Update app.py

app.py

@@ -27,6 +27,31 @@ def debug_print(message):
     if DEBUG_MODE:
         print(f"[DEBUG] {message}")
 
+# 添加网络连接测试函数
+def test_hf_connection():
+    """测试Hugging Face API连接"""
+    try:
+        import requests
+        test_url = "https://api-inference.huggingface.co/v1/models"
+        headers = {"Authorization": f"Bearer {HF_TOKEN}"}
+        
+        debug_print("测试Hugging Face API连接...")
+        response = requests.get(test_url, headers=headers, timeout=10)
+        debug_print(f"连接测试响应状态: {response.status_code}")
+        
+        if response.status_code == 200:
+            debug_print("Hugging Face API连接成功")
+            return True
+        elif response.status_code == 401:
+            print("错误: Hugging Face API密钥无效")
+            return False
+        else:
+            print(f"警告: Hugging Face API连接测试返回状态码 {response.status_code}")
+            return True  # 其他状态码可能仍然可以工作
+    except Exception as e:
+        print(f"警告: 无法连接到Hugging Face API: {e}")
+        return False
+
 def extract_json_from_response(response_text):
     """
     从模型响应中提取并解析JSON（修复UTF-8编码问题）
@@ -45,6 +70,19 @@ def extract_json_from_response(response_text):
         except Exception:
             pass
 
+    # 清理响应文本，移除可能的前缀或后缀
+    response_text = response_text.strip()
+    
+    # 移除可能的 Markdown 代码块标记
+    if response_text.startswith("```json"):
+        response_text = response_text[7:]
+    if response_text.startswith("```"):
+        response_text = response_text[3:]
+    if response_text.endswith("```"):
+        response_text = response_text[:-3]
+        
+    response_text = response_text.strip()
+
     # 方法1: 直接解析
     try:
         return json.loads(response_text)
@@ -83,7 +121,8 @@ def extract_json_from_response(response_text):
         except (json.JSONDecodeError, UnicodeDecodeError) as e:
             debug_print(f"清理后的JSON解析失败: {e}")
 
-    raise ValueError(f"无法从响应中提取有效的JSON。")
+    # 如果所有方法都失败，抛出更详细的错误信息
+    raise ValueError(f"无法从响应中提取有效的JSON。响应长度: {len(response_text)}, 响应前缀: {response_text[:100] if len(response_text) > 100 else response_text}")
 
 # =======================
 # 工具函数：调用模型
@@ -105,7 +144,23 @@ def query_model(prompt, model_name="Qwen/Qwen3-4B-Thinking-2507:nscale"):
         api_key=HF_TOKEN  # 使用你的 Token
     )
 
