/
githubmirror
/
transformers
Обзор
Документация
Войти
/
githubmirror
/
transformers
Код
Запросы
0
Пакеты
0
Релизы
0
Аналитика
Безопасность
main
tests/models/electra/test_modeling_electra.py
487 строк
17 KB
kaixuanliu
fix bug embedding_size mismatch with hidden_size in electra model test (#44657)
18 мар 2026, 13:12
Не верифицирован
18 мар 2026, 13:12
3bf9200
Код
Авторство
О чём код?
# 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 transformers import ElectraConfig, is_torch_available from transformers.models.auto import get_values from transformers.testing_utils import require_torch, slow, torch_device from ...generation.test_utils import GenerationTesterMixin from ...test_configuration_common import ConfigTester from ...test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor, random_attention_mask from ...test_pipeline_mixin import PipelineTesterMixin if is_torch_available(): import torch from transformers import ( MODEL_FOR_PRETRAINING_MAPPING, ElectraForCausalLM, ElectraForMaskedLM, ElectraForMultipleChoice, ElectraForPreTraining, ElectraForQuestionAnswering, ElectraForSequenceClassification, ElectraForTokenClassification, ElectraModel, ) class ElectraModelTester: 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, 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.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 = random_attention_mask([self.batch_size, self.seq_length]) 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) fake_token_labels = ids_tensor([self.batch_size, self.seq_length], 1) config = self.get_config() return ( config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels, fake_token_labels, ) def get_config(self): return ElectraConfig( vocab_size=self.vocab_size, embedding_size=self.hidden_size, # Match hidden_size to avoid dimension expansion amplifying BF16 rounding errors 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, is_decoder=False, initializer_range=self.initializer_range, ) def prepare_config_and_inputs_for_decoder(self): ( config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels, _, ) = self.prepare_config_and_inputs() config.is_decoder = True encoder_hidden_states = floats_tensor([self.batch_size, self.seq_length, self.hidden_size]) encoder_attention_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2) return ( config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels, encoder_hidden_states, encoder_attention_mask, ) def create_and_check_electra_model( self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels, fake_token_labels, ): model = ElectraModel(config=config) model.to(torch_device) 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.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size)) def create_and_check_electra_model_as_decoder( self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels, encoder_hidden_states, encoder_attention_mask, ): config.add_cross_attention = True model = ElectraModel(config) model.to(torch_device) model.eval() result = model( input_ids, attention_mask=input_mask, token_type_ids=token_type_ids, encoder_hidden_states=encoder_hidden_states, encoder_attention_mask=encoder_attention_mask, ) result = model( input_ids, attention_mask=input_mask, token_type_ids=token_type_ids, encoder_hidden_states=encoder_hidden_states, ) result = model(input_ids, attention_mask=input_mask, token_type_ids=token_type_ids) self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size)) def create_and_check_electra_for_masked_lm( self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels, fake_token_labels, ): model = ElectraForMaskedLM(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_electra_for_causal_lm( self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels, encoder_hidden_states, encoder_attention_mask, ): model = ElectraForCausalLM(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_electra_for_token_classification( self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels, fake_token_labels, ): config.num_labels = self.num_labels model = ElectraForTokenClassification(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_electra_for_pretraining( self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels, fake_token_labels, ): config.num_labels = self.num_labels model = ElectraForPreTraining(config=config) model.to(torch_device) model.eval() result = model(input_ids, attention_mask=input_mask, token_type_ids=token_type_ids, labels=fake_token_labels) self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length)) def create_and_check_electra_for_sequence_classification( self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels, fake_token_labels, ): config.num_labels = self.num_labels model = ElectraForSequenceClassification(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.assertEqual(result.logits.shape, (self.batch_size, self.num_labels)) def create_and_check_electra_for_question_answering( self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels, fake_token_labels, ): model = ElectraForQuestionAnswering(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_electra_for_multiple_choice( self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels, fake_token_labels, ): config.num_choices = self.num_choices model = ElectraForMultipleChoice(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, fake_token_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 ElectraModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixin, unittest.TestCase): all_model_classes = ( ( ElectraModel, ElectraForPreTraining, ElectraForMaskedLM, ElectraForCausalLM, ElectraForMultipleChoice, ElectraForTokenClassification, ElectraForSequenceClassification, ElectraForQuestionAnswering, ) if is_torch_available() else () ) pipeline_model_mapping = ( { "feature-extraction": ElectraModel, "fill-mask": ElectraForMaskedLM, "text-classification": ElectraForSequenceClassification, "text-generation": ElectraForCausalLM, "token-classification": ElectraForTokenClassification, "zero-shot": ElectraForSequenceClassification, } if is_torch_available() else {} ) # Overwriting to add `is_decoder` flag def prepare_config_and_inputs_for_generate(self, batch_size=2): config, inputs = super().prepare_config_and_inputs_for_generate(batch_size) config.is_decoder = True return config, inputs # special case for ForPreTraining model def _prepare_for_class(self, inputs_dict, model_class, return_labels=False): inputs_dict = super()._prepare_for_class(inputs_dict, model_class, return_labels=return_labels) if return_labels: if model_class in get_values(MODEL_FOR_PRETRAINING_MAPPING): inputs_dict["labels"] = torch.zeros( (self.model_tester.batch_size, self.model_tester.seq_length), dtype=torch.long, device=torch_device ) return inputs_dict def setUp(self): self.model_tester = ElectraModelTester(self) self.config_tester = ConfigTester(self, config_class=ElectraConfig, hidden_size=32) def test_config(self): self.config_tester.run_common_tests() def test_electra_model(self): config_and_inputs = self.model_tester.prepare_config_and_inputs() self.model_tester.create_and_check_electra_model(*config_and_inputs) def test_electra_model_as_decoder(self): config_and_inputs = self.model_tester.prepare_config_and_inputs_for_decoder() self.model_tester.create_and_check_electra_model_as_decoder(*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_electra_for_masked_lm(*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_electra_for_token_classification(*config_and_inputs) def test_for_pre_training(self): config_and_inputs = self.model_tester.prepare_config_and_inputs() self.model_tester.create_and_check_electra_for_pretraining(*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_electra_for_sequence_classification(*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_electra_for_question_answering(*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_electra_for_multiple_choice(*config_and_inputs) @slow def test_model_from_pretrained(self): model_name = "google/electra-small-generator" model = ElectraModel.from_pretrained(model_name) self.assertIsNotNone(model) def test_for_causal_lm(self): config_and_inputs = self.model_tester.prepare_config_and_inputs_for_decoder() self.model_tester.create_and_check_electra_for_causal_lm(*config_and_inputs) @require_torch class ElectraModelIntegrationTest(unittest.TestCase): @slow def test_inference_no_head_absolute_embedding(self): model = ElectraModel.from_pretrained("google/electra-small-discriminator") input_ids = torch.tensor([[0, 345, 232, 328, 740, 140, 1695, 69, 6078, 1588, 2]]) attention_mask = torch.tensor([[0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]]) output = model(input_ids, attention_mask=attention_mask)[0] expected_shape = torch.Size((1, 11, 256)) self.assertEqual(output.shape, expected_shape) expected_slice = torch.tensor( [[[0.4471, 0.6821, -0.3265], [0.4627, 0.5255, -0.3668], [0.4532, 0.3313, -0.4344]]] ) torch.testing.assert_close(output[:, 1:4, 1:4], expected_slice, rtol=1e-4, atol=1e-4)