import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer

# 加载模型和分词器
model_name = "nvidia/NVLM-D-72B"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

def generate_response(prompt):
    # 编码输入
    inputs = tokenizer(prompt, return_tensors="pt")
    
    # 生成输出
    outputs = model.generate(**inputs)
    
    # 解码输出
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return response

# 创建 Gradio 接口
iface = gr.Interface(
    fn=generate_response,
    inputs=gr.inputs.Textbox(lines=2, placeholder="输入你的问题..."),
    outputs="text",
    title="NVLM-D-72B 交互式问答",
    description="使用 NVIDIA 的 NVLM-D-72B 模型进行问答。"
)

# 启动应用
if __name__ == "__main__":
    iface.launch()