diff --git a/README.md b/README.md index 31a93bf8c67..b1de5c7033a 100644 --- a/README.md +++ b/README.md @@ -380,7 +380,7 @@ Set `LITELLM_PROXY_API_BASE` and `LITELLM_PROXY_API_KEY` and every model call th | [Cerebras (`cerebras`)](https://docs.litellm.ai/docs/providers/cerebras) | ✅ | ✅ | ✅ | | | | | | | | | [Clarifai (`clarifai`)](https://docs.litellm.ai/docs/providers/clarifai) | ✅ | ✅ | ✅ | | | | | | | | | [CLF AI Gateway (`clf_ai_gateway`)](https://docs.litellm.ai/docs/providers/clf_ai_gateway) | ✅ | ✅ | ✅ | | | | | | | | -| [Cloudflare AI Workers (`cloudflare`)](https://docs.litellm.ai/docs/providers/cloudflare_workers) | ✅ | ✅ | ✅ | | | | | | | | +| [Cloudflare AI Workers (`cloudflare`)](https://docs.litellm.ai/docs/providers/cloudflare_workers) | ✅ | ✅ | ✅ | ✅ | | | | | | ✅ | | [Codestral (`codestral`)](https://docs.litellm.ai/docs/providers/codestral) | ✅ | ✅ | ✅ | | | | | | | | | [Cognition (`cognition`)](https://docs.litellm.ai/docs/providers/cognition) | ✅ | ✅ | ✅ | | | | | | | | | [Cohere (`cohere`)](https://docs.litellm.ai/docs/providers/cohere) | ✅ | ✅ | ✅ | ✅ | | | | | | ✅ | diff --git a/litellm/__init__.py b/litellm/__init__.py index 8e5dd68a186..711a7f8ecc7 100644 --- a/litellm/__init__.py +++ b/litellm/__init__.py @@ -1841,6 +1841,12 @@ if TYPE_CHECKING: from .llms.cloudflare.chat.transformation import ( CloudflareChatConfig as CloudflareChatConfig, ) + from .llms.cloudflare.embedding.transformation import ( + CloudflareEmbeddingConfig as CloudflareEmbeddingConfig, + ) + from .llms.cloudflare.rerank.transformation import ( + CloudflareRerankConfig as CloudflareRerankConfig, + ) from .llms.novita.chat.transformation import NovitaConfig as NovitaConfig from .llms.petals.completion.transformation import PetalsConfig as PetalsConfig from .llms.ollama.chat.transformation import OllamaChatConfig as OllamaChatConfig diff --git a/litellm/_lazy_imports_registry.py b/litellm/_lazy_imports_registry.py index bf903e23bc0..1ef819c76c3 100644 --- a/litellm/_lazy_imports_registry.py +++ b/litellm/_lazy_imports_registry.py @@ -192,6 +192,8 @@ LLM_CONFIG_NAMES: Final = ( "LlamaAPIConfig", "TogetherAITextCompletionConfig", "CloudflareChatConfig", + "CloudflareEmbeddingConfig", + "CloudflareRerankConfig", "NovitaConfig", "PetalsConfig", "OllamaChatConfig", @@ -797,6 +799,14 @@ _LLM_CONFIGS_IMPORT_MAP: Final = { ".llms.cloudflare.chat.transformation", "CloudflareChatConfig", ), + "CloudflareEmbeddingConfig": ( + ".llms.cloudflare.embedding.transformation", + "CloudflareEmbeddingConfig", + ), + "CloudflareRerankConfig": ( + ".llms.cloudflare.rerank.transformation", + "CloudflareRerankConfig", + ), "NovitaConfig": (".llms.novita.chat.transformation", "NovitaConfig"), "PetalsConfig": (".llms.petals.completion.transformation", "PetalsConfig"), "OllamaChatConfig": (".llms.ollama.chat.transformation", "OllamaChatConfig"), diff --git a/litellm/llms/cloudflare/chat/transformation.py b/litellm/llms/cloudflare/chat/transformation.py index 2bbfd2010cc..51763876bdf 100644 --- a/litellm/llms/cloudflare/chat/transformation.py +++ b/litellm/llms/cloudflare/chat/transformation.py @@ -38,7 +38,7 @@ class CloudflareChatConfig(OpenAIGPTConfig): stream: bool | None = None, ) -> str: return super().get_complete_url( - api_base=self._resolve_api_base(api_base), + api_base=self.resolve_api_base(api_base), api_key=api_key, model=model, optional_params=optional_params, @@ -47,7 +47,7 @@ class CloudflareChatConfig(OpenAIGPTConfig): ) @staticmethod - def _resolve_api_base(api_base: str | None) -> str: + def resolve_api_base(api_base: str | None) -> str: if not api_base: account_id: Final = normalize_nonempty_secret_str(get_secret_str("CLOUDFLARE_ACCOUNT_ID")) if account_id is None: diff --git a/litellm/llms/cloudflare/embedding/transformation.py b/litellm/llms/cloudflare/embedding/transformation.py new file mode 100644 index 00000000000..2d75c8dab98 --- /dev/null +++ b/litellm/llms/cloudflare/embedding/transformation.py @@ -0,0 +1,31 @@ +from collections.abc import Mapping +from typing import Final + +import httpx + +from litellm.llms.cloudflare.chat.transformation import CloudflareChatConfig, CloudflareError +from litellm.llms.vercel_ai_gateway.embedding.transformation import VercelAIGatewayEmbeddingConfig + + +class CloudflareEmbeddingConfig(VercelAIGatewayEmbeddingConfig): + def get_complete_url( + self, + api_base: str | None, + api_key: str | None, + model: str, + optional_params: Mapping[object, object], + litellm_params: Mapping[object, object], + stream: bool | None = None, + ) -> str: + resolved_base: Final = CloudflareChatConfig.resolve_api_base(api_base).rstrip("/") + if resolved_base.endswith("/embeddings"): + return