- 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
1660 lines
52 KiB
JSON
1660 lines
52 KiB
JSON
{
|
|
"candidates": [
|
|
"V2-128 v6 best",
|
|
"V2-128 v6 final",
|
|
"XL-128 v6 best",
|
|
"XL-128 v6 final",
|
|
"V2-512 v6 best",
|
|
"XL-512 v6 best"
|
|
],
|
|
"baselines": [
|
|
"V2-128 deployed(v3)"
|
|
],
|
|
"passed_axes": 117,
|
|
"failed_axes": 15,
|
|
"skipped_axes": 0,
|
|
"dominates": false,
|
|
"per_candidate": {
|
|
"V2-128 v6 best": {
|
|
"passed": 22,
|
|
"failed": 0,
|
|
"dominates": true,
|
|
"failed_axes": []
|
|
},
|
|
"V2-128 v6 final": {
|
|
"passed": 22,
|
|
"failed": 0,
|
|
"dominates": true,
|
|
"failed_axes": []
|
|
},
|
|
"XL-128 v6 best": {
|
|
"passed": 18,
|
|
"failed": 4,
|
|
"dominates": false,
|
|
"failed_axes": [
|
|
"单查询延迟 ms (GPU)",
|
|
"路由 QPS (GPU, 单查询)",
|
|
"批量 QPS (GPU, batch=64)",
|
|
"批量 QPS (GPU, batch=256)"
|
|
]
|
|
},
|
|
"XL-128 v6 final": {
|
|
"passed": 18,
|
|
"failed": 4,
|
|
"dominates": false,
|
|
"failed_axes": [
|
|
"单查询延迟 ms (GPU)",
|
|
"路由 QPS (GPU, 单查询)",
|
|
"批量 QPS (GPU, batch=64)",
|
|
"批量 QPS (GPU, batch=256)"
|
|
]
|
|
},
|
|
"V2-512 v6 best": {
|
|
"passed": 20,
|
|
"failed": 2,
|
|
"dominates": false,
|
|
"failed_axes": [
|
|
"批量 QPS (GPU, batch=256)",
|
|
"地址字节/记录"
|
|
]
|
|
},
|
|
"XL-512 v6 best": {
|
|
"passed": 17,
|
|
"failed": 5,
|
|
"dominates": false,
|
|
"failed_axes": [
|
|
"单查询延迟 ms (GPU)",
|
|
"路由 QPS (GPU, 单查询)",
|
|
"批量 QPS (GPU, batch=64)",
|
|
"批量 QPS (GPU, batch=256)",
|
|
"地址字节/记录"
|
|
]
|
|
}
|
|
},
|
|
"rows": [
|
|
{
|
|
"axis": "Top-1 正确率",
|
|
"candidate": "V2-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.4112495789828225,
|
|
"candidate_value": 0.946165936903559,
|
|
"delta": 0.5349163579207366,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 130.07098007098008
|
|
},
|
|
{
|
|
"axis": "Top-1 正确率",
|
|
"candidate": "V2-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.4112495789828225,
|
|
"candidate_value": 0.9436959694622207,
|
|
"delta": 0.5324463904793981,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 129.47037947037944
|
|
},
|
|
{
|
|
"axis": "Top-1 正确率",
|
|
"candidate": "XL-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.4112495789828225,
|
|
"candidate_value": 0.9566071629055799,
|
|
"delta": 0.5453575839227573,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 132.6098826098826
|
|
},
|
|
{
|
|
"axis": "Top-1 正确率",
|
|
"candidate": "XL-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.4112495789828225,
|
|
"candidate_value": 0.9518917705175705,
|
|
"delta": 0.5406421915347479,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 131.46328146328145
|
|
},
|
|
{
|
|
"axis": "Top-1 正确率",
|
|
"candidate": "V2-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.4112495789828225,
|
|
"candidate_value": 0.950488379925901,
|
|
"delta": 0.5392388009430784,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 131.1220311220311
|
|
},
|
|
{
|
|
"axis": "Top-1 正确率",
|
|
"candidate": "XL-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.4112495789828225,
|
|
"candidate_value": 0.957842146626249,
|
|
"delta": 0.5465925676434265,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 132.9101829101829
|
|
},
|
|
{
|
|
"axis": "Recall@1",
|
|
"candidate": "V2-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.37027057370607386,
|
|
"candidate_value": 0.8919108566296171,
|
|
"delta": 0.5216402829235433,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 140.8808368708308
|
|
},
|
|
{
|
|
"axis": "Recall@1",
|
|
"candidate": "V2-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.37027057370607386,
|
|
"candidate_value": 0.8882059054676097,
|
|
"delta": 0.5179353317615358,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 139.88023044269252
|
|
},
|
|
{
|
|
"axis": "Recall@1",
|
|
"candidate": "XL-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.37027057370607386,
|
|
"candidate_value": 0.9027730998091389,
|
|
"delta": 0.532502526103065,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 143.81443298969074
|
|
},
|
|
{
|
|
"axis": "Recall@1",
|
|
"candidate": "XL-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.37027057370607386,
|
|
"candidate_value": 0.8963736387111261,
|
|
"delta": 0.5261030650050522,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 142.0861127956337
|
|
},
|
|
{
|
|
"axis": "Recall@1",
|
|
"candidate": "V2-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.37027057370607386,
|
|
"candidate_value": 0.8971034018187942,
|
|
"delta": 0.5268328281127204,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 142.28320194057005
|
|
},
|
|
{
|
|
"axis": "Recall@1",
|
|
"candidate": "XL-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.37027057370607386,
|
|
"candidate_value": 0.9024082182553048,
|
|
"delta": 0.532137644549231,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 143.71588841722257
|
|
},
|
|
{
|
|
