Zero-shot Common Sense Reasoning Suite (PIQA, HellaSwag, WinoGrande, ARC-e, ARC-c, OBQA)
81.12PIQAQwen2.5-14B
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Qwen2.5-14BPruning Ratio=0%, Evaluation Protocol=Zero-shot2026.01 | 81.12 | 63.38 | 75.3 | 82.49 | 55.8 | 34.4 | 65.42 | |
| Qwen2.5-7BPruning Ratio=0%, Zero-shot=true2026.01 | 78.56 | 59.95 | 72.85 | 80.47 | 47.61 | 33.8 | 62.21 | |
| Wanda-sp(w_mix)Pruning Ratio=25%, Evaluation Protocol=Zero-shot2026.01 | 75.14 | 46.89 | 58.25 | 63.59 | 37.12 | 25 | 51 | |
| Ours(w_mix)Pruning Ratio=25%, Evaluation Protocol=Zero-shot2026.01 | 74.97 | 46.85 | 68.67 | 68.77 | 39.76 | 24.6 | 53.94 | |
| FLAP(w_mix)Pruning Ratio=25%, Evaluation Protocol=Zero-shot2026.01 | 74.86 | 47.42 | 64.72 | 68.39 | 39.51 | 23.8 | 53.12 | |
| Wanda-sp(w_mix)Pruning Ratio=25%, Zero-shot=true2026.01 | 73.23 | 43.45 | 54.06 | 63.22 | 33.62 | 23.8 | 48.56 | |
| Iterative Structured Pruning (w_mix)Pruning Ratio=25%, Zero-shot=true2026.01 | 72.85 | 43.12 | 60.54 | 65.45 | 34.04 | 24.6 | 50.1 | |
| FLAP(w_mix)Pruning Ratio=25%, Zero-shot=true2026.01 | 72.31 | 41.75 | 59.59 | 62.33 | 32.08 | 21.4 | 48.24 | |
| ChatGLM-6BPruning Ratio=0%2023.05 | 67.95 | 46.37 | 52.33 | 48.36 | 29.95 | 37.4 | 47.05 | |
| LLM-Pruner (Vector)Pruning Ratio=10%, Tuning=with tuning, Aggregation Strategy=Vector2023.05 | 67.74 | 46.35 | 53.99 | 51.01 | 29.95 | 35 | 47.34 | |
| LLM-Pruner (Vector)Pruning Ratio=10%, Tuning=without tuning, Aggregation Strategy=Vector2023.05 | 66.32 | 43.51 | 53.04 | 47.56 | 30.72 | 35.8 | 46.16 | |
| RandomPruning Ratio=10%, Tuning=without tuning2023.05 | 65.29 | 43.18 | 51.3 | 47.52 | 29.52 | 34.6 | 45.24 | |
| Ours(w_mix)Pruning Ratio=50%, Evaluation Protocol=Zero-shot2026.01 | 62.62 | 30.49 | 53.67 | 39.52 | 21.42 | 16.4 | 37.35 | |
| L2Pruning Ratio=10%, Tuning=without tuning2023.05 | 61.97 | 37.22 | 49.72 | 42.05 | 28.24 | 35.4 | 42.43 | |
| Iterative Structured Pruning (w_mix)Pruning Ratio=50%, Zero-shot=true2026.01 | 60.77 | 29.35 | 50.2 | 35.4 | 18.86 | 12.4 | 34.49 | |
| FLAP(w_mix)Pruning Ratio=50%, Zero-shot=true2026.01 | 56.09 | 27.17 | 49.01 | 29.97 | 19.37 | 12.2 | 32.3 | |
| FLAP(w_mix)Pruning Ratio=50%, Evaluation Protocol=Zero-shot2026.01 | 56.09 | 26.26 | 49.49 | 26.22 | 20.99 | 11.4 | 31.74 | |
| LLM-Pruner (Element)Pruning Ratio=10%, Tuning=without tuning, Aggregation Strategy=Element2023.05 | 54.35 | 28.07 | 50.59 | 27.82 | 24.66 | 33.2 | 36.45 | |
| Wanda-sp(w_mix)Pruning Ratio=50%, Evaluation Protocol=Zero-shot2026.01 | 54.08 | 25.73 | 49.41 | 27.23 | 21.5 | 14.6 | 32.09 | |
| Wanda-sp(w_mix)Pruning Ratio=50%, Zero-shot=true2026.01 | 51.85 | 25.64 | 51.78 | 25.59 | 21.67 | 14.6 | 31.85 |