nanoGPT speedrun
Train GPT-2 small to a fixed validation loss in as little time as possible (placeholder)
| # | 路线 | minutes to 3.28 ↓ | 对比基线 |
|---|---|---|---|
| 1 | 3.95 | 好 91% | |
| 6 | 15.9 | 好 65% | |
| 7 | 22.3 | 好 50% | |
| 8 | 26.1 | 好 42% | |
| 9 | 31.4 | 好 30% | |
| — | 基线 · llm.c GPT-2 (124M) | 45 | 条形从这里开始 |
Train GPT-2 small to a fixed validation loss in as little time as possible (placeholder)
| # | 路线 | minutes to 3.28 ↓ | 对比基线 |
|---|---|---|---|
| 1 | 3.95 | 好 91% | |
| 6 | 15.9 | 好 65% | |
| 7 | 22.3 | 好 50% | |
| 8 | 26.1 | 好 42% | |
| 9 | 31.4 | 好 30% | |
| — | 基线 · llm.c GPT-2 (124M) | 45 | 条形从这里开始 |