[Unittest]Add roformer unittest (#2991)

* add roformer unittest

* add roformer unittest

* update test_modeling

* use relative import

* reduce model config to accelerate testing

* remove input_embedding from pretrained model

* revert slow tag

* update local branch

* update get_vocab method

* update get_vocab method

* update test_chinese method

* change absolute import

* update unittest

* update chinese test case

Co-authored-by: Guo Sheng <[email protected]>

Author: wj-Mcat

paddlenlp/transformers/roformer/modeling.py

@@ -671,6 +671,12 @@ def forward(
         else:
             return sequence_output, pooled_output
 
+    def get_input_embeddings(self) -> nn.Embedding:
+        return self.embeddings.word_embeddings
+
+    def set_input_embeddings(self, embedding: nn.Embedding):
+        self.embeddings.word_embeddings = embedding
+
 
 class RoFormerForQuestionAnswering(RoFormerPretrainedModel):
     """

paddlenlp/transformers/roformer/tokenizer.py

@@ -35,10 +35,18 @@ class JiebaBasicTokenizer(BasicTokenizer):
             Defaults to `True`.
     """
 
-    def __init__(self, vocab, do_lower_case=True):
+    def __init__(self,
+                 vocab,
+                 do_lower_case=True,
+                 never_split=None,
+                 tokenize_chinese_chars=True,
+                 strip_accents=None):
         """Constructs a JiebaBasicTokenizer."""
+        super().__init__(never_split=never_split,
+                         do_lower_case=do_lower_case,
+                         tokenize_chinese_chars=tokenize_chinese_chars,
+                         strip_accents=strip_accents)
         self.vocab = vocab
-        self.do_lower_case = do_lower_case
 
     def _tokenize_chinese_chars(self, text):
         output = []
@@ -140,6 +148,18 @@ class RoFormerTokenizer(PretrainedTokenizer):
             "https://bj.bcebos.com/paddlenlp/models/transformers/roformer/roformer-english-small-generator/vocab.txt",
         }
     }
+    max_model_input_sizes = {
+        "roformer-chinese-small": 512,
+        "roformer-chinese-base": 1536,
+        "roformer-chinese-char-small": 512,
+        "roformer-chinese-char-base": 512,
+        "roformer-chinese-sim-char-ft-small": 512,
+        "roformer-chinese-sim-char-ft-base": 512,
+        "roformer-chinese-sim-char-small": 512,
+        "roformer-chinese-sim-char-base": 512,
+        "roformer-english-small-discriminator": 128,
+        "roformer-english-small-generator": 128,
+    }
     pretrained_init_configuration = {
         "roformer-chinese-small": {
             "do_lower_case": True,
@@ -397,3 +417,6 @@ def get_special_tokens_mask(self,
             return [1] + ([0] * len(token_ids_0)) + [1] + (
                 [0] * len(token_ids_1)) + [1]
         return [1] + ([0] * len(token_ids_0)) + [1]
+
+    def get_vocab(self):
+        return dict(self.vocab.token_to_idx, **self.added_tokens_encoder)

