Losita/mai-bot
0
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat；优化了记忆检索的速度和token消耗（将question提出交给planenr)

Browse Source
This commit is contained in:
SengokuCola
2025-12-24 18:43:32 +08:00
parent 490589b0ad
commit 0852af49f9
15 changed files with 448 additions and 152 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/chat/heart_flow/heartFC_chat.py
View File

@@ -687,7 +687,7 @@ class HeartFChatting:
return {
"action_type": "reply",
"success": True,
"result": f"你回复内容{reply_text}",
"result": f"你使用reply动作，对' {action_planner_info.action_message.processed_plain_text} '这句话进行了回复，回复内容为: '{reply_text}'",
"loop_info": loop_info,
}

29
src/chat/planner_actions/planner.py
View File

@@ -53,7 +53,7 @@ reply
4.不要选择回复你自己发送的消息
5.不要单独对表情包进行回复
6.将上下文中所有含义不明的，疑似黑话的，缩写词均写入unknown_words中
7.用一句简单的话来描述当前回复场景，不超过10个字
7.如果你对上下文存在疑问，有需要查询的问题，写入question中
{reply_action_example}
no_reply
@@ -224,6 +224,25 @@ class ActionPlanner:
else:
reasoning = "未提供原因"
action_data = {key: value for key, value in action_json.items() if key not in ["action"]}
# 验证和清理 question
if "question" in action_data:
q = action_data.get("question")
if isinstance(q, str):
cleaned_q = q.strip()
if cleaned_q:
action_data["question"] = cleaned_q
else:
# 如果清理后为空字符串，移除该字段
action_data.pop("question", None)
elif q is None:
# 如果为 None，移除该字段
action_data.pop("question", None)
else:
# 如果不是字符串类型，记录警告并移除
logger.warning(f"{self.log_prefix}question 格式不正确，应为字符串类型，已忽略")
action_data.pop("question", None)
# 非no_reply动作需要target_message_id
target_message = None
@@ -503,18 +522,20 @@ class ActionPlanner:
name_block = f"你的名字是{bot_name}{bot_nickname}，请注意哪些是你自己的发言。"
# 根据 think_mode 配置决定 reply action 的示例 JSON
# 在 JSON 中直接作为 action 参数携带 unknown_words
# 在 JSON 中直接作为 action 参数携带 unknown_words 和 question
if global_config.chat.think_mode == "classic":
reply_action_example = (
'{{"action":"reply", "target_message_id":"消息id(m+数字)", '
'"unknown_words":["词语1","词语2"]}}'
'"unknown_words":["词语1","词语2"], '
'"question":"需要查询的问题"}'
)
else:
reply_action_example = (
"5.think_level表示思考深度，0表示该回复不需要思考和回忆，1表示该回复需要进行回忆和思考\n"
+ '{{"action":"reply", "think_level":数值等级(0或1), '
'"target_message_id":"消息id(m+数字)", '
'"unknown_words":["词语1","词语2"]}}'
'"unknown_words":["词语1","词语2"], '
'"question":"需要查询的问题"}'
)
planner_prompt_template = await global_prompt_manager.get_prompt_async("planner_prompt")

14
src/chat/replyer/group_generator.py
View File

@@ -947,6 +947,18 @@ class DefaultReplyer:
chat_id, message_list_before_short, chat_talking_prompt_short, unknown_words
)
# 从 chosen_actions 中提取 question（仅在 reply 动作中）
question = None
if chosen_actions:
for action_info in chosen_actions:
if action_info.action_type == "reply" and isinstance(action_info.action_data, dict):
q = action_info.action_data.get("question")
if isinstance(q, str):
cleaned_q = q.strip()
if cleaned_q:
question = cleaned_q
break
# 并行执行构建任务（包括黑话解释，可配置关闭）
task_results = await asyncio.gather(
self._time_and_run_task(
@@ -961,7 +973,7 @@ class DefaultReplyer:
self._time_and_run_task(self.build_personality_prompt(), "personality_prompt"),
self._time_and_run_task(
build_memory_retrieval_prompt(
chat_talking_prompt_short, sender, target, self.chat_stream, think_level=think_level
chat_talking_prompt_short, sender, target, self.chat_stream, think_level=think_level, unknown_words=unknown_words, question=question
),
"memory_retrieval",
),

