SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

revert black formatting to conform with lib style

Browse Source
This commit is contained in:
Rémi Louf
2019-10-17 09:27:56 +02:00
parent 9b71fc9a18
commit 624a5644cc
1 changed files with 2 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
transformers/modeling_seq2seq.py
View File

@@ -43,13 +43,7 @@ class PreTrainedSeq2seq(PreTrainedModel):
self.decoder = decoder self.decoder = decoder
@classmethod @classmethod
def from_pretrained( def from_pretrained(cls, encoder_pretrained_model_name_or_path, decoder_pretrained_model_name_or_path, *model_args, **kwargs):
cls,
encoder_pretrained_model_name_or_path,
decoder_pretrained_model_name_or_path,
*model_args,
**kwargs
):
r""" Instantiates an encoder and a decoder from one or two base classes r""" Instantiates an encoder and a decoder from one or two base classes
of the library from pre-trained model checkpoints. of the library from pre-trained model checkpoints.
@@ -190,7 +184,7 @@ class Model2Model(PreTrainedSeq2seq):
def tie_weights(self): def tie_weights(self):
""" Tying the encoder and decoders' embeddings together. """ Tying the encoder and decoders' embeddings together.
We need for each to get down to the embedding weights. However the We need for each to get down to the embedding weights. However the
different model classes are inconsistent to that respect: different model classes are inconsistent to that respect:
- BertModel: embeddings.word_embeddings - BertModel: embeddings.word_embeddings
- RoBERTa: embeddings.word_embeddings - RoBERTa: embeddings.word_embeddings