l5_hw_translation_all_full_gpt · variant overview
trajectories
16
finished
16
correct
6
compiled
8
avg speedup (correct)
0.55x
sweep-wide eval metricsaveraged over correct kernels
outcomeserror 8 · correct 6 · incorrect 2
avg speedup0.55x (over 6 correct)
avg memory ratio (kernel / ref)0.76x (6 samples)
avg fusion ratio (ref launches / kernel launches)7.93x (6 samples)
avg energy ratio (ref / kernel)0.05x (1 samples)
avg SOL score (0–1)0.321 (5 samples)
avg DRAM util %6.8%
avg occupancy %123.4%
avg max abs error1.33e-02
modelncorrectcompiled avg speedup
gpt-5.5-priority 16 6 8 0.55x