Long-context language understanding on SCROLLS (test)
47.4Average ScoreCOLT5-XL
Evaluation Results
| Method | Links | |||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| COLT5-XLModel Scale=XL, Inference Speed (inf)=2.3 samples/s, Fine-tuning Speed (fn)=0.5 samples/s2023.03 | 47.4 | — | — | 40.5 | — | — | 20 | — | — | 22.5 | — | 31.1 | — | — | 88.4 | 86.1 | 53.9 | 48.1 | 36.1 | |
| 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/s2023.03 | 46.6 | — | — | 40.5 | — | — | 19.4 | — | — | 21.3 | — | 29.3 | — | — | 88.2 | 85.3 | 53.1 | 46 | 35.9 | |
| 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/s2023.03 | 45.3 | — | — | 39.5 | — | — | 19.1 | — | — | 21.4 | — | 27.2 | — | — | 87.3 | 84.2 | 52.3 | 40.6 | 35.7 | |
| COLT5-LModel Scale=Large, Inference Speed (inf)=5.0 samples/s, Fine-tuning Speed (fn)=2.0 samples/s2023.03 | 45.3 | — | — | 39.7 | — | — | 20.5 | — | — | 21 | — | 27.7 | — | — | 88.7 | 84.5 | 49.8 | 39.9 | 35.9 | |
| 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/s2023.03 | 43.1 | — | — | 37.7 | — | — | 19.2 | — | — | 20.4 | — | 23 | — | — | 85.6 | 82.2 | 46.6 | 37.9 | 35.4 | |
| COLT5-BModel Scale=Base, Inference Speed (inf)=11.2 samples/s, Fine-tuning Speed (fn)=6.5 samples/s2023.03 | 42.4 | — | — | 37.9 | — | — | 18.7 | — | — | 18.4 | — | 23.3 | — | — | 86.5 | 82.4 | 42.1 | 36.5 | 35.3 | |
| LongT5XLChunk/Input=255/16K, #Params=3B2022.08 | 41.9 | 54.7 | 28.2 | 30.2 | 35.8 | 9.6 | 21.1 | 34.9 | 11.8 | 23.5 | 53.1 | 29.3 | 46 | 42.1 | 88.2 | — | — | — | — | |
| LongT5largeChunk/Input=255/16K, #Params=770M2022.08 | 40.5 | 54.2 | 27.8 | 29.8 | 35.6 | 9.2 | 21.2 | 35.1 | 12 | 23.3 | 52.3 | 27.2 | 40.6 | 38.6 | 87.3 | — | — | — | — | |
| LongT5baseChunk/Input=255/16K, #Params=220M2022.08 | 38.2 | 53.5 | 27.3 | 29.3 | 34.8 | 9.6 | 21.1 | 33.9 | 11 | 22.8 | 46.6 | 23 | 37.9 | 36.6 | 85.6 | — | — | — | — | |
| BARTlarge-SLEDChunk/Input=256/16K, #Params=406M2022.08 | 38 | 57.5 | 26.3 | 27.4 | 35.2 | 8.7 | 19.4 | 34.2 | 11 | 22 | 46.9 | 24.1 | 34.8 | 34.8 | 87.3 | — | — | — | — | |
| UL2Chunk/Input=2K/2K, #Params=20B2022.08 | 37.9 | 53.6 | 26.1 | 28.8 | 32.9 | 7.8 | 19.4 | 31.1 | 8.5 | 20.4 | 37.6 | 24.2 | 45.8 | 40.7 | 88.7 | — | — | — | — | |
| BARTbase-SLEDChunk/Input=256/16K, #Params=139M2022.08 | 35.4 | 54.7 | 24.4 | 25.4 | 32.7 | 7.9 | 19.1 | 33.8 | 11.7 | 22.6 | 41.1 | 21.5 | 29.7 | 30.4 | 85.6 | — | — | — | — | |
| LEDbaseChunk/Input=256/16K, #Params=162M2022.08 | 33.6 | 56.8 | 27.3 | 29.2 | 30 | 6 | 17.5 | 31.3 | 8.6 | 20.5 | 34.8 | 21 | 28.5 | 28.3 | 82.9 | — | — | — | — | |
| T5base-SLEDChunk/Input=256/16K, #Params=220M2022.08 | 33.3 | 46.6 | 20.1 | 25.1 | 24.5 | 4.6 | 16.5 | 28.4 | 8.7 | 20.5 | 43 | 18.9 | 31.2 | 29.4 | 81.4 | — | — | — | — | |
| BARTlargeChunk/Input=1K/1K, #Params=406M2022.08 | 32.1 | 50.7 | 20.1 | 23.5 | 31.6 | 6.8 | 18.5 | 32 | 9.1 | 20.8 | 29.2 | 18.3 | 34.8 | 33.9 | 79.7 | — | — | — | — | |
| BARTbaseChunk/Input=1K/1K, #Params=139M2022.08 | 30.6 | 48 | 19.1 | 22.7 | 30.1 | 6.6 | 18.1 | 31.2 | 9.1 | 20.3 | 27.6 | 16 | 32.5 | 31.6 | 77.1 | — | — | — | — | |
| LED_SCROLLSChunk/Input=1K/16K, #Params=162M2022.08 | 29.2 | 56.2 | 26.6 | 28.8 | 24.2 | 4.5 | 15.4 | 25.1 | 6.7 | 18.8 | 26.6 | 18.5 | 25.8 | 25.4 | 71.5 | — | — | — | — | |
| T5baseChunk/Input=1K/1K, #Params=220M2022.08 | 26.3 | 33.2 | 12.1 | 20.4 | 21.4 | 3.6 | 15 | 24.2 | 5.9 | 18.6 | 16.3 | 15 | 31.9 | 28.6 | 76.3 | — | — | — | — |