1.5e-08]], [null, 10000000, [6e-08, 2.4e-07, 6e-08, 6e-08, 1.5e-08]]]}, +{"info": {"input_cost_per_token": 3.3e-05, "input_cost_per_token_above_272k_tokens": 6.6e-05, "input_cost_per_token_above_272k_tokens_batches": 3.3e-05, "input_cost_per_token_batches": 1.65e-05, "output_cost_per_token": 0.000198, "output_cost_per_token_above_272k_tokens": 0.000297, "output_cost_per_token_above_272k_tokens_batches": 0.0001485, "output_cost_per_token_batches": 9.9e-05}, "inclusive": false, "checks": [[null, 1, [3.3e-05, 0.000198, 3.3e-05, 3.3e-05, 3.3e-05]], [null, 272000, [3.3e-05, 0.000198, 3.3e-05, 3.3e-05, 3.3e-05]], [null, 272001, [6.6e-05, 0.000297, 6.6e-05, 6.6e-05, 6.6e-05]], [null, 10000000, [6.6e-05, 0.000297, 6.6e-05, 6.6e-05, 6.6e-05]]]}, +{"info": {"cache_read_input_token_cost": 7.5e-08, "cache_read_input_token_cost_priority": 1.3125e-07, "input_cost_per_token": 3e-07, "input_cost_per_token_batches": 1.5e-07, "input_cost_per_token_priority": 5.25e-07, "output_cost_per_token": 2.5e-06, "output_cost_per_token_batches": 1.25e-06, "output_cost_per_token_priority": 4.375e-06}, "inclusive": false, "checks": [[null, 1, [3e-07, 2.5e-06, 3e-07, 3e-07, 7.5e-08]], [null, 10000000, [3e-07, 2.5e-06, 3e-07, 3e-07, 7.5e-08]], ["priority", 1, [5.25e-07, 4.375e-06, 5.25e-07, 5.25e-07, 1.3125e-07]], ["priority", 10000000, [5.25e-07, 4.375e-06, 5.25e-07, 5.25e-07, 1.3125e-07]]]}, +{"info": {"input_cost_per_token": 1.25e-06, "output_cost_per_token": 2.5e-06, "cache_read_input_token_cost": 2e-07, "input_cost_per_token_above_200k_tokens": 2.5e-06, "output_cost_per_token_above_200k_tokens": 5e-06}, "inclusive": false, "checks": [[null, 1, [1.25e-06, 2.5e-06, 1.25e-06, 1.25e-06, 2e-07]], [null, 200000, [1.25e-06, 2.5e-06, 1.25e-06, 1.25e-06, 2e-07]], [null, 200001, [2.5e-06, 5e-06, 2.5e-06, 2.5e-06, 2e-07]], [null, 10000000, [2.5e-06, 5e-06, 2.5e-06, 2.5e-06, 2e-07]]]}, +{"info": {"input_cost_per_token": 5e-06, "output_cost_per_token": 3e-05, "cache_read_input_token_cost": 5e-07, "input_cost_per_token_above_272k_tokens": 1e-05, "output_cost_per_token_above_272k_tokens": 4.5e-05}, "inclusive": false, "checks": [[null, 1, [5e-06, 3e-05, 5e-06, 5e-06, 5e-07]], [null, 272000, [5e-06, 3e-05, 5e-06, 5e-06, 5e-07]], [null, 272001, [1e-05, 4.5e-05, 1e-05, 1e-05, 5e-07]], [null, 10000000, [1e-05, 4.5e-05, 1e-05, 1e-05, 5e-07]]]}, +{"info": {"input_cost_per_token": 1.5e-07, "input_cost_per_token_batches": 7.5e-08, "input_cost_per_token_flex": 7.5e-08, "input_cost_per_token_priority": 2.625e-07, "output_cost_per_token": 6e-07, "output_cost_per_token_batches": 3e-07, "output_cost_per_token_flex": 3e-07, "output_cost_per_token_priority": 1.05e-06}, "inclusive": false, "checks": [[null, 1, [1.5e-07, 6e-07, 1.5e-07, 1.5e-07, 1.5e-07]], [null, 10000000, [1.5e-07, 6e-07, 1.5e-07, 1.5e-07, 1.5e-07]], ["priority", 1, [2.625e-07, 1.05e-06, 2.625e-07, 2.625e-07, 2.625e-07]], ["priority", 10000000, [2.625e-07, 1.05e-06, 2.625e-07, 2.625e-07, 2.625e-07]], ["flex", 1, [7.5e-08, 3e-07, 7.5e-08, 7.5e-08, 7.5e-08]], ["flex", 10000000, [7.5e-08, 3e-07, 7.5e-08, 7.5e-08, 7.5e-08]]]}, +{"info": {"tiered_pricing": [{"input_cost_per_token": 1.2e-06, "output_cost_per_token": 6e-06, "range": [0, 32000.0]}, {"input_cost_per_token": 2.4e-06, "output_cost_per_token": 1.2e-05, "range": [32000.0, 128000.0]}, {"input_cost_per_token": 3e-06, "output_cost_per_token": 1.5e-05, "range": [128000.0, 252000.0]}]}, "inclusive": false, "checks": [[null, 1, [1.2e-06, 6e-06, 1.2e-06, 1.2e-06, 1.2e-06]], [null, 32000, [1.2e-06, 6e-06, 1.2e-06, 1.2e-06, 1.2e-06]], [null, 32001, [2.4e-06, 1.2e-05, 2.4e-06, 2.4e-06, 2.4e-06]], [null, 128000, [2.4e-06, 1.2e-05, 2.4e-06, 2.4e-06, 2.4e-06]], [null, 128001, [3e-06, 1.5e-05, 3e-06, 3e-06, 3e-06]], [null, 10000000, [3e-06, 1.5e-05, 3e-06, 3e-06, 3e-06]]]}, +{"info": {"cache_creation_input_token_cost": 1.25e-06, "cache_creation_input_token_cost_above_1hr": 2e-06, "cache_read_input_token_cost": 1e-07, "input_cost_per_token": 1e-06, "output_cost_per_token": 5e-06, "input_cost_per_token_batches": 5e-07, "output_cost_per_token_batches": 2.5e-06}, "inclusive": false, "checks": [[null, 1, [1e-06, 5e-06, 1.25e-06, 2e-06, 1e-07]], [null, 10000000, [1e-06, 5e-06, 1.25e-06, 2e-06, 1e-07]]]}, +{"info": {"cache_read_input_token_cost": 2e-07, "cache_read_input_token_cost_above_200k_tokens": 4e-07, "input_cost_per_token": 2e-06, "input_cost_per_token_above_200k_tokens": 4e-06, "output_cost_per_token": 1.2e-05, "output_cost_per_token_above_200k_tokens": 1.8e-05, "cache_read_input_token_cost_above_200k_tokens_priority": 7.2e-07, "cache_read_input_token_cost_priority": 3.6e-07, "input_cost_per_token_above_200k_tokens_priority": 7.2e-06, "input_cost_per_token_batches": 1e-06, "input_cost_per_token_priority": 3.6e-06, "output_cost_per_token_above_200k_tokens_priority": 3.24e-05, "output_cost_per_token_batches": 6e-06, "output_cost_per_token_priority": 2.16e-05, "cache_read_input_token_cost_flex": 2e-07, "input_cost_per_token_flex": 1e-06, "output_cost_per_token_flex": 6e-06}, "inclusive": false, "checks": [[null, 1, [2e-06, 1.2e-05, 2e-06, 2e-06, 2e-07]], [null, 200000, [2e-06, 1.2e-05, 2e-06, 2e-06, 2e-07]], [null, 200001, [4e-06, 1.8e-05, 4e-06, 4e-06, 4e-07]], [null, 10000000, [4e-06, 1.8e-05, 4e-06, 4e-06, 4e-07]], ["priority", 1, [3.6e-06, 2.16e-05, 3.6e-06, 3.6e-06, 3.6e-07]], ["priority", 200000, [3.6e-06, 2.16e-05, 3.6e-06, 3.6e-06, 3.6e-07]], ["priority", 200001, [7.2e-06, 3.24e-05, 7.2e-06, 7.2e-06, 7.2e-07]], ["priority", 10000000, [7.2e-06, 3.24e-05, 7.2e-06, 7.2e-06, 7.2e-07]], ["flex", 1, [1e-06, 6e-06, 1e-06, 1e-06, 2e-07]], ["flex", 200000, [1e-06, 6e-06, 1e-06, 1e-06, 2e-07]], ["flex", 200001, [4e-06, 1.8e-05, 4e-06, 4e-06, 4e-07]], ["flex", 10000000, [4e-06, 1.8e-05, 4e-06, 4e-06, 4e-07]]]}, +{"info": {"input_cost_per_token": 5e-06, "input_cost_per_token_batches": 2.5e-06, "input_cost_per_token_priority": 8.75e-06, "output_cost_per_token": 1.5e-05, "output_cost_per_token_batches": 7.5e-06, "output_cost_per_token_priority": 2.625e-05}, "inclusive": false, "checks": [[null, 1, [5e-06, 1.5e-05, 5e-06, 5e-06, 5e-06]], [null, 10000000, [5e-06, 1.5e-05, 5e-06, 5e-06, 5e-06]], ["priority", 1, [8.75e-06, 2.625e-05, 8.75e-06, 8.75e-06, 8.75e-06]], ["priority", 10000000, [8.75e-06, 2.625e-05, 8.75e-06, 8.75e-06, 8.75e-06]]]}, +{"info": {"cache_read_input_token_cost": 2e-07, "cache_read_input_token_cost_priority": 3.6e-07, "cache_read_input_token_cost_above_200k_tokens": 4e-07, "cache_read_input_token_cost_above_200k_tokens_priority": 7.2e-07, "cache_read_input_token_cost_batches": 1e-07, "input_cost_per_token": 2e-06, "input_cost_per_token_priority": 3.6e-06, "input_cost_per_token_above_200k_tokens": 4e-06, "input_cost_per_token_above_200k_tokens_priority": 7.2e-06, "input_cost_per_token_batches": 1e-06, "output_cost_per_token": 1.2e-05, "output_cost_per_token_priority": 2.16e-05, "output_cost_per_token_above_200k_tokens": 1.8e-05, "output_cost_per_token_above_200k_tokens_priority": 3.24e-05, "output_cost_per_token_batches": 6e-06}, "inclusive": false, "checks": [[null, 1, [2e-06, 1.2e-05, 2e-06, 2e-06, 2e-07]], [null, 200000, [2e-06, 1.2e-05, 2e-06, 2e-06, 2e-07]], [null, 200001, [4e-06, 1.8e-05, 4e-06, 4e-06, 4e-07]], [null, 10000000, [4e-06, 1.8e-05, 4e-06, 4e-06, 4e-07]], ["priority", 1, [3.6e-06, 2.16e-05, 3.6e-06, 3.6e-06, 3.6e-07]], ["priority", 200000, [3.6e-06, 2.16e-05, 3.6e-06, 3.6e-06, 3.6e-07]], ["priority", 200001, [7.2e-06, 3.24e-05, 7.2e-06, 7.2e-06, 7.2e-07]], ["priority", 10000000, [7.2e-06, 3.24e-05, 7.2e-06, 7.2e-06, 7.2e-07]]]}, +{"info": {"cache_read_input_token_cost": 2e-07, "cache_read_input_token_cost_above_200k_tokens": 4e-07, "cache_read_input_token_cost_above_200k_tokens_priority": 7.2e-07, "cache_read_input_token_cost_batches": 1e-07, "cache_read_input_token_cost_flex": 1e-07, "cache_read_input_token_cost_priority": 3.6e-07, "input_cost_per_token": 2e-06, "input_cost_per_token_above_200k_tokens": 4e-06, "input_cost_per_token_above_200k_tokens_priority": 7.2e-06, "input_cost_per_token_batches": 1e-06, "input_cost_per_token_flex": 1e-06, "input_cost_per_token_priority": 3.6e-06, "output_cost_per_token": 1.2e-05, "output_cost_per_token_above_200k_tokens": 1.8e-05, "output_cost_per_token_above_200k_tokens_priority": 3.24e-05, "output_cost_per_token_batches": 6e-06, "output_cost_per_token_flex": 6e-06, "output_cost_per_token_priority": 2.16e-05}, "inclusive": false, "checks": [[null, 1, [2e-06, 1.2e-05, 2e-06, 2e-06, 2e-07]], [null, 200000, [2e-06, 1.2e-05, 2e-06, 2e-06, 2e-07]], [null, 200001, [4e-06, 1.8e-05, 4e-06, 4e-06, 4e-07]], [null, 10000000, [4e-06, 1.8e-05, 4e-06, 4e-06, 4e-07]], ["priority", 1, [3.6e-06, 2.16e-05, 3.6e-06, 3.6e-06, 3.6e-07]], ["priority", 200000, [3.6e-06, 2.16e-05, 3.6e-06, 3.6e-06, 3.6e-07]], ["priority", 200001, [7.2e-06, 3.24e-05, 7.2e-06, 7.2e-06, 7.2e-07]], ["priority", 10000000, [7.2e-06, 3.24e-05, 7.2e-06, 7.2e-06, 7.2e-07]], ["flex", 1, [1e-06, 6e-06, 1e-06, 1e-06, 1e-07]], ["flex", 200000, [1e-06, 6e-06, 1e-06, 1e-06, 1e-07]], ["flex", 200001, [4e-06, 1.8e-05, 4e-06, 4e-06, 4e-07]], ["flex", 10000000, [4e-06, 1.8e-05, 4e-06, 4e-06, 4e-07]]]}, +{"info": {"input_cost_per_token": 3e-07, "input_cost_per_token_batches": 1.5e-07, "input_cost_per_token_flex": 1.5e-07, "output_cost_per_token": 2.5e-06, "output_cost_per_token_batches": 1.25e-06, "output_cost_per_token_flex": 1.25e-06}, "inclusive": false, "checks": [[null, 1, [3e-07, 2.5e-06, 3e-07, 3e-07, 3e-07]], [null, 10000000, [3e-07, 2.5e-06, 3e-07, 3e-07, 3e-07]], ["flex", 1, [1.5e-07, 1.25e-06, 1.5e-07, 1.5e-07, 1.5e-07]], ["flex", 10000000, [1.5e-07, 1.25e-06, 1.5e-07, 1.5e-07, 1.5e-07]]]}, +{"info": {"cache_creation_input_token_cost": 3.75e-06, "cache_creation_input_token_cost_above_1hr": 6e-06, "cache_creation_input_token_cost_above_1hr_above_200k_tokens": 1.2e-05, "cache_read_input_token_cost": 3e-07, "input_cost_per_token": 3e-06, "input_cost_per_token_above_200k_tokens": 6e-06, "output_cost_per_token_above_200k_tokens": 2.25e-05, "cache_creation_input_token_cost_above_200k_tokens": 7.5e-06, "cache_read_input_token_cost_above_200k_tokens": 6e-07, "cache_read_input_token_cost_batches": 1.5e-07, "input_cost_per_token_batches": 1.5e-06, "output_cost_per_token": 1.5e-05, "output_cost_per_token_batches": 7.5e-06, "cache_creation_input_token_cost_batches": 