From 476743427ac0c4a716c49a803ffd88cd1e648b97 Mon Sep 17 00:00:00 2001 From: SengokuCola <1026294844@qq.com> Date: Mon, 2 Jun 2025 20:50:05 +0800 Subject: [PATCH] =?UTF-8?q?better=EF=BC=9A=E4=BC=98=E5=8C=96=E5=B7=A5?= =?UTF-8?q?=E4=BD=9C=E8=AE=B0=E5=BF=86=E5=A4=84=E7=90=86=E5=99=A8?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- .../working_memory_processor.py | 14 +- .../working_memory/memory_manager.py | 224 ++++-------------- template/bot_config_template.toml | 2 +- 3 files changed, 54 insertions(+), 186 deletions(-) diff --git a/src/chat/focus_chat/info_processors/working_memory_processor.py b/src/chat/focus_chat/info_processors/working_memory_processor.py index da7203989..faef24288 100644 --- a/src/chat/focus_chat/info_processors/working_memory_processor.py +++ b/src/chat/focus_chat/info_processors/working_memory_processor.py @@ -108,9 +108,7 @@ class WorkingMemoryProcessor(BaseProcessor): memory_summary = memory.summary memory_id = memory.id memory_brief = memory_summary.get("brief") - # memory_detailed = memory_summary.get("detailed") - memory_keypoints = memory_summary.get("keypoints") - memory_events = memory_summary.get("events") + memory_keypoints = memory_summary.get("key_points", []) memory_single_prompt = f"记忆id:{memory_id},记忆摘要:{memory_brief}\n" memory_prompts.append(memory_single_prompt) @@ -165,15 +163,9 @@ class WorkingMemoryProcessor(BaseProcessor): memory_summary = memory.summary memory_id = memory.id memory_brief = memory_summary.get("brief") - # memory_detailed = memory_summary.get("detailed") - memory_keypoints = memory_summary.get("keypoints") - memory_events = memory_summary.get("events") + memory_keypoints = memory_summary.get("key_points", []) for keypoint in memory_keypoints: memory_str += f"记忆要点:{keypoint}\n" - for event in memory_events: - memory_str += f"记忆事件:{event}\n" - # memory_str += f"记忆摘要:{memory_detailed}\n" - # memory_str += f"记忆主题:{memory_brief}\n" working_memory_info = WorkingMemoryInfo() if memory_str: @@ -225,7 +217,7 @@ class WorkingMemoryProcessor(BaseProcessor): logger.debug(f"{self.log_prefix} 异步合并记忆成功: {memory_id1} 和 {memory_id2}...") logger.debug(f"{self.log_prefix} 合并后的记忆梗概: {merged_memory.summary.get('brief')}") logger.debug(f"{self.log_prefix} 合并后的记忆详情: {merged_memory.summary.get('detailed')}") - logger.debug(f"{self.log_prefix} 合并后的记忆要点: {merged_memory.summary.get('keypoints')}") + logger.debug(f"{self.log_prefix} 合并后的记忆要点: {merged_memory.summary.get('key_points')}") logger.debug(f"{self.log_prefix} 合并后的记忆事件: {merged_memory.summary.get('events')}") except Exception as e: diff --git a/src/chat/focus_chat/working_memory/memory_manager.py b/src/chat/focus_chat/working_memory/memory_manager.py index bdbb429e7..1015207aa 100644 --- a/src/chat/focus_chat/working_memory/memory_manager.py +++ b/src/chat/focus_chat/working_memory/memory_manager.py @@ -224,39 +224,27 @@ class MemoryManager: Returns: 包含总结、概括、关键概念和事件的字典 """ - prompt = f"""请对以下内容进行总结，总结成记忆，输出四部分： + prompt = f"""请对以下内容进行总结，总结成记忆，输出两部分： 1. 记忆内容主题（精简，20字以内）：让用户可以一眼看出记忆内容是什么 -2. 记忆内容概括（200字以内）：让用户可以了解记忆内容的大致内容 -3. 关键概念和知识（keypoints）：多条，提取关键的概念、知识点和关键词，要包含对概念的解释 -4. 事件描述（events）：多条，描述谁（人物）在什么时候（时间）做了什么（事件） +2. key_points：多条，包含关键的概念、事件，每条都要包含解释或描述，谁在什么时候干了什么 内容： {content} 请按以下JSON格式输出： -```json {{ "brief": "记忆内容主题（20字以内）", - "detailed": "记忆内容概括（200字以内）", - "keypoints": [ - "概念1：解释", - "概念2：解释", - ... - ], - "events": [ - "事件1：谁在什么时候做了什么", - "事件2：谁在什么时候做了什么", + "key_points": [ + "要点1：解释或描述", + "要点2：解释或描述", ... ] }} -``` 请确保输出是有效的JSON格式，不要添加任何额外的说明或解释。 """ default_summary = { "brief": "主题未知的记忆", - "detailed": "大致内容未知的记忆", - "keypoints": ["未知的概念"], - "events": ["未知的事件"], + "key_points": ["未知的要点"], } try: @@ -288,29 +276,14 @@ class MemoryManager: if "brief" not in json_result or not isinstance(json_result["brief"], str): json_result["brief"] = "主题未知的记忆" - if "detailed" not in json_result or not isinstance(json_result["detailed"], str): - json_result["detailed"] = "大致内容未知的记忆" - - # 处理关键概念 - if "keypoints" not in json_result or not isinstance(json_result["keypoints"], list): - json_result["keypoints"] = ["未知的概念"] + # 处理关键要点 + if "key_points" not in json_result or not isinstance(json_result["key_points"], list): + json_result["key_points"] = ["未知的要点"] else: - # 确保keypoints中的每个项目都是字符串 - json_result["keypoints"] = [str(point) for point in json_result["keypoints"] if point is not None] - if not json_result["keypoints"]: - json_result["keypoints"] = ["未知的概念"] - - # 处理事件 - if "events" not in json_result or not isinstance(json_result["events"], list): - json_result["events"] = ["未知的事件"] - else: - # 确保events中的每个项目都是字符串 - json_result["events"] = [str(event) for event in json_result["events"] if event is not None] - if not json_result["events"]: - json_result["events"] = ["未知的事件"] - - # 兼容旧版，将keypoints和events合并到key_points中 - json_result["key_points"] = json_result["keypoints"] + json_result["events"] + # 确保key_points中的每个项目都是字符串 + json_result["key_points"] = [str(point) for point in json_result["key_points"] if point is not None] + if not json_result["key_points"]: + json_result["key_points"] = ["未知的要点"] return json_result @@ -348,52 +321,31 @@ class MemoryManager: # 使用LLM根据要求对总结、概括和要点进行精简修改 prompt = f""" -请根据以下要求，对记忆内容的主题、概括、关键概念和事件进行精简，模拟记忆的遗忘过程： +请根据以下要求，对记忆内容的主题和关键要点进行精简，模拟记忆的遗忘过程： 要求：{requirements} -你可以随机对关键概念和事件进行压缩，模糊或者丢弃，修改后，同样修改主题和概括 +你可以随机对关键要点进行压缩，模糊或者丢弃，修改后，同样修改主题 目前主题：{summary["brief"]} -目前概括：{summary["detailed"]} +目前关键要点： +{chr(10).join([f"- {point}" for point in summary.get("key_points", [])])} -目前关键概念： -{chr(10).join([f"- {point}" for point in summary.get("keypoints", [])])} - -目前事件： -{chr(10).join([f"- {point}" for point in summary.get("events", [])])} - -请生成修改后的主题、概括、关键概念和事件，遵循以下格式： +请生成修改后的主题和关键要点，遵循以下格式： ```json {{ "brief": "修改后的主题（20字以内）", - "detailed": "修改后的概括（200字以内）", - "keypoints": [ - "修改后的概念1：解释", - "修改后的概念2：解释" - ], - "events": [ - "修改后的事件1：谁在什么时候做了什么", - "修改后的事件2：谁在什么时候做了什么" + "key_points": [ + "修改后的要点1：解释或描述", + "修改后的要点2：解释或描述" ] }} ``` 请确保输出是有效的JSON格式，不要添加任何额外的说明或解释。 """ - # 检查summary中是否有旧版结构，转换为新版结构 - if "keypoints" not in summary and "events" not in summary and "key_points" in summary: - # 尝试区分key_points中的keypoints和events - # 简单地将前半部分视为keypoints，后半部分视为events - key_points = summary.get("key_points", []) - halfway = len(key_points) // 2 - summary["keypoints"] = key_points[:halfway] or ["未知的概念"] - summary["events"] = key_points[halfway:] or ["未知的事件"] - # 定义默认的精简结果 default_refined = { "brief": summary["brief"], - "detailed": summary["detailed"], - "keypoints": summary.get("keypoints", ["未知的概念"])[:1], # 默认只保留第一个关键概念 - "events": summary.get("events", ["未知的事件"])[:1], # 默认只保留第一个事件 + "key_points": summary.get("key_points", ["未知的要点"])[:1], # 默认只保留第一个要点 } try: @@ -421,30 +373,17 @@ class MemoryManager: logger.error(f"修复后的JSON不是字典类型: {type(refined_data)}") refined_data = default_refined - # 更新总结、概括 + # 更新总结 summary["brief"] = refined_data.get("brief", "主题未知的记忆") - summary["detailed"] = refined_data.get("detailed", "大致内容未知的记忆") - # 更新关键概念 - keypoints = refined_data.get("keypoints", []) - if isinstance(keypoints, list) and