Average 7 Commonsense Reasoning Tasks
72.04Avg Accuracyw/o pruning
Evaluation Results
| Method | Links | |
|---|---|---|
| w/o pruningModel=Mixtral-8x7B, Pruning Ratio=0%2025.02 | 72.04 | |
| w/o pruningModel=Qwen3-8B, Pruning Ratio=0%2025.02 | 71.73 | |
| PASERModel=Mixtral-8x7B, Pruning Ratio=25%2025.02 | 70.87 | |
| w/o pruningModel=LLaMA3.1-8B, Pruning Ratio=0%2025.02 | 70.81 | |
| PASERModel=Qwen3-8B, Pruning Ratio=25%2025.02 | 70.58 | |
| w/o pruningModel=Qwen2.5-7B, Pruning Ratio=0%2025.02 | 70.49 | |
| PASERModel=LLaMA3.1-8B, Pruning Ratio=25%2025.02 | 70.05 | |
| PASERModel=Qwen2.5-7B, Pruning Ratio=25%2025.02 | 69.14 | |
| Full DataModel=Qwen3-8B, Pruning Ratio=25%2025.02 | 68.62 | |
| NuggetsModel=Qwen3-8B, Pruning Ratio=25%2025.02 | 68.24 | |
| Full DataModel=LLaMA3.1-8B, Pruning Ratio=25%2025.02 | 67.95 | |
| IFDModel=Qwen3-8B, Pruning Ratio=25%2025.02 | 67.95 | |
| NuggetsModel=LLaMA3.1-8B, Pruning Ratio=25%2025.02 | 67.58 | |
| NuggetsModel=Qwen2.5-7B, Pruning Ratio=25%2025.02 | 67.42 | |
| NuggetsModel=Mixtral-8x7B, Pruning Ratio=25%2025.02 | 67.24 | |
| Full DataModel=Qwen2.5-7B, Pruning Ratio=25%2025.02 | 67.08 | |
| IFDModel=Qwen2.5-7B, Pruning Ratio=25%2025.02 | 66.93 | |
| IFDModel=LLaMA3.1-8B, Pruning Ratio=25%2025.02 | 66.82 | |
| IFDModel=Mixtral-8x7B, Pruning Ratio=25%2025.02 | 66.48 | |
| RandomModel=Qwen3-8B, Pruning Ratio=25%2025.02 | 66.37 | |
| Instruction MiningModel=Qwen3-8B, Pruning Ratio=25%2025.02 | 65.81 | |
| RandomModel=LLaMA3.1-8B, Pruning Ratio=25%2025.02 | 65.73 | |
| Instruction MiningModel=LLaMA3.1-8B, Pruning Ratio=25%2025.02 | 65.34 | |
| Instruction MiningModel=Mixtral-8x7B, Pruning Ratio=25%2025.02 | 65.21 | |
| PASERQuantization=GPTQ 4 bits, Backbone=LLaMA2-13B2025.02 | 65.13 | |
| Full DataModel=Mixtral-8x7B, Pruning Ratio=25%2025.02 | 64.83 | |
| w/o trainingQuantization=w/o Quant, Backbone=LLaMA2-13B2025.02 | 64.78 | |
| RandomModel=Qwen2.5-7B, Pruning Ratio=25%2025.02 | 64.29 | |
| NuggetsQuantization=GPTQ 4 bits, Backbone=LLaMA2-13B2025.02 | 64.28 | |
| PASERQuantization=RTN 4 bits, Backbone=LLaMA2-13B2025.02 | 64.25 | |
| w/o TrainingModel=Qwen3-8B, Pruning Ratio=25%2025.02 | 64.25 | |
| w/o TrainingModel=LLaMA3.1-8B, Pruning Ratio=25%2025.02 | 63.92 | |
| IFDQuantization=GPTQ 4 bits, Backbone=LLaMA2-13B2025.02 | 63.85 | |
| Instruction MiningModel=Qwen2.5-7B, Pruning Ratio=25%2025.02 | 63.75 | |
| RandomModel=Mixtral-8x7B, Pruning Ratio=25%2025.02 | 63.75 | |
| Instruction MiningQuantization=GPTQ 4 bits, Backbone=LLaMA2-13B2025.02 | 63.63 | |
| NuggetsQuantization=RTN 4 bits, Backbone=LLaMA2-13B2025.02 | 62.98 | |
| w/o TrainingModel=Qwen2.5-7B, Pruning Ratio=25%2025.02 | 62.71 | |
| IFDQuantization=RTN 4 bits, Backbone=LLaMA2-13B2025.02 | 62.3 | |
