nolib1
Step 1 (shared with hero3)
| Rollout | Reward | Test | Edited accuracy | Hit cap | Tensors edited | Turns | Minutes | Forward passes | Python calls | Ended |
|---|
| 1 | +1.2 | +0.6 | 8.6% | 92.6% | 1 | 76 | 105.6 | 15 | 55 | editor finished |
| 2 | +9.4 | +11.1 | 16.8% | 82.4% | 48 | 66 | 52.7 | 7 | 54 | editor finished |
| 3 | +18.8 | +16.7 | 26.2% | 78.5% | 6 | 87 | 88.4 | 11 | 73 | editor finished |
| 4 | +15.6 | +15.2 | 23.0% | 75.4% | 6 | 77 | 85.9 | 11 | 63 | editor finished |
| 5 | +33.6 | +32.8 | 41.0% | 39.8% | 48 | 84 | 66.1 | 9 | 72 | editor finished |
| 6 | +0.0 | no edit | – | – | 0 | 27 | 8.6 | 0 | 19 | no tool call in five replies |
| 7 | +2.7 | +2.5 | 10.2% | 90.6% | 1 | 96 | 106.2 | 14 | 79 | editor finished |
| 8 | +27.3 | +29.4 | 34.8% | 58.6% | 48 | 86 | 59.9 | 14 | 69 | editor finished |
| Mean | +13.6 | +15.5 | 22.9% | 74.0% | 19.8 | 74.9 | 71.7 | 10.1 | 60.5 | |
Step 2
| Rollout | Reward | Test | Edited accuracy | Hit cap | Tensors edited | Turns | Minutes | Forward passes | Python calls | Ended |
|---|
| 1 | +21.1 | +19.0 | 28.5% | 70.3% | 48 | 81 | 47.1 | 9 | 68 | editor finished |
| 2 | +11.3 | +12.3 | 18.8% | 80.5% | 4 | 69 | 73.0 | 12 | 54 | editor finished |
| 3 | +16.8 | +18.7 | 24.2% | 74.6% | 48 | 61 | 58.8 | 12 | 47 | editor finished |
| 4 | +21.1 | +21.6 | 28.5% | 69.1% | 1 | 91 | 88.8 | 13 | 73 | editor finished |
| 5 | +0.4 | +0.4 | 7.8% | 93.0% | 1 | 63 | 80.2 | 9 | 50 | editor finished |
| 6 | +13.3 | +10.0 | 20.7% | 83.2% | 18 | 72 | 119.5 | 13 | 53 | editor finished |
| 7 | +21.1 | +21.4 | 28.5% | 69.5% | 1 | 84 | 78.0 | 14 | 67 | editor finished |
| 8 | +3.1 | +4.4 | 10.5% | 90.2% | 48 | 68 | 59.0 | 8 | 56 | editor finished |
| Mean | +13.5 | +13.5 | 20.9% | 78.8% | 21.1 | 73.6 | 75.5 | 11.2 | 58.5 | |
Step 3
| Rollout | Reward | Test | Edited accuracy | Hit cap | Tensors edited | Turns | Minutes | Forward passes | Python calls | Ended |
|---|
| 1 | +32.8 | – | 40.2% | 47.7% | 48 | 86 | 66.6 | 13 | 69 | editor finished |
| 2 | +27.7 | – | 35.2% | 60.2% | 48 | 97 | 85.6 | 15 | 81 | editor finished |
| 3 | +0.0 | – | – | – | 0 | 46 | 13.4 | 4 | 30 | no tool call in five replies |
| 4 | +16.8 | – | 24.2% | 74.6% | 30 | 73 | 51.3 | 10 | 60 | editor finished |
| 5 | +14.4 | – | 21.9% | 80.5% | 6 | 78 | 68.6 | 10 | 66 | editor finished |
| 6 | +18.4 | – | 25.8% | 69.9% | 24 | 97 | 89.3 | 15 | 78 | editor finished |
| 7 | +25.0 | – | 32.4% | 62.5% | 24 | 105 | 119.1 | 18 | 84 | editor finished |
| 8 | +14.4 | – | 21.9% | 76.2% | 48 | 81 | 79.8 | 13 | 65 | editor finished |
| Mean | +18.7 | – | 28.8% | 67.4% | 28.5 | 82.9 | 71.7 | 12.2 | 66.6 | |
Reward: edited minus original accuracy on the reward set (256 answers, original 7.4%), in points. Test: the same on MATH-500 (original 6.6%), not used for training; scored after a step finishes, – until then. Hit cap: share of answers reaching 8,192 tokens (original 93%).