All TF RoBERTa models pass tests

pytorch_transformers/modeling_tf_roberta.py

@@ -0,0 +1,160 @@
+# coding=utf-8
+# Copyright 2018 The Google AI Language Team Authors and The HuggingFace Inc. team.
+# Copyright (c) 2018, NVIDIA CORPORATION.  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.
+""" TF 2.0 BERT model. """
+
+from __future__ import absolute_import, division, print_function, unicode_literals
+
+import json
+import logging
+import math
+import os
+import sys
+from io import open
+
+import numpy as np
+import tensorflow as tf
+
+from pytorch_transformers import RobertaConfig, ROBERTA_PRETRAINED_MODEL_ARCHIVE_MAP, TFSharedEmbeddings
+from pytorch_transformers.modeling_tf_bert import TFBertEmbeddings, TFBertModel, gelu, TFBertPreTrainedModel
+from .configuration_bert import BertConfig
+from .modeling_tf_utils import TFPreTrainedModel
+from .file_utils import add_start_docstrings
+
+logger = logging.getLogger(__name__)
+
+TF_ROBERTA_PRETRAINED_MODEL_ARCHIVE_MAP = {
+    'roberta-base': "https://s3.amazonaws.com/models.huggingface.co/bert/roberta-base-pytorch_model.bin",
+    'roberta-large': "https://s3.amazonaws.com/models.huggingface.co/bert/roberta-large-pytorch_model.bin",
+    'roberta-large-mnli': "https://s3.amazonaws.com/models.huggingface.co/bert/roberta-large-mnli-pytorch_model.bin",
+}
+
+class TFRobertaModel(TFBertModel):
+    config_class = RobertaConfig
+    pretrained_model_archive_map = ROBERTA_PRETRAINED_MODEL_ARCHIVE_MAP
+    base_model_prefix = "roberta"
+
+    def __init__(self, config):
+        super(TFRobertaModel, self).__init__(config)
+
+        self.embeddings = TFSharedEmbeddings(config.vocab_size, config.hidden_size, name="embeddings")
+
+    def call(self, inputs, training=False):
+        if not isinstance(inputs, (dict, tuple, list)):
+            input_ids = inputs
+        else:
+            if isinstance(inputs, (tuple, list)):
+                input_ids = inputs[0]
+            else:
+                input_ids = inputs["input_ids"]
+
+        if tf.reduce_sum(input_ids[:, 0]) == 0:
+            logger.warning("A sequence with no special tokens has been passed to the RoBERTa model. "
+                           "This model requires special tokens in order to work. "
+                           "Please specify add_special_tokens=True in your encoding.")
+
+        return super(TFRobertaModel, self).call(inputs, training=training)
+
+
+class TFRobertaForMaskedLM(TFBertPreTrainedModel):
+    config_class = RobertaConfig
+    pretrained_model_archive_map = ROBERTA_PRETRAINED_MODEL_ARCHIVE_MAP
+    base_model_prefix = "roberta"
+
+    def __init__(self, config):
+        super(TFRobertaForMaskedLM, self).__init__(config)
+
+        self.roberta = TFRobertaModel(config)
+        self.lm_head = TFRobertaLMHead(config)
+
+    def call(self, inputs, training=False):
+
+        outputs = self.roberta(inputs, training=training)
+        sequence_output = outputs[0]
+        predictions = self.lm_head(sequence_output)
+        prediction_scores = self.roberta.embeddings(predictions, mode='linear')
+
+        outputs = (prediction_scores,) + outputs[2:]  # Add hidden states and attention if they are here
+
+        return outputs  # prediction_scores, (hidden_states), (attentions)
+
+
+class TFRobertaLMHead(tf.keras.layers.Layer):
+    """Roberta Head for masked language modeling."""
+
+    def __init__(self, config):
+        super(TFRobertaLMHead, self).__init__()
+        self.dense = tf.keras.layers.Dense(config.hidden_size)
+        self.layer_norm = tf.keras.layers.LayerNormalization(epsilon=config.layer_norm_eps, name='LayerNorm')
+
+    def call(self, features, **kwargs):
+        x = self.dense(features)
+        x = gelu(x)
+        x = self.layer_norm(x)
+        return x
+
+
+class TFRobertaForSequenceClassification(TFBertPreTrainedModel):
+    config_class = RobertaConfig
+    pretrained_model_archive_map = ROBERTA_PRETRAINED_MODEL_ARCHIVE_MAP
+    base_model_prefix = "roberta"
+
+    def __init__(self, config):
+        super(TFRobertaForSequenceClassification, self).__init__(config)
+        self.num_labels = config.num_labels
+
+        self.roberta = TFRobertaModel(config)
+        self.classifier = TFRobertaClassificationHead(config)
+
+
+    def call(self, inputs, training=False):
+        outputs = self.roberta(inputs, training=training)
+        sequence_output = outputs[0]
+        logits = self.classifier(sequence_output, training=training)
+        outputs = (logits,) + outputs[2:]
+        return outputs  # (loss), logits, (hidden_states), (attentions)
+
+
+class TFRobertaClassificationHead(tf.keras.layers.Layer):
+    """Head for sentence-level classification tasks."""
+
+    def __init__(self, config):
+        super(TFRobertaClassificationHead, self).__init__()
+        self.dense = tf.keras.layers.Dense(config.hidden_size)
+        self.dropout = tf.keras.layers.Dropout(config.hidden_dropout_prob)
+        self.out_proj = tf.keras.layers.Dense(config.num_labels)
+
+    def call(self, features, training,  **kwargs):
+        x = features[:, 0, :]  # take <s> token (equiv. to [CLS])
+        x = self.dropout(x, training=training)
+        x = self.dense(x)
+        x = tf.tanh(x)
+        x = self.dropout(x, training=training)
+        x = self.out_proj(x)
+        return x
+
+
+
+
+
+
+
+
+
+
+
+
+
+

