Files
natural-memory-nm21/nm2_1_gate_battery_summary.json
T
WpyQwq 643e22ecb9 Natural Memory NM2.1: 记忆路由器分叉、数据集缺陷修复与全轴评测证据
- 引入 MemoryRouterXL 与 v5/v6 流式多线程训练/编码管线
- 修复 prepare_memory_router_dataset 候选池重建缺陷(mega 家族 3568x 加速,输出逐字节相同)
- 修复 v5 被破坏的拒答与多跳标签(train 未知样本 319 -> 16319,multi_hop 平均正例 1.00 -> 2.00)
- 同存储预算下 V2-128 v6 逐轴 22/22 通过:Top-1 41.12% -> 94.62%,未知拒答 0.00% -> 100.00%
- 记录三条被实测推翻的显然优化(logits_to_keep=1 反而慢 55%、XL 容量未带来收益)
- 记忆手术跨架构可移植性 14/14,读写关闭时与原生模型逐位相同
2026-09-19 11:11:31 +08:00

2429 lines
111 KiB
JSON
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
{
"package": "qwen3_5_4b_natural_memory_v2_1",
"label": "NM2.1+gate",
"tag": "nm2_1_gate",
"built_at": "2026-09-12T21:35:14",
"per_category": 10,
"overlap_cases": 48,
"results": {
"A_end_to_end": [
"| NM2.1+gate | 110 | 16.36% | 6.25% | 43.33% | 7.50% | 437s |",
"wrote nm2_1_gate_e2e.md",
"wrote nm2_1_gate_e2e.json",
{
"NM2.1+gate": {
"summary": {
"router": "NM2.1+gate",
"cases": 110,
"accuracy_pct": 16.363636363636363,
"answerable_cases": 80,
"answerable_accuracy_pct": 6.25,
"unknown_cases": 30,
"unknown_refusal_pct": 43.333333333333336,
"wrong_abstention_pct": 7.5,
"seconds": 437.1,
"per_category": {
"(policy)": {
"cases": 10,
"accuracy_pct": 0.0,
"answerable": 0,
"answerable_accuracy_pct": 0.0,
"unknown": 10,
"unknown_refusal_pct": 0.0
},
"conflict_update": {
"cases": 10,
"accuracy_pct": 0.0,
"answerable": 10,
"answerable_accuracy_pct": 0.0,
"unknown": 0,
"unknown_refusal_pct": 0.0
},
"distractor_128": {
"cases": 10,
"accuracy_pct": 0.0,
"answerable": 10,
"answerable_accuracy_pct": 0.0,
"unknown": 0,
"unknown_refusal_pct": 0.0
},
"distractor_32": {
"cases": 10,
"accuracy_pct": 0.0,
"answerable": 10,
"answerable_accuracy_pct": 0.0,
"unknown": 0,
"unknown_refusal_pct": 0.0
},
"forget_correction": {
"cases": 10,
"accuracy_pct": 30.0,
"answerable": 0,
"answerable_accuracy_pct": 0.0,
"unknown": 10,
"unknown_refusal_pct": 30.0
},
"long_context": {
"cases": 10,
"accuracy_pct": 0.0,
"answerable": 10,
"answerable_accuracy_pct": 0.0,
"unknown": 0,
"unknown_refusal_pct": 0.0
},
"multi_hop": {
"cases": 10,
"accuracy_pct": 0.0,
"answerable": 10,
"answerable_accuracy_pct": 0.0,
"unknown": 0,
"unknown_refusal_pct": 0.0
},
"paraphrase": {
"cases": 10,
"accuracy_pct": 50.0,
"answerable": 10,
"answerable_accuracy_pct": 50.0,
"unknown": 0,
"unknown_refusal_pct": 0.0
},
"random_position": {
"cases": 10,
"accuracy_pct": 0.0,
"answerable": 10,
"answerable_accuracy_pct": 0.0,
"unknown": 0,
"unknown_refusal_pct": 0.0
},
"single_fact": {
"cases": 10,
"accuracy_pct": 0.0,
"answerable": 10,
"answerable_accuracy_pct": 0.0,
"unknown": 0,
"unknown_refusal_pct": 0.0
},
"unknown_abstention": {
"cases": 10,
"accuracy_pct": 100.0,
"answerable": 0,
"answerable_accuracy_pct": 0.0,
"unknown": 10,
"unknown_refusal_pct": 100.0
}
}
},
"rows": [
{
"category": "(policy)",
"query": "请问对象测试实体0000的驻地代号为?",
"reply": "关于对象测试实体 0000 的驻地代号,现有证据不足,无法提供确切信息。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "(policy)",
"query": "查询:测试实体0001的负责人代号是什么?",
"reply": "关于测试实体 0001 的负责人代号,当前长期记忆中无相关记录,无法提供确切答案。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "(policy)",
"query": "查询:测试实体0002的维护日代号是什么?",
"reply": "关于测试实体 0002 的维护日代号,当前记忆库中无相关记录,无法提供确切答案。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "(policy)",
"query": "请问对象测试实体0003的安全级别代号为?",
"reply": "关于对象测试实体 0003 的安全级别代号,当前记忆库中无相关记录,无法提供确切答案。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "(policy)",
"query": "请问对象测试实体0004的档案类别代号为?",
"reply": "关于对象测试实体 0004 的档案类别代号,当前记忆库中无相关记录,无法提供确切答案。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "(policy)",
"query": "查询:测试实体0005的驻地代号是什么?",
"reply": "关于测试实体 0005 的驻地代号,当前长期记忆中无相关记录,无法提供确切答案。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "(policy)",
"query": "查询:测试实体0006的负责人代号是什么?",
"reply": "测试实体 0006 的负责人代号未知,当前长期记忆中仅包含项目项目-P-DRG2XUA1 的负责人是成员-M-W5075MN3 这一事实,未涉及测试",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "(policy)",
"query": "根据已记信息,测试实体0007的维护日代号是?",
"reply": "根据已记信息,测试实体 0007 的维护日代号是 N-7HEQZ5KK。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "(policy)",
"query": "请问对象测试实体0008的安全级别代号为?",
"reply": "关于对象测试实体 0008 的安全级别代号,当前记忆库中无相关记录,无法提供确切答案。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "(policy)",
"query": "查询:测试实体0009的档案类别代号是什么?",
"reply": "关于测试实体 0009 的档案类别代号,当前记忆库中无相关记录,无法提供确切答案。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "conflict_update",
"query": "我的提醒时间最新值是什么?",
"reply": "作为一个人工智能助手,我**无法访问您的设备、系统或任何外部数据**,因此不知道您具体的“提醒时间”设置是什么。\n\n这通常取决于您正在使用的设备或应用(例如:手机日历、电脑",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": true
},
{
"category": "conflict_update",
"query": "我的备用联系人最新值是什么?",
"reply": "作为一个人工智能助手,我**无法访问您的个人设备、手机通讯录或任何本地数据**,因此无法得知您设备中“备用联系人”的具体数值或名称。\n\n这个信息完全取决于您使用的设备(如 iPhone",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "conflict_update",
"query": "我的备用联系人最新值是什么?",
"reply": "作为一个人工智能助手,我**无法访问您的个人设备、手机通讯录或任何本地数据**,因此无法得知您设备中“备用联系人”的具体数值或名称。\n\n这个信息完全取决于您使用的设备(如 iPhone",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "conflict_update",
"query": "我的最喜欢的水果最新值是什么?",
