List your company

Natural Language Understanding on GLUE (All Tasks)

95.18SST-2SIFT
49.149661.099873.0585.0002Apr 6, 2025Jun 1, 2025Jul 27, 2025Sep 21, 2025Nov 16, 2025Jan 11, 2026Mar 8, 2026

Evaluation results

MethodLinks
SIFTPre-trained Backbone=RoBERTa-Base, Evaluation Protocol=Fine-tuning2026.0395.1862.3992.7377.9890.2892.5988.72——85.9187.4——————
RoBERTaPrecision (W/A)=32-322025.1195.18—90.4480.51—92.6891.687.7587.2386.4——62.593.188.7891.0490.72
AdamWPre-trained Backbone=RoBERTa-Base, Evaluation Protocol=Fine-tuning, Parameters=Full params2026.0394.8464.1692.0780.5190.8192.9389.12——86.5587.97——————
LISA-wor-no-scalePre-trained Backbone=RoBERTa-Base, Evaluation Protocol=Fine-tuning, Sampling=Without Replacement, Gradient Scaling=False2026.0394.8462.3592.3678.3490.4592.5988.73——85.987.55——————
GoLorePre-trained Backbone=RoBERTa-Base, Evaluation Protocol=Fine-tuning2026.0394.7262.6291.9578.790.4992.3587.83——85.7587.33——————
LISA-worPre-trained Backbone=RoBERTa-Base, Evaluation Protocol=Fine-tuning, Sampling=Without Replacement, Gradient Scaling=True2026.0394.7262.9892.8279.0690.4992.8888.73——86.1887.72——————
LISAPre-trained Backbone=RoBERTa-Base, Evaluation Protocol=Fine-tuning2026.0394.561.7692.2578.3490.1992.6888.77——85.7587.54——————
LISA-scalePre-trained Backbone=RoBERTa-Base, Evaluation Protocol=Fine-tuning, Gradient Scaling=True2026.0394.2761.5191.9176.1790.292.7188.81——85.3987.55——————
MASS2025.0494.274.185.884.988.990.985.1——85.983.299.4—————
Finetuned2025.0493.67587.585.988.791.585.4——86.483.4100—————
SMILE-22025.0493.573.28584.887.391.384.9——85.584.299—————
BERTPrecision (W/A)=32-322025.1193.35—87.7572.56—91.8490.9184.9484.7683.83——59.691.3587.8289.789.22
SMILE-12025.0493.37284.384.18791.384.7——85.184.298.5—————
QUARKBackbone=RoBERTa, Precision (W/A)=6-62025.1192.72—83.4278.75—86.9889.0383.3382.6881.77——49.4987.6586.3490.0289.92
BERT-BaseTruncation Ratio=0%2025.1292.556.789.865.885.790.387.682.783.281.6———————
Task Arithmetic2025.0492.370.578.480.577.890.283.6——78.957.891.3—————
Outlier Sup.Backbone=RoBERTa, Precision (W/A)=6-62025.1192.2—83.3367.51—86.8290.0183.3883.3279.62——46.3587.586.8686.8386.93
Outlier Sup.Backbone=BERT, Precision (W/A)=6-62025.1191.86—83.3370.76—89.8288.9482.0281.6981.19——54.487.4584.6988.6588.55
Weight Averaging2025.0491.769.179.481.273.289.883.9——78.962.691.3—————
Ties-Merging2025.0491.770.378.981.678.390.283.5——79.96592.5—————
bert2BERTModel=RoBERTa-B, FLOPs Savings=7.2%, Walltime Savings=6.6%2026.0391.6341.0486.03—85.28990.21——80.7382——————
Prehab-SVDTruncation Ratio=20%2025.1291.449.290.161.485.287.685.381.281.879———————
QUARKBackbone=BERT, Precision (W/A)=6-62025.1191.09—83.5670.15—90.289.8382.7881.6182.25——61.4188.1986.8888.3887.5
MangoModel=RoBERTa-B, FLOPs Savings=22.2%, Walltime Savings=20.4%2026.0390.8344.3684.31—85.8790.4390.23——81.1581.99——————
ScratchModel=RoBERTa-B, FLOPs Savings=0.0%, Walltime Savings=0.0%2026.0390.7145.8783.33—85.4489.7590.38——81.1582.59——————
