app.py

import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
import os

MODEL_ID = "piphany/Qwen2.5-7B-Law-Assistant"

def main():
    # --- 1. 加载模型和分词器 (从HF Hub加载) ---
    print(f"开始从Hugging Face Hub加载模型: {MODEL_ID}")
    

    # 加载分词器
    tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, trust_remote_code=True)
    tokenizer.pad_token = tokenizer.eos_token

    
    # 加载完整的、已合并的模型 (已移除quantization_config)
    # Load the full, merged model (quantization_config has been removed)
    model = AutoModelForCausalLM.from_pretrained(
        MODEL_ID,
        torch_dtype=torch.float16,
        device_map="auto",
        trust_remote_code=True
    )

    model.eval() 
    print("模型加载完成，准备启动Web界面...")

    # --- 2. 定义推理函数 ---
    def predict(case_fact, history):
        instruction = "你是一个专业的法律助手。请根据以下提供的案件事实，草拟一份判决书初稿。"
        user_content = f"{instruction}\n\n案件事实如下：\n{case_fact}"
        
        chat_history_for_model = [{"role": "user", "content": user_content}]
        
        messages = tokenizer.apply_chat_template(
            chat_history_for_model,
            tokenize=False,
            add_generation_prompt=True
        )
        
        model_inputs = tokenizer([messages], return_tensors="pt").to(model.device)
        
        generated_ids = model.generate(
            model_inputs.input_ids,
            max_new_tokens=2048,
            do_sample=True,
            top_p=0.9,
            temperature=0.5,
            repetition_penalty=1.1,
            eos_token_id=tokenizer.eos_token_id,
        )

        generated_ids = [
            output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
        ]
        response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
        
        return response

    # --- 3. 启动Gradio Web UI ---
    iface = gr.Interface(
        fn=predict,
        inputs=gr.Textbox(lines=15, label="请输入案件事实", placeholder="在这里粘贴完整的案情描述..."),
        outputs=gr.Textbox(lines=20, label="生成的判决书初稿"),
        title="智能判案辅助系统 Demo",
        description=f"本系统基于模型 {MODEL_ID} 微调。请输入案情事实，系统将为您生成一份判决书初稿。",
        theme="soft",
        allow_flagging="never"
    )

    print("\nGradio应用正在启动...")
    # 在HF Spaces上，share=True是默认的，不需要特别设置
    iface.launch()


if __name__ == '__main__':
    main()