| outcomes | correct 218 · incorrect 33 · compile_fail 14 · error 2 |
| avg speedup | 561.33x (over 218 correct) |
| avg memory ratio (kernel / ref) | 0.91x (218 samples) |
| avg fusion ratio (ref launches / kernel launches) | 2559.10x (218 samples) |
| avg energy ratio (ref / kernel) | 14.26x (27 samples) |
| avg SOL score (0–1) | 0.280 (218 samples) |
| avg DRAM util % | 20.2% |
| avg occupancy % | 168.1% |
| avg max abs error | 2.06e-04 |
| model | n | correct | compiled | avg speedup |
|---|---|---|---|---|
| FW-GLM-5-1 | 89 | 70 | 87 | 312.82x |
| gpt-5.5-priority | 89 | 89 | 89 | 1062.82x |
| grok-4.3 | 89 | 59 | 75 | 99.70x |