Update src/promptings/DebateCoder.py

src/promptings/DebateCoder.py

@@ -16,7 +16,6 @@ try:
 except ImportError:
     print("Warning: Could not import Pangu model. Debugging might fail if Pangu is not defined.")
 
-from models.Qwen import Qwen
 from .Base import BaseStrategy
 from models.Base import BaseModel
 from results.Results import Results
@@ -57,7 +56,7 @@ class DebateCoder(BaseStrategy):
     
     def log_response(self, response: str, stage: str, item: dict):
         """记录响应到日志文件"""
-        log_file = os.path.join(self.log_dir, f"DebateCoder_{self.model.__class__.__name__}_responses.log")
+        log_file = os.path.join(self.log_dir, f"MultiAgentDebate_{self.model.__class__.__name__}_responses.log")
         
         try:
             with open(log_file, "a", encoding="utf-8") as f:
@@ -102,7 +101,6 @@ Your responsibility is to ensure:
         """生成初始计划（第一轮）"""
         role_prompt = self.get_agent_role_prompt(role)
         problem_description = self.data.get_prompt(item)
-        
         messages = [
             {
                 "role": "system",
@@ -126,7 +124,7 @@ Provide a numbered list of steps to solve this problem."""
         print(f"\n{'='*60}")
         print(f"[Round 1] {role} - Initial Planning")
         print(f"{'='*60}")
-        
+        print("messages:", messages)
         response, pr_tok, com_tok = self.gpt_chat(messages)
         item['api_calls'] = item.get('api_calls', 0) + 1
         
@@ -435,7 +433,7 @@ Output ONLY a number between 0-100, nothing else."""
         return code_str
 
 
-    def Reviewer_pangu1b(self,problem_prompt: str, code: str, test_log: str, task_id: str = "unknown") -> Tuple[str, int, int]:
+    def Reviewer_pangu1b(self,problem_prompt: str, plan: str, code: str, test_log: str, task_id: str = "unknown") -> Tuple[str, int, int]:
         """
         Reviewer 角色：分析代码失败原因并提供修复计划。
         
@@ -453,7 +451,7 @@ Output ONLY a number between 0-100, nothing else."""
         reviewer_input = [
             {
                 "role": "user",
-                "content": f"You are an expert programmer. The following code was generated to solve a problem but failed sample test cases.\n\n## Problem:\n{problem_prompt}\n\n## Generated Code:\n```\n{code}\n```\n\n## Test Report:\n{test_log}\n\nPlease analyze why the code failed and provide a specific plan to fix it. Do not generate the full code, just the analysis and fix plan."
+                "content": f"You are an expert programmer. The following code was generated to solve a problem but failed sample test cases.\n\n## Problem:\n{problem_prompt}\n\n## Plan:\n{plan}\n \n## Generated Code:\n```\n{code}\n```\n\n## Test Report:\n{test_log}\n\nPlease analyze why the code failed and provide a specific plan to fix it. Do not generate the full code, just the analysis and fix plan."
             }
         ]
         
@@ -477,54 +475,11 @@ Output ONLY a number between 0-100, nothing else."""
             return "Code failed sample tests. Please check logic and edge cases.", 0, 0
 
 
-    def Reviewer_qwen32b(self,problem_prompt: str, code: str, test_log: str, task_id: str = "unknown") -> Tuple[str, int, int]:
-        """
-        Reviewer 角色：分析代码失败原因并提供修复计划。
-        
-        Args:
-            problem_prompt (str): 题目描述。
-            code (str):生成的代码。
-            test_log (str): 测试失败的日志报告。
-            task_id (str): 当前任务的ID，用于日志打印 (对应原代码中的 i)。
-            
-        Returns:
-            Tuple[str, int, int]: 返回 (分析结果, prompt_token消耗, completion_token消耗)
-        """
-        
-        # 构造 Prompt
-        reviewer_input = [
-            {
-                "role": "user",
-                "content": f"You are an expert programmer. The following code was generated to solve a problem but failed sample test cases.\n\n## Problem:\n{problem_prompt}\n\n## Generated Code:\n```\n{code}\n```\n\n## Test Report:\n{test_log}\n\nPlease analyze why the code failed and provide a specific plan to fix it. Do not generate the full code, just the analysis and fix plan."
-            }
-        ]
-        
-        print(f"Input for Reviewer analysis: {task_id}")
-
-        try:
-            qwen_model = Qwen() 
-            analysis, q_pr_tok, q_com_tok = qwen_model.prompt(reviewer_input)
-            
-            print(f"Reviewer Analysis: {analysis}", flush=True)
-            return analysis, q_pr_tok, q_com_tok
-
-        except NameError:
-            print("Error: Qwen model class not found. Skipping detailed analysis.")
-            return "Code failed sample tests. Please check logic and edge cases.", 0, 0
-            
-        except Exception as e:
-            print(f"Error during Reviewer analysis: {e}")
-            return "Code failed sample tests. Please check logic and edge cases.", 0, 0
-
-
-
-
-
     def debugging(self, plan: list, code: str, item: dict, algorithm_prompt: str) -> str:
         passed = False
         planning, _, _ = plan
         
