| outcomes | in_progress 47 · correct 43 · incorrect 39 · error 18 · compile_fail 5 |
| avg speedup | 1258.95x (over 43 correct) |
| avg memory ratio (kernel / ref) | 0.94x (43 samples) |
| avg fusion ratio (ref launches / kernel launches) | 2721.27x (43 samples) |
| avg energy ratio (ref / kernel) | 79.24x (6 samples) |
| avg SOL score (0–1) | 0.216 (43 samples) |
| avg DRAM util % | 15.3% |
| avg occupancy % | 119.5% |
| avg max abs error | 8.59e-04 |
| model | n | correct | compiled | avg speedup |
|---|---|---|---|---|
| FW-GLM-5-1 | 53 | 11 | 30 | 714.12x |
| gpt-5.5-priority | 52 | 21 | 27 | 1674.29x |
| grok-4.3 | 47 | 11 | 25 | 1672.16x |