Spaces:
Build error
Build error
File size: 1,305 Bytes
62b3774 71f6464 db56cf5 71f6464 db56cf5 a405ea7 db56cf5 62b3774 db56cf5 a405ea7 2a8b0bb a405ea7 2a8b0bb a405ea7 2b62b02 a405ea7 db56cf5 71f6464 1ba4a0c 71f6464 2a8b0bb 71f6464 f35ab4d 71f6464 b932faf |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 |
import os
import pickle
from typing import Union
from fastapi import Request
import torch
from transformers import pipeline
from fastapi import FastAPI
from contextlib import asynccontextmanager
DEVICE = os.getenv('DEVICE', 'mps')
ATTN_IMPLEMENTATION = os.getenv('ATTN_IMPLEMENTATION', "sdpa")
@asynccontextmanager
async def lifespan(app: FastAPI):
app.state.transcribe_pipeline = pipeline(
"automatic-speech-recognition",
model="openai/whisper-large-v3",
torch_dtype=torch.float16 if ATTN_IMPLEMENTATION == "sdpa" else torch.bfloat16,
device=DEVICE,
model_kwargs={"attn_implementation": ATTN_IMPLEMENTATION},
)
app.state.transcribe_pipeline.model.to('cuda')
yield
app = FastAPI(lifespan=lifespan)
@app.get("/")
def read_root():
return {"status": "ok"}
@app.post("/transcribe")
async def transcribe(request: Request):
body = await request.body()
audio_chunk = pickle.loads(body)
outputs = app.state.transcribe_pipeline(
audio_chunk,
chunk_length_s=30,
batch_size=24,
generate_kwargs={
'task': 'transcribe',
'language': 'english'
},
return_timestamps='word'
)
text = outputs["text"].strip()
return {"transcribe": text, "outputs": outputs}
|