{ "bench_version": 1, "track": "correlated_token_moe_regression", "model": "custom_moe", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.006, "aux": 0.01, "balanced": false, "K": 4 }, "sweep": [ { "cfg": { "lr": 0.0015, "aux": 0.0, "balanced": false, "K": 4 }, "mean": 0.0866615679115057, "per_seed": [ 0.08718974888324738, 0.09299269318580627, 0.08348067104816437, 0.08298315852880478 ] }, { "cfg": { "lr": 0.0015, "aux": 0.01, "balanced": false, "K": 4 }, "mean": 0.08026662468910217, "per_seed": [ 0.06962176412343979, 0.09358783811330795, 0.08609369397163391, 0.07176320254802704 ] }, { "cfg": { "lr": 0.0015, "aux": 0.1, "balanced": false, "K": 4 }, "mean": 0.0802982971072197, "per_seed": [ 0.06983038038015366, 0.09358862042427063, 0.08610628545284271, 0.07166790217161179 ] }, { "cfg": { "lr": 0.003, "aux": 0.0, "balanced": false, "K": 4 }, "mean": 0.07278427481651306, "per_seed": [ 0.07727840542793274, 0.07310336083173752, 0.07483050227165222, 0.06592483073472977 ] }, { "cfg": { "lr": 0.003, "aux": 0.01, "balanced": false, "K": 4 }, "mean": 0.06707096565514803, "per_seed": [ 0.06202782317996025, 0.07177857309579849, 0.07393620163202286, 0.060541264712810516 ] }, { "cfg": { "lr": 0.003, "aux": 0.1, "balanced": false, "K": 4 }, "mean": 0.06711035128682852, "per_seed": [ 0.062259428203105927, 0.07183852046728134, 0.0737687274813652, 0.0605747289955616 ] }, { "cfg": { "lr": 0.006, "aux": 0.0, "balanced": false, "K": 4 }, "mean": 0.05864693410694599, "per_seed": [ 0.062149204313755035, 0.05799993872642517, 0.06060343608260155, 0.05383515730500221 ] }, { "cfg": { "lr": 0.006, "aux": 0.01, "balanced": false, "K": 4 }, "mean": 0.056222427636384964, "per_seed": [ 0.049452465027570724, 0.058252543210983276, 0.05263408645987511, 0.06455061584711075 ] }, { "cfg": { "lr": 0.006, "aux": 0.1, "balanced": false, "K": 4 }, "mean": 0.0565244285389781, "per_seed": [ 0.04950065538287163, 0.0584375374019146, 0.052471455186605453, 0.06568806618452072 ] } ], "full": { "per_seed": [ 0.049452465027570724, 0.058252543210983276, 0.05263408645987511, 0.06455061584711075, 0.053144052624702454, 0.07924187928438187, 0.04448141157627106, 0.05137491226196289 ], "mean": 0.056641495786607265, "records": [ { "seed": 0, "metric": 0.049452465027570724, "routing": { "load_cv": 0.0913184674641444, "loads": [ 346.0, 426.0, 354.0, 395.0, 422.0, 452.0, 373.0, 432.0 ], "max_shell_count_spread": 311, "reassigned_fraction": 0.0, "shells": 4 } }, { "seed": 1, "metric": 0.058252543210983276, "routing": { "load_cv": 0.13504628836069504, "loads": [ 382.0, 403.0, 373.0, 531.0, 386.0, 400.0, 395.0, 330.0 ], "max_shell_count_spread": 184, "reassigned_fraction": 0.0, "shells": 4 } }, { "seed": 2, "metric": 0.05263408645987511, "routing": { "load_cv": 0.12644168616401794, "loads": [ 397.0, 428.0, 348.0, 306.0, 433.0, 424.0, 385.0, 479.0 ], "max_shell_count_spread": 159, "reassigned_fraction": 0.0, "shells": 4 } }, { "seed": 3, "metric": 0.06455061584711075, "routing": { "load_cv": 0.06754628043053149, "loads": [ 398.0, 374.0, 455.0, 379.0, 394.0, 387.0, 433.0, 380.0 ], "max_shell_count_spread": 269, "reassigned_fraction": 0.0, "shells": 4 } }, { "seed": 