fxi

src/server/main.py

@@ -14,7 +14,7 @@ from pydantic_settings import BaseSettings
 from slowapi import Limiter
 from slowapi.util import get_remote_address
 import torch
-from transformers import AutoModelForSeq2SeqLM, AutoTokenizer, AutoProcessor, BitsAndBytesConfig, Gemma3ForConditionalGeneration
+from transformers import AutoModelForSeq2SeqLM, AutoTokenizer, AutoProcessor, BitsAndBytesConfig, Gemma3ForConditionalGeneration, AutoModel
 from IndicTransToolkit import IndicProcessor
 import json
 import asyncio
@@ -25,6 +25,7 @@ import requests
 from starlette.responses import StreamingResponse
 from logging_config import logger
 from tts_config import SPEED, ResponseFormat, config as tts_config
+import torchaudio
 
 # Device setup
 if torch.cuda.is_available():
@@ -815,7 +816,6 @@ async def transcribe_audio(file: UploadFile = File(...), language: str = Query(.
     if not asr_manager.model:
         raise HTTPException(status_code=503, detail="ASR model still loading, please try again later")
     try:
-        import torchaudio
         wav, sr = torchaudio.load(file.file)
         wav = torch.mean(wav, dim=0, keepdim=True)
         target_sample_rate = 16000