List your company

Zero-shot NLP Reasoning Suite (BoolQ, HellaSwag, ARC, PIQA, OBQA)

68.59Average Accuracy (Zero-Shot Suite)LLaMA-7B
33.510842.617951.72560.8321May 19, 2023Nov 11, 2023May 6, 2024Oct 30, 2024Apr 25, 2025Oct 19, 2025Apr 14, 2026

Evaluation results

MethodLinks
LLaMA-7BPruning Ratio=0%, Evaluation Prompts=Official2023.0568.5976.579.876.170.172.847.657.2—
Original Llama-7BCR=0%, # Params=6.74B, Re-trained=N/A2024.1066.07————————
LLaMA-13B*Pruning Ratio=0%2023.0564.9768.4778.8976.2470.0974.5844.5442—
BlockPruning Ratio=20%, Post-training=w/ tune, Importance Estimation=Element¹2023.0564.0270.3177.9175.1667.8871.0942.4143.4—
LLaMA-7B*Pruning Ratio=0%, Evaluation Prompts=Modified2023.0563.2573.1878.3572.9967.0167.4541.3842.4—
BLAST16CR=20%, # Params=5.41B, Re-trained=No2024.1062.94————————
L2Pruning Ratio=20%, Post-training=w/ tune2023.0562.1273.2476.7771.8664.6467.5939.9340.8—
BlockPruning Ratio=20%, Post-training=w/o tune, Importance Estimation=Element¹2023.0561.7967.6877.1573.4165.1168.3538.442.4—
ChannelPruning Ratio=20%, Post-training=w/ tune, Importance Estimation=Element¹2023.0560.2469.276.5568.8966.3862.0838.9939.6—
Element2Pruning Ratio=20%, Post-training=Tune2023.0560.0769.5476.4468.1165.1163.4337.8840—
RandomPruning Ratio=20%, Post-training=w/ tune2023.0559.7864.1976.0668.8963.366.8838.3140.8—
Low-RankCR=20%, # Params=5.41B, Re-trained=No2024.1059.57————————
Element1Pruning Ratio=20%, Post-training=Tune2023.0559.2364.6277.268.863.1464.3136.7739.8—
VectorPruning Ratio=20%, Post-training=Tune2023.0557.2365.7574.764.5259.3560.6536.2639.4—
RandomPruning Ratio=20%, Post-training=w/o tune2023.0557.0963.3373.1863.5460.8564.4436.2638—
Element2Pruning Ratio=20%, Post-training=None2023.0556.8259.3975.5765.3461.3359.1837.1239.8—
Element1Pruning Ratio=20%, Post-training=None2023.0556.6957.0675.6866.859.8360.9436.5240—
BLAST16CR=50%, # Params=3.56B, Re-trained=Yes2024.1056.22————————
ChannelPruning Ratio=20%, Post-training=Tune2023.0555.5759.0873.3964.0260.5457.9535.5838.4—
Llama-3.2-1B-InstructSparsity=Dense2026.0453.88————————
VectorPruning Ratio=20%, Post-training=None2023.0553.2561.4471.7157.2754.2255.7733.9638.4—
RandomPruning Ratio=20%, Post-training=None2023.0552.6961.8371.3356.2654.4657.0732.8535—
L2Pruning Ratio=20%, Post-training=w/o tune2023.0551.0861.567.5752.957.5450.1331.1436.8—
ChannelPruning Ratio=20%, Post-training=w/o tune, Importance Estimation=Element¹2023.0550.2962.3966.8749.1758.9649.6231.8333.2—
Monarch (BLR)CR=20%, # Params=5.41B, Re-trained=No2024.1048.91————————
Low-RankCR=50%, # Params=3.51B, Re-trained=Yes2024.1048.4————————
MOONSHOT-SparseGPTSparsity=0.5, kp=0.92026.0447.85———————61.9
MOONSHOT-SparseGPTSparsity=0.5, kp=1.02026.0447.7————————
MOONSHOT-SparseGPTSparsity=0.5, kp=0.52026.0447.64———————38.1
MOONSHOT-SparseGPTSparsity=0.5, kp=0.752026.0447.56———————47.62
MOONSHOT-SparseGPTSparsity=0.5, kp=0.12026.0447.4———————38.1
MOONSHOT-SparseGPTSparsity=0.5, kp=0.252026.0447.4———————42.86
MOONSHOT-SparseGPTSparsity=0.5, kp=0.02026.0446.16———————0
ChannelPruning Ratio=20%, Post-training=None2023.0545.3862.7562.7341.451.0741.3827.930.4—
MOONSHOT-SparseGPTSparsity=0.6, kp=0.92026.0443.03———————71.43
MOONSHOT-SparseGPTSparsity=0.6, kp=0.752026.0443.02———————76.19
MOONSHOT-SparseGPTSparsity=0.6, kp=0.52026.0442.84———————57.14
L2Pruning Ratio=20%, Post-training=None2023.0542.6559.665837.0452.4133.1228.5829.8—
MOONSHOT-SparseGPTSparsity=0.6, kp=0.12026.0442.56———————42.86
MOONSHOT-SparseGPTSparsity=0.6, kp=0.252026.0442.55———————47.62
MOONSHOT-SparseGPTSparsity=0.6, kp=1.02026.0442.51————————
MOONSHOT-SparseGPTSparsity=0.6, kp=0.02026.0440.8———————14.29
MOONSHOT-SparseGPTSparsity=0.7, kp=0.752026.0438.79———————85.71
MOONSHOT-SparseGPTSparsity=0.7, kp=0.92026.0438.68———————90.48
MOONSHOT-SparseGPTSparsity=0.7, kp=0.52026.0438.57———————80.95
MOONSHOT-SparseGPTSparsity=0.7, kp=0.12026.0438.43———————80.95
MOONSHOT-SparseGPTSparsity=0.7, kp=0.252026.0438.33———————80.95
MOONSHOT-SparseGPTSparsity=0.7, kp=1.02026.0437.37————————
MOONSHOT-SparseGPTSparsity=0.7, kp=0.02026.0435.66———————4.76
Monarch (BLR)CR=50%, # Params=3.50B, Re-trained=Yes2024.1035.03————————
Block-DiagonalCR=50%, # Params=3.50B, Re-trained=Yes2024.1034.86————————