l5_hw_translation_default_final_gpt · variant overview
trajectories
64
finished
64
correct
13
compiled
31
avg speedup (correct)
2.32x
sweep-wide eval metricsaveraged over correct kernels
outcomescompile_fail 23 · incorrect 18 · correct 13 · error 10
avg speedup2.32x (over 13 correct)
avg memory ratio (kernel / ref)0.79x (13 samples)
avg fusion ratio (ref launches / kernel launches)8.65x (13 samples)
avg energy ratio (ref / kernel)— (0 samples)
avg SOL score (0–1)0.263 (11 samples)
avg DRAM util %15.7%
avg occupancy %67.8%
avg max abs error6.20e-03
modelncorrectcompiled avg speedup
FW-GLM-5-1 16 3 12 2.98x
Llama-4-Maverick-17B-128E-Instruct-FP8 16 0 0 0.00x
gpt-5.5-priority 16 10 13 2.12x
grok-4.3 16 0 6 0.00x