1.88e-06}, "inclusive": false, "checks": [[null, 1, [3e-06, 1.5e-05, 3.75e-06, 6e-06, 3e-07]], [null, 200000, [3e-06, 1.5e-05, 3.75e-06, 6e-06, 3e-07]], [null, 200001, [6e-06, 2.25e-05, 7.5e-06, 1.2e-05, 6e-07]], [null, 10000000, [6e-06, 2.25e-05, 7.5e-06, 1.2e-05, 6e-07]]]}, +{"info": {"input_cost_per_token": 7.4e-07, "input_cost_per_token_flex": 3.7e-07, "output_cost_per_token": 2.22e-06, "output_cost_per_token_flex": 1.11e-06}, "inclusive": false, "checks": [[null, 1, [7.4e-07, 2.22e-06, 7.4e-07, 7.4e-07, 7.4e-07]], [null, 10000000, [7.4e-07, 2.22e-06, 7.4e-07, 7.4e-07, 7.4e-07]], ["flex", 1, [3.7e-07, 1.11e-06, 3.7e-07, 3.7e-07, 3.7e-07]], ["flex", 10000000, [3.7e-07, 1.11e-06, 3.7e-07, 3.7e-07, 3.7e-07]]]}, +{"info": {"cache_creation_input_token_cost": 3.75e-06, "cache_read_input_token_cost": 3e-07, "input_cost_per_token": 3e-06, "input_cost_per_token_above_200k_tokens": 6e-06, "output_cost_per_token_above_200k_tokens": 2.25e-05, "cache_creation_input_token_cost_above_200k_tokens": 7.5e-06, "cache_read_input_token_cost_above_200k_tokens": 6e-07, "output_cost_per_token": 1.5e-05}, "inclusive": false, "checks": [[null, 1, [3e-06, 1.5e-05, 3.75e-06, 3.75e-06, 3e-07]], [null, 200000, [3e-06, 1.5e-05, 3.75e-06, 3.75e-06, 3e-07]], [null, 200001, [6e-06, 2.25e-05, 7.5e-06, 7.5e-06, 6e-07]], [null, 10000000, [6e-06, 2.25e-05, 7.5e-06, 7.5e-06, 6e-07]]]}, +{"info": {"cache_creation_input_token_cost": 4.5e-06, "cache_creation_input_token_cost_above_1hr": 7.2e-06, "cache_read_input_token_cost": 3.6e-07, "input_cost_per_token": 3.6e-06, "input_cost_per_token_above_200k_tokens": 7.2e-06, "output_cost_per_token_above_200k_tokens": 2.7e-05, "cache_creation_input_token_cost_above_200k_tokens": 9e-06, "cache_creation_input_token_cost_above_1hr_above_200k_tokens": 1.44e-05, "cache_read_input_token_cost_above_200k_tokens": 7.2e-07, "output_cost_per_token": 1.8e-05}, "inclusive": false, "checks": [[null, 1, [3.6e-06, 1.8e-05, 4.5e-06, 7.2e-06, 3.6e-07]], [null, 200000, [3.6e-06, 1.8e-05, 4.5e-06, 7.2e-06, 3.6e-07]], [null, 200001, [7.2e-06, 2.7e-05, 9e-06, 1.44e-05, 7.2e-07]], [null, 10000000, [7.2e-06, 2.7e-05, 9e-06, 1.44e-05, 7.2e-07]]]}, +{"info": {"cache_creation_input_token_cost": 3.75e-06, "cache_creation_input_token_cost_above_1hr": 6e-06, "cache_creation_input_token_cost_above_1hr_above_200k_tokens": 1.2e-05, "cache_read_input_token_cost": 3e-07, "input_cost_per_token": 3e-06, "input_cost_per_token_above_200k_tokens": 6e-06, "output_cost_per_token_above_200k_tokens": 2.25e-05, "cache_creation_input_token_cost_above_200k_tokens": 7.5e-06, "cache_read_input_token_cost_above_200k_tokens": 6e-07, "output_cost_per_token": 1.5e-05, "input_cost_per_token_batches": 1.5e-06, "output_cost_per_token_batches": 7.5e-06}, "inclusive": false, "checks": [[null, 1, [3e-06, 1.5e-05, 3.75e-06, 6e-06, 3e-07]], [null, 200000, [3e-06, 1.5e-05, 3.75e-06, 6e-06, 3e-07]], [null, 200001, [6e-06, 2.25e-05, 7.5e-06, 1.2e-05, 6e-07]], [null, 10000000, [6e-06, 2.25e-05, 7.5e-06, 1.2e-05, 6e-07]]]}, +{"info": {"input_cost_per_token": 1.25e-06, "input_cost_per_token_above_200k_tokens": 2.5e-06, "output_cost_per_token": 1e-05, "output_cost_per_token_above_200k_tokens": 1.5e-05}, "inclusive": false, "checks": [[null, 1, [1.25e-06, 1e-05, 1.25e-06, 1.25e-06, 1.25e-06]], [null, 200000, [1.25e-06, 1e-05, 1.25e-06, 1.25e-06, 1.25e-06]], [null, 200001, [2.5e-06, 1.5e-05, 2.5e-06, 2.5e-06, 2.5e-06]], [null, 10000000, [2.5e-06, 1.5e-05, 2.5e-06, 2.5e-06, 2.5e-06]]]}, +{"info": {"cache_read_input_token_cost": 1.25e-06, "input_cost_per_token": 5e-06}, "inclusive": false, "checks": [[null, 1, [5e-06, 0.0, 5e-06, 5e-06, 1.25e-06]], [null, 10000000, [5e-06, 0.0, 5e-06, 5e-06, 1.25e-06]]]}, +{"info": {"input_cost_per_token": 3e-07, "input_cost_per_token_above_512k_tokens": 6e-07, "output_cost_per_token": 1.2e-06, "output_cost_per_token_above_512k_tokens": 2.4e-06, "cache_read_input_token_cost": 6e-08, "cache_read_input_token_cost_above_512k_tokens": 1.2e-07}, "inclusive": false, "checks": [[null, 1, [3e-07, 1.2e-06, 3e-07, 3e-07, 6e-08]], [null, 512000, [3e-07, 1.2e-06, 3e-07, 3e-07, 6e-08]], [null, 512001, [6e-07, 2.4e-06, 6e-07, 6e-07, 1.2e-07]], [null, 10000000, [6e-07, 2.4e-06, 6e-07, 6e-07, 1.2e-07]]]}, +{"info": {"cache_read_input_token_cost": 1.25e-06, "cache_read_input_token_cost_batches": 6.25e-07, "input_cost_per_token": 5e-06, "input_cost_per_token_batches": 2.5e-06}, "inclusive": false, "checks": [[null, 1, [5e-06, 0.0, 5e-06, 5e-06, 1.25e-06]], [null, 10000000, [5e-06, 0.0, 5e-06, 5e-06, 1.25e-06]]]}, +{"info": {"cache_read_input_token_cost": 5e-09, "cache_read_input_token_cost_batches": 2.5e-09, "cache_read_input_token_cost_flex": 2.5e-09, "input_cost_per_token": 5e-08, "input_cost_per_token_batches": 2.5e-08, "input_cost_per_token_flex": 2.5e-08, "input_cost_per_token_priority": 2.5e-06, "output_cost_per_token": 4e-07, "output_cost_per_token_batches": 2e-07, "output_cost_per_token_flex": 2e-07}, "inclusive": false, "checks": [[null, 1, [5e-08, 4e-07, 5e-08, 5e-08, 5e-09]], [null, 10000000, [5e-08, 4e-07, 5e-08, 5e-08, 5e-09]], ["priority", 1, [2.5e-06, 4e-07, 2.5e-06, 2.5e-06, 5e-09]], ["priority", 10000000, [2.5e-06, 4e-07, 2.5e-06, 2.5e-06, 5e-09]], ["flex", 1, [2.5e-08, 2e-07, 2.5e-08, 2.5e-08, 2.5e-09]], ["flex", 10000000, [2.5e-08, 2e-07, 2.5e-08, 2.5e-08, 2.5e-09]]]}, +{"info": {"cache_read_input_token_cost": 2e-08, "cache_read_input_token_cost_batches": 1e-08, "cache_read_input_token_cost_flex": 1e-08, "input_cost_per_token": 2e-07, "output_cost_per_token": 1.25e-06, "input_cost_per_token_batches": 1e-07, "input_cost_per_token_flex": 1e-07, "output_cost_per_token_batches": 6.25e-07, "output_cost_per_token_flex": 6.25e-07}, "inclusive": false, "checks": [[null, 1, [2e-07, 1.25e-06, 2e-07, 2e-07, 2e-08]], [null, 10000000, [2e-07, 1.25e-06, 2e-07, 2e-07, 2e-08]], ["flex", 1, [1e-07, 6.25e-07, 1e-07, 1e-07, 1e-08]], ["flex", 10000000, [1e-07, 6.25e-07, 1e-07, 1e-07, 1e-08]]]}, +{"info": {"input_cost_per_token": 3e-05, "input_cost_per_token_above_272k_tokens": 6e-05, "input_cost_per_token_flex": 1.5e-05, "input_cost_per_token_batches": 1.5e-05, "input_cost_per_token_above_272k_tokens_batches": 3e-05, "output_cost_per_token": 0.00018, "output_cost_per_token_above_272k_tokens": 0.00027, "output_cost_per_token_flex": 9e-05, "output_cost_per_token_batches": 9e-05, "output_cost_per_token_above_272k_tokens_batches": 0.000135, "input_cost_per_token_above_272k_tokens_flex": 3e-05, "output_cost_per_token_above_272k_tokens_flex": 0.000135}, "inclusive": false, "checks": [[null, 1, [3e-05, 0.00018, 3e-05, 3e-05, 3e-05]], [null, 272000, [3e-05, 0.00018, 3e-05, 3e-05, 3e-05]], [null, 272001, [6e-05, 0.00027, 6e-05, 6e-05, 6e-05]], [null, 10000000, [6e-05, 0.00027, 6e-05, 6e-05, 6e-05]], ["flex", 1, [1.5e-05, 9e-05, 1.5e-05, 1.5e-05, 1.5e-05]], ["flex", 272000, [1.5e-05, 9e-05, 1.5e-05, 1.5e-05, 1.5e-05]], ["flex", 272001, [3e-05, 0.000135, 3e-05, 3e-05, 3e-05]], ["flex", 10000000, [3e-05, 0.000135, 3e-05, 3e-05, 3e-05]]]}, +{"info": {"cache_read_input_token_cost": 5e-07, "cache_read_input_token_cost_above_272k_tokens": 1e-06, "cache_read_input_token_cost_batches": 2.5e-07, "cache_read_input_token_cost_above_272k_tokens_batches": 5e-07, "cache_read_input_token_cost_flex": 2.5e-07, "cache_read_input_token_cost_priority": 1.25e-06, "input_cost_per_token": 5e-06, "input_cost_per_token_above_272k_tokens": 1e-05, "input_cost_per_token_flex": 2.5e-06, "input_cost_per_token_batches": 2.5e-06, "input_cost_per_token_above_272k_tokens_batches": 5e-06, "input_cost_per_token_priority": 1.25e-05, "output_cost_per_token": 3e-05, "output_cost_per_token_above_272k_tokens": 4.5e-05, "output_cost_per_token_flex": 1.5e-05, "output_cost_per_token_batches": 1.5e-05, "output_cost_per_token_above_272k_tokens_batches": 2.25e-05, "output_cost_per_token_priority": 7.5e-05, "input_cost_per_token_above_272k_tokens_flex": 5e-06, "output_cost_per_token_above_272k_tokens_flex": 2.25e-05, "cache_read_input_token_cost_above_272k_tokens_flex": 5e-07}, "inclusive": false, "checks": [[null, 1, [5e-06, 3e-05, 5e-06, 5e-06, 5e-07]], [null, 272000, [5e-06, 3e-05, 5e-06, 5e-06, 5e-07]], [null, 272001, [1e-05, 4.5e-05, 1e-05, 1e-05, 1e-06]], [null, 10000000, [1e-05, 4.5e-05, 1e-05, 1e-05, 1e-06]], ["priority", 1, [1.25e-05, 7.5e-05, 1.25e-05, 1.25e-05, 1.25e-06]], ["priority", 272000, [1.25e-05, 7.5e-05, 1.25e-05, 1.25e-05, 1.25e-06]], ["priority", 272001, [1e-05, 4.5e-05, 1e-05, 1e-05, 1e-06]], ["priority", 10000000, [1e-05, 4.5e-05, 1e-05, 1e-05, 1e-06]], ["flex", 1, [2.5e-06, 1.5e-05, 2.5e-06, 2.5e-06, 2.5e-07]], ["flex", 272000, [2.5e-06, 1.5e-05, 2.5e-06, 2.5e-06, 2.5e-07]], ["flex", 272001, [5e-06, 2.25e-05, 5e-06, 5e-06, 5e-07]], ["flex", 10000000, [5e-06, 2.25e-05, 5e-06, 5e-06, 5e-07]]]}, +{"info": {"input_cost_per_token": 3e-05, "input_cost_per_token_above_272k_tokens": 6e-05, "input_cost_per_token_flex": 1.5e-05, "input_cost_per_token_batches": 1.5e-05, "input_cost_per_token_above_272k_tokens_batches": 3e-05, "output_cost_per_token": 0.00018, "output_cost_per_token_above_272k_tokens": 0.00027, "output_cost_per_token_flex": 9e-05, "output_cost_per_token_batches": 9e-05, "output_cost_per_token_above_272k_tokens_batches": 0.000135}, "inclusive": false, "checks": [[null, 1, [3e-05, 0.00018, 3e-05, 3e-05, 3e-05]], [null, 272000, [3e-05, 0.00018, 3e-05, 3e-05, 3e-05]], [null, 272001, [6e-05, 0.00027, 6e-05, 6e-05, 6e-05]], [null, 10000000, [6e-05, 0.00027, 6e-05, 6e-05, 6e-05]], ["flex", 1, [1.5e-05, 9e-05, 1.5e-05, 1.5e-05, 1.5e-05]], ["flex", 272000, [1.5e-05, 9e-05, 1.5e-05, 1.5e-05, 1.5e-05]], ["flex", 272001, [6e-05, 0.00027, 6e-05, 6e-05, 6e-05]], ["flex", 10000000, [6e-05, 0.00027, 6e-05, 6e-05, 6e-05]]]}, +{"info": {"cache_creation_input_token_cost": 1.25e-07, "cache_creation_input_token_cost_decisions": 1.25e-07, "cache_creation_input_token_cost_above_272k_tokens": 2.5e-07, "cache_creation_input_token_cost_above_272k_tokens_decisions": 2.5e-07, "cache_creation_input_token_cost_above_272k_tokens_flex": 1.25e-07, "cache_creation_input_token_cost_above_272k_tokens_priority": 5e-07, "cache_creation_input_token_cost_flex": 6.25e-08, "cache_creation_input_token_cost_priority": 2.5e-07, "cache_read_input_token_cost": 1e-08, "cache_read_input_token_cost_decisions": 1e-08, "cache_read_input_token_cost_above_272k_tokens": 