# 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. from __future__ import annotations import tempfile import unittest import numpy as np import paddle from parameterized import parameterized from paddlenlp.transformers import ( DistilBertForMaskedLM, DistilBertForQuestionAnswering, DistilBertForSequenceClassification, DistilBertForTokenClassification, DistilBertModel, ) from paddlenlp.transformers.distilbert.configuration import DistilBertConfig from ...testing_utils import require_package, slow from ..test_configuration_common import ConfigTester from ..test_modeling_common import ( ModelTesterMixin, ModelTesterPretrainedMixin, ids_tensor, random_attention_mask, ) class DistilBertModelTester: def __init__( self, parent: DistilBertModelTest, batch_size=13, seq_length=7, is_training=True, use_input_mask=True, use_labels=True, vocab_size=99, hidden_size=32, num_hidden_layers=5, 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, initializer_range=0.02, pad_token_id=0, pool_act="tanh", fuse=False, type_sequence_label_size=2, num_labels=3, num_choices=4, scope=None, dropout=0.56, return_dict=False, ): self.parent: DistilBertModelTest = 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_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.initializer_range = initializer_range self.pad_token_id = pad_token_id self.pool_act = pool_act self.fuse = fuse self.type_sequence_label_size = type_sequence_label_size self.num_labels = num_labels self.num_choices = num_choices self.scope = scope self.dropout = dropout self.return_dict = return_dict 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 = random_attention_mask([self.batch_size, self.seq_length]) 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, input_mask, sequence_labels, token_labels, choice_labels def get_config(self) -> DistilBertConfig: return DistilBertConfig( 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, pad_token_id=self.pad_token_id, pool_act=self.pool_act, fuse=self.fuse, num_labels=self.num_labels, num_choices=self.num_choices, ) def create_and_check_model( self, config: DistilBertConfig, input_ids, input_mask, sequence_labels, token_labels, choice_labels ): model = DistilBertModel(config) model.eval() result = model(input_ids, attention_mask=input_mask) self.parent.assertEqual(result.shape, [self.batch_size, self.seq_length, self.hidden_size]) result = model(input_ids) self.parent.assertEqual(result.shape, [self.batch_size, self.seq_length, self.hidden_size]) def create_and_check_for_masked_lm( self, config: DistilBertConfig, input_ids, input_mask, sequence_labels, token_labels, choice_labels, ): model = DistilBertForMaskedLM(config) model.eval() result = model(input_ids, attention_mask=input_mask) self.parent.assertEqual(result.shape, [self.batch_size, self.seq_length, self.vocab_size]) def create_and_check_for_question_answering( self, config, input_ids, input_mask, sequence_labels, token_labels, choice_labels, ): model = DistilBertForQuestionAnswering(config) model.eval() result = model( input_ids, attention_mask=input_mask, ) self.parent.assertEqual(result[0].shape, [self.batch_size, self.seq_length]) self.parent.assertEqual(result[1].shape, [self.batch_size, self.seq_length]) def create_and_check_for_sequence_classification( self, config: DistilBertConfig, input_ids, input_mask, sequence_labels, token_labels, choice_labels, ): model = DistilBertForSequenceClassification(config) model.eval() result = model(input_ids, attention_mask=input_mask) self.parent.assertEqual(result.shape, [self.batch_size, self.num_labels]) def create_and_check_for_token_classification( self, config, input_ids, input_mask, sequence_labels, token_labels, choice_labels, ): model = DistilBertForTokenClassification(config) model.eval() result = model(input_ids, attention_mask=input_mask) self.parent.assertEqual(result.shape, [self.batch_size, self.seq_length, self.num_labels]) def test_addition_params(self, config: DistilBertConfig, *args, **kwargs): config.num_labels = 7 config.classifier_dropout = 0.98 model = DistilBertForTokenClassification(config) model.eval() self.parent.assertEqual(model.classifier.weight.shape, [config.hidden_size, 7]) self.parent.assertEqual(model.dropout.p, 0.98) def prepare_config_and_inputs_for_common(self): config_and_inputs = self.prepare_config_and_inputs() ( config, input_ids, input_mask, sequence_labels, token_labels, choice_labels, ) = config_and_inputs inputs_dict = {"input_ids": input_ids, "attention_mask": input_mask} return config, inputs_dict class DistilBertModelTest(ModelTesterMixin, unittest.TestCase): base_model_class = DistilBertModel return_dict = False use_labels = False test_resize_embeddings = False all_model_classes = ( DistilBertModel, DistilBertForMaskedLM, DistilBertForQuestionAnswering, DistilBertForSequenceClassification, DistilBertForTokenClassification, ) def setUp(self): super().setUp() self.model_tester = DistilBertModelTester(self) self.config_tester = ConfigTester(self, config_class=DistilBertConfig, vocab_size=256, hidden_size=24) 