Merge branch 'huggingface/master'

transformers/tests/modeling_common_test.py

@@ -17,8 +17,10 @@ from __future__ import division
 from __future__ import print_function
 
 import copy
+import sys
 import os
 import shutil
+import tempfile
 import json
 import random
 import uuid
@@ -31,6 +33,7 @@ from transformers import is_torch_available
 
 if is_torch_available():
     import torch
+    import numpy as np
 
     from transformers import (PretrainedConfig, PreTrainedModel,
                                     BertModel, BertConfig, BERT_PRETRAINED_MODEL_ARCHIVE_MAP,
@@ -38,6 +41,20 @@ if is_torch_available():
 else:
     pytestmark = pytest.mark.skip("Require Torch")
 
+if sys.version_info[0] == 2:
+    import cPickle as pickle
+
+    class TemporaryDirectory(object):
+        """Context manager for tempfile.mkdtemp() so it's usable with "with" statement."""
+        def __enter__(self):
+            self.name = tempfile.mkdtemp()
+            return self.name
+        def __exit__(self, exc_type, exc_value, traceback):
+            shutil.rmtree(self.name)
+else:
+    import pickle
+    TemporaryDirectory = tempfile.TemporaryDirectory
+    unicode = str
 
 def _config_zero_init(config):
     configs_no_init = copy.deepcopy(config)
@@ -57,6 +74,29 @@ class CommonTestCases:
         test_resize_embeddings = True
         test_head_masking = True
 
+        def test_save_load(self):
+            config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
+
+            for model_class in self.all_model_classes:
+                model = model_class(config)
+                model.eval()
+                with torch.no_grad():
+                    outputs = model(**inputs_dict)
+
+                with TemporaryDirectory() as tmpdirname:
+                    model.save_pretrained(tmpdirname)
+                    model = model_class.from_pretrained(tmpdirname)
+                    with torch.no_grad():
+                        after_outputs = model(**inputs_dict)
+
+                    # Make sure we don't have nans
+                    out_1 = after_outputs[0].numpy()
+                    out_2 = outputs[0].numpy()
+                    out_1 = out_1[~np.isnan(out_1)]
+                    out_2 = out_2[~np.isnan(out_2)]
+                    max_diff = np.amax(np.abs(out_1 - out_2))
+                    self.assertLessEqual(max_diff, 1e-5)
+
         def test_initialization(self):
             config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
 

transformers/tests/modeling_ctrl_test.py

+# coding=utf-8
+# Copyright 2018 Salesforce and HuggingFace Inc. team.
+# 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 pytest
+import shutil
+import pdb
+
+from transformers import is_torch_available
+
+if is_torch_available():
+    from transformers import (CTRLConfig, CTRLModel, CTRL_PRETRAINED_MODEL_ARCHIVE_MAP,
+                                    CTRLLMHeadModel)
+else:
+    pytestmark = pytest.mark.skip("Require Torch")
+
+from .modeling_common_test import (CommonTestCases, ids_tensor)
+from .configuration_common_test import ConfigTester
+
+
+class CTRLModelTest(CommonTestCases.CommonModelTester):
+
+    all_model_classes = (CTRLModel, CTRLLMHeadModel) if is_torch_available() else ()
+    test_pruning = False
+    test_torchscript = False
+    test_resize_embeddings = False
+    test_head_masking = False
+
+    class CTRLModelTester(object):
+
+        def __init__(self,
+                     parent,
+                     batch_size=13,
+                     seq_length=7,
+                     is_training=True,
+                     use_token_type_ids=True,
+                     use_input_mask=True,
+                     use_labels=True,
+                     use_mc_token_ids=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_token_type_ids = use_token_type_ids
+            self.use_input_mask = use_input_mask
+            self.use_labels = use_labels
+            self.use_mc_token_ids = use_mc_token_ids
+            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)
+
+            mc_token_ids = None
+            if self.use_mc_token_ids:
+                mc_token_ids = ids_tensor([self.batch_size, self.num_choices], 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 = CTRLConfig(
+                vocab_size_or_config_json_file=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,
+                n_ctx=self.max_position_embeddings
+                # type_vocab_size=self.type_vocab_size,
+                # initializer_range=self.initializer_range
+            )
+
+            head_mask = ids_tensor([self.num_hidden_layers, self.num_attention_heads], 2)
+
+            return config, input_ids, input_mask, head_mask, token_type_ids, mc_token_ids, sequence_labels, token_labels, choice_labels
+
+        def check_loss_output(self, result):
+            self.parent.assertListEqual(
+                list(result["loss"].size()),
+                [])
+
+        def create_and_check_ctrl_model(self, config, input_ids, input_mask, head_mask, token_type_ids, *args):
+            model = CTRLModel(config=config)
+            model.eval()
+
+            model(input_ids, token_type_ids=token_type_ids, head_mask=head_mask)
+            model(input_ids, token_type_ids=token_type_ids)
+            sequence_output, presents = model(input_ids)
+
+            result = {
+                "sequence_output": sequence_output,
+                "presents": presents,
+            }
+            self.parent.assertListEqual(
+                list(result["sequence_output"].size()),
+                [self.batch_size, self.seq_length, self.hidden_size])
+            self.parent.assertEqual(len(result["presents"]), config.n_layer)
+
+        def create_and_check_lm_head_model(self, config, input_ids, input_mask, head_mask, token_type_ids, *args):
+            model = CTRLLMHeadModel(config)
+            model.eval()
+
+            loss, lm_logits, _ = model(input_ids, token_type_ids=token_type_ids, labels=input_ids)
+
+            result = {
+                "loss": loss,
+                "lm_logits": lm_logits
+            }
+            self.parent.assertListEqual(
+                list(result["loss"].size()),
+                [])
+            self.parent.assertListEqual(
+                list(result["lm_logits"].size()),
+                [self.batch_size, self.seq_length, self.vocab_size])
+
+
+        def prepare_config_and_inputs_for_common(self):
+            config_and_inputs = self.prepare_config_and_inputs()
+
+            (config, input_ids, input_mask, head_mask, token_type_ids,
+             mc_token_ids, sequence_labels, token_labels, choice_labels) = config_and_inputs
+
+            inputs_dict = {
+                'input_ids': input_ids,
+                'token_type_ids': token_type_ids,
+                'head_mask': head_mask
+            }
+
+            return config, inputs_dict
+
+    def setUp(self):
+        self.model_tester = CTRLModelTest.CTRLModelTester(self)
+        self.config_tester = ConfigTester(self, config_class=CTRLConfig, n_embd=37)
+
+    def test_config(self):
+        self.config_tester.run_common_tests()
+
+    def test_ctrl_model(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_ctrl_model(*config_and_inputs)
+
+    def test_ctrl_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)
+
+    @pytest.mark.slow
+    def test_model_from_pretrained(self):
+        cache_dir = "/tmp/transformers_test/"
+        for model_name in list(CTRL_PRETRAINED_MODEL_ARCHIVE_MAP.keys())[:1]:
+            model = CTRLModel.from_pretrained(model_name, cache_dir=cache_dir)
+            shutil.rmtree(cache_dir)
+            self.assertIsNotNone(model)
+
+
+if __name__ == "__main__":
+    unittest.main()

transformers/tests/modeling_tf_common_test.py

@@ -14,6 +14,7 @@
 # limitations under the License.
 from __future__ import absolute_import, division, print_function
 
+import os
 import copy
 import json
 import logging
@@ -22,6 +23,7 @@ import random
 import shutil
 import unittest
 import uuid
+import tempfile
 
 import pytest
 import sys
@@ -36,6 +38,20 @@ if is_tf_available():
 else:
     pytestmark = pytest.mark.skip("Require TensorFlow")
 
+if sys.version_info[0] == 2:
+    import cPickle as pickle
+
+    class TemporaryDirectory(object):
+        """Context manager for tempfile.mkdtemp() so it's usable with "with" statement."""
+        def __enter__(self):
+            self.name = tempfile.mkdtemp()
+            return self.name
+        def __exit__(self, exc_type, exc_value, traceback):
+            shutil.rmtree(self.name)
+else:
+    import pickle
+    TemporaryDirectory = tempfile.TemporaryDirectory
+    unicode = str
 
 def _config_zero_init(config):
     configs_no_init = copy.deepcopy(config)
@@ -66,11 +82,31 @@ class TFCommonTestCases:
             #             self.assertIn(param.data.mean().item(), [0.0, 1.0],
             #             msg="Parameter {} of model {} seems not properly initialized".format(name, model_class))
 
