{ "package": "qwen3_5_4b_natural_memory_v2_1", "label": "NM2.1 合体", "tag": "nm2_1_fix_C", "built_at": "2026-09-12T23:17:34", "per_category": 10, "overlap_cases": 48, "results": { "C_same_shape_and_unknown": [ "", "| blend | 用例 | 可回答 | 回答正确率 | 答成别的属性 | 未知泄漏率 | 平均选中记录 | 活跃记录 min/max |", "|---|---:|---:|---:|---:|---:|---:|---|", "| prior=1.00 blend=0.50 | 48 | 40 | 70.00% | 27.50% | 0.00% | 5.2083 | 23/24 |", "", "## 检索最高分分布(原始分,非百分比;用于判断阈值能否分开『未知』与『可回答』)", "", "| 组 | n | min | p10 | p50 | p90 | max |", "|---|---:|---:|---:|---:|---:|---:|", "| prior=1.00 blend=0.50 · 可回答 | 40 | 0.84 | 1.19 | 2.73 | 8.52 | 8.52 |", "| prior=1.00 blend=0.50 · 不可回答 | 8 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 |", "", { "eval_file": "data/zero_overlap/eval.jsonl", "episodes": 48, "candidate_count": 24, "router": "checkpoints\\router_replay_v7_v2_128\\memory_router_v2.pt", "results": { "prior=1.00 blend=0.50": { "router": "prior=1.00 blend=0.50", "cases": 48, "answerable_cases": 40, "unknown_cases": 8, "accuracy_pct": 70.0, "wrong_attribute_pct": 27.5, "unknown_leak_pct": 0.0, "mean_records_selected": 5.208333333333333, "records_active_min": 23, "records_active_max": 24, "stop_reasons": { "evidence_found": 40, "attribute_not_covered": 8 }, "top_score_answerable": { "n": 40, "min": 0.84, "p10": 1.19, "p50": 2.73, "p90": 8.52, "max": 8.52 }, "top_score_unknown": { "n": 8, "min": 0.0, "p10": 0.0, "p50": 0.0, "p90": 0.0, "max": 0.0 }, "seconds": 288.5 } }, "rows": { "prior=1.00 blend=0.50": [ { "id": "zero-overlap-000259", "attribute": "出生城市", "answerable": true, "expected": "VAL-M9VSJAEX", "reply": "您的出生城市是 VAL-M9VSJAEX。", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 2.313681895152214, "need_memory": true }, { "id": "zero-overlap-000219", "attribute": "咖啡口味", "answerable": true, "expected": "VAL-M3W7MRJR", "reply": "关于您早上那杯咖啡的豆子偏好,当前记忆中没有相关记录。已知信息仅包含:user;午餐偏好;VAL-Y4Z5H2FG。", "correct": false, "wrong_attribute": true, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 1, "top_score": 8.515033109300843, "need_memory": true }, { "id": "zero-overlap-000193", "attribute": "午餐偏好", "answerable": true, "expected": "VAL-HG84H5TD", "reply": "您的午餐偏好是 user;午餐偏好;VAL-HG84H5TD。", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 0.953831069469452, "need_memory": true }, { "id": "zero-overlap-000080", "attribute": "入职年份", "answerable": true, "expected": "VAL-GFT4HRX2", "reply": "关于您何时开始在这家公司上班,当前记忆中没有相关记录。已知信息:user;办公城市;VAL-TCNMN9HW。", "correct": false, "wrong_attribute": true, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 1, "top_score": 8.51429933919514, "need_memory": true }, { "id": "zero-overlap-000285", "attribute": "起床时间", "answerable": true, "expected": "VAL-4K3RQQKF", "reply": "我的闹钟设定时间是 VAL-4K3RQQKF。", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 3.2757300050727958, "need_memory": true }, { "id": "zero-overlap-000246", "attribute": "常住城市", "answerable": false, "expected": "VAL-6BWJYDE4", "reply": "作为一个人工智能助手,我并没有“小时候”的概念,也无法知道您具体待过哪里。\n\n不过,如果您是想询问**您自己**小时候待的地方,或者想聊聊关于某个特定地点的童年回忆,我很乐意", "correct": false, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 23, "stop_reason": "attribute_not_covered", "records_selected": 0, "top_score": 0.0, "need_memory": false }, { "id": "zero-overlap-000204", "attribute": "阅读工具", "answerable": true, "expected": "VAL-ZRLSPU5V", "reply": "我翻书时依赖的长期记忆中没有关于“翻书”习惯的记录,因此无法回答您依赖什么。", "correct": false, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 2.989592991362775, "need_memory": true }, { "id": "zero-overlap-000078", "attribute": "常住城市", "answerable": true, "expected": "VAL-TQBY65U8", "reply": "您的常住城市是 VAL-TQBY65U8。", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 1.4848385143280027, "need_memory": true }, { "id": "zero-overlap-000104", "attribute": "午餐偏好", "answerable": true, "expected": "VAL-SRFC842D", "reply": "您的午餐偏好是 user;午餐偏好;VAL-SRFC842D。", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 1.1867529082298276, "need_memory": true }, { "id": "zero-overlap-000281", "attribute": "咖啡口味", "answerable": true, "expected": "VAL-3GP4STZX", "reply": "关于您早上饮用的咖啡偏好哪种豆子,当前记忆中没有相关记录。