Zero-shot Commonsense Reasoning on Commonsense Reasoning Suite
73.18BoolQ AccuracyLLaMA1-7B
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| LLaMA1-7BRatio=100%2026.03 | 73.18 | 78.35 | 73 | 66.93 | 67.3 | 41.47 | 42.4 | 63.23 | |
| Wanda-spRatio=80%2026.03 | 71.59 | 75.63 | 67.86 | 65.67 | 61.78 | 38.05 | 38.6 | 59.88 | |
| LLaMA2-7BRetention Ratio=100%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-shot2026.03 | 71.1 | 78.4 | 72.96 | 67.25 | 69.36 | 40.53 | 40.8 | 62.91 | |
| GPrune-LLMRatio=80%, Base Method=Wanda-sp2026.03 | 70.98 | 75.3 | 68.28 | 66.93 | 64.35 | 38.57 | 38 | 60.34 | |
| GPrune-LLMRatio=80%, Base Method=FLAP2026.03 | 70.21 | 75.63 | 66.17 | 64 | 64.52 | 37.29 | 40.8 | 59.8 | |
| FLAPRatio=80%2026.03 | 68.72 | 74.86 | 64.97 | 64.88 | 63.09 | 37.3 | 39.4 | 59.03 | |
| GPrune-LLMRatio=70%, Base Method=FLAP2026.03 | 65.9 | 72.14 | 59.39 | 59.83 | 61.95 | 34.81 | 39.8 | 56.26 | |
| GPrune-LLM (FLAP)Retention Ratio=80%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-shot2026.03 | 65.66 | 74.65 | 66.55 | 59.98 | 63.76 | 35.24 | 39.2 | 57.86 | |
| FLAPRatio=70%2026.03 | 64.8 | 70.78 | 58.3 | 59.04 | 57.15 | 34.56 | 37.6 | 54.6 | |
| LLM-PrunerRatio=80%2026.03 | 64.19 | 75.68 | 66.56 | 59.19 | 61.91 | 36.6 | 40.8 | 57.85 | |
| Wanda-spRatio=70%2026.03 | 63.98 | 71.33 | 59.48 | 60.69 | 57.15 | 34.3 | 38.2 | 55.02 | |
| GPrune-LLM (FLAP)Retention Ratio=70%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-shot2026.03 | 63.24 | 72.36 | 60.71 | 56.91 | 58.12 | 33.96 | 39 | 54.9 | |
| GPrune-LLM (Wanda-sp)Retention Ratio=80%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-shot2026.03 | 62.69 | 75.52 | 68.42 | 65.27 | 63.97 | 37.46 | 39.6 | 58.99 | |
| LLM-PrunerRetention Ratio=80%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-shot2026.03 | 62.26 | 76.28 | 65.82 | 60.85 | 63.85 | 37.46 | 39.2 | 57.96 | |
| GPrune-LLMRatio=50%, Base Method=FLAP2026.03 | 62.17 | 64.42 | 43.1 | 52.88 | 47.18 | 31.83 | 35.8 | 48.2 | |
| FLAPRatio=50%2026.03 | 61.8 | 63 | 41.78 | 55.01 | 45.79 | 30.03 | 35.4 | 47.54 | |
| GPrune-LLMRatio=50%, Base Method=Wanda-sp2026.03 | 61.65 | 55.77 | 29.76 | 50.2 | 32.7 | 22.1 | 31.8 | 40.57 | |
| GPrune-LLMRatio=70%, Base Method=Wanda-sp2026.03 | 61.07 | 71.55 | 60.74 | 60.06 | 59.72 | 34.39 | 39.6 | 55.3 | |
| LLM-PrunerRetention Ratio=50%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-shot2026.03 | 60.61 | 58.43 | 30.56 | 49.33 | 31.19 | 27.05 | 31.4 | 41.22 | |
| Wanda-spRetention Ratio=50%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-shot2026.03 | 59.82 | 54.13 | 26.54 | 49.41 | 29 | 23.89 | 32.8 | 39.37 | |
| Wanda-spRetention Ratio=70%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-shot2026.03 | 58.78 | 71.49 | 59.44 | 55.56 | 60.27 | 37.03 | 39.4 | 54.57 | |
| GPrune-LLM (Wanda-sp)Retention Ratio=70%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-shot2026.03 | 57.74 | 72.52 | 63.08 | 59.51 | 60.82 | 37.2 | 38.8 | 55.67 | |
| GPrune-LLM (Wanda-sp)Retention Ratio=50%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-shot2026.03 | 56.7 | 63.11 | 43.84 | 51.46 | 40.74 | 30.63 | 35.6 | 46.01 | |
| LLM-PrunerRatio=50%2026.03 | 56.48 | 60.99 | 35.78 | 52.01 | 34.13 | 27.73 | 34.2 | 43.05 | |
| LLM-PrunerRatio=70%2026.03 | 53.49 | 72.2 | 59.05 | 57.06 | 54.59 | 34.98 | 38.4 | 52.82 | |
| FLAPRetention Ratio=70%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-shot2026.03 | 53.39 | 69.48 | 55.49 | 60.3 | 54.71 | 32.59 | 38.2 | 52.02 | |
| FLAPRetention Ratio=80%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-shot2026.03 | 52.45 | 73.61 | 62.56 | 62.12 | 59.13 | 34.81 | 39.2 | 54.84 | |
| GPrune-LLM (FLAP)Retention Ratio=50%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-shot2026.03 | 51.13 | 63.6 | 45.01 | 52.49 | 43.52 | 29.35 | 34.8 | 45.7 | |
| LLM-PrunerRetention Ratio=70%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-shot2026.03 | 50.15 | 71.82 | 55.64 | 55.49 | 54.67 | 31.57 | 36.4 | 50.82 | |
| Wanda-spRetention Ratio=80%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-shot2026.03 | 46.79 | 76.5 | 68.13 | 65.43 | 65.95 | 39.76 | 38.2 | 57.25 | |
| Wanda-spRatio=50%2026.03 | 40.7 | 54.46 | 28.31 | 49.72 | 28.62 | 23.46 | 30.6 | 36.55 | |
| FLAPRetention Ratio=50%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-shot2026.03 | 39.63 | 62.13 | 39.54 | 50.36 | 35.52 | 28.24 | 35.2 | 41.52 |