Merge remote-tracking branch 'origin/main' into litellm_responses_fallback_prefill

# Conflicts:
#	litellm/responses/streaming_iterator.py
This commit is contained in:
mateo-berri 2026-10-20 21:34:04 -07:00
commit 2143c6c263
267 changed files with 19685 additions and 3621 deletions

View file

@ -1291,7 +1291,7 @@ jobs:
name: Install MCP SDK1 peer
command: |
uv venv --python 3.12 .venv-mcp-peer
uv pip install --python .venv-mcp-peer 'mcp==1.28.1' 'langchain-mcp-adapters==0.2.1'
uv pip install --no-config --exclude-newer "3 days" --python .venv-mcp-peer 'mcp==1.28.1' 'langchain-mcp-adapters==0.2.1'
echo "export MCP_TEST_PEER_PYTHON=$PWD/.venv-mcp-peer/bin/python" >> "$BASH_ENV"
- when:
condition:

View file

@ -33,7 +33,7 @@ while IFS= read -r file || [ -n "$file" ]; do
*) has_backend=true ;;
esac
case "$file" in
model_prices_and_context_window.json | litellm/model_prices_and_context_window_backup.json | model_prices_and_context_window.schema.json)
model_prices_and_context_window.json | litellm/model_prices_and_context_window_backup.json)
has_cost_map=true ;;
tests/unit/proxy/*) : ;;
*) outside_cost_map_set=true ;;

View file

@ -139,6 +139,7 @@ jobs:
test-path: |-
tests/unit/batch_completion
tests/unit/decisions
tests/unit/extraction
tests/unit/litellm_core_utils
tests/unit/systemone
python-version: "3.12"

View file

@ -4,8 +4,7 @@ Every pull request whose diff against its merge base touches one of the two cost
checks: the files parse and the backup copy matches the root file. A pull request that leaves both untouched
skips them, since merging it keeps the base branch's copies and its head tree only carries whatever state the
branch was cut from. Pull requests from the cost map sync bot (branches named litellm_cost_map_sync_*) always
get the file checks and additionally may only touch those two files or the legacy schema
path and may only add or update models.
get the file checks and additionally may only touch those two files and may only add or update models.
"""
from __future__ import annotations
@ -21,7 +20,6 @@ from typing import Final
COST_MAP_PATH: Final = "model_prices_and_context_window.json"
BACKUP_PATH: Final = "litellm/model_prices_and_context_window_backup.json"
LEGACY_SCHEMA_PATH: Final = "model_prices_and_context_window.schema.json"
GUARDED_PATHS: Final = (COST_MAP_PATH, BACKUP_PATH)
BOT_BRANCH_PREFIX: Final = "litellm_cost_map_sync_"
SPECIAL_ROOT_KEYS: Final = frozenset({"sample_spec", "fallback_generalizations"})
@ -110,7 +108,7 @@ def _bot_failures(base: Snapshot, head_map: CostMap, changed_files: Sequence[str
*(
f"bot PRs may only change the cost map files, not {path}"
for path in changed_files
if path not in GUARDED_PATHS and path != LEGACY_SCHEMA_PATH
if path not in GUARDED_PATHS
),
*(f"bot PRs may not remove models: {key}" for key in base_map if key not in head_map),
*(f"bot PRs may not remove fields: {ref}" for ref in removed_fields),

View file

@ -14,6 +14,7 @@ from typing import TYPE_CHECKING, Final, List, Literal, Optional, Protocol, Tupl
from litellm_enterprise.proxy.common_utils.managed_object_attribution import (
ManagedObjectCreatorAttribution,
ManagedObjectRow,
row_metadata,
)
import litellm
@ -90,11 +91,6 @@ def _team_table(prisma_client: "PrismaClient") -> "TableActions[prisma_models.Li
return TeamRepository(prisma_client).table
def _row_metadata(row: object) -> dict[str, object]:
metadata: Final = getattr(row, "metadata", None)
return cast(dict[str, object], metadata) if isinstance(metadata, dict) else {} # cast-ok: prisma Json object column
class CheckBatchCost:
def __init__(
self,
@ -1063,7 +1059,9 @@ class CheckBatchCost:
if (has_key_row and key_row is None) or (team_id and team_row is None):
return True
settings: Final = get_dynamic_logging_metadata(
UserAPIKeyAuth(metadata=_row_metadata(key_row), team_metadata=_row_metadata(team_row), team_id=team_id),
UserAPIKeyAuth(
metadata=dict(row_metadata(key_row)), team_metadata=dict(row_metadata(team_row)), team_id=team_id
),
proxy_config,
)
callback_vars: Final = (settings.callback_vars if settings is not None else None) or {}

View file

@ -7,6 +7,12 @@ attribution, priced against the deployment that served it. The poll's status che
with BACKGROUND_RESPONSE_STATUS_POLL_CALL_ORIGIN and client reads are marked managed by the
managed files hook, so both stay free. The spend event carries the id the client holds, the row's
unified_object_id, so its spend log is found under that id with the _background_cost suffix.
The spend event reaches the creating key's or team's own logging callbacks the way that creator's
requests do. Its message content is redacted when the creator's key or team logging turns message
logging off, else as the proxy is set, the way the batch poll decides it. The row is claimed before
the creator is read, so reads that lose the race skip the lookups. A creator whose lookup fails, or
whose logging settings the logger rejects, is still billed, through the proxy's own callbacks only
and with its message content redacted.
A read that finds no row, because the create could not store one, stores it under the reading key
and bills it the same way, so a response the create left untracked is still billed once. A streamed
background create is free on its own; its stream stores the row under the creating key and bills
@ -29,6 +35,7 @@ from typing import TYPE_CHECKING, Final, cast
from litellm_enterprise.proxy.common_utils.managed_object_attribution import (
ManagedObjectCreatorAttribution,
ManagedObjectRow,
ResponseCreator,
)
from pydantic import BaseModel, ConfigDict, TypeAdapter
@ -54,11 +61,13 @@ from litellm.types.utils import (
if TYPE_CHECKING:
from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLogging
from litellm.proxy.proxy_server import ProxyConfig
from litellm.proxy.utils import PrismaClient, ProxyLogging
from litellm.repositories.prisma_protocols import TableActions
from litellm.router import Router
_STORED_RESPONSE_ADAPTER: Final[TypeAdapter[dict[str, object]]] = TypeAdapter(dict[str, object])
_CALLBACK_LIST_ADAPTER: Final[TypeAdapter[tuple[object, ...]]] = TypeAdapter(tuple[object, ...])
class _RoutedModelInfo(BaseModel):
@ -73,6 +82,8 @@ class _RoutedDeployment(BaseModel):
_ROUTED_DEPLOYMENTS: Final = TypeAdapter(tuple[_RoutedDeployment, ...])
_RETRIEVE_RESPONSE_CALL_TYPE: Final = "aget_responses"
_CALLBACK_LIST_FIELDS: Final = frozenset({"success_callback", "failure_callback", "litellm_disabled_callbacks"})
_REJECTED_CREATOR_LOGGING: Final[Mapping[str, object]] = MappingProxyType({"turn_off_message_logging": True})
@dataclass(frozen=True, slots=True)
@ -89,6 +100,7 @@ class BilledBackgroundResponse:
custom_llm_provider: str | None
litellm_params: Mapping[str, object]
metadata: Mapping[str, object]
creator_logging: Mapping[str, object]
@dataclass(frozen=True, slots=True)
@ -107,11 +119,51 @@ def _without_read_price(response: ResponsesAPIResponse) -> ResponsesAPIResponse:
return unpriced
async def log_billed_background_response(billed: BilledBackgroundResponse) -> BilledCharge:
from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLogging
from litellm.litellm_core_utils.logging_worker import GLOBAL_LOGGING_WORKER
def _string_callbacks(callbacks: object) -> tuple[str, ...]:
if not isinstance(callbacks, list):
return ()
return tuple(
callback for callback in _CALLBACK_LIST_ADAPTER.validate_python(callbacks) if isinstance(callback, str)
)
logging_obj: Final = LiteLLMLogging(
def _running_proxy_config() -> "ProxyConfig":
from litellm.proxy.proxy_server import proxy_config
return proxy_config
def creator_logging_fields(creator: ResponseCreator, proxy_config: "ProxyConfig | None") -> Mapping[str, object]:
from litellm.proxy._types import UserAPIKeyAuth
from litellm.proxy.litellm_pre_call_utils import key_and_team_logging_fields
if not creator.settings_read:
return _REJECTED_CREATOR_LOGGING
try:
return key_and_team_logging_fields(
user_api_key_dict=UserAPIKeyAuth(
metadata=dict(creator.key_metadata),
team_id=creator.team_id,
team_metadata=dict(creator.team_metadata),
),
proxy_config=proxy_config if proxy_config is not None else _running_proxy_config(),
)
except Exception as e:
verbose_proxy_logger.error(
f"CheckResponsesCost: could not read the creator's logging settings, so only the proxy's "
f"callbacks get the billed event: {e}"
)
return _REJECTED_CREATOR_LOGGING
def _billed_logging_obj(billed: BilledBackgroundResponse, creator_logging: Mapping[str, object]) -> "LiteLLMLogging":
from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLogging
from litellm.utils import split_dynamic_success_callbacks
sync_callbacks, async_callbacks = split_dynamic_success_callbacks(
_string_callbacks(creator_logging.get("success_callback"))
)
return LiteLLMLogging(
model=billed.model,
messages=[{"role": "user", "content": "<retrieve_response>"}],
stream=False,
@ -119,7 +171,27 @@ async def log_billed_background_response(billed: BilledBackgroundResponse) -> Bi
start_time=datetime.now(),
litellm_call_id=str(uuid.uuid4()),
function_id=str(uuid.uuid4()),
dynamic_success_callbacks=list(sync_callbacks),
dynamic_async_success_callbacks=list(async_callbacks) or None,
kwargs={key: value for key, value in creator_logging.items() if key not in _CALLBACK_LIST_FIELDS},
)
def _creator_or_proxy_logging_obj(billed: BilledBackgroundResponse) -> "LiteLLMLogging":
try:
return _billed_logging_obj(billed, billed.creator_logging)
except Exception as e:
verbose_proxy_logger.error(
f"CheckResponsesCost: the creator's logging settings were rejected, so only the proxy's "
f"callbacks get the billed event: {e}"
)
return _billed_logging_obj(billed, _REJECTED_CREATOR_LOGGING)
async def log_billed_background_response(billed: BilledBackgroundResponse) -> BilledCharge:
from litellm.litellm_core_utils.logging_worker import GLOBAL_LOGGING_WORKER
logging_obj: Final = _creator_or_proxy_logging_obj(billed)
logging_obj.update_environment_variables(
litellm_params={**billed.litellm_params, "metadata": dict(billed.metadata)},
optional_params={},
@ -187,6 +259,7 @@ class CheckResponsesCost:
prisma_client: "PrismaClient",
llm_router: "Router",
log_billed_response: BilledResponseLogger = log_billed_background_response,
proxy_config: "ProxyConfig | None" = None,
):
from litellm.proxy.utils import PrismaClient, ProxyLogging
from litellm.router import Router
@ -196,6 +269,7 @@ class CheckResponsesCost:
self.llm_router: Router = llm_router
self._creator_attribution: Final = ManagedObjectCreatorAttribution(prisma_client, "CheckResponsesCost")
self._log_billed_response: Final = log_billed_response
self._proxy_config: Final = proxy_config
def _resolve_deployment(self, response_id: str) -> Deployment | None:
model_id: str | None = ResponsesAPIRequestUtils.get_model_id_from_response_id(response_id)
@ -307,11 +381,12 @@ class CheckResponsesCost:
data={"status": job.status or PENDING_RESPONSE_STATUSES[0]},
)
async def _billed_response(
def _billed_response(
self,
job: ManagedObjectRow,
target: _PollTarget,
response: ResponsesAPIResponse,
creator: ResponseCreator,
) -> BilledBackgroundResponse:
deployment: Final = target.deployment
return BilledBackgroundResponse(
@ -323,13 +398,14 @@ class CheckResponsesCost:
litellm_params=_deployment_litellm_params(deployment) if deployment is not None else MappingProxyType({}),
metadata=MappingProxyType(
{
**(await self._creator_attribution.response_metadata(job, job.unified_object_id)),
**creator.metadata,
INTERNAL_CALL_ORIGIN_METADATA_KEY: BACKGROUND_RESPONSE_COST_POLL_CALL_ORIGIN,
**target.model_metadata,
**(_deployment_metadata(deployment) if deployment is not None else {}),
**self._requested_model_group(target, job.team_id),
}
),
creator_logging=creator_logging_fields(creator, self._proxy_config),
)
def _requested_model_group(self, target: _PollTarget, team_id: str | None) -> Mapping[str, object]:
@ -354,7 +430,8 @@ class CheckResponsesCost:
verbose_proxy_logger.info(f"Response {job.unified_object_id} was already billed elsewhere, skipping")
return None
try:
charge: Final = await self._log_billed_response(await self._billed_response(job, target, response))
creator: Final = await self._creator_attribution.response_creator(job, job.unified_object_id)
charge: Final = await self._log_billed_response(self._billed_response(job, target, response, creator))
except BaseException:
await self._release_billing_claim(job)
raise

View file

@ -10,12 +10,16 @@ from __future__ import annotations
from collections.abc import Mapping
from dataclasses import dataclass
from datetime import datetime
from typing import TYPE_CHECKING, Final, Protocol, cast
from typing import TYPE_CHECKING, Final, Protocol, TypeGuard, cast
from pydantic import TypeAdapter
from litellm._logging import verbose_proxy_logger
from litellm.constants import CLI_SESSION_KEY_PREFIX
from litellm.repositories.table_repositories import (
DeletedVerificationTokenRepository,
DeprecatedVerificationTokenRepository,
)
from litellm.repositories.team_repository import TeamRepository
from litellm.repositories.user_repository import UserRepository
from litellm.repositories.verification_token_repository import VerificationTokenRepository
@ -70,6 +74,7 @@ class ManagedObjectRow(Protocol):
_TAG_LIST_ADAPTER: Final[TypeAdapter[list[object]]] = TypeAdapter(list[object])
_KEY_OWNER_SCOPE_PREFIX: Final = "key:"
_MAX_ROTATION_HOPS: Final = 16
def _string_tags(request_tags: object) -> list[str] | None:
@ -96,6 +101,20 @@ def _team_table(prisma_client: PrismaClient) -> TableActions[prisma_models.LiteL
return TeamRepository(prisma_client).table
def _rotated_token_table(
prisma_client: PrismaClient,
) -> TableActions[prisma_models.LiteLLM_DeprecatedVerificationToken]:
return DeprecatedVerificationTokenRepository(prisma_client).table
def _deleted_token_table(prisma_client: PrismaClient) -> TableActions[prisma_models.LiteLLM_DeletedVerificationToken]:
return DeletedVerificationTokenRepository(prisma_client).table
def _deleted_team_table(prisma_client: PrismaClient) -> TableActions[prisma_models.LiteLLM_DeletedTeamTable]:
return TeamRepository(prisma_client).deleted_table
@dataclass(frozen=True, slots=True)
class _CreatorRows:
key: prisma_models.LiteLLM_VerificationToken | None
@ -103,6 +122,15 @@ class _CreatorRows:
user: prisma_models.LiteLLM_UserTable | None
@dataclass(frozen=True, slots=True)
class ResponseCreator:
metadata: Mapping[str, object]
key_metadata: Mapping[str, object]
team_id: str | None
team_metadata: Mapping[str, object]
settings_read: bool = True
class ManagedObjectCreatorAttribution:
"""Rebuilds the spend-tracking metadata of whoever created a managed object, so the poll
job's spend row is attributed like the creating request: key hash, team, organization,
@ -117,7 +145,12 @@ class ManagedObjectCreatorAttribution:
A background response also carries its end user, the creating key's project and its alias,
and the key, team, and user budgets and per-model budgets, the way the read that billed it
before did, and never names a key-scoped creator as its user. A key that sets no budget of
its own takes its budget tier's, the same rule auth applies.
its own takes its budget tier's, the same rule auth applies. Its creator's key and team
metadata come along too, so the billed event reaches the key's and team's logging callbacks
like that read did. A rotated key's metadata is read from the live key its rotations lead
to, and a deleted key's or team's from its deleted row. A lookup that fails leaves the
creator's settings unread and the attribution to the rows that still load, so the
response still bills and still counts toward their budgets.
"""
def __init__(self, prisma_client: PrismaClient, poller_name: str) -> None:
@ -128,31 +161,70 @@ class ManagedObjectCreatorAttribution:
rows: Final = await self._rows(job, object_id)
return {"user_api_key_user_id": job.created_by, **_shared_metadata(job, rows)}
async def response_metadata(self, job: ManagedObjectRow, object_id: str) -> Mapping[str, object]:
rows: Final = await self._rows(job, object_id)
end_user_id: Final = getattr(job, "end_user_id", None)
return {
"user_api_key_user_id": _creator_user_id(job.created_by),
**_shared_metadata(job, rows),
**({"user_api_key_end_user_id": end_user_id} if end_user_id else {}),
"user_api_key_project_id": getattr(rows.key, "project_id", None),
"user_api_key_project_alias": getattr(
getattr(rows.key, "litellm_project_table", None), "project_alias", None
async def response_creator(self, job: ManagedObjectRow, object_id: str) -> ResponseCreator:
try:
return await self._read_response_creator(job, object_id)
except Exception as e:
verbose_proxy_logger.error(
f"{self._poller_name}: could not read the creator of {object_id}, so it bills without their "
f"settings: {e}"
)
return ResponseCreator(
metadata=_response_creator_metadata(job, await self._rows(job, object_id)),
key_metadata={},
team_id=getattr(job, "team_id", None),
team_metadata={},
settings_read=False,
)
async def _read_response_creator(self, job: ManagedObjectRow, object_id: str) -> ResponseCreator:
api_key: Final = getattr(job, "api_key", None)
team_id: Final = getattr(job, "team_id", None)
rows: Final = _CreatorRows(
key=await self._live_key_row(api_key, job.created_by),
team=await self._live_team_row(team_id),
user=await self._user_row(object_id, _creator_user_id(job.created_by)),
)
return ResponseCreator(
metadata=_response_creator_metadata(job, rows),
key_metadata=row_metadata(rows.key) if rows.key is not None else await self._retired_key_metadata(api_key),
team_id=team_id,
team_metadata=(
row_metadata(rows.team) if rows.team is not None else await self._deleted_team_metadata(team_id)
),
**_key_budget_fields(rows.key),
**_budget_fields(
rows.team,
"user_api_key_team_spend",
"user_api_key_team_max_budget",
"user_api_key_team_model_max_budget",
),
**_budget_fields(
rows.user,
"user_api_key_user_spend",
"user_api_key_user_max_budget",
"user_api_key_user_model_max_budget",
),
}
)
async def _retired_key_metadata(self, api_key: str | None) -> Mapping[str, object]:
if not _names_a_key_row(api_key):
return {}
current_token: Final = await self._token_rotated_to(api_key, _MAX_ROTATION_HOPS)
current_key: Final = (
await _token_table(self._prisma_client).find_unique(where={"token": current_token})
if current_token != api_key
else None
)
if current_key is not None:
return row_metadata(current_key)
return row_metadata(
await _deleted_token_table(self._prisma_client).find_first(
where={"token": current_token}, order={"deleted_at": "desc"}
)
)
async def _token_rotated_to(self, token: str, hops_left: int) -> str:
rotation: Final = await _rotated_token_table(self._prisma_client).find_unique(where={"token": token})
if rotation is None or hops_left == 0:
return token
return await self._token_rotated_to(rotation.active_token_id, hops_left - 1)
async def _deleted_team_metadata(self, team_id: str | None) -> Mapping[str, object]:
if not team_id:
return {}
return row_metadata(
await _deleted_team_table(self._prisma_client).find_first(
where={"team_id": team_id}, order={"deleted_at": "desc"}
)
)
async def _rows(self, job: ManagedObjectRow, object_id: str) -> _CreatorRows:
return _CreatorRows(
@ -175,25 +247,57 @@ class ManagedObjectCreatorAttribution:
async def _key_row(
self, object_id: str, api_key: str | None, created_by: str | None
) -> prisma_models.LiteLLM_VerificationToken | None:
if not api_key or _is_cli_session_key(api_key, created_by):
return None
try:
return await _token_table(self._prisma_client).find_unique(
where={"token": api_key}, include={"litellm_budget_table": True, "litellm_project_table": True}
)
return await self._live_key_row(api_key, created_by)
except Exception as e:
verbose_proxy_logger.error(f"{self._poller_name}: could not look up the creating key for {object_id}: {e}")
return None
async def _team_row(self, object_id: str, team_id: str | None) -> prisma_models.LiteLLM_TeamTable | None:
if not team_id:
async def _live_key_row(
self, api_key: str | None, created_by: str | None
) -> prisma_models.LiteLLM_VerificationToken | None:
if not api_key or _is_cli_session_key(api_key, created_by):
return None
return await _token_table(self._prisma_client).find_unique(
where={"token": api_key}, include={"litellm_budget_table": True, "litellm_project_table": True}
)
async def _team_row(self, object_id: str, team_id: str | None) -> prisma_models.LiteLLM_TeamTable | None:
try:
return await _team_table(self._prisma_client).find_unique(where={"team_id": team_id})
return await self._live_team_row(team_id)
except Exception as e:
verbose_proxy_logger.error(f"{self._poller_name}: could not look up team {team_id} for {object_id}: {e}")
return None
async def _live_team_row(self, team_id: str | None) -> prisma_models.LiteLLM_TeamTable | None:
if not team_id:
return None
return await _team_table(self._prisma_client).find_unique(where={"team_id": team_id})
def _response_creator_metadata(job: ManagedObjectRow, rows: _CreatorRows) -> Mapping[str, object]:
end_user_id: Final = getattr(job, "end_user_id", None)
return {
"user_api_key_user_id": _creator_user_id(job.created_by),
**_shared_metadata(job, rows),
**({"user_api_key_end_user_id": end_user_id} if end_user_id else {}),
"user_api_key_project_id": getattr(rows.key, "project_id", None),
"user_api_key_project_alias": getattr(getattr(rows.key, "litellm_project_table", None), "project_alias", None),
**_key_budget_fields(rows.key),
**_budget_fields(
rows.team,
"user_api_key_team_spend",
"user_api_key_team_max_budget",
"user_api_key_team_model_max_budget",
),
**_budget_fields(
rows.user,
"user_api_key_user_spend",
"user_api_key_user_max_budget",
"user_api_key_user_model_max_budget",
),
}
def _shared_metadata(job: ManagedObjectRow, rows: _CreatorRows) -> Mapping[str, object]:
api_key: Final = getattr(job, "api_key", None)
@ -221,6 +325,19 @@ def _shared_metadata(job: ManagedObjectRow, rows: _CreatorRows) -> Mapping[str,
}
def row_metadata(row: object) -> Mapping[str, object]:
metadata: Final = getattr(row, "metadata", None)
if not isinstance(metadata, dict):
return {}
return cast(Mapping[str, object], metadata) # cast-ok: prisma Json object column
def _names_a_key_row(api_key: object) -> TypeGuard[str]:
from litellm.litellm_core_utils.litellm_logging import is_valid_sha256_hash
return isinstance(api_key, str) and is_valid_sha256_hash(api_key)
def _is_cli_session_key(api_key: str, created_by: str | None) -> bool:
return bool(created_by) and api_key == f"{CLI_SESSION_KEY_PREFIX}-{created_by}"

View file

@ -1,6 +1,6 @@
[project]
name = "litellm-enterprise"
version = "0.1.78"
version = "0.1.79"
description = "Package for LiteLLM Enterprise features"
readme = "README.md"
requires-python = ">=3.9"
@ -26,7 +26,7 @@ required-version = ">=0.10.9"
module-root = ""
[tool.commitizen]
version = "0.1.78"
version = "0.1.79"
version_files = [
"pyproject.toml:^version",
"../pyproject.toml:litellm-enterprise==",

View file

@ -1,7 +1,7 @@
"""Path allowlist for the gateway component.
The gateway exposes the LLM data-plane surface: chat/completions, embeddings,
audio, batches, files, fine-tuning, rerank, decisions, ocr, rag, video, search, image,
audio, batches, files, fine-tuning, rerank, decisions, extraction, ocr, rag, video, search, image,
responses, vector stores, passthrough providers, realtime websockets, MCP
tool-call endpoints, and operational endpoints (/health, /metrics, and the
/debug/memory/summary read of the serving worker's RSS).
@ -64,6 +64,8 @@ GATEWAY_PATH_PREFIXES: tuple[str, ...] = (
"/decisions",
"/v1/systemone",
"/systemone",
"/v1/extraction",
"/extraction",
"/v1/ocr",
"/ocr",
"/v1/rag/",

View file

@ -1,6 +1,6 @@
[project]
name = "litellm-proxy-extras"
version = "0.4.110"
version = "0.4.111"
description = "Additional files for the LiteLLM Proxy. Reduces the size of the main litellm package."
readme = "README.md"
requires-python = ">=3.9"
@ -30,7 +30,7 @@ required-version = ">=0.10.9"
module-root = ""
[tool.commitizen]
version = "0.4.110"
version = "0.4.111"
version_files = [
"pyproject.toml:^version",
"../pyproject.toml:litellm-proxy-extras==",

View file

@ -1505,6 +1505,7 @@ from .rag.main import *
from .sandbox.main import *
from .decisions.main import *
from .systemone.main import *
from .extraction.main import *
from .tool_loop import ToolLoopMaxRoundsExceeded, arun_tool_loop, run_tool_loop
from .search.main import *
from .realtime_api.main import (
@ -1711,6 +1712,9 @@ if TYPE_CHECKING:
from .llms.cloudflare.decisions.transformation import (
CloudflareDecisionsConfig as CloudflareDecisionsConfig,
)
from .llms.cloudflare.extraction.transformation import (
CloudflareExtractionConfig as CloudflareExtractionConfig,
)
from .llms.strands_decider.decisions.transformation import (
StrandsDeciderDecisionsConfig as StrandsDeciderDecisionsConfig,
)

View file

@ -166,6 +166,7 @@ LLM_CONFIG_NAMES: Final = (
"OpenRouterDecisionsConfig",
"FireworksAIDecisionsConfig",
"CloudflareDecisionsConfig",
"CloudflareExtractionConfig",
"StrandsDeciderDecisionsConfig",
"DatabricksDecisionsConfig",
"HostedVLLMDecisionsConfig",
@ -735,6 +736,7 @@ _LLM_CONFIGS_IMPORT_MAP: Final = {
"OpenRouterDecisionsConfig": (".llms.openrouter.decisions.transformation", "OpenRouterDecisionsConfig"),
"FireworksAIDecisionsConfig": (".llms.fireworks_ai.decisions.transformation", "FireworksAIDecisionsConfig"),
"CloudflareDecisionsConfig": (".llms.cloudflare.decisions.transformation", "CloudflareDecisionsConfig"),
"CloudflareExtractionConfig": (".llms.cloudflare.extraction.transformation", "CloudflareExtractionConfig"),
"StrandsDeciderDecisionsConfig": (
".llms.strands_decider.decisions.transformation",
"StrandsDeciderDecisionsConfig",

View file

@ -95,7 +95,10 @@ class InMemoryCache(BaseCache):
"""
Check if a specific key is expired
"""
return key in self.ttl_dict and self._clock() > self.ttl_dict[key]
try:
return self._clock() > self.ttl_dict[key]
except KeyError:
return False
def _cache_keys(self) -> tuple[str, ...]:
return tuple(self.cache_dict) # pyright: ignore[reportUnknownVariableType] # cache_dict is untyped legacy state
@ -227,7 +230,10 @@ class InMemoryCache(BaseCache):
if key in self.cache_dict:
if self.evict_element_if_expired(key):
return None
original_cached_response: Final = self.cache_dict[key]
try:
original_cached_response: Final = self.cache_dict[key]
except KeyError:
return None
try:
cached_response = json.loads(original_cached_response)
except Exception:

View file

@ -156,6 +156,9 @@ DEFAULT_REPLICATE_POLLING_RETRIES: Final = get_env_int("DEFAULT_REPLICATE_POLLIN
DEFAULT_REPLICATE_POLLING_DELAY_SECONDS: Final = get_env_int("DEFAULT_REPLICATE_POLLING_DELAY_SECONDS", 1)
DEFAULT_IMAGE_TOKEN_COUNT: Final = get_env_int("DEFAULT_IMAGE_TOKEN_COUNT", 250)
HF_CONFIG_FETCH_TIMEOUT_SECONDS: Final = 10.0
HUGGINGFACE_TOKENIZER_LOAD_RETRY_SECONDS: Final = get_env_int_in_range(
"HUGGINGFACE_TOKENIZER_LOAD_RETRY_SECONDS", 300, 1, 86400
)
# Maximum wall-clock seconds a streaming response is allowed to run.
# Streams exceeding this duration are terminated with a Timeout error.
@ -258,6 +261,9 @@ MCP_PER_USER_TOKEN_EXPIRY_BUFFER_SECONDS: Final = get_env_int("MCP_PER_USER_TOKE
# MCP timeout defaults (seconds). Override via env vars for slow/custom MCP servers.
MCP_CLIENT_TIMEOUT: Final = get_env_float("LITELLM_MCP_CLIENT_TIMEOUT", 60.0)
MCP_MAX_SSE_EVENT_SIZE_BYTES: Final = get_env_int_in_range(
"LITELLM_MCP_MAX_SSE_EVENT_SIZE_BYTES", 64 * 1024 * 1024, 1, sys.maxsize
)
MCP_TOOL_LISTING_TIMEOUT: Final = get_env_float("LITELLM_MCP_TOOL_LISTING_TIMEOUT", 30.0)
MCP_UPSTREAM_RATE_LIMIT_MAX_RETRIES: Final = get_env_int("LITELLM_MCP_UPSTREAM_RATE_LIMIT_MAX_RETRIES", 3)
MCP_UPSTREAM_RATE_LIMIT_BASE_DELAY_SECONDS: Final = get_env_float(
@ -1651,6 +1657,7 @@ BAGGAGE_MAPPED_TAGS_METADATA_KEY: Final = "_baggage_mapped_tags"
VECTOR_STORE_CALLER_METADATA_KWARG: Final = "litellm_vector_store_caller_metadata"
ROUTER_USAGE_COUNTED_TOKENS_METADATA_KEY: Final = "_litellm_router_usage_counted_tokens"
INTERNAL_CALL_ORIGIN_METADATA_KEY: Final = "internal_call_origin"
BILLED_RETRY_ATTEMPT_METADATA_KEY: Final = "billed_retry_attempt"
MANAGED_BACKGROUND_RESPONSE_METADATA_KEY: Final = "_managed_background_response"
SESSION_ID_GENERATED_METADATA_KEY: Final = "litellm_session_id_generated"
SESSION_ID_OMITTED_METADATA_KEY: Final = "litellm_session_id_omitted"

View file

@ -112,6 +112,7 @@ from litellm.llms.xai.cost_calculator import cost_per_token as xai_cost_per_toke
from litellm.responses.utils import ResponseAPILoggingUtils
from litellm.types.agents import LiteLLMSendMessageResponse
from litellm.types.decisions import DecisionsResponse, DecisionsUsage, OpenAIDecisionResponse, OpenAIDecisionUsage
from litellm.types.extraction import ExtractionResponse, ExtractionUsage
from litellm.types.llms.base import CachedTokensDetails, LiteLLMBaseModel
from litellm.types.llms.openai import (
HttpxBinaryResponseContent,
@ -1171,6 +1172,8 @@ def get_usage_object(
if isinstance(completion_response, (DecisionsResponse, OpenAIDecisionResponse)):
return None if completion_response.usage is None else _decisions_usage(completion_response.usage)
if isinstance(completion_response, ExtractionResponse):
return _extraction_usage(completion_response.usage)
if isinstance(completion_response, HttpxBinaryResponseContent):
return completion_response.usage
if usage_obj is None:
@ -1227,6 +1230,14 @@ def _decisions_usage(usage: DecisionsUsage | OpenAIDecisionUsage) -> Usage:
)
def _extraction_usage(usage: ExtractionUsage) -> Usage:
return Usage(
prompt_tokens=usage.input_tokens,
completion_tokens=usage.output_tokens,
total_tokens=usage.input_tokens + usage.output_tokens,
)
def _is_known_usage_objects(usage_obj):
"""Returns True if the usage obj is a known Usage type"""
return (
@ -1234,6 +1245,7 @@ def _is_known_usage_objects(usage_obj):
or isinstance(usage_obj, ResponseAPIUsage)
or isinstance(usage_obj, DecisionsUsage)
or isinstance(usage_obj, OpenAIDecisionUsage)
or isinstance(usage_obj, ExtractionUsage)
or TranscriptionUsageObjectTransformation.is_transcription_usage_object(usage_obj)
)
@ -1263,6 +1275,8 @@ def infer_call_type(call_type: CallTypesLiteral | None, completion_response: obj
return "send_message"
elif isinstance(completion_response, (DecisionsResponse, OpenAIDecisionResponse)):
return "adecisions"
elif isinstance(completion_response, ExtractionResponse):
return "aextraction"
return call_type
@ -1659,8 +1673,10 @@ def completion_cost(
verbose_logger.debug("selected model name for cost calculation: %s", model)
if completion_response is not None and (
isinstance(completion_response, BaseModel) or isinstance(completion_response, dict)
): # tts returns a custom class
isinstance(completion_response, (BaseModel, dict))
or isinstance(completion_response, HttpxBinaryResponseContent)
and completion_response.usage is not None
):
if isinstance(completion_response, dict):
usage_obj: dict | Usage | None = completion_response.get("usage", {})
else:
@ -1674,6 +1690,8 @@ def completion_cost(
)
if isinstance(usage_obj, (DecisionsUsage, OpenAIDecisionUsage)):
_usage = _decisions_usage(usage_obj).model_dump()
elif isinstance(usage_obj, ExtractionUsage):
_usage = _extraction_usage(usage_obj).model_dump()
elif usage_obj is None:
_usage = {}
elif isinstance(usage_obj, BaseModel):
@ -2199,7 +2217,8 @@ def response_cost_calculator(
| Response
| SearchResponse
| DecisionsResponse
| OpenAIDecisionResponse,
| OpenAIDecisionResponse
| ExtractionResponse,
model: str,
custom_llm_provider: str | None,
call_type: Literal[
@ -2225,6 +2244,8 @@ def response_cost_calculator(
"adecisions",
"systemone",
"asystemone",
"extraction",
"aextraction",
],
optional_params: dict,
cache_hit: bool | None = None,

View file

@ -122,6 +122,7 @@ class SpeechToCompletionBridgeHandler:
if isinstance(result, ModelResponse):
return self.transformation_handler.transform_response(
model_response=result,
model=model,
response_format=requested_response_format if isinstance(requested_response_format, str) else None,
)
else:

View file

@ -5,6 +5,7 @@ from typing import TYPE_CHECKING, Final, cast
from typing_extensions import NotRequired, ReadOnly, TypedDict
from litellm.constants import OPENAI_CHAT_COMPLETION_PARAMS
from litellm.litellm_core_utils.core_helpers import drops_unsupported_params
from litellm.litellm_core_utils.hidden_params import get_hidden_params, served_from_cache, set_hidden_param
if TYPE_CHECKING:
@ -38,6 +39,11 @@ class ChatAudioParam(TypedDict):
format: ReadOnly[NotRequired[str]]
class ChatSpeechConfigAudioParam(TypedDict):
speech_config: ReadOnly[Mapping[str, object]]
format: ReadOnly[str]
class SpeechToCompletionBridgeTransformationHandler:
def _validate_response_format(
self, model: str, custom_llm_provider: str, optional_params: Mapping[str, object]
@ -49,6 +55,20 @@ class SpeechToCompletionBridgeTransformationHandler:
return
from litellm.exceptions import BadRequestError
if custom_llm_provider.startswith("vertex_ai"):
from litellm.llms.vertex_ai.text_to_speech.transformation import VertexAITextToSpeechConfig
vertex_supported: Final = ", ".join(
sorted({*GEMINI_TTS_SUPPORTED_RESPONSE_FORMATS, *VertexAITextToSpeechConfig.GEMINI_FORMAT_MAPPINGS})
)
raise BadRequestError(
message=(
f"Vertex AI Gemini TTS does not support response_format='{response_format}'."
f" Supported response formats: {vertex_supported}."
),
model=model,
llm_provider=custom_llm_provider,
)
supported: Final = ", ".join(sorted(GEMINI_TTS_SUPPORTED_RESPONSE_FORMATS))
raise BadRequestError(
message=(
@ -75,8 +95,27 @@ class SpeechToCompletionBridgeTransformationHandler:
return response_format if isinstance(response_format, str) else None
def _chat_audio_param(
self, model: str, voice: str | Mapping[str, object] | None, optional_params: Mapping[str, object]
) -> ChatAudioParam | None:
self,
model: str,
voice: str | Mapping[str, object] | None,
optional_params: Mapping[str, object],
drop_params: bool,
custom_llm_provider: str,
) -> ChatAudioParam | ChatSpeechConfigAudioParam | None:
import litellm
if isinstance(voice, Mapping) and self._is_gemini_tts_model(litellm.model_alias_map.get(model, model)):
from litellm.llms.vertex_ai.gemini.vertex_and_google_ai_studio_gemini import (
normalize_gemini_speech_config,
reject_unknown_gemini_tts_voice_keys,
)
reject_unknown_gemini_tts_voice_keys(model, voice, drop_params, custom_llm_provider)
speech_audio: Final[ChatSpeechConfigAudioParam] = {
"speech_config": normalize_gemini_speech_config(voice, drops_unsupported_params(drop_params)),
"format": GEMINI_TTS_CHAT_AUDIO_FORMAT,
}
return speech_audio
if not isinstance(voice, str):
return None
audio_format: Final = self._chat_audio_format(model, optional_params)
@ -90,9 +129,9 @@ class SpeechToCompletionBridgeTransformationHandler:
self,
model: str,
input: str,
voice: str | dict | None,
optional_params: dict,
litellm_params: dict,
voice: str | Mapping[str, object] | None,
optional_params: Mapping[str, object],
litellm_params: Mapping[str, object],
headers: dict,
litellm_logging_obj: "LiteLLMLoggingObj",
custom_llm_provider: str,
@ -104,7 +143,13 @@ class SpeechToCompletionBridgeTransformationHandler:
"messages": [user_message],
"modalities": ["audio"],
**self._chat_completion_params(optional_params),
"audio": self._chat_audio_param(model, voice, optional_params),
"audio": self._chat_audio_param(
model,
voice,
optional_params,
drops_unsupported_params(litellm_params.get("drop_params")),
custom_llm_provider,
),
**litellm_params,
"headers": headers,
"litellm_logging_obj": litellm_logging_obj,
@ -153,8 +198,9 @@ class SpeechToCompletionBridgeTransformationHandler:
return wav_header + pcm_data
def _is_gemini_tts_model(self, model: str) -> bool:
"""Check if the model is a Gemini TTS model that returns PCM16 data."""
return "gemini" in model.lower() and ("tts" in model.lower() or "preview-tts" in model.lower())
from litellm.utils import is_gemini_tts_model
return is_gemini_tts_model(model)
def _gemini_tts_response_body(self, decoded_audio: bytes, response_format: str | None) -> tuple[bytes, str]:
if response_format == GEMINI_TTS_RAW_RESPONSE_FORMAT:
@ -162,7 +208,7 @@ class SpeechToCompletionBridgeTransformationHandler:
return self._convert_pcm16_to_wav(decoded_audio), "audio/wav"
def transform_response(
self, model_response: "ModelResponse", response_format: str | None
self, model_response: "ModelResponse", model: str, response_format: str | None
) -> "HttpxBinaryResponseContent":
import base64
@ -176,10 +222,9 @@ class SpeechToCompletionBridgeTransformationHandler:
raise ValueError("No audio part found in the response")
decoded_audio: Final = base64.b64decode(audio_part.data)
model: Final = getattr(model_response, "model", "")
content, content_type = (
self._gemini_tts_response_body(decoded_audio, response_format)
if self._is_gemini_tts_model(model)
if self._is_gemini_tts_model(model) or self._is_gemini_tts_model(model_response.model or "")
else (decoded_audio, "audio/mpeg")
)
response: Final = httpx.Response(

View file

@ -95,6 +95,7 @@ from pydantic import AnyUrl, TypeAdapter
from litellm._logging import verbose_logger
from litellm.constants import (
MCP_CLIENT_TIMEOUT,
MCP_MAX_SSE_EVENT_SIZE_BYTES,
MCP_NPM_CACHE_DIR,
MCP_TOOL_LISTING_MAX_PAGES,
MCP_TOOL_LISTING_TIMEOUT,
@ -647,6 +648,7 @@ class MCPClient:
transport_ctx: Final = streamable_http_client(
url=self.server_url,
http_client=http_client,
max_sse_event_size=MCP_MAX_SSE_EVENT_SIZE_BYTES,
)
return transport_ctx, http_client

View file

@ -0,0 +1,3 @@
from litellm.extraction.main import aextraction, extraction
__all__ = ["aextraction", "extraction"]

351
litellm/extraction/main.py Normal file
View file

@ -0,0 +1,351 @@
import asyncio
from collections.abc import Mapping, Sequence
from dataclasses import dataclass
from functools import partial
from types import MappingProxyType
from typing import Final, TypeAlias
import httpx
from pydantic import TypeAdapter, ValidationError
import litellm
from litellm.decisions.call import log_caller_request
from litellm.litellm_core_utils.core_helpers import get_metadata_variable_name_from_kwargs
from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
from litellm.litellm_core_utils.tool_fields import DropPolicy, drop_policy
from litellm.llms.base_llm.chat.transformation import BaseLLMException
from litellm.llms.base_llm.extraction.transformation import BaseExtractionConfig
from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler, HTTPHandler
from litellm.llms.custom_httpx.llm_http_handler import BaseLLMHTTPHandler
from litellm.types.extraction import (
EXTRACTION_STATE_REWRITTEN_METADATA_KEY,
ExtractionAnswer,
ExtractionBase64Image,
ExtractionField,
ExtractionOptions,
ExtractionRequestBody,
ExtractionResponse,
)
from litellm.types.router import GenericLiteLLMParams
from litellm.types.utils import LlmProviders, is_litellm_owned_kwarg
from litellm.utils import (
ProviderConfigManager,
client, # pyright: ignore[reportUnknownVariableType] # client is an untyped decorator
get_model_cost_key,
)
ExtractionFields: TypeAlias = Mapping[str, ExtractionField | Mapping[str, object]]
ExtractionImages: TypeAlias = Sequence[str | ExtractionBase64Image | Mapping[str, object]]
ExtractionOptionsParam: TypeAlias = ExtractionOptions | Mapping[str, object]
_REQUEST_ADAPTER: Final[TypeAdapter[ExtractionRequestBody]] = TypeAdapter(ExtractionRequestBody)
_METADATA_ADAPTER: Final[TypeAdapter[Mapping[str, object]]] = TypeAdapter(Mapping[str, object])
_HANDLER: Final = BaseLLMHTTPHandler()
_LOCAL_KWARGS: Final = frozenset(
(
*GenericLiteLLMParams.model_fields,
"extra_headers",
"base_url",
"default_headers",
"deployment_id",
"additional_drop_params",
)
)
@dataclass(frozen=True, slots=True, repr=False)
class _ExtractionCall:
model: str
requested_model: str
custom_llm_provider: str
provider_config: BaseExtractionConfig
body: Mapping[str, object]
api_base: str
api_key: str | None
logging_obj: LiteLLMLoggingObj | None
headers: Mapping[str, str]
timeout: float | httpx.Timeout | None
litellm_params: Mapping[str, object]
client: HTTPHandler | AsyncHTTPHandler | None
def supported_extraction_providers() -> tuple[str, ...]:
return tuple(
provider.value
for provider in LlmProviders
if ProviderConfigManager.get_provider_extraction_config(model="", provider=provider) is not None
)
def _provider_config(model: str, custom_llm_provider: str) -> BaseExtractionConfig:
provider: Final = next((member for member in LlmProviders if member.value == custom_llm_provider), None)
provider_config: Final = (
None if provider is None else ProviderConfigManager.get_provider_extraction_config(model, provider)
)
if provider_config is None:
raise litellm.BadRequestError(
message=(
f"Unknown Extraction provider '{custom_llm_provider}'. "
f"Supported providers: {', '.join(supported_extraction_providers())}"
),
model=model,
llm_provider=custom_llm_provider,
)
return provider_config
def _request_fields(
*,
state: str | None,
fields: ExtractionFields | None,
images: ExtractionImages | None,
reference_date: str | None,
options: ExtractionOptionsParam | None,
) -> Mapping[str, object]:
sent: Final[Mapping[str, object | None]] = MappingProxyType(
{"state": state, "fields": fields, "images": images, "reference_date": reference_date, "options": options}
)
return MappingProxyType({name: value for name, value in sent.items() if value is not None})
def _is_unsupported(name: str, value: object, policy: DropPolicy) -> bool:
return (
value is not None
and value is not False
and name not in _LOCAL_KWARGS
and not is_litellm_owned_kwarg(name)
and not policy.drops(name)
)
def _unsupported_params(kwargs: Mapping[str, object]) -> tuple[str, ...]:
policy: Final = drop_policy(kwargs)
return tuple(sorted(name for name, value in kwargs.items() if _is_unsupported(name, value, policy)))
def _prepare_call(
*,
model: str,
request_fields: Mapping[str, object],
api_key: str | None,
api_base: str | None,
timeout: float | httpx.Timeout | None,
custom_llm_provider: str | None,
extra_headers: Mapping[str, str] | None,
kwargs: Mapping[str, object],
) -> _ExtractionCall:
log_caller_request(kwargs, request_fields)
upstream_model, provider, dynamic_api_key, dynamic_api_base = litellm.get_llm_provider(
model=model,
custom_llm_provider=custom_llm_provider,
api_base=api_base,
api_key=api_key,
)
provider_config: Final = _provider_config(model, provider)
unsupported: Final = _unsupported_params(kwargs)
if unsupported:
raise litellm.UnsupportedParamsError(
message=(
f"{provider} does not support parameters: {list(unsupported)}, for model={model}. "
"To drop these, set `litellm.drop_params=True` or for proxy:\n\n"
"`litellm_settings:\n drop_params: true`\n"
),
model=model,
llm_provider=provider,
)
try:
request: Final = _REQUEST_ADAPTER.validate_python(request_fields)
except ValidationError as error:
raise litellm.BadRequestError(
message=f"Invalid Extraction request: {error}", model=model, llm_provider=provider
) from error
canonical_model: Final = provider_config.canonical_model(upstream_model)
resolved_api_base: Final = provider_config.resolve_api_base(dynamic_api_base or api_base)
if resolved_api_base is None:
raise litellm.BadRequestError(
message=provider_config.missing_api_base_message(provider), model=model, llm_provider=provider
)
resolved_api_key: Final = provider_config.resolve_api_key(dynamic_api_key or api_key)
if resolved_api_key is None:
raise litellm.BadRequestError(
message=f"Missing API key for Extraction provider '{provider}'", model=model, llm_provider=provider
)
logging_obj: Final = kwargs.get("litellm_logging_obj")
http_client: Final = kwargs.get("client")
if isinstance(logging_obj, LiteLLMLoggingObj):
logging_obj.update_from_kwargs( # pyright: ignore[reportUnknownMemberType] # Logging takes untyped dicts
kwargs=dict(kwargs),
model=canonical_model,
litellm_params={
"litellm_call_id": kwargs.get("litellm_call_id"),
"api_base": provider_config.get_complete_url(resolved_api_base, canonical_model),
},
custom_llm_provider=provider,
)
return _ExtractionCall(
model=canonical_model,
requested_model=model,
custom_llm_provider=provider,
provider_config=provider_config,
body=provider_config.transform_extraction_request(canonical_model, request),
api_base=resolved_api_base,
api_key=resolved_api_key,
logging_obj=logging_obj if isinstance(logging_obj, LiteLLMLoggingObj) else None,
headers=MappingProxyType(dict(extra_headers or {})),
timeout=timeout,
litellm_params=kwargs,
client=http_client if isinstance(http_client, HTTPHandler | AsyncHTTPHandler) else None,
)
def _state_rewritten_by_guardrail(litellm_params: Mapping[str, object]) -> bool:
metadata: Final = litellm_params.get(get_metadata_variable_name_from_kwargs(litellm_params))
return (
isinstance(metadata, Mapping)
and _METADATA_ADAPTER.validate_python(metadata).get(EXTRACTION_STATE_REWRITTEN_METADATA_KEY) is True
)
def _without_spans(answers: Mapping[str, ExtractionAnswer]) -> Mapping[str, ExtractionAnswer]:
return {name: answer.model_copy(update={"span": None, "spans": None}) for name, answer in answers.items()}
def _formatted_response(response: ExtractionResponse, call: _ExtractionCall) -> ExtractionResponse:
spans_index_hidden_text: Final = response.state_text is None and _state_rewritten_by_guardrail(call.litellm_params)
formatted: Final = response.model_copy(
update={
"model": call.requested_model,
**({"answers": _without_spans(response.answers)} if spans_index_hidden_text else {}),
}
)
formatted.set_hidden_params(
{
"model": f"{call.custom_llm_provider}/{call.model}",
"custom_llm_provider": call.custom_llm_provider,
"provider_response_model": (
get_model_cost_key(f"{call.custom_llm_provider}/{response.model}")
or f"{call.custom_llm_provider}/{call.model}"
),
}
)
return formatted
def _upstream_exception(error: Exception, call: _ExtractionCall) -> Exception:
if isinstance(error, BaseLLMException) and error.status_code_is_synthesized:
return litellm.APIConnectionError(
message=f"{call.custom_llm_provider.capitalize()}Exception - {error.message}",
llm_provider=call.custom_llm_provider,
model=f"{call.custom_llm_provider}/{call.model}",
)
return litellm.exception_type( # pyright: ignore[reportUnknownMemberType, reportUnknownVariableType] # untyped
model=f"{call.custom_llm_provider}/{call.model}",
custom_llm_provider=call.custom_llm_provider,
original_exception=error,
)
async def _asend(call: _ExtractionCall) -> ExtractionResponse:
try:
response: Final = await _HANDLER.aextraction(
model=call.model,
custom_llm_provider=call.custom_llm_provider,
logging_obj=call.logging_obj,
provider_config=call.provider_config,
body=call.body,
api_base=call.api_base,
api_key=call.api_key,
headers=call.headers,
timeout=call.timeout,
litellm_params=call.litellm_params,
client=call.client if isinstance(call.client, AsyncHTTPHandler) else None,
)
except Exception as error:
raise _upstream_exception(error, call) from error
return _formatted_response(response, call)
def _send(call: _ExtractionCall) -> ExtractionResponse:
try:
response: Final = _HANDLER.extraction(
model=call.model,
custom_llm_provider=call.custom_llm_provider,
logging_obj=call.logging_obj,
provider_config=call.provider_config,
body=call.body,
api_base=call.api_base,
api_key=call.api_key,
headers=call.headers,
timeout=call.timeout,
litellm_params=call.litellm_params,
client=call.client if isinstance(call.client, HTTPHandler) else None,
)
except Exception as error:
raise _upstream_exception(error, call) from error
return _formatted_response(response, call)
@client
async def aextraction(
model: str,
fields: ExtractionFields,
state: str | None = None,
images: ExtractionImages | None = None,
reference_date: str | None = None,
options: ExtractionOptionsParam | None = None,
api_key: str | None = None,
api_base: str | None = None,
timeout: float | httpx.Timeout | None = None,
custom_llm_provider: str | None = None,
extra_headers: Mapping[str, str] | None = None,
**kwargs: object, # kwargs-ok: litellm_params forwarded by @client
) -> ExtractionResponse:
call: Final = await asyncio.to_thread(
partial(
_prepare_call,
model=model,
request_fields=_request_fields(
state=state, fields=fields, images=images, reference_date=reference_date, options=options
),
api_key=api_key,
api_base=api_base,
timeout=timeout,
custom_llm_provider=custom_llm_provider,
extra_headers=extra_headers,
kwargs=kwargs,
)
)
return await _asend(call)
@client
def extraction(
model: str,
fields: ExtractionFields,
state: str | None = None,
images: ExtractionImages | None = None,
reference_date: str | None = None,
options: ExtractionOptionsParam | None = None,
api_key: str | None = None,
api_base: str | None = None,
timeout: float | httpx.Timeout | None = None,
custom_llm_provider: str | None = None,
extra_headers: Mapping[str, str] | None = None,
**kwargs: object, # kwargs-ok: litellm_params forwarded by @client
) -> ExtractionResponse:
call: Final = _prepare_call(
model=model,
request_fields=_request_fields(
state=state, fields=fields, images=images, reference_date=reference_date, options=options
),
api_key=api_key,
api_base=api_base,
timeout=timeout,
custom_llm_provider=custom_llm_provider,
extra_headers=extra_headers,
kwargs=kwargs,
)
return _send(call)
__all__ = ["aextraction", "extraction"]

