Zero-shot Classification on (BoolQ, HellaSwag, WinoGrande, ARC-e, ARC-c, PIQA, OBQA)
81.65BoolQ AccuracyDense
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| DenseSparsity=-, Zero-shot=true2026.04 | 81.65 | 60.71 | 71.11 | 77.53 | 46.16 | 77.91 | 35.2 | 64.32 | — | |
| Llama-3.2-3B-Instruct (Dense)Sparsity=0.02026.04 | 78.47 | 52.24 | 67.4 | 73.99 | 43.43 | 75.79 | 27.4 | 59.82 | — | |
| Dense BaselineSparsity=Dense, Backbone=Llama-3.2-3B-Instruct2026.04 | 78.47 | 52.24 | 67.4 | 73.99 | 43.43 | 75.79 | 27.4 | 59.82 | — | |
| MOONSHOT-SparseGPTSparsity=0.5, lambda=0.5, Backbone=Llama-3.2-3B-Instruct, Zero-shot=true2026.04 | 76.13 | 44.87 | 63.04 | 65.74 | 34.33 | 71.84 | 21.93 | 53.98 | 42.86 | |
| MOONSHOT-SparseGPTSparsity=0.5, lambda=1.0, Backbone=Llama-3.2-3B-Instruct, Zero-shot=true2026.04 | 76.06 | 45.08 | 64.59 | 64.52 | 34.22 | 71.16 | 23.33 | 54.14 | — | |
| MOONSHOT-SparseGPTSparsity=0.5, lambda=0.75, Backbone=Llama-3.2-3B-Instruct, Zero-shot=true2026.04 | 75.95 | 44.88 | 62.93 | 66.11 | 34.07 | 71.47 | 22.13 | 53.94 | 38.1 | |
| MOONSHOT-SparseGPTSparsity=0.5, lambda=0.9, Backbone=Llama-3.2-3B-Instruct, Zero-shot=true2026.04 | 75.81 | 44.89 | 63.35 | 66.16 | 34.33 | 71.65 | 22.33 | 54.08 | 42.86 | |
| MOONSHOT-SparseGPTSparsity=0.5, lambda=0.25, Backbone=Llama-3.2-3B-Instruct, Zero-shot=true2026.04 | 75.77 | 44.81 | 63.27 | 65.5 | 34.3 | 71.8 | 22.13 | 53.94 | 38.1 | |
| MOONSHOT-SparseGPTSparsity=0.7, Zero-shot=true2026.04 | 75.71 | 40.01 | 61.12 | 57.41 | 28.16 | 68.64 | 21.73 | 50.4 | 80.95 | |
| MOONSHOT-SparseGPTSparsity=0.5, lambda=0.1, Backbone=Llama-3.2-3B-Instruct, Zero-shot=true2026.04 | 75.52 | 44.79 | 62.88 | 65.42 | 34.3 | 71.85 | 22 | 53.82 | 38.1 | |
| MOONSHOT-SparseGPTSparsity=0.5, lambda=0.0, Backbone=Llama-3.2-3B-Instruct, Zero-shot=true2026.04 | 74.55 | 43.43 | 61.64 | 64.37 | 33.25 | 70.51 | 21.73 | 52.78 | 4.76 | |
| MOONSHOT-WandaSparsity=0.5, Lambda (λ)=1.0, Backbone Model=Llama-3.2-3B-Instruct2026.04 | 74.28 | 43.43 | 63.85 | 64.66 | 34.5 | 70.26 | 21.27 | 53.18 | — | |
| MOONSHOT-WandaSparsity=0.5, Lambda (λ)=0.75, Backbone Model=Llama-3.2-3B-Instruct2026.04 | 73.86 | 42.9 | 62.67 | 66.22 | 35.01 | 71.4 | 21.93 | 53.43 | 61.9 | |
| MOONSHOT-WandaSparsity=0.5, Lambda (λ)=0.9, Backbone Model=Llama-3.2-3B-Instruct2026.04 | 73.86 | 43.2 | 63.59 | 66.4 | 35.32 | 71.25 | 21.6 | 53.6 | 52.38 | |
