first commit

tests/models/openai/test_modeling_openai.py

@@ -0,0 +1,328 @@
+# 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 is_torch_available
+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, ids_tensor
+from ...test_pipeline_mixin import PipelineTesterMixin
+
+
+if is_torch_available():
+    import torch
+
+    from transformers import (
+        OpenAIGPTConfig,
+        OpenAIGPTDoubleHeadsModel,
+        OpenAIGPTForSequenceClassification,
+        OpenAIGPTLMHeadModel,
+        OpenAIGPTModel,
+    )
+
+
+class OpenAIGPTModelTester:
+    def __init__(
+        self,
+        parent,
+        batch_size=13,
+        seq_length=7,
+        is_training=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_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
+        self.pad_token_id = self.vocab_size - 1
+
+    def prepare_config_and_inputs(self):
+        input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size)
+
+        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 = OpenAIGPTConfig(
+            vocab_size=self.vocab_size,
+            n_embd=self.hidden_size,
+            n_layer=self.num_hidden_layers,
+            n_head=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,
+            n_positions=self.max_position_embeddings,
+            # type_vocab_size=self.type_vocab_size,
+            # initializer_range=self.initializer_range
+            pad_token_id=self.pad_token_id,
+        )
+
+        return (
+            config,
+            input_ids,
+            token_type_ids,
+            sequence_labels,
+            token_labels,
+            choice_labels,
+        )
+
+    def create_and_check_openai_gpt_model(self, config, input_ids, token_type_ids, *args):
+        model = OpenAIGPTModel(config=config)
+        model.to(torch_device)
+        model.eval()
+
+        result = model(input_ids, 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_lm_head_model(self, config, input_ids, token_type_ids, *args):
+        model = OpenAIGPTLMHeadModel(config)
+        model.to(torch_device)
+        model.eval()
+
+        result = model(input_ids, token_type_ids=token_type_ids, labels=input_ids)
+        self.parent.assertEqual(result.loss.shape, ())
+        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size))
+
+    def create_and_check_double_lm_head_model(self, config, input_ids, token_type_ids, *args):
+        model = OpenAIGPTDoubleHeadsModel(config)
+        model.to(torch_device)
+        model.eval()
+
+        result = model(input_ids, token_type_ids=token_type_ids, labels=input_ids)
+        self.parent.assertEqual(result.loss.shape, ())
+        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size))
+
+    def create_and_check_openai_gpt_for_sequence_classification(self, config, input_ids, token_type_ids, *args):
+        config.num_labels = self.num_labels
+        model = OpenAIGPTForSequenceClassification(config)
+        model.to(torch_device)
+        model.eval()
+
+        sequence_labels = ids_tensor([self.batch_size], self.type_sequence_label_size)
+        result = model(input_ids, token_type_ids=token_type_ids, labels=sequence_labels)
+        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.num_labels))
+
+    def prepare_config_and_inputs_for_common(self):
+        config_and_inputs = self.prepare_config_and_inputs()
+        (
+            config,
+            input_ids,
+            token_type_ids,
+            sequence_labels,
+            token_labels,
+            choice_labels,
+        ) = config_and_inputs
+        inputs_dict = {
+            "input_ids": input_ids,
+            "token_type_ids": token_type_ids,
+        }
+
+        return config, inputs_dict
+
+
+@require_torch
+class OpenAIGPTModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixin, unittest.TestCase):
+    all_model_classes = (
+        (OpenAIGPTModel, OpenAIGPTLMHeadModel, OpenAIGPTDoubleHeadsModel, OpenAIGPTForSequenceClassification)
+        if is_torch_available()
+        else ()
+    )
+    pipeline_model_mapping = (
+        {
+            "feature-extraction": OpenAIGPTModel,
+            "text-classification": OpenAIGPTForSequenceClassification,
+            "text-generation": OpenAIGPTLMHeadModel,
+            "zero-shot": OpenAIGPTForSequenceClassification,
+        }
+        if is_torch_available()
+        else {}
+    )
+
+    # TODO: Fix the failed tests
+    def is_pipeline_test_to_skip(
+        self,
+        pipeline_test_case_name,
+        config_class,
+        model_architecture,
+        tokenizer_name,
+        image_processor_name,
+        feature_extractor_name,
+        processor_name,
+    ):
+        if pipeline_test_case_name == "ZeroShotClassificationPipelineTests":
+            # Get `tokenizer does not have a padding token` error for both fast/slow tokenizers.
