modify assistant

modified_assistant.py

@@ -128,77 +128,8 @@ class ModifiedAssistant(Assistant):
                         used_any_tool = True
                 
                 # 如果使用了工具，继续循环
-                if used_any_tool:
+                if not used_any_tool:
-                    continue
-                # 如果没有使用工具，无调用次数，已经执行了2次以上循环，退出循环
-                if num_llm_calls_available ==0 or total_num_llm_calls_available - num_llm_calls_available >=2:
                     break
-                # 如果没有使用工具，还有调用次数，并只执行了1次，就需要调用模型判断回答是否完整（修复部分case不执行工具调用就停止的问题）
-                if num_llm_calls_available > 0:
-                    # 构建判断消息 - 使用英文系统提示词，并在用户提示中包含具体内容
-                    user_question = messages[-1].content if messages[-1].content else str(messages[-1])
-                    assistant_response = output[-1].content if hasattr(output[-1], 'content') and output[-1].content else str(output[-1])
-                    
-                    judge_messages = [
-                        Message(role='system', content='''You are a professional conversation completeness evaluator. Your task is to determine whether an AI assistant has provided a complete and sufficient answer to a user's question based on strict evaluation criteria.
-
-## Evaluation Criteria
-
-### Complete Answer Characteristics (mark as "complete" if ANY of these are met):
-1. **Information Completeness**: Provides core information users need, including specific parameters, specifications, data, or detailed descriptions
-2. **Problem Resolution**: Directly addresses the user's question with clear answers or solutions
-3. **Recommendation Adequacy**: When providing product/service recommendations, includes key information (model, parameters, price, features, etc.)
-4. **Conclusion Clarity**: Provides clear conclusions, advice, or summaries
-
-### Incomplete Answer Characteristics (mark as "incomplete" if ANY of these are met):
-1. **Information Gaps**: Obviously lacks key information needed to answer the question
-2. **Vague Responses**: Uses evasive language and avoids specific questions
-3. **Unfulfilled Promises**: Promises to provide more information but doesn't deliver
-4. **Needs Follow-up**: Clearly indicates more information is needed to continue answering
-
-## Special Evaluation Rules
-- **Recommendation Questions**: Consider complete if specific recommendations with key information are provided
-- **Technical Questions**: Consider complete if technical details and specific data are provided  
-- **Comparison Questions**: Consider complete if valuable comparative information is provided
-- **Inquiry Questions**: Consider complete if best effort answer is given based on available information
-
-## Output Format
-Only output "complete" or "incomplete" with no additional text or explanation.
-
-Adhere strictly to these standards for objective evaluation.'''),
-                        Message(role='user', content=f'''Please evaluate whether the following assistant response is complete based on the evaluation criteria.
-
-USER QUESTION:
-{user_question}
-
-ASSISTANT RESPONSE:
-{assistant_response}
-
-Based on the evaluation criteria, is this response complete? Only reply with "complete" or "incomplete".''')
-                    ]
-                    # 调用模型进行判断（不使用工具）
-                    judge_stream = self._call_llm_with_retry(messages=judge_messages, functions=[], extra_generate_cfg={'lang': 'en'})
-                    judge_output = []
-                    for judge_msg in judge_stream:
-                        if judge_msg:
-                            judge_output.extend(judge_msg)
-                    
-                    # 分析判断结果 - 使用正则匹配检测 complete
-                    is_complete = False
-                    if judge_output:
-                        judge_content = judge_output[-1].content if hasattr(judge_output[-1], 'content') else str(judge_output[-1])
-                        judge_content = judge_content.lower().strip()
-                        print(judge_content)
-                        # 使用正则匹配检测 complete
-                        import re
-                        
-                        if re.search(r'incomplete', judge_content):
-                            is_complete = False
-                        else:
-                            is_complete = True
-                    # 如果回答完整，退出循环；否则继续
-                    if is_complete:
-                        break
                     
         yield response