app.py

import gradio as gr
from transformers import AutoModelForCausalLM, AutoProcessor
from PIL import Image
import torch

model_id = "skt/A.X-4.0-VL-Light"

model = AutoModelForCausalLM.from_pretrained(
    model_id, trust_remote_code=True, torch_dtype=torch.bfloat16
).to("cuda")

processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)

def ask_image_question(image, text):
    messages = [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": text}]}]
    inputs = processor(images=[image], conversations=[messages], return_tensors="pt").to("cuda")

    generation_kwargs = {
        "max_new_tokens": 256,
        "top_p": 0.8,
        "temperature": 0.5,
        "top_k": 20,
        "repetition_penalty": 1.05,
        "do_sample": True,
    }

    generated_ids = model.generate(**inputs, **generation_kwargs)
    output = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
    return output

gr.Interface(fn=ask_image_question, 
             inputs=[gr.Image(type="pil"), gr.Textbox(label="질문")], 
             outputs="text").launch()