sl5_hw_translation_8xh100_all_gpt · variant overview
trajectories
45
finished
45
correct
2
compiled
23
avg speedup (correct)
3.18x
sweep-wide eval metricsaveraged over correct kernels
outcomescompile_fail 22 · incorrect 21 · correct 2
avg speedup3.18x (over 2 correct)
avg memory ratio (kernel / ref)0.69x (2 samples)
avg fusion ratio (ref launches / kernel launches)3.52x (2 samples)
avg energy ratio (ref / kernel)— (0 samples)
avg SOL score (0–1)0.236 (2 samples)
avg DRAM util %23.6%
avg occupancy %52.0%
avg max abs error5.86e-03
modelncorrectcompiled avg speedup
FW-GLM-5-1 15 1 9 1.12x
gpt-5.5-priority 15 1 7 5.25x
grok-4.3 15 0 7 0.00x