l1234_popcorn_default_gpt · variant overview
trajectories
61
finished
42
correct
10
compiled
15
avg speedup (correct)
4.92x
sweep-wide eval metricsaveraged over correct kernels
outcomescompile_fail 25 · in_progress 19 · correct 10 · incorrect 5 · error 2
avg speedup4.92x (over 10 correct)
avg memory ratio (kernel / ref)0.85x (10 samples)
avg fusion ratio (ref launches / kernel launches)2.75x (10 samples)
avg energy ratio (ref / kernel)1.60x (9 samples)
avg SOL score (0–1)0.720 (10 samples)
avg DRAM util %75.1%
avg occupancy %209.3%
avg max abs error9.06e-04
modelncorrectcompiled avg speedup
FW-GLM-5-1 7 2 4 0.66x
Llama-4-Maverick-17B-128E-Instruct-FP8 18 1 1 0.84x
gpt-5.5-priority 19 4 7 10.60x
grok-4.3 17 3 3 1.56x