| MOONSHOT-WandaSparsity=0.5, Lambda (λ)=0.25, Backbone Model=Llama-3.2-3B-Instruct2026.04 | 73.79 | 42.7 | 61.88 | 66.22 | 34.19 | 71.42 | 21.93 | 53.16 | 47.62 | |
| MOONSHOT-WandaSparsity=0.5, Lambda (λ)=0.5, Backbone Model=Llama-3.2-3B-Instruct2026.04 | 73.37 | 42.88 | 62.77 | 66.33 | 34.7 | 71.27 | 22.07 | 53.34 | 57.14 | |
| MOONSHOT-WandaSparsity=0.5, Lambda (λ)=0.1, Backbone Model=Llama-3.2-3B-Instruct2026.04 | 73.21 | 42.57 | 62.04 | 66.02 | 33.79 | 70.96 | 22.33 | 52.99 | 42.86 | |
| MOONSHOT-WandaSparsity=0.5, Lambda (λ)=0.0, Backbone Model=Llama-3.2-3B-Instruct2026.04 | 73.04 | 42.38 | 60.8 | 66.26 | 33.73 | 70.98 | 21.47 | 52.67 | 38.1 | |
| DenseSparsity=None, Model=Llama-3.2-3B2026.04 | 72.72 | 55.3 | 69.22 | 74.37 | 42.41 | 76.71 | 31.2 | 60.28 | — | |
| SparseGPTSparsity=0.7, Zero-shot=true2026.04 | 71.86 | 38.59 | 59.93 | 53.65 | 28.21 | 67.16 | 22.73 | 48.88 | 19.05 | |
| MOONSHOT-SparseGPTSparsity=0.6, lambda=0.9, Backbone=Llama-3.2-3B-Instruct, Zero-shot=true2026.04 | 70.37 | 36.9 | 59.67 | 55.57 | 27.9 | 65.96 | 17.33 | 47.67 | 76.19 | |
| MOONSHOT-SparseGPTSparsity=0.6, lambda=0.75, Backbone=Llama-3.2-3B-Instruct, Zero-shot=true2026.04 | 70.32 | 36.9 | 59.69 | 55.84 | 27.36 | 66.25 | 17.47 | 47.69 | 76.19 | |
| MOONSHOT-SparseGPTSparsity=0.6, lambda=0.1, Backbone=Llama-3.2-3B-Instruct, Zero-shot=true2026.04 | 70.11 | 36.57 | 59.54 | 54.11 | 26.68 | 65.81 | 17.8 | 47.23 | 61.9 | |
| MOONSHOT-SparseGPTSparsity=0.6, lambda=0.25, Backbone=Llama-3.2-3B-Instruct, Zero-shot=true2026.04 | 70.08 | 36.84 | 58.83 | 55.3 | 27.13 | 66.43 | 18.07 | 47.53 | 76.19 | |
| MOONSHOT-SparseGPTSparsity=0.6, lambda=0.5, Backbone=Llama-3.2-3B-Instruct, Zero-shot=true2026.04 | 69.87 | 36.9 | 59.93 | 55.23 | 27.36 | 66.41 | 17.93 | 47.66 | 90.48 | |
| MOONSHOT-SparseGPTSparsity=0.6, lambda=1.0, Backbone=Llama-3.2-3B-Instruct, Zero-shot=true2026.04 | 69.73 | 36.61 | 59.3 | 53.27 | 26.22 | 65.16 | 17.27 | 46.79 | — | |
| MOONSHOT-SparseGPTSparsity=0.6 (Alpha Pruning), Model=Llama-3.2-3B2026.04 | 68.73 | 39.34 | 62.27 | 56.52 | 26.93 | 69.13 | 20.33 | 49.04 | 85.71 | |
| SparseGPTSparsity=0.6 (Alpha Pruning), Model=Llama-3.2-3B2026.04 | 68.17 | 38.62 | 61.98 | 53.37 | 26 | 68.06 | 19.8 | 48 | 14.29 | |
| MOONSHOT-SparseGPTSparsity=0.6, Model=Llama-3.2-3B2026.04 | 67.76 | 39.13 | 61.01 | 57.59 | 27.79 | 69.44 | 20 | 48.96 | 95.24 | |
| MOONSHOT-SparseGPTSparsity=0.6, lambda=0.0, Backbone=Llama-3.2-3B-Instruct, Zero-shot=true2026.04 | 67.55 | 34.8 | 56.41 | 49.13 | 24.06 | 64.78 | 16.73 | 44.78 | 9.52 | |
| MOONSHOT-SparseGPTSparsity=0.6 (OWL), Model=Llama-3.2-3B2026.04 | 67.41 | 40.64 | 61.62 | 57.39 | 28.16 | 69.73 | 20.6 | 49.36 | 71.43 | |
| WandaSparsity=0.6 (OWL), Model=Llama-3.2-3B2026.04 | 67.26 | 37.18 | 59.06 | 51.89 | 25.54 | 66.47 | 17.2 | 46.37 | 33.33 | |
| SparseGPTSparsity=0.6, Model=Llama-3.2-3B2026.04 | 66.82 | 38.14 | 60.91 | 53.89 | 26.28 | 67.75 | 18.47 | 47.47 | 4.76 | |
| MOONSHOT-WandaSparsity=0.7, Zero-shot=true2026.04 | 66.68 | 34.11 | 54.75 | 48.22 | 21.16 | 64.4 | 14.53 | 43.41 | 80.95 | |
| SparseGPTSparsity=0.6 (OWL), Model=Llama-3.2-3B2026.04 | 66.64 | 39.89 | 61.96 | 55.57 | 27.53 | 68.72 | 21.2 | 48.79 | 28.57 | |
| MOONSHOT-WandaSparsity=0.6, Lambda (λ)=0.9, Backbone Model=Llama-3.2-3B-Instruct2026.04 | 66.41 | 34.49 | 57.51 | 51.46 | 25.65 | 64.54 | 16.2 | 45.18 | 80.95 | |
| WandaSparsity=0.6 (Alpha Pruning), Model=Llama-3.2-3B2026.04 | 65.93 | 35.95 | 57.51 | 51.09 | 24.32 | 66.03 | 16.87 | 45.39 | 38.1 | |
| MOONSHOT-WandaSparsity=0.6, Lambda (λ)=0.75, Backbone Model=Llama-3.2-3B-Instruct2026.04 | 65.89 | 34.25 | 56.99 | 51.5 | 25.74 | 64.33 | 16.67 | 45.05 | 76.19 | |
| SparseGPTSparsity=2:4, Model=Llama-3.2-3B2026.04 | 65.64 | 38.31 | 59.93 | 55.63 | 26.14 | 68.34 | 20.87 | 47.83 | 52.38 | |
| MOONSHOT-SparseGPTSparsity=2:4, Model=Llama-3.2-3B2026.04 | 65.58 | 38.06 | 59.3 | 55.99 | 26.56 | 67.94 | 20.93 | 47.77 | 47.62 | |
| MOONSHOT-WandaSparsity=0.6, Lambda (λ)=1.0, Backbone Model=Llama-3.2-3B-Instruct2026.04 | 65.54 | 34.33 | 57.14 | 48.72 | 25.06 | 63.98 | 16.67 | 44.49 | — | |
| MOONSHOT-WandaSparsity=0.6, Lambda (λ)=0.5, Backbone Model=Llama-3.2-3B-Instruct2026.04 | 65.38 | 33.92 | 56.91 | 51.02 | 25.17 | 64.25 | 15.87 | 44.65 | 47.62 | |
| MOONSHOT-OSSCARSparsity=0.1 (structured), Model=Llama-3.2-3B2026.04 | 64.91 | 49.5 | 61.43 | 66.39 | 34.41 | 75.54 | 26.8 | 54.14 | 66.67 | |
| WandaSparsity=0.6, Model=Llama-3.2-3B2026.04 | 64.82 | 35.12 | 56.56 | 50.58 | 23.83 | 65.58 | 16.93 | 44.77 | 38.1 | |
