Commonsense Reasoning on Commonsense Reasoning (WG, RTE, OBQA, HS, BoolQ, ARC)
80.66WinoGrande Accuracy (WG)SparseGPT+SPP
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| SparseGPT+SPPLLaMA=V1-13B, Pruning=SparseGPT, Sparsity level=0.62025.05 | 80.66 | 56.68 | 35 | 57.53 | 79.78 | 75.76 | 45.22 | 61.52 | |
| SparseGPT+SEFTLLaMA=V1-13B, Pruning=SparseGPT, Sparsity level=0.62025.05 | 80.19 | 57.4 | 33.2 | 56.81 | 83.52 | 77.02 | 45.9 | 62.01 | |
| Wanda+SPPLLaMA=V1-13B, Pruning=Wanda, Sparsity level=0.62025.05 | 80.03 | 62.09 | 32.6 | 57.08 | 76.73 | 75.88 | 44.71 | 61.3 | |
| SparseGPT+SQFTLLaMA=V1-13B, Pruning=SparseGPT, Sparsity level=0.62025.05 | 79.63 | 57.28 | 34.2 | 57.5 | 83.45 | 76.01 | 44.79 | 61.84 | |
| SparseGPT+LoRA*LLaMA=V3-8B, Pruning=SparseGPT, Sparsity level=0.62025.05 | 79.48 | 57.04 | 29.4 | 53.78 | 83.3 | 74.2 | 43.34 | 60.07 | |
| Wanda+SQFTLLaMA=V1-13B, Pruning=Wanda, Sparsity level=0.62025.05 | 79.48 | 58.84 | 32.4 | 56.84 | 82.54 | 76.05 | 46.42 | 61.79 | |
| SparseGPT+SQFTLLaMA=V3-8B, Pruning=SparseGPT, Sparsity level=0.62025.05 | 79.43 | 63.17 | 31.4 | 53.83 | 82.09 | 73.19 | 41.97 | 60.73 | |
| Wanda+SEFTLLaMA=V1-13B, Pruning=Wanda, Sparsity level=0.62025.05 | 79.32 | — | 33 | 56.58 | 83.58 | 77.02 | 47.44 | 62.31 | |
| SparseGPT+SEFTLLaMA=V3-8B, Pruning=SparseGPT, Sparsity level=0.62025.05 | 78.37 | 62.82 | 31.2 | 53.99 | 83 | 73.4 | 43.69 | 60.92 | |
| SparseGPT+LoRA*LLaMA=V1-13B, Pruning=SparseGPT, Sparsity level=0.62025.05 | 78.37 | 59.93 | 31.2 | 55.23 | 79.6 | 76.01 | 43.26 | 60.51 | |
| Wanda+LoRA*LLaMA=V1-13B, Pruning=Wanda, Sparsity level=0.62025.05 | 77.11 | 59.93 | 33.6 | 56.37 | 82.38 | 76.56 | 44.88 | 61.54 | |
| Wanda+SEFTLLaMA=V3-8B, Pruning=Wanda, Sparsity level=0.62025.05 | 76.87 | 57.76 | 28.6 | 52.2 | 80.83 | 71.46 | 41.38 | 58.44 | |
| SparseGPT+SPPLLaMA=V3-8B, Pruning=SparseGPT, Sparsity level=0.62025.05 | 76.87 | 57.76 | 31.2 | 53.45 | 80.67 | 74.12 | 43.09 | 59.6 | |
| Wanda+SEFTLLaMA=V2-7B, Pruning=Wanda, Sparsity level=0.62025.05 | 75.85 | 56.32 | 29.8 | 52.34 | 67.65 | 72.6 | 41.21 | 56.54 | |
| Wanda+SQFTLLaMA=V3-8B, Pruning=Wanda, Sparsity level=0.62025.05 | 75.84 | 60.26 | 26.4 | 50.85 | 77.61 | 70.24 | 40.01 | 57.32 | |
| Wanda+SQFTLLaMA=V2-7B, Pruning=Wanda, Sparsity level=0.62025.05 | 75.61 | 54.87 | 29 | 52.86 | 70.46 | 72.51 | 39.07 | 56.34 | |
| SparseGPT+SEFTLLaMA=V2-7B, Pruning=SparseGPT, Sparsity level=0.62025.05 | 75.53 | 55.23 | 30.8 | 52.99 | 77.98 | 72.64 | 41.13 | 58.04 | |
| SparseGPT+SQFTLLaMA=V2-7B, Pruning=SparseGPT, Sparsity level=0.62025.05 | 75.37 | 53.43 | 30 | 52.38 | 74.95 | 72.89 | 41.38 | 57.2 | |
| Wanda+LoRA*LLaMA=V3-8B, Pruning=Wanda, Sparsity level=0.62025.05 | 73.87 | 56.68 | 27.2 | 50.69 | 69.91 | 71.08 | 39.51 | 55.56 | |
| Wanda+SPPLLaMA=V2-7B, Pruning=Wanda, Sparsity level=0.62025.05 | 73.79 | 54.15 | 27.4 | 51.97 | 71.8 | 70.54 | 37.63 | 55.32 | |
| Wanda+LoRA*LLaMA=V2-7B, Pruning=Wanda, Sparsity level=0.62025.05 | 73.48 | 54.51 | 29 | 51.49 | 70.98 | 72.81 | 39.07 | 55.91 | |
| Wanda+SPPLLaMA=V3-8B, Pruning=Wanda, Sparsity level=0.62025.05 | 73.32 | 54.51 | 27.4 | 50.37 | 70.46 | 70.75 | 37.97 | 54.97 | |
| SparseGPT+LoRA*LLaMA=V2-7B, Pruning=SparseGPT, Sparsity level=0.62025.05 | 73.16 | 53.79 | 28.6 | 51.13 | 77.73 | 72.34 | 38.48 | 56.46 | |
| SparseGPT+SPPLLaMA=V2-7B, Pruning=SparseGPT, Sparsity level=0.62025.05 | 72.69 | 58.48 | 28.8 | 52.63 | 70.55 | 69.91 | 39.16 | 56.03 |