WpyQwq
643e22ecb9
- 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
493 lines
11 KiB
JSON
493 lines
11 KiB
JSON
{
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"thresholds": [
|
|
"thr0.30",
|
|
"thr0.40",
|
|
"thr0.50",
|
|
"thr0.60",
|
|
"thr0.70",
|
|
"thr0.80"
|
|
],
|
|
"candidates": {
|
|
"V2-128 v6 final": {
|
|
"rows": [
|
|
{
|
|
"axis": "need F1",
|
|
"direction": "max",
|
|
"values_pct": [
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "need 召回",
|
|
"direction": "max",
|
|
"values_pct": [
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "need 精确率",
|
|
"direction": "max",
|
|
"values_pct": [
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "未知拒答率",
|
|
"direction": "max",
|
|
"values_pct": [
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "已知问题被误拒率",
|
|
"direction": "min",
|
|
"values_pct": [
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "未知问题被误读率",
|
|
"direction": "min",
|
|
"values_pct": [
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "仲裁准确率",
|
|
"direction": "max",
|
|
"values_pct": [
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%"
|
|
],
|
|
"axis_passed": true
|
|
}
|
|
],
|
|
"sweep_dominates": true
|
|
},
|
|
"REPLAY-128 v7 final": {
|
|
"rows": [
|
|
{
|
|
"axis": "need F1",
|
|
"direction": "max",
|
|
"values_pct": [
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "need 召回",
|
|
"direction": "max",
|
|
"values_pct": [
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "need 精确率",
|
|
"direction": "max",
|
|
"values_pct": [
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "未知拒答率",
|
|
"direction": "max",
|
|
"values_pct": [
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "已知问题被误拒率",
|
|
"direction": "min",
|
|
"values_pct": [
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "未知问题被误读率",
|
|
"direction": "min",
|
|
"values_pct": [
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "仲裁准确率",
|
|
"direction": "max",
|
|
"values_pct": [
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%"
|
|
],
|
|
"axis_passed": true
|
|
}
|
|
],
|
|
"sweep_dominates": true
|
|
},
|
|
"V2-512 v6 final": {
|
|
"rows": [
|
|
{
|
|
"axis": "need F1",
|
|
"direction": "max",
|
|
"values_pct": [
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "need 召回",
|
|
"direction": "max",
|
|
"values_pct": [
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "need 精确率",
|
|
"direction": "max",
|
|
"values_pct": [
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "未知拒答率",
|
|
"direction": "max",
|
|
"values_pct": [
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "已知问题被误拒率",
|
|
"direction": "min",
|
|
"values_pct": [
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "未知问题被误读率",
|
|
"direction": "min",
|
|
"values_pct": [
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "仲裁准确率",
|
|
"direction": "max",
|
|
"values_pct": [
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%"
|
|
],
|
|
"axis_passed": true
|
|
}
|
|
],
|
|
"sweep_dominates": true
|
|
},
|
|
"XL-512 v6 final": {
|
|
"rows": [
|
|
{
|
|
"axis": "need F1",
|
|
"direction": "max",
|
|
"values_pct": [
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "need 召回",
|
|
"direction": "max",
|
|
"values_pct": [
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "need 精确率",
|
|
"direction": "max",
|
|
"values_pct": [
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "未知拒答率",
|
|
"direction": "max",
|
|
"values_pct": [
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "已知问题被误拒率",
|
|
"direction": "min",
|
|
"values_pct": [
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "未知问题被误读率",
|
|
"direction": "min",
|
|
"values_pct": [
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "仲裁准确率",
|
|
"direction": "max",
|
|
"values_pct": [
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%"
|
|
],
|
|
"axis_passed": true
|
|
}
|
|
],
|
|
"sweep_dominates": true
|
|
},
|
|
"XL-128 v6 final": {
|
|
"rows": [
|
|
{
|
|
"axis": "need F1",
|
|
"direction": "max",
|
|
"values_pct": [
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "need 召回",
|
|
"direction": "max",
|
|
"values_pct": [
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "need 精确率",
|
|
"direction": "max",
|
|
"values_pct": [
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "未知拒答率",
|
|
"direction": "max",
|
|
"values_pct": [
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "已知问题被误拒率",
|
|
"direction": "min",
|
|
"values_pct": [
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "未知问题被误读率",
|
|
"direction": "min",
|
|
"values_pct": [
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%",
|
|
"0.00%"
|
|
],
|
|
"axis_passed": true
|
|
},
|
|
{
|
|
"axis": "仲裁准确率",
|
|
"direction": "max",
|
|
"values_pct": [
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%",
|
|
"100.00%"
|
|
],
|
|
"axis_passed": true
|
|
}
|
|
],
|
|
"sweep_dominates": true
|
|
}
|
|
}
|
|
} |