2e-08, "cache_read_input_token_cost_above_272k_tokens_decisions": 2e-08, "cache_read_input_token_cost_above_272k_tokens_flex": 1e-08, "cache_read_input_token_cost_above_272k_tokens_priority": 4e-08, "cache_read_input_token_cost_batches": 5e-09, "cache_read_input_token_cost_above_272k_tokens_batches": 1e-08, "cache_creation_input_token_cost_batches": 6.25e-08, "cache_creation_input_token_cost_above_272k_tokens_batches": 1.25e-07, "cache_read_input_token_cost_flex": 5e-09, "cache_read_input_token_cost_priority": 2e-08, "input_cost_per_token": 1e-07, "input_cost_per_token_decisions": 1e-07, "input_cost_per_token_above_272k_tokens": 2e-07, "input_cost_per_token_above_272k_tokens_decisions": 2e-07, "input_cost_per_token_above_272k_tokens_flex": 1e-07, "input_cost_per_token_above_272k_tokens_priority": 4e-07, "input_cost_per_token_batches": 5e-08, "input_cost_per_token_above_272k_tokens_batches": 1e-07, "input_cost_per_token_flex": 5e-08, "input_cost_per_token_priority": 2e-07, "output_cost_per_token": 5e-07, "output_cost_per_token_decisions": 0.0, "output_cost_per_token_above_272k_tokens": 7.5e-07, "output_cost_per_token_above_272k_tokens_decisions": 0.0, "output_cost_per_token_above_272k_tokens_flex": 3.75e-07, "output_cost_per_token_above_272k_tokens_priority": 1.5e-06, "output_cost_per_token_batches": 2.5e-07, "output_cost_per_token_above_272k_tokens_batches": 3.75e-07, "output_cost_per_token_flex": 2.5e-07, "output_cost_per_token_priority": 1e-06}, "inclusive": false, "checks": [[null, 1, [1e-07, 5e-07, 1.25e-07, 1.25e-07, 1e-08]], [null, 272000, [1e-07, 5e-07, 1.25e-07, 1.25e-07, 1e-08]], [null, 272001, [2e-07, 7.5e-07, 2.5e-07, 2.5e-07, 2e-08]], [null, 10000000, [2e-07, 7.5e-07, 2.5e-07, 2.5e-07, 2e-08]], ["priority", 1, [2e-07, 1e-06, 2.5e-07, 2.5e-07, 2e-08]], ["priority", 272000, [2e-07, 1e-06, 2.5e-07, 2.5e-07, 2e-08]], ["priority", 272001, [4e-07, 1.5e-06, 5e-07, 5e-07, 4e-08]], ["priority", 10000000, [4e-07, 1.5e-06, 5e-07, 5e-07, 4e-08]], ["flex", 1, [5e-08, 2.5e-07, 6.25e-08, 6.25e-08, 5e-09]], ["flex", 272000, [5e-08, 2.5e-07, 6.25e-08, 6.25e-08, 5e-09]], ["flex", 272001, [1e-07, 3.75e-07, 1.25e-07, 1.25e-07, 1e-08]], ["flex", 10000000, [1e-07, 3.75e-07, 1.25e-07, 1.25e-07, 1e-08]]]}, +{"info": {"cache_creation_input_token_cost": 1.25e-07, "cache_creation_input_token_cost_above_100k_tokens": 6.25e-07, "cache_read_input_token_cost": 1e-08, "cache_read_input_token_cost_above_100k_tokens": 5e-08, "input_cost_per_token": 1e-07, "input_cost_per_token_above_100k_tokens": 5e-07, "output_cost_per_token": 5e-07, "output_cost_per_token_above_100k_tokens": 2.5e-06}, "inclusive": false, "checks": [[null, 1, [1e-07, 5e-07, 1.25e-07, 1.25e-07, 1e-08]], [null, 100000, [1e-07, 5e-07, 1.25e-07, 1.25e-07, 1e-08]], [null, 100001, [5e-07, 2.5e-06, 6.25e-07, 6.25e-07, 5e-08]], [null, 10000000, [5e-07, 2.5e-06, 6.25e-07, 6.25e-07, 5e-08]]]}, +{"info": {"input_cost_per_token": 0, "input_cost_per_token_above_128k_tokens": 0, "output_cost_per_token": 0, "output_cost_per_token_above_128k_tokens": 0}, "inclusive": false, "checks": [[null, 1, [0.0, 0.0, 0.0, 0.0, 0.0]], [null, 128000, [0.0, 0.0, 0.0, 0.0, 0.0]], [null, 128001, [0.0, 0.0, 0.0, 0.0, 0.0]], [null, 10000000, [0.0, 0.0, 0.0, 0.0, 0.0]]]}, +{"info": {"cache_read_input_token_cost": 5e-08, "cache_read_input_token_cost_priority": 9e-08, "input_cost_per_token": 5e-07, "input_cost_per_token_batches": 2.5e-07, "input_cost_per_token_flex": 2.5e-07, "input_cost_per_token_priority": 9e-07, "output_cost_per_token": 3e-06, "output_cost_per_token_batches": 1.5e-06, "output_cost_per_token_flex": 1.5e-06, "output_cost_per_token_priority": 5.4e-06}, "inclusive": false, "checks": [[null, 1, [5e-07, 3e-06, 5e-07, 5e-07, 5e-08]], [null, 10000000, [5e-07, 3e-06, 5e-07, 5e-07, 5e-08]], ["priority", 1, [9e-07, 5.4e-06, 9e-07, 9e-07, 9e-08]], ["priority", 10000000, [9e-07, 5.4e-06, 9e-07, 9e-07, 9e-08]], ["flex", 1, [2.5e-07, 1.5e-06, 2.5e-07, 2.5e-07, 5e-08]], ["flex", 10000000, [2.5e-07, 1.5e-06, 2.5e-07, 2.5e-07, 5e-08]]]}, +{"info": {"cache_read_input_token_cost": 3e-08, "input_cost_per_token": 3e-07, "input_cost_per_token_batches": 1.5e-07, "input_cost_per_token_flex": 1.5e-07, "input_cost_per_token_priority": 5.4e-07, "output_cost_per_token": 2.5e-06}, "inclusive": false, "checks": [[null, 1, [3e-07, 2.5e-06, 3e-07, 3e-07, 3e-08]], [null, 10000000, [3e-07, 2.5e-06, 3e-07, 3e-07, 3e-08]], ["priority", 1, [5.4e-07, 2.5e-06, 5.4e-07, 5.4e-07, 3e-08]], ["priority", 10000000, [5.4e-07, 2.5e-06, 5.4e-07, 5.4e-07, 3e-08]], ["flex", 1, [1.5e-07, 2.5e-06, 1.5e-07, 1.5e-07, 3e-08]], ["flex", 10000000, [1.5e-07, 2.5e-06, 1.5e-07, 1.5e-07, 3e-08]]]}, +{"info": {"cache_read_input_token_cost": 8.3e-08, "input_cost_per_token": 1.65e-07, "input_cost_per_token_batches": 8.3e-08, "output_cost_per_token": 6.6e-07}, "inclusive": false, "checks": [[null, 1, [1.65e-07, 6.6e-07, 1.65e-07, 1.65e-07, 8.3e-08]], [null, 10000000, [1.65e-07, 6.6e-07, 1.65e-07, 1.65e-07, 8.3e-08]]]}, +{"info": {"cache_read_input_token_cost": 2e-07, "cache_read_input_token_cost_above_200k_tokens": 4e-07, "cache_creation_input_token_cost_above_200k_tokens": 2.5e-07, "input_cost_per_token": 2e-06, "input_cost_per_token_above_200k_tokens": 4e-06, "input_cost_per_token_batches": 1e-06, "output_cost_per_token": 1.2e-05, "output_cost_per_token_above_200k_tokens": 1.8e-05, "output_cost_per_token_batches": 6e-06, "input_cost_per_token_priority": 3.6e-06, "input_cost_per_token_above_200k_tokens_priority": 7.2e-06, "output_cost_per_token_priority": 2.16e-05, "output_cost_per_token_above_200k_tokens_priority": 3.24e-05, "cache_read_input_token_cost_priority": 3.6e-07, "cache_read_input_token_cost_above_200k_tokens_priority": 7.2e-07}, "inclusive": false, "checks": [[null, 1, [2e-06, 1.2e-05, 2e-06, 2e-06, 2e-07]], [null, 200000, [2e-06, 1.2e-05, 2e-06, 2e-06, 2e-07]], [null, 200001, [4e-06, 1.8e-05, 2.5e-07, 2.5e-07, 4e-07]], [null, 10000000, [4e-06, 1.8e-05, 2.5e-07, 2.5e-07, 4e-07]], ["priority", 1, [3.6e-06, 2.16e-05, 3.6e-06, 3.6e-06, 3.6e-07]], ["priority", 200000, [3.6e-06, 2.16e-05, 3.6e-06, 3.6e-06, 3.6e-07]], ["priority", 200001, [7.2e-06, 3.24e-05, 2.5e-07, 2.5e-07, 7.2e-07]], ["priority", 10000000, [7.2e-06, 3.24e-05, 2.5e-07, 2.5e-07, 7.2e-07]]]}, +{"info": {"cache_creation_input_token_cost_above_200k_tokens": 2.5e-07, "cache_read_input_token_cost": 2e-07, "cache_read_input_token_cost_above_200k_tokens": 4e-07, "cache_read_input_token_cost_above_200k_tokens_priority": 7.2e-07, "cache_read_input_token_cost_priority": 3.6e-07, "input_cost_per_token": 2e-06, "input_cost_per_token_above_200k_tokens": 4e-06, "input_cost_per_token_above_200k_tokens_priority": 7.2e-06, "input_cost_per_token_batches": 1e-06, "input_cost_per_token_flex": 1e-06, "input_cost_per_token_priority": 3.6e-06, "output_cost_per_token": 1.2e-05, "output_cost_per_token_above_200k_tokens": 1.8e-05, "output_cost_per_token_above_200k_tokens_priority": 3.24e-05, "output_cost_per_token_batches": 6e-06, "output_cost_per_token_flex": 6e-06, "output_cost_per_token_priority": 2.16e-05}, "inclusive": false, "checks": [[null, 1, [2e-06, 1.2e-05, 2e-06, 2e-06, 2e-07]], [null, 200000, [2e-06, 1.2e-05, 2e-06, 2e-06, 2e-07]], [null, 200001, [4e-06, 1.8e-05, 2.5e-07, 2.5e-07, 4e-07]], [null, 10000000, [4e-06, 1.8e-05, 2.5e-07, 2.5e-07, 4e-07]], ["priority", 1, [3.6e-06, 2.16e-05, 3.6e-06, 3.6e-06, 3.6e-07]], ["priority", 200000, [3.6e-06, 2.16e-05, 3.6e-06, 3.6e-06, 3.6e-07]], ["priority", 200001, [7.2e-06, 3.24e-05, 2.5e-07, 2.5e-07, 7.2e-07]], ["priority", 10000000, [7.2e-06, 3.24e-05, 2.5e-07, 2.5e-07, 7.2e-07]], ["flex", 1, [1e-06, 6e-06, 1e-06, 1e-06, 2e-07]], ["flex", 200000, [1e-06, 6e-06, 1e-06, 1e-06, 2e-07]], ["flex", 200001, [4e-06, 1.8e-05, 2.5e-07, 2.5e-07, 4e-07]], ["flex", 10000000, [4e-06, 1.8e-05, 2.5e-07, 2.5e-07, 4e-07]]]}, +{"info": {"cache_read_input_token_cost": 2e-07, "cache_read_input_token_cost_above_200k_tokens": 4e-07, "cache_creation_input_token_cost_above_200k_tokens": 2.5e-07, "input_cost_per_token": 2e-06, "input_cost_per_token_above_200k_tokens": 4e-06, "input_cost_per_token_batches": 1e-06, "output_cost_per_token": 1.2e-05, "output_cost_per_token_above_200k_tokens": 1.8e-05, "output_cost_per_token_batches": 6e-06}, "inclusive": false, "checks": [[null, 1, [2e-06, 1.2e-05, 2e-06, 2e-06, 2e-07]], [null, 200000, [2e-06, 1.2e-05, 2e-06, 2e-06, 2e-07]], [null, 200001, [4e-06, 1.8e-05, 2.5e-07, 2.5e-07, 4e-07]], [null, 10000000, [4e-06, 1.8e-05, 2.5e-07, 2.5e-07, 4e-07]]]}, +{"info": {"cache_read_input_token_cost": 1.25e-07, "cache_read_input_token_cost_above_200k_tokens": 2.5e-07, "cache_creation_input_token_cost_above_200k_tokens": 2.5e-07, "cache_read_input_token_cost_above_200k_tokens_priority": 4.5e-07, "cache_read_input_token_cost_flex": 1.25e-07, "cache_read_input_token_cost_priority": 2.25e-07, "input_cost_per_token": 1.25e-06, "input_cost_per_token_above_200k_tokens": 2.5e-06, "output_cost_per_token": 1e-05, "output_cost_per_token_above_200k_tokens": 1.5e-05, "input_cost_per_token_above_200k_tokens_priority": 4.5e-06, "input_cost_per_token_batches": 6.25e-07, "input_cost_per_token_flex": 6.25e-07, "input_cost_per_token_priority": 2.25e-06, "output_cost_per_token_above_200k_tokens_priority": 2.7e-05, "output_cost_per_token_batches": 5e-06, "output_cost_per_token_flex": 5e-06, "output_cost_per_token_priority": 1.8e-05}, "inclusive": false, "checks": [[null, 1, [1.25e-06, 1e-05, 1.25e-06, 1.25e-06, 1.25e-07]], [null, 200000, [1.25e-06, 1e-05, 1.25e-06, 1.25e-06, 1.25e-07]], [null, 200001, [2.5e-06, 1.5e-05, 2.5e-07, 2.5e-07, 2.5e-07]], [null, 10000000, [2.5e-06, 1.5e-05, 2.5e-07, 2.5e-07, 2.5e-07]], ["priority", 1, [2.25e-06, 1.8e-05, 2.25e-06, 2.25e-06, 2.25e-07]], ["priority", 200000, [2.25e-06, 1.8e-05, 2.25e-06, 2.25e-06, 2.25e-07]], ["priority", 200001, [4.5e-06, 2.7e-05, 2.5e-07, 2.5e-07, 4.5e-07]], ["priority", 10000000, [4.5e-06, 2.7e-05, 2.5e-07, 2.5e-07, 4.5e-07]], ["flex", 1, [6.25e-07, 5e-06, 6.25e-07, 6.25e-07, 1.25e-07]], ["flex", 200000, [6.25e-07, 5e-06, 6.25e-07, 6.25e-07, 1.25e-07]], ["flex", 200001, [2.5e-06, 1.5e-05, 2.5e-07, 2.5e-07, 2.5e-07]], ["flex", 10000000, [2.5e-06, 1.5e-05, 2.5e-07, 2.5e-07, 2.5e-07]]]}, +{"info": {"cache_creation_input_token_cost": 1.875e-06, "input_cost_per_token": 3.75e-06, "output_cost_per_token": 1.5e-05}, "inclusive": false, "checks": [[null, 