l1234_popcorn_default_gpt · variant overview
trajectories
30
finished
19
correct
5
compiled
10
avg speedup (correct)
58.15x
sweep-wide eval metricsaveraged over correct kernels
outcomesin_progress 11 · compile_fail 8 · incorrect 5 · correct 5 · error 1
avg speedup58.15x (over 5 correct)
avg memory ratio (kernel / ref)0.80x (5 samples)
avg fusion ratio (ref launches / kernel launches)204.02x (5 samples)
avg energy ratio (ref / kernel)23.47x (1 samples)
avg SOL score (0–1)0.408 (5 samples)
avg DRAM util %36.6%
avg occupancy %70.2%
avg max abs error5.51e-06
modelncorrectcompiled avg speedup
FW-GLM-5-1 7 1 4 2.39x
Llama-4-Maverick-17B-128E-Instruct-FP8 8 0 0 0.00x
gpt-5.5-priority 8 3 5 95.76x
grok-4.3 7 1 1 1.09x