"reply": "作为一个人工智能助手,我并没有个人喜好,因此也没有“最喜欢的水果”。\n\n不过,如果您是想询问**最新上市**或**近期热门**的水果,目前市场上比较受欢迎的包括:\n* **车厘子**",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "conflict_update",
"query": "我的常用语言最新值是什么?",
"reply": "您好!作为 AI 助手,我并没有直接访问您个人设备或历史数据的能力,因此无法得知您的“常用语言最新值”。不过,如果您是在询问我与您对话时使用的语言,目前我们是用**中文**进行交流",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "conflict_update",
"query": "我的最喜欢的水果最新值是什么?",
"reply": "作为一个人工智能助手,我并没有个人喜好,因此也没有“最喜欢的水果”。\n\n不过,如果您是想询问**最新上市**或**近期热门**的水果,目前市场上比较受欢迎的包括:\n* **车厘子**",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "conflict_update",
"query": "我的默认输出风格最新值是什么?",
"reply": "作为 Qwen3.5,我没有直接访问您本地设置或历史对话记录的权限,因此无法得知您的“默认输出风格”具体是什么。不过,我可以帮您调整或设定输出风格!例如:\n\n- 如果您希望",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "conflict_update",
"query": "我的提醒时间最新值是什么?",
"reply": "作为一个人工智能助手,我**无法访问您的设备、系统或任何外部数据**,因此不知道您具体的“提醒时间”设置是什么。\n\n这通常取决于您正在使用的设备或应用(例如:手机日历、电脑",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": true
},
{
"category": "conflict_update",
"query": "我的备用联系人最新值是什么?",
"reply": "作为一个人工智能助手,我**无法访问您的个人设备、手机通讯录或任何本地数据**,因此无法得知您设备中“备用联系人”的具体数值或名称。\n\n这个信息完全取决于您使用的设备(如 iPhone",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "conflict_update",
"query": "我的常用语言最新值是什么?",
"reply": "您好!作为 AI 助手,我并没有直接访问您个人设备或历史数据的能力,因此无法得知您的“常用语言最新值”。不过,如果您是在询问我与您对话时使用的语言,目前我们是用**中文**进行交流",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "distractor_128",
"query": "在大量无关信息中,只读取我的常用语言,它的值是什么?",
"reply": "您的常用语言是**中文**。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "distractor_128",
"query": "在大量无关信息中,只读取我的最喜欢的水果,它的值是什么?",
"reply": "您最喜欢的水果是什么?",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "distractor_128",
"query": "在大量无关信息中,只读取我的工作地点,它的值是什么?",
"reply": "您提供的消息中**没有包含任何关于工作地点的信息**。\n\n如果您有具体的文本或上下文想要我从中提取工作地点,请提供该内容,我会立即为您提取。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "distractor_128",
"query": "在大量无关信息中,只读取我的常用时区,它的值是什么?",
"reply": "在大量无关信息中,只读取您的常用时区,它的值是 **UTC+8**(中国标准时间)。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "distractor_128",
"query": "在大量无关信息中,只读取我的常用时区,它的值是什么?",
"reply": "在大量无关信息中,只读取您的常用时区,它的值是 **UTC+8**(中国标准时间)。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "distractor_128",
"query": "在大量无关信息中,只读取我的默认输出风格,它的值是什么?",
"reply": "您的默认输出风格是**专业、简洁、直接且乐于助人**。\n\n作为一个人工智能助手,我的核心设计原则是:\n1. **准确**:提供基于事实或逻辑的回答。\n2. **清晰",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "distractor_128",
"query": "在大量无关信息中,只读取我的默认输出风格,它的值是什么?",
"reply": "您的默认输出风格是**专业、简洁、直接且乐于助人**。\n\n作为一个人工智能助手,我的核心设计原则是:\n1. **准确**:提供基于事实或逻辑的回答。\n2. **清晰",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "distractor_128",
"query": "在大量无关信息中,只读取我的项目代号,它的值是什么?",
"reply": "您的项目代号是:**Project Phoenix**(凤凰计划)。\n\n> 注:由于您未提供具体的“大量无关信息”文本,此回答基于常见代号命名习惯及上下文逻辑推断。若您有具体文本内容,请",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "distractor_128",
"query": "在大量无关信息中,只读取我的备用联系人,它的值是什么?",
"reply": "作为一个人工智能助手,我**无法访问您的个人设备、手机通讯录、系统设置或任何私人数据**。因此,我无法读取您的“备用联系人”列表,也无法告诉您该值具体是什么。\n\n这涉及您的隐私",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "distractor_128",
"query": "在大量无关信息中,只读取我的工作地点,它的值是什么?",
"reply": "您提供的消息中**没有包含任何关于工作地点的信息**。\n\n如果您有具体的文本或上下文想要我从中提取工作地点,请提供该内容,我会立即为您提取。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "distractor_32",
"query": "在大量无关信息中,只读取我的工作地点,它的值是什么?",
"reply": "您提供的消息中**没有包含任何关于工作地点的信息**。\n\n如果您有具体的文本或上下文想要我从中提取工作地点,请提供该内容,我会立即为您提取。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "distractor_32",
"query": "在大量无关信息中,只读取我的备用联系人,它的值是什么?",
"reply": "作为一个人工智能助手,我**无法访问您的个人设备、手机通讯录、系统设置或任何私人数据**。因此,我无法读取您的“备用联系人”列表,也无法告诉您该值具体是什么。\n\n这涉及您的隐私",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "distractor_32",
"query": "在大量无关信息中,只读取我的备用联系人,它的值是什么?",
"reply": "作为一个人工智能助手,我**无法访问您的个人设备、手机通讯录、系统设置或任何私人数据**。因此,我无法读取您的“备用联系人”列表,也无法告诉您该值具体是什么。\n\n这涉及您的隐私",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "distractor_32",
"query": "在大量无关信息中,只读取我的常用时区,它的值是什么?",
"reply": "在大量无关信息中,只读取您的常用时区,它的值是 **UTC+8**(中国标准时间)。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "distractor_32",
"query": "在大量无关信息中,只读取我的备用联系人,它的值是什么?",
"reply": "作为一个人工智能助手,我**无法访问您的个人设备、手机通讯录、系统设置或任何私人数据**。因此,我无法读取您的“备用联系人”列表,也无法告诉您该值具体是什么。\n\n这涉及您的隐私",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "distractor_32",
"query": "在大量无关信息中,只读取我的常用语言,它的值是什么?",
"reply": "您的常用语言是**中文**。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "distractor_32",
"query": "在大量无关信息中,只读取我的常用语言,它的值是什么?",
"reply": "您的常用语言是**中文**。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "distractor_32",
"query": "在大量无关信息中,只读取我的工作地点,它的值是什么?",
"reply": "您提供的消息中**没有包含任何关于工作地点的信息**。\n\n如果您有具体的文本或上下文想要我从中提取工作地点,请提供该内容,我会立即为您提取。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "distractor_32",
"query": "在大量无关信息中,只读取我的常用时区,它的值是什么?",
"reply": "在大量无关信息中,只读取您的常用时区,它的值是 **UTC+8**(中国标准时间)。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "distractor_32",
"query": "在大量无关信息中,只读取我的默认输出风格,它的值是什么?",
"reply": "您的默认输出风格是**专业、简洁、直接且乐于助人**。\n\n作为一个人工智能助手,我的核心设计原则是:\n1. **准确**:提供基于事实或逻辑的回答。\n2. **清晰",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "forget_correction",
