transformers/tests/models/convbert/test_modeling_tf_convbert.py

# coding=utf-8
# Copyright 2021 The HuggingFace Inc. 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 os
import tempfile
import unittest

from transformers import ConvBertConfig, is_tf_available
from transformers.testing_utils import require_tf, slow

from ...test_configuration_common import ConfigTester
from ...test_modeling_tf_common import TFModelTesterMixin, ids_tensor, random_attention_mask
from ...test_pipeline_mixin import PipelineTesterMixin


if is_tf_available():
    import tensorflow as tf

    from transformers import (
        TFConvBertForMaskedLM,
        TFConvBertForMultipleChoice,
        TFConvBertForQuestionAnswering,
        TFConvBertForSequenceClassification,
        TFConvBertForTokenClassification,
        TFConvBertModel,
    )


class TFConvBertModelTester:
    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 = 13
        self.seq_length = 7
        self.is_training = True
        self.use_input_mask = True
        self.use_token_type_ids = True
        self.use_labels = True
        self.vocab_size = 99
        self.hidden_size = 384
        self.num_hidden_layers = 2
        self.num_attention_heads = 4
        self.intermediate_size = 37
        self.hidden_act = "gelu"
        self.hidden_dropout_prob = 0.1
        self.attention_probs_dropout_prob = 0.1
        self.max_position_embeddings = 512
        self.type_vocab_size = 16
        self.type_sequence_label_size = 2
        self.initializer_range = 0.02
        self.num_labels = 3
        self.num_choices = 4
        self.embedding_size = 128
        self.head_ratio = 2
        self.conv_kernel_size = 9
        self.num_groups = 1
        self.scope = None

    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)

        config = ConvBertConfig(
            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,
            return_dict=True,
        )

        return config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels

    def create_and_check_model(
        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
    ):
        model = TFConvBertModel(config=config)
        inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}

        inputs = [input_ids, input_mask]
        result = model(inputs)

        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_for_masked_lm(
        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
    ):
        model = TFConvBertForMaskedLM(config=config)
        inputs = {
            "input_ids": input_ids,
            "attention_mask": input_mask,
            "token_type_ids": token_type_ids,
        }
        result = model(inputs)
        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size))

    def create_and_check_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 = TFConvBertForSequenceClassification(config=config)
        inputs = {
            "input_ids": input_ids,
            "attention_mask": input_mask,
            "token_type_ids": token_type_ids,
        }

        result = model(inputs)
        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.num_labels))

    def create_and_check_for_multiple_choice(
        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
    ):
        config.num_choices = self.num_choices
        model = TFConvBertForMultipleChoice(config=config)
        multiple_choice_inputs_ids = tf.tile(tf.expand_dims(input_ids, 1), (1, self.num_choices, 1))
        multiple_choice_input_mask = tf.tile(tf.expand_dims(input_mask, 1), (1, self.num_choices, 1))
        multiple_choice_token_type_ids = tf.tile(tf.expand_dims(token_type_ids, 1), (1, self.num_choices, 1))
        inputs = {
            "input_ids": multiple_choice_inputs_ids,
            "attention_mask": multiple_choice_input_mask,
            "token_type_ids": multiple_choice_token_type_ids,
        }
        result = model(inputs)
        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.num_choices))

    def create_and_check_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 = TFConvBertForTokenClassification(config=config)
        inputs = {
            "input_ids": input_ids,
            "attention_mask": input_mask,
            "token_type_ids": token_type_ids,
        }
        result = model(inputs)
        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.num_labels))

    def create_and_check_for_question_answering(
        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
    ):
        model = TFConvBertForQuestionAnswering(config=config)
        inputs = {
            "input_ids": input_ids,
            "attention_mask": input_mask,
            "token_type_ids": token_type_ids,
        }

        result = model(inputs)
        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 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_tf
class TFConvBertModelTest(TFModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
    all_model_classes = (
        (
            TFConvBertModel,
            TFConvBertForMaskedLM,
            TFConvBertForQuestionAnswering,
            TFConvBertForSequenceClassification,
            TFConvBertForTokenClassification,
            TFConvBertForMultipleChoice,
        )
        if is_tf_available()
        else ()
    )
    pipeline_model_mapping = (
        {
            "feature-extraction": TFConvBertModel,
            "fill-mask": TFConvBertForMaskedLM,
            "question-answering": TFConvBertForQuestionAnswering,
            "text-classification": TFConvBertForSequenceClassification,
            "token-classification": TFConvBertForTokenClassification,
            "zero-shot": TFConvBertForSequenceClassification,
        }
        if is_tf_available()
        else {}
    )
    test_pruning = False
    test_head_masking = False
    test_onnx = False

    def setUp(self):
        self.model_tester = TFConvBertModelTester(self)
        self.config_tester = ConfigTester(self, config_class=ConvBertConfig, hidden_size=37)

    def test_config(self):
        self.config_tester.run_common_tests()