-        plan_code_response = f"## Planning: {planning}\n## Code:\n```\n{code}\n```"
+        # plan_code_response = f"## Planning: {planning}\n## Code:\n```\n{code}\n```"
 
         if type(self.data) == APPSDataset or type(self.data) == CodeContestDataset or type(self.data) == XCodeDataset:
             std_input_prompt = "## Note: Strictly follow the input and output format. The input should be taken from Standard input and output should be given to standard output. If you are writing a function then after the function definition take input using `input()` function then call the function with specified parameters and finally print the output of the function. Do not add extra print statement otherwise it will failed the test cases."
@@ -545,13 +500,14 @@ Output ONLY a number between 0-100, nothing else."""
             problem_prompt = self.data.get_prompt(item)
 
             # 调用 Reviewer
-            analysis, pr_cost, com_cost = self.Reviewer_qwen32b(
+            analysis, pr_cost, com_cost = self.Reviewer_pangu1b(
                 problem_prompt=problem_prompt,
+                plan = planning,
                 code=code,
                 test_log=test_log,
                 task_id=i
             )
-
+            print(f"pr_cost: {pr_cost}, com_cost: {com_cost}")
             # 更新 Token 计数
             self.pr_tok += pr_cost
             self.com_tok += com_cost
@@ -560,8 +516,7 @@ Output ONLY a number between 0-100, nothing else."""
             input_for_improving_code = [
                 {
                     "role": "user",
-                    "content": f"Given a competitive programming problem you have generated {self.language} code to solve the problem. But the generated code can not pass sample test cases.\n\nHere is an analysis of the failure and a fix plan provided by an expert:\n{analysis}\n\nImprove your code to solve the problem correctly based on this analysis.\n{algorithm_prompt}\n## Problem to be solved:\n{self.data.get_prompt(item)}\n{plan_code_response}\n## Test Report:\n{test_log}\n## Modified Planning:\n## Let's think step by step to modify {self.language} Code for solving this problem.\n\n----------------\nImportant:\n{std_input_prompt}\n## Your response must contain the modified planning and then the {self.language} code inside ``` block to solve this problem."
-                }
+                    "content": f" You are an expert competitive programmer. Your task is to fix the provided {self.language} code based on the Expert Analysis. The original code failed sample test cases.\n\n### Problem Description: \n{self.data.get_prompt(item)}\n ### Original Code (Buggy): \n```\n{code}\n```\n### Expert Analysis & Fix Plan:\n{analysis}\n\nImprove your code to solve the problem correctly based on this analysis.\n### Requirement: 1. Read the Expert Analysis carefully. 2. {std_input_prompt} 3. Generate the corrected code.\n\n----------------\nImportant:\n{std_input_prompt}\n## Your response must contain the modified planning and then the {self.language} code inside ``` block to solve this problem."                }
             ]
 
             print("\n\n________________________")
@@ -611,6 +566,9 @@ Output ONLY a number between 0-100, nothing else."""
 
     def run_single_pass(self, item: dict):
         """执行单个问题的多智能体辩论流程"""
+        self.pr_tok = 0
+        self.com_tok = 0
+        
         print("\n" + "="*80)
         print(f"Processing: {item.get('task_id', item.get('name', 'unknown'))}")
         print("="*80)