Commonsense Reasoning on Winogrande (0-shot and 32-shot evaluation)
73.7Accuracy (0-shot)OPT 175B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| OPT 175BScale=175B2022.12 | 73.7 | 77.6 | |
| OPT-IML 175BScale=175B2022.12 | 73 | 74.4 | |
| DenseSparsity=0%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 72.14 | — | |
| ALPSSparsity=50%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 72.14 | — | |
| L-ADMMSparsity=50%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 71.98 | — | |
| ALPSSparsity=60%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 71.9 | — | |
| WandaSparsity=50%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 71.43 | — | |
| SparseGPTSparsity=50%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 71.35 | — | |
| SparseGPTSparsity=60%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 70.56 | — | |
| L-ADMMSparsity=60%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 70.32 | — | |
| ELSASparsity=50%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 70 | — | |
| OPT 30BScale=30B2022.12 | 69.7 | 71.7 | |
| SAFESparsity=50%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 69.14 | — | |
| WandaSparsity=60%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 68.19 | — | |
| SAFESparsity=60%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 67.88 | — | |
| OPT-IML 30BScale=30B2022.12 | 67.8 | 69 | |
| ELSASparsity=60%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 67.8 | — | |
| MagnitudeSparsity=50%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 65.35 | — | |
| L-ADMMSparsity=70%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 65.35 | — | |
| ALPSSparsity=70%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 64.56 | — | |
| SparseGPTSparsity=70%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 60.93 | — | |
| ELSASparsity=70%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 60.38 | — | |
| OPT 1.3BScale=1.3B2022.12 | 59.5 | 56.4 | |
| SAFESparsity=70%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 58.96 | — | |
| OPT-IML 1.3BScale=1.3B2022.12 | 58 | 58.7 | |
| MagnitudeSparsity=60%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 57.46 | — | |
| ALPSSparsity=80%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 54.93 | — | |
| ELSASparsity=80%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 53.75 | — | |
| SAFESparsity=90%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 53.43 | — | |
| L-ADMMSparsity=80%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 53.28 | — | |
| MagnitudeSparsity=90%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 51.78 | — | |
| WandaSparsity=90%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 51.54 | — | |
| WandaSparsity=70%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 50.83 | — | |
| SAFESparsity=80%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 50.51 | — | |
| ELSASparsity=90%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 50.43 | — | |
| WandaSparsity=80%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 50.04 | — | |
| L-ADMMSparsity=90%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 50.04 | — | |
| SparseGPTSparsity=90%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 49.49 | — | |
| SparseGPTSparsity=80%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 49.33 | — | |
| MagnitudeSparsity=70%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 49.25 | — | |
| MagnitudeSparsity=80%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 49.25 | — | |
| ALPSSparsity=90%, Backbone=Llama-2 13B, Protocol=Zero-shot2025.10 | 48.3 | — |