resolved_base + return f"{resolved_base}/embeddings" + + def get_error_class( + self, + error_message: str, + status_code: int, + headers: Mapping[str, str] | httpx.Headers, + ) -> CloudflareError: + return CloudflareError(status_code=status_code, message=error_message) diff --git a/litellm/llms/cloudflare/rerank/transformation.py b/litellm/llms/cloudflare/rerank/transformation.py new file mode 100644 index 00000000000..4cc9b4f374a --- /dev/null +++ b/litellm/llms/cloudflare/rerank/transformation.py @@ -0,0 +1,286 @@ +import json +from collections.abc import Mapping, Sequence +from typing import Final, cast # noqa: TID251 # json boundaries only expose untyped mappings + +import httpx +from typing_extensions import NotRequired, ReadOnly, TypedDict + +from litellm._uuid import uuid +from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj +from litellm.litellm_core_utils.url_utils import encode_url_path_segments +from litellm.llms.base_llm.chat.transformation import BaseLLMException +from litellm.llms.base_llm.rerank.transformation import BaseRerankConfig +from litellm.secret_managers.main import get_secret_str, normalize_nonempty_secret_str +from litellm.types.rerank import RerankResponse, RerankResponseDocument, RerankResponseResult + +from ..chat.transformation import CloudflareError + + +class CloudflareRerankContext(TypedDict): + text: ReadOnly[str] + + +class CloudflareRerankRequest(TypedDict): + query: ReadOnly[str] + contexts: ReadOnly[Sequence[CloudflareRerankContext]] + top_k: NotRequired[ReadOnly[object]] + + +class CohereRerankParams(TypedDict): + query: ReadOnly[str] + documents: ReadOnly[Sequence[str | Mapping[str, object]]] + top_n: NotRequired[ReadOnly[int]] + return_documents: NotRequired[ReadOnly[bool]] + + +class LoggingAdditionalArgs(TypedDict): + complete_input_dict: ReadOnly[Mapping[str, object]] + + +class EmptyRequestData(TypedDict, total=False): + pass + + +EMPTY_REQUEST_DATA: Final[Mapping[str, object]] = EmptyRequestData() + + +class CloudflareRerankConfig(BaseRerankConfig): + @staticmethod + def _default_api_base() -> str: + account_id: Final = normalize_nonempty_secret_str(get_secret_str("CLOUDFLARE_ACCOUNT_ID")) + if account_id is None: + raise ValueError("Missing CLOUDFLARE_ACCOUNT_ID - set CLOUDFLARE_ACCOUNT_ID or pass api_base explicitly") + return f"https://api.cloudflare.com/client/v4/accounts/{account_id}/ai/run" + + @staticmethod + def _document_to_context( + document: object, + ) -> CloudflareRerankContext: + if isinstance(document, str): + return CloudflareRerankContext(text=document) + if not isinstance(document, Mapping): + raise TypeError("Cloudflare rerank documents must be strings or dictionaries") + + document_mapping: Final = cast(Mapping[str, object], document) # cast-ok: isinstance proves only a bare Mapping + text: Final = document_mapping.get("text") + return CloudflareRerankContext(text=text if isinstance(text, str) else json.dumps(document_mapping)) + + @staticmethod + def _response_items( + response_json: Mapping[str, object], + status_code: int, + ) -> tuple[object, ...]: + if response_json.get("success") is False: + raise CloudflareError( + status_code=status_code, + message=str(response_json.get("errors") or response_json), + ) + + result: Final = response_json.get("result", response_json) + result_mapping: Final = ( + cast(Mapping[str, object], result) # cast-ok: isinstance proves only a bare Mapping + if isinstance(result, Mapping) + else None + ) + response: Final = result_mapping.get("response") if result_mapping is not None else None + if not isinstance(response, list): + raise CloudflareError( + status_code=status_code, + message=f"No response in Cloudflare rerank result: {response_json}", + ) + return tuple(cast(Sequence[object], response)) # cast-ok: JSON list narrows only to list[Unknown] + + @staticmethod + def _transform_response_item( + item: object, + documents: Sequence[object], + return_documents: bool, + ) -> RerankResponseResult: + if not isinstance(item, Mapping): + raise TypeError("Invalid item in Cloudflare rerank response") + + item_mapping: Final = cast(Mapping[str, object], item) # cast-ok: isinstance proves only a bare Mapping + index: Final = item_mapping.get("id") + score: Final = item_mapping.get("score") + if not isinstance(index, int) or not isinstance(score, (int, float)): + raise TypeError("Invalid item in Cloudflare rerank response") + + if return_documents and index < len(documents): + document: Final = CloudflareRerankConfig._document_to_context(documents[index]) + return RerankResponseResult( + index=index, + relevance_score=float(score), + document=RerankResponseDocument(text=document["text"]), + ) + return