Triton Kernel Generation on KernelBench LEVEL2 1.0
80.9Fast1 ScoreDR. KERNEL-14B-STTS
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| DR. KERNEL-14B-STTSscaling_method=sequential test-time scaling (STTS), selection_strategy=best turn selection2026.02 | 80.9 | 47.8 | 23.6 | 12.5 | |
| DR. KERNEL-14B-STTSscaling_method=sequential test-time scaling (STTS)2026.02 | 59.8 | 31.6 | 9.6 | 3 | |
| Claude-4.5-Sonnet2026.02 | 50 | 26.7 | 9.2 | 1.8 | |
| DR. KERNEL-14B2026.02 | 49.2 | 25.6 | 7.4 | 2.1 | |
| GPT-52026.02 | 46.7 | 28.6 | 13.1 | 3 | |
| DR. KERNEL-8B2026.02 | 46 | 20 | 5 | 1.5 | |
| AutoTriton2026.02 | 30.6 | 9.2 | 2.6 | 0.5 | |
| GLM-4.72026.02 | 30 | 20.5 | 8.5 | 3.5 | |
| DR. KERNEL-14BEvaluation Engine=torch.compile2026.02 | 23.5 | 1.9 | 0 | 0 | |
| GPT-5Evaluation Engine=torch.compile2026.02 | 22.1 | 3.6 | 1.5 | 1 | |
| DR. KERNEL-8BEvaluation Engine=torch.compile2026.02 | 20.6 | 0.8 | 0 | 0 | |
| Claude-4.5-SonnetEvaluation Engine=torch.compile2026.02 | 20.5 | 3 | 0 | 0 | |
| Qwen3-32B2026.02 | 14 | 9.4 | 2.4 | 0.2 | |
| Qwen3-8B2026.02 | 13 | 5.6 | 2 | 1.1 | |
| Qwen3-Coder-A30BA32026.02 | 12.6 | 5 | 1.5 | 0.3 | |
| Deepseek-V3.2-Thinking2026.02 | 11 | 6.5 | 2.5 | 0.5 | |
| Cold-Start-8Btraining=cold-start data2026.02 | 8.8 | 5.6 | 1.8 | 0.4 |