diff --git a/litellm/constants.py b/litellm/constants.py index 2901b7d8948..242f0e2a80c 100644 --- a/litellm/constants.py +++ b/litellm/constants.py @@ -342,6 +342,7 @@ DEFAULT_REASONING_EFFORT_MINIMAL_THINKING_BUDGET: Final = get_env_int( # Provider-specific API base URLs XAI_API_BASE: Final = "https://api.x.ai/v1" PLACEHOLDER_API_KEY: Final = "fake-api-key" +PLACEHOLDER_API_KEYS: Final = frozenset({PLACEHOLDER_API_KEY, "dummy-key"}) OPEN_SANDBOX_API_BASE_ENV_VAR: Final = "OPEN_SANDBOX_API_BASE" OPEN_SANDBOX_API_KEY_ENV_VAR: Final = "OPEN_SANDBOX_API_KEY" OPEN_SANDBOX_DEFAULT_TEMPLATE: Final = "opensandbox/code-interpreter:v1.1.0" diff --git a/litellm/main.py b/litellm/main.py index 23eac8ce0aa..e515f0b5e65 100644 --- a/litellm/main.py +++ b/litellm/main.py @@ -66,6 +66,7 @@ from litellm.constants import ( DEFAULT_MOCK_RESPONSE_PROMPT_TOKEN_COUNT, NADIR_DEFAULT_API_BASE, OPENAI_AUDIO_TRANSCRIPTION_PROVIDERS, + PLACEHOLDER_API_KEYS, ) from litellm.exceptions import LiteLLMUnknownProvider from litellm.integrations.custom_logger import CustomLogger @@ -8629,6 +8630,7 @@ def speech( model, custom_llm_provider, dynamic_api_key, api_base = get_llm_provider( model=model, custom_llm_provider=custom_llm_provider, api_base=api_base or base_url ) + provider_api_key: Final = None if dynamic_api_key in PLACEHOLDER_API_KEYS else dynamic_api_key kwargs.pop("tags", []) optional_params = {} @@ -8718,9 +8720,11 @@ def speech( # set API KEY api_key = ( api_key - or litellm.api_key # for deepinfra/perplexity/anyscale we check in get_llm_provider and pass in the api key from there + or provider_api_key # for deepinfra/perplexity/anyscale we check in get_llm_provider and pass in the api key from there + or litellm.api_key or litellm.openai_key or get_secret("OPENAI_API_KEY") + or dynamic_api_key ) organization = ( @@ -8803,6 +8807,7 @@ def speech( api_key = ( api_key + or provider_api_key or litellm.api_key or litellm.azure_key or get_secret("AZURE_OPENAI_API_KEY") diff --git a/tests/unit/test_main.py b/tests/unit/test_main.py index 70b268bf2a2..d3df49269ec 100644 --- a/tests/unit/test_main.py +++ b/tests/unit/test_main.py @@ -5129,6 +5129,121 @@ def test_speech_mistral_routes_to_configured_api_base(respx_mock: respx.MockRout assert response.content == audio_bytes + +TTS_AUDIO: Final = b"ID3\x04\x00\x00\x00fake-mp3-payload" + + +def _clear_global_tts_keys(monkeypatch: pytest.MonkeyPatch) -> None: + for attribute in ("api_key", "openai_key", "azure_key", "api_base"): + monkeypatch.setattr(litellm, attribute, None) + + +def _set_or_clear_env(monkeypatch: pytest.MonkeyPatch, values: dict[str, str | None]) -> None: + for name, value in values.items(): + if value is None: + monkeypatch.delenv(name, raising=False) + else: + monkeypatch.setenv(name, value) + + +@pytest.mark.parametrize( + ("provider_key", "openai_key", "sent_authorization"), + [ + ("hosted-vllm-scoped-key", None, "Bearer hosted-vllm-scoped-key"), + ("hosted-vllm-scoped-key", "generic-openai-key", "Bearer hosted-vllm-scoped-key"), + (None, "generic-openai-key", "Bearer generic-openai-key"), + (None, None, "Bearer fake-api-key"), + ], + ids=["provider-key-alone", "provider-key-beats-generic", "generic-key-beats-placeholder", "placeholder-when-unset"], +) +def test_speech_openai_compatible_sends_the_provider_scoped_key( + respx_mock: respx.MockRouter, + monkeypatch: pytest.MonkeyPatch, + provider_key: str | None, + openai_key: str | None, + sent_authorization: str, +): + _clear_global_tts_keys(monkeypatch) + _set_or_clear_env(monkeypatch, {"HOSTED_VLLM_API_KEY": provider_key, "OPENAI_API_KEY": openai_key}) + route: Final = respx_mock.post("https://vllm.internal/v1/audio/speech").mock( + return_value=httpx.Response(200, content=TTS_AUDIO, headers={"content-type": "audio/mpeg"}) + ) + + response: Final = litellm.speech( + model="hosted_vllm/tts-1", input="which key goes out", voice="alloy", api_base="https://vllm.internal/v1" + ) + + assert response.content == TTS_AUDIO + assert route.call_count == 1 + assert route.calls.last.request.headers["authorization"] == sent_authorization + + +@pytest.mark.parametrize( + ("generic_azure_key", "sent_api_key"), + [(None, "azure-ai-scoped-key"), ("generic-azure-key", "azure-ai-scoped-key")], + ids=["provider-key-alone", "provider-key-beats-generic"], +) +def test_speech_azure_sends_the_provider_scoped_key( + respx_mock: respx.MockRouter, monkeypatch: pytest.MonkeyPatch, generic_azure_key: str | None, sent_api_key: str +): + _clear_global_tts_keys(monkeypatch) + _set_or_clear_env( + monkeypatch, + {"AZURE_AI_API_KEY": "azure-ai-scoped-key", "AZURE_OPENAI_API_KEY": None, "AZURE_API_KEY": generic_azure_key}, + ) + route: Final = respx_mock.post( + url__regex=r"https://foundry\.internal/openai/deployments/tts-1/audio/speech\?api-version=.+" + ).mock(return_value=httpx.Response(200, content=TTS_AUDIO, headers={"content-type": "audio/mpeg"})) + + response: Final = litellm.speech( + model="azure_ai/tts-1", + input="which key goes out", + voice="alloy", + api_base="https://foundry.internal", + api_version="2024-05-01-preview", + ) + + assert response.content == TTS_AUDIO + assert route.call_count == 1 + assert route.calls.last.request.headers["api-key"] == sent_api_key + + +def test_speech_docker_model_runner_placeholder_does_not_beat_a_real_key( + respx_mock: respx.MockRouter, monkeypatch: pytest.MonkeyPatch +): + _clear_global_tts_keys(monkeypatch) + _set_or_clear_env(monkeypatch, {"DOCKER_MODEL_RUNNER_API_KEY": None, "OPENAI_API_KEY": "generic-openai-key"}) + route: Final = respx_mock.post("https://dmr.internal/v1/audio/speech").mock( + return_value=httpx.Response(200, content=TTS_AUDIO, headers={"content-type": "audio/mpeg"}) + ) + + litellm.speech( + model="docker_model_runner/tts-1", input="which key goes out", voice="alloy", api_base="https://dmr.internal/v1" + ) + + assert route.calls.last.request.headers["authorization"] == "Bearer generic-openai-key" + + +def test_speech_keeps_an_explicitly_passed_api_key_over_the_provider_scoped_one( + respx_mock: respx.MockRouter, monkeypatch: pytest.MonkeyPatch +): + _clear_global_tts_keys(monkeypatch) + monkeypatch.setenv("HOSTED_VLLM_API_KEY", "hosted-vllm-scoped-key") + route: Final = respx_mock.post("https://vllm.internal/v1/audio/speech").mock( + return_value=httpx.Response(200, content=TTS_AUDIO, headers={"content-type": "audio/mpeg"}) + ) + + litellm.speech( + model="hosted_vllm/tts-1", + input="which key goes out", + voice="alloy", + api_base="https://vllm.internal/v1", + api_key="caller-supplied-key", + ) + + assert route.calls.last.request.headers["authorization"] == "Bearer caller-supplied-key" + + FOUNDRY_HOST: Final = "https://my-project.services.ai.azure.com"