+    # 添加模型名称验证和备用模型
+    available_models = [
+        "Qwen/Qwen3-4B-Thinking-2507:nscale",
+        "Qwen/Qwen2.5-7B-Instruct:nscale",
+        "meta-llama/Llama-3.1-8B-Instruct:nscale",
+        "mistralai/Mistral-Nemo-Instruct-2407:nscale"
+    ]
+    
+    # 如果指定的模型不在可用列表中，使用默认模型
+    if model_name not in available_models:
+        print(f"警告: 模型 {model_name} 不在推荐列表中，将使用默认模型")
+        model_name = "Qwen/Qwen3-4B-Thinking-2507:nscale"
+
     try:
+        debug_print(f"正在调用模型: {model_name}")
+        debug_print(f"提示词长度: {len(prompt)} 字符")
+        
         response = client.chat.completions.create(
             model=model_name,
             messages=[
@@ -114,21 +169,70 @@ def query_model(prompt, model_name="Qwen/Qwen3-4B-Thinking-2507:nscale"):
             max_tokens=1000,
             temperature=0.7
         )
+        
+        debug_print(f"API响应状态: 成功")
+        
         if response.choices:
             content = response.choices[0].message.content.strip()
+            debug_print(f"响应内容长度: {len(content) if content else 0} 字符")
+            
+            # 检查空响应
+            if not content:
+                print("警告: 模型返回空内容")
+                return None
+                
             try:
                 content = content.encode('utf-8').decode('utf-8')
             except Exception:
                 pass
             return content
         else:
+            print("警告: API响应中没有choices字段")
             return None
     except Exception as e:
-        print(f"Error calling model: {e}")
+        error_msg = f"Error calling model: {e}"
+        print(error_msg)
+        debug_print(f"错误详情: {type(e).__name__}: {str(e)}")
+        
+        # 更详细的错误处理
         if "401" in str(e):
             print("错误：API密钥无效。请检查HF_TOKEN环境变量。")
         elif "404" in str(e):
-            print(f"错误：模型 {model_name} 未找到。")
+            print(f"错误：模型 {model_name} 未找到。尝试使用备用模型...")
+            # 尝试备用模型
+            for backup_model in available_models:
+                if backup_model != model_name:
+                    print(f"尝试备用模型: {backup_model}")
+                    try:
+                        response = client.chat.completions.create(
+                            model=backup_model,
+                            messages=[
+                                {"role": "user", "content": prompt}
+                            ],
+                            max_tokens=1000,
+                            temperature=0.7
+                        )
+                        if response.choices:
+                            content = response.choices[0].message.content.strip()
+                            # 检查空响应
+                            if not content:
+                                print(f"备用模型 {backup_model} 返回空内容")
+                                continue
+                                
+                            try:
+                                content = content.encode('utf-8').decode('utf-8')
+                            except Exception:
+                                pass
+                            print(f"备用模型 {backup_model} 调用成功")
+                            return content
+                    except Exception as backup_e:
+                        print(f"备用模型 {backup_model} 调用失败: {backup_e}")
+                        continue
+            print("所有模型都调用失败")
+        elif "timeout" in str(e).lower() or "time out" in str(e).lower():
+            print("错误：API调用超时，请检查网络连接。")
+        elif "connection" in str(e).lower():
+            print("错误：网络连接问题，请检查网络设置。")
         return None
 
 # =======================
@@ -165,15 +269,27 @@ def plan_tasks(user_request, filename=None):
     planning_response = query_model(planning_prompt)
     debug_print(f"模型响应: {planning_response[:200] if planning_response else '空响应'}")
 
+    # 增强空响应处理
+    if not planning_response:
+        print("任务规划失败: 模型返回空响应")
+        print("可能的原因:")
+        print("1. Hugging Face API密钥无效")
+        print("2. 网络连接问题")
+        print("3. 模型当前不可用")
+        print("4. 请求超时")
+        raise Exception("任务规划失败: 模型返回空响应")
+
     try:
-        if not planning_response:
-            raise ValueError("模型返回空响应")
         plan = extract_json_from_response(planning_response)
         debug_print("JSON解析成功")
         return plan
     except Exception as e:
         error_msg = f"规划解析失败: {e}\n模型响应内容: {planning_response[:500] if planning_response else '空响应'}"
         print(error_msg)
+        # 添加更多调试信息
+        if planning_response:
+            print(f"响应长度: {len(planning_response)} 字符")
+            print(f"响应前100字符: {planning_response[:100] if len(planning_response) > 100 else planning_response}")
         raise Exception("任务规划失败，请重新描述您的需求。")
 
 # =======================
@@ -311,7 +427,21 @@ def ai_agent_master(uploaded_file, user_request):
         chat_history.append({"role": "assistant", "content": "正在分析需求并分解任务..."})
         yield chat_history, None
 
-        task_plan = plan_tasks(user_request, uploaded_file.name if uploaded_file else None)
+        # 添加重试机制
+        max_retries = 3
+        retry_count = 0
+        task_plan = None
+        
+        while retry_count < max_retries and task_plan is None:
+            try:
+                task_plan = plan_tasks(user_request, uploaded_file.name if uploaded_file else None)
+            except Exception as e:
+                retry_count += 1
+                if retry_count < max_retries:
+                    print(f"任务规划失败，{retry_count}秒后重试... (第{retry_count}次)")
+                    time.sleep(retry_count)  # 递增延迟
+                else:
+                    raise e  # 达到最大重试次数，抛出异常
 
         if not task_plan or "steps" not in task_plan:
             raise Exception("任务规划返回无效格式，请重试。")
@@ -330,10 +460,28 @@ def ai_agent_master(uploaded_file, user_request):
             chat_history.append({"role": "assistant", "content": f"正在调用模型 {step_model} 执行步骤 {step_num}: {step_desc}..."})
             yield chat_history, None
 