"query": "我的项目代号是什么?如果已经删除,请回答没有记录。",
"reply": "作为一个人工智能助手,我**没有访问权限**来查询或记录您的个人项目信息。我的设计原则是保护用户隐私,不会存储或追踪任何用户的特定项目代号。\n\n如果您之前提到过某个项目代号,",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "forget_correction",
"query": "我的常用时区是什么?如果已经删除,请回答没有记录。",
"reply": "作为一个人工智能助手,我**没有访问您个人设备设置、浏览器历史或系统记录的权限**,因此无法得知您的“常用时区”是什么。\n\n如果您之前设置过某个时区,但现在已经删除或从未",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "forget_correction",
"query": "我的备用联系人是什么?如果已经删除,请回答没有记录。",
"reply": "作为一个人工智能助手,我**无法访问您的手机、电脑或任何设备的通讯录数据**,因此无法查询您的“备用联系人”具体是谁。\n\n如果您是在询问自己设备上的设置,请尝试以下方法:\n1",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "forget_correction",
"query": "我的工作地点是什么?如果已经删除,请回答没有记录。",
"reply": "作为一个人工智能助手,我**没有访问权限**来查看您的个人数据、工作信息或任何已删除的记录。我的设计原则是保护用户隐私,不会存储或检索用户的个人信息。\n\n如果您之前设置过某个特定的",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "forget_correction",
"query": "我的提醒时间是什么?如果已经删除,请回答没有记录。",
"reply": "没有记录。",
"written": 6,
"correct": true,
"matched": [],
"abstained": true,
"wrongly_abstained": false
},
{
"category": "forget_correction",
"query": "我的默认输出风格是什么?如果已经删除,请回答没有记录。",
"reply": "没有记录。",
"written": 6,
"correct": true,
"matched": [],
"abstained": true,
"wrongly_abstained": false
},
{
"category": "forget_correction",
"query": "我的工作地点是什么?如果已经删除,请回答没有记录。",
"reply": "作为一个人工智能助手,我**没有访问权限**来查看您的个人数据、工作信息或任何已删除的记录。我的设计原则是保护用户隐私,不会存储或检索用户的个人信息。\n\n如果您之前设置过某个特定的",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "forget_correction",
"query": "我的提醒时间是什么?如果已经删除,请回答没有记录。",
"reply": "没有记录。",
"written": 6,
"correct": true,
"matched": [],
"abstained": true,
"wrongly_abstained": false
},
{
"category": "forget_correction",
"query": "我的常用语言是什么?如果已经删除,请回答没有记录。",
"reply": "您好!作为一个人工智能助手,我**没有访问或存储您个人设备、浏览器或账户中“常用语言”设置的能力**。\n\n如果您是在询问:\n1. **当前对话使用的语言**:我们目前使用的是**",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "forget_correction",
"query": "我的常用时区是什么?如果已经删除,请回答没有记录。",
"reply": "作为一个人工智能助手,我**没有访问您个人设备设置、浏览器历史或系统记录的权限**,因此无法得知您的“常用时区”是什么。\n\n如果您之前设置过某个时区,但现在已经删除或从未",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "long_context",
"query": "在长上下文压缩之后,检索我的项目代号并回答。",
"reply": "您好!由于我无法访问您之前的对话历史或上下文(每次对话对我来说都是全新的开始),目前我**不知道您的项目代号是什么**。\n\n为了帮您准确回答,请提供以下信息:\n1. **项目代号**(例如",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": true
},
{
"category": "long_context",
"query": "在长上下文压缩之后,检索我的常用时区并回答。",
"reply": "您好!由于我无法访问您的个人数据或历史记录,因此无法直接检索您的常用时区。如果您能提供以下信息,我可以帮您更准确地回答:\n\n1. 您常用的时区名称(如“北京时间”、“美国东部",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "long_context",
"query": "在长上下文压缩之后,检索我的工作地点并回答。",
"reply": "您好!由于我无法访问您的个人数据或之前的对话历史(长上下文压缩后,具体信息可能已丢失),请告诉我您的工作地点,我将立即为您查找相关信息并回答您的问题。例如: \n- 如果您在**北京**",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "long_context",
"query": "在长上下文压缩之后,检索我的提醒时间并回答。",
"reply": "您好!由于我无法直接访问您的长上下文压缩后的内容(例如压缩后的摘要、笔记或文档),请您提供以下信息之一:\n\n1. **提醒时间**(例如:“明天上午 10 点”或“",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "long_context",
"query": "在长上下文压缩之后,检索我的提醒时间并回答。",
"reply": "您好!由于我无法直接访问您的长上下文压缩后的内容(例如压缩后的摘要、笔记或文档),请您提供以下信息之一:\n\n1. **提醒时间**(例如:“明天上午 10 点”或“",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "long_context",
"query": "在长上下文压缩之后,检索我的项目代号并回答。",
"reply": "您好!由于我无法访问您之前的对话历史或上下文(每次对话对我来说都是全新的开始),目前我**不知道您的项目代号是什么**。\n\n为了帮您准确回答,请提供以下信息:\n1. **项目代号**(例如",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": true
},
{
"category": "long_context",
"query": "在长上下文压缩之后,检索我的常用语言并回答。",
"reply": "您好!在长上下文压缩后,我已准备好为您服务。请问您希望我使用哪种语言进行交流?(例如:中文、英文、法语等)",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "long_context",
"query": "在长上下文压缩之后,检索我的工作地点并回答。",
"reply": "您好!由于我无法访问您的个人数据或之前的对话历史(长上下文压缩后,具体信息可能已丢失),请告诉我您的工作地点,我将立即为您查找相关信息并回答您的问题。例如: \n- 如果您在**北京**",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "long_context",
"query": "在长上下文压缩之后,检索我的备用联系人并回答。",
"reply": "您好!我注意到您提到了“长上下文压缩”和“检索备用联系人”,但当前对话中**没有提供具体的联系人列表或上下文数据**。为了准确帮助您,请补充以下信息:\n\n1. **需要检索的联系人",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "long_context",
"query": "在长上下文压缩之后,检索我的常用时区并回答。",
"reply": "您好!由于我无法访问您的个人数据或历史记录,因此无法直接检索您的常用时区。如果您能提供以下信息,我可以帮您更准确地回答:\n\n1. 您常用的时区名称(如“北京时间”、“美国东部",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "multi_hop",
"query": "请通过项目负责人关系,找出项目项目-P-KP2F9GHO负责人的工作代号。",
"reply": "无法通过项目负责人关系找出项目-P-KP2F9GHO负责人的工作代号,因为当前提供的长期记忆证据中不包含该项目负责人或相关项目信息的记录。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "multi_hop",
"query": "请通过项目负责人关系,找出项目项目-P-DG818670负责人的工作代号。",
"reply": "无法通过项目负责人关系找出项目-P-DG818670负责人的工作代号,因为当前提供的长期记忆证据中不包含该项目负责人及其工作代号的关联信息。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "multi_hop",
"query": "请通过项目负责人关系,找出项目项目-P-LXAVDTVY负责人的工作代号。",
"reply": "无法通过项目负责人关系找出项目-P-LXAVDTVY负责人的工作代号,因为当前提供的长期记忆证据中不包含该项目负责人及其工作代号的任何信息。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "multi_hop",
"query": "请通过项目负责人关系,找出项目项目-P-F1BVFMQ5负责人的工作代号。",
"reply": "无法通过项目负责人关系找出项目-P-F1BVFMQ5负责人的工作代号,因为当前提供的长期记忆中不包含该项目负责人或相关项目信息的证据。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "multi_hop",
"query": "请通过项目负责人关系,找出项目项目-P-CY5VYJU9负责人的工作代号。",
"reply": "项目项目-P-CY5VYJU9的负责人是成员-M-ISLCPRNN,其工作代号为M-ISLCPRNN。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "multi_hop",
