Mathematical Reasoning on MATH (test) (Domain Accuracy + Learning Progress)
9.9Domain AccuracyLaLoRA
Evaluation results
| Method | Links | ||
|---|---|---|---|
| LaLoRAlambda=10^22025.12 | 9.9 | 2.1 | |
| MIGUt=0.92025.12 | 9.9 | 2.1 | |
| LaLoRAlambda=102025.12 | 9.3 | 1.5 | |
| LaLoRAlambda=1.5 * 10^22025.12 | 8.8 | 1 | |
| L2 Regularizationstrength=10^-42025.12 | 8.6 | 0.8 | |
| L2 Regularizationstrength=10^-52025.12 | 8.3 | 0.5 | |
| LaLoRAlambda=2.5 * 10^22025.12 | 8.1 | 0.3 | |
| LoRAlambda=02025.12 | 7.8 | 0 | |
| L2 Regularizationstrength=10^-32025.12 | 7.8 | 0 | |
| MIGUt=0.852025.12 | 7.6 | -0.2 | |
| MIGUt=0.72025.12 | 7.6 | -0.2 | |
| MIGUt=0.6, variant=b2025.12 | 7.6 | -0.2 | |
| L2 Regularizationstrength=0.12025.12 | 7.6 | -0.2 | |
| MIGUt=0.952025.12 | 7.5 | -0.3 | |
| L2 Regularizationstrength=10^-62025.12 | 7.1 | -0.7 | |
| LaLoRAlambda=10^32025.12 | 7 | -0.8 | |
| LaLoRAlambda=12025.12 | 6.9 | -0.9 | |
| MIGUt=0.82025.12 | 6.8 | -1 | |
| L2 Regularizationstrength=0.052025.12 | 6.7 | -1.1 | |
| LaLoRAlambda=5 * 10^32025.12 | 6.6 | -1.2 | |
| L2 Regularizationstrength=0.52025.12 | 6.5 | -1.3 | |
| MiLoRA2025.12 | 6.1 | -1.7 | |
| MIGUt=0.6, variant=a2025.12 | 5.9 | -1.9 |