Files
natural-memory-nm21/data/_probe_cap1/eval.jsonl
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

2 lines
13 KiB
JSON

{"id":"benchmark_qa:测试实体-0000:query","group_id":"benchmark_qa:测试实体-0000","source":"dynamic_memory_lab/data/benchmark_eval.jsonl","family":"benchmark_qa","query":"请问对象测试实体0000的驻地代号为?","candidates":[{"id":"mega_validation:10298fe27ce978d23bc3","text":"这是普通对话噪声:验证用户-0008515暂时提到一个无关编号N-0G4UVIBJ,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-distractor_128-0008515","family":"mega_validation","kind":"noise","entity":"验证用户-0008515","attribute":"干扰属性13"},{"id":"mega_validation:d549e1a25dbbb7dcd2cf","text":"我的备用联系人是OLD-T3D6RO49。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-conflict_update-0009470","family":"mega_validation","kind":"fact","entity":"验证用户-0009470","attribute":"备用联系人"},{"id":"mega_validation:5bd0f998ab024560e522","text":"这是普通对话噪声:验证用户-0009230暂时提到一个无关编号N-L0Z2L0TX,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-distractor_128-0009230","family":"mega_validation","kind":"noise","entity":"验证用户-0009230","attribute":"干扰属性58"},{"id":"mega_validation:5c6158f5ad6bb6047f76","text":"这是普通对话噪声:验证用户-0009431暂时提到一个无关编号N-FD4GMCX5,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-long_context-0009431","family":"mega_validation","kind":"noise","entity":"验证用户-0009431","attribute":"干扰属性54"},{"id":"mega_validation:4629b7f46c7856cdbd9f","text":"这是普通对话噪声:验证用户-0009891暂时提到一个无关编号N-FZSU7B2O,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-random_position-0009891","family":"mega_validation","kind":"noise","entity":"验证用户-0009891","attribute":"干扰属性2"},{"id":"mega_validation:d34fa5417676877a1820","text":"这是普通对话噪声:验证用户-0008647暂时提到一个无关编号N-VCMT4CVH,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-distractor_128-0008647","family":"mega_validation","kind":"noise","entity":"验证用户-0008647","attribute":"干扰属性65"},{"id":"mega_validation:5a3d7ed7669d35956bf9","text":"这是普通对话噪声:验证用户-0008152暂时提到一个无关编号N-TH1G8CJF,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-long_context-0008152","family":"mega_validation","kind":"noise","entity":"验证用户-0008152","attribute":"干扰属性84"},{"id":"mega_validation:08fad5659d7ed9413f63","text":"这是普通对话噪声:验证用户-0008487暂时提到一个无关编号N-ODG725WF,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-distractor_128-0008487","family":"mega_validation","kind":"noise","entity":"验证用户-0008487","attribute":"干扰属性15"},{"id":"mega_validation:458132b278201b6288de","text":"这是普通对话噪声:验证用户-0008990暂时提到一个无关编号N-RRR4NJKZ,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-distractor_32-0008990","family":"mega_validation","kind":"noise","entity":"验证用户-0008990","attribute":"干扰属性11"},{"id":"mega_validation:b48dcc9435941af089ad","text":"这是普通对话噪声:验证用户-0008442暂时提到一个无关编号N-V71O69AU,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-distractor_128-0008442","family":"mega_validation","kind":"noise","entity":"验证用户-0008442","attribute":"干扰属性65"},{"id":"mega_validation:45ea30fa6d9e67f30f30","text":"这是普通对话噪声:验证用户-0008253暂时提到一个无关编号N-CWBUE53G,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-distractor_128-0008253","family":"mega_validation","kind":"noise","entity":"验证用户-0008253","attribute":"干扰属性63"},{"id":"mega_validation:a9af4a2dd9d3d5a172fa","text":"这是普通对话噪声:验证用户-0008920暂时提到一个无关编号N-K5AMVVQW,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-long_context-0008920","family":"mega_validation","kind":"noise","entity":"验证用户-0008920","attribute":"干扰属性67"},{"id":"mega_validation:ac388eea71a02b681706","text":"这是普通对话噪声:验证用户-0008783暂时提到一个无关编号N-A7IFNXIC,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-long_context-0008783","family":"mega_validation","kind":"noise","entity":"验证用户-0008783","attribute":"干扰属性94"},{"id":"mega_validation:89c3b5d2f9301d05582d","text":"这是普通对话噪声:验证用户-0009311暂时提到一个无关编号N-E03EM80Y,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-long_context-0009311","family":"mega_validation","kind":"noise","entity":"验证用户-0009311","attribute":"干扰属性62"},{"id":"mega_validation:e15943c3a4d3b97c2d32","text":"这是普通对话噪声:验证用户-0009534暂时提到一个无关编号N-NZ0N73B0,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-distractor_32-0009534","family":"mega_validation","kind":"noise","entity":"验证用户-0009534","attribute":"干扰属性17"},{"id":"mega_validation:e1fa176e208c3583e1f8","text":"这是普通对话噪声:验证用户-0008681暂时提到一个无关编号N-UDKI5VIK,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-distractor_128-0008681","family":"mega_validation","kind":"noise","entity":"验证用户-0