examples/transcribe.py

#!/usr/bin/env python3
from __future__ import annotations

import argparse
from pathlib import Path

import torch
from transformers import AutoModelForCausalLM, AutoProcessor


PROMPT = "Please transcribe this audio."


def build_conversation(audio_path: Path) -> list[dict]:
    return [
        {
            "role": "user",
            "content": [
                {"type": "audio", "path": str(audio_path)},
                {"type": "text", "text": PROMPT},
            ],
        }
    ]


def main() -> None:
    parser = argparse.ArgumentParser(description="Transcribe one audio file with audio8-asr-0.1B.")
    parser.add_argument("audio", type=Path)
    parser.add_argument("--model", default=".")
    parser.add_argument("--device", default="cuda" if torch.cuda.is_available() else "cpu")
    parser.add_argument("--max_new_tokens", type=int, default=128)
    parser.add_argument("--max_audio_seconds", type=int, default=30)
    args = parser.parse_args()

    device = torch.device(args.device)
    dtype = torch.bfloat16 if device.type == "cuda" else torch.float32
    processor = AutoProcessor.from_pretrained(args.model, trust_remote_code=True)
    model = AutoModelForCausalLM.from_pretrained(
        args.model,
        trust_remote_code=True,
        torch_dtype=dtype,
        attn_implementation="eager",
    ).to(device)
    model.eval()

    batch = processor.apply_chat_template(
        build_conversation(args.audio),
        return_tensors="pt",
        sampling_rate=16000,
        audio_padding="longest",
        add_generation_prompt=True,
        audio_max_length=int(args.max_audio_seconds) * 16000,
        text_kwargs={"padding": "longest", "truncation": True, "max_length": 1000},
    )
    batch = {key: value.to(device) if hasattr(value, "to") else value for key, value in dict(batch).items()}
    with torch.inference_mode():
        output_ids = model.generate(**batch, max_new_tokens=args.max_new_tokens, do_sample=False)
    prompt_len = int(batch["input_ids"].shape[1])
    text = processor.decode(output_ids[0, prompt_len:], skip_special_tokens=True).strip()
    print(text)


if __name__ == "__main__":
    main()