+        def test_save_load(self):
+            config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
+
+            for model_class in self.all_model_classes:
+                model = model_class(config)
+                outputs = model(inputs_dict)
+
+                with TemporaryDirectory() as tmpdirname:
+                    model.save_pretrained(tmpdirname)
+                    model = model_class.from_pretrained(tmpdirname)
+                    after_outputs = model(inputs_dict)
+
+                    # Make sure we don't have nans
+                    out_1 = after_outputs[0].numpy()
+                    out_2 = outputs[0].numpy()
+                    out_1 = out_1[~np.isnan(out_1)]
+                    out_2 = out_2[~np.isnan(out_2)]
+                    max_diff = np.amax(np.abs(out_1 - out_2))
+                    self.assertLessEqual(max_diff, 1e-5)
 
         def test_pt_tf_model_equivalence(self):
             if not is_torch_available():
                 return
 
+            import torch
             import transformers
 
             config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
@@ -79,12 +115,71 @@ class TFCommonTestCases:
                 pt_model_class_name = model_class.__name__[2:]  # Skip the "TF" at the beggining
                 pt_model_class = getattr(transformers, pt_model_class_name)
 
+                config.output_hidden_states = True
                 tf_model = model_class(config)
                 pt_model = pt_model_class(config)
 
+                # Check we can load pt model in tf and vice-versa with model => model functions
                 tf_model = transformers.load_pytorch_model_in_tf2_model(tf_model, pt_model, tf_inputs=inputs_dict)
                 pt_model = transformers.load_tf2_model_in_pytorch_model(pt_model, tf_model)
 
+                # Check predictions on first output (logits/hidden-states) are close enought given low-level computational differences
+                pt_model.eval()
+                pt_inputs_dict = dict((name, torch.from_numpy(key.numpy()).to(torch.long))
+                                      for name, key in inputs_dict.items())
+                with torch.no_grad():
+                    pto = pt_model(**pt_inputs_dict)
+                tfo = tf_model(inputs_dict)
+                max_diff = np.amax(np.abs(tfo[0].numpy() - pto[0].numpy()))
+                self.assertLessEqual(max_diff, 2e-2)
+
+                # Check we can load pt model in tf and vice-versa with checkpoint => model functions
+                with TemporaryDirectory() as tmpdirname:
+                    pt_checkpoint_path = os.path.join(tmpdirname, 'pt_model.bin')
+                    torch.save(pt_model.state_dict(), pt_checkpoint_path)
+                    tf_model = transformers.load_pytorch_checkpoint_in_tf2_model(tf_model, pt_checkpoint_path)
+
+                    tf_checkpoint_path = os.path.join(tmpdirname, 'tf_model.h5')
+                    tf_model.save_weights(tf_checkpoint_path)
+                    pt_model = transformers.load_tf2_checkpoint_in_pytorch_model(pt_model, tf_checkpoint_path)
+
+                # Check predictions on first output (logits/hidden-states) are close enought given low-level computational differences
+                pt_model.eval()
+                pt_inputs_dict = dict((name, torch.from_numpy(key.numpy()).to(torch.long))
+                                      for name, key in inputs_dict.items())
+                with torch.no_grad():
+                    pto = pt_model(**pt_inputs_dict)
+                tfo = tf_model(inputs_dict)
+                max_diff = np.amax(np.abs(tfo[0].numpy() - pto[0].numpy()))
+                self.assertLessEqual(max_diff, 2e-2)
+
+        def test_compile_tf_model(self):
+            config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
+
+            input_ids = tf.keras.Input(batch_shape=(2, 2000), name='input_ids', dtype='int32')
+            optimizer = tf.keras.optimizers.Adam(learning_rate=3e-5, epsilon=1e-08, clipnorm=1.0)
+            loss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)
+            metric = tf.keras.metrics.SparseCategoricalAccuracy('accuracy')
+
+            for model_class in self.all_model_classes:
+                # Prepare our model
+                model = model_class(config)
+                
+                # Let's load it from the disk to be sure we can use pretrained weights
+                with TemporaryDirectory() as tmpdirname:
+                    outputs = model(inputs_dict)  # build the model
+                    model.save_pretrained(tmpdirname)
+                    model = model_class.from_pretrained(tmpdirname)
+
+                outputs_dict = model(input_ids)
+                hidden_states = outputs_dict[0]
+
+                # Add a dense layer on top to test intetgration with other keras modules
+                outputs = tf.keras.layers.Dense(2, activation='softmax', name='outputs')(hidden_states)
+
+                # Compile extended model
+                extended_model = tf.keras.Model(inputs=[input_ids], outputs=[outputs])
+                extended_model.compile(optimizer=optimizer, loss=loss, metrics=[metric])
 
         def test_keyword_and_dict_args(self):
             config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()

transformers/tests/modeling_tf_ctrl_test.py

+# 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 transformers import CTRLConfig, is_tf_available
+
+if is_tf_available():
+    import tensorflow as tf
+    from transformers.modeling_tf_ctrl import (TFCTRLModel, TFCTRLLMHeadModel,
+                                                TF_CTRL_PRETRAINED_MODEL_ARCHIVE_MAP)
+else:
+    pytestmark = pytest.mark.skip("Require TensorFlow")
+
+
+class TFCTRLModelTest(TFCommonTestCases.TFCommonModelTester):
+
+    all_model_classes = (TFCTRLModel, TFCTRLLMHeadModel) if is_tf_available() else ()
+
+    class TFCTRLModelTester(object):
+
+        def __init__(self,
+                     parent,
+                     batch_size=13,
+                     seq_length=7,
+                     is_training=True,
+                     use_token_type_ids=True,
+                     use_input_mask=True,
+                     use_labels=True,
+                     use_mc_token_ids=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_token_type_ids = use_token_type_ids
+            self.use_input_mask = use_input_mask
+            self.use_labels = use_labels
+            self.use_mc_token_ids = use_mc_token_ids
+            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)
+
+            mc_token_ids = None
+            if self.use_mc_token_ids:
+                mc_token_ids = ids_tensor([self.batch_size, self.num_choices], 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 = CTRLConfig(
+                vocab_size_or_config_json_file=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,
+                n_ctx=self.max_position_embeddings
+                # type_vocab_size=self.type_vocab_size,
+                # initializer_range=self.initializer_range
+            )
+
+            head_mask = ids_tensor([self.num_hidden_layers, self.num_attention_heads], 2)
+
+            return config, input_ids, input_mask, head_mask, token_type_ids, mc_token_ids, sequence_labels, token_labels, choice_labels
+
+        def create_and_check_ctrl_model(self, config, input_ids, input_mask, head_mask, token_type_ids, *args):
+            model = TFCTRLModel(config=config)
+            inputs = {'input_ids': input_ids,
+                      'attention_mask': input_mask,
+                      'token_type_ids': token_type_ids}
+            sequence_output = model(inputs)[0]
+
+            inputs = [input_ids, None, input_mask]  # None is the input for 'past'
+            sequence_output = model(inputs)[0]
+
+            sequence_output = model(input_ids)[0]
+
+            result = {
+                "sequence_output": sequence_output.numpy(),
+            }
+            self.parent.assertListEqual(
+                list(result["sequence_output"].shape),
+                [self.batch_size, self.seq_length, self.hidden_size])
+
+
+        def create_and_check_ctrl_lm_head(self, config, input_ids, input_mask, head_mask, token_type_ids, *args):
+            model = TFCTRLLMHeadModel(config=config)
+            inputs = {'input_ids': input_ids,
+                      'attention_mask': input_mask,
+                      'token_type_ids': token_type_ids}
+            prediction_scores = model(inputs)[0]
+            result = {
+                "prediction_scores": prediction_scores.numpy(),
+            }
+            self.parent.assertListEqual(
+                list(result["prediction_scores"].shape),
+                [self.batch_size, self.seq_length, self.vocab_size])
+
+        def prepare_config_and_inputs_for_common(self):
+            config_and_inputs = self.prepare_config_and_inputs()
+
+            (config, input_ids, input_mask, head_mask, token_type_ids,
+             mc_token_ids, 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 = TFCTRLModelTest.TFCTRLModelTester(self)
+        self.config_tester = ConfigTester(self, config_class=CTRLConfig, n_embd=37)
+
+    def test_config(self):
+        self.config_tester.run_common_tests()
+
+    def test_ctrl_model(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_ctrl_model(*config_and_inputs)
+
+    def test_ctrl_lm_head(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_ctrl_lm_head(*config_and_inputs)
+
+    @pytest.mark.slow
+    def test_model_from_pretrained(self):
+        cache_dir = "/tmp/transformers_test/"
+        for model_name in list(TF_CTRL_PRETRAINED_MODEL_ARCHIVE_MAP.keys())[:1]:
+            model = TFCTRLModel.from_pretrained(model_name, cache_dir=cache_dir)
+            shutil.rmtree(cache_dir)
+            self.assertIsNotNone(model)
+
+if __name__ == "__main__":
+    unittest.main()
+