"axis": "Recall@3",
|
|
"candidate": "V2-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.4673009992141013,
|
|
"candidate_value": 0.9780509711462895,
|
|
"delta": 0.5107499719321882,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 109.2978557270707
|
|
},
|
|
{
|
|
"axis": "Recall@3",
|
|
"candidate": "V2-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.4673009992141013,
|
|
"candidate_value": 0.9639609296059279,
|
|
"delta": 0.49665993039182665,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 106.28265961919634
|
|
},
|
|
{
|
|
"axis": "Recall@3",
|
|
"candidate": "XL-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.4673009992141013,
|
|
"candidate_value": 0.9874817559223082,
|
|
"delta": 0.520180756708207,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 111.31599495465193
|
|
},
|
|
{
|
|
"axis": "Recall@3",
|
|
"candidate": "XL-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.4673009992141013,
|
|
"candidate_value": 0.9745424946671157,
|
|
"delta": 0.5072414954530144,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 108.54705988347648
|
|
},
|
|
{
|
|
"axis": "Recall@3",
|
|
"candidate": "V2-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.4673009992141013,
|
|
"candidate_value": 0.9884079937128102,
|
|
"delta": 0.5211069944987089,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 111.51420505736081
|
|
},
|
|
{
|
|
"axis": "Recall@3",
|
|
"candidate": "XL-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.4673009992141013,
|
|
"candidate_value": 0.9758897496351184,
|
|
"delta": 0.5085887504210171,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 108.83536548741665
|
|
},
|
|
{
|
|
"axis": "Recall@5",
|
|
"candidate": "V2-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.4891377568204783,
|
|
"candidate_value": 0.984001347254968,
|
|
"delta": 0.49486359043448974,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 101.1705973489413
|
|
},
|
|
{
|
|
"axis": "Recall@5",
|
|
"candidate": "V2-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.4891377568204783,
|
|
"candidate_value": 0.973981138430448,
|
|
"delta": 0.4848433816099697,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 99.12205198829403
|
|
},
|
|
{
|
|
"axis": "Recall@5",
|
|
"candidate": "XL-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.4891377568204783,
|
|
"candidate_value": 0.9922532839339845,
|
|
"delta": 0.5031155271135062,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 102.85763470476846
|
|
},
|
|
{
|
|
"axis": "Recall@5",
|
|
"candidate": "XL-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.4891377568204783,
|
|
"candidate_value": 0.9820366004266308,
|
|
"delta": 0.4928988436061525,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 100.76892178803007
|
|
},
|
|
{
|
|
"axis": "Recall@5",
|
|
"candidate": "V2-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.4891377568204783,
|
|
"candidate_value": 0.9934882676546536,
|
|
"delta": 0.5043505108341753,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 103.11011648591264
|
|
},
|
|
{
|
|
"axis": "Recall@5",
|
|
"candidate": "XL-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.4891377568204783,
|
|
"candidate_value": 0.9834399910183003,
|
|
"delta": 0.494302234197822,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 101.05583290296667
|
|
},
|
|
{
|
|
"axis": "MRR",
|
|
"candidate": "V2-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.476908110965675,
|
|
"candidate_value": 0.9648813461092199,
|
|
"delta": 0.48797323514354496,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 102.32017949023022
|
|
},
|
|
{
|
|
"axis": "MRR",
|
|
"candidate": "V2-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.476908110965675,
|
|
"candidate_value": 0.9597054333626427,
|
|
"delta": 0.4827973223969677,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 101.23487340556399
|
|
},
|
|
{
|
|
"axis": "MRR",
|
|
"candidate": "XL-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.476908110965675,
|
|
"candidate_value": 0.9737093504776227,
|
|
"delta": 0.49680123951194777,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 104.17127075192573
|
|
},
|
|
{
|
|
"axis": "MRR",
|
|
"candidate": "XL-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.476908110965675,
|
|
"candidate_value": 0.9670023550641874,
|
|
"delta": 0.49009424409851243,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 102.76492113043273
|
|
},
|
|
{
|
|
"axis": "MRR",
|
|
"candidate": "V2-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.476908110965675,
|
|
"candidate_value": 0.9711571604054701,
|
|
"delta": 0.4942490494397951,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 103.63611733065454
|
|
},
|
|
{
|
|
"axis": "MRR",
|
|
"candidate": "XL-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.476908110965675,
|
|
"candidate_value": 0.9724377931581535,
|
|
