List your company

Zero-shot Reasoning on Reasoning Suite (ARC-e, ARC-c, HellaSwag, PIQA, Winogrande)

6,540Average Reasoning ScoreRIA+SQ+VC+EBFT
-199.01281,550.53863,300.095,049.6414Jul 3, 2025Aug 27, 2025Oct 22, 2025Dec 17, 2025Feb 10, 2026Apr 7, 2026Jun 2, 2026

Evaluation results

MethodLinks
RIA+SQ+VC+EBFTOutliers=16:256, Sparsity=8:16, Calibration Set=WikiText2, Model=LLaMA2-13B2025.076,540——————
RIA+SQOutliers=16:256, Sparsity=8:16, Calibration Set=C4, Model=LLaMA2-13B2025.076,523——————
RIA+SQ+VC+EBFTOutliers=16:256, Sparsity=8:16, Calibration Set=C4, Model=LLaMA2-13B2025.076,513——————
RIA+SQOutliers=16:256, Sparsity=8:16, Calibration Set=WikiText2, Model=LLaMA2-13B2025.076,503——————
RIA+SQ+VC+EBFTOutliers=16:256, Sparsity=2:4, Calibration Set=WikiText2, Model=LLaMA2-13B2025.076,401——————
RIA+SQ+VC+EBFTOutliers=16:256, Sparsity=2:4, Calibration Set=C4, Model=LLaMA2-13B2025.076,334——————
RIA+SQOutliers=16:256, Sparsity=2:4, Calibration Set=WikiText2, Model=LLaMA2-13B2025.076,303——————
RIA+SQOutliers=16:256, Sparsity=2:4, Calibration Set=C4, Model=LLaMA2-13B2025.076,296——————
FP16Model=3-70, type=-, bits=-2026.0675.3386.9960.4166.3682.3780.51—
LiftUQModel=3-70, type=UQ, bits=3.022026.0674.2585.8658.8765.3282.4378.77—
AWQ-g128Model=3-70, type=UQ, bits=3.132026.0673.6584.5158.3664.2682.2678.85—
EPTQ-g128Model=3-70, type=UQ, bits=3.132026.0672.4283.1255.1265.5380.5277.82—
FP16Model=2-70, type=-, bits=-2026.0672.4182.754.4464.7782.1577.98—
QuIP#Model=2-70, type=VQ, bits=3.002026.0672.1382.1155.8964.2282.2176.24—
LiftUQModel=2-70, type=UQ, bits=3.022026.0671.9682.5854.6163.9881.577.11—
BF16Bits=BF162026.0371.6481.0656.4874.9677.6968.03100
BATQuantBits=W4A8KV162026.0371.1479.9256.1473.177.9768.5999.34
GPTQBits=W4A8KV162026.0370.8180.655.8973.1677.3167.0998.82
RTNBits=W4A8KV162026.0370.7780.8155.7273.2977.0966.9398.75
GPTQBits=W4A8KV82026.0370.7480.6856.0672.9577.5366.4698.72
BATQuantBits=W4A8KV82026.0370.5779.855.6373.1577.0967.1798.5
RTNBits=W4A8KV82026.0370.5180.5155.7272.8676.5566.9398.42
SmoothQuantBits=W4A8KV82026.0370.2879.2155.0372.7676.9967.498.08
FlatQuantBits=W4A8KV162026.0370.2179.6355.7272.6676.8266.2298.01
QuaRotBits=W4A8KV82026.0370.1779.8455.3872.5476.8866.2297.92
SmoothQuantBits=W4A8KV162026.0370.1279.0455.872.3876.5566.8597.93
BRQBits=W4A8KV162026.0369.8878.8753.6773.2776.6666.9397.43
SpinQuantBits=W4A8KV162026.0369.8776.9854.6972.7678.0766.8597.52
BRQBits=W4A8KV82026.0369.7778.1152.9973.0976.8867.897.26
QuaRotBits=W4A8KV162026.0369.7378.755.272.7776.8865.1197.31
FlatQuantBits=W4A8KV82026.0369.4877.152.5672.4677.0968.1996.86
SpinQuantBits=W4A8KV82026.0369.4377.6553.572.5677.5365.996.8
Baseline (Uncompressed)Model=LLaMA3-8B, Zero-shot=true2025.0769.23——————
BATQuantBits=W4A4KV162026.0368.777.5353.3371.1275.366.2295.84
BATQuantBits=W4A8KV42026.0368.6578.5453.3369.5376.6665.1995.71
FP16Model=3-8, type=-, bits=-2026.0668.5880.0950.4360.1779.672.61—
