← Subnet 114 · test bench

ablA · bashother

stage-1 fail

compressor variants/ablation/ablate.py · md5 e105c3628c8c · 18/18 runs graded · tasks: dask__dask-12430, vega__altair-4019 · other arms: listing

Score
+0.000
90% CI +0.00 … +0.00
Weighted saving
-14.8%
pooled per run vs baseline
Resolved
0/6
baseline 0/6
Stage gates
S1 fail · S2 fail
tasks 0/2 (need 1)
Cost
$0.42
6 graded runs

Tokens per run vs baseline

bar = arm · tick = baseline · mean per graded run
Weighted376K of 328K↑ 14.8%
Fresh input32K of 30K↑ 4.9%
Cached input2.27M of 1.95M↑ 16.7%
Output39K of 34K↑ 14.1%
Steps60.7 of 54.2↑ 12.0%

Tasks

x = baseline resolves · y = arm resolves · r = log2(W base / W arm)
TaskBase x/nArm y/nW baseW armrZoneSteps b / aOut/step b / a
dask__dask-124300/30/3324K477K-0.558none56 / 71638 / 717
vega__altair-40190/30/3332K276K+0.266none53 / 51614 / 525

Runs

12 runs incl. baseline
TaskArmRepResultSteps FreshCachedOutputWeightedTimeCost
dask__dask-12430baseliner1failed6332K2.32M42K388K1355s$0.073
dask__dask-12430baseliner2failed3822K1.01M27K203K354s$0.045
dask__dask-12430baseliner3failed6632K2.39M36K381K1065s$0.068
dask__dask-12430bashotherr1failed7943K4.07M69K655K1271s$0.120
dask__dask-12430bashotherr2failed6428K2.39M48K412K486s$0.081
dask__dask-12430bashotherr3failed6928K2.25M37K363K746s$0.066
vega__altair-4019baseliner1failed5237K1.84M36K327K1153s$0.065
vega__altair-4019baseliner2failed2821K634K14K127K233s$0.027
vega__altair-4019baseliner3failed7839K3.49M51K541K1323s$0.094
vega__altair-4019bashotherr1failed5037K1.95M34K334K719s$0.064
vega__altair-4019bashotherr2failed4221K1.04M16K172K1165s$0.031
vega__altair-4019bashotherr3failed6034K1.92M32K321K1167s$0.060