Zero-shot Evaluation on Tasks Zero-shot (mean)
76.57mAccDense
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| DenseSparsity=0%, Backbone=LLaMA-2 70B2026.02 | 76.57 | 299 | 1 | 1,718.4 | 1 | |
| FlattenGPTSparsity=20%, Pruning Strategy=Depth, Backbone=LLaMA-2 70B2026.02 | 73.94 | 381 | 1.27 | 1,364.1 | 1.26 | |
| SLEBSparsity=10%, Pruning Strategy=Depth, Backbone=LLaMA-2 70B2026.02 | 73.14 | 336 | 1.12 | 1,529.1 | 1.12 | |
| WandaSparsity=50%, Pruning Strategy=2:4, Backbone=LLaMA-2 70B2026.02 | 72.34 | 293 | 0.98 | 1,555.5 | 1.1 | |
| SliceGPTSparsity=20%, Pruning Strategy=Width, Backbone=LLaMA-2 70B2026.02 | 72.34 | 314 | 1.05 | 1,658.7 | 1.04 | |
| SparseGPTSparsity=50%, Pruning Strategy=2:4, Backbone=LLaMA-2 70B2026.02 | 71.87 | 293 | 0.98 | 1,555.5 | 1.1 | |
| DenseSparsity=0%, Backbone=LLaMA-2 13B2026.02 | 71.76 | — | — | — | — | |
| SLEBSparsity=20%, Pruning Strategy=Depth, Backbone=LLaMA-2 70B2026.02 | 70.81 | 381 | 1.27 | 1,364.1 | 1.26 | |
| SliceGPTSparsity=25%, Pruning Strategy=Width, Backbone=LLaMA-2 70B2026.02 | 69.75 | 331 | 1.11 | 1,440.7 | 1.19 | |
| DenseSparsity=0%, Backbone=LLaMA-2 7B2026.02 | 69 | — | — | — | — | |
| FlattenGPTSparsity=20%, Pruning Strategy=Depth, Backbone=LLaMA-2 13B2026.02 | 68.27 | — | — | — | — | |
| LLM-PrunerSparsity=20%, Pruning Strategy=Width, Backbone=LLaMA-2 13B2026.02 | 67.72 | — | — | — | — | |
| SLEBSparsity=10%, Pruning Strategy=Depth, Backbone=LLaMA-2 13B2026.02 | 66.77 | — | — | — | — | |
| SliceGPTSparsity=30%, Pruning Strategy=Width, Backbone=LLaMA-2 70B2026.02 | 66.11 | 343 | 1.15 | 1,364.2 | 1.26 | |
| SliceGPTSparsity=20%, Pruning Strategy=Width, Backbone=LLaMA-2 13B2026.02 | 63.45 | — | — | — | — | |
| SparseGPTSparsity=50%, Pruning Strategy=2:4, Backbone=LLaMA-2 13B2026.02 | 63.06 | — | — | — | — | |
| SLEBSparsity=20%, Pruning Strategy=Depth, Backbone=LLaMA-2 13B2026.02 | 62.96 | — | — | — | — | |
| FlattenGPTSparsity=20%, Pruning Strategy=Depth, Backbone=LLaMA-2 7B2026.02 | 62.49 | — | — | — | — | |
| SLEBSparsity=10%, Pruning Strategy=Depth, Backbone=LLaMA-2 7B2026.02 | 62.24 | — | — | — | — | |
| LLM-PrunerSparsity=20%, Pruning Strategy=Width, Backbone=LLaMA-2 7B2026.02 | 62.15 | — | — | — | — | |
| WandaSparsity=50%, Pruning Strategy=2:4, Backbone=LLaMA-2 13B2026.02 | 61.23 | — | — | — | — | |
| SparseGPTSparsity=50%, Pruning Strategy=2:4, Backbone=LLaMA-2 7B2026.02 | 58.23 | — | — | — | — | |
| SliceGPTSparsity=20%, Pruning Strategy=Width, Backbone=LLaMA-2 7B2026.02 | 58.17 | — | — | — | — | |
| SLEBSparsity=20%, Pruning Strategy=Depth, Backbone=LLaMA-2 7B2026.02 | 56.8 | — | — | — | — | |
| WandaSparsity=50%, Pruning Strategy=2:4, Backbone=LLaMA-2 7B2026.02 | 55.59 | — | — | — | — |