"query": "请通过项目负责人关系,找出项目项目-P-Q1Y92XOA负责人的工作代号。",
"reply": "无法通过项目负责人关系找出项目-P-Q1Y92XOA负责人的工作代号,因为当前提供的长期记忆中不包含该项目负责人或相关项目信息的证据。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "multi_hop",
"query": "请通过项目负责人关系,找出项目项目-P-7Y96TFNY负责人的工作代号。",
"reply": "根据现有长期记忆,无法通过项目负责人关系找到项目-P-7Y96TFNY负责人的工作代号,因为记忆中仅包含用户的项目代号信息,不包含项目与负责人之间的关联数据。已知信息:用户;项目代号;",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "multi_hop",
"query": "请通过项目负责人关系,找出项目项目-P-8SKCFVHH负责人的工作代号。",
"reply": "项目项目-P-8SKCFVHH的负责人是成员-M-3JUKA95S,其工作代号为成员-M-3JUKA95S。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "multi_hop",
"query": "请通过项目负责人关系,找出项目项目-P-5ECU7LJU负责人的工作代号。",
"reply": "项目项目-P-5ECU7LJU的负责人是成员-M-8X6VMANP,其工作代号为成员-M-8X6VMANP。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "multi_hop",
"query": "请通过项目负责人关系,找出项目项目-P-THNH77VT负责人的工作代号。",
"reply": "无法通过项目负责人关系找出项目-P-THNH77VT负责人的工作代号,因为当前提供的长期记忆证据中不包含该项目负责人或相关项目信息的记录。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "paraphrase",
"query": "我之前登记的最喜欢的水果内容是什么?",
"reply": "您好,作为一个人工智能助手,我**无法访问或查看您个人的历史数据、账户信息或之前登记的任何内容**。\n\n我的设计原则是保护用户隐私,因此我不会存储或记忆您个人的偏好记录。如果您之前",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "paraphrase",
"query": "不要猜,回忆一下我在最喜欢的水果上的设置。",
"reply": "我在最喜欢的水果上的设置为 V-I4RYZQXJ。",
"written": 6,
"correct": true,
"matched": [
"V-I4RYZQXJ"
],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "paraphrase",
"query": "关于常用时区,你保存的用户信息是哪一个?",
"reply": "关于常用时区,我保存的用户信息是:user;常用时区偏好/设置;V-XV6N3LW0。",
"written": 6,
"correct": true,
"matched": [
"V-XV6N3LW0"
],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "paraphrase",
"query": "关于工作地点,你保存的用户信息是哪一个?",
"reply": "关于工作地点,我保存的用户信息是:无。当前长期记忆中未包含任何关于用户工作地点的有效记录,仅存在无关编号(如 N-AJBU9TVL 或 N-RJX85JOK)且",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "paraphrase",
"query": "关于常用时区,你保存的用户信息是哪一个?",
"reply": "关于常用时区,我保存的用户信息是:user;常用时区偏好/设置;V-VSBMYZQZ。",
"written": 6,
"correct": true,
"matched": [
"V-VSBMYZQZ"
],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "paraphrase",
"query": "我之前登记的常用语言内容是什么?",
"reply": "您好!作为 AI 助手,我**无法访问或查看您的个人历史数据、设置或之前登记的语言偏好**。我的设计原则是保护用户隐私,不会存储或记录任何用户的个人信息。\n\n如果您想切换语言,可以直接",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "paraphrase",
"query": "我之前登记的项目代号内容是什么?",
"reply": "作为一个人工智能助手,我**无法访问您的个人数据、历史记录或您之前登记的任何项目信息**。我的设计原则是保护用户隐私,因此我不会存储或回忆您个人的具体项目代号。\n\n如果您之前是在某个特定的",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "paraphrase",
"query": "我之前登记的备用联系人内容是什么?",
"reply": "作为一个人工智能助手,我**无法访问您的个人设备、手机通讯录或任何已登记的备用联系人信息**。这类数据属于您的个人隐私,仅存储在您的手机、电脑或相关通讯软件(如微信、QQ、短信服务等",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "paraphrase",
"query": "关于备用联系人,你保存的用户信息是哪一个?",
"reply": "关于备用联系人,我保存的用户信息是 V-HBH1S9LW。",
"written": 6,
"correct": true,
"matched": [
"V-HBH1S9LW"
],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "paraphrase",
"query": "不要猜,回忆一下我在默认输出风格上的设置。",
"reply": "根据长期记忆证据,您在默认输出风格上的设置值为 V-F269K2O1。",
"written": 6,
"correct": true,
"matched": [
"V-F269K2O1"
],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "random_position",
"query": "随机位置事实测试:我的常用语言是什么?",
"reply": "您的常用语言是**中文**。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "random_position",
"query": "随机位置事实测试:我的默认输出风格是什么?",
"reply": "我的默认输出风格是**专业、简洁、准确且乐于助人**。\n\n具体来说,我会:\n1. **直接回答问题**:不添加不必要的寒暄或冗余信息,直奔主题。\n2. **保持客观中立**",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "random_position",
"query": "随机位置事实测试:我的项目代号是什么?",
"reply": "您的项目代号是:**Project Phoenix**(凤凰计划)。\n\n这是一个虚构的代号,用于本次随机位置事实测试。如果您有特定的项目或需要其他类型的测试,请随时告诉我!",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "random_position",
"query": "随机位置事实测试:我的默认输出风格是什么?",
"reply": "我的默认输出风格是**专业、简洁、准确且乐于助人**。\n\n具体来说,我会:\n1. **直接回答问题**:不添加不必要的寒暄或冗余信息,直奔主题。\n2. **保持客观中立**",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "random_position",
"query": "随机位置事实测试:我的备用联系人是什么?",
"reply": "作为一个人工智能助手,我**无法访问您的个人设备、通讯录或任何与您的私人数据相关的信息**。因此,我不知道您的备用联系人是谁。\n\n如果您需要查找或管理备用联系人,建议您:\n1. 打开",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": true
},
{
"category": "random_position",
"query": "随机位置事实测试:我的提醒时间是什么?",
"reply": "抱歉,我无法回答这个问题,因为我没有记录您的提醒时间。作为 AI 助手,我无法访问您的个人设备或应用数据。建议您查看手机或电脑上的提醒设置,或检查您使用的日历/任务管理应用来获取准确信息",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": true
},
{
"category": "random_position",
"query": "随机位置事实测试:我的常用语言是什么?",
"reply": "您的常用语言是**中文**。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "random_position",
"query": "随机位置事实测试:我的常用语言是什么?",
"reply": "您的常用语言是**中文**。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "random_position",
"query": "随机位置事实测试:我的常用语言是什么?",
"reply": "您的常用语言是**中文**。",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "random_position",
"query": "随机位置事实测试:我的常用时区是什么?",
"reply": "作为一个人工智能,我**无法知道您的常用时区**,因为我没有访问您个人设备、网络设置或地理位置数据的权限。\n\n时区信息通常取决于:\n1. 您设备(手机/电脑)的系统",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "single_fact",
"query": "跨对话后请回答:我的默认输出风格是什么?",
"reply": "作为 Qwen3.5,我的默认输出风格是**专业、简洁、准确且富有同理心**的。我会根据对话内容灵活调整语气,确保信息清晰易懂,同时保持逻辑严谨和表达自然。如果您有",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "single_fact",
"query": "跨对话后请回答:我的项目代号是什么?",
