pop_l123_default_gpt_stress_gpt · variant overview
trajectories
267
finished
266
correct
54
compiled
96
avg speedup (correct)
187.15x
sweep-wide eval metricsaveraged over correct kernels
outcomescompile_fail 170 · correct 54 · incorrect 32 · in_progress 6 · pending 5
avg speedup187.15x (over 54 correct)
avg memory ratio (kernel / ref)0.89x (54 samples)
avg fusion ratio (ref launches / kernel launches)989.75x (54 samples)
avg energy ratio (ref / kernel)11.78x (12 samples)
avg SOL score (0–1)0.313 (53 samples)
avg DRAM util %25.2%
avg occupancy %190.2%
avg max abs error4.52e-05
modelncorrectcompiled avg speedup
FW-GLM-5-1 89 27 40 108.96x
gpt-5.5-priority 89 0 2 0.00x
grok-4.3 89 27 54 193.74x