mirror of
https://github.com/BerriAI/litellm.git
synced 2026-11-16 09:22:22 +00:00
fix(azure): keep the deployment route for v1 api_version values on audio transcriptions
Azure's v1 route answers DeploymentNotFound for transcription deployments, so preview, latest and v1 fall back to the dated default on the per-deployment route, which also stops the sync path from raising on a v1 client
This commit is contained in:
parent
bcc5f903bb
commit
73e244b101
2 changed files with 10 additions and 46 deletions
|
|
@ -1,7 +1,7 @@
|
|||
from collections.abc import Coroutine
|
||||
from typing import TYPE_CHECKING, Final
|
||||
|
||||
from openai import AsyncOpenAI, OpenAI
|
||||
from openai import AsyncAzureOpenAI, AzureOpenAI
|
||||
from pydantic import BaseModel
|
||||
|
||||
from litellm._uuid import uuid
|
||||
|
|
@ -40,6 +40,7 @@ class AzureAudioTranscription(AzureChatCompletion):
|
|||
custom_llm_provider: str = "azure",
|
||||
) -> TranscriptionResponse | Coroutine[object, object, TranscriptionResponse]:
|
||||
data: Final = {"model": model, "file": audio_file, **optional_params}
|
||||
deployment_api_version: Final = None if self.is_azure_v1_api_version(api_version) else api_version
|
||||
|
||||
if atranscription is True:
|
||||
return self.async_audio_transcriptions(
|
||||
|
|
@ -49,7 +50,7 @@ class AzureAudioTranscription(AzureChatCompletion):
|
|||
timeout=timeout,
|
||||
api_key=api_key,
|
||||
api_base=api_base,
|
||||
api_version=api_version,
|
||||
api_version=deployment_api_version,
|
||||
client=client,
|
||||
max_retries=max_retries,
|
||||
logging_obj=logging_obj,
|
||||
|
|
@ -59,7 +60,7 @@ class AzureAudioTranscription(AzureChatCompletion):
|
|||
)
|
||||
|
||||
azure_client: Final = self.get_azure_openai_client(
|
||||
api_version=api_version,
|
||||
api_version=deployment_api_version,
|
||||
api_base=api_base,
|
||||
api_key=api_key,
|
||||
model=model,
|
||||
|
|
@ -67,10 +68,10 @@ class AzureAudioTranscription(AzureChatCompletion):
|
|||
client=client,
|
||||
litellm_params=litellm_params,
|
||||
)
|
||||
if not isinstance(azure_client, OpenAI):
|
||||
if not isinstance(azure_client, AzureOpenAI):
|
||||
raise AzureOpenAIError(
|
||||
status_code=500,
|
||||
message="azure_client is not an instance of OpenAI",
|
||||
message="azure_client is not an instance of AzureOpenAI",
|
||||
)
|
||||
|
||||
## LOGGING
|
||||
|
|
@ -138,10 +139,10 @@ class AzureAudioTranscription(AzureChatCompletion):
|
|||
client=client,
|
||||
litellm_params=litellm_params,
|
||||
)
|
||||
if not isinstance(async_azure_client, AsyncOpenAI):
|
||||
if not isinstance(async_azure_client, AsyncAzureOpenAI):
|
||||
raise AzureOpenAIError(
|
||||
status_code=500,
|
||||
message="async_azure_client is not an instance of AsyncOpenAI",
|
||||
message="async_azure_client is not an instance of AsyncAzureOpenAI",
|
||||
)
|
||||
|
||||
## LOGGING
|
||||
|
|
|
|||
|
|
@ -10,7 +10,6 @@ from openai import AsyncAzureOpenAI, AzureOpenAI
|
|||
import litellm
|
||||
from litellm.cost_calculator import completion_cost
|
||||
from litellm.litellm_core_utils.audio_utils.utils import calculate_request_duration
|
||||
from litellm.llms.azure.common_utils import AzureOpenAIError
|
||||
from litellm.types.utils import TranscriptionResponse
|
||||
|
||||
AUDIO_FILE: Final = Path(__file__).parents[3] / "gettysburg.wav"
|
||||
|
|
@ -177,7 +176,7 @@ async def test_azure_transcription_parses_response_formats(
|
|||
("api_version", "expected_path", "expected_query_version"),
|
||||
[
|
||||
("2025-03-01-preview", "/openai/deployments/whisper-1/audio/transcriptions", "2025-03-01-preview"),
|
||||
("preview", "/openai/v1/audio/transcriptions", None),
|
||||
("preview", "/openai/deployments/whisper-1/audio/transcriptions", litellm.AZURE_DEFAULT_API_VERSION),
|
||||
],
|
||||
)
|
||||
async def test_azure_transcription_sends_the_configured_api_version(
|
||||
|
|
@ -191,6 +190,7 @@ async def test_azure_transcription_sends_the_configured_api_version(
|
|||
return httpx.Response(200, json={"text": str(request.url)})
|
||||
|
||||
transport: Final = httpx.MockTransport(echo_url)
|
||||
monkeypatch.delenv("AZURE_API_VERSION", raising=False)
|
||||
monkeypatch.setattr(litellm, "aclient_session", httpx.AsyncClient(transport=transport))
|
||||
monkeypatch.setattr(litellm, "client_session", httpx.Client(transport=transport))
|
||||
request_kwargs: Final = {
|
||||
|
|
@ -206,40 +206,3 @@ async def test_azure_transcription_sends_the_configured_api_version(
|
|||
sent_url: Final = httpx.URL(response.text)
|
||||
assert sent_url.path == expected_path
|
||||
assert sent_url.params.get("api-version") == expected_query_version
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
@pytest.mark.parametrize("is_async", [True, False])
|
||||
async def test_azure_transcription_rejects_a_client_of_the_other_kind(is_async: bool) -> None:
|
||||
def unreachable(request: httpx.Request) -> httpx.Response:
|
||||
raise AssertionError(f"no request expected, got {request.url}")
|
||||
|
||||
transport: Final = httpx.MockTransport(unreachable)
|
||||
wrong_client: Final = (
|
||||
AzureOpenAI(
|
||||
api_key="test-key",
|
||||
api_version="2024-06-01",
|
||||
azure_endpoint="https://example.cognitiveservices.azure.com",
|
||||
http_client=httpx.Client(transport=transport),
|
||||
)
|
||||
if is_async
|
||||
else AsyncAzureOpenAI(
|
||||
api_key="test-key",
|
||||
api_version="2024-06-01",
|
||||
azure_endpoint="https://example.cognitiveservices.azure.com",
|
||||
http_client=httpx.AsyncClient(transport=transport),
|
||||
)
|
||||
)
|
||||
request_kwargs: Final = {
|
||||
"model": "azure/whisper-1",
|
||||
"file": _audio_file(),
|
||||
"api_base": "https://example.openai.azure.com",
|
||||
"api_key": "test-key",
|
||||
"api_version": "2024-06-01",
|
||||
"client": wrong_client,
|
||||
}
|
||||
|
||||
with pytest.raises((litellm.APIError, AzureOpenAIError), match="is not an instance of") as raised:
|
||||
await _transcribe(is_async, request_kwargs)
|
||||
|
||||
assert raised.value.status_code == 500
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue