Zero-shot Evaluation on MMLU, PIQA, Arc-E, Arc-C, Wino, and OpenQA
77.62MMLUDense
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| DenseBackbone=Qwen 2.5 14B, Evaluation Protocol=Zero-shot, Sparsity=0%2025.12 | 77.62 | 81.28 | 82.24 | 55.8 | 75.14 | 34.4 | 67.75 | |
| DenseModel=Qwen 2.5 14B, Mask Selection=Dense, Weight Update=-2025.12 | 77.62 | 81.28 | 82.24 | 55.8 | 75.14 | 34.4 | 67.75 | |
| DenseModel=Qwen 2.5 14B, Mask Selection=Dense, Weight Update=None, Sparsity=0%, Evaluation Protocol=Zero-shot2025.12 | 77.62 | 81.28 | 82.24 | 55.8 | 75.14 | 34.4 | 67.75 | |
| DenseBackbone=Qwen 2.5 7B, Evaluation Protocol=Zero-shot, Sparsity=0%2025.12 | 71.76 | 78.73 | 80.51 | 48.38 | 72.61 | 33.4 | 64.23 | |
| DenseModel=Qwen 2.5 7B, Mask Selection=Dense, Weight Update=-2025.12 | 71.76 | 78.73 | 80.51 | 48.38 | 72.61 | 33.4 | 64.23 | |
| DenseModel=Qwen 2.5 7B, Mask Selection=Dense, Weight Update=None, Sparsity=0%, Evaluation Protocol=Zero-shot2025.12 | 71.76 | 78.73 | 80.51 | 48.38 | 72.61 | 33.4 | 64.23 | |
| ThanosModel=Qwen 2.5 14B, Mask Selection=Thanos, Weight Update=Thanos, Sparsity=50% unstructured, Evaluation Protocol=Zero-shot2025.12 | 70.57 | 79.87 | 80.18 | 49.15 | 73.09 | 32.2 | 64.17 | |
| Thanos + OPTIMAModel=Qwen 2.5 14B, Mask Selection=Thanos, Weight Update=OPTIMA, Sparsity=50% unstructured, Evaluation Protocol=Zero-shot2025.12 | 70.16 | 79.6 | 81.57 | 51.37 | 73.48 | 33 | 64.86 | |
| WandaModel=Qwen 2.5 14B, Mask Selection=Wanda, Weight Update=None, Sparsity=50% unstructured, Evaluation Protocol=Zero-shot2025.12 | 69.84 | 79.16 | 81.02 | 51.28 | 73.72 | 34.6 | 64.94 | |
| SparseGPTModel=Qwen 2.5 14B, Mask Selection=SparseGPT, Weight Update=SparseGPT, Sparsity=50% unstructured, Evaluation Protocol=Zero-shot2025.12 | 69.83 | 79.6 | 80.98 | 51.02 | 72.93 | 32.8 | 64.53 | |
| SparseGPT + OPTIMAModel=Qwen 2.5 14B, Mask Selection=SparseGPT, Weight Update=OPTIMA, Sparsity=50% unstructured, Evaluation Protocol=Zero-shot2025.12 | 69.71 | 79.54 | 81.19 | 51.79 | 73.8 | 33.6 | 64.94 | |
| Wanda + OPTIMAModel=Qwen 2.5 14B, Mask Selection=Wanda, Weight Update=OPTIMA, Sparsity=50% unstructured, Evaluation Protocol=Zero-shot2025.12 | 69.29 | 79.43 | 81.19 | 52.3 | 73.8 | 33.8 | 64.97 | |
| SparseGPTModel=Qwen 2.5 14B, Mask Selection=SparseGPT, Weight Update=SparseGPT, Sparsity=2:4, Pruning Scope=MLP2025.12 | 67.45 | 77.58 | 77.61 | 44.62 | 73.88 | 32.6 | 62.29 | |
| SparseGPT + OPTIMAModel=Qwen 2.5 14B, Mask Selection=SparseGPT, Weight Update=OPTIMA, Sparsity=2:4, Pruning Scope=MLP2025.12 | 67.33 | 77.58 | 77.82 | 44.28 | 73.88 | 32 | 62.15 | |
| Wanda + OPTIMAModel=Qwen 2.5 14B, Mask Selection=Wanda, Weight Update=OPTIMA, Sparsity=2:4, Pruning Scope=MLP2025.12 | 67.03 | 77.31 | 77.82 | 46.76 | 74.27 | 32.6 | 62.63 | |
| ThanosModel=Qwen 2.5 14B, Mask Selection=Thanos, Weight Update=Thanos, Sparsity=2:4, Pruning Scope=MLP2025.12 | 66.31 | 77.64 | 77.9 | 46.25 | 72.77 | 31.2 | 62.01 | |
| Thanos + OPTIMAModel=Qwen 2.5 7B, Mask Selection=Thanos, Weight Update=OPTIMA, Sparsity=50% unstructured, Evaluation Protocol=Zero-shot2025.12 | 66.23 | 77.58 | 76.22 | 44.45 | 71.19 | 31.2 | 61.15 | |
| ThanosModel=Qwen 2.5 7B, Mask Selection=Thanos, Weight Update=Thanos, Sparsity=50% unstructured, Evaluation Protocol=Zero-shot2025.12 | 66.21 | 77.86 | 74.71 | 42.32 | 70.17 | 30.4 | 60.28 | |
| Wanda + OPTIMAModel=Qwen 2.5 7B, Mask Selection=Wanda, Weight Update=OPTIMA, Sparsity=50% unstructured, Evaluation Protocol=Zero-shot2025.12 | 66.17 | 77.69 | 76.43 | 42.66 | 71.27 | 30.6 | 60.8 | |
| SparseGPTModel=Qwen 2.5 7B, Mask Selection=SparseGPT, Weight Update=SparseGPT, Sparsity=50% unstructured, Evaluation Protocol=Zero-shot2025.12 | 66.09 | 78.07 | 75.34 | 42.75 | 71.11 | 31 | 60.73 | |
| Thanos + OPTIMAModel=Qwen 2.5 14B, Mask Selection=Thanos, Weight Update=OPTIMA, Sparsity=2:4, Pruning Scope=MLP2025.12 | 66.07 | 77.97 | 77.44 | 45.65 | 72.93 | 31.8 | 61.98 | |
| WandaModel=Qwen 2.5 7B, Mask Selection=Wanda, Weight Update=None, Sparsity=50% unstructured, Evaluation Protocol=Zero-shot2025.12 | 65.89 | 77.31 | 75.08 | 40.53 | 70.17 | 30.8 | 59.96 | |
| WandaModel=Qwen 2.5 14B, Mask Selection=Wanda, Weight Update=-, Sparsity=2:4, Pruning Scope=MLP2025.12 | 65.82 | 76.99 | 76.89 | 45.39 | 73.56 | 31.8 | 61.74 | |
| SparseGPT + OPTIMAModel=Qwen 2.5 7B, Mask Selection=SparseGPT, Weight Update=OPTIMA, Sparsity=50% unstructured, Evaluation Protocol=Zero-shot2025.12 | 65.78 | 77.64 | 75.63 | 42.92 | 71.51 | 31.6 | 60.85 | |
| DenseBackbone=Qwen 2.5 3B, Evaluation Protocol=Zero-shot, Sparsity=0%2025.12 | 65 | 78.35 | 77.31 | 44.88 | 68.43 | 29.2 | 60.53 | |
| DenseModel=Qwen 2.5 3B, Mask Selection=Dense, Weight Update=-2025.12 | 65 | 78.35 | 77.31 | 44.88 | 68.43 | 29.2 | 60.53 | |
| DenseModel=Qwen 2.5 3B, Mask Selection=Dense, Weight Update=None, Sparsity=0%, Evaluation Protocol=Zero-shot2025.12 | 65 | 78.35 | 77.31 | 44.88 | 68.43 | 29.2 | 60.53 | |
| DenseModel=LLaMA 3.1 8B, Weight Update=-, Evaluation Protocol=Zero-shot, Sparsity Type=2:42025.12 | 63.57 | 80.09 | 81.44 | 51.37 | 73.48 | 33.4 | 63.89 | |
| DenseModel=LLaMA-3.1-8B, Mask Selection=Dense, Weight Update=None, Sparsity=0%2025.12 | 63.57 | 80.09 | 81.44 | 51.37 | 73.48 | 33.4 | 63.89 | |
| Dense BaselineModel=LLaMA 3.1 8B, Mask Selection=Dense, Weight Update=-, Sparsity=0%, Evaluation Protocol=zero-shot2025.12 | 63.57 | 80.09 | 81.44 | 51.37 | 73.48 | 33.4 | 63.89 | |
| ThanosBackbone=Qwen 2.5 14B, Evaluation Protocol=Zero-shot, Sparsity=60%2025.12 | 63.03 | 77.2 | 76.05 | 43.77 | 71.98 | 29.8 | 60.31 | |
| SparseGPTBackbone=Qwen 2.5 14B, Evaluation Protocol=Zero-shot, Sparsity=60%2025.12 | 62.83 | 76.66 | 76.18 | 44.45 | 72.14 | 29.6 | 60.31 | |
| SparseGPT + OPTIMABackbone=Qwen 2.5 14B, Evaluation Protocol=Zero-shot, Sparsity=60%2025.12 | 62.22 | 76.93 | 76.47 | 44.54 | 71.67 | 29 | 60.14 | |
| Wanda + OPTIMABackbone=Qwen 2.5 14B, Evaluation Protocol=Zero-shot, Sparsity=60%2025.12 | 61.74 | 75.57 | 75.34 | 41.98 | 73.09 | 29.4 | 59.52 | |
| WandaBackbone=Qwen 2.5 14B, Evaluation Protocol=Zero-shot, Sparsity=60%2025.12 | 61.38 | 75.41 | 74.12 | 42.15 | 71.51 | 29.2 | 58.96 | |
| WandaModel=Qwen 2.5 7B, Mask Selection=Wanda, Weight Update=-, Sparsity=2:4, Pruning Scope=MLP2025.12 | 61.03 | 74.48 | 75.34 | 42.15 | 68.82 | 27.8 | 58.27 | |
| Thanos + OPTIMABackbone=Qwen 2.5 14B, Evaluation Protocol=Zero-shot, Sparsity=60%2025.12 | 60.3 | 76.39 | 76.3 | 43.77 | 72.14 | 30.6 | 59.92 | |
| SparseGPTModel=Qwen 2.5 7B, Mask Selection=SparseGPT, Weight Update=SparseGPT, Sparsity=2:4, Pruning Scope=MLP2025.12 | 60.3 | 75.57 | 75.59 | 41.38 | 71.51 | 28.2 | 58.76 | |
| ThanosModel=Qwen 2.5 7B, Mask Selection=Thanos, Weight Update=Thanos, Sparsity=2:4, Pruning Scope=MLP2025.12 | 60.12 | 74.54 | 75.08 | 41.13 | 69.93 | 28.8 | 58.27 | |
| DenseBackbone=Qwen 2.5 1.5B, Evaluation Protocol=Zero-shot, Sparsity=0%2025.12 | 59.7 | 75.73 | 75.34 | 40.96 | 63.14 | 32.2 | 57.84 | |
| DenseModel=Qwen 2.5 1.5B, Mask Selection=Dense, Weight Update=-2025.12 | 59.7 | 75.73 | 75.34 | 40.96 | 63.14 | 32.2 | 57.84 | |
| DenseModel=Qwen 2.5 1.5B, Mask Selection=Dense, Weight Update=None, Sparsity=0%, Evaluation Protocol=Zero-shot2025.12 | 59.7 | 75.73 | 75.34 | 40.96 | 63.14 | 32.2 | 57.84 | |
| ThanosBackbone=Qwen 2.5 7B, Evaluation Protocol=Zero-shot, Sparsity=60%2025.12 | 59.54 | 74.7 | 73.44 | 40.44 | 69.22 | 26.4 | 57.29 | |
| Thanos + OPTIMABackbone=Qwen 2.5 7B, Evaluation Protocol=Zero-shot, Sparsity=60%2025.12 | 58.9 | 75.35 | 72.69 | 40.1 | 69.46 | 27 | 57.25 | |
| SparseGPTBackbone=Qwen 2.5 7B, Evaluation Protocol=Zero-shot, Sparsity=60%2025.12 | 56.63 | 74.92 | 73.36 | 40.61 | 67.25 | 25.8 | 56.43 | |
| SparseGPT + OPTIMABackbone=Qwen 2.5 7B, Evaluation Protocol=Zero-shot, Sparsity=60%2025.12 | 55.55 | 75.46 | 73.78 | 40.7 | 66.93 | 26.6 | 56.5 | |
| Wanda + OPTIMABackbone=Qwen 2.5 7B, Evaluation Protocol=Zero-shot, Sparsity=60%2025.12 | 55.49 | 73.99 | 73.86 | 37.88 | 67.96 | 26.2 | 55.9 | |
| Thanos + OPTIMAModel=Qwen 2.5 7B, Mask Selection=Thanos, Weight Update=OPTIMA, Sparsity=2:4, Pruning Scope=MLP2025.12 | 54.9 | 73.94 | 70.83 | 35.15 | 69.06 | 26 | 54.98 | |
| WandaBackbone=Qwen 2.5 7B, Evaluation Protocol=Zero-shot, Sparsity=60%2025.12 | 54.58 | 72.03 | 71.68 | 37.03 | 66.46 | 25.4 | 54.53 | |
| DenseModel=LLaMA 3.2 3B, Weight Update=-, Evaluation Protocol=Zero-shot, Sparsity Type=2:42025.12 | 54.13 | 76.55 | 74.28 | 42.75 | 69.38 | 30.6 | 57.95 | |
| DenseModel=LLaMA-3.2-3B, Mask Selection=Dense, Weight Update=None, Sparsity=0%2025.12 | 54.13 | 76.55 | 74.28 | 42.75 | 69.38 | 30.6 | 57.95 | |
| Dense BaselineModel=LLaMA 3.2 3B, Mask Selection=Dense, Weight Update=-, Sparsity=0%, Evaluation Protocol=zero-shot2025.12 | 54.13 | 76.55 | 74.28 | 42.75 | 69.38 | 30.6 | 57.95 | |
| SparseGPT + OPTIMAModel=Qwen 2.5 3B, Mask Selection=SparseGPT, Weight Update=OPTIMA, Sparsity=50% unstructured, Evaluation Protocol=Zero-shot2025.12 | 53.92 | 75.35 | 70.83 | 38.31 | 66.14 | 29.6 | 55.69 | |
| Wanda + OPTIMAModel=Qwen 2.5 7B, Mask Selection=Wanda, Weight Update=OPTIMA, Sparsity=2:4, Pruning Scope=MLP2025.12 | 53.9 | 74.1 | 71.97 | 38.05 | 67.96 | 26.4 | 55.4 | |
| SparseGPT + OPTIMAModel=Qwen 2.5 7B, Mask Selection=SparseGPT, Weight Update=OPTIMA, Sparsity=2:4, Pruning Scope=MLP2025.12 | 53.9 | 74.32 | 72.43 | 37.46 | 69.61 | 27.8 | 55.92 | |
| ThanosModel=Qwen 2.5 3B, Mask Selection=Thanos, Weight Update=Thanos, Sparsity=50% unstructured, Evaluation Protocol=Zero-shot2025.12 | 52.61 | 75.52 | 70.54 | 36.69 | 66.61 | 28.4 | 55.06 | |
| SparseGPTModel=Qwen 2.5 3B, Mask Selection=SparseGPT, Weight Update=SparseGPT, Sparsity=50% unstructured, Evaluation Protocol=Zero-shot2025.12 | 52.49 | 74.65 | 71.34 | 36.86 | 64.64 | 28.2 | 54.7 | |
| Thanos + OPTIMAModel=Qwen 2.5 3B, Mask Selection=Thanos, Weight Update=OPTIMA, Sparsity=50% unstructured, Evaluation Protocol=Zero-shot2025.12 | 52.11 | 75.46 | 70.12 | 37.29 | 66.69 | 28.2 | 54.98 | |
| Wanda + OPTIMAModel=Qwen 2.5 3B, Mask Selection=Wanda, Weight Update=OPTIMA, Sparsity=50% unstructured, Evaluation Protocol=Zero-shot2025.12 | 52 | 74.37 | 72.18 | 38.05 | 66.77 | 28.6 | 55.33 | |
| SparseGPTModel=LLaMA-3.1-8B, Mask Selection=SparseGPT, Weight Update=SparseGPT, Sparsity=50% unstructured2025.12 | 51.32 | 76.19 | 73.02 | 41.27 | 70.88 | 29.4 | 57.01 | |
| ThanosModel=LLaMA-3.1-8B, Mask Selection=Thanos, Weight Update=Thanos, Sparsity=50% unstructured2025.12 | 50.36 | 77.04 | 74.92 | 42.58 | 70.96 | 30 | 57.64 | |
| Thanos with OPTIMAModel=LLaMA-3.1-8B, Mask Selection=Thanos, Weight Update=OPTIMA, Sparsity=50% unstructured2025.12 | 50.17 | 76.5 | 74.16 | 41.89 | 70.24 | 28.4 | 56.89 | |
| DenseModel=Gemma 2 2B, Weight Update=-, Evaluation Protocol=Zero-shot, Sparsity Type=2:42025.12 | 49.33 | 78.24 | 80.22 | 46.93 | 68.82 | 31.4 | 59.16 | |
| DenseModel=Gemma-2-2B, Mask Selection=Dense, Weight Update=None, Sparsity=0%2025.12 | 49.33 | 78.24 | 80.22 | 46.93 | 68.82 | 31.4 | 59.16 | |
| Dense BaselineModel=Gemma 2 2B, Mask Selection=Dense, Weight Update=-, Sparsity=0%, Evaluation Protocol=zero-shot2025.12 | 49.33 | 78.24 | 80.22 | 46.93 | 68.82 | 31.4 | 59.16 | |
| SparseGPT with OPTIMAModel=LLaMA-3.1-8B, Mask Selection=SparseGPT, Weight Update=OPTIMA, Sparsity=50% unstructured2025.12 | 49.31 | 76.61 | 74.28 | 42.83 | 70.88 | 28.2 | 57.02 | |
| WandaModel=Qwen 2.5 3B, Mask Selection=Wanda, Weight Update=None, Sparsity=50% unstructured, Evaluation Protocol=Zero-shot2025.12 | 49.09 | 73.23 | 71.46 | 38.48 | 65.43 | 26.4 | 54.02 | |
| Wanda with OPTIMAModel=LLaMA-3.1-8B, Mask Selection=Wanda, Weight Update=OPTIMA, Sparsity=50% unstructured2025.12 | 48.85 | 76.71 | 73.82 | 42.32 | 70.32 | 28.2 | 56.7 | |
| ThanosModel=Qwen 2.5 1.5B, Mask Selection=Thanos, Weight Update=Thanos, Sparsity=50% unstructured, Evaluation Protocol=Zero-shot2025.12 | 48.4 | 71.76 | 66.84 | 33.7 | 62.83 | 27.2 | 51.79 | |
| Thanos + OPTIMAModel=Qwen 2.5 1.5B, Mask Selection=Thanos, Weight Update=OPTIMA, Sparsity=50% unstructured, Evaluation Protocol=Zero-shot2025.12 | 48.21 | 72.03 | 67.26 | 33.53 | 62.04 | 26.2 | 51.55 | |
| WandaModel=LLaMA-3.1-8B, Mask Selection=Wanda, Weight Update=None, Sparsity=50% unstructured2025.12 | 47.79 | 75.68 | 72.56 | 40.7 | 70.09 | 27.4 | 55.7 | |
| DenseBackbone=Qwen 2.5 0.5B, Evaluation Protocol=Zero-shot, Sparsity=0%2025.12 | 47.36 | 69.97 | 64.18 | 29.18 | 55.8 | 24.4 | 48.48 | |
| DenseModel=Qwen 2.5 0.5B, Mask Selection=Dense, Weight Update=-2025.12 | 47.36 | 69.97 | 64.18 | 29.18 | 55.8 | 24.4 | 48.48 | |
| DenseModel=Qwen 2.5 0.5B, Mask Selection=Dense, Weight Update=None, Sparsity=0%, Evaluation Protocol=Zero-shot2025.12 | 47.36 | 69.97 | 64.18 | 29.18 | 55.8 | 24.4 | 48.48 | |
| SparseGPT + OPTIMAModel=Qwen 2.5 1.5B, Mask Selection=SparseGPT, Weight Update=OPTIMA, Sparsity=50% unstructured, Evaluation Protocol=Zero-shot2025.12 | 46.96 | 71.82 | 65.45 | 33.02 | 61.8 | 26.2 | 50.87 | |
| SparseGPTModel=Qwen 2.5 1.5B, Mask Selection=SparseGPT, Weight Update=SparseGPT, Sparsity=50% unstructured, Evaluation Protocol=Zero-shot2025.12 | 46.8 | 71.65 | 66.75 | 33.62 | 62.27 | 25.6 | 51.12 | |
| WandaModel=Qwen 2.5 3B, Mask Selection=Wanda, Weight Update=-, Sparsity=2:4, Pruning Scope=MLP2025.12 | 46.68 | 70.08 | 64.77 | 31.66 | 61.48 | 22.2 | 49.48 | |
| Wanda + OPTIMAModel=Qwen 2.5 3B, Mask Selection=Wanda, Weight Update=OPTIMA, Sparsity=2:4, Pruning Scope=MLP2025.12 | 46.55 | 71.44 | 64.9 | 31.31 | 64.17 | 25.4 | 50.63 | |
| SparseGPTModel=LLaMA 3.1 8B, Weight Update=OPTIMA, Evaluation Protocol=Zero-shot, Sparsity Type=2:42025.12 | 46.04 | 73.72 | 69.95 | 36.77 | 69.61 | 27 | 53.85 | |
| SparseGPTModel=LLaMA 3.1 8B, Weight Update=SparseGPT, Evaluation Protocol=Zero-shot, Sparsity Type=2:42025.12 | 45.62 | 73.83 | 69.15 | 35.84 | 69.22 | 25.6 | 53.21 | |
| Wanda + OPTIMAModel=Qwen 2.5 1.5B, Mask Selection=Wanda, Weight Update=OPTIMA, Sparsity=50% unstructured, Evaluation Protocol=Zero-shot2025.12 | 45.61 | 72.36 | 66.62 | 32.34 | 61.8 | 24.6 | 50.55 | |
| WandaModel=LLaMA 3.1 8B, Weight Update=OPTIMA, Evaluation Protocol=Zero-shot, Sparsity Type=2:42025.12 | 45.45 | 73.39 | 69.57 | 36.18 | 68.9 | 25.2 | 53.12 | |
| WandaModel=Qwen 2.5 1.5B, Mask Selection=Wanda, Weight Update=None, Sparsity=50% unstructured, Evaluation Protocol=Zero-shot2025.12 | 44.76 | 71.22 | 66.62 | 31.74 | 59.91 | 24.8 | 49.84 | |
| ThanosModel=LLaMA 3.1 8B, Weight Update=OPTIMA, Evaluation Protocol=Zero-shot, Sparsity Type=2:42025.12 | 44.41 | 74.05 | 69.95 | 36.43 | 68.59 | 25.6 | 53.17 | |
| ThanosModel=LLaMA 3.1 8B, Weight Update=Thanos, Evaluation Protocol=Zero-shot, Sparsity Type=2:42025.12 | 44.39 | 73.94 | 69.57 | 36.18 | 68.9 | 25.2 | 53.03 | |
