Zero-shot Evaluation on Arc-e, PIQA, Hellaswag, OpenBookQA, Winogrande, MMLU, and BoolQ (test)
31.63Arc-e AccuracyPeri-LN
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Peri-LNModel=Llama-1B, Evaluation Protocol=Zero-shot2025.12 | 31.63 | 63.07 | 32.05 | 32.4 | 49.41 | 24.91 | 58.05 | 41.65 | |
| LNSModel=Llama-1B, Evaluation Protocol=Zero-shot2025.12 | 31.39 | 62.94 | 32.8 | 33.5 | 50.37 | 24.88 | 56.67 | 41.79 | |
| RMSNormModel=Llama-1B, Evaluation Protocol=Zero-shot2025.12 | 30.97 | 62.89 | 32.77 | 32.32 | 50.54 | 25.7 | 56.26 | 41.64 | |
| BHyTModel=Llama-1B, Evaluation Protocol=Zero-shot2025.12 | 30.5 | 62.42 | 32.11 | 33.88 | 50.15 | 25.19 | 61.86 | 42.3 | |
| DyTModel=Llama-1B, Evaluation Protocol=Zero-shot2025.12 | 29.28 | 58.84 | 26.51 | 30.85 | 50.34 | 24.9 | 39.08 | 37.11 |