fix defaults for layer drop things (#918)

Summary:
recent layerdrop related changes break existing models because they assume presence of certain args
Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/918

Reviewed By: huihuifan

Differential Revision: D18578572

Pulled By: alexeib

fbshipit-source-id: 368c2d5b3add55864bf59516820807303aac6001

fairseq/models/roberta/model.py

@@ -383,6 +383,8 @@ def base_architecture(args):
     args.attention_dropout = getattr(args, 'attention_dropout', 0.1)
     args.activation_dropout = getattr(args, 'activation_dropout', 0.0)
     args.pooler_dropout = getattr(args, 'pooler_dropout', 0.0)
+    args.encoder_layers_to_keep = getattr(args, 'encoder_layers_to_keep', None)
+    args.encoder_layerdrop = getattr(args, 'encoder_layerdrop', 0.0)
 
 
 @register_model_architecture('roberta', 'roberta_base')