    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_saved_model_creation_extended(self):
        config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
        config.output_hidden_states = True
        config.output_attentions = True

        if hasattr(config, "use_cache"):
            config.use_cache = True

        encoder_seq_length = getattr(self.model_tester, "encoder_seq_length", self.model_tester.seq_length)
        encoder_key_length = getattr(self.model_tester, "key_length", encoder_seq_length)

        for model_class in self.all_model_classes:
            class_inputs_dict = self._prepare_for_class(inputs_dict, model_class)
            model = model_class(config)
            num_out = len(model(class_inputs_dict))

            with tempfile.TemporaryDirectory() as tmpdirname:
                model.save_pretrained(tmpdirname, saved_model=True)
                saved_model_dir = os.path.join(tmpdirname, "saved_model", "1")
                model = tf.keras.models.load_model(saved_model_dir)
                outputs = model(class_inputs_dict)

                if self.is_encoder_decoder:
                    output_hidden_states = outputs["encoder_hidden_states"]
                    output_attentions = outputs["encoder_attentions"]
                else:
                    output_hidden_states = outputs["hidden_states"]
                    output_attentions = outputs["attentions"]

                self.assertEqual(len(outputs), num_out)

                expected_num_layers = getattr(
                    self.model_tester, "expected_num_hidden_layers", self.model_tester.num_hidden_layers + 1
                )

                self.assertEqual(len(output_hidden_states), expected_num_layers)
                self.assertListEqual(
                    list(output_hidden_states[0].shape[-2:]),
                    [self.model_tester.seq_length, self.model_tester.hidden_size],
                )

                self.assertEqual(len(output_attentions), self.model_tester.num_hidden_layers)
                self.assertListEqual(
                    list(output_attentions[0].shape[-3:]),
                    [self.model_tester.num_attention_heads / 2, encoder_seq_length, encoder_key_length],
                )

    @slow
    def test_model_from_pretrained(self):
        model = TFConvBertModel.from_pretrained("YituTech/conv-bert-base")
        self.assertIsNotNone(model)

    def test_attention_outputs(self):
        config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
        config.return_dict = True
        decoder_seq_length = getattr(self.model_tester, "decoder_seq_length", self.model_tester.seq_length)
        encoder_seq_length = getattr(self.model_tester, "encoder_seq_length", self.model_tester.seq_length)
        decoder_key_length = getattr(self.model_tester, "key_length", decoder_seq_length)
        encoder_key_length = getattr(self.model_tester, "key_length", encoder_seq_length)

        def check_decoder_attentions_output(outputs):
            out_len = len(outputs)
            self.assertEqual(out_len % 2, 0)
            decoder_attentions = outputs.decoder_attentions
            self.assertEqual(len(decoder_attentions), self.model_tester.num_hidden_layers)
            self.assertListEqual(
                list(decoder_attentions[0].shape[-3:]),
                [self.model_tester.num_attention_heads / 2, decoder_seq_length, decoder_key_length],
            )

        def check_encoder_attentions_output(outputs):
            attentions = [
                t.numpy() for t in (outputs.encoder_attentions if config.is_encoder_decoder else outputs.attentions)
            ]
            self.assertEqual(len(attentions), self.model_tester.num_hidden_layers)
            self.assertListEqual(
                list(attentions[0].shape[-3:]),
                [self.model_tester.num_attention_heads / 2, encoder_seq_length, encoder_key_length],
            )

        for model_class in self.all_model_classes:
            inputs_dict["output_attentions"] = True
            config.output_hidden_states = False
            model = model_class(config)
            outputs = model(self._prepare_for_class(inputs_dict, model_class))
            out_len = len(outputs)
            self.assertEqual(config.output_hidden_states, False)
            check_encoder_attentions_output(outputs)

            if self.is_encoder_decoder:
                model = model_class(config)
                outputs = model(self._prepare_for_class(inputs_dict, model_class))
                self.assertEqual(config.output_hidden_states, False)
                check_decoder_attentions_output(outputs)

            # Check that output attentions can also be changed via the config
            del inputs_dict["output_attentions"]
            config.output_attentions = True
            model = model_class(config)
            outputs = model(self._prepare_for_class(inputs_dict, model_class))
            self.assertEqual(config.output_hidden_states, False)
            check_encoder_attentions_output(outputs)

            # Check attention is always last and order is fine
            inputs_dict["output_attentions"] = True
            config.output_hidden_states = True
            model = model_class(config)
            outputs = model(self._prepare_for_class(inputs_dict, model_class))

            self.assertEqual(out_len + (2 if self.is_encoder_decoder else 1), len(outputs))
            self.assertEqual(model.config.output_hidden_states, True)
            check_encoder_attentions_output(outputs)


@require_tf
class TFConvBertModelIntegrationTest(unittest.TestCase):
    @slow
    def test_inference_masked_lm(self):
        model = TFConvBertModel.from_pretrained("YituTech/conv-bert-base")
        input_ids = tf.constant([[0, 1, 2, 3, 4, 5]])
        output = model(input_ids)[0]

        expected_shape = [1, 6, 768]
        self.assertEqual(output.shape, expected_shape)

        expected_slice = tf.constant(
            [
                [
                    [-0.03475493, -0.4686034, -0.30638832],
                    [0.22637248, -0.26988646, -0.7423424],
                    [0.10324868, -0.45013508, -0.58280784],
                ]
            ]
        )
        tf.debugging.assert_near(output[:, :3, :3], expected_slice, atol=1e-4)