| MOONSHOT-WandaSparsity=0.6, Lambda (λ)=0.25, Backbone Model=Llama-3.2-3B-Instruct2026.04 | 64.69 | 33.55 | 56.75 | 51.32 | 24.91 | 64.24 | 15.33 | 44.4 | 33.33 | |
| MOONSHOT-WandaSparsity=0.6, Lambda (λ)=0.1, Backbone Model=Llama-3.2-3B-Instruct2026.04 | 64.48 | 33.09 | 55.62 | 50.74 | 24.94 | 64.18 | 15 | 44.01 | 23.81 | |
| OSSCARSparsity=0.1 (structured), Model=Llama-3.2-3B2026.04 | 64.38 | 49.03 | 61.3 | 65.82 | 34.1 | 75.7 | 27.33 | 53.95 | 33.33 | |
| WandaSparsity=2:4, Model=Llama-3.2-3B2026.04 | 64.29 | 34.16 | 55.88 | 50.88 | 25.28 | 65.25 | 17.13 | 44.7 | 38.1 | |
| WandaSparsity=0.7, Zero-shot=true2026.04 | 64.05 | 32.17 | 54.22 | 43.95 | 20.71 | 61.35 | 17 | 41.92 | 19.05 | |
| DenseBase Model=Llama-3.2-1B, Sparsity=None, Evaluation Protocol=Zero-shot2026.04 | 64.01 | 47.73 | 60.14 | 65.15 | 31.23 | 74.32 | 26.4 | 52.71 | — | |
| MOONSHOT-WandaSparsity=0.6, Lambda (λ)=0.0, Backbone Model=Llama-3.2-3B-Instruct2026.04 | 63.9 | 32.57 | 55.67 | 49.89 | 23.83 | 63.64 | 14.47 | 43.42 | 19.05 | |
| MOONSHOT-WandaSparsity=0.6 (OWL), Model=Llama-3.2-3B2026.04 | 63.73 | 37.4 | 58.93 | 54.31 | 25.68 | 67.05 | 17 | 46.3 | 66.67 | |
| MOONSHOT-WandaSparsity=0.6 (Alpha Pruning), Model=Llama-3.2-3B2026.04 | 63.38 | 36.05 | 57.51 | 54.15 | 25.43 | 66.96 | 16.27 | 45.68 | 61.9 | |
| MOONSHOT-SparseGPTSparsity=0.7, lambda=1.0, Backbone=Llama-3.2-3B-Instruct, Zero-shot=true2026.04 | 62.92 | 28.22 | 52.12 | 34.69 | 18.86 | 57.29 | 12.73 | 38.12 | — | |
| MOONSHOT-SparseGPTSparsity=0.7, lambda=0.25, Backbone=Llama-3.2-3B-Instruct, Zero-shot=true2026.04 | 62.7 | 28.57 | 51.51 | 35.17 | 19 | 58.29 | 11.87 | 38.16 | 57.14 | |
| MOONSHOT-SparseGPTSparsity=0.7, lambda=0.9, Backbone=Llama-3.2-3B-Instruct, Zero-shot=true2026.04 | 62.63 | 28.83 | 52.33 | 35.69 | 19.2 | 58.14 | 12.2 | 38.43 | 71.43 | |
| MOONSHOT-SparseGPTSparsity=0.7, lambda=0.75, Backbone=Llama-3.2-3B-Instruct, Zero-shot=true2026.04 | 62.6 | 28.77 | 51.83 | 36.27 | 18.69 | 58.23 | 12.33 | 38.39 | 61.9 | |
| MOONSHOT-SparseGPTSparsity=0.7, lambda=0.1, Backbone=Llama-3.2-3B-Instruct, Zero-shot=true2026.04 | 62.59 | 28.53 | 51.83 | 35.16 | 19.11 | 58.12 | 12 | 38.19 | 57.14 | |
| MOONSHOT-SparseGPTSparsity=0.7, lambda=0.5, Backbone=Llama-3.2-3B-Instruct, Zero-shot=true2026.04 | 62.59 | 28.79 | 52.01 | 35.97 | 19.2 | 58.65 | 12 | 38.46 | 66.67 | |