tests/transformers/roformer/__init__.py

@@ -0,0 +1,336 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+# Copyright 2020 The HuggingFace Team. All rights reserved.
+#
+# 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 typing import Optional, Tuple
+from dataclasses import dataclass, fields, Field
+
+import paddle
+
+from paddlenlp.transformers import (
+    RoFormerModel, RoFormerPretrainedModel, RoFormerForPretraining,
+    RoFormerForSequenceClassification, RoFormerForTokenClassification,
+    RoFormerForQuestionAnswering, RoFormerForMultipleChoice,
+    RoFormerForMaskedLM)
+
+from ..test_modeling_common import ids_tensor, floats_tensor, random_attention_mask, ModelTesterMixin
+from ...testing_utils import slow
+
+
+@dataclass
+class RoFormerModelTestModelConfig:
+    """RoFormerModel model config which keep consist with pretrained_init_configuration sub fields
+    """
+    vocab_size: int = 200
+    embedding_size: int = 50
+    hidden_size: int = 36
+    num_hidden_layers: int = 6
+    num_attention_heads: int = 6
+    intermediate_size: int = 16
+    hidden_act: str = "gelu"
+    hidden_dropout_prob: float = 0.1
+    attention_probs_dropout_prob: float = 0.1
+    max_position_embeddings: int = 20
+    type_vocab_size: int = 2
+    initializer_range: float = 0.02
+    pad_token_id: int = 0
+    rotary_value: bool = False
+
+    @property
+    def model_kwargs(self) -> dict:
+        """get the model kwargs configuration to init the model"""
+        model_config_fields: Tuple[Field,
+                                   ...] = fields(RoFormerModelTestModelConfig)
+        return {
+            field.name: getattr(self, field.name)
+            for field in model_config_fields
+        }
+
+
+@dataclass
+class RoFormerModelTestConfig(RoFormerModelTestModelConfig):
+    """train config under unittest code"""
+    batch_size: int = 2
+    seq_length: int = 7
+    is_training: bool = False
+    use_input_mask: bool = False
+    use_token_type_ids: bool = True
+
+    # used for sequence classification
+    num_classes: int = 3
+    num_choices: int = 3
+
+
+class RoFormerModelTester:
+
+    def __init__(
+        self,
+        parent,
+        config: Optional[RoFormerModelTestConfig] = None,
+    ):
+        self.parent = parent
+        self.config: RoFormerModelTestConfig = config or RoFormerModelTestConfig(
+        )
+
+        self.is_training = self.config.is_training
+        self.num_classes = self.config.num_classes
+        self.num_choices = self.config.num_choices
+
+    def prepare_config_and_inputs(self):
+        config = self.config
+        input_ids = ids_tensor([config.batch_size, config.seq_length],
+                               config.vocab_size)
+
+        input_mask = None
+        if self.config.use_input_mask:
+            input_mask = random_attention_mask(
+                [config.batch_size, config.seq_length])
+
+        token_type_ids = None
+        if self.config.use_token_type_ids:
+            token_type_ids = ids_tensor([config.batch_size, config.seq_length],
+                                        config.type_vocab_size)
+
+        config = self.get_config()
+        return config, input_ids, token_type_ids, input_mask
+
+    def get_config(self) -> dict:
+        return self.config.model_kwargs
+
+    def create_and_check_model(
+        self,
+        config,
+        input_ids,
+        token_type_ids,
+        input_mask,
+    ):
+        model = RoFormerModel(**config)
+        model.eval()
+        result = model(input_ids,
+                       attention_mask=input_mask,
+                       token_type_ids=token_type_ids)
+        result = model(input_ids, token_type_ids=token_type_ids)
+        result = model(input_ids)
+        self.parent.assertEqual(result[0].shape, [
+            self.config.batch_size, self.config.seq_length,
+            self.config.hidden_size
+        ])
+        self.parent.assertEqual(
+            result[1].shape, [self.config.batch_size, self.config.hidden_size])
+
+    def create_and_check_for_multiple_choice(
+        self,
+        config,
+        input_ids,
+        token_type_ids,
+        input_mask,
+    ):
+        model = RoFormerForMultipleChoice(RoFormerModel(**config),
+                                          num_choices=self.config.num_choices)
+        model.eval()
+        multiple_choice_inputs_ids = input_ids.unsqueeze(1).expand(
+            [-1, self.config.num_choices, -1])
+
+        if token_type_ids is not None:
+            token_type_ids = token_type_ids.unsqueeze(1).expand(
+                [-1, self.config.num_choices, -1])
+
+        if input_mask is not None:
+            input_mask = input_mask.unsqueeze(1).expand(
+                [-1, self.config.num_choices, -1])
+
+        result = model(
+            multiple_choice_inputs_ids,
+            attention_mask=input_mask,
+            token_type_ids=token_type_ids,
+        )
+        self.parent.assertEqual(
+            result.shape, [self.config.batch_size, self.config.num_choices])
+
+    def create_and_check_for_question_answering(self, config, input_ids,
+                                                token_type_ids, input_mask):
+        model = RoFormerForQuestionAnswering(RoFormerModel(**config))
+        model.eval()
+        result = model(
+            input_ids,
+            attention_mask=input_mask,
