From 9c4681805f3be6cb9b68c5130ea03988432fe83e Mon Sep 17 00:00:00 2001 From: SengokuCola <1026294844@qq.com> Date: Sat, 13 Sep 2025 22:08:29 +0800 Subject: [PATCH 1/6] =?UTF-8?q?better:=20=E4=BC=98=E5=8C=96=E6=A8=A1?= =?UTF-8?q?=E5=9E=8B=E9=85=8D=E7=BD=AE=E5=92=8Cmood?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- .../heart_flow/heartflow_message_processor.py | 7 ++- src/config/api_ada_configs.py | 3 -- src/mood/mood_manager.py | 13 +++--- template/model_config_template.toml | 45 +++++++++---------- 4 files changed, 29 insertions(+), 39 deletions(-) diff --git a/src/chat/heart_flow/heartflow_message_processor.py b/src/chat/heart_flow/heartflow_message_processor.py index bd9f0f81..1d90e468 100644 --- a/src/chat/heart_flow/heartflow_message_processor.py +++ b/src/chat/heart_flow/heartflow_message_processor.py @@ -72,16 +72,15 @@ class HeartFCMessageReceiver: chat = message.chat_stream # 2. 兴趣度计算与更新 - interested_rate, keywords = await _calculate_interest(message) + _, keywords = await _calculate_interest(message) await self.storage.store_message(message, chat) heartflow_chat: HeartFChatting = await heartflow.get_or_create_heartflow_chat(chat.stream_id) # type: ignore - # subheartflow.add_message_to_normal_chat_cache(message, interested_rate, is_mentioned) if global_config.mood.enable_mood: chat_mood = mood_manager.get_mood_by_chat_id(heartflow_chat.stream_id) - asyncio.create_task(chat_mood.update_mood_by_message(message, interested_rate)) + asyncio.create_task(chat_mood.update_mood_by_message(message)) # 3. 日志记录 mes_name = chat.group_info.group_name if chat.group_info else "私聊" @@ -109,7 +108,7 @@ class HeartFCMessageReceiver: replace_bot_name=True, ) - logger.info(f"[{mes_name}]{userinfo.user_nickname}:{processed_plain_text}[{interested_rate:.2f}]") # type: ignore + logger.info(f"[{mes_name}]{userinfo.user_nickname}:{processed_plain_text}") # type: ignore _ = Person.register_person( platform=message.message_info.platform, # type: ignore diff --git a/src/config/api_ada_configs.py b/src/config/api_ada_configs.py index bd881bfd..3fc9c878 100644 --- a/src/config/api_ada_configs.py +++ b/src/config/api_ada_configs.py @@ -102,9 +102,6 @@ class ModelTaskConfig(ConfigBase): replyer: TaskConfig """normal_chat首要回复模型模型配置""" - emotion: TaskConfig - """情绪模型配置""" - vlm: TaskConfig """视觉语言模型配置""" diff --git a/src/mood/mood_manager.py b/src/mood/mood_manager.py index 16784230..ed4b5f56 100644 --- a/src/mood/mood_manager.py +++ b/src/mood/mood_manager.py @@ -62,11 +62,11 @@ class ChatMood: self.regression_count: int = 0 - self.mood_model = LLMRequest(model_set=model_config.model_task_config.emotion, request_type="mood") + self.mood_model = LLMRequest(model_set=model_config.model_task_config.utils, request_type="mood") self.last_change_time: float = 0 - async def update_mood_by_message(self, message: MessageRecv, interested_rate: float): + async def update_mood_by_message(self, message: MessageRecv): self.regression_count = 0 during_last_time = message.message_info.time - self.last_change_time # type: ignore @@ -74,10 +74,9 @@ class ChatMood: base_probability = 0.05 time_multiplier = 4 * (1 - math.exp(-0.01 * during_last_time)) - if interested_rate <= 0: - interest_multiplier = 0 - else: - interest_multiplier = 2 * math.pow(interested_rate, 0.25) + # 基于消息长度计算基础兴趣度 + message_length = len(message.message_content.content or "") + interest_multiplier = min(2.0, 1.0 + message_length / 100) logger.debug( f"base_probability: {base_probability}, time_multiplier: {time_multiplier}, interest_multiplier: {interest_multiplier}" @@ -90,7 +89,7 @@ class ChatMood: return logger.debug( - f"{self.log_prefix} 更新情绪状态，感兴趣度: {interested_rate:.2f}, 更新概率: {update_probability:.2f}" + f"{self.log_prefix} 更新情绪状态，更新概率: {update_probability:.2f}" ) message_time: float = message.message_info.time # type: ignore diff --git a/template/model_config_template.toml b/template/model_config_template.toml index 76f09955..f7be4325 100644 --- a/template/model_config_template.toml +++ b/template/model_config_template.toml @@ -1,5 +1,5 @@ [inner] -version = "1.6.0" +version = "1.7.0" # 配置文件版本号迭代规则同bot_config.toml @@ -12,14 +12,14 @@ max_retry = 2 # 最大重试次数（单个模型API timeout = 30 # API请求超时时间（单位：秒） retry_interval = 10 # 重试间隔时间（单位：秒） -[[api_providers]] # SiliconFlow的API服务商配置 -name = "SiliconFlow" -base_url = "https://api.siliconflow.cn/v1" -api_key = "your-siliconflow-api-key" +[[api_providers]] # 阿里 百炼 API服务商配置 +name = "BaiLian" +base_url = "https://dashscope.aliyuncs.com/compatible-mode/v1" +api_key = "your-bailian-key" client_type = "openai" max_retry = 2 -timeout = 30 -retry_interval = 10 +timeout = 15 +retry_interval = 5 [[api_providers]] # 特殊：Google的Gimini使用特殊API，与OpenAI格式不兼容，需要配置client为"gemini" name = "Google" @@ -30,14 +30,14 @@ max_retry = 2 timeout = 30 retry_interval = 10 -[[api_providers]] # 阿里 百炼 API服务商配置 -name = "BaiLian" -base_url = "https://dashscope.aliyuncs.com/compatible-mode/v1" -api_key = "your-bailian-key" +[[api_providers]] # SiliconFlow的API服务商配置 +name = "SiliconFlow" +base_url = "https://api.siliconflow.cn/v1" +api_key = "your-siliconflow-api-key" client_type = "openai" max_retry = 2 -timeout = 15 -retry_interval = 5 +timeout = 60 +retry_interval = 10 [[models]] # 模型（可以配置多个） @@ -93,8 +93,8 @@ price_in = 0 price_out = 0 -[model_task_config.utils] # 在麦麦的一些组件中使用的模型，例如表情包模块，取名模块，关系模块，是麦麦必须的模型 -model_list = ["siliconflow-deepseek-v3"] # 使用的模型列表，每个子项对应上面的模型名称(name) +[model_task_config.utils] # 在麦麦的一些组件中使用的模型，例如表情包模块，取名模块，关系模块，麦麦的情绪变化等，是麦麦必须的模型 +model_list = ["siliconflow-deepseek-v3","qwen3-30b"] # 使用的模型列表，每个子项对应上面的模型名称(name) temperature = 0.2 # 模型温度，新V3建议0.1-0.3 max_tokens = 800 # 最大输出token数 @@ -103,6 +103,11 @@ model_list = ["qwen3-8b","qwen3-30b"] temperature = 0.7 max_tokens = 800 +[model_task_config.tool_use] #工具调用模型，需要使用支持工具调用的模型 +model_list = ["qwen3-30b"] +temperature = 0.7 +max_tokens = 800 + [model_task_config.replyer] # 首要回复模型，还用于表达器和表达方式学习 model_list = ["siliconflow-deepseek-v3"] temperature = 0.3 # 模型温度，新V3建议0.1-0.3 @@ -113,11 +118,6 @@ model_list = ["siliconflow-deepseek-v3"] temperature = 0.3 max_tokens = 800 -[model_task_config.emotion] #负责麦麦的情绪变化 -model_list = ["qwen3-30b"] -temperature = 0.7 -max_tokens = 800 - [model_task_config.vlm] # 图像识别模型 model_list = ["qwen2.5-vl-72b"] max_tokens = 800 @@ -125,11 +125,6 @@ max_tokens = 800 [model_task_config.voice] # 语音识别模型 model_list = ["sensevoice-small"] -[model_task_config.tool_use] #工具调用模型，需要使用支持工具调用的模型 -model_list = ["qwen3-30b"] -temperature = 0.7 -max_tokens = 800 - #嵌入模型 [model_task_config.embedding] model_list = ["bge-m3"] From 25cd19d8ece74334b4f603401d3577630d939d83 Mon Sep 17 00:00:00 2001 From: SengokuCola <1026294844@qq.com> Date: Sun, 14 Sep 2025 17:10:07 +0800 Subject: [PATCH 2/6] =?UTF-8?q?better=EF=BC=9A=E5=AF=B9replyer=E8=BF=9B?= =?UTF-8?q?=E8=A1=8C=E6=B7=B1=E5=BA=A6=E4=BC=98=E5=8C=96?