nanoGPT speedrun
Train GPT-2 small to a fixed validation loss in as little time as possible (placeholder)
检验
score
+1 −1 routes/adam-mini/README.md 查看文件
| @@ -7,7 +7,7 @@ Adam-mini in place of Muon, for memory. | ||
| 7 | 7 | |
| 8 | 8 | ## What happened |
| 9 | 9 | |
| 10 | -- Started. | |
| 10 | +- 2026-09-06: 26.1 minutes to 3.28 (−18.9 on the baseline) | |
| 11 | 11 | |
| 12 | 12 | ## Next |
| 13 | 13 | |