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

v0.5.9

Browse Source 
修复了记忆刷屏 加入了又新又好错别字生成器 增加了记忆过滤
This commit is contained in:
SengokuCola
2025-03-07 00:09:36 +08:00
parent 21d1a69b6e
commit 8ef00ee571
8 changed files with 883 additions and 547 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
src/plugins/memory_system/memory.py
View File

@@ -181,13 +181,19 @@ class Hippocampus:
topic_num = self.calculate_topic_num(input_text, compress_rate)
topics_response = await self.llm_model_get_topic.generate_response(self.find_topic_llm(input_text, topic_num))
# 修改话题处理逻辑
print(f"话题: {topics_response[0]}")
topics = [topic.strip() for topic in topics_response[0].replace("，", ",").replace("、", ",").replace(" ", ",").split(",") if topic.strip()]
print(f"话题: {topics}")
# 定义需要过滤的关键词
filter_keywords = ['表情包', '图片', '回复', '聊天记录']
# 创建所有话题的请求任务
# 过滤topics
topics = [topic.strip() for topic in topics_response[0].replace("，", ",").replace("、", ",").replace(" ", ",").split(",") if topic.strip()]
filtered_topics = [topic for topic in topics if not any(keyword in topic for keyword in filter_keywords)]
# print(f"原始话题: {topics}")
print(f"过滤后话题: {filtered_topics}")
# 使用过滤后的话题继续处理
tasks = []
for topic in topics:
for topic in filtered_topics:
topic_what_prompt = self.topic_what(input_text, topic)
# 创建异步任务
task = self.llm_model_summary.generate_response_async(topic_what_prompt)
@@ -501,9 +507,9 @@ class Hippocampus:
list: 识别出的主题列表
"""
topics_response = await self.llm_model_get_topic.generate_response(self.find_topic_llm(text, 5))
print(f"话题: {topics_response[0]}")
# print(f"话题: {topics_response[0]}")
topics = [topic.strip() for topic in topics_response[0].replace("，", ",").replace("、", ",").replace(" ", ",").split(",") if topic.strip()]
print(f"话题: {topics}")
# print(f"话题: {topics}")
return topics
@@ -579,7 +585,7 @@ class Hippocampus:
print(f"\033[1;32m[记忆激活]\033[0m 识别出的主题: {identified_topics}")
if not identified_topics:
print(f"\033[1;32m[记忆激活]\033[0m 未识别出主题,返回0")
# print(f"\033[1;32m[记忆激活]\033[0m 未识别出主题,返回0")
return 0
# 查找相似主题
@@ -644,7 +650,7 @@ class Hippocampus:
return int(activation)
async def get_relevant_memories(self, text: str, max_topics: int = 5, similarity_threshold: float = 0.4) -> list:
async def get_relevant_memories(self, text: str, max_topics: int = 5, similarity_threshold: float = 0.4, max_memory_num: int = 5) -> list:
"""根据输入文本获取相关的记忆内容"""
# 识别主题
identified_topics = await self._identify_topics(text)
@@ -665,6 +671,9 @@ class Hippocampus:
# 获取该主题的记忆内容
first_layer, _ = self.memory_graph.get_related_item(topic, depth=1)
if first_layer:
# 如果记忆条数超过限制，随机选择指定数量的记忆
if len(first_layer) > max_memory_num:
first_layer = random.sample(first_layer, max_memory_num)
# 为每条记忆添加来源主题和相似度信息
for memory in first_layer:
relevant_memories.append({

33
src/plugins/memory_system/memory_manual_build.py
View File

@@ -234,16 +234,22 @@ class Hippocampus:
async def memory_compress(self, input_text, compress_rate=0.1):
print(input_text)
#获取topics
topic_num = self.calculate_topic_num(input_text, compress_rate)
topics_response = await self.llm_model_get_topic.generate_response_async(self.find_topic_llm(input_text, topic_num))
topics_response = self.llm_model_get_topic.generate_response(self.find_topic_llm(input_text, topic_num))
# 修改话题处理逻辑
# 定义需要过滤的关键词
filter_keywords = ['表情包', '图片', '回复', '聊天记录']
# 过滤topics
topics = [topic.strip() for topic in topics_response[0].replace("，", ",").replace("、", ",").replace(" ", ",").split(",") if topic.strip()]
print(f"话题: {topics}")
filtered_topics = [topic for topic in topics if not any(keyword in topic for keyword in filter_keywords)]
# print(f"原始话题: {topics}")
print(f"过滤后话题: {filtered_topics}")
# 创建所有话题的请求任务
tasks = []
for topic in topics:
for topic in filtered_topics:
topic_what_prompt = self.topic_what(input_text, topic)
# 创建异步任务
task = self.llm_model_small.generate_response_async(topic_what_prompt)
@@ -650,7 +656,22 @@ def visualize_graph_lite(memory_graph: Memory_graph, color_by_memory: bool = Fal
G = memory_graph.G
# 创建一个新图用于可视化
H = G.copy()
H = G.copy()
# 过滤掉内容数量小于2的节点
nodes_to_remove = []
for node in H.nodes():
memory_items = H.nodes[node].get('memory_items', [])
memory_count = len(memory_items) if isinstance(memory_items, list) else (1 if memory_items else 0)
if memory_count < 2:
nodes_to_remove.append(node)
H.remove_nodes_from(nodes_to_remove)
# 如果没有符合条件的节点，直接返回
if len(H.nodes()) == 0:
print("没有找到内容数量大于等于2的节点")
return
# 计算节点大小和颜色
node_colors = []
@@ -704,7 +725,7 @@ def visualize_graph_lite(memory_graph: Memory_graph, color_by_memory: bool = Fal
edge_color='gray',
width=1.5) # 统一的边宽度
title = '记忆图谱可视化 - 节点大小表示记忆数量\n节点颜色：蓝(弱连接)到红(强连接)渐变，边的透明度表示连接强度\n连接强度越大的节点距离越近'
title = '记忆图谱可视化（仅显示内容≥2的节点）\n节点大小表示记忆数量\n节点颜色：蓝(弱连接)到红(强连接)渐变，边的透明度表示连接强度\n连接强度越大的节点距离越近'
plt.title(title, fontsize=16, fontfamily='SimHei')
plt.show()