已知信息仅包含:user;午餐偏好;VAL-XNN7T8EK。", "correct": false, "wrong_attribute": true, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 1, "top_score": 8.515033109300843, "need_memory": true }, { "id": "zero-overlap-000106", "attribute": "咖啡口味", "answerable": false, "expected": "VAL-837DYYWV", "reply": "作为一个人工智能助手,我**无法直接访问您的设备、查看您的闹钟设置或获取您的实时时间信息**。\n\n要找到您的闹钟设定时间,您可以尝试以下方法:\n\n1. **查看手机/电脑屏幕", "correct": false, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 23, "stop_reason": "attribute_not_covered", "records_selected": 0, "top_score": 0.0, "need_memory": false }, { "id": "zero-overlap-000091", "attribute": "运动习惯", "answerable": true, "expected": "VAL-92MQHEDL", "reply": "您的运动习惯是 VAL-92MQHEDL。", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 0.8434395718574524, "need_memory": true }, { "id": "zero-overlap-000088", "attribute": "设备型号", "answerable": true, "expected": "VAL-JY6S7EYF", "reply": "关于您使用的机器档次,当前长期记忆中未包含此信息,无法回答。已知信息仅涉及:user;常用编辑器;VAL-3KPKRM4J。", "correct": false, "wrong_attribute": true, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 1, "top_score": 8.516791512356487, "need_memory": true }, { "id": "zero-overlap-000244", "attribute": "午餐偏好", "answerable": false, "expected": "VAL-2VLHC26V", "reply": "您好!作为一个人工智能助手,我并没有直接服务于特定某一家公司或机构。我的设计目标是**为所有用户**提供通用、安全、有帮助的信息和服务。\n\n不过,如果您是在某个具体平台(如企业微信", "correct": false, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 23, "stop_reason": "attribute_not_covered", "records_selected": 0, "top_score": 0.0, "need_memory": false }, { "id": "zero-overlap-000282", "attribute": "咖啡口味", "answerable": true, "expected": "VAL-Q2AQQMUP", "reply": "关于您早上那杯咖啡的豆子偏好,当前记忆中没有相关记录。已知信息仅包含:user;午餐偏好;VAL-5ATEHVSM。", "correct": false, "wrong_attribute": true, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 1, "top_score": 8.515033109300843, "need_memory": true }, { "id": "zero-overlap-000056", "attribute": "主管姓名", "answerable": true, "expected": "VAL-VDL8H7Y9", "reply": "您的主管姓名是 VAL-VDL8H7Y9。", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 2.049178622612938, "need_memory": true }, { "id": "zero-overlap-000184", "attribute": "项目代号", "answerable": true, "expected": "VAL-3H5TQEC4", "reply": "您的项目代号是 VAL-3H5TQEC4。", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 3.4711107313126073, "need_memory": true }, { "id": "zero-overlap-000198", "attribute": "手机尾号", "answerable": true, "expected": "VAL-7T3XVMVH", "reply": "您随身那台设备的末端数字是 4N,该设备的完整型号为 VAL-9MWPDS4N。", "correct": false, "wrong_attribute": true, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 1, "top_score": 8.513663491707547, "need_memory": true }, { "id": "zero-overlap-000258", "attribute": "阅读工具", "answerable": true, "expected": "VAL-SHRY3GNG", "reply": "我翻书时依赖的实体是 user,属性是阅读工具,已确认值为 VAL-SHRY3GNG。", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 2.9794869642364628, "need_memory": true }, { "id": "zero-overlap-000030", "attribute": "课程名称", "answerable": true, "expected": "VAL-BQY3K5EV", "reply": "您的课程名称是 VAL-BQY3K5EV。", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 1.1944098997116086, "need_memory": true }, { "id": "zero-overlap-000196", "attribute": "手机尾号", "answerable": true, "expected": "VAL-ZHN2NMZV", "reply": "您随身那台设备的末端数字是 W9,该设备的完整型号为 VAL-5VBJ22W9。", "correct": false, "wrong_attribute": true, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 1, "top_score": 8.513663491707547, "need_memory": true }, { "id": "zero-overlap-000268", "attribute": "宿舍楼号", "answerable": true, "expected": "VAL-QVYMWTLK", "reply": "关于您睡觉的地儿是第几排,当前长期记忆中未包含此信息。