"reply": "作为一个人工智能助手,我**没有权限访问您的个人身份信息、项目档案或任何外部数据库**,因此无法得知您的“项目代号”具体是什么。\n\n项目代号通常由您或您的团队内部定义,属于特定",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "single_fact",
"query": "跨对话后请回答:我的默认输出风格是什么?",
"reply": "作为 Qwen3.5,我的默认输出风格是**专业、简洁、准确且富有同理心**的。我会根据对话内容灵活调整语气,确保信息清晰易懂,同时保持逻辑严谨和表达自然。如果您有",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "single_fact",
"query": "跨对话后请回答:我的项目代号是什么?",
"reply": "作为一个人工智能助手,我**没有权限访问您的个人身份信息、项目档案或任何外部数据库**,因此无法得知您的“项目代号”具体是什么。\n\n项目代号通常由您或您的团队内部定义,属于特定",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "single_fact",
"query": "跨对话后请回答:我的常用语言是什么?",
"reply": "您好!作为一个人工智能助手,我并没有直接访问您个人设备或历史记录的能力,因此无法确切知道您“常用”的语言是什么。\n\n不过,从我们当前的对话来看,您使用的是**中文**。如果您希望我用其他",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "single_fact",
"query": "跨对话后请回答:我的最喜欢的水果是什么?",
"reply": "作为一个人工智能助手,我并没有个人偏好或味觉体验,因此无法知道您“最喜欢”的水果是什么。\n\n不过,如果您愿意告诉我,我很乐意:\n1. 记录您的喜好(例如:您最喜欢吃苹果",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "single_fact",
"query": "跨对话后请回答:我的提醒时间是什么?",
"reply": "作为一个人工智能助手,我**没有访问您设备本地设置(如手机或电脑日历)的权限**,因此无法直接告诉您具体的“提醒时间”。\n\n不过,如果您是在询问**当前系统时间**,我可以",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "single_fact",
"query": "跨对话后请回答:我的备用联系人是什么?",
"reply": "作为一个人工智能助手,我**无法访问您的个人设备、通讯录或任何与您的私人数据相关的信息**。因此,我无法知道您的“备用联系人”是谁。\n\n要查找您的备用联系人,您可以尝试以下方法:",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "single_fact",
"query": "跨对话后请回答:我的常用语言是什么?",
"reply": "您好!作为一个人工智能助手,我并没有直接访问您个人设备或历史记录的能力,因此无法确切知道您“常用”的语言是什么。\n\n不过,从我们当前的对话来看,您使用的是**中文**。如果您希望我用其他",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "single_fact",
"query": "跨对话后请回答:我的提醒时间是什么?",
"reply": "作为一个人工智能助手,我**没有访问您设备本地设置(如手机或电脑日历)的权限**,因此无法直接告诉您具体的“提醒时间”。\n\n不过,如果您是在询问**当前系统时间**,我可以",
"written": 6,
"correct": false,
"matched": [],
"abstained": false,
"wrongly_abstained": false
},
{
"category": "unknown_abstention",
"query": "我的不存在的个人属性-0008000是什么?如果没有记录,请明确说不知道。",
"reply": "不知道。\n\n“不存在的个人属性 -0008000\"并非任何已知的系统、数据库、游戏或现实世界中的有效标识符。在没有任何上下文或记录的情况下,我无法提供关于该属性的具体信息",
"written": 6,
"correct": true,
"matched": [],
"abstained": true,
"wrongly_abstained": false
},
{
"category": "unknown_abstention",
"query": "我的不存在的个人属性-0008001是什么?如果没有记录,请明确说不知道。",
"reply": "不知道。\n\n“不存在的个人属性 -0008001\"并非任何已知的系统、数据库、公开记录或通用标准中存在的概念。如果您是在某个特定游戏、小说、虚构作品或内部系统中看到",
"written": 6,
"correct": true,
"matched": [],
"abstained": true,
"wrongly_abstained": false
},
{
"category": "unknown_abstention",
"query": "我的不存在的个人属性-0008002是什么?如果没有记录,请明确说不知道。",
"reply": "不知道。\n\n“不存在的个人属性 -0008002\"并非任何已知的系统、数据库、公开记录或通用标准中存在的概念。如果您是在某个特定游戏、小说、虚构作品或内部系统中看到",
"written": 6,
"correct": true,
"matched": [],
"abstained": true,
"wrongly_abstained": false
},
{
"category": "unknown_abstention",
"query": "我的不存在的个人属性-0008003是什么?如果没有记录,请明确说不知道。",
"reply": "不知道。\n\n“不存在的个人属性 -0008003\"并非任何已知的系统、数据库、公开记录或通用标准中存在的概念。如果您是在某个特定上下文(如游戏、小说、特定软件或",
"written": 6,
"correct": true,
"matched": [],
"abstained": true,
"wrongly_abstained": false
},
{
"category": "unknown_abstention",
"query": "我的不存在的个人属性-0008004是什么?如果没有记录,请明确说不知道。",
"reply": "不知道。\n\n“不存在的个人属性 -0008004\"并非任何已知的系统、数据库、游戏或现实世界中的有效标识符。在没有任何上下文或记录的情况下,我无法提供关于该属性的具体信息",
"written": 6,
"correct": true,
"matched": [],
"abstained": true,
"wrongly_abstained": false
},
{
"category": "unknown_abstention",
"query": "我的不存在的个人属性-0008005是什么?如果没有记录,请明确说不知道。",
"reply": "不知道。\n\n“不存在的个人属性 -0008005\"并非任何已知的系统、数据库、游戏或现实世界中的有效标识符。在没有任何上下文或记录的情况下,我无法提供关于该属性的具体信息",
"written": 6,
"correct": true,
"matched": [],
"abstained": true,
"wrongly_abstained": false
},
{
"category": "unknown_abstention",
"query": "我的不存在的个人属性-0008006是什么?如果没有记录,请明确说不知道。",
"reply": "不知道。\n\n“不存在的个人属性 -0008006\"并非任何已知的公开数据、系统记录或通用概念。在缺乏具体上下文(如特定的游戏、软件、数据库或虚构作品)的情况下,",
"written": 6,
"correct": true,
"matched": [],
"abstained": true,
"wrongly_abstained": false
},
{
"category": "unknown_abstention",
"query": "我的不存在的个人属性-0008007是什么?如果没有记录,请明确说不知道。",
"reply": "不知道。\n\n“不存在的个人属性 -0008007\"并非任何已知的系统、数据库、游戏或现实世界中的有效标识符。在没有任何上下文或记录的情况下,我无法提供关于该属性的具体信息",
"written": 6,
"correct": true,
"matched": [],
"abstained": true,
"wrongly_abstained": false
},
{
"category": "unknown_abstention",
"query": "我的不存在的个人属性-0008008是什么?如果没有记录,请明确说不知道。",
"reply": "不知道。\n\n“不存在的个人属性 -0008008\"并非任何已知的系统、数据库、公开记录或通用标准中存在的概念。如果您是在某个特定上下文(如游戏、小说、特定软件或",
"written": 6,
"correct": true,
"matched": [],
"abstained": true,
"wrongly_abstained": false
},
{
"category": "unknown_abstention",
"query": "我的不存在的个人属性-0008009是什么?如果没有记录,请明确说不知道。",
"reply": "不知道。\n\n“不存在的个人属性 -0008009\"并非任何已知的系统、数据库、游戏或现实世界中的有效标识符。在没有任何上下文或记录的情况下,我无法提供关于该属性的具体信息",
"written": 6,
"correct": true,
"matched": [],
"abstained": true,
"wrongly_abstained": false
}
]
}
}
],
"B_zero_overlap_paraphrase": [
"| NM2.1+gate | 16 | 75.00% | 12.50% | 81.25% | 6/16 | 5.06 | attribute_not_covered:2, evidence_found:13, router_abstained:1 |",
"wrote nm2_1_gate_critical_e2e.md",
"wrote nm2_1_gate_critical_e2e.json",
{
"NM2.1+gate": {
"summary": {
"router": "NM2.1+gate",
"cases": 16,
"accuracy_pct": 75.0,
"wrong_attribute_pct": 12.5,
"read_pct": 81.25,
"cases_with_rule_hits": 6,
"stop_reasons": {
"evidence_found": 13,
