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

feat：实现具有向量和元数据索引的统一内存存储系统

Browse Source 
- 添加了 UnifiedMemoryStorage 类，用于管理带向量嵌入的内存块。
- 集成了 FAISS，以实现高效的向量存储和搜索。
- 实现了内存缓存、关键字、类型和用户索引。
- 增加了内存遗忘和自动保存存储数据的支持。
- 包含用于存储、搜索和遗忘记忆的方法。
- 引入了存储行为和性能的配置选项。
- 实现了从磁盘加载和保存内存及向量数据。
This commit is contained in:
Windpicker-owo
2025-10-01 18:02:42 +08:00
parent d46475ca8c
commit afb1a75ebf
28 changed files with 1883 additions and 499 deletions
Download Patch File Download Diff File Expand all files Collapse all files

104
src/chat/replyer/default_generator.py
View File

@@ -491,10 +491,10 @@ class DefaultReplyer:
running_memories = []
instant_memory = None
if global_config.memory.enable_instant_memory:
if global_config.memory.enable_memory:
try:
# 使用新的增强记忆系统
from src.chat.memory_system.enhanced_memory_integration import recall_memories, remember_message
# 使用新的统一记忆系统
from src.chat.memory_system import get_memory_system
stream = self.chat_stream
user_info_obj = getattr(stream, "user_info", None)
@@ -588,33 +588,37 @@ class DefaultReplyer:
memory_context = {key: value for key, value in memory_context.items() if value}
# 获取记忆系统实例
memory_system = get_memory_system()
# 检索相关记忆
enhanced_memories = await recall_memories(
enhanced_memories = await memory_system.retrieve_relevant_memories(
query=target,
user_id=memory_user_id,
chat_id=stream.stream_id,
context=memory_context
scope_id=stream.stream_id,
context=memory_context,
limit=10
)
# 注意：记忆存储已迁移到回复生成完成后进行，不在查询阶段执行
# 转换格式以兼容现有代码
running_memories = []
if enhanced_memories and enhanced_memories.get("has_memories"):
for memory in enhanced_memories.get("memories", []):
if enhanced_memories:
for memory_chunk in enhanced_memories:
running_memories.append({
"content": memory.get("content", ""),
"memory_type": memory.get("type", "unknown"),
"confidence": memory.get("confidence"),
"importance": memory.get("importance"),
"relevance": memory.get("relevance"),
"source": memory.get("source"),
"structure": memory.get("structure"),
"content": memory_chunk.display or memory_chunk.text_content or "",
"memory_type": memory_chunk.memory_type.value,
"confidence": memory_chunk.metadata.confidence.value,
"importance": memory_chunk.metadata.importance.value,
"relevance": getattr(memory_chunk, 'relevance_score', 0.5),
"source": memory_chunk.metadata.source,
"structure": memory_chunk.content_structure.value if memory_chunk.content_structure else "unknown",
})
# 构建瞬时记忆字符串
if enhanced_memories and enhanced_memories.get("has_memories"):
top_memory = enhanced_memories.get("memories", [])[:1]
if running_memories:
top_memory = running_memories[:1]
if top_memory:
instant_memory = top_memory[0].get("content", "")
@@ -639,26 +643,45 @@ class DefaultReplyer:
rounded = int(value)
return mapping.get(rounded, f"{value:.2f}")
# 构建记忆字符串，即使某种记忆为空也要继续
# 构建记忆字符串，使用方括号格式
memory_str = ""
has_any_memory = False
# 添加长期记忆（来自增强记忆系统）
if running_memories:
if not memory_str:
memory_str = "以下是当前在聊天中，你回忆起的记忆：\n"
for running_memory in running_memories:
details = []
details.append(f"类型: {running_memory.get('memory_type', 'unknown')}")
if running_memory.get("confidence") is not None:
details.append(f"置信度: {_format_confidence_label(running_memory.get('confidence'))}")
if running_memory.get("importance") is not None:
details.append(f"重要性: {_format_importance_label(running_memory.get('importance'))}")
if running_memory.get("relevance") is not None:
details.append(f"相关度: {running_memory['relevance']:.2f}")
# 使用方括号格式
memory_parts = ["### 🧠 相关记忆 (Relevant Memories)", ""]
detail_text = f" （{'，'.join(details)}）" if details else ""
memory_str += f"- {running_memory['content']}{detail_text}\n"
# 按相关度排序，并记录相关度信息用于调试
sorted_memories = sorted(running_memories, key=lambda x: x.get('relevance', 0.0), reverse=True)
# 调试相关度信息
relevance_info = [(m.get('memory_type', 'unknown'), m.get('relevance', 0.0)) for m in sorted_memories]
logger.debug(f"记忆相关度信息: {relevance_info}")
for running_memory in sorted_memories:
content = running_memory.get('content', '')
memory_type = running_memory.get('memory_type', 'unknown')
# 映射记忆类型到中文标签
type_mapping = {
"personal_fact": "个人事实",
"preference": "偏好",
"event": "事件",
"opinion": "观点",
"relationship": "个人事实",
"unknown": "未知"
}
chinese_type = type_mapping.get(memory_type, "未知")
# 提取纯净内容（如果包含旧格式的元数据）
clean_content = content
if "（类型:" in content and "）" in content:
clean_content = content.split("（类型:")[0].strip()
memory_parts.append(f"- **[{chinese_type}]** {clean_content}")
memory_str = "\n".join(memory_parts) + "\n"
has_any_memory = True
# 添加瞬时记忆
@@ -1790,11 +1813,11 @@ class DefaultReplyer:
reply_message: 回复的原始消息
"""
try:
if not global_config.memory.enable_memory or not global_config.memory.enable_instant_memory:
if not global_config.memory.enable_memory:
return
# 使用增强记忆系统存储记忆
from src.chat.memory_system.enhanced_memory_integration import remember_message
# 使用统一记忆系统存储记忆
from src.chat.memory_system import get_memory_system
stream = self.chat_stream
user_info_obj = getattr(stream, "user_info", None)
@@ -1904,12 +1927,15 @@ class DefaultReplyer:
)
# 异步存储聊天历史（完全非阻塞）
memory_system = get_memory_system()
asyncio.create_task(
remember_message(
message=chat_history,
user_id=memory_user_id,
chat_id=stream.stream_id,
context=memory_context
memory_system.process_conversation_memory(
context={
"conversation_text": chat_history,
"user_id": memory_user_id,
"scope_id": stream.stream_id,
**memory_context
}
)
)