已知信息如下:user;宿舍楼号;VAL-QVYMWTLK;常住城市;VAL-NW2E2ZWE;", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 2.7256882035478824, "need_memory": true }, { "id": "zero-overlap-000006", "attribute": "档案标识", "answerable": true, "expected": "VAL-8EFN3KGR", "reply": "您的档案标识是 VAL-8EFN3KGR。user;档案标识;VAL-8EFN3KGR", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 3.7116483740847537, "need_memory": true }, { "id": "zero-overlap-000047", "attribute": "咖啡口味", "answerable": true, "expected": "VAL-PW46K9K9", "reply": "关于您早上饮用的咖啡偏好哪种豆子,当前记忆中没有相关记录。已知信息仅包含您的午餐偏好是 VAL-BFW4LVTV。", "correct": false, "wrong_attribute": true, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 1, "top_score": 8.515033109300843, "need_memory": true }, { "id": "zero-overlap-000173", "attribute": "客户名称", "answerable": false, "expected": "VAL-P84HTKZR", "reply": "作为一个人工智能助手,我**无法知道您随身设备的型号或数字**,因为我没有访问您个人设备信息的权限,也无法看到您正在使用的具体设备。\n\n如果您指的是:\n- **手机/电脑型号**:", "correct": false, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 23, "stop_reason": "attribute_not_covered", "records_selected": 0, "top_score": 0.0, "need_memory": false }, { "id": "zero-overlap-000000", "attribute": "出生城市", "answerable": true, "expected": "VAL-PHB8FNF9", "reply": "您的出生城市是 VAL-PHB8FNF9。", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 2.455323672186914, "need_memory": true }, { "id": "zero-overlap-000266", "attribute": "出生城市", "answerable": true, "expected": "VAL-K7FT5XFB", "reply": "您的出生城市是 VAL-K7FT5XFB。", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 2.1730912274276046, "need_memory": true }, { "id": "zero-overlap-000051", "attribute": "项目代号", "answerable": false, "expected": "VAL-KG4RZYH5", "reply": "作为一个人工智能助手,我并没有实体,也没有“待得最久”的物理空间或地理位置。我的“存在”体现在每一次与您的对话中。\n\n不过,如果您是在寻找**您自己**待得最久的地方", "correct": false, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 23, "stop_reason": "attribute_not_covered", "records_selected": 0, "top_score": 0.0, "need_memory": false }, { "id": "zero-overlap-000061", "attribute": "常住城市", "answerable": true, "expected": "VAL-3SUNPNRD", "reply": "您的常住城市是 VAL-3SUNPNRD,办公城市是 VAL-XYRY8BUQ,出生城市是 VAL-GE828D22,通勤方式是 VAL-2CXTE3TB,入职年份", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 1.6610547590255735, "need_memory": true }, { "id": "zero-overlap-000214", "attribute": "客户名称", "answerable": true, "expected": "VAL-MV382EEF", "reply": "您的服务对象是 VAL-MV382EEF。", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 1.2588214325904845, "need_memory": true }, { "id": "zero-overlap-000092", "attribute": "宿舍楼号", "answerable": true, "expected": "VAL-PLZZ659B", "reply": "关于您睡觉的地儿是第几排,当前长期记忆中未包含此信息。已知信息如下:user;宿舍楼号;VAL-PLZZ659B。", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 2.809071942542481, "need_memory": true }, { "id": "zero-overlap-000140", "attribute": "宿舍楼号", "answerable": true, "expected": "VAL-RMDSB297", "reply": "关于您睡觉的地儿是第几排,当前长期记忆中未包含此信息。已知信息如下:user;宿舍楼号;VAL-RMDSB297。", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 2.8702512824281925, "need_memory": true }, { "id": "zero-overlap-000164", "attribute": "客户名称", "answerable": true, "expected": "VAL-BB55YHW5", "reply": "您的客户名称是 VAL-BB55YHW5。", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 1.2650055932998656, "need_memory": true }, { "id": "zero-overlap-000213", "attribute": "通勤方式", "answerable": true, "expected": "VAL-M58DDQX2", "reply": "您的通勤方式是 VAL-M58DDQX2。", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 2.0363681795533712, "need_memory": true }, { "id": "zero-overlap-000122", "attribute": "通勤方式", "answerable": true, "expected": "VAL-VULSNGJ3", "reply": "您的通勤方式是 VAL-VULSNGJ3。