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- src/chat/replyer/default_generator.py | 141 +++++--------------------- src/chat/replyer/lpmm_prompt.py | 53 ++++++++++ src/chat/replyer/replyer_prompt.py | 114 +++++++++++++++++++++ template/bot_config_template.toml | 4 +- 4 files changed, 195 insertions(+), 117 deletions(-) create mode 100644 src/chat/replyer/lpmm_prompt.py create mode 100644 src/chat/replyer/replyer_prompt.py diff --git a/src/chat/replyer/default_generator.py b/src/chat/replyer/default_generator.py index e19bb018..36b3df7d 100644 --- a/src/chat/replyer/default_generator.py +++ b/src/chat/replyer/default_generator.py @@ -31,108 +31,15 @@ from src.person_info.person_info import Person, is_person_known from src.plugin_system.base.component_types import ActionInfo, EventType from src.plugin_system.apis import llm_api +from src.chat.replyer.lpmm_prompt import init_lpmm_prompt +from src.chat.replyer.replyer_prompt import init_replyer_prompt + +init_lpmm_prompt() +init_replyer_prompt() + logger = get_logger("replyer") - -def init_prompt(): - Prompt("你正在qq群里聊天，下面是群里在聊的内容：", "chat_target_group1") - Prompt("你正在和{sender_name}聊天，这是你们之前聊的内容：", "chat_target_private1") - Prompt("在群里聊天", "chat_target_group2") - Prompt("和{sender_name}聊天", "chat_target_private2") - - Prompt( - """ -{expression_habits_block} -{relation_info_block} - -{chat_target} -{time_block} -{chat_info} -{identity} - -你现在的心情是：{mood_state} -你正在{chat_target_2},{reply_target_block} -你想要对上述的发言进行回复，回复的具体内容（原句）是：{raw_reply} -原因是：{reason} -现在请你将这条具体内容改写成一条适合在群聊中发送的回复消息。 -你需要使用合适的语法和句法，参考聊天内容，组织一条日常且口语化的回复。请你修改你想表达的原句，符合你的表达风格和语言习惯 -{reply_style} -你可以完全重组回复，保留最基本的表达含义就好，但重组后保持语意通顺。 -{keywords_reaction_prompt} -{moderation_prompt} -不要输出多余内容(包括前后缀，冒号和引号，括号，表情包，emoji,at或 @等 )，只输出一条回复就好。 -现在，你说： -""", - "default_expressor_prompt", - ) - - # s4u 风格的 prompt 模板 - Prompt( - """{identity} -你正在群聊中聊天，你想要回复 {sender_name} 的发言。同时，也有其他用户会参与聊天，你可以参考他们的回复内容，但是你现在想回复{sender_name}的发言。 - -{time_block} -{background_dialogue_prompt} -{core_dialogue_prompt} - -{expression_habits_block}{tool_info_block} -{knowledge_prompt}{relation_info_block} -{extra_info_block} - -{reply_target_block} -你的心情：{mood_state} -{reply_style} -注意不要复读你说过的话 -{keywords_reaction_prompt} -请注意不要输出多余内容(包括前后缀，冒号和引号，at或 @等 )。只输出回复内容。 -{moderation_prompt} -不要输出多余内容(包括前后缀，冒号和引号，括号()，表情包，emoji,at或 @等 )。只输出一条回复就好 -现在，你说：""", - "replyer_prompt", - ) - - Prompt( - """{identity} -{time_block} -你现在正在一个QQ群里聊天，以下是正在进行的聊天内容： -{background_dialogue_prompt} - -{expression_habits_block}{tool_info_block} -{knowledge_prompt}{relation_info_block} -{extra_info_block} - -你现在想补充说明你刚刚自己的发言内容：{target}，原因是{reason} -请你根据聊天内容，组织一条新回复。注意，{target} 是刚刚你自己的发言，你要在这基础上进一步发言，请按照你自己的角度来继续进行回复。 -注意保持上下文的连贯性。 -你现在的心情是：{mood_state} -{reply_style} -{keywords_reaction_prompt} -请注意不要输出多余内容(包括前后缀，冒号和引号，at或 @等 )。只输出回复内容。 -{moderation_prompt} -不要输出多余内容(包括前后缀，冒号和引号，括号()，表情包，emoji,at或 @等 )。只输出一条回复就好 -现在，你说： -""", - "replyer_self_prompt", - ) - - Prompt( - """ -你是一个专门获取知识的助手。你的名字是{bot_name}。现在是{time_now}。 -群里正在进行的聊天内容： -{chat_history} - -现在，{sender}发送了内容:{target_message},你想要回复ta。 -请仔细分析聊天内容，考虑以下几点： -1. 内容中是否包含需要查询信息的问题 -2. 是否有明确的知识获取指令 - -If you need to use the search tool, please directly call the function "lpmm_search_knowledge". If you do not need to use any tool, simply output "No tool needed". -""", - name="lpmm_get_knowledge_prompt", - ) - - class DefaultReplyer: def __init__( self, @@ -368,7 +275,7 @@ class DefaultReplyer: expression_habits_title = "" if style_habits_str.strip(): expression_habits_title = ( - "你可以参考以下的语言习惯，当情景合适就使用，但不要生硬使用，以合理的方式结合到你的回复中：" + "在回复时,你可以参考以下的语言习惯，当情景合适就使用，但不要生硬使用，以合理的方式结合到你的回复中：" ) expression_habits_block += f"{style_habits_str}\n" @@ -556,18 +463,6 @@ class DefaultReplyer: except Exception as e: logger.error(f"处理消息记录时出错: {msg}, 错误: {e}") - # 构建背景对话 prompt - all_dialogue_prompt = "" - if message_list_before_now: - latest_25_msgs = message_list_before_now[-int(global_config.chat.max_context_size) :] - all_dialogue_prompt_str = build_readable_messages( - latest_25_msgs, - replace_bot_name=True, - timestamp_mode="normal_no_YMD", - truncate=True, - ) - all_dialogue_prompt = f"所有用户的发言：\n{all_dialogue_prompt_str}" - # 构建核心对话 prompt core_dialogue_prompt = "" if core_dialogue_list: @@ -600,6 +495,22 @@ class DefaultReplyer: -------------------------------- """ + + # 构建背景对话 prompt + all_dialogue_prompt = "" + if message_list_before_now: + latest_25_msgs = message_list_before_now[-int(global_config.chat.max_context_size) :] + all_dialogue_prompt_str = build_readable_messages( + latest_25_msgs, + replace_bot_name=True, + timestamp_mode="normal_no_YMD", + truncate=True, + ) + if core_dialogue_prompt: + all_dialogue_prompt = f"所有用户的发言：\n{all_dialogue_prompt_str}" + else: + all_dialogue_prompt = f"{all_dialogue_prompt_str}" + return core_dialogue_prompt, all_dialogue_prompt def build_mai_think_context( @@ -842,11 +753,11 @@ class DefaultReplyer: if sender: if is_group_chat: reply_target_block = ( - f"现在{sender}说的:{target}。引起了你的注意，你想要在群里发言或者回复这条消息。原因是{reply_reason}" + f"现在{sender}说的:{target}。引起了你的注意" ) else: # private chat reply_target_block = ( - f"现在{sender}说的:{target}。引起了你的注意，针对这条消息回复。原因是{reply_reason}" + f"现在{sender}说的:{target}。引起了你的注意" ) else: reply_target_block = "" @@ -1138,4 +1049,4 @@ def weighted_sample_no_replacement(items, weights, k) -> list: return selected -init_prompt() + diff --git a/src/chat/replyer/lpmm_prompt.py b/src/chat/replyer/lpmm_prompt.py new file mode 100644 index 00000000..67baf027 --- /dev/null +++ b/src/chat/replyer/lpmm_prompt.py @@ -0,0 +1,53 @@ +import traceback +import time +import asyncio +import random +import re + +from typing import List, Optional, Dict, Any, Tuple +from datetime import datetime +from src.mais4u.mai_think import mai_thinking_manager +from src.common.logger import get_logger +from src.common.data_models.database_data_model import DatabaseMessages +from src.common.data_models.info_data_model import ActionPlannerInfo +from src.common.data_models.llm_data_model import LLMGenerationDataModel +from src.config.config import global_config, model_config +from src.llm_models.utils_model import LLMRequest +from src.chat.message_receive.message import UserInfo, Seg, MessageRecv, MessageSending +from src.chat.message_receive.chat_stream import ChatStream +from src.chat.message_receive.uni_message_sender import UniversalMessageSender +from src.chat.utils.timer_calculator import Timer # <--- Import Timer +from src.chat.utils.utils import get_chat_type_and_target_info +from src.chat.utils.prompt_builder import Prompt, global_prompt_manager +from src.chat.utils.chat_message_builder import ( + build_readable_messages, + get_raw_msg_before_timestamp_with_chat, + replace_user_references, +) +from src.chat.express.expression_selector import expression_selector +# from src.chat.memory_system.memory_activator import MemoryActivator +from src.mood.mood_manager import mood_manager +from src.person_info.person_info import Person, is_person_known +from src.plugin_system.base.component_types import ActionInfo, EventType +from src.plugin_system.apis import llm_api + + + +def init_lpmm_prompt(): + Prompt( + """ +你是一个专门获取知识的助手。你的名字是{bot_name}。现在是{time_now}。 +群里正在进行的聊天内容： +{chat_history} + +现在，{sender}发送了内容:{target_message},你想要回复ta。 +请仔细分析聊天内容，考虑以下几点： +1. 内容中是否包含需要查询信息的问题 +2. 