diff --git a/litellm/proxy/management_endpoints/usage_endpoints/ai_usage_chat.py b/litellm/proxy/management_endpoints/usage_endpoints/ai_usage_chat.py index 04ea99e1baf..9f43080534c 100644 --- a/litellm/proxy/management_endpoints/usage_endpoints/ai_usage_chat.py +++ b/litellm/proxy/management_endpoints/usage_endpoints/ai_usage_chat.py @@ -8,7 +8,7 @@ from collections.abc import AsyncGenerator, AsyncIterator, Awaitable, Callable, from datetime import date from itertools import chain from types import MappingProxyType -from typing import TYPE_CHECKING, Final, Literal, NamedTuple, Protocol, TypeAlias, cast, overload +from typing import TYPE_CHECKING, Final, Literal, NamedTuple, Protocol, cast, overload from fastapi import HTTPException from pydantic import BaseModel, ConfigDict, ValidationError, field_validator @@ -471,23 +471,12 @@ def _resolve_model(model: str | None) -> str: return (model or "").strip() or DEFAULT_COMPETITOR_DISCOVERY_MODEL -_RouterRoute: TypeAlias = Literal["model_group", "deployment"] - - -def _router_route( - router: "Router", model: str, caller_is_proxy_admin: bool, request_team_id: str | None -) -> _RouterRoute | None: +def _is_router_model(router: "Router", model: str, caller_is_proxy_admin: bool) -> bool: if router.is_recognized_model(model) or (caller_is_proxy_admin and model in router.team_public_model_names): - return "model_group" + return True if router.router_general_settings.pass_through_all_models: - return None - if router.default_deployment is not None or router.pattern_router.route(model) is not None: # pyright: ignore[reportUnknownMemberType] # PatternMatchDeployments.route returns untyped dicts - return "model_group" - deployments: Final = router._get_deployment_by_litellm_model(model=model) # pyright: ignore[reportPrivateUsage] # the lookup the router's specific_deployment path runs - serves_model: Final = any( - router.deployment_usable_by_team(deployment, request_team_id) for deployment in deployments - ) - return "deployment" if serves_model else None + return False + return router.default_deployment is not None or router.pattern_router.route(model) is not None # pyright: ignore[reportUnknownMemberType] # PatternMatchDeployments.route returns untyped dicts class _TaggedMetadata(BaseModel): @@ -610,7 +599,7 @@ async def _acompletion( router_settings: Mapping[str, object] | None = None, ) -> ModelResponse | CustomStreamWrapper: from litellm.proxy.proxy_server import llm_router - from litellm.router_utils.common_utils import get_request_team_id, is_proxy_admin_request + from litellm.router_utils.common_utils import is_proxy_admin_request from litellm.types.llms.openai import AllMessageValues completion_messages: Final = cast( # cast-ok: acompletion accepts the same Mapping messages it already receives @@ -619,13 +608,9 @@ async def _acompletion( caller_metadata: Final[Mapping[str, object]] = ( _caller_metadata(user_api_key_dict) if user_api_key_dict is not None else MappingProxyType({}) ) - router_request: Final = MappingProxyType({"metadata": caller_metadata}) - route: Final = ( - _router_route(llm_router, model, is_proxy_admin_request(router_request), get_request_team_id(router_request)) - if llm_router is not None - else None - ) - if llm_router is not None and route is not None: + if llm_router is not None and _is_router_model( + llm_router, model, is_proxy_admin_request(MappingProxyType({"metadata": caller_metadata})) + ): router_kwargs: Final = _per_request_router_kwargs(router_settings) if stream: return await llm_router.acompletion( # pyright: ignore[reportUnknownMemberType] # Router.acompletion overloads carry untyped **kwargs @@ -637,7 +622,6 @@ async def _acompletion( drop_params=True, allowed_openai_params=list(USAGE_AI_REQUIRED_PARAMS), metadata=dict(caller_metadata), - specific_deployment=route == "deployment", **router_kwargs, ) return await llm_router.acompletion( # pyright: ignore[reportUnknownMemberType] # Router.acompletion overloads carry untyped **kwargs @@ -649,7 +633,6 @@ async def _acompletion( drop_params=True, allowed_openai_params=list(USAGE_AI_REQUIRED_PARAMS), metadata=dict(caller_metadata), - specific_deployment=route == "deployment", **router_kwargs, ) return await litellm.acompletion( diff --git a/tests/unit/proxy/management_endpoints/usage_endpoints/test_ai_usage_chat.py b/tests/unit/proxy/management_endpoints/usage_endpoints/test_ai_usage_chat.py index 0eea3dfa6dd..42e71f15fdd 100644 --- a/tests/unit/proxy/management_endpoints/usage_endpoints/test_ai_usage_chat.py +++ b/tests/unit/proxy/management_endpoints/usage_endpoints/test_ai_usage_chat.py @@ -646,46 +646,13 @@ class TestUsageAiChatProviderRequest: ] @pytest.mark.asyncio - async def test_deployment_name_is_called_with_that_deployment_credentials( + async def test_a_deployment_model_string_runs_on_the_environment_key( self, respx_mock: respx.MockRouter, monkeypatch: pytest.MonkeyPatch ) -> None: self._mock_openai(respx_mock) await self._ask("openai/gpt-5.6", monkeypatch, self._router()) - assert respx_mock.calls.last.request.headers["Authorization"] == "Bearer sk-gpt5-deployment" - - @pytest.mark.asyncio - async def test_deleted_deployment_name_is_called_with_the_environment_key( - self, respx_mock: respx.MockRouter, monkeypatch: pytest.MonkeyPatch - ) -> None: - self._mock_openai(respx_mock) - router: Final = self._router() - router.delete_deployment(id=router.get_model_ids(model_name="gpt-5-prod")[0]) - - await self._ask("openai/gpt-5.6", monkeypatch, router) - - assert respx_mock.calls.last.request.headers["Authorization"] == "Bearer sk-env" - - @pytest.mark.asyncio - async def test_team_owned_deployment_name_is_called_with_the_environment_key( - self, respx_mock: respx.MockRouter, monkeypatch: pytest.MonkeyPatch - ) -> None: - from litellm.router import Router - - self._mock_openai(respx_mock) - router: Final = Router( - model_list=[ - { - "model_name": "team-gpt-5", - "litellm_params": {"model": "openai/gpt-5.6", "api_key": "sk-team-deployment"}, - "model_info": {"team_id": "team-1"}, - } - ] - ) - - await self._ask("openai/gpt-5.6", monkeypatch, router) - assert respx_mock.calls.last.request.headers["Authorization"] == "Bearer sk-env" @pytest.mark.asyncio