This commit is contained in:
V0v1kkk's Assistant 2026-10-30 18:01:08 +01:00 • committed by GitHub
commit a6e42e12c5
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 191 additions and 11 deletions

View file

@ -204,6 +204,7 @@ from litellm.types.utils import (
TextCompletionResponse,
TokenCountMessagesFormat,
TokenCountResponse,
TranscriptionResponse,
echoed_cost_map_pricing_fields,
is_server_derived_pricing_key,
pricing_override_fields,
@ -1104,6 +1105,7 @@ from fastapi.responses import (
HTMLResponse,
JSONResponse,
ORJSONResponse,
PlainTextResponse,
RedirectResponse,
StreamingResponse,
)
@ -13542,6 +13544,7 @@ async def audio_transcriptions(
try:
# Use orjson to parse JSON data, orjson speeds up requests significantly
form_data: Final = await get_form_data(request)
requested_response_format: Final = form_data.get("response_format")
data = {**_form_fields_without_file(form_data), **data}
# Include original request and headers in the data
@ -13640,6 +13643,16 @@ async def audio_transcriptions(
if callback_headers:
fastapi_response.headers.update(callback_headers)
if requested_response_format == "text":
response_headers: Final = {
key: value
for key, value in fastapi_response.headers.items()
if key.lower() not in {"content-type", "content-length"}
}
if not isinstance(response, TranscriptionResponse):
raise TypeError("Audio transcription response must be a TranscriptionResponse")
return PlainTextResponse(content=response.text or "", headers=response_headers)
return response
except Exception as e:
await proxy_logging_obj.post_call_failure_hook(