transformers/tests/modeling_tf_gpt2_test.py

@@ -222,7 +222,7 @@ class TFGPT2ModelTest(TFCommonTestCases.TFCommonModelTester):
     @pytest.mark.slow
     def test_model_from_pretrained(self):
         cache_dir = "/tmp/transformers_test/"
-        for model_name in list(TF_gpt2_PRETRAINED_MODEL_ARCHIVE_MAP.keys())[:1]:
+        for model_name in list(TF_GPT2_PRETRAINED_MODEL_ARCHIVE_MAP.keys())[:1]:
             model = TFGPT2Model.from_pretrained(model_name, cache_dir=cache_dir)
             shutil.rmtree(cache_dir)
             self.assertIsNotNone(model)

transformers/tests/modeling_tf_xlnet_test.py

@@ -161,6 +161,11 @@ class TFXLNetModelTest(TFCommonTestCases.TFCommonModelTester):
                 "outputs": outputs.numpy(),
             }
 
+            config.mem_len = 0
+            model = TFXLNetModel(config)
+            no_mems_outputs = model(inputs)
+            self.parent.assertEqual(len(no_mems_outputs), 1)
+
             self.parent.assertListEqual(
                 list(result["outputs"].shape),
                 [self.batch_size, self.seq_length, self.hidden_size])

transformers/tests/modeling_xlnet_test.py

@@ -150,6 +150,12 @@ class XLNetModelTest(CommonTestCases.CommonModelTester):
                 "outputs": outputs,
             }
 
+            config.mem_len = 0
+            model = XLNetModel(config)
+            model.eval()
+            no_mems_outputs = model(input_ids_1)
+            self.parent.assertEqual(len(no_mems_outputs), 1)
+
             self.parent.assertListEqual(
                 list(result["outputs"].size()),
                 [self.batch_size, self.seq_length, self.hidden_size])

transformers/tests/tokenization_bert_test.py

@@ -131,8 +131,8 @@ class BertTokenizationTest(CommonTestCases.CommonTokenizerTester):
         text = tokenizer.encode("sequence builders")
         text_2 = tokenizer.encode("multi-sequence build")
 
-        encoded_sentence = tokenizer.add_special_tokens_single_sequence(text)
-        encoded_pair = tokenizer.add_special_tokens_sequence_pair(text, text_2)
+        encoded_sentence = tokenizer.build_inputs_with_special_tokens(text)
+        encoded_pair = tokenizer.build_inputs_with_special_tokens(text, text_2)
 
         assert encoded_sentence == [101] + text + [102]
         assert encoded_pair == [101] + text + [102] + text_2 + [102]

transformers/tests/tokenization_ctrl_test.py

+# coding=utf-8
+# Copyright 2018 Salesforce and HuggingFace Inc. team.
+# 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, division, print_function, unicode_literals
+
+import os
+import unittest
+import json
+from io import open
+
+from transformers.tokenization_ctrl import CTRLTokenizer, VOCAB_FILES_NAMES
+
+from .tokenization_tests_commons import CommonTestCases
+
+class CTRLTokenizationTest(CommonTestCases.CommonTokenizerTester):
+
+    tokenizer_class = CTRLTokenizer
+
+    def setUp(self):
+        super(CTRLTokenizationTest, self).setUp()
+
+        # Adapted from Sennrich et al. 2015 and https://github.com/rsennrich/subword-nmt
+        vocab = ['adapt', 're@@', 'a@@', 'apt', 'c@@', 't', '<unk>']
+        vocab_tokens = dict(zip(vocab, range(len(vocab))))
+        merges = ["#version: 0.2", 'a p', 'ap t</w>', 'r e', 'a d', 'ad apt</w>', '']
+        self.special_tokens_map = {"unk_token": "<unk>"}
+
+        self.vocab_file = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES['vocab_file'])
+        self.merges_file = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES['merges_file'])
+        with open(self.vocab_file, "w", encoding="utf-8") as fp:
+            fp.write(json.dumps(vocab_tokens) + "\n")
+        with open(self.merges_file, "w", encoding="utf-8") as fp:
+            fp.write("\n".join(merges))
+
+    def get_tokenizer(self, **kwargs):
+        kwargs.update(self.special_tokens_map)
+        return CTRLTokenizer.from_pretrained(self.tmpdirname, **kwargs)
+
+    def get_input_output_texts(self):
+        input_text = u"adapt react readapt apt"
+        output_text = u"adapt react readapt apt"
+        return input_text, output_text
+
+    def test_full_tokenizer(self):
+        tokenizer = CTRLTokenizer(self.vocab_file, self.merges_file, **self.special_tokens_map)
+        text = "adapt react readapt apt"
+        bpe_tokens = 'adapt re@@ a@@ c@@ t re@@ adapt apt'.split()
+        tokens = tokenizer.tokenize(text)
+        self.assertListEqual(tokens, bpe_tokens)
+
+        input_tokens = tokens + [tokenizer.unk_token]
+
+        input_bpe_tokens = [0, 1, 2, 4, 5, 1, 0, 3, 6]
+        self.assertListEqual(
+            tokenizer.convert_tokens_to_ids(input_tokens), input_bpe_tokens)
+
+
+if __name__ == '__main__':
+    unittest.main()

transformers/tests/tokenization_distilbert_test.py

@@ -36,8 +36,8 @@ class DistilBertTokenizationTest(BertTokenizationTest):
         text = tokenizer.encode("sequence builders")
         text_2 = tokenizer.encode("multi-sequence build")
 
-        encoded_sentence = tokenizer.add_special_tokens_single_sequence(text)
-        encoded_pair = tokenizer.add_special_tokens_sequence_pair(text, text_2)
+        encoded_sentence = tokenizer.build_inputs_with_special_tokens(text)
+        encoded_pair = tokenizer.build_inputs_with_special_tokens(text, text_2)
 
         assert encoded_sentence == [tokenizer.cls_token_id] + text + [tokenizer.sep_token_id]
         assert encoded_pair == [tokenizer.cls_token_id] + text + [tokenizer.sep_token_id] + \

transformers/tests/tokenization_roberta_test.py

@@ -87,8 +87,8 @@ class RobertaTokenizationTest(CommonTestCases.CommonTokenizerTester):
         encoded_text_from_decode = tokenizer.encode("sequence builders", add_special_tokens=True)
         encoded_pair_from_decode = tokenizer.encode("sequence builders", "multi-sequence build", add_special_tokens=True)
 
-        encoded_sentence = tokenizer.add_special_tokens_single_sequence(text)
-        encoded_pair = tokenizer.add_special_tokens_sequence_pair(text, text_2)
+        encoded_sentence = tokenizer.build_inputs_with_special_tokens(text)
+        encoded_pair = tokenizer.build_inputs_with_special_tokens(text, text_2)
 
         assert encoded_sentence == encoded_text_from_decode
         assert encoded_pair == encoded_pair_from_decode

transformers/tests/tokenization_tests_commons.py

@@ -193,12 +193,12 @@ class CommonTestCases:
 
             tokenizer = self.get_tokenizer()
 
-            if tokenizer.add_special_tokens_sequence_pair.__qualname__.split('.')[0] != "PreTrainedTokenizer":
+            if tokenizer.build_inputs_with_special_tokens.__qualname__.split('.')[0] != "PreTrainedTokenizer":
                 seq_0 = "Test this method."
                 seq_1 = "With these inputs."
                 information = tokenizer.encode_plus(seq_0, seq_1, add_special_tokens=True)
                 sequences, mask = information["input_ids"], information["token_type_ids"]
-                assert len(sequences) == len(mask)
+                self.assertEqual(len(sequences), len(mask))
 
         def test_number_of_added_tokens(self):
             tokenizer = self.get_tokenizer()
@@ -211,7 +211,7 @@ class CommonTestCases:
 
             # Method is implemented (e.g. not GPT-2)
             if len(attached_sequences) != 2:
-                assert tokenizer.num_added_tokens(pair=True) == len(attached_sequences) - len(sequences)
+                self.assertEqual(tokenizer.num_added_tokens(pair=True), len(attached_sequences) - len(sequences))
 
         def test_maximum_encoding_length_single_input(self):
             tokenizer = self.get_tokenizer()
@@ -227,10 +227,10 @@ class CommonTestCases:
             truncated_sequence = information["input_ids"]
             overflowing_tokens = information["overflowing_tokens"]
 