是否有明确的知识获取指令 + +If you need to use the search tool, please directly call the function "lpmm_search_knowledge". If you do not need to use any tool, simply output "No tool needed". +""", + name="lpmm_get_knowledge_prompt", + ) + + diff --git a/src/chat/replyer/replyer_prompt.py b/src/chat/replyer/replyer_prompt.py new file mode 100644 index 00000000..c726f247 --- /dev/null +++ b/src/chat/replyer/replyer_prompt.py @@ -0,0 +1,114 @@ +import traceback +import time +import asyncio +import random +import re + +from typing import List, Optional, Dict, Any, Tuple +from datetime import datetime +from src.mais4u.mai_think import mai_thinking_manager +from src.common.logger import get_logger +from src.common.data_models.database_data_model import DatabaseMessages +from src.common.data_models.info_data_model import ActionPlannerInfo +from src.common.data_models.llm_data_model import LLMGenerationDataModel +from src.config.config import global_config, model_config +from src.llm_models.utils_model import LLMRequest +from src.chat.message_receive.message import UserInfo, Seg, MessageRecv, MessageSending +from src.chat.message_receive.chat_stream import ChatStream +from src.chat.message_receive.uni_message_sender import UniversalMessageSender +from src.chat.utils.timer_calculator import Timer # <--- Import Timer +from src.chat.utils.utils import get_chat_type_and_target_info +from src.chat.utils.prompt_builder import Prompt, global_prompt_manager +from src.chat.utils.chat_message_builder import ( + build_readable_messages, + get_raw_msg_before_timestamp_with_chat, + replace_user_references, +) +from src.chat.express.expression_selector import expression_selector +# from src.chat.memory_system.memory_activator import MemoryActivator +from src.mood.mood_manager import mood_manager +from src.person_info.person_info import Person, is_person_known +from src.plugin_system.base.component_types import ActionInfo, EventType +from src.plugin_system.apis import llm_api + + + +def init_replyer_prompt(): + Prompt("你正在qq群里聊天，下面是群里正在聊的内容:", "chat_target_group1") + Prompt("你正在和{sender_name}聊天，这是你们之前聊的内容：", "chat_target_private1") + Prompt("正在群里聊天", "chat_target_group2") + Prompt("和{sender_name}聊天", "chat_target_private2") + + Prompt( + """ +{expression_habits_block} +{relation_info_block} + +{chat_target} +{time_block} +{chat_info} +{identity} + +你现在的心情是：{mood_state} +你正在{chat_target_2},{reply_target_block} +你想要对上述的发言进行回复，回复的具体内容（原句）是：{raw_reply} +原因是：{reason} +现在请你将这条具体内容改写成一条适合在群聊中发送的回复消息。 +你需要使用合适的语法和句法，参考聊天内容，组织一条日常且口语化的回复。请你修改你想表达的原句，符合你的表达风格和语言习惯 +{reply_style} +你可以完全重组回复，保留最基本的表达含义就好，但重组后保持语意通顺。 +{keywords_reaction_prompt} +{moderation_prompt} +不要输出多余内容(包括前后缀，冒号和引号，括号，表情包，emoji,at或 @等 )，只输出一条回复就好。 +现在，你说： +""", + "default_expressor_prompt", + ) + + + + + Prompt( +"""{knowledge_prompt}{relation_info_block}{tool_info_block}{extra_info_block} +{expression_habits_block} + +你正在qq群里聊天，下面是群里正在聊的内容: +{time_block} +{background_dialogue_prompt} +{core_dialogue_prompt} + +{reply_target_block}。 +{identity} +你正在群里聊天,现在请你读读之前的聊天记录，然后给出日常且口语化的回复，平淡一些， +尽量简短一些。{keywords_reaction_prompt}请注意把握聊天内容，不要回复的太有条理，可以有个性。 +{reply_style} +请注意不要输出多余内容(包括前后缀，冒号和引号，括号，表情等)，只输出回复内容。 +{moderation_prompt}不要输出多余内容(包括前后缀，冒号和引号，括号，表情包，at或 @等 )。""", + "replyer_prompt", + ) + + + + Prompt( + """{identity} +{time_block} +你现在正在一个QQ群里聊天，以下是正在进行的聊天内容： +{background_dialogue_prompt} + +{expression_habits_block}{tool_info_block} +{knowledge_prompt}{relation_info_block} +{extra_info_block} + +你现在想补充说明你刚刚自己的发言内容：{target}，原因是{reason} +请你根据聊天内容，组织一条新回复。注意，{target} 是刚刚你自己的发言，你要在这基础上进一步发言，请按照你自己的角度来继续进行回复。 +注意保持上下文的连贯性。 +你现在的心情是：{mood_state} +{reply_style} +{keywords_reaction_prompt} +请注意不要输出多余内容(包括前后缀，冒号和引号，at或 @等 )。只输出回复内容。 +{moderation_prompt} +不要输出多余内容(包括前后缀，冒号和引号，括号()，表情包，emoji,at或 @等 )。只输出一条回复就好 +现在，你说： +""", + "replyer_self_prompt", + ) \ No newline at end of file diff --git a/template/bot_config_template.toml b/template/bot_config_template.toml index bb362db4..2b3ff61a 100644 --- a/template/bot_config_template.toml +++ b/template/bot_config_template.toml @@ -19,10 +19,10 @@ alias_names = ["麦叠", "牢麦"] # 麦麦的别名 [personality] # 建议120字以内，描述人格特质 和 身份特征 -personality = "是一个女大学生，现在在读大二，会刷贴吧。有时候说话不过脑子,有时候会喜欢说一些奇怪的话。年龄为19岁,有黑色的短发。" +personality = "是一个女大学生，现在在读大二，会刷贴吧。" #アイデンティティがない 生まれないらららら # 描述麦麦说话的表达风格，表达习惯，如要修改，可以酌情新增内容 -reply_style = "回复可以简短一些。可以参考贴吧，知乎和微博的回复风格，回复不要浮夸，不要用夸张修辞，平淡一些。不要浮夸，不要夸张修辞。" +reply_style = "请回复的平淡一些，简短一些，说中文，不要刻意突出自身学科背景。可以参考贴吧，知乎和微博的回复风格。" # 情感特征，影响情绪的变化情况 emotion_style = "情绪较为稳定，但遭遇特定事件的时候起伏较大" # 麦麦的兴趣，会影响麦麦对什么话题进行回复 From 328bc796dad56cbe2bb082f02c8da688a5316675 Mon Sep 17 00:00:00 2001 From: SengokuCola <1026294844@qq.com> Date: Sun, 14 Sep 2025 17:14:34 +0800 Subject: [PATCH 3/6] =?UTF-8?q?better=EF=BC=9A=E5=8D=87=E7=BA=A7=E8=87=AA?= =?UTF-8?q?=E6=88=91=E5=9B=9E=E5=A4=8D=E7=9A=84prompt?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- src/chat/replyer/default_generator.py | 2 + src/chat/replyer/replyer_prompt.py | 58 ++++++----------------- src/chat/replyer/rewrite_prompt.py | 66 +++++++++++++++++++++++++++ 3 files changed, 82 insertions(+), 44 deletions(-) create mode 100644 src/chat/replyer/rewrite_prompt.py diff --git a/src/chat/replyer/default_generator.py b/src/chat/replyer/default_generator.py index 5263517c..64383d4e 100644 --- a/src/chat/replyer/default_generator.py +++ b/src/chat/replyer/default_generator.py @@ -34,9 +34,11 @@ from src.plugin_system.apis import llm_api from src.chat.replyer.lpmm_prompt import init_lpmm_prompt from src.chat.replyer.replyer_prompt import init_replyer_prompt +from src.chat.replyer.rewrite_prompt import init_rewrite_prompt init_lpmm_prompt() init_replyer_prompt() +init_rewrite_prompt() logger = get_logger("replyer") diff --git a/src/chat/replyer/replyer_prompt.py b/src/chat/replyer/replyer_prompt.py index c726f247..3abfcd6a 100644 --- a/src/chat/replyer/replyer_prompt.py +++ b/src/chat/replyer/replyer_prompt.py @@ -38,36 +38,8 @@ def init_replyer_prompt(): Prompt("你正在和{sender_name}聊天，这是你们之前聊的内容：", "chat_target_private1") Prompt("正在群里聊天", "chat_target_group2") Prompt("和{sender_name}聊天", "chat_target_private2") - - Prompt( - """ -{expression_habits_block} -{relation_info_block} - -{chat_target} -{time_block} -{chat_info} -{identity} - -你现在的心情是：{mood_state} -你正在{chat_target_2},{reply_target_block} -你想要对上述的发言进行回复，回复的具体内容（原句）是：{raw_reply} -原因是：{reason} -现在请你将这条具体内容改写成一条适合在群聊中发送的回复消息。 -你需要使用合适的语法和句法，参考聊天内容，组织一条日常且口语化的回复。请你修改你想表达的原句，符合你的表达风格和语言习惯 -{reply_style} -你可以完全重组回复，保留最基本的表达含义就好，但重组后保持语意通顺。 -{keywords_reaction_prompt} -{moderation_prompt} -不要输出多余内容(包括前后缀，冒号和引号，括号，表情包，emoji,at或 @等 )，只输出一条回复就好。 -现在，你说： -""", - "default_expressor_prompt", - ) - - - - + + Prompt( """{knowledge_prompt}{relation_info_block}{tool_info_block}{extra_info_block} {expression_habits_block} @@ -90,25 +62,23 @@ def init_replyer_prompt(): Prompt( - """{identity} + """{knowledge_prompt}{relation_info_block}{tool_info_block}{extra_info_block} +{expression_habits_block} + + + + +你正在qq群里聊天，下面是群里正在聊的内容: {time_block} -你现在正在一个QQ群里聊天，以下是正在进行的聊天内容： {background_dialogue_prompt} -{expression_habits_block}{tool_info_block} -{knowledge_prompt}{relation_info_block} -{extra_info_block} - 你现在想补充说明你刚刚自己的发言内容：{target}，原因是{reason} -请你根据聊天内容，组织一条新回复。注意，{target} 是刚刚你自己的发言，你要在这基础上进一步发言，请按照你自己的角度来继续进行回复。 -注意保持上下文的连贯性。 -你现在的心情是：{mood_state} +请你根据聊天内容，组织一条新回复。注意，{target} 是刚刚你自己的发言，你要在这基础上进一步发言，请按照你自己的角度来继续进行回复。注意保持上下文的连贯性。 +{identity} +尽量简短一些。{keywords_reaction_prompt}请注意把握聊天内容，不要回复的太有条理，可以有个性。 {reply_style} -{keywords_reaction_prompt} -请注意不要输出多余内容(包括前后缀，冒号和引号，at或 @等 )。只输出回复内容。 -{moderation_prompt} -不要输出多余内容(包括前后缀，冒号和引号，括号()，表情包，emoji,at或 @等 )。只输出一条回复就好 -现在，你说： +请注意不要输出多余内容(包括前后缀，冒号和引号，括号，表情等)，只输出回复内容。 +{moderation_prompt}不要输出多余内容(包括前后缀，冒号和引号，括号，表情包，at或 @等 )。 """, "replyer_self_prompt", ) \ No newline at end of file diff --git a/src/chat/replyer/rewrite_prompt.py b/src/chat/replyer/rewrite_prompt.py new file mode 100644 index 00000000..bff94585 --- /dev/null +++ b/src/chat/replyer/rewrite_prompt.py @@ -0,0 +1,66 @@ +import traceback +import time +import asyncio +import random +import re + +from typing import List, Optional, Dict, Any, Tuple +from datetime import datetime +from src.mais4u.mai_think import mai_thinking_manager +from src.common.logger import get_logger +from src.common.data_models.database_data_model import DatabaseMessages +from src.common.data_models.info_data_model import ActionPlannerInfo +from src.common.data_models.llm_data_model import LLMGenerationDataModel +from src.config.config import global_config, model_config +from src.llm_models.utils_model import LLMRequest +from src.chat.message_receive.message import UserInfo, Seg, MessageRecv, MessageSending +from src.chat.message_receive.chat_stream import ChatStream +from src.chat.message_receive.uni_message_sender import UniversalMessageSender +from src.chat.utils.timer_calculator import Timer # <--- Import Timer +from src.chat.utils.utils import get_chat_type_and_target_info +from src.chat.utils.prompt_builder import Prompt, global_prompt_manager +from src.chat.utils.chat_message_builder import ( + build_readable_messages, + get_raw_msg_before_timestamp_with_chat, + replace_user_references, +) +from src.chat.express.expression_selector import expression_selector +# from src.chat.memory_system.memory_activator import MemoryActivator +from src.mood.mood_manager import mood_manager +from src.person_info.person_info import Person, is_person_known +from src.plugin_system.base.component_types import ActionInfo, EventType +from src.plugin_system.apis import llm_api + + + +def init_rewrite_prompt(): + Prompt("你正在qq群里聊天，下面是群里正在聊的内容:", "chat_target_group1") + Prompt("你正在和{sender_name}聊天，这是你们之前聊的内容：", "chat_target_private1") + Prompt("正在群里聊天", "chat_target_group2") + Prompt("和{sender_name}聊天", "chat_target_private2") + + Prompt( + """ +{expression_habits_block} +{relation_info_block} + +{chat_target} +{time_block} +{chat_info} +{identity} + +你现在的心情是：{mood_state} +你正在{chat_target_2},{reply_target_block} +你想要对上述的发言进行回复，回复的具体内容（原句）是：{raw_reply} +原因是：{reason} +现在请你将这条具体内容改写成一条适合在群聊中发送的回复消息。 +你需要使用合适的语法和句法，参考聊天内容，组织一条日常且口语化的回复。请你修改你想表达的原句，符合你的表达风格和语言习惯 +{reply_style} +你可以完全重组回复，保留最基本的表达含义就好，但重组后保持语意通顺。 +{keywords_reaction_prompt} +{moderation_prompt} +不要输出多余内容(包括前后缀，冒号和引号，括号，表情包，emoji,at或 @等 )，只输出一条回复就好。 +现在，你说： +""", + "default_expressor_prompt", + ) \ No newline at end of file From d693a2dabb59a1b48fc743c593369627bf1b64e1 Mon Sep 17 00:00:00 2001 From: SengokuCola <1026294844@qq.com> Date: Sun, 14 Sep 2025 17:17:44 +0800 Subject: [PATCH 4/6] Update expression_learner.py --- src/chat/express/expression_learner.py | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) diff --git a/src/chat/express/expression_learner.py b/src/chat/express/expression_learner.py index d245ee25..e36d4d57 100644 --- a/src/chat/express/expression_learner.py +++ b/src/chat/express/expression_learner.py @@ -16,7 +16,7 @@ from src.chat.message_receive.chat_stream import get_chat_manager MAX_EXPRESSION_COUNT = 300 -DECAY_DAYS = 30 # 30天衰减到0.01 +DECAY_DAYS = 15 # 30天衰减到0.01 DECAY_MIN = 0.01 # 最小衰减值 logger = get_logger("expressor") @@ -45,10 +45,10 @@ def init_prompt() -> None: 例如：当"AAAAA"时，可以"BBBBB", AAAAA代表某个具体的场景，不超过20个字。BBBBB代表对应的语言风格，特定句式或表达方式，不超过20个字。 例如： -当"对某件事表示十分惊叹，有些意外"时，使用"我嘞个xxxx" -当"表示讽刺的赞同，不想讲道理"时，使用"对对对" -当"想说明某个具体的事实观点，但懒得明说，或者不便明说，或表达一种默契"，使用"懂的都懂" -当"当涉及游戏相关时，表示意外的夸赞，略带戏谑意味"时，使用"这么强！" +当"对某件事表示十分惊叹"时，使用"我嘞个xxxx" +当"表示讽刺的赞同，不讲道理"时，使用"对对对" +当"想说明某个具体的事实观点，但懒得明说，使用"懂的都懂" +当"当涉及游戏相关时，夸赞，略带戏谑意味"时，使用"这么强！" 请注意：不要总结你自己（SELF）的发言，尽量保证总结内容的逻辑性 现在请你概括 From 3c7e868d6dd658958c6ce7f1e3774af2851cad90 Mon Sep 17 00:00:00 2001 From: SengokuCola <1026294844@qq.com> Date: Sun, 14 Sep 2025 17:25:40 +0800 Subject: [PATCH 5/6] fix ruff --- scripts/test_interest_embedding.py | 613 ------------------ .../frequency_control/frequency_control.py | 1 - .../talk_frequency_control.py | 272 -------- src/chat/heart_flow/heartFC_chat.py | 5 +- .../heart_flow/heartflow_message_processor.py | 1 - src/chat/replyer/default_generator.py | 2 +- src/chat/replyer/lpmm_prompt.py | 31 +- src/chat/replyer/replyer_prompt.py | 31 +- src/chat/replyer/rewrite_prompt.py | 31 +- src/plugins/built_in/memory/plugin.py | 2 +- template/bot_config_template.toml | 2 +- 11 files changed, 8 insertions(+), 983 deletions(-) delete mode 100644 scripts/test_interest_embedding.py delete mode 100644 src/chat/frequency_control/talk_frequency_control.py diff --git a/scripts/test_interest_embedding.py b/scripts/test_interest_embedding.py deleted file mode 100644 index f5fcd195..00000000 --- a/scripts/test_interest_embedding.py +++ /dev/null @@ -1,613 +0,0 @@ -#!/usr/bin/env python3 -""" -基于Embedding的兴趣度计算测试脚本 -使用MaiBot-Core的EmbeddingStore计算兴趣描述与目标文本的关联度 -""" - -import sys -import os -sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__)))) - -from typing import List, Dict, Tuple, Optional -import time -import json -import asyncio -from src.chat.knowledge.embedding_store import EmbeddingStore, cosine_similarity -from src.chat.knowledge.embedding_store import EMBEDDING_DATA_DIR_STR -from src.llm_models.utils_model import LLMRequest -from src.config.config import model_config - - -class InterestScorer: - """基于Embedding的兴趣度计算器""" - - def __init__(self, namespace: str = "interest_test"): - """初始化兴趣度计算器""" - self.embedding_store = EmbeddingStore(namespace, EMBEDDING_DATA_DIR_STR) - - async def get_embedding(self, text: str) -> Tuple[Optional[List[float]], float]: - """获取文本的嵌入向量""" - start_time = time.time() - try: - # 直接使用异步方式获取嵌入 - from src.llm_models.utils_model import LLMRequest - from src.config.config import model_config - - llm = LLMRequest(model_set=model_config.model_task_config.embedding, request_type="embedding") - embedding, _ = await llm.get_embedding(text) - - end_time = time.time() - elapsed = end_time - start_time - - if embedding and len(embedding) > 0: - return embedding, elapsed - return None, elapsed - except Exception as e: - print(f"获取嵌入向量失败: {e}") - return None, 0.0 - - async def calculate_similarity(self, text1: str, text2: str) -> Tuple[float, float, float]: - """计算两段文本的余弦相似度，返回(相似度, 文本1耗时, 文本2耗时)""" - emb1, time1 = await self.get_embedding(text1) - emb2, time2 = await self.get_embedding(text2) - - if emb1 is None or emb2 is None: - return 0.0, time1, time2 - - return cosine_similarity(emb1, emb2), time1, time2 - - async def calculate_interest_score(self, interest_text: str, target_text: str) -> Dict: - """ - 计算兴趣度分数 - - Args: - interest_text: 兴趣描述文本 - target_text: 目标文本 - - Returns: - 包含各种分数的字典 - """ - # 只计算语义相似度（嵌入分数） - semantic_score, interest_time, target_time = await self.calculate_similarity(interest_text, target_text) - - # 直接使用语义相似度作为最终分数 - final_score = semantic_score - - return { - "final_score": final_score, - "semantic_score": semantic_score, - "timing": { - "interest_embedding_time": interest_time, - "target_embedding_time": target_time, - "total_time": interest_time + target_time - } - } - - async def batch_calculate(self, interest_text: str, target_texts: List[str]) -> List[Dict]: - """批量计算兴趣度""" - results = [] - total_start_time = time.time() - - print(f"开始批量计算兴趣度...") - print(f"兴趣文本: {interest_text}") - print(f"目标文本数量: {len(target_texts)}") - - # 获取兴趣文本的嵌入向量（只需要一次） - interest_embedding, interest_time = await self.get_embedding(interest_text) - if interest_embedding is None: - print("无法获取兴趣文本的嵌入向量") - return [] - - print(f"兴趣文本嵌入计算耗时: {interest_time:.3f}秒") - - total_target_time = 0.0 - - for i, target_text in enumerate(target_texts): - print(f"处理第 {i+1}/{len(target_texts)} 个文本...") - - # 获取目标文本的嵌入向量 - target_embedding, target_time = await self.get_embedding(target_text) - total_target_time += target_time - - if target_embedding is None: - semantic_score = 0.0 - else: - semantic_score = cosine_similarity(interest_embedding, target_embedding) - - # 直接使用语义相似度作为最终分数 - final_score = semantic_score - - results.append({ - "target_text": target_text, - "final_score": final_score, - "semantic_score": semantic_score, - "timing": { - "target_embedding_time": target_time, - "item_total_time": target_time - } - }) - - # 按分数排序 - results.sort(key=lambda x: x["final_score"], reverse=True) - - total_time = time.time() - total_start_time - avg_target_time = total_target_time / len(target_texts) if target_texts else 0 - - print(f"\n=== 性能统计 ===") - print(f"兴趣文本嵌入计算耗时: {interest_time:.3f}秒") - print(f"目标文本嵌入计算总耗时: {total_target_time:.3f}秒") - print(f"目标文本嵌入计算平均耗时: {avg_target_time:.3f}秒") - print(f"总耗时: {total_time:.3f}秒") - print(f"平均每个目标文本处理耗时: {total_time / len(target_texts):.3f}秒") - - return results - - async def generate_paraphrases(self, original_text: str, num_sentences: int = 5) -> List[str]: - """ - 使用LLM生成近义句子 - - Args: - original_text: 原始文本 - num_sentences: 生成句子数量 - - Returns: - 近义句子列表 - """ - try: - # 创建LLM请求实例 - llm_request = LLMRequest( - model_set=model_config.model_task_config.replyer, - request_type="paraphrase_generator" - ) - - # 构建生成近义句子的提示词 - prompt = f"""请为以下兴趣描述生成{num_sentences}个意义相近但表达不同的句子： - -原始兴趣描述：{original_text} - -要求： -1. 