"delta": 0.4955296821924785,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 103.90464552785595
|
|
},
|
|
{
|
|
"axis": "nDCG@3",
|
|
"candidate": "V2-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.44220238482445295,
|
|
"candidate_value": 0.9649494672503449,
|
|
"delta": 0.522747082425892,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 118.21444215716157
|
|
},
|
|
{
|
|
"axis": "nDCG@3",
|
|
"candidate": "V2-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.44220238482445295,
|
|
"candidate_value": 0.9530820526370456,
|
|
"delta": 0.5108796678125926,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 115.53073555118962
|
|
},
|
|
{
|
|
"axis": "nDCG@3",
|
|
"candidate": "XL-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.44220238482445295,
|
|
"candidate_value": 0.975110976145454,
|
|
"delta": 0.532908591321001,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 120.51237388341016
|
|
},
|
|
{
|
|
"axis": "nDCG@3",
|
|
"candidate": "XL-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.44220238482445295,
|
|
"candidate_value": 0.9642642772784563,
|
|
"delta": 0.5220618924540034,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 118.05949275041866
|
|
},
|
|
{
|
|
"axis": "nDCG@3",
|
|
"candidate": "V2-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.44220238482445295,
|
|
"candidate_value": 0.9723689530227545,
|
|
"delta": 0.5301665681983015,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 119.89229058743518
|
|
},
|
|
{
|
|
"axis": "nDCG@3",
|
|
"candidate": "XL-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.44220238482445295,
|
|
"candidate_value": 0.9664212737428025,
|
|
"delta": 0.5242188889183496,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 118.54727765126273
|
|
},
|
|
{
|
|
"axis": "多跳证据全中(Top-3)",
|
|
"candidate": "V2-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.43426518468620184,
|
|
"candidate_value": 0.9774896149096216,
|
|
"delta": 0.5432244302234197,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 125.09048603929678
|
|
},
|
|
{
|
|
"axis": "多跳证据全中(Top-3)",
|
|
"candidate": "V2-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.43426518468620184,
|
|
"candidate_value": 0.9595262153362524,
|
|
"delta": 0.5252610306500505,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 120.95398138572907
|
|
},
|
|
{
|
|
"axis": "多跳证据全中(Top-3)",
|
|
"candidate": "XL-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.43426518468620184,
|
|
"candidate_value": 0.9869765353093073,
|
|
"delta": 0.5527113506231054,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 127.27507755946226
|
|
},
|
|
{
|
|
"axis": "多跳证据全中(Top-3)",
|
|
"candidate": "XL-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.43426518468620184,
|
|
"candidate_value": 0.9728864937689458,
|
|
"delta": 0.5386213090827439,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 124.03050672182006
|
|
},
|
|
{
|
|
"axis": "多跳证据全中(Top-3)",
|
|
"candidate": "V2-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.43426518468620184,
|
|
"candidate_value": 0.9863590434489727,
|
|
"delta": 0.5520938587627708,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 127.13288521199586
|
|
},
|
|
{
|
|
"axis": "多跳证据全中(Top-3)",
|
|
"candidate": "XL-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.43426518468620184,
|
|
"candidate_value": 0.9686201863702706,
|
|
"delta": 0.5343550016840688,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 123.04808686659776
|
|
},
|
|
{
|
|
"axis": "多跳证据全中(仅多正例)",
|
|
"candidate": "V2-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.3715,
|
|
"candidate_value": 0.99,
|
|
"delta": 0.6185,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 166.4872139973082
|
|
},
|
|
{
|
|
"axis": "多跳证据全中(仅多正例)",
|
|
"candidate": "V2-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.3715,
|
|
"candidate_value": 0.921,
|
|
"delta": 0.5495000000000001,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 147.91386271870797
|
|
},
|
|
{
|
|
"axis": "多跳证据全中(仅多正例)",
|
|
"candidate": "XL-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.3715,
|
|
"candidate_value": 0.991,
|
|
"delta": 0.6194999999999999,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 166.75639300134588
|
|
},
|
|
{
|
|
"axis": "多跳证据全中(仅多正例)",
|
|
"candidate": "XL-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.3715,
|
|
"candidate_value": 0.9705,
|
|
"delta": 0.599,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 161.23822341857334
|
|
},
|
|
{
|
|
"axis": "多跳证据全中(仅多正例)",
|
|
"candidate": "V2-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.3715,
|
|
"candidate_value": 0.9635,
|
|
"delta": 0.5920000000000001,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 159.3539703903096
|
|
},
|
|
{
|
|
"axis": "多跳证据全中(仅多正例)",
|
|
"candidate": "XL-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.3715,
|
|
"candidate_value": 0.8705,
|
|