"attribute_not_covered": 2,
"router_abstained": 1
},
"mean_records_selected": 5.0625,
"seconds": 74.4
},
"rows": [
{
"attribute": "常住城市",
"wanted": "CITY-A1B2C3",
"reply": "您的常住城市是 CITY-A1B2C3。",
"correct": true,
"wrong_attribute_code": false,
"rule_address_hit": false,
"rule_lexical_hits": 0,
"stop_reason": "evidence_found",
"need_memory": true,
"top_score": 3.384649144503837,
"records_selected": 8
},
{
"attribute": "出生城市",
"wanted": "CITY-D4E5F6",
"reply": "根据现有记忆,您的老家信息未知。已知事实包括:我的工位在 3 号楼;我的常住城市是 CITY-A1B2C3;我的出生城市是 CITY-D4E5F6;我的办公城市",
"correct": true,
"wrong_attribute_code": false,
"rule_address_hit": false,
"rule_lexical_hits": 0,
"stop_reason": "evidence_found",
"need_memory": true,
"top_score": 2.6641891400334443,
"records_selected": 8
},
{
"attribute": "办公城市",
"wanted": "CITY-G7H8J9",
"reply": "您的办公城市是 CITY-G7H8J9。user;办公城市;CITY-G7H8J9",
"correct": true,
"wrong_attribute_code": false,
"rule_address_hit": false,
"rule_lexical_hits": 20,
"stop_reason": "evidence_found",
"need_memory": true,
"top_score": 8.520044145734458,
"records_selected": 8
},
{
"attribute": "档案代号",
"wanted": "CODE-K1L2M3",
"reply": "您的主管姓名是 BOSS-X4Y5Z6,团队名称是 TEAM-D1E2F3。",
"correct": false,
"wrong_attribute_code": true,
"rule_address_hit": false,
"rule_lexical_hits": 0,
"stop_reason": "attribute_not_covered",
"need_memory": false,
"top_score": 0.0,
"records_selected": 0
},
{
"attribute": "常用编辑器",
"wanted": "EDIT-N4P5Q6",
"reply": "您写代码时使用的软件是 EDIT-N4P5Q6。user;常用编辑器;EDIT-N4P5Q6",
"correct": true,
"wrong_attribute_code": false,
"rule_address_hit": false,
"rule_lexical_hits": 3,
"stop_reason": "evidence_found",
"need_memory": true,
"top_score": 8.515892826960929,
"records_selected": 3
},
{
"attribute": "默认语言",
"wanted": "LANG-R7S8T9",
"reply": "我的默认语言是 LANG-R7S8T9。user;默认语言;LANG-R7S8T9",
"correct": true,
"wrong_attribute_code": false,
"rule_address_hit": false,
"rule_lexical_hits": 0,
"stop_reason": "evidence_found",
"need_memory": true,
"top_score": 2.865009269951279,
"records_selected": 8
},
{
"attribute": "通勤方式",
"wanted": "COMMU-U1V2W3",
"reply": "我的通勤方式是 COMMU-U1V2W3。",
"correct": true,
"wrong_attribute_code": false,
"rule_address_hit": false,
"rule_lexical_hits": 0,
"stop_reason": "evidence_found",
"need_memory": true,
"top_score": 3.389946369992935,
"records_selected": 8
},
{
"attribute": "主管姓名",
"wanted": "BOSS-X4Y5Z6",
"reply": "这句话听起来带着几分疲惫、迷茫,或者是对现状的无力感。能感受到你此刻可能正经历一些难以独自应对的困境。\n\n“谁管我?”这个问题背后,或许藏着几个不同的需求,我想为你拆解一下",
"correct": false,
"wrong_attribute_code": false,
"rule_address_hit": false,
"rule_lexical_hits": 0,
"stop_reason": "router_abstained",
"need_memory": false,
"top_score": 0.0,
"records_selected": 0
},
{
"attribute": "工位楼层",
"wanted": "FLOOR-A7B8C9",
"reply": "我的工位楼层是 FLOOR-A7B8C9。",
"correct": true,
"wrong_attribute_code": false,
"rule_address_hit": false,
"rule_lexical_hits": 0,
"stop_reason": "evidence_found",
"need_memory": true,
"top_score": 4.038213959597986,
"records_selected": 8
},
{
"attribute": "团队名称",
"wanted": "TEAM-D1E2F3",
"reply": "您属于的小组是 TEAM-D1E2F3。\nuser;团队名称;TEAM-D1E2F3",
"correct": true,
"wrong_attribute_code": false,
"rule_address_hit": false,
"rule_lexical_hits": 0,
"stop_reason": "evidence_found",
"need_memory": true,
"top_score": 3.7658000497788082,
"records_selected": 8
},
{
"attribute": "邮箱域名",
"wanted": "MAIL-G4H5J6",
"reply": "关于别人发信需要写的后缀,当前记忆中没有相关信息,无法回答。已知信息仅包含:user;紧急联系人姓氏;KIN-X1Y2Z3。",
"correct": false,
"wrong_attribute_code": true,
"rule_address_hit": false,
"rule_lexical_hits": 1,
"stop_reason": "evidence_found",
"need_memory": true,
"top_score": 8.513982709418043,
"records_selected": 1
},
{
"attribute": "手机尾号",
"wanted": "PHONE-K7L8M9",
"reply": "您的电话号码最后几位是 PHONE-K7L8M9。user;手机尾号;PHONE-K7L8M9",
"correct": true,
"wrong_attribute_code": false,
"rule_address_hit": false,
"rule_lexical_hits": 0,
"stop_reason": "evidence_found",
"need_memory": true,
"top_score": 4.0849550882345795,
"records_selected": 8
},
{
"attribute": "项目代号",
"wanted": "PROJ-N1P2Q3",
"reply": "关于您正在做的工程名称,当前长期记忆中没有相关记录,无法回答。",
"correct": false,
"wrong_attribute_code": false,
"rule_address_hit": false,
"rule_lexical_hits": 2,
"stop_reason": "evidence_found",
"need_memory": true,
"top_score": 8.514652462471835,
"records_selected": 2
},
{
"attribute": "入职年份",
"wanted": "YEAR-R4S5T6",
"reply": "您的入职年份是 YEAR-R4S5T6。",
"correct": true,
"wrong_attribute_code": false,
"rule_address_hit": false,
"rule_lexical_hits": 0,
"stop_reason": "attribute_not_covered",
"need_memory": false,
"top_score": 0.0,
"records_selected": 0
},
{
"attribute": "办公楼层",
"wanted": "DESK-U7V8W9",
"reply": "您的办公楼层是 DESK-U7V8W9。",
"correct": true,
"wrong_attribute_code": false,
"rule_address_hit": false,
"rule_lexical_hits": 3,
"stop_reason": "evidence_found",
"need_memory": true,
"top_score": 8.524232190583632,
"records_selected": 3
},
{
"attribute": "紧急联系人姓氏",
"wanted": "KIN-X1Y2Z3",
"reply": "您的紧急联系人姓氏是 KIN-X1Y2Z3。",
"correct": true,
"wrong_attribute_code": false,
"rule_address_hit": false,
"rule_lexical_hits": 20,
"stop_reason": "evidence_found",
"need_memory": true,
"top_score": 8.522833668863877,
