Question Answering on Commonsense QA Suite (WinoGrande, HellaSwag, ARC-e, ARC-c, PIQA)
72.7Avg AccuracyLLaMA-3 8B
Evaluation Results
| Method | Links | |
|---|---|---|
| LLaMA-3 8BPruning Ratio=0%, Base Model=LLaMA-3 8B, Evaluation Protocol=Zero-shot2026.02 | 72.7 | |
| LLaMA-2 7BPruning Ratio=0%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.02 | 68.99 | |
| MoPPruning Ratio=20%, Base Model=LLaMA-3 8B, Evaluation Protocol=Zero-shot2026.02 | 65.44 | |
| MoPPruning Ratio=20%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.02 | 65.22 | |
| AmoebaLLMPruning Ratio=20%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.02 | 64.9 | |
| SlimLLMPruning Ratio=20%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.02 | 64 | |
| AMPPruning Ratio=20%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.02 | 63.48 | |
| CoMePruning Ratio=20%, Base Model=LLaMA-3 8B, Evaluation Protocol=Zero-shot2026.02 | 62.93 | |
| LINEARPATCHPruning Ratio=20%, Base Model=LLaMA-3 8B, Evaluation Protocol=Zero-shot2026.02 | 62.75 | |
| AMPPruning Ratio=20%, Base Model=LLaMA-3 8B, Evaluation Protocol=Zero-shot2026.02 | 62.72 | |
| LINEARPATCHPruning Ratio=20%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.02 | 62.51 | |
| CoMePruning Ratio=20%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.02 | 62.41 | |
| PruneNetPruning Ratio=20%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.02 | 62.34 | |
| MoPPruning Ratio=30%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.02 | 61.54 | |
| AMPPruning Ratio=30%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.02 | 61.02 | |
| ModHiFi-PPruning Ratio=20%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.02 | 60.62 | |
| CoMePruning Ratio=30%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.02 | 60.47 | |
| MoPPruning Ratio=30%, Base Model=LLaMA-3 8B, Evaluation Protocol=Zero-shot2026.02 | 60.43 | |
| DISP-LLMPruning Ratio=30%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.02 | 59.53 | |
| LINEARPATCHPruning Ratio=30%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.02 | 59.39 | |
| SliceGPTPruning Ratio=20%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.02 | 58.18 | |
| AMPPruning Ratio=30%, Base Model=LLaMA-3 8B, Evaluation Protocol=Zero-shot2026.02 | 58.01 | |
| PruneNetPruning Ratio=30%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.02 | 57.08 | |
| Yang et al.Pruning Ratio=30%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.02 | 56.65 | |
| MoPPruning Ratio=40%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.02 | 55.37 | |
| MoPPruning Ratio=40%, Base Model=LLaMA-3 8B, Evaluation Protocol=Zero-shot2026.02 | 55.34 | |
| Yang et al.Pruning Ratio=30%, Base Model=LLaMA-3 8B, Evaluation Protocol=Zero-shot2026.02 | 54.34 | |
| AMPPruning Ratio=40%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.02 | 53.06 | |
| AMPPruning Ratio=40%, Base Model=LLaMA-3 8B, Evaluation Protocol=Zero-shot2026.02 | 52.65 | |
| CoMePruning Ratio=30%, Base Model=LLaMA-3 8B, Evaluation Protocol=Zero-shot2026.02 | 52.5 | |
| ModHiFi-PPruning Ratio=30%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.02 | 52.44 | |
| Shortened LLaMAPruning Ratio=30%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.02 | 51.56 | |
| SliceGPTPruning Ratio=30%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.02 | 51.5 | |
| Yang et al.Pruning Ratio=40%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.02 | 50.39 | |
| Yang et al.Pruning Ratio=40%, Base Model=LLaMA-3 8B, Evaluation Protocol=Zero-shot2026.02 | 47.8 |