app.py

import gradio as gr
import spaces
import torch
from peft import PeftModel
import os
from datasets import load_dataset
from transformers import AutoModel, AutoTokenizer, AutoModelForCausalLM, LlamaForCausalLM, LlamaTokenizerFast


# 获取HF token（Spaces会自动提供）
hf_token = os.environ.get("HF_TOKEN") or os.environ.get("HUGGING_FACE_HUB_TOKEN")


model = None
tokenizer = None

try:
    base_model = "meta-llama/Meta-Llama-3-8B"
    peft_model = "FinGPT/fingpt-mt_llama3-8b_lora"
    tokenizer = LlamaTokenizerFast.from_pretrained(base_model, trust_remote_code=True)
    tokenizer.pad_token = tokenizer.eos_token
    model = LlamaForCausalLM.from_pretrained(base_model, trust_remote_code=True, device_map="cuda:0")
    model = PeftModel.from_pretrained(model, peft_model)
    model = model.eval()
    # model = AutoModel.from_pretrained("tuananhle/fingpt-forecaster_dow30_qwen3-8b_lora_250814_v3", dtype="auto")
    # base_model = AutoModelForCausalLM.from_pretrained(
    #     'meta-llama/Llama-2-7b-chat-hf',
    #     trust_remote_code=True,
    #     device_map="auto",
    #     torch_dtype=torch.float16,   # optional if you have enough VRAM
    #     token=hf_token,
    # )
    # tokenizer = AutoTokenizer.from_pretrained('meta-llama/Llama-2-7b-chat-hf',
    #                                           trust_remote_code=True,
    #                                           token=hf_token,
    #                                           )
    # model = PeftModel.from_pretrained(base_model, 'FinGPT/fingpt-forecaster_dow30_llama2-7b_lora')
    # model = model.eval()

except Exception as e:
    print("\n" + "=" * 50)
    print("❌ 模型加载失败！")
    print(f"错误信息: {e}")
    raise


@spaces.GPU
def chat(message, history):
    """
    聊天函数，处理用户消息并返回模型响应
    """
    if model is None or tokenizer is None:
        return "❌ 模型未正确加载，请检查Spaces日志获取详细错误信息。"

    try:
        # 构建对话历史
        conversation = []
        for user_msg, bot_msg in history:
            conversation.append(f"User: {user_msg}")
            conversation.append(f"Assistant: {bot_msg}")

        conversation.append(f"User: {message}")
        conversation.append("Assistant:")

        prompt = "\n".join(conversation)

        # 编码输入
        inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=2048)
        inputs = {k: v.to(model.device) for k, v in inputs.items()}

        # 生成响应
        with torch.no_grad():
            outputs = model.generate(
                **inputs,
                max_new_tokens=512,
                temperature=0.7,
                top_p=0.9,
                do_sample=True,
                pad_token_id=tokenizer.eos_token_id
            )

        # 解码输出
        response = tokenizer.decode(outputs[0], skip_special_tokens=True)

        # 提取助手的回复
        if "Assistant:" in response:
            response = response.split("Assistant:")[-1].strip()

        return response
    except Exception as e:
        return f"❌ 生成回复时出错: {str(e)}"


# 创建Gradio Chatbot界面
with gr.Blocks(theme=gr.themes.Soft()) as demo:
    gr.Markdown(
        """
        # 🤖 FinGPT Chatbot
        
        这是一个基于 **FinGPT/fingpt-forecaster_dow30_llama2-7b_lora** 模型的金融对话助手。
        
        您可以询问关于金融市场、投资、经济分析等问题。
        """
    )

    chatbot = gr.Chatbot(
        label="聊天记录",
        height=500,
        bubble_full_width=False
    )

    with gr.Row():
        msg = gr.Textbox(
            label="输入您的消息",
            placeholder="在这里输入您的问题...",
            scale=4
        )
        submit = gr.Button("发送", scale=1, variant="primary")

    clear = gr.Button("清空对话历史")

    gr.Examples(
        examples=[
            "什么是量化宽松政策？",
            "如何评估一只股票的价值？",
            "请解释一下技术分析中的MACD指标",
            "当前市场环境下应该如何配置资产？"
        ],
        inputs=msg
    )

    # 事件处理
    def user_message(user_msg, history):
        return "", history + [[user_msg, None]]

    def bot_message(history):
        user_msg = history[-1][0]
        bot_response = chat(user_msg, history[:-1])
        history[-1][1] = bot_response
        return history

    msg.submit(user_message, [msg, chatbot], [msg, chatbot], queue=False).then(
        bot_message, chatbot, chatbot
    )
    submit.click(user_message, [msg, chatbot], [msg, chatbot], queue=False).then(
        bot_message, chatbot, chatbot
    )
    clear.click(lambda: None, None, chatbot, queue=False)

if __name__ == "__main__":
    demo.launch()