From 3cac867fac3f8717b25e3026b97b456a4e748039 Mon Sep 17 00:00:00 2001 From: Sam Shleifer Date: Fri, 28 Aug 2020 15:22:33 -0400 Subject: [PATCH] t5 model should make decoder_attention_mask (#6800) --- src/transformers/tokenization_t5.py | 1 - tests/test_tokenization_t5.py | 2 -- 2 files changed, 3 deletions(-) diff --git a/src/transformers/tokenization_t5.py b/src/transformers/tokenization_t5.py index 0c9966a48e..571fabe690 100644 --- a/src/transformers/tokenization_t5.py +++ b/src/transformers/tokenization_t5.py @@ -372,6 +372,5 @@ class T5Tokenizer(PreTrainedTokenizer): **kwargs, ) model_inputs["labels"] = labels_and_decoder_mask["input_ids"] - model_inputs["decoder_attention_mask"] = labels_and_decoder_mask["attention_mask"] self.prefix_tokens = [] return model_inputs diff --git a/tests/test_tokenization_t5.py b/tests/test_tokenization_t5.py index 16bf536b25..130680a57d 100644 --- a/tests/test_tokenization_t5.py +++ b/tests/test_tokenization_t5.py @@ -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 ) self.assertEqual(32, batch["labels"].shape[1]) - self.assertEqual(32, batch["decoder_attention_mask"].shape[1]) # test None max_target_length batch = tokenizer.prepare_seq2seq_batch( 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["decoder_attention_mask"].shape[1]) def test_outputs_not_longer_than_maxlen(self): tokenizer = self.t5_base_tokenizer