| MOONSHOT-SparseGPTBase Model=Llama-3.2-1B, Sparsity=0.6, Evaluation Protocol=Zero-shot2026.04 | 62.36 | 32.49 | 53.09 | 46.49 | 21.3 | 63.22 | 15.73 | 42.1 | 57.14 | |
| MOONSHOT-SparseGPTBase Model=Llama-3.2-1B, Sparsity=0.6 (Alpha Pruning), Evaluation Protocol=Zero-shot2026.04 | 62.29 | 32.53 | 53.7 | 46.3 | 21.99 | 63.13 | 16.6 | 42.36 | 66.67 | |
| SparseGPTBase Model=Llama-3.2-1B, Sparsity=0.6 (OWL), Evaluation Protocol=Zero-shot2026.04 | 62.2 | 32.86 | 53.67 | 44.14 | 23.46 | 62.59 | 18 | 42.42 | 33.33 | |
| MOONSHOT-SparseGPTBase Model=Llama-3.2-1B, Sparsity=0.6 (OWL), Evaluation Protocol=Zero-shot2026.04 | 62.2 | 33.66 | 53.54 | 46.37 | 23.41 | 64.04 | 16.4 | 42.8 | 66.67 | |
| SparseGPTBase Model=Llama-3.2-1B, Sparsity=0.6 (Alpha Pruning), Evaluation Protocol=Zero-shot2026.04 | 62.08 | 32 | 53.88 | 45.29 | 22.01 | 62.02 | 17.6 | 42.13 | 33.33 | |
| MOONSHOT-SparseGPTBase Model=Llama-3.2-1B, Sparsity=2:4, Evaluation Protocol=Zero-shot2026.04 | 61.98 | 31.47 | 53.09 | 45.37 | 20.28 | 62.13 | 14.33 | 41.24 | 42.86 | |
| MOONSHOT-SparseGPTSparsity=0.7, lambda=0.0, Backbone=Llama-3.2-3B-Instruct, Zero-shot=true2026.04 | 61.95 | 27.16 | 49.99 | 30.89 | 18.03 | 56.31 | 12.27 | 36.66 | 9.52 | |
| MOONSHOT-WandaBase Model=Llama-3.2-1B, Sparsity=0.6 (OWL), Evaluation Protocol=Zero-shot2026.04 | 61.81 | 30.47 | 51.51 | 40.92 | 20.71 | 60.36 | 13.27 | 39.86 | 66.67 | |
| MOONSHOT-WandaBase Model=Llama-3.2-1B, Sparsity=0.6 (Alpha Pruning), Evaluation Protocol=Zero-shot2026.04 | 61.69 | 29.56 | 52.33 | 39.28 | 19.65 | 60.32 | 12.07 | 39.27 | 85.71 | |
| MOONSHOT-WandaSparsity=2:4, Model=Llama-3.2-3B2026.04 | 61.6 | 34.41 | 55.51 | 52.53 | 25.6 | 65.4 | 17.2 | 44.61 | 61.9 | |
| MOONSHOT-WandaBase Model=Llama-3.2-1B, Sparsity=0.6, Evaluation Protocol=Zero-shot2026.04 | 61.56 | 29.53 | 51.64 | 40.4 | 19.6 | 61.12 | 13.4 | 39.61 | 80.95 | |
| SparseGPTBase Model=Llama-3.2-1B, Sparsity=2:4, Evaluation Protocol=Zero-shot2026.04 | 61.42 | 31.68 | 53.83 | 44.04 | 21.47 | 61.79 | 15 | 41.32 | 57.14 | |
| MOONSHOT-WandaSparsity=0.6, Model=Llama-3.2-3B2026.04 | 61.33 | 35.53 | 54.83 | 52.53 | 24.69 | 66.81 | 16.6 | 44.62 | 61.9 | |
| WandaBase Model=Llama-3.2-1B, Sparsity=0.6 (OWL), Evaluation Protocol=Zero-shot2026.04 | 61.28 | 29.88 | 52.07 | 40.22 | 20.82 | 60.03 | 14.8 | 39.87 | 33.33 | |
