diff --git a/litellm/llms/anthropic/messages/native/handler.py b/litellm/llms/anthropic/messages/native/handler.py index 51b28c9ead8..27ac0501160 100644 --- a/litellm/llms/anthropic/messages/native/handler.py +++ b/litellm/llms/anthropic/messages/native/handler.py @@ -153,7 +153,7 @@ def _bridge_target(requested: _BridgeTarget, resolved: _BridgeTarget, custom_llm return requested return _BridgeTarget( model=f"{custom_llm_provider}/{resolved.model}", - api_key=requested.api_key or resolved.api_key, + api_key=requested.api_key if requested.api_base else requested.api_key or resolved.api_key, api_base=resolved.api_base, ) diff --git a/tests/unit/llms/anthropic/messages/native/test_anthropic_messages_handler.py b/tests/unit/llms/anthropic/messages/native/test_anthropic_messages_handler.py index eabc18e2550..4033105a6be 100644 --- a/tests/unit/llms/anthropic/messages/native/test_anthropic_messages_handler.py +++ b/tests/unit/llms/anthropic/messages/native/test_anthropic_messages_handler.py @@ -1279,10 +1279,15 @@ async def test_bridged_messages_send_the_deployments_key_for_azure_models_served {"azure_username": "user", "azure_password": "password", "client_id": "client"}, ], ) -async def test_bridged_messages_keep_the_deployments_entra_credential_over_the_env_key(monkeypatch, entra_credential): +@pytest.mark.parametrize("api_base_from_env", [False, True], ids=["api_base_in_params", "api_base_from_env"]) +async def test_bridged_messages_keep_the_deployments_entra_credential_over_the_env_key( + monkeypatch, entra_credential, api_base_from_env +): monkeypatch.setattr(litellm, "disable_aiohttp_transport", True) + monkeypatch.setenv("AZURE_AI_API_BASE", "https://example-resource.cognitiveservices.azure.com") monkeypatch.setenv("AZURE_AI_API_KEY", "azure-env-key") litellm.in_memory_llm_clients_cache.flush_cache() + api_base: Final = {} if api_base_from_env else {"api_base": "https://example-resource.cognitiveservices.azure.com"} with ( patch("litellm.llms.azure.common_utils.get_azure_ad_token_from_entra_id", return_value=lambda: "entra-token"), # test-quality-ok: stands in for the Entra token fetch so the test can see which credential reaches Azure @@ -1298,8 +1303,8 @@ async def test_bridged_messages_keep_the_deployments_entra_credential_over_the_e max_tokens=100, messages=[{"role": "user", "content": "ping"}], model="azure_ai/gpt-6-luna", - api_base="https://example-resource.cognitiveservices.azure.com", tools=[{"name": "get_weather", "input_schema": {"type": "object", "properties": {}}}], + **api_base, **entra_credential, ) @@ -1307,6 +1312,47 @@ async def test_bridged_messages_keep_the_deployments_entra_credential_over_the_e assert {call.request.headers.get("api-key") for call in upstream.calls} == {None} +@pytest.mark.asyncio +@pytest.mark.parametrize( + "azure_env_credential, expected_header, expected_value", + [ + ({"AZURE_API_KEY": "azure-env-key"}, "api-key", "azure-env-key"), + ( + {"AZURE_TENANT_ID": "tenant", "AZURE_CLIENT_ID": "client", "AZURE_CLIENT_SECRET": "secret"}, + "authorization", + "Bearer entra-token", + ), + ], + ids=["azure_env_key", "azure_env_entra"], +) +async def test_bridged_messages_keep_the_azure_env_credential_when_the_deployment_sets_its_own_api_base( + azure_env_credential, expected_header, expected_value, monkeypatch +): + monkeypatch.setattr(litellm, "disable_aiohttp_transport", True) + for name in ("AZURE_API_KEY", "AZURE_OPENAI_API_KEY", "AZURE_AD_TOKEN", "AZURE_TENANT_ID", "AZURE_CLIENT_ID"): + monkeypatch.delenv(name, raising=False) + monkeypatch.setenv("AZURE_AI_API_KEY", "key-for-another-resource") + for name, value in azure_env_credential.items(): + monkeypatch.setenv(name, value) + litellm.in_memory_llm_clients_cache.flush_cache() + + with ( + patch("litellm.llms.azure.common_utils.get_azure_ad_token_from_entra_id", return_value=lambda: "entra-token"), # test-quality-ok: stands in for the Entra token fetch so the test can see which credential reaches Azure + respx.mock(assert_all_called=True) as respx_mock, + ): + upstream: Final = respx_mock.route().mock(side_effect=httpx.ConnectError("cut at the wire")) + with pytest.raises(openai.APIError): + await litellm.anthropic.messages.acreate( + max_tokens=100, + messages=[{"role": "user", "content": "ping"}], + model="azure_ai/gpt-6-luna", + api_base="https://example-resource.cognitiveservices.azure.com", + tools=[{"name": "get_weather", "input_schema": {"type": "object", "properties": {}}}], + ) + + assert {call.request.headers.get(expected_header) for call in upstream.calls} == {expected_value} + + @pytest.mark.asyncio async def test_bridged_messages_still_route_through_the_litellm_proxy_when_asked(monkeypatch): monkeypatch.setattr(litellm, "disable_aiohttp_transport", True)