NLP tasks
Benchmarks
TaskDatasetSOTA resultTrendResults
7 NLP Tasks (test)
88.9Average Accuracy
38
7 NLP tasks (test)
79.2Accuracy
22
NLP tasks (QASC, WikiQA, etc.)
82.4Accuracy
19
8 NLP Tasks (avg)
82.21Accuracy
10
eleven NLP tasks
73.1Average Accuracy
10
11 NLP tasks Symbol-Tuning (held-out)
86.4Accuracy
9
14 standard NLP tasks suite (held-out)
86.9StoryCloze
8
NLP Tasks CSQA CosmosQA SocialIQA
74.94CSQA Accuracy
3
10 NLP tasks (BoolQ, CB, COPA, H-SWAG, MultiRC, RTE, Story Cloze, WiC, Winogrande, WSC) (test)
92.1BoolQ
3