pop_l123_popcorn2_bench_st_gpt · variant overview
trajectories
102
finished
102
correct
13
compiled
51
avg speedup (correct)
895.11x
sweep-wide eval metricsaveraged over correct kernels
outcomesincorrect 38 · compile_fail 33 · error 18 · correct 13
avg speedup895.11x (over 13 correct)
avg memory ratio (kernel / ref)0.92x (13 samples)
avg fusion ratio (ref launches / kernel launches)3496.75x (13 samples)
avg energy ratio (ref / kernel)28.84x (5 samples)
avg SOL score (0–1)0.238 (13 samples)
avg DRAM util %15.8%
avg occupancy %144.1%
avg max abs error2.60e-06
modelncorrectcompiled avg speedup
FW-GLM-5-1 34 5 15 1203.01x
gpt-5.5-priority 34 6 21 857.06x
grok-4.3 34 2 15 239.47x