View file

@ -37,6 +37,7 @@ from litellm.harness.context import GatewayTarget
from litellm.harness.errors import HarnessError, HarnessInstallFailed
from litellm.harness.types import Harness, Usage
from litellm.llms.custom_httpx.http_handler import get_async_httpx_client
from litellm.types.litellm_params import STREAM_USAGE_INJECTED_FIELD
from litellm.types.llms.custom_http import httpxSpecialProvider
if TYPE_CHECKING:
@ -690,18 +691,21 @@ class ModelEndpoint:
async def _invoke_sdk(
self,
route: str,
kwargs: dict[str, Any], # mutable-ok: injects stream_options into the SDK kwargs
kwargs: dict[str, Any], # mutable-ok: SDK call kwargs built by _sdk_kwargs
) -> object:
if route == ROUTE_MESSAGES:
return await litellm.anthropic.messages.acreate(**kwargs)
if route == ROUTE_CHAT:
if kwargs.get("stream"):
stream_options = kwargs.get("stream_options") or {} # mutable-ok: empty default for a JSON field
kwargs["stream_options"] = { # mutable-ok: JSON field sent to litellm.acompletion
"include_usage": True,
**stream_options,
if not kwargs.get("stream"):
return await litellm.acompletion(**kwargs)
stream_options = kwargs.get("stream_options") or {} # mutable-ok: empty default for a JSON field
return await litellm.acompletion(
**{
**kwargs,
"stream_options": {"include_usage": True, **stream_options},
STREAM_USAGE_INJECTED_FIELD: "include_usage" not in stream_options,
}
return await litellm.acompletion(**kwargs)
)
return await litellm.aresponses(**kwargs)
async def _call_sdk(self, route: str, body: Mapping[str, object]) -> Response:

View file

@ -4,6 +4,7 @@ Helper functions for health check calls.
import base64
from collections.abc import Awaitable, Callable, Mapping
from types import MappingProxyType
from typing import TYPE_CHECKING, Final, Literal
from litellm.llms.base_llm.ocr.transformation import DocumentType
@ -25,6 +26,9 @@ IMAGE_EDIT_HEALTH_CHECK_PROMPT: Final = (
)
ANTHROPIC_MESSAGES_HEALTH_CHECK_MAX_TOKENS: Final = 16
EXTRACTION_HEALTH_CHECK_FIELDS: Final[Mapping[str, Mapping[str, str]]] = MappingProxyType(
{"reachable": MappingProxyType({"type": "text", "instructions": "Any word from the text"})}
)
def native_health_check_mode(model: str, custom_llm_provider: str | None) -> Literal["anthropic_messages"] | None:
@ -215,6 +219,7 @@ class HealthCheckHelpers:
"anthropic_messages",
"ocr",
"decisions",
"extraction",
],
Callable,
]:
@ -237,7 +242,11 @@ class HealthCheckHelpers:
from litellm.litellm_core_utils.audio_utils.utils import (
get_audio_file_for_health_check,
)
from litellm.litellm_core_utils.health_check_utils import DECISIONS_CALL_PARAMS, _filter_model_params
from litellm.litellm_core_utils.health_check_utils import (
DECISIONS_CALL_PARAMS,
EXTRACTION_CALL_PARAMS,
_filter_model_params,
)
from litellm.realtime_api.main import realtime_health_check
return {
@ -322,4 +331,13 @@ class HealthCheckHelpers:
}
)
),
"extraction": lambda: litellm.aextraction(
**EXTRACTION_CALL_PARAMS.validate_python(
{
"state": prompt or "health check",
"fields": EXTRACTION_HEALTH_CHECK_FIELDS,
**_filter_model_params(model_params=model_params),
}
)
),
}

View file

@ -8,8 +8,10 @@ from typing import Final
from pydantic import TypeAdapter
from litellm.types.decisions import DecisionsCallParams
from litellm.types.extraction import ExtractionCallParams
DECISIONS_CALL_PARAMS: Final[TypeAdapter[DecisionsCallParams]] = TypeAdapter(DecisionsCallParams)
EXTRACTION_CALL_PARAMS: Final[TypeAdapter[ExtractionCallParams]] = TypeAdapter(ExtractionCallParams)
OPTIONAL_STR: Final[TypeAdapter[str | None]] = TypeAdapter(str | None)

View file

@ -12,7 +12,7 @@ import subprocess
import sys
import time
import traceback
from collections.abc import Awaitable, Callable, Iterable, Iterator, Mapping, Sequence
from collections.abc import Awaitable, Callable, Coroutine, Iterable, Iterator, Mapping, Sequence
from dataclasses import dataclass
from datetime import datetime as dt_object
from functools import lru_cache
@ -132,11 +132,13 @@ from litellm.responses.utils import ResponseAPILoggingUtils
from litellm.types.agents import LiteLLMSendMessageResponse
from litellm.types.containers.main import ContainerObject
from litellm.types.decisions import DecisionsResponse, OpenAIDecisionResponse
from litellm.types.extraction import ExtractionResponse
from litellm.types.integrations.s3_v2 import S3PartitionGranularity
from litellm.types.interactions import (
InteractionsAPIResponse,
InteractionsAPIStreamingResponse,
)
from litellm.types.litellm_params import STREAM_USAGE_INJECTED_FIELD
from litellm.types.llms.openai import (
AllMessageValues,
Batch,
@ -573,6 +575,27 @@ class UncachedPrice:
cost_breakdown: CostBreakdown | None
@dataclass(frozen=True, slots=True)
class BilledRoundsSpend:
usage: Usage
cost: float
cost_breakdown: CostBreakdown | None
@dataclass(frozen=True, slots=True)
class BilledRoundsDispatch:
"""A parked success log for a stream bridged through chat completions that billed several rounds, and the spend
its failure log bills when a post-call hook rejects the response, as it does for a rejected one-round bridged
stream."""
dispatch: Coroutine[object, object, None]
spend: BilledRoundsSpend
def _fixed_margin(cost_breakdown: CostBreakdown | None) -> float:
return 0.0 if cost_breakdown is None else cost_breakdown.get("margin_fixed_amount", 0.0)
def _timestamp_seconds(moment: object) -> float | None:
if isinstance(moment, datetime.datetime):
return moment.timestamp()
@ -594,6 +617,7 @@ class Logging(LiteLLMLoggingBaseClass):
prometheusLogger, \
slack_app
custom_pricing: bool = False
replaced_global_callback_keys: frozenset[str] = frozenset()
stream_options = None
litellm_request_debug: bool = False
streamed_anthropic_message_id: str | None = None
@ -648,11 +672,11 @@ class Logging(LiteLLMLoggingBaseClass):
litellm_call_id: str,
function_id: str,
litellm_trace_id: str | None = None,
dynamic_input_callbacks: list[str | Callable | CustomLogger] | None = None,
dynamic_input_callbacks: list[str | Callable[..., object] | CustomLogger] | None = None,
dynamic_success_callbacks: list[str | Callable[..., object] | CustomLogger] | None = None,
dynamic_async_success_callbacks: list[str | Callable[..., object] | CustomLogger] | None = None,
dynamic_failure_callbacks: list[str | Callable | CustomLogger] | None = None,
dynamic_async_failure_callbacks: list[str | Callable | CustomLogger] | None = None,
dynamic_failure_callbacks: list[str | Callable[..., object] | CustomLogger] | None = None,
dynamic_async_failure_callbacks: list[str | Callable[..., object] | CustomLogger] | None = None,
applied_guardrails: list[str] | None = None,
kwargs: dict | None = None,
log_raw_request_response: bool = False,
@ -697,6 +721,7 @@ class Logging(LiteLLMLoggingBaseClass):
self._pre_call_session_id: str = session_id_var.get()
_sid: Final = kwargs.get("litellm_session_id") if kwargs else None
self.litellm_session_id: str = str(_sid) if _sid else ""
self.stream_usage_injected: Final = kwargs is not None and kwargs.get(STREAM_USAGE_INJECTED_FIELD) is True
# supports_correlation_logging is False for calls originating from the
# sync client entry point (wrapper() in utils.py): a plain OS thread
# has no per-call context isolation the way an asyncio Task does, and
@ -725,15 +750,17 @@ class Logging(LiteLLMLoggingBaseClass):
self.raw_request_only = raw_request_only
# Initialize dynamic callbacks
self.dynamic_input_callbacks: list[str | Callable | CustomLogger] | None = dynamic_input_callbacks
self.dynamic_input_callbacks: list[str | Callable[..., object] | CustomLogger] | None = dynamic_input_callbacks
self.dynamic_success_callbacks: list[str | Callable[..., object] | CustomLogger] | None = (
dynamic_success_callbacks
)
self.dynamic_async_success_callbacks: list[str | Callable[..., object] | CustomLogger] | None = (
dynamic_async_success_callbacks
)
self.dynamic_failure_callbacks: list[str | Callable | CustomLogger] | None = dynamic_failure_callbacks
self.dynamic_async_failure_callbacks: list[str | Callable | CustomLogger] | None = (
self.dynamic_failure_callbacks: list[str | Callable[..., object] | CustomLogger] | None = (
dynamic_failure_callbacks
)
self.dynamic_async_failure_callbacks: list[str | Callable[..., object] | CustomLogger] | None = (
dynamic_async_failure_callbacks
)
@ -850,10 +877,32 @@ class Logging(LiteLLMLoggingBaseClass):
self.dynamic_async_failure_callbacks, callback
)
def replace_global_callback(self, callback: CustomLogger) -> None:
self.add_dynamic_callback(callback)
self.replaced_global_callback_keys = self.replaced_global_callback_keys | {
litellm.logging_callback_manager.get_custom_logger_key(callback)
}
def add_dynamic_function_callbacks(
self,
*,
async_success: Callable[..., object],
failure: Callable[..., object],
async_failure: Callable[..., object],
) -> None:
self.dynamic_async_success_callbacks = self._with_dynamic_callback(
self.dynamic_async_success_callbacks, async_success
)
self.dynamic_failure_callbacks = self._with_dynamic_callback(self.dynamic_failure_callbacks, failure)
self.dynamic_async_failure_callbacks = self._with_dynamic_callback(
self.dynamic_async_failure_callbacks, async_failure
)
@staticmethod
def _with_dynamic_callback(
callbacks: Sequence[str | Callable | CustomLogger] | None, callback: CustomLogger
) -> list[str | Callable | CustomLogger]:
callbacks: Sequence[str | Callable[..., object] | CustomLogger] | None,
callback: Callable[..., object] | CustomLogger,
) -> list[str | Callable[..., object] | CustomLogger]:
existing: Final = tuple(callbacks or ())
return [*existing, *(() if callback in existing else (callback,))]
@ -890,9 +939,9 @@ class Logging(LiteLLMLoggingBaseClass):
def _process_dynamic_callback_list(
self,
callback_list: list[str | Callable | CustomLogger] | None,
callback_list: list[str | Callable[..., object] | CustomLogger] | None,
dynamic_callbacks_type: Literal["input", "success", "failure", "async_success", "async_failure"],
) -> list[str | Callable | CustomLogger] | None:
) -> list[str | Callable[..., object] | CustomLogger] | None:
"""
Helper function to initialize CustomLogger compatible callbacks in self.dynamic_* callbacks
@ -904,7 +953,7 @@ class Logging(LiteLLMLoggingBaseClass):
if callback_list is None:
return None
processed_list: Final[list[str | Callable | CustomLogger]] = []
processed_list: Final[list[str | Callable[..., object] | CustomLogger]] = []
for callback in callback_list:
if isinstance(callback, str) and callback in litellm._known_custom_logger_compatible_callbacks:
for callback_instance in self._resolve_dynamic_callback_string(callback):
@ -1565,7 +1614,10 @@ class Logging(LiteLLMLoggingBaseClass):
"api_call_start_time"
]
# Input Integration Logging -> If you want to log the fact that an attempt to call the model was made
callbacks: Final = litellm.input_callback + (self.dynamic_input_callbacks or [])
callbacks: Final = (
*self._without_replaced_global_callbacks(litellm.input_callback),
*(self.dynamic_input_callbacks or ()),
)
for callback in callbacks:
try:
if callback == "supabase" and supabaseClient is not None:
@ -1738,7 +1790,10 @@ class Logging(LiteLLMLoggingBaseClass):
)
# Input Integration Logging -> If you want to log the fact that an attempt to call the model was made
callbacks: Final = litellm.input_callback + (self.dynamic_input_callbacks or [])
callbacks: Final = (
*self._without_replaced_global_callbacks(litellm.input_callback),
*(self.dynamic_input_callbacks or ()),
)
for callback in callbacks:
try:
if callback == "sentry" and add_breadcrumb:
@ -2209,20 +2264,23 @@ class Logging(LiteLLMLoggingBaseClass):
usage=usage,
)
def custom_pricing_for(self, result: object) -> bool:
litellm_params: Final = getattr(self, "litellm_params", None)
def spilled_over_ptu_for(self, result: object) -> bool:
result_hidden_params: Final = getattr(result, HIDDEN_PARAMS_ATTR, None) or MappingProxyType({})
additional_headers: Final = (
result_hidden_params.get("additional_headers")
if isinstance(result_hidden_params, dict)
else getattr(result_hidden_params, "additional_headers", None)
)
spilled_over: Final = is_spilled_over_ptu_request(
model_info=_deployment_model_info(litellm_params),
return is_spilled_over_ptu_request(
model_info=_deployment_model_info(getattr(self, "litellm_params", None)),
response_headers=self.model_call_details.get("response_headers"),
additional_headers=additional_headers,
)
return False if spilled_over else use_custom_pricing_for_model(litellm_params=litellm_params)
def custom_pricing_for(self, result: object) -> bool:
if self.spilled_over_ptu_for(result):
return False
return use_custom_pricing_for_model(litellm_params=getattr(self, "litellm_params", None))
def _prompt_for_cost_calculation(self) -> str:
"""
@ -2384,6 +2442,31 @@ class Logging(LiteLLMLoggingBaseClass):
self.model_call_details["combined_usage_object"] = usage
self.model_call_details["response_cost"] = response_cost
def record_failure_after_earlier_rounds(
self, earlier: BilledRoundsSpend, failed_usage: Usage, failed_cost: float
) -> None:
"""Bill a failed round on the failure log after the rounds of its request before it were billed on the success
log.
The fixed margin is a per-request fee, so it is charged once: the failed round's margin is dropped when the
earlier rounds already paid one."""
from litellm.responses.mcp.round_billing import breakdown_less_repeated_fixed_margin
failed_breakdown: Final = None if self.cost_breakdown is earlier.cost_breakdown else self.cost_breakdown
repeated_fixed_margin: Final = (
0.0
if failed_breakdown is None
else min(_fixed_margin(earlier.cost_breakdown), _fixed_margin(failed_breakdown), earlier.cost, failed_cost)
)
self.model_call_details["combined_usage_object"] = failed_usage
self.model_call_details["response_cost"] = failed_cost - repeated_fixed_margin
self.cost_breakdown = breakdown_less_repeated_fixed_margin(failed_breakdown, repeated_fixed_margin)
def record_billed_rounds_for_failure(self, spend: BilledRoundsSpend) -> None:
"""Bill every round of a fully streamed request on the failure log when a post-call hook rejects it."""
self.record_partial_usage_for_failure(spend.usage, spend.cost)
self.cost_breakdown = spend.cost_breakdown
def attempt_scoped_copy(self) -> "Logging":
"""A copy with its own model_call_details, so logging one attempt of a request through it leaves
the state the request's other attempts share untouched."""
@ -2919,6 +3002,7 @@ class Logging(LiteLLMLoggingBaseClass):
or isinstance(logging_result, SearchResponse) # Search API
or isinstance(logging_result, DecisionsResponse)
or isinstance(logging_result, OpenAIDecisionResponse)
or isinstance(logging_result, ExtractionResponse)
or (
isinstance(logging_result, InteractionsAPIResponse)
and logging_result.usage is not None
@ -4511,14 +4595,26 @@ class Logging(LiteLLMLoggingBaseClass):
_filtered_failure_callbacks = self._remove_internal_litellm_callbacks(_filtered_failure_callbacks)
return len(_filtered_failure_callbacks) > 0
def _without_replaced_global_callbacks(
self, global_callbacks: Iterable[str | Callable[..., object] | CustomLogger]
) -> tuple[str | Callable[..., object] | CustomLogger, ...]:
if not self.replaced_global_callback_keys:
return tuple(global_callbacks)
return tuple(
litellm.logging_callback_manager.without_custom_logger_keys(
global_callbacks, self.replaced_global_callback_keys
)
)
def get_combined_callback_list(
self,
dynamic_success_callbacks: list[str | Callable[..., object] | CustomLogger] | None,
global_callbacks: list[str | Callable[..., object] | CustomLogger],
) -> list[str | Callable[..., object] | CustomLogger]:
kept_global_callbacks: Final = self._without_replaced_global_callbacks(global_callbacks)
if dynamic_success_callbacks is None:
return list(global_callbacks)
return list(dict.fromkeys(dynamic_success_callbacks + global_callbacks))
return list(kept_global_callbacks)
return list(dict.fromkeys((*dynamic_success_callbacks, *kept_global_callbacks)))
def _remove_internal_litellm_callbacks(self, callbacks: list) -> list:
"""
@ -7245,13 +7341,16 @@ def get_standard_logging_object_payload(
saved_cache_cost: Final = (uncached_price.cost if uncached_price is not None else None) or 0.0
## Get model cost information ##
base_model = get_base_model_from_metadata(model_call_details=kwargs)
# The router overrides completion_response.model to the model-group alias before
# this payload is built, so cost-map lookup via that alias always misses.
# Fall back to the actual deployment model set by the router in metadata.
if base_model is None:
base_model = metadata.get("deployment")
custom_pricing: Final = logging_obj.custom_pricing_for(init_response_obj)
configured_base_model: Final = get_base_model_from_metadata(model_call_details=kwargs)
spilled_over: Final = logging_obj.spilled_over_ptu_for(init_response_obj)
# The router overrides completion_response.model to the model-group alias before this payload is
# built, so cost-map lookup via that alias misses and the deployment model stands in for base_model.
# A PTU request Azure spilled onto pay-as-you-go capacity is billed at the served model's rates,
# which the deployment's zeroed PTU entry would misreport, so that fallback is skipped there.
base_model: Final = (
metadata.get("deployment") if configured_base_model is None and not spilled_over else configured_base_model
)
custom_pricing: Final = False if spilled_over else logging_obj.custom_pricing_for(init_response_obj)
raw_response_cost: Final = kwargs.get("response_cost")
llm_response_cost: Final[float] = raw_response_cost or 0.0
guardrail_cost: Final = guardrail_information_cost(metadata.get("standard_logging_guardrail_information"))

View file

@ -4,6 +4,7 @@ from functools import reduce
from typing import Final
import httpx
from pydantic import TypeAdapter, ValidationError
from litellm.constants import LITELLM_DETAILED_TIMING
from litellm.litellm_core_utils.core_helpers import get_litellm_metadata_from_kwargs, process_response_headers
@ -101,6 +102,32 @@ def response_timing_metrics(
return {"_response_ms": total_response_time_ms, "litellm_overhead_time_ms": overhead_ms}
_HIDDEN_PARAMS_DICT: Final = TypeAdapter(dict[str, object])
def _hidden_params_mapping(hidden_params: object) -> Mapping[str, object]:
try:
return _HIDDEN_PARAMS_DICT.validate_python(hidden_params)
except ValidationError:
return {}
def _existing_provider_response_model(result: object) -> str | None:
"""The pricing model a response already carries, like the decisions path sets, wins over the response model."""
hidden_params: Final[object] = getattr(result, "_hidden_params", None)
carried: Final[object] = _hidden_params_mapping(hidden_params).get("provider_response_model") or getattr(
hidden_params, "provider_response_model", None
)
return carried if isinstance(carried, str) and carried else None
def _provider_response_model(result: object) -> str | None:
"""The model the provider answered with, kept because the proxy later rewrites ``result.model`` to the alias the
client asked for and cost-map lookups for the standard logging payload would otherwise miss."""
model: Final = getattr(result, "model", None) if isinstance(result, ModelResponse) else None
return model if isinstance(model, str) and model else None
class ResponseMetadata:
"""
Handles setting and managing `_hidden_params`, `response_time_ms`, and `litellm_overhead_time_ms` for LiteLLM responses
@ -147,6 +174,8 @@ class ResponseMetadata:
preserve_litellm_internal_headers=True,
),
"litellm_model_name": model,
"provider_response_model": _existing_provider_response_model(self.result)
or _provider_response_model(self.result),
}
self._update_hidden_params(new_params)

View file

@ -1,4 +1,6 @@
from collections.abc import Callable
import threading
from collections.abc import Callable, Iterable, Iterator
from types import MethodType
from typing import TYPE_CHECKING, Final
import litellm
@ -15,6 +17,7 @@ else:
_custom_logger_compatible_callbacks_literal = str
_generic_api_logger_cache: Final[dict[str, GenericAPILogger]] = {}
_callback_lists_lock: Final = threading.RLock()
class LoggingCallbackManager:
@ -43,16 +46,18 @@ class LoggingCallbackManager:
Add a input callback to litellm.input_callback.
Auto-routes async callbacks to litellm._async_input_callback.
"""
if not isinstance(callback, str) and self._is_async_callable(callback):
self._safe_add_callback_to_list(callback=callback, parent_list=litellm._async_input_callback)
else:
self._safe_add_callback_to_list(callback=callback, parent_list=litellm.input_callback)
with _callback_lists_lock:
if not isinstance(callback, str) and self._is_async_callable(callback):
self._safe_add_callback_to_list(callback=callback, parent_list=litellm._async_input_callback)
else:
self._safe_add_callback_to_list(callback=callback, parent_list=litellm.input_callback)
def add_litellm_service_callback(self, callback: CustomLogger | str | Callable[..., object]):
"""
Add a service callback to litellm.service_callback
"""
self._safe_add_callback_to_list(callback=callback, parent_list=litellm.service_callback)
with _callback_lists_lock:
self._safe_add_callback_to_list(callback=callback, parent_list=litellm.service_callback)
def add_litellm_callback(self, callback: CustomLogger | str | Callable[..., object]):
"""
@ -60,10 +65,11 @@ class LoggingCallbackManager:
Ensures no duplicates are added.
"""
self._safe_add_callback_to_list(
callback=callback,
parent_list=litellm.callbacks,
)
with _callback_lists_lock:
self._safe_add_callback_to_list(
callback=callback,
parent_list=litellm.callbacks,
)
def add_litellm_success_callback(self, callback: CustomLogger | str | Callable[..., object]):
"""
@ -71,37 +77,41 @@ class LoggingCallbackManager:
Auto-routes async callbacks to litellm._async_success_callback.
Special-cases 'dynamodb' and 'openmeter' as async callbacks.
"""
if (
isinstance(callback, str)
and callback in ("dynamodb", "openmeter")
or not isinstance(callback, str)
and self._is_async_callable(callback)
):
self._safe_add_callback_to_list(callback=callback, parent_list=litellm._async_success_callback)
else:
self._safe_add_callback_to_list(callback=callback, parent_list=litellm.success_callback)
with _callback_lists_lock:
if (
isinstance(callback, str)
and callback in ("dynamodb", "openmeter")
or not isinstance(callback, str)
and self._is_async_callable(callback)
):
self._safe_add_callback_to_list(callback=callback, parent_list=litellm._async_success_callback)
else:
self._safe_add_callback_to_list(callback=callback, parent_list=litellm.success_callback)
def add_litellm_failure_callback(self, callback: CustomLogger | str | Callable[..., object]):
"""
Add a failure callback to `litellm.failure_callback`.
Auto-routes async callbacks to litellm._async_failure_callback.
"""
if not isinstance(callback, str) and self._is_async_callable(callback):
self._safe_add_callback_to_list(callback=callback, parent_list=litellm._async_failure_callback)
else:
self._safe_add_callback_to_list(callback=callback, parent_list=litellm.failure_callback)
with _callback_lists_lock:
if not isinstance(callback, str) and self._is_async_callable(callback):
self._safe_add_callback_to_list(callback=callback, parent_list=litellm._async_failure_callback)
else:
self._safe_add_callback_to_list(callback=callback, parent_list=litellm.failure_callback)
def add_litellm_async_success_callback(self, callback: CustomLogger | Callable[..., object] | str):
"""
Add a success callback to litellm._async_success_callback
"""
self._safe_add_callback_to_list(callback=callback, parent_list=litellm._async_success_callback)
with _callback_lists_lock:
self._safe_add_callback_to_list(callback=callback, parent_list=litellm._async_success_callback)
def add_litellm_async_failure_callback(self, callback: CustomLogger | Callable[..., object] | str):
"""
Add a failure callback to litellm._async_failure_callback
"""
self._safe_add_callback_to_list(callback=callback, parent_list=litellm._async_failure_callback)
with _callback_lists_lock:
self._safe_add_callback_to_list(callback=callback, parent_list=litellm._async_failure_callback)
def remove_callback_from_list_by_object(self, callback_list, obj, require_self=True):
"""
@ -110,13 +120,14 @@ class LoggingCallbackManager:
if not isinstance(callback_list, list): # Not list -> do nothing
return
if require_self:
remove_list = [c for c in callback_list if hasattr(c, "__self__") and c.__self__ == obj]
else:
remove_list = [c for c in callback_list if c == obj]
with _callback_lists_lock:
if require_self:
remove_list = [c for c in callback_list if hasattr(c, "__self__") and c.__self__ == obj]
else:
remove_list = [c for c in callback_list if c == obj]
for c in remove_list:
callback_list.remove(c)
for c in remove_list:
callback_list.remove(c)
def remove_callbacks_by_type(self, callback_list, callback_type):
"""
@ -134,10 +145,11 @@ class LoggingCallbackManager:
if not isinstance(callback_list, list):
return
remove_list: Final = [c for c in callback_list if isinstance(c, callback_type)]
with _callback_lists_lock:
remove_list: Final = [c for c in callback_list if isinstance(c, callback_type)]
for c in remove_list:
callback_list.remove(c)
for c in remove_list:
callback_list.remove(c)
def _add_string_callback_to_list(
self, callback: str, parent_list: list[CustomLogger | Callable[..., object] | str]
@ -336,6 +348,55 @@ class LoggingCallbackManager:
return "-".join(key_parts)
get_custom_logger_key = _get_custom_logger_key
callback_lists_lock: Final = _callback_lists_lock
def drop_callbacks(self, callbacks: Iterable[object], method_owners: Iterable[object] = ()) -> frozenset[str]:
dropped_ids: Final = frozenset(map(id, callbacks))
owner_ids: Final = frozenset(map(id, method_owners))
def keep(callback: object) -> bool:
if id(callback) in dropped_ids:
return False
return not (isinstance(callback, MethodType) and id(callback.__self__) in owner_ids)
with _callback_lists_lock:
vacated_keys: Final = frozenset(
self._get_custom_logger_key(callback)
for callback in litellm.callbacks
if isinstance(callback, CustomLogger) and id(callback) in dropped_ids
)
litellm.callbacks = [callback for callback in litellm.callbacks if keep(callback)]
litellm.input_callback = [callback for callback in litellm.input_callback if keep(callback)]
litellm.success_callback = [callback for callback in litellm.success_callback if keep(callback)]
litellm.failure_callback = [callback for callback in litellm.failure_callback if keep(callback)]
litellm.service_callback = [callback for callback in litellm.service_callback if keep(callback)]
litellm._async_success_callback = [
callback for callback in litellm._async_success_callback if keep(callback)
]
litellm._async_failure_callback = [
callback for callback in litellm._async_failure_callback if keep(callback)
]
return vacated_keys
def with_custom_logger_keys(
self, candidates: Iterable[CustomLogger], keys: frozenset[str]
) -> tuple[CustomLogger, ...]:
if not keys:
return ()
class_names: Final = frozenset(key.split("-", 1)[0] for key in keys)
same_class: Final = (candidate for candidate in candidates if type(candidate).__name__ in class_names)
return tuple(candidate for candidate in same_class if self._get_custom_logger_key(candidate) in keys)
def without_custom_logger_keys(
self, callbacks: Iterable[str | Callable[..., object] | CustomLogger], keys: frozenset[str]
) -> Iterator[str | Callable[..., object] | CustomLogger]:
return (
callback
for callback in callbacks
if not (isinstance(callback, CustomLogger) and self._get_custom_logger_key(callback) in keys)
)
def _reset_all_callbacks(self):
"""
Reset all callbacks to an empty list
@ -363,19 +424,21 @@ class LoggingCallbackManager:
_get_all_callbacks = get_all_callbacks
def remove_callback_from_all_lists(self, obj, require_self=False) -> None:
def remove_callback_from_all_lists(self, obj: object, require_self: bool = False) -> None:
"""
Remove a callback object from every callback list it may have been
promoted into, so a re-initialized callback leaves no stale instance behind.
"""
for callback_list in (
litellm.callbacks,
litellm.success_callback,
litellm.failure_callback,
litellm._async_success_callback,
litellm._async_failure_callback,
):
self.remove_callback_from_list_by_object(callback_list, obj, require_self=require_self)
with _callback_lists_lock:
for callback_list in (
litellm.callbacks,
litellm.input_callback,
litellm.success_callback,
litellm.failure_callback,
litellm._async_success_callback,
litellm._async_failure_callback,
):
self.remove_callback_from_list_by_object(callback_list, obj, require_self=require_self)
def get_active_additional_logging_utils_from_custom_logger(
self,

View file

@ -10,7 +10,7 @@ import traceback
from collections.abc import AsyncIterator, Callable, Iterable, Iterator, Mapping, Sequence
from dataclasses import dataclass
from types import MappingProxyType
from typing import Any, Final, NoReturn, Protocol, TypeVar, cast
from typing import TYPE_CHECKING, Any, Final, NoReturn, Protocol, TypeVar, cast
import anyio
import httpx
@ -60,6 +60,9 @@ from .exception_mapping_utils import exception_type, with_deployment_api_base
from .llm_response_utils.get_api_base import get_api_base
from .rules import Rules
if TYPE_CHECKING:
from .litellm_logging import BilledRoundsSpend
# Constants for special delta attribute names
AUDIO_ATTRIBUTE: Final = "audio"
IMAGE_ATTRIBUTE: Final = "images"
@ -259,6 +262,8 @@ def _provider_hidden_params(
class CustomStreamWrapper:
_earlier_rounds_spend: "BilledRoundsSpend | None" = None
@property
def hidden_params(self) -> dict[str, object]: # mutable-ok: API requires mutation
return self._hidden_params
@ -342,6 +347,10 @@ class CustomStreamWrapper:
self.messages = getattr(logging_obj, "messages", None)
self.sent_stream_usage = False
self.send_stream_usage = True if self.check_send_stream_usage(self.stream_options) else False
self.caller_requested_stream_usage: Final = (
self.check_send_stream_usage(getattr(logging_obj, "stream_options", None))
and getattr(logging_obj, "stream_usage_injected", False) is not True
)
self.tool_call = False
# keep track of the returned chunks - used for calculating the input/output tokens for stream options
self.chunks: list[ModelResponseStream] = []
@ -2070,7 +2079,7 @@ class CustomStreamWrapper:
except StopIteration:
if self.sent_last_chunk is True:
complete_streaming_response: Final = self._build_complete_streaming_response()
response = self.model_response_creator()
response = self._new_usage_chunk()
if complete_streaming_response is not None:
self._propagate_usage_cost_to_hidden_params(complete_streaming_response, self.custom_llm_provider)
@ -2222,27 +2231,10 @@ class CustomStreamWrapper:
)
if _has_usage:
# Store a copy ONLY when usage stripping below will mutate
# the chunk. For non-usage chunks (vast majority), store
# directly to avoid expensive model_copy() per chunk.
self.chunks.append(processed_chunk.model_copy())
# Strip usage from the outgoing chunk so it's not sent twice
# (once in the chunk, once in _hidden_params).
obj_dict = processed_chunk.model_dump()
if "usage" in obj_dict:
del obj_dict["usage"]
processed_chunk = self.model_response_creator(
chunk=obj_dict,
hidden_params=cast( # cast-ok: preserve mapping operations on dynamic storage
Mapping[str, object], getattr(processed_chunk, HIDDEN_PARAMS_ATTR)
),
)
is_empty = is_model_response_stream_empty(
model_response=cast(ModelResponseStream, processed_chunk)
)
if is_empty:
usage_free_chunk = self._record_and_strip_usage_chunk(processed_chunk)
if usage_free_chunk is None:
continue
processed_chunk = usage_free_chunk
else:
# No usage data — safe to store directly without copying
self.chunks.append(processed_chunk)
@ -2273,13 +2265,18 @@ class CustomStreamWrapper:
if processed_chunk is None:
continue
choice = processed_chunk.choices[0]
if isinstance(choice, StreamingChoices):
self._accumulate_streamed_delta(choice.delta)
if processed_chunk.choices:
choice = processed_chunk.choices[0]
if isinstance(choice, StreamingChoices):
self._accumulate_streamed_delta(choice.delta)
self.rules.post_call_rules(input=self.response_uptil_now, model=self.model)
# RETURN RESULT
self.chunks.append(processed_chunk)
return processed_chunk
if not self.send_stream_usage or getattr(processed_chunk, "usage", None) is None:
self.chunks.append(processed_chunk)
return processed_chunk
usage_free_chunk = self._record_and_strip_usage_chunk(processed_chunk)
if usage_free_chunk is None:
continue
return usage_free_chunk
except (StopAsyncIteration, StopIteration):
return await self._finalize_completed_stream(cache_hit=cache_hit)
except httpx.TimeoutException as e: # if httpx read timeout error occues
@ -2306,7 +2303,7 @@ class CustomStreamWrapper:
async def _finalize_completed_stream(self, cache_hit: bool) -> "ModelResponseStream":
if self.sent_last_chunk is True:
complete_streaming_response: Final = await self._async_build_complete_streaming_response()
response: Final = self.model_response_creator()
response: Final = self._new_usage_chunk()
if complete_streaming_response is not None:
self._propagate_usage_cost_to_hidden_params(complete_streaming_response, self.custom_llm_provider)
@ -2407,6 +2404,9 @@ class CustomStreamWrapper:
return
await asyncio.wait((self._failure_logging,), timeout=timeout)
def bill_failure_after_earlier_rounds(self, earlier: "BilledRoundsSpend") -> None:
self._earlier_rounds_spend = earlier
def _record_partial_usage_for_failure(self) -> None:
"""
A stream that breaks mid-flight still billed the provider for the chunks
@ -2433,10 +2433,12 @@ class CustomStreamWrapper:
if self.model:
partial_response.model = self.model
backfill_missing_cache_usage_fields(usage)
self.logging_obj.model_call_details["combined_usage_object"] = usage
self.logging_obj.model_call_details["response_cost"] = (
self.logging_obj.response_cost_calculator(result=partial_response) or 0.0
)
partial_cost: Final = self.logging_obj.response_cost_calculator(result=partial_response) or 0.0
if self._earlier_rounds_spend is None:
self.logging_obj.model_call_details["combined_usage_object"] = usage
self.logging_obj.model_call_details["response_cost"] = partial_cost
return
self.logging_obj.record_failure_after_earlier_rounds(self._earlier_rounds_spend, usage, partial_cost)
except Exception as recover_error:
verbose_logger.debug(
"could not recover partial usage for interrupted stream: %s",
@ -2463,6 +2465,22 @@ class CustomStreamWrapper:
return False
return any(get(field) for field in cls._CONTINUATION_DISQUALIFYING_DELTA_FIELDS)
def _new_usage_chunk(self) -> ModelResponseStream:
usage_chunk: Final = self.model_response_creator()
if self.caller_requested_stream_usage:
usage_chunk.choices = []
return usage_chunk
def _record_and_strip_usage_chunk(self, chunk: ModelResponseStream) -> ModelResponseStream | None:
self.chunks.append(chunk.model_copy())
usage_free_chunk: Final = self.model_response_creator(
chunk={key: value for key, value in chunk.model_dump().items() if key != "usage"},
hidden_params=cast( # cast-ok: preserve mapping operations on dynamic storage
Mapping[str, object], getattr(chunk, HIDDEN_PARAMS_ATTR)
),
)
return None if is_model_response_stream_empty(model_response=usage_free_chunk) else usage_free_chunk
def _accumulate_streamed_delta(self, delta: object) -> None:
# Shared by the sync, async, and non-aiohttp iteration sites so answer
# text and the disqualifying-content latch stay in step across all three.

View file

@ -0,0 +1,3 @@
from .transformation import BaseExtractionConfig
__all__ = ["BaseExtractionConfig"]

View file

@ -0,0 +1,86 @@
from abc import ABC, abstractmethod
from collections.abc import Mapping
from typing import Final
import httpx
from pydantic import TypeAdapter, ValidationError
from litellm.llms.base_llm.chat.transformation import BaseLLMException
from litellm.secret_managers.main import get_secret_str
from litellm.types.extraction import ExtractionRequestBody, ExtractionResponse
_PAYLOAD_ADAPTER: Final[TypeAdapter[object]] = TypeAdapter(object)
_RESPONSE_ADAPTER: Final[TypeAdapter[ExtractionResponse]] = TypeAdapter(ExtractionResponse)
_RESERVED_HEADERS: Final[frozenset[str]] = frozenset({"authorization", "content-type"})
class BaseExtractionConfig(ABC):
api_key_env: tuple[str, ...] = ()
api_base_env: tuple[str, ...] = ()
def get_default_api_base(self) -> str | None:
return None
def missing_api_base_message(self, custom_llm_provider: str) -> str:
return f"api_base is required for Extraction provider '{custom_llm_provider}'"
def resolve_api_base(self, api_base: str | None) -> str | None:
return api_base or self._first_secret(self.api_base_env) or self.get_default_api_base()
def resolve_api_key(self, api_key: str | None) -> str | None:
return api_key or self._first_secret(self.api_key_env)
@staticmethod
def _first_secret(names: tuple[str, ...]) -> str | None:
return next((value for value in (get_secret_str(name) for name in names) if value), None)
def canonical_model(self, model: str) -> str:
return model
def request_model(self, model: str) -> str:
return model
def validate_environment(
self,
headers: Mapping[str, str],
model: str,
api_key: str | None,
litellm_params: Mapping[str, object],
) -> Mapping[str, str]:
return {
**{name: value for name, value in headers.items() if name.lower() not in _RESERVED_HEADERS},
**({"Authorization": f"Bearer {api_key}"} if api_key is not None else {}),
"Content-Type": "application/json",
}
@abstractmethod
def get_complete_url(self, api_base: str, model: str) -> str: ...
def transform_extraction_request(self, model: str, request: ExtractionRequestBody) -> Mapping[str, object]:
return {**request.model_dump(mode="json", exclude_unset=True), "model": self.request_model(model)}
def unwrap_response(self, payload: object) -> object:
return payload
def transform_extraction_response(
self,
model: str,
custom_llm_provider: str,
raw_response: httpx.Response,
) -> ExtractionResponse:
payload: Final[object] = _PAYLOAD_ADAPTER.validate_json(raw_response.content)
try:
return _RESPONSE_ADAPTER.validate_python(self.unwrap_response(payload))
except ValidationError as error:
raise BaseLLMException(
status_code=500,
message=f"Extraction provider '{custom_llm_provider}' returned an unexpected response: {error}",
) from error
def get_error_class(
self,
error_message: str,
status_code: int,
headers: Mapping[str, str] | httpx.Headers,
) -> BaseLLMException:
return BaseLLMException(status_code=status_code, message=error_message, headers=httpx.Headers(headers))

View file

@ -0,0 +1,53 @@
from collections.abc import Mapping
from typing import Final
from pydantic import TypeAdapter, ValidationError
from litellm.secret_managers.main import (
get_secret_str,
normalize_nonempty_secret_str,
)
CLOUDFLARE_MISSING_API_BASE_MESSAGE: Final = (
"Missing CLOUDFLARE_ACCOUNT_ID - set CLOUDFLARE_ACCOUNT_ID or pass api_base explicitly"
)
_RESPONSE_MAPPING_ADAPTER: Final[TypeAdapter[Mapping[str, object]]] = TypeAdapter(Mapping[str, object])
def workers_ai_default_api_base() -> str | None:
account_id: Final = normalize_nonempty_secret_str(get_secret_str("CLOUDFLARE_ACCOUNT_ID"))
if account_id is None:
return None
return f"https://api.cloudflare.com/client/v4/accounts/{account_id}/ai/run"
def workers_ai_canonical_model(model: str) -> str:
if model.startswith("@cf/"):
return model
return f"@cf/cloudflare/{model}"
def workers_ai_request_model(model: str) -> str:
return model.rsplit("/", maxsplit=1)[-1]
def workers_ai_run_url(api_base: str, model: str) -> str:
normalized_api_base: Final = api_base.rstrip("/")
if normalized_api_base.endswith("/ai/v1"):
return f"{normalized_api_base.removesuffix('/ai/v1')}/ai/run/{model}"
if normalized_api_base.endswith("/ai/run"):
return f"{normalized_api_base}/{model}"
return f"{normalized_api_base}/ai/run/{model}"
def unwrap_workers_ai_result(payload: object) -> object:
try:
response_mapping: Final = _RESPONSE_MAPPING_ADAPTER.validate_python(payload)
except ValidationError:
return payload
if "answers" in response_mapping:
return response_mapping
try:
return _RESPONSE_MAPPING_ADAPTER.validate_python(response_mapping.get("result"))
except ValidationError:
return response_mapping

View file

@ -1,16 +1,13 @@
from collections.abc import Mapping
from typing import Final
from pydantic import TypeAdapter, ValidationError
from litellm.llms.base_llm.decisions.transformation import BaseDecisionsConfig
from litellm.secret_managers.main import (
get_secret_str,
normalize_nonempty_secret_str,
from litellm.llms.cloudflare.common_utils import (
CLOUDFLARE_MISSING_API_BASE_MESSAGE,
unwrap_workers_ai_result,
workers_ai_canonical_model,
workers_ai_default_api_base,
workers_ai_request_model,
workers_ai_run_url,
)
_RESPONSE_MAPPING_ADAPTER: Final[TypeAdapter[Mapping[str, object]]] = TypeAdapter(Mapping[str, object])
class CloudflareDecisionsConfig(BaseDecisionsConfig):
api_key_env = ("CLOUDFLARE_API_KEY",)
@ -23,38 +20,19 @@ class CloudflareDecisionsConfig(BaseDecisionsConfig):
inlines_remote_images = True
def get_default_api_base(self) -> str | None:
account_id: Final = normalize_nonempty_secret_str(get_secret_str("CLOUDFLARE_ACCOUNT_ID"))
if account_id is None:
return None
return f"https://api.cloudflare.com/client/v4/accounts/{account_id}/ai/run"
return workers_ai_default_api_base()
def missing_api_base_message(self, custom_llm_provider: str) -> str:
return "Missing CLOUDFLARE_ACCOUNT_ID - set CLOUDFLARE_ACCOUNT_ID or pass api_base explicitly"
return CLOUDFLARE_MISSING_API_BASE_MESSAGE
def canonical_model(self, model: str) -> str:
if model.startswith("@cf/"):
return model
return f"@cf/cloudflare/{model}"
return workers_ai_canonical_model(model)
def request_model(self, model: str) -> str:
return model.rsplit("/", maxsplit=1)[-1]
return workers_ai_request_model(model)
def get_complete_url(self, api_base: str, model: str) -> str:
normalized_api_base: Final = api_base.rstrip("/")
if normalized_api_base.endswith("/ai/v1"):
return f"{normalized_api_base.removesuffix('/ai/v1')}/ai/run/{model}"
if normalized_api_base.endswith("/ai/run"):
return f"{normalized_api_base}/{model}"
return f"{normalized_api_base}/ai/run/{model}"
return workers_ai_run_url(api_base, model)
def unwrap_response(self, payload: object) -> object:
try:
response_mapping: Final = _RESPONSE_MAPPING_ADAPTER.validate_python(payload)
except ValidationError:
return payload
if "answers" in response_mapping:
return response_mapping
try:
return _RESPONSE_MAPPING_ADAPTER.validate_python(response_mapping.get("result"))
except ValidationError:
return response_mapping
return unwrap_workers_ai_result(payload)

View file

@ -0,0 +1,11 @@
from typing import Final
from litellm.llms.cloudflare.extraction.guardrail_translation.handler import ExtractionHandler
from litellm.types.utils import CallTypes
guardrail_translation_mappings: Final = {
CallTypes.extraction: ExtractionHandler,
CallTypes.aextraction: ExtractionHandler,
}
__all__ = ["ExtractionHandler", "guardrail_translation_mappings"]

View file

@ -0,0 +1,276 @@
from collections.abc import Iterable, Mapping, Sequence
from dataclasses import dataclass
from itertools import chain, groupby
from types import MappingProxyType
from typing import TYPE_CHECKING, Annotated, Final, Literal, TypeAlias, TypeVar
from pydantic import Field, JsonValue, TypeAdapter
from litellm.litellm_core_utils.core_helpers import get_metadata_variable_name_from_kwargs
from litellm.llms.base_llm.guardrail_translation.base_translation import BaseTranslation
from litellm.types.extraction import (
EXTRACTION_STATE_REWRITTEN_METADATA_KEY,
ExtractionAnswer,
ExtractionBase64Image,
ExtractionResponse,
)
from litellm.types.utils import GenericGuardrailAPIInputs
if TYPE_CHECKING:
from litellm.integrations.custom_guardrail import CustomGuardrail
from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
from litellm.types.proxy.auth.user_api_key_auth import UserAPIKeyAuth
_FIELDS_ADAPTER: Final[TypeAdapter[Mapping[str, Mapping[str, object]]]] = TypeAdapter(
Mapping[str, Mapping[str, object]]
)
_METADATA_ADAPTER: Final[TypeAdapter[Mapping[str, object]]] = TypeAdapter(Mapping[str, object])
_IMAGES_ADAPTER: Final[TypeAdapter[tuple[ExtractionBase64Image | JsonValue, ...]]] = TypeAdapter(
tuple[Annotated[ExtractionBase64Image | JsonValue, Field(union_mode="left_to_right")], ...]
)
_AnswerKey: TypeAlias = Literal["value", "guess", "mean", "median", "p10", "p90", "quantiles"]
_ESTIMATE_KEYS: Final[tuple[_AnswerKey, ...]] = ("value", "guess", "mean", "median", "p10", "p90", "quantiles")
_STATISTIC_KEYS: Final[frozenset[_AnswerKey]] = frozenset(("mean", "median", "p10", "p90", "quantiles"))
_NULLED_ANSWER: Final[Mapping[str, None]] = MappingProxyType(
dict.fromkeys(("value", "guess", "mean", "median", "p10", "p90", "quantiles", "steps", "span", "spans"), None)
)
@dataclass(frozen=True, slots=True)
class _AnswerText:
name: str
key: _AnswerKey
item: int | None
text: str
_SPEC_TEXT_KEYS: Final = ("instructions", "unit")
@dataclass(frozen=True, slots=True)
class _SpecText:
name: str
key: str
text: str
def _spec_texts_of(name: str, spec: Mapping[str, object]) -> tuple[_SpecText, ...]:
return tuple(
_SpecText(name=name, key=key, text=text) for key in _SPEC_TEXT_KEYS if isinstance(text := spec.get(key), str)
)
def _spec_texts(fields: Mapping[str, Mapping[str, object]]) -> tuple[_SpecText, ...]:
return tuple(chain.from_iterable(_spec_texts_of(name, spec) for name, spec in fields.items()))
_NamedText = TypeVar("_NamedText", _AnswerText, _SpecText)
def _scans_by_name(scanned: Iterable[tuple[_NamedText, str]]) -> Mapping[str, tuple[tuple[_NamedText, str], ...]]:
return {name: tuple(group) for name, group in groupby(scanned, key=lambda pair: pair[0].name)}
def _guardrailed_spec(spec: Mapping[str, object], scanned: Sequence[tuple[_SpecText, str]]) -> Mapping[str, object]:
return {**spec, **{text.key: guardrailed for text, guardrailed in scanned}}
def _metadata_key(data: Mapping[str, object]) -> Literal["metadata", "litellm_metadata"]:
return get_metadata_variable_name_from_kwargs(data)
def _metadata(data: Mapping[str, object]) -> Mapping[str, object]:
return _METADATA_ADAPTER.validate_python(data.get(_metadata_key(data)) or {})
def _is_text_answer(answer: ExtractionAnswer, spec: Mapping[str, object] | None) -> bool:
return answer.type == "text" or (answer.type == "list" and (spec is None or spec.get("items") == "text"))
def _image_urls(images: object) -> tuple[str, ...]:
if not isinstance(images, list):
return ()
return tuple(
f"data:{image.content_type};base64,{image.base64}" if isinstance(image, ExtractionBase64Image) else image
for image in _IMAGES_ADAPTER.validate_python(images)
if isinstance(image, (str, ExtractionBase64Image))
)
def _scalar_text(value: JsonValue, whole: bool) -> str | None:
if isinstance(value, bool):
return None
if isinstance(value, float) and whole:
return format(value, ".0f")
return str(value) if isinstance(value, (str, int, float)) else None
def _item_texts(items: Sequence[JsonValue], whole: bool) -> tuple[tuple[int, str], ...]:
return tuple((index, text) for index, item in enumerate(items) if (text := _scalar_text(item, whole)) is not None)
def _value_texts(name: str, key: _AnswerKey, value: JsonValue) -> tuple[_AnswerText, ...]:
whole: Final = key in _STATISTIC_KEYS
if isinstance(value, list):
return tuple(
_AnswerText(name=name, key=key, item=index, text=text) for index, text in _item_texts(value, whole)
)
text: Final = _scalar_text(value, whole)
return () if text is None else (_AnswerText(name=name, key=key, item=None, text=text),)
def _step_texts(name: str, step: JsonValue) -> tuple[_AnswerText, ...]:
if not isinstance(step, dict):
return ()
return tuple(chain.from_iterable(_value_texts(name, key, step.get(key)) for key in _ESTIMATE_KEYS))
def _answer_texts(name: str, answer: ExtractionAnswer) -> tuple[_AnswerText, ...]:
return (
*_value_texts(name, "value", answer.value),
*_value_texts(name, "guess", answer.guess),
*_value_texts(name, "mean", answer.mean),
*_value_texts(name, "median", answer.median),
*_value_texts(name, "p10", answer.p10),
*_value_texts(name, "p90", answer.p90),
*_value_texts(name, "quantiles", list(answer.quantiles or ())),
*chain.from_iterable(_step_texts(name, step) for step in answer.steps or ()),
)
def _guardrailed_value(value: JsonValue, scanned: Mapping[int | None, str]) -> JsonValue:
if isinstance(value, list):
return [scanned.get(index, item) for index, item in enumerate(value)]
return scanned.get(None, value)
def _scans_of(key: _AnswerKey, scanned: Sequence[tuple[_AnswerText, str]]) -> Mapping[int | None, str]:
return {text.item: guardrailed for text, guardrailed in scanned if text.key == key}
def _guardrailed_answer(
answer: ExtractionAnswer,
scanned: Sequence[tuple[_AnswerText, str]],
drop_spans: bool,
is_text: bool,
) -> ExtractionAnswer:
if not is_text and any(text.text != guardrailed for text, guardrailed in scanned):
return answer.model_copy(update=_NULLED_ANSWER)
value_scans: Final = _scans_of("value", scanned if is_text else ())
guess_scans: Final = _scans_of("guess", scanned if is_text else ())
update: Final = {
**({"value": _guardrailed_value(answer.value, value_scans)} if value_scans else {}),
**({"guess": _guardrailed_value(answer.guess, guess_scans)} if guess_scans else {}),
**({"span": None, "spans": None} if drop_spans else {}),
}
return answer.model_copy(update=update) if update else answer
async def _scanned_texts(
texts: Sequence[str],
model: object,
guardrail_to_apply: "CustomGuardrail",
request_data: dict[str, object], # mutable-ok: CustomGuardrail.apply_guardrail takes the request as a plain dict
input_type: Literal["request", "response"],
litellm_logging_obj: "LiteLLMLoggingObj | None",
images: Sequence[str] = (),
) -> tuple[str, ...]:
unique_texts: Final = tuple(dict.fromkeys(texts))
inputs: Final = GenericGuardrailAPIInputs(texts=list(unique_texts), model=model if isinstance(model, str) else None)
if images:
inputs["images"] = list(images)
guardrailed: Final = await guardrail_to_apply.apply_guardrail( # pyright: ignore[reportUnknownMemberType] # CustomGuardrail.apply_guardrail types request_data as a bare dict
inputs=inputs,
request_data=request_data,
input_type=input_type,
logging_obj=litellm_logging_obj,
)
returned: Final = guardrailed.get("texts", [])
guardrailed_by_text: Final = MappingProxyType(
{text: returned[index] if index < len(returned) else text for index, text in enumerate(unique_texts)}
)
return tuple(guardrailed_by_text[text] for text in texts)
class ExtractionHandler(BaseTranslation):
async def process_input_messages(
self,
data: dict[str, object], # mutable-ok: BaseTranslation hands over the proxy's request dict
guardrail_to_apply: "CustomGuardrail",
litellm_logging_obj: "LiteLLMLoggingObj | None" = None,
) -> Mapping[str, object]:
state: Final = data.get("state")
states: Final = (state,) if isinstance(state, str) and state else ()
fields: Final = _FIELDS_ADAPTER.validate_python(data.get("fields") or {})
spec_texts: Final = _spec_texts(fields)
images: Final = _image_urls(data.get("images"))
if not states and not spec_texts and not images:
return data
scanned: Final = await _scanned_texts(
(*states, *(text.text for text in spec_texts)),
data.get("model"),
guardrail_to_apply,
data,
"request",
litellm_logging_obj,
images,
)
spec_scans: Final = _scans_by_name(zip(spec_texts, scanned[len(states) :]))
state_rewritten: Final = bool(states) and scanned[0] != states[0]
return {
**data,
**({"state": scanned[0]} if states else {}),
**(
{_metadata_key(data): {**_metadata(data), EXTRACTION_STATE_REWRITTEN_METADATA_KEY: True}}
if state_rewritten
else {}
),
"fields": {name: _guardrailed_spec(spec, spec_scans.get(name, ())) for name, spec in fields.items()},
}
async def process_output_response(
self,
response: object,
guardrail_to_apply: "CustomGuardrail",
litellm_logging_obj: "LiteLLMLoggingObj | None" = None,
user_api_key_dict: "UserAPIKeyAuth | None" = None,
request_data: dict[str, object] | None = None, # mutable-ok: BaseTranslation hands over the request dict
) -> object:
if not isinstance(response, ExtractionResponse):
return response
texts: Final = tuple(
chain.from_iterable(_answer_texts(name, answer) for name, answer in response.answers.items())
)
state_texts: Final = (response.state_text,) if response.state_text else ()
request: Final = request_data or {}
scanned: Final = (
await _scanned_texts(
(*(text.text for text in texts), *state_texts),
response.model,
guardrail_to_apply,
request,
"response",
litellm_logging_obj,
)
if texts or state_texts
else ()
)
answer_scans: Final = _scans_by_name(zip(texts, scanned))
state_text: Final = next(iter(scanned[len(texts) :]), response.state_text)
specs: Final = _FIELDS_ADAPTER.validate_python(request.get("fields") or {})
return response.model_copy(
update={
"answers": {
name: _guardrailed_answer(
answer,
answer_scans.get(name, ()),
drop_spans=state_text != response.state_text,
is_text=_is_text_answer(answer, specs.get(name)),
)
for name, answer in response.answers.items()
},
**({"state_text": state_text} if state_texts else {}),
}
)