"delta": 0.49900000000000005,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 134.32032301480484
|
|
},
|
|
{
|
|
"axis": "hop 正确率",
|
|
"candidate": "V2-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.7322992700729927,
|
|
"candidate_value": 1.0,
|
|
"delta": 0.26770072992700733,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 36.556192374781965
|
|
},
|
|
{
|
|
"axis": "hop 正确率",
|
|
"candidate": "V2-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.7322992700729927,
|
|
"candidate_value": 1.0,
|
|
"delta": 0.26770072992700733,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 36.556192374781965
|
|
},
|
|
{
|
|
"axis": "hop 正确率",
|
|
"candidate": "XL-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.7322992700729927,
|
|
"candidate_value": 0.9995437956204379,
|
|
"delta": 0.26724452554744527,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 36.493894841764266
|
|
},
|
|
{
|
|
"axis": "hop 正确率",
|
|
"candidate": "XL-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.7322992700729927,
|
|
"candidate_value": 1.0,
|
|
"delta": 0.26770072992700733,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 36.556192374781965
|
|
},
|
|
{
|
|
"axis": "hop 正确率",
|
|
"candidate": "V2-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.7322992700729927,
|
|
"candidate_value": 1.0,
|
|
"delta": 0.26770072992700733,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 36.556192374781965
|
|
},
|
|
{
|
|
"axis": "hop 正确率",
|
|
"candidate": "XL-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.7322992700729927,
|
|
"candidate_value": 1.0,
|
|
"delta": 0.26770072992700733,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 36.556192374781965
|
|
},
|
|
{
|
|
"axis": "hop 欠预测率",
|
|
"candidate": "V2-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.046817110138093636,
|
|
"candidate_value": 0.0,
|
|
"delta": -0.046817110138093636,
|
|
"direction": "min",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": -100.0
|
|
},
|
|
{
|
|
"axis": "hop 欠预测率",
|
|
"candidate": "V2-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.046817110138093636,
|
|
"candidate_value": 0.0,
|
|
"delta": -0.046817110138093636,
|
|
"direction": "min",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": -100.0
|
|
},
|
|
{
|
|
"axis": "hop 欠预测率",
|
|
"candidate": "XL-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.046817110138093636,
|
|
"candidate_value": 0.0,
|
|
"delta": -0.046817110138093636,
|
|
"direction": "min",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": -100.0
|
|
},
|
|
{
|
|
"axis": "hop 欠预测率",
|
|
"candidate": "XL-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.046817110138093636,
|
|
"candidate_value": 0.0,
|
|
"delta": -0.046817110138093636,
|
|
"direction": "min",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": -100.0
|
|
},
|
|
{
|
|
"axis": "hop 欠预测率",
|
|
"candidate": "V2-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.046817110138093636,
|
|
"candidate_value": 0.0,
|
|
"delta": -0.046817110138093636,
|
|
"direction": "min",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": -100.0
|
|
},
|
|
{
|
|
"axis": "hop 欠预测率",
|
|
"candidate": "XL-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.046817110138093636,
|
|
"candidate_value": 0.0,
|
|
"delta": -0.046817110138093636,
|
|
"direction": "min",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": -100.0
|
|
},
|
|
{
|
|
"axis": "need F1 (门槛 0.50)",
|
|
"candidate": "V2-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.895773512669387,
|
|
"candidate_value": 1.0,
|
|
"delta": 0.10422648733061302,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 11.635361601619609
|
|
},
|
|
{
|
|
"axis": "need F1 (门槛 0.50)",
|
|
"candidate": "V2-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.895773512669387,
|
|
"candidate_value": 1.0,
|
|
"delta": 0.10422648733061302,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 11.635361601619609
|
|
},
|
|
{
|
|
"axis": "need F1 (门槛 0.50)",
|
|
"candidate": "XL-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.895773512669387,
|
|
"candidate_value": 0.9995230747650443,
|
|
"delta": 0.10374956209565733,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 11.582119880558393
|
|
},
|
|
{
|
|
"axis": "need F1 (门槛 0.50)",
|
|
"candidate": "XL-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.895773512669387,
|
|
"candidate_value": 1.0,
|
|
"delta": 0.10422648733061302,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 11.635361601619609
|
|
},
|
|
{
|
|
"axis": "need F1 (门槛 0.50)",
|
|
"candidate": "V2-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.895773512669387,
|
|
"candidate_value": 0.999775507913346,
|
|
"delta": 0.10400199524395903,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 11.61030034634929
|
|
},
|
|
{
|
|
"axis": "need F1 (门槛 0.50)",
|
|
"candidate": "XL-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.895773512669387,
