app.py

from transformers import AutoConfig, AutoModel
import gradio as gr
import spaces

model_path = "turing-motors/Heron-NVILA-Lite-15B"

# or directly from_pretrained
model = AutoModel.from_pretrained(model_path, trust_remote_code=True, device_map="auto")

# examples generate using generation_config
from PIL import Image
import requests
from transformers import GenerationConfig
generation_config = {
    "max_new_tokens": 2048,
    "temperature": 0.5,
    "do_sample": True,
}
generation_config = GenerationConfig(**generation_config)

DESCRIPTION = '''
<div>
<h1 style="text-align: center;">非公式Heron-NVILA-Lite-15B</h1>
<p>Heron-NVILA-Lite-15Bの非公式デモだよ。 <a href="https://huggingface.co/turing-motors/Heron-NVILA-Lite-15B"><b>turing-motors/Heron-NVILA-Lite-15B</b></a>.</p>
</div>
'''

@spaces.GPU()
def infer(image):
    response = model.generate_content(
        [image, "画像を可能な限り詳細に説明してください。"],
        generation_config=generation_config
    )

    return response

css = """
h1 {
  text-align: center;
  display: block;
}
#col-container {
    margin: 0 auto;
    max-width: 520px;
}
"""

with gr.Blocks(fill_height=True, css=css) as demo:
    with gr.Column(elem_id="col-container"):
        gr.Markdown(DESCRIPTION)
        image_input=gr.Image(type="pil", label="画像をアップロード")
        run_button = gr.Button("画像の説明文を生成する", scale=0)
        text_output=gr.Textbox(label="生成されたテキスト")
    run_button.click(
        fn = infer,
        inputs = image_input,
        outputs = text_output,
    )
    
if __name__ == "__main__":
    demo.launch(mcp_server=True)