View file

@ -0,0 +1,32 @@
from litellm.llms.base_llm.extraction.transformation import BaseExtractionConfig
from litellm.llms.cloudflare.common_utils import (
CLOUDFLARE_MISSING_API_BASE_MESSAGE,
unwrap_workers_ai_result,
workers_ai_canonical_model,
workers_ai_default_api_base,
workers_ai_request_model,
workers_ai_run_url,
)
class CloudflareExtractionConfig(BaseExtractionConfig):
api_key_env = ("CLOUDFLARE_API_KEY",)
api_base_env = ("CLOUDFLARE_API_BASE",)
def get_default_api_base(self) -> str | None:
return workers_ai_default_api_base()
def missing_api_base_message(self, custom_llm_provider: str) -> str:
return CLOUDFLARE_MISSING_API_BASE_MESSAGE
def canonical_model(self, model: str) -> str:
return workers_ai_canonical_model(model)
def request_model(self, model: str) -> str:
return workers_ai_request_model(model)
def get_complete_url(self, api_base: str, model: str) -> str:
return workers_ai_run_url(api_base, model)
def unwrap_response(self, payload: object) -> object:
return unwrap_workers_ai_result(payload)

View file

@ -76,6 +76,7 @@ from litellm.llms.base_llm.containers.transformation import BaseContainerConfig
from litellm.llms.base_llm.decisions.transformation import BaseDecisionsConfig
from litellm.llms.base_llm.embedding.transformation import BaseEmbeddingConfig
from litellm.llms.base_llm.evals.transformation import BaseEvalsAPIConfig
from litellm.llms.base_llm.extraction.transformation import BaseExtractionConfig
from litellm.llms.base_llm.files.transformation import (
BaseFilesConfig,
BaseFileUploadStream,
@ -128,6 +129,7 @@ from litellm.types.containers.main import (
DeleteContainerResult,
)
from litellm.types.decisions import DecisionsIRRequest, DecisionsIRResponse
from litellm.types.extraction import ExtractionResponse
from litellm.types.files import StreamingMediaUploadConfig, TwoStepFileUploadConfig
from litellm.types.integrations.custom_logger import (
NON_CODE_INTERPRETER_INTERCEPTION_INTERNAL_PREFIXES,
@ -1700,6 +1702,115 @@ class BaseLLMHTTPHandler:
model=model, custom_llm_provider=custom_llm_provider, raw_response=response, request=request
)
def _prepare_extraction_request(
self,
model: str,
logging_obj: LiteLLMLoggingObj | None,
provider_config: BaseExtractionConfig,
body: Mapping[str, object],
api_base: str,
api_key: str | None,
headers: Mapping[str, str],
litellm_params: Mapping[str, object],
) -> tuple[str, Mapping[str, str]]:
outbound_headers: Final = provider_config.validate_environment(
headers=headers,
model=model,
api_key=api_key,
litellm_params=litellm_params,
)
url: Final = provider_config.get_complete_url(api_base=api_base, model=model)
if logging_obj is not None:
logging_obj.pre_call( # pyright: ignore[reportUnknownMemberType] # Logging.pre_call is untyped
input=dict(body),
api_key=api_key,
model=model,
additional_args={"api_base": url, "complete_input_dict": dict(body), "headers": outbound_headers},
)
return url, outbound_headers
def extraction(
self,
model: str,
custom_llm_provider: str,
logging_obj: LiteLLMLoggingObj | None,
provider_config: BaseExtractionConfig,
body: Mapping[str, object],
api_base: str,
api_key: str | None,
headers: Mapping[str, str],
timeout: float | httpx.Timeout | None,
litellm_params: Mapping[str, object],
client: HTTPHandler | None = None,
) -> ExtractionResponse:
url, outbound_headers = self._prepare_extraction_request(
model=model,
logging_obj=logging_obj,
provider_config=provider_config,
body=body,
api_base=api_base,
api_key=api_key,
headers=headers,
litellm_params=litellm_params,
)
sync_httpx_client: Final = client if client is not None else get_httpx_client()
try:
response: Final = sync_httpx_client.post( # pyright: ignore[reportUnknownMemberType] # untyped
url,
json=dict(body),
headers=dict(outbound_headers),
timeout=timeout,
logging_obj=logging_obj,
)
except httpx.HTTPError as e:
raise self._handle_error(e=e, provider_config=provider_config)
return provider_config.transform_extraction_response(
model=model, custom_llm_provider=custom_llm_provider, raw_response=response
)
async def aextraction(
self,
model: str,
custom_llm_provider: str,
logging_obj: LiteLLMLoggingObj | None,
provider_config: BaseExtractionConfig,
body: Mapping[str, object],
api_base: str,
api_key: str | None,
headers: Mapping[str, str],
timeout: float | httpx.Timeout | None,
litellm_params: Mapping[str, object],
client: AsyncHTTPHandler | None = None,
) -> ExtractionResponse:
url, outbound_headers = self._prepare_extraction_request(
model=model,
logging_obj=logging_obj,
provider_config=provider_config,
body=body,
api_base=api_base,
api_key=api_key,
headers=headers,
litellm_params=litellm_params,
)
async_httpx_client: Final = (
client
if client is not None
else get_async_httpx_client(llm_provider=litellm.LlmProviders(custom_llm_provider))
)
try:
response: Final = await async_httpx_client.post( # pyright: ignore[reportUnknownMemberType] # untyped
url,
json=dict(body),
headers=dict(outbound_headers),
timeout=timeout,
logging_obj=logging_obj,
)
except httpx.HTTPError as e:
raise self._handle_error(e=e, provider_config=provider_config)
return provider_config.transform_extraction_response(
model=model, custom_llm_provider=custom_llm_provider, raw_response=response
)
def _prepare_audio_transcription_request(
self,
model: str,
@ -6462,6 +6573,7 @@ class BaseLLMHTTPHandler:
provider_config: Union[
BaseConfig,
BaseDecisionsConfig,
BaseExtractionConfig,
BaseRerankConfig,
BaseResponsesAPIConfig,
BaseImageEditConfig,

View file

@ -82,9 +82,9 @@ class GoogleAIStudioGeminiConfig(VertexGeminiConfig):
return "tts" in model
def get_supported_openai_params(self, model: str) -> list[str]:
sampling_params: Final = ("temperature", "top_p") if self.supports_sampling_params(model) else ()
supported_params: Final = [
"temperature",
"top_p",
*sampling_params,
"max_tokens",
"max_completion_tokens",
"stream",

View file

@ -16,7 +16,9 @@ The handler is intentionally a thin orchestration layer:
3. Construct ``riva.client.Auth`` honoring NVCF (function-id metadata + TLS)
vs self-hosted (any host:port, optional TLS) modes.
4. Stream the audio through Riva's ``streaming_response_generator`` and
aggregate ``is_final`` results into a single transcript.
aggregate ``is_final`` results into a single transcript, or, when
``riva_offline`` is set, send the whole buffer through the unary
``offline_recognize`` RPC for NIMs that only ship offline engines.
5. Return a normalized ``TranscriptionResponse`` with ``duration`` exposed
on ``_hidden_params`` so cost calculation works.
@ -26,13 +28,19 @@ without the optional STT extras installed.
import asyncio
import inspect
from collections.abc import Callable, Iterable
from collections.abc import Callable, Iterable, Mapping
from types import MappingProxyType
from typing import TYPE_CHECKING, Any, Final, Protocol
import httpx
import litellm
from litellm.constants import COMPLETION_HTTP_FALLBACK_SECONDS
from litellm.litellm_core_utils.audio_utils.utils import (
get_audio_file_name,
process_audio_file,
)
from litellm.litellm_core_utils.completion_timeout import CompletionTimeout
from litellm.llms.nvidia_riva.audio_transcription.audio_utils import (
resample_to_riva_pcm,
)
@ -45,6 +53,7 @@ from litellm.llms.nvidia_riva.common_utils import (
NvidiaRivaException,
grpc_error_to_litellm_exception,
)
from litellm.secret_managers.main import str_to_bool
from litellm.types.utils import FileTypes, TranscriptionResponse
from litellm.utils import convert_to_model_response_object
@ -67,10 +76,20 @@ class _RivaAuth(Protocol):
"""Opaque ``riva.client.Auth`` handle."""
class _RecognizeFuture(Protocol):
def result(self, timeout: float | None = None) -> object: ...
def done(self) -> bool: ...
def cancel(self) -> bool: ...
class _AsrService(Protocol):
@property
def streaming_response_generator(self) -> Callable[..., Iterable[object]]: ...
def offline_recognize(self, audio_bytes: bytes, config: object, future: bool) -> _RecognizeFuture: ...
class _EndpointingConfig(Protocol):
"""Opaque ``EndpointingConfig`` protobuf message."""
@ -128,6 +147,38 @@ def _audio_encoding(riva_asr_module: _RivaAsrModule) -> _AudioEncoding:
return riva_asr_module.AudioEncoding
def _parse_riva_offline(value: object, drop_params: bool) -> bool:
match value:
case None:
return False
case bool():
return value
case str() if (parsed := str_to_bool(value)) is not None:
return parsed
case _ if drop_params:
return False
case _:
raise NvidiaRivaException(
status_code=400,
message=f"riva_offline must be true or false, got {value!r}. Set drop_params to ignore it.",
)
def _deadline_seconds(timeout: float | str | httpx.Timeout, drop_params: bool) -> float:
try:
limit: Final = CompletionTimeout.as_limit(timeout)
except ValueError:
if drop_params:
return COMPLETION_HTTP_FALLBACK_SECONDS
raise NvidiaRivaException(
status_code=400,
message=f"timeout must be a number of seconds, got {timeout!r}. Set drop_params to ignore it.",
) from None
if not isinstance(limit, httpx.Timeout):
return limit
return COMPLETION_HTTP_FALLBACK_SECONDS if limit.read is None else limit.read
class NvidiaRivaAudioTranscription:
"""Sync + async entry point for Riva ASR."""
@ -138,7 +189,7 @@ class NvidiaRivaAudioTranscription:
optional_params: dict,
litellm_params: dict,
model_response: TranscriptionResponse,
timeout: float,
timeout: float | str | httpx.Timeout | None,
logging_obj: "LiteLLMLoggingObj",
api_key: str | None,
api_base: str | None,
@ -183,7 +234,7 @@ class NvidiaRivaAudioTranscription:
optional_params: dict,
litellm_params: dict,
model_response: TranscriptionResponse,
timeout: float,
timeout: float | str | httpx.Timeout | None,
logging_obj: "LiteLLMLoggingObj",
api_key: str | None,
api_base: str | None,
@ -213,7 +264,7 @@ class NvidiaRivaAudioTranscription:
optional_params: dict,
litellm_params: dict,
model_response: TranscriptionResponse,
timeout: float,
timeout: float | str | httpx.Timeout | None,
logging_obj: "LiteLLMLoggingObj",
api_key: str | None,
api_base: str | None,
@ -259,6 +310,10 @@ class NvidiaRivaAudioTranscription:
response_format: Final = request_payload.get("response_format") or "json"
timestamp_granularities: Final = request_payload.get("timestamp_granularities")
drop_params: Final = (
"drop_params" in litellm_params and litellm_params["drop_params"] is True
) or litellm.drop_params is True
riva_offline: Final = _parse_riva_offline(optional_params.get("riva_offline"), drop_params=drop_params)
riva_module, riva_asr_module = _import_riva()
auth_obj: Final = self._construct_auth(
@ -272,10 +327,6 @@ class NvidiaRivaAudioTranscription:
riva_asr_module=riva_asr_module,
recognition_config_dict=recognition_config_dict,
)
streaming_config: Final[_StreamingRecognitionConfig] = riva_asr_module.StreamingRecognitionConfig(
config=recognition_config, interim_results=False
)
logging_obj.pre_call(
input=None,
api_key=api_key,
@ -286,24 +337,31 @@ class NvidiaRivaAudioTranscription:
"recognition_config": recognition_config_dict,
"nvcf_function_id_set": bool(optional_params.get("nvcf_function_id")),
"use_ssl": optional_params.get("use_ssl"),
"riva_offline": riva_offline,
},
},
)
try:
asr_service: Final[_AsrService] = riva_module.ASRService(auth_obj)
audio_chunks: Final = self._iter_audio_chunks(resampled.pcm_bytes)
stream_kwargs: Final[dict[str, object]] = {
"audio_chunks": audio_chunks,
"streaming_config": streaming_config,
}
# Forward the deadline so the stream cannot block forever if the
# server stalls. Older riva-client versions do not accept a
# ``timeout`` kwarg, so pass it only when supported.
if timeout is not None and self._supports_timeout_kwarg(asr_service.streaming_response_generator):
stream_kwargs["timeout"] = float(timeout)
stream: Final = asr_service.streaming_response_generator(**stream_kwargs)
final_results: Final = self._collect_final_results(stream)
responses: Final = (
self._offline_responses(
asr_service=asr_service,
pcm_bytes=resampled.pcm_bytes,
recognition_config=recognition_config,
timeout=None if timeout is None else _deadline_seconds(timeout, drop_params),
)
if riva_offline
else self._streaming_responses(
asr_service=asr_service,
riva_asr_module=riva_asr_module,
pcm_bytes=resampled.pcm_bytes,
recognition_config=recognition_config,
timeout=timeout,
drop_params=drop_params,
)
)
final_results: Final = self._collect_final_results(responses, require_final=not riva_offline)
except NvidiaRivaException:
raise
except Exception as e:
@ -409,6 +467,54 @@ class NvidiaRivaAudioTranscription:
return config
def _streaming_responses(
self,
asr_service: _AsrService,
riva_asr_module: _RivaAsrModule,
pcm_bytes: bytes,
recognition_config: _RecognitionConfig,
timeout: float | str | httpx.Timeout | None,
drop_params: bool,
) -> Iterable[object]:
streaming_config: Final[_StreamingRecognitionConfig] = riva_asr_module.StreamingRecognitionConfig(
config=recognition_config, interim_results=False
)
return asr_service.streaming_response_generator(
audio_chunks=self._iter_audio_chunks(pcm_bytes),
streaming_config=streaming_config,
**self._timeout_kwargs(asr_service.streaming_response_generator, timeout, drop_params),
)
def _offline_responses(
self,
asr_service: _AsrService,
pcm_bytes: bytes,
recognition_config: _RecognitionConfig,
timeout: float | None,
) -> Iterable[object]:
pending: Final = asr_service.offline_recognize(audio_bytes=pcm_bytes, config=recognition_config, future=True)
try:
return (pending.result(timeout=timeout),)
except Exception as e:
if pending.done():
raise
pending.cancel()
raise NvidiaRivaException(
status_code=408,
message=f"NVIDIA Riva offline recognition did not finish within {timeout} seconds",
) from e
@classmethod
def _timeout_kwargs(
cls, rpc: Callable[..., object], timeout: float | str | httpx.Timeout | None, drop_params: bool
) -> Mapping[str, float]:
# Forward the deadline so the call cannot block forever if the
# server stalls. Older riva-client versions do not accept a
# ``timeout`` kwarg, so pass it only when supported.
if timeout is None or not cls._supports_timeout_kwarg(rpc):
return MappingProxyType({})
return MappingProxyType({"timeout": _deadline_seconds(timeout, drop_params)})
@staticmethod
def _supports_timeout_kwarg(callable_obj: Callable[..., object]) -> bool:
try:
@ -429,9 +535,9 @@ class NvidiaRivaAudioTranscription:
yield chunk
@staticmethod
def _collect_final_results(stream) -> list[dict[str, object]]:
def _collect_final_results(stream: Iterable[object], require_final: bool = True) -> list[dict[str, object]]:
"""
Walk the gRPC stream, ignore empty / non-final chunks, and return a
Walk the gRPC responses, ignore empty / non-final chunks, and return a
list of normalized final-result dicts. Matching the user's note: the
``id`` blocks with no ``results`` are streaming heartbeats and must
be skipped.
@ -440,7 +546,7 @@ class NvidiaRivaAudioTranscription:
for response in stream:
results = getattr(response, "results", None) or []
for result in results:
if not getattr(result, "is_final", False):
if require_final and not getattr(result, "is_final", False):
continue
alternatives = getattr(result, "alternatives", None) or []
if not alternatives:

View file

@ -41,6 +41,9 @@ _GRPC_STATUS_CODE_TO_HTTP: Final[dict] = {
}
_GRPC_MESSAGE_TOO_LARGE: Final = "message larger than max"
def _extract_grpc_status_name(error: object) -> str | None:
"""
Best-effort extraction of a gRPC StatusCode name from an arbitrary error.
@ -79,8 +82,16 @@ def grpc_error_to_litellm_exception(error: Exception) -> NvidiaRivaException:
a ``NvidiaRivaException`` with an appropriate HTTP-equivalent status code.
"""
status_name: Final = _extract_grpc_status_name(error)
http_status: Final = _GRPC_STATUS_CODE_TO_HTTP.get(status_name or "", 500)
detail: Final = _extract_grpc_details(error) or str(error)
if status_name == "RESOURCE_EXHAUSTED" and _GRPC_MESSAGE_TOO_LARGE in detail:
return NvidiaRivaException(
status_code=413,
message=(
f"NVIDIA Riva gRPC error ({status_name}): {detail}. The audio is too large for one offline "
"request to this Riva server: send shorter audio, or drop riva_offline on models that support streaming."
),
)
http_status: Final = _GRPC_STATUS_CODE_TO_HTTP.get(status_name or "", 500)
message = f"NVIDIA Riva gRPC error ({status_name}): {detail}" if status_name else f"NVIDIA Riva error: {detail}"
return NvidiaRivaException(status_code=http_status, message=message)

View file

@ -8,6 +8,7 @@ import time
from collections.abc import AsyncGenerator, Callable, Iterable, Iterator, Mapping
from contextlib import aclosing
from dataclasses import dataclass
from functools import partial
from types import MappingProxyType
from typing import IO, Any, Final, TypedDict
from urllib.parse import quote, unquote
@ -30,6 +31,7 @@ from litellm.litellm_core_utils.cloud_storage_security import (
split_configured_cloud_bucket_name,
validate_managed_cloud_file_id,
)
from litellm.litellm_core_utils.core_helpers import drops_unsupported_params
from litellm.litellm_core_utils.litellm_logging import Logging
from litellm.litellm_core_utils.prompt_templates.common_utils import (
extract_file_data,
@ -1103,21 +1105,26 @@ class VertexAIFilesConfig(VertexBase, BaseFilesConfig):
def _map_openai_to_vertex_params(
self,
openai_request_body: dict[str, Any],
deployment_drop_params: object = None,
) -> dict[str, object]:
"""
wrapper to call VertexGeminiConfig.map_openai_params
"""
from litellm.llms.vertex_ai.gemini.vertex_and_google_ai_studio_gemini import (
VertexGeminiConfig,
)
config: Final = VertexGeminiConfig()
_model: Final = openai_request_body.get("model", "")
raw_model: Final = openai_request_body.get("model", "")
model: Final = raw_model if isinstance(raw_model, str) else ""
sampling_params: Final = tuple(
p for p in ("temperature", "top_p", "top_k") if openai_request_body.get(p) is not None
)
if sampling_params and not VertexGeminiConfig.supports_sampling_params(model):
VertexGeminiConfig.reject_unsupported_sampling_param(
model=model, param=", ".join(sampling_params), drop_params=False
)
vertex_params: Final = config.map_openai_params(
model=_model,
model=model,
non_default_params=openai_request_body,
optional_params={},
drop_params=False,
drop_params=drops_unsupported_params(openai_request_body.get("drop_params", deployment_drop_params)),
allow_integer_enums=False,
)
return vertex_params
@ -1153,11 +1160,12 @@ class VertexAIFilesConfig(VertexBase, BaseFilesConfig):
create_file_data=create_file_data,
content_type=content_type,
):
deployment_drop_params: Final[object] = litellm_params.get("drop_params")
return {
"streaming_media_upload": StreamingMediaUploadConfig(
body_stream=_OpenAIToVertexBatchUploadStream(
file_data,
self._map_openai_to_vertex_params,
partial(self._map_openai_to_vertex_params, deployment_drop_params=deployment_drop_params),
),
content_type="application/json",
)

View file

@ -2,17 +2,20 @@ import json
from collections.abc import Awaitable, Callable, Mapping, Sequence
from dataclasses import dataclass
from itertools import accumulate, chain
from typing import Final, TypeAlias, TypeVar, cast # noqa: TID251 # sum_usage_metadata casts computed keys
from typing import Final, Literal, TypeAlias, TypeVar, cast # noqa: TID251 # sum_usage_metadata casts computed keys
import httpx
from pydantic import TypeAdapter, ValidationError
from litellm._logging import verbose_logger
from litellm.exceptions import Timeout
from litellm.litellm_core_utils.env_utils import get_env_int_or_none
from litellm.types.llms.vertex_ai import (
Candidates,
Citation,
CitationMetadata,
GenerateContentResponseBody,
GenerationConfig,
GroundingMetadata,
GroundingSegment,
GroundingSupport,
@ -48,8 +51,14 @@ HOP_FAILURES: Final = (VertexAIError, httpx.HTTPError, Timeout)
HOP_END_FINISH_REASON: Final = "CONTINUATION"
CUT_OFF_FINISH_REASONS: Final = frozenset({"MAX_TOKENS", HOP_END_FINISH_REASON})
DEFAULT_OUTPUT_BUDGET_ENV: Final = "LITELLM_GEMINI_DEFAULT_MAX_OUTPUT_TOKENS"
_OUTPUT_LIMIT_KEYS: Final = frozenset({"max_output_tokens", "maxOutputTokens"})
_REQUEST_KEYS_THAT_SKIP_THE_DEFAULT_BUDGET: Final = frozenset({"generation_config", "continuationToken"})
_TEXT_ONLY_MODALITIES: Final = TypeAdapter(list[Literal["TEXT"]])
_VERTEX_V1_PATH: Final = "/v1/projects/"
_VERTEX_PATH_ACCEPTING_CONTINUATION_TOKEN: Final = "/v1beta1/projects/"
_CANDIDATE_COUNT_KEYS: Final = ("candidate_count", "candidateCount")
LongDecodeHops: TypeAlias = tuple[GenerateContentResponseBody, ...]
AsyncHopPost: TypeAlias = Callable[[RequestBody], Awaitable[httpx.Response]]
@ -76,15 +85,53 @@ def _int_or_none(value: object) -> int | None:
return value if isinstance(value, int) else None
def _asks_for_one_candidate(generation_config: Mapping[str, object]) -> bool:
return all(
_int_or_none(generation_config[key] or 1) == 1 for key in _CANDIDATE_COUNT_KEYS if key in generation_config
)
def continuation_cap_of(request_body: RequestBody) -> int | None:
generation_config: Final = request_body.get("generationConfig")
if generation_config is None or "max_output_tokens" not in generation_config:
return None
if "candidate_count" in generation_config and _int_or_none(generation_config["candidate_count"]) != 1:
if not _asks_for_one_candidate(generation_config):
return None
return _int_or_none(generation_config["max_output_tokens"])
def _asks_for_text_only(generation_config: Mapping[str, object]) -> bool:
if "response_modalities" in generation_config:
return False
try:
_ = _TEXT_ONLY_MODALITIES.validate_python(generation_config.get("responseModalities", []))
except ValidationError:
return False
return True
def with_default_output_budget(
request_body: RequestBody, model_max_output_tokens: Callable[[], int | None]
) -> RequestBody:
generation_config: Final = request_body.get("generationConfig", GenerationConfig())
request_json: Final[Mapping[str, object]] = request_body
budget: Final = get_env_int_or_none(DEFAULT_OUTPUT_BUDGET_ENV)
if budget is None or budget <= 0:
return request_body
if not _REQUEST_KEYS_THAT_SKIP_THE_DEFAULT_BUDGET.isdisjoint(request_body):
return request_body
if not isinstance(request_json.get("generationConfig", {}), dict):
return request_body
if not _OUTPUT_LIMIT_KEYS.isdisjoint(generation_config):
return request_body
if not _asks_for_text_only(generation_config):
return request_body
model_max: Final = model_max_output_tokens()
if model_max is None:
return request_body
return {**request_body, "generationConfig": {**generation_config, "max_output_tokens": min(budget, model_max)}}
def client_continuation_token(candidate: Candidates) -> str | None:
if candidate.get("finishReason") not in CUT_OFF_FINISH_REASONS:
return None

View file

@ -7,15 +7,19 @@ Why separate file? Make it easy to see how transformation works
import os
import re
from collections.abc import Callable, Mapping
from functools import partial
from typing import TYPE_CHECKING, Any, Final, Literal, cast
from urllib.parse import quote
import httpx
from pydantic import BaseModel, TypeAdapter
from pydantic import BaseModel, PositiveInt, TypeAdapter, ValidationError
from typing_extensions import NotRequired, ReadOnly, TypedDict
import litellm
from litellm._logging import verbose_logger
from litellm.litellm_core_utils.asyncify import asyncify
from litellm.litellm_core_utils.core_helpers import drops_unsupported_params
from litellm.litellm_core_utils.get_model_cost_map import GetModelCostMap
from litellm.litellm_core_utils.prompt_templates.common_utils import (
get_image_mime_type_from_url,
)
@ -64,6 +68,7 @@ from ..common_utils import (
get_supports_response_schema,
get_supports_system_message,
)
from .long_decode import with_default_output_budget
# VertexBase is imported only for type checking to avoid introducing a
# module-load-time cyclic import to vertex_llm_base. The instance is lazily
@ -868,7 +873,7 @@ def gemini_convert_messages_with_history(
"""
has_text_in_content = check_text_in_content(user_content)
if has_text_in_content is False:
verbose_logger.warning(
verbose_logger.debug(
"No text in user content. Adding a blank text to user content, to ensure Gemini doesn't fail the request. Relevant Issue - https://github.com/BerriAI/litellm/issues/5515"
)
user_content.append(
@ -1189,7 +1194,18 @@ def transform_request_body(
# labels: optional explicit param and/or metadata.requester_metadata (OpenAI metadata)
labels: Final = pop_vertex_request_labels(optional_params, litellm_params)
filtered_params = {k: v for k, v in optional_params.items() if _get_equivalent_key(k, set(config_fields))}
from .vertex_and_google_ai_studio_gemini import VertexGeminiConfig
drops_top_k: Final = "top_k" in optional_params and not VertexGeminiConfig.supports_sampling_params(model)
if drops_top_k:
request_params: Final = cast(Mapping[str, object], litellm_params) # cast-ok: untyped shared signature
VertexGeminiConfig.reject_unsupported_sampling_param(
model=model,
param="top_k",
drop_params=drops_unsupported_params(request_params.get("drop_params")),
)
config_keys: Final = set(config_fields) - {"top_k"} if drops_top_k else set(config_fields)
filtered_params = {k: v for k, v in optional_params.items() if _get_equivalent_key(k, config_keys)}
generation_config: Final[GenerationConfig | None] = GenerationConfig(**filtered_params)
@ -1292,13 +1308,75 @@ def sync_transform_request_body(
vertex_auth_header=vertex_auth_header,
)
return transform_request_body(
messages=messages,
model=model,
custom_llm_provider=custom_llm_provider,
litellm_params=litellm_params,
cached_content=cached_content,
optional_params=optional_params,
return _with_default_output_budget(
transform_request_body(
messages=messages,
model=model,
custom_llm_provider=custom_llm_provider,
litellm_params=litellm_params,
cached_content=cached_content,
optional_params=optional_params,
),
model,
custom_llm_provider,
litellm_params.get("model_info"), # pyright: ignore[reportUnknownMemberType, reportUnknownArgumentType] # litellm_params is an untyped dict
)
class _CostMapOutputModalities(TypedDict):
supported_output_modalities: NotRequired[ReadOnly[tuple[str, ...]]]
_COST_MAP_OUTPUT_MODALITIES: Final = TypeAdapter(_CostMapOutputModalities)
def _answers_text(cost_map_key: str) -> bool:
try:
row: Final = _COST_MAP_OUTPUT_MODALITIES.validate_python(litellm.model_cost.get(cost_map_key)) # pyright: ignore[reportUnknownMemberType] # cost map rows are untyped JSON
except ValidationError:
return False
return "text" in row.get("supported_output_modalities", ("text",))
class _OutputLimit(TypedDict):
max_output_tokens: NotRequired[ReadOnly[PositiveInt | None]]
_OUTPUT_LIMIT: Final = TypeAdapter(_OutputLimit)
def _output_limit(fields: object) -> int | None:
try:
return _OUTPUT_LIMIT.validate_python(fields).get("max_output_tokens")
except ValidationError:
return None
def _catalog_capped_output_limit(cost_map_key: str, override: object) -> int | None:
limits: Final = (
_output_limit(GetModelCostMap.loaded_model_cost_map().get(cost_map_key)),
_output_limit(override),
)
return min((limit for limit in limits if limit is not None), default=None)
def _text_chat_model_max_output_tokens(
model: str, custom_llm_provider: str, deployment_model_info: object
) -> int | None:
info: Final = litellm.utils.lookup_model_info_for_registration(model, custom_llm_provider)
if isinstance(info, litellm.utils.ModelInfoLookupMiss) or info.get("mode") != "chat":
return None
if not _answers_text(info["key"]):
return None
return _catalog_capped_output_limit(info["key"], deployment_model_info or info)
def _with_default_output_budget(
request_body: RequestBody, model: str, custom_llm_provider: str, deployment_model_info: object
) -> RequestBody:
return with_default_output_budget(
request_body,
partial(_text_chat_model_max_output_tokens, model, custom_llm_provider, deployment_model_info),
)
@ -1368,21 +1446,8 @@ async def async_transform_request_body(
messages, should_inline=should_inline_remote_media(custom_llm_provider)
)
if _openai_messages_may_need_sync_gcs_metadata_fetch(inlined_messages):
# _transform_request_body may issue a sync httpx.get (up to 5s timeout)
# via _get_gcs_object_content_type to fetch GCS object metadata. Run the
# whole sync transformation on a worker thread so it does not block the
# async event loop.
return await asyncify(transform_request_body)(
messages=inlined_messages,
model=model,
custom_llm_provider=custom_llm_provider,
litellm_params=litellm_params,
cached_content=cached_content,
optional_params=optional_params,
)
return transform_request_body(
transform: Final = partial(
transform_request_body,
messages=inlined_messages,
model=model,
custom_llm_provider=custom_llm_provider,
@ -1390,6 +1455,21 @@ async def async_transform_request_body(
cached_content=cached_content,
optional_params=optional_params,
)
# _transform_request_body may issue a sync httpx.get (up to 5s timeout)
# via _get_gcs_object_content_type to fetch GCS object metadata. Run the
# whole sync transformation on a worker thread so it does not block the
# async event loop.
request_body: Final[RequestBody] = (
await asyncify(transform)()
if _openai_messages_may_need_sync_gcs_metadata_fetch(inlined_messages)
else transform()
)
return _with_default_output_budget(
request_body,
model,
custom_llm_provider,
litellm_params.get("model_info"), # pyright: ignore[reportUnknownMemberType, reportUnknownArgumentType] # litellm_params is an untyped dict
)
def _default_user_message_when_system_message_passed() -> ChatCompletionUserMessage:

View file

@ -3,15 +3,16 @@
## Initial implementation - covers gemini + image gen calls
import json
import re
import threading
import time
from collections.abc import Awaitable, Callable, Mapping, Sequence
from copy import deepcopy
from functools import partial
from functools import lru_cache, partial
from types import MappingProxyType
from typing import TYPE_CHECKING, Any, Final, Literal, Optional, Protocol, Union, cast, get_args
import httpx
from pydantic import JsonValue
from pydantic import JsonValue, TypeAdapter
import litellm
from litellm import verbose_logger
@ -27,7 +28,8 @@ from litellm.constants import (
DEFAULT_REASONING_EFFORT_MINIMAL_THINKING_BUDGET_GEMINI_2_5_PRO,
)
from litellm.exceptions import UnsupportedParamsError
from litellm.litellm_core_utils.core_helpers import normalize_drop_params
from litellm.litellm_core_utils.core_helpers import as_str_mapping, drops_unsupported_params, normalize_drop_params
from litellm.litellm_core_utils.get_model_cost_map import GetModelCostMap
from litellm.litellm_core_utils.json_fragment_accumulator import JSONFragmentAccumulator
from litellm.litellm_core_utils.prompt_templates.factory import (
encode_tool_call_id_with_signature,
@ -196,6 +198,29 @@ def _unsupported_reasoning_effort(reasoning_effort: str) -> UnsupportedParamsErr
)
_SAMPLING_DEPRECATION_UNMAPPED_MODELS: Final = "models outside litellm's model map"
_sampling_deprecation_lock: Final = threading.Lock()
@lru_cache(maxsize=None)
def _log_gemini_sampling_params_deprecated(model: str) -> None:
verbose_logger.warning(
"DeprecationWarning: `temperature`, `top_p`, and `top_k` continue to function for Gemini 3+ (%s) but are planned for removal in a future release. Move sampling guidance into the `system` instructions instead.",
model,
)
def _sampling_deprecation_notice_key(model: str) -> str:
catalog: Final = GetModelCostMap.loaded_model_cost_map()
is_cataloged: Final = any(name in catalog for name in (model, f"vertex_ai/{model}", f"gemini/{model}"))
return model if is_cataloged else _SAMPLING_DEPRECATION_UNMAPPED_MODELS
def _warn_gemini_sampling_params_deprecated(model: str) -> None:
with _sampling_deprecation_lock:
_log_gemini_sampling_params_deprecated(_sampling_deprecation_notice_key(model))
def _cache_creation_tokens(logging_obj: LoggingClass) -> int:
recorded: Final[object] = getattr(logging_obj, "vertex_ai_cache_creation_input_tokens", 0)
return recorded if isinstance(recorded, int) and recorded > 0 else 0
@ -207,6 +232,274 @@ def _served_model_name(model_version: object) -> str | None:
return model_version.split("@", 1)[0]
_GEMINI_SPEECH_CONFIG_KEY_MAP: Final[Mapping[str, str]] = MappingProxyType(
{
"multi_speaker_voice_config": "multiSpeakerVoiceConfig",
"speaker_voice_configs": "speakerVoiceConfigs",
"voice_config": "voiceConfig",
"prebuilt_voice_config": "prebuiltVoiceConfig",
"voice_name": "voiceName",
"replicated_voice_config": "replicatedVoiceConfig",
"mime_type": "mimeType",
"voice_sample_audio": "voiceSampleAudio",
"language_code": "languageCode",
"speaker_alias": "speakerAlias",
"speaker_id": "speakerId",
}
)
_SPEECH_CONFIG_ITEMS: Final = TypeAdapter(list[object])
def _normalize_gemini_speech_config_item(value: Mapping[str, object]) -> Mapping[str, object]:
normalized: Final[dict[str, object]] = {} # mutable-ok: iterative traversal fills nested output without recursion
pending: Final[ # mutable-ok: the explicit worklist must grow while avoiding recursive traversal
list[tuple[dict[str, object] | list[object], Mapping[str, object] | list[object]]]
] = [(normalized, value)]
while pending:
target, source = pending.pop()
items = source.items() if isinstance(source, Mapping) else enumerate(source)
for key, item in items:
if isinstance(source, Mapping):
if not isinstance(key, str):
continue
normalized_key: str | int = _GEMINI_SPEECH_CONFIG_KEY_MAP.get(key, key)
else:
normalized_key = len(target)
child: object
if (mapping_item := as_str_mapping(item)) is not None:
pending.append((child_mapping := dict[str, object](), mapping_item))
child = child_mapping
elif isinstance(item, list):
pending.append((child_items := list[object](), _SPEECH_CONFIG_ITEMS.validate_python(item)))
child = child_items
else:
child = (
OPENAI_TO_GEMINI_TTS_VOICES.get(item.lower(), item)
if normalized_key == "voiceName" and isinstance(item, str)
else item
)
if isinstance(target, dict):
assert isinstance(normalized_key, str)
target[normalized_key] = child
else:
target.append(child)
return normalized
def _gemini_speaker_voice_config(item: object) -> object:
speaker_config: Final = as_str_mapping(item)
if speaker_config is None:
return item
speaker: Final = speaker_config.get("speakerAlias")
voice_name: Final = speaker_config.get("speakerId")
if not isinstance(speaker, str) or not isinstance(voice_name, str):
return item
return {
**{key: value for key, value in speaker_config.items() if key not in ("speakerAlias", "speakerId")},
"speaker": speaker,
"voiceConfig": {
"prebuiltVoiceConfig": {"voiceName": OPENAI_TO_GEMINI_TTS_VOICES.get(voice_name.lower(), voice_name)}
},
}
def _with_gemini_speaker_voice_configs(config: Mapping[str, object]) -> Mapping[str, object]:
multi_speaker_config: Final = as_str_mapping(config.get("multiSpeakerVoiceConfig"))
if multi_speaker_config is None:
return config
speaker_voice_configs: Final = multi_speaker_config.get("speakerVoiceConfigs")
if not isinstance(speaker_voice_configs, list):
return config
return {
**config,
"multiSpeakerVoiceConfig": {
**multi_speaker_config,
"speakerVoiceConfigs": [
_gemini_speaker_voice_config(item)
for item in _SPEECH_CONFIG_ITEMS.validate_python(speaker_voice_configs)
],
},
}
_GEMINI_SPEECH_CONFIG_FIELDS: Final = frozenset({"voiceConfig", "multiSpeakerVoiceConfig", "languageCode"})
_GEMINI_VOICE_CONFIG_FIELDS: Final = MappingProxyType(
{
"prebuiltVoiceConfig": frozenset({"voiceName"}),
"replicatedVoiceConfig": frozenset({"mimeType", "voiceSampleAudio"}),
"voice": None,
}
)
_GEMINI_SPEAKER_VOICE_CONFIG_FIELDS: Final = frozenset({"speaker", "voiceConfig"})
def _known_leaf_fields(value: object, fields: frozenset[str] | None) -> object:
mapping: Final = as_str_mapping(value)
if fields is None or mapping is None:
return value
return {key: item for key, item in mapping.items() if key in fields}
def _known_voice_config_fields(value: object) -> object:
mapping: Final = as_str_mapping(value)
if mapping is None:
return value
return {
key: _known_leaf_fields(item, _GEMINI_VOICE_CONFIG_FIELDS[key])
for key, item in mapping.items()
if key in _GEMINI_VOICE_CONFIG_FIELDS
}
def _known_speaker_voice_config_fields(value: object) -> object:
mapping: Final = as_str_mapping(value)
if mapping is None:
return value
return {
key: _known_voice_config_fields(item) if key == "voiceConfig" else item
for key, item in mapping.items()
if key in _GEMINI_SPEAKER_VOICE_CONFIG_FIELDS
}
def _known_speaker_voice_configs(value: object) -> object:
if not isinstance(value, list):
return value
return [_known_speaker_voice_config_fields(item) for item in _SPEECH_CONFIG_ITEMS.validate_python(value)]
def _known_multi_speaker_voice_config_fields(value: object) -> object:
mapping: Final = as_str_mapping(value)
if mapping is None:
return value
return {key: _known_speaker_voice_configs(item) for key, item in mapping.items() if key == "speakerVoiceConfigs"}
def _known_speech_config_field(key: str, value: object) -> object:
match key:
case "voiceConfig":
return _known_voice_config_fields(value)
case "multiSpeakerVoiceConfig":
return _known_multi_speaker_voice_config_fields(value)
case _:
return value
def _known_gemini_speech_fields(
config: Mapping[str, object],
) -> dict[str, object]: # mutable-ok: provider request serialization requires a concrete dict
return {
key: _known_speech_config_field(key, item)
for key, item in config.items()
if key in _GEMINI_SPEECH_CONFIG_FIELDS
}
_GEMINI_TTS_VOICE_KEYS: Final = frozenset(
{
"format",
"name",
"voice",
"voiceName",
"voice_name",
"languageCode",
"language_code",
"modelName",
"model_name",
"speechConfig",
"speech_config",
"voiceConfig",
"voice_config",
"multiSpeakerVoiceConfig",
"multi_speaker_voice_config",
}
)
_NAMED_UNKNOWN_VOICE_KEYS: Final = 10
_NAMED_VOICE_KEY_CHARS: Final = 100
def _gemini_speech_config_wrapper(value: Mapping[str, object]) -> Mapping[str, object] | None:
camel_config: Final = as_str_mapping(value.get("speechConfig"))
return camel_config if camel_config is not None else as_str_mapping(value.get("speech_config"))
def named_voice_keys(keys: Sequence[str]) -> str:
named: Final = ", ".join(key[:_NAMED_VOICE_KEY_CHARS] for key in keys[:_NAMED_UNKNOWN_VOICE_KEYS])
unnamed_count: Final = len(keys) - _NAMED_UNKNOWN_VOICE_KEYS
return f"{named}, and {unnamed_count} more" if unnamed_count > 0 else named
def reject_unknown_gemini_tts_voice_keys(
model: str, voice: Mapping[str, object], drop_params: bool, llm_provider: str | None
) -> None:
nested_config: Final = _gemini_speech_config_wrapper(voice) or {}
ignored_wrapper: Final = (
("speech_config",)
if as_str_mapping(voice.get("speechConfig")) is not None and voice.get("speech_config") is not None
else ()
)
unknown_keys: Final = sorted(
{*(key for key in (*voice, *nested_config) if key not in _GEMINI_TTS_VOICE_KEYS), *ignored_wrapper}
)
if not unknown_keys or drops_unsupported_params(drop_params):
return
raise UnsupportedParamsError(
message=(
f"{model} does not support the voice keys {named_voice_keys(unknown_keys)}. Remove them, or set "
"drop_params to ignore them."
),
model=model,
llm_provider=llm_provider,
)
def normalize_gemini_speech_config(
value: Mapping[str, object], drop_unknown_fields: bool = False
) -> dict[str, object]: # mutable-ok: provider request serialization requires a concrete dict
speech_config: Final = {**_with_gemini_speaker_voice_configs(mapped_gemini_speech_config(value))}
return _known_gemini_speech_fields(speech_config) if drop_unknown_fields else speech_config
def _merge_gemini_speech_config_wrapper(value: Mapping[str, object]) -> Mapping[str, object]:
nested_config: Final = _gemini_speech_config_wrapper(value)
if nested_config is None:
return value
return {
**{key: item for key, item in value.items() if key not in ("speechConfig", "speech_config")},
**nested_config,
}
def mapped_gemini_speech_config(
value: Mapping[str, object],
) -> dict[str, object]: # mutable-ok: provider request serialization requires a concrete dict
normalized: Final = _normalize_gemini_speech_config_item(_merge_gemini_speech_config_wrapper(value))
speech_config: Final = {key: item for key, item in normalized.items() if key in _GEMINI_SPEECH_CONFIG_FIELDS}
voice_name: Final = next(
(normalized[key] for key in ("name", "voiceName", "voice") if isinstance(normalized.get(key), str)),
None,
)
if "voiceConfig" in speech_config or "multiSpeakerVoiceConfig" in speech_config or not isinstance(voice_name, str):
return speech_config
return {
**speech_config,
"voiceConfig": {
"prebuiltVoiceConfig": {"voiceName": OPENAI_TO_GEMINI_TTS_VOICES.get(voice_name.lower(), voice_name)}
},
}
class VertexAIBaseConfig:
def get_mapped_special_auth_params(self) -> dict:
"""
@ -404,11 +697,43 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig):
return False
return True
@staticmethod
def supports_sampling_params(model: str) -> bool:
"""Whether Google still honors ``temperature`` / ``top_p`` / ``top_k`` for ``model``.
Driven by ``supports_sampling_params`` in the model map: Gemini 3.6 Flash, 3.5 Flash-Lite
and later accept the fields but ignore them, and a later generation will 400 on them.
A model the map says nothing about keeps forwarding them. ``temperature`` and ``top_p``
leave ``get_supported_openai_params`` when the flag is off, so the shared unsupported-param
check raises or drops them and ``allowed_openai_params`` still forwards them verbatim;
``top_k`` is a provider-specific kwarg that bypasses that check and is gated while the
request body is built."""
rows: Final = cast(Mapping[str, Mapping[str, object]], litellm.model_cost) # cast-ok: JSON rows by model id
bare: Final = model.split("/")[-1]
for candidate in (model, bare, f"vertex_ai/{bare}", f"gemini/{bare}"):
if isinstance(flag := rows.get(candidate, {}).get("supports_sampling_params"), bool):
return flag
return True
@staticmethod
def reject_unsupported_sampling_param(model: str, param: str, drop_params: bool) -> None:
"""Return under ``drop_params`` so the caller drops ``param``, else raise the client-side 400
that names the escape hatch."""
if litellm.drop_params or drop_params:
return
raise UnsupportedParamsError(
message=(
f"{model} does not support {param}: the model map flags it `supports_sampling_params: false` "
"because Google ignores sampling params on this model. "
"To drop unsupported params, set `litellm.drop_params = True`."
),
status_code=400,
)
def get_supported_openai_params(self, model: str) -> list[str]:
sampling_params: Final = ("temperature", "top_p", "top_k") if self.supports_sampling_params(model) else ()
supported_params: Final = [
"temperature",
"top_p",
"top_k",
*sampling_params,
"max_tokens",
"max_completion_tokens",
"stream",
@ -1180,7 +1505,9 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig):
status_code=400,
)
def _map_audio_params(self, value: dict) -> dict:
def _map_audio_params(
self, value: Mapping[str, object], drop_unknown_fields: bool = False
) -> dict[str, object]: # mutable-ok: the public forwarding interface returns a concrete request dictionary
"""
Expected input:
{
@ -1199,12 +1526,6 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig):
languageCode: "en-US",
}
"""
from litellm.types.llms.vertex_ai import (
PrebuiltVoiceConfig,
SpeechConfig,
VoiceConfig,
)
# Validate audio format - Gemini TTS only supports pcm16
audio_format: Final = value.get("format")
if audio_format is not None and audio_format != "pcm16":
@ -1214,22 +1535,46 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig):
f"Please set audio format to 'pcm16'."
)
# Map OpenAI audio parameter to Gemini speech config
speech_config: Final[SpeechConfig] = {}
camel_speech_config: Final = value.get("speechConfig")
snake_speech_config: Final = value.get("speech_config")
speech_config_value: Final = (
camel_speech_config if isinstance(camel_speech_config, Mapping) else snake_speech_config
)
language_code: Final = value.get("language_code", value.get("languageCode"))
if isinstance(speech_config_value, Mapping):
return normalize_gemini_speech_config(value, drop_unknown_fields)
if "voice" in value:
voice: Final = value["voice"]
voice_name: Final = (
OPENAI_TO_GEMINI_TTS_VOICES.get(voice.lower(), voice) if isinstance(voice, str) else voice
)
prebuilt_voice_config: Final[PrebuiltVoiceConfig] = {"voiceName": voice_name}
voice_config: Final[VoiceConfig] = {"prebuiltVoiceConfig": prebuilt_voice_config}
speech_config["voiceConfig"] = voice_config
voice: Final = value.get("voice")
speech_config_without_language: Final[ # mutable-ok: provider request serialization requires a concrete dict
dict[str, object]
] = (
{
"voiceConfig": {
"prebuiltVoiceConfig": {
"voiceName": OPENAI_TO_GEMINI_TTS_VOICES.get(voice.lower(), voice)
if isinstance(voice, str)
else voice,
}
}
}
if "voice" in value
else {}
)
if "language_code" in value:
speech_config["languageCode"] = value["language_code"]
if language_code is not None:
return {
**speech_config_without_language,
"languageCode": language_code,
}
return cast(dict, speech_config)
return speech_config_without_language
def _map_chat_audio_params(
self, model: str, value: Mapping[str, object], drop_params: bool
) -> dict[str, object]: # mutable-ok: the public forwarding interface returns a concrete request dictionary
if _gemini_speech_config_wrapper(value) is not None:
reject_unknown_gemini_tts_voice_keys(model, value, drop_params, self.custom_llm_provider)
return self._map_audio_params(value, drop_unknown_fields=drops_unsupported_params(drop_params))
def map_audio_params(
self,
@ -1266,8 +1611,10 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig):
allow_integer_enums: bool = True,
) -> dict:
self._apply_include_server_side_tool_invocations(non_default_params, optional_params)
gemini_sampling_params_warned: bool = False
supports_sampling_params: Final = self.supports_sampling_params(model)
for param, value in non_default_params.items():
if param in ("temperature", "top_p", "top_k") and not supports_sampling_params:
continue
if param == "temperature":
if VertexGeminiConfig._is_gemini_3_or_newer(model):
if value is not None and value < 1.0:
@ -1275,28 +1622,15 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig):
"Warning: Setting temperature < 1.0 for Gemini 3 models (%s) can cause infinite loops, degraded reasoning performance, and failure on complex tasks. Strongly recommended to use temperature = 1.0 (default).",
model,
)
if not gemini_sampling_params_warned:
verbose_logger.warning(
"DeprecationWarning: `temperature`, `top_p`, and `top_k` continue to function for Gemini 3+ (%s) but are planned for removal in a future release. Move sampling guidance into the `system` instructions instead.",
model,
)
gemini_sampling_params_warned = True
_warn_gemini_sampling_params_deprecated(model)
optional_params["temperature"] = value
elif param == "top_p":
if VertexGeminiConfig._is_gemini_3_or_newer(model) and not gemini_sampling_params_warned:
verbose_logger.warning(
"DeprecationWarning: `temperature`, `top_p`, and `top_k` continue to function for Gemini 3+ (%s) but are planned for removal in a future release. Move sampling guidance into the `system` instructions instead.",
model,
)
gemini_sampling_params_warned = True
if VertexGeminiConfig._is_gemini_3_or_newer(model):
_warn_gemini_sampling_params_deprecated(model)
optional_params["top_p"] = value
elif param == "top_k":
if VertexGeminiConfig._is_gemini_3_or_newer(model) and not gemini_sampling_params_warned:
verbose_logger.warning(
"DeprecationWarning: `temperature`, `top_p`, and `top_k` continue to function for Gemini 3+ (%s) but are planned for removal in a future release. Move sampling guidance into the `system` instructions instead.",
model,
)
gemini_sampling_params_warned = True
if VertexGeminiConfig._is_gemini_3_or_newer(model):
_warn_gemini_sampling_params_deprecated(model)
optional_params["top_k"] = value
elif (
param == "stream" and value is True
@ -1305,7 +1639,7 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig):
elif param == "n":
optional_params["candidate_count"] = value
elif param == "audio" and isinstance(value, dict):
optional_params["speechConfig"] = self._map_audio_params(value)
optional_params["speechConfig"] = self._map_chat_audio_params(model, value, drop_params)
elif param == "stop":
if isinstance(value, str):
optional_params["stop_sequences"] = [value]