保持原意不变，但尽量自由发挥，使用不同的表达方式，内容也可以有差异 -2. 句子结构要有所变化 -3. 可以适当调整语气和重点 -4. 每个句子都要完整且自然 -5. 只返回句子，不要编号，每行一个句子 - -生成的近义句子：""" - - print(f"正在生成近义句子...") - content, (reasoning, model_name, tool_calls) = await llm_request.generate_response_async(prompt) - - # 解析生成的句子 - sentences = [] - for line in content.strip().split('\n'): - line = line.strip() - if line and not line.startswith('生成') and not line.startswith('近义'): - sentences.append(line) - - # 确保返回指定数量的句子 - sentences = sentences[:num_sentences] - print(f"成功生成 {len(sentences)} 个近义句子") - print(f"使用的模型: {model_name}") - - return sentences - - except Exception as e: - print(f"生成近义句子失败: {e}") - return [] - - async def evaluate_all_paraphrases(self, original_text: str, target_texts: List[str], num_sentences: int = 5) -> Dict: - """ - 评估原始文本和所有近义句子的兴趣度 - - Args: - original_text: 原始兴趣描述文本 - target_texts: 目标文本列表 - num_sentences: 生成近义句子数量 - - Returns: - 包含所有评估结果的字典 - """ - print(f"\n=== 开始近义句子兴趣度评估 ===") - print(f"原始兴趣描述: {original_text}") - print(f"目标文本数量: {len(target_texts)}") - print(f"生成近义句子数量: {num_sentences}") - - # 生成近义句子 - paraphrases = await self.generate_paraphrases(original_text, num_sentences) - if not paraphrases: - print("生成近义句子失败，使用原始文本进行评估") - paraphrases = [] - - # 所有待评估的文本（原始文本 + 近义句子） - all_texts = [original_text] + paraphrases - - # 对每个文本进行兴趣度评估 - evaluation_results = {} - - for i, text in enumerate(all_texts): - text_type = "原始文本" if i == 0 else f"近义句子{i}" - print(f"\n--- 评估 {text_type} ---") - print(f"文本内容: {text}") - - # 计算兴趣度 - results = await self.batch_calculate(text, target_texts) - evaluation_results[text_type] = { - "text": text, - "results": results, - "top_score": results[0]["final_score"] if results else 0.0, - "average_score": sum(r["final_score"] for r in results) / len(results) if results else 0.0 - } - - return { - "original_text": original_text, - "paraphrases": paraphrases, - "evaluations": evaluation_results, - "summary": self._generate_summary(evaluation_results, target_texts) - } - - def _generate_summary(self, evaluation_results: Dict, target_texts: List[str]) -> Dict: - """生成评估摘要 - 关注目标句子的表现""" - summary = { - "best_performer": None, - "worst_performer": None, - "average_scores": {}, - "max_scores": {}, - "rankings": [], - "target_stats": {}, - "target_rankings": [] - } - - scores = [] - - for text_type, data in evaluation_results.items(): - scores.append({ - "text_type": text_type, - "text": data["text"], - "top_score": data["top_score"], - "average_score": data["average_score"] - }) - - # 按top_score排序 - scores.sort(key=lambda x: x["top_score"], reverse=True) - - summary["rankings"] = scores - summary["best_performer"] = scores[0] if scores else None - summary["worst_performer"] = scores[-1] if scores else None - - # 计算原始文本统计 - original_score = next((s for s in scores if s["text_type"] == "原始文本"), None) - if original_score: - summary["average_scores"]["original"] = original_score["average_score"] - summary["max_scores"]["original"] = original_score["top_score"] - - # 计算目标句子的统计信息 - target_stats = {} - for i, target_text in enumerate(target_texts): - target_key = f"目标{i+1}" - scores_for_target = [] - - # 收集所有兴趣描述对该目标文本的分数 - for text_type, data in evaluation_results.items(): - for result in data["results"]: - if result["target_text"] == target_text: - scores_for_target.append(result["final_score"]) - - if scores_for_target: - target_stats[target_key] = { - "target_text": target_text, - "scores": scores_for_target, - "average": sum(scores_for_target) / len(scores_for_target), - "max": max(scores_for_target), - "min": min(scores_for_target), - "std": (sum((x - sum(scores_for_target) / len(scores_for_target)) ** 2 for x in scores_for_target) / len(scores_for_target)) ** 0.5 - } - - summary["target_stats"] = target_stats - - # 按平均分对目标文本排序 - target_rankings = [] - for target_key, stats in target_stats.items(): - target_rankings.append({ - "target_key": target_key, - "target_text": stats["target_text"], - "average_score": stats["average"], - "max_score": stats["max"], - "min_score": stats["min"], - "std_score": stats["std"] - }) - - target_rankings.sort(key=lambda x: x["average_score"], reverse=True) - summary["target_rankings"] = target_rankings - - # 计算目标文本的整体统计 - if target_rankings: - all_target_averages = [t["average_score"] for t in target_rankings] - all_target_scores = [] - for stats in target_stats.values(): - all_target_scores.extend(stats["scores"]) - - summary["target_overall"] = { - "avg_of_averages": sum(all_target_averages) / len(all_target_averages), - "overall_max": max(all_target_scores), - "overall_min": min(all_target_scores), - "best_target": target_rankings[0]["target_text"], - "worst_target": target_rankings[-1]["target_text"] - } - - return summary - - -async def run_single_test(): - """运行单个测试""" - print("单个兴趣度测试") - print("=" * 40) - - # 输入兴趣文本 - # interest_text = input("请输入兴趣描述文本: ").strip() - # if not interest_text: - # print("兴趣描述不能为空") - # return - - interest_text ="对技术相关话题，游戏和动漫相关话题感兴趣，也对日常话题感兴趣，不喜欢太过沉重严肃的话题" - - # 输入目标文本 - print("请输入目标文本 (输入空行结束):") - import random - target_texts = [ - "AveMujica非常好看，你看了吗", - "明日方舟这个游戏挺好玩的", - "你能不能说点正经的", - "明日方舟挺好玩的", - "你的名字非常好看，你看了吗", - "《你的名字》非常好看，你看了吗", - "我们来聊聊苏联政治吧", - "轻音少女非常好看，你看了吗", - "我还挺喜欢打游戏的", - "我嘞个原神玩家啊", - "我心买了PlayStation5", - "直接Steam", - "有没有R" - ] - random.shuffle(target_texts) - # while True: - # line = input().strip() - # if not line: - # break - # target_texts.append(line) - - # if not target_texts: - # print("目标文本不能为空") - # return - - # 计算兴趣度 - scorer = InterestScorer() - results = await scorer.batch_calculate(interest_text, target_texts) - - # 显示结果 - print(f"\n兴趣度排序结果:") - print("-" * 80) - print(f"{'排名':<4} {'最终分数':<10} {'语义分数':<10} {'耗时(秒)':<10} {'目标文本'}") - print("-" * 80) - - for j, result in enumerate(results): - target_text = result['target_text'] - if len(target_text) > 40: - target_text = target_text[:37] + "..." - - timing = result.get('timing', {}) - item_time = timing.get('item_total_time', 0.0) - - print(f"{j+1:<4} {result['final_score']:<10.3f} {result['semantic_score']:<10.3f} " - f"{item_time:<10.3f} {target_text}") - - -async def run_paraphrase_test(): - """运行近义句子测试""" - print("近义句子兴趣度对比测试") - print("=" * 40) - - # 输入兴趣文本 - interest_text = "对技术相关话题，游戏和动漫相关话题感兴趣，比如明日方舟和原神，也对日常话题感兴趣，不喜欢太过沉重严肃的话题" - - # 输入目标文本 - print("请输入目标文本 (输入空行结束):") - # target_texts = [] - # while True: - # line = input().strip() - # if not line: - # break - # target_texts.append(line) - target_texts = [ - "AveMujica非常好看，你看了吗", - "明日方舟这个游戏挺好玩的", - "你能不能说点正经的", - "明日方舟挺好玩的", - "你的名字非常好看，你看了吗", - "《你的名字》非常好看，你看了吗", - "我们来聊聊苏联政治吧", - "轻音少女非常好看，你看了吗", - "我还挺喜欢打游戏的", - "刚加好友就视奸空间14条", - "可乐老大加我好友，我先日一遍空间", - "鸟一茬茬的", - "可乐可以是m，群友可以是s" - ] - - if not target_texts: - print("目标文本不能为空") - return - - # 创建评估器 - scorer = InterestScorer() - - # 运行评估 - result = await scorer.evaluate_all_paraphrases(interest_text, target_texts, num_sentences=5) - - # 显示结果 - display_paraphrase_results(result, target_texts) - - -def display_paraphrase_results(result: Dict, target_texts: List[str]): - """显示近义句子评估结果""" - print("\n" + "=" * 80) - print("近义句子兴趣度评估结果") - print("=" * 80) - - # 显示目标文本 - print(f"\n📋 目标文本列表:") - print("-" * 40) - for i, target in enumerate(target_texts): - print(f"{i+1}. {target}") - - # 显示生成的近义句子 - print(f"\n📝 生成的近义句子 (作为兴趣描述):") - print("-" * 40) - for i, paraphrase in enumerate(result["paraphrases"]): - print(f"{i+1}. {paraphrase}") - - # 显示摘要 - summary = result["summary"] - print(f"\n📊 评估摘要:") - print("-" * 40) - - if summary["best_performer"]: - print(f"最佳表现: {summary['best_performer']['text_type']} (最高分: {summary['best_performer']['top_score']:.3f})") - - if summary["worst_performer"]: - print(f"最差表现: {summary['worst_performer']['text_type']} (最高分: {summary['worst_performer']['top_score']:.3f})") - - print(f"原始文本平均分: {summary['average_scores'].get('original', 0):.3f}") - - # 显示目标文本的整体统计 - if "target_overall" in summary: - overall = summary["target_overall"] - print(f"\n📈 目标文本整体统计:") - print("-" * 40) - print(f"目标文本数量: {len(summary['target_rankings'])}") - print(f"平均分的平均值: {overall['avg_of_averages']:.3f}") - print(f"所有匹配中的最高分: {overall['overall_max']:.3f}") - print(f"所有匹配中的最低分: {overall['overall_min']:.3f}") - print(f"最佳匹配目标: {overall['best_target'][:50]}...") - print(f"最差匹配目标: {overall['worst_target'][:50]}...") - - # 显示目标文本排名 - if "target_rankings" in summary and summary["target_rankings"]: - print(f"\n🏆 目标文本排名 (按平均分):") - print("-" * 80) - print(f"{'排名':<4} {'平均分':<8} {'最高分':<8} {'最低分':<8} {'标准差':<8} {'目标文本'}") - print("-" * 80) - - for i, target in enumerate(summary["target_rankings"]): - target_text = target["target_text"][:40] + "..." if len(target["target_text"]) > 40 else target["target_text"] - print(f"{i+1:<4} {target['average_score']:<8.3f} {target['max_score']:<8.3f} {target['min_score']:<8.3f} {target['std_score']:<8.3f} {target_text}") - - # 显示每个目标文本的详细分数分布 - if "target_stats" in summary: - print(f"\n📊 目标文本详细分数分布:") - print("-" * 80) - - for target_key, stats in summary["target_stats"].items(): - print(f"\n{target_key}: {stats['target_text']}") - print(f" 平均分: {stats['average']:.3f}") - print(f" 最高分: {stats['max']:.3f}") - print(f" 最低分: {stats['min']:.3f}") - print(f" 标准差: {stats['std']:.3f}") - print(f" 所有分数: {[f'{s:.3f}' for s in stats['scores']]}") - - # 显示最佳和最差兴趣描述的目标表现对比 - if summary["best_performer"] and summary["worst_performer"]: - print(f"\n🔍 最佳 vs 最差兴趣描述对比:") - print("-" * 80) - - best_data = result["evaluations"][summary["best_performer"]["text_type"]] - worst_data = result["evaluations"][summary["worst_performer"]["text_type"]] - - print(f"最佳兴趣描述: {summary['best_performer']['text']}") - print(f"最差兴趣描述: {summary['worst_performer']['text']}") - print(f"") - print(f"{'目标文本':<30} {'最佳分数':<10} {'最差分数':<10} {'差值'}") - print("-" * 60) - - for best_result, worst_result in zip(best_data["results"], worst_data["results"]): - if best_result["target_text"] == worst_result["target_text"]: - diff = best_result["final_score"] - worst_result["final_score"] - target_text = best_result["target_text"][:27] + "..." if len(best_result["target_text"]) > 30 else best_result["target_text"] - print(f"{target_text:<30} {best_result['final_score']:<10.3f} {worst_result['final_score']:<10.3f} {diff:+.3f}") - - # 显示排名 - print(f"\n🏆 兴趣描述性能排名:") - print("-" * 80) - print(f"{'排名':<4} {'文本类型':<10} {'最高分':<8} {'平均分':<8} {'兴趣描述内容'}") - print("-" * 80) - - for i, item in enumerate(summary["rankings"]): - text_content = item["text"][:40] + "..." if len(item["text"]) > 40 else item["text"] - print(f"{i+1:<4} {item['text_type']:<10} {item['top_score']:<8.3f} {item['average_score']:<8.3f} {text_content}") - - # 显示每个兴趣描述的详细结果 - print(f"\n🔍 详细结果:") - print("-" * 80) - - for text_type, data in result["evaluations"].items(): - print(f"\n--- {text_type} ---") - print(f"兴趣描述: {data['text']}") - print(f"最高分: {data['top_score']:.3f}") - print(f"平均分: {data['average_score']:.3f}") - - # 显示前3个匹配结果 - top_results = data["results"][:3] - print(f"前3个匹配的目标文本:") - for j, result_item in enumerate(top_results): - print(f" {j+1}. 分数: {result_item['final_score']:.3f} - {result_item['target_text']}") - - # 显示对比表格 - print(f"\n📈 兴趣描述对比表格:") - print("-" * 100) - header = f"{'兴趣描述':<20}" - for i, target in enumerate(target_texts): - target_name = f"目标{i+1}" - header += f" {target_name:<12}" - print(header) - print("-" * 100) - - # 原始文本行 - original_line = f"{'原始文本':<20}" - original_data = result["evaluations"]["原始文本"]["results"] - for i in range(len(target_texts)): - if i < len(original_data): - original_line += f" {original_data[i]['final_score']:<12.3f}" - else: - original_line += f" {'-':<12}" - print(original_line) - - # 近义句子行 - for i, paraphrase in enumerate(result["paraphrases"]): - text_type = f"近义句子{i+1}" - line = f"{text_type:<20}" - paraphrase_data = result["evaluations"][text_type]["results"] - for j in range(len(target_texts)): - if j < len(paraphrase_data): - line += f" {paraphrase_data[j]['final_score']:<12.3f}" - else: - line += f" {'-':<12}" - print(line) - - -def main(): - """主函数""" - print("基于Embedding的兴趣度计算测试工具") - print("1. 单个兴趣度测试") - print("2. 近义句子兴趣度对比测试") - - choice = input("\n请选择 (1/2): ").strip() - - if choice == "1": - asyncio.run(run_single_test()) - elif choice == "2": - asyncio.run(run_paraphrase_test()) - else: - print("无效选择") - - -if __name__ == "__main__": - main() \ No newline at end of file diff --git a/src/chat/frequency_control/frequency_control.py b/src/chat/frequency_control/frequency_control.py index 323368d5..361ad128 100644 --- a/src/chat/frequency_control/frequency_control.py +++ b/src/chat/frequency_control/frequency_control.py @@ -3,7 +3,6 @@ from typing import Optional, Dict from src.plugin_system.apis import message_api from src.chat.message_receive.chat_stream import ChatStream, get_chat_manager from src.common.logger import get_logger -from src.chat.frequency_control.talk_frequency_control import get_config_base_talk_frequency from src.chat.frequency_control.focus_value_control import get_config_base_focus_value logger = get_logger("frequency_control") diff --git a/src/chat/frequency_control/talk_frequency_control.py b/src/chat/frequency_control/talk_frequency_control.py deleted file mode 100644 index b0733bb3..00000000 --- a/src/chat/frequency_control/talk_frequency_control.py +++ /dev/null @@ -1,272 +0,0 @@ -from typing import Optional -from datetime import datetime, timedelta -import statistics -from src.config.config import global_config -from src.chat.frequency_control.utils import parse_stream_config_to_chat_id -from src.common.database.database_model import Messages - - -def get_config_base_talk_frequency(chat_id: Optional[str] = None) -> float: - """ - 根据当前时间和聊天流获取对应的 talk_frequency - - Args: - chat_stream_id: 聊天流ID，格式为 "platform:chat_id:type" - - Returns: - float: 对应的频率值 - """ - if not global_config.chat.talk_frequency_adjust: - return global_config.chat.talk_frequency - - # 优先检查聊天流特定的配置 - if chat_id: - stream_frequency = get_stream_specific_frequency(chat_id) - if stream_frequency is not None: - return stream_frequency - - # 检查全局时段配置（第一个元素为空字符串的配置） - global_frequency = get_global_frequency() - return global_config.chat.talk_frequency if global_frequency is None else global_frequency - - -def get_time_based_frequency(time_freq_list: list[str]) -> Optional[float]: - """ - 根据时间配置列表获取当前时段的频率 - - Args: - time_freq_list: 时间频率配置列表，格式为 ["HH:MM,frequency", ...] - - Returns: - float: 频率值，如果没有配置则返回 None - """ - from datetime import datetime - - current_time = datetime.now().strftime("%H:%M") - current_hour, current_minute = map(int, current_time.split(":")) - current_minutes = current_hour * 60 + current_minute - - # 解析时间频率配置 - time_freq_pairs = [] - for time_freq_str in time_freq_list: - try: - time_str, freq_str = time_freq_str.split(",") - hour, minute = map(int, time_str.split(":")) - frequency = float(freq_str) - minutes = hour * 60 + minute - time_freq_pairs.append((minutes, frequency)) - except (ValueError, IndexError): - continue - - if not time_freq_pairs: - return None - - # 按时间排序 - time_freq_pairs.sort(key=lambda x: x[0]) - - # 查找当前时间对应的频率 - current_frequency = None - for minutes, frequency in time_freq_pairs: - if current_minutes >= minutes: - current_frequency = frequency - else: - break - - # 如果当前时间在所有配置时间之前，使用最后一个时间段的频率（跨天逻辑） - if current_frequency is None and time_freq_pairs: - current_frequency = time_freq_pairs[-1][1] - - return current_frequency - - -def get_stream_specific_frequency(chat_stream_id: str): - """ - 获取特定聊天流在当前时间的频率 - - Args: - chat_stream_id: 聊天流ID（哈希值） - - Returns: - float: 频率值，如果没有配置则返回 None - """ - # 查找匹配的聊天流配置 - for config_item in global_config.chat.talk_frequency_adjust: - if not config_item or len(config_item) < 2: - continue - - stream_config_str = config_item[0] # 例如 "qq:1026294844:group" - - # 解析配置字符串并生成对应的 chat_id - config_chat_id = parse_stream_config_to_chat_id(stream_config_str) - if config_chat_id is None: - continue - - # 比较生成的 chat_id - if config_chat_id != chat_stream_id: - continue - - # 使用通用的时间频率解析方法 - return get_time_based_frequency(config_item[1:]) - - return None - - -def get_global_frequency() -> Optional[float]: - """ - 获取全局默认频率配置 - - Returns: - float: 频率值，如果没有配置则返回 None - """ - for config_item in global_config.chat.talk_frequency_adjust: - if not config_item or len(config_item) < 2: - continue - - # 检查是否为全局默认配置（第一个元素为空字符串） - if config_item[0] == "": - return get_time_based_frequency(config_item[1:]) - - return None - - -def get_weekly_hourly_message_stats(chat_id: str): - """ - 计算指定聊天最近一周每个小时的消息数量和用户数量 - - Args: - chat_id: 聊天ID（对应 Messages 表的 chat_id 字段） - - Returns: - dict: 包含24个小时统计数据，格式为: - { - "0": {"message_count": [5, 8, 3, 12, 6, 9, 7], "message_std_dev": 2.1}, - "1": {"message_count": [10, 15, 8, 20, 12, 18, 14], "message_std_dev": 3.2}, - ... - } - """ - # 计算一周前的时间戳 - one_week_ago = datetime.now() - timedelta(days=7) - one_week_ago_timestamp = one_week_ago.timestamp() - - # 初始化数据结构：按小时存储每天的消息计数 - hourly_data = {} - for hour in range(24): - hourly_data[f"hour_{hour}"] = {"daily_counts": []} - - try: - # 查询指定聊天最近一周的消息 - messages = Messages.select().where( - (Messages.time >= one_week_ago_timestamp) & - (Messages.chat_id == chat_id) - ) - - # 统计每个小时的数据 - for message in messages: - # 将时间戳转换为datetime - msg_time = datetime.fromtimestamp(message.time) - hour = msg_time.hour - - # 记录每天的消息计数（按日期分组） - day_key = msg_time.strftime("%Y-%m-%d") - hour_key = f"{hour}" - - # 为该小时添加当天的消息计数 - found = False - for day_count in hourly_data[hour_key]["daily_counts"]: - if day_count["date"] == day_key: - day_count["count"] += 1 - found = True - break - - if not found: - hourly_data[hour_key]["daily_counts"].append({"date": day_key, "count": 1}) - - - except Exception as e: - # 如果查询失败，返回空的统计结果 - print(f"Error getting weekly hourly message stats for chat {chat_id}: {e}") - hourly_stats = {} - for hour in range(24): - hourly_stats[f"hour_{hour}"] = { - "message_count": [], - "message_std_dev": 0.0 - } - return hourly_stats - - # 计算每个小时的统计结果 - hourly_stats = {} - for hour in range(24): - hour_key = f"hour_{hour}" - daily_counts = [day["count"] for day in hourly_data[hour_key]["daily_counts"]] - - # 计算总消息数 - total_messages = sum(daily_counts) - - # 计算标准差 - message_std_dev = 0.0 - if len(daily_counts) > 1: - message_std_dev = statistics.stdev(daily_counts) - elif len(daily_counts) == 1: - message_std_dev = 0.0 - - # 按日期排序每日消息计数 - daily_counts_sorted = sorted(hourly_data[hour_key]["daily_counts"], key=lambda x: x["date"]) - - hourly_stats[hour_key] = { - "message_count": [day["count"] for day in daily_counts_sorted], - "message_std_dev": message_std_dev - } - - return hourly_stats - -def get_recent_15min_stats(chat_id: str): - """ - 获取最近15分钟指定聊天的消息数量和发言人数 - - Args: - chat_id: 聊天ID（对应 Messages 表的 chat_id 字段） - - Returns: - dict: 包含消息数量和发言人数，格式为: - { - "message_count": 25, - "user_count": 8, - "time_range": "2025-01-01 14:30:00 - 2025-01-01 14:45:00" - } - """ - # 计算15分钟前的时间戳 - fifteen_min_ago = datetime.now() - timedelta(minutes=15) - fifteen_min_ago_timestamp = fifteen_min_ago.timestamp() - current_time = datetime.now() - - # 初始化统计结果 - message_count = 0 - user_set = set() - - try: - # 查询最近15分钟的消息 - messages = Messages.select().where( - (Messages.time >= fifteen_min_ago_timestamp) & - (Messages.chat_id == chat_id) - ) - - # 统计消息数量和用户 - for message in messages: - message_count += 1 - if message.user_id: - user_set.add(message.user_id) - - except Exception as e: - # 如果查询失败，返回空结果 - print(f"Error getting recent 15min stats for chat {chat_id}: {e}") - return { - "message_count": 0, - "user_count": 0, - "time_range": f"{fifteen_min_ago.strftime('%Y-%m-%d %H:%M:%S')} - {current_time.strftime('%Y-%m-%d %H:%M:%S')}" - } - - return { - "message_count": message_count, - "user_count": len(user_set), - "time_range": f"{fifteen_min_ago.strftime('%Y-%m-%d %H:%M:%S')} - {current_time.strftime('%Y-%m-%d %H:%M:%S')}" - } diff --git a/src/chat/heart_flow/heartFC_chat.py b/src/chat/heart_flow/heartFC_chat.py index 54057c8a..9aab694b 100644 --- a/src/chat/heart_flow/heartFC_chat.py +++ b/src/chat/heart_flow/heartFC_chat.py @@ -1,7 +1,6 @@ import asyncio import time import traceback -import math import random from typing import List, Optional, Dict, Any, Tuple, TYPE_CHECKING from rich.traceback import install @@ -20,7 +19,7 @@ from src.chat.heart_flow.hfc_utils import CycleDetail from src.chat.heart_flow.hfc_utils import send_typing, stop_typing from src.chat.express.expression_learner import expression_learner_manager from src.person_info.person_info import Person -from src.plugin_system.base.component_types import ChatMode, EventType, ActionInfo +from src.plugin_system.base.component_types import EventType, ActionInfo from src.plugin_system.core import events_manager from src.plugin_system.apis import generator_api, send_api, message_api, database_api from src.mais4u.mai_think import mai_thinking_manager @@ -376,7 +375,7 @@ class HeartFChatting: action_success = False action_reply_text = "" - for i, result in enumerate(results): + for result in results: if isinstance(result, BaseException): logger.error(f"{self.log_prefix} 动作执行异常: {result}") continue diff --git a/src/chat/heart_flow/heartflow_message_processor.py b/src/chat/heart_flow/heartflow_message_processor.py index 1d90e468..ce4d033f 100644 --- a/src/chat/heart_flow/heartflow_message_processor.py +++ b/src/chat/heart_flow/heartflow_message_processor.py @@ -1,6 +1,5 @@ import asyncio import re -import math import traceback from typing import Tuple, TYPE_CHECKING diff --git a/src/chat/replyer/default_generator.py b/src/chat/replyer/default_generator.py index 