MIT Press

Table 2:

Dataset statistics for the machine translation, code generation, and image captioning tasks. Length represents the average number of tokens for target sentences of the training set.

Dataset	Train	Dev	Test	Length
WMT16 Ro-En	620k	2000	2000	26.48
WMT18 En-Tr	207k	3007	3000	25.81
KFTT En-Ja	405k	1166	1160	27.51
Django	16k	1000	1801	8.87
MS-COCO	567k	5000	5000	12.52

Dataset	Train	Dev	Test	Length
WMT16 Ro-En	620k	2000	2000	26.48
WMT18 En-Tr	207k	3007	3000	25.81
KFTT En-Ja	405k	1166	1160	27.51
Django	16k	1000	1801	8.87
MS-COCO	567k	5000	5000	12.52