RerankResponseResult(index=index, relevance_score=float(score)) + + def validate_environment( # pyright: ignore[reportIncompatibleMethodOverride] # base annotates the return as dict; the read-only Mapping is intentional + self, + headers: Mapping[str, object], + model: str, + api_key: str | None = None, + optional_params: Mapping[str, object] | None = None, + litellm_params: Mapping[str, object] | None = None, + ) -> Mapping[str, object]: + resolved_api_key: Final = api_key or get_secret_str("CLOUDFLARE_API_KEY") + if resolved_api_key is None: + raise ValueError("Missing Cloudflare API Key - set CLOUDFLARE_API_KEY or pass api_key explicitly") + default_headers: Final = { + "Authorization": f"Bearer {resolved_api_key}", + "accept": "application/json", + "content-type": "application/json", + } + return {**default_headers, **headers} + + def get_complete_url( + self, + api_base: str | None, + model: str, + optional_params: Mapping[str, object] | None = None, + ) -> str: + cleaned: Final = (api_base or self._default_api_base()).rstrip("/") + encoded_model: Final = encode_url_path_segments(model, field_name="model") + if cleaned.endswith(f"/{encoded_model}"): + return cleaned + if cleaned.endswith("/ai/v1"): + return f"{cleaned[: -len('/ai/v1')]}/ai/run/{encoded_model}" + if cleaned.endswith("/ai/run"): + return f"{cleaned}/{encoded_model}" + return f"{cleaned}/ai/run/{encoded_model}" + + def get_supported_cohere_rerank_params( # pyright: ignore[reportIncompatibleMethodOverride] # base annotates the return as list; a read-only tuple is intentional + self, + model: str, + ) -> Sequence[str]: + return ( + "query", + "documents", + "top_n", + "return_documents", + ) + + def map_cohere_rerank_params( # pyright: ignore[reportIncompatibleMethodOverride] # base annotates the return as dict; the read-only Mapping is intentional + self, + non_default_params: Mapping[str, object], + model: str, + drop_params: bool, + query: str, + documents: Sequence[str | Mapping[str, object]], + custom_llm_provider: str | None = None, + top_n: int | None = None, + rank_fields: Sequence[str] | None = None, + return_documents: bool | None = True, + max_chunks_per_doc: int | None = None, + max_tokens_per_doc: int | None = None, + instruction: str | None = None, + ) -> Mapping[str, object]: + if top_n is None and return_documents is None: + return CohereRerankParams( + query=query, + documents=documents, + ) + if top_n is None: + assert return_documents is not None + return CohereRerankParams( + query=query, + documents=documents, + return_documents=return_documents, + ) + if return_documents is None: + return CohereRerankParams( + query=query, + documents=documents, + top_n=top_n, + ) + return CohereRerankParams( + query=query, + documents=documents, + top_n=top_n, + return_documents=return_documents, + ) + + def transform_rerank_request( # pyright: ignore[reportIncompatibleMethodOverride] # base annotates the return as dict; the read-only Mapping is intentional + self, + model: str, + optional_rerank_params: Mapping[str, object], + headers: Mapping[str, object], + litellm_params: Mapping[str, object] | None = None, + ) -> Mapping[str, object]: + query: Final = optional_rerank_params.get("query") + raw_documents: Final = optional_rerank_params.get("documents") + if not isinstance(query, str) or not query: + raise ValueError("query is required for Cloudflare rerank") + if not isinstance(raw_documents, Sequence) or isinstance(raw_documents, str) or not raw_documents: + raise ValueError("documents is required for Cloudflare rerank") + + documents: Final = cast(Sequence[object], raw_documents) # cast-ok: isinstance proves only a bare Sequence + contexts: Final = tuple(self._document_to_context(document) for document in documents) + top_n: Final = optional_rerank_params.get("top_n") + if top_n is None: + return CloudflareRerankRequest(query=query, contexts=contexts) + return CloudflareRerankRequest(query=query, contexts=contexts, top_k=top_n) + + def transform_rerank_response( + self, + model: str, + raw_response: httpx.Response, + model_response: RerankResponse, + logging_obj: LiteLLMLoggingObj, + api_key: str | None = None, + request_data: Mapping[str, object] | None = None, + optional_params: Mapping[str, object] | None = None, + litellm_params: Mapping[str, object] | None = None, + ) -> RerankResponse: + resolved_request_data: Final = request_data or EMPTY_REQUEST_DATA + try: + response_json: Final = cast(object, raw_response.json()) # cast-ok: httpx Response.json is untyped + except ValueError: + raise CloudflareError( + status_code=raw_response.status_code, + message=raw_response.text, + ) + if not