1, [3.75e-06, 1.5e-05, 1.875e-06, 1.875e-06, 3.75e-06]], [null, 10000000, [3.75e-06, 1.5e-05, 1.875e-06, 1.875e-06, 3.75e-06]]]}, +{"info": {"cache_read_input_token_cost": 2.1e-07, "cache_read_input_token_cost_flex": 1.05e-07, "input_cost_per_token": 8.4e-07, "input_cost_per_token_flex": 4.2e-07, "output_cost_per_token": 3.36e-06, "output_cost_per_token_flex": 1.68e-06}, "inclusive": false, "checks": [[null, 1, [8.4e-07, 3.36e-06, 8.4e-07, 8.4e-07, 2.1e-07]], [null, 10000000, [8.4e-07, 3.36e-06, 8.4e-07, 8.4e-07, 2.1e-07]], ["flex", 1, [4.2e-07, 1.68e-06, 4.2e-07, 4.2e-07, 1.05e-07]], ["flex", 10000000, [4.2e-07, 1.68e-06, 4.2e-07, 4.2e-07, 1.05e-07]]]}, +{"info": {"input_cost_per_token": 2.5e-06, "output_cost_per_token": 1.25e-05, "cache_read_input_token_cost": 2.5e-07, "input_cost_per_token_balanced": 2e-06, "output_cost_per_token_balanced": 1e-05, "cache_read_input_token_cost_balanced": 2e-07, "input_cost_per_token_flex": 1.25e-06, "output_cost_per_token_flex": 6.25e-06, "cache_read_input_token_cost_flex": 1.5e-07}, "inclusive": false, "checks": [[null, 1, [2.5e-06, 1.25e-05, 2.5e-06, 2.5e-06, 2.5e-07]], [null, 10000000, [2.5e-06, 1.25e-05, 2.5e-06, 2.5e-06, 2.5e-07]], ["flex", 1, [1.25e-06, 6.25e-06, 1.25e-06, 1.25e-06, 1.5e-07]], ["flex", 10000000, [1.25e-06, 6.25e-06, 1.25e-06, 1.25e-06, 1.5e-07]], ["balanced", 1, [2e-06, 1e-05, 2e-06, 2e-06, 2e-07]], ["balanced", 10000000, [2e-06, 1e-05, 2e-06, 2e-06, 2e-07]]]}, +{"info": {"tiered_pricing": [{"input_cost_per_query": 0.00166, "max_results_range": [1, 10]}, {"input_cost_per_query": 0.00332, "max_results_range": [11, 20]}, {"input_cost_per_query": 0.00498, "max_results_range": [21, 30]}, {"input_cost_per_query": 0.00664, "max_results_range": [31, 40]}, {"input_cost_per_query": 0.0083, "max_results_range": [41, 50]}, {"input_cost_per_query": 0.00996, "max_results_range": [51, 60]}, {"input_cost_per_query": 0.01162, "max_results_range": [61, 70]}, {"input_cost_per_query": 0.01328, "max_results_range": [71, 80]}, {"input_cost_per_query": 0.01494, "max_results_range": [81, 90]}, {"input_cost_per_query": 0.0166, "max_results_range": [91, 100]}]}, "inclusive": false, "checks": [[null, 1, [0.0, 0.0, 0.0, 0.0, 0.0]], [null, 10000000, [0.0, 0.0, 0.0, 0.0, 0.0]]]}, +{"info": {"tiered_pricing": [{"input_cost_per_query": 0.005, "max_results_range": [0, 25]}, {"input_cost_per_query": 0.025, "max_results_range": [26, 100]}]}, "inclusive": false, "checks": [[null, 1, [0.0, 0.0, 0.0, 0.0, 0.0]], [null, 10000000, [0.0, 0.0, 0.0, 0.0, 0.0]]]}, +{"info": {"tiered_pricing": [{"cache_read_input_token_cost": 8e-08, "input_cost_per_token": 4e-07, "output_cost_per_token": 1.6e-06, "range": [0, 256000.0]}, {"cache_read_input_token_cost": 2.4e-07, "input_cost_per_token": 1.2e-06, "output_cost_per_token": 4.8e-06, "range": [256000.0, 1000000.0]}]}, "inclusive": false, "checks": [[null, 1, [4e-07, 1.6e-06, 4e-07, 4e-07, 8e-08]], [null, 256000, [4e-07, 1.6e-06, 4e-07, 4e-07, 8e-08]], [null, 256001, [1.2e-06, 4.8e-06, 1.2e-06, 1.2e-06, 2.4e-07]], [null, 10000000, [1.2e-06, 4.8e-06, 1.2e-06, 1.2e-06, 2.4e-07]]]}, +{"info": {"tiered_pricing": [{"input_cost_per_token": 5e-08, "output_cost_per_token": 4e-07, "range": [0, 256000.0]}, {"input_cost_per_token": 2.5e-07, "output_cost_per_token": 2e-06, "range": [256000.0, 1000000.0]}]}, "inclusive": false, "checks": [[null, 1, [5e-08, 4e-07, 5e-08, 5e-08, 5e-08]], [null, 256000, [5e-08, 4e-07, 5e-08, 5e-08, 5e-08]], [null, 256001, [2.5e-07, 2e-06, 2.5e-07, 2.5e-07, 2.5e-07]], [null, 10000000, [2.5e-07, 2e-06, 2.5e-07, 2.5e-07, 2.5e-07]]]}, +{"info": {"tiered_pricing": [{"input_cost_per_token": 3e-07, "output_cost_per_token": 1.5e-06, "range": [0, 32000.0]}, {"input_cost_per_token": 5e-07, "output_cost_per_token": 2.5e-06, "range": [32000.0, 128000.0]}, {"input_cost_per_token": 8e-07, "output_cost_per_token": 4e-06, "range": [128000.0, 256000.0]}, {"input_cost_per_token": 1.6e-06, "output_cost_per_token": 9.6e-06, "range": [256000.0, 1000000.0]}]}, "inclusive": false, "checks": [[null, 1, [3e-07, 1.5e-06, 3e-07, 3e-07, 3e-07]], [null, 32000, [3e-07, 1.5e-06, 3e-07, 3e-07, 3e-07]], [null, 32001, [5e-07, 2.5e-06, 5e-07, 5e-07, 5e-07]], [null, 128000, [5e-07, 2.5e-06, 5e-07, 5e-07, 5e-07]], [null, 128001, [8e-07, 4e-06, 8e-07, 8e-07, 8e-07]], [null, 256000, [8e-07, 4e-06, 8e-07, 8e-07, 8e-07]], [null, 256001, [1.6e-06, 9.6e-06, 1.6e-06, 1.6e-06, 1.6e-06]], [null, 10000000, [1.6e-06, 9.6e-06, 1.6e-06, 1.6e-06, 1.6e-06]]]}, +{"info": {"tiered_pricing": [{"cache_read_input_token_cost": 8e-08, "input_cost_per_token": 3e-07, "output_cost_per_token": 1.5e-06, "range": [0, 32000.0]}, {"cache_read_input_token_cost": 1.2e-07, "input_cost_per_token": 5e-07, "output_cost_per_token": 2.5e-06, "range": [32000.0, 128000.0]}, {"cache_read_input_token_cost": 2e-07, "input_cost_per_token": 8e-07, "output_cost_per_token": 4e-06, "range": [128000.0, 256000.0]}, {"cache_read_input_token_cost": 4e-07, "input_cost_per_token": 1.6e-06, "output_cost_per_token": 9.6e-06, "range": [256000.0, 1000000.0]}]}, "inclusive": false, "checks": [[null, 1, [3e-07, 1.5e-06, 3e-07, 3e-07, 8e-08]], [null, 32000, [3e-07, 1.5e-06, 3e-07, 3e-07, 8e-08]], [null, 32001, [5e-07, 2.5e-06, 5e-07, 5e-07, 1.2e-07]], [null, 128000, [5e-07, 2.5e-06, 5e-07, 5e-07, 1.2e-07]], [null, 128001, [8e-07, 4e-06, 8e-07, 8e-07, 2e-07]], [null, 256000, [8e-07, 4e-06, 8e-07, 8e-07, 2e-07]], [null, 256001, [1.6e-06, 9.6e-06, 1.6e-06, 1.6e-06, 4e-07]], [null, 10000000, [1.6e-06, 9.6e-06, 1.6e-06, 1.6e-06, 4e-07]]]}, +{"info": {"tiered_pricing": [{"input_cost_per_token": 4e-07, "output_cost_per_reasoning_token": 4e-06, "output_cost_per_token": 1.2e-06, "range": [0, 256000.0]}, {"input_cost_per_token": 1.2e-06, "output_cost_per_reasoning_token": 1.2e-05, "output_cost_per_token": 3.6e-06, "range": [256000.0, 1000000.0]}]}, "inclusive": false, "checks": [[null, 1, [4e-07, 1.2e-06, 4e-07, 4e-07, 4e-07]], [null, 256000, [4e-07, 1.2e-06, 4e-07, 4e-07, 4e-07]], [null, 256001, [1.2e-06, 3.6e-06, 1.2e-06, 1.2e-06, 1.2e-06]], [null, 10000000, [1.2e-06, 3.6e-06, 1.2e-06, 1.2e-06, 1.2e-06]]]}, +{"info": {"cache_creation_input_token_cost": 1.375e-06, "cache_read_input_token_cost": 1.1e-07, "input_cost_per_token": 1.1e-06, "output_cost_per_token": 5.5e-06, "input_cost_per_token_batches": 5.5e-07, "output_cost_per_token_batches": 2.75e-06}, "inclusive": false, "checks": [[null, 1, [1.1e-06, 5.5e-06, 1.375e-06, 1.375e-06, 1.1e-07]], [null, 10000000, [1.1e-06, 5.5e-06, 1.375e-06, 1.375e-06, 1.1e-07]]]}, +{"info": {"cache_read_input_token_cost": 2.75e-07, "cache_read_input_token_cost_above_272k_tokens": 5.5e-07, "cache_read_input_token_cost_above_272k_tokens_batches": 2.75e-07, "cache_read_input_token_cost_batches": 1.43e-07, "cache_read_input_token_cost_priority": 5.5e-07, "input_cost_per_token": 2.75e-06, "input_cost_per_token_priority": 5.5e-06, "output_cost_per_token": 1.65e-05, "output_cost_per_token_priority": 3.3e-05, "input_cost_per_token_above_272k_tokens": 5.5e-06, "input_cost_per_token_above_272k_tokens_batches": 2.75e-06, "input_cost_per_token_batches": 1.375e-06, "output_cost_per_token_above_272k_tokens": 2.475e-05, "output_cost_per_token_above_272k_tokens_batches": 1.2375e-05, "output_cost_per_token_batches": 8.25e-06}, "inclusive": false, "checks": [[null, 1, [2.75e-06, 1.65e-05, 2.75e-06, 2.75e-06, 2.75e-07]], [null, 272000, [2.75e-06, 1.65e-05, 2.75e-06, 2.75e-06, 2.75e-07]], [null, 272001, [5.5e-06, 2.475e-05, 5.5e-06, 5.5e-06, 5.5e-07]], [null, 10000000, [5.5e-06, 2.475e-05, 5.5e-06, 5.5e-06, 5.5e-07]], ["priority", 1, [5.5e-06, 3.3e-05, 5.5e-06, 5.5e-06, 5.5e-07]], ["priority", 272000, [5.5e-06, 3.3e-05, 5.5e-06, 5.5e-06, 5.5e-07]], ["priority", 272001, [5.5e-06, 2.475e-05, 5.5e-06, 5.5e-06, 5.5e-07]], ["priority", 10000000, [5.5e-06, 2.475e-05, 5.5e-06, 5.5e-06, 5.5e-07]]]}, +{"info": {"cache_read_input_token_cost": 5e-07, "cache_read_input_token_cost_above_272k_tokens": 1e-06, "cache_read_input_token_cost_above_272k_tokens_batches": 5e-07, "cache_read_input_token_cost_priority": 1.25e-06, "cache_read_input_token_cost_above_272k_tokens_priority": 2e-06, "cache_read_input_token_cost_batches": 2.5e-07, "cache_read_input_token_cost_flex": 2.5e-07, "input_cost_per_token": 5e-06, "input_cost_per_token_above_272k_tokens": 1e-05, "input_cost_per_token_above_272k_tokens_batches": 5e-06, "input_cost_per_token_above_272k_tokens_flex": 5e-06, "input_cost_per_token_priority": 1.25e-05, "input_cost_per_token_above_272k_tokens_priority": 2e-05, "input_cost_per_token_batches": 2.5e-06, "input_cost_per_token_flex": 2.5e-06, "output_cost_per_token": 3e-05, "output_cost_per_token_above_272k_tokens": 4.5e-05, "output_cost_per_token_above_272k_tokens_batches": 2.25e-05, "output_cost_per_token_priority": 7.5e-05, "output_cost_per_token_above_272k_tokens_priority": 9e-05, "output_cost_per_token_batches": 1.5e-05}, "inclusive": false, "checks": [[null, 1, [5e-06, 3e-05, 5e-06, 5e-06, 5e-07]], [null, 272000, [5e-06, 3e-05, 5e-06, 5e-06, 5e-07]], [null, 272001, [1e-05, 4.5e-05, 1e-05, 1e-05, 1e-06]], [null, 10000000, [1e-05, 4.5e-05, 1e-05, 1e-05, 1e-06]], ["priority", 1, [1.25e-05, 7.5e-05, 1.25e-05, 1.25e-05, 1.25e-06]], ["priority", 272000, [1.25e-05, 7.5e-05, 1.25e-05, 1.25e-05, 1.25e-06]], ["priority", 272001, [2e-05, 9e-05, 2e-05, 2e-05, 2e-06]], ["priority", 10000000, [2e-05, 9e-05, 2e-05, 2e-05, 2e-06]], ["flex", 1, [2.5e-06, 3e-05, 2.5e-06, 2.5e-06, 2.5e-07]], ["flex", 272000, [2.5e-06, 3e-05, 2.5e-06, 2.5e-06, 2.5e-07]], ["flex", 272001, [5e-06, 4.5e-05, 5e-06, 5e-06, 1e-06]], ["flex", 10000000, [5e-06, 4.5e-05, 5e-06, 5e-06, 1e-06]]]}, +{"info": {"cache_creation_input_token_cost": 5e-06, "cache_creation_input_token_cost_above_272k_tokens": 1e-05, "cache_creation_input_token_cost_priority": 1e-05, "cache_creation_input_token_cost_above_272k_tokens_priority": 2e-05, "cache_read_input_token_cost": 4e-07, "cache_read_input_token_cost_above_272k_tokens": 8e-07, "cache_read_input_token_cost_priority": 8e-07, "cache_read_input_token_cost_above_272k_tokens_priority": 1.6e-06, "input_cost_per_token": 4e-06, "input_cost_per_token_above_272k_tokens": 8e-06, "input_cost_per_token_priority": 8e-06, "input_cost_per_token_above_272k_tokens_priority": 1.6e-05, "output_cost_per_token": 2e-05, "output_cost_per_token_above_272k_tokens": 