nanoGPT speedrun
Train GPT-2 small to a fixed validation loss in as little time as possible (placeholder)
Score: 2026-10-14: 4.9 minutes to 3.28 (−40.1 on the baseline)
Browse filesChecks
score
+1 −0 routes/sparse-attn-schedule/README.md View file
| @@ -10,6 +10,7 @@ Block-sparse attention with a batch-size schedule. | ||
| 10 | 10 | ## What happened |
| 11 | 11 | |
| 12 | 12 | - 2026-10-11: 4.66 minutes to 3.28 (−40.34 on the baseline) |
| 13 | +- 2026-10-14: 4.9 minutes to 3.28 (−40.1 on the baseline) | |
| 13 | 14 | |
| 14 | 15 | ## Next |
| 15 | 16 | |