|
|
"candidate_value": 1.0,
|
|
"delta": 0.10422648733061302,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 11.635361601619609
|
|
},
|
|
{
|
|
"axis": "need 召回 (门槛 0.50)",
|
|
"candidate": "V2-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.998203660042663,
|
|
"candidate_value": 1.0,
|
|
"delta": 0.0017963399573369676,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 0.1799572601507184
|
|
},
|
|
{
|
|
"axis": "need 召回 (门槛 0.50)",
|
|
"candidate": "V2-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.998203660042663,
|
|
"candidate_value": 1.0,
|
|
"delta": 0.0017963399573369676,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 0.1799572601507184
|
|
},
|
|
{
|
|
"axis": "need 召回 (门槛 0.50)",
|
|
"candidate": "XL-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.998203660042663,
|
|
"candidate_value": 1.0,
|
|
"delta": 0.0017963399573369676,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 0.1799572601507184
|
|
},
|
|
{
|
|
"axis": "need 召回 (门槛 0.50)",
|
|
"candidate": "XL-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.998203660042663,
|
|
"candidate_value": 1.0,
|
|
"delta": 0.0017963399573369676,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 0.1799572601507184
|
|
},
|
|
{
|
|
"axis": "need 召回 (门槛 0.50)",
|
|
"candidate": "V2-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.998203660042663,
|
|
"candidate_value": 1.0,
|
|
"delta": 0.0017963399573369676,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 0.1799572601507184
|
|
},
|
|
{
|
|
"axis": "need 召回 (门槛 0.50)",
|
|
"candidate": "XL-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.998203660042663,
|
|
"candidate_value": 1.0,
|
|
"delta": 0.0017963399573369676,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 0.1799572601507184
|
|
},
|
|
{
|
|
"axis": "未知拒答率 (门槛 0.50)",
|
|
"candidate": "V2-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.0,
|
|
"candidate_value": 1.0,
|
|
"delta": 1.0,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": null
|
|
},
|
|
{
|
|
"axis": "未知拒答率 (门槛 0.50)",
|
|
"candidate": "V2-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.0,
|
|
"candidate_value": 1.0,
|
|
"delta": 1.0,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": null
|
|
},
|
|
{
|
|
"axis": "未知拒答率 (门槛 0.50)",
|
|
"candidate": "XL-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.0,
|
|
"candidate_value": 0.995859717486605,
|
|
"delta": 0.995859717486605,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": null
|
|
},
|
|
{
|
|
"axis": "未知拒答率 (门槛 0.50)",
|
|
"candidate": "XL-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.0,
|
|
"candidate_value": 1.0,
|
|
"delta": 1.0,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": null
|
|
},
|
|
{
|
|
"axis": "未知拒答率 (门槛 0.50)",
|
|
"candidate": "V2-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.0,
|
|
"candidate_value": 0.9980516317584024,
|
|
"delta": 0.9980516317584024,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": null
|
|
},
|
|
{
|
|
"axis": "未知拒答率 (门槛 0.50)",
|
|
"candidate": "XL-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.0,
|
|
"candidate_value": 1.0,
|
|
"delta": 1.0,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": null
|
|
},
|
|
{
|
|
"axis": "已知问题被误拒率",
|
|
"candidate": "V2-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.001796339957336926,
|
|
"candidate_value": 0.0,
|
|
"delta": -0.001796339957336926,
|
|
"direction": "min",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": -100.0
|
|
},
|
|
{
|
|
"axis": "已知问题被误拒率",
|
|
"candidate": "V2-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.001796339957336926,
|
|
"candidate_value": 0.0,
|
|
"delta": -0.001796339957336926,
|
|
"direction": "min",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": -100.0
|
|
},
|
|
{
|
|
"axis": "已知问题被误拒率",
|
|
"candidate": "XL-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.001796339957336926,
|
|
"candidate_value": 0.0,
|
|
"delta": -0.001796339957336926,
|
|
"direction": "min",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": -100.0
|
|
},
|
|
{
|
|
"axis": "已知问题被误拒率",
|
|
"candidate": "XL-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.001796339957336926,
|
|
"candidate_value": 0.0,
|
|
"delta": -0.001796339957336926,
|
|
"direction": "min",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": -100.0
|
|
},
|
|
{
|
|
"axis": "已知问题被误拒率",
|
|
"candidate": "V2-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.001796339957336926,
|
|
"candidate_value": 0.0,
|
|
"delta": -0.001796339957336926,
|
|
"direction": "min",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": -100.0
|
|
},
|
|
{
|
|
"axis": "已知问题被误拒率",
|
|
"candidate": "XL-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.001796339957336926,
|
|
"candidate_value": 0.0,
|
|
"delta": -0.001796339957336926,
|
|
"direction": "min",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": -100.0