64383d4e..832767d6 100644 --- a/src/chat/replyer/default_generator.py +++ b/src/chat/replyer/default_generator.py @@ -18,7 +18,7 @@ from src.chat.message_receive.chat_stream import ChatStream from src.chat.message_receive.uni_message_sender import UniversalMessageSender from src.chat.utils.timer_calculator import Timer # <--- Import Timer from src.chat.utils.utils import get_chat_type_and_target_info -from src.chat.utils.prompt_builder import Prompt, global_prompt_manager +from src.chat.utils.prompt_builder import global_prompt_manager from src.chat.utils.chat_message_builder import ( build_readable_messages, get_raw_msg_before_timestamp_with_chat, diff --git a/src/chat/replyer/lpmm_prompt.py b/src/chat/replyer/lpmm_prompt.py index 67baf027..d5d02664 100644 --- a/src/chat/replyer/lpmm_prompt.py +++ b/src/chat/replyer/lpmm_prompt.py @@ -1,35 +1,6 @@ -import traceback -import time -import asyncio -import random -import re -from typing import List, Optional, Dict, Any, Tuple -from datetime import datetime -from src.mais4u.mai_think import mai_thinking_manager -from src.common.logger import get_logger -from src.common.data_models.database_data_model import DatabaseMessages -from src.common.data_models.info_data_model import ActionPlannerInfo -from src.common.data_models.llm_data_model import LLMGenerationDataModel -from src.config.config import global_config, model_config -from src.llm_models.utils_model import LLMRequest -from src.chat.message_receive.message import UserInfo, Seg, MessageRecv, MessageSending -from src.chat.message_receive.chat_stream import ChatStream -from src.chat.message_receive.uni_message_sender import UniversalMessageSender -from src.chat.utils.timer_calculator import Timer # <--- Import Timer -from src.chat.utils.utils import get_chat_type_and_target_info -from src.chat.utils.prompt_builder import Prompt, global_prompt_manager -from src.chat.utils.chat_message_builder import ( - build_readable_messages, - get_raw_msg_before_timestamp_with_chat, - replace_user_references, -) -from src.chat.express.expression_selector import expression_selector +from src.chat.utils.prompt_builder import Prompt # from src.chat.memory_system.memory_activator import MemoryActivator -from src.mood.mood_manager import mood_manager -from src.person_info.person_info import Person, is_person_known -from src.plugin_system.base.component_types import ActionInfo, EventType -from src.plugin_system.apis import llm_api diff --git a/src/chat/replyer/replyer_prompt.py b/src/chat/replyer/replyer_prompt.py index 3abfcd6a..d3de1935 100644 --- a/src/chat/replyer/replyer_prompt.py +++ b/src/chat/replyer/replyer_prompt.py @@ -1,35 +1,6 @@ -import traceback -import time -import asyncio -import random -import re -from typing import List, Optional, Dict, Any, Tuple -from datetime import datetime -from src.mais4u.mai_think import mai_thinking_manager -from src.common.logger import get_logger -from src.common.data_models.database_data_model import DatabaseMessages -from src.common.data_models.info_data_model import ActionPlannerInfo -from src.common.data_models.llm_data_model import LLMGenerationDataModel -from src.config.config import global_config, model_config -from src.llm_models.utils_model import LLMRequest -from src.chat.message_receive.message import UserInfo, Seg, MessageRecv, MessageSending -from src.chat.message_receive.chat_stream import ChatStream -from src.chat.message_receive.uni_message_sender import UniversalMessageSender -from src.chat.utils.timer_calculator import Timer # <--- Import Timer -from src.chat.utils.utils import get_chat_type_and_target_info -from src.chat.utils.prompt_builder import Prompt, global_prompt_manager -from src.chat.utils.chat_message_builder import ( - build_readable_messages, - get_raw_msg_before_timestamp_with_chat, - replace_user_references, -) -from src.chat.express.expression_selector import expression_selector +from src.chat.utils.prompt_builder import Prompt # from src.chat.memory_system.memory_activator import MemoryActivator -from src.mood.mood_manager import mood_manager -from src.person_info.person_info import Person, is_person_known -from src.plugin_system.base.component_types import ActionInfo, EventType -from src.plugin_system.apis import llm_api diff --git a/src/chat/replyer/rewrite_prompt.py b/src/chat/replyer/rewrite_prompt.py index bff94585..35ca767c 100644 --- a/src/chat/replyer/rewrite_prompt.py +++ b/src/chat/replyer/rewrite_prompt.py @@ -1,35 +1,6 @@ -import traceback -import time -import asyncio -import random -import re -from typing import List, Optional, Dict, Any, Tuple -from datetime import datetime -from src.mais4u.mai_think import mai_thinking_manager -from src.common.logger import get_logger -from src.common.data_models.database_data_model import DatabaseMessages -from src.common.data_models.info_data_model import ActionPlannerInfo -from src.common.data_models.llm_data_model import LLMGenerationDataModel -from src.config.config import global_config, model_config -from src.llm_models.utils_model import LLMRequest -from src.chat.message_receive.message import UserInfo, Seg, MessageRecv, MessageSending -from src.chat.message_receive.chat_stream import ChatStream -from src.chat.message_receive.uni_message_sender import UniversalMessageSender -from src.chat.utils.timer_calculator import Timer # <--- Import Timer -from src.chat.utils.utils import get_chat_type_and_target_info -from src.chat.utils.prompt_builder import Prompt, global_prompt_manager -from src.chat.utils.chat_message_builder import ( - build_readable_messages, - get_raw_msg_before_timestamp_with_chat, - replace_user_references, -) -from src.chat.express.expression_selector import expression_selector +from src.chat.utils.prompt_builder import Prompt # from src.chat.memory_system.memory_activator import MemoryActivator -from src.mood.mood_manager import mood_manager -from src.person_info.person_info import Person, is_person_known -from src.plugin_system.base.component_types import ActionInfo, EventType -from src.plugin_system.apis import llm_api diff --git a/src/plugins/built_in/memory/plugin.py b/src/plugins/built_in/memory/plugin.py index 2ab1ec52..25f95448 100644 --- a/src/plugins/built_in/memory/plugin.py +++ b/src/plugins/built_in/memory/plugin.py @@ -1,7 +1,7 @@ from typing import List, Tuple, Type # 导入新插件系统 -from src.plugin_system import BasePlugin, register_plugin, ComponentInfo +from src.plugin_system import BasePlugin, ComponentInfo from src.plugin_system.base.config_types import ConfigField # 导入依赖的系统组件 diff --git a/template/bot_config_template.toml b/template/bot_config_template.toml index 2b3ff61a..36471bdb 100644 --- a/template/bot_config_template.toml +++ b/template/bot_config_template.toml @@ -1,5 +1,5 @@ [inner] -version = "6.12.0" +version = "6.13.0" #----以下是给开发人员阅读的，如果你只是部署了麦麦，不需要阅读---- #如果你想要修改配置文件，请递增version的值 From a55ce70c0da27948764699502833b66a36f08a20 Mon Sep 17 00:00:00 2001 From: SengokuCola <1026294844@qq.com> Date: Sun, 14 Sep 2025 17:26:09 +0800 Subject: [PATCH 6/6] Update config.py --- src/config/config.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/src/config/config.py b/src/config/config.py index 82cebf52..920a155c 100644 --- a/src/config/config.py +++ b/src/config/config.py @@ -54,7 +54,7 @@ TEMPLATE_DIR = os.path.join(PROJECT_ROOT, "template") # 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 # 对该字段的更新，请严格参照语义化版本规范：https://semver.org/lang/zh-CN/ -MMC_VERSION = "0.10.3-snapshot.3" +MMC_VERSION = "0.10.3-snapshot.4" def get_key_comment(toml_table, key):