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_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) def test_for_custom_params(self): config_and_inputs = self.model_tester.prepare_config_and_inputs() self.model_tester.test_addition_params(*config_and_inputs) def test_model_name_list(self): config = self.model_tester.get_config() model = self.base_model_class(config) self.assertTrue(len(model.model_name_list) != 0) @slow def test_params_compatibility_of_init_method(self): """test initing model with different params""" model: DistilBertForTokenClassification = DistilBertForTokenClassification.from_pretrained( "distilbert-base-uncased", num_classes=4, dropout=0.3 ) assert model.config.num_labels == 4 assert model.config.dropout == 0.3 class DistilBertModelCompatibilityTest(unittest.TestCase): model_id = "hf-internal-testing/tiny-random-DistilBertModel" @require_package("transformers", "torch") def test_distilBert_converter(self): with tempfile.TemporaryDirectory() as tempdir: # 1. create input input_ids = np.random.randint(100, 200, [1, 20]) # 2. forward the paddle model from paddlenlp.transformers import DistilBertModel paddle_model = DistilBertModel.from_pretrained(self.model_id, from_hf_hub=True, cache_dir=tempdir) paddle_model.eval() paddle_logit = paddle_model(paddle.to_tensor(input_ids))[0] # 3. forward the torch model import torch from transformers import DistilBertModel torch_model = DistilBertModel.from_pretrained(self.model_id, cache_dir=tempdir) torch_model.eval() torch_logit = torch_model(torch.tensor(input_ids))[0] # 4. compare results self.assertTrue( np.allclose( paddle_logit.detach().cpu().reshape([-1])[:9].numpy(), torch_logit.detach().cpu().reshape([-1])[:9].numpy(), rtol=1e-4, ) ) @require_package("transformers", "torch") def test_distilBert_converter_from_local_dir(self): with tempfile.TemporaryDirectory() as tempdir: # 1. create common input input_ids = np.random.randint(100, 200, [1, 20]) # 2. forward the torch model import torch from transformers import DistilBertModel torch_model = DistilBertModel.from_pretrained(self.model_id) torch_model.eval() torch_model.save_pretrained(tempdir) torch_logit = torch_model(torch.tensor(input_ids))[0] # 2. forward the paddle model from paddlenlp.transformers import DistilBertModel paddle_model = DistilBertModel.from_pretrained(tempdir, convert_from_torch=True) paddle_model.eval() paddle_logit = paddle_model(paddle.to_tensor(input_ids))[0] self.assertTrue( np.allclose( paddle_logit.detach().cpu().reshape([-1])[:9].numpy(), torch_logit.detach().cpu().reshape([-1])[:9].numpy(), rtol=1e-4, ) ) @parameterized.expand( [ ("DistilBertModel",), ("DistilBertForQuestionAnswering",), ("DistilBertForSequenceClassification",), ("DistilBertForTokenClassification",), ] ) @require_package("transformers", "torch") def test_distilBert_classes_from_local_dir(self, class_name, pytorch_class_name=None): pytorch_class_name = pytorch_class_name or class_name with tempfile.TemporaryDirectory() as tempdir: # 1. create common input input_ids = np.random.randint(100, 200, [1, 20]) # 2. forward the torch model import torch import transformers torch_model_class = getattr(transformers, pytorch_class_name) torch_model = torch_model_class.from_pretrained(self.model_id) torch_model.eval() torch_model.save_pretrained(tempdir) torch_logit = torch_model(torch.tensor(input_ids))[0] # 3. forward the paddle model from paddlenlp import transformers paddle_model_class = getattr(transformers, class_name) paddle_model = paddle_model_class.from_pretrained(tempdir, convert_from_torch=True) paddle_model.eval() paddle_logit = paddle_model(paddle.to_tensor(input_ids))[0] self.assertTrue( np.allclose( paddle_logit.detach().cpu().reshape([-1])[:9].numpy(), torch_logit.detach().cpu().reshape([-1])[:9].numpy(), atol=1e-3, ) ) class DistilBertModelIntegrationTest(ModelTesterPretrainedMixin, unittest.TestCase): base_model_class = DistilBertModel @slow def test_inference_no_attention(self): model = DistilBertModel.from_pretrained("__internal_testing__/tiny-random-distilbert") 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) expected_shape = [1, 11, 8] self.assertEqual(output.shape, expected_shape) expected_slice = paddle.to_tensor( [ [ [0.50366199, -1.33068442, -1.73558784], [1.72435653, 1.08600891, -0.28388503], [-0.19172087, -0.56781638, 0.51192915], ] ] ) self.assertTrue(paddle.allclose(output[:, 1:4, 1:4], expected_slice, atol=1e-4)) @slow def test_inference_with_attention(self): model = DistilBertModel.from_pretrained("__internal_testing__/tiny-random-distilbert") 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) expected_shape = [1, 11, 8] self.assertEqual(output.shape, expected_shape) expected_slice = paddle.to_tensor( [ [ [0.50366199, -1.33068442, -1.73558784], [1.72435653, 1.08600891, -0.28388503], [-0.19172087, -0.56781638, 0.51192915], ] ] ) self.assertTrue(paddle.allclose(output[:, 1:4, 1:4], expected_slice, atol=1e-4)) if __name__ == "__main__": unittest.main()