app.py

import torch
import gradio as gr
from transformers import AutoProcessor, Qwen3VLForConditionalGeneration
import os

MODEL_ID = "yuyan-chen/Lep-Description-Qwen3-VL-2B-Instruct"

token = os.environ.get("HF_TOKEN")

device = "cuda" if torch.cuda.is_available() else "cpu"

processor = AutoProcessor.from_pretrained(MODEL_ID, trust_remote_code=True, token=token)
model = Qwen3VLForConditionalGeneration.from_pretrained(
    MODEL_ID,
    torch_dtype=torch.float16 if device == "cuda" else torch.float32,
    device_map="auto",
    trust_remote_code=True,
    token=token
)

def infer(image, text):
    messages = [
        {
            "role": "user",
            "content": [
                {"type": "image", "image": image},
                {"type": "text", "text": text},
            ],
        }
    ]

    inputs = processor(
        messages,
        return_tensors="pt"
    ).to(device)

    with torch.no_grad():
        output_ids = model.generate(
            **inputs,
            max_new_tokens=512,
            do_sample=True,
            temperature=0.7
        )

    return processor.decode(
        output_ids[0],
        skip_special_tokens=True
    )

demo = gr.Interface(
    fn=infer,
    inputs=[
        gr.Image(type="pil", label="Image"),
        gr.Textbox(label="Prompt")
    ],
    outputs=gr.Textbox(label="Response"),
    title="Qwen3-VL-2B (Finetuned)",
    description="Upload an image and ask a question."
)

demo.launch()