+            # `OpenAIGPTConfig` was never used in pipeline tests, either because of a missing checkpoint or because a
+            # tiny config could not be created.
+            return True
+
+        return False
+
+    def _greedy_generate(self, *args, use_cache=False, **kwargs):
+        """Same as the general one, with `use_cache=False` explicitly as openai cannot use cache at all."""
+        return super()._greedy_generate(*args, use_cache=use_cache, **kwargs)
+
+    def _sample_generate(self, *args, use_cache=False, **kwargs):
+        """Same as the general one, with `use_cache=False` explicitly as openai cannot use cache at all."""
+        return super()._sample_generate(*args, use_cache=use_cache, **kwargs)
+
+    def _beam_search_generate(self, *args, use_cache=False, **kwargs):
+        """Same as the general one, with `use_cache=False` explicitly as openai cannot use cache at all."""
+        return super()._beam_search_generate(*args, use_cache=use_cache, **kwargs)
+
+    def _beam_sample_generate(self, *args, use_cache=False, **kwargs):
+        """Same as the general one, with `use_cache=False` explicitly as openai cannot use cache at all."""
+        return super()._beam_sample_generate(*args, use_cache=use_cache, **kwargs)
+
+    # special case for DoubleHeads 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.__name__ == "OpenAIGPTDoubleHeadsModel":
+                inputs_dict["labels"] = torch.zeros(
+                    (self.model_tester.batch_size, self.model_tester.num_choices, self.model_tester.seq_length),
+                    dtype=torch.long,
+                    device=torch_device,
+                )
+                inputs_dict["input_ids"] = inputs_dict["labels"]
+                inputs_dict["token_type_ids"] = inputs_dict["labels"]
+                inputs_dict["mc_token_ids"] = torch.zeros(
+                    (self.model_tester.batch_size, self.model_tester.num_choices),
+                    dtype=torch.long,
+                    device=torch_device,
+                )
+                inputs_dict["mc_labels"] = torch.zeros(
+                    self.model_tester.batch_size, dtype=torch.long, device=torch_device
+                )
+        return inputs_dict
+
+    def setUp(self):
+        self.model_tester = OpenAIGPTModelTester(self)
+        self.config_tester = ConfigTester(self, config_class=OpenAIGPTConfig, n_embd=37)
+
+    def test_config(self):
+        self.config_tester.run_common_tests()
+
+    def test_openai_gpt_model(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_openai_gpt_model(*config_and_inputs)
+
+    def test_openai_gpt_lm_head_model(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_lm_head_model(*config_and_inputs)
+
+    def test_openai_gpt_double_lm_head_model(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_double_lm_head_model(*config_and_inputs)
+
+    def test_openai_gpt_classification_model(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_openai_gpt_for_sequence_classification(*config_and_inputs)
+
+    @slow
+    def test_model_from_pretrained(self):
+        model_name = "openai-community/openai-gpt"
+        model = OpenAIGPTModel.from_pretrained(model_name)
+        self.assertIsNotNone(model)
+
+    @unittest.skip("Openai cannot use a cache correctly and this test sets it to True explicitly")
+    def test_generate_methods_with_logits_to_keep(self):
+        pass
+
+    @unittest.skip("Openai cannot use a cache correctly and this test sets it to True explicitly")
+    def test_generate_with_and_without_position_ids(self):
+        pass
+
+
+@require_torch
+class OPENAIGPTModelLanguageGenerationTest(unittest.TestCase):
+    @slow
+    def test_lm_generate_openai_gpt(self):
+        model = OpenAIGPTLMHeadModel.from_pretrained("openai-community/openai-gpt")
+        model.to(torch_device)
+        input_ids = torch.tensor([[481, 4735, 544]], dtype=torch.long, device=torch_device)  # the president is
+        expected_output_ids = [
+            481,
+            4735,
+            544,
+            246,
+            963,
+            870,
+            762,
+            239,
+            244,
+            40477,
+            244,
+            249,
+            719,
+            881,
+            487,
+            544,
+            240,
+            244,
+            603,
+            481,
+        ]  # the president is a very good man. " \n " i\'m sure he is, " said the
+
+        output_ids = model.generate(input_ids, do_sample=False, max_length=20)
+        self.assertListEqual(output_ids[0].tolist(), expected_output_ids)

tests/models/openai/test_tokenization_openai.py

@@ -0,0 +1,32 @@
+# Copyright 2018 The Google AI Language Team Authors.