Baseline (Uncompressed)Model=Mistral, Zero-shot=true2025.0768.55——————
GPTQBits=W4A8KV42026.0368.2576.5252.3971.2575.7365.3595.15
GPTQBits=W4A4KV162026.0367.8276.9851.2870.4775.7964.5694.44
FP16Model=2-13, type=-, bits=-2026.0667.8179.4248.2960.0779.0572.22—
RTNBits=W4A4KV162026.0367.7576.8952.4770.4474.1664.894.49
FlatQuantBits=W4A4KV162026.0367.6778.250.670.3675.6363.5494.13
RTNBits=W4A8KV42026.0367.6276.8951.9670.5475.0863.6194.22
FlatQuantBits=W4A8KV42026.0367.4677.952.1369.675.1462.5193.97
LiftUQModel=3-8, type=UQ, bits=3.022026.0667.378.8346.5958.4278.7373.95—
QuaRotBits=W4A8KV42026.0367.2976.4752.7370.1574.8162.2793.82
SmoothQuantBits=W4A4KV162026.0366.9975.7250.0970.1574.3764.6493.29
SmoothQuantBits=W4A8KV42026.0366.9373.2349.7469.6175.2466.8593.28
BRQBits=W4A8KV42026.0366.8275.9750.6870.3874.6562.4393.04
LiftUQModel=2-13, type=UQ, bits=3.022026.0666.6777.9946.2559.1678.8471.11—
SpinQuantBits=W4A8KV42026.0366.4974.0749.3269.8275.9563.392.53
VPTQModel=2-13, type=VQ, bits=3.032026.0666.3778.8346.558.578.1869.85—
VPTQModel=3-8, type=VQ, bits=3.032026.0666.3278.4544.857.8578.7871.74—
QuIP#Model=2-13, type=VQ, bits=3.002026.0666.2677.944.6258.2678.0772.45—
BRQBits=W4A4KV162026.0366.2474.7148.5568.7975.2463.9392.14
QuaRotBits=W4A4KV162026.0365.8674.2850.4367.5573.6763.3891.81
RIA+SQ+EBFTModel=Mistral, Outliers=16:256, Sparsity=8:16, Calibration Set=WikiText2, Zero-shot=true2025.0765.27——————
RIA+SQ+VC+EBFTModel=LLaMA3-8B, Outliers=16:256, Sparsity=8:16, Calibration Set=WikiText2, Zero-shot=true2025.0764.89——————
RIA+SQModel=Mistral, Outliers=16:256, Sparsity=8:16, Calibration Set=WikiText2, Zero-shot=true2025.0764.87——————
FP16Model=2-7, type=-, bits=-2026.0664.8576.2643.5257.1678.0769.22—
RIA+SQOutliers=8:256, Sparsity=8:16, Calibration Set=C4, Model=LLaMA2-13B2025.0764.52——————
RIA+SQ+VC+EBFTOutliers=4:256, Sparsity=8:16, Calibration Set=C4, Model=LLaMA2-13B2025.0764.52——————
RIA+SQ+EBFTModel=Mistral, Outliers=8:256, Sparsity=8:16, Calibration Set=WikiText2, Zero-shot=true2025.0764.48——————
RIA+SQ+VC+EBFTOutliers=8:256, Sparsity=8:16, Calibration Set=C4, Model=LLaMA2-13B2025.0764.16——————
RIA+SQModel=Mistral, Outliers=8:256, Sparsity=8:16, Calibration Set=WikiText2, Zero-shot=true2025.0764.09——————
RIA+SQ+EBFTModel=Mistral, Outliers=4:256, Sparsity=8:16, Calibration Set=WikiText2, Zero-shot=true2025.0764.07——————
RIA+SQOutliers=8:256, Sparsity=8:16, Calibration Set=WikiText2, Model=LLaMA2-13B2025.0764.03——————
RIA+SQ+VCModel=LLaMA3-8B, Outliers=16:256, Sparsity=8:16, Calibration Set=WikiText2, Zero-shot=true2025.0763.98——————
RIA+SQOutliers=4:256, Sparsity=8:16, Calibration Set=C4, Model=LLaMA2-13B2025.0763.94——————
RIA+SQModel=Mistral, Outliers=4:256, Sparsity=8:16, Calibration Set=WikiText2, Zero-shot=true2025.0763.81——————
RIA+SQ+VC+EBFTOutliers=8:256, Sparsity=8:16, Calibration Set=WikiText2, Model=LLaMA2-13B2025.0763.76——————
RIA+SQModel=LLaMA3-8B, Outliers=16:256, Sparsity=8:16, Calibration Set=WikiText2, Zero-shot=true2025.0763.75——————
