better：新增config项目

src/do_tool/tool_use.py

@@ -19,7 +19,7 @@ logger = get_module_logger("tool_use", config=tool_use_config)
 class ToolUser:
     def __init__(self):
         self.llm_model_tool = LLM_request(
-            model=global_config.llm_heartflow, temperature=0.2, max_tokens=1000, request_type="tool_use"
+            model=global_config.llm_tool_use, temperature=0.2, max_tokens=1000, request_type="tool_use"
         )
 
     async def _build_tool_prompt(self, message_txt: str, sender_name: str, chat_stream: ChatStream, subheartflow: SubHeartflow = None):

src/heart_flow/observation.py

@@ -29,17 +29,10 @@ class ChattingObservation(Observation):
         self.name = global_config.BOT_NICKNAME
         self.nick_name = global_config.BOT_ALIAS_NAMES
 
-        self.observe_times = 0
+        self.max_now_obs_len = global_config.observation_context_size
-
+        self.overlap_len = global_config.compressed_length
-        self.summary_count = 0  # 30秒内的更新次数
-        self.max_update_in_30s = 2  # 30秒内最多更新2次
-        self.last_summary_time = 0  # 上次更新summary的时间
-
-        self.sub_observe = None
-        self.max_now_obs_len = 20
-        self.overlap_len = 5
         self.mid_memorys = []
-        self.max_mid_memory_len = 5
+        self.max_mid_memory_len = global_config.compress_length_limit
         self.mid_memory_info = ""
         self.now_message_info = ""
         

src/heart_flow/sub_heartflow.py

@@ -79,7 +79,7 @@ class SubHeartflow:
         self.past_mind = []
         self.current_state: CurrentState = CurrentState()
         self.llm_model = LLM_request(
-            model=global_config.llm_sub_heartflow, temperature=0.2, max_tokens=600, request_type="sub_heart_flow"
+            model=global_config.llm_sub_heartflow, temperature=global_config.llm_sub_heartflow["temp"], max_tokens=600, request_type="sub_heart_flow"
         )
 
         self.main_heartflow_info = ""

src/plugins/PFC/action_planner.py

@@ -24,7 +24,7 @@ class ActionPlanner:
 
     def __init__(self, stream_id: str):
         self.llm = LLM_request(
-            model=global_config.llm_normal, temperature=0.2, max_tokens=1000, request_type="action_planning"
+            model=global_config.llm_normal, temperature=global_config.llm_normal["temp"], max_tokens=1000, request_type="action_planning"
         )
         self.personality_info = Individuality.get_instance().get_prompt(type="personality", x_person=2, level=2)
         self.name = global_config.BOT_NICKNAME

src/plugins/PFC/pfc_KnowledgeFetcher.py

@@ -13,7 +13,7 @@ class KnowledgeFetcher:
 
     def __init__(self):
         self.llm = LLM_request(
-            model=global_config.llm_normal, temperature=0.7, max_tokens=1000, request_type="knowledge_fetch"
+            model=global_config.llm_normal, temperature=global_config.llm_normal["temp"], max_tokens=1000, request_type="knowledge_fetch"
         )
 
     async def fetch(self, query: str, chat_history: List[Message]) -> Tuple[str, str]:

src/plugins/PFC/reply_generator.py

@@ -16,7 +16,7 @@ class ReplyGenerator:
 
     def __init__(self, stream_id: str):
         self.llm = LLM_request(
-            model=global_config.llm_normal, temperature=0.2, max_tokens=300, request_type="reply_generation"
+            model=global_config.llm_normal, temperature=global_config.llm_normal["temp"], max_tokens=300, request_type="reply_generation"
         )
         self.personality_info = Individuality.get_instance().get_prompt(type="personality", x_person=2, level=2)
         self.name = global_config.BOT_NICKNAME

src/plugins/chat_module/reasoning_chat/reasoning_generator.py

@@ -29,7 +29,7 @@ class ResponseGenerator:
             request_type="response_reasoning",
         )
         self.model_normal = LLM_request(
-            model=global_config.llm_normal, temperature=0.8, max_tokens=256, request_type="response_reasoning"
+            model=global_config.llm_normal, temperature=global_config.llm_normal["temp"], max_tokens=256, request_type="response_reasoning"
         )
 
