- 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
124 lines
2.8 KiB
JSON
124 lines
2.8 KiB
JSON
{
|
|
"attributes": 24,
|
|
"chance_pct": 4.17,
|
|
"train_queries": 1200,
|
|
"eval_queries": 300,
|
|
"distinct_train_queries": null,
|
|
"train_accuracy_pct": 83.42,
|
|
"eval_accuracy_pct": 83.33,
|
|
"eval_accuracy_answerable_pct": 100.0,
|
|
"note": "eval queries are different paraphrases of the same attributes, so this is generalisation to unseen wording",
|
|
"per_attribute": {
|
|
"主管姓名": {
|
|
"eval_queries": 12,
|
|
"correct_pct": 100.0
|
|
},
|
|
"入职年份": {
|
|
"eval_queries": 11,
|
|
"correct_pct": 100.0
|
|
},
|
|
"出生城市": {
|
|
"eval_queries": 9,
|
|
"correct_pct": 100.0
|
|
},
|
|
"办公城市": {
|
|
"eval_queries": 9,
|
|
"correct_pct": 100.0
|
|
},
|
|
"午餐偏好": {
|
|
"eval_queries": 12,
|
|
"correct_pct": 100.0
|
|
},
|
|
"咖啡口味": {
|
|
"eval_queries": 19,
|
|
"correct_pct": 100.0
|
|
},
|
|
"团队名称": {
|
|
"eval_queries": 10,
|
|
"correct_pct": 100.0
|
|
},
|
|
"客户名称": {
|
|
"eval_queries": 12,
|
|
"correct_pct": 100.0
|
|
},
|
|
"宿舍楼号": {
|
|
"eval_queries": 14,
|
|
"correct_pct": 100.0
|
|
},
|
|
"工位楼层": {
|
|
"eval_queries": 6,
|
|
"correct_pct": 100.0
|
|
},
|
|
"常住城市": {
|
|
"eval_queries": 13,
|
|
"correct_pct": 100.0
|
|
},
|
|
"常用编辑器": {
|
|
"eval_queries": 6,
|
|
"correct_pct": 100.0
|
|
},
|
|
"手机尾号": {
|
|
"eval_queries": 11,
|
|
"correct_pct": 100.0
|
|
},
|
|
"档案标识": {
|
|
"eval_queries": 16,
|
|
"correct_pct": 100.0
|
|
},
|
|
"紧急联系人姓氏": {
|
|
"eval_queries": 7,
|
|
"correct_pct": 100.0
|
|
},
|
|
"设备型号": {
|
|
"eval_queries": 7,
|
|
"correct_pct": 100.0
|
|
},
|
|
"课程名称": {
|
|
"eval_queries": 12,
|
|
"correct_pct": 100.0
|
|
},
|
|
"起床时间": {
|
|
"eval_queries": 12,
|
|
"correct_pct": 100.0
|
|
},
|
|
"运动习惯": {
|
|
"eval_queries": 6,
|
|
"correct_pct": 100.0
|
|
},
|
|
"通勤方式": {
|
|
"eval_queries": 8,
|
|
"correct_pct": 100.0
|
|
},
|
|
"邮箱域名": {
|
|
"eval_queries": 7,
|
|
"correct_pct": 100.0
|
|
},
|
|
"阅读工具": {
|
|
"eval_queries": 7,
|
|
"correct_pct": 100.0
|
|
},
|
|
"项目代号": {
|
|
"eval_queries": 12,
|
|
"correct_pct": 100.0
|
|
},
|
|
"默认语言": {
|
|
"eval_queries": 12,
|
|
"correct_pct": 100.0
|
|
}
|
|
},
|
|
"coverage_rule": {
|
|
"rule": "predict the asked-about attribute; refuse when the candidate set lacks it",
|
|
"answerable_inside_pct": 100.0,
|
|
"unknown_inside_pct": 0.0,
|
|
"answerable_cases": 250,
|
|
"unknown_cases": 50,
|
|
"unknown_refusal_pct": 100.0,
|
|
"known_false_refusal_pct": 0.0,
|
|
"counts": {
|
|
"tp": 50,
|
|
"fn": 0,
|
|
"fp": 0,
|
|
"tn": 250
|
|
}
|
|
}
|
|
} |