BoTModel=RoBERTa-B, FLOPs Savings=54.7%, Walltime Savings=51.5%2026.0390.64484.82—85.790.1390.19——81.1982.91——————
Prehab-SVDTruncation Ratio=40%2025.1290.542.188.758.182.181.883.87978.975.8———————
LiGOModel=RoBERTa-B, FLOPs Savings=22.1%, Walltime Savings=20.3%2026.0390.3744.9683.82—85.1690.1990.18——81.0182.38——————
QUARKBackbone=RoBERTa, Precision (W/A)=4-42025.1190.28—66.7273.95—52.4558.3936.2637.3359.02——10.4975.4440.7588.7187.03
SVD-LLMTruncation Ratio=20%2025.1289.845.288.452.784.787.184.780.881.276.7———————
ScratchModel=RoBERTa-S, FLOPs Savings=0.0%, Walltime Savings=0.0%2026.0389.5620.3979.41—82.5984.988.32——74.5476.64——————
EasyQuantBackbone=RoBERTa, Precision (W/A)=6-62025.1189.56—76.4762.45—74.0489.1274.9675.8770.01——9.2884.3185.5280.8982.38
BoTModel=RoBERTa-S, FLOPs Savings=35.7%, Walltime Savings=35.8%2026.0389.3323.1876.47—82.9384.6188.71——74.5776.78——————
SVD-LLMTruncation Ratio=40%2025.1289.14186.948.680.978.482.978.177.273.2———————
FWSVDTruncation Ratio=20%2025.128928.782.555.282.186.283.870.569.872.2———————
GFWSVDTruncation Ratio=20%2025.1288.839.985.552.364.482.882.871.872.371.1———————
WeMoE2025.0488.672.551.981.582.189.369.6——76.87988.9—————
WSModel=RoBERTa-S, FLOPs Savings=32.4%, Walltime Savings=32.3%2026.0388.521.6278.19—81.4383.5488.15——74.0476.85——————
PercentileBackbone=RoBERTa, Precision (W/A)=6-62025.1188.19—78.4362.82—77.1687.4472.2373.6870.98——20.7384.4382.2179.4179.64
Prehab-SVDTruncation Ratio=60%2025.1287.730.984.756.768.167.177.169.869.567.7———————
PercentileBackbone=BERT, Precision (W/A)=6-62025.1187.27—79.961.73—79.3780.1972.471.6972.93——37.3285.0972.5886.3887.29
EasyQuantBackbone=BERT, Precision (W/A)=6-62025.1187.27—77.4565.34—84.9481.8175.8275.6674.49——38.1682.5175.3185.586.33
GFWSVDTruncation Ratio=40%2025.1285.221.972.446.260.180.980.963.263.363.8———————
QUARKBackbone=BERT, Precision (W/A)=4-42025.1184.39—67.1251.77—67.0872.7639.1540.6459.91——43.7680.4335.9963.6764.82
SVDTruncation Ratio=20%2025.1284.12.168.45261.671.772.757.958.358.8———————
SVD-LLMTruncation Ratio=60%2025.1283.625.9835266.265.176.365.865.964.7———————
FWSVDTruncation Ratio=40%2025.1282.8082.350.28181.780.154.654.964.1———————
FWSVDTruncation Ratio=60%2025.1279.9081.151.66866.77144.544.758.4———————
SVDTruncation Ratio=40%2025.1278.907.550.943.764.559.645.145.643.8———————
GFWSVDTruncation Ratio=60%2025.1273.77.20.746.967.7676847.248.547.4———————
Outlier Sup.Backbone=BERT, Precision (W/A)=4-42025.1170.41—60.7852.71—59.6465.8727.0926.5539.5——-1.636823.340.48-0.29
SVDTruncation Ratio=60%2025.1262.70052.311.456.149.835.635.233.5———————
Outlier Sup.Backbone=RoBERTa, Precision (W/A)=4-42025.1151.15—63.8452.35—49.560.3632.7532.3735.98——-3.2476.7615.75-3.15-2.52
PercentileBackbone=BERT, Precision (W/A)=4-42025.1150.92—68.3847.29—50.3660.7132.7332.9535.64——-4.1581.2223.16-9.18-8.63
PercentileBackbone=RoBERTa, Precision (W/A)=4-42025.1150.92—63.3847.29—50.5452.9231.8231.7236.89——0.0177.2238.15-1.52-0.99