3e-05, "output_cost_per_token_priority": 4e-05, "output_cost_per_token_above_272k_tokens_priority": 6e-05}, "inclusive": false, "checks": [[null, 1, [4e-06, 2e-05, 5e-06, 5e-06, 4e-07]], [null, 272000, [4e-06, 2e-05, 5e-06, 5e-06, 4e-07]], [null, 272001, [8e-06, 3e-05, 1e-05, 1e-05, 8e-07]], [null, 10000000, [8e-06, 3e-05, 1e-05, 1e-05, 8e-07]], ["priority", 1, [8e-06, 4e-05, 1e-05, 1e-05, 8e-07]], ["priority", 272000, [8e-06, 4e-05, 1e-05, 1e-05, 8e-07]], ["priority", 272001, [1.6e-05, 6e-05, 2e-05, 2e-05, 1.6e-06]], ["priority", 10000000, [1.6e-05, 6e-05, 2e-05, 2e-05, 1.6e-06]]]}, +{"info": {"cache_creation_input_token_cost": 5e-06, "cache_creation_input_token_cost_above_272k_tokens": 1e-05, "cache_creation_input_token_cost_above_272k_tokens_flex": 5e-06, "cache_creation_input_token_cost_priority": 1e-05, "cache_creation_input_token_cost_above_272k_tokens_priority": 2e-05, "cache_creation_input_token_cost_flex": 2.5e-06, "cache_read_input_token_cost": 4e-07, "cache_read_input_token_cost_above_272k_tokens": 8e-07, "cache_read_input_token_cost_above_272k_tokens_flex": 4e-07, "cache_read_input_token_cost_priority": 8e-07, "cache_read_input_token_cost_above_272k_tokens_priority": 1.6e-06, "cache_read_input_token_cost_flex": 2e-07, "input_cost_per_token": 4e-06, "input_cost_per_token_above_272k_tokens": 8e-06, "input_cost_per_token_above_272k_tokens_flex": 4e-06, "input_cost_per_token_priority": 8e-06, "input_cost_per_token_above_272k_tokens_priority": 1.6e-05, "input_cost_per_token_flex": 2e-06, "output_cost_per_token": 2e-05, "output_cost_per_token_above_272k_tokens": 3e-05, "output_cost_per_token_above_272k_tokens_flex": 1.5e-05, "output_cost_per_token_priority": 4e-05, "output_cost_per_token_above_272k_tokens_priority": 6e-05, "output_cost_per_token_flex": 1e-05}, "inclusive": false, "checks": [[null, 1, [4e-06, 2e-05, 5e-06, 5e-06, 4e-07]], [null, 272000, [4e-06, 2e-05, 5e-06, 5e-06, 4e-07]], [null, 272001, [8e-06, 3e-05, 1e-05, 1e-05, 8e-07]], [null, 10000000, [8e-06, 3e-05, 1e-05, 1e-05, 8e-07]], ["priority", 1, [8e-06, 4e-05, 1e-05, 1e-05, 8e-07]], ["priority", 272000, [8e-06, 4e-05, 1e-05, 1e-05, 8e-07]], ["priority", 272001, [1.6e-05, 6e-05, 2e-05, 2e-05, 1.6e-06]], ["priority", 10000000, [1.6e-05, 6e-05, 2e-05, 2e-05, 1.6e-06]], ["flex", 1, [2e-06, 1e-05, 2.5e-06, 2.5e-06, 2e-07]], ["flex", 272000, [2e-06, 1e-05, 2.5e-06, 2.5e-06, 2e-07]], ["flex", 272001, [4e-06, 1.5e-05, 5e-06, 5e-06, 4e-07]], ["flex", 10000000, [4e-06, 1.5e-05, 5e-06, 5e-06, 4e-07]]]}, +{"info": {"cache_read_input_token_cost": 3e-06, "cache_read_input_token_cost_above_272k_tokens": 6e-06, "input_cost_per_token": 3e-05, "input_cost_per_token_above_272k_tokens": 6e-05, "input_cost_per_token_above_272k_tokens_batches": 3e-05, "input_cost_per_token_above_272k_tokens_flex": 3e-05, "input_cost_per_token_batches": 1.5e-05, "input_cost_per_token_flex": 1.5e-05, "output_cost_per_token": 0.00018, "output_cost_per_token_above_272k_tokens": 0.00027, "output_cost_per_token_above_272k_tokens_batches": 0.000135, "output_cost_per_token_above_272k_tokens_flex": 0.000135, "output_cost_per_token_batches": 9e-05, "output_cost_per_token_flex": 9e-05}, "inclusive": false, "checks": [[null, 1, [3e-05, 0.00018, 3e-05, 3e-05, 3e-06]], [null, 272000, [3e-05, 0.00018, 3e-05, 3e-05, 3e-06]], [null, 272001, [6e-05, 0.00027, 6e-05, 6e-05, 6e-06]], [null, 10000000, [6e-05, 0.00027, 6e-05, 6e-05, 6e-06]], ["flex", 1, [1.5e-05, 9e-05, 1.5e-05, 1.5e-05, 3e-06]], ["flex", 272000, [1.5e-05, 9e-05, 1.5e-05, 1.5e-05, 3e-06]], ["flex", 272001, [3e-05, 0.000135, 3e-05, 3e-05, 6e-06]], ["flex", 10000000, [3e-05, 0.000135, 3e-05, 3e-05, 6e-06]]]}, +{"info": {"cache_read_input_token_cost": 2.5e-07, "cache_read_input_token_cost_above_272k_tokens": 5e-07, "cache_read_input_token_cost_above_272k_tokens_batches": 2.5e-07, "cache_read_input_token_cost_above_272k_tokens_flex": 2.5e-07, "cache_read_input_token_cost_priority": 5e-07, "cache_read_input_token_cost_above_272k_tokens_priority": 1e-06, "cache_read_input_token_cost_batches": 1.3e-07, "input_cost_per_token": 2.5e-06, "input_cost_per_token_above_272k_tokens": 5e-06, "input_cost_per_token_priority": 5e-06, "input_cost_per_token_above_272k_tokens_priority": 1e-05, "output_cost_per_token": 1.5e-05, "output_cost_per_token_above_272k_tokens": 2.25e-05, "output_cost_per_token_priority": 3e-05, "output_cost_per_token_above_272k_tokens_priority": 4.5e-05, "input_cost_per_token_above_272k_tokens_batches": 2.5e-06, "input_cost_per_token_above_272k_tokens_flex": 2.5e-06, "input_cost_per_token_batches": 1.25e-06, "input_cost_per_token_flex": 1.25e-06, "output_cost_per_token_above_272k_tokens_batches": 1.125e-05, "output_cost_per_token_above_272k_tokens_flex": 1.125e-05, "output_cost_per_token_batches": 7.5e-06, "output_cost_per_token_flex": 7.5e-06}, "inclusive": false, "checks": [[null, 1, [2.5e-06, 1.5e-05, 2.5e-06, 2.5e-06, 2.5e-07]], [null, 272000, [2.5e-06, 1.5e-05, 2.5e-06, 2.5e-06, 2.5e-07]], [null, 272001, [5e-06, 2.25e-05, 5e-06, 5e-06, 5e-07]], [null, 10000000, [5e-06, 2.25e-05, 5e-06, 5e-06, 5e-07]], ["priority", 1, [5e-06, 3e-05, 5e-06, 5e-06, 5e-07]], ["priority", 272000, [5e-06, 3e-05, 5e-06, 5e-06, 5e-07]], ["priority", 272001, [1e-05, 4.5e-05, 1e-05, 1e-05, 1e-06]], ["priority", 10000000, [1e-05, 4.5e-05, 1e-05, 1e-05, 1e-06]], ["flex", 1, [1.25e-06, 7.5e-06, 1.25e-06, 1.25e-06, 2.5e-07]], ["flex", 272000, [1.25e-06, 7.5e-06, 1.25e-06, 1.25e-06, 2.5e-07]], ["flex", 272001, [2.5e-06, 1.125e-05, 2.5e-06, 2.5e-06, 2.5e-07]], ["flex", 10000000, [2.5e-06, 1.125e-05, 2.5e-06, 2.5e-06, 2.5e-07]]]}, +{"info": {"cache_read_input_token_cost": 5e-07, "cache_read_input_token_cost_above_272k_tokens": 1e-06, "cache_read_input_token_cost_priority": 1e-06, "cache_read_input_token_cost_above_272k_tokens_priority": 2e-06, "input_cost_per_token": 5e-06, "input_cost_per_token_above_272k_tokens": 1e-05, "input_cost_per_token_priority": 1e-05, "input_cost_per_token_above_272k_tokens_priority": 2e-05, "output_cost_per_token": 3e-05, "output_cost_per_token_above_272k_tokens": 4.5e-05, "output_cost_per_token_priority": 6e-05, "output_cost_per_token_above_272k_tokens_priority": 9e-05}, "inclusive": false, "checks": [[null, 1, [5e-06, 3e-05, 5e-06, 5e-06, 5e-07]], [null, 272000, [5e-06, 3e-05, 5e-06, 5e-06, 5e-07]], [null, 272001, [1e-05, 4.5e-05, 1e-05, 1e-05, 1e-06]], [null, 10000000, [1e-05, 4.5e-05, 1e-05, 1e-05, 1e-06]], ["priority", 1, [1e-05, 6e-05, 1e-05, 1e-05, 1e-06]], ["priority", 272000, [1e-05, 6e-05, 1e-05, 1e-05, 1e-06]], ["priority", 272001, [2e-05, 9e-05, 2e-05, 2e-05, 2e-06]], ["priority", 10000000, [2e-05, 9e-05, 2e-05, 2e-05, 2e-06]]]}, +{"info": {"cache_read_input_token_cost": 3.125e-07, "input_cost_per_token": 1.25e-06, "input_cost_per_token_batches": 6.25e-07, "input_cost_per_token_priority": 2.1875e-06, "output_cost_per_token": 1e-05, "output_cost_per_token_batches": 5e-06, "output_cost_per_token_priority": 1.75e-05}, "inclusive": false, "checks": [[null, 1, [1.25e-06, 1e-05, 1.25e-06, 1.25e-06, 3.125e-07]], [null, 10000000, [1.25e-06, 1e-05, 1.25e-06, 1.25e-06, 3.125e-07]], ["priority", 1, [2.1875e-06, 1.75e-05, 2.1875e-06, 2.1875e-06, 3.125e-07]], ["priority", 10000000, [2.1875e-06, 1.75e-05, 2.1875e-06, 2.1875e-06, 3.125e-07]]]} +]} diff --git a/litellm-rust/crates/model-catalog/tests/registry_validation.rs b/litellm-rust/crates/model-catalog/tests/registry_validation.rs deleted file mode 100644 index 3b65ef13d78..00000000000 --- a/litellm-rust/crates/model-catalog/tests/registry_validation.rs +++ /dev/null @@ -1,93 +0,0 @@ -use std::path::{Path, PathBuf}; - -use litellm_model_catalog::{ - Catalog, FallbackGeneralizations, Provenance, validate_model_entry, validate_registry, -}; -use rstest::{fixture, rstest}; -use serde_json::{Map, Value}; - -#[fixture] -fn repo_root() -> PathBuf { - Path::new(env!("CARGO_MANIFEST_DIR")).join("../../..") -} - -#[rstest] -#[case("model_prices_and_context_window.json")] -#[case("litellm/model_prices_and_context_window_backup.json")] -#[ignore] -fn checked_in_registry_passes_strict_validation(repo_root: PathBuf, #[case] filename: &str) { - let body = std::fs::read(repo_root.join(filename)).unwrap(); - let catalog = Catalog::parse(&body, Provenance::default()).unwrap(); - validate_registry(&catalog).unwrap(); -} - -#[rstest] -#[ignore] -fn fallback_generalizations_are_typed(repo_root: PathBuf) { - let body = std::fs::read(repo_root.join("model_prices_and_context_window.json")).unwrap(); - let document: Map = serde_json::from_slice(&body).unwrap(); - let Some(raw_rules) = document.get("fallback_generalizations") else { - return; - }; - let _: FallbackGeneralizations = serde_json::from_value(raw_rules.clone()).unwrap(); - let catalog = Catalog::parse(&body, Provenance::default()).unwrap(); - assert!( - catalog - .fallback_rules() - .is_some_and(|rules| !rules.is_empty()) - ); -} - -#[rstest] -#[case::missing_provider(serde_json::json!({"mode": "chat"}), "litellm_provider")] -#[case::unknown_field(serde_json::json!({"litellm_provider": "test", "typo": true}), "unknown field")] -#[case::negative_price(serde_json::json!({"litellm_provider": "test", "input_cost_per_token": -1}), "nonnegative")] -#[case::negative_nested_price(serde_json::json!({"litellm_provider": "test", "guardrail_cost_per_unit": {"unit": -1}}), "nonnegative")] -#[case::invalid_mode(serde_json::json!({"litellm_provider": "test", "mode": "invalid"}), "unknown variant")] -#[case::invalid_date(serde_json::json!({"litellm_provider": "test", "deprecation_date": "2026-02-31"}), "deprecation_date")] -#[case::invalid_hours(serde_json::json!({"litellm_provider": "test", "off_peak_pricing": {"hours_utc": "25:00-01:00"}}), "hours_utc")] -#[case::empty_windows(serde_json::json!({"litellm_provider": "test", "off_peak_pricing": {"windows": []}}), "windows is empty")] -#[case::invalid_weekday(serde_json::json!({"litellm_provider": "test", "off_peak_pricing": {"windows": [{"hours_utc": "00:00-01:00", "weekdays": [0]}]}}), "weekdays is invalid")] -#[case::invalid_aliases(serde_json::json!({"litellm_provider": "test", "aliases": ["good", 7]}), "aliases must contain strings")] -#[case::null_aliases(serde_json::json!({"litellm_provider": "test", "aliases": null}), "aliases must be an array")] -#[ignore] -fn registry_validation_rejects_malformed_entries(#[case] entry: Value, #[case] expected: &str) { - let error = validate_model_entry("test", &entry).unwrap_err(); - assert!