4, "metric": 0.053144052624702454, "routing": { "load_cv": 0.1465595015684756, "loads": [ 412.0, 382.0, 296.0, 394.0, 490.0, 373.0, 372.0, 481.0 ], "max_shell_count_spread": 297, "reassigned_fraction": 0.0, "shells": 4 } }, { "seed": 5, "metric": 0.07924187928438187, "routing": { "load_cv": 0.17972200755611428, "loads": [ 437.0, 391.0, 372.0, 424.0, 316.0, 562.0, 353.0, 345.0 ], "max_shell_count_spread": 321, "reassigned_fraction": 0.0, "shells": 4 } }, { "seed": 6, "metric": 0.04448141157627106, "routing": { "load_cv": 0.08075812652606547, "loads": [ 360.0, 407.0, 417.0, 424.0, 348.0, 376.0, 445.0, 423.0 ], "max_shell_count_spread": 264, "reassigned_fraction": 0.0, "shells": 4 } }, { "seed": 7, "metric": 0.05137491226196289, "routing": { "load_cv": 0.10753633107001559, "loads": [ 341.0, 418.0, 375.0, 380.0, 347.0, 427.0, 465.0, 447.0 ], "max_shell_count_spread": 262, "reassigned_fraction": 0.0, "shells": 4 } } ], "cfg": { "lr": 0.006, "aux": 0.01, "balanced": false, "K": 4 } } }, "idea": { "per_seed": [ 0.05467335134744644, 0.05477738380432129, 0.056872498244047165, 0.051517780870199203, 0.05167350545525551, 0.06755945831537247, 0.04719924181699753, 0.04966289922595024 ], "mean": 0.05424201488494873, "records": [ { "seed": 0, "metric": 0.05467335134744644, "routing": { "load_cv": 0.0, "loads": [ 400.0, 400.0, 400.0, 400.0, 400.0, 400.0, 400.0, 400.0 ], "max_shell_count_spread": 0, "reassigned_fraction": 0.61625, "shells": 8 } }, { "seed": 1, "metric": 0.05477738380432129, "routing": { "load_cv": 0.0, "loads": [ 400.0, 400.0, 400.0, 400.0, 400.0, 400.0, 400.0, 400.0 ], "max_shell_count_spread": 0, "reassigned_fraction": 0.3775, "shells": 8 } }, { "seed": 2, "metric": 0.056872498244047165, "routing": { "load_cv": 0.0, "loads": [ 400.0, 400.0, 400.0, 400.0, 400.0, 400.0, 400.0, 400.0 ], "max_shell_count_spread": 0, "reassigned_fraction": 0.5353125, "shells": 8 } }, { "seed": 3, "metric": 0.051517780870199203, "routing": { "load_cv": 0.0, "loads": [ 400.0, 400.0, 400.0, 400.0, 400.0, 400.0, 400.0, 400.0 ], "max_shell_count_spread": 0, "reassigned_fraction": 0.5540625, "shells": 8 } }, { "seed": 4, "metric": 0.05167350545525551, "routing": { "load_cv": 0.0, "loads": [ 400.0, 400.0, 400.0, 400.0, 400.0, 400.0, 400.0, 400.0 ], "max_shell_count_spread": 0, "reassigned_fraction": 0.4, "shells": 8 } }, { "seed": 5, "metric": 0.06755945831537247, "routing": { "load_cv": 0.0, "loads": [ 400.0, 400.0, 400.0, 400.0, 400.0, 400.0, 400.0, 400.0 ], "max_shell_count_spread": 0, "reassigned_fraction": 0.5334375, "shells": 8 } }, { "seed": 6, "metric": 0.04719924181699753, "routing": { "load_cv": 0.0, "loads": [ 400.0, 400.0, 400.0, 400.0, 400.0, 400.0, 400.0, 400.0 ], "max_shell_count_spread": 0, "reassigned_fraction": 0.534375, "shells": 8 } }, { "seed": 7, "metric": 0.04966289922595024, "routing": { "load_cv": 0.0, "loads": [ 400.0, 400.0, 400.0, 400.0, 400.0, 400.0, 400.0, 400.0 ], "max_shell_count_spread": 0, "reassigned_fraction": 0.4796875, "shells": 8 } } ], "cfg": { "lr": 0.006, "balanced": true, "K": 8 } }, "comparison": { "delta_mean": -0.002399480901658535, "idea_wins": 5, "n_pairs": 8, "per_seed_diffs": [ 