| SparseGPTModel=Qwen 2.5 3B, Mask Selection=SparseGPT, Weight Update=SparseGPT, Sparsity=2:4, Pruning Scope=MLP2025.12 | 43.79 | 72.03 | 66.16 | 31.31 | 64.96 | 25.2 | 50.58 | |
| Thanos + OPTIMAModel=Qwen 2.5 3B, Mask Selection=Thanos, Weight Update=OPTIMA, Sparsity=2:4, Pruning Scope=MLP2025.12 | 43.56 | 70.73 | 64.06 | 30.97 | 64.09 | 25.4 | 49.8 | |
| WandaModel=LLaMA 3.1 8B, Weight Update=-, Evaluation Protocol=Zero-shot, Sparsity Type=2:42025.12 | 43.42 | 73.18 | 69.23 | 35.32 | 67.32 | 25.8 | 52.38 | |
| SparseGPT + OPTIMAModel=Qwen 2.5 3B, Mask Selection=SparseGPT, Weight Update=OPTIMA, Sparsity=2:4, Pruning Scope=MLP2025.12 | 43.36 | 71.44 | 66.62 | 31.91 | 65.27 | 27 | 50.93 | |
| SparseGPTBackbone=Qwen 2.5 3B, Evaluation Protocol=Zero-shot, Sparsity=60%2025.12 | 43.16 | 71.6 | 64.35 | 32.59 | 63.3 | 23.2 | 49.7 | |
| Wanda with OPTIMAModel=LLaMA-3.2-3B, Mask Selection=Wanda, Weight Update=OPTIMA, Sparsity=50% unstructured2025.12 | 43.11 | 72.47 | 66.5 | 33.53 | 66.38 | 26.2 | 51.37 | |
| ThanosModel=Qwen 2.5 1.5B, Mask Selection=Thanos, Weight Update=Thanos, Sparsity=2:4, Pruning Scope=MLP2025.12 | 42.83 | 68.55 | 61.53 | 29.1 | 58.96 | 21.4 | 47.06 | |
| Thanos + OPTIMABackbone=Qwen 2.5 3B, Evaluation Protocol=Zero-shot, Sparsity=60%2025.12 | 42.58 | 71.27 | 64.73 | 32.68 | 63.61 | 25 | 49.98 | |
| Thanos with OPTIMAModel=LLaMA-3.2-3B, Mask Selection=Thanos, Weight Update=OPTIMA, Sparsity=50% unstructured2025.12 | 41.51 | 73.23 | 65.07 | 34.39 | 67.25 | 27 | 51.41 | |
| ThanosModel=Qwen 2.5 3B, Mask Selection=Thanos, Weight Update=Thanos, Sparsity=2:4, Pruning Scope=MLP2025.12 | 41.41 | 71.49 | 61.7 | 29.27 | 64.01 | 25.2 | 48.85 | |
| SparseGPTModel=Qwen 2.5 1.5B, Mask Selection=SparseGPT, Weight Update=SparseGPT, Sparsity=2:4, Pruning Scope=MLP2025.12 | 41.05 | 69.64 | 63.09 | 30.63 | 61.17 | 23 | 48.1 | |
| WandaModel=LLaMA-3.2-3B, Mask Selection=Wanda, Weight Update=None, Sparsity=50% unstructured2025.12 | 40.79 | 72.03 | 65.45 | 32.34 | 63.69 | 25.4 | 49.95 | |
| ThanosBackbone=Qwen 2.5 3B, Evaluation Protocol=Zero-shot, Sparsity=60%2025.12 | 40.76 | 71.38 | 63.26 | 30.63 | 61.25 | 23.4 | 48.45 | |
| SparseGPT with OPTIMAModel=LLaMA-3.2-3B, Mask Selection=SparseGPT, Weight Update=OPTIMA, Sparsity=50% unstructured2025.12 | 40.54 | 73.45 | 66.37 | 35.07 | 66.69 | 26.2 | 51.39 | |
| Thanos + OPTIMAModel=Qwen 2.5 1.5B, Mask Selection=Thanos, Weight Update=OPTIMA, Sparsity=2:4, Pruning Scope=MLP2025.12 | 40.31 | 69.75 | 62.67 | 30.38 | 58.56 | 25 | 47.78 |