first commit

tests/models/deberta_v2/test_modeling_deberta_v2.py

@@ -0,0 +1,313 @@
+# Copyright 2018 Microsoft Authors and the HuggingFace Inc. team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import unittest
+
+from transformers import DebertaV2Config, is_torch_available
+from transformers.testing_utils import require_sentencepiece, require_tokenizers, require_torch, slow, torch_device
+
+from ...test_configuration_common import ConfigTester
+from ...test_modeling_common import ModelTesterMixin, ids_tensor
+from ...test_pipeline_mixin import PipelineTesterMixin
+
+
+if is_torch_available():
+    import torch
+
+    from transformers import (
+        DebertaV2ForMaskedLM,
+        DebertaV2ForMultipleChoice,
+        DebertaV2ForQuestionAnswering,
+        DebertaV2ForSequenceClassification,
+        DebertaV2ForTokenClassification,
+        DebertaV2Model,
+    )
+
+
+class DebertaV2ModelTester:
+    def __init__(
+        self,
+        parent,
+        batch_size=13,
+        seq_length=7,
+        is_training=True,
+        use_input_mask=True,
+        use_token_type_ids=True,
+        use_labels=True,
+        vocab_size=99,
+        hidden_size=32,
+        num_hidden_layers=2,
+        num_attention_heads=4,
+        intermediate_size=37,
+        hidden_act="gelu",
+        hidden_dropout_prob=0.1,
+        attention_probs_dropout_prob=0.1,
+        max_position_embeddings=512,
+        type_vocab_size=16,
+        type_sequence_label_size=2,
+        initializer_range=0.02,
+        relative_attention=False,
+        position_biased_input=True,
+        pos_att_type="None",
+        num_labels=3,
+        num_choices=4,
+        scope=None,
+    ):
+        self.parent = parent
+        self.batch_size = batch_size
+        self.seq_length = seq_length
+        self.is_training = is_training
+        self.use_input_mask = use_input_mask
+        self.use_token_type_ids = use_token_type_ids
+        self.use_labels = use_labels
+        self.vocab_size = vocab_size
+        self.hidden_size = hidden_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.intermediate_size = intermediate_size
+        self.hidden_act = hidden_act
+        self.hidden_dropout_prob = hidden_dropout_prob
+        self.attention_probs_dropout_prob = attention_probs_dropout_prob
+        self.max_position_embeddings = max_position_embeddings
+        self.type_vocab_size = type_vocab_size
+        self.type_sequence_label_size = type_sequence_label_size
+        self.initializer_range = initializer_range
+        self.num_labels = num_labels
+        self.num_choices = num_choices
+        self.relative_attention = relative_attention
+        self.position_biased_input = position_biased_input
+        self.pos_att_type = pos_att_type
+        self.scope = scope
+
+    def prepare_config_and_inputs(self):
+        input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size)
+
+        input_mask = None
+        if self.use_input_mask:
+            input_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
+
+        token_type_ids = None
+        if self.use_token_type_ids:
+            token_type_ids = ids_tensor([self.batch_size, self.seq_length], self.type_vocab_size)
+
+        sequence_labels = None
+        token_labels = None
+        choice_labels = None
+        if self.use_labels:
+            sequence_labels = ids_tensor([self.batch_size], self.type_sequence_label_size)
+            token_labels = ids_tensor([self.batch_size, self.seq_length], self.num_labels)
+            choice_labels = ids_tensor([self.batch_size], self.num_choices)
+
+        config = self.get_config()
+
+        return config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+
+    def get_config(self):
+        return DebertaV2Config(
+            vocab_size=self.vocab_size,
+            hidden_size=self.hidden_size,