"records_selected": 8
}
]
}
}
],
"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 |",
"|---|---:|---:|---:|---:|---:|---:|---|",
"| 0.00 | 48 | 40 | 65.00% | 35.00% | 62.50% | 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 |",
"|---|---:|---:|---:|---:|---:|---:|",
"| 0.00 · 可回答 | 40 | 0.7 | 0.77 | 2.85 | 8.52 | 8.52 |",
"| 0.00 · 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": {
"0.00": {
"router": "blend=0.00",
"cases": 48,
"answerable_cases": 40,
"unknown_cases": 8,
"accuracy_pct": 65.0,
"wrong_attribute_pct": 35.0,
"unknown_leak_pct": 62.5,
"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": 0.7,
"p10": 0.77,
"p50": 2.85,
"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": 336.0
}
},
"rows": {
"0.00": [
{
"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": 2.564332926884773,
"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": 0.6998433045297862,
"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": 3.359343914853775,
"need_memory": true
},
{
"id": "zero-overlap-000246",
"attribute": "常住城市",
"answerable": false,
"expected": "VAL-6BWJYDE4",
"reply": "关于您小时候待的地方,当前长期记忆中仅包含您常住城市的信息(user;常住城市;VAL-BXJKK64G),没有关于您小时候居住地的记录,因此不知道。",
"correct": false,
"wrong_attribute": false,
"leaked": true,
"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": "您的阅读工具是 VAL-ZRLSPU5V。",
"correct": true,
"wrong_attribute": false,
"leaked": false,
"records_written": 24,
"records_active": 24,
"stop_reason": "evidence_found",
"records_selected": 8,
"top_score": 2.9886870104974497,
"need_memory": true
},
{
"id": "zero-overlap-000078",
"attribute": "常住城市",
"answerable": true,
"expected": "VAL-TQBY65U8",
"reply": "我的常住城市是 VAL-TQBY65U8。",
"correct": true,
"wrong_attribute": false,
"leaked": false,
"records_written": 24,
"records_active": 24,
"stop_reason": "evidence_found",
"records_selected": 8,
"top_score": 1.2650369691848753,
"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 VAL-SRFC842D。",
"correct": true,
"wrong_attribute": false,
"leaked": false,
"records_written": 24,
"records_active": 24,
"stop_reason": "evidence_found",
"records_selected": 8,
"top_score": 0.7423390823602677,
"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": "关于您的闹钟设置时间,当前长期记忆中未包含相关证据,无法回答。",
"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": 1.0058205354213714,
"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 user;主管姓名;VAL-VDL8H7Y9。",
"correct": true,
"wrong_attribute": false,
"leaked": false,
"records_written": 24,
"records_active": 24,
"stop_reason": "evidence_found",
"records_selected": 8,
"top_score": 1.9513123025616852,
"need_memory": true
},
{
"id": "zero-overlap-000184",
"attribute": "项目代号",
"answerable": true,
"expected": "VAL-3H5TQEC4",
"reply": "您的团队名称是 VAL-Z98JC7BG。",
"correct": false,
"wrong_attribute": true,
"leaked": false,
"records_written": 24,
"records_active": 24,
"stop_reason": "evidence_found",
"records_selected": 8,
"top_score": 3.191310502301591,
"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": "我翻书时依赖的阅读工具是 VAL-SHRY3GNG。",
"correct": true,
"wrong_attribute": false,
"leaked": false,
"records_written": 24,
"records_active": 24,
"stop_reason": "evidence_found",
"records_selected": 8,
"top_score": 2.9932317454522837,
"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": 0.7088197509944439,
"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。",
"correct": true,
"wrong_attribute": false,
"leaked": false,
"records_written": 24,
"records_active": 24,
"stop_reason": "evidence_found",
"records_selected": 8,
"top_score": 2.7659836388747943,
"need_memory": true
},
{
"id": "zero-overlap-000006",
"attribute": "档案标识",
"answerable": true,
"expected": "VAL-8EFN3KGR",
"reply": "您的系统分配字符包括:紧急联系人姓氏 VAL-KLPQW2L2;课程名称 VAL-PVMQLCL9;团队名称 VAL-V4BVFB2N;客户名称 VAL-92LH7CHG",
"correct": false,
"wrong_attribute": true,
"leaked": false,
"records_written": 24,
"records_active": 24,
"stop_reason": "evidence_found",
"records_selected": 8,
"top_score": 4.169416844133658,
"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": "您的课程名称是 VAL-DCFZV57V,您的阅读工具是 VAL-YJF3JEVA。",
"correct": false,
"wrong_attribute": false,
"leaked": true,
"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。user;出生城市;VAL-PHB8FNF9",
"correct": true,
"wrong_attribute": false,
"leaked": false,
"records_written": 24,
"records_active": 24,
"stop_reason": "evidence_found",
"records_selected": 8,
"top_score": 2.8505235558902053,
"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": 2.3051346188937454,
"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,通勤方式是 VAL-2CXTE3TB,出生城市是 VAL-GE828D22,办公城市是 VAL-XYRY8BUQ,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": 1.6213454890251158,
"need_memory": true
},
{
"id": "zero-overlap-000214",
"attribute": "客户名称",
"answerable": true,
"expected": "VAL-MV382EEF",
"reply": "我服务的对象是 VAL-MV382EEF。",
"correct": true,
"wrong_attribute": false,
"leaked": false,
"records_written": 24,
"records_active": 24,
"stop_reason": "evidence_found",
"records_selected": 8,
"top_score": 0.7693164377659559,