View file

@ -6,20 +6,25 @@ Reference: https://cloud.google.com/text-to-speech/docs/reference/rest/v1/text/s
"""
import base64
from collections.abc import Coroutine, Mapping
import math
from collections.abc import Coroutine, Iterator, Mapping, Sequence
from types import MappingProxyType
from typing import TYPE_CHECKING, Any, Final, TypeAlias, Union
from typing import TYPE_CHECKING, Any, ClassVar, Final, TypeAlias, Union
import httpx
from pydantic import ConfigDict, TypeAdapter
from pydantic import BaseModel, ConfigDict, TypeAdapter, ValidationError
import litellm
from litellm._logging import verbose_logger
from litellm.exceptions import UnsupportedParamsError
from litellm.litellm_core_utils.audio_utils.utils import (
DEFAULT_SPEECH_MEDIA_TYPE,
calculate_request_duration,
speech_media_type_from_audio_bytes,
)
from litellm.litellm_core_utils.core_helpers import as_str_mapping, drops_unsupported_params
from litellm.litellm_core_utils.url_utils import encode_url_path_segment
from litellm.llms.base_llm.chat.transformation import BaseLLMException
from litellm.llms.base_llm.text_to_speech.transformation import (
BaseTextToSpeechConfig,
TextToSpeechRequestData,
@ -33,6 +38,7 @@ from litellm.types.llms.vertex_ai import VERTEX_CREDENTIALS_TYPES
from litellm.types.llms.vertex_ai_text_to_speech import (
VertexTextToSpeechAudioConfig,
VertexTextToSpeechInput,
VertexTextToSpeechSpeakerVoiceConfig,
VertexTextToSpeechVoice,
)
@ -49,6 +55,166 @@ _LyriaVoice: TypeAlias = str | dict | None
_JSON_OBJECT: Final = TypeAdapter(Mapping[str, object], config=ConfigDict(hide_input_in_errors=True))
_JSON_ARRAY: Final = TypeAdapter(tuple[object, ...], config=ConfigDict(hide_input_in_errors=True))
_STR: Final = TypeAdapter(str, config=ConfigDict(hide_input_in_errors=True))
_GEMINI_CLOUD_TTS_LOCATIONS: Final = MappingProxyType(
{
**dict.fromkeys(
("us", "us-central1", "us-east1", "us-east4", "us-east5", "us-south1", "us-west1", "us-west4"), "us"
),
**dict.fromkeys(
("eu", "europe-central2", "europe-north1", "europe-southwest1", "europe-west1", "europe-west4"), "eu"
),
"northamerica-northeast1": "northamerica-northeast1",
}
)
_DEFAULT_GEMINI_TTS_VOICE: Final = "Kore"
class _GeminiCloudTtsInput(BaseModel):
model_config = ConfigDict(frozen=True)
text: str | None = None
ssml: str | None = None
prompt: str | None = None
class _GeminiCloudTtsAudioConfig(BaseModel):
model_config = ConfigDict(frozen=True)
audioEncoding: str = ""
sampleRateHertz: int | None = None
class _GeminiCloudTtsRequest(BaseModel):
model_config = ConfigDict(frozen=True)
input: _GeminiCloudTtsInput
audioConfig: _GeminiCloudTtsAudioConfig
def _fallback_gemini_tts_audio_duration(audio: bytes, encoding: str, sample_rate: int) -> float | None:
if encoding == "PCM":
return len(audio) / (2 * sample_rate) if sample_rate > 0 else None
if encoding in ("ALAW", "MULAW") and not (audio[:4] == b"RIFF" and audio[8:12] == b"WAVE"):
return len(audio) / sample_rate if sample_rate > 0 else None
if encoding in ("LINEAR16", "ALAW", "MULAW") and audio[:4] == b"RIFF" and audio[8:12] == b"WAVE":
format_offset: Final = audio.find(b"fmt ", 12)
data_offset: Final = audio.find(b"data", 12)
if format_offset >= 0 and data_offset >= 0 and format_offset + 20 <= len(audio):
byte_rate: Final = int.from_bytes(audio[format_offset + 16 : format_offset + 20], "little")
data_size: Final = int.from_bytes(audio[data_offset + 4 : data_offset + 8], "little")
if byte_rate > 0 and data_size <= len(audio) - data_offset - 8:
return data_size / byte_rate
if encoding == "MP3" and speech_media_type_from_audio_bytes(audio) == "audio/mpeg":
return len(audio) / 4000
if encoding == "OGG_OPUS" and audio[:4] == b"OggS":
opus_header: Final = audio.find(b"OpusHead")
last_page: Final = audio.rfind(b"OggS")
if opus_header >= 0 and opus_header + 12 <= len(audio) and last_page + 14 <= len(audio):
pre_skip: Final = int.from_bytes(audio[opus_header + 10 : opus_header + 12], "little")
granule: Final = int.from_bytes(audio[last_page + 6 : last_page + 14], "little")
if pre_skip <= granule < 2**64 - 1:
return (granule - pre_skip) / 48000
return None
_SPEAKER_CONFIG_ITEMS: Final = TypeAdapter(list[object])
_CloudTtsVoiceFields: TypeAlias = "Mapping[str, _CloudTtsVoiceFields]"
_CLOUD_TTS_VOICE_LEAF: Final[_CloudTtsVoiceFields] = MappingProxyType({})
_CLOUD_TTS_PREBUILT_VOICE_FIELDS: Final[_CloudTtsVoiceFields] = MappingProxyType(
{"prebuiltVoiceConfig": MappingProxyType({"voiceName": _CLOUD_TTS_VOICE_LEAF})}
)
_CLOUD_TTS_SPEECH_CONFIG_FIELDS: Final[_CloudTtsVoiceFields] = MappingProxyType(
{
"languageCode": _CLOUD_TTS_VOICE_LEAF,
"voiceConfig": _CLOUD_TTS_PREBUILT_VOICE_FIELDS,
"multiSpeakerVoiceConfig": MappingProxyType(
{
"speakerVoiceConfigs": MappingProxyType(
{
"speaker": _CLOUD_TTS_VOICE_LEAF,
"speakerAlias": _CLOUD_TTS_VOICE_LEAF,
"speakerId": _CLOUD_TTS_VOICE_LEAF,
"voiceConfig": _CLOUD_TTS_PREBUILT_VOICE_FIELDS,
}
)
}
),
}
)
_CLOUD_TTS_LIST_FIELDS: Final = frozenset({"speakerVoiceConfigs"})
_GEMINI_TTS_VOICE_NAME_KEYS: Final = ("name", "voiceName", "voice_name", "voice")
def _fields_cloud_tts_cannot_send(
config: Mapping[str, object], fields: _CloudTtsVoiceFields, path: str
) -> Iterator[str]:
for key, value in config.items():
yield from _field_cloud_tts_cannot_send(f"{path}{key}", value, fields.get(key), key in _CLOUD_TTS_LIST_FIELDS)
def _field_cloud_tts_cannot_send(
path: str, value: object, fields: "_CloudTtsVoiceFields | None", is_list_field: bool
) -> Iterator[str]:
if fields is None:
yield path
return
if value is None:
return
if not fields:
yield from () if isinstance(value, str) else (path,)
return
items: Final = (
_SPEAKER_CONFIG_ITEMS.validate_python(value) if is_list_field and isinstance(value, list) else (value,)
)
for item in items:
mapping = as_str_mapping(item)
yield from (path,) if mapping is None else _fields_cloud_tts_cannot_send(mapping, fields, f"{path}.")
def _non_string_voice_names(config: Mapping[str, object], path: str) -> Iterator[str]:
return (
f"{path}{key}"
for key in _GEMINI_TTS_VOICE_NAME_KEYS
if config.get(key) is not None and not isinstance(config.get(key), str)
)
def _non_mapping_speech_config_wrappers(voice: Mapping[str, object]) -> Iterator[str]:
return (
key
for key in ("speechConfig", "speech_config")
if voice.get(key) is not None and as_str_mapping(voice.get(key)) is None
)
def _speaker_entries(config: Mapping[str, object]) -> Iterator[Mapping[str, object]]:
multi_speaker_config: Final = as_str_mapping(config.get("multiSpeakerVoiceConfig"))
speakers: Final = multi_speaker_config.get("speakerVoiceConfigs") if multi_speaker_config is not None else None
items: Final = _SPEAKER_CONFIG_ITEMS.validate_python(speakers) if isinstance(speakers, list) else ()
return (entry for item in items if (entry := as_str_mapping(item)) is not None)
def _prebuilt_voices_replaced_by_speaker_id(config: Mapping[str, object]) -> Iterator[str]:
return (
"multiSpeakerVoiceConfig.speakerVoiceConfigs.voiceConfig.prebuiltVoiceConfig"
for entry in _speaker_entries(config)
if entry.get("speakerId") is not None
and (voice_config := as_str_mapping(entry.get("voiceConfig"))) is not None
and voice_config.get("prebuiltVoiceConfig") is not None
)
class VertexAITextToSpeechConfig(BaseTextToSpeechConfig, VertexBase):
@ -81,14 +247,19 @@ class VertexAITextToSpeechConfig(BaseTextToSpeechConfig, VertexBase):
}
# Response format mappings from OpenAI to Google Cloud audio encoding
FORMAT_MAPPINGS = {
"mp3": "MP3",
"opus": "OGG_OPUS",
"aac": "MP3", # Google doesn't have AAC, use MP3
"flac": "FLAC",
"wav": "LINEAR16",
"pcm": "LINEAR16",
}
FORMAT_MAPPINGS: ClassVar[Mapping[str, str]] = MappingProxyType(
{
"mp3": "MP3",
"opus": "OGG_OPUS",
"aac": "MP3", # Google doesn't have AAC, use MP3
"flac": "FLAC",
"wav": "LINEAR16",
"pcm": "LINEAR16",
}
)
GEMINI_FORMAT_MAPPINGS: ClassVar[Mapping[str, str]] = MappingProxyType(
{"mp3": "MP3", "opus": "OGG_OPUS", "alaw": "ALAW", "mulaw": "MULAW"}
)
def __init__(self) -> None:
BaseTextToSpeechConfig.__init__(self)
@ -96,7 +267,9 @@ class VertexAITextToSpeechConfig(BaseTextToSpeechConfig, VertexBase):
def _map_voice_to_vertex_format(
self,
voice: str | dict | None,
voice: str | dict[str, object] | None,
model: str | None = None,
drop_params: bool = False,
) -> tuple[str | None, dict | None]:
"""
Map voice to Vertex AI format.
@ -111,6 +284,9 @@ class VertexAITextToSpeechConfig(BaseTextToSpeechConfig, VertexBase):
- voice_str: Original string voice (for interface compatibility)
- voice_dict: Vertex AI format dict with languageCode and name
"""
if model is not None and self._is_gemini_tts_model(model):
return self._map_gemini_tts_voice_to_vertex_format(model=model, voice=voice, drop_params=drop_params)
if voice is None:
return None, None
@ -142,11 +318,225 @@ class VertexAITextToSpeechConfig(BaseTextToSpeechConfig, VertexBase):
return voice_str, voice_dict
@staticmethod
def _is_gemini_tts_model(model: str) -> bool:
from litellm.utils import is_gemini_tts_model
return is_gemini_tts_model(model)
@staticmethod
def _get_str_value(
source: Mapping[str, object],
*keys: str,
) -> str | None:
for key in keys:
value = source.get(key)
if isinstance(value, str):
return value
return None
@staticmethod
def _get_dict_value(
source: Mapping[str, object],
*keys: str,
) -> Mapping[str, object] | None:
for key in keys:
value = source.get(key)
if (mapping := as_str_mapping(value)) is not None:
return mapping
return None
@staticmethod
def _resolve_gemini_tts_voice_name(model: str, voice_name: str, drop_params: bool) -> str:
from litellm.llms.vertex_ai.gemini.vertex_and_google_ai_studio_gemini import OPENAI_TO_GEMINI_TTS_VOICES
gemini_voice_name: Final = OPENAI_TO_GEMINI_TTS_VOICES.get(voice_name.lower(), voice_name)
if "-" not in gemini_voice_name:
return gemini_voice_name
if drop_params:
return _DEFAULT_GEMINI_TTS_VOICE
raise UnsupportedParamsError(
message=(
f"{model} speaks Gemini voices such as {_DEFAULT_GEMINI_TTS_VOICE}, and {voice_name} is a Cloud "
"Text-to-Speech voice. Use a Gemini voice, or set drop_params to use the default Gemini voice."
),
model=model,
llm_provider="vertex_ai",
)
@staticmethod
def _reject_malformed_gemini_tts_voice(model: str, drop_params: bool) -> None:
if drop_params:
return
raise UnsupportedParamsError(
message=(
f"{model} needs voice to be a Gemini voice name such as {_DEFAULT_GEMINI_TTS_VOICE} or a speech "
"config object. Send one, or set drop_params to use the default Gemini voice."
),
model=model,
llm_provider="vertex_ai",
)
@staticmethod
def _reject_voice_fields_cloud_tts_cannot_send(model: str, voice: Mapping[str, object], drop_params: bool) -> None:
if drop_params:
return
from litellm.llms.vertex_ai.gemini.vertex_and_google_ai_studio_gemini import (
mapped_gemini_speech_config,
named_voice_keys,
)
speech_config: Final = VertexAITextToSpeechConfig._get_dict_value(voice, "speechConfig", "speech_config")
mapped_config: Final = mapped_gemini_speech_config(voice)
unsendable: Final = sorted(
{
*_fields_cloud_tts_cannot_send(mapped_config, _CLOUD_TTS_SPEECH_CONFIG_FIELDS, ""),
*_prebuilt_voices_replaced_by_speaker_id(mapped_config),
*_non_mapping_speech_config_wrappers(voice),
*_non_string_voice_names(voice, ""),
*_non_string_voice_names(speech_config or {}, "speechConfig."),
}
)
if not unsendable:
return
raise UnsupportedParamsError(
message=(
f"{model} cannot send the voice fields {named_voice_keys(unsendable)} as mp3, opus, alaw, or mulaw. "
"Request wav or pcm, remove them, or set drop_params to ignore them."
),
model=model,
llm_provider="vertex_ai",
)
@staticmethod
def _reject_malformed_gemini_tts_speaker(model: str, drop_params: bool) -> None:
if drop_params:
return
raise UnsupportedParamsError(
message=(
f"{model} needs each entry of speaker_voice_configs to name a speaker and a voice, as speaker with "
"voice_config.prebuilt_voice_config.voice_name, or speakerAlias with speakerId."
),
model=model,
llm_provider="vertex_ai",
)
def _extract_gemini_tts_speaker_configs(
self,
model: str,
speech_config: Mapping[str, object],
drop_params: bool,
) -> Sequence[VertexTextToSpeechSpeakerVoiceConfig]:
multi_speaker_config: Final = self._get_dict_value(speech_config, "multiSpeakerVoiceConfig")
if multi_speaker_config is None:
return []
raw_speaker_configs: Final = multi_speaker_config.get("speakerVoiceConfigs", [])
try:
validated_speaker_configs: Final = _SPEAKER_CONFIG_ITEMS.validate_python(raw_speaker_configs, strict=True)
except ValidationError:
self._reject_malformed_gemini_tts_speaker(model, drop_params)
return []
speaker_configs: Final[ # mutable-ok: validated payloads are accumulated for serialization
list[VertexTextToSpeechSpeakerVoiceConfig]
] = []
for raw_item in validated_speaker_configs:
if (raw_config := as_str_mapping(raw_item)) is None:
self._reject_malformed_gemini_tts_speaker(model, drop_params)
continue
speaker_alias = self._get_str_value(raw_config, "speakerAlias", "speaker")
speaker_id = self._get_str_value(raw_config, "speakerId") or self._prebuilt_voice_name(raw_config)
if speaker_alias is None or speaker_id is None:
self._reject_malformed_gemini_tts_speaker(model, drop_params)
continue
speaker_configs.append(
{
"speakerAlias": speaker_alias,
"speakerId": self._resolve_gemini_tts_voice_name(model, speaker_id, drop_params),
}
)
return speaker_configs
def _prebuilt_voice_name(self, config: Mapping[str, object]) -> str | None:
voice_config: Final = self._get_dict_value(config, "voiceConfig")
prebuilt_voice_config: Final = (
self._get_dict_value(voice_config, "prebuiltVoiceConfig") if voice_config is not None else None
)
if prebuilt_voice_config is None:
return None
return self._get_str_value(prebuilt_voice_config, "voiceName")
def _map_gemini_tts_voice_to_vertex_format(
self,
model: str,
voice: object,
drop_params: bool = False,
) -> tuple[str | None, dict[str, object]]: # mutable-ok: provider request serialization requires a concrete dict
from litellm.llms.vertex_ai.gemini.vertex_and_google_ai_studio_gemini import (
mapped_gemini_speech_config,
reject_unknown_gemini_tts_voice_keys,
)
default_voice: Final[dict[str, object]] = {
"languageCode": self.DEFAULT_LANGUAGE_CODE,
"modelName": model,
"name": _DEFAULT_GEMINI_TTS_VOICE,
}
if voice is None:
return None, default_voice
if isinstance(voice, str) and voice:
return voice, {
"languageCode": self.DEFAULT_LANGUAGE_CODE,
"modelName": model,
"name": self._resolve_gemini_tts_voice_name(model, voice, drop_params),
}
voice_config: Final = as_str_mapping(voice)
if voice_config is None:
self._reject_malformed_gemini_tts_voice(model, drop_params)
return None, default_voice
reject_unknown_gemini_tts_voice_keys(model, voice_config, drop_params, "vertex_ai")
speech_config: Final = mapped_gemini_speech_config(voice_config)
language_code: Final = self._get_str_value(speech_config, "languageCode") or self.DEFAULT_LANGUAGE_CODE
model_name: Final = model
speaker_configs: Final = self._extract_gemini_tts_speaker_configs(model, speech_config, drop_params)
self._reject_voice_fields_cloud_tts_cannot_send(model, voice_config, drop_params)
if speaker_configs:
return None, {
"languageCode": language_code,
"modelName": model_name,
"multiSpeakerVoiceConfig": {
"speakerVoiceConfigs": speaker_configs,
},
}
voice_name: Final = self._prebuilt_voice_name(speech_config)
if voice_name is not None:
return None, {
"languageCode": language_code,
"modelName": model_name,
"name": self._resolve_gemini_tts_voice_name(model, voice_name, drop_params),
}
return None, {
"languageCode": language_code,
"modelName": model_name,
"name": _DEFAULT_GEMINI_TTS_VOICE,
}
@staticmethod
def _dispatch_voice_name(voice: str | Mapping[str, object] | None) -> str | None:
if isinstance(voice, str):
return voice
if not isinstance(voice, Mapping):
return None
name: Final = voice.get("name")
return name if isinstance(name, str) else None
def dispatch_text_to_speech(
self,
model: str,
input: str,
voice: str | dict | None,
voice: str | dict[str, object] | None,
optional_params: dict[str, object],
litellm_params_dict: dict[str, object],
logging_obj: "LiteLLMLoggingObj",
@ -175,14 +565,17 @@ class VertexAITextToSpeechConfig(BaseTextToSpeechConfig, VertexBase):
vertex_project: Final = self.safe_get_vertex_ai_project(litellm_params_dict)
vertex_location: Final = self.safe_get_vertex_ai_location(litellm_params_dict)
# Convert voice to string if it's a dict (extract name)
# Actual voice mapping happens in map_openai_params
voice_str: str | None = None
if isinstance(voice, str):
voice_str = voice
elif isinstance(voice, dict):
# Extract voice name from dict if needed
voice_str = voice.get("name") if voice else None
mapped_voice, mapped_params = (
self.map_openai_params(
model=model,
voice=voice,
optional_params=optional_params,
drop_params=drops_unsupported_params(kwargs.get("drop_params")),
kwargs=kwargs,
)
if self._is_gemini_tts_model(model)
else (self._dispatch_voice_name(voice), optional_params)
)
# Store credentials in litellm_params for use in transform methods
litellm_params_dict.update(
@ -198,9 +591,11 @@ class VertexAITextToSpeechConfig(BaseTextToSpeechConfig, VertexBase):
response: Final = base_llm_http_handler.text_to_speech_handler(
model=model,
input=input,
voice=voice_str,
text_to_speech_provider_config=self,
text_to_speech_optional_params=optional_params,
voice=mapped_voice,
text_to_speech_provider_config=(
VertexAIGeminiCloudTextToSpeechConfig() if self._is_gemini_tts_model(model) else self
),
text_to_speech_optional_params=mapped_params,
custom_llm_provider="vertex_ai",
litellm_params=litellm_params_dict,
logging_obj=logging_obj,
@ -219,13 +614,13 @@ class VertexAITextToSpeechConfig(BaseTextToSpeechConfig, VertexBase):
Note: Vertex AI also supports additional parameters like audioConfig
which can be passed but are not part of the OpenAI spec
"""
return ["voice", "response_format", "speed"]
return ["voice", "response_format", "speed", *(["instructions"] if self._is_gemini_tts_model(model) else [])]
def map_openai_params(
self,
model: str,
optional_params: dict,
voice: str | dict | None = None,
voice: str | dict[str, object] | None = None,
drop_params: bool = False,
kwargs: dict = {},
) -> tuple[str | None, dict]:
@ -244,19 +639,28 @@ class VertexAITextToSpeechConfig(BaseTextToSpeechConfig, VertexBase):
Tuple of (mapped_voice_str, mapped_params)
"""
mapped_params: Final[dict[str, object]] = {}
openai_params: Final = _JSON_OBJECT.validate_python(optional_params)
drops_params: Final = drop_params or litellm.drop_params is True
##########################################################
# Map voice using helper
##########################################################
mapped_voice_str, voice_dict = self._map_voice_to_vertex_format(voice)
mapped_voice_str, voice_dict = self._map_voice_to_vertex_format(
voice=voice,
model=model,
drop_params=drops_params,
)
if voice_dict is not None:
mapped_params["vertex_voice_dict"] = voice_dict
# Map response format
if "response_format" in optional_params:
format_name: Final = optional_params["response_format"]
if format_name in self.FORMAT_MAPPINGS:
mapped_params["audioEncoding"] = self.FORMAT_MAPPINGS[format_name]
format_mappings: Final = (
self.GEMINI_FORMAT_MAPPINGS if self._is_gemini_tts_model(model) else self.FORMAT_MAPPINGS
)
if format_name in format_mappings:
mapped_params["audioEncoding"] = format_mappings[format_name]
else:
# Try to use it directly as Google Cloud format
mapped_params["audioEncoding"] = format_name
@ -265,10 +669,19 @@ class VertexAITextToSpeechConfig(BaseTextToSpeechConfig, VertexBase):
mapped_params["audioEncoding"] = self.DEFAULT_AUDIO_ENCODING
# Map speed (OpenAI: 0.25-4.0, Vertex AI: speakingRate 0.25-4.0)
if "speed" in optional_params:
speed: Final = optional_params["speed"]
if speed is not None:
mapped_params["speakingRate"] = str(speed)
speed: Final = openai_params.get("speed")
if speed is not None:
mapped_params["speakingRate"] = str(speed)
instructions: Final = openai_params.get("instructions")
if isinstance(instructions, str) and self._is_gemini_tts_model(model):
mapped_params["prompt"] = instructions
elif instructions is not None and self._is_gemini_tts_model(model) and not drops_params:
raise UnsupportedParamsError(
message=f"{model} needs instructions to be a string. Send one, or set drop_params to ignore it.",
model=model,
llm_provider="vertex_ai",
)
# Pass through Vertex AI-specific parameters from kwargs
if "audioConfig" in kwargs:
@ -311,15 +724,48 @@ class VertexAITextToSpeechConfig(BaseTextToSpeechConfig, VertexBase):
Google Cloud TTS endpoint: https://texttospeech.googleapis.com/v1/text:synthesize
"""
if self._is_gemini_tts_model(model):
return self._get_gemini_tts_url(
model=model, api_base=api_base, litellm_params=_JSON_OBJECT.validate_python(litellm_params)
)
if api_base:
return api_base
return self.TTS_API_URL
def _get_gemini_tts_url(self, model: str, api_base: str | None, litellm_params: Mapping[str, object]) -> str:
if api_base:
raise UnsupportedParamsError(
message=(
f"{model} serves mp3, opus, alaw and mulaw through Cloud Text-to-Speech, which does not use "
"this deployment's api_base. Request wav or pcm, or use a deployment without api_base."
),
model=model,
llm_provider="vertex_ai",
)
location: Final = litellm_params.get("vertex_location")
vertex_location: Final = location if isinstance(location, str) else self.get_default_vertex_location()
if vertex_location == "global":
return self.TTS_API_URL
cloud_tts_location: Final = _GEMINI_CLOUD_TTS_LOCATIONS.get(vertex_location)
if cloud_tts_location is None:
raise UnsupportedParamsError(
message=(
f"{model} serves mp3, opus, alaw and mulaw through Cloud Text-to-Speech, which has no Gemini TTS "
f"location for vertex_location {vertex_location}. Use a deployment in global, us, eu, "
"northamerica-northeast1, or a US or EU Vertex AI region."
),
model=model,
llm_provider="vertex_ai",
)
return f"https://{cloud_tts_location}-texttospeech.googleapis.com/v1/text:synthesize"
def _validate_vertex_input(
self,
input_data: VertexTextToSpeechInput,
optional_params: dict,
detect_ssml: bool,
) -> VertexTextToSpeechInput:
"""
Validate and transform input for Vertex AI TTS
@ -343,7 +789,7 @@ class VertexAITextToSpeechConfig(BaseTextToSpeechConfig, VertexBase):
else:
# LiteLLM will auto-detect if text is in ssml format
# check if "text" is an ssml - in this case we should pass it as ssml instead of text
if input_data:
if detect_ssml and input_data:
_text: Final = input_data.get("text", None) or ""
if "<speak>" in _text:
input_data["ssml"] = input_data.pop("text")
@ -403,8 +849,13 @@ class VertexAITextToSpeechConfig(BaseTextToSpeechConfig, VertexBase):
headers["x-goog-user-project"] = vertex_project
####### Build the request ################
vertex_input = VertexTextToSpeechInput(text=input)
vertex_input = self._validate_vertex_input(vertex_input, optional_params)
prompt: Final = _JSON_OBJECT.validate_python(optional_params).get("prompt")
validated_input: Final = self._validate_vertex_input(
VertexTextToSpeechInput(text=input), optional_params, detect_ssml=not self._is_gemini_tts_model(model)
)
vertex_input: Final[VertexTextToSpeechInput] = (
{**validated_input, "prompt": prompt} if isinstance(prompt, str) else validated_input
)
# Build voice configuration
# Check for voice dict stored in:
@ -412,8 +863,8 @@ class VertexAITextToSpeechConfig(BaseTextToSpeechConfig, VertexBase):
# 2. optional_params by map_openai_params
voice_dict: Final = litellm_params.get("vertex_voice_dict") or optional_params.get("vertex_voice_dict")
if voice_dict is not None and isinstance(voice_dict, dict):
vertex_voice = VertexTextToSpeechVoice(**voice_dict)
elif voice is not None and isinstance(voice, str):
vertex_voice = _JSON_OBJECT.validate_python(voice_dict)
elif voice is not None:
# Handle string voice (shouldn't normally happen if dispatch was called)
parts: Final = voice.split("-")
if len(parts) >= 2:
@ -485,8 +936,44 @@ class VertexAITextToSpeechConfig(BaseTextToSpeechConfig, VertexBase):
content=binary_data,
)
# Initialize the HttpxBinaryResponseContent instance
return HttpxBinaryResponseContent(response)
binary_response: Final = HttpxBinaryResponseContent(response)
if self._is_gemini_tts_model(model):
from litellm.types.utils import CompletionTokensDetailsWrapper, Usage
request: Final = _GeminiCloudTtsRequest.model_validate_json(raw_response.request.content)
input_data: Final = request.input
audio_config: Final = request.audioConfig
container_duration: Final = calculate_request_duration(binary_data)
sample_rate: Final = audio_config.sampleRateHertz or 24000
duration: Final = (
container_duration
if container_duration is not None
else _fallback_gemini_tts_audio_duration(binary_data, audio_config.audioEncoding, sample_rate)
)
if duration is None:
verbose_logger.warning(
"Could not measure Gemini TTS audio duration for %s, so its audio tokens are not billed", model
)
return binary_response
input_text: Final = " ".join(
value for value in (input_data.text, input_data.ssml, input_data.prompt) if value
)
prompt_tokens: Final = litellm.token_counter(model=model, text=input_text)
audio_tokens: Final = math.ceil(duration * 25)
binary_response.usage = Usage(
prompt_tokens=prompt_tokens,
completion_tokens=audio_tokens,
total_tokens=prompt_tokens + audio_tokens,
completion_tokens_details=CompletionTokensDetailsWrapper(audio_tokens=audio_tokens),
)
return binary_response
class VertexAIGeminiCloudTextToSpeechConfig(VertexAITextToSpeechConfig):
def get_error_class(self, error_message: str, status_code: int, headers: dict[str, str]) -> BaseLLMException:
return super().get_error_class(
error_message=error_message, status_code=400 if status_code == 404 else status_code, headers=headers
)
class VertexAILyriaTextToSpeechConfig(VertexAITextToSpeechConfig):

View file

@ -8846,8 +8846,11 @@ def speech(
generic_optional_params: Final = GenericLiteLLMParams.model_validate(kwargs)
# Handle Gemini models separately (they use speech_to_completion_bridge)
if "gemini" in model:
needs_cloud_text_to_speech: Final = (
litellm.utils.is_gemini_tts_model(model)
and response_format in VertexAITextToSpeechConfig.GEMINI_FORMAT_MAPPINGS
)
if "gemini" in model and not needs_cloud_text_to_speech:
from .endpoints.speech.speech_to_completion_bridge.handler import (
speech_to_completion_bridge_handler,
)

View file

@ -16580,6 +16580,24 @@
"source": "https://developers.cloudflare.com/workers-ai/platform/pricing/",
"supports_vision": true
},
"cloudflare/clairic": {
"input_cost_per_token": 1e-07,
"litellm_provider": "cloudflare",
"max_input_tokens": 65536,
"mode": "extraction",
"output_cost_per_token": 0.0,
"source": "https://developers.cloudflare.com/workers-ai/models/clairic/",
"supports_vision": true
},
"cloudflare/@cf/cloudflare/clairic": {
"input_cost_per_token": 1e-07,
"litellm_provider": "cloudflare",
"max_input_tokens": 65536,
"mode": "extraction",
"output_cost_per_token": 0.0,
"source": "https://developers.cloudflare.com/workers-ai/models/clairic/",
"supports_vision": true
},
"cloudflare/@cf/meta/llama-2-7b-chat-fp16": {
"input_cost_per_token": 1.923e-06,
"litellm_provider": "cloudflare",
@ -21903,6 +21921,7 @@
],
"supports_function_calling": true,
"supports_reasoning": true,
"supports_sampling_params": false,
"supports_tool_choice": true,
"supports_vision": true
},
@ -21927,6 +21946,7 @@
],
"supports_function_calling": true,
"supports_reasoning": true,
"supports_sampling_params": false,
"supports_tool_choice": true,
"supports_vision": true
},
@ -21958,6 +21978,7 @@
"supports_function_calling": true,
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_sampling_params": false,
"supports_tool_choice": true,
"supports_vision": true
},
@ -22021,6 +22042,7 @@
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_tool_choice": true,
"supports_vision": true
},
@ -29417,6 +29439,7 @@
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_url_context": true,
@ -30139,6 +30162,7 @@
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_url_context": true,
@ -30200,6 +30224,7 @@
"supports_reasoning": true,
"supports_minimal_reasoning_effort": false,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_url_context": true,
@ -30268,6 +30293,7 @@
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_url_context": true,
@ -30317,6 +30343,7 @@
"supports_reasoning": true,
"supports_minimal_reasoning_effort": false,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": false,
"supports_url_context": false,
@ -31743,6 +31770,7 @@
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_url_context": true,
@ -31927,6 +31955,7 @@
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_url_context": true,
@ -31989,6 +32018,7 @@
"supports_reasoning": true,
"supports_minimal_reasoning_effort": false,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_url_context": true,
@ -32051,6 +32081,7 @@
"supports_reasoning": true,
"supports_minimal_reasoning_effort": false,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_url_context": true,
@ -32420,6 +32451,7 @@
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_url_context": true,
@ -32481,6 +32513,7 @@
"supports_reasoning": true,
"supports_minimal_reasoning_effort": false,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_url_context": true,
@ -32549,6 +32582,7 @@
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_url_context": true,
@ -32598,6 +32632,7 @@
"supports_reasoning": true,
"supports_minimal_reasoning_effort": false,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": false,
"supports_url_context": false,
@ -33192,6 +33227,7 @@
"supports_function_calling": true,
"supports_parallel_function_calling": true,
"supports_reasoning": true,
"supports_sampling_params": false,
"supports_vision": true
},
"github_copilot/gemini-3.8-flash": {
@ -33210,6 +33246,7 @@
"supports_function_calling": true,
"supports_parallel_function_calling": true,
"supports_reasoning": true,
"supports_sampling_params": false,
"supports_vision": true
},
"github_copilot/gpt-3.5-turbo": {
@ -52605,6 +52642,7 @@
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_url_context": true,
@ -65823,6 +65861,7 @@
"supports_tool_choice": true,
"supports_response_schema": true,
"supports_reasoning": true,
"supports_sampling_params": false,
"supports_vision": true,
"source": "https://deepinfra.com/pricing",
"supports_audio_input": true
@ -69264,6 +69303,7 @@
"perplexity/google/gemini-3.5-flash-lite": {
"litellm_provider": "perplexity",
"mode": "responses",
"supports_sampling_params": false,
"supports_web_search": true,
"supports_function_calling": true,
"input_cost_per_token": 3e-07,
@ -69274,6 +69314,7 @@
"perplexity/google/gemini-3.6-flash": {
"litellm_provider": "perplexity",
"mode": "responses",
"supports_sampling_params": false,
"supports_web_search": true,
"supports_function_calling": true,
"input_cost_per_token": 1.5e-06,
@ -69284,6 +69325,7 @@
"perplexity/google/gemini-3.7-flash": {
"litellm_provider": "perplexity",
"mode": "responses",
"supports_sampling_params": false,
"supports_web_search": true,
"supports_function_calling": true,
"input_cost_per_token": 7.5e-07,
@ -71375,6 +71417,7 @@
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_vision": true,
"supports_web_search": true
},
@ -71392,6 +71435,7 @@
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_vision": true,
"supports_web_search": true
},
@ -74322,6 +74366,7 @@
"perplexity/google/gemini-3.8-flash": {
"litellm_provider": "perplexity",
"mode": "responses",
"supports_sampling_params": false,
"input_cost_per_token": 7.5e-07,
"output_cost_per_token": 3.75e-06,
"cache_read_input_token_cost": 7.5e-08,
@ -75064,12 +75109,10 @@
"gemini-nano-banana-2.1": {
"cache_read_input_token_cost": 1.5e-07,
"cache_read_input_token_cost_batches": 7.5e-08,
"cache_read_input_token_cost_flex": 7.5e-08,
"cache_read_input_token_cost_priority": 2.7e-07,
"input_cost_per_image": 0.00168,
"input_cost_per_token": 1.5e-06,
"input_cost_per_token_batches": 7.5e-07,
"input_cost_per_token_flex": 7.5e-07,
"input_cost_per_token_priority": 2.7e-06,
"litellm_provider": "vertex_ai-language-models",
"max_input_tokens": 131072,
@ -75080,7 +75123,6 @@
"output_cost_per_image_token": 3e-05,
"output_cost_per_token": 7.5e-06,
"output_cost_per_token_batches": 3.75e-06,
"output_cost_per_token_flex": 3.75e-06,
"output_cost_per_token_priority": 1.35e-05,
"search_context_cost_per_query": {"search_context_size_low":0.014,"search_context_size_medium":0.014,"search_context_size_high":0.014},
"source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing",
@ -75113,12 +75155,10 @@
"vertex_ai/gemini-nano-banana-2.1": {
"cache_read_input_token_cost": 1.5e-07,
"cache_read_input_token_cost_batches": 7.5e-08,
"cache_read_input_token_cost_flex": 7.5e-08,
"cache_read_input_token_cost_priority": 2.7e-07,
"input_cost_per_image": 0.00168,
"input_cost_per_token": 1.5e-06,
"input_cost_per_token_batches": 7.5e-07,
"input_cost_per_token_flex": 7.5e-07,
"input_cost_per_token_priority": 2.7e-06,
"litellm_provider": "vertex_ai-language-models",
"max_input_tokens": 131072,
@ -75129,7 +75169,6 @@
"output_cost_per_image_token": 3e-05,
"output_cost_per_token": 7.5e-06,
"output_cost_per_token_batches": 3.75e-06,
"output_cost_per_token_flex": 3.75e-06,
"output_cost_per_token_priority": 1.35e-05,
"search_context_cost_per_query": {"search_context_size_low":0.014,"search_context_size_medium":0.014,"search_context_size_high":0.014},
"source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing",
@ -76929,5 +76968,83 @@
"supported_output_modalities": [
"text"
]
},
"fireworks_ai/accounts/fireworks/models/deepseek-v4p1-flash-1017-us": {
"cache_read_input_token_cost": 9e-09,
"cache_read_input_token_cost_priority": 1.125e-08,
"input_cost_per_token": 4.5e-07,
"input_cost_per_token_priority": 5.625e-07,
"litellm_provider": "fireworks_ai",
"max_input_tokens": 1048576,
"max_output_tokens": 393216,
"max_tokens": 393216,
"mode": "chat",
"output_cost_per_token": 1.8e-06,
"output_cost_per_token_priority": 2.25e-06,
"source": "https://docs.fireworks.ai/serverless/us-only-serverless",
"supports_function_calling": true,
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_tool_choice": true,
"supports_vision": true
},
"wandb/reflection/Beam-1.0-501B-A23B": {
"cache_read_input_token_cost": 1e-07,
"input_cost_per_token": 1e-06,
"litellm_provider": "wandb",
"max_input_tokens": 1049000,
"mode": "chat",
"output_cost_per_token": 3e-06,
"source": "https://www.coreweave.com/products/coreweave-forge/serverless-inference",
"supports_function_calling": true,
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_tool_choice": true,
"supports_vision": false
},
"wandb/zai-org/GLM-5.3": {
"cache_read_input_token_cost": 1.9e-07,
"input_cost_per_token": 1.2e-06,
"litellm_provider": "wandb",
"max_input_tokens": 1049000,
"mode": "chat",
"output_cost_per_token": 4e-06,
"source": "https://www.coreweave.com/products/coreweave-forge/serverless-inference",
"supports_function_calling": true,
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_tool_choice": true,
"supports_vision": false
},
"azure_ai/MAI-Code-1.1-Flash": {
"cache_read_input_token_cost": 2e-08,
"deprecation_date": "2026-12-09",
"input_cost_per_token": 2e-07,
"litellm_provider": "azure_ai",
"max_input_tokens": 256000,
"max_output_tokens": 64000,
"max_tokens": 64000,
"mode": "chat",
"output_cost_per_token": 1.2e-06,
"source": "https://prices.azure.com/api/retail/prices?$filter=serviceName%20eq%20'Foundry%20Models'%20and%20armRegionName%20eq%20'eastus'%20and%20priceType%20eq%20'Consumption'",
"supported_endpoints": [
"/v1/chat/completions",
"/v1/responses"
],
"supported_modalities": [
"text",
"image"
],
"supported_output_modalities": [
"text"
],
"supports_function_calling": true,
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_tool_choice": true,
"supports_vision": true
}
}

View file

@ -4361,6 +4361,7 @@ class MCPServerManager:
raw_headers: Mapping[str, str] | None = None,
client_ip: str | None = None,
protocol_version_override: MCPUpstreamProtocol | None = None,
for_tool_listing: bool = False,
) -> MCPClient:
self.catalog.assert_current(server)
record_auth_resolution(server.server_id, AuthResolution.unresolved)
@ -4394,6 +4395,7 @@ class MCPServerManager:
else None
),
upstream_transport=self._upstream_transport,
for_tool_listing=for_tool_listing,
)
_create_mcp_client = create_mcp_client
@ -4540,6 +4542,7 @@ class MCPServerManager:
user_api_key_auth=user_api_key_auth,
raw_headers=raw_headers,
client_ip=client_ip,
for_tool_listing=True,
)
## HANDLE OPENAPI TOOLS

View file

@ -460,6 +460,8 @@ def validate_static_credential(
headers: Mapping[str, str],
upstream_token_header: str | None = None,
static_header_names: Iterable[str] = (),
*,
allow_absent: bool = False,
) -> Result[None, CredError]:
if auth_type not in _STATIC_MODES:
return Ok(None)
@ -475,17 +477,23 @@ def validate_static_credential(
)
)
values: Final = tuple((name.lower(), value.strip()) for name, value in headers.items() if name.lower() in slots)
if allow_absent and not values:
return Ok(None)
if any(_usable_credential_value(auth_type, name, value) for name, value in values):
return Ok(None)
return Error(CredError.of_misconfigured(f"{auth_type} requires a usable upstream credential"))
async def prepare_mcp_client(server: MCPServer, client: MCPClient) -> MCPClient:
async def prepare_mcp_client(server: MCPServer, client: MCPClient, *, for_tool_listing: bool = False) -> MCPClient:
if server.auth_type not in _STATIC_MODES or client.transport_type == MCPTransport.stdio:
return client
request: Final = await client.prepare_request_auth()
match validate_static_credential(
server.auth_type, request.headers, server.upstream_token_header, server.static_headers or ()
server.auth_type,
request.headers,
server.upstream_token_header,
server.static_headers or (),
allow_absent=for_tool_listing and server.auth_type == MCPAuth.authorization,
):
case Error(error):
raise_public(error)

View file

@ -208,6 +208,7 @@ async def prepare_upstream_client(
sampling_callback: SamplingCallback | None = None,
elicitation_callback: ElicitationCallback | None = None,
upstream_transport: httpx2.AsyncBaseTransport | None = None,
for_tool_listing: bool = False,
) -> MCPClient:
transport: Final[MCPTransportType] = server.transport or MCPTransport.sse
server_spec: Final = None if transport == MCPTransport.stdio else to_server_spec_fail_closed(server)
@ -272,6 +273,7 @@ async def prepare_upstream_client(
sampling_callback=sampling_callback,
elicitation_callback=elicitation_callback,
),
for_tool_listing=for_tool_listing,
)
aws_auth: Final = (
MCPSigV4Auth(
@ -316,4 +318,5 @@ async def prepare_upstream_client(
sampling_callback=sampling_callback,
elicitation_callback=elicitation_callback,
),
for_tool_listing=for_tool_listing,
)

View file

@ -275,7 +275,7 @@ LAZY_FEATURES: Final[tuple[LazyFeature, ...]] = (
LazyFeature(
name="decisions",
module_path="litellm.proxy.decisions_endpoints.endpoints",
path_prefixes=("/v1/decisions", "/decisions", "/v1/systemone", "/systemone"),
path_prefixes=("/v1/decisions", "/decisions", "/v1/systemone", "/systemone", "/v1/extraction", "/extraction"),
),
LazyFeature(
name="claude_code_marketplace",

File diff suppressed because it is too large Load diff

View file

@ -377,6 +377,8 @@ class LiteLLMRoutes(enum.Enum):
"/v1/decisions",
"/systemone",
"/v1/systemone",
"/extraction",
"/v1/extraction",
# OCR
"/ocr",
"/v1/ocr",
@ -3729,6 +3731,7 @@ class SpendLogsMetadata(TypedDict):
routing_decision: StandardLoggingRoutingDecision | None
kubernetes_pod_routing: ReadOnly[StandardLoggingKubernetesPodRouting | None]
internal_call_origin: InternalCallOrigin | None
billed_retry_attempt: ReadOnly[NotRequired[bool | None]]
litellm_roi_estimator: ReadOnly[NotRequired[bool | None]]
guardrail_information: list[StandardLoggingGuardrailInformation] | None
eval_information: Any | None

View file

@ -32,6 +32,7 @@ _MANAGED_MODEL_ROUTES: Final = frozenset(
"rerank",
"decisions",
"systemone",
"extraction",
"ocr",
),
)
@ -77,6 +78,7 @@ _MODEL_ROUTE_KINDS: Final[
"/rerank": "body",
"/decisions": "body",
"/systemone": "body",
"/extraction": "body",
"/messages/count_tokens": "body",
":countTokens": "path",
}

View file

@ -446,6 +446,7 @@ _BANNED_REQUEST_BODY_PARAMS: Final[tuple[str, ...]] = (
# the request away from the admin's pinned configuration.
"nvcf_function_id",
"use_ssl",
"riva_offline",
# Per-deployment opt-in that hands the whole call to the Rust core. It is a
# deployment decision, not a request one: the Rust path uses its own client
# rather than the one the deployment configured, and reports no post_call,

