| outcomes | correct 169 · error 16 · compile_fail 12 · incorrect 3 |
| avg speedup | 22.34x (over 169 correct) |
| avg memory ratio (kernel / ref) | 0.82x (169 samples) |
| avg fusion ratio (ref launches / kernel launches) | 5.21x (169 samples) |
| avg energy ratio (ref / kernel) | 1.60x (132 samples) |
| avg SOL score (0–1) | 0.732 (167 samples) |
| avg DRAM util % | 61.8% |
| avg occupancy % | 126.4% |
| avg max abs error | 8.03e-03 |
| model | n | correct | compiled | avg speedup |
|---|---|---|---|---|
| gpt-5.5-priority | 200 | 169 | 172 | 22.34x |