+            num_hidden_layers=self.num_hidden_layers,
+            num_attention_heads=self.num_attention_heads,
+            intermediate_size=self.intermediate_size,
+            hidden_act=self.hidden_act,
+            hidden_dropout_prob=self.hidden_dropout_prob,
+            attention_probs_dropout_prob=self.attention_probs_dropout_prob,
+            max_position_embeddings=self.max_position_embeddings,
+            type_vocab_size=self.type_vocab_size,
+            initializer_range=self.initializer_range,
+            relative_attention=self.relative_attention,
+            position_biased_input=self.position_biased_input,
+            pos_att_type=self.pos_att_type,
+        )
+
+    def check_loss_output(self, result):
+        self.parent.assertListEqual(list(result.loss.size()), [])
+
+    def create_and_check_deberta_model(
+        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        model = DebertaV2Model(config=config)
+        model.to(torch_device)
+        model.eval()
+        sequence_output = model(input_ids, attention_mask=input_mask, token_type_ids=token_type_ids)[0]
+        sequence_output = model(input_ids, token_type_ids=token_type_ids)[0]
+        sequence_output = model(input_ids)[0]
+
+        self.parent.assertListEqual(list(sequence_output.size()), [self.batch_size, self.seq_length, self.hidden_size])
+
+    def create_and_check_deberta_for_masked_lm(
+        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        model = DebertaV2ForMaskedLM(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(input_ids, attention_mask=input_mask, token_type_ids=token_type_ids, labels=token_labels)
+
+        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size))
+
+    def create_and_check_deberta_for_sequence_classification(
+        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        config.num_labels = self.num_labels
+        model = DebertaV2ForSequenceClassification(config)
+        model.to(torch_device)
+        model.eval()
+        result = model(input_ids, attention_mask=input_mask, token_type_ids=token_type_ids, labels=sequence_labels)
+        self.parent.assertListEqual(list(result.logits.size()), [self.batch_size, self.num_labels])
+        self.check_loss_output(result)
+
+    def create_and_check_deberta_for_token_classification(
+        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        config.num_labels = self.num_labels
+        model = DebertaV2ForTokenClassification(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(input_ids, attention_mask=input_mask, token_type_ids=token_type_ids, labels=token_labels)
+        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.num_labels))
+
+    def create_and_check_deberta_for_question_answering(
+        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        model = DebertaV2ForQuestionAnswering(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(
+            input_ids,
+            attention_mask=input_mask,
+            token_type_ids=token_type_ids,
+            start_positions=sequence_labels,
+            end_positions=sequence_labels,
+        )
+        self.parent.assertEqual(result.start_logits.shape, (self.batch_size, self.seq_length))
+        self.parent.assertEqual(result.end_logits.shape, (self.batch_size, self.seq_length))
+
+    def create_and_check_deberta_for_multiple_choice(
+        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        model = DebertaV2ForMultipleChoice(config=config)
+        model.to(torch_device)
+        model.eval()
+        multiple_choice_inputs_ids = input_ids.unsqueeze(1).expand(-1, self.num_choices, -1).contiguous()
+        multiple_choice_token_type_ids = token_type_ids.unsqueeze(1).expand(-1, self.num_choices, -1).contiguous()
+        multiple_choice_input_mask = input_mask.unsqueeze(1).expand(-1, self.num_choices, -1).contiguous()
