Accuracy on LLaMA2-7B zero-shot evaluation suite
67.18Zero-Shot AccuracyQTIP
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| QTIPVariant=2-bit, Model Size=2.15 GB, Data=12.58M2026.02 | 67.18 | — | — | — | — | — | — | — | — | |
| AQLM + PVVariant=2-bit-1x16, Model Size=2.38 GB, Data=8.00M2026.02 | 64.04 | — | — | — | — | — | — | — | — | |
| AQLM + PVVariant=2-bit-2x8, Model Size=2.15 GB, Data=8.00M2026.02 | 63.68 | — | — | — | — | — | — | — | — | |
| AQLMVariant=2-bit-1x16, Model Size=2.38 GB, Data=8.00M2026.02 | 63.46 | — | — | — | — | — | — | — | — | |
| AQLMVariant=2-bit-2x8, Model Size=2.15 GB, Data=8.00M2026.02 | 59.53 | — | — | — | — | — | — | — | — | |
| NANOQUANTVariant=2-bit, Model Size=1.68 GB, Data=1.05M2026.02 | 56.9 | — | — | — | — | — | — | — | — | |
| DenseComp. Rate=0%, Base Model=LLaMA2-7B, Fine-tuning=None2025.12 | 54.79 | — | — | — | — | — | — | — | — | |
| SkipCatComp. Rate=20%, Base Model=LLaMA2-7B, Fine-tuning=LoRA (Alpaca)2025.12 | 54.41 | — | — | — | — | — | — | — | 0.39 | |
| SVD-LLMComp. Rate=20%, Base Model=LLaMA2-7B, Fine-tuning=LoRA (Alpaca)2025.12 | 51.02 | — | — | — | — | — | — | — | 3.78 | |
| NANOQUANTVariant=1-bit, Model Size=1.24 GB, Data=1.05M2026.02 | 51.01 | — | — | — | — | — | — | — | — | |
| SkipCatComp. Rate=40%, Base Model=LLaMA2-7B, Fine-tuning=LoRA (Alpaca)2025.12 | 48.65 | — | — | — | — | — | — | — | 6.15 | |
| SVD-LLMComp. Rate=40%, Base Model=LLaMA2-7B, Fine-tuning=LoRA (Alpaca)2025.12 | 46.91 | — | — | — | — | — | — | — | 7.88 | |
| SkipCatComp. Rate=60%, Base Model=LLaMA2-7B, Fine-tuning=LoRA (Alpaca)2025.12 | 41.16 | — | — | — | — | — | — | — | 13.64 | |
| SVD-LLMComp. Rate=60%, Base Model=LLaMA2-7B, Fine-tuning=LoRA (Alpaca)2025.12 | 39.5 | — | — | — | — | — | — | — | 15.29 | |
| SkipCatComp. Rate=80%, Base Model=LLaMA2-7B, Fine-tuning=LoRA (Alpaca)2025.12 | 32.82 | — | — | — | — | — | — | — | 21.97 | |
| SVD-LLMComp. Rate=80%, Base Model=LLaMA2-7B, Fine-tuning=LoRA (Alpaca)2025.12 | 32.33 | — | — | — | — | — | — | — | 22.47 | |
| DenseSparsity=0%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-Shot2025.12 | — | 77.74 | 76.02 | 68.98 | 74.58 | 46.25 | 44.2 | 79.05 | — | |
| FLAPSparsity=30%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-Shot2025.12 | — | 44.71 | 56.58 | 61.72 | 53.83 | 31.23 | 37 | 71.27 | — | |
| LLM surgeonSparsity=30%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-Shot2025.12 | — | — | 60.72 | 61.09 | 63.09 | 36.69 | — | 73.56 | — | |
| LLM surgeonSparsity=40%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-Shot2025.12 | — | — | 48.04 | 54.38 | 52.31 | 30.29 | — | 69.26 | — | |
| ModeGPTSparsity=30%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-Shot2025.12 | — | — | 63.26 | 67.32 | 63.26 | 38.73 | — | 70.4 | — | |
| ModeGPTSparsity=40%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-Shot2025.12 | — | — | 53.01 | 61.96 | 49.45 | 30.03 | — | 64.96 | — | |
| O-FANGSparsity=30%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-Shot2025.12 | — | 63.06 | 67.33 | 67.32 | 67.34 | 39.93 | 40.4 | 73.45 | — | |
| O-FANGSparsity=40%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-Shot2025.12 | — | 61.19 | 58.72 | 64.48 | 58.16 | 33.11 | 40.2 | 68.82 | — | |
| OBCSparsity=30%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-Shot2025.12 | — | 64.68 | 64.34 | 66.06 | 62.88 | 36.43 | 38.2 | 72.58 | — | |
| OBCSparsity=40%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-Shot2025.12 | — | 63 | 52.04 | 60.62 | 55.68 | 31.14 | 37.6 | 67.19 | — | |
| SliceGPTSparsity=30%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-Shot2025.12 | — | 38.32 | 49.09 | 60.69 | 49.58 | 31.74 | 32.6 | 64.91 | — | |
| SliceGPTSparsity=40%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-Shot2025.12 | — | — | 34.8 | 53.43 | 36.49 | 24.57 | — | 54.9 | — | |
| SoBPSparsity=30%, Backbone=LLaMA2-7B, Evaluation Protocol=Zero-Shot2025.12 | — | 71.19 | 67.27 | 66.22 | 59.81 | 37.63 | 38.4 | 73.5 | — |