fix

README.md

@@ -1,6 +1,6 @@
 ---
-title: FinGPT Chatbot
-emoji: 💰
+title: FinGPT Sentiment Analysis
+emoji: 📊
 colorFrom: blue
 colorTo: green
 sdk: gradio
@@ -11,40 +11,48 @@ license: mit
 hf_oauth: true
 ---
 
-# 🤖 FinGPT Chatbot
+# 📊 FinGPT 金融新闻情感分析
 
-这是一个基于 **FinGPT/fingpt-mt_llama3-8b_lora** 模型的金融对话助手Spaces应用。
+这是一个基于 **FinGPT/fingpt-mt_llama3-8b_lora** 模型的金融新闻情感分析工具。
 
 ## ⚠️ 重要配置
 
 由于使用了Llama 3基础模型，需要在Spaces设置中配置访问权限：
 
 1. 确保你的HF账号已经获得 [Meta-Llama-3-8B](https://huggingface.co/meta-llama/Meta-Llama-3-8B) 的访问权限
-2. 在Spaces的Settings中添加 `HF_TOKEN` secret（使用你的Hugging Face访问令牌）
+2. 在Spaces的Settings中启用 `hf_oauth: true`（已在README配置中启用）
 
 ## 功能特性
 
-- 💬 智能对话：基于金融领域微调的对话系统
-- 🧠 金融专业：使用Llama 3-8B + LoRA适配器
-- 🚀 GPU加速：使用Hugging Face Spaces的GPU支持
-- 💾 智能缓存：模型文件本地缓存，加速启动
+- 📰 **情感分析**：自动分析金融新闻的情感倾向（positive/neutral/negative）
+- 🧠 **专业模型**：基于Llama 3-8B在金融市场趋势分析任务上微调
+- 🚀 **GPU加速**：使用Hugging Face Spaces的GPU支持快速推理
+- 🌐 **英文支持**：针对英文金融新闻优化
 
 ## 使用说明
 
-1. 在输入框中输入您的金融相关问题
-2. 点击"发送"按钮或按Enter键
-3. 等待模型生成回复
-4. 可以点击"清空对话历史"重新开始
+1. 在输入框中粘贴或输入金融新闻文本
+2. 点击"分析情感"按钮或按Enter键
+3. 查看AI分析的情感结果：
+   - **positive（积极）**：利好消息，市场看涨信号
+   - **neutral（中性）**：中性报道，无明显倾向
+   - **negative（消极）**：利空消息，市场看跌信号
 
 ## 模型信息
 
 - 基础模型：meta-llama/Meta-Llama-3-8B
 - LoRA适配器：FinGPT/fingpt-mt_llama3-8b_lora
-- 任务类型：金融对话和分析
+- 任务类型：Market Trend Analysis（市场趋势分析）
+- 输出类别：positive / neutral / negative
 
-## 示例问题
+## 应用场景
 
-- 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？
+- 📈 量化交易：情感信号作为交易策略输入
+- 📊 市场研究：批量分析新闻情绪趋势
+- 🔍 风险监控：实时监测负面新闻
+- 💼 投资决策：辅助判断市场情绪
+
+## 参考
+
+- [FinGPT项目](https://github.com/AI4Finance-Foundation/FinGPT)
+- [模型页面](https://huggingface.co/FinGPT/fingpt-mt_llama3-8b_lora)

app.py

@@ -3,147 +3,135 @@ import spaces
 import torch
 from peft import PeftModel
 import os
-from datasets import load_dataset
-from transformers import AutoModel, AutoTokenizer, AutoModelForCausalLM, LlamaForCausalLM, LlamaTokenizerFast
 
 
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers.generation.utils import GenerationConfig
 # 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")
 
+print("=" * 50)
+print("开始加载FinGPT情感分析模型...")
+print("=" * 50)
 
 model = None
 tokenizer = None
+device = 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()
+
+    tokenizer = AutoTokenizer.from_pretrained("Go4miii/DISC-FinLLM", use_fast=False, trust_remote_code=True)
+    model = AutoModelForCausalLM.from_pretrained("Go4miii/DISC-FinLLM", device_map="auto", torch_dtype=torch.float16, trust_remote_code=True)
+    model.generation_config = GenerationConfig.from_pretrained("Go4miii/DISC-FinLLM")
+
 
 except Exception as e:
     print("\n" + "=" * 50)
     print("❌ 模型加载失败！")
     print(f"错误信息: {e}")
+    print("=" * 50)
     raise
 