0.005220886319875717, -0.0034751594066619873, 0.004238411784172058, -0.013032834976911545, -0.0014705471694469452, -0.0116824209690094, 0.002717830240726471, -0.0017120130360126495 ], "p_value": 0.38035, "mde": 0.00574154155483705, "mde_rel_pct": 10.136634767676144, "verdict": "no significant win", "system_worked": false }, "mechanism_signature": { "custom_track": { "name": "correlated_token_moe_regression", "file": "shell_moe_bench_track.py", "domain": "moe-routing" }, "prediction": "per-shell expert count spread <= 1", "baseline_mean_load_cv": 0.11686608614250746, "idea_mean_load_cv": 0.0, "baseline_mean_shell_spread": 258.375, "idea_mean_shell_spread": 0.0, "idea_mean_reassigned_fraction": 0.5038281250000001, "confirmed": true }, "idea_trials": [ { "cfg": { "lr": 0.0015, "balanced": true, "K": 2 }, "mean": 0.08199466113001108, "per_seed": [ 0.07786157727241516, 0.0871136337518692, 0.08436500281095505, 0.07892747968435287, 0.07485600560903549, 0.10391078889369965, 0.07056240737438202, 0.07836039364337921 ] }, { "cfg": { "lr": 0.0015, "balanced": true, "K": 4 }, "mean": 0.08325883187353611, "per_seed": [ 0.07747699320316315, 0.09112893790006638, 0.08618856221437454, 0.07941422611474991, 0.07391377538442612, 0.10588844120502472, 0.07102843374013901, 0.08103128522634506 ] }, { "cfg": { "lr": 0.0015, "balanced": true, "K": 8 }, "mean": 0.08230428863316774, "per_seed": [ 0.07840844988822937, 0.08882392942905426, 0.08409003913402557, 0.08046981692314148, 0.07284295558929443, 0.10753589123487473, 0.06986328959465027, 0.07639993727207184 ] }, { "cfg": { "lr": 0.003, "balanced": true, "K": 2 }, "mean": 0.06749503640457988, "per_seed": [ 0.06871458142995834, 0.06850986182689667, 0.07267368584871292, 0.06359779834747314, 0.06259164959192276, 0.08258448541164398, 0.05686591938138008, 0.06442230939865112 ] }, { "cfg": { "lr": 0.003, "balanced": true, "K": 4 }, "mean": 0.06823915103450418, "per_seed": [ 0.0692436695098877, 0.07023227959871292, 0.07335231453180313, 0.061882417649030685, 0.06339849531650543, 0.08482187241315842, 0.057540178298950195, 0.06544198095798492 ] }, { "cfg": { "lr": 0.003, "balanced": true, "K": 8 }, "mean": 0.06749762082472444, "per_seed": [ 0.06905151903629303, 0.06874477118253708, 0.07229793071746826, 0.06277107447385788, 0.061685409396886826, 0.08680634945631027, 0.05603644251823425, 0.06258746981620789 ] }, { "cfg": { "lr": 0.006, "balanced": true, "K": 2 }, "mean": 0.054870990104973316, "per_seed": [ 0.05411909893155098, 0.057356927543878555, 0.05673087760806084, 0.055093344300985336, 0.051891688257455826, 0.06572796404361725, 0.04767368361353874, 0.050374336540699005 ] }, { "cfg": { "lr": 0.006, "balanced": true, "K": 4 }, "mean": 0.05486724339425564, "per_seed": [ 0.05602085962891579, 0.056902941316366196, 0.05770782381296158, 0.052157238125801086, 0.054022110998630524, 0.06586991250514984, 0.046959299594163895, 0.0492977611720562 ] }, { "cfg": { "lr": 0.006, "balanced": true, "K": 8 }, "mean": 0.05424201488494873, "per_seed": [ 0.05467335134744644, 0.05477738380432129, 0.056872498244047165, 0.051517780870199203, 0.05167350545525551, 0.06755945831537247, 0.04719924181699753, 0.04966289922595024 ] } ] }