hf_token

README.md

@@ -8,12 +8,20 @@ sdk_version: 5.9.1
 app_file: app.py
 pinned: false
 license: mit
+hf_oauth: true
 ---
 
 # 🤖 FinGPT Chatbot
 
 这是一个基于 **FinGPT/fingpt-mt_llama3-8b_lora** 模型的金融对话助手Spaces应用。
 
+## ⚠️ 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）
+
 ## 功能特性
 
 - 💬 实时对话：支持多轮对话，保持上下文

app.py

@@ -3,26 +3,41 @@ import spaces
 import torch
 from transformers import AutoTokenizer, AutoModelForCausalLM
 from peft import PeftModel
+import os
 
 # 加载模型和tokenizer
 model_name = "meta-llama/Meta-Llama-3-8B"
 adapter_name = "FinGPT/fingpt-mt_llama3-8b_lora"
 
+# 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("正在加载模型...")
-tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
-tokenizer.pad_token = tokenizer.eos_token
+try:
+    tokenizer = AutoTokenizer.from_pretrained(
+        model_name,
+        trust_remote_code=True,
+        token=hf_token
+    )
+    tokenizer.pad_token = tokenizer.eos_token
 
-base_model = AutoModelForCausalLM.from_pretrained(
-    model_name,
-    torch_dtype=torch.float16,
-    device_map="auto",
-    trust_remote_code=True
-)
+    base_model = AutoModelForCausalLM.from_pretrained(
+        model_name,
+        torch_dtype=torch.float16,
+        device_map="auto",
+        trust_remote_code=True,
+        token=hf_token
+    )
 
-model = PeftModel.from_pretrained(base_model, adapter_name)
-model = model.eval()
+    model = PeftModel.from_pretrained(base_model, adapter_name)
+    model = model.eval()
+
+    print("模型加载完成！")
+except Exception as e:
+    print(f"模型加载错误: {e}")
+    print("请确保在Spaces设置中添加了HF_TOKEN")
+    raise
 
-print("模型加载完成！")
 
 @spaces.GPU
 def chat(message, history):
@@ -34,16 +49,16 @@ def chat(message, history):
     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(
@@ -54,16 +69,17 @@ def chat(message, history):
             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
 
+
 # 创建Gradio Chatbot界面
 with gr.Blocks(theme=gr.themes.Soft()) as demo:
     gr.Markdown(
@@ -75,13 +91,13 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
         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",
@@ -89,9 +105,9 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
             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？",
@@ -101,17 +117,17 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
         ],
         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
     )