( - litellm_error::Report::new(&error) - .to_string() - .contains(expected) - ); -} - -#[test] -#[ignore] -fn checked_in_catalog_and_backup_match() { - let root = Path::new(env!("CARGO_MANIFEST_DIR")).join("../../.."); - let current = std::fs::read(root.join("model_prices_and_context_window.json")).unwrap(); - let backup = - std::fs::read(root.join("litellm/model_prices_and_context_window_backup.json")).unwrap(); - assert_eq!(current, backup); - let catalog = Catalog::parse(¤t, Provenance::default()).unwrap(); - assert!( - catalog.alias_issues().is_empty(), - "invalid registry aliases" - ); -} - -#[rstest] -#[case::input("input_cost_per_token_above_100k_tokens")] -#[case::input_batches("input_cost_per_token_above_100k_tokens_batches")] -#[case::output("output_cost_per_token_above_100k_tokens")] -#[case::output_batches("output_cost_per_token_above_100k_tokens_batches")] -#[case::cache_creation("cache_creation_input_token_cost_above_100k_tokens")] -#[case::cache_creation_batches("cache_creation_input_token_cost_above_100k_tokens_batches")] -#[case::cache_creation_1hr("cache_creation_input_token_cost_above_1hr_above_100k_tokens")] -#[case::cache_read("cache_read_input_token_cost_above_100k_tokens")] -#[case::cache_read_batches("cache_read_input_token_cost_above_100k_tokens_batches")] -fn registry_validation_keeps_above_100k_tier_rates(#[case] field: &str) { - let mut entry = Map::new(); - entry.insert("litellm_provider".into(), "anthropic".into()); - entry.insert(field.into(), 5e-7.into()); - validate_model_entry("test", &Value::Object(entry)).unwrap(); -} diff --git a/litellm-rust/crates/model-catalog/tests/schema.rs b/litellm-rust/crates/model-catalog/tests/schema.rs index f9a028a78ab..464cf069cac 100644 --- a/litellm-rust/crates/model-catalog/tests/schema.rs +++ b/litellm-rust/crates/model-catalog/tests/schema.rs @@ -1,16 +1,17 @@ #![cfg(feature = "schema")] -use std::collections::BTreeSet; -use std::path::Path; - -use litellm_model_catalog::{model_entry_json_schema, registry_json_schema}; +use litellm_model_catalog::{BUNDLED_REGISTRY, model_entry_json_schema, registry_json_schema}; use rstest::rstest; -use serde_json::{Value, json}; +use serde_json::{Map, Value, json}; fn schema() -> Value { serde_json::to_value(model_entry_json_schema()).expect("generated schema serializes") } +fn checked_in_catalog() -> Value { + serde_json::from_slice(BUNDLED_REGISTRY).unwrap() +} + fn registry_validator() -> jsonschema::Validator { let schema = serde_json::to_value(registry_json_schema()).unwrap(); jsonschema::options() @@ -19,26 +20,21 @@ fn registry_validator() -> jsonschema::Validator { .expect("generated registry schema is valid") } -#[rstest] -#[case("model_prices_and_context_window.json")] -#[case("litellm/model_prices_and_context_window_backup.json")] -#[ignore] -fn generated_registry_schema_validates_checked_in_catalog(#[case] path: &str) { - let root = Path::new(env!("CARGO_MANIFEST_DIR")).join("../../.."); - let catalog: Value = serde_json::from_slice(&std::fs::read(root.join(path)).unwrap()).unwrap(); +#[test] +fn generated_registry_schema_validates_checked_in_catalog() { + let catalog = checked_in_catalog(); let validator = registry_validator(); let errors: Vec<_> = validator .iter_errors(&catalog) .map(|error| error.to_string()) .collect(); - assert!(errors.is_empty(), "{path}: {errors:?}"); + assert!(errors.is_empty(), "{errors:?}"); } #[rstest] #[case(json!({"example": {"litellm_provider": "test"}}))] #[case(json!({"example": {"litellm_provider": "test", "future_field": true}}))] #[case(json!({"sample_spec": {"litellm_provider": "placeholder"}}))] -#[ignore] fn generated_registry_schema_keeps_reader_compatibility(#[case] document: Value) { assert!(registry_validator().is_valid(&document)); } @@ -59,47 +55,55 @@ fn generated_registry_schema_keeps_reader_compatibility(#[case] document: Value) #[case::invalid_weekday(json!({"litellm_provider": "test", "off_peak_pricing": {"windows": [{"hours_utc": "00:00-01:00", "weekdays": [0]}]}}))] #[case::invalid_aliases(json!({"litellm_provider": "test", "aliases": "wrong"}))] #[case::non_object_model(json!(4))] -#[ignore] fn generated_registry_schema_rejects_invalid_entries(#[case] entry: Value) { assert!(!registry_validator().is_valid(&json!({"example": entry}))); } -#[rstest] -#[case("model_prices_and_context_window.json")] -#[case("litellm/model_prices_and_context_window_backup.json")] -#[ignore] -fn generated_schema_covers_catalog_fields(#[case] path: &str) { - let root = Path::new(env!("CARGO_MANIFEST_DIR")).join("../../.."); - let catalog: Value = serde_json::from_slice(&std::fs::read(root.join(path)).unwrap()).unwrap(); +#[test] +fn generated_schema_covers_catalog_fields() { + let catalog = checked_in_catalog(); let schema = schema(); - let properties = schema["properties"] + let any_value = |keys: &Value| -> Map { + keys.as_object() + .expect("schema keys are an object") + .keys() + .map(|key| (key.clone(), json!({}))) + .collect() + }; + let keys_only = json!({ + "type": "object", + "properties": any_value(&schema["properties"]), + "patternProperties": any_value(&schema["patternProperties"]), + "additionalProperties": false, + }); + let registry = serde_json::to_value(registry_json_schema()).unwrap(); + let sections = registry["properties"] .as_object() - .expect("ModelInfo schema has properties"); - let fields: BTreeSet<&str> = catalog + .expect("registry has sections"); + let fields: Map = catalog .as_object() .expect("catalog is an object") .iter() - .filter(|(name, _)| *name != "sample_spec" && *name != "fallback_generalizations") + .filter(|(name, _)| !sections.contains_key(*name)) .flat_map(|(_, entry)| entry.as_object().expect("model entry is an object").keys()) - .map(String::as_str) - .filter(|name| *name != "aliases") + .map(|name| (name.clone(), Value::Null)) .collect(); - let missing: Vec<_> = fields - .into_iter() - .filter(|name| !properties.contains_key(*name)) + let missing: Vec<_> = jsonschema::validator_for(&keys_only) + .unwrap() + .iter_errors(&Value::Object(fields)) + .map(|error| error.to_string()) .collect(); assert!( missing.is_empty(), - "{path}: fields missing from schema: {missing:?}" + "fields missing from schema: {missing:?}" ); } #[rstest] #[case("Mode", "chat")] -#[case("ReasoningEffort", "high")] +#[case("EffortLevel", "high")] #[case("InputModality", "image")] -#[ignore] fn generated_schema_includes_enum_values(#[case] definition: &str, #[case] value: &str) { let schema = schema(); let variants = schema["$defs"][definition]["enum"] @@ -110,7 +114,6 @@ fn generated_schema_includes_enum_values(#[case] definition: &str, #[case] value } #[test] -#[ignore] fn generated_schema_includes_nested_pricing_types() { let schema = schema(); let definitions = schema["$defs"].as_object().expect("schema has definitions"); From e4932a453953340e9df7b839ea20a9eff28429b5 Mon Sep 17 00:00:00 2001 From: Kerry Lu Date: Wed, 28 Oct 2026 18:35:54 -0700 Subject: [PATCH 39/41] test(router): set string fallback targets after Router init in the sdk cooldown cases (#48954) Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- .../sdk/test_router_cooldown_fallback_sdk.py | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) diff --git a/tests/integration/sdk/test_router_cooldown_fallback_sdk.py b/tests/integration/sdk/test_router_cooldown_fallback_sdk.py index e7508a2670e..0a5bfd339fc 100644 --- a/tests/integration/sdk/test_router_cooldown_fallback_sdk.py +++ b/tests/integration/sdk/test_router_cooldown_fallback_sdk.py @@ -122,12 +122,12 @@ async def test_cooldown_skips_only_the_failed_duplicate_model_deployment() -> No @pytest.mark.parametrize( - "router_kwargs", - [{"fallbacks": [{"primary": "backup"}]}, {"default_fallbacks": "backup"}], - ids=["string_fallback_target", "string_default_fallbacks"], + "fallbacks", + [[{"primary": "backup"}], [{"*": "backup"}]], + ids=["string_fallback_target", "string_catch_all_target"], ) @pytest.mark.asyncio -async def test_string_valued_fallbacks_still_cool_down_the_failing_primary(router_kwargs: dict[str, object]) -> None: +async def test_string_valued_fallbacks_still_cool_down_the_failing_primary(fallbacks: list[dict[str, object]]) -> None: with wire_server(_peer) as wire: router: Final = Router( model_list=[ @@ -136,8 +136,8 @@ async def test_string_valued_fallbacks_still_cool_down_the_failing_primary(route ], allowed_fails=0, num_retries=0, - **router_kwargs, ) + router.fallbacks = fallbacks with pytest.raises(litellm.AuthenticationError): await router.acompletion( model="primary", From 7aa7948a472f0d5dbb455ec5ced740cb3338819a Mon Sep 17 00:00:00 2001 From: Mateo Wang <277851410+mateo-berri@users.noreply.github.com> Date: Wed, 28 Oct 2026 18:37:29 -0700 Subject: [PATCH 40/41] fix(health): ask OpenAI and Azure audio chat models for audio output in the health check probe (#47730) * fix(health): ask OpenAI and Azure audio chat models for audio output in the health check probe * fix(health): check the probed model when health_check_model is set, and mark Azure audio models as audio output A deployment with an audio base_model and a text health_check_model got the audio fields sent to its text probe model. The Azure gpt-audio and gpt-4o audio preview cost-map entries now set supports_audio_output, so a dated Azure name like azure/gpt-audio-1.5-2026-02-23 gets the audio probe too * fix(health): read litellm_params.base_model, and drop the stored base_model when Test Connect swaps the model * fix(health): judge the audio probe on a swapped-in health_check_params model, and keep Test Connect's stored base_model * fix(health): look up a provider-prefixed base_model under its own provider and keep text-only operator modalities --- ...odel_prices_and_context_window_backup.json | 6 + litellm/proxy/health_check.py | 68 +++++++++- .../health_endpoints/_health_endpoints.py | 2 + model_prices_and_context_window.json | 6 + .../health_endpoints/test_health_endpoints.py | 47 ++++++- .../proxy/test_health_check_max_tokens.py | 119 ++++++++++++++++++ 6 files changed, 243 insertions(+), 5 deletions(-) diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index 8719565a96f..60385d940d8 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -6101,6 +6101,7 @@ "text", "audio" ], + "supports_audio_output": true, "supports_function_calling": true, "supports_native_streaming": true, "supports_parallel_function_calling": true, @@ -6133,6 +6134,7 @@ "text", "audio" ], + "supports_audio_output": true, "supports_function_calling": true, "supports_native_streaming": true, "supports_parallel_function_calling": true, @@ -6164,6 +6166,7 @@ "text", "audio" ], + "supports_audio_output": true, "supports_function_calling": true, "supports_native_streaming": true, "supports_parallel_function_calling": true, @@ -6230,6 +6233,7 @@ "text", "audio" ], + "supports_audio_output": true, "supports_function_calling": true, "supports_native_streaming": true, "supports_parallel_function_calling": true, @@ -6293,6 +6297,7 @@ "text", "audio" ], + "supports_audio_output": true, "supports_function_calling": true, "supports_native_streaming": true, "supports_parallel_function_calling": true, @@ -6369,6 +6374,7 @@ "text", "audio" ], + "supports_audio_output": true, "supports_function_calling": true, "supports_native_streaming": true, "supports_parallel_function_calling": true, diff --git a/litellm/proxy/health_check.py b/litellm/proxy/health_check.py index be32405154c..60d2abe0ab4 100644 --- a/litellm/proxy/health_check.py +++ b/litellm/proxy/health_check.py @@ -131,6 +131,52 @@ def _should_inject_health_check_max_tokens(model_info: Mapping[str, object], mod return (mode or "chat") in _MAX_TOKEN_SUPPORT_MODES +_AUDIO_OUTPUT_PROBE_PROVIDERS: Final[frozenset[str]] = frozenset({"openai", "azure"}) + + +def _health_check_requires_audio_output( + model_info: Mapping[str, object], probe_params: Mapping[str, object], mode: str | None, deployment_model: str | None +) -> bool: + if (mode or "chat") != "chat": + return False + operator_modalities: Final = probe_params.get("modalities") + if isinstance(operator_modalities, list) and "audio" not in operator_modalities: + return False + model: Final = probe_params.get("model") + if not isinstance(model, str): + return False + provider_param: Final = probe_params.get("custom_llm_provider") + try: + _, provider, _, _ = litellm.get_llm_provider( + model=model, custom_llm_provider=provider_param if isinstance(provider_param, str) else None + ) + except litellm.exceptions.BadRequestError: + return False + if provider not in _AUDIO_OUTPUT_PROBE_PROVIDERS: + return False + base_model: Final = model_info.get("base_model") or probe_params.get("base_model") + checked_model: Final = base_model if isinstance(base_model, str) and model == deployment_model else model + checked_model_prefix: Final = checked_model.partition("/")[0] + return litellm.supports_audio_output( + model=checked_model, + custom_llm_provider=checked_model_prefix if checked_model_prefix in _AUDIO_OUTPUT_PROBE_PROVIDERS else provider, + ) + + +def _health_check_audio_output_params( + model_info: Mapping[str, object], probe_params: Mapping[str, object], mode: str | None, deployment_model: str | None +) -> Mapping[str, object]: + if not _health_check_requires_audio_output(model_info, probe_params, mode, deployment_model): + return MappingProxyType({}) + voice: Final = model_info.get("health_check_voice") + return MappingProxyType( + { + "modalities": ["text", "audio"], + "audio": {"voice": voice if isinstance(voice, str) else "alloy", "format": "wav"}, + } + ) + + # Health-check modes that forward `reasoning_effort` to the provider (chat-style calls). _HEALTH_CHECK_MODES_SUPPORTING_REASONING_EFFORT: Final = frozenset((None, "chat", "completion")) @@ -782,9 +828,12 @@ def _resolve_health_check_max_tokens(model_info: dict, litellm_params: dict) -> _PROBE_PARAMS_ADAPTER: Final = TypeAdapter(dict[str, object]) +_MODEL_INFO_ADAPTER: Final = TypeAdapter(Mapping[str, object]) -def update_litellm_params_for_health_check(model_info: dict, litellm_params: dict) -> dict: +def update_litellm_params_for_health_check( + model_info: dict, litellm_params: dict, deployment_model: str | None = None +) -> dict: """ Update the litellm params for health check. @@ -800,7 +849,15 @@ def update_litellm_params_for_health_check(model_info: dict, litellm_params: dic - updates the `model` param with the `health_check_model` if it exists Doc: https://docs.litellm.ai/docs/proxy/health#wildcard-routes - updates the `voice` param with the `health_check_voice` for `audio_speech` mode if it exists Doc: https://docs.litellm.ai/docs/proxy/health#text-to-speech-models - for Bedrock models with region routing (bedrock/region/model), strips the litellm routing prefix but preserves the model ID, and pins `custom_llm_provider` to `bedrock` (only when the deployment hasn't already set one, so an explicit `bedrock_converse` survives) so the bare model id still resolves to the provider (e.g. cross-region ids like `us.cohere.embed-v4:0`) + - asks OpenAI and Azure chat models that support audio output for `modalities` text + audio and a wav + `audio` in the `health_check_voice` (default "alloy"), since those models reject text-only requests. + `base_model` (from `model_info`, then `litellm_params`) decides it only while the probe still targets + `deployment_model` (default: the incoming `litellm_params` model), so a `health_check_model` or a `model` + in `health_check_params` is judged on its own. Keys already set through `health_check_params` are kept, + and `modalities` set there without "audio" turns the audio fields off """ + incoming_model: Final = _PROBE_PARAMS_ADAPTER.validate_python(litellm_params).get("model") + base_model_owner: Final = deployment_model or (incoming_model if isinstance(incoming_model, str) else None) mode: Final = resolve_health_check_mode( model_info, litellm_params, # any-ok: untyped router config dict @@ -875,7 +932,14 @@ def update_litellm_params_for_health_check(model_info: dict, litellm_params: dic ) probe_params: Final = _PROBE_PARAMS_ADAPTER.validate_python(litellm_params) - return {key: value for key, value in probe_params.items() if not is_model_capability_field(key)} + audio_output_params: Final = _health_check_audio_output_params( + _MODEL_INFO_ADAPTER.validate_python(model_info), probe_params, mode, base_model_owner + ) + return { + key: value + for key, value in {**audio_output_params, **probe_params}.items() + if not is_model_capability_field(key) + } _update_litellm_params_for_health_check: Final = update_litellm_params_for_health_check diff --git a/litellm/proxy/health_endpoints/_health_endpoints.py b/litellm/proxy/health_endpoints/_health_endpoints.py index 53c54fe15bf..869e4953df0 100644 --- a/litellm/proxy/health_endpoints/_health_endpoints.py +++ b/litellm/proxy/health_endpoints/_health_endpoints.py @@ -2330,9 +2330,11 @@ async def test_model_connection( stored_params=_OBJECT_MAPPING.validate_python(config_litellm_params), request_params=_OBJECT_MAPPING.validate_python(request_litellm_params), ) + stored_model: Final = _OBJECT_MAPPING.validate_python(config_litellm_params).get("model") litellm_params = update_litellm_params_for_health_check( model_info=dict(probe_model_info), litellm_params=litellm_params, + deployment_model=stored_model if isinstance(stored_model, str) else None, ) ## Auth check, on the final probe params so health_check_params cannot retarget it afterwards diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index 8719565a96f..60385d940d8 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -6101,6 +6101,7 @@ "text", "audio" ], + "supports_audio_output": true, "supports_function_calling": true, "supports_native_streaming": true, "supports_parallel_function_calling": true, @@ -6133,6 +6134,7 @@ "text", "audio" ], + "supports_audio_output": true, "supports_function_calling": true, "supports_native_streaming": true, "supports_parallel_function_calling": true, @@ -6164,6 +6166,7 @@ "text", "audio" ], + "supports_audio_output": true, "supports_function_calling": true, "supports_native_streaming": true, "supports_parallel_function_calling": true, @@ -6230,6 +6233,7 @@ "text", "audio" ], + "supports_audio_output": true, "supports_function_calling": true, "supports_native_streaming": true, "supports_parallel_function_calling": true, @@ -6293,6 +6297,7 @@ "text", "audio" ], + "supports_audio_output": true, "supports_function_calling": true, "supports_native_streaming": true, "supports_parallel_function_calling": true, @@ -6369,6 +6374,7 @@ "text", "audio" ], + "supports_audio_output": true, "supports_function_calling": true, "supports_native_streaming": true, "supports_parallel_function_calling": true, diff --git a/tests/unit/proxy/health_endpoints/test_health_endpoints.py b/tests/unit/proxy/health_endpoints/test_health_endpoints.py index 67b280eb379..49183a143c0 100644 --- a/tests/unit/proxy/health_endpoints/test_health_endpoints.py +++ b/tests/unit/proxy/health_endpoints/test_health_endpoints.py @@ -394,7 +394,7 @@ async def test_test_model_connection_loads_config_from_router(): mock_run_with_timeout = AsyncMock(return_value=mock_health_check_result) # Mock _update_litellm_params_for_health_check - def mock_update_params(model_info, litellm_params): + def mock_update_params(model_info, litellm_params, deployment_model: str | None = None): # Just return params with messages added params = litellm_params.copy() params["messages"] = [{"role": "user", "content": "test"}] @@ -545,7 +545,7 @@ async def test_test_model_connection_uses_model_info_id_to_disambiguate_duplicat mock_ahealth_check = AsyncMock(return_value=mock_health_check_result) mock_run_with_timeout = AsyncMock(return_value=mock_health_check_result) - def mock_update_params(model_info, litellm_params): + def mock_update_params(model_info, litellm_params, deployment_model: str | None = None): params = litellm_params.copy() params["messages"] = [{"role": "user", "content": "test"}] return params @@ -656,7 +656,7 @@ async def test_test_model_connection_falls_back_to_deployments_zero_without_id() mock_ahealth_check = AsyncMock(return_value=mock_health_check_result) mock_run_with_timeout = AsyncMock(return_value=mock_health_check_result) - def mock_update_params(model_info, litellm_params): + def mock_update_params(model_info, litellm_params, deployment_model: str | None = None): params = litellm_params.copy() params["messages"] = [{"role": "user", "content": "test"}] return params @@ -902,6 +902,47 @@ async def test_test_model_connection_overridden_model_probe_params_follow_the_pr