View file

@ -494,7 +494,7 @@ Each command resolves your LiteLLM key (logging in via SSO when none is stored a
The right variables are picked per agent. Claude Code gets `ANTHROPIC_BASE_URL` (the proxy root, so it appends `/v1/messages`) and `ANTHROPIC_AUTH_TOKEN`, with any stray `ANTHROPIC_API_KEY` cleared so the proxy token wins, and `ENABLE_TOOL_SEARCH=true` (unless you already set it) so Claude Code keeps tool search on even though the base URL is a proxy rather than a first-party Anthropic host. It also gets `CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY=1` (again unless you already set it) so Claude Code v2.1.129+ fills its `/model` picker from the proxy's `/v1/models`; Claude Code only lists entries whose id contains `claude` or `anthropic`, so the proxy lists every other group to Claude Code as `claude-router-<UTF-8 hex of the group name>` and marks a group whose input window reaches 1M with `[1m]`, and a request on such an id is served by the group. Older Claude Code versions ignore the variable. Export it as `0` to turn discovery off. Codex and OpenCode get `OPENAI_BASE_URL` (the proxy plus `/v1`) and `OPENAI_API_KEY`. Codex ignores `OPENAI_BASE_URL`, so it is additionally pointed at the proxy through a custom provider passed as `-c` config overrides (HTTP/SSE Responses transport, since the proxy does not speak the Responses WebSocket protocol). OpenCode additionally gets `OPENCODE_CONFIG_CONTENT` holding a generated `litellm` provider (`@ai-sdk/openai-compatible`, the proxy `/v1` URL, `{env:OPENAI_API_KEY}`) with one model entry per chat model your key can see on `/v1/models`, so its model picker mirrors the proxy without a hand-maintained `opencode.json`; OpenCode merges that over your own config files, and if you already export `OPENCODE_CONFIG_CONTENT` yours is left alone. When the list cannot be fetched, `lite opencode` says so on stderr and launches anyway. Codex gets the same list as a catalog file, `$CODEX_HOME/litellm-models.json` (default `~/.codex/`), passed as `-c model_catalog_json=<path>` so `/model` lists exactly the proxy's chat models; a proxy model the installed Codex already knows (`gpt-5.5`, say) keeps that Codex's own entry, reasoning levels and prompt included, and only its place in the picker comes from the proxy, while a model Codex does not know gets the plain entry Codex uses for an unknown `-m` slug. Before launching, `lite codex` asks the installed Codex for its own list and then has it read the written file back (both through `codex debug models`), and when the fetch, either of those or the write fails (Codex releases older than 0.130 have no such command) it says so on stderr and launches with Codex's built-in catalog, leaving a rejected file in place.
pi ignores base-URL environment variables entirely, so `lite pi` (kept out of the `lite --help` command listing for now, but fully functional) wires it up differently: before handoff it fetches the models your key can use from the proxy's `/v1/models` (plus each model's context window and output cap from `/model_group/info`, when available) and syncs them into a `litellm` provider entry in pi's `~/.pi/agent/models.json` (honoring `PI_CODING_AGENT_DIR`), then starts pi on that provider's first model via an injected `--model litellm/<id>`. Only that one provider entry is rewritten; the rest of the file, including any other custom providers, is left alone. The entry references the key as `$LITELLM_PROXY_API_KEY`, which the wrapper exports for the session, so the token itself never lands on disk and plain `pi` outside the wrapper simply shows the litellm models as unavailable. Your own flags come after the injected pin, so `lite pi --model litellm/<other-id>` wins, and inside the TUI the `/model` picker lists every synced litellm model.
pi ignores base-URL environment variables entirely, so `lite pi` (kept out of the `lite --help` command listing for now, but fully functional) wires it up differently: before handoff it fetches the models your key can use from the proxy's `/v1/models` (plus each model's context window and output cap from `/model_group/info`, when available) and syncs them into a `litellm` provider entry in pi's `~/.pi/agent/models.json` (honoring `PI_CODING_AGENT_DIR`), then starts pi on that provider's first model via an injected `--model litellm/<id>`. Only that one provider entry is rewritten; the rest of the file, including any other custom providers, is left alone. The entry references the key as `$LITELLM_PROXY_API_KEY`, which the wrapper exports for the session, so the token itself never lands on disk and plain `pi` outside the wrapper simply shows the litellm models as unavailable. The entry also turns on pi's session header (`x-session-id`, sent by pi 0.80.7 and later), so every chat turn of one pi session shares one session id in the logs and sticks to one deployment when session affinity is on. A custom `pi --session-id` must be at least 8 letters, digits, `_`, or `-` for the proxy to read it. Your own flags come after the injected pin, so `lite pi --model litellm/<other-id>` wins, and inside the TUI the `/model` picker lists every synced litellm model.
Options (these belong to the wrapper, so put them before the agent's own flags):

View file

@ -191,7 +191,7 @@ def provider_block(
Real contextWindow/maxTokens matter: pi otherwise assumes 128k/16384, which
breaks compaction thresholds and over-asks models with smaller output caps.
pi sniffs compat from the base URL, and one gateway URL fronts models with
different capabilities, so both flags are pinned off.
different capabilities, so the store and long-cache flags are pinned off.
"""
return {
"baseUrl": base_url.rstrip("/") + "/v1",
@ -199,6 +199,8 @@ def provider_block(
"compat": {
"supportsStore": False,
"supportsLongCacheRetention": False,
"sendSessionAffinityHeaders": True,
"sessionAffinityFormat": "openrouter",
},
"apiKey": f"${LITELLM_PROXY_API_KEY_ENV}",
"models": [_model_entry(model_id, limits) for model_id in model_ids],

View file

@ -67,6 +67,7 @@ from litellm.litellm_core_utils.get_supported_openai_params import (
get_supported_openai_params,
)
from litellm.litellm_core_utils.internal_call_metadata import is_unbilled_non_inference_call_from_params
from litellm.litellm_core_utils.litellm_logging import BilledRoundsDispatch
from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
from litellm.litellm_core_utils.llm_cost_calc.guardrail_cost import guardrail_information_cost
from litellm.litellm_core_utils.llm_response_utils.get_headers import (
@ -133,12 +134,14 @@ from litellm.router import Router
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
from litellm.types.router import RouterRateLimitError
from litellm.types.router_weights import validate_router_weights
_LateResponseT = TypeVar("_LateResponseT", bound=Response)
_LlmCallT = TypeVar("_LlmCallT")
_ResponseT: Final = TypeVar("_ResponseT")
_ChunkT: Final = TypeVar("_ChunkT")
KNOWN_PROXY_ROUTES: Final = frozenset(
@ -190,6 +193,7 @@ ProxyRouteType: TypeAlias = Literal[
"asearch",
"adecisions",
"asystemone",
"aextraction",
"avideo_generation",
"avideo_list",
"avideo_status",
@ -262,6 +266,7 @@ from litellm.proxy.litellm_pre_call_utils import (
sanitize_for_log,
)
from litellm.proxy.policy_engine.response_retrieval import attach_post_call_pipelines_to_retrieval
from litellm.types.litellm_params import STREAM_USAGE_INJECTED_FIELD
from litellm.types.utils import (
ModelResponse,
ModelResponseStream,
@ -548,22 +553,29 @@ def model_deployments_support_stream_options(
return all(_litellm_model_supports_stream_options(m) for m in candidate_models)
_STREAM_USAGE_FLAGS: Final = ("_litellm_strip_stream_usage", STREAM_USAGE_INJECTED_FIELD)
def stream_usage_tracking_updates(
data: Mapping[str, object],
general_settings: Mapping[str, object],
route_type: str,
supports_stream_options: Callable[[], bool],
) -> Mapping[str, object]:
scrub: Final = {"_litellm_strip_stream_usage": False} if "_litellm_strip_stream_usage" in data else {}
scrub: Final = {flag: False for flag in _STREAM_USAGE_FLAGS if flag in data}
if data.get("stream", False) is not True:
return scrub
always_include: Final = general_settings.get("always_include_stream_usage")
stream_options: Final = data.get("stream_options")
if always_include is True:
if "stream_options" not in data:
return {**scrub, "stream_options": {"include_usage": True}}
return {**scrub, "stream_options": {"include_usage": True}, STREAM_USAGE_INJECTED_FIELD: True}
if isinstance(stream_options, dict) and "include_usage" not in stream_options:
return {**scrub, "stream_options": {**stream_options, "include_usage": True}}
return {
**scrub,
"stream_options": {**stream_options, "include_usage": True},
STREAM_USAGE_INJECTED_FIELD: True,
}
return scrub
if always_include is False and route_type == "agenerate_content_stream" and "stream_options" not in data:
return {**scrub, "stream_options": {"include_usage": False}}
@ -577,6 +589,7 @@ def stream_usage_tracking_updates(
return {
"stream_options": {**merged_stream_options, "include_usage": True},
"_litellm_strip_stream_usage": True,
STREAM_USAGE_INJECTED_FIELD: True,
}
@ -1481,11 +1494,11 @@ _is_azure_model_router_request: Final = is_azure_model_router_request
def _override_openai_response_model(
*,
response_obj: object,
response_obj: _ResponseT,
requested_model: str,
log_context: str,
return_raw_model_name: bool = False,
) -> None:
) -> _ResponseT:
"""
Force the OpenAI-compatible `model` field in the response to match what the client requested.
@ -1509,7 +1522,7 @@ def _override_openai_response_model(
model group name instead of the comma-separated list the client sent.
"""
if return_raw_model_name or not requested_model:
return
return response_obj
hidden_params: Final = get_hidden_params_dict(response_obj)
if isinstance(hidden_params, dict):
@ -1522,7 +1535,7 @@ def _override_openai_response_model(
log_context,
attempted_fallbacks,
)
return
return response_obj
# For fastest_response batch completions, use the winning model's group
# name rather than the comma-separated list the client sent.
@ -1541,7 +1554,7 @@ def _override_openai_response_model(
"%s: fastest_response detected but no model group header found, preserving actual model from response.",
log_context,
)
return
return response_obj
# Check if this is an Azure Model Router request - if so, preserve the actual model used
if is_azure_model_router_request(requested_model, hidden_params):
@ -1549,11 +1562,12 @@ def _override_openai_response_model(
"%s: Azure Model Router detected - preserving actual model used from response instead of overriding to router model.",
log_context,
)
return
return response_obj
unchanged_response: Final = response_obj
if isinstance(response_obj, dict):
if "model" not in response_obj:
return
return unchanged_response
downstream_model = response_obj.get("model")
if downstream_model != requested_model:
verbose_proxy_logger.debug(
@ -1563,10 +1577,10 @@ def _override_openai_response_model(
downstream_model,
)
response_obj["model"] = requested_model
return
return unchanged_response
if not hasattr(response_obj, "model"):
return
return response_obj
downstream_model = getattr(response_obj, "model", None)
if downstream_model != requested_model:
@ -1577,6 +1591,9 @@ def _override_openai_response_model(
downstream_model,
)
if isinstance(response_obj, ExtractionResponse):
return response_obj.model_copy(update={"model": requested_model})
try:
setattr(response_obj, "model", requested_model)
except Exception as e:
@ -1588,6 +1605,7 @@ def _override_openai_response_model(
str(e),
exc_info=True,
)
return response_obj
_METADATA_BUCKET_KEYS: Final = ("metadata", "litellm_metadata")
@ -2200,6 +2218,7 @@ class ProxyBaseLLMRequestProcessing:
"asearch",
"adecisions",
"asystemone",
"aextraction",
"avideo_generation",
"avideo_list",
"avideo_status",
@ -3226,7 +3245,7 @@ class ProxyBaseLLMRequestProcessing:
# Always return the client-requested model name (not provider-prefixed internal identifiers)
# for OpenAI-compatible responses.
if requested_model_from_client:
_override_openai_response_model(
response = _override_openai_response_model( # rebind-ok: a frozen response comes back as a copy
response_obj=response,
requested_model=requested_model_from_client,
log_context=f"litellm_call_id={logging_obj.litellm_call_id}",
@ -3978,6 +3997,8 @@ class ProxyBaseLLMRequestProcessing:
async def _on_deferred_native_stream_complete(*args: object) -> None:
match args:
case (BilledRoundsDispatch(dispatch=logging_coroutine),):
GLOBAL_LOGGING_WORKER.ensure_initialized_and_enqueue(async_coroutine=logging_coroutine)
case (logging_coroutine,) if asyncio.iscoroutine(logging_coroutine):
GLOBAL_LOGGING_WORKER.ensure_initialized_and_enqueue(async_coroutine=logging_coroutine)
case (assembled_response, cache_hit):

View file

@ -29,6 +29,7 @@ from litellm._internal_context import service_target, with_service_target
from litellm._logging import verbose_proxy_logger
from litellm.caching import RedisCache
from litellm.constants import (
BILLED_RETRY_ATTEMPT_METADATA_KEY,
DB_DAILY_TAG_SPEND_UPDATE_JOB_NAME,
DB_SPEND_UPDATE_JOB_NAME,
INTERNAL_CALL_ORIGIN_METADATA_KEY,
@ -312,6 +313,11 @@ def _spend_tables_left_to_send(
return cast(DBSpendUpdateTransactions, remaining) if any(remaining.values()) else None
def _is_billed_retry_attempt(metadata: SpendLogsMetadata) -> bool:
"""An earlier billed retry attempt's own spend row, not a request the caller sent."""
return BILLED_RETRY_ATTEMPT_METADATA_KEY in metadata and metadata[BILLED_RETRY_ATTEMPT_METADATA_KEY] is True
def _timed_request_duration_ms(
payload: dict | SpendLogsPayload,
request_status: Literal["success", "failure"],
@ -1651,6 +1657,8 @@ class DBSpendUpdateWriter:
window_spend_update_queue=self.window_spend_update_queue,
)
await self._flush_tool_discovery_queue(prisma_client=prisma_client)
# Only commit from redis to db if this pod is the leader
if await self.pod_lock_manager.acquire_lock(
cronjob_id=DB_SPEND_UPDATE_JOB_NAME,
@ -2803,11 +2811,12 @@ class DBSpendUpdateWriter:
if not date:
return
metadata: Final[SpendLogsMetadata] = json.loads(payload["metadata"])
internal_origin: Final = bool(metadata.get(INTERNAL_CALL_ORIGIN_METADATA_KEY)) # pyright: ignore[reportUnknownMemberType] # TypedDict.get overloads carry Any defaults
request_error_accumulator.record(
payload=payload,
request_status=prisma_client.get_request_status(payload), # pyright: ignore[reportUnknownMemberType] # legacy payload union; this caller supplies a typed spend payload
date=date.split("T")[0],
is_internal_call=bool(metadata.get(INTERNAL_CALL_ORIGIN_METADATA_KEY)), # pyright: ignore[reportUnknownMemberType] # TypedDict.get overloads carry Any defaults
is_internal_call=internal_origin or _is_billed_retry_attempt(metadata),
)
async def _common_add_spend_log_transaction_to_daily_transaction(
@ -2872,6 +2881,7 @@ class DBSpendUpdateWriter:
endpoint = ROUTE_ENDPOINT_MAPPING.get(call_type, None)
is_internal_call: Final = bool(_metadata.get(INTERNAL_CALL_ORIGIN_METADATA_KEY))
caller_request: Final = not is_internal_call and not _is_billed_retry_attempt(_metadata)
cache_read_input_tokens: Final = extract_cache_read_tokens(usage_obj)
compression_saved_tokens: Final = extract_compression_saved_tokens(_metadata)
savings_spend: Final = compute_savings_spend(
@ -2906,9 +2916,9 @@ class DBSpendUpdateWriter:
# requests the caller made: counting them inflates request-volume
# readers, and an auto-router savings figure computed on a shadow
# duplicate credits savings for traffic no user sent.
api_requests=0 if is_internal_call else 1,
successful_requests=1 if not is_internal_call and request_status == "success" else 0,
failed_requests=1 if not is_internal_call and request_status != "success" else 0,
api_requests=1 if caller_request else 0,
successful_requests=1 if caller_request and request_status == "success" else 0,
failed_requests=1 if caller_request and request_status != "success" else 0,
cache_read_input_tokens=cache_read_input_tokens,
cache_creation_input_tokens=extract_cache_creation_tokens(usage_obj),
compression_saved_tokens=compression_saved_tokens,

View file

@ -10,8 +10,9 @@ worker on every call. ``SELECT DISTINCT ON`` keeps the transfer at one row per
from __future__ import annotations
import json
from collections.abc import Mapping, Sequence
from collections.abc import Iterable, Mapping, Sequence
from datetime import datetime, timezone
from types import MappingProxyType
from typing import TYPE_CHECKING, Final, Protocol
from pydantic import ConfigDict, JsonValue, TypeAdapter, field_validator
@ -113,3 +114,13 @@ async def fetch_latest_health_checks_for_models(
except Exception as query_err: # noqa: BLE001 # a paged model list must not fail on its health decoration
verbose_proxy_logger.error("Error getting latest health checks for models: %s", query_err)
return ()
def latest_by_model_name(rows: Iterable[LatestHealthCheckRow]) -> Mapping[str, LatestHealthCheckRow]:
"""The newest row per model name.
The queries above return one row per (model_id, model_name), so a model group with several
deployments comes back as several rows, plus one without a model_id if it was ever checked by
name. They arrive in model_id order with NULL last, which says nothing about which check is newest.
"""
return MappingProxyType({row.model_name: row for row in sorted(rows, key=lambda row: row.checked_at)})

View file

@ -1,6 +1,6 @@
from collections.abc import Mapping
from types import MappingProxyType
from typing import Annotated, Final, Literal
from typing import Annotated, Final, Literal, TypeAlias
from fastapi import APIRouter, Depends, Request, Response
from fastapi.responses import ORJSONResponse # pyright: ignore[reportDeprecated] # required endpoint contract
@ -18,6 +18,7 @@ from litellm.types.decisions import (
OpenAIDecisionRequestBody,
OpenAIDecisionResponse,
)
from litellm.types.extraction import ExtractionRequest, ExtractionRequestBody, ExtractionResponse
router: Final = APIRouter()
_REQUEST_DATA_ADAPTER: Final[TypeAdapter[dict[str, object]]] = TypeAdapter(dict[str, object])
@ -25,6 +26,7 @@ _DECISIONS_REQUEST_BODY_ADAPTER: Final[TypeAdapter[DecisionsRequestBody]] = Type
_OPENAI_DECISION_REQUEST_BODY_ADAPTER: Final[TypeAdapter[OpenAIDecisionRequestBody]] = TypeAdapter(
OpenAIDecisionRequestBody
)
_EXTRACTION_REQUEST_BODY_ADAPTER: Final[TypeAdapter[ExtractionRequestBody]] = TypeAdapter(ExtractionRequestBody)
_GENERAL_SETTINGS_ADAPTER: Final[TypeAdapter[dict[str, object]]] = TypeAdapter(dict[str, object])
_OPTIONAL_STRING_ADAPTER: Final[TypeAdapter[str | None]] = TypeAdapter(str | None)
_OPTIONAL_FLOAT_ADAPTER: Final[TypeAdapter[float | None]] = TypeAdapter(float | None)
@ -73,16 +75,35 @@ _DECISIONS_REQUEST_BODY: Final = inline_request_body(
],
},
)
_EXTRACTION_REQUEST_BODY: Final = inline_request_body(
ExtractionRequest,
{
"model": "clairic",
"state": "Invoice #4471 from Acme Tools, dated 2026-03-03. 12 hammers. Total $174. Payment due in 30 days.",
"fields": {
"vendor": {"type": "text", "instructions": "Who sent the invoice?"},
"invoice_date": {"type": "date"},
"total": {"type": "number", "unit": "USD"},
"items": {"type": "list", "items": "text", "instructions": "What was bought?"},
"days_to_pay": {"type": "estimate", "min": 0, "max": 90},
},
},
)
_ApiName: TypeAlias = Literal["Decisions", "Extraction"]
_RouteType: TypeAlias = Literal["adecisions", "asystemone", "aextraction"]
async def _invalid_request(
raw_data: Mapping[str, object], error: ValidationError, user_api_key_dict: UserAPIKeyAuth
raw_data: Mapping[str, object],
error: ValidationError,
user_api_key_dict: UserAPIKeyAuth,
api_name: _ApiName,
) -> Exception:
from litellm.proxy.proxy_server import proxy_logging_obj, version
return await ProxyBaseLLMRequestProcessing(data=dict(raw_data)).handle_llm_api_exception(
e=BadRequestError(
message=f"Invalid Decisions request: {error}",
message=f"Invalid {api_name} request: {error}",
model=str(raw_data.get("model", "")),
llm_provider="",
),
@ -95,19 +116,21 @@ async def _invalid_request(
_FIELDS_CHECKED_AFTER_ROUTING: Final = frozenset({"safety_identifier", "images", "assets", "audio", "videos"})
def _fields_checked_before_routing(data: Mapping[str, object]) -> Mapping[str, object]:
def _fields_checked_before_routing(data: Mapping[str, object], route_type: _RouteType) -> Mapping[str, object]:
if route_type == "aextraction":
return data
return MappingProxyType({key: value for key, value in data.items() if key not in _FIELDS_CHECKED_AFTER_ROUTING})
_DEPLOYMENT_ONLY_FIELDS: Final = frozenset({"model_info"})
async def _request_data(request: Request, user_api_key_dict: UserAPIKeyAuth) -> dict[str, object]:
async def _request_data(request: Request, user_api_key_dict: UserAPIKeyAuth, api_name: _ApiName) -> dict[str, object]:
body: Final = await request.body()
try:
data: Final = _REQUEST_DATA_ADAPTER.validate_json(body)
except ValidationError as error:
raise await _invalid_request(raw_data={}, error=error, user_api_key_dict=user_api_key_dict)
raise await _invalid_request(raw_data={}, error=error, user_api_key_dict=user_api_key_dict, api_name=api_name)
return {key: value for key, value in data.items() if key not in _DEPLOYMENT_ONLY_FIELDS}
@ -115,8 +138,10 @@ async def _process_decisions(
request: Request,
fastapi_response: Response,
user_api_key_dict: UserAPIKeyAuth,
body_adapter: TypeAdapter[DecisionsRequestBody] | TypeAdapter[OpenAIDecisionRequestBody],
route_type: Literal["adecisions", "asystemone"],
body_adapter: (
TypeAdapter[DecisionsRequestBody] | TypeAdapter[OpenAIDecisionRequestBody] | TypeAdapter[ExtractionRequestBody]
),
route_type: _RouteType,
) -> object:
from litellm.proxy.proxy_server import (
general_settings as proxy_general_settings,
@ -139,15 +164,17 @@ async def _process_decisions(
user_temperature as proxy_user_temperature,
)
data: Final = await _request_data(request, user_api_key_dict)
api_name: Final[_ApiName] = "Extraction" if route_type == "aextraction" else "Decisions"
data: Final = await _request_data(request, user_api_key_dict, api_name)
try:
body_adapter.validate_python(_fields_checked_before_routing(data))
body_adapter.validate_python(_fields_checked_before_routing(data, route_type))
except ValidationError as error:
raise await _invalid_request(raw_data=data, error=error, user_api_key_dict=user_api_key_dict)
raise await _invalid_request(raw_data=data, error=error, user_api_key_dict=user_api_key_dict, api_name=api_name)
general_settings: Final = _GENERAL_SETTINGS_ADAPTER.validate_python(proxy_general_settings)
user_api_base: Final = _OPTIONAL_STRING_ADAPTER.validate_python(proxy_user_api_base)
user_model: Final = _OPTIONAL_STRING_ADAPTER.validate_python(proxy_user_model)
user_temperature: Final = _OPTIONAL_FLOAT_ADAPTER.validate_python(proxy_user_temperature)
takes_generation_settings: Final = route_type != "aextraction"
processor: Final = ProxyBaseLLMRequestProcessing(data=data)
try:
return await processor.base_process_llm_request(
@ -162,9 +189,9 @@ async def _process_decisions(
select_data_generator=None,
model=None,
user_model=user_model,
user_temperature=user_temperature,
user_temperature=user_temperature if takes_generation_settings else None,
user_request_timeout=user_request_timeout,
user_max_tokens=user_max_tokens,
user_max_tokens=user_max_tokens if takes_generation_settings else None,
user_api_base=user_api_base,
version=version,
)
@ -263,3 +290,53 @@ async def decisions(
body_adapter=_OPENAI_DECISION_REQUEST_BODY_ADAPTER,
route_type="adecisions",
)
@router.post(
"/v1/extraction",
dependencies=[Depends(user_api_key_auth)],
response_class=ORJSONResponse, # pyright: ignore[reportDeprecated] # required endpoint contract
tags=["extraction"],
summary="Extract typed fields from a state",
responses={200: {"model": ExtractionResponse}},
response_model=None,
openapi_extra={"requestBody": _EXTRACTION_REQUEST_BODY},
)
@router.post(
"/extraction",
dependencies=[Depends(user_api_key_auth)],
response_class=ORJSONResponse, # pyright: ignore[reportDeprecated] # required endpoint contract
tags=["extraction"],
summary="Extract typed fields from a state",
responses={200: {"model": ExtractionResponse}},
response_model=None,
openapi_extra={"requestBody": _EXTRACTION_REQUEST_BODY},
)
async def extraction(
request: Request,
fastapi_response: Response,
user_api_key_dict: Annotated[UserAPIKeyAuth, Depends(user_api_key_auth)],
) -> object:
"""
Read a `state` text and up to 4 images and get one typed answer per named field in `fields`.
Field types are `text`, `number`, `date`, `time`, `datetime`, `duration`, `list` (set `items`), `estimate`
(set `min` and `max`) and `forecast` (set `min`, `max` and `horizon`). Each answer carries its `value`, which is
null when the input does not determine it, plus `confidence`, `p_null`, text `span`s into the state and, for
estimates, `mean`, `median`, `p10` and `p90`. Images are data URLs or `{content_type, base64}` objects, remote
URLs are not accepted. An unknown field type, a field name that is not 1 to 64 letters, digits, `_`, `.` or `-`
starting with a letter or `_`, a remote image URL or more than 4 images is rejected with a 400 before any
provider is called. Any other top-level key, `stream: true` and OpenAI generation settings such as `temperature`
included, is rejected with a 400 naming it unless `drop_params` or the deployment's `additional_drop_params`
drops it. Settings inside `fields` and `options` go to the provider as sent, so a field missing a setting its
type needs or an empty `state` with no images comes back with the provider's own error. `model` is any
extraction model in the proxy model list, such as Cloudflare Clairic. The request and the answers keep the
provider's format.
"""
return await _process_decisions(
request=request,
fastapi_response=fastapi_response,
user_api_key_dict=user_api_key_dict,
body_adapter=_EXTRACTION_REQUEST_BODY_ADAPTER,
route_type="aextraction",
)

View file

@ -494,7 +494,7 @@ class LLMShieldProxyGuardrail(CustomGuardrail):
sse: Final = AnthropicSSERestorer(step)
events: Final = ResponsesStreamRestorer(step, rehydrate)
carries: Final[CarryWindows] = {}
last_chunk = None # rebind-ok: tracks the most recent chunk for the final flush.
last_chunk = None # rebind-ok: tracks the most recent chunk with choices for the final flush.
async for chunk in response:
if isinstance(chunk, (bytes, str)):
@ -506,7 +506,8 @@ class LLMShieldProxyGuardrail(CustomGuardrail):
yield event
continue
restored_chunk = detached(chunk)
last_chunk = restored_chunk
if getattr(restored_chunk, "choices", None):
last_chunk = restored_chunk
for choice in attribute_items(restored_chunk, "choices"):
await self._restore_choice(choice, carries, session_id)
yield restored_chunk

View file

@ -2155,6 +2155,7 @@ async def test_model_connection(
"ocr",
"evaluation",
"decisions",
"extraction",
]
| None = fastapi.Body(
None,

View file

@ -254,39 +254,45 @@ class KeyManagementEventHooks:
litellm_changed_by: str | None = None,
) -> None:
from litellm.proxy.management_helpers.audit_logs import (
create_audit_log_for_update,
create_audit_logs_for_updates,
get_audit_log_changed_by,
is_audit_logging_enabled,
track_audit_task,
)
from litellm.proxy.proxy_server import litellm_proxy_admin_name
if not is_audit_logging_enabled():
return
for key in keys_being_deleted:
key_row = LiteLLM_VerificationToken.model_validate(key, from_attributes=True)
if key_row.token is None:
continue
_key_row = key_row.model_dump_json(exclude_none=True)
asyncio.create_task(
create_audit_log_for_update(
request_data=LiteLLM_AuditLogs(
id=str(uuid.uuid4()),
updated_at=datetime.now(timezone.utc),
changed_by=get_audit_log_changed_by(
litellm_changed_by=litellm_changed_by,
user_api_key_dict=user_api_key_dict,
litellm_proxy_admin_name=litellm_proxy_admin_name,
),
changed_by_api_key=user_api_key_dict.token,
table_name=LitellmTableNames.KEY_TABLE_NAME,
object_id=key_row.token,
action="deleted",
updated_values="{}",
before_value=_key_row,
)
)
changed_by: Final = get_audit_log_changed_by(
litellm_changed_by=litellm_changed_by,
user_api_key_dict=user_api_key_dict,
litellm_proxy_admin_name=litellm_proxy_admin_name,
)
key_rows: Final = (
LiteLLM_VerificationToken.model_validate(key, from_attributes=True) for key in keys_being_deleted
)
audit_rows: Final = tuple(
LiteLLM_AuditLogs(
id=str(uuid.uuid4()),
updated_at=datetime.now(timezone.utc),
changed_by=changed_by,
changed_by_api_key=user_api_key_dict.token,
table_name=LitellmTableNames.KEY_TABLE_NAME,
object_id=key_row.token,
action="deleted",
updated_values="{}",
before_value=key_row.model_dump_json(exclude_none=True),
)
for key_row in key_rows
if key_row.token is not None
)
if not audit_rows:
return
track_audit_task(
asyncio.create_task(
create_audit_logs_for_updates(audit_rows), name=f"{len(audit_rows)} deleted key audit logs"
)
)
@staticmethod
async def _store_virtual_key_in_secret_manager(secret_name: str, secret_token: str, team_id: str | None = None):

View file

@ -374,6 +374,8 @@ class _PROXY_VirtualKeyModelMaxBudgetLimiter(RouterBudgetLimiting):
self._redis_increment_queue_lock = asyncio.Lock()
self._redis_increment_flush_lock = asyncio.Lock()
self._detached_increment_operations = None
self._retired = False
self._flush_after_retire = None
self.deployment_budget_config = None
@with_service_target("model_budgets")

View file

@ -7,6 +7,7 @@ This is currently in development and not yet ready for production.
import asyncio
import binascii
import itertools
import json
import logging
import os
import uuid
@ -42,6 +43,7 @@ from litellm.caching.redis_batch import (
)
from litellm.caching.redis_cache import log_redis_failure
from litellm.constants import DYNAMIC_RATE_LIMIT_ERROR_THRESHOLD_PER_MINUTE, INTERNAL_CALL_ORIGIN_METADATA_KEY
from litellm.cost_calculator import get_usage_object
from litellm.integrations.custom_logger import CustomLogger
from litellm.litellm_core_utils.prompt_templates.common_utils import (
get_str_from_messages,
@ -78,6 +80,7 @@ from litellm.router_utils.add_retry_fallback_headers import (
from litellm.router_utils.common_utils import resolve_model_group_alias
from litellm.router_utils.ptu_shares import PTUTeamCeiling
from litellm.types.caching import RedisPipelineIncrementOperation
from litellm.types.extraction import ExtractionResponse
from litellm.types.llms.openai import BaseLiteLLMOpenAIResponseObject, ResponseAPIUsage
from litellm.types.mcp_server.mcp_server_manager import MCPServer
from litellm.types.passthrough_endpoints.pass_through_endpoints import EndpointType
@ -490,6 +493,7 @@ RESPONSES_API_CALL_TYPES: Final = ("aresponses", "responses")
EMBEDDING_API_CALL_TYPES: Final = ("aembedding", "embedding")
TEXT_COMPLETION_API_CALL_TYPES: Final = ("atext_completion", "text_completion")
RERANK_API_CALL_TYPES: Final = (CallTypes.rerank.value, CallTypes.arerank.value)
EXTRACTION_API_CALL_TYPES: Final = (CallTypes.extraction.value, CallTypes.aextraction.value)
GOOGLE_GENAI_NATIVE_CALL_TYPES: Final = (
CallTypes.generate_content.value,
CallTypes.agenerate_content.value,
@ -691,6 +695,7 @@ class RequestRateLimiterStash:
owner_litellm_call_id: str | None = None
rate_limit_response: RateLimitResponse | None = None
parallel_slot: ParallelSlotAcquisition | None = None
parallel_slot_reused_at: datetime | None = None
parallel_slot_release_lock: asyncio.Lock = field(default_factory=asyncio.Lock, repr=False, compare=False)
reserved_tokens: int = 0
reserved_model: RateLimitedModel | None = None
@ -787,6 +792,19 @@ def claim_request_stash_for_data(data: dict) -> RequestRateLimiterStash:
return stash
def _ended_before_slot_reuse(ended_at: datetime | None, reused_at: datetime | None) -> bool:
if ended_at is None or reused_at is None or (ended_at.tzinfo is None) != (reused_at.tzinfo is None):
return False
return ended_at <= reused_at
def _callback_kwargs_mapping(kwargs: object) -> Mapping[str, object]:
try:
return _REQUEST_RATE_LIMIT_DATA.validate_python(kwargs)
except ValidationError:
return MappingProxyType({})
def get_request_stash_for_call(litellm_call_id: str | None) -> RequestRateLimiterStash | None:
stash: Final = _request_stash.get()
if stash is None:
@ -1094,6 +1112,7 @@ class _PROXY_MaxParallelRequestsHandler_v3(CustomLogger):
or PROXY_MaxParallelRequestsHandler_v3._has_explicit_output_cap(data, call_type)
or is_embedding
or endpoint_type == EndpointType.DECISIONS
or call_type in EXTRACTION_API_CALL_TYPES
):
return
effective_cap: Final = max(capped_floor, configured_output_tokens or 0)
@ -1218,6 +1237,8 @@ class _PROXY_MaxParallelRequestsHandler_v3(CustomLogger):
if call_type in RESPONSES_API_CALL_TYPES or call_type in EMBEDDING_API_CALL_TYPES
else (None, estimable_data.get("prompt"), None)
if call_type in TEXT_COMPLETION_API_CALL_TYPES
else (None, self._extraction_input_text(_REQUEST_RATE_LIMIT_DATA.validate_python(estimable_data)), None)
if call_type in EXTRACTION_API_CALL_TYPES
else (estimable_data.get("messages"), None, None)
if call_type
else (
@ -1245,7 +1266,9 @@ class _PROXY_MaxParallelRequestsHandler_v3(CustomLogger):
estimated_input_tokens: Final = max(1, total_chars // DEFAULT_CHARS_PER_TOKEN) if total_chars > 0 else 0
explicit_max_tokens: Final = self._get_explicit_output_cap(data, call_type)
is_embedding: Final = self._is_embedding_request(data, call_type)
generates_no_output: Final = (
self._is_embedding_request(data, call_type) or call_type in EXTRACTION_API_CALL_TYPES
)
base_output_floor: Final = self.no_max_tokens_output_floor(min_configured_tpm_limit)
output_floor: Final = (
@ -1255,7 +1278,8 @@ class _PROXY_MaxParallelRequestsHandler_v3(CustomLogger):
)
max_tokens_estimate: Final = (
0
if is_embedding or (explicit_max_tokens is None and total_chars == 0 and configured_output_tokens is None)
if generates_no_output
or (explicit_max_tokens is None and total_chars == 0 and configured_output_tokens is None)
else explicit_max_tokens
if explicit_max_tokens is not None
else configured_output_tokens
@ -1642,6 +1666,7 @@ class _PROXY_MaxParallelRequestsHandler_v3(CustomLogger):
read_only: bool = False,
skip_tpm_check: bool = False,
parallel_slot_id: str | None = None,
skip_parallel_check: bool = False,
) -> RateLimitResponse:
"""
Check if any of the rate limit descriptors should be rate limited.
@ -1658,6 +1683,9 @@ class _PROXY_MaxParallelRequestsHandler_v3(CustomLogger):
``reserve_tpm_tokens`` reservation path should set this to
avoid the +1-per-key Lua / in-memory increment double-charging
the tokens counter.
skip_parallel_check: If True, ``max_parallel_requests`` gauges are
neither checked nor acquired, for a call that runs inside a
request already holding its slot.
``max_parallel_requests`` descriptors are enforced by the dedicated
concurrency-gauge path (``_check_parallel_request_gauges``), never by
@ -1750,7 +1778,7 @@ class _PROXY_MaxParallelRequestsHandler_v3(CustomLogger):
if windowed_response["overall_code"] == "OVER_LIMIT":
return windowed_response
if not gauges:
if not gauges or skip_parallel_check:
return windowed_response
gauge_response: Final = await self._check_parallel_request_gauges(
@ -2013,12 +2041,13 @@ class _PROXY_MaxParallelRequestsHandler_v3(CustomLogger):
parent_otel_span: Span | None,
*,
in_logging_callback: bool = False,
round_ended_at: datetime | None = None,
) -> None:
if stash is None:
return
async with stash.parallel_slot_release_lock:
acquisition: Final = stash.parallel_slot
if acquisition is None:
if acquisition is None or _ended_before_slot_reuse(round_ended_at, stash.parallel_slot_reused_at):
return
deferred: Final = in_logging_callback and await self._defer_parallel_slot_release(
acquisition, parent_otel_span
@ -3834,6 +3863,11 @@ class _PROXY_MaxParallelRequestsHandler_v3(CustomLogger):
if isinstance(part, (str, dict))
)
@staticmethod
def _extraction_input_text(data: Mapping[str, object]) -> str:
state: Final = data.get("state")
return f"{state if isinstance(state, str) else ''}\n{json.dumps(data.get('fields'), default=str)}"
def _estimate_precise_input_tokens(self, data: object, model: str | None, call_type: str | None = None) -> int:
"""
Model-aware input token estimate for the project ITPM reservation,
@ -3892,6 +3926,8 @@ class _PROXY_MaxParallelRequestsHandler_v3(CustomLogger):
if call_type in RERANK_API_CALL_TYPES
else (None, prompt, data.get("tools"), data.get("tool_choice"))
if call_type in TEXT_COMPLETION_API_CALL_TYPES
else (None, self._extraction_input_text(_REQUEST_RATE_LIMIT_DATA.validate_python(data)), None, None)
if call_type in EXTRACTION_API_CALL_TYPES
else (data.get("messages"), fallback_text, data.get("tools"), data.get("tool_choice"))
)
messages, selected_text, countable_tools, countable_tool_choice = selected_inputs
@ -3930,6 +3966,7 @@ class _PROXY_MaxParallelRequestsHandler_v3(CustomLogger):
tpm_reservation_amount: int,
call_type: str | None = None,
endpoint_type: EndpointType = EndpointType.GENERIC,
reuses_request_slot: bool = False,
) -> None:
"""
Reserve project-scoped ITPM/OTPM tokens (Bedrock Mantle-style
@ -4009,7 +4046,8 @@ class _PROXY_MaxParallelRequestsHandler_v3(CustomLogger):
parent_otel_span=user_api_key_dict.parent_otel_span,
)
stash.reservation_released = True
await self._release_stashed_parallel_slot(stash, user_api_key_dict.parent_otel_span)
if not reuses_request_slot:
await self._release_stashed_parallel_slot(stash, user_api_key_dict.parent_otel_span)
self._handle_rate_limit_error(
response=io_response,
descriptors=descriptors,
@ -4170,6 +4208,9 @@ class _PROXY_MaxParallelRequestsHandler_v3(CustomLogger):
verbose_proxy_logger.debug("Inside Rate Limit Pre-Call Hook")
stash: Final = claim_request_stash_for_data(data)
reuses_request_slot: Final = call_type == CallTypes.call_mcp_tool.value and stash.parallel_slot is not None
if reuses_request_slot:
stash.parallel_slot_reused_at = self._get_current_time()
#########################################################
# Check if the call type has a specific rate limiter
@ -4214,11 +4255,15 @@ class _PROXY_MaxParallelRequestsHandler_v3(CustomLogger):
# then because _reserve_project_io_tokens_or_raise below charges
# them unconditionally and counting them here too would
# double-charge every request.
parallel_counter_keys: Final = [
self.create_rate_limit_keys(d["key"], d["value"], "max_parallel_requests")
for d in descriptors
if (d.get("rate_limit") or {}).get("max_parallel_requests") is not None
]
parallel_counter_keys: Final = (
[]
if reuses_request_slot
else [
self.create_rate_limit_keys(d["key"], d["value"], "max_parallel_requests")
for d in descriptors
if (d.get("rate_limit") or {}).get("max_parallel_requests") is not None
]
)
parallel_slot_id: Final = uuid.uuid4().hex if parallel_counter_keys else None
first_pass_descriptors: Final = (
@ -4233,6 +4278,7 @@ class _PROXY_MaxParallelRequestsHandler_v3(CustomLogger):
parent_otel_span=user_api_key_dict.parent_otel_span,
skip_tpm_check=self.tpm_reservation_enabled,
parallel_slot_id=parallel_slot_id,
skip_parallel_check=reuses_request_slot,
)
if response["overall_code"] == "OVER_LIMIT":
@ -4242,7 +4288,8 @@ class _PROXY_MaxParallelRequestsHandler_v3(CustomLogger):
requested_model=requested_model,
)
else:
stash.rate_limit_response = response
if not reuses_request_slot:
stash.rate_limit_response = response
if parallel_slot_id is not None:
stash.parallel_slot = ParallelSlotAcquisition(
slot_id=parallel_slot_id,
@ -4340,7 +4387,8 @@ class _PROXY_MaxParallelRequestsHandler_v3(CustomLogger):
raw_estimate=estimated_tokens,
)
if stash.ptu_ceiling is not None and ptu_estimated_tokens > stash.ptu_ceiling.tpm_limit:
await self._release_stashed_parallel_slot(stash, user_api_key_dict.parent_otel_span)
if not reuses_request_slot:
await self._release_stashed_parallel_slot(stash, user_api_key_dict.parent_otel_span)
raise PTUShareExceededError(stash.ptu_ceiling, ptu_estimated_tokens)
tpm_response: Final = await self.reserve_tpm_tokens(
descriptors=descriptors,
@ -4350,7 +4398,8 @@ class _PROXY_MaxParallelRequestsHandler_v3(CustomLogger):
)
if tpm_response["overall_code"] == "OVER_LIMIT":
await self._release_stashed_parallel_slot(stash, user_api_key_dict.parent_otel_span)
if not reuses_request_slot:
await self._release_stashed_parallel_slot(stash, user_api_key_dict.parent_otel_span)
self._handle_rate_limit_error(
response=tpm_response,
descriptors=descriptors,
@ -4401,6 +4450,7 @@ class _PROXY_MaxParallelRequestsHandler_v3(CustomLogger):
tpm_reservation_amount=tpm_reservation_amount,
call_type=call_type,
endpoint_type=endpoint_type,
reuses_request_slot=reuses_request_slot,
)
def _create_pipeline_operations(
@ -4512,6 +4562,8 @@ class _PROXY_MaxParallelRequestsHandler_v3(CustomLogger):
) -> Usage | ResponseAPIUsage | Mapping[str, object] | None:
if isinstance(response_obj, (Usage, ResponseAPIUsage)):
return response_obj
if isinstance(response_obj, ExtractionResponse):
return get_usage_object(response_obj)
if isinstance(
response_obj,
(ModelResponse, EmbeddingResponse, TextCompletionResponse, BaseLiteLLMOpenAIResponseObject),
@ -5143,21 +5195,24 @@ class _PROXY_MaxParallelRequestsHandler_v3(CustomLogger):
# than parsed out of the body) carry their usage in
# ``combined_usage_object`` instead, and would otherwise never charge
# the TPM window.
_usage: Usage | dict[str, object] | None = None
if isinstance(
response_obj,
(
ModelResponse,
EmbeddingResponse,
TextCompletionResponse,
BaseLiteLLMOpenAIResponseObject,
),
):
_usage = getattr(response_obj, "usage", None)
else:
_combined_usage: Final = kwargs.get("combined_usage_object")
if isinstance(_combined_usage, Usage):
_usage = _combined_usage
_combined_usage: Final = kwargs.get("combined_usage_object")
_usage: Final[Usage | dict[str, object] | None] = (
getattr(response_obj, "usage", None)
if isinstance(
response_obj,
(
ModelResponse,
EmbeddingResponse,
TextCompletionResponse,
BaseLiteLLMOpenAIResponseObject,
),
)
else get_usage_object(response_obj)
if isinstance(response_obj, ExtractionResponse)
else _combined_usage
if isinstance(_combined_usage, Usage)
else None
)
total_tokens = self._get_total_tokens_from_usage(usage=_usage, rate_limit_type=rate_limit_type)
if total_tokens == 0:
total_tokens = self._aggregate_only_total_tokens(usage=_usage)
@ -5310,8 +5365,16 @@ class _PROXY_MaxParallelRequestsHandler_v3(CustomLogger):
try:
verbose_proxy_logger.debug("INSIDE parallel request limiter ASYNC SUCCESS LOGGING")
stash: Final = get_request_stash_for_call(_call_id_from_callback_kwargs(kwargs))
await self._release_stashed_parallel_slot(stash, litellm_parent_otel_span, in_logging_callback=True)
callback_kwargs: Final = _callback_kwargs_mapping(kwargs)
stash: Final = get_request_stash_for_call(_call_id_from_callback_kwargs(callback_kwargs))
await self._release_stashed_parallel_slot(
stash,
litellm_parent_otel_span,
in_logging_callback=True,
round_ended_at=end_time
if isinstance(end_time, datetime) and not callback_kwargs.get("stream")
else None,
)
pipeline_operations: Final = self._build_success_event_pipeline_operations(
kwargs=kwargs,

View file

@ -5,14 +5,14 @@ from contextlib import contextmanager
from contextvars import ContextVar
from datetime import datetime, timezone
from types import MappingProxyType
from typing import TYPE_CHECKING, Final, Protocol, cast
from typing import TYPE_CHECKING, Final, Protocol, TypeAlias, cast
from pydantic import TypeAdapter, ValidationError
import litellm
from litellm._logging import verbose_proxy_logger
from litellm.batches.batch_utils import batch_cost_is_final
from litellm.constants import BACKGROUND_INTERACTION_COST_POLLING_ENABLED
from litellm.constants import BACKGROUND_INTERACTION_COST_POLLING_ENABLED, BILLED_RETRY_ATTEMPT_METADATA_KEY
from litellm.integrations.custom_logger import CustomLogger
from litellm.litellm_core_utils.core_helpers import (
budget_reservation_from_metadata,
@ -54,11 +54,13 @@ from litellm.proxy.spend_tracking.spend_log_error_logger import (
)
from litellm.proxy.spend_tracking.spend_tracking_utils import ( # noqa: F401 # legacy module exports
_sanitize_error_information_for_spend_logs, # pyright: ignore[reportPrivateUsage,reportUnusedImport] # backwards-compatible package export
deployment_params_provider,
get_request_model_access_groups,
sanitize_error_information_for_spend_logs,
should_store_prompts_and_responses_in_spend_logs,
)
from litellm.proxy.utils import BilledFailedAttempt, ProxyUpdateSpend
from litellm.types.router import Deployment
from litellm.types.utils import (
CallTypes,
LiteLLMBatch,
@ -187,6 +189,14 @@ def _proxy_spend_writer() -> DBSpendUpdateWriter:
return proxy_logging_obj.db_spend_update_writer
DeploymentLookup: TypeAlias = Callable[[str], Deployment | None]
def _proxy_deployment(model_id: str) -> Deployment | None:
router: Final = get_llm_router()
return None if router is None else router.get_deployment(model_id=model_id)
def _billed_failed_attempts(request_data: Mapping[str, object]) -> tuple[BilledFailedAttempt, ...]:
"""The earlier billed attempts ``_failure_fields_to_lift`` stashed on the request, if any."""
attempts: Final = request_data.get("billed_failed_attempts")
@ -197,7 +207,10 @@ def _billed_failed_attempts(request_data: Mapping[str, object]) -> tuple[BilledF
def _failure_spend_rows(
request_data: Mapping[str, object], final_cost: float, late_guardrails_snapshot: object
request_data: Mapping[str, object],
final_cost: float,
late_guardrails_snapshot: object,
deployment_lookup: DeploymentLookup,
) -> tuple[tuple[float, Mapping[str, object]], ...]:
"""One ``(response_cost, kwargs)`` per spend row a failure writes: the final attempt under the
request id, then every earlier attempt the provider also billed under its own row."""
@ -209,25 +222,59 @@ def _failure_spend_rows(
return (
(final_cost, final_row),
*(
(attempt.response_cost, _attempt_spend_row_kwargs(request_data, attempt, position))
(
attempt.response_cost,
_attempt_spend_row_kwargs(
request_data, attempt, position, _attempt_deployment(attempt, deployment_lookup)
),
)
for position, attempt in enumerate(_billed_failed_attempts(request_data))
),
)
def _attempt_deployment(attempt: BilledFailedAttempt, deployment_lookup: DeploymentLookup) -> Deployment | None:
return None if attempt.deployment_id is None else deployment_lookup(attempt.deployment_id)
def _deployment_attribution(deployment: Deployment) -> Mapping[str, object]:
"""The spend-log attribution fields of the deployment that served an attempt: the model string it
was called with, its provider, its configured ``api_base`` and its model group."""
return MappingProxyType(
{
"model": deployment.litellm_params.model,
"custom_llm_provider": deployment_params_provider(deployment.litellm_params),
"api_base": deployment.litellm_params.api_base,
"model_group": deployment.model_name,
}
)
_NO_ATTRIBUTION: Final[Mapping[str, object]] = MappingProxyType({})
def _attempt_spend_row_kwargs(
request_data: Mapping[str, object], attempt: BilledFailedAttempt, position: int
request_data: Mapping[str, object], attempt: BilledFailedAttempt, position: int, deployment: Deployment | None
) -> Mapping[str, object]:
"""The final attempt's row kwargs re-keyed for one earlier billed attempt: its own
``request_id`` (``<id>-attempt-<position>``, since ``request_id`` is the spend-log primary key
and the retry count restarts after a fallback hop), its estimated usage and cost, its retry
count under ``attempted_retries`` and the deployment that billed it under ``model_info.id``."""
count under ``attempted_retries``, the deployment that billed it under ``model_info.id`` and,
when the router still knows that deployment, its model, provider, ``api_base`` and model group.
The row is stamped ``billed_retry_attempt`` so the daily spend tables keep its spend, tokens
and model usage without counting it as a request the caller sent."""
suffix: Final = f"-attempt-{position}"
attribution: Final = _NO_ATTRIBUTION if deployment is None else _deployment_attribution(deployment)
raw_params: Final = cast("Mapping[str, object]", request_data.get("litellm_params") or {}) # cast-ok: request dict
litellm_params: Final = dict(raw_params)
litellm_params: Final = {
**raw_params,
**({} if deployment is None else {"api_base": attribution["api_base"]}),
}
metadata: Final = {
**cast("Mapping[str, object]", litellm_params.get("metadata") or {}), # cast-ok: request dict
"attempted_retries": attempt.attempted_retries,
BILLED_RETRY_ATTEMPT_METADATA_KEY: True,
**({} if deployment is None else {"model_group": attribution["model_group"]}),
}
metadata.pop("previous_models", None)
if attempt.deployment_id is not None:
@ -238,6 +285,7 @@ def _attempt_spend_row_kwargs(
litellm_params["metadata"] = metadata
row: Final = {
**request_data,
**({} if deployment is None else {k: attribution[k] for k in ("model", "custom_llm_provider")}),
"litellm_params": litellm_params,
"combined_usage_object": attempt.usage,
"response_cost": attempt.response_cost,
@ -250,7 +298,11 @@ def _attempt_spend_row_kwargs(
if isinstance(standard_logging_object, dict):
payload: Final = cast("Mapping[str, object]", standard_logging_object) # cast-ok: StandardLoggingPayload dict
if isinstance(payload.get("id"), str):
row["standard_logging_object"] = {**payload, "id": f"{payload['id']}{suffix}"}
row["standard_logging_object"] = {
**payload,
"id": f"{payload['id']}{suffix}",
**({} if deployment is None else {**attribution, "model_id": attempt.deployment_id}),
}
return row
@ -262,10 +314,12 @@ class _ProxyDBLogger(CustomLogger):
turn_off_message_logging: bool = False,
message_logging: bool = True,
spend_writer: Callable[[], DBSpendUpdateWriter] = _proxy_spend_writer,
deployment_lookup: DeploymentLookup = _proxy_deployment,
) -> None:
super().__init__(turn_off_message_logging=turn_off_message_logging, message_logging=message_logging)
self.spend_event_producer = spend_event_producer
self._spend_writer: Final = spend_writer
self._deployment_lookup: Final = deployment_lookup
async def async_log_success_event(
self, kwargs: ObjectMapping, response_obj: object, start_time: datetime, end_time: datetime
@ -439,6 +493,7 @@ class _ProxyDBLogger(CustomLogger):
cast("Mapping[str, object]", request_data), # cast-ok: request dict, copied per row below
recovered_response_cost,
failure_snapshot_with_late_guardrails,
self._deployment_lookup,
),
)