|
|
},
|
|
{
|
|
"axis": "未知问题被误读率",
|
|
"candidate": "V2-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 1.0,
|
|
"candidate_value": 0.0,
|
|
"delta": -1.0,
|
|
"direction": "min",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": -100.0
|
|
},
|
|
{
|
|
"axis": "未知问题被误读率",
|
|
"candidate": "V2-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 1.0,
|
|
"candidate_value": 0.0,
|
|
"delta": -1.0,
|
|
"direction": "min",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": -100.0
|
|
},
|
|
{
|
|
"axis": "未知问题被误读率",
|
|
"candidate": "XL-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 1.0,
|
|
"candidate_value": 0.004140282513395032,
|
|
"delta": -0.995859717486605,
|
|
"direction": "min",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": -99.5859717486605
|
|
},
|
|
{
|
|
"axis": "未知问题被误读率",
|
|
"candidate": "XL-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 1.0,
|
|
"candidate_value": 0.0,
|
|
"delta": -1.0,
|
|
"direction": "min",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": -100.0
|
|
},
|
|
{
|
|
"axis": "未知问题被误读率",
|
|
"candidate": "V2-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 1.0,
|
|
"candidate_value": 0.001948368241597662,
|
|
"delta": -0.9980516317584024,
|
|
"direction": "min",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": -99.80516317584024
|
|
},
|
|
{
|
|
"axis": "未知问题被误读率",
|
|
"candidate": "XL-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 1.0,
|
|
"candidate_value": 0.0,
|
|
"delta": -1.0,
|
|
"direction": "min",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": -100.0
|
|
},
|
|
{
|
|
"axis": "仲裁准确率",
|
|
"candidate": "V2-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.8112226277372263,
|
|
"candidate_value": 1.0,
|
|
"delta": 0.1887773722627737,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 23.27072320323923
|
|
},
|
|
{
|
|
"axis": "仲裁准确率",
|
|
"candidate": "V2-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.8112226277372263,
|
|
"candidate_value": 1.0,
|
|
"delta": 0.1887773722627737,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 23.27072320323923
|
|
},
|
|
{
|
|
"axis": "仲裁准确率",
|
|
"candidate": "XL-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.8112226277372263,
|
|
"candidate_value": 0.9992244525547446,
|
|
"delta": 0.18800182481751826,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 23.175120908784162
|
|
},
|
|
{
|
|
"axis": "仲裁准确率",
|
|
"candidate": "XL-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.8112226277372263,
|
|
"candidate_value": 1.0,
|
|
"delta": 0.1887773722627737,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 23.27072320323923
|
|
},
|
|
{
|
|
"axis": "仲裁准确率",
|
|
"candidate": "V2-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.8112226277372263,
|
|
"candidate_value": 0.9996350364963503,
|
|
"delta": 0.18841240875912402,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 23.225733888201542
|
|
},
|
|
{
|
|
"axis": "仲裁准确率",
|
|
"candidate": "XL-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 0.8112226277372263,
|
|
"candidate_value": 1.0,
|
|
"delta": 0.1887773722627737,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 23.27072320323923
|
|
},
|
|
{
|
|
"axis": "平均分数余量",
|
|
"candidate": "V2-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": -9.29892828391031,
|
|
"candidate_value": 14.959203617273067,
|
|
"delta": 24.258131901183376,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": -260.87019020413976
|
|
},
|
|
{
|
|
"axis": "平均分数余量",
|
|
"candidate": "V2-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": -9.29892828391031,
|
|
"candidate_value": 51.315328869764656,
|
|
"delta": 60.61425715367497,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": -651.8413230323995
|
|
},
|
|
{
|
|
"axis": "平均分数余量",
|
|
"candidate": "XL-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": -9.29892828391031,
|
|
"candidate_value": 16.798009918724365,
|
|
"delta": 26.096938202634675,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": -280.6445797392536
|
|
},
|
|
{
|
|
"axis": "平均分数余量",
|
|
"candidate": "XL-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": -9.29892828391031,
|
|
"candidate_value": 200.33457717921095,
|
|
"delta": 209.63350546312125,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": -2254.3835059556764
|
|
},
|
|
{
|
|
"axis": "平均分数余量",
|
|
"candidate": "V2-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": -9.29892828391031,
|
|
"candidate_value": 12.36926158705623,
|
|
"delta": 21.66818987096654,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": -233.0181415471118
|
|
},
|
|
{
|
|
"axis": "平均分数余量",
|
|
"candidate": "XL-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": -9.29892828391031,
|
|
"candidate_value": 45.56406442258217,
|
|
"delta": 54.86299270649248,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": -589.9926425007542