LiftUQModel=2-7, type=UQ, bits=3.022026.0663.7175.0741.0256.5777.8967.97—
RIA+SQOutliers=4:256, Sparsity=8:16, Calibration Set=WikiText2, Model=LLaMA2-13B2025.0763.7——————
RIA+SQ+EBFTModel=Mistral, Outliers=-, Sparsity=8:16, Calibration Set=WikiText2, Zero-shot=true2025.0763.63——————
RIA+SQ+VC+EBFTOutliers=4:256, Sparsity=8:16, Calibration Set=WikiText2, Model=LLaMA2-13B2025.0763.63——————
SpinQuantBits=W4A4KV162026.0363.5368.1845.6567.4174.2162.1988.36
QuIP#Model=2-7, type=VQ, bits=3.002026.0663.5274.6241.8955.8577.0468.19—
RIA+SQ+VC+EBFTModel=LLaMA3-8B, Outliers=8:256, Sparsity=8:16, Calibration Set=WikiText2, Zero-shot=true2025.0763.2——————
RIA+SQModel=Mistral, Outliers=-, Sparsity=8:16, Calibration Set=WikiText2, Zero-shot=true2025.0763.04——————
RIA+SQ+VC+EBFTModel=LLaMA3-8B, Outliers=4:256, Sparsity=8:16, Calibration Set=WikiText2, Zero-shot=true2025.0762.71——————
RIA+SQ+VCModel=LLaMA3-8B, Outliers=8:256, Sparsity=8:16, Calibration Set=WikiText2, Zero-shot=true2025.0762.29——————
RIA+SQ+VC+EBFTOutliers=8:256, Sparsity=2:4, Calibration Set=C4, Model=LLaMA2-13B2025.0762.18——————
RIA+SQ+EBFTModel=Mistral, Outliers=16:256, Sparsity=2:4, Calibration Set=WikiText2, Zero-shot=true2025.0762.11——————
RIA+SQ+VC+EBFTModel=LLaMA3-8B, Outliers=-, Sparsity=8:16, Calibration Set=WikiText2, Zero-shot=true2025.0762.01——————
RIA+SQ+VC+EBFTOutliers=8:256, Sparsity=2:4, Calibration Set=WikiText2, Model=LLaMA2-13B2025.0761.84——————
RIA+SQModel=LLaMA3-8B, Outliers=8:256, Sparsity=8:16, Calibration Set=WikiText2, Zero-shot=true2025.0761.81——————
RIA+SQ+VC+EBFTModel=LLaMA3-8B, Outliers=16:256, Sparsity=2:4, Calibration Set=WikiText2, Zero-shot=true2025.0761.75——————
VPTQModel=2-7, type=VQ, bits=3.022026.0661.769.139.354.977.368—
RIA+SQ+VC+EBFTOutliers=4:256, Sparsity=2:4, Calibration Set=C4, Model=LLaMA2-13B2025.0761.5——————
RIA+SQModel=Mistral, Outliers=16:256, Sparsity=2:4, Calibration Set=WikiText2, Zero-shot=true2025.0761.38——————
RIA+SQ+VC+EBFTOutliers=4:256, Sparsity=2:4, Calibration Set=WikiText2, Model=LLaMA2-13B2025.0761.38——————
RIA+SQOutliers=8:256, Sparsity=2:4, Calibration Set=C4, Model=LLaMA2-13B2025.0761.19——————
RIA+SQ+VCModel=LLaMA3-8B, Outliers=4:256, Sparsity=8:16, Calibration Set=WikiText2, Zero-shot=true2025.0760.98——————
RIA+SQOutliers=8:256, Sparsity=2:4, Calibration Set=WikiText2, Model=LLaMA2-13B2025.0760.92——————
RIA+SQ+EBFTModel=Mistral, Outliers=8:256, Sparsity=2:4, Calibration Set=WikiText2, Zero-shot=true2025.0760.68——————
RIA+SQModel=LLaMA3-8B, Outliers=4:256, Sparsity=8:16, Calibration Set=WikiText2, Zero-shot=true2025.0760.6——————
RIA+SQOutliers=4:256, Sparsity=2:4, Calibration Set=WikiText2, Model=LLaMA2-13B2025.0760.28——————
RIA+SQ+VCModel=LLaMA3-8B, Outliers=-, Sparsity=8:16, Calibration Set=WikiText2, Zero-shot=true2025.0760.19——————
RIA+SQOutliers=4:256, Sparsity=2:4, Calibration Set=C4, Model=LLaMA2-13B2025.0760.18——————