Fix positions for LM

Summary: Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/622

Differential Revision: D15572555

Pulled By: myleott

fbshipit-source-id: 2b81f22207b4c894ffe645af0b45c70ac0a80612

fairseq/models/transformer.py

@@ -27,6 +27,9 @@ from fairseq.modules import (
     SinusoidalPositionalEmbedding,
 )
 
+DEFAULT_MAX_SOURCE_POSITIONS = 1024
+DEFAULT_MAX_TARGET_POSITIONS = 1024
+
 
 @register_model('transformer')
 class TransformerModel(FairseqEncoderDecoderModel):
@@ -112,9 +115,9 @@ class TransformerModel(FairseqEncoderDecoderModel):
         base_architecture(args)
 
         if not hasattr(args, 'max_source_positions'):
-            args.max_source_positions = 1024
+            args.max_source_positions = DEFAULT_MAX_SOURCE_POSITIONS
         if not hasattr(args, 'max_target_positions'):
-            args.max_target_positions = 1024
+            args.max_target_positions = DEFAULT_MAX_TARGET_POSITIONS
 
         src_dict, tgt_dict = task.source_dictionary, task.target_dictionary
 

fairseq/models/transformer_lm.py

@@ -20,7 +20,6 @@ from fairseq.modules import (
     CharacterTokenEmbedder,
 )
 
-DEFAULT_MAX_SOURCE_POSITIONS = 1024
 DEFAULT_MAX_TARGET_POSITIONS = 1024
 
 
@@ -103,9 +102,7 @@ class TransformerLanguageModel(FairseqLanguageModel):
             # backward compatibility
             args.tie_adaptive_proj = True
 
-        if not hasattr(args, 'max_source_positions'):
-            args.max_source_positions = getattr(args, 'tokens_per_sample', DEFAULT_MAX_SOURCE_POSITIONS)
-        if not hasattr(args, 'max_target_positions'):
+        if getattr(args, 'max_target_positions', None) is None:
             args.max_target_positions = getattr(args, 'tokens_per_sample', DEFAULT_MAX_TARGET_POSITIONS)
 
         if args.character_embeddings:

fairseq/tasks/language_modeling.py

@@ -81,7 +81,7 @@ class LanguageModelingTask(FairseqTask):
                             help='include past target')
         parser.add_argument('--add-bos-token', action='store_true',
                             help='prepend beginning of sentence token (<s>)')
-        parser.add_argument('--max-target-positions', default=1024, type=int, metavar='N',
+        parser.add_argument('--max-target-positions', type=int, metavar='N',
                             help='max number of tokens in the target sequence')
         # fmt: on