View file

@ -29,6 +29,7 @@ from litellm._service_logger import ServiceLogging
from litellm._uuid import uuid
from litellm.constants import (
BAGGAGE_MAPPED_TAGS_METADATA_KEY,
BILLED_RETRY_ATTEMPT_METADATA_KEY,
CALLER_SESSION_ID_METADATA_KEY,
CLIENT_OUTPUT_CEILING_METADATA_KEY,
CONSUMED_REQUEST_TAGS_METADATA_KEY,
@ -267,7 +268,7 @@ def sanitize_for_log(value: object) -> str:
from litellm.router import Router
from litellm.secret_managers.main import get_secret_bool
from litellm.types.litellm_params import AGENTIC_LOOP_KWARG_NAMES
from litellm.types.litellm_params import AGENTIC_LOOP_KWARG_NAMES, STREAM_USAGE_INJECTED_FIELD
from litellm.types.llms.anthropic import ANTHROPIC_API_HEADERS
from litellm.types.router import BaggageMappedTagsStamp, holds_api_key_secret_reference
from litellm.types.services import ServiceTypes
@ -426,6 +427,7 @@ _UNTRUSTED_METADATA_CONTROL_FIELDS: Final = (
CONSUMED_REQUEST_TAGS_METADATA_KEY,
ROUTING_REQUEST_TAGS_METADATA_KEY,
INTERNAL_CALL_ORIGIN_METADATA_KEY,
BILLED_RETRY_ATTEMPT_METADATA_KEY,
MANAGED_BACKGROUND_RESPONSE_METADATA_KEY,
"standard_logging_object",
"litellm_roi_estimator",
@ -1218,6 +1220,32 @@ def get_dynamic_logging_metadata(
return callback_settings_obj
def _callback_settings_fields(callback_settings_obj: TeamCallbackMetadata) -> Mapping[str, object]:
callback_vars: Final = callback_settings_obj.callback_vars
return {
"success_callback": callback_settings_obj.success_callback,
"failure_callback": callback_settings_obj.failure_callback,
**({**callback_vars, TRUSTED_CALLBACK_VARS_FIELD: callback_vars} if callback_vars is not None else {}),
}
def key_and_team_logging_fields(user_api_key_dict: UserAPIKeyAuth, proxy_config: ProxyConfig) -> Mapping[str, object]:
callback_settings_obj: Final = get_dynamic_logging_metadata(
user_api_key_dict=user_api_key_dict, proxy_config=proxy_config
)
disabled_callbacks: Final = (user_api_key_dict.metadata or {}).get("litellm_disabled_callbacks")
return MappingProxyType(
{
**(_callback_settings_fields(callback_settings_obj) if callback_settings_obj is not None else {}),
**(
{"litellm_disabled_callbacks": disabled_callbacks}
if disabled_callbacks and isinstance(disabled_callbacks, list)
else {}
),
}
)
_get_dynamic_logging_metadata: Final = get_dynamic_logging_metadata
@ -2342,7 +2370,15 @@ def refresh_proxy_server_request_body_snapshot(
if not isinstance(proxy_server_request, dict):
return
_body_snapshot_exclude: Final = (
frozenset({"secret_fields", "proxy_server_request", "litellm_logging_obj", VECTOR_STORE_CALLER_METADATA_KWARG})
frozenset(
{
"secret_fields",
"proxy_server_request",
"litellm_logging_obj",
VECTOR_STORE_CALLER_METADATA_KWARG,
STREAM_USAGE_INJECTED_FIELD,
}
)
| _TRANSPORT_ONLY_CREDENTIAL_KEYS
| _CALLBACK_CREDENTIAL_KEYS
)
@ -2841,27 +2877,7 @@ async def add_litellm_data_to_request(
request=request, request_data=data, metadata_key=_metadata_variable_name
)
# Team Callbacks controls
callback_settings_obj: Final = get_dynamic_logging_metadata(
user_api_key_dict=user_api_key_dict, proxy_config=proxy_config
)
if callback_settings_obj is not None:
data["success_callback"] = callback_settings_obj.success_callback
data["failure_callback"] = callback_settings_obj.failure_callback
if callback_settings_obj.callback_vars is not None:
# unpack callback_vars in data
for k, v in callback_settings_obj.callback_vars.items():
data[k] = v
# Callbacks that must not honour request-supplied credentials read this
# proxy-owned field instead of the raw request kwargs.
data[TRUSTED_CALLBACK_VARS_FIELD] = callback_settings_obj.callback_vars
# Add disabled callbacks from key metadata
if user_api_key_dict.metadata and "litellm_disabled_callbacks" in user_api_key_dict.metadata:
disabled_callbacks: Final = user_api_key_dict.metadata["litellm_disabled_callbacks"]
if disabled_callbacks and isinstance(disabled_callbacks, list):
data["litellm_disabled_callbacks"] = disabled_callbacks
data.update(key_and_team_logging_fields(user_api_key_dict=user_api_key_dict, proxy_config=proxy_config))
# Guardrails from key/team metadata and policy engine
await move_guardrails_to_metadata(

View file

@ -11,7 +11,7 @@ from dataclasses import dataclass
from functools import partial, reduce
from itertools import chain, groupby
from types import MappingProxyType
from typing import TYPE_CHECKING, Final, NamedTuple, NoReturn, Protocol, TypeAlias, overload
from typing import TYPE_CHECKING, Annotated, Final, NamedTuple, NoReturn, Protocol, TypeAlias, overload
from fastapi import (
APIRouter,
@ -53,7 +53,12 @@ from litellm.proxy.auth.auth_checks import ( # noqa: F401 # legacy module expo
delete_cache_team_object,
)
from litellm.proxy.auth.user_api_key_auth import user_api_key_auth
from litellm.proxy.common_utils.auth_cache_invalidation_pubsub import evict_and_broadcast
from litellm.proxy.common_utils.auth_cache_invalidation_pubsub import (
broadcast_evictions,
evict_and_broadcast,
evict_local,
evict_shared,
)
from litellm.proxy.common_utils.http_parsing_utils import ( # noqa: F401 # legacy module exports
_safe_get_request_headers, # pyright: ignore[reportPrivateUsage,reportUnusedImport] # backwards-compatible package export
safe_get_request_headers,
@ -64,6 +69,7 @@ from litellm.proxy.management_endpoints.scim.scim_transformations import (
ScimTransformations,
)
from litellm.proxy.management_endpoints.team_endpoints import (
delete_validated_team_rows,
new_team,
team_member_add,
team_member_delete,
@ -104,6 +110,7 @@ from litellm.types.llms.base import LiteLLMBaseModel
from litellm.types.proxy.management_endpoints.scim_v2 import *
if TYPE_CHECKING:
from prisma import Prisma
from prisma.models import LiteLLM_VerificationToken as PrismaVerificationToken
@ -317,11 +324,15 @@ async def _check_team_exists(team_id: str) -> LiteLLM_TeamTable:
team: Final = await _table(TeamRepository(prisma_client)).find_unique(where={"team_id": team_id})
if not team:
raise HTTPException(status_code=404, detail={"error": f"Group not found with ID: {team_id}"})
raise _group_not_found(team_id)
return team
def _group_not_found(team_id: str) -> HTTPException:
return HTTPException(status_code=404, detail={"error": f"Group not found with ID: {team_id}"})
def _extract_scim_user_data(user: SCIMUser) -> ScimUserData:
"""Extract common data from SCIMUser object."""
user_email = None
@ -482,10 +493,7 @@ async def _recompute_scim_member_roles(prisma_client: PrismaClient, user_ids: It
Recompute and persist each user's global proxy role from their resulting team
membership. No-op unless scim_admin_group is configured, so a SCIM group write
that drops a member from the admin group demotes them just like the user
endpoints do, and the role is left untouched when the feature is off. The users'
rows are locked through ``USER_ROWS_LOCK_SQL`` before their teams are read, the
same lock a roster write takes, so a concurrent group push cannot change a
member's teams between the read and the role write.
endpoints do, and the role is left untouched when the feature is off.
"""
admin_group: Final = await _get_scim_admin_group()
if admin_group is None:
@ -494,28 +502,46 @@ async def _recompute_scim_member_roles(prisma_client: PrismaClient, user_ids: It
ids: Final = tuple(dict.fromkeys(user_ids))
if not ids:
return
default_role: Final = _default_scim_user_role()
async with prisma_client.tx(timeout=ROSTER_TX_TIMEOUT) as tx:
await tx.query_raw(USER_ROWS_LOCK_SQL, sorted(ids))
rows: Final = await find_many_in(user_tx_db(tx), "user_id", ids)
team_ids: Final = tuple(dict.fromkeys(chain.from_iterable(row.teams for row in rows)))
teams: Final = await find_many_in(team_tx_db(tx), "team_id", team_ids)
alias_of: Final = MappingProxyType({team.team_id: team.team_alias for team in teams})
resolved: Final = tuple(
(
row.user_id,
_resolve_scim_user_role(
[SCIMUserGroup(value=team_id, display=alias_of.get(team_id)) for team_id in row.teams],
admin_group,
default_role,
),
)
for row in rows
await _write_scim_member_roles_tx(tx, ids, admin_group=admin_group)
from litellm.proxy.proxy_server import user_api_key_cache
evict_local(cache_keys=ids, user_api_key_cache=user_api_key_cache)
await evict_shared(cache_keys=ids, cache=user_api_key_cache)
await broadcast_evictions(cache_keys=ids)
async def _write_scim_member_roles_tx(tx: "Prisma", user_ids: Sequence[str], *, admin_group: str) -> None:
"""
Write each user's role resolved from the teams they are in, on ``tx``. The users'
rows are locked through ``USER_ROWS_LOCK_SQL`` before their teams are read, the
same lock a roster write takes, so a concurrent group push cannot change a
member's teams between the read and the role write.
"""
if not user_ids:
return
default_role: Final = _default_scim_user_role()
await tx.query_raw(USER_ROWS_LOCK_SQL, sorted(user_ids))
rows: Final = await find_many_in(user_tx_db(tx), "user_id", user_ids)
team_ids: Final = tuple(dict.fromkeys(chain.from_iterable(row.teams for row in rows)))
teams: Final = await find_many_in(team_tx_db(tx), "team_id", team_ids)
alias_of: Final = MappingProxyType({team.team_id: team.team_alias for team in teams})
resolved: Final = tuple(
(
row.user_id,
_resolve_scim_user_role(
[SCIMUserGroup(value=team_id, display=alias_of.get(team_id)) for team_id in row.teams],
admin_group,
default_role,
),
)
for row in rows
)
for role in dict.fromkeys(role for _, role in resolved):
await tx.execute_raw(
_SET_USER_ROLE_SQL, role, sorted(user_id for user_id, user_role in resolved if user_role == role)
)
for role in dict.fromkeys(role for _, role in resolved):
await tx.execute_raw(
_SET_USER_ROLE_SQL, role, sorted(user_id for user_id, user_role in resolved if user_role == role)
)
class _ResolvedUserMember(NamedTuple):
@ -3051,7 +3077,8 @@ async def update_group(
dependencies=[Depends(user_api_key_auth)],
)
async def delete_group(
group_id: str = Path(..., title="Group ID"),
group_id: Annotated[str, Path(title="Group ID")],
user_api_key_dict: Annotated[UserAPIKeyAuth, Depends(user_api_key_auth)],
):
"""
Delete a group according to SCIM v2 protocol
@ -3064,23 +3091,22 @@ async def delete_group(
existing_team: Final = await _check_team_exists(group_id)
raise_if_teams_hold_ptu_shares((group_id,), llm_router)
member_ids: Final = await _get_team_member_user_ids_from_team(existing_team)
# For each member, remove this team from their teams list
for member_id in member_ids:
user = await _table(UserRepository(prisma_client)).find_unique(where={"user_id": member_id})
if user:
current_teams = user.teams or []
if group_id in current_teams:
new_teams = [t for t in current_teams if t != group_id]
await _table(UserRepository(prisma_client)).update(
where={"user_id": member_id}, data={"teams": new_teams}
)
await _recompute_scim_member_roles(prisma_client, member_ids)
# Delete team
await _table(TeamRepository(prisma_client)).delete(where={"team_id": group_id})
admin_group: Final = await _get_scim_admin_group()
deleted: Final = await delete_validated_team_rows(
teams=(LiteLLM_TeamTable.model_validate(existing_team.model_dump()),),
prisma_client=prisma_client,
user_api_key_dict=UserAPIKeyAuth(
user_role=LitellmUserRoles.PROXY_ADMIN,
api_key=user_api_key_dict.api_key,
user_id=user_api_key_dict.user_id,
),
litellm_changed_by=None,
before_commit=(
None if admin_group is None else partial(_write_scim_member_roles_tx, admin_group=admin_group)
),
)
if group_id not in deleted.team_ids:
raise _group_not_found(group_id)
return Response(status_code=204)