View file

@ -10,17 +10,23 @@ Pins (PR2):
from __future__ import annotations
import io
from collections.abc import Callable
from contextlib import AbstractContextManager
from typing import Final
from unittest.mock import AsyncMock, MagicMock
import httpx
import pytest
from fastapi import HTTPException
from fastapi.testclient import TestClient
from litellm import Router
from litellm.proxy import proxy_server
from litellm.proxy._types import ProxyErrorTypes, ProxyException
from litellm.proxy.common_utils.http_parsing_utils import get_form_data
from litellm.types.llms.openai import HttpxBinaryResponseContent
from litellm.types.router import RouterErrors
from litellm.types.utils import TranscriptionResponse
_TAG_ROUTING_DENIAL = f"{RouterErrors.no_deployments_with_tag_routing.value}. Passed model=tts-1 and tags=['x']"
@ -125,9 +131,7 @@ def patched_transcription(monkeypatch):
return data
monkeypatch.setattr(proxy_server, "add_litellm_data_to_request", _add_data)
monkeypatch.setattr(
proxy_server, "check_file_size_under_limit", lambda **kwargs: True
)
monkeypatch.setattr(proxy_server, "check_file_size_under_limit", lambda **kwargs: True)
async def _form_data(request):
from starlette.datastructures import FormData, UploadFile
@ -151,12 +155,65 @@ def patched_transcription(monkeypatch):
@pytest.fixture
def patched_transcription_error(monkeypatch, patched_transcription):
def patched_transcription_error(monkeypatch: pytest.MonkeyPatch, patched_transcription: None) -> AsyncMock:
monkeypatch.setattr(proxy_server, "get_form_data", get_form_data)
async def _raise(*args, **kwargs):
raise ValueError("transcription boom")
monkeypatch.setattr(proxy_server, "route_request", _raise)
yield
route_request: Final[AsyncMock] = AsyncMock(side_effect=_raise)
monkeypatch.setattr(proxy_server, "route_request", route_request)
return route_request
@pytest.fixture
def patched_transcription_http(monkeypatch: pytest.MonkeyPatch, request: pytest.FixtureRequest):
transcript: Final = getattr(request, "param", "hello world")
router: Final = MagicMock()
router.model_names = ["whisper-1"]
router.get_deployment_by_model_group_name.return_value = None
monkeypatch.setattr(proxy_server, "llm_router", router)
async def _response_headers(data: dict[str, object], **kwargs: object) -> dict[str, str]:
return {
"x-callback-trace": "callback-value",
**(
{"Content-Type": "application/json", "Content-Length": "999"}
if data.get("response_format") == "text"
else {}
),
}
monkeypatch.setattr(
proxy_server,
"proxy_logging_obj",
MagicMock(
pre_call_hook=AsyncMock(side_effect=lambda **kw: kw["data"]),
post_call_failure_hook=AsyncMock(),
post_call_response_headers_hook=AsyncMock(side_effect=_response_headers),
update_request_status=AsyncMock(),
),
)
async def _add_data(data: dict[str, object], **kwargs: object) -> dict[str, object]:
return data
monkeypatch.setattr(proxy_server, "add_litellm_data_to_request", _add_data)
async def _llm_call() -> TranscriptionResponse | dict[str, str]:
if not isinstance(transcript, str):
return {"text": "not a TranscriptionResponse"}
response: Final = TranscriptionResponse(text=transcript)
response.hidden_params = {
"litellm_call_id": "transcription-call-id",
"model_id": "transcription-model-id",
"cache_key": "transcription-cache-key",
}
return response
route_request: Final[AsyncMock] = AsyncMock(return_value=_llm_call())
monkeypatch.setattr(proxy_server, "route_request", route_request)
return transcript, route_request
@pytest.mark.parametrize("path", ["/v1/audio/speech", "/audio/speech"])
@ -351,11 +408,121 @@ def test_audio_transcription_keeps_the_status_of_a_proxy_rejection(
@pytest.mark.parametrize("path", ["/v1/audio/transcriptions", "/audio/transcriptions"])
def test_audio_transcription_error(client, auth_as, patched_transcription_error, path):
@pytest.mark.parametrize("response_format", [None, "text"], ids=["default", "text"])
def test_audio_transcription_error(
client: TestClient,
auth_as: Callable[..., AbstractContextManager[None]],
patched_transcription_error: AsyncMock,
path: str,
response_format: str | None,
) -> None:
"""Pins ``POST /v1/audio/transcriptions`` / ``POST /audio/transcriptions`` (error)."""
files = {"file": ("audio.mp3", b"\x00\x01\x02", "audio/mpeg")}
data = {"model": "whisper-1"}
files: Final = {"file": ("audio.mp3", b"\x00\x01\x02", "audio/mpeg")}
data: Final[dict[str, str]] = {"model": "whisper-1"} | (
{"response_format": response_format} if response_format is not None else {}
)
with auth_as():
response = client.post(path, files=files, data=data)
response: Final = client.post(
path,
files=files,
data=data,
headers={"x-litellm-call-id": "transcription-request-id"},
)
assert response.status_code == 500
assert len(response.content) > 0
assert response.headers["content-type"] == "application/json"
assert response.headers["x-litellm-call-id"] == "transcription-request-id"
assert response.json() == {
"error": {
"message": "transcription boom",
"type": "internal_server_error",
"param": None,
"code": "500",
}
}
route_request_call: Final = patched_transcription_error.await_args
assert route_request_call is not None
assert route_request_call.kwargs["data"].get("response_format") == response_format
@pytest.mark.parametrize("path", ["/v1/audio/transcriptions", "/audio/transcriptions"], ids=["v1", "alias"])
@pytest.mark.parametrize(
"patched_transcription_http",
["hello world", "Grüße, 世界\nline two\n", ""],
indirect=True,
ids=["plain", "unicode-multiline", "empty"],
)
def test_audio_transcription_text_format_returns_plain_text(client, auth_as, patched_transcription_http, path):
transcript, route_request = patched_transcription_http
files: Final = {"file": ("audio.mp3", b"\x00\x01\x02", "audio/mpeg")}
data: Final = {"model": "whisper-1", "response_format": "text"}
with auth_as():
response: Final = client.post(
path,
files=files,
data=data,
headers={"x-litellm-call-id": "request-call-id"},
)
assert response.status_code == 200
assert response.headers["content-type"] == "text/plain; charset=utf-8"
assert response.content == transcript.encode("utf-8")
assert response.headers["content-length"] == str(len(transcript.encode("utf-8")))
assert response.headers["x-litellm-call-id"] == "transcription-call-id"
assert response.headers["x-litellm-model-id"] == "transcription-model-id"
assert response.headers["x-litellm-cache-key"] == "transcription-cache-key"
assert response.headers["x-callback-trace"] == "callback-value"
assert route_request.await_args.kwargs["data"]["response_format"] == "text"
@pytest.mark.parametrize("path", ["/v1/audio/transcriptions", "/audio/transcriptions"], ids=["v1", "alias"])
@pytest.mark.parametrize("response_format", [None, "json"], ids=["default", "json"])
def test_audio_transcription_json_formats_keep_body_and_headers(
client, auth_as, patched_transcription_http, path, response_format
):
transcript, route_request = patched_transcription_http
files: Final = {"file": ("audio.mp3", b"\x00\x01\x02", "audio/mpeg")}
data: Final = {"model": "whisper-1"} | ({"response_format": response_format} if response_format is not None else {})
with auth_as():
response: Final = client.post(
path,
files=files,
data=data,
headers={"x-litellm-call-id": "request-call-id"},
)
assert response.status_code == 200
assert response.headers["content-type"] == "application/json"
assert response.json() == {"text": transcript, "usage": None}
assert response.headers["x-litellm-call-id"] == "transcription-call-id"
assert response.headers["x-litellm-model-id"] == "transcription-model-id"
assert response.headers["x-litellm-cache-key"] == "transcription-cache-key"
assert response.headers["x-callback-trace"] == "callback-value"
assert route_request.await_args.kwargs["data"].get("response_format") == response_format
@pytest.mark.parametrize("path", ["/v1/audio/transcriptions", "/audio/transcriptions"], ids=["v1", "alias"])
@pytest.mark.parametrize("patched_transcription_http", [None], indirect=True, ids=["malformed-response"])
def test_audio_transcription_text_format_rejects_malformed_response(
client: TestClient,
auth_as: Callable[..., AbstractContextManager[None]],
patched_transcription_http: tuple[object, AsyncMock],
path: str,
):
files: Final = {"file": ("audio.mp3", b"\x00\x01\x02", "audio/mpeg")}
data: Final = {"model": "whisper-1", "response_format": "text"}
with auth_as():
response: Final = client.post(path, files=files, data=data)
assert response.status_code == 500
assert response.headers["content-type"] == "application/json"
assert response.json() == {
"error": {
"message": "Audio transcription response must be a TranscriptionResponse",
"type": "internal_server_error",
"param": None,
"code": "500",
}
}
logging_obj: Final = proxy_server.proxy_logging_obj
assert isinstance(logging_obj, MagicMock)
failure_hook: Final = logging_obj.post_call_failure_hook
assert isinstance(failure_hook, AsyncMock)
assert failure_hook.await_count == 1