user;通勤方式;VAL-VULSNGJ3", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 2.0330638098057325, "need_memory": true }, { "id": "zero-overlap-000168", "attribute": "常住城市", "answerable": true, "expected": "VAL-8TQ8737C", "reply": "您的常住城市是 VAL-8TQ8737C。", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 1.4606248426437376, "need_memory": true }, { "id": "zero-overlap-000175", "attribute": "客户名称", "answerable": true, "expected": "VAL-QJNRRE8X", "reply": "您的客户名称是 VAL-QJNRRE8X。", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 1.2628431367874144, "need_memory": true }, { "id": "zero-overlap-000144", "attribute": "入职年份", "answerable": false, "expected": "VAL-GLHQMRN6", "reply": "作为一个人工智能助手,我**无法知道您睡觉的具体位置**,因为我无法感知现实世界中的物理环境,也无法获取您的个人生活信息。\n\n如果您是在询问某个特定场所(如学校教室、电影院、酒店或", "correct": false, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 23, "stop_reason": "attribute_not_covered", "records_selected": 0, "top_score": 0.0, "need_memory": false }, { "id": "zero-overlap-000002", "attribute": "午餐偏好", "answerable": true, "expected": "VAL-HEE2QMH2", "reply": "您的午餐偏好是 user;午餐偏好;VAL-HEE2QMH2。", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 0.9400680589675904, "need_memory": true }, { "id": "zero-overlap-000171", "attribute": "常住城市", "answerable": true, "expected": "VAL-KZA246K9", "reply": "您的常住城市是 VAL-KZA246K9,办公城市是 VAL-Z8QBE5HQ,出生城市是 VAL-N8SWTE6N,宿舍楼号是 VAL-XZY4WPYH,通勤", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 1.49963095664978, "need_memory": true }, { "id": "zero-overlap-000154", "attribute": "入职年份", "answerable": true, "expected": "VAL-7Z9CBUX8", "reply": "关于您何时开始在这家公司上班,当前记忆中没有相关记录。已知信息:user;办公城市;VAL-BP7253WL。", "correct": false, "wrong_attribute": true, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 1, "top_score": 8.51429933919514, "need_memory": true }, { "id": "zero-overlap-000234", "attribute": "工位楼层", "answerable": true, "expected": "VAL-87M7HVC7", "reply": "您的工位楼层是 VAL-87M7HVC7。", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 3.1934859675651266, "need_memory": true }, { "id": "zero-overlap-000083", "attribute": "办公城市", "answerable": true, "expected": "VAL-YBVYV97X", "reply": "您的办公城市是 VAL-YBVYV97X。", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 2.3717270611639645, "need_memory": true }, { "id": "zero-overlap-000079", "attribute": "出生城市", "answerable": true, "expected": "VAL-7R8WVAFG", "reply": "您的出生城市是 VAL-7R8WVAFG。", "correct": true, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 2.4271069115554122, "need_memory": true }, { "id": "zero-overlap-000180", "attribute": "客户名称", "answerable": false, "expected": "VAL-3F87H6UT", "reply": "作为一个人工智能助手,我**无法知道您随身设备的型号或数字**,因为我没有访问您个人设备信息的权限,也无法看到您正在使用的具体设备。\n\n如果您指的是:\n- **手机/电脑型号**:", "correct": false, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 23, "stop_reason": "attribute_not_covered", "records_selected": 0, "top_score": 0.0, "need_memory": false }, { "id": "zero-overlap-000025", "attribute": "紧急联系人姓氏", "answerable": true, "expected": "VAL-STECQ3PX", "reply": "当遇到突发问题或需要联系支持时,您可以直接联系您的专属支持团队。具体联系渠道包括:\n- 默认语言:VAL-PRVTUCP7\n- 出生城市:VAL-KYBRBBKV", "correct": false, "wrong_attribute": true, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 8, "top_score": 8.516015553315961, "need_memory": true }, { "id": "zero-overlap-000187", "attribute": "手机尾号", "answerable": true, "expected": "VAL-PXRHJ49F", "reply": "您随身那台设备的末端数字是 H9,该设备的完整型号为 VAL-ZPUPV9H9。", "correct": false, "wrong_attribute": true, "leaked": false, "records_written": 24, "records_active": 24, "stop_reason": "evidence_found", "records_selected": 1, "top_score": 8.513663491707547, "need_memory": true }, { "id": "zero-overlap-000045", "attribute": "办公城市", "answerable": false, "expected": "VAL-CG4X6ZWD", "reply": "作为一个人工智能助手,我并没有实体,也没有“待得最久”的物理空间或地理位置。我的“存在”体现在每一次与您的对话中。\n\n不过,如果您是在寻找**您自己**待得最久的地方", "correct": false, "wrong_attribute": false, "leaked": false, "records_written": 24, "records_active": 23, "stop_reason": "attribute_not_covered", "records_selected": 0, "top_score": 0.0, "need_memory": false } ] } } ] } }