-            assert len(overflowing_tokens) == 2 + stride
-            assert overflowing_tokens == sequence[-(2 + stride):]
-            assert len(truncated_sequence) == total_length - 2
-            assert truncated_sequence == tokenizer.add_special_tokens_single_sequence(sequence[:-2])
+            self.assertEqual(len(overflowing_tokens), 2 + stride)
+            self.assertEqual(overflowing_tokens, sequence[-(2 + stride):])
+            self.assertEqual(len(truncated_sequence), total_length - 2)
+            self.assertEqual(truncated_sequence, tokenizer.build_inputs_with_special_tokens(sequence[:-2]))
 
         def test_maximum_encoding_length_pair_input(self):
             tokenizer = self.get_tokenizer()
@@ -243,26 +243,26 @@ class CommonTestCases:
             sequence_1_no_special_tokens = tokenizer.encode(seq_1)
 
             sequence = tokenizer.encode(seq_0, seq_1, add_special_tokens=True)
-            truncated_second_sequence = tokenizer.add_special_tokens_sequence_pair(
+            truncated_second_sequence = tokenizer.build_inputs_with_special_tokens(
                 tokenizer.encode(seq_0),
                 tokenizer.encode(seq_1)[:-2]
             )
 
             information = tokenizer.encode_plus(seq_0, seq_1, max_length=len(sequence) - 2, add_special_tokens=True,
-                                                stride=stride, truncate_first_sequence=False)
+                                                stride=stride, truncation_strategy='only_second')
             information_first_truncated = tokenizer.encode_plus(seq_0, seq_1, max_length=len(sequence) - 2,
                                                                 add_special_tokens=True, stride=stride,
-                                                                truncate_first_sequence=True)
+                                                                truncation_strategy='only_first')
 
             truncated_sequence = information["input_ids"]
             overflowing_tokens = information["overflowing_tokens"]
             overflowing_tokens_first_truncated = information_first_truncated["overflowing_tokens"]
 
-            assert len(overflowing_tokens) == 2 + stride
-            assert overflowing_tokens == sequence_1_no_special_tokens[-(2 + stride):]
-            assert overflowing_tokens_first_truncated == sequence_0_no_special_tokens[-(2 + stride):]
-            assert len(truncated_sequence) == len(sequence) - 2
-            assert truncated_sequence == truncated_second_sequence
+            self.assertEqual(len(overflowing_tokens), 2 + stride)
+            self.assertEqual(overflowing_tokens, sequence_1_no_special_tokens[-(2 + stride):])
+            self.assertEqual(overflowing_tokens_first_truncated, sequence_0_no_special_tokens[-(2 + stride):])
+            self.assertEqual(len(truncated_sequence), len(sequence) - 2)
+            self.assertEqual(truncated_sequence, truncated_second_sequence)
 
         def test_encode_input_type(self):
             tokenizer = self.get_tokenizer()
@@ -273,5 +273,43 @@ class CommonTestCases:
             input_ids = tokenizer.convert_tokens_to_ids(tokens)
             formatted_input = tokenizer.encode(sequence, add_special_tokens=True)
 
-            assert tokenizer.encode(tokens, add_special_tokens=True) == formatted_input
-            assert tokenizer.encode(input_ids, add_special_tokens=True) == formatted_input
+            self.assertEqual(tokenizer.encode(tokens, add_special_tokens=True), formatted_input)
+            self.assertEqual(tokenizer.encode(input_ids, add_special_tokens=True), formatted_input)
+
+        def test_special_tokens_mask(self):
+            tokenizer = self.get_tokenizer()
+
+            sequence_0 = "Encode this."
+            sequence_1 = "This one too please."
+
+            # Testing single inputs
+            encoded_sequence = tokenizer.encode(sequence_0)
+            encoded_sequence_dict = tokenizer.encode_plus(sequence_0, add_special_tokens=True)
+            encoded_sequence_w_special = encoded_sequence_dict["input_ids"]
+            special_tokens_mask = encoded_sequence_dict["special_tokens_mask"]
+            self.assertEqual(len(special_tokens_mask), len(encoded_sequence_w_special))
+
+            filtered_sequence = [(x if not special_tokens_mask[i] else None) for i, x in enumerate(encoded_sequence_w_special)]
+            filtered_sequence = [x for x in filtered_sequence if x is not None]
+            self.assertEqual(encoded_sequence, filtered_sequence)
+
+            # Testing inputs pairs
+            encoded_sequence = tokenizer.encode(sequence_0) + tokenizer.encode(sequence_1)
+            encoded_sequence_dict = tokenizer.encode_plus(sequence_0, sequence_1, add_special_tokens=True)
+            encoded_sequence_w_special = encoded_sequence_dict["input_ids"]
+            special_tokens_mask = encoded_sequence_dict["special_tokens_mask"]
+            self.assertEqual(len(special_tokens_mask), len(encoded_sequence_w_special))
+
+            filtered_sequence = [(x if not special_tokens_mask[i] else None) for i, x in enumerate(encoded_sequence_w_special)]
+            filtered_sequence = [x for x in filtered_sequence if x is not None]
+            self.assertEqual(encoded_sequence, filtered_sequence)
+
+            # Testing with already existing special tokens
+            if tokenizer.cls_token_id == tokenizer.unk_token_id and tokenizer.cls_token_id == tokenizer.unk_token_id:
+                tokenizer.add_special_tokens({'cls_token': '</s>', 'sep_token': '<s>'})
+            encoded_sequence_dict = tokenizer.encode_plus(sequence_0, add_special_tokens=True)
+            encoded_sequence_w_special = encoded_sequence_dict["input_ids"]
+            special_tokens_mask_orig = encoded_sequence_dict["special_tokens_mask"]
+            special_tokens_mask = tokenizer.get_special_tokens_mask(encoded_sequence_w_special, already_has_special_tokens=True)
+            self.assertEqual(len(special_tokens_mask), len(encoded_sequence_w_special))
+            self.assertEqual(special_tokens_mask_orig, special_tokens_mask)

transformers/tests/tokenization_xlm_test.py

@@ -72,8 +72,8 @@ class XLMTokenizationTest(CommonTestCases.CommonTokenizerTester):
         text = tokenizer.encode("sequence builders")
         text_2 = tokenizer.encode("multi-sequence build")
 
-        encoded_sentence = tokenizer.add_special_tokens_single_sequence(text)
-        encoded_pair = tokenizer.add_special_tokens_sequence_pair(text, text_2)
+        encoded_sentence = tokenizer.build_inputs_with_special_tokens(text)
+        encoded_pair = tokenizer.build_inputs_with_special_tokens(text, text_2)
 
         assert encoded_sentence == [1] + text + [1]
         assert encoded_pair == [1] + text + [1] + text_2 + [1]

transformers/tests/tokenization_xlnet_test.py

@@ -95,8 +95,8 @@ class XLNetTokenizationTest(CommonTestCases.CommonTokenizerTester):
         text = tokenizer.encode("sequence builders")
         text_2 = tokenizer.encode("multi-sequence build")
 
-        encoded_sentence = tokenizer.add_special_tokens_single_sequence(text)
-        encoded_pair = tokenizer.add_special_tokens_sequence_pair(text, text_2)
+        encoded_sentence = tokenizer.build_inputs_with_special_tokens(text)
+        encoded_pair = tokenizer.build_inputs_with_special_tokens(text, text_2)
 
         assert encoded_sentence == text + [4, 3]
         assert encoded_pair == text + [4] + text_2 + [4, 3]