-            step_result, step_file = execute_step(step, uploaded_file)
+            # 添加步骤执行重试机制
+            step_retry_count = 0
+            step_result = None
+            step_file = None
+            
+            while step_retry_count < max_retries and step_result is None:
+                try:
+                    step_result, step_file = execute_step(step, uploaded_file)
+                except Exception as e:
+                    step_retry_count += 1
+                    if step_retry_count < max_retries:
+                        print(f"步骤执行失败，{step_retry_count}秒后重试... (第{step_retry_count}次)")
+                        time.sleep(step_retry_count)
+                    else:
+                        raise e  # 达到最大重试次数，抛出异常
+            
             if step_file:
                 final_generated_file = step_file
 
+            if not step_result:
+                raise Exception(f"步骤 {step_num} 执行返回空结果")
+
             display_result = step_result if len(step_result) < 500 else step_result[:500] + "... (结果过长已截断)"
             chat_history.append({"role": "assistant", "content": f"步骤 {step_num} 结果: {display_result}"})
             yield chat_history, None
@@ -341,12 +489,30 @@ def ai_agent_master(uploaded_file, user_request):
             chat_history.append({"role": "assistant", "content": f"正在验证步骤 {step_num}..."})
             yield chat_history, None
 
-            is_valid, validation_feedback = validate_step(step, step_result, user_request)
+            # 添加验证重试机制
+            validation_retry_count = 0
+            is_valid = False
+            validation_feedback = ""
+            
+            while validation_retry_count < max_retries and not is_valid:
+                try:
+                    is_valid, validation_feedback = validate_step(step, step_result, user_request)
+                except Exception as e:
+                    validation_retry_count += 1
+                    if validation_retry_count < max_retries:
+                        print(f"步骤验证失败，{validation_retry_count}秒后重试... (第{validation_retry_count}次)")
+                        time.sleep(validation_retry_count)
+                    else:
+                        # 验证失败不终止整个流程，但记录警告
+                        print(f"步骤 {step_num} 验证失败: {e}")
+                        is_valid = True  # 继续执行而不是终止
+                        validation_feedback = "验证过程出现异常，但继续执行下一步"
 
             if not is_valid:
                 chat_history.append({"role": "assistant", "content": f"步骤 {step_num} 未通过验证: {validation_feedback}"})
                 yield chat_history, None
-                raise Exception(f"步骤 {step_num} 验证失败，任务中止。")
+                # 不再抛出异常终止整个流程，而是记录警告继续
+                print(f"警告: 步骤 {step_num} 验证失败，但继续执行下一步")
 
             chat_history.append({"role": "assistant", "content": f"步骤 {step_num} 验证通过！反馈: {validation_feedback}"})
             yield chat_history, None
@@ -357,7 +523,9 @@ def ai_agent_master(uploaded_file, user_request):
         yield chat_history, final_generated_file
 
     except Exception as e:
-        chat_history.append({"role": "assistant", "content": f"执行过程中出现错误: {str(e)}"})
+        error_msg = f"执行过程中出现错误: {str(e)}"
+        print(f"详细错误信息: {type(e).__name__}: {str(e)}")
+        chat_history.append({"role": "assistant", "content": error_msg})
         yield chat_history, None
 
 # =======================
@@ -389,4 +557,15 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
 
 # 启动
 if __name__ == "__main__":
+    print("正在初始化AI智能体...")
+    if DEBUG_MODE:
+        print("调试模式已启用")
+    
+    # 测试Hugging Face连接
+    if test_hf_connection():
+        print("Hugging Face API连接测试通过")
+    else:
+        print("警告: Hugging Face API连接测试失败，但仍将继续启动...")
+    
+    print("启动Gradio界面...")
     demo.launch()