General Language Modeling on General Benchmarks Llama 3.1 8B
66.5Generation Quality ScoreBaseline
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| BaselineLR=-2025.05 | 66.5 | 43.2 | 36.4 | 62.3 | 70 | |
| GALR=2e-62025.05 | 66.2 | 43 | 36.7 | 62.8 | 69 | |
| MELR=3e-62025.05 | 66.2 | 43.2 | 35.5 | 62.4 | 70 | |
| UndialLR=2e-72025.05 | 66.1 | 43.4 | 35.6 | 62.9 | 70 | |
| UndialLR=6e-72025.05 | 66.1 | 43.4 | 35.3 | 62.7 | 69 | |
| UndialLR=5e-62025.05 | 66 | 41.9 | 36.1 | 59.8 | 69 | |
| NPOLR=2e-62025.05 | 65.9 | 43.2 | 35.3 | 63.2 | 67 | |
| UnilogitLR=2e-72025.05 | 65.7 | 41.9 | 35.1 | 63.3 | 67 | |
| UnilogitLR=5e-72025.05 | 65.2 | 28.1 | 29.1 | 57 | 58 | |
| UndialLR=1e-52025.05 | 65 | 41 | 35.4 | 48.5 | 66 | |
| RKLDLR=1e-52025.05 | 63.1 | 20.5 | 35.6 | 13.9 | 63 |