+        result = model(
+            multiple_choice_inputs_ids,
+            attention_mask=multiple_choice_input_mask,
+            token_type_ids=multiple_choice_token_type_ids,
+            labels=choice_labels,
+        )
+        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.num_choices))
+
+    def prepare_config_and_inputs_for_common(self):
+        config_and_inputs = self.prepare_config_and_inputs()
+        (
+            config,
+            input_ids,
+            token_type_ids,
+            input_mask,
+            sequence_labels,
+            token_labels,
+            choice_labels,
+        ) = config_and_inputs
+        inputs_dict = {"input_ids": input_ids, "token_type_ids": token_type_ids, "attention_mask": input_mask}
+        return config, inputs_dict
+
+
+@require_torch
+class DebertaV2ModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
+    all_model_classes = (
+        (
+            DebertaV2Model,
+            DebertaV2ForMaskedLM,
+            DebertaV2ForSequenceClassification,
+            DebertaV2ForTokenClassification,
+            DebertaV2ForQuestionAnswering,
+            DebertaV2ForMultipleChoice,
+        )
+        if is_torch_available()
+        else ()
+    )
+    pipeline_model_mapping = (
+        {
+            "feature-extraction": DebertaV2Model,
+            "fill-mask": DebertaV2ForMaskedLM,
+            "text-classification": DebertaV2ForSequenceClassification,
+            "token-classification": DebertaV2ForTokenClassification,
+            "zero-shot": DebertaV2ForSequenceClassification,
+        }
+        if is_torch_available()
+        else {}
+    )
+
+    is_encoder_decoder = False
+
+    def setUp(self):
+        self.model_tester = DebertaV2ModelTester(self)
+        self.config_tester = ConfigTester(self, config_class=DebertaV2Config, hidden_size=32)
+
+    def test_config(self):
+        self.config_tester.run_common_tests()
+
+    def test_deberta_model(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_deberta_model(*config_and_inputs)
+
+    def test_for_sequence_classification(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_deberta_for_sequence_classification(*config_and_inputs)
+
+    def test_for_masked_lm(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_deberta_for_masked_lm(*config_and_inputs)
+
+    def test_for_question_answering(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_deberta_for_question_answering(*config_and_inputs)
+
+    def test_for_token_classification(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_deberta_for_token_classification(*config_and_inputs)
+
+    def test_for_multiple_choice(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_deberta_for_multiple_choice(*config_and_inputs)
+
+    @slow
+    def test_model_from_pretrained(self):
+        model_name = "microsoft/deberta-v2-xlarge"
+        model = DebertaV2Model.from_pretrained(model_name)
+        self.assertIsNotNone(model)
+
+
+@require_torch
+@require_sentencepiece
+@require_tokenizers
+class DebertaV2ModelIntegrationTest(unittest.TestCase):
+    @unittest.skip(reason="Model not available yet")
+    def test_inference_masked_lm(self):
+        pass
+
+    @slow
+    def test_inference_no_head(self):
+        model = DebertaV2Model.from_pretrained("microsoft/deberta-v2-xlarge", dtype=torch.float32)
+
+        input_ids = torch.tensor([[0, 31414, 232, 328, 740, 1140, 12695, 69, 46078, 1588, 2]])
+        attention_mask = torch.tensor([[0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]])
+        with torch.no_grad():
+            output = model(input_ids, attention_mask=attention_mask)[0]
+        # compare the actual values for a slice.
+        expected_slice = torch.tensor(
+            [[[0.2356, 0.1948, 0.0369], [-0.1063, 0.3586, -0.5152], [-0.6399, -0.0259, -0.2525]]]
+        )
