List your company

Long-context language understanding on SCROLLS (test)

47.4Average ScoreCOLT5-XL
25.45631.15336.8542.547Aug 1, 2022Sep 8, 2022Oct 16, 2022Nov 23, 2022Dec 31, 2022Feb 7, 2023Mar 17, 2023

Evaluation results

MethodLinks
COLT5-XLModel Scale=XL, Inference Speed (inf)=2.3 samples/s, Fine-tuning Speed (fn)=0.5 samples/s2023.0347.4——40.5——20——22.5—31.1——88.486.153.948.136.1
LONGT5-XLModel Scale=XL, Inference Speed (no MQA)=0.2 samples/s, Inference Speed (with MQA)=1.2 samples/s, Fine-tuning Speed (fn)=0.4 samples/s2023.0346.6——40.5——19.4——21.3—29.3——88.285.353.14635.9
LONGT5-LModel Scale=Large, Inference Speed (no MQA)=0.3 samples/s, Inference Speed (with MQA)=3.0 samples/s, Fine-tuning Speed (fn)=1.3 samples/s2023.0345.3——39.5——19.1——21.4—27.2——87.384.252.340.635.7
COLT5-LModel Scale=Large, Inference Speed (inf)=5.0 samples/s, Fine-tuning Speed (fn)=2.0 samples/s2023.0345.3——39.7——20.5——21—27.7——88.784.549.839.935.9
LONGT5-BModel Scale=Base, Inference Speed (no MQA)=0.6 samples/s, Inference Speed (with MQA)=7.4 samples/s, Fine-tuning Speed (fn)=3.7 samples/s2023.0343.1——37.7——19.2——20.4—23——85.682.246.637.935.4
COLT5-BModel Scale=Base, Inference Speed (inf)=11.2 samples/s, Fine-tuning Speed (fn)=6.5 samples/s2023.0342.4——37.9——18.7——18.4—23.3——86.582.442.136.535.3
LongT5XLChunk/Input=255/16K, #Params=3B2022.0841.954.728.230.235.89.621.134.911.823.553.129.34642.188.2————
LongT5largeChunk/Input=255/16K, #Params=770M2022.0840.554.227.829.835.69.221.235.11223.352.327.240.638.687.3————
LongT5baseChunk/Input=255/16K, #Params=220M2022.0838.253.527.329.334.89.621.133.91122.846.62337.936.685.6————
BARTlarge-SLEDChunk/Input=256/16K, #Params=406M2022.083857.526.327.435.28.719.434.2112246.924.134.834.887.3————
UL2Chunk/Input=2K/2K, #Params=20B2022.0837.953.626.128.832.97.819.431.18.520.437.624.245.840.788.7————
BARTbase-SLEDChunk/Input=256/16K, #Params=139M2022.0835.454.724.425.432.77.919.133.811.722.641.121.529.730.485.6————
LEDbaseChunk/Input=256/16K, #Params=162M2022.0833.656.827.329.230617.531.38.620.534.82128.528.382.9————
T5base-SLEDChunk/Input=256/16K, #Params=220M2022.0833.346.620.125.124.54.616.528.48.720.54318.931.229.481.4————
BARTlargeChunk/Input=1K/1K, #Params=406M2022.0832.150.720.123.531.66.818.5329.120.829.218.334.833.979.7————
BARTbaseChunk/Input=1K/1K, #Params=139M2022.0830.64819.122.730.16.618.131.29.120.327.61632.531.677.1————
LED_SCROLLSChunk/Input=1K/16K, #Params=162M2022.0829.256.226.628.824.24.515.425.16.718.826.618.525.825.471.5————
T5baseChunk/Input=1K/1K, #Params=220M2022.0826.333.212.120.421.43.61524.25.918.616.31531.928.676.3————