18
src/chat/replyer/private_generator.py
View File

@@ -110,6 +110,7 @@ class PrivateReplyer:
enable_tool=enable_tool,
reply_message=reply_message,
reply_reason=reply_reason,
unknown_words=unknown_words,
)
llm_response.prompt = prompt
llm_response.selected_expressions = selected_expressions
@@ -611,6 +612,7 @@ class PrivateReplyer:
available_actions: Optional[Dict[str, ActionInfo]] = None,
chosen_actions: Optional[List[ActionPlannerInfo]] = None,
enable_tool: bool = True,
unknown_words: Optional[List[str]] = None,
) -> Tuple[str, List[int]]:
"""
构建回复器上下文
@@ -709,12 +711,24 @@ class PrivateReplyer:
else:
jargon_coroutine = self._build_disabled_jargon_explanation()
# 从 chosen_actions 中提取 question（仅在 reply 动作中）
question = None
if chosen_actions:
for action_info in chosen_actions:
if action_info.action_type == "reply" and isinstance(action_info.action_data, dict):
q = action_info.action_data.get("question")
if isinstance(q, str):
cleaned_q = q.strip()
if cleaned_q:
question = cleaned_q
break
# 并行执行九个构建任务（包括黑话解释，可配置关闭）
task_results = await asyncio.gather(
self._time_and_run_task(
self.build_expression_habits(chat_talking_prompt_short, target, reply_reason), "expression_habits"
),
self._time_and_run_task(self.build_relation_info(chat_talking_prompt_short, sender), "relation_info"),
# self._time_and_run_task(self.build_relation_info(chat_talking_prompt_short, sender), "relation_info"),
self._time_and_run_task(
self.build_tool_info(chat_talking_prompt_short, sender, target, enable_tool=enable_tool), "tool_info"
),
@@ -723,7 +737,7 @@ class PrivateReplyer:
self._time_and_run_task(self.build_personality_prompt(), "personality_prompt"),
self._time_and_run_task(
build_memory_retrieval_prompt(
chat_talking_prompt_short, sender, target, self.chat_stream, self.tool_executor
chat_talking_prompt_short, sender, target, self.chat_stream, think_level=1, unknown_words=unknown_words, question=question
),
"memory_retrieval",
),

