gardel/Mofox-Core
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

better：优化normal的 prompt

Browse Source
This commit is contained in:
SengokuCola
2025-06-08 21:29:20 +08:00
parent d45ef07ea8
commit 52cb3ed273
6 changed files with 643 additions and 48 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
src/chat/focus_chat/planners/actions/reply_action.py
View File

@@ -25,7 +25,8 @@ class ReplyAction(BaseAction):
action_name: str = "reply"
action_description: str = "当你想要参与回复或者聊天"
action_parameters: dict[str:str] = {
"target": "如果你要明确回复特定某人的某句话，请在target参数中中指定那句话的原始文本（非必须，仅文本，不包含发送者)（可选）",
"reply_to": "如果是明确回复某个人的发言，请在reply_to参数中指定，格式：（用户名:发言内容），如果不是，reply_to的值设为none",
"emoji": "如果你想用表情包辅助你的回答，请在emoji参数中用文字描述你想要发送的表情包内容，如果没有，值设为空",
}
action_require: list[str] = [
"你想要闲聊或者随便附和",

2
src/chat/focus_chat/planners/planner_simple.py
View File

@@ -181,7 +181,7 @@ class ActionPlanner(BasePlanner):
prompt = f"{prompt}"
llm_content, (reasoning_content, _) = await self.planner_llm.generate_response_async(prompt=prompt)
logger.debug(f"{self.log_prefix}规划器原始提示词: {prompt}")
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}")