008681","attribute":"干扰属性113"},{"id":"mega_validation:1404004e29d99e558e87","text":"这是普通对话噪声:验证用户-0008500暂时提到一个无关编号N-KHOF7Y2E,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-random_position-0008500","family":"mega_validation","kind":"noise","entity":"验证用户-0008500","attribute":"干扰属性44"},{"id":"mega_validation:30b301aac603d409387c","text":"这是普通对话噪声:验证用户-0008850暂时提到一个无关编号N-EWIEB1L1,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-distractor_32-0008850","family":"mega_validation","kind":"noise","entity":"验证用户-0008850","attribute":"干扰属性14"},{"id":"mega_validation:6a2514902ace28170929","text":"这是普通对话噪声:验证用户-0009906暂时提到一个无关编号N-4SZTVH9P,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-long_context-0009906","family":"mega_validation","kind":"noise","entity":"验证用户-0009906","attribute":"干扰属性69"},{"id":"mega_validation:37aaff204711a3f8d18e","text":"这是普通对话噪声:验证用户-0008100暂时提到一个无关编号N-7RBTPEHU,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-distractor_128-0008100","family":"mega_validation","kind":"noise","entity":"验证用户-0008100","attribute":"干扰属性98"},{"id":"mega_validation:7bf8b629017f5b09408d","text":"这是普通对话噪声:验证用户-0008882暂时提到一个无关编号N-WG0YZI6J,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-random_position-0008882","family":"mega_validation","kind":"noise","entity":"验证用户-0008882","attribute":"干扰属性20"},{"id":"mega_validation:0835b835dd71078f106d","text":"这是普通对话噪声:验证用户-0009189暂时提到一个无关编号N-UJLV87SQ,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-random_position-0009189","family":"mega_validation","kind":"noise","entity":"验证用户-0009189","attribute":"干扰属性57"},{"id":"mega_validation:3463720b5b37f8396023","text":"这是普通对话噪声:验证用户-0008275暂时提到一个无关编号N-8ZNZZXH5,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-long_context-0008275","family":"mega_validation","kind":"noise","entity":"验证用户-0008275","attribute":"干扰属性81"},{"id":"mega_validation:4948be4ebe97474ceb1f","text":"这是普通对话噪声:验证用户-0009973暂时提到一个无关编号N-1OSL249L,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-distractor_32-0009973","family":"mega_validation","kind":"noise","entity":"验证用户-0009973","attribute":"干扰属性20"},{"id":"mega_validation:5c210c0dd1c0a940bb03","text":"这是普通对话噪声:验证用户-0008147暂时提到一个无关编号N-KHE5TQIT,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-distractor_32-0008147","family":"mega_validation","kind":"noise","entity":"验证用户-0008147","attribute":"干扰属性18"},{"id":"mega_validation:0c2d085f804d2a90de41","text":"这是普通对话噪声:验证用户-0008536暂时提到一个无关编号N-6LC4U0RR,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-random_position-0008536","family":"mega_validation","kind":"noise","entity":"验证用户-0008536","attribute":"干扰属性22"},{"id":"mega_validation:148dd18b1451bd2c2be4","text":"这是普通对话噪声:验证用户-0008249暂时提到一个无关编号N-P2X33P7Y,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-random_position-0008249","family":"mega_validation","kind":"noise","entity":"验证用户-0008249","attribute":"干扰属性28"},{"id":"mega_validation:177cc9c88e28f00e252f","text":"这是普通对话噪声:验证用户-0009951暂时提到一个无关编号N-3O1Y7MEY,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-distractor_128-0009951","family":"mega_validation","kind":"noise","entity":"验证用户-0009951","attribute":"干扰属性58"},{"id":"mega_validation:d065c96dde6af220fe71","text":"这是普通对话噪声:验证用户-0008730暂时提到一个无关编号N-80A9FH33,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-long_context-0008730","family":"mega_validation","kind":"noise","entity":"验证用户-0008730","attribute":"干扰属性31"},{"id":"mega_validation:e28e87409c6daf7a4b49","text":"这是普通对话噪声:验证用户-0008832暂时提到一个无关编号N-219EQH6T,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-distractor_128-0008832","family":"mega_validation","kind":"noise","entity":"验证用户-0008832","attribute":"干扰属性16"},{"id":"mega_validation:9713b372ed5e4596994d","text":"这是普通对话噪声:验证用户-0008012暂时提到一个无关编号N-ZYJMF7NS,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-distractor_128-0008012","family":"mega_validation","kind":"noise","entity":"验证用户-0008012","attribute":"干扰属性123"},{"id":"mega_validation:07aa3cba9c2c822547a0","text":"这是普通对话噪声:验证用户-0008269暂时提到一个无关编号N-065ENCLW,不需要长期保存。","source":"dynamic_memory_lab/data/mega_validation/memory_validation_20k_eval.jsonl","group_id":"mega_validation:mega-long_context-0008269","family":"mega_validation","kind":"noise","entity":"验证用户-0008269","attribute":"干扰属性39"}],"positive_indices":[],"positive_index":-1,"need_memory":0.0,"hop":0,"metadata":{"answer":"H","subject":"测试实体0000","attribute":"驻地","task":"single_fact_retrieval"}}