33
src/chat/utils/statistic.py
View File

@@ -743,13 +743,13 @@ class StatisticOutputTask(AsyncTask):
"""
if stats[TOTAL_REQ_CNT] <= 0:
return ""
data_fmt = "{:<32} {:>10} {:>12} {:>12} {:>12} {:>9.2f}¥ {:>10.1f} {:>10.1f} {:>12} {:>12}"
data_fmt = "{:<32} {:>10} {:>12} {:>12} {:>12} {:>9.2f}¥ {:>10.1f} {:>10.1f} {:>12} {:>12} {:>12}"
total_replies = stats.get(TOTAL_REPLY_CNT, 0)
output = [
"按模型分类统计:",
" 模型名称 调用次数 输入Token 输出Token Token总量 累计花费 平均耗时(秒) 标准差(秒) 每次回复平均调用次数 每次回复平均Token数",
" 模型名称 调用次数 输入Token 输出Token Token总量 累计花费 平均耗时(秒) 标准差(秒) 每次回复平均调用次数 每次回复平均Token数 每次调用平均Token",
]
for model_name, count in sorted(stats[REQ_CNT_BY_MODEL].items()):
name = f"{model_name[:29]}..." if len(model_name) > 32 else model_name
@@ -764,6 +764,9 @@ class StatisticOutputTask(AsyncTask):
avg_count_per_reply = count / total_replies if total_replies > 0 else 0.0
avg_tokens_per_reply = tokens / total_replies if total_replies > 0 else 0.0
# 计算每次调用平均token
avg_tokens_per_call = tokens / count if count > 0 else 0.0
# 格式化大数字
formatted_count = _format_large_number(count)
formatted_in_tokens = _format_large_number(in_tokens)
@@ -771,6 +774,7 @@ class StatisticOutputTask(AsyncTask):
formatted_tokens = _format_large_number(tokens)
formatted_avg_count = _format_large_number(avg_count_per_reply) if total_replies > 0 else "N/A"
formatted_avg_tokens = _format_large_number(avg_tokens_per_reply) if total_replies > 0 else "N/A"
formatted_avg_tokens_per_call = _format_large_number(avg_tokens_per_call) if count > 0 else "N/A"
output.append(
data_fmt.format(
@@ -784,6 +788,7 @@ class StatisticOutputTask(AsyncTask):
std_time_cost,
formatted_avg_count,
formatted_avg_tokens,
formatted_avg_tokens_per_call,
)
)
@@ -797,13 +802,13 @@ class StatisticOutputTask(AsyncTask):
"""
if stats[TOTAL_REQ_CNT] <= 0:
return ""
data_fmt = "{:<32} {:>10} {:>12} {:>12} {:>12} {:>9.2f}¥ {:>10.1f} {:>10.1f} {:>12} {:>12}"
data_fmt = "{:<32} {:>10} {:>12} {:>12} {:>12} {:>9.2f}¥ {:>10.1f} {:>10.1f} {:>12} {:>12} {:>12}"
total_replies = stats.get(TOTAL_REPLY_CNT, 0)
output = [
"按模块分类统计:",
" 模块名称 调用次数 输入Token 输出Token Token总量 累计花费 平均耗时(秒) 标准差(秒) 每次回复平均调用次数 每次回复平均Token数",
" 模块名称 调用次数 输入Token 输出Token Token总量 累计花费 平均耗时(秒) 标准差(秒) 每次回复平均调用次数 每次回复平均Token数 每次调用平均Token",
]
for module_name, count in sorted(stats[REQ_CNT_BY_MODULE].items()):
name = f"{module_name[:29]}..." if len(module_name) > 32 else module_name
@@ -818,6 +823,9 @@ class StatisticOutputTask(AsyncTask):
avg_count_per_reply = count / total_replies if total_replies > 0 else 0.0
avg_tokens_per_reply = tokens / total_replies if total_replies > 0 else 0.0
# 计算每次调用平均token
avg_tokens_per_call = tokens / count if count > 0 else 0.0
# 格式化大数字
formatted_count = _format_large_number(count)
formatted_in_tokens = _format_large_number(in_tokens)
@@ -825,6 +833,7 @@ class StatisticOutputTask(AsyncTask):
formatted_tokens = _format_large_number(tokens)
formatted_avg_count = _format_large_number(avg_count_per_reply) if total_replies > 0 else "N/A"
formatted_avg_tokens = _format_large_number(avg_tokens_per_reply) if total_replies > 0 else "N/A"
formatted_avg_tokens_per_call = _format_large_number(avg_tokens_per_call) if count > 0 else "N/A"
output.append(
data_fmt.format(
@@ -838,6 +847,7 @@ class StatisticOutputTask(AsyncTask):
std_time_cost,
formatted_avg_count,
formatted_avg_tokens,
formatted_avg_tokens_per_call,
)
)
@@ -935,11 +945,12 @@ class StatisticOutputTask(AsyncTask):
f"<td>{stat_data[STD_TIME_COST_BY_MODEL][model_name]:.1f} 秒</td>"
f"<td>{_format_large_number(count / total_replies, html=True) if total_replies > 0 else 'N/A'}</td>"