135
src/chat/focus_chat/replyer/default_replyer.py
View File

@@ -23,6 +23,9 @@ from src.chat.focus_chat.expressors.exprssion_learner import expression_learner
import random
from datetime import datetime
import re
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.metrics.pairwise import cosine_similarity
import numpy as np
logger = get_logger("replyer")
@@ -32,6 +35,7 @@ def init_prompt():
"""
你可以参考以下的语言习惯，如果情景合适就使用，不要盲目使用,不要生硬使用，而是结合到表达中：
{style_habbits}
请你根据情景使用以下句法：
{grammar_habbits}
@@ -40,15 +44,10 @@ def init_prompt():
{relation_info_block}
{time_block}
你现在正在群里聊天，以下是群里正在进行的聊天内容：
{chat_info}
以上是聊天内容，你需要了解聊天记录中的内容
{chat_target}
{identity}，在这聊天中，"{target_message}"引起了你的注意，你想要在群里发言或者回复这条消息。
{chat_info}
{reply_target_block}
{identity}
你需要使用合适的语言习惯和句法，参考聊天内容，组织一条日常且口语化的回复。注意不要复读你说过的话。
{config_expression_style}，请注意不要输出多余内容(包括前后缀，冒号和引号，括号()，表情包，at或 @等 )。只输出回复内容。
{keywords_reaction_prompt}
@@ -61,20 +60,17 @@ def init_prompt():
Prompt(
"""
{extra_info_block}
{time_block}
你现在正在聊天，以下是你和对方正在进行的聊天内容：
{chat_info}
以上是聊天内容，你需要了解聊天记录中的内容
{chat_target}
{identity}，在这聊天中，"{target_message}"引起了你的注意，你想要发言或者回复这条消息。
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。注意不要复读你说过的话。
你可以参考以下的语言习惯和句法，如果情景合适就使用，不要盲目使用,不要生硬使用，而是结合到表达中：
{style_habbits}
{grammar_habbits}
{extra_info_block}
{time_block}
{chat_target}
{chat_info}
现在"{sender_name}"说的:{target_message}。引起了你的注意，你想要发言或者回复这条消息。
{identity}，
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。注意不要复读你说过的话。
你可以参考以下的语言习惯和句法，如果情景合适就使用，不要盲目使用,不要生硬使用，而是结合到表达中：
{config_expression_style}，请注意不要输出多余内容(包括前后缀，冒号和引号，括号()，表情包，at或 @等 )。只输出回复内容。
{keywords_reaction_prompt}
@@ -155,11 +151,12 @@ class DefaultReplyer:
action_data=action_data,
)
# with Timer("选择表情", cycle_timers):
# emoji_keyword = action_data.get("emojis", [])
# emoji_base64 = await self._choose_emoji(emoji_keyword)
# if emoji_base64:
# reply.append(("emoji", emoji_base64))
with Timer("选择表情", cycle_timers):
emoji_keyword = action_data.get("emoji", "")
if emoji_keyword:
emoji_base64 = await self._choose_emoji(emoji_keyword)
if emoji_base64:
reply.append(("emoji", emoji_base64))
if reply:
with Timer("发送消息", cycle_timers):
@@ -251,23 +248,22 @@ class DefaultReplyer:
# 2. 获取信息捕捉器
info_catcher = info_catcher_manager.get_info_catcher(thinking_id)
# --- Determine sender_name for private chat ---
sender_name_for_prompt = "某人" # Default for group or if info unavailable
if not self.is_group_chat and self.chat_target_info:
# Prioritize person_name, then nickname
sender_name_for_prompt = (
self.chat_target_info.get("person_name")
or self.chat_target_info.get("user_nickname")
or sender_name_for_prompt
)
# --- End determining sender_name ---
target_message = action_data.get("target", "")
reply_to = action_data.get("reply_to", "none")
sender = ""
targer = ""
if ":" in reply_to or "：" in reply_to:
# 使用正则表达式匹配中文或英文冒号
parts = re.split(pattern=r'[:：]', string=reply_to, maxsplit=1)
if len(parts) == 2:
sender = parts[0].strip()
targer = parts[1].strip()
identity = action_data.get("identity", "")
extra_info_block = action_data.get("extra_info_block", "")
relation_info_block = action_data.get("relation_info_block", "")
# 3. 构建 Prompt
with Timer("构建Prompt", {}): # 内部计时器，可选保留
prompt = await self.build_prompt_focus(
@@ -277,8 +273,8 @@ class DefaultReplyer:
extra_info_block=extra_info_block,
relation_info_block=relation_info_block,
reason=reason,
sender_name=sender_name_for_prompt, # Pass determined name
target_message=target_message,
sender_name=sender, # Pass determined name
target_message=targer,
config_expression_style=global_config.expression.expression_style,
)
@@ -340,6 +336,7 @@ class DefaultReplyer:
identity,
target_message,
config_expression_style,
# stuation,
) -> str:
is_group_chat = bool(chat_stream.group_info)
@@ -368,15 +365,16 @@ class DefaultReplyer:
grammar_habbits = []
# 1. learnt_expressions加权随机选3条
if learnt_style_expressions:
weights = [expr["count"] for expr in learnt_style_expressions]
selected_learnt = weighted_sample_no_replacement(learnt_style_expressions, weights, 4)
for expr in selected_learnt:
# 使用相似度匹配选择最相似的表达
similar_exprs = find_similar_expressions(target_message, learnt_style_expressions, 3)
for expr in similar_exprs:
print(f"expr: {expr}")
if isinstance(expr, dict) and "situation" in expr and "style" in expr:
style_habbits.append(f"当{expr['situation']}时，使用 {expr['style']}")
# 2. learnt_grammar_expressions加权随机选3条
# 2. learnt_grammar_expressions加权随机选2条
if learnt_grammar_expressions:
weights = [expr["count"] for expr in learnt_grammar_expressions]
selected_learnt = weighted_sample_no_replacement(learnt_grammar_expressions, weights, 4)
selected_learnt = weighted_sample_no_replacement(learnt_grammar_expressions, weights, 2)
for expr in selected_learnt:
if isinstance(expr, dict) and "situation" in expr and "style" in expr:
grammar_habbits.append(f"当{expr['situation']}时，使用 {expr['style']}")
@@ -419,6 +417,16 @@ class DefaultReplyer:
time_block = f"当前时间：{datetime.now().strftime('%Y-%m-%d %H:%M:%S')}"
# logger.debug("开始构建 focus prompt")
if sender_name:
reply_target_block = f"现在{sender_name}说的:{target_message}。引起了你的注意，你想要在群里发言或者回复这条消息。"
elif target_message:
reply_target_block = f"现在{target_message}引起了你的注意，你想要在群里发言或者回复这条消息。"
else:
reply_target_block = "现在，你想要在群里发言或者回复消息。"
# --- Choose template based on chat type ---
if is_group_chat:
@@ -436,6 +444,7 @@ class DefaultReplyer:
extra_info_block=extra_info_block,
relation_info_block=relation_info_block,
time_block=time_block,
reply_target_block=reply_target_block,
# bot_name=global_config.bot.nickname,
# prompt_personality="",
# reason=reason,
@@ -443,6 +452,7 @@ class DefaultReplyer:
keywords_reaction_prompt=keywords_reaction_prompt,
identity=identity,
target_message=target_message,
sender_name=sender_name,
config_expression_style=config_expression_style,
)
else: # Private chat
@@ -457,6 +467,7 @@ class DefaultReplyer:
extra_info_block=extra_info_block,
relation_info_block=relation_info_block,
time_block=time_block,
reply_target_block=reply_target_block,
# bot_name=global_config.bot.nickname,
# prompt_personality="",
# reason=reason,
@@ -464,6 +475,7 @@ class DefaultReplyer:
keywords_reaction_prompt=keywords_reaction_prompt,
identity=identity,
target_message=target_message,
sender_name=sender_name,
config_expression_style=config_expression_style,
)
@@ -659,4 +671,35 @@ def weighted_sample_no_replacement(items, weights, k) -> list:
return selected
def find_similar_expressions(input_text: str, expressions: List[Dict], top_k: int = 3) -> List[Dict]:
"""使用TF-IDF和余弦相似度找出与输入文本最相似的top_k个表达方式"""
if not expressions:
return []
# 准备文本数据
texts = [expr['situation'] for expr in expressions]
texts.append(input_text) # 添加输入文本
# 使用TF-IDF向量化
vectorizer = TfidfVectorizer()
tfidf_matrix = vectorizer.fit_transform(texts)
# 计算余弦相似度
similarity_matrix = cosine_similarity(tfidf_matrix)
# 获取输入文本的相似度分数（最后一行）
scores = similarity_matrix[-1][:-1] # 排除与自身的相似度
# 获取top_k的索引
top_indices = np.argsort(scores)[::-1][:top_k]
# 获取相似表达
similar_exprs = []
for idx in top_indices:
if scores[idx] > 0: # 只保留有相似度的
similar_exprs.append(expressions[idx])
return similar_exprs
init_prompt()