fix：提供更自定义的max_token配置

弱智api服务商输出</think>输出一半被max_tokens截断了

src/chat/focus_chat/planners/planner_simple.py

@@ -78,13 +78,11 @@ class ActionPlanner(BasePlanner):
         # LLM规划器配置
         self.planner_llm = LLMRequest(
             model=global_config.model.planner,
-            max_tokens=1000,
             request_type="focus.planner",  # 用于动作规划
         )
 
         self.utils_llm = LLMRequest(
             model=global_config.model.utils_small,
-            max_tokens=1000,
             request_type="focus.planner",  # 用于动作规划
         )
 
@@ -188,6 +186,12 @@ class ActionPlanner(BasePlanner):
                 llm_content, (reasoning_content, _) = await self.planner_llm.generate_response_async(prompt=prompt)
 
                 logger.debug(f"{self.log_prefix}LLM 原始理由响应: {reasoning_content}")
+                
+                logger.info(f"{self.log_prefix}规划器原始提示词: {prompt}")
+                logger.info(f"{self.log_prefix}规划器原始响应: {llm_content}")
+                logger.info(f"{self.log_prefix}规划器推理: {reasoning_content}")
+                
+                
             except Exception as req_e:
                 logger.error(f"{self.log_prefix}LLM 请求执行失败: {req_e}")
                 reasoning = f"LLM 请求失败，你的模型出现问题: {req_e}"

src/chat/normal_chat/normal_chat_planner.py

@@ -43,8 +43,7 @@ def init_prompt():
 
 基于以上聊天上下文和用户的最新消息，选择最合适的action。
 
-请以动作的输出要求，以严格的 JSON 格式输出，且仅包含 JSON 内容。
+请以动作的输出要求，以严格的 JSON 格式输出，且仅包含 JSON 内容。不要有任何其他文字或解释：
-请输出你提取的JSON，不要有任何其他文字或解释：
 """,
         "normal_chat_planner_prompt",
     )
@@ -70,7 +69,6 @@ class NormalChatPlanner:
         # LLM规划器配置
         self.planner_llm = LLMRequest(
             model=global_config.model.planner,
-            max_tokens=1000,
             request_type="normal_chat.planner",  # 用于normal_chat动作规划
         )
 

src/config/official_configs.py

@@ -439,7 +439,6 @@ class ModelConfig(ConfigBase):
     focus_working_memory: dict[str, Any] = field(default_factory=lambda: {})
     """专注工作记忆模型配置"""
 
-
     focus_tool_use: dict[str, Any] = field(default_factory=lambda: {})
     """专注工具使用模型配置"""
 

src/llm_models/utils_model.py

@@ -123,6 +123,8 @@ class LLMRequest:
         self.stream = model.get("stream", False)
         self.pri_in = model.get("pri_in", 0)
         self.pri_out = model.get("pri_out", 0)
+        self.max_tokens = model.get("max_tokens", global_config.model.model_max_output_length)
+        # print(f"max_tokens: {self.max_tokens}")
 
         # 获取数据库实例
         self._init_database()
@@ -243,6 +245,26 @@ class LLMRequest:
         if stream_mode:
             payload["stream"] = stream_mode
             
+        if self.temp != 0.7:
+            payload["temperature"] = self.temp
+
+        # 添加enable_thinking参数（如果不是默认值False）
+        if not self.enable_thinking:
+            payload["enable_thinking"] = False
+
+        if self.thinking_budget != 4096:
+            payload["thinking_budget"] = self.thinking_budget
+
+        if self.max_tokens:
+            payload["max_tokens"] = self.max_tokens
+        
+        # if "max_tokens" not in payload and "max_completion_tokens" not in payload:
+            # payload["max_tokens"] = global_config.model.model_max_output_length
+        # 如果 payload 中依然存在 max_tokens 且需要转换，在这里进行再次检查
+        if self.model_name.lower() in self.MODELS_NEEDING_TRANSFORMATION and "max_tokens" in payload:
+            payload["max_completion_tokens"] = payload.pop("max_tokens")    
+            
+
         return {
             "policy": policy,
             "payload": payload,
@@ -631,6 +653,7 @@ class LLMRequest:
             ]
         else:
             messages = [{"role": "user", "content": prompt}]
+        
         payload = {
             "model": self.model_name,
             "messages": messages,
@@ -648,8 +671,11 @@ class LLMRequest:
         if self.thinking_budget != 4096:
             payload["thinking_budget"] = self.thinking_budget
 
-        if "max_tokens" not in payload and "max_completion_tokens" not in payload:
+        if self.max_tokens:
-            payload["max_tokens"] = global_config.model.model_max_output_length
+            payload["max_tokens"] = self.max_tokens
+        
+        # if "max_tokens" not in payload and "max_completion_tokens" not in payload:
+            # payload["max_tokens"] = global_config.model.model_max_output_length
         # 如果 payload 中依然存在 max_tokens 且需要转换，在这里进行再次检查
         if self.model_name.lower() in self.MODELS_NEEDING_TRANSFORMATION and "max_tokens" in payload:
             payload["max_completion_tokens"] = payload.pop("max_tokens")