transformers/tokenization_auto.py

@@ -21,6 +21,7 @@ import logging
 from .tokenization_bert import BertTokenizer
 from .tokenization_openai import OpenAIGPTTokenizer
 from .tokenization_gpt2 import GPT2Tokenizer
+from .tokenization_ctrl import CTRLTokenizer
 from .tokenization_transfo_xl import TransfoXLTokenizer
 from .tokenization_xlnet import XLNetTokenizer
 from .tokenization_xlm import XLMTokenizer
@@ -45,6 +46,7 @@ class AutoTokenizer(object):
             - contains `bert`: BertTokenizer (Bert model)
             - contains `openai-gpt`: OpenAIGPTTokenizer (OpenAI GPT model)
             - contains `gpt2`: GPT2Tokenizer (OpenAI GPT-2 model)
+            - contains `ctrl`: CTRLTokenizer (Salesforce CTRL model)
             - contains `transfo-xl`: TransfoXLTokenizer (Transformer-XL model)
             - contains `xlnet`: XLNetTokenizer (XLNet model)
             - contains `xlm`: XLMTokenizer (XLM model)
@@ -67,6 +69,7 @@ class AutoTokenizer(object):
             - contains `bert`: BertTokenizer (Bert model)
             - contains `openai-gpt`: OpenAIGPTTokenizer (OpenAI GPT model)
             - contains `gpt2`: GPT2Tokenizer (OpenAI GPT-2 model)
+            - contains `ctrl`: CTRLTokenizer (Salesforce CTRL model)
             - contains `transfo-xl`: TransfoXLTokenizer (Transformer-XL model)
             - contains `xlnet`: XLNetTokenizer (XLNet model)
             - contains `xlm`: XLMTokenizer (XLM model)
@@ -114,7 +117,8 @@ class AutoTokenizer(object):
             return XLNetTokenizer.from_pretrained(pretrained_model_name_or_path, *inputs, **kwargs)
         elif 'xlm' in pretrained_model_name_or_path:
             return XLMTokenizer.from_pretrained(pretrained_model_name_or_path, *inputs, **kwargs)
-
+        elif 'ctrl' in pretrained_model_name_or_path:
+            return CTRLTokenizer.from_pretrained(pretrained_model_name_or_path, *inputs, **kwargs)
         raise ValueError("Unrecognized model identifier in {}. Should contains one of "
                          "'bert', 'openai-gpt', 'gpt2', 'transfo-xl', 'xlnet', "
-                         "'xlm', 'roberta'".format(pretrained_model_name_or_path))
+                         "'xlm', 'roberta', 'ctrl'".format(pretrained_model_name_or_path))

transformers/tokenization_bert.py

@@ -44,6 +44,8 @@ PRETRAINED_VOCAB_FILES_MAP = {
         'bert-large-uncased-whole-word-masking-finetuned-squad': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-uncased-whole-word-masking-finetuned-squad-vocab.txt",
         'bert-large-cased-whole-word-masking-finetuned-squad': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-cased-whole-word-masking-finetuned-squad-vocab.txt",
         'bert-base-cased-finetuned-mrpc': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-cased-finetuned-mrpc-vocab.txt",
+        'bert-base-german-dbmdz-cased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-german-dbmdz-cased-vocab.txt",
+        'bert-base-german-dbmdz-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-german-dbmdz-uncased-vocab.txt",
     }
 }
 
@@ -61,6 +63,8 @@ PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
     'bert-large-uncased-whole-word-masking-finetuned-squad': 512,
     'bert-large-cased-whole-word-masking-finetuned-squad': 512,
     'bert-base-cased-finetuned-mrpc': 512,
+    'bert-base-german-dbmdz-cased': 512,
+    'bert-base-german-dbmdz-uncased': 512,
 }
 
 PRETRAINED_INIT_CONFIGURATION = {
@@ -77,6 +81,8 @@ PRETRAINED_INIT_CONFIGURATION = {
     'bert-large-uncased-whole-word-masking-finetuned-squad': {'do_lower_case': True},
     'bert-large-cased-whole-word-masking-finetuned-squad': {'do_lower_case': False},
     'bert-base-cased-finetuned-mrpc': {'do_lower_case': False},
+    'bert-base-german-dbmdz-cased': {'do_lower_case': False},
+    'bert-base-german-dbmdz-uncased': {'do_lower_case': True},
 }
 
 
@@ -187,33 +193,59 @@ class BertTokenizer(PreTrainedTokenizer):
         out_string = ' '.join(tokens).replace(' ##', '').strip()
         return out_string
 
-    def add_special_tokens_single_sequence(self, token_ids):
+    def build_inputs_with_special_tokens(self, token_ids_0, token_ids_1=None):
         """
-        Adds special tokens to the a sequence for sequence classification tasks.
-        A BERT sequence has the following format: [CLS] X [SEP]
+        Build model inputs from a sequence or a pair of sequence for sequence classification tasks
+        by concatenating and adding special tokens.
+        A BERT sequence has the following format:
+            single sequence: [CLS] X [SEP]
+            pair of sequences: [CLS] A [SEP] B [SEP]
         """
-        return [self.cls_token_id] + token_ids + [self.sep_token_id]
+        if token_ids_1 is None:
+            return [self.cls_token_id] + token_ids_0 + [self.sep_token_id]
+        cls = [self.cls_token_id]
+        sep = [self.sep_token_id]
+        return cls + token_ids_0 + sep + token_ids_1 + sep
 
-    def add_special_tokens_sequence_pair(self, token_ids_0, token_ids_1):
+    def get_special_tokens_mask(self, token_ids_0, token_ids_1=None, already_has_special_tokens=False):
         """
-        Adds special tokens to a sequence pair for sequence classification tasks.
-        A BERT sequence pair has the following format: [CLS] A [SEP] B [SEP]
+        Retrieves sequence ids from a token list that has no special tokens added. This method is called when adding
+        special tokens using the tokenizer ``prepare_for_model`` or ``encode_plus`` methods.
+
+        Args:
+            token_ids_0: list of ids (must not contain special tokens)
+            token_ids_1: Optional list of ids (must not contain special tokens), necessary when fetching sequence ids
+                for sequence pairs
+            already_has_special_tokens: (default False) Set to True if the token list is already formated with
+                special tokens for the model
+
+        Returns:
+            A list of integers in the range [0, 1]: 0 for a special token, 1 for a sequence token.
         """
-        sep = [self.sep_token_id]
-        cls = [self.cls_token_id]
 
-        return cls + token_ids_0 + sep + token_ids_1 + sep
+        if already_has_special_tokens:
+            if token_ids_1 is not None:
+                raise ValueError("You should not supply a second sequence if the provided sequence of "
+                                 "ids is already formated with special tokens for the model.")
+            return list(map(lambda x: 1 if x in [self.sep_token_id, self.cls_token_id] else 0, token_ids_0))
 
-    def create_token_type_ids_from_sequences(self, token_ids_0, token_ids_1):
+        if token_ids_1 is not None:
+            return [1] + ([0] * len(token_ids_0)) + [1] + ([0] * len(token_ids_1)) + [1]
+        return [1] + ([0] * len(token_ids_0)) + [1]
+
+    def create_token_type_ids_from_sequences(self, token_ids_0, token_ids_1=None):
         """
         Creates a mask from the two sequences passed to be used in a sequence-pair classification task.
         A BERT sequence pair mask has the following format:
         0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1
         | first sequence    | second sequence
+
+        if token_ids_1 is None, only returns the first portion of the mask (0's).
         """
         sep = [self.sep_token_id]
         cls = [self.cls_token_id]
-
+        if token_ids_1 is None:
+            return len(cls + token_ids_0 + sep) * [0]
         return len(cls + token_ids_0 + sep) * [0] + len(token_ids_1 + sep) * [1]
 
     def save_vocabulary(self, vocab_path):