         self.model_sum = LLM_request(

src/plugins/chat_module/think_flow_chat/think_flow_generator.py

@@ -26,7 +26,7 @@ logger = get_module_logger("llm_generator", config=llm_config)
 class ResponseGenerator:
     def __init__(self):
         self.model_normal = LLM_request(
-            model=global_config.llm_normal, temperature=0.15, max_tokens=256, request_type="response_heartflow"
+            model=global_config.llm_normal, temperature=global_config.llm_normal["temp"], max_tokens=256, request_type="response_heartflow"
         )
 
         self.model_sum = LLM_request(
@@ -49,7 +49,7 @@ class ResponseGenerator:
             if random.random() > 0:
                 checked = False
                 current_model = self.model_normal
-                current_model.temperature = 0.3 * arousal_multiplier  # 激活度越高，温度越高
+                current_model.temperature = global_config.llm_normal["temp"] * arousal_multiplier  # 激活度越高，温度越高
                 model_response = await self._generate_response_with_model(
                     message, current_model, thinking_id, mode="normal"
                 )
@@ -58,13 +58,13 @@ class ResponseGenerator:
             else:
                 checked = True
                 current_model = self.model_normal
-                current_model.temperature = 0.3 * arousal_multiplier  # 激活度越高，温度越高
+                current_model.temperature = global_config.llm_normal["temp"] * arousal_multiplier  # 激活度越高，温度越高
                 print(f"生成{message.processed_plain_text}回复温度是：{current_model.temperature}")
                 model_response = await self._generate_response_with_model(
                     message, current_model, thinking_id, mode="simple"
                 )
 
-                current_model.temperature = 0.3
+                current_model.temperature = global_config.llm_normal["temp"]
                 model_checked_response = await self._check_response_with_model(
                     message, model_response, current_model, thinking_id
                 )

src/plugins/config/config.py

@@ -26,9 +26,9 @@ config_config = LogConfig(
 logger = get_module_logger("config", config=config_config)
 
 # 考虑到，实际上配置文件中的mai_version是不会自动更新的,所以采用硬编码
-is_test = True
+is_test = False
 mai_version_main = "0.6.2"
-mai_version_fix = "snapshot-2"
+mai_version_fix = ""
 
 if mai_version_fix:
     if is_test:
@@ -196,6 +196,9 @@ class BotConfig:
     sub_heart_flow_freeze_time: int = 120  # 子心流冻结时间，超过这个时间没有回复，子心流会冻结，间隔 单位秒
     sub_heart_flow_stop_time: int = 600  # 子心流停止时间，超过这个时间没有回复，子心流会停止，间隔 单位秒
     heart_flow_update_interval: int = 300  # 心流更新频率，间隔 单位秒
+    observation_context_size: int = 20  # 心流观察到的最长上下文大小，超过这个值的上下文会被压缩
+    compressed_length: int = 5  # 不能大于observation_context_size,心流上下文压缩的最短压缩长度，超过心流观察到的上下文长度，会压缩，最短压缩长度为5
+    compress_length_limit: int = 5  #最多压缩份数，超过该数值的压缩上下文会被删除
 
     # willing
     willing_mode: str = "classical"  # 意愿模式
@@ -440,6 +443,16 @@ class BotConfig:
             config.heart_flow_update_interval = heartflow_config.get(
                 "heart_flow_update_interval", config.heart_flow_update_interval
             )
+            if config.INNER_VERSION in SpecifierSet(">=1.3.0"):
+                config.observation_context_size = heartflow_config.get(
+                    "observation_context_size", config.observation_context_size
+                )
+                config.compressed_length = heartflow_config.get(
+                    "compressed_length", config.compressed_length
+                )
+                config.compress_length_limit = heartflow_config.get(
+                    "compress_length_limit", config.compress_length_limit
+                )
 
         def willing(parent: dict):
             willing_config = parent["willing"]
@@ -477,7 +490,7 @@ class BotConfig:
                 "llm_emotion_judge",
                 "vlm",
                 "embedding",
-                "moderation",
+                "llm_tool_use",
                 "llm_observation",
                 "llm_sub_heartflow",
                 "llm_heartflow",
@@ -489,7 +502,7 @@ class BotConfig:
 
                     # base_url 的例子： SILICONFLOW_BASE_URL
                     # key 的例子： SILICONFLOW_KEY
-                    cfg_target = {"name": "", "base_url": "", "key": "", "stream": False, "pri_in": 0, "pri_out": 0}
+                    cfg_target = {"name": "", "base_url": "", "key": "", "stream": False, "pri_in": 0, "pri_out": 0, "temp": 0.7}
 
                     if config.INNER_VERSION in SpecifierSet("<=0.0.0"):
                         cfg_target = cfg_item
@@ -502,6 +515,7 @@ class BotConfig:
                             stable_item.append("stream")
 
