app/model.py

import torch
from transformers import pipeline

MODEL_NAME = "openai/whisper-small"

def load_model():
    torch.set_num_threads(2)  # Important for HF free CPU

    device = 0 if torch.cuda.is_available() else -1

    asr_pipeline = pipeline(
    "automatic-speech-recognition",
    model=MODEL_NAME,
    device=device,
    chunk_length_s=20,
    stride_length_s=5,
)


    return asr_pipeline