transformers/tokenization_ctrl.py

+# coding=utf-8
+# Copyright 2018 Salesforce and The HuggingFace Inc. team.
+#
+# 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.
+"""Tokenization classes for Salesforce CTRL."""
+from __future__ import (absolute_import, division, print_function,
+                        unicode_literals)
+
+import json
+import logging
+import os
+import regex as re
+from io import open
+
+from .tokenization_utils import PreTrainedTokenizer
+
+logger = logging.getLogger(__name__)
+
+VOCAB_FILES_NAMES = {
+    'vocab_file': 'vocab.json',
+    'merges_file': 'merges.txt',
+}
+
+PRETRAINED_VOCAB_FILES_MAP = {
+    'vocab_file':
+    {
+        'ctrl': "https://raw.githubusercontent.com/salesforce/ctrl/master/ctrl-vocab.json",
+    },
+    'merges_file':
+    {
+        'ctrl': "https://raw.githubusercontent.com/salesforce/ctrl/master/ctrl-merges.txt",
+    },
+}
+
+PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
+    'ctrl': 256,
+}
+
+def get_pairs(word):
+    """Return set of symbol pairs in a word.
+
+    Word is represented as tuple of symbols (symbols being variable-length strings).
+    """
+    pairs = set()
+    prev_char = word[0]
+    for char in word[1:]:
+        pairs.add((prev_char, char))
+        prev_char = char
+
+    pairs = set(pairs)
+    return pairs
+
+class CTRLTokenizer(PreTrainedTokenizer):
+    """
+    CTRL BPE tokenizer. Peculiarities:
+        - Byte-level Byte-Pair-Encoding
+        - Requires a space to start the input string => the encoding methods should be called with the
+          ``add_prefix_space`` flag set to ``True``.
+          Otherwise, this tokenizer ``encode`` and ``decode`` method will not conserve
+          the absence of a space at the beginning of a string: `tokenizer.decode(tokenizer.encode("Hello")) = " Hello"`
+    """
+    vocab_files_names = VOCAB_FILES_NAMES
+    pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
+    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
+
+    def __init__(self, vocab_file, merges_file, unk_token="<unk>", **kwargs):
+        super(CTRLTokenizer, self).__init__(unk_token=unk_token, **kwargs)
+        self.max_len_single_sentence = self.max_len # no default special tokens - you can update this value if you add special tokens
+        self.max_len_sentences_pair = self.max_len # no default special tokens - you can update this value if you add special tokens
+
+        self.encoder = json.load(open(vocab_file, encoding="utf-8"))
+        self.decoder = {v:k for k,v in self.encoder.items()}
+        merges = open(merges_file, encoding='utf-8').read().split('\n')[1:-1]
+        merges = [tuple(merge.split()) for merge in merges]
+        self.bpe_ranks = dict(zip(merges, range(len(merges))))
+        self.cache = {}
+
+    @property
+    def vocab_size(self):
+        return len(self.encoder)
+
+    def bpe(self, token):
+        if token in self.cache:
+            return self.cache[token]
+        word = tuple(token)
+        word = tuple(list(word[:-1]) + [word[-1]+'</w>'])
+        pairs = get_pairs(word)
+
+        if not pairs:
+            return token
+
+        while True:
+            bigram = min(pairs, key = lambda pair: self.bpe_ranks.get(pair, float('inf')))
+            if bigram not in self.bpe_ranks:
+                break
+            first, second = bigram
+            new_word = []
+            i = 0
+            while i < len(word):
+                try:
+                    j = word.index(first, i)
+                    new_word.extend(word[i:j])
+                    i = j
+                except:
+                    new_word.extend(word[i:])
+                    break
+
+                if word[i] == first and i < len(word)-1 and word[i+1] == second:
+                    new_word.append(first+second)
+                    i += 2
+                else:
+                    new_word.append(word[i])
+                    i += 1
+            new_word = tuple(new_word)
+            word = new_word
+            if len(word) == 1:
+                break
+            else:
+                pairs = get_pairs(word)
+        word = '@@ '.join(word)
+        word = word[:-4]
+        self.cache[token] = word
+        return word
+
+    def _tokenize(self, text):
+        """ Tokenize a string.
+        """
+        split_tokens = []
+
+        text = text.split(' ')
+
+        for token in text:
+            split_tokens.extend([t for t in self.bpe(token).split(' ')])
+        return split_tokens
+
+    def _convert_token_to_id(self, token):
+        """ Converts a token (str/unicode) in an id using the vocab. """
+        return self.encoder.get(token, self.encoder.get(self.unk_token))
+
+    def _convert_id_to_token(self, index):
+        """Converts an index (integer) in a token (string/unicode) using the vocab."""
+        return self.decoder.get(index, self.unk_token)
+
+    def convert_tokens_to_string(self, tokens):
+        """ Converts a sequence of tokens (string) in a single string. """
+        out_string = ' '.join(tokens).replace('@@ ', '').strip()
+        return out_string
+
+    def save_vocabulary(self, save_directory):
+        """Save the tokenizer vocabulary and merge files to a directory."""
+        if not os.path.isdir(save_directory):
+            logger.error("Vocabulary path ({}) should be a directory".format(save_directory))
+            return
+        vocab_file = os.path.join(save_directory, VOCAB_FILES_NAMES['vocab_file'])
+        merge_file = os.path.join(save_directory, VOCAB_FILES_NAMES['merges_file'])
+
+        with open(vocab_file, 'w', encoding='utf-8') as f:
+            f.write(json.dumps(self.encoder, ensure_ascii=False))
+
+        index = 0
+        with open(merge_file, "w", encoding="utf-8") as writer:
+            writer.write(u'#version: 0.2\n')
+            for bpe_tokens, token_index in sorted(self.bpe_ranks.items(), key=lambda kv: kv[1]):
+                if index != token_index:
+                    logger.warning("Saving vocabulary to {}: BPE merge indices are not consecutive."
+                                   " Please check that the tokenizer is not corrupted!".format(merge_file))
+                    index = token_index
+                writer.write(' '.join(bpe_tokens) + u'\n')
+                index += 1
+
+        return vocab_file, merge_file
+
+    # def decode(self, token_ids, skip_special_tokens=False, clean_up_tokenization_spaces=True):
+    #     filtered_tokens = ' '.join(self.convert_ids_to_tokens(token_ids, skip_special_tokens=skip_special_tokens))
+    #     tokens_generated_so_far = re.sub('(@@ )', '', string=filtered_tokens)
+    #     tokens_generated_so_far = re.sub('(@@ ?$)', '', string=tokens_generated_so_far)
+    #     return ''.join(tokens_generated_so_far)

transformers/tokenization_gpt2.py

@@ -46,12 +46,14 @@ PRETRAINED_VOCAB_FILES_MAP = {
         'gpt2': "https://s3.amazonaws.com/models.huggingface.co/bert/gpt2-vocab.json",
         'gpt2-medium': "https://s3.amazonaws.com/models.huggingface.co/bert/gpt2-medium-vocab.json",
         'gpt2-large': "https://s3.amazonaws.com/models.huggingface.co/bert/gpt2-large-vocab.json",
+        'distilgpt2': "https://s3.amazonaws.com/models.huggingface.co/bert/distilgpt2-vocab.json",
     },
     'merges_file':
     {
         'gpt2': "https://s3.amazonaws.com/models.huggingface.co/bert/gpt2-merges.txt",
         'gpt2-medium': "https://s3.amazonaws.com/models.huggingface.co/bert/gpt2-medium-merges.txt",
         'gpt2-large': "https://s3.amazonaws.com/models.huggingface.co/bert/gpt2-large-merges.txt",
+        'distilgpt2': "https://s3.amazonaws.com/models.huggingface.co/bert/distilgpt2-merges.txt",
     },
 }
 
@@ -59,6 +61,7 @@ PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
     'gpt2': 1024,
     'gpt2-medium': 1024,
     'gpt2-large': 1024,
+    'distilgpt2': 1024,
 }
 
 @lru_cache()

transformers/tokenization_roberta.py

@@ -46,12 +46,14 @@ PRETRAINED_VOCAB_FILES_MAP = {
         'roberta-base': "https://s3.amazonaws.com/models.huggingface.co/bert/roberta-base-vocab.json",
         'roberta-large': "https://s3.amazonaws.com/models.huggingface.co/bert/roberta-large-vocab.json",
         'roberta-large-mnli': "https://s3.amazonaws.com/models.huggingface.co/bert/roberta-large-mnli-vocab.json",
+        'distilroberta-base': "https://s3.amazonaws.com/models.huggingface.co/bert/distilroberta-base-vocab.json",
     },
     'merges_file':
     {
         'roberta-base': "https://s3.amazonaws.com/models.huggingface.co/bert/roberta-base-merges.txt",
         'roberta-large': "https://s3.amazonaws.com/models.huggingface.co/bert/roberta-large-merges.txt",
         'roberta-large-mnli': "https://s3.amazonaws.com/models.huggingface.co/bert/roberta-large-mnli-merges.txt",
+        'distilroberta-base': "https://s3.amazonaws.com/models.huggingface.co/bert/distilroberta-base-merges.txt",
     },
 }
 
@@ -59,6 +61,7 @@ PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
     'roberta-base': 512,
     'roberta-large': 512,
     'roberta-large-mnli': 512,
+    'distilroberta-base': 512,
 }
 
 
@@ -84,30 +87,57 @@ class RobertaTokenizer(GPT2Tokenizer):
         self.max_len_single_sentence = self.max_len - 2  # take into account special tokens
         self.max_len_sentences_pair = self.max_len - 4  # take into account special tokens
 
-    def add_special_tokens_single_sequence(self, token_ids):
+    def build_inputs_with_special_tokens(self, token_ids_0, token_ids_1=None):
         """
-        Adds special tokens to a sequence for sequence classification tasks.
-        A RoBERTa sequence has the following format: <s> X </s>
+        Build model inputs from a sequence or a pair of sequence for sequence classification tasks
+        by concatenating and adding special tokens.
+        A RoBERTa sequence has the following format:
+            single sequence: <s> X </s>
+            pair of sequences: <s> A </s></s> B </s>
         """
-        return [self.cls_token_id] + token_ids + [self.sep_token_id]
+        if token_ids_1 is None:
+            return [self.cls_token_id] + token_ids_0 + [self.sep_token_id]
+        cls = [self.cls_token_id]
+        sep = [self.sep_token_id]
+        return cls + token_ids_0 + sep + sep + token_ids_1 + sep
 
