Table 8: 

Ablation model results across three tasks.

ModelXNLI (Accuracy)TyDiQA-GoldP (F1)GEM-XSum (BLEU)
ByT5-Large (1.23B) 79.7 87.7 11.5 
mT5-Large (1.23B) 81.1 85.3 10.1 
 
(a) ByT5-36/12-668M 78.3 87.8 12.3 
(b) ByT5-24/24-718M 75.4 83.0 7.1 
(c) ByT5-12/36-768M 73.5 83.1 8.3 
 
(d) mT5-36/12-1.18B 81.5 87.1 10.8 
 
(e) ByT5-Large-Span3 79.4 87.4 10.2 
(f) ByT5-Large-Span40 78.9 88.3 12.6 
 
(g) CharT5-36/12-1.23B 79.0 87.6 11.2 
ModelXNLI (Accuracy)TyDiQA-GoldP (F1)GEM-XSum (BLEU)
ByT5-Large (1.23B) 79.7 87.7 11.5 
mT5-Large (1.23B) 81.1 85.3 10.1 
 
(a) ByT5-36/12-668M 78.3 87.8 12.3 
(b) ByT5-24/24-718M 75.4 83.0 7.1 
(c) ByT5-12/36-768M 73.5 83.1 8.3 
 
(d) mT5-36/12-1.18B 81.5 87.1 10.8 
 
(e) ByT5-Large-Span3 79.4 87.4 10.2 
(f) ByT5-Large-Span40 78.9 88.3 12.6 
 
(g) CharT5-36/12-1.23B 79.0 87.6 11.2 
Close Modal

or Create an Account

Close Modal
Close Modal