List your company

Zero-shot Commonsense Reasoning on PIQA, HellaSwag, WinoGrande, ARC-Easy, OpenBookQA, and MathQA (test)

59Zero-shot AccuracyOriginal
0.7615.883146.12May 15, 2026

Evaluation results

MethodLinks
OriginalModel Backbone=LLaMA-13B, Evaluation Protocol=Zero-shot2026.0559
OriginalModel Backbone=Vicuna-7B, Evaluation Protocol=Zero-shot2026.0556
ZS-SVDModel Backbone=LLaMA-13B, Pruning Ratio=20%, Evaluation Protocol=Zero-shot2026.0556
IO-SVDModel Backbone=LLaMA-13B, Pruning Ratio=20%, Evaluation Protocol=Zero-shot2026.0556
SVDLLMModel Backbone=LLaMA-13B, Pruning Ratio=20%, Evaluation Protocol=Zero-shot2026.0555
ZS-SVDModel Backbone=Vicuna-7B, Pruning Ratio=20%, Evaluation Protocol=Zero-shot2026.0554
ASVDModel Backbone=LLaMA-13B, Pruning Ratio=20%, Evaluation Protocol=Zero-shot2026.0554
IO-SVDModel Backbone=Vicuna-7B, Pruning Ratio=20%, Evaluation Protocol=Zero-shot2026.0553
OriginalModel Backbone=OPT-6.7B, Evaluation Protocol=Zero-shot2026.0552
ZS-SVDModel Backbone=OPT-6.7B, Pruning Ratio=20%, Evaluation Protocol=Zero-shot2026.0551
IO-SVDModel Backbone=OPT-6.7B, Pruning Ratio=20%, Evaluation Protocol=Zero-shot2026.0551
SVDLLMModel Backbone=Vicuna-7B, Pruning Ratio=20%, Evaluation Protocol=Zero-shot2026.0551
FWSVDModel Backbone=LLaMA-13B, Pruning Ratio=20%, Evaluation Protocol=Zero-shot2026.0543
SVDLLMModel Backbone=OPT-6.7B, Pruning Ratio=20%, Evaluation Protocol=Zero-shot2026.0541
ASVDModel Backbone=Vicuna-7B, Pruning Ratio=20%, Evaluation Protocol=Zero-shot2026.0533
ASVDModel Backbone=OPT-6.7B, Pruning Ratio=20%, Evaluation Protocol=Zero-shot2026.0532
SVDModel Backbone=LLaMA-13B, Pruning Ratio=20%, Evaluation Protocol=Zero-shot2026.0521
FWSVDModel Backbone=Vicuna-7B, Pruning Ratio=20%, Evaluation Protocol=Zero-shot2026.059
FWSVDModel Backbone=OPT-6.7B, Pruning Ratio=20%, Evaluation Protocol=Zero-shot2026.056
SVDModel Backbone=Vicuna-7B, Pruning Ratio=20%, Evaluation Protocol=Zero-shot2026.055
SVDModel Backbone=OPT-6.7B, Pruning Ratio=20%, Evaluation Protocol=Zero-shot2026.053