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

t5 model should make decoder_attention_mask (#6800)

Browse Source
This commit is contained in:
Sam Shleifer
2020-08-28 15:22:33 -04:00
committed by GitHub
parent 20f7786453
commit 3cac867fac
2 changed files with 0 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
src/transformers/tokenization_t5.py
View File

@@ -372,6 +372,5 @@ class T5Tokenizer(PreTrainedTokenizer):
**kwargs, **kwargs,
) )
model_inputs["labels"] = labels_and_decoder_mask["input_ids"] model_inputs["labels"] = labels_and_decoder_mask["input_ids"]
model_inputs["decoder_attention_mask"] = labels_and_decoder_mask["attention_mask"]
self.prefix_tokens = [] self.prefix_tokens = []
return model_inputs return model_inputs

2
tests/test_tokenization_t5.py
View File

@@ -162,14 +162,12 @@ class T5TokenizationTest(TokenizerTesterMixin, unittest.TestCase):
src_text, tgt_texts=tgt_text, max_target_length=32, padding="max_length", return_tensors=FRAMEWORK src_text, tgt_texts=tgt_text, max_target_length=32, padding="max_length", return_tensors=FRAMEWORK
) )
self.assertEqual(32, batch["labels"].shape[1]) self.assertEqual(32, batch["labels"].shape[1])
self.assertEqual(32, batch["decoder_attention_mask"].shape[1])
# test None max_target_length # test None max_target_length
batch = tokenizer.prepare_seq2seq_batch( batch = tokenizer.prepare_seq2seq_batch(
src_text, tgt_texts=tgt_text, max_length=32, padding="max_length", return_tensors=FRAMEWORK src_text, tgt_texts=tgt_text, max_length=32, padding="max_length", return_tensors=FRAMEWORK
) )
self.assertEqual(32, batch["labels"].shape[1]) self.assertEqual(32, batch["labels"].shape[1])
self.assertEqual(32, batch["decoder_attention_mask"].shape[1])
def test_outputs_not_longer_than_maxlen(self): def test_outputs_not_longer_than_maxlen(self):
tokenizer = self.t5_base_tokenizer tokenizer = self.t5_base_tokenizer