Table 1.

Model selection for the best 5 models on the development set.

Parameter settings			BLEU score
n_block	p_drop	d_model
4	0.3	512	12.11
6		512	12.04
8		512	11.59
6		256	11.62
8		256	11.38

The n_blocks are the number of layers engaged by the Transformer.