-    def add_special_tokens_sequence_pair(self, token_ids_0, token_ids_1):
+    def get_special_tokens_mask(self, token_ids_0, token_ids_1=None, already_has_special_tokens=False):
         """
-        Adds special tokens to a sequence pair for sequence classification tasks.
-        A RoBERTa sequence pair has the following format: <s> A </s></s> B </s>
+        Retrieves sequence ids from a token list that has no special tokens added. This method is called when adding
+        special tokens using the tokenizer ``prepare_for_model`` or ``encode_plus`` methods.
+
+        Args:
+            token_ids_0: list of ids (must not contain special tokens)
+            token_ids_1: Optional list of ids (must not contain special tokens), necessary when fetching sequence ids
+                for sequence pairs
+            already_has_special_tokens: (default False) Set to True if the token list is already formated with
+                special tokens for the model
+
+        Returns:
+            A list of integers in the range [0, 1]: 0 for a special token, 1 for a sequence token.
         """
-        sep = [self.sep_token_id]
-        cls = [self.cls_token_id]
-        return cls + token_ids_0 + sep + sep + token_ids_1 + sep
+        if already_has_special_tokens:
+            if token_ids_1 is not None:
+                raise ValueError("You should not supply a second sequence if the provided sequence of "
+                                 "ids is already formated with special tokens for the model.")
+            return list(map(lambda x: 1 if x in [self.sep_token_id, self.cls_token_id] else 0, token_ids_0))
 
-    def create_token_type_ids_from_sequences(self, token_ids_0, token_ids_1):
+        if token_ids_1 is None:
+            return [1] + ([0] * len(token_ids_0)) + [1]
+        return [1] + ([0] * len(token_ids_0)) + [1, 1] + ([0] * len(token_ids_1)) + [1]
+
+    def create_token_type_ids_from_sequences(self, token_ids_0, token_ids_1=None):
         """
         Creates a mask from the two sequences passed to be used in a sequence-pair classification task.
         A RoBERTa sequence pair mask has the following format:
         0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1
         | first sequence    | second sequence
+
+        if token_ids_1 is None, only returns the first portion of the mask (0's).
         """
         sep = [self.sep_token_id]
         cls = [self.cls_token_id]
 
-        return len(cls + token_ids_0 + sep + sep) * [0] + len(token_ids_1 + sep) * [1]
\ No newline at end of file
+        if token_ids_1 is None:
+            return len(cls + token_ids_0 + sep) * [0]
+        return len(cls + token_ids_0 + sep + sep) * [0] + len(token_ids_1 + sep) * [1]

transformers/tokenization_utils.py

@@ -337,13 +337,13 @@ class PreTrainedTokenizer(object):
                 vocab_files[file_id] = full_file_name
 
             if all(full_file_name is None for full_file_name in vocab_files.values()):
-                logger.error(
-                    "Model name '{}' was not found in model name list ({}). "
-                    "We assumed '{}' was a path or url but couldn't find tokenizer files"
-                    "at this path or url.".format(
+                raise EnvironmentError(
+                    "Model name '{}' was not found in tokenizers model name list ({}). "
+                    "We assumed '{}' was a path or url to a directory containing vocabulary files "
+                    "named {} but couldn't find such vocabulary files at this path or url.".format(
                         pretrained_model_name_or_path, ', '.join(s3_models),
-                        pretrained_model_name_or_path, ))
-                return None
+                        pretrained_model_name_or_path, 
+                        list(cls.vocab_files_names.values())))
 
         # Get files from url, cache, or disk depending on the case
         try:
@@ -353,17 +353,18 @@ class PreTrainedTokenizer(object):
                     resolved_vocab_files[file_id] = None
                 else:
                     resolved_vocab_files[file_id] = cached_path(file_path, cache_dir=cache_dir, force_download=force_download, proxies=proxies)
-        except EnvironmentError as e:
+        except EnvironmentError:
             if pretrained_model_name_or_path in s3_models:
-                logger.error("Couldn't reach server to download vocabulary.")
+                msg = "Couldn't reach server at '{}' to download vocabulary files."
             else:
-                logger.error(
-                    "Model name '{}' was not found in model name list ({}). "
-                    "We assumed '{}' was a path or url but couldn't find files {} "
-                    "at this path or url.".format(
+                msg = "Model name '{}' was not found in tokenizers model name list ({}). " \
+                    "We assumed '{}' was a path or url to a directory containing vocabulary files " \
+                    "named {}, but couldn't find such vocabulary files at this path or url.".format(
                         pretrained_model_name_or_path, ', '.join(s3_models),
-                        pretrained_model_name_or_path, str(vocab_files.keys())))
-            raise e
+                        pretrained_model_name_or_path,
+                        list(cls.vocab_files_names.values()))
+
+            raise EnvironmentError(msg)
 
         for file_id, file_path in vocab_files.items():
             if file_path == resolved_vocab_files[file_id]:
@@ -512,7 +513,8 @@ class PreTrainedTokenizer(object):
         for token in new_tokens:
             assert isinstance(token, str) or (six.PY2 and isinstance(token, unicode))
             if token != self.unk_token and \
-                    self.convert_tokens_to_ids(token) == self.convert_tokens_to_ids(self.unk_token):
+                    self.convert_tokens_to_ids(token) == self.convert_tokens_to_ids(self.unk_token) and \
+                    token not in to_add_tokens:
                 to_add_tokens.append(token)
                 logger.info("Adding %s to the vocabulary", token)
 
@@ -538,15 +540,9 @@ class PreTrainedTokenizer(object):
         Returns:
             Number of tokens added to sequences
         """
-
-        if pair:
-            initial_tokens_len = len(self.encode("This is a sequence") + self.encode("This is another"))
-            final_tokens_len = len(self.encode("This is a sequence", "This is another", add_special_tokens=True))
-        else:
-            initial_tokens_len = len(self.encode("This is a sequence"))
-            final_tokens_len = len(self.encode("This is a sequence", add_special_tokens=True))
-
-        return final_tokens_len - initial_tokens_len
+        token_ids_0 = []
+        token_ids_1 = []
+        return len(self.build_inputs_with_special_tokens(token_ids_0, token_ids_1 if pair else None))
 
     def add_special_tokens(self, special_tokens_dict):
         """
@@ -698,7 +694,7 @@ class PreTrainedTokenizer(object):
                 add_special_tokens=False,
                 max_length=None,
                 stride=0,
-                truncate_first_sequence=True,
+                truncation_strategy='longest_first',
                 return_tensors=None,
                 **kwargs):
         """
@@ -718,9 +714,13 @@ class PreTrainedTokenizer(object):
             max_length: if set to a number, will limit the total sequence returned so that it has a maximum length.
                 If there are overflowing tokens, those will be added to the returned dictionary
             stride: if set to a number along with max_length, the overflowing tokens returned will contain some tokens
-                from the main sequence returned. The value of this argument defined the number of additional tokens.
-            truncate_first_sequence: if there is a specified max_length, this flag will choose which sequence
-                will be truncated.
+                from the main sequence returned. The value of this argument defines the number of additional tokens.
+            truncation_strategy: string selected in the following options:
+                - 'longest_first' (default) Iteratively reduce the inputs sequence until the input is under max_length
+                    starting from the longest one at each token (when there is a pair of input sequences)
+                - 'only_first': Only truncate the first sequence
+                - 'only_second': Only truncate the second sequence
+                - 'do_not_truncate': Does not truncate (raise an error if the input sequence is longer than max_length)
             return_tensors: (optional) can be set to 'tf' or 'pt' to return respectively TensorFlow tf.constant
                 or PyTorch torch.Tensor instead of a list of python integers.
             **kwargs: passed to the `self.tokenize()` method
@@ -730,7 +730,7 @@ class PreTrainedTokenizer(object):
                                           max_length=max_length,
                                           add_special_tokens=add_special_tokens,
                                           stride=stride,
-                                          truncate_first_sequence=truncate_first_sequence,
+                                          truncation_strategy=truncation_strategy,
                                           return_tensors=return_tensors,
                                           **kwargs)
 