View file

@ -14,7 +14,7 @@ import copy
import json
import math
import traceback
from collections.abc import Iterable, Mapping, Sequence
from collections.abc import Awaitable, Callable, Iterable, Iterator, Mapping, Sequence
from collections.abc import Set as AbstractSet
from dataclasses import dataclass
from datetime import datetime, timedelta, timezone
@ -34,7 +34,7 @@ from typing import (
import fastapi
from fastapi import APIRouter, Depends, Header, HTTPException, Request, status
from pydantic import BaseModel, Field, JsonValue, TypeAdapter, ValidationError
from pydantic import BaseModel, ConfigDict, Field, JsonValue, TypeAdapter, ValidationError
from typing_extensions import ReadOnly, TypedDict, assert_never
import litellm
@ -102,6 +102,7 @@ from litellm.proxy.auth.auth_checks import ( # noqa: F401 # legacy module expo
OrganizationNotFoundError,
_cache_team_object, # pyright: ignore[reportPrivateUsage,reportUnusedImport] # backwards-compatible package export
allowed_route_check_inside_route,
await_inflight_team_membership_load,
cache_team_object,
can_org_access_model,
delete_cache_key_objects,
@ -112,6 +113,7 @@ from litellm.proxy.auth.auth_checks import ( # noqa: F401 # legacy module expo
get_team_object,
get_user_object,
invalidate_team_member_spend_state,
jwt_key_mapping_cache_key,
)
from litellm.proxy.auth.auth_utils import (
enforce_batch_limits_are_admin_only,
@ -119,12 +121,19 @@ from litellm.proxy.auth.auth_utils import (
)
from litellm.proxy.auth.user_api_key_auth import user_api_key_auth
from litellm.proxy.common_utils.auth_cache_invalidation_pubsub import (
broadcast_evictions,
evict_and_broadcast,
evict_local,
evict_shared,
publish_auth_cache_invalidation,
)
from litellm.proxy.common_utils.callback_utils import encrypt_callback_vars
from litellm.proxy.common_utils.json_merge_patch import apply_json_merge_patch
from litellm.proxy.common_utils.user_api_key_cache import UserApiKeyCache
from litellm.proxy.common_utils.user_api_key_cache import (
UserApiKeyCache,
team_membership_auth_cache_key,
team_membership_reservation_cache_key,
)
from litellm.proxy.db.db_span import db_span
from litellm.proxy.hooks.key_management_event_hooks import KeyManagementEventHooks
from litellm.proxy.hooks.model_max_budget_limiter import (
@ -207,7 +216,7 @@ from litellm.proxy.management_helpers.utils import (
)
from litellm.proxy.utils import PrismaClient, ProxyLogging, handle_exception_on_proxy
from litellm.repositories.budget_repository import BudgetRepository
from litellm.repositories.chunked_in import find_many_in
from litellm.repositories.chunked_in import delete_many_in, find_many_in
from litellm.repositories.organization_repository import OrganizationRepository
from litellm.repositories.prisma_protocols import TableActions
from litellm.repositories.table_repositories import (
@ -417,6 +426,23 @@ class _TeamDeleteTx(AccessGroupSyncTx, Protocol):
@property
def litellm_teammembership(self) -> "TableActions[prisma_models.LiteLLM_TeamMembership]": ...
@property
def litellm_deletedteamtable(self) -> "TableActions[prisma_models.LiteLLM_DeletedTeamTable]": ...
@property
def litellm_verificationtoken(self) -> "TableActions[prisma_models.LiteLLM_VerificationToken]": ...
@property
def litellm_jwtkeymapping(self) -> "TableActions[prisma_models.LiteLLM_JWTKeyMapping]": ...
def _team_delete_tx(tx: "Prisma") -> _TeamDeleteTx:
return tx # pyright: ignore[reportReturnType] # TableActions widens the generated inputs to Mapping, as the repositories do
_LOCK_DELETED_TEAM_MEMBERS_SQL: Final = """
SELECT user_id, teams FROM "LiteLLM_UserTable" WHERE teams && $1::text[] ORDER BY user_id COLLATE "C" FOR UPDATE
"""
_STRIP_DELETED_TEAMS_FROM_USERS_SQL: Final = """
UPDATE "LiteLLM_UserTable"
@ -430,6 +456,22 @@ WHERE user_id IN (
)
"""
class _DetachedMember(LiteLLMBaseModel):
model_config = ConfigDict(frozen=True)
user_id: str
teams: tuple[str, ...]
_DetachedMembers: Final = TypeAdapter(tuple[_DetachedMember, ...])
@dataclass(frozen=True, slots=True)
class DeletedTeams:
team_ids: tuple[str, ...]
_INCLUDE_MODEL_TABLE: Final = MappingProxyType({"litellm_model_table": True})
@ -4526,18 +4568,7 @@ async def delete_team(
}'
```
"""
from litellm.proxy.management_helpers.audit_logs import (
get_audit_log_changed_by,
is_audit_logging_enabled,
)
from litellm.proxy.proxy_server import (
create_audit_log_for_update,
litellm_proxy_admin_name,
llm_router,
prisma_client,
proxy_logging_obj,
user_api_key_cache,
)
from litellm.proxy.proxy_server import llm_router, prisma_client
if prisma_client is None:
raise HTTPException(status_code=500, detail={"error": "No db connected"})
@ -4567,47 +4598,32 @@ async def delete_team(
raise_if_teams_hold_ptu_shares(data.team_ids, llm_router)
await _persist_deleted_team_records(
await delete_validated_teams(
teams=team_rows,
prisma_client=prisma_client,
user_api_key_dict=user_api_key_dict,
litellm_changed_by=litellm_changed_by,
)
deleted_teams: Final[_DeletedTeamsResult] = {"deleted_teams": data.team_ids}
return deleted_teams
# we do this after the first for loop, since first for loop is for validation. we only want this inserted after validation passes
if is_audit_logging_enabled():
# make an audit log for each team deleted
for team_id in data.team_ids:
team_row: LiteLLM_TeamTable | None = await prisma_client.get_data(
team_id=team_id, table_name="team", query_type="find_unique"
)
if team_row is None:
continue
async def delete_validated_teams(
teams: Sequence[LiteLLM_TeamTable],
prisma_client: PrismaClient,
user_api_key_dict: UserAPIKeyAuth,
litellm_changed_by: str | None,
) -> DeletedTeams:
"""
What ``/team/delete`` does once the teams exist and the caller may delete them: every key
and BYOK model of the teams, then ``delete_validated_team_rows``.
"""
from litellm.proxy.proxy_server import (
proxy_logging_obj,
user_api_key_cache,
)
_team_row = team_row.json(exclude_none=True)
asyncio.create_task(
create_audit_log_for_update(
request_data=LiteLLM_AuditLogs(
id=str(uuid.uuid4()),
updated_at=datetime.now(timezone.utc),
changed_by=get_audit_log_changed_by(
litellm_changed_by=litellm_changed_by,
user_api_key_dict=user_api_key_dict,
litellm_proxy_admin_name=litellm_proxy_admin_name,
),
changed_by_api_key=user_api_key_dict.api_key,
table_name=LitellmTableNames.TEAM_TABLE_NAME,
object_id=team_id,
action="deleted",
updated_values="{}",
before_value=_team_row,
)
)
)
# End of Audit logging
team_ids: Final = [team.team_id for team in teams]
## DELETE ASSOCIATED KEYS
# Fetch keys before deletion to persist them
@ -4615,7 +4631,7 @@ async def delete_team(
persist_deleted_verification_tokens,
)
keys_to_delete: Final = await _tokens_db(prisma_client).find_many(where={"team_id": {"in": data.team_ids}})
keys_to_delete: Final = await _tokens_db(prisma_client).find_many(where={"team_id": {"in": team_ids}})
jwt_mapping_cache_keys: Final = await get_jwt_key_mapping_cache_keys_for_tokens(
hashed_tokens=tuple(key.token for key in keys_to_delete),
prisma_client=prisma_client,
@ -4629,7 +4645,7 @@ async def delete_team(
litellm_changed_by=litellm_changed_by,
)
await prisma_client.delete_data(team_id_list=data.team_ids, table_name="key")
await prisma_client.delete_data(team_id_list=team_ids, table_name="key")
if keys_to_delete:
KeyManagementEventHooks.create_key_deleted_audit_logs(
@ -4654,18 +4670,41 @@ async def delete_team(
from litellm.proxy.proxy_server import llm_router
await delete_team_models(
team_ids=data.team_ids,
team_ids=team_ids,
prisma_client=prisma_client,
llm_router=llm_router,
)
await _sweep_deleted_team_references(team_ids=data.team_ids, prisma_client=prisma_client)
member_ids_per_team: Final = await _resolve_deleted_team_member_user_ids(
teams=team_rows,
return await delete_validated_team_rows(
teams=teams,
prisma_client=prisma_client,
user_api_key_dict=user_api_key_dict,
litellm_changed_by=litellm_changed_by,
)
async def delete_validated_team_rows(
teams: Sequence[LiteLLM_TeamTable],
prisma_client: PrismaClient,
user_api_key_dict: UserAPIKeyAuth,
litellm_changed_by: str | None,
before_commit: Callable[["Prisma", tuple[str, ...]], Awaitable[None]] | None = None,
) -> DeletedTeams:
"""
Under every team's advisory lock, delete the teams still there with the keys their rostered
members hold on them, their membership rows and user references, and write their deleted-team
records; then evict caches and write audit rows. Keys of users off the roster and the teams'
models stay. ``team_ids`` names only the teams this call deleted. ``before_commit`` runs on the
delete's transaction with the members of the rosters read under the lock, so its writes commit
or roll back with the delete.
"""
from litellm.proxy.proxy_server import (
proxy_logging_obj,
user_api_key_cache,
)
team_ids: Final = [team.team_id for team in teams]
## DELETE TEAMS
# Both the delete and the reconcile sweep run under every team's advisory lock
# (TEAM_ADVISORY_LOCK_SQL, the same one /team/member_add takes before its own writes),
@ -4674,14 +4713,28 @@ async def delete_team(
# the lock before this transaction starts, in which case this sweep reaches what it wrote,
# or is still waiting on the lock, in which case its own re-read happens after this commits
# and sees the row gone before it writes anything.
delete_filter: Final[_TeamIdInFilter] = {"team_id": {"in": data.team_ids}}
async with prisma_client.tx(timeout=timedelta(seconds=TEAM_DELETE_TRANSACTION_TIMEOUT_SECONDS)) as tx:
for team_id in sorted(data.team_ids):
tx_budget: Final = timedelta(seconds=TEAM_DELETE_TRANSACTION_TIMEOUT_SECONDS)
async with prisma_client.tx(max_wait=tx_budget, timeout=tx_budget) as tx:
for team_id in sorted(team_ids):
await tx.query_raw(TEAM_ADVISORY_LOCK_SQL, team_id)
await tx.litellm_teamtable.delete_many(where=delete_filter)
await _sweep_deleted_team_references_tx(team_ids=data.team_ids, tx=tx)
team_tx: Final = _team_delete_tx(tx)
locked: Final = await _locked_teams_tx(team_ids=team_ids, tx=team_tx)
member_keys, jwt_mapping_cache_keys = await _delete_roster_member_keys_tx(
locked=locked,
tx=tx,
prisma_client=prisma_client,
user_api_key_dict=user_api_key_dict,
litellm_changed_by=litellm_changed_by,
)
await _delete_locked_teams_tx(
locked=locked, tx=team_tx, user_api_key_dict=user_api_key_dict, litellm_changed_by=litellm_changed_by
)
detached_per_team: Final = await _sweep_deleted_team_references_tx(team_ids=team_ids, tx=team_tx)
if before_commit is not None:
await before_commit(tx, _locked_roster_user_ids(locked))
deleted_teams: Final[_DeletedTeamsResult] = {"deleted_teams": data.team_ids}
rostered_per_team: Final = await _resolve_deleted_team_member_user_ids(teams=locked, prisma_client=prisma_client)
member_ids_per_team: Final = _members_per_deleted_team(rostered_per_team, detached_per_team)
# Evict AFTER the rows are gone. Both writers of these keys (`_cache_team_object` and
# `get_team_object_by_alias`) hydrate from the db, so evicting first leaves a window where a
@ -4689,7 +4742,7 @@ async def delete_team(
# invalidated anything. Nothing fallible runs between the delete and this, or a failure there
# would strand the deleted team in cache.
await _invalidate_deleted_team_cache(
teams=team_rows,
teams=locked,
user_api_key_cache=user_api_key_cache,
proxy_logging_obj=proxy_logging_obj,
)
@ -4697,44 +4750,167 @@ async def delete_team(
member_ids_per_team=member_ids_per_team,
user_api_key_cache=user_api_key_cache,
)
await _invalidate_deleted_key_cache(
keys=member_keys,
user_api_key_cache=user_api_key_cache,
proxy_logging_obj=proxy_logging_obj,
)
await evict_and_broadcast(cache_keys=jwt_mapping_cache_keys, user_api_key_cache=user_api_key_cache)
for deleted_team in team_rows:
_schedule_deleted_team_audit_rows(
teams=locked, user_api_key_dict=user_api_key_dict, litellm_changed_by=litellm_changed_by
)
if member_keys:
KeyManagementEventHooks.create_key_deleted_audit_logs(
keys_being_deleted=member_keys,
user_api_key_dict=user_api_key_dict,
litellm_changed_by=litellm_changed_by,
)
for deleted_team in locked:
_emit_team_members_metric(deleted_team.model_copy(update={"members_with_roles": ()}))
await sync_team_access_group_membership(prisma_client=prisma_client, team_id=deleted_team.team_id)
return deleted_teams
return DeletedTeams(team_ids=tuple(team.team_id for team in locked))
async def _sweep_deleted_team_references(team_ids: Sequence[str], prisma_client: PrismaClient) -> None:
def _locked_roster_user_ids(locked: Sequence[LiteLLM_TeamTable]) -> tuple[str, ...]:
return tuple(dict.fromkeys(chain.from_iterable(_roster_user_ids(team) for team in locked)))
def _members_per_deleted_team(
rostered: Sequence[tuple[str, Sequence[str]]], detached: Sequence[tuple[str, Sequence[str]]]
) -> tuple[tuple[str, tuple[str, ...]], ...]:
detached_by_team: Final = MappingProxyType({team_id: user_ids for team_id, user_ids in detached})
return tuple(
(team_id, tuple(dict.fromkeys((*user_ids, *detached_by_team[team_id])))) for team_id, user_ids in rostered
)
def _roster_user_ids(team: LiteLLM_TeamTable) -> tuple[str, ...]:
return tuple(member.user_id for member in team.members_with_roles if member.user_id is not None)
def _is_roster_member_key(
key: "prisma_models.LiteLLM_VerificationToken", roster_by_team: Mapping[str, frozenset[str]]
) -> bool:
return key.team_id is not None and key.user_id in roster_by_team.get(key.team_id, frozenset())
async def _locked_teams_tx(team_ids: Sequence[str], tx: _TeamDeleteTx) -> tuple[LiteLLM_TeamTable, ...]:
rows: Final = await find_many_in(tx.litellm_teamtable, "team_id", team_ids)
return tuple(LiteLLM_TeamTable.model_validate(row.model_dump()) for row in rows)
async def _delete_roster_member_keys_tx(
locked: Sequence[LiteLLM_TeamTable],
tx: "Prisma",
prisma_client: PrismaClient,
user_api_key_dict: UserAPIKeyAuth,
litellm_changed_by: str | None,
) -> tuple[tuple["prisma_models.LiteLLM_VerificationToken", ...], tuple[str, ...]]:
from litellm.proxy.management_endpoints.key_management_endpoints import (
persist_deleted_verification_tokens,
)
team_tx: Final = _team_delete_tx(tx)
roster_by_team: Final = MappingProxyType({team.team_id: frozenset(_roster_user_ids(team)) for team in locked})
team_keys: Final = await find_many_in(team_tx.litellm_verificationtoken, "team_id", tuple(roster_by_team))
member_keys: Final = tuple(key for key in team_keys if _is_roster_member_key(key, roster_by_team))
if not member_keys:
return (), ()
tokens: Final = tuple(key.token for key in member_keys)
jwt_mappings: Final = await find_many_in(team_tx.litellm_jwtkeymapping, "token", tokens)
await persist_deleted_verification_tokens(
keys=member_keys, # pyright: ignore[reportArgumentType] # generated row model carries the same columns as LiteLLM_VerificationToken
prisma_client=prisma_client,
user_api_key_dict=user_api_key_dict,
litellm_changed_by=litellm_changed_by,
tx=tx,
)
await delete_many_in(team_tx.litellm_verificationtoken, "token", tokens, atomicity="caller_transaction")
return member_keys, tuple(
jwt_key_mapping_cache_key(mapping.jwt_claim_name, mapping.jwt_claim_value, mapping.jwt_issuer)
for mapping in jwt_mappings
)
async def _delete_locked_teams_tx(
locked: Sequence[LiteLLM_TeamTable],
tx: _TeamDeleteTx,
user_api_key_dict: UserAPIKeyAuth,
litellm_changed_by: str | None,
) -> None:
await _persist_deleted_team_records(
teams=locked,
deleted_teams=tx.litellm_deletedteamtable,
user_api_key_dict=user_api_key_dict,
litellm_changed_by=litellm_changed_by,
)
await delete_many_in(
tx.litellm_teamtable,
"team_id",
tuple(team.team_id for team in locked),
atomicity="caller_transaction",
)
def _schedule_deleted_team_audit_rows(
teams: Sequence[LiteLLM_TeamTable],
user_api_key_dict: UserAPIKeyAuth,
litellm_changed_by: str | None,
) -> None:
from litellm.proxy.management_helpers.audit_logs import (
get_audit_log_changed_by,
is_audit_logging_enabled,
)
from litellm.proxy.proxy_server import (
create_audit_log_for_update,
litellm_proxy_admin_name,
)
if not is_audit_logging_enabled():
return
for team in teams:
asyncio.create_task(
create_audit_log_for_update(
request_data=LiteLLM_AuditLogs(
id=str(uuid.uuid4()),
updated_at=datetime.now(timezone.utc),
changed_by=get_audit_log_changed_by(
litellm_changed_by=litellm_changed_by,
user_api_key_dict=user_api_key_dict,
litellm_proxy_admin_name=litellm_proxy_admin_name,
),
changed_by_api_key=user_api_key_dict.api_key,
table_name=LitellmTableNames.TEAM_TABLE_NAME,
object_id=team.team_id,
action="deleted",
updated_values="{}",
before_value=team.model_dump_json(exclude_none=True),
)
)
)
async def _sweep_deleted_team_references_tx(
team_ids: Sequence[str], tx: _TeamDeleteTx
) -> tuple[tuple[str, tuple[str, ...]], ...]:
"""
Strip the deleted team ids from every user row and team-membership row that still references them.
The per-member `team_member_delete` pass above only reaches users listed in the team's
`members_with_roles`, so a user row that outlived its roster entry is invisible to it and keeps
surfacing the team on `/user/info` after the team is gone.
#36839 closed the route that created that drift, by resolving member removal off the roster
entry's `user_id` rather than the identifier the caller happened to pass. It does not backfill
rows that already drifted, which is the state this was reported against, so the sweep still has
to run on delete.
One statement filters the array in place rather than read-filter-write: rewriting the whole
array from a snapshot read outside a transaction drops any team a concurrent `/team/member_add`
appended in between. It locks the user rows in the order `USER_ROWS_LOCK_SQL` takes them, so it
never waits on a SCIM group push in a cycle.
Strip the deleted team ids from every user and team-membership row still referencing them, on
the transaction holding the teams' advisory locks. User rows are locked first in
``USER_ROWS_LOCK_SQL`` order so the sweep never deadlocks with a SCIM group push, and one
statement filters each array in place so a team a concurrent ``/team/member_add`` appended is
kept. Returns, per team, the users read under the lock as referencing it.
"""
_ = await prisma_client.db.execute_raw(_STRIP_DELETED_TEAMS_FROM_USERS_SQL, list(team_ids))
_ = await _team_membership_db(prisma_client).delete_many(where=_TeamIdInFilter(team_id={"in": tuple(team_ids)}))
async def _sweep_deleted_team_references_tx(team_ids: Sequence[str], tx: _TeamDeleteTx) -> None:
"""Same sweep as `_sweep_deleted_team_references`, run on the transaction that holds
every id's advisory lock and deletes the team rows, so it commits or rolls back with them."""
locked: Final = _DetachedMembers.validate_python(await tx.query_raw(_LOCK_DELETED_TEAM_MEMBERS_SQL, list(team_ids)))
_ = await tx.execute_raw(_STRIP_DELETED_TEAMS_FROM_USERS_SQL, list(team_ids))
membership_filter: Final[_TeamIdInFilter] = {"team_id": {"in": tuple(team_ids)}}
_ = await tx.litellm_teammembership.delete_many(where=membership_filter)
return tuple(
(team_id, tuple(member.user_id for member in locked if team_id in member.teams)) for team_id in team_ids
)
async def _invalidate_deleted_key_cache(
@ -4791,17 +4967,20 @@ async def _evict_deleted_team_member_cache(
member_user_ids: Sequence[str],
user_api_key_cache: UserApiKeyCache,
) -> None:
await evict_and_broadcast(cache_keys=tuple(member_user_ids), user_api_key_cache=user_api_key_cache)
await asyncio.gather(
*(
invalidate_team_member_spend_state(
user_id=user_id,
team_id=team_id,
user_api_key_cache=user_api_key_cache,
)
for user_id in member_user_ids
)
_ = await asyncio.gather(
*(await_inflight_team_membership_load(user_id=user_id, team_id=team_id) for user_id in member_user_ids)
)
cache_keys: Final = tuple(member_cache_keys(team_id, member_user_ids))
evict_local(cache_keys=cache_keys, user_api_key_cache=user_api_key_cache)
await evict_shared(cache_keys=cache_keys, cache=user_api_key_cache)
await broadcast_evictions(cache_keys=cache_keys)
def member_cache_keys(team_id: str, user_ids: Sequence[str]) -> Iterator[str]:
for user_id in user_ids:
yield user_id
yield team_membership_auth_cache_key(team_id=team_id, user_id=user_id)
yield team_membership_reservation_cache_key(user_id=user_id, team_id=team_id)
async def _resolve_deleted_team_member_user_ids(
@ -4832,7 +5011,7 @@ async def _deleted_team_member_user_ids(team: LiteLLM_TeamTable, prisma_client:
def _transform_teams_to_deleted_records(
teams: list[LiteLLM_TeamTable],
teams: Sequence[LiteLLM_TeamTable],
user_api_key_dict: UserAPIKeyAuth,
litellm_changed_by: str | None = None,
) -> list[dict[str, object]]:
@ -4881,17 +5060,17 @@ def _transform_teams_to_deleted_records(
async def _save_deleted_team_records(
records: list[dict[str, object]],
prisma_client: PrismaClient,
deleted_teams: "TableActions[prisma_models.LiteLLM_DeletedTeamTable]",
) -> None:
"""Save deleted team records to the database."""
if not records:
return
await _deleted_team_db(prisma_client).create_many(data=records)
await deleted_teams.create_many(data=records)
async def _persist_deleted_team_records(
teams: list[LiteLLM_TeamTable],
prisma_client: PrismaClient,
teams: Sequence[LiteLLM_TeamTable],
deleted_teams: "TableActions[prisma_models.LiteLLM_DeletedTeamTable]",
user_api_key_dict: UserAPIKeyAuth,
litellm_changed_by: str | None = None,
) -> None:
@ -4903,7 +5082,7 @@ async def _persist_deleted_team_records(
)
await _save_deleted_team_records(
records=records,
prisma_client=prisma_client,
deleted_teams=deleted_teams,
)

View file

@ -4,6 +4,7 @@ Functions to create audit logs for LiteLLM Proxy
import asyncio
import json
from collections.abc import Awaitable, Sequence
from datetime import datetime, timezone
from typing import Final, TypeVar
@ -258,12 +259,19 @@ async def create_object_audit_log(
)
async def _insert_audit_log_row(repository: AuditLogRepository, request_data: LiteLLM_AuditLogs) -> None:
async def _insert_audit_logs(insert: Awaitable[object], description: str) -> None:
try:
await repository.table.create(data={**request_data.model_dump(exclude_none=True)})
await insert
except Exception as e:
# [Non-Blocking Exception. Do not allow blocking LLM API call]
verbose_proxy_logger.error("Failed Creating audit log %s", e)
verbose_proxy_logger.error("Failed Creating %s %s", description, e)
def _json_encoded(request_data: LiteLLM_AuditLogs) -> LiteLLM_AuditLogs:
values: Final = (("updated_values", request_data.updated_values), ("before_value", request_data.before_value))
return request_data.model_copy(
update={field: json.dumps(value) for field, value in values if isinstance(value, dict)}
)
async def create_audit_log_for_update(request_data: LiteLLM_AuditLogs):
@ -279,15 +287,10 @@ async def create_audit_log_for_update(request_data: LiteLLM_AuditLogs):
return
verbose_proxy_logger.debug("creating audit log for %s", request_data)
if isinstance(request_data.updated_values, dict):
request_data.updated_values = json.dumps(request_data.updated_values)
if isinstance(request_data.before_value, dict):
request_data.before_value = json.dumps(request_data.before_value)
encoded: Final = _json_encoded(request_data)
# Dispatch to external audit log callbacks regardless of DB availability
await _dispatch_audit_log_to_callbacks(request_data)
await _dispatch_audit_log_to_callbacks(encoded)
if prisma_client is None:
verbose_proxy_logger.error("prisma_client is None, cannot write audit log to DB")
@ -295,7 +298,41 @@ async def create_audit_log_for_update(request_data: LiteLLM_AuditLogs):
await track_audit_task(
asyncio.create_task(
_insert_audit_log_row(AuditLogRepository(prisma_client), request_data),
name=_audit_task_name(request_data, "the database"),
_insert_audit_logs(
AuditLogRepository(prisma_client).table.create(data={**encoded.model_dump(exclude_none=True)}),
"audit log",
),
name=_audit_task_name(encoded, "the database"),
)
)
async def create_audit_logs_for_updates(rows: Sequence[LiteLLM_AuditLogs]) -> None:
"""Create audit logs for many objects with one insert."""
if not rows or not is_audit_logging_enabled():
return
from litellm.proxy.proxy_server import premium_user, prisma_client
if premium_user is not True:
return
encoded: Final = tuple(_json_encoded(row) for row in rows)
for row in encoded:
await _dispatch_audit_log_to_callbacks(row)
if prisma_client is None:
verbose_proxy_logger.error("prisma_client is None, cannot write audit log to DB")
return
await track_audit_task(
asyncio.create_task(
_insert_audit_logs(
AuditLogRepository(prisma_client).table.create_many(
data=tuple(row.model_dump(exclude_none=True) for row in encoded)
),
f"{len(encoded)} audit logs",
),
name=f"{len(encoded)} audit logs to the database",
)
)

View file

@ -0,0 +1,68 @@
import math
from collections.abc import Iterator, Mapping, Sequence
from types import MappingProxyType
from typing import Final, TypeAlias
from pydantic import TypeAdapter
CustomMetadataValue: TypeAlias = str | int | float | bool
_OBJECT_MAPPING_ADAPTER: Final = TypeAdapter(Mapping[object, object])
_OBJECT_LIST_ADAPTER: Final = TypeAdapter(list[object])
def _string_tags(tags: object) -> Iterator[str]:
if isinstance(tags, list):
yield from (tag for tag in _OBJECT_LIST_ADAPTER.validate_python(tags) if isinstance(tag, str))
def _is_non_finite(value: object) -> bool:
return isinstance(value, float) and not math.isfinite(value)
def _scalar_metadata(metadata: object) -> Iterator[tuple[str, CustomMetadataValue]]:
if isinstance(metadata, dict):
yield from (
(key, value)
for key, value in _OBJECT_MAPPING_ADAPTER.validate_python(metadata).items()
if isinstance(key, str) and isinstance(value, (str, int, float, bool)) and not _is_non_finite(value)
)
def _custom_tag_values(model_infos: Sequence[Mapping[object, object]]) -> Iterator[str]:
for model_info in model_infos:
yield from _string_tags(model_info.get("custom_tags"))
def _custom_metadata_entries(
model_infos: Sequence[Mapping[object, object]],
) -> Iterator[tuple[str, CustomMetadataValue]]:
for model_info in model_infos:
yield from _scalar_metadata(model_info.get("custom_metadata"))
def _merged_custom_tags(model_infos: Sequence[Mapping[object, object]]) -> tuple[str, ...] | None:
return tuple(dict.fromkeys(_custom_tag_values(model_infos))) or None
def _merged_custom_metadata(
model_infos: Sequence[Mapping[object, object]],
) -> Mapping[str, CustomMetadataValue] | None:
entries: Final = tuple(_custom_metadata_entries(model_infos))
first_seen: Final = MappingProxyType(dict(reversed(entries)))
return MappingProxyType({key: first_seen[key] for key in dict.fromkeys(key for key, _ in entries)}) or None
def _model_info(deployment: Mapping[str, object]) -> Mapping[object, object]:
model_info: Final = deployment.get("model_info")
return _OBJECT_MAPPING_ADAPTER.validate_python(model_info) if isinstance(model_info, dict) else MappingProxyType({})
def model_group_custom_fields(deployments: Sequence[Mapping[str, object]]) -> Mapping[str, object]:
model_infos: Final = tuple(_model_info(deployment) for deployment in deployments)
return MappingProxyType(
{
"custom_tags": _merged_custom_tags(model_infos),
"custom_metadata": _merged_custom_metadata(model_infos),
}
)

View file

@ -1,6 +1,6 @@
import asyncio
import json
from collections.abc import Iterator, Mapping, Sequence
from collections.abc import Mapping, Sequence
from dataclasses import dataclass
from datetime import timedelta
from types import MappingProxyType
@ -23,8 +23,6 @@ from litellm.proxy.common_utils.auth_cache_invalidation_pubsub import (
)
from litellm.proxy.common_utils.user_api_key_cache import (
UserApiKeyCache,
team_membership_auth_cache_key,
team_membership_reservation_cache_key,
)
from litellm.proxy.hooks.key_management_event_hooks import KeyManagementEventHooks
from litellm.proxy.management_endpoints.key_management_endpoints import (
@ -33,6 +31,7 @@ from litellm.proxy.management_endpoints.key_management_endpoints import (
from litellm.proxy.management_endpoints.team_endpoints import (
_emit_team_members_metric, # pyright: ignore[reportPrivateUsage] # same gauge /team/member_add moves
_schedule_team_membership_audit_log, # pyright: ignore[reportPrivateUsage] # same audit row /team/member_add writes
member_cache_keys,
)
from litellm.proxy.management_helpers.access_group_team_sync import (
DETACH_TEAM_FROM_USERS_SQL,
@ -277,7 +276,7 @@ async def _settle_caches(
evicted: Final = (
*deleted_tokens,
*removal.jwt_mapping_cache_keys,
*_member_cache_keys(team.team_id, changed_user_ids),
*member_cache_keys(team.team_id, changed_user_ids),
)
evict_local(cache_keys=evicted, user_api_key_cache=user_api_key_cache)
await evict_shared(cache_keys=evicted, cache=user_api_key_cache)
@ -292,13 +291,6 @@ async def _settle_caches(
await broadcast_evictions(cache_keys=evicted)
def _member_cache_keys(team_id: str, user_ids: Sequence[str]) -> Iterator[str]:
for user_id in user_ids:
yield user_id
yield team_membership_auth_cache_key(team_id=team_id, user_id=user_id)
yield team_membership_reservation_cache_key(user_id=user_id, team_id=team_id)
async def sync_team_roster(
prisma_client: PrismaClient,
team_id: str,

View file

@ -477,6 +477,18 @@ async def get_authorized_credentials_for_model(
)
def deployment_drop_params(
llm_router: Optional["Router"], model_id: str, request: Mapping[str, object]
) -> Mapping[str, bool | str]:
deployment: Final = (
llm_router.get_credential_deployment(model_id=model_id)
if llm_router is not None and "drop_params" not in request
else None
)
drop_params: Final = deployment.litellm_params.drop_params if deployment is not None else None
return MappingProxyType({}) if drop_params is None else MappingProxyType({"drop_params": drop_params})
def _key_allowlist_covers(public_model_name: str, key_model_allowlist: Sequence[str]) -> bool:
from litellm.proxy.auth.auth_checks import model_matches_any_wildcard_pattern_in_list

View file

@ -76,6 +76,7 @@ from litellm.proxy.openai_files_endpoints.common_utils import ( # noqa: F401 #
add_internal_model_credentials,
apply_team_provider_credentials,
authorize_model_for_key,
deployment_drop_params,
encode_file_id_with_model,
extract_file_creation_params,
get_authorized_credentials_for_model,
@ -457,7 +458,7 @@ async def route_create_file(
# Merge credentials into the request
prepare_data_with_credentials(
data=_create_file_request,
credentials=credentials,
credentials={**credentials, **deployment_drop_params(llm_router, model, _create_file_request)},
)
# Create the file with model credentials

View file

@ -880,8 +880,8 @@ def _build_passthrough_failure_request_payload(
@dataclass(frozen=True, slots=True)
class _TeamCallbackWiring:
success_callbacks: "list[str | Callable | CustomLogger] | None" = None # mutable-ok: Logging.__init__ arg
failure_callbacks: "list[str | Callable | CustomLogger] | None" = None # mutable-ok: Logging.__init__ arg
success_callbacks: "list[str | Callable[..., object] | CustomLogger] | None" = None # mutable-ok: Logging init arg
failure_callbacks: "list[str | Callable[..., object] | CustomLogger] | None" = None # mutable-ok: Logging init arg
logging_kwargs: dict[str, str | dict[str, str]] | None = None # mutable-ok: Logging.__init__ arg

View file

@ -166,6 +166,7 @@ from litellm.proxy.common_utils.token_count_settings import (
resolve_token_count_settings,
)
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,
@ -975,6 +976,7 @@ from litellm.secret_managers.main import (
)
from litellm.tracing.config import is_lens_tracing_enabled
from litellm.types.integrations.slack_alerting import AlertType, SlackAlertingArgs
from litellm.types.litellm_params import STREAM_USAGE_INJECTED_FIELD
from litellm.types.llms.anthropic import (
AnthropicMessagesRequest,
AnthropicResponse,
@ -17747,10 +17749,7 @@ def get_model_group_info(
) -> list[ModelGroupInfoProxy]:
model_groups: Final[list[ModelGroupInfoProxy]] = []
unique_models: Final = []
for model in all_models_str:
if model not in unique_models:
unique_models.append(model)
unique_models: Final = tuple(dict.fromkeys(all_models_str))
for model in unique_models:
if model_group is not None and model_group != model:
@ -17759,7 +17758,18 @@ def get_model_group_info(
_model_group_info = llm_router.get_model_group_info(model_group=model, team_id=team_id)
if _model_group_info is not None:
model_groups.append(ModelGroupInfoProxy.model_validate(_model_group_info.model_dump()))
model_groups.append(
ModelGroupInfoProxy.model_validate(
{
**_model_group_info.model_dump(),
**model_group_custom_fields(
llm_router.get_model_list_of_routed_group(
resolve_model_group_alias(llm_router.model_group_alias, model) or model
)
),
}
)
)
else:
model_group_info = ModelGroupInfoProxy(
model_group=model,
@ -18210,6 +18220,7 @@ async def async_queue_request(
try:
data = await request.json()
data.pop("_litellm_strip_stream_usage", None)
data.pop(STREAM_USAGE_INJECTED_FIELD, None)
# Include original request and headers in the data
data["proxy_server_request"] = {

View file

@ -22,6 +22,7 @@ from litellm.litellm_core_utils.get_model_cost_map import GetModelCostMap
from litellm.proxy._types import (
CommonProxyErrors,
)
from litellm.proxy.db.health_check_latest import latest_by_model_name
from litellm.proxy.utils import get_custom_url
from litellm.repositories.table_repositories import ClaudeCodePluginRepository
from litellm.router_strategy.complexity_router.fuse_presets import FusePresetCatalog, get_fuse_presets
@ -295,13 +296,8 @@ async def public_model_hub():
if prisma_client is not None:
try:
latest_checks: Final = await prisma_client.get_all_latest_health_checks()
for check in latest_checks:
key = check.model_id if check.model_id else check.model_name
if key:
health_check_dict = convert_health_check_to_dict(check)
health_checks_map[key] = health_check_dict
if check.model_name:
health_checks_map[check.model_name] = health_check_dict
for model_name, check in latest_by_model_name(latest_checks).items():
health_checks_map[model_name] = convert_health_check_to_dict(check)
except Exception:
pass

View file

@ -12,6 +12,7 @@ import litellm
from litellm._logging import verbose_proxy_logger
from litellm.proxy._types import CommonProxyErrors, UserAPIKeyAuth
from litellm.proxy.auth.user_api_key_auth import user_api_key_auth
from litellm.proxy.db.health_check_latest import latest_by_model_name
from litellm.proxy.list_api.common import PROBLEM_TYPE_BASE, ManagementProblem
from litellm.proxy.list_api.in_memory import Cells, InMemoryListExecutor
from litellm.proxy.list_api.list_framework import (
@ -59,12 +60,12 @@ class PrismaHealthSnapshotLookup:
checks: Final = await self.prisma_client.get_latest_health_checks_for_models(model_groups)
return MappingProxyType(
{
check.model_name: HealthSnapshot(
model_group: HealthSnapshot(
status=check.status,
response_time_ms=check.response_time_ms,
checked_at=check.checked_at.isoformat() if check.checked_at else None,
)
for check in checks
for model_group, check in latest_by_model_name(checks).items()
}
)

View file

@ -1,8 +1,13 @@
import asyncio
from collections.abc import Mapping
import hashlib
import json
from collections import OrderedDict
from collections.abc import Callable, Mapping
from dataclasses import dataclass
from datetime import datetime
from types import MappingProxyType
from typing import TYPE_CHECKING, Annotated, Any, Final, Literal
from uuid import uuid4
import httpx
from fastapi import HTTPException, status
@ -10,6 +15,7 @@ from pydantic import BeforeValidator, TypeAdapter, ValidationError
import litellm
from litellm.litellm_core_utils.core_helpers import normalize_drop_params
from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
from litellm.proxy._types import ProxyException, UserAPIKeyAuth
from litellm.router_utils.common_utils import is_proxy_admin_request
@ -54,7 +60,34 @@ _ROUTER_ONLY_SETTINGS: Final = frozenset(
)
def _route_user_config_request(data: dict, route_type: str):
class UserConfigRouters:
def __init__(self, max_size: int) -> None:
self._max_size: Final = max_size
self._routers: Final[OrderedDict[str, LitellmRouter]] = OrderedDict() # mutable-ok: LRU evicts in place
def get(self, config: Mapping[str, object], build: Callable[[], LitellmRouter]) -> LitellmRouter:
key: Final = hashlib.sha256(json.dumps(config, sort_keys=True, default=str).encode()).hexdigest()
cached: Final = self._routers.get(key)
if cached is not None:
self._routers.move_to_end(key)
return cached
router: Final = build()
router.serve_through_request_logging()
self._routers[key] = router
if len(self._routers) > self._max_size:
self._routers.popitem(last=False)[1].discard()
return router
def clear(self) -> None:
for router in self._routers.values():
router.discard()
self._routers.clear()
user_config_routers: Final = UserConfigRouters(max_size=32)
async def _route_user_config_request(data: dict, route_type: str):
"""Route a request using the user-provided router config."""
router_config: Final = data.pop("user_config")
@ -63,10 +96,24 @@ def _route_user_config_request(data: dict, route_type: str):
valid_args: Final = litellm.Router.get_valid_args()
filtered_config: Final = {k: v for k, v in router_config.items() if k in valid_args}
user_router: Final = litellm.Router(**filtered_config)
ret_val: Final = getattr(user_router, f"{route_type}")(**data)
user_router.discard()
return ret_val
logging_obj: Final = _request_logging_obj(JSON_OBJECT_ADAPTER.validate_python(data), route_type)
cached_router: Final = user_config_routers.get(filtered_config, lambda: litellm.Router(**filtered_config))
cached_router.bind_to_request_logging(logging_obj)
return await getattr(cached_router, f"{route_type}")(**{**data, "litellm_logging_obj": logging_obj})
def _request_logging_obj(data: Mapping[str, object], route_type: str) -> LiteLLMLoggingObj:
arrived_with: Final = data.get("litellm_logging_obj")
if isinstance(arrived_with, LiteLLMLoggingObj):
return arrived_with
logging_obj, _ = litellm.utils.function_setup(
original_function=route_type,
rules_obj=litellm.utils.Rules(),
start_time=datetime.now(), # noqa: DTZ005 # Logging subtracts it from naive end times
is_async_call=True,
**{"litellm_call_id": str(uuid4()), **data},
)
return logging_obj
def _is_a2a_agent_model(model_name: object) -> bool:
@ -111,6 +158,7 @@ ROUTE_ENDPOINT_MAPPING: Final = {
"asearch": "/search",
"adecisions": "/decisions",
"asystemone": "/systemone",
"aextraction": "/extraction",
"avideo_generation": "/videos",
"avideo_list": "/videos",
"avideo_status": "/videos/{video_id}",
@ -573,6 +621,7 @@ RouteType = Literal[
"asearch",
"adecisions",
"asystemone",
"aextraction",
"avideo_generation",
"avideo_list",
"avideo_status",

View file

@ -15,6 +15,7 @@ from pydantic import BaseModel, JsonValue, TypeAdapter, ValidationError
import litellm
from litellm._logging import verbose_proxy_logger
from litellm.constants import (
BILLED_RETRY_ATTEMPT_METADATA_KEY,
CALLER_SESSION_ID_METADATA_KEY,
CLI_SESSION_KEY_PREFIX,
EMPTY_MAPPING,
@ -187,6 +188,7 @@ _STAMPED_METADATA_KEYS: Final = frozenset(
"used_client_oauth_token",
KUBERNETES_POD_ROUTING_KEY,
"litellm_roi_estimator",
BILLED_RETRY_ATTEMPT_METADATA_KEY,
"user_api_key_user_email",
)
)
@ -249,6 +251,7 @@ def _get_spend_logs_metadata(
guardrail_information=None,
internal_call_origin=None,
litellm_roi_estimator=False,
billed_retry_attempt=False,
eval_information=None,
cold_storage_object_key=cold_storage_object_key,
litellm_overhead_time_ms=None,
@ -270,6 +273,8 @@ def _get_spend_logs_metadata(
)
verbose_proxy_logger.debug("getting payload for SpendLogs, available keys in metadata: %s", [*metadata])
retry_key: Final = BILLED_RETRY_ATTEMPT_METADATA_KEY
billed_retry_attempt: Final = retry_key in metadata and metadata[retry_key] is True
# Filter the metadata dictionary to include only the specified keys
clean_metadata: Final = SpendLogsMetadata(
**MappingProxyType(
@ -283,6 +288,7 @@ def _get_spend_logs_metadata(
used_client_oauth_token=used_client_oauth_token,
kubernetes_pod_routing=kubernetes_pod_routing,
litellm_roi_estimator=metadata.get("litellm_roi_estimator") is True,
billed_retry_attempt=billed_retry_attempt,
user_api_key_user_email=_stored_user_email(metadata),
)
_raw_key: Final = clean_metadata.get("user_api_key")
@ -453,7 +459,12 @@ def _sl_attribution_fallback(
def _deployment_provider(deployment: DeploymentTypedDict) -> str | None:
litellm_params: Final = LiteLLM_Params.model_validate(deployment["litellm_params"])
litellm_params: Final[LiteLLM_Params] = LiteLLM_Params.model_validate(deployment["litellm_params"])
return deployment_params_provider(litellm_params)
def deployment_params_provider(litellm_params: LiteLLM_Params) -> str | None:
"""The provider a deployment's ``litellm_params`` authenticate against, None when its model cannot be resolved."""
if litellm.LiteLLMProxyChatConfig.should_use_litellm_proxy_by_default(litellm_params=litellm_params):
return LlmProviders.LITELLM_PROXY.value
declared: Final = declared_authenticating_provider(litellm_params.model, litellm_params.custom_llm_provider)

View file

@ -158,7 +158,7 @@ from litellm.litellm_core_utils.core_helpers import (
)
from litellm.litellm_core_utils.duration_parser import get_budget_window_start
from litellm.litellm_core_utils.env_utils import get_env_float, get_env_int
from litellm.litellm_core_utils.litellm_logging import Logging, deployment_pricing_model_info
from litellm.litellm_core_utils.litellm_logging import BilledRoundsDispatch, Logging, deployment_pricing_model_info
from litellm.litellm_core_utils.llm_cost_calc.utils import generic_cost_per_token
from litellm.litellm_core_utils.safe_json_dumps import safe_dumps
from litellm.litellm_core_utils.safe_json_loads import safe_json_loads
@ -1161,6 +1161,43 @@ def _timed_out_earlier_records(
return tuple(record for identity, record in by_identity.items() if identity != final_identity)
def _record_pricing(
model_call_details: Mapping[str, object], record: Mapping[str, object], base_model: str | None
) -> tuple[Mapping[str, object], str | None]:
"""The ``(model_call_details, base_model)`` to price one earlier attempt with: the record's own
deployment model and ``base_model`` when it served a different model than the final attempt
(the provider is then inferred from that model), else the final attempt's."""
record_model: Final = record.get("model")
if not isinstance(record_model, str) or record_model == model_call_details.get("model"):
return model_call_details, base_model
record_base_model: Final = record.get("base_model")
return (
{**model_call_details, "model": record_model, "custom_llm_provider": None},
record_base_model if isinstance(record_base_model, str) else None,
)
def _billed_earlier_attempt(
model_call_details: Mapping[str, object],
record: Mapping[str, object],
index: int,
estimated_usage: Usage,
base_model: str | None,
) -> BilledFailedAttempt:
pricing_details, pricing_base_model = _record_pricing(model_call_details, record, base_model)
return BilledFailedAttempt(
attempted_retries=_attempt_number(record, index),
deployment_id=_record_deployment_id(record),
usage=estimated_usage,
response_cost=_estimated_input_cost(
pricing_details,
estimated_usage.prompt_tokens,
deployment_id=_record_deployment_id(record),
base_model=pricing_base_model,
),
)
def _billed_earlier_attempts(
model_call_details: Mapping[str, object],
litellm_params: Mapping[str, object] | None,
@ -1174,23 +1211,14 @@ def _billed_earlier_attempts(
request's own deployment and retry count (``attempted_retries`` restarts at 0 after a fallback
hop, so the count alone cannot tell the groups apart). An earlier attempt that ended in a
``Timeout`` had the same prompt accepted by its deployment, so it is billed like the final
attempt, at that deployment's input rate (else the cost map's rate for ``base_model`` or the
model). Only the exception class survives on the record, so an earlier connect timeout is
billed too; any other failure class is not."""
attempt, at that deployment's input rate (else the cost map's rate for the record's own
``base_model`` or model, which a fallback hop can make a different model than the final one).
Only the exception class survives on the record, so an earlier connect timeout is billed too;
any other failure class is not."""
final_attempt: Final = _retry_metadata_bucket(litellm_params).get("attempted_retries")
final_identity: Final = (deployment_id, final_attempt if isinstance(final_attempt, int) else None)
return tuple(
BilledFailedAttempt(
attempted_retries=_attempt_number(record, index),
deployment_id=_record_deployment_id(record),
usage=estimated_usage,
response_cost=_estimated_input_cost(
model_call_details,
estimated_usage.prompt_tokens,
deployment_id=_record_deployment_id(record),
base_model=base_model,
),
)
_billed_earlier_attempt(model_call_details, record, index, estimated_usage, base_model)
for index, record in enumerate(_timed_out_earlier_records(litellm_params, final_identity))
)
@ -4704,19 +4732,25 @@ class ProxyLogging:
@staticmethod
def _discard_deferred_stream_logging_for_failure(request_data: Mapping[str, object], error: Exception) -> bool:
"""Drop the parked success dispatch for an assembled chat stream that ends in an error
``post_call_failure_hook`` logs as a failure, billing its usage on the failure row instead.
"""Drop the parked success dispatch for an assembled chat stream or a stream that billed several rounds when it
ends in an error ``post_call_failure_hook`` logs as a failure, billing its usage on the failure row instead.
Returns False when the parked dispatch should still be flushed by the caller."""
logging_obj: Final = request_data.get("litellm_logging_obj")
if not isinstance(logging_obj, Logging):
return False
_args: Final[tuple[object, ...] | None] = getattr(logging_obj, "_deferred_stream_complete_args", None)
assembled: Final = _args[0] if _args else None
if not isinstance(error, _PROXY_ONLY_LLM_API_ERRORS) or not isinstance(assembled, ModelResponse):
if not isinstance(error, _PROXY_ONLY_LLM_API_ERRORS):
return False
match _args:
case (ModelResponse() as assembled, *_):
logging_obj.record_assembled_response_for_failure(assembled)
case (BilledRoundsDispatch() as billed,):
billed.dispatch.close()
logging_obj.record_billed_rounds_for_failure(billed.spend)
case _:
return False
logging_obj._on_deferred_stream_complete = None
logging_obj._deferred_stream_complete_args = None
logging_obj.record_assembled_response_for_failure(assembled)
return True
async def arelease_max_parallel_requests_on_disconnect(
@ -4804,6 +4838,7 @@ def jsonify_object(data: dict) -> dict:
_deprecated_key_cache: Final[LimitedSizeOrderedDict] = LimitedSizeOrderedDict(max_size=1000)
_DEPRECATED_KEY_CACHE_TTL_SECONDS: Final = 60
_PRISMA_DEFAULT_TX_TIMEOUT: Final = timedelta(seconds=5)
_PRISMA_DEFAULT_TX_MAX_WAIT: Final = timedelta(seconds=2)
async def _lookup_deprecated_key(
@ -5210,13 +5245,17 @@ class PrismaClient:
return self.db.read_target
return self.db
def tx(self, *, timeout: timedelta = _PRISMA_DEFAULT_TX_TIMEOUT) -> "TransactionManager":
def tx(
self, *, timeout: timedelta = _PRISMA_DEFAULT_TX_TIMEOUT, max_wait: timedelta = _PRISMA_DEFAULT_TX_MAX_WAIT
) -> "TransactionManager":
"""Open an interactive transaction on the writer.
Callers go through this instead of reaching into ``self.db`` so writer
selection and read-replica routing stay encapsulated in the wrapper.
"""
return cast("TransactionManager", self.db.tx(timeout=timeout)) # cast-ok: untyped __getattr__ delegate
return cast( # cast-ok: untyped __getattr__ delegate
"TransactionManager", self.db.tx(max_wait=max_wait, timeout=timeout)
)
def get_request_status(self, payload: dict | SpendLogsPayload) -> Literal["success", "failure"]:
"""

View file

@ -5,7 +5,8 @@ from itertools import filterfalse
from typing import Any, Final, cast
import litellm
from litellm.litellm_core_utils.hidden_params import get_or_create_hidden_params
from litellm.litellm_core_utils.hidden_params import get_or_create_hidden_params, served_from_cache
from litellm.litellm_core_utils.litellm_logging import BilledRoundsSpend
from litellm.main import stream_chunk_builder
from litellm.responses.litellm_completion_transformation.custom_tools import (
build_tool_call_item_kwargs,
@ -1364,6 +1365,17 @@ class LiteLLMCompletionStreamingIterator(ResponsesAPIStreamingIterator):
return False
return not any(getattr(part, "text", None) for part in getattr(item, "content", None) or ())
def finished_round_event(self) -> ResponseCompletedEvent | ResponseIncompleteEvent | None:
if not isinstance(self.litellm_model_response, ModelResponse):
return None
finished: Final = self._emit_terminal_response_event(self.litellm_model_response)
if finished is not None and served_from_cache(self.litellm_custom_stream_wrapper):
finished.response.hidden_params = {**finished.response.hidden_params, "cache_hit": True}
return finished
def bill_failure_after_earlier_rounds(self, earlier: BilledRoundsSpend) -> None:
self.litellm_custom_stream_wrapper.bill_failure_after_earlier_rounds(earlier)
def partial_response_on_disconnect(self) -> ResponseIncompleteEvent | None:
"""
The response.incomplete event to bill when the client went away mid-stream.

View file

@ -3,7 +3,9 @@ from typing import TYPE_CHECKING, Any, Final, cast
from litellm._logging import verbose_logger
from litellm._uuid import uuid
from litellm.responses.litellm_completion_transformation.streaming_iterator import LiteLLMCompletionStreamingIterator
from litellm.responses.mcp.request_context import MCPRequestContext
from litellm.responses.mcp.stream_round_billing import StreamedRoundBilling
from litellm.responses.streaming_iterator import BaseResponsesAPIStreamingIterator
from litellm.types.llms.openai import (
BaseLiteLLMOpenAIResponseObject,
@ -352,6 +354,9 @@ class MCPEnhancedStreamingIterator(BaseResponsesAPIStreamingIterator):
self._round_max_output_index = -1
self._composed_output: list[object] = [] # mutable-ok: grows as each round finishes
self._pending_mcp_call_items: list[dict[str, object]] = [] # mutable-ok: grows per executed tool
self._iteration_started = False
self._round_billing: StreamedRoundBilling | None = None
self._bridged_round_billed: LiteLLMCompletionStreamingIterator | None = None
def _extract_mcp_headers_from_params(self) -> None:
"""Extract MCP headers from original request params to pass to tool calls"""
@ -436,7 +441,15 @@ class MCPEnhancedStreamingIterator(BaseResponsesAPIStreamingIterator):
return self
async def __anext__(self) -> ResponsesAPIStreamingResponse:
chunk: Final = await self._anext_impl()
if not self._iteration_started:
self._iteration_started = True
self._round_billing = self._bill_rounds_on_one_row()
try:
chunk: Final = await self._anext_impl()
except Exception:
if self._round_billing is not None:
self._round_billing.finish()
raise
sequence_number: Final = getattr(chunk, "sequence_number", None)
if isinstance(sequence_number, int):
if sequence_number <= self._last_sequence_number and self._last_sequence_number > 0:
@ -446,6 +459,49 @@ class MCPEnhancedStreamingIterator(BaseResponsesAPIStreamingIterator):
self._last_sequence_number = max(self._last_sequence_number, sequence_number)
return chunk
def __del__(self) -> None:
round_billing: Final[object] = getattr(self, "_round_billing", None)
if isinstance(round_billing, StreamedRoundBilling):
round_billing.release_abandoned()
def _bill_rounds_on_one_row(self) -> StreamedRoundBilling | None:
if not self.should_auto_execute:
return None
if isinstance(self.base_iterator, LiteLLMCompletionStreamingIterator):
return StreamedRoundBilling(self.base_iterator.litellm_logging_obj)
if isinstance(self.base_iterator, BaseResponsesAPIStreamingIterator):
return StreamedRoundBilling(self.base_iterator.logging_obj)
return None
def _restore_round_headers(self, logged_event: ResponsesAPIStreamingResponse) -> None:
if isinstance(self.base_iterator, LiteLLMCompletionStreamingIterator):
return
if isinstance(self.base_iterator, BaseResponsesAPIStreamingIterator):
self.base_iterator._restore_provider_response_headers(logged_event)
def _round_billed_early(self) -> bool:
return self._bridged_round_billed is not None and self._bridged_round_billed is self.base_iterator
def _bill_finished_bridged_round(self) -> None:
base: Final = self.base_iterator
if self._round_billing is None or not isinstance(base, LiteLLMCompletionStreamingIterator):
return
if self._round_billed_early():
return
finished: Final = base.finished_round_event()
if finished is None:
return
self._bridged_round_billed = base
round_output: Final = _output_items(finished.response)
self._round_billing.complete_round(
finished,
finished.response,
round_output
if finished.type == ResponsesAPIStreamEvents.RESPONSE_INCOMPLETE
else (*self._composed_output, *round_output),
self._restore_round_headers,
)
async def _anext_impl(self) -> ResponsesAPIStreamingResponse:
"""
Phase-based streaming:
@ -617,6 +673,7 @@ class MCPEnhancedStreamingIterator(BaseResponsesAPIStreamingIterator):
follow-up rounds past the items already emitted, and lists every
round's items on the final response.completed.
"""
self._bill_finished_bridged_round()
chunk_type: Final[object] = getattr(chunk, "type", None)
if self._round_index > 0 and chunk_type in (
ResponsesAPIStreamEvents.RESPONSE_CREATED,
@ -630,12 +687,21 @@ class MCPEnhancedStreamingIterator(BaseResponsesAPIStreamingIterator):
if self._output_index_offset:
_set_event_field(chunk, "output_index", output_index + self._output_index_offset)
if self.should_auto_execute and chunk_type == ResponsesAPIStreamEvents.RESPONSE_INCOMPLETE:
return self._charge_incomplete_round(chunk)
if not (self.should_auto_execute and self._is_response_completed(chunk)):
return chunk
response_obj: Final[object] = getattr(chunk, "response", None)
if isinstance(response_obj, ResponsesAPIResponse):
self.collected_response = response_obj
if self._round_billing is not None and not self._round_billed_early():
self._round_billing.complete_round(
chunk,
response_obj,
(*self._composed_output, *_output_items(response_obj)),
self._restore_round_headers,
)
# Move to tool execution phase after this chunk
self.phase = "tool_execution"
await self._generate_tool_execution_events()
@ -645,13 +711,33 @@ class MCPEnhancedStreamingIterator(BaseResponsesAPIStreamingIterator):
if self._follow_up_pending():
self._absorb_round(response_obj)
return None
if self._composed_output:
merged_output: Final[list[object]] = [ # mutable-ok: the response model declares output as a list
*self._composed_output,
*_output_items(response_obj),
]
merged_response: Final = response_obj.model_copy(update={"output": merged_output})
_set_event_field(chunk, "response", merged_response)
merged_response: Final = (
response_obj.model_copy(update={"output": [*self._composed_output, *_output_items(response_obj)]})
if self._composed_output
else response_obj
)
client_response: Final = (
merged_response
if self._round_billing is None
else self._round_billing.charge_final_round(chunk, merged_response, self._restore_round_headers)
)
if client_response is not response_obj:
_set_event_field(chunk, "response", client_response)
return chunk
def _charge_incomplete_round(self, chunk: ResponsesAPIStreamingResponse) -> ResponsesAPIStreamingResponse:
response_obj: Final[object] = getattr(chunk, "response", None)
if self._round_billing is None or not isinstance(response_obj, ResponsesAPIResponse):
return chunk
if not self._round_billed_early():
self._round_billing.complete_round(
chunk, response_obj, _output_items(response_obj), self._restore_round_headers
)
client_response: Final = self._round_billing.charge_final_round(
chunk, response_obj, self._restore_round_headers
)
if client_response is not response_obj:
_set_event_field(chunk, "response", client_response)
return chunk
async def _process_base_iterator_chunk(self) -> ResponsesAPIStreamingResponse:
@ -949,6 +1035,8 @@ class MCPEnhancedStreamingIterator(BaseResponsesAPIStreamingIterator):
self.collected_response = None
self._cached_response_id = None
self._round_index += 1
if self._round_billing is not None:
self._round_billing.record_billed_rounds(follow_up_response)
except Exception as e:
verbose_logger.error("Error creating follow-up iterator: %s", e)

View file

@ -4,7 +4,7 @@ from contextlib import contextmanager
from types import MappingProxyType
from typing import Final, TypeVar
from pydantic import TypeAdapter
from pydantic import BaseModel, TypeAdapter
from pydantic_core import to_jsonable_python
from litellm.litellm_core_utils.hidden_params import get_hidden_params, served_from_cache, set_hidden_params
@ -46,15 +46,18 @@ def _summed_mapping(first: Mapping[str, object], final: Mapping[str, object]) ->
return MappingProxyType({key: _summed_entry(key, first, final) for key in keys})
def _summed_usage(first: BaseModel, final: BaseModel) -> object:
return to_jsonable_python(_summed_mapping(first.model_dump(), final.model_dump()), fallback=dict)
def merged_round_usage(first: ResponseAPIUsage | None, final: ResponseAPIUsage | None) -> ResponseAPIUsage | None:
if first is None or final is None:
return final if first is None else first
return ResponseAPIUsage.model_validate(_summed_mapping(first.model_dump(), final.model_dump()))
return ResponseAPIUsage.model_validate(_summed_usage(first, final))
def _merged_chat_usage(first: Usage, final: Usage) -> Usage:
summed: Final = _summed_mapping(first.model_dump(), final.model_dump())
return Usage.model_validate(to_jsonable_python(summed, fallback=dict))
def merged_chat_usage(first: Usage, final: Usage) -> Usage:
return Usage.model_validate(_summed_usage(first, final))
def _chat_usage(round_response: ResponsesAPIResponse | ModelResponse) -> Usage | None:
@ -82,7 +85,7 @@ def _usage_of_both_rounds(
final_usage: Final = _chat_usage(final)
if first_usage is None or final_usage is None:
return final_usage if first_usage is None else first_usage
return _merged_chat_usage(first_usage, final_usage)
return merged_chat_usage(first_usage, final_usage)
def _repeated_fixed_margin(
@ -105,6 +108,14 @@ def _less_repeated_fixed_margin(summed: Mapping[str, object], repeated: float) -
return MappingProxyType({**summed, **charged_once})
def breakdown_less_repeated_fixed_margin(
breakdown: CostBreakdown | None, repeated_fixed_margin: float
) -> CostBreakdown | None:
if breakdown is None:
return None
return _COST_BREAKDOWN.validate_python(_less_repeated_fixed_margin(breakdown, repeated_fixed_margin))
def summed_cost_breakdown(
first: ResponsesAPIResponse | ModelResponse,
first_breakdown: CostBreakdown | None,

View file

@ -0,0 +1,376 @@
import asyncio
from collections.abc import Callable, Coroutine, Sequence
from dataclasses import dataclass
from datetime import datetime
from typing import Final, Protocol, TypeAlias
from litellm.litellm_core_utils.hidden_params import get_hidden_params, served_from_cache
from litellm.litellm_core_utils.litellm_logging import BilledRoundsDispatch, BilledRoundsSpend, UncachedPrice
from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
from litellm.responses.mcp.round_billing import billed_for_every_round, merged_chat_usage, summed_cost_breakdown
from litellm.responses.streaming_iterator import BaseResponsesAPIStreamingIterator, spawn_logging_task
from litellm.responses.utils import ResponseAPILoggingUtils
from litellm.types.llms.openai import ResponsesAPIResponse, ResponsesAPIStreamingResponse
from litellm.types.utils import CostBreakdown, ModelResponse, Usage
Parked: TypeAlias = tuple[object, ...]
RestoresHeaders: TypeAlias = Callable[[ResponsesAPIStreamingResponse], None]
_PARKED_DISPATCH: Final = "_deferred_stream_complete_args"
_DISPATCH_RELEASE: Final = "_on_deferred_stream_complete"
_SERVED_FROM_CACHE: Final = UncachedPrice(cost=0.0, cost_breakdown=None)
class _DispatchesSuccessHandlers(Protocol):
async def dispatch_success_handlers(
self,
result: object = None,
start_time: object = None,
end_time: object = None,
cache_hit: object = None,
prefer_async_handlers: bool = False,
) -> None: ...
def _success_dispatcher(logger: LiteLLMLoggingObj) -> _DispatchesSuccessHandlers:
return logger
def _event_loop_running() -> bool:
try:
asyncio.get_running_loop()
except RuntimeError:
return False
return True
def _priced(response: ResponsesAPIResponse, price: UncachedPrice) -> ResponsesAPIResponse:
priced: Final = response.model_copy()
priced.hidden_params = {**response.hidden_params, "response_cost": price.cost}
return priced
def _assembled(parked: Parked | None) -> ModelResponse | None:
match parked:
case (ModelResponse() as assembled, bool()):
return assembled
case _:
return None
def _bridged_round_usage(parked: Parked | None) -> Usage | None:
usage: Final[object] = getattr(_assembled(parked), "usage", None)
return usage if isinstance(usage, Usage) else None
def _usage_of_bridged_rounds(billed: Usage | None, round_usage: Usage | None, cost: object) -> Usage | None:
if billed is None or round_usage is None:
return None
return merged_chat_usage(billed, round_usage).model_copy(update={"cost": cost})
def _logged_cache_hit(response: ResponsesAPIResponse, chat_usage: Usage | None) -> bool | None:
if chat_usage is not None:
return served_from_cache(response)
return served_from_cache(response) or None
@dataclass(frozen=True, slots=True)
class _RoundRow:
logged: ResponsesAPIStreamingResponse | ModelResponse
cache_hit: bool | None
end_time: datetime
@dataclass(frozen=True, slots=True)
class _SuccessLog:
logger: LiteLLMLoggingObj
row: _RoundRow
chat_usage: Usage | None
def carry_bridged_usage(self) -> None:
if self.chat_usage is not None:
self.logger.model_call_details["combined_usage_object"] = self.chat_usage
def _round_row(
completed: ResponsesAPIStreamingResponse,
response: ResponsesAPIResponse,
chat_usage: Usage | None,
restore_headers: RestoresHeaders,
) -> _RoundRow:
logged_response: Final = response.model_copy()
logged_response.hidden_params = {**response.hidden_params}
logged: Final = completed.model_copy(update={"response": logged_response})
restore_headers(logged)
return _RoundRow(
logged,
_logged_cache_hit(response, chat_usage),
datetime.now(), # noqa: DTZ005 # naive to match the logging pipeline start_time
)
@dataclass(frozen=True, slots=True)
class _BilledRounds:
response: ResponsesAPIResponse
breakdown: CostBreakdown | None
chat_usage: Usage | None
rounds: int = 1
@property
def stamped_cost(self) -> float | None:
cost: Final = self.response.hidden_params.get("response_cost")
return float(cost) if isinstance(cost, (int, float)) else None
@property
def cost(self) -> float:
return self.stamped_cost or 0.0
@property
def spend(self) -> BilledRoundsSpend:
usage: Final = (
ResponseAPILoggingUtils.transform_response_api_usage_to_chat_usage(self.response.usage)
if self.chat_usage is None
else self.chat_usage
)
return BilledRoundsSpend(usage, self.cost, self.breakdown)
def with_round(
self, priced: ResponsesAPIResponse, breakdown: CostBreakdown | None, chat_usage: Usage | None
) -> "_BilledRounds":
merged: Final = billed_for_every_round(self.response, self.breakdown, priced, breakdown)
total: Final = merged.hidden_params.get("response_cost")
charged: Final = merged.model_copy(
update={"usage": None if merged.usage is None else merged.usage.model_copy(update={"cost": total})}
)
charged.hidden_params = {
**merged.hidden_params,
"cache_hit": served_from_cache(self.response) and served_from_cache(priced),
}
return _BilledRounds(
charged,
summed_cost_breakdown(self.response, self.breakdown, priced, breakdown),
_usage_of_bridged_rounds(self.chat_usage, chat_usage, total),
self.rounds + 1,
)
def row(
self,
completed: ResponsesAPIStreamingResponse,
client_output: Sequence[object],
restore_headers: RestoresHeaders,
) -> _RoundRow:
return _round_row(
completed,
self.response.model_copy(update={"output": list(client_output)}),
self.chat_usage,
restore_headers,
)
def bridged_row(self, assembled: ModelResponse | None, logged_id: str) -> _RoundRow | None:
if assembled is None or self.chat_usage is None:
return None
logged: Final = assembled.model_copy(update={"id": logged_id, "usage": self.chat_usage})
logged.hidden_params = {
**(get_hidden_params(assembled) or {}),
"response_cost": self.stamped_cost,
}
return _RoundRow(
logged,
served_from_cache(self.response),
datetime.now(), # noqa: DTZ005 # naive to match the logging pipeline start_time
)
class StreamedRoundBilling:
"""One success row per streamed auto-execute request, billed for every round it ran.
Each round's iterator parks its success dispatch on the shared logging object rather than sending it, either as a
ready dispatch or, for a stream bridged through chat completions, as the assembled response and its cache hit.
When a round completes, before its tools run, it is folded into the rounds before it and a dispatch for every
round so far stays parked, so a client disconnect or a later round that fails still bills them. Once a later round
starts, that dispatch logs from a copy of the logging state taken then, so the later round's failure, logged on the
request's own state where the proxy reads its failure row, never reaches their success row. Only the latest
dispatch is sent, through the proxy's deferred release when the proxy armed one."""
def __init__(self, logging_obj: LiteLLMLoggingObj) -> None:
self._logging_obj: Final = logging_obj
self._billed: _BilledRounds | None = None
self._banked: Parked | None = None
self._banked_row: _RoundRow | None = None
if getattr(logging_obj, _DISPATCH_RELEASE, None) is None:
setattr(logging_obj, _DISPATCH_RELEASE, self._run_parked)
def complete_round(
self,
completed: ResponsesAPIStreamingResponse,
response: ResponsesAPIResponse,
client_output: Sequence[object],
restore_headers: RestoresHeaders,
) -> None:
round_parked: Final = self._take_parked()
price: Final = self._price(response, round_parked)
priced: Final = _priced(response, price)
chat_usage: Final = _bridged_round_usage(round_parked)
if self._billed is None:
first_round: Final = _BilledRounds(priced, price.cost_breakdown, chat_usage)
self._billed = first_round
self._bank(
round_parked,
first_round.bridged_row(_assembled(round_parked), priced.id)
or first_round.row(completed, client_output, restore_headers),
)
return
billed: Final = self._billed.with_round(priced, price.cost_breakdown, chat_usage)
self._close(round_parked)
self._close(self._banked)
self._logging_obj.model_call_details.pop("combined_usage_object", None)
self._logging_obj.model_call_details.pop("response_cost", None)
self._billed = billed
self._logging_obj.cost_breakdown = billed.breakdown
row: Final = billed.bridged_row(_assembled(round_parked), billed.response.id) or billed.row(
completed, client_output, restore_headers
)
self._bank(None if round_parked is None else (self._dispatch(row, billed, self._logging_obj),), row)
def record_billed_rounds(self, follow_up: object) -> None:
if self._billed is None or not isinstance(follow_up, BaseResponsesAPIStreamingIterator):
return
follow_up.bill_failure_after_earlier_rounds(self._billed.spend)
self._bank_apart_from_later_rounds(self._billed)
def charge_final_round(
self, completed: ResponsesAPIStreamingResponse, response: ResponsesAPIResponse, restore_headers: RestoresHeaders
) -> ResponsesAPIResponse:
parked: Final = self._take_parked()
self._banked = None
if self._billed is None or self._billed.rounds == 1:
if parked is not None:
self._release(parked)
return response
self._close(parked)
charged: Final = response.model_copy(update={"usage": self._billed.response.usage})
charged.hidden_params = {**self._billed.response.hidden_params}
if parked is not None:
final_row: Final = self._billed.bridged_row(self._latest_assembled(), charged.id) or _round_row(
completed, charged, self._billed.chat_usage, restore_headers
)
self._release((self._dispatch(final_row, self._billed, self._logging_obj),))
return charged