|
|
},
|
|
{
|
|
"axis": "单查询延迟 ms (GPU)",
|
|
"candidate": "V2-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 1.155849999122438,
|
|
"candidate_value": 1.167899999927613,
|
|
"delta": 0.012050000805174932,
|
|
"direction": "min",
|
|
"pass": true,
|
|
"relative_tolerance_applied": true,
|
|
"relative_delta_pct": 1.0425228891572191
|
|
},
|
|
{
|
|
"axis": "单查询延迟 ms (GPU)",
|
|
"candidate": "V2-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 1.155849999122438,
|
|
"candidate_value": 1.165349998700549,
|
|
"delta": 0.009499999578110874,
|
|
"direction": "min",
|
|
"pass": true,
|
|
"relative_tolerance_applied": true,
|
|
"relative_delta_pct": 0.8219059207789599
|
|
},
|
|
{
|
|
"axis": "单查询延迟 ms (GPU)",
|
|
"candidate": "XL-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 1.155849999122438,
|
|
"candidate_value": 1.578349998453632,
|
|
"delta": 0.422499999331194,
|
|
"direction": "min",
|
|
"pass": false,
|
|
"relative_tolerance_applied": true,
|
|
"relative_delta_pct": 36.55318593692701
|
|
},
|
|
{
|
|
"axis": "单查询延迟 ms (GPU)",
|
|
"candidate": "XL-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 1.155849999122438,
|
|
"candidate_value": 1.5958999993017642,
|
|
"delta": 0.44005000017932616,
|
|
"direction": "min",
|
|
"pass": false,
|
|
"relative_tolerance_applied": true,
|
|
"relative_delta_pct": 38.07154912085717
|
|
},
|
|
{
|
|
"axis": "单查询延迟 ms (GPU)",
|
|
"candidate": "V2-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 1.155849999122438,
|
|
"candidate_value": 1.162250000561471,
|
|
"delta": 0.006400001439033076,
|
|
"direction": "min",
|
|
"pass": true,
|
|
"relative_tolerance_applied": true,
|
|
"relative_delta_pct": 0.553705190456563
|
|
},
|
|
{
|
|
"axis": "单查询延迟 ms (GPU)",
|
|
"candidate": "XL-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 1.155849999122438,
|
|
"candidate_value": 1.5855000019655563,
|
|
"delta": 0.4296500028431183,
|
|
"direction": "min",
|
|
"pass": false,
|
|
"relative_tolerance_applied": true,
|
|
"relative_delta_pct": 37.17177861913948
|
|
},
|
|
{
|
|
"axis": "路由 QPS (GPU, 单查询)",
|
|
"candidate": "V2-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 865.1641655571528,
|
|
"candidate_value": 856.2376916362535,
|
|
"delta": -8.926473920899298,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": true,
|
|
"relative_delta_pct": -1.031766487363792
|
|
},
|
|
{
|
|
"axis": "路由 QPS (GPU, 单查询)",
|
|
"candidate": "V2-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 865.1641655571528,
|
|
"candidate_value": 858.1112979920828,
|
|
"delta": -7.052867565070073,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": true,
|
|
"relative_delta_pct": -0.8152056968896917
|
|
},
|
|
{
|
|
"axis": "路由 QPS (GPU, 单查询)",
|
|
"candidate": "XL-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 865.1641655571528,
|
|
"candidate_value": 633.5730357523598,
|
|
"delta": -231.591129804793,
|
|
"direction": "max",
|
|
"pass": false,
|
|
"relative_tolerance_applied": true,
|
|
"relative_delta_pct": -26.768460718163446
|
|
},
|
|
{
|
|
"axis": "路由 QPS (GPU, 单查询)",
|
|
"candidate": "XL-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 865.1641655571528,
|
|
"candidate_value": 626.6056773215856,
|
|
"delta": -238.5584882355672,
|
|
"direction": "max",
|
|
"pass": false,
|
|
"relative_tolerance_applied": true,
|
|
"relative_delta_pct": -27.573782841773053
|
|
},
|
|
{
|
|
"axis": "路由 QPS (GPU, 单查询)",
|
|
"candidate": "V2-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 865.1641655571528,
|
|
"candidate_value": 860.4000856243581,
|
|
"delta": -4.764079932794743,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": true,
|
|
"relative_delta_pct": -0.550656178614003
|
|
},
|
|
{
|
|
"axis": "路由 QPS (GPU, 单查询)",
|
|
"candidate": "XL-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 865.1641655571528,
|
|
"candidate_value": 630.7158617220388,
|
|
"delta": -234.44830383511407,
|
|
"direction": "max",
|
|
"pass": false,
|
|
"relative_tolerance_applied": true,
|
|
"relative_delta_pct": -27.09870717820737
|
|
},
|
|
{
|
|
"axis": "批量 QPS (GPU, batch=64)",
|
|
"candidate": "V2-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 54038.53621622001,
|
|
"candidate_value": 54149.18098951274,
|
|
"delta": 110.64477329272631,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": true,
|
|
"relative_delta_pct": 0.20475161068392445
|
|
},
|
|
{
|
|
"axis": "批量 QPS (GPU, batch=64)",
|
|
"candidate": "V2-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 54038.53621622001,
|
|
"candidate_value": 53471.46796235134,
|
|
"delta": -567.0682538686742,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": true,
|
|
"relative_delta_pct": -1.0493775249568382
|
|
},
|
|
{
|
|
"axis": "批量 QPS (GPU, batch=64)",
|
|
"candidate": "XL-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 54038.53621622001,
|
|
"candidate_value": 38630.547104467114,