+            token_type_ids=token_type_ids,
+        )
+        self.parent.assertEqual(
+            result[0].shape, [self.config.batch_size, self.config.seq_length])
+        self.parent.assertEqual(
+            result[1].shape, [self.config.batch_size, self.config.seq_length])
+
+    def create_and_check_for_token_classification(
+        self,
+        config,
+        input_ids,
+        token_type_ids,
+        input_mask,
+    ):
+        model = RoFormerForTokenClassification(RoFormerModel(**config),
+                                               num_classes=self.num_classes)
+        model.eval()
+        result = model(input_ids,
+                       attention_mask=input_mask,
+                       token_type_ids=token_type_ids)
+        self.parent.assertEqual(result.shape, [
+            self.config.batch_size, self.config.seq_length,
+            self.config.num_classes
+        ])
+
+    def create_and_check_for_masked_lm(
+        self,
+        config,
+        input_ids,
+        token_type_ids,
+        input_mask,
+    ):
+        model = RoFormerForMaskedLM(RoFormerModel(**config))
+        model.eval()
+        result = model(input_ids,
+                       attention_mask=input_mask,
+                       token_type_ids=token_type_ids)
+        self.parent.assertEqual(result.shape, [
+            self.config.batch_size, self.config.seq_length,
+            self.config.vocab_size
+        ])
+
+    def create_and_check_for_sequence_classification(
+        self,
+        config,
+        input_ids,
+        token_type_ids,
+        input_mask,
+    ):
+        model = RoFormerForSequenceClassification(
+            RoFormerModel(**config), num_classes=self.config.num_classes)
+        model.eval()
+        result = model(
+            input_ids,
+            attention_mask=input_mask,
+            token_type_ids=token_type_ids,
+        )
+        self.parent.assertEqual(
+            result.shape, [self.config.batch_size, self.config.num_classes])
+
+    def prepare_config_and_inputs_for_common(self):
+        config_and_inputs = self.prepare_config_and_inputs()
+        (
+            config,
+            input_ids,
+            token_type_ids,
+            input_mask,
+        ) = config_and_inputs
+        inputs_dict = {
+            "input_ids": input_ids,
+            "token_type_ids": token_type_ids,
+            "attention_mask": input_mask
+        }
+        return config, inputs_dict
+
+
+class RoFormerModelTest(ModelTesterMixin, unittest.TestCase):
+    base_model_class = RoFormerModel
+
+    all_model_classes = (
+        RoFormerModel,
+        RoFormerForMultipleChoice,
+        RoFormerForPretraining,
+        RoFormerForSequenceClassification,
+    )
+
+    def setUp(self):
+        self.model_tester = RoFormerModelTester(self)
+
+    def test_model(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_model(*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_for_masked_lm(*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_for_multiple_choice(
+            *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_for_question_answering(
+            *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_for_sequence_classification(
+            *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_for_token_classification(
+            *config_and_inputs)
+
+    @slow
+    def test_model_from_pretrained(self):
+        for model_name in list(
+                RoFormerPretrainedModel.pretrained_init_configuration)[:1]:
+            model = RoFormerModel.from_pretrained(model_name)
+            self.assertIsNotNone(model)
+
+
+class RoFormerModelIntegrationTest(unittest.TestCase):
+
+    @slow
+    def test_inference_no_attention(self):
+        model = RoFormerModel.from_pretrained("roformer-chinese-small")
+        model.eval()
+        input_ids = paddle.to_tensor(
+            [[0, 345, 232, 328, 740, 140, 1695, 69, 6078, 1588, 2]])
+        with paddle.no_grad():
+            output = model(input_ids)[0]
+        expected_shape = [1, 11, 384]
+        self.assertEqual(output.shape, expected_shape)
+
+        expected_slice = paddle.to_tensor(
+            [[[0.17788891, -2.17795515, 0.28824317],
+              [-1.70342600, -2.84062195, -0.53377795],
+              [-0.16374627, -0.67967212, -0.37192002]]])
+
+        self.assertTrue(
+            paddle.allclose(output[:, 1:4, 1:4], expected_slice, atol=1e-4))
+
+    @slow
+    def test_inference_with_attention(self):
+        model = RoFormerModel.from_pretrained("roformer-chinese-small")
+        model.eval()
+        input_ids = paddle.to_tensor(
+            [[0, 345, 232, 328, 740, 140, 1695, 69, 6078, 1588, 2]])
+        attention_mask = paddle.to_tensor([[0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]])
+        with paddle.no_grad():
+            output = model(input_ids, attention_mask=attention_mask)[0]
+        expected_shape = [1, 11, 384]
+        self.assertEqual(output.shape, expected_shape)
+
+        expected_slice = paddle.to_tensor(
+            [[[0.17788891, -2.17795515, 0.28824317],
+              [-1.70342600, -2.84062195, -0.53377795],
+              [-0.16374627, -0.67967212, -0.37192002]]])
+        self.assertTrue(
+            paddle.allclose(output[:, 1:4, 1:4], expected_slice, atol=1e-4))
+
+
+if __name__ == "__main__":
+    unittest.main()