isinstance(response_json, Mapping): + raise CloudflareError( + status_code=raw_response.status_code, + message=f"Invalid Cloudflare rerank response: {response_json}", + ) + response_data: Final = cast(Mapping[str, object], response_json) # cast-ok: bare Mapping + + logging_obj.post_call( # pyright: ignore[reportUnknownMemberType] # logging object is legacy-untyped + input=resolved_request_data.get("query"), + api_key=api_key, + additional_args=LoggingAdditionalArgs(complete_input_dict=resolved_request_data), + original_response=response_data, + ) + + resolved_optional_params: Final = optional_params or EMPTY_REQUEST_DATA + raw_documents: Final = resolved_optional_params.get("documents") + documents: Final = ( + cast(Sequence[object], raw_documents) # cast-ok: isinstance proves only a bare Sequence + if isinstance(raw_documents, Sequence) and not isinstance(raw_documents, str) + else () + ) + return_documents: Final = resolved_optional_params.get("return_documents") is not False + results: Final = tuple( + self._transform_response_item( + item, + documents=documents, + return_documents=return_documents, + ) + for item in self._response_items( + response_json=response_data, + status_code=raw_response.status_code, + ) + ) + return RerankResponse( + id=str(response_data.get("id") or uuid.uuid4()), + results=list(results), + ) + + def get_error_class( + self, + error_message: str, + status_code: int, + headers: Mapping[str, object] | httpx.Headers, + ) -> BaseLLMException: + return CloudflareError(status_code=status_code, message=error_message) diff --git a/litellm/llms/custom_httpx/llm_http_handler.py b/litellm/llms/custom_httpx/llm_http_handler.py index df82891f789..48405b040d5 100644 --- a/litellm/llms/custom_httpx/llm_http_handler.py +++ b/litellm/llms/custom_httpx/llm_http_handler.py @@ -1611,6 +1611,9 @@ class BaseLLMHTTPHandler: api_key=api_key, timeout=timeout, client=client, + optional_rerank_params=cast( # cast-ok: untyped caller dict + Mapping[str, object], optional_rerank_params + ), ) if client is None or not isinstance(client, HTTPHandler): @@ -1638,6 +1641,7 @@ class BaseLLMHTTPHandler: logging_obj=logging_obj, api_key=api_key, request_data=data, + optional_params=optional_rerank_params, ) async def arerank( @@ -1653,6 +1657,7 @@ class BaseLLMHTTPHandler: api_key: str | None = None, timeout: float | httpx.Timeout | None = None, client: HTTPHandler | AsyncHTTPHandler | None = None, + optional_rerank_params: Mapping[str, object] | None = None, ) -> RerankResponse: if client is None or not isinstance(client, AsyncHTTPHandler): async_httpx_client = get_async_httpx_client(llm_provider=litellm.LlmProviders(custom_llm_provider)) @@ -1669,6 +1674,7 @@ class BaseLLMHTTPHandler: except Exception as e: raise self._handle_error(e=e, provider_config=provider_config) + resolved_optional_rerank_params: Final = dict(optional_rerank_params or {}) return provider_config.transform_rerank_response( model=model, raw_response=response, @@ -1676,6 +1682,7 @@ class BaseLLMHTTPHandler: logging_obj=logging_obj, api_key=api_key, request_data=request_data, + optional_params=resolved_optional_rerank_params, ) def _prepare_decisions_request( diff --git a/litellm/main.py b/litellm/main.py index bdb35823757..dae7880fc8b 100644 --- a/litellm/main.py +++ b/litellm/main.py @@ -6814,6 +6814,30 @@ def _embedding_headers(param: str, configured: object, model: str, drop_params: ) +def _resolve_vercel_or_cloudflare_embedding_credentials( + custom_llm_provider: str, + api_base: str | None, + api_key: str | None, +) -> tuple[str | None, str | None]: + if custom_llm_provider == "cloudflare": + resolved_api_key: Final = ( + api_key or litellm.cloudflare_api_key or litellm.api_key or get_secret_str("CLOUDFLARE_API_KEY") + ) + if resolved_api_key is None: + raise ValueError("Missing Cloudflare API Key - no key is set in the environment or request parameters") + return api_base or litellm.api_base or get_secret_str("CLOUDFLARE_API_BASE"), resolved_api_key + return ( + api_base + or litellm.api_base + or get_secret_str("VERCEL_AI_GATEWAY_API_BASE") + or "https://ai-gateway.vercel.sh/v1", + api_key + or litellm.api_key + or get_secret_str("VERCEL_AI_GATEWAY_API_KEY") + or get_secret_str("VERCEL_OIDC_TOKEN"), + ) + + # fmt: off # Overload for when aembedding=True (returns coroutine) @@ -7275,21 +7299,10 @@ def embedding( litellm_params=litellm_params_dict, headers=headers, ) - elif custom_llm_provider == "vercel_ai_gateway": - api_base = ( - api_base - or litellm.api_base - or get_secret_str("VERCEL_AI_GATEWAY_API_BASE") - or "https://ai-gateway.vercel.sh/v1" + elif custom_llm_provider in ("vercel_ai_gateway", "cloudflare"): + api_base, api_key = _resolve_vercel_or_cloudflare_embedding_credentials( + custom_llm_provider, api_base, api_key ) - - api_key = ( - api_key - or litellm.api_key - or get_secret_str("VERCEL_AI_GATEWAY_API_KEY") - or get_secret_str("VERCEL_OIDC_TOKEN") - ) - response = base_llm_http_handler.embedding( model=model, input=input, diff --git a/litellm/rerank_api/main.py b/litellm/rerank_api/main.py index e344ce46be7..8e224f5db1b 100644 --- a/litellm/rerank_api/main.py +++ b/litellm/rerank_api/main.py @@ -39,7 +39,7 @@ async def arerank( query: str, documents: list[str | dict[str, object]], custom_llm_provider: ( - Literal["cohere", "together_ai", "deepinfra", "fireworks_ai", "voyage", "watsonx"] | None + Literal["cohere", "together_ai", "deepinfra", "fireworks_ai", "voyage", "watsonx", "cloudflare"] | None ) = None, top_n: int | None = None, rank_fields: list[str] | None = None, @@ -112,6 +112,7 @@ def rerank( "fireworks_ai", "voyage", "watsonx", + "cloudflare", ] | None ) = None, diff --git a/litellm/utils.py b/litellm/utils.py index 13f22ddc36e..331635f9734 100644 --- a/litellm/utils.py +++ b/litellm/utils.py @@ -10221,6 +10221,8 @@ class ProviderConfigManager: ) return VercelAIGatewayEmbeddingConfig() + elif litellm.LlmProviders.CLOUDFLARE == provider: + return litellm.CloudflareEmbeddingConfig() elif litellm.LlmProviders.GIGACHAT == provider: return litellm.GigaChatEmbeddingConfig() elif litellm.LlmProviders.HOSTED_VLLM == provider: @@ -10238,7 +10240,7 @@ class ProviderConfigManager: return None @staticmethod - def get_provider_rerank_config( + def get_provider_rerank_config( # noqa: C901 # provider dispatch; one branch per rerank provider model: str, provider: LlmProviders, api_base: str | None, @@ -10289,6 +10291,8 @@ class ProviderConfigManager: ) return get_dashscope_family_rerank_config(provider.value) + elif litellm.LlmProviders.CLOUDFLARE == provider: + return litellm.CloudflareRerankConfig() return litellm.CohereRerankConfig() @staticmethod diff --git a/provider_endpoints_support.json b/provider_endpoints_support.json index 00fee0a0d5a..9fe38cf68e1 100644 --- a/provider_endpoints_support.json +++ b/provider_endpoints_support.json @@ -860,12 +860,12 @@ "chat_completions": true, "responses": true, "text_completion": true, - "embeddings": false, + "embeddings": true, "image_generations": false, "image_edits": false, "audio_transcriptions": true, "audio_speech": false, - "rerank": false, + "rerank": true, "ocr": false, "search": false, "moderations": false, diff --git a/tests/unit/llms/cloudflare/test_cloudflare_embedding_transformation.py b/tests/unit/llms/cloudflare/test_cloudflare_embedding_transformation.py new file mode 100644 index 00000000000..c17eed9b913 --- /dev/null +++ b/tests/unit/llms/cloudflare/test_cloudflare_embedding_transformation.py @@ -0,0 +1,151 @@ +import json +from unittest.mock import Mock + +import pytest + +import litellm +from litellm.llms.cloudflare.chat.transformation import CloudflareError +from litellm.llms.cloudflare.embedding.transformation import CloudflareEmbeddingConfig +from litellm.llms.custom_httpx.http_handler import HTTPHandler +from litellm.utils import ProviderConfigManager + + +class _RecordingHTTPHandler(HTTPHandler): + def __init__(self, response): + super().__init__() + self.response = response + self.requests = [] + + def post(self, url: str, **kwargs): + self.requests.append({"url": url, **kwargs}) + return self.response + + +def test_provider_config_manager_returns_cloudflare_embedding_config(): + config = ProviderConfigManager.get_provider_embedding_config( + model="@cf/baai/bge-large-en-v1.5", + provider=litellm.LlmProviders.CLOUDFLARE, + ) + + assert isinstance(config, CloudflareEmbeddingConfig) + + +def test_get_complete_url_defaults_to_openai_compatible_endpoint(monkeypatch): + monkeypatch.setenv("CLOUDFLARE_ACCOUNT_ID", "acct") + config = CloudflareEmbeddingConfig() + + url = config.get_complete_url( + api_base=None, + api_key="cf-key", + model="@cf/baai/bge-large-en-v1.5", + optional_params={}, + litellm_params={}, + ) + + assert url == "https://api.cloudflare.com/client/v4/accounts/acct/ai/v1/embeddings" + + +def test_get_complete_url_is_idempotent_for_full_endpoint(): + config = CloudflareEmbeddingConfig() + + url = config.get_complete_url( + api_base="https://api.cloudflare.com/client/v4/accounts/acct/ai/v1/embeddings", + api_key="cf-key", + model="@cf/baai/bge-large-en-v1.5", + optional_params={}, + litellm_params={}, + ) + + assert url == "https://api.cloudflare.com/client/v4/accounts/acct/ai/v1/embeddings" + + +def test_get_complete_url_migrates_legacy_ai_run_base(): + config = CloudflareEmbeddingConfig() + + url = config.get_complete_url( + api_base="https://api.cloudflare.com/client/v4/accounts/acct/ai/run/", + api_key="cf-key", + model="@cf/baai/bge-large-en-v1.5", + optional_params={}, + litellm_params={}, + ) + + assert url == "https://api.cloudflare.com/client/v4/accounts/acct/ai/v1/embeddings" + + +def test_validate_environment_preserves_extra_headers(): + config = CloudflareEmbeddingConfig() + + headers = config.validate_environment( + headers={"X-Test": "value"}, + model="@cf/baai/bge-large-en-v1.5", + messages=[], + optional_params={}, + litellm_params={}, + api_key="cf-key", + ) + + assert headers == { + "Authorization": "Bearer cf-key", + "Content-Type": "application/json", + "X-Test": "value", + } + + +def test_get_error_class(): + error = CloudflareEmbeddingConfig().get_error_class("failed", 400, {}) + + assert isinstance(error, CloudflareError) + assert error.status_code == 400 + + +def test_embedding_routes_to_cloudflare_openai_compatible_endpoint(monkeypatch): + monkeypatch.setenv("CLOUDFLARE_ACCOUNT_ID", "acct") + response_json = { + "object": "list", + "data": [ + { + "object": "embedding", + "embedding": [0.1, 0.2, 0.3], + "index": 0, + } + ], + "model": "@cf/baai/bge-large-en-v1.5", + "usage": {"prompt_tokens": 2, "total_tokens": 2}, + } + raw_response = Mock() + raw_response.status_code = 200 + raw_response.headers = {"content-type": "application/json"} + raw_response.json.return_value = response_json + raw_response.text = json.dumps(response_json) + client = _RecordingHTTPHandler(raw_response) + + response = litellm.embedding( + model="cloudflare/@cf/baai/bge-large-en-v1.5", + input=["hello"], + api_key="cf-key", + client=client, + caching=False, + ) + + request = client.requests[0] + body = json.loads(request["data"]) + assert request["url"] == "https://api.cloudflare.com/client/v4/accounts/acct/ai/v1/embeddings" + assert request["headers"]["Authorization"] == "Bearer cf-key" + assert body == { + "model": "@cf/baai/bge-large-en-v1.5", + "input": ["hello"], + } + assert response.data[0]["embedding"] == [0.1, 0.2, 0.3] + + +def test_embedding_requires_cloudflare_api_key(monkeypatch): + monkeypatch.setenv("CLOUDFLARE_ACCOUNT_ID", "acct") + monkeypatch.delenv("CLOUDFLARE_API_KEY", raising=False) + + with pytest.raises(litellm.APIConnectionError, match="Missing Cloudflare API Key"): + litellm.embedding( + model="cloudflare/@cf/baai/bge-large-en-v1.5", + input=["hello"], + caching=False, + ) diff --git a/tests/unit/llms/cloudflare/test_cloudflare_rerank_transformation.py b/tests/unit/llms/cloudflare/test_cloudflare_rerank_transformation.py new file mode 100644 index 00000000000..d6e63a9df78 --- /dev/null +++ b/tests/unit/llms/cloudflare/test_cloudflare_rerank_transformation.py @@ -0,0 +1,413 @@ +import json +from unittest.mock import MagicMock, Mock + +import httpx +import pytest + +import litellm +from litellm.llms.cloudflare.chat.transformation import CloudflareError +from litellm.llms.cloudflare.rerank.transformation import CloudflareRerankConfig +from litellm.llms.custom_httpx.http_handler import HTTPHandler +from litellm.types.rerank import RerankResponse +from litellm.utils import ProviderConfigManager + + +class _RecordingHTTPHandler(HTTPHandler): + def __init__(self, response): + super().__init__() + self.response = response + self.requests = [] + + def post(self, url: str, **kwargs): + self.requests.append({"url": url, **kwargs}) + return self.response + + +def test_provider_config_manager_returns_cloudflare_rerank_config(): + config = ProviderConfigManager.get_provider_rerank_config( + model="@cf/baai/bge-reranker-base", + provider=litellm.LlmProviders.CLOUDFLARE, + api_base=None, + present_version_params=[], + ) + + assert isinstance(config, CloudflareRerankConfig) + + +def test_get_complete_url_uses_native_workers_ai_endpoint(monkeypatch): + monkeypatch.setenv("CLOUDFLARE_ACCOUNT_ID", "account-id") + config = CloudflareRerankConfig() + + url = config.get_complete_url( + api_base=None, + model="@cf/baai/bge-reranker-base", + ) + + assert url == ("https://api.cloudflare.com/client/v4/accounts/account-id/ai/run/%40cf/baai/bge-reranker-base") + + +def test_get_complete_url_rewrites_openai_compatible_base(): + config = CloudflareRerankConfig() + + url = config.get_complete_url( + api_base="https://api.cloudflare.com/client/v4/accounts/account-id/ai/v1", + model="@cf/baai/bge-reranker-base", + ) + + assert url.endswith("/accounts/account-id/ai/run/%40cf/baai/bge-reranker-base") + + +@pytest.mark.parametrize( + ("api_base", "expected"), + [ + ( + "https://api.cloudflare.com/client/v4/accounts/account-id/ai/run/%40cf/baai/bge-reranker-base", + "https://api.cloudflare.com/client/v4/accounts/account-id/ai/run/%40cf/baai/bge-reranker-base", + ), + ( + "https://api.cloudflare.com/client/v4/accounts/account-id/ai/run", + "https://api.cloudflare.com/client/v4/accounts/account-id/ai/run/%40cf/baai/bge-reranker-base", + ), + ( + "https://example.com", + "https://example.com/ai/run/%40cf/baai/bge-reranker-base", + ), + ], +) +def test_get_complete_url_handles_supported_base_shapes(api_base, expected): + config = CloudflareRerankConfig() + + assert config.get_complete_url(api_base, "@cf/baai/bge-reranker-base") == expected + + +def test_get_complete_url_requires_account_id(monkeypatch): + monkeypatch.delenv("CLOUDFLARE_ACCOUNT_ID", raising=False) + config = CloudflareRerankConfig() + + with pytest.raises(ValueError, match="CLOUDFLARE_ACCOUNT_ID"): + config.get_complete_url(None, "@cf/baai/bge-reranker-base") + + +@pytest.mark.parametrize( + "model,error_match", + ( + ("../graphql", "cannot be a dot path segment"), + ("@cf/baai/../graphql", "cannot be a dot path segment"), + ("/@cf/baai/bge-reranker-base", "model is required"), + ), +) +def test_get_complete_url_rejects_path_traversal(model, error_match): + config = CloudflareRerankConfig() + + with pytest.raises(ValueError, match=error_match): + config.get_complete_url( + "https://api.cloudflare.com/client/v4/accounts/account-id/ai/run", + model, + ) + + +def test_get_complete_url_encodes_model_path_segments(): + config = CloudflareRerankConfig() + + url = config.get_complete_url( + "https://api.cloudflare.com/client/v4/accounts/account-id/ai/run", + "@cf/baai/model?debug=1", + ) + + assert url.endswith("/ai/run/%40cf/baai/model%3Fdebug%3D1") + + +def test_validate_environment_and_supported_params(): + config = CloudflareRerankConfig() + + headers = config.validate_environment( + headers={"X-Test": "value"}, + model="@cf/baai/bge-reranker-base", + api_key="cf-key", + ) + + assert headers["Authorization"] == "Bearer cf-key" + assert headers["X-Test"] == "value" + assert config.get_supported_cohere_rerank_params("@cf/baai/bge-reranker-base") == ( + "query", + "documents", + "top_n", + "return_documents", + ) + + +def test_validate_environment_requires_api_key(monkeypatch): + monkeypatch.delenv("CLOUDFLARE_API_KEY", raising=False) + config = CloudflareRerankConfig() + + with pytest.raises(ValueError, match="Cloudflare API Key"): + config.validate_environment({}, "@cf/baai/bge-reranker-base") + + +def test_map_cohere_rerank_params_without_top_n(): + config = CloudflareRerankConfig() + + assert config.map_cohere_rerank_params( + non_default_params={}, + model="@cf/baai/bge-reranker-base", + drop_params=False, + query="query", + documents=("document",), + ) == { + "query": "query", + "documents": ("document",), + "return_documents": True, + } + + +@pytest.mark.parametrize( + ("top_n", "return_documents", "expected"), + [ + (None, None, {"query": "query", "documents": ("document",)}), + ( + 1, + None, + {"query": "query", "documents": ("document",), "top_n": 1}, + ), + ], +) +def test_map_cohere_rerank_params_handles_optional_values(top_n, return_documents, expected): + config = CloudflareRerankConfig() + + assert ( + config.map_cohere_rerank_params( + non_default_params={}, + model="@cf/baai/bge-reranker-base", + drop_params=False, + query="query", + documents=("document",), + top_n=top_n, + return_documents=return_documents, + ) + == expected + ) + + +def test_transform_rerank_request(): + config = CloudflareRerankConfig() + + request = config.transform_rerank_request( + model="@cf/baai/bge-reranker-base", + optional_rerank_params={ + "query": "Which animal is faster?", + "documents": ["A cheetah can sprint.", {"text": "A turtle walks."}], + "top_n": 1, + }, + headers={}, + ) + + assert request == { + "query": "Which animal is faster?", + "contexts": ( + {"text": "A cheetah can sprint."}, + {"text": "A turtle walks."}, + ), + "top_k": 1, + } + + +@pytest.mark.parametrize( + "params,error_match", + ( + ({"documents": ("document",)}, "query is required for Cloudflare rerank"), + ({"query": "query"}, "documents is required for Cloudflare rerank"), + ( + {"query": "query", "documents": "document"}, + "documents is required for Cloudflare rerank", + ), + ({"query": "query", "documents": ()}, "documents is required for Cloudflare rerank"), + ), +) +def test_transform_rerank_request_validates_required_params(params, error_match): + config = CloudflareRerankConfig() + + with pytest.raises(ValueError, match=error_match): + config.transform_rerank_request( + model="@cf/baai/bge-reranker-base", + optional_rerank_params=params, + headers={}, + ) + + +def test_transform_rerank_request_without_top_n(): + config = CloudflareRerankConfig() + + request = config.transform_rerank_request( + model="@cf/baai/bge-reranker-base", + optional_rerank_params={ + "query": "query", + "documents": ({"title": "LiteLLM"},), + }, + headers={}, + ) + + assert request == { + "query": "query", + "contexts": ({"text": '{"title": "LiteLLM"}'},), + } + + +def test_transform_rerank_request_rejects_invalid_document(): + config = CloudflareRerankConfig() + + with pytest.raises(TypeError, match="strings or dictionaries"): + config.transform_rerank_request( + model="@cf/baai/bge-reranker-base", + optional_rerank_params={ + "query": "query", + "documents": (123,), + }, + headers={}, + ) + + +def test_transform_rerank_response_handles_rest_envelope(): + config = CloudflareRerankConfig() + logging_obj = MagicMock() + raw_response = httpx.Response( + status_code=200, + json={ + "result": { + "response": [ + {"id": 1, "score": 0.91}, + {"id": 0, "score": 0.42}, + ] + }, + "success": True, + "errors": [], + "messages": [], + }, + ) + + response = config.transform_rerank_response( + model="@cf/baai/bge-reranker-base", + raw_response=raw_response, + model_response=RerankResponse(), + logging_obj=logging_obj, + request_data={"query": "query", "contexts": [{"text": "a"}, {"text": "b"}]}, + ) + + assert response.results == [ + {"index": 1, "relevance_score": 0.91}, + {"index": 0, "relevance_score": 0.42}, + ] + + +@pytest.mark.parametrize("return_documents", [True, False]) +def test_transform_rerank_response_honors_return_documents(return_documents): + config = CloudflareRerankConfig() + raw_response = httpx.Response( + status_code=200, + json={"result": {"response": [{"id": 0, "score": 0.91}]}}, + ) + + response = config.transform_rerank_response( + model="@cf/baai/bge-reranker-base", + raw_response=raw_response, + model_response=RerankResponse(), + logging_obj=MagicMock(), + optional_params={ + "documents": ("LiteLLM is an LLM gateway.",), + "return_documents": return_documents, + }, + ) + + assert response.results is not None + if return_documents: + assert response.results[0]["document"] == {"text": "LiteLLM is an LLM gateway."} + else: + assert "document" not in response.results[0] + + +@pytest.mark.parametrize( + ("response_json", "error_type"), + [ + ({"success": False, "errors": ("failed",)}, CloudflareError), + ({"result": {}}, CloudflareError), + ({"result": {"response": ["invalid"]}}, TypeError), + ({"result": {"response": [{"id": "0", "score": 1}]}}, TypeError), + ], +) +def test_transform_rerank_response_rejects_invalid_responses(response_json, error_type): + config = CloudflareRerankConfig() + raw_response = httpx.Response(status_code=400, json=response_json) + + with pytest.raises(error_type): + config.transform_rerank_response( + model="@cf/baai/bge-reranker-base", + raw_response=raw_response, + model_response=RerankResponse(), + logging_obj=MagicMock(), + ) + + +@pytest.mark.parametrize( + "raw_response", + [ + httpx.Response(status_code=500, text="not json"), + httpx.Response(status_code=200, json=("not", "an", "object")), + ], +) +def test_transform_rerank_response_rejects_invalid_json(raw_response): + config = CloudflareRerankConfig() + + with pytest.raises(CloudflareError): + config.transform_rerank_response( + model="@cf/baai/bge-reranker-base", + raw_response=raw_response, + model_response=RerankResponse(), + logging_obj=MagicMock(), + ) + + +def test_get_error_class(): + error = CloudflareRerankConfig().get_error_class("failed", 400, {}) + + assert isinstance(error, CloudflareError) + assert error.status_code == 400 + + +def test_litellm_rerank_sends_cloudflare_request(): + response_json = { + "result": {"response": [{"id": 0, "score": 0.98}]}, + "success": True, + } + raw_response = Mock() + raw_response.status_code = 200 + raw_response.json.return_value = response_json + raw_response.text = json.dumps(response_json) + client = _RecordingHTTPHandler(raw_response) + + response = litellm.rerank( + model="cloudflare/@cf/baai/bge-reranker-base", + query="What is LiteLLM?", + documents=["LiteLLM is an LLM gateway.", "A recipe for soup."], + top_n=1, + api_key="test-key", + api_base="https://api.cloudflare.com/client/v4/accounts/account-id/ai/run", + client=client, + ) + + request = client.requests[0] + assert request["url"].endswith("/ai/run/%40cf/baai/bge-reranker-base") + assert request["headers"]["Authorization"] == "Bearer test-key" + assert json.loads(request["data"]) == { + "query": "What is LiteLLM?", + "contexts": [ + {"text": "LiteLLM is an LLM gateway."}, + {"text": "A recipe for soup."}, + ], + "top_k": 1, + } + assert response.results == [ + { + "index": 0, + "relevance_score": 0.98, + "document": {"text": "LiteLLM is an LLM gateway."}, + } + ]