"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": 2.8819549524530643,
"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": 2.942847951625275,
"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": 0.77552460052073,
"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": 2.1366783114608343,
"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": 2.130069571965557,
"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": 1.4585979866981504,
"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": 0.7712558816373348,
"need_memory": true
},
{
"id": "zero-overlap-000144",
"attribute": "入职年份",
"answerable": false,
"expected": "VAL-GLHQMRN6",
"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。已知信息包括:user;常住城市;VAL-M7H3ZUK5;user;入职年份;VAL-GLHQMRN6",
"correct": false,
"wrong_attribute": false,
"leaked": true,
"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": 0.7196446585655213,
"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。",
"correct": true,
"wrong_attribute": false,
"leaked": false,
"records_written": 24,
"records_active": 24,
"stop_reason": "evidence_found",
"records_selected": 8,
"top_score": 1.286561994552612,
"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": "关于您坐在哪一间房的信息,当前长期记忆中未包含该数据。已知信息如下:user;宿舍楼号;VAL-JTSHA7PM。",
"correct": false,
"wrong_attribute": true,
"leaked": false,
"records_written": 24,
"records_active": 24,
"stop_reason": "evidence_found",
"records_selected": 8,
"top_score": 3.1937497879965546,
"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": 2.8081361769552853,
"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": 2.7214163786326675,
"need_memory": true
},
{
"id": "zero-overlap-000180",
"attribute": "客户名称",
"answerable": false,
"expected": "VAL-3F87H6UT",
"reply": "您的课程名称是 VAL-HMSTQFKW,您的阅读工具是 VAL-2SP3QMVD。",
"correct": false,
"wrong_attribute": false,
"leaked": true,
"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": "根据已保存的长期记忆,当出现问题时,您可以联系以下实体:user;档案标识;VAL-9XT5ERHC。",
"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": "关于您待得最久的那块地儿,当前记忆库中未包含此信息。已知事实仅确认您的出生城市是 VAL-4XKTN2EE,但无法推断出您待得最久的地点。",
"correct": false,
"wrong_attribute": false,
"leaked": true,
"records_written": 24,
"records_active": 23,
"stop_reason": "attribute_not_covered",
"records_selected": 0,
"top_score": 0.0,
"need_memory": false
}
]
}
}
],
"D_restart_persistence": [
" \"enabled\": true,",
" \"router_ready\": true",
" }",
"}",
{
"model_path": "qwen3_5_4b_natural_memory_v2_1",
"history_passed_to_restart": false,
"fact": "我正在开发一个长期项目,项目内部代号是NM-V2-RESTART,使用中文。",
"query": "我正在开发的长期项目内部代号是什么?只回答代号。",
"expected": "NM-V2-RESTART",
"automatic_write": true,
"write_changed": true,
"before_save": {
"pages": 1,
"max_pages": 32768,
"page_capacity": 32,
"capacity_records": 1048576,
"open_pages": 1,
"records": 1,
"active_records": 1,
"superseded_records": 0,
"retracted_records": 0,
"quarantined_records": 0,
"hot_pages": 1,
"warm_pages": 0,
"cold_pages": 0,
"active_conflict_keys": 0,
"coarse_index_buckets": 1,
"last_coarse_candidates": 0,
"storage_mode": "embedded",
"resident_records": 1,
"resident_pages": 1,
"gpu_cache_records": 1,
"gpu_cache_tokens": 141,
"gpu_cache_bytes": 2293,
"gpu_cache_hits": 0,
"gpu_cache_misses": 1,
"gpu_cache_fallbacks": 0,
"gpu_cache_alloc_failures": 0,
"gpu_cache_device": "cuda:0",
"gpu_cache_limit_records": 256,
"gpu_cache_limit_tokens": 131072,
"gpu_cache_reserve_mb": 2048,
"gpu_cache_adaptive": true,
"gpu_cache_last_free_mb": 7685.0,
"kv_budget": {
"max_tokens": 32768,
"trigger_tokens": 29491
},
"enabled": true,
"router_ready": true
},
"after_restart": {
"response": "项目内部代号是 NM-V2-RESTART。",
"prefix_length": 141,
"prefix_used": true,
"decision": [
{
"need_memory": true,
"page_ids": [
"page_00000001"
],
"record_ids": [
"mem_30077f4c30229692"
],
"page_scores": [
3.549997901916504
],
"record_scores": [
8.436647681137504
],
"hop_count": 0,
"hop_trace": [],
"confidence": 1.0,
"stop_reason": "evidence_found",
"top_score": 8.436647681137504,
"score_margin": 8.436647681137504,
"evidence_score": 1.0
}
],
"prefix_shape": [
1,
141
],
"prefix_mask_shape": [
1,
141
]
},
"router_recalled_after_restart": true,
"generated_contains_expected": true,
"cleanup_applied": true,
"after_cleanup": {
"pages": 0,
"max_pages": 32768,
"page_capacity": 32,
"capacity_records": 1048576,
"open_pages": 0,
"records": 0,
"active_records": 0,
"superseded_records": 0,
"retracted_records": 0,
"quarantined_records": 0,
"hot_pages": 0,
"warm_pages": 0,
"cold_pages": 0,
"active_conflict_keys": 0,
"coarse_index_buckets": 0,
"last_coarse_candidates": 0,
"storage_mode": "embedded",
"resident_records": 0,
"resident_pages": 0,
"gpu_cache_records": 0,
"gpu_cache_tokens": 0,
"gpu_cache_bytes": 0,
"gpu_cache_hits": 0,
"gpu_cache_misses": 0,
"gpu_cache_fallbacks": 0,
"gpu_cache_alloc_failures": 0,
"gpu_cache_device": "cuda:0",
"gpu_cache_limit_records": 256,
"gpu_cache_limit_tokens": 131072,
"gpu_cache_reserve_mb": 2048,
"gpu_cache_adaptive": true,
"gpu_cache_last_free_mb": null,
"kv_budget": {
"max_tokens": 32768,
"trigger_tokens": 29491
},
"enabled": true,
"router_ready": true
}
}
]
}
}