f"<td>{_format_large_number(stat_data[TOTAL_TOK_BY_MODEL][model_name] / total_replies, html=True) if total_replies > 0 else 'N/A'}</td>"
f"<td>{_format_large_number(stat_data[TOTAL_TOK_BY_MODEL][model_name] / count, html=True) if count > 0 else 'N/A'}</td>"
f"</tr>"
for model_name, count in sorted(stat_data[REQ_CNT_BY_MODEL].items())
]
if stat_data[REQ_CNT_BY_MODEL]
else ["<tr><td colspan='10' style='text-align: center; color: #999;'>暂无数据</td></tr>"]
else ["<tr><td colspan='11' style='text-align: center; color: #999;'>暂无数据</td></tr>"]
)
# 按请求类型分类统计
type_rows = "\n".join(
@@ -955,11 +966,12 @@ class StatisticOutputTask(AsyncTask):
f"<td>{stat_data[STD_TIME_COST_BY_TYPE][req_type]:.1f} 秒</td>"
f"<td>{_format_large_number(count / total_replies, html=True) if total_replies > 0 else 'N/A'}</td>"
f"<td>{_format_large_number(stat_data[TOTAL_TOK_BY_TYPE][req_type] / total_replies, html=True) if total_replies > 0 else 'N/A'}</td>"
f"<td>{_format_large_number(stat_data[TOTAL_TOK_BY_TYPE][req_type] / count, html=True) if count > 0 else 'N/A'}</td>"
f"</tr>"
for req_type, count in sorted(stat_data[REQ_CNT_BY_TYPE].items())
]
if stat_data[REQ_CNT_BY_TYPE]
else ["<tr><td colspan='10' style='text-align: center; color: #999;'>暂无数据</td></tr>"]
else ["<tr><td colspan='11' style='text-align: center; color: #999;'>暂无数据</td></tr>"]
)
# 按模块分类统计
module_rows = "\n".join(
@@ -975,11 +987,12 @@ class StatisticOutputTask(AsyncTask):
f"<td>{stat_data[STD_TIME_COST_BY_MODULE][module_name]:.1f} 秒</td>"
f"<td>{_format_large_number(count / total_replies, html=True) if total_replies > 0 else 'N/A'}</td>"
f"<td>{_format_large_number(stat_data[TOTAL_TOK_BY_MODULE][module_name] / total_replies, html=True) if total_replies > 0 else 'N/A'}</td>"
f"<td>{_format_large_number(stat_data[TOTAL_TOK_BY_MODULE][module_name] / count, html=True) if count > 0 else 'N/A'}</td>"
f"</tr>"
for module_name, count in sorted(stat_data[REQ_CNT_BY_MODULE].items())
]
if stat_data[REQ_CNT_BY_MODULE]
else ["<tr><td colspan='10' style='text-align: center; color: #999;'>暂无数据</td></tr>"]
else ["<tr><td colspan='11' style='text-align: center; color: #999;'>暂无数据</td></tr>"]
)
# 聊天消息统计
@@ -1054,7 +1067,7 @@ class StatisticOutputTask(AsyncTask):
<h2>按模型分类统计</h2>
<div class=\"table-wrap\">
<table>
<thead><tr><th>模型名称</th><th>调用次数</th><th>输入Token</th><th>输出Token</th><th>Token总量</th><th>累计花费</th><th>平均耗时(秒)</th><th>标准差(秒)</th><th>每次回复平均调用次数</th><th>每次回复平均Token数</th></tr></thead>
<thead><tr><th>模型名称</th><th>调用次数</th><th>输入Token</th><th>输出Token</th><th>Token总量</th><th>累计花费</th><th>平均耗时(秒)</th><th>标准差(秒)</th><th>每次回复平均调用次数</th><th>每次回复平均Token数</th><th>每次调用平均Token</th></tr></thead>
<tbody>
{model_rows}
</tbody>
@@ -1065,7 +1078,7 @@ class StatisticOutputTask(AsyncTask):
<div class=\"table-wrap\">
<table>
<thead>
<tr><th>模块名称</th><th>调用次数</th><th>输入Token</th><th>输出Token</th><th>Token总量</th><th>累计花费</th><th>平均耗时(秒)</th><th>标准差(秒)</th><th>每次回复平均调用次数</th><th>每次回复平均Token数</th></tr>
<tr><th>模块名称</th><th>调用次数</th><th>输入Token</th><th>输出Token</th><th>Token总量</th><th>累计花费</th><th>平均耗时(秒)</th><th>标准差(秒)</th><th>每次回复平均调用次数</th><th>每次回复平均Token数</th><th>每次调用平均Token</th></tr>
</thead>
<tbody>
{module_rows}
@@ -1077,7 +1090,7 @@ class StatisticOutputTask(AsyncTask):
<div class=\"table-wrap\">
<table>
<thead>
<tr><th>请求类型</th><th>调用次数</th><th>输入Token</th><th>输出Token</th><th>Token总量</th><th>累计花费</th><th>平均耗时(秒)</th><th>标准差(秒)</th><th>每次回复平均调用次数</th><th>每次回复平均Token数</th></tr>
<tr><th>请求类型</th><th>调用次数</th><th>输入Token</th><th>输出Token</th><th>Token总量</th><th>累计花费</th><th>平均耗时(秒)</th><th>标准差(秒)</th><th>每次回复平均调用次数</th><th>每次回复平均Token数</th><th>每次调用平均Token</th></tr>
</thead>
<tbody>
{type_rows}