| RandomQuantization=GPTQ 4 bits, Backbone=LLaMA2-13B2025.02 | 62.1 | |
| w/o trainingQuantization=GPTQ 4 bits, Backbone=LLaMA2-13B2025.02 | 61.97 | |
| w/o TrainingModel=Mixtral-8x7B, Pruning Ratio=25%2025.02 | 61.56 | |
| Full DataQuantization=GPTQ 4 bits, Backbone=LLaMA2-13B2025.02 | 60.98 | |
| Full DataQuantization=RTN 4 bits, Backbone=LLaMA2-13B2025.02 | 60.97 | |
| w/o trainingQuantization=RTN 4 bits, Backbone=LLaMA2-13B2025.02 | 60.57 | |
| Instruction MiningQuantization=RTN 4 bits, Backbone=LLaMA2-13B2025.02 | 60.19 | |
| BaselineRatio=1.0, Backbone=LLaMA-3-8B, Evaluation Protocol=zero-shot2026.04 | 60 | |
| BaselineRatio=1.0, Backbone=Qwen-2.5-7B, Evaluation Protocol=zero-shot2026.04 | 60 | |
| RandomQuantization=RTN 4 bits, Backbone=LLaMA2-13B2025.02 | 59.45 | |
| BaselineRatio=1.0, Backbone=LLaMA-2-13B, Evaluation Protocol=zero-shot2026.04 | 58 | |
| BaselineRatio=1.0, Backbone=LLaMA-2-7B, Evaluation Protocol=zero-shot2026.04 | 55 | |
| AA-SVDRatio=0.8, Backbone=LLaMA-2-13B, Evaluation Protocol=zero-shot2026.04 | 53 | |
| AA-SVDRatio=0.8, Backbone=Qwen-2.5-7B, Evaluation Protocol=zero-shot2026.04 | 53 | |
| AA-SVDRatio=0.8, Backbone=LLaMA-2-7B, Evaluation Protocol=zero-shot2026.04 | 50 | |
| AA-SVDRatio=0.8, Backbone=LLaMA-3-8B, Evaluation Protocol=zero-shot2026.04 | 50 | |
| BaselineRatio=1.0, Backbone=LLaMA-3-1B, Evaluation Protocol=zero-shot2026.04 | 48 | |
| SVD-LLMRatio=0.8, Backbone=LLaMA-2-13B, Evaluation Protocol=zero-shot2026.04 | 48 | |
| SVD-LLMRatio=0.8, Backbone=Qwen-2.5-7B, Evaluation Protocol=zero-shot2026.04 | 47 | |
| AA-SVDRatio=0.6, Backbone=LLaMA-2-13B, Evaluation Protocol=zero-shot2026.04 | 46 | |
| SVD-LLMRatio=0.8, Backbone=LLaMA-3-8B, Evaluation Protocol=zero-shot2026.04 | 44 | |
| AA-SVDRatio=0.6, Backbone=LLaMA-2-7B, Evaluation Protocol=zero-shot2026.04 | 44 | |
| AA-SVDRatio=0.6, Backbone=Qwen-2.5-7B, Evaluation Protocol=zero-shot2026.04 | 44 | |
| SVD-LLMRatio=0.8, Backbone=LLaMA-2-7B, Evaluation Protocol=zero-shot2026.04 | 43 | |
| AA-SVDRatio=0.6, Backbone=LLaMA-3-8B, Evaluation Protocol=zero-shot2026.04 | 41 | |
| AA-SVDRatio=0.8, Backbone=LLaMA-3-1B, Evaluation Protocol=zero-shot2026.04 | 39 | |
| SVD-LLMRatio=0.6, Backbone=LLaMA-2-13B, Evaluation Protocol=zero-shot2026.04 | 38 | |
| SVD-LLMRatio=0.6, Backbone=LLaMA-2-7B, Evaluation Protocol=zero-shot2026.04 | 35 | |
| AA-SVDRatio=0.6, Backbone=LLaMA-3-1B, Evaluation Protocol=zero-shot2026.04 | 35 | |
| SVD-LLMRatio=0.6, Backbone=Qwen-2.5-7B, Evaluation Protocol=zero-shot2026.04 | 33 | |
| SVD-LLMRatio=0.8, Backbone=LLaMA-3-1B, Evaluation Protocol=zero-shot2026.04 | 32 | |
| SVD-LLMRatio=0.6, Backbone=LLaMA-3-8B, Evaluation Protocol=zero-shot2026.04 | 32 | |
| SVD-LLMRatio=0.6, Backbone=LLaMA-3-1B, Evaluation Protocol=zero-shot2026.04 | 30 |