Zero-shot Evaluation on ARC (c/e), BoolQ, CrowS-Pairs, OBQA, PIQA, RACE, SiQA, TruthfulQA, WinoGrande
52.1ARC-C AccuracyHFPrune
Evaluation Results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| HFPruneModel=Qwen2.5-7B, Ratio=20%2026.03 | 52.1 | 74.5 | 84.6 | 62.9 | 45.4 | 78.1 | 41.2 | 48.1 | 52 | 64.8 | 60.4 | |
| OriginalModel=Qwen2.5-7B, Ratio=0%2026.03 | 51.1 | 77.5 | 84.7 | 64 | 46.8 | 79.8 | 41.4 | 54.8 | 56.4 | 73.2 | 63 | |
| HFPruneModel=Qwen2.5-7B, Ratio=30%2026.03 | 48.2 | 70.3 | 83.2 | 60.4 | 41.6 | 76.1 | 40.7 | 48.9 | 48.2 | 62.8 | 58 | |
| SDMPruneModel=Qwen2.5-7B, Ratio=20%2026.03 | 48.1 | 69.7 | 84.6 | 64.2 | 44.4 | 77.6 | 41.7 | 48 | 51.8 | 67.2 | 59.7 | |
| HFPruneRatio=20%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot, Finetuning Dataset=LaMini2026.03 | 47.1 | 73.8 | 85.2 | 60.2 | 43.2 | 77.3 | 43.3 | 49.5 | 44.7 | 66.2 | 59 | |
| OriginalModel=Qwen2.5-1.5B, Ratio=0%2026.03 | 45.4 | 72 | 73 | 60.9 | 40.4 | 76 | 36.5 | 48.9 | 46.6 | 63.5 | 56.3 | |
| Llama-2-7BRatio=0%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot, Finetuning Dataset=LaMini2026.03 | 45.1 | 73.8 | 79.4 | 67.4 | 44.2 | 78.7 | 40.1 | 46.5 | 38.8 | 69.3 | 58.3 | |
| SDMPruneRatio=20%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot, Finetuning Dataset=LaMini2026.03 | 43.9 | 72.3 | 81.7 | 62.1 | 42 | 77 | 41.3 | 48.5 | 44.9 | 68.4 | 58.2 | |
| OriginalModel=Qwen3-1.7B, Ratio=0%2026.03 | 42.8 | 70 | 77.6 | 55.2 | 37.8 | 71.9 | 36.8 | 44.9 | 46 | 61.6 | 54.4 | |
| HFPruneRatio=30%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot, Finetuning Dataset=LaMini2026.03 | 41.9 | 70.2 | 82.9 | 58.1 | 40 | 75.2 | 39.5 | 48.8 | 44.2 | 62.4 | 56.3 | |
| HFPruneModel=Qwen2.5-1.5B, Ratio=20%2026.03 | 41.8 | 68.8 | 79.4 | 55.3 | 39.4 | 74.1 | 38.7 | 46.4 | 42.2 | 59.8 | 54.6 | |
| LoRAPruneRatio=20%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot, Finetuning Dataset=LaMini2026.03 | 41.6 | 71 | 81.7 | 58.7 | 41.4 | 76.7 | 40.4 | 44 | 65.9 | 65.9 | 56.7 | |
| SDMPruneModel=Qwen2.5-7B, Ratio=30%2026.03 | 40.5 | 63.6 | 81.1 | 62.2 | 38.4 | 74.2 | 39.4 | 44.5 | 49.1 | 60.2 | 55.3 | |
| LLM-prunerRatio=20%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot, Finetuning Dataset=LaMini2026.03 | 40.4 | 70.1 | 80.2 | 61.7 | 38.8 | 75.8 | 39 | 47.1 | 43.9 | 64.3 | 56.1 | |
| SDMPruneRatio=30%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot, Finetuning Dataset=LaMini2026.03 | 39.6 | 67.9 | 80.4 | 58.5 | 37.2 | 75.2 | 40 | 47.8 | 43.7 | 65.4 | 55.6 | |
| HFPruneModel=Qwen3-1.7B, Ratio=20%2026.03 | 39.1 | 69.4 | 78.9 | 55.8 | 36.2 | 72.4 | 39.7 | 46.4 | 46.4 | 58.2 | 54.3 | |
| LoRAPruneRatio=30%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot, Finetuning Dataset=LaMini2026.03 | 38.6 | 65.1 | 74.1 | 61.4 | 37.4 | 72.9 | 39 | 46.3 | 44.8 | 66.5 | 54.6 | |
| LoRAPRatio=20%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot, Finetuning Dataset=LaMini2026.03 | 38.5 | 66 | 70.9 | — | 39.6 | 78.1 | — | — | — | 65.7 | — | |
| LLM-prunerRatio=30%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot, Finetuning Dataset=LaMini2026.03 | 38 | 64.8 | 75.6 | 62.3 | 36.4 | 73.4 | 35.7 | 47.3 | 42.3 | 62.9 | 53.9 | |
| LoRAPRatio=30%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot, Finetuning Dataset=LaMini2026.03 | 35.5 | 60.6 | 69.6 | — | 37.8 | 76.7 | — | — | — | 63 | — | |
| HFPruneModel=Qwen2.5-1.5B, Ratio=30%2026.03 | 35.2 | 62.6 | 76 | 54 | 34.6 | 70.8 | 36.2 | 44 | 44.5 | 58.6 | 51.7 | |
| HFPruneModel=Qwen3-1.7B, Ratio=30%2026.03 | 34.5 | 61.9 | 78.2 | 55.5 | 33.8 | 70.8 | 36.3 | 44.1 | 45.2 | 56.4 | 51.7 | |
| SDMPruneModel=Qwen2.5-1.5B, Ratio=20%2026.03 | 32.3 | 59.2 | 72.1 | 56.2 | 35.2 | 72 | 37.7 | 43.6 | 44.7 | 58.2 | 51.1 | |
| SDMPruneModel=Qwen3-1.7B, Ratio=20%2026.03 | 31.3 | 58.5 | 70.8 | 53.7 | 33.4 | 71.4 | 37.1 | 43.8 | 44.7 | 58.6 | 50.3 | |
| SDMPruneModel=Qwen2.5-1.5B, Ratio=30%2026.03 | 26.3 | 49.5 | 62.9 | 55 | 32.8 | 68.6 | 34.9 | 42.1 | 43.8 | 57.9 | 47.4 | |
| SDMPruneModel=Qwen3-1.7B, Ratio=30%2026.03 | 25.9 | 45.1 | 62.3 | 55.1 | 29.8 | 66.2 | 35.3 | 40.5 | 43.4 | 56.7 | 46 |