File size: 2,174 Bytes
e2e1acd
 
 
15a6c65
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
# Install required dependency
# !pip install mistral-common

import gradio as gr
import torch
import tempfile
import os
from typing import List, Tuple
from transformers import VoxtralForConditionalGeneration, AutoProcessor

device = "cuda" if torch.cuda.is_available() else "cpu"
repo_id = "mistralai/Voxtral-Mini-3B-2507"

processor = AutoProcessor.from_pretrained(repo_id)
model = VoxtralForConditionalGeneration.from_pretrained(
    repo_id,
    torch_dtype=torch.bfloat16,
    device_map=device,
)

def respond(audio_files: List[str], question: str) -> Tuple[str, List[str]]:
    if not audio_files:
        return "Please upload at least one audio file.", []

    conversation = [
        {
            "role": "user",
            "content": [
                {"type": "audio", "path": path} for path in audio_files
            ] + [{"type": "text", "text": question}],
        }
    ]

    inputs = processor.apply_chat_template(conversation)
    inputs = inputs.to(device, dtype=torch.bfloat16)

    with torch.no_grad():
        outputs = model.generate(**inputs, max_new_tokens=500)
    decoded = processor.batch_decode(
        outputs[:, inputs.input_ids.shape[1]:],
        skip_special_tokens=True,
    )
    return decoded[0], audio_files

demo = gr.Interface(
    fn=respond,
    inputs=[
        gr.Audio(type="filepath", label="Audio files", file_count="multiple"),
        gr.Textbox(lines=2, placeholder="Ask something about the audio(s)...", label="Question"),
    ],
    outputs=[
        gr.Textbox(label="Answer"),
        gr.Gallery(label="Uploaded audio files"),
    ],
    title="Voxtral-Mini-3B-2507 Audio Q&A",
    description="Upload one or more audio files and ask any question about them.",
    examples=[
        [
            [
                "https://huggingface.co/datasets/hf-internal-testing/dummy-audio-samples/resolve/main/mary_had_lamb.mp3",
                "https://huggingface.co/datasets/hf-internal-testing/dummy-audio-samples/resolve/main/winning_call.mp3",
            ],
            "What sport and what nursery rhyme are referenced?",
        ]
    ],
    cache_examples=False,
)

if __name__ == "__main__":
    demo.launch()