Zero-shot Common Sense Reasoning on (PIQA, HellaSwag, WSC, BoolQ, RACE-H)
71.22PIQALLMPruner
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| LLMPrunerBackbone=Llama-2-7B, Pruning Ratio=27.00%, Evaluation Protocol=Zero-shot, Post-training=Excluded2026.01 | 71.22 | 56.46 | 36.54 | 55.2 | 22.56 | 48.4 | |
| LaCoBackbone=Llama-2-7B, Pruning Ratio=27.10%, Evaluation Protocol=Zero-shot2026.01 | 69.8 | 55.69 | 40.38 | 64.07 | 22.61 | 50.51 | |
| TAPPABackbone=Llama-2-7B, Pruning Ratio=28.10%, Evaluation Protocol=Zero-shot2026.01 | 66.76 | 55.97 | 68.13 | 62.17 | 33.88 | 57.38 | |
| ShortGPTBackbone=Llama-2-7B, Pruning Ratio=27.10%, Evaluation Protocol=Zero-shot, Post-training=Included2026.01 | 66.43 | 53.02 | 52.46 | 74.71 | 32.25 | 55.77 | |
| SliceGPTBackbone=Llama-2-7B, Pruning Ratio=26.40%, Evaluation Protocol=Zero-shot, Post-training=Included2026.01 | 66.21 | 50.27 | 36.54 | 38.32 | 21.07 | 42.48 |