@@ -742,7 +742,7 @@ class PreTrainedTokenizer(object):
                     add_special_tokens=False,
                     max_length=None,
                     stride=0,
-                    truncate_first_sequence=True,
+                    truncation_strategy='longest_first',
                     return_tensors=None,
                     **kwargs):
         """
@@ -761,9 +761,13 @@ class PreTrainedTokenizer(object):
             max_length: if set to a number, will limit the total sequence returned so that it has a maximum length.
                 If there are overflowing tokens, those will be added to the returned dictionary
             stride: if set to a number along with max_length, the overflowing tokens returned will contain some tokens
-                from the main sequence returned. The value of this argument defined the number of additional tokens.
-            truncate_first_sequence: if there is a specified max_length, this flag will choose which sequence
-                will be truncated.
+                from the main sequence returned. The value of this argument defines the number of additional tokens.
+            truncation_strategy: string selected in the following options:
+                - 'longest_first' (default) Iteratively reduce the inputs sequence until the input is under max_length
+                    starting from the longest one at each token (when there is a pair of input sequences)
+                - 'only_first': Only truncate the first sequence
+                - 'only_second': Only truncate the second sequence
+                - 'do_not_truncate': Does not truncate (raise an error if the input sequence is longer than max_length)
             return_tensors: (optional) can be set to 'tf' or 'pt' to return respectively TensorFlow tf.constant
                 or PyTorch torch.Tensor instead of a list of python integers.
             **kwargs: passed to the `self.tokenize()` method
@@ -787,12 +791,11 @@ class PreTrainedTokenizer(object):
                                       max_length=max_length,
                                       add_special_tokens=add_special_tokens,
                                       stride=stride,
-                                      truncate_first_sequence=truncate_first_sequence,
+                                      truncation_strategy=truncation_strategy,
                                       return_tensors=return_tensors)
 
-
     def prepare_for_model(self, ids, pair_ids=None, max_length=None, add_special_tokens=False, stride=0,
-                          truncate_first_sequence=True, return_tensors=None):
+                          truncation_strategy='longest_first', return_tensors=None):
         """
         Prepares a sequence of input id, or a pair of sequences of inputs ids so that it can be used by the model.
         It adds special tokens, truncates
@@ -809,41 +812,50 @@ class PreTrainedTokenizer(object):
                 to their model.
             stride: window stride for overflowing tokens. Can be useful for edge effect removal when using sequential
                 list of inputs.
-            truncate_first_sequence: if set to `True` and an optional second list of input ids is provided,
-                alongside a specified `max_length`, will truncate the first sequence if the total size is superior
-                than the specified `max_length`. If set to `False`, will truncate the second sequence instead.
+            truncation_strategy: string selected in the following options:
+                - 'longest_first' (default) Iteratively reduce the inputs sequence until the input is under max_length
+                    starting from the longest one at each token (when there is a pair of input sequences)
+                - 'only_first': Only truncate the first sequence
+                - 'only_second': Only truncate the second sequence
+                - 'do_not_truncate': Does not truncate (raise an error if the input sequence is longer than max_length)
             return_tensors: (optional) can be set to 'tf' or 'pt' to return respectively TensorFlow tf.constant
                 or PyTorch torch.Tensor instead of a list of python integers.
 
         Return:
-            a dictionary containing the `input_ids` as well as the `overflowing_tokens` if a `max_length` was given.
+            A Dictionary of shape::
+
+                {
+                    input_ids: list[int],
+                    overflowing_tokens: list[int] if a ``max_length`` is specified, else None
+                    special_tokens_mask: list[int] if ``add_special_tokens`` if set to ``True``
+                }
+
+            With the fields:
+                ``input_ids``: list of tokens to be fed to a model
+
+                ``overflowing_tokens``: list of overflowing tokens if a max length is specified.
+
+                ``special_tokens_mask``: if adding special tokens, this is a list of [0, 1], with 0 specifying special added
+                tokens and 1 specifying sequence tokens.
         """
         pair = bool(pair_ids is not None)
         len_ids = len(ids)
         len_pair_ids = len(pair_ids) if pair else 0
 
         encoded_inputs = {}
-        if max_length:
-            n_added_tokens = self.num_added_tokens(pair=pair) if add_special_tokens else 0
-            if pair and n_added_tokens + (len_pair_ids if truncate_first_sequence else len_ids) >= max_length:
-                logger.warning(
-                    "You supplied a pair of sequence in which the sequence that will not be truncated is longer than the maximum specified length."
-                    "This pair of sequences will not be truncated.")
-            else:
-                if n_added_tokens + len_ids + len_pair_ids > max_length:
-                    if truncate_first_sequence or not pair:
-                        encoded_inputs["overflowing_tokens"] = ids[max_length - len_pair_ids - n_added_tokens - stride:]
-                        ids = ids[:max_length - len_pair_ids - n_added_tokens]
-                    elif not truncate_first_sequence and pair:
-                        encoded_inputs["overflowing_tokens"] = pair_ids[max_length - len_ids - n_added_tokens - stride:]
-                        pair_ids = pair_ids[:max_length - len_ids - n_added_tokens]
-                    else:
-                        logger.warning(
-                            "Cannot truncate second sequence as it is not provided. No truncation.")
+        total_len = len_ids + len_pair_ids + (self.num_added_tokens(pair=pair) if add_special_tokens else 0)
+        if max_length and total_len > max_length:
+            ids, pair_ids, overflowing_tokens = self.truncate_sequences(ids, pair_ids=pair_ids,
+                                                                        num_tokens_to_remove=total_len-max_length,
+                                                                        truncation_strategy=truncation_strategy,
+                                                                        stride=stride)
+            encoded_inputs["overflowing_tokens"] = overflowing_tokens
+            encoded_inputs["num_truncated_tokens"] = total_len - max_length
 
         if add_special_tokens:
-            sequence = self.add_special_tokens_sequence_pair(ids, pair_ids) if pair else self.add_special_tokens_single_sequence(ids)
-            token_type_ids = self.create_token_type_ids_from_sequences(ids, pair_ids) if pair else [0] * len(sequence)
+            sequence = self.build_inputs_with_special_tokens(ids, pair_ids)
+            token_type_ids = self.create_token_type_ids_from_sequences(ids, pair_ids)
+            encoded_inputs["special_tokens_mask"] = self.get_special_tokens_mask(ids, pair_ids)
         else:
             sequence = ids + pair_ids if pair else ids
             token_type_ids = [0] * len(ids) + ([1] * len(pair_ids) if pair else [])
@@ -860,20 +872,89 @@ class PreTrainedTokenizer(object):
         encoded_inputs["input_ids"] = sequence
         encoded_inputs["token_type_ids"] = token_type_ids
 
+        if max_length and len(encoded_inputs["input_ids"]) > max_length:
+            encoded_inputs["input_ids"] = encoded_inputs["input_ids"][:max_length]
+            encoded_inputs["token_type_ids"] = encoded_inputs["token_type_ids"][:max_length]
+            encoded_inputs["special_tokens_mask"] = encoded_inputs["special_tokens_mask"][:max_length]
+
         return encoded_inputs
 
-    def create_token_type_ids_from_sequences(self, token_ids_0, token_ids_1):
+    def truncate_sequences(self, ids, pair_ids=None, num_tokens_to_remove=0, truncation_strategy='longest_first', stride=0):
+        """Truncates a sequence pair in place to the maximum length.
+            truncation_strategy: string selected in the following options:
+                - 'longest_first' (default) Iteratively reduce the inputs sequence until the input is under max_length
+                    starting from the longest one at each token (when there is a pair of input sequences).
+                    Overflowing tokens only contains overflow from the first sequence.
+                - 'only_first': Only truncate the first sequence. raise an error if the first sequence is shorter or equal to than num_tokens_to_remove.
+                - 'only_second': Only truncate the second sequence
+                - 'do_not_truncate': Does not truncate (raise an error if the input sequence is longer than max_length)
+        """
+        if num_tokens_to_remove <= 0:
+            return ids, pair_ids, []
+
+        if truncation_strategy == 'longest_first':
+            overflowing_tokens = []
+            for _ in range(num_tokens_to_remove):
+                if pair_ids is None or len(ids) > len(pair_ids):
+                    overflowing_tokens = [ids[-1]] + overflowing_tokens
+                    ids = ids[:-1]
+                else:
+                    pair_ids = pair_ids[:-1]
+            window_len = min(len(ids), stride)
+            if window_len > 0:
+                overflowing_tokens = ids[-window_len:] + overflowing_tokens
+        elif truncation_strategy == 'only_first':
+            assert len(ids) > num_tokens_to_remove
+            window_len = min(len(ids), stride + num_tokens_to_remove)
+            overflowing_tokens = ids[-window_len:]
+            ids = ids[:-num_tokens_