SCROLLS: Standardized CompaRison Over Long Language Sequences

Benchmark Model Rank Results
long-range-modeling-on-scrollsLED Base#7Avg.: 29.16GovRep: 56.2 / 26.6 / 28.8SumScr: 24.2 / 4.5 / 15.4…
long-range-modeling-on-scrollsBART Base#8Avg.: 29.01GovRep: 47.9 / 18.6 / 22.7SumScr: 27.2 / 4.9 / 16.7…
long-range-modeling-on-scrollsNaive#9Avg.: 19.35GovRep: 45.3 / 17.9 / 20.8SumScr: 19.6 / 1.8 / 11.0…