keypoints: - # 确保所有关键概念都是字符串 - summary["keypoints"] = [str(point) for point in keypoints if point is not None] + # 更新关键要点 + key_points = refined_data.get("key_points", []) + if isinstance(key_points, list) and key_points: + # 确保所有要点都是字符串 + summary["key_points"] = [str(point) for point in key_points if point is not None] else: - # 如果keypoints不是列表或为空，使用默认值 - summary["keypoints"] = ["主要概念已遗忘"] - - # 更新事件 - events = refined_data.get("events", []) - if isinstance(events, list) and events: - # 确保所有事件都是字符串 - summary["events"] = [str(event) for event in events if event is not None] - else: - # 如果events不是列表或为空，使用默认值 - summary["events"] = ["事件细节已遗忘"] - - # 兼容旧版，维护key_points - summary["key_points"] = summary["keypoints"] + summary["events"] + # 如果key_points不是列表或为空，使用默认值 + summary["key_points"] = ["主要要点已遗忘"] except Exception as e: logger.error(f"精简记忆出错: {str(e)}") @@ -452,9 +391,7 @@ class MemoryManager: # 出错时使用简化的默认精简 summary["brief"] = summary["brief"] + " (已简化)" - summary["keypoints"] = summary.get("keypoints", ["未知的概念"])[:1] - summary["events"] = summary.get("events", ["未知的事件"])[:1] - summary["key_points"] = summary["keypoints"] + summary["events"] + summary["key_points"] = summary.get("key_points", ["未知的要点"])[:1] except Exception as e: logger.error(f"精简记忆调用LLM出错: {str(e)}") @@ -573,27 +510,11 @@ class MemoryManager: # 如果有摘要信息，添加到提示中 if summary1: prompt += f"记忆1主题：{summary1['brief']}\n" - prompt += f"记忆1概括：{summary1['detailed']}\n" - - if "keypoints" in summary1: - prompt += "记忆1关键概念：\n" + "\n".join([f"- {point}" for point in summary1["keypoints"]]) + "\n\n" - - if "events" in summary1: - prompt += "记忆1事件：\n" + "\n".join([f"- {point}" for point in summary1["events"]]) + "\n\n" - elif "key_points" in summary1: - prompt += "记忆1要点：\n" + "\n".join([f"- {point}" for point in summary1["key_points"]]) + "\n\n" + prompt += "记忆1关键要点：\n" + "\n".join([f"- {point}" for point in summary1.get("key_points", [])]) + "\n\n" if summary2: prompt += f"记忆2主题：{summary2['brief']}\n" - prompt += f"记忆2概括：{summary2['detailed']}\n" - - if "keypoints" in summary2: - prompt += "记忆2关键概念：\n" + "\n".join([f"- {point}" for point in summary2["keypoints"]]) + "\n\n" - - if "events" in summary2: - prompt += "记忆2事件：\n" + "\n".join([f"- {point}" for point in summary2["events"]]) + "\n\n" - elif "key_points" in summary2: - prompt += "记忆2要点：\n" + "\n".join([f"- {point}" for point in summary2["key_points"]]) + "\n\n" + prompt += "记忆2关键要点：\n" + "\n".join([f"- {point}" for point in summary2.get("key_points", [])]) + "\n\n" # 添加记忆原始内容 prompt += f""" @@ -608,15 +529,10 @@ class MemoryManager: {{ "content": "合并后的记忆内容文本（尽可能保留原信息，但去除重复）", "brief": "合并后的主题（20字以内）", - "detailed": "合并后的概括（200字以内）", - "keypoints": [ - "合并后的概念1：解释", - "合并后的概念2：解释", - "合并后的概念3：解释" - ], - "events": [ - "合并后的事件1：谁在什么时候做了什么", - "合并后的事件2：谁在什么时候做了什么" + "key_points": [ + "合并后的要点1：解释或描述", + "合并后的要点2：解释或描述", + "合并后的要点3：解释或描述" ] }} ``` @@ -627,40 +543,18 @@ class MemoryManager: default_merged = { "content": f"{content1}\n\n{content2}", "brief": f"合并：{summary1['brief']} + {summary2['brief']}", - "detailed": f"合并了两个记忆：{summary1['detailed']} 以及 {summary2['detailed']}", - "keypoints": [], - "events": [], + "key_points": [], } - # 合并旧版key_points + # 合并key_points if "key_points" in summary1: - default_merged["keypoints"].extend(summary1.get("keypoints", [])) - default_merged["events"].extend(summary1.get("events", [])) - # 如果没有新的结构，尝试从旧结构分离 - if not default_merged["keypoints"] and not default_merged["events"] and "key_points" in summary1: - key_points = summary1["key_points"] - halfway = len(key_points) // 2 - default_merged["keypoints"].extend(key_points[:halfway]) - default_merged["events"].extend(key_points[halfway:]) - + default_merged["key_points"].extend(summary1["key_points"]) if "key_points" in summary2: - default_merged["keypoints"].extend(summary2.get("keypoints", [])) - default_merged["events"].extend(summary2.get("events", [])) - # 如果没有新的结构，尝试从旧结构分离 - if not default_merged["keypoints"] and not default_merged["events"] and "key_points" in summary2: - key_points = summary2["key_points"] - halfway = len(key_points) // 2 - default_merged["keypoints"].extend(key_points[:halfway]) - default_merged["events"].extend(key_points[halfway:]) + default_merged["key_points"].extend(summary2["key_points"]) # 确保列表不为空 - if not default_merged["keypoints"]: - default_merged["keypoints"] = ["合并的关键概念"] - if not default_merged["events"]: - default_merged["events"] = ["合并的事件"] - - # 添加key_points兼容 - default_merged["key_points"] = default_merged["keypoints"] + default_merged["events"] + if not default_merged["key_points"]: + default_merged["key_points"] = ["合并的要点"] try: # 调用LLM合并记忆 @@ -694,29 +588,14 @@ class MemoryManager: if "brief" not in merged_data or not isinstance(merged_data["brief"], str): merged_data["brief"] = default_merged["brief"] - if "detailed" not in merged_data or not isinstance(merged_data["detailed"], str): - merged_data["detailed"] = default_merged["detailed"] - - # 处理关键概念 - if "keypoints" not in merged_data or not isinstance(merged_data["keypoints"], list): - merged_data["keypoints"] = default_merged["keypoints"] + # 处理关键要点 + if "key_points" not in merged_data or not isinstance(merged_data["key_points"], list): + merged_data["key_points"] = default_merged["key_points"] else: - # 确保keypoints中的每个项目都是字符串 - merged_data["keypoints"] = [str(point) for point in merged_data["keypoints"] if point is not None] - if not merged_data["keypoints"]: - merged_data["keypoints"] = ["合并的关键概念"] - - # 处理事件 - if "events" not in merged_data or not isinstance(merged_data["events"], list): - merged_data["events"] = default_merged["events"] - else: - # 确保events中的每个项目都是字符串 - merged_data["events"] = [str(event) for event in merged_data["events"] if event is not None] - if not merged_data["events"]: - merged_data["events"] = ["合并的事件"] - - # 添加key_points兼容 - merged_data["key_points"] = merged_data["keypoints"] + merged_data["events"] + # 确保key_points中的每个项目都是字符串 + merged_data["key_points"] = [str(point) for point in merged_data["key_points"] if point is not None] + if not merged_data["key_points"]: + merged_data["key_points"] = ["合并的要点"] except Exception as e: logger.error(f"合并记忆时处理JSON出错: {str(e)}") @@ -744,9 +623,6 @@ class MemoryManager: # 设置合并后的摘要 summary = { "brief": merged_data["brief"], - "detailed": merged_data["detailed"], - "keypoints": merged_data["keypoints"], - "events": merged_data["events"], "key_points": merged_data["key_points"], } merged_memory.set_summary(summary) diff --git a/template/bot_config_template.toml b/template/bot_config_template.toml index dfac98bcb..9d69827c4 100644 --- a/template/bot_config_template.toml +++ b/template/bot_config_template.toml @@ -98,7 +98,7 @@ parallel_processing = true # 是否并行处理回忆和处理器阶段，可以 processor_max_time = 25 # 处理器最大时间，单位秒，如果超过这个时间，处理器会自动停止 -observation_context_size = 16 # 观察到的最长上下文大小 +observation_context_size = 20 # 观察到的最长上下文大小 compressed_length = 8 # 不能大于observation_context_size,心流上下文压缩的最短压缩长度，超过心流观察到的上下文长度，会压缩，最短压缩长度为5 compress_length_limit = 4 #最多压缩份数，超过该数值的压缩上下文会被删除