|
|
"delta": -15407.989111752897,
|
|
"direction": "max",
|
|
"pass": false,
|
|
"relative_tolerance_applied": true,
|
|
"relative_delta_pct": -28.512965358836073
|
|
},
|
|
{
|
|
"axis": "批量 QPS (GPU, batch=64)",
|
|
"candidate": "XL-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 54038.53621622001,
|
|
"candidate_value": 36843.43841344889,
|
|
"delta": -17195.097802771124,
|
|
"direction": "max",
|
|
"pass": false,
|
|
"relative_tolerance_applied": true,
|
|
"relative_delta_pct": -31.820065839625588
|
|
},
|
|
{
|
|
"axis": "批量 QPS (GPU, batch=64)",
|
|
"candidate": "V2-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 54038.53621622001,
|
|
"candidate_value": 53757.11860745415,
|
|
"delta": -281.4176087658634,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": true,
|
|
"relative_delta_pct": -0.5207720794653836
|
|
},
|
|
{
|
|
"axis": "批量 QPS (GPU, batch=64)",
|
|
"candidate": "XL-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 54038.53621622001,
|
|
"candidate_value": 39909.70428801726,
|
|
"delta": -14128.831928202751,
|
|
"direction": "max",
|
|
"pass": false,
|
|
"relative_tolerance_applied": true,
|
|
"relative_delta_pct": -26.145845016360553
|
|
},
|
|
{
|
|
"axis": "批量 QPS (GPU, batch=256)",
|
|
"candidate": "V2-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 199554.12120495233,
|
|
"candidate_value": 196829.20449441153,
|
|
"delta": -2724.916710540798,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": true,
|
|
"relative_delta_pct": -1.365502598536749
|
|
},
|
|
{
|
|
"axis": "批量 QPS (GPU, batch=256)",
|
|
"candidate": "V2-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 199554.12120495233,
|
|
"candidate_value": 198025.9290094514,
|
|
"delta": -1528.1921955009166,
|
|
"direction": "max",
|
|
"pass": true,
|
|
"relative_tolerance_applied": true,
|
|
"relative_delta_pct": -0.7658033751813047
|
|
},
|
|
{
|
|
"axis": "批量 QPS (GPU, batch=256)",
|
|
"candidate": "XL-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 199554.12120495233,
|
|
"candidate_value": 141500.569342867,
|
|
"delta": -58053.551862085325,
|
|
"direction": "max",
|
|
"pass": false,
|
|
"relative_tolerance_applied": true,
|
|
"relative_delta_pct": -29.091632641583658
|
|
},
|
|
{
|
|
"axis": "批量 QPS (GPU, batch=256)",
|
|
"candidate": "XL-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 199554.12120495233,
|
|
"candidate_value": 134976.95897715187,
|
|
"delta": -64577.162227800465,
|
|
"direction": "max",
|
|
"pass": false,
|
|
"relative_tolerance_applied": true,
|
|
"relative_delta_pct": -32.360725921303526
|
|
},
|
|
{
|
|
"axis": "批量 QPS (GPU, batch=256)",
|
|
"candidate": "V2-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 199554.12120495233,
|
|
"candidate_value": 154085.05975293368,
|
|
"delta": -45469.06145201865,
|
|
"direction": "max",
|
|
"pass": false,
|
|
"relative_tolerance_applied": true,
|
|
"relative_delta_pct": -22.785328199420942
|
|
},
|
|
{
|
|
"axis": "批量 QPS (GPU, batch=256)",
|
|
"candidate": "XL-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 199554.12120495233,
|
|
"candidate_value": 94647.25411335309,
|
|
"delta": -104906.86709159924,
|
|
"direction": "max",
|
|
"pass": false,
|
|
"relative_tolerance_applied": true,
|
|
"relative_delta_pct": -52.570634200961706
|
|
},
|
|
{
|
|
"axis": "地址字节/记录",
|
|
"candidate": "V2-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 512,
|
|
"candidate_value": 512,
|
|
"delta": 0,
|
|
"direction": "min",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 0.0
|
|
},
|
|
{
|
|
"axis": "地址字节/记录",
|
|
"candidate": "V2-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 512,
|
|
"candidate_value": 512,
|
|
"delta": 0,
|
|
"direction": "min",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 0.0
|
|
},
|
|
{
|
|
"axis": "地址字节/记录",
|
|
"candidate": "XL-128 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 512,
|
|
"candidate_value": 512,
|
|
"delta": 0,
|
|
"direction": "min",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 0.0
|
|
},
|
|
{
|
|
"axis": "地址字节/记录",
|
|
"candidate": "XL-128 v6 final",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 512,
|
|
"candidate_value": 512,
|
|
"delta": 0,
|
|
"direction": "min",
|
|
"pass": true,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 0.0
|
|
},
|
|
{
|
|
"axis": "地址字节/记录",
|
|
"candidate": "V2-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 512,
|
|
"candidate_value": 2048,
|
|
"delta": 1536,
|
|
"direction": "min",
|
|
"pass": false,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 300.0
|
|
},
|
|
{
|
|
"axis": "地址字节/记录",
|
|
"candidate": "XL-512 v6 best",
|
|
"baseline": "V2-128 deployed(v3)",
|
|
"baseline_value": 512,
|
|
"candidate_value": 2048,
|
|
"delta": 1536,
|
|
"direction": "min",
|
|
"pass": false,
|
|
"relative_tolerance_applied": false,
|
|
"relative_delta_pct": 300.0
|
|
}
|
|
]
|
|
} |