| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | |
| 4 | |
| 5 | from http import HTTPStatus |
| 6 | from typing import Annotated |
| 7 | |
| 8 | from fastapi import APIRouter, Form, Request |
| 9 | from fastapi.responses import JSONResponse, StreamingResponse |
| 10 | |
| 11 | from vllm.entrypoints.serve.utils.api_utils import ( |
| 12 | load_aware_call, |
| 13 | with_cancellation, |
| 14 | ) |
| 15 | from vllm.entrypoints.speech_to_text.base.utils import read_upload_with_limit |
| 16 | from vllm.logger import init_logger |
| 17 | |
| 18 | from ...serve.engine.protocol import ErrorResponse |
| 19 | from .protocol import TranscriptionRequest, TranscriptionResponseVariant |
| 20 | from .serving import OpenAIServingTranscription |
| 21 | |
| 22 | logger = init_logger(__name__) |
| 23 | |
| 24 | router = APIRouter() |
| 25 | |
| 26 | |
| 27 | def transcription(request: Request) -> OpenAIServingTranscription: |
| 28 | return request.app.state.openai_serving_transcription |
| 29 | |
| 30 | |
| 31 | @router.post( |
| 32 | "/v1/audio/transcriptions", |
| 33 | responses={ |
| 34 | HTTPStatus.OK.value: {"content": {"text/event-stream": {}}}, |
| 35 | HTTPStatus.BAD_REQUEST.value: {"model": ErrorResponse}, |
| 36 | HTTPStatus.UNPROCESSABLE_ENTITY.value: {"model": ErrorResponse}, |
| 37 | HTTPStatus.INTERNAL_SERVER_ERROR.value: {"model": ErrorResponse}, |
| 38 | }, |
| 39 | ) |
| 40 | @with_cancellation |
| 41 | @load_aware_call |
| 42 | async def create_transcriptions( |
| 43 | raw_request: Request, request: Annotated[TranscriptionRequest, Form()] |
| 44 | ): |
| 45 | handler = transcription(raw_request) |
| 46 | if handler is None: |
| 47 | raise NotImplementedError("The model does not support Transcriptions API") |
| 48 | |
| 49 | audio_data = await read_upload_with_limit(request.file) |
| 50 | |
| 51 | generator = await handler.create_transcription(audio_data, request, raw_request) |
| 52 | |
| 53 | if isinstance(generator, ErrorResponse): |
| 54 | return JSONResponse( |
| 55 | content=generator.model_dump(), status_code=generator.error.code |
| 56 | ) |
| 57 | |
| 58 | elif isinstance(generator, TranscriptionResponseVariant): |
| 59 | return JSONResponse(content=generator.model_dump()) |
| 60 | |
| 61 | return StreamingResponse(content=generator, media_type="text/event-stream") |