+#
+# 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 OpenAIGPTTokenizer
+from transformers.testing_utils import require_tokenizers
+
+from ...test_tokenization_common import TokenizerTesterMixin
+
+
+@require_tokenizers
+class OpenAIGPTTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
+    from_pretrained_id = "openai-community/openai-gpt"
+
+    tokenizer_class = OpenAIGPTTokenizer
+    integration_expected_tokens = ['this</w>', 'is</w>', 'a</w>', 'test</w>', '<unk>', 'i</w>', 'was</w>', 'born</w>', 'in</w>', '9', '2000</w>', ',</w>', 'and</w>', 'this</w>', 'is</w>', 'false</w>', '.</w>', '<unk>', '<unk>', '<unk>', '<unk>', '<unk>', '<unk>', 'hi</w>', 'hello</w>', 'hi</w>', 'hello</w>', 'hello</w>', '<</w>', 's</w>', '></w>', 'hi</w>', '<</w>', 's</w>', '></w>', 'there</w>', 'the</w>', 'following</w>', 'string</w>', 'should</w>', 'be</w>', 'properly</w>', 'en', 'coded</w>', ':</w>', 'hello</w>', '.</w>', 'but</w>', 'ird</w>', 'and</w>', '<unk>', 'ird</w>', '<unk>', 'hey</w>', 'how</w>', 'are</w>', 'you</w>', 'doing</w>']  # fmt: skip
+    integration_expected_token_ids = [616, 544, 246, 2345, 0, 249, 509, 3105, 500, 53, 28654, 240, 488, 616, 544, 6843, 239, 0, 0, 0, 0, 0, 0, 3569, 3570, 3569, 3570, 3570, 295, 252, 290, 3569, 295, 252, 290, 655, 481, 2890, 6422, 994, 580, 6506, 496, 20925, 271, 3570, 239, 568, 13926, 488, 0, 13926, 0, 2229, 718, 640, 512, 1273]  # fmt: skip
+    expected_tokens_from_ids = ['this</w>', 'is</w>', 'a</w>', 'test</w>', '<unk>', 'i</w>', 'was</w>', 'born</w>', 'in</w>', '9', '2000</w>', ',</w>', 'and</w>', 'this</w>', 'is</w>', 'false</w>', '.</w>', '<unk>', '<unk>', '<unk>', '<unk>', '<unk>', '<unk>', 'hi</w>', 'hello</w>', 'hi</w>', 'hello</w>', 'hello</w>', '<</w>', 's</w>', '></w>', 'hi</w>', '<</w>', 's</w>', '></w>', 'there</w>', 'the</w>', 'following</w>', 'string</w>', 'should</w>', 'be</w>', 'properly</w>', 'en', 'coded</w>', ':</w>', 'hello</w>', '.</w>', 'but</w>', 'ird</w>', 'and</w>', '<unk>', 'ird</w>', '<unk>', 'hey</w>', 'how</w>', 'are</w>', 'you</w>', 'doing</w>']  # fmt: skip
+    integration_expected_decoded_text = "this is a test <unk>i was born in 92000 , and this is false . <unk><unk><unk><unk><unk><unk>hi hello hi hello hello < s > hi < s > there the following string should be properly encoded : hello . but ird and <unk>ird <unk>hey how are you doing"