                         pricing_item = ["pri_in", "pri_out"]
+                        
                         # 从配置中原始拷贝稳定字段
                         for i in stable_item:
                             # 如果 字段 属于计费项 且获取不到，那默认值是 0
@@ -519,6 +533,13 @@ class BotConfig:
                                     logger.error(f"{item} 中的必要字段不存在，请检查")
                                     raise KeyError(f"{item} 中的必要字段 {e} 不存在，请检查") from e
 
+                        # 如果配置中有temp参数，就使用配置中的值
+                        if "temp" in cfg_item:
+                            cfg_target["temp"] = cfg_item["temp"]
+                        else:
+                            # 如果没有temp参数，就删除默认值
+                            cfg_target.pop("temp", None)
+
                         provider = cfg_item.get("provider")
                         if provider is None:
                             logger.error(f"provider 字段在模型配置 {item} 中不存在，请检查")

template/bot_config_template.toml

@@ -1,5 +1,5 @@
 [inner]
-version = "1.2.7"
+version = "1.3.0"
 
 
 #以下是给开发人员阅读的，一般用户不需要阅读
@@ -77,7 +77,11 @@ model_v3_probability = 0.3 # 麦麦回答时选择次要回复模型2 模型的
 sub_heart_flow_update_interval = 60 # 子心流更新频率，间隔 单位秒
 sub_heart_flow_freeze_time = 100 # 子心流冻结时间，超过这个时间没有回复，子心流会冻结，间隔 单位秒
 sub_heart_flow_stop_time = 500 # 子心流停止时间，超过这个时间没有回复，子心流会停止，间隔 单位秒
-heart_flow_update_interval = 300 # 心流更新频率，间隔 单位秒
+heart_flow_update_interval = 600 # 心流更新频率，间隔 单位秒
+
+observation_context_size = 20 # 心流观察到的最长上下文大小，超过这个值的上下文会被压缩
+compressed_length = 5 # 不能大于observation_context_size,心流上下文压缩的最短压缩长度，超过心流观察到的上下文长度，会压缩，最短压缩长度为5
+compress_length_limit = 5 #最多压缩份数，超过该数值的压缩上下文会被删除
 
 
 [message]
@@ -189,11 +193,12 @@ pri_out = 16 #模型的输出价格（非必填，可以记录消耗）
 
 #非推理模型
 
-[model.llm_normal] #V3 回复模型1 主要回复模型，默认temp 0.2 如果你使用的是老V3或者其他模型，请自己修改代码中的temp参数
+[model.llm_normal] #V3 回复模型1 主要回复模型，默认temp 0.2 如果你使用的是老V3或者其他模型，请自己修改temp参数
 name = "Pro/deepseek-ai/DeepSeek-V3"
 provider = "SILICONFLOW"
 pri_in = 2 #模型的输入价格（非必填，可以记录消耗）
 pri_out = 8 #模型的输出价格（非必填，可以记录消耗）
+temp = 0.2 #模型的温度，新V3建议0.1-0.3
 
 [model.llm_emotion_judge] #表情包判断 
 name = "Qwen/Qwen2.5-14B-Instruct"
@@ -213,11 +218,11 @@ provider = "SILICONFLOW"
 pri_in = 1.26
 pri_out = 1.26
 
-[model.moderation] #内容审核，开发中
+[model.llm_tool_use] #工具调用模型，需要使用支持工具调用的模型，建议使用qwen2.5 32b
-name = ""
+name = "Qwen/Qwen2.5-32B-Instruct"
 provider = "SILICONFLOW"
-pri_in = 1.0
+pri_in = 1.26
-pri_out = 2.0
+pri_out = 1.26
 
 # 识图模型
 
@@ -247,6 +252,7 @@ name = "Pro/deepseek-ai/DeepSeek-V3"
 provider = "SILICONFLOW"
 pri_in = 2
 pri_out = 8
+temp = 0.2 #模型的温度，新V3建议0.1-0.3
 
 [model.llm_heartflow] #心流：建议使用qwen2.5 32b
 # name = "Pro/Qwen/Qwen2.5-7B-Instruct"