 
 @spaces.GPU
-def chat(message, history):
+def analyze_sentiment(news_text):
     """
-    聊天函数，处理用户消息并返回模型响应
+    分析金融新闻的情感倾向
     """
     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。"
+        return "❌ 模型未正确加载，请检查Spaces日志。"
 
     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)
+        # 构建prompt（按照FinGPT的格式）
+        prompt = f'''Instruction: What is the sentiment of this news? Please choose an answer from {{negative/neutral/positive}}
+Input: {news_text}
+Answer: '''
 
         # 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()}
+        tokens = tokenizer(
+            prompt,
+            return_tensors='pt',
+            padding=True,
+            max_length=512,
+            truncation=True
+        ).to(device)
 
         # 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,
+            res = model.generate(
+                **tokens,
+                max_length=512,
                 pad_token_id=tokenizer.eos_token_id
             )
 
         # 解码输出
-        response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+        res_sentence = tokenizer.decode(res[0], skip_special_tokens=True)
+
+        # 提取答案
+        if "Answer: " in res_sentence:
+            sentiment = res_sentence.split("Answer: ")[1].strip()
+            # 清理多余的换行和空格
+            sentiment = sentiment.split('\n')[0].strip()
+        else:
+            sentiment = res_sentence
 
-        # 提取助手的回复
-        if "Assistant:" in response:
-            response = response.split("Assistant:")[-1].strip()
+        return sentiment
 
-        return response
     except Exception as e:
-        return f"❌ 生成回复时出错: {str(e)}"
+        return f"❌ 分析出错: {str(e)}"
 
 
-# 创建Gradio Chatbot界面
+# 创建Gradio界面
 with gr.Blocks(theme=gr.themes.Soft()) as demo:
     gr.Markdown(
         """
-        # 🤖 FinGPT Chatbot
+        # 📊 FinGPT 金融新闻情���分析
         
-        这是一个基于 **FinGPT/fingpt-forecaster_dow30_llama2-7b_lora** 模型的金融对话助手。
+        基于 **FinGPT/fingpt-mt_llama3-8b_lora** 模型的金融新闻情感分析工具。
         
-        您可以询问关于金融市场、投资、经济分析等问题。
+        输入金融新闻文本，AI将分析其情感倾向：**positive（积极）** / **neutral（中性）** / **negative（消极）**
         """
     )
 
-    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")
+        with gr.Column():
+            news_input = gr.Textbox(
+                label="📰 输入金融新闻",
+                placeholder="粘贴或输入金融新闻内容...",
+                lines=6
+            )
+            analyze_btn = gr.Button("🔍 分析情感", variant="primary", size="lg")
 
-    clear = gr.Button("清空对话历史")
+        with gr.Column():
+            sentiment_output = gr.Textbox(
+                label="😊 情感分析结果",
+                lines=2
+            )
 
     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？"
+            "FINANCING OF ASPOCOMP 'S GROWTH Aspocomp is aggressively pursuing its growth strategy by increasingly focusing on technologically more demanding HDI printed circuit boards PCBs.",
+            "According to Gran, the company has no plans to move all production to Russia, although that is where the company is growing.",
+            "Apple Inc. reported record quarterly revenue of $123.9 billion, up 11% year over year, and quarterly earnings per diluted share of $2.10.",
+            "The Federal Reserve announced a 0.75 percentage point interest rate increase, the largest since 1994, to combat rising inflation.",
+            "Tesla shares tumbled 12% after the company missed delivery expectations for the third consecutive quarter.",
+            "Microsoft and OpenAI announced a multi-year partnership to develop advanced AI technologies, with Microsoft investing $10 billion.",
         ],
-        inputs=msg
+        inputs=news_input,
+        label="📋 示例新闻（点击使用）"
     )
 
     # 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
+    analyze_btn.click(
+        fn=analyze_sentiment,
+        inputs=news_input,
+        outputs=sentiment_output
     )
-    submit.click(user_message, [msg, chatbot], [msg, chatbot], queue=False).then(
-        bot_message, chatbot, chatbot
+
+    news_input.submit(
+        fn=analyze_sentiment,
+        inputs=news_input,
+        outputs=sentiment_output
     )
     clear.click(lambda: None, None, chatbot, queue=False)