Commit 58e2c449 authored by alexeib's avatar alexeib Committed by Myle Ott
Browse files

default dropout to correct value for big transformer

parent bf47b956
Loading
Loading
Loading
Loading
+1 −0
Changes for fairseq/models/transformer.py: 1 added line, 0 removed lines.
Original line number Diff line number Diff line
@@ -435,6 +435,7 @@ def transformer_vaswani_wmt_en_de_big(args):
    args.decoder_ffn_embed_dim = 4096
    args.decoder_layers = 6
    args.decoder_attention_heads = 16
    args.dropout = 0.3


@register_model_architecture('transformer', 'transformer_wmt_en_de_big')