assert "max_tokens" not in ahealth_check.call_args.kwargs["model_params"] +@pytest.mark.asyncio +@pytest.mark.parametrize( + ("deployment_params", "deployment_model_info"), + [ + ({"model": "azure/my-audio-deployment"}, {"id": "audio-id", "base_model": "azure/gpt-audio-1.5"}), + ({"model": "azure/my-audio-deployment", "base_model": "azure/gpt-audio-1.5"}, {"id": "audio-id"}), + ], + ids=["model_info.base_model", "litellm_params.base_model"], +) +@pytest.mark.parametrize( + ("request_params", "expects_audio_probe"), + [({}, True), ({"model": "azure/my-text-deployment"}, False)], + ids=["own model", "swapped model"], +) +async def test_test_model_connection_stored_audio_base_model_follows_the_stored_model( + deployment_params: Mapping[str, str], + deployment_model_info: Mapping[str, str], + request_params: Mapping[str, str], + expects_audio_probe: bool, +): + deployment: Final = MappingProxyType( + { + "model_name": "custom-name-audio", + "litellm_params": {**deployment_params, "api_key": "fake-azure-key", "api_base": "https://x.openai.azure.com"}, + "model_info": deployment_model_info, + } + ) + with _test_connection_probe(deployment) as ahealth_check: + await health_test_model_connection( + request=MagicMock(), + mode=None, + litellm_params=dict(request_params), + model_info={"id": "audio-id"}, + user_api_key_dict=UserAPIKeyAuth(user_id="test-user", token="test-token"), + ) + + probe_params: Final = ahealth_check.call_args.kwargs["model_params"] + assert ("modalities" in probe_params) is expects_audio_probe + assert probe_params.get("base_model") == deployment_params.get("base_model") + + @pytest.mark.asyncio @pytest.mark.parametrize("params_mode", [123, ["chat"], {"mode": "chat"}, False], ids=["int", "list", "dict", "bool"]) async def test_test_model_connection_non_string_params_mode_is_a_bad_request(params_mode: object): diff --git a/tests/unit/proxy/test_health_check_max_tokens.py b/tests/unit/proxy/test_health_check_max_tokens.py index 24117f45106..e48bf4edc09 100644 --- a/tests/unit/proxy/test_health_check_max_tokens.py +++ b/tests/unit/proxy/test_health_check_max_tokens.py @@ -840,6 +840,125 @@ def test_health_check_params_apply_to_non_chat_modes(): assert "max_tokens" not in updated +@pytest.mark.parametrize( + ("model_info", "litellm_params"), + [ + ({}, {"model": "azure/gpt-audio-mini"}), + ({}, {"model": "openai/gpt-audio-mini"}), + ({"base_model": "azure/gpt-audio-mini"}, {"model": "azure/my-audio-deployment"}), + ({}, {"model": "azure/my-audio-deployment", "base_model": "azure/gpt-audio-1.5"}), + ( + {"base_model": "azure/gpt-4o-mini", "health_check_model": "azure/gpt-audio-mini"}, + {"model": "azure/my-text-deployment"}, + ), + ( + {"base_model": "gpt-4o-mini", "health_check_params": {"model": "openai/gpt-audio-mini"}}, + {"model": "openai/my-text-model"}, + ), + ({"base_model": "openai/gpt-audio-1.5"}, {"model": "azure/my-audio-deployment"}), + ], +) +def test_openai_and_azure_audio_chat_probe_asks_for_audio_output( + model_info: dict[str, str], litellm_params: dict[str, str] +): + updated: Final = update_litellm_params_for_health_check(model_info, litellm_params) + + assert updated["modalities"] == ["text", "audio"] + assert updated["audio"] == {"voice": "alloy", "format": "wav"} + assert updated["max_tokens"] == 16 + + +@pytest.mark.parametrize( + ("model_info", "litellm_params"), + [ + ({}, {"model": "azure/gpt-4o-mini"}), + ({}, {"model": "gemini/lyria-3-clip-preview"}), + ({}, {"model": "azure/my-audio-deployment"}), + ({}, {"model": "unmapped-model-name"}), + ({"mode": "responses"}, {"model": "openai/gpt-audio-mini"}), + ({"mode": "completion"}, {"model": "openai/gpt-audio-mini"}), + ({"base_model": "gemini/lyria-3-clip-preview"}, {"model": "azure/my-audio-deployment"}), + ( + {"base_model": "azure/gpt-audio-mini", "health_check_model": "azure/gpt-4o-mini"}, + {"model": "azure/my-audio-deployment"}, + ), + ( + {"health_check_model": "azure/gpt-4o-mini"}, + {"model": "azure/my-audio-deployment", "base_model": "azure/gpt-audio-mini"}, + ), + ( + {"base_model": "gpt-audio-mini", "health_check_params": {"model": "openai/gpt-5.4-mini"}}, + {"model": "openai/gpt-audio-mini"}, + ), + ( + {"health_check_params": {"model": "openai/gpt-5.4-mini"}}, + {"model": "openai/gpt-audio-mini", "base_model": "gpt-audio-mini"}, + ), + ], +) +def test_probe_without_openai_audio_chat_model_asks_for_no_audio_output( + model_info: dict[str, str], litellm_params: dict[str, str] +): + updated: Final = update_litellm_params_for_health_check(model_info, litellm_params) + + assert "modalities" not in updated + assert "audio" not in updated + + +@pytest.mark.parametrize( + "operator_params", + [ + {"modalities": ["audio", "text"], "audio": {"voice": "verse", "format": "mp3"}}, + {"audio": {"voice": "verse", "format": "mp3"}}, + ], +) +def test_audio_chat_probe_keeps_audio_params_set_in_health_check_params(operator_params: dict[str, object]): + updated: Final = update_litellm_params_for_health_check( + {"health_check_params": operator_params}, {"model": "azure/gpt-audio-mini"} + ) + + assert {key: updated[key] for key in operator_params} == operator_params + assert updated["modalities"] == operator_params.get("modalities", ["text", "audio"]) + + +_AUDIO_OUTPUT_CHAT_MODELS: Final = tuple( + sorted( + name + for name, info in litellm.model_cost.items() + if info.get("litellm_provider") in ("openai", "azure") + and info.get("mode") == "chat" + and "audio" in (info.get("supported_output_modalities") or []) + ) +) + + +@pytest.mark.parametrize("cost_map_model", _AUDIO_OUTPUT_CHAT_MODELS) +def test_every_audio_output_chat_model_in_the_cost_map_gets_the_audio_probe(cost_map_model: str): + provider: Final = litellm.model_cost[cost_map_model]["litellm_provider"] + updated: Final = update_litellm_params_for_health_check( + {"base_model": cost_map_model}, {"model": f"{provider}/my-audio-deployment"} + ) + + assert updated["modalities"] == ["text", "audio"] + + +def test_text_only_modalities_in_health_check_params_get_no_default_audio(): + updated: Final = update_litellm_params_for_health_check( + {"health_check_params": {"modalities": ["text"]}}, {"model": "openai/gpt-audio-mini"} + ) + + assert updated["modalities"] == ["text"] + assert "audio" not in updated + + +def test_audio_chat_probe_uses_health_check_voice(): + updated: Final = update_litellm_params_for_health_check( + {"health_check_voice": "verse"}, {"model": "openai/gpt-audio-mini"} + ) + + assert updated["audio"] == {"voice": "verse", "format": "wav"} + + async def _pegasus_health_check_request_body( model_info: dict[str, object], monkeypatch: pytest.MonkeyPatch ) -> dict[str, object]: From d9c9d842982afc3129f3295709e0bb8c7ad7942c Mon Sep 17 00:00:00 2001 From: ryan-crabbe-berri Date: Wed, 28 Oct 2026 18:45:24 -0700 Subject: [PATCH 41/41] fix(proxy): stop log lines from leaving terminal colors on (#48966) * fix(proxy): stop log lines from leaving terminal colors on * test(proxy): drop source scan for unclosed terminal colors Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> --------- Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- litellm/proxy/common_request_processing.py | 4 ++-- litellm/proxy/example_config_yaml/custom_callbacks.py | 3 +-- litellm/proxy/proxy_server.py | 6 +++--- 3 files changed, 6 insertions(+), 7 deletions(-) diff --git a/litellm/proxy/common_request_processing.py b/litellm/proxy/common_request_processing.py index 9bb451495c8..89886c43525 100644 --- a/litellm/proxy/common_request_processing.py +++ b/litellm/proxy/common_request_processing.py @@ -4234,7 +4234,7 @@ class ProxyBaseLLMRequestProcessing: e = transformed_exception litellm_debug_info: Final = getattr(e, "litellm_debug_info", "") verbose_proxy_logger.debug( - "\033[1;31mAn error occurred: %s %s\n\n Debug this by setting `--debug`, e.g. `litellm --model gpt-3.5-turbo --debug`", + "An error occurred: %s %s", e, litellm_debug_info, ) @@ -4597,7 +4597,7 @@ class ProxyBaseLLMRequestProcessing: if transformed_exception is not None: e = transformed_exception verbose_proxy_logger.debug( - "\x1b[1;31mAn error occurred: %s\n\n Debug this by setting `--debug`, e.g. `litellm --model gpt-3.5-turbo --debug`", + "An error occurred: %s", e, ) diff --git a/litellm/proxy/example_config_yaml/custom_callbacks.py b/litellm/proxy/example_config_yaml/custom_callbacks.py index 5ae55060930..d11e27e0718 100644 --- a/litellm/proxy/example_config_yaml/custom_callbacks.py +++ b/litellm/proxy/example_config_yaml/custom_callbacks.py @@ -28,7 +28,7 @@ class MyCustomHandler(CustomLogger): def __init__(self): blue_color_code: Final = "\033[94m" reset_color_code: Final = "\033[0m" - print_verbose(f"{blue_color_code}Initialized LiteLLM custom logger") + print_verbose(f"{blue_color_code}Initialized LiteLLM custom logger{reset_color_code}") try: print_verbose("Logger Initialized with following methods:") methods: Final = [ @@ -40,7 +40,6 @@ class MyCustomHandler(CustomLogger): # Pretty print_verbose the methods for method in methods: print_verbose(f" - {method}") - print_verbose(f"{reset_color_code}") except Exception: pass diff --git a/litellm/proxy/proxy_server.py b/litellm/proxy/proxy_server.py index b263412ce0f..42816e7db61 100644 --- a/litellm/proxy/proxy_server.py +++ b/litellm/proxy/proxy_server.py @@ -6606,7 +6606,7 @@ class ProxyConfig: reset_color_code: Final = "\033[0m" for key, value in litellm_settings.items(): if key == "cache" and value is True: - print_startup_message(f"{blue_color_code}\nSetting Cache on Proxy") + print_startup_message(f"{blue_color_code}\nSetting Cache on Proxy{reset_color_code}") from litellm.caching.caching import Cache cache_params = {} @@ -9825,7 +9825,7 @@ async def async_assistants_data_generator(response, user_api_key_dict: UserAPIKe request_data=request_data, ) verbose_proxy_logger.debug( - "\x1b[1;31mAn error occurred: %s\n\n Debug this by setting `--debug`, e.g. `litellm --model gpt-3.5-turbo --debug`", + "An error occurred: %s", e, ) if isinstance(e, HTTPException): @@ -10655,7 +10655,7 @@ async def async_data_generator( ) client_error: Final[Exception] = e if replacement is None else replacement verbose_proxy_logger.debug( - "\x1b[1;31mAn error occurred: %s\n\n Debug this by setting `--debug`, e.g. `litellm --model gpt-3.5-turbo --debug`", + "An error occurred: %s", e, )