| MOONSHOT-WandaBase Model=Llama-3.2-1B, Sparsity=2:4, Evaluation Protocol=Zero-shot2026.04 | 61.08 | 28.51 | 50.62 | 38.41 | 19.43 | 59.36 | 12.67 | 38.58 | 71.43 | |
| SparseGPTBase Model=Llama-3.2-1B, Sparsity=0.6, Evaluation Protocol=Zero-shot2026.04 | 60.67 | 32.16 | 54.46 | 44.94 | 21.47 | 62.21 | 17.07 | 41.85 | 42.86 | |
| WandaBase Model=Llama-3.2-1B, Sparsity=0.6, Evaluation Protocol=Zero-shot2026.04 | 58.96 | 28.86 | 51.35 | 38.82 | 18.94 | 59.05 | 13.93 | 38.56 | 19.05 | |
| WandaBase Model=Llama-3.2-1B, Sparsity=0.6 (Alpha Pruning), Evaluation Protocol=Zero-shot2026.04 | 58.01 | 28.92 | 51.22 | 37.56 | 19.51 | 59.32 | 13.4 | 38.28 | 14.29 | |
| WandaBase Model=Llama-3.2-1B, Sparsity=2:4, Evaluation Protocol=Zero-shot2026.04 | 57.28 | 28.32 | 51.51 | 35.76 | 18.34 | 58.41 | 13.67 | 37.61 | 28.57 | |
| MOONSHOT-OSSCARBase Model=Llama-3.2-1B, Sparsity=0.1 (structured), Evaluation Protocol=Zero-shot2026.04 | 56.55 | 37.89 | 55.51 | 58.84 | 28.58 | 71.49 | 18.93 | 46.83 | 80.95 | |
| MOONSHOT-WandaSparsity=0.7, Lambda (λ)=1.0, Backbone Model=Llama-3.2-3B-Instruct2026.04 | 55.83 | 27.42 | 48.8 | 31 | 17.83 | 56.84 | 12.2 | 35.7 | — | |
| OSSCARBase Model=Llama-3.2-1B, Sparsity=0.1 (structured), Evaluation Protocol=Zero-shot2026.04 | 52.61 | 35.82 | 54.17 | 45.19 | 24.23 | 65.65 | 15.93 | 41.94 | 19.05 | |
| MOONSHOT-WandaSparsity=0.7, Lambda (λ)=0.75, Backbone Model=Llama-3.2-3B-Instruct2026.04 | 47.18 | 27.26 | 50.3 | 31.75 | 18 | 56.66 | 11.73 | 34.7 | 52.38 | |
| MOONSHOT-WandaSparsity=0.7, Lambda (λ)=0.5, Backbone Model=Llama-3.2-3B-Instruct2026.04 | 46.76 | 27.1 | 49.96 | 31.59 | 17.61 | 56.57 | 11.87 | 34.49 | 42.86 | |
| MOONSHOT-WandaSparsity=0.7, Lambda (λ)=0.9, Backbone Model=Llama-3.2-3B-Instruct2026.04 | 45.51 | 27.38 | 50.3 | 31.8 | 17.78 | 57.05 | 11.47 | 34.47 | 52.38 | |
| MOONSHOT-WandaSparsity=0.7, Lambda (λ)=0.25, Backbone Model=Llama-3.2-3B-Instruct2026.04 | 45.08 | 26.93 | 49.83 | 31.14 | 17.95 | 56.31 | 11.4 | 34.09 | 38.1 | |
| MOONSHOT-WandaSparsity=0.7, Lambda (λ)=0.1, Backbone Model=Llama-3.2-3B-Instruct2026.04 | 41.1 | 26.86 | 49.7 | 31.1 | 18.15 | 56.33 | 12 | 33.6 | 47.62 | |
| MOONSHOT-WandaSparsity=0.7, Lambda (λ)=0.0, Backbone Model=Llama-3.2-3B-Instruct2026.04 | 39.43 | 26.81 | 49.46 | 31.14 | 18.54 | 55.82 | 11.67 | 33.27 | 42.86 |