def finish(self) -> None:
parked: Final = self._take_parked()
self._banked = None
if parked is not None:
self._release(parked)
return
self._disarm()
def release_abandoned(self) -> None:
if not self._owns_release():
return
if _event_loop_running():
self.finish()
return
self.abandon()
def abandon(self) -> None:
self._close(self._take_parked())
self._close(self._banked)
self._banked = None
self._disarm()
def _latest_assembled(self) -> ModelResponse | None:
row: Final = self._banked_row
return row.logged if row is not None and isinstance(row.logged, ModelResponse) else None
def _price(self, response: ResponsesAPIResponse, round_parked: Parked | None) -> UncachedPrice:
if served_from_cache(response):
return _SERVED_FROM_CACHE
match round_parked:
case (ModelResponse() as assembled, bool()):
return self._logging_obj.uncached_price(assembled)
case _:
return self._logging_obj.uncached_price(response)
def _bank(self, parked: Parked | None, row: _RoundRow) -> None:
self._banked = parked
self._banked_row = row
if parked is not None:
setattr(self._logging_obj, _PARKED_DISPATCH, parked)
def _bank_apart_from_later_rounds(self, billed: _BilledRounds) -> None:
row: Final = self._banked_row
if row is None:
return
parked: Final = self._take_parked()
if parked is None:
return
self._close(parked)
self._bank((self._dispatch(row, billed, self._logging_obj.attempt_scoped_copy()),), row)
def _dispatch(
self, row: _RoundRow, billed: _BilledRounds, logger: LiteLLMLoggingObj
) -> BilledRoundsDispatch | Coroutine[object, object, None]:
dispatch: Final = self._log_success(_SuccessLog(logger, row, billed.chat_usage))
if billed.chat_usage is None:
return dispatch
return BilledRoundsDispatch(dispatch, billed.spend)
@staticmethod
async def _log_success(success: _SuccessLog) -> None:
success.carry_bridged_usage()
await _success_dispatcher(success.logger).dispatch_success_handlers(
success.row.logged,
start_time=success.logger.start_time,
end_time=success.row.end_time,
cache_hit=success.row.cache_hit,
prefer_async_handlers=True,
)
def _take_parked(self) -> Parked | None:
parked: Final[Parked | None] = getattr(self._logging_obj, _PARKED_DISPATCH, None)
if not isinstance(parked, tuple):
return None
setattr(self._logging_obj, _PARKED_DISPATCH, None)
return parked
async def _run_parked(self, *parked: object) -> None:
match parked:
case (BilledRoundsDispatch(dispatch=dispatch),):
await dispatch
case (dispatch,) if asyncio.iscoroutine(dispatch):
await dispatch
case (assembled, cache_hit):
await _success_dispatcher(self._logging_obj).dispatch_success_handlers(
assembled, cache_hit=cache_hit, start_time=None, end_time=None, prefer_async_handlers=True
)
case _:
return
def _owns_release(self) -> bool:
release: Final[object] = getattr(self._logging_obj, _DISPATCH_RELEASE, None)
return release == self._run_parked
def _release(self, parked: Parked) -> None:
if getattr(self._logging_obj, _DISPATCH_RELEASE, None) is not None and not self._owns_release():
setattr(self._logging_obj, _PARKED_DISPATCH, parked)
return
setattr(self._logging_obj, _DISPATCH_RELEASE, None)
spawn_logging_task(
asyncio.get_running_loop(), self._run_parked(*parked), task_name="Responses MCP stream success logging"
)
def _disarm(self) -> None:
if self._owns_release():
setattr(self._logging_obj, _DISPATCH_RELEASE, None)
@staticmethod
def _close(parked: Parked | None) -> None:
match parked:
case (BilledRoundsDispatch(dispatch=dispatch),):
dispatch.close()
case (dispatch,) if asyncio.iscoroutine(dispatch):
dispatch.close()
case _:
return

View file

@ -42,6 +42,7 @@ from litellm.litellm_core_utils.asyncify import run_async_function
from litellm.litellm_core_utils.core_helpers import process_response_headers
from litellm.litellm_core_utils.exception_mapping_utils import with_deployment_api_base
from litellm.litellm_core_utils.internal_call_metadata import without_budget_reservation
from litellm.litellm_core_utils.litellm_logging import BilledRoundsSpend
from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
from litellm.litellm_core_utils.llm_response_utils.get_api_base import get_api_base
from litellm.litellm_core_utils.llm_response_utils.get_headers import provider_request_id_headers
@ -364,6 +365,7 @@ class BaseResponsesAPIStreamingIterator:
_pending_logging_tasks: tuple[asyncio.Task[object], ...] = ()
_started_tool_call: bool = False
_earlier_rounds_spend: BilledRoundsSpend | None = None
def __init__(
self,
@ -708,18 +710,22 @@ class BaseResponsesAPIStreamingIterator:
if usage_obj is None:
return
try:
self.logging_obj.model_call_details["combined_usage_object"] = (
ResponseAPILoggingUtils.transform_response_api_usage_to_chat_usage(usage_obj)
)
failed_usage: Final = ResponseAPILoggingUtils.transform_response_api_usage_to_chat_usage(usage_obj)
except (TypeError, ValueError) as usage_error:
verbose_logger.debug(
"could not record usage for failed responses stream: %s",
usage_error,
)
return
self.logging_obj.model_call_details["response_cost"] = (
self.logging_obj.response_cost_calculator(result=response_obj) or 0.0
)
failed_cost: Final = self.logging_obj.response_cost_calculator(result=response_obj) or 0.0
if self._earlier_rounds_spend is None:
self.logging_obj.model_call_details["combined_usage_object"] = failed_usage
self.logging_obj.model_call_details["response_cost"] = failed_cost
return
self.logging_obj.record_failure_after_earlier_rounds(self._earlier_rounds_spend, failed_usage, failed_cost)
def bill_failure_after_earlier_rounds(self, earlier: BilledRoundsSpend) -> None:
self._earlier_rounds_spend = earlier
def _map_error_event_exception(self, error_obj: object) -> Exception:
return _map_stream_error_to_exception(

View file

@ -38,6 +38,7 @@ from contextvars import ContextVar
from dataclasses import dataclass
from datetime import datetime, timezone
from functools import lru_cache, partial
from itertools import chain
from types import MappingProxyType
from typing import TYPE_CHECKING, Any, Final, Generic, Literal, NoReturn, Optional, TypeAlias, TypeVar, Union, cast
@ -366,10 +367,11 @@ from litellm.router_utils.routing_groups import (
validate_routing_strategy,
)
from litellm.router_utils.routing_read_batch import ROUTER_USAGE_TARGET, RoutingPrefetch, RoutingReadBatch
from litellm.router_utils.stream_usage_injection import router_config_requests_stream_usage
from litellm.scheduler import FlowItem, Scheduler
from litellm.secret_managers.main import str_to_bool
from litellm.types.guardrails import GuardrailEventHooks
from litellm.types.litellm_params import RoutingStrategyName
from litellm.types.litellm_params import STREAM_USAGE_INJECTED_FIELD, RoutingStrategyName
from litellm.types.llms.openai import (
AllMessageValues,
ChatCompletionToolParam,
@ -557,6 +559,7 @@ _PreRoutingStrategyT = TypeVar("_PreRoutingStrategyT")
_CallbackT = TypeVar("_CallbackT")
_DeploymentT = TypeVar("_DeploymentT", bound=Mapping[str, object])
_ALIAS_PARAMS_NEVER_FORWARDED: Final = frozenset(
{"model", "api_base", "api_key", "api_version", "responses_api_version"}
)
@ -1051,6 +1054,8 @@ class RoutingArgs(enum.Enum):
# as the per-request one built from a caller-supplied user_config, drops out on its
# own rather than leaving entries behind that nothing can withdraw.
_live_routers: Final["weakref.WeakSet[Router]"] = weakref.WeakSet()
_routers_holding_global_callbacks: Final["weakref.WeakSet[Router]"] = weakref.WeakSet()
_routers_holding_global_callbacks_lock: Final = threading.Lock()
def _replay_live_router_model_cost() -> None:
@ -1583,22 +1588,18 @@ class Router:
self._init_routing_groups(self._routing_groups_input)
self._override_selectors: dict[str, RouterStrategySelector | None] = {}
self._override_selectors_lock = threading.Lock()
self._request_scoped_selectors: tuple[CustomLogger, ...] = ()
self._request_scoped_policies: tuple[CustomLogger, ...] = ()
self.access_groups = None
with _routers_holding_global_callbacks_lock:
_routers_holding_global_callbacks.add(self)
## USAGE TRACKING ##
if isinstance(litellm._async_success_callback, list):
litellm.logging_callback_manager.add_litellm_async_success_callback(self.deployment_callback_on_success)
else:
litellm.logging_callback_manager.add_litellm_async_success_callback(self.deployment_callback_on_success)
litellm.logging_callback_manager.add_litellm_async_success_callback(self.deployment_callback_on_success)
if isinstance(litellm.success_callback, list):
litellm.logging_callback_manager.add_litellm_success_callback(self.sync_deployment_callback_on_success)
else:
litellm.success_callback = [self.sync_deployment_callback_on_success]
if isinstance(litellm._async_failure_callback, list):
litellm.logging_callback_manager.add_litellm_async_failure_callback(
self.async_deployment_callback_on_failure
)
else:
litellm._async_failure_callback = [self.async_deployment_callback_on_failure]
litellm.logging_callback_manager.add_litellm_async_failure_callback(self.async_deployment_callback_on_failure)
## COOLDOWNS ##
if isinstance(litellm.failure_callback, list):
litellm.logging_callback_manager.add_litellm_failure_callback(self.deployment_callback_on_failure)
@ -1694,20 +1695,116 @@ class Router:
# Stop contributing to cost-map rebuilds straight away rather than waiting
# for this router to be collected.
_live_routers.discard(self)
litellm.logging_callback_manager.remove_callback_from_list_by_object(litellm._async_success_callback, self)
litellm.logging_callback_manager.remove_callback_from_list_by_object(litellm.success_callback, self)
litellm.logging_callback_manager.remove_callback_from_list_by_object(litellm._async_failure_callback, self)
litellm.logging_callback_manager.remove_callback_from_list_by_object(litellm.failure_callback, self)
litellm.logging_callback_manager.remove_callback_from_list_by_object(litellm.input_callback, self)
litellm.logging_callback_manager.remove_callback_from_list_by_object(litellm.service_callback, self)
litellm.logging_callback_manager.remove_callback_from_list_by_object(litellm.callbacks, self)
self.retire_routing_state()
self._release_global_callbacks()
# Remove ForwardClientSideHeadersByModelGroup if it exists
if self.optional_callbacks is not None:
for callback in self.optional_callbacks:
litellm.logging_callback_manager.remove_callback_from_list_by_object(
litellm.callbacks, callback, require_self=False
)
def _release_global_callbacks(self) -> None:
with _routers_holding_global_callbacks_lock:
_routers_holding_global_callbacks.discard(self)
slack_alerting_loggers: Final = self._slack_alerting_loggers()
adaptive_router_hooks: Final = self._adaptive_router_hooks()
vacated_keys: Final = litellm.logging_callback_manager.drop_callbacks(
(
*self._callback_selectors(),
*self._override_selectors.values(),
*(self.optional_callbacks or ()),
*slack_alerting_loggers,
*adaptive_router_hooks,
),
method_owners=(self, *slack_alerting_loggers),
)
with _routers_holding_global_callbacks_lock:
survivors: Final = tuple(_routers_holding_global_callbacks)
candidates: Final = chain.from_iterable(router._global_callback_candidates() for router in survivors)
for candidate in litellm.logging_callback_manager.with_custom_logger_keys(candidates, vacated_keys):
self._register_router_selector(candidate)
adaptive_keys: Final = frozenset(
map(litellm.logging_callback_manager.get_custom_logger_key, adaptive_router_hooks)
)
adaptive_survivor: Final = next((router for router in survivors if router.adaptive_routers), None)
if adaptive_survivor is not None and vacated_keys & adaptive_keys:
adaptive_survivor._sync_adaptive_router_hooks()
def serve_through_request_logging(self) -> None:
self._request_scoped_selectors = (*self._callback_selectors(), *self._request_adaptive_router_hooks())
self._request_scoped_policies = (*self._optional_loggers(), *self._slack_alerting_loggers())
self._release_global_callbacks()
def bind_to_request_logging(self, logging_obj: LiteLLMLogging) -> None:
logging_obj.add_dynamic_function_callbacks(
async_success=self.deployment_callback_on_success,
failure=self.deployment_callback_on_failure,
async_failure=self.async_deployment_callback_on_failure,
)
for selector in self._request_scoped_selectors:
logging_obj.replace_global_callback(selector)
for policy in self._request_policies_not_set_globally():
logging_obj.add_dynamic_callback(policy)
def _request_policies_not_set_globally(self) -> tuple[CustomLogger, ...]:
global_keys: Final = frozenset(
map(
litellm.logging_callback_manager.get_custom_logger_key,
(callback for callback in litellm.callbacks if isinstance(callback, CustomLogger)),
)
)
return tuple(
policy
for policy in self._request_scoped_policies
if litellm.logging_callback_manager.get_custom_logger_key(policy) not in global_keys
)
def _request_adaptive_router_hooks(self) -> tuple[CustomLogger, ...]:
from litellm.router_strategy.adaptive_router.hooks import (
AdaptiveRouterPostCallHook,
)
return tuple(
AdaptiveRouterPostCallHook(adaptive_router=tagged.strategy)
for tagged in chain.from_iterable(self.adaptive_routers.values())
)
def _routing_callbacks(self) -> Iterable[object]:
if not (self._request_scoped_selectors or self._request_scoped_policies):
return litellm.callbacks
replaced_keys: Final = frozenset(
map(litellm.logging_callback_manager.get_custom_logger_key, self._request_scoped_selectors)
)
return chain(
litellm.logging_callback_manager.without_custom_logger_keys(litellm.callbacks, replaced_keys),
self._request_scoped_selectors,
self._request_policies_not_set_globally(),
)
def retire_routing_state(self) -> None:
for strategy in (*self._global_callback_candidates(), *self._override_selectors.values()):
if isinstance(strategy, BaseRoutingStrategy | RouterBudgetLimiting):
strategy.retire()
def _slack_alerting_loggers(self) -> tuple[CustomLogger, ...]:
slack_alerting_logger: Final = getattr(self, "slack_alerting_logger", None)
return (slack_alerting_logger,) if isinstance(slack_alerting_logger, CustomLogger) else ()
def _adaptive_router_hooks(self) -> tuple[CustomLogger, ...]:
from litellm.router_strategy.adaptive_router.hooks import (
AdaptiveRouterPostCallHook,
)
own_strategy_ids: Final = frozenset(
id(tagged.strategy) for tagged in chain.from_iterable(self.adaptive_routers.values())
)
hooks: Final = litellm.logging_callback_manager.get_custom_loggers_for_type(AdaptiveRouterPostCallHook)
return tuple(
hook
for hook in hooks
if isinstance(hook, AdaptiveRouterPostCallHook) and id(hook.adaptive_router) in own_strategy_ids
)
def _optional_loggers(self) -> tuple[CustomLogger, ...]:
return tuple(callback for callback in self.optional_callbacks or () if isinstance(callback, CustomLogger))
def _global_callback_candidates(self) -> tuple[CustomLogger, ...]:
return (*self._callback_selectors(), *self._optional_loggers(), *self._slack_alerting_loggers())
@staticmethod
def _create_redis_cache(
@ -1813,7 +1910,7 @@ class Router:
return selector
@staticmethod
def _register_router_selector(selector: RouterStrategySelector) -> None:
def _register_router_selector(selector: CustomLogger) -> None:
if isinstance(selector, LeastBusyLoggingHandler):
if isinstance(litellm.input_callback, list):
litellm.logging_callback_manager.add_litellm_input_callback(selector)
@ -1822,23 +1919,28 @@ class Router:
if isinstance(litellm.callbacks, list):
litellm.logging_callback_manager.add_litellm_callback(selector)
def _callback_selectors(self) -> tuple[RouterStrategySelector, ...]:
return tuple(
selector
for selector in (
*(getattr(self, attr, None) for attr in self._DEFAULT_SELECTOR_ATTR_BY_STRATEGY.values()),
*chain.from_iterable(selectors.values() for selectors in self._group_selectors.values()),
)
if selector is not None
)
def _unregister_router_selectors(self, selectors: Sequence[object]) -> None:
"""
Drop router-owned strategy selectors from litellm's global callback
lists by identity. Used before re-init (`routing_strategy_init` /
`_init_routing_groups`) so repeated `update_settings` calls don't
accumulate dead selectors that keep receiving callback events.
lists. Used before re-init (`routing_strategy_init` /
`_init_routing_groups`) and by `discard` so dead selectors stop
receiving callback events. `function_setup` copies `litellm.callbacks`
into the per-event lists on every request, so those are purged too.
"""
for selector in selectors:
if isinstance(selector, BaseRoutingStrategy):
selector.retire()
selector_ids: Final = {id(s) for s in selectors if s is not None}
if not selector_ids:
return
if isinstance(litellm.callbacks, list):
litellm.callbacks = [c for c in litellm.callbacks if id(c) not in selector_ids]
if isinstance(litellm.input_callback, list):
litellm.input_callback = [c for c in litellm.input_callback if id(c) not in selector_ids]
litellm.logging_callback_manager.drop_callbacks(selectors)
def _apply_updated_routing_strategy_args(self) -> None:
"""
@ -2530,6 +2632,11 @@ class Router:
self.asystemone = self.factory_function(asystemone, call_type="asystemone") # pyright: ignore[reportUnknownMemberType] # factory_function takes a bare Callable
self.systemone = self.factory_function(systemone, call_type="systemone") # pyright: ignore[reportUnknownMemberType] # factory_function takes a bare Callable
from litellm.extraction import aextraction, extraction
self.aextraction = self.factory_function(aextraction, call_type="aextraction") # pyright: ignore[reportUnknownMemberType] # factory_function takes a bare Callable
self.extraction = self.factory_function(extraction, call_type="extraction") # pyright: ignore[reportUnknownMemberType] # factory_function takes a bare Callable
def _initialize_video_endpoints(self):
"""Initialize video endpoints."""
from litellm.videos import (
@ -4657,7 +4764,8 @@ class Router:
deployment=deployment, forwarded_keys=kwargs.pop(_ALIAS_MARKER_FORWARDED_PARAMS_KWARG, ())
):
kwargs.pop(key, None)
self._merge_tools_from_deployment(deployment=deployment, kwargs=kwargs, defaults=self.default_litellm_params)
defaults: Final = _LITELLM_PARAMS_ADAPTER.validate_python(self.default_litellm_params)
self._merge_tools_from_deployment(deployment=deployment, kwargs=kwargs, defaults=defaults)
model_info = deployment.get("model_info", {}).copy()
deployment_litellm_model_name = deployment["litellm_params"]["model"]
@ -4801,6 +4909,19 @@ class Router:
data=_DEPLOYMENT_PARAMS_ADAPTER.validate_python(deployment["litellm_params"]),
)
deployment_litellm_params: Final = _DEPLOYMENT_PARAMS_ADAPTER.validate_python(deployment["litellm_params"])
if router_config_requests_stream_usage(
_REQUEST_KWARGS_ADAPTER.validate_python(kwargs),
(
_LITELLM_PARAMS_ADAPTER.validate_python(
deployment_litellm_params
if isinstance(deployment_litellm_params, Mapping)
else deployment_litellm_params.model_dump(exclude_none=True)
),
defaults,
),
):
kwargs[STREAM_USAGE_INJECTED_FIELD] = True
self._update_kwargs_with_default_litellm_params(kwargs=kwargs, metadata_variable_name=metadata_variable_name)
hop_kwargs: Final = _FALLBACK_HOP_ADAPTER.validate_python(kwargs)
if _is_fallback_hop(hop_kwargs):
@ -7699,6 +7820,8 @@ class Router:
"decisions",
"asystemone",
"systemone",
"aextraction",
"extraction",
"aadapter_generate_content",
"avideo_generation",
"video_generation",
@ -7774,6 +7897,7 @@ class Router:
"search",
"decisions",
"systemone",
"extraction",
"video_generation",
"video_list",
"video_status",
@ -7943,6 +8067,7 @@ class Router:
"ocr",
"adecisions",
"asystemone",
"aextraction",
"avideo_generation",
"avideo_list",
"avideo_status",
@ -9492,17 +9617,24 @@ class Router:
When a retry or fallback happens, record which model group, deployment and attempt just failed and why,
and count it toward the request-wide num_retries_per_request cap
"""
from litellm.litellm_core_utils.get_litellm_params import get_base_model_from_litellm_call_metadata
from litellm.types.router import RetryAttemptRecord
_metadata_var: Final = "litellm_metadata" if "litellm_metadata" in kwargs else "metadata"
request_metadata: Final[Mapping[str, object]] = kwargs[_metadata_var]
request_metadata: Final = cast("Mapping[str, object]", kwargs[_metadata_var]) # cast-ok: untyped kwargs
model_group: Final = kwargs.get("model")
model_info: Final = request_metadata.get("model_info")
deployment_id: Final = model_info.get("id") if isinstance(model_info, Mapping) else None
base_model: Final = get_base_model_from_litellm_call_metadata(
metadata=request_metadata
) or self._deployment_params_base_model(deployment_id if isinstance(deployment_id, str) else None)
deployment_model: Final = request_metadata.get("deployment")
attempted_retries: Final = request_metadata.get("attempted_retries")
attempt_record: Final[RetryAttemptRecord] = {
"model_group": model_group if isinstance(model_group, str) else None,
"deployment_id": deployment_id if isinstance(deployment_id, str) else None,
"model": deployment_model if isinstance(deployment_model, str) else None,
"base_model": base_model if isinstance(base_model, str) else None,
"exception_type": type(e).__name__,
"exception_string": str(e),
"attempted_retries": attempted_retries if type(attempted_retries) is int else None,
@ -9533,6 +9665,18 @@ class Router:
kwargs[_metadata_var]["request_retry_count"] = request_retry_count # rebind-ok: same dict, read by the cap
return kwargs
def _deployment_params_base_model(self, deployment_id: str | None) -> str | None:
"""The ``base_model`` a deployment declares under ``litellm_params`` (the Azure named-deployment
shape), the same key the success path's ``get_base_model_from_metadata`` reads."""
deployment: Final = self.get_deployment(model_id=deployment_id) if deployment_id is not None else None
if deployment is None:
return None
extra_params: Final = cast( # cast-ok: pydantic extras are untyped, base_model is isinstance-checked below
"Mapping[str, object]", deployment.litellm_params.model_extra or {}
)
base_model: Final = extra_params.get("base_model")
return base_model if isinstance(base_model, str) else None
def _update_usage(self, deployment_id: str, parent_otel_span: Span | None) -> int:
"""
Update deployment rpm for that minute
@ -9755,7 +9899,7 @@ class Router:
Raises:
- Rate Limit Exception - If the deployment is over it's tpm/rpm limits
"""
for _callback in litellm.callbacks:
for _callback in self._routing_callbacks():
if isinstance(_callback, CustomLogger):
_callback.pre_call_check(deployment)
@ -9776,7 +9920,7 @@ class Router:
Raises:
- Rate Limit Exception - If the deployment is over it's tpm/rpm limits
"""
for _callback in litellm.callbacks:
for _callback in self._routing_callbacks():
if isinstance(_callback, CustomLogger):
try:
await _callback.async_pre_call_check(deployment, parent_otel_span)
@ -9913,7 +10057,7 @@ class Router:
- Rate Limit Exception - If the deployment is over it's tpm/rpm limits
"""
returned_healthy_deployments = healthy_deployments
for _callback in litellm.callbacks:
for _callback in self._routing_callbacks():
if isinstance(_callback, CustomLogger):
try:
returned_healthy_deployments = await _callback.async_filter_deployments(
@ -10568,6 +10712,7 @@ class Router:
changes, otherwise a released router keeps recording turns through its hook."""
from litellm.router_strategy.adaptive_router.hooks import (
AdaptiveRouterPostCallHook,
AdaptiveRouterResponseHeaderHook,
)
for callback in litellm.logging_callback_manager.get_custom_loggers_for_type(AdaptiveRouterPostCallHook):
@ -10577,6 +10722,8 @@ class Router:
litellm.logging_callback_manager.add_litellm_callback(
AdaptiveRouterPostCallHook(adaptive_router=tagged.strategy)
)
if self.adaptive_routers:
litellm.logging_callback_manager.add_litellm_callback(AdaptiveRouterResponseHeaderHook())
def init_adaptive_router_deployment(self, deployment: Deployment) -> None:
"""

View file

@ -1,6 +1,9 @@
"""Adaptive router strategy. See README.md for design overview."""
from litellm.router_strategy.adaptive_router.adaptive_router import AdaptiveRouter
from litellm.router_strategy.adaptive_router.hooks import AdaptiveRouterPostCallHook
from litellm.router_strategy.adaptive_router.hooks import (
AdaptiveRouterPostCallHook,
AdaptiveRouterResponseHeaderHook,
)
__all__ = ["AdaptiveRouter", "AdaptiveRouterPostCallHook"]
__all__ = ["AdaptiveRouter", "AdaptiveRouterPostCallHook", "AdaptiveRouterResponseHeaderHook"]

View file

@ -170,12 +170,7 @@ def _assistant_content_and_tool_calls(response_obj: Any) -> tuple[object, Sequen
return content, tool_calls
class AdaptiveRouterPostCallHook(CustomLogger):
"""One hook instance per AdaptiveRouter. Registered into litellm.callbacks."""
def __init__(self, adaptive_router: AdaptiveRouter) -> None:
self.adaptive_router = adaptive_router
class AdaptiveRouterResponseHeaderHook(CustomLogger):
async def async_post_call_response_headers_hook(
self,
data: Mapping[str, object],
@ -200,6 +195,13 @@ class AdaptiveRouterPostCallHook(CustomLogger):
return None
return {ADAPTIVE_ROUTER_RESPONSE_HEADER: chosen}
class AdaptiveRouterPostCallHook(CustomLogger):
"""One hook instance per AdaptiveRouter. Registered into litellm.callbacks."""
def __init__(self, adaptive_router: AdaptiveRouter) -> None:
self.adaptive_router = adaptive_router
async def async_log_success_event(self, kwargs, response_obj, start_time, end_time):
await self._record(kwargs, response_obj, response_status=200)

View file

@ -24,6 +24,8 @@ class BaseRoutingStrategy(ABC):
self.dual_cache = dual_cache
self.redis_increment_operation_queue: list[RedisPipelineIncrementOperation] = []
self._sync_task: asyncio.Task[None] | None = None
self._retired = False
self._flush_after_retire: asyncio.Task[None] | None = None
if should_batch_redis_writes:
self.setup_sync_task(default_sync_interval)
@ -47,13 +49,25 @@ class BaseRoutingStrategy(ABC):
def retire(self) -> None:
self.cancel_sync_task()
if not self.redis_increment_operation_queue:
self._retired = True
self._schedule_flush_after_retire()
def _schedule_flush_after_retire(self) -> None:
if self._flush_after_retire is not None and not self._flush_after_retire.done():
return
if not self.redis_increment_operation_queue or self.dual_cache.redis_cache is None:
return
try:
loop: Final = asyncio.get_running_loop()
except RuntimeError:
return
loop.create_task(self._push_in_memory_increments_to_redis())
self._flush_after_retire = loop.create_task(self._drain_increments_after_retire())
async def _drain_increments_after_retire(self) -> None:
if not self.redis_increment_operation_queue:
return
await self._push_in_memory_increments_to_redis()
await self._drain_increments_after_retire()
async def cleanup(self):
"""Cleanup method to be called when shutting down"""
@ -98,6 +112,8 @@ class BaseRoutingStrategy(ABC):
self.redis_increment_operation_queue.append(increment_op)
self.add_to_in_memory_keys_to_update(key=key)
if self._retired:
self._schedule_flush_after_retire()
return result
@with_service_target("router_usage")

View file

@ -128,7 +128,9 @@ class RouterBudgetLimiting(CustomLogger):
self._redis_increment_queue_lock = asyncio.Lock()
self._redis_increment_flush_lock = asyncio.Lock()
self._detached_increment_operations: tuple[RedisPipelineIncrementOperation, ...] | None = None
asyncio.create_task(self.periodic_sync_in_memory_spend_with_redis())
self._sync_task: asyncio.Task[None] = asyncio.create_task(self.periodic_sync_in_memory_spend_with_redis())
self._retired = False
self._flush_after_retire: asyncio.Task[None] | None = None
self.provider_budget_config: GenericBudgetConfigType | None = provider_budget_config
self.deployment_budget_config: GenericBudgetConfigType | None = None
self.tag_budget_config: GenericBudgetConfigType | None = None
@ -140,6 +142,30 @@ class RouterBudgetLimiting(CustomLogger):
if isinstance(litellm.callbacks, list):
litellm.logging_callback_manager.add_litellm_callback(self)
def retire(self) -> None:
self._sync_task.cancel()
self._retired = True
self._schedule_flush_after_retire()
def _schedule_flush_after_retire(self) -> None:
if self._flush_after_retire is not None and not self._flush_after_retire.done():
return
if not self._has_unwritten_increments():
return
try:
loop: Final = asyncio.get_running_loop()
except RuntimeError:
return
self._flush_after_retire = loop.create_task(self._drain_increments_after_retire())
def _has_unwritten_increments(self) -> bool:
return bool(self.redis_increment_operation_queue) or self._detached_increment_operations is not None
async def _drain_increments_after_retire(self) -> None:
while self._has_unwritten_increments():
if not await self._push_in_memory_increments_to_redis():
await asyncio.sleep(DEFAULT_REDIS_SYNC_INTERVAL)
@with_service_target("router_budgets")
async def async_filter_deployments(
self,
@ -427,6 +453,8 @@ class RouterBudgetLimiting(CustomLogger):
if self.dual_cache.redis_cache is None:
return
self.redis_increment_operation_queue.append(increment_op)
if self._retired:
self._schedule_flush_after_retire()
def _get_redis_increment_queue_lock(self) -> asyncio.Lock:
return self._redis_increment_queue_lock

View file

@ -0,0 +1,17 @@
from collections.abc import Mapping, Sequence
def router_config_requests_stream_usage(
request_kwargs: Mapping[str, object], configured_params: Sequence[Mapping[str, object]]
) -> bool:
if request_kwargs.get("stream") is not True or "stream_options" in request_kwargs:
return False
return any(_includes_usage(params.get("stream_options")) for params in configured_params)
def _includes_usage(stream_options: object) -> bool:
match stream_options:
case {"include_usage": True}:
return True
case _:
return False

162
litellm/types/extraction.py Normal file
View file

@ -0,0 +1,162 @@
from collections.abc import Mapping, Sequence
from typing import Annotated, Final, Literal, TypeAlias
from pydantic import (
ConfigDict,
Field,
JsonValue,
PrivateAttr,
StrictBool,
StrictFloat,
StrictInt,
StrictStr,
model_validator,
with_config,
)
from typing_extensions import ReadOnly, Required, Self, TypedDict
from litellm.types.llms.base import LiteLLMPydanticObjectBase
ExtractionFieldType: TypeAlias = Literal[
"text", "number", "date", "time", "datetime", "duration", "list", "estimate", "forecast"
]
ExtractionFieldName: TypeAlias = Annotated[str, Field(pattern=r"^[A-Za-z_][A-Za-z0-9_.\-]{0,63}$")]
JsonNumber: TypeAlias = StrictInt | StrictFloat
ExtractionBound: TypeAlias = JsonNumber | StrictStr
EXTRACTION_STATE_REWRITTEN_METADATA_KEY: Final = "extraction_state_rewritten_by_guardrail"
def _is_none(value: object) -> bool:
return value is None
_OMITTED_WHEN_NONE: Final = Field(exclude_if=_is_none)
OptionalJson: TypeAlias = Annotated[JsonValue, _OMITTED_WHEN_NONE]
OptionalFloat: TypeAlias = Annotated[float | None, _OMITTED_WHEN_NONE]
OptionalInt: TypeAlias = Annotated[int | None, _OMITTED_WHEN_NONE]
class _ClairicObject(LiteLLMPydanticObjectBase):
model_config = ConfigDict(extra="allow", frozen=True)
class ExtractionField(_ClairicObject):
type: Annotated[ExtractionFieldType, Field(description="The answer type for this field")]
instructions: Annotated[str | None, Field(description="What the field should contain")] = None
unit: Annotated[str | None, Field(description="Unit label. For durations, seconds through years")] = None
integer: Annotated[StrictBool | None, Field(description="Numbers and durations only")] = None
decimals: Annotated[JsonNumber | None, Field(description="Numbers, durations and number lists only, 0 to 6")] = None
items: Annotated[
Literal["text", "number", "date"] | None, Field(description="Item type, required for list fields")
] = None
min: Annotated[ExtractionBound | None, Field(description="Estimate and forecast lower bound")] = None
max: Annotated[ExtractionBound | None, Field(description="Estimate and forecast upper bound")] = None
step: Annotated[ExtractionBound | None, Field(description="Estimate grid step, or month or year")] = None
scale: Literal["linear", "log"] | None = None
null_threshold: Annotated[
JsonNumber | None, Field(description="Probability of null above which the value is null")
] = None
horizon: Annotated[JsonNumber | None, Field(description="Forecast only, number of steps")] = None
interval: Literal["step", "minute", "hour", "day", "week", "month", "quarter", "year"] | None = None
@model_validator(mode="after")
def _list_fields_name_their_item_type(self) -> Self:
if self.type == "list" and self.items is None:
raise ValueError("list fields need items set to text, number or date")
return self
class ExtractionBase64Image(_ClairicObject):
content_type: Annotated[str, Field(description="The image media type, such as image/png")]
base64: Annotated[str, Field(description="Base64-encoded image bytes")]
ExtractionImage: TypeAlias = Annotated[str, Field(pattern=r"^[Dd][Aa][Tt][Aa]:")] | ExtractionBase64Image
class ExtractionOptions(_ClairicObject):
include_quantiles: Annotated[StrictBool | None, Field(description="Add 99 percentiles to estimates")] = None
return_state_text: Annotated[StrictBool | None, Field(description="Return the text that spans index into")] = None
class ExtractionRequestBody(_ClairicObject):
state: Annotated[
str,
Field(description="The text to read. Must be non-empty unless images are sent. Text answers are spans of it"),
] = ""
fields: Annotated[
Mapping[ExtractionFieldName, ExtractionField],
Field(min_length=1, description="Field name to typed field spec. Each name becomes a key in the answers"),
]
images: Annotated[
tuple[ExtractionImage, ...] | None,
Field(max_length=4, description="Up to 4 images as data URLs or content_type plus base64, no remote URLs"),
] = None
reference_date: Annotated[str | None, Field(description="ISO date that anchors dates written without a year")] = (
None
)
options: ExtractionOptions | None = None
class ExtractionRequest(ExtractionRequestBody):
model: Annotated[str, Field(description="An extraction model from the proxy model_list")]
class ExtractionAnswer(_ClairicObject):
type: str
value: Annotated[JsonValue, Field(description="The answer, null when the input does not determine it")]
guess: OptionalJson = None
p_null: OptionalFloat = None
confidence: OptionalFloat = None
span: Annotated[tuple[int, ...] | None, _OMITTED_WHEN_NONE] = None
spans: Annotated[tuple[tuple[int, ...], ...] | None, _OMITTED_WHEN_NONE] = None
mean: OptionalJson = None
median: OptionalJson = None
p10: OptionalJson = None
p90: OptionalJson = None
p_below: OptionalFloat = None
p_above: OptionalFloat = None
quantiles: Annotated[tuple[JsonValue, ...] | None, _OMITTED_WHEN_NONE] = None
steps: Annotated[tuple[JsonValue, ...] | None, _OMITTED_WHEN_NONE] = None
class ExtractionUsage(_ClairicObject):
input_tokens: int
output_tokens: int
state_tokens: OptionalInt = None
image_tokens: OptionalInt = None
schema_tokens: OptionalInt = None
passes: OptionalInt = None
class ExtractionResponse(_ClairicObject):
model: str
answers: Mapping[str, ExtractionAnswer]
truncated: bool
state_text: Annotated[str | None, _OMITTED_WHEN_NONE] = None
usage: ExtractionUsage
_hidden_params: dict[str, object] = PrivateAttr( # mutable-ok: @client and the proxy write cost into it in place
default_factory=dict
)
@property
def hidden_params(self) -> dict[str, object]: # mutable-ok: API requires mutation
return self._hidden_params
def set_hidden_params(self, params: Mapping[str, object]) -> None:
self._hidden_params.update(params)
@with_config(ConfigDict(extra="allow"))
class ExtractionCallParams(TypedDict, total=False):
model: Required[ReadOnly[str]]
state: Required[ReadOnly[str]]
fields: Required[ReadOnly[Mapping[str, Mapping[str, object]]]]
images: ReadOnly[Sequence[str | Mapping[str, object]] | None]
reference_date: ReadOnly[str | None]
api_key: ReadOnly[str | None]
api_base: ReadOnly[str | None]
timeout: ReadOnly[float | None]
custom_llm_provider: ReadOnly[str | None]
extra_headers: ReadOnly[Mapping[str, str] | None]

View file

@ -54,6 +54,7 @@ RoutingStrategyName: TypeAlias = Literal[
TRUSTED_CALLBACK_VARS_FIELD: Final = "litellm_trusted_callback_vars"
ADDRESSED_RESPONSE_ID_FIELD: Final = "_litellm_addressed_response_id"
STREAM_USAGE_INJECTED_FIELD: Final = "_litellm_stream_usage_injected"
WIRE_NAME: Final = "wire_name"
@ -397,6 +398,7 @@ class ProxyRequestState:
trusted_callback_vars: Mapping[str, str] | None = field(default=None, metadata=wire(TRUSTED_CALLBACK_VARS_FIELD))
addressed_response_id: str | None = field(default=None, metadata=wire(ADDRESSED_RESPONSE_ID_FIELD))
strip_stream_usage: bool | None = field(default=None, metadata=wire("_litellm_strip_stream_usage"))
stream_usage_injected: bool | None = field(default=None, metadata=wire(STREAM_USAGE_INJECTED_FIELD))
client_side_timeout: bool | None = None
model_file_id_mapping: Mapping[str, Mapping[str, str]] | None = None

View file

@ -239,8 +239,18 @@ class VoiceConfig(TypedDict):
prebuiltVoiceConfig: PrebuiltVoiceConfig
class SpeakerVoiceConfig(TypedDict):
speaker: ReadOnly[str]
voiceConfig: ReadOnly[VoiceConfig]
class MultiSpeakerVoiceConfig(TypedDict):
speakerVoiceConfigs: ReadOnly[Sequence[SpeakerVoiceConfig]]
class SpeechConfig(TypedDict, total=False):
voiceConfig: VoiceConfig
multiSpeakerVoiceConfig: ReadOnly[MultiSpeakerVoiceConfig]
languageCode: str

View file

@ -4,7 +4,9 @@ Type definitions for Vertex AI Text-to-Speech API
Reference: https://cloud.google.com/text-to-speech/docs/reference/rest/v1/text/synthesize
"""
from typing_extensions import TypedDict
from collections.abc import Sequence
from typing_extensions import ReadOnly, TypedDict
class VertexTextToSpeechInput(TypedDict, total=False):
@ -16,6 +18,16 @@ class VertexTextToSpeechInput(TypedDict, total=False):
text: str | None
ssml: str | None
prompt: ReadOnly[str | None]
class VertexTextToSpeechSpeakerVoiceConfig(TypedDict):
speakerAlias: ReadOnly[str]
speakerId: ReadOnly[str]
class VertexTextToSpeechMultiSpeakerVoiceConfig(TypedDict):
speakerVoiceConfigs: ReadOnly[Sequence[VertexTextToSpeechSpeakerVoiceConfig]]
class VertexTextToSpeechVoice(TypedDict, total=False):
@ -29,6 +41,8 @@ class VertexTextToSpeechVoice(TypedDict, total=False):
languageCode: str
name: str
modelName: ReadOnly[str]
multiSpeakerVoiceConfig: ReadOnly[VertexTextToSpeechMultiSpeakerVoiceConfig]
class VertexTextToSpeechAudioConfig(TypedDict, total=False):
@ -42,6 +56,7 @@ class VertexTextToSpeechAudioConfig(TypedDict, total=False):
audioEncoding: str
speakingRate: str
sampleRateHertz: ReadOnly[int]
class VertexTextToSpeechRequest(TypedDict, total=False):

View file

@ -1,3 +1,4 @@
from collections.abc import Mapping
from datetime import datetime
from typing import Any
@ -13,6 +14,8 @@ class ModelGroupInfoProxy(ModelGroupInfo):
health_status: str | None = Field(default=None)
health_response_time: float | None = Field(default=None)
health_checked_at: str | None = Field(default=None)
custom_tags: tuple[str, ...] | None = None
custom_metadata: Mapping[str, str | int | float | bool] | None = None
class UpdateUsefulLinksRequest(LiteLLMBaseModel):

View file

@ -1184,6 +1184,8 @@ class RouterModelGroupAliasItem(TypedDict):
class RetryAttemptRecord(TypedDict):
model_group: ReadOnly[str | None]
deployment_id: ReadOnly[str | None]
model: ReadOnly[str | None]
base_model: ReadOnly[str | None]
exception_type: ReadOnly[str]
exception_string: ReadOnly[str]
attempted_retries: ReadOnly[int | None]

View file

@ -212,6 +212,7 @@ class ProviderSpecificModelInfo(TypedDict, total=False):
supports_parallel_function_calling: bool | None
supports_web_search: bool | None
supports_reasoning: bool | None
supports_sampling_params: ReadOnly[bool | None]
supports_adaptive_thinking: bool | None
supports_legacy_thinking: ReadOnly[bool | None]
thinking_always_on: ReadOnly[bool | None]
@ -458,6 +459,7 @@ class ModelInfoBase(ProviderSpecificModelInfo, total=False):
"responses",
"decisions",
"evaluation",
"extraction",
"ocr",
"realtime",
]
@ -518,6 +520,8 @@ class CallTypes(str, Enum):
adecisions = "adecisions"
systemone = "systemone"
asystemone = "asystemone"
extraction = "extraction"
aextraction = "aextraction"
arealtime = "_arealtime"
aresponses_websocket = "_aresponses_websocket"
create_batch = "create_batch"
@ -708,6 +712,8 @@ CallTypesLiteral = Literal[
"adecisions",
"systemone",
"asystemone",
"extraction",
"aextraction",
"_arealtime",
"_aresponses_websocket",
"create_batch",
@ -821,6 +827,8 @@ API_ROUTE_TO_CALL_TYPES: Final[Mapping[str, Sequence[CallTypes]]] = {
"/v1/decisions": [CallTypes.adecisions, CallTypes.decisions],
"/systemone": [CallTypes.asystemone, CallTypes.systemone],
"/v1/systemone": [CallTypes.asystemone, CallTypes.systemone],
"/extraction": [CallTypes.aextraction, CallTypes.extraction],
"/v1/extraction": [CallTypes.aextraction, CallTypes.extraction],
# Batches
"/batches": [CallTypes.acreate_batch, CallTypes.create_batch],
"/v1/batches": [CallTypes.acreate_batch, CallTypes.create_batch],

View file

@ -71,6 +71,7 @@ from litellm.constants import (
DEFAULT_TRIM_RATIO,
FUNCTION_DEFINITION_TOKEN_COUNT,
HF_CONFIG_FETCH_TIMEOUT_SECONDS,
HUGGINGFACE_TOKENIZER_LOAD_RETRY_SECONDS,
INITIAL_RETRY_DELAY,
JITTER,
MAX_RETRY_DELAY,
@ -390,6 +391,7 @@ if TYPE_CHECKING:
from litellm.llms.base_llm.containers.transformation import BaseContainerConfig
from litellm.llms.base_llm.decisions.transformation import BaseDecisionsConfig
from litellm.llms.base_llm.embedding.transformation import BaseEmbeddingConfig
from litellm.llms.base_llm.extraction.transformation import BaseExtractionConfig
from litellm.llms.base_llm.files.transformation import BaseFilesConfig
from litellm.llms.base_llm.google_genai.transformation import (
BaseGoogleGenAIGenerateContentConfig,
@ -858,6 +860,11 @@ def _is_gemini_model(model: str | None, custom_llm_provider: str | None) -> bool
return model is not None and "gemini" in model.lower()
def is_gemini_tts_model(model: str) -> bool:
model_name: Final = model.lower()
return "gemini" in model_name and "tts" in model_name
def _remove_thought_signature_from_id(tool_call_id: str, separator: str) -> str:
"""
Remove thought signature from a tool call ID.
@ -1124,83 +1131,90 @@ def function_setup(
dynamic_callbacks: Final[list[str | Callable | CustomLogger] | None] = kwargs.pop("callbacks", None)
all_callbacks: Final = get_dynamic_callbacks(dynamic_callbacks=dynamic_callbacks)
if len(all_callbacks) > 0:
for callback in all_callbacks:
# check if callback is a string - e.g. "lago", "openmeter"
if isinstance(callback, str):
callback = litellm.litellm_core_utils.litellm_logging._init_custom_logger_compatible_class(
callback,
internal_usage_cache=None,
llm_router=None,
)
if callback is None or any(
type(cb) is type(callback) for cb in litellm._async_success_callback
): # don't double add a callback
continue
if callback not in litellm.input_callback:
litellm.input_callback.append(callback)
if callback not in litellm.success_callback:
litellm.logging_callback_manager.add_litellm_success_callback(callback)
if callback not in litellm.failure_callback:
litellm.logging_callback_manager.add_litellm_failure_callback(callback)
if callback not in litellm._async_success_callback:
litellm.logging_callback_manager.add_litellm_async_success_callback(callback)
if callback not in litellm._async_failure_callback:
litellm.logging_callback_manager.add_litellm_async_failure_callback(callback)
print_verbose(f"Initialized litellm callbacks, Async Success Callbacks: {litellm._async_success_callback}")
with litellm.logging_callback_manager.callback_lists_lock:
if len(all_callbacks) > 0:
for callback in all_callbacks:
# check if callback is a string - e.g. "lago", "openmeter"
if isinstance(callback, str):
callback = litellm.litellm_core_utils.litellm_logging._init_custom_logger_compatible_class(
callback,
internal_usage_cache=None,
llm_router=None,
)
if callback is None or any(
type(cb) is type(callback) for cb in litellm._async_success_callback
): # don't double add a callback
continue
if callback not in litellm.input_callback:
litellm.input_callback.append(callback)
if callback not in litellm.success_callback:
litellm.logging_callback_manager.add_litellm_success_callback(callback)
if callback not in litellm.failure_callback:
litellm.logging_callback_manager.add_litellm_failure_callback(callback)
if callback not in litellm._async_success_callback:
litellm.logging_callback_manager.add_litellm_async_success_callback(callback)
if callback not in litellm._async_failure_callback:
litellm.logging_callback_manager.add_litellm_async_failure_callback(callback)
print_verbose(
f"Initialized litellm callbacks, Async Success Callbacks: {litellm._async_success_callback}"
)
if (
len(litellm.input_callback) > 0 or len(litellm.success_callback) > 0 or len(litellm.failure_callback) > 0
) and len(callback_list) == 0:
callback_list = list(set(litellm.input_callback + litellm.success_callback + litellm.failure_callback))
callbacks_to_set_up: Final = (
list(set(litellm.input_callback + litellm.success_callback + litellm.failure_callback))
if len(callback_list) == 0
else []
)
if callbacks_to_set_up:
callback_list = callbacks_to_set_up
## ASYNC CALLBACKS - safety net for callbacks added via direct append
if len(litellm.input_callback) > 0:
removed_async_items = []
for index, callback in enumerate(litellm.input_callback):
if coroutine_checker.is_async_callable(callback):
litellm._async_input_callback.append(callback)
removed_async_items.append(index)
# Pop the async items from input_callback in reverse order to avoid index issues
for index in reversed(removed_async_items):
litellm.input_callback.pop(index)
if len(litellm.success_callback) > 0:
removed_async_items = []
for index, callback in enumerate(litellm.success_callback):
if coroutine_checker.is_async_callable(callback):
litellm.logging_callback_manager.add_litellm_async_success_callback(callback)
removed_async_items.append(index)
elif callback == "dynamodb" or callback == "openmeter":
# dynamo is an async callback, it's used for the proxy and needs to be async
# we only support async dynamo db logging for acompletion/aembedding since that's used on proxy
litellm.logging_callback_manager.add_litellm_async_success_callback(callback)
removed_async_items.append(index)
elif callback in litellm._known_custom_logger_compatible_callbacks and isinstance(callback, str):
add_custom_logger_callback_to_specific_event(callback, "success")
# Pop the async items from success_callback in reverse order to avoid index issues
for index in reversed(removed_async_items):
litellm.success_callback.pop(index)
if len(litellm.failure_callback) > 0:
removed_async_items = []
for index, callback in enumerate(litellm.failure_callback):
if coroutine_checker.is_async_callable(callback):
litellm.logging_callback_manager.add_litellm_async_failure_callback(callback)
removed_async_items.append(index)
elif callback in litellm._known_custom_logger_compatible_callbacks and isinstance(callback, str):
add_custom_logger_callback_to_specific_event(callback, "failure")
# Pop the async items from failure_callback in reverse order to avoid index issues
for index in reversed(removed_async_items):
litellm.failure_callback.pop(index)
if callbacks_to_set_up:
get_set_callbacks: Final[Callable[[], Callable[..., None]]] = litellm_utils.get_set_callbacks
get_set_callbacks()(callback_list=callback_list, function_id=function_id)
## ASYNC CALLBACKS - safety net for callbacks added via direct append
if len(litellm.input_callback) > 0:
removed_async_items = []
for index, callback in enumerate(litellm.input_callback):
if coroutine_checker.is_async_callable(callback):
litellm._async_input_callback.append(callback)
removed_async_items.append(index)
# Pop the async items from input_callback in reverse order to avoid index issues
for index in reversed(removed_async_items):
litellm.input_callback.pop(index)
if len(litellm.success_callback) > 0:
removed_async_items = []
for index, callback in enumerate(litellm.success_callback):
if coroutine_checker.is_async_callable(callback):
litellm.logging_callback_manager.add_litellm_async_success_callback(callback)
removed_async_items.append(index)
elif callback == "dynamodb" or callback == "openmeter":
# dynamo is an async callback, it's used for the proxy and needs to be async
# we only support async dynamo db logging for acompletion/aembedding since that's used on proxy
litellm.logging_callback_manager.add_litellm_async_success_callback(callback)
removed_async_items.append(index)
elif callback in litellm._known_custom_logger_compatible_callbacks and isinstance(callback, str):
add_custom_logger_callback_to_specific_event(callback, "success")
# Pop the async items from success_callback in reverse order to avoid index issues
for index in reversed(removed_async_items):
litellm.success_callback.pop(index)
if len(litellm.failure_callback) > 0:
removed_async_items = []
for index, callback in enumerate(litellm.failure_callback):
if coroutine_checker.is_async_callable(callback):
litellm.logging_callback_manager.add_litellm_async_failure_callback(callback)
removed_async_items.append(index)
elif callback in litellm._known_custom_logger_compatible_callbacks and isinstance(callback, str):
add_custom_logger_callback_to_specific_event(callback, "failure")
# Pop the async items from failure_callback in reverse order to avoid index issues
for index in reversed(removed_async_items):
litellm.failure_callback.pop(index)
get_set_callbacks()(callback_list=callbacks_to_set_up, function_id=function_id)
### DYNAMIC CALLBACKS ###
dynamic_success_callbacks: list[str | Callable | CustomLogger] | None = None
dynamic_async_success_callbacks: list[str | Callable | CustomLogger] | None = None
dynamic_failure_callbacks: list[str | Callable | CustomLogger] | None = None
dynamic_async_failure_callbacks: Final[list[str | Callable | CustomLogger] | None] = None
dynamic_success_callbacks: list[str | Callable[..., object] | CustomLogger] | None = None
dynamic_async_success_callbacks: list[str | Callable[..., object] | CustomLogger] | None = None
dynamic_failure_callbacks: list[str | Callable[..., object] | CustomLogger] | None = None
dynamic_async_failure_callbacks: Final[list[str | Callable[..., object] | CustomLogger] | None] = None
if kwargs.get("success_callback", None) is not None and isinstance(kwargs["success_callback"], list):
removed_async_items = []
for index, callback in enumerate(kwargs["success_callback"]):
@ -1579,6 +1593,22 @@ def _get_wrapper_timeout(kwargs: dict[str, object]) -> object:
return kwargs.get("timeout") if stream_timeout is None else stream_timeout
def _runs_as_async_success_callback(callback: str | Callable[..., object] | CustomLogger) -> bool:
get_coroutine_checker: Final = litellm_utils.get_coroutine_checker
return get_coroutine_checker().is_async_callable(callback) or callback in ("dynamodb", "s3")
def split_dynamic_success_callbacks(
callbacks: Sequence[str | Callable[..., object] | CustomLogger],
) -> tuple[
tuple[str | Callable[..., object] | CustomLogger, ...], tuple[str | Callable[..., object] | CustomLogger, ...]
]:
return (
tuple(callback for callback in callbacks if not _runs_as_async_success_callback(callback)),
tuple(callback for callback in callbacks if _runs_as_async_success_callback(callback)),
)
def check_coroutine(value) -> bool:
get_coroutine_checker: Final = litellm_utils.get_coroutine_checker
return get_coroutine_checker().is_async_callable(value)
@ -2166,7 +2196,11 @@ def client(original_function):
# Sync logging_obj.stream after deployment hooks (they may convert it).
_hook_stream: Final = kwargs.get("stream")
if _hook_stream is not None and logging_obj.stream != _hook_stream:
if (
_hook_stream is not None
and logging_obj.stream != _hook_stream
and call_type not in _NEVER_STREAMING_CALL_TYPES
):
logging_obj.stream = _hook_stream
kwargs["litellm_logging_obj"] = logging_obj
@ -2510,6 +2544,15 @@ _STREAMING_CALL_TYPES: Final = frozenset(
}
)
_NEVER_STREAMING_CALL_TYPES: Final = frozenset(
{
CallTypes.extraction,
CallTypes.aextraction,
CallTypes.extraction.value,
CallTypes.aextraction.value,
}
)
def _is_streaming_request(
kwargs: dict[str, object],
@ -2521,6 +2564,8 @@ def _is_streaming_request(
- if "stream=True" in kwargs (litellm chat completion, litellm text completion, litellm messages)
- if call_type is generate_content_stream or agenerate_content_stream (litellm google genai)
"""
if call_type in _NEVER_STREAMING_CALL_TYPES:
return False
if "stream" in kwargs and kwargs["stream"] is True:
return True
return call_type in _STREAMING_CALL_TYPES
@ -2565,19 +2610,14 @@ def _select_tokenizer_helper(model: str) -> SelectTokenizerResponse:
try:
result: Final = _return_huggingface_tokenizer(model)
if result is not None:
if result is not None and not isinstance(result, _HuggingFaceTokenizerLoadFailure):
return result
except Exception as e:
from litellm.rust_bridge.fork_guard import ForkedAfterNativeRuntimeStarted, ProcessReservedForForking
if isinstance(e, (ForkedAfterNativeRuntimeStarted, ProcessReservedForForking)):
raise
verbose_logger.warning(
"Falling back to tiktoken for %s; token counts may be approximate. "
"For Python Hugging Face tokenization, install tokenizers and huggingface-hub. Error: %s",
json.dumps(redact_secrets(model)),
json.dumps(redact_secrets(str(e))),
)
_warn_huggingface_tokenizer_fallback(model, str(e))
# default - tiktoken
return _return_openai_tokenizer(model)
@ -2617,15 +2657,41 @@ def huggingface_tokenizer_kind(model: str) -> HuggingFaceTokenizerKind | None:
return None
def _return_huggingface_tokenizer(model: str) -> SelectTokenizerResponse | None:
@dataclass(frozen=True, slots=True)
class _HuggingFaceTokenizerLoadFailure:
reason: str
def _warn_huggingface_tokenizer_fallback(model: str, reason: str) -> None:
verbose_logger.warning(
"Falling back to tiktoken for %s; token counts may be approximate. "
"For Python Hugging Face tokenization, install tokenizers and huggingface-hub. Error: %s",
json.dumps(redact_secrets(model)),
json.dumps(redact_secrets(reason)),
)
def _return_huggingface_tokenizer(model: str) -> SelectTokenizerResponse | _HuggingFaceTokenizerLoadFailure | None:
kind: Final = huggingface_tokenizer_kind(model)
if kind is None:
return None
backend: Final = _huggingface_tokenizer_backend()
key: Final = _huggingface_tokenizer_failure_key(kind, backend)
with _BUNDLED_TOKENIZER_LOAD_LOCKS.get(kind, nullcontext()):
tokenizer: Final = _load_huggingface_tokenizer(kind, _huggingface_tokenizer_backend())
cached_failure: Final = _HUGGINGFACE_TOKENIZER_LOAD_FAILURES.get_cache(key)
if isinstance(cached_failure, _HuggingFaceTokenizerLoadFailure):
return cached_failure
tokenizer: Final = _load_huggingface_tokenizer_recording_failure(kind, backend, key)
if isinstance(tokenizer, _HuggingFaceTokenizerLoadFailure):
_warn_huggingface_tokenizer_fallback(model, tokenizer.reason)
return tokenizer
return {"type": "huggingface_tokenizer", "tokenizer": tokenizer}
def _huggingface_tokenizer_failure_key(kind: HuggingFaceTokenizerKind, backend: Decision) -> str:
return f"{kind}:{backend.name}"
@lru_cache(maxsize=DEFAULT_MAX_LRU_CACHE_SIZE)
def _load_huggingface_tokenizer(kind: HuggingFaceTokenizerKind, backend: Decision) -> HuggingFace:
"""One tokenizer per kind and backend; `backend` is the cache key, the dispatch re-derives it."""
@ -2648,6 +2714,21 @@ def _load_huggingface_tokenizer(kind: HuggingFaceTokenizerKind, backend: Decisio
assert_never(kind)
def _load_huggingface_tokenizer_recording_failure(
kind: HuggingFaceTokenizerKind, backend: Decision, key: str
) -> HuggingFace | _HuggingFaceTokenizerLoadFailure:
try:
return _load_huggingface_tokenizer(kind, backend)
except Exception as error:
from litellm.rust_bridge.fork_guard import ForkedAfterNativeRuntimeStarted, ProcessReservedForForking
if isinstance(error, (ForkedAfterNativeRuntimeStarted, ProcessReservedForForking)):
raise
failure: Final = _HuggingFaceTokenizerLoadFailure(reason=str(error))
_HUGGINGFACE_TOKENIZER_LOAD_FAILURES.set_cache(key, failure, ttl=HUGGINGFACE_TOKENIZER_LOAD_RETRY_SECONDS)
return failure
def encode(model="", text="", custom_tokenizer: dict | None = None):
"""
Encodes the given text using the specified model.
@ -7893,12 +7974,16 @@ class TextCompletionStreamWrapper:
text_choices: Final = TextChoices()
if isinstance(chunk, Choices): # chunk should always be of type StreamingChoices
raise Exception
delta: Final = chunk["choices"][0]["delta"]
text_choices["text"] = delta["content"]
text_choices["reasoning_content"] = delta.get("reasoning_content")
text_choices["index"] = chunk["choices"][0]["index"]
text_choices["finish_reason"] = chunk["choices"][0]["finish_reason"]
response["choices"] = [text_choices]
if chunk["choices"]:
delta: Final = chunk["choices"][0]["delta"]
text_choices["text"] = delta["content"]
text_choices["reasoning_content"] = delta.get("reasoning_content")
text_choices["index"] = chunk["choices"][0]["index"]
text_choices["finish_reason"] = chunk["choices"][0]["finish_reason"]
response["choices"] = [text_choices]
else:
response["choices"] = []
# only pass usage when stream_options["include_usage"] is True
if self.stream_options and self.stream_options.get("include_usage", False) is True:
@ -8263,6 +8348,26 @@ def trim_messages(
from litellm.caching.in_memory_cache import InMemoryCache
class _HuggingFaceTokenizerLoadFailureCache(Protocol):
def get_cache(self, key: str) -> object | None: ...
def set_cache(self, key: str, value: object, ttl: float | None = None) -> None: ...
def flush_cache(self) -> None: ...
_HUGGINGFACE_TOKENIZER_LOAD_FAILURES: Final[_HuggingFaceTokenizerLoadFailureCache] = (
cast( # cast-ok: cache methods are untyped
_HuggingFaceTokenizerLoadFailureCache,
InMemoryCache(
max_size_in_memory=16,
default_ttl=HUGGINGFACE_TOKENIZER_LOAD_RETRY_SECONDS,
clock=lambda: time.monotonic(),
),
)
)
class AvailableModelsCache(InMemoryCache):
def __init__(self, ttl_seconds: int = 300, max_size: int = 1000):
super().__init__(ttl_seconds, max_size)
@ -9539,6 +9644,12 @@ class ProviderConfigManager:
return litellm.AzureAIDecisionsConfig()
return None
@staticmethod
def get_provider_extraction_config(model: str, provider: LlmProviders) -> BaseExtractionConfig | None:
if provider == LlmProviders.CLOUDFLARE:
return litellm.CloudflareExtractionConfig()
return None
@staticmethod
def get_provider_anthropic_messages_config(
model: str,

View file

@ -16580,6 +16580,24 @@
"source": "https://developers.cloudflare.com/workers-ai/platform/pricing/",
"supports_vision": true
},
"cloudflare/clairic": {
"input_cost_per_token": 1e-07,
"litellm_provider": "cloudflare",
"max_input_tokens": 65536,
"mode": "extraction",
"output_cost_per_token": 0.0,
"source": "https://developers.cloudflare.com/workers-ai/models/clairic/",
"supports_vision": true
},
"cloudflare/@cf/cloudflare/clairic": {
"input_cost_per_token": 1e-07,
"litellm_provider": "cloudflare",
"max_input_tokens": 65536,
"mode": "extraction",
"output_cost_per_token": 0.0,
"source": "https://developers.cloudflare.com/workers-ai/models/clairic/",
"supports_vision": true
},
"cloudflare/@cf/meta/llama-2-7b-chat-fp16": {
"input_cost_per_token": 1.923e-06,
"litellm_provider": "cloudflare",
@ -21903,6 +21921,7 @@
],
"supports_function_calling": true,
"supports_reasoning": true,
"supports_sampling_params": false,
"supports_tool_choice": true,
"supports_vision": true
},
@ -21927,6 +21946,7 @@
],
"supports_function_calling": true,
"supports_reasoning": true,
"supports_sampling_params": false,
"supports_tool_choice": true,
"supports_vision": true
},
@ -21958,6 +21978,7 @@
"supports_function_calling": true,
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_sampling_params": false,
"supports_tool_choice": true,
"supports_vision": true
},
@ -22021,6 +22042,7 @@
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_tool_choice": true,
"supports_vision": true
},
@ -29417,6 +29439,7 @@
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_url_context": true,
@ -30139,6 +30162,7 @@
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_url_context": true,
@ -30200,6 +30224,7 @@
"supports_reasoning": true,
"supports_minimal_reasoning_effort": false,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_url_context": true,
@ -30268,6 +30293,7 @@
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_url_context": true,
@ -30317,6 +30343,7 @@
"supports_reasoning": true,
"supports_minimal_reasoning_effort": false,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": false,
"supports_url_context": false,
@ -31743,6 +31770,7 @@
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_url_context": true,
@ -31927,6 +31955,7 @@
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_url_context": true,
@ -31989,6 +32018,7 @@
"supports_reasoning": true,
"supports_minimal_reasoning_effort": false,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_url_context": true,
@ -32051,6 +32081,7 @@
"supports_reasoning": true,
"supports_minimal_reasoning_effort": false,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_url_context": true,
@ -32420,6 +32451,7 @@
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_url_context": true,
@ -32481,6 +32513,7 @@
"supports_reasoning": true,
"supports_minimal_reasoning_effort": false,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_url_context": true,
@ -32549,6 +32582,7 @@
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_url_context": true,
@ -32598,6 +32632,7 @@
"supports_reasoning": true,
"supports_minimal_reasoning_effort": false,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": false,
"supports_url_context": false,
@ -33192,6 +33227,7 @@
"supports_function_calling": true,
"supports_parallel_function_calling": true,
"supports_reasoning": true,
"supports_sampling_params": false,
"supports_vision": true
},
"github_copilot/gemini-3.8-flash": {
@ -33210,6 +33246,7 @@
"supports_function_calling": true,
"supports_parallel_function_calling": true,
"supports_reasoning": true,
"supports_sampling_params": false,
"supports_vision": true
},
"github_copilot/gpt-3.5-turbo": {
@ -52605,6 +52642,7 @@
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_url_context": true,
@ -65823,6 +65861,7 @@
"supports_tool_choice": true,
"supports_response_schema": true,
"supports_reasoning": true,
"supports_sampling_params": false,
"supports_vision": true,
"source": "https://deepinfra.com/pricing",
"supports_audio_input": true
@ -69264,6 +69303,7 @@
"perplexity/google/gemini-3.5-flash-lite": {
"litellm_provider": "perplexity",
"mode": "responses",
"supports_sampling_params": false,
"supports_web_search": true,
"supports_function_calling": true,
"input_cost_per_token": 3e-07,
@ -69274,6 +69314,7 @@
"perplexity/google/gemini-3.6-flash": {
"litellm_provider": "perplexity",
"mode": "responses",
"supports_sampling_params": false,
"supports_web_search": true,
"supports_function_calling": true,
"input_cost_per_token": 1.5e-06,
@ -69284,6 +69325,7 @@
"perplexity/google/gemini-3.7-flash": {
"litellm_provider": "perplexity",
"mode": "responses",
"supports_sampling_params": false,
"supports_web_search": true,
"supports_function_calling": true,
"input_cost_per_token": 7.5e-07,
@ -71375,6 +71417,7 @@
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_vision": true,
"supports_web_search": true
},
@ -71392,6 +71435,7 @@
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_vision": true,
"supports_web_search": true
},
@ -74322,6 +74366,7 @@
"perplexity/google/gemini-3.8-flash": {
"litellm_provider": "perplexity",
"mode": "responses",
"supports_sampling_params": false,
"input_cost_per_token": 7.5e-07,
"output_cost_per_token": 3.75e-06,
"cache_read_input_token_cost": 7.5e-08,
@ -75064,12 +75109,10 @@
"gemini-nano-banana-2.1": {
"cache_read_input_token_cost": 1.5e-07,
"cache_read_input_token_cost_batches": 7.5e-08,
"cache_read_input_token_cost_flex": 7.5e-08,
"cache_read_input_token_cost_priority": 2.7e-07,
"input_cost_per_image": 0.00168,
"input_cost_per_token": 1.5e-06,
"input_cost_per_token_batches": 7.5e-07,
"input_cost_per_token_flex": 7.5e-07,
"input_cost_per_token_priority": 2.7e-06,
"litellm_provider": "vertex_ai-language-models",
"max_input_tokens": 131072,
@ -75080,7 +75123,6 @@
"output_cost_per_image_token": 3e-05,
"output_cost_per_token": 7.5e-06,
"output_cost_per_token_batches": 3.75e-06,
"output_cost_per_token_flex": 3.75e-06,
"output_cost_per_token_priority": 1.35e-05,
"search_context_cost_per_query": {"search_context_size_low":0.014,"search_context_size_medium":0.014,"search_context_size_high":0.014},
"source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing",
@ -75113,12 +75155,10 @@
"vertex_ai/gemini-nano-banana-2.1": {
"cache_read_input_token_cost": 1.5e-07,
"cache_read_input_token_cost_batches": 7.5e-08,
"cache_read_input_token_cost_flex": 7.5e-08,
"cache_read_input_token_cost_priority": 2.7e-07,
"input_cost_per_image": 0.00168,
"input_cost_per_token": 1.5e-06,
"input_cost_per_token_batches": 7.5e-07,
"input_cost_per_token_flex": 7.5e-07,
"input_cost_per_token_priority": 2.7e-06,
"litellm_provider": "vertex_ai-language-models",
"max_input_tokens": 131072,
@ -75129,7 +75169,6 @@
"output_cost_per_image_token": 3e-05,
"output_cost_per_token": 7.5e-06,
"output_cost_per_token_batches": 3.75e-06,
"output_cost_per_token_flex": 3.75e-06,
"output_cost_per_token_priority": 1.35e-05,
"search_context_cost_per_query": {"search_context_size_low":0.014,"search_context_size_medium":0.014,"search_context_size_high":0.014},
"source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing",
@ -76929,5 +76968,83 @@
"supported_output_modalities": [
"text"
]
},
"fireworks_ai/accounts/fireworks/models/deepseek-v4p1-flash-1017-us": {
"cache_read_input_token_cost": 9e-09,
"cache_read_input_token_cost_priority": 1.125e-08,
"input_cost_per_token": 4.5e-07,
"input_cost_per_token_priority": 5.625e-07,
"litellm_provider": "fireworks_ai",
"max_input_tokens": 1048576,
"max_output_tokens": 393216,
"max_tokens": 393216,
"mode": "chat",
"output_cost_per_token": 1.8e-06,
"output_cost_per_token_priority": 2.25e-06,
"source": "https://docs.fireworks.ai/serverless/us-only-serverless",
"supports_function_calling": true,
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_tool_choice": true,
"supports_vision": true
},
"wandb/reflection/Beam-1.0-501B-A23B": {
"cache_read_input_token_cost": 1e-07,
"input_cost_per_token": 1e-06,
"litellm_provider": "wandb",
"max_input_tokens": 1049000,
"mode": "chat",
"output_cost_per_token": 3e-06,
"source": "https://www.coreweave.com/products/coreweave-forge/serverless-inference",
"supports_function_calling": true,
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_tool_choice": true,
"supports_vision": false
},
"wandb/zai-org/GLM-5.3": {
"cache_read_input_token_cost": 1.9e-07,
"input_cost_per_token": 1.2e-06,
"litellm_provider": "wandb",
"max_input_tokens": 1049000,
"mode": "chat",
"output_cost_per_token": 4e-06,
"source": "https://www.coreweave.com/products/coreweave-forge/serverless-inference",
"supports_function_calling": true,
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_tool_choice": true,
"supports_vision": false
},
"azure_ai/MAI-Code-1.1-Flash": {
"cache_read_input_token_cost": 2e-08,
"deprecation_date": "2026-12-09",
"input_cost_per_token": 2e-07,
"litellm_provider": "azure_ai",
"max_input_tokens": 256000,
"max_output_tokens": 64000,
"max_tokens": 64000,
"mode": "chat",
"output_cost_per_token": 1.2e-06,
"source": "https://prices.azure.com/api/retail/prices?$filter=serviceName%20eq%20'Foundry%20Models'%20and%20armRegionName%20eq%20'eastus'%20and%20priceType%20eq%20'Consumption'",
"supported_endpoints": [
"/v1/chat/completions",
"/v1/responses"
],
"supported_modalities": [
"text",
"image"
],
"supported_output_modalities": [
"text"
],
"supports_function_calling": true,
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_tool_choice": true,
"supports_vision": true
}
}

Some files were not shown because too many files have changed in this diff Show more