+        torch.testing.assert_close(output[:, 1:4, 1:4], expected_slice, rtol=1e-4, atol=1e-4)

tests/models/deberta_v2/test_tokenization_deberta_v2.py

@@ -0,0 +1,136 @@
+# Copyright 2019 Hugging Face inc.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import unittest
+
+from transformers import DebertaV2Tokenizer
+from transformers.testing_utils import get_tests_dir, require_sentencepiece, require_tokenizers
+from transformers.tokenization_utils_sentencepiece import SentencePieceExtractor
+
+from ...test_tokenization_common import TokenizerTesterMixin
+
+
+SAMPLE_VOCAB = get_tests_dir("fixtures/spiece.model")
+
+
+@require_sentencepiece
+@require_tokenizers
+class DebertaV2TokenizationTest(TokenizerTesterMixin, unittest.TestCase):
+    from_pretrained_id = "microsoft/deberta-v2-xlarge"
+    tokenizer_class = DebertaV2Tokenizer
+
+    integration_expected_tokens = ['▁This', '▁is', '▁a', '▁test', '▁😊', '▁I', '▁was', '▁born', '▁in', '▁9', '2000', ',', '▁and', '▁this', '▁is', '▁fal', 's', 'é', '.', '▁', '生', '活', '的', '真', '谛', '是', '▁Hi', '▁Hello', '▁Hi', '▁Hello', '▁Hello', '▁<', 's', '>', '▁hi', '<', 's', '>', 'there', '▁The', '▁following', '▁string', '▁should', '▁be', '▁properly', '▁encoded', ':', '▁Hello', '.', '▁But', '▁i', 'rd', '▁and', '▁', 'ป', 'ี', '▁i', 'rd', '▁', 'ด', '▁Hey', '▁how', '▁are', '▁you', '▁doing']  # fmt: skip
+    integration_expected_token_ids = [69, 13, 10, 711, 112100, 16, 28, 1022, 11, 728, 16135, 6, 7, 32, 13, 46426, 12, 5155, 4, 250, 40289, 102080, 8593, 98226, 3, 29213, 2302, 4800, 2302, 4800, 4800, 2318, 12, 2259, 8133, 9475, 12, 2259, 7493, 23, 524, 3664, 146, 26, 2141, 23085, 43, 4800, 4, 167, 306, 1893, 7, 250, 86501, 70429, 306, 1893, 250, 51857, 4839, 100, 24, 17, 381]  # fmt: skip
+    expected_tokens_from_ids = ['▁This', '▁is', '▁a', '▁test', '▁😊', '▁I', '▁was', '▁born', '▁in', '▁9', '2000', ',', '▁and', '▁this', '▁is', '▁fal', 's', 'é', '.', '▁', '生', '活', '的', '真', '[UNK]', '是', '▁Hi', '▁Hello', '▁Hi', '▁Hello', '▁Hello', '▁<', 's', '>', '▁hi', '<', 's', '>', 'there', '▁The', '▁following', '▁string', '▁should', '▁be', '▁properly', '▁encoded', ':', '▁Hello', '.', '▁But', '▁i', 'rd', '▁and', '▁', 'ป', 'ี', '▁i', 'rd', '▁', 'ด', '▁Hey', '▁how', '▁are', '▁you', '▁doing']  # fmt: skip
+    integration_expected_decoded_text = "This is a test 😊 I was born in 92000, and this is falsé. 生活的真[UNK]是 Hi Hello Hi Hello Hello <s> hi<s>there The following string should be properly encoded: Hello. But ird and ปี ird ด Hey how are you doing"
+
+    def test_do_lower_case(self):
+        # fmt: off
+        sequence = " \tHeLLo!how  \n Are yoU?  "
+        tokens_target = ["▁hello", "!", "how", "▁are", "▁you", "?"]
+        # fmt: on
+
+        extractor = SentencePieceExtractor(SAMPLE_VOCAB)
+        vocab, vocab_scores, merges = extractor.extract()
+        tokenizer = DebertaV2Tokenizer(vocab=vocab_scores, unk_token="<unk>", do_lower_case=True)
+        tokens = tokenizer.convert_ids_to_tokens(tokenizer.encode(sequence, add_special_tokens=False))
+
+        self.assertListEqual(tokens, tokens_target)
+
+    def test_split_by_punct(self):
+        # fmt: off
+        sequence = "I was born in 92000, and this is falsé!"
+        tokens_target = ["▁", "<unk>", "▁was", "▁born", "▁in", "▁9", "2000", "▁", ",", "▁and", "▁this", "▁is", "▁fal", "s", "<unk>", "▁", "!", ]
+        # fmt: on
+
+        extractor = SentencePieceExtractor(SAMPLE_VOCAB)
+        vocab, vocab_scores, merges = extractor.extract()
+        tokenizer = DebertaV2Tokenizer(vocab=vocab_scores, merges=merges, unk_token="<unk>", split_by_punct=True)
+        tokens = tokenizer.convert_ids_to_tokens(tokenizer.encode(sequence, add_special_tokens=False))
+
+        self.assertListEqual(tokens, tokens_target)
+
+    def test_do_lower_case_split_by_punct(self):
+        # fmt: off
+        sequence = "I was born in 92000, and this is falsé!"