pytorch_transformers/tests/modeling_tf_roberta_test.py

@@ -0,0 +1,228 @@
+# coding=utf-8
+# 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.
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import unittest
+import shutil
+import pytest
+import sys
+
+from .modeling_tf_common_test import (TFCommonTestCases, ids_tensor)
+from .configuration_common_test import ConfigTester
+
+from pytorch_transformers import RobertaConfig, is_tf_available
+
+try:
+    import tensorflow as tf
+    from pytorch_transformers.modeling_tf_roberta import (TFRobertaModel, TFRobertaForMaskedLM,
+                                                          TFRobertaForSequenceClassification,
+                                                          TF_ROBERTA_PRETRAINED_MODEL_ARCHIVE_MAP)
+except ImportError:
+    pytestmark = pytest.mark.skip("Require TensorFlow")
+
+
+class TFRobertaModelTest(TFCommonTestCases.TFCommonModelTester):
+
+    all_model_classes = (TFRobertaModel,) if is_tf_available() else ()
+
+    class TFRobertaModelTester(object):
+
+        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=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,
+                     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 = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
+
+            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 = RobertaConfig(
+                vocab_size_or_config_json_file=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 config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+
+        def create_and_check_roberta_model(self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels):
+            model = TFRobertaModel(config=config)
+            inputs = {'input_ids': input_ids,
+                      'attention_mask': input_mask,
+                      'token_type_ids': token_type_ids}
+            sequence_output, pooled_output = model(inputs)
+
+            inputs = [input_ids, input_mask]
+            sequence_output, pooled_output = model(inputs)
+
+            sequence_output, pooled_output = model(input_ids)
+
+            result = {
+                "sequence_output": sequence_output.numpy(),
+                "pooled_output": pooled_output.numpy(),
+            }
+            self.parent.assertListEqual(
+                list(result["sequence_output"].shape),
+                [self.batch_size, self.seq_length, self.hidden_size])
+            self.parent.assertListEqual(list(result["pooled_output"].shape), [self.batch_size, self.hidden_size])
+
+
+        def create_and_check_roberta_for_masked_lm(self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels):
+            model = TFRobertaForMaskedLM(config=config)
+            inputs = {'input_ids': input_ids,
+                      'attention_mask': input_mask,
+                      'token_type_ids': token_type_ids}
+            prediction_scores, = model(inputs)
+            result = {
+                "prediction_scores": prediction_scores.numpy(),
+            }
+            self.parent.assertListEqual(
+                list(result["prediction_scores"].shape),
+                [self.batch_size, self.seq_length, self.vocab_size])
+
+        def create_and_check_roberta_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 = TFRobertaForSequenceClassification(config=config)
+            inputs = {'input_ids': input_ids,
+                      'attention_mask': input_mask,
+                      'token_type_ids': token_type_ids}
+            logits, = model(inputs)
+            result = {
+                "logits": logits.numpy(),
+            }
+            self.parent.assertListEqual(
+                list(result["logits"].shape),
+                [self.batch_size, self.num_labels])
+
+        # def create_and_check_bert_for_question_answering(self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels):
+        #     model = TFBertForQuestionAnswering(config=config)
+        #     inputs = {'input_ids': input_ids,
+        #               'attention_mask': input_mask,
+        #               'token_type_ids': token_type_ids}
+        #     start_logits, end_logits = model(inputs)
+        #     result = {
+        #         "start_logits": start_logits.numpy(),
+        #         "end_logits": end_logits.numpy(),
+        #     }
+        #     self.parent.assertListEqual(
+        #         list(result["start_logits"].shape),
+        #         [self.batch_size, self.seq_length])
+        #     self.parent.assertListEqual(
+        #         list(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
+
+    def setUp(self):
+        self.model_tester = TFRobertaModelTest.TFRobertaModelTester(self)
+        self.config_tester = ConfigTester(self, config_class=RobertaConfig, hidden_size=37)
+
+    def test_config(self):
+        self.config_tester.run_common_tests()
+
+    def test_roberta_model(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_roberta_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_roberta_for_masked_lm(*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_roberta_for_sequence_classification(*config_and_inputs)
+
+    # @pytest.mark.slow
+    # def test_model_from_pretrained(self):
+    #     cache_dir = "/tmp/pytorch_transformers_test/"
+    #     for model_name in list(TF_ROBERTA_PRETRAINED_MODEL_ARCHIVE_MAP.keys())[:1]:
+    #         model = TFRobertaModel.from_pretrained(model_name, cache_dir=cache_dir)
+    #         shutil.rmtree(cache_dir)
+    #         self.assertIsNotNone(model)
+
+if __name__ == "__main__":
+    unittest.main()
+