+        tokens_target = ["▁i", "▁was", "▁born", "▁in", "▁9", "2000", "▁", ",", "▁and", "▁this", "▁is", "▁fal", "s", "<unk>", "▁", "!", ]
+        # fmt: on
+
+        extractor = SentencePieceExtractor(SAMPLE_VOCAB)
+        vocab, vocab_scores, merges = extractor.extract()
+        tokenizer = DebertaV2Tokenizer(
+            vocab=vocab_scores, merges=merges, unk_token="<unk>", do_lower_case=True, split_by_punct=True
+        )
+        tokens = tokenizer.convert_ids_to_tokens(tokenizer.encode(sequence, add_special_tokens=False))
+        self.assertListEqual(tokens, tokens_target)
+
+    def test_do_lower_case_split_by_punct_false(self):
+        # fmt: off
+        sequence = "I was born in 92000, and this is falsé!"
+        tokens_target = ["▁i", "▁was", "▁born", "▁in", "▁9", "2000", ",", "▁and", "▁this", "▁is", "▁fal", "s", "<unk>", "!", ]
+        # fmt: on
+
+        extractor = SentencePieceExtractor(SAMPLE_VOCAB)
+        vocab, vocab_scores, merges = extractor.extract()
+        tokenizer = DebertaV2Tokenizer(
+            vocab=vocab_scores, merges=merges, unk_token="<unk>", do_lower_case=True, split_by_punct=False
+        )
+        tokens = tokenizer.convert_ids_to_tokens(tokenizer.encode(sequence, add_special_tokens=False))
+
+        self.assertListEqual(tokens, tokens_target)
+
+    def test_do_lower_case_false_split_by_punct(self):
+        # fmt: off
+        sequence = "I was born in 92000, and this is falsé!"
+        tokens_target = ["▁", "<unk>", "▁was", "▁born", "▁in", "▁9", "2000", "▁", ",", "▁and", "▁this", "▁is", "▁fal", "s", "<unk>", "▁", "!", ]
+        # fmt: on
+        extractor = SentencePieceExtractor(SAMPLE_VOCAB)
+        vocab, vocab_scores, merges = extractor.extract()
+        tokenizer = DebertaV2Tokenizer(
+            vocab=vocab_scores, merges=merges, unk_token="<unk>", do_lower_case=False, split_by_punct=True
+        )
+        tokens = tokenizer.convert_ids_to_tokens(tokenizer.encode(sequence, add_special_tokens=False))
+
+        self.assertListEqual(tokens, tokens_target)
+
+    def test_do_lower_case_false_split_by_punct_false(self):
+        # fmt: off
+        sequence = " \tHeLLo!how  \n Are yoU?  "
+        tokens_target = ["▁", "<unk>", "e", "<unk>", "o", "!", "how", "▁", "<unk>", "re", "▁yo", "<unk>", "?"]
+        # fmt: on
+        extractor = SentencePieceExtractor(SAMPLE_VOCAB)
+        vocab, vocab_scores, merges = extractor.extract()
+        tokenizer = DebertaV2Tokenizer(
+            vocab=vocab_scores, merges=merges, unk_token="<unk>", do_lower_case=False, split_by_punct=False
+        )
+        tokens = tokenizer.convert_ids_to_tokens(tokenizer.encode(sequence, add_special_tokens=False))
+
+        self.assertListEqual(tokens, tokens_target)
+
+    def test_post_processor_adds_special_tokens(self):
+        extractor = SentencePieceExtractor(SAMPLE_VOCAB)
+        vocab, vocab_scores, merges = extractor.extract()
+        tokenizer = DebertaV2Tokenizer(vocab=vocab_scores, unk_token="<unk>")
+
+        encoding = tokenizer("Hello world")
+        tokens = tokenizer.convert_ids_to_tokens(encoding["input_ids"])
+        self.assertEqual(tokens[0], "[CLS]")
+        self.assertEqual(tokens[-1], "[SEP]")
+
+        encoding_pair = tokenizer("Hello", "World")
+        tokens_pair = tokenizer.convert_ids_to_tokens(encoding_pair["input_ids"])
+        self.assertEqual(tokens_pair[0], "[CLS]")
+        sep_indices = [i for i, t in enumerate(tokens_pair) if t == "[SEP]"]
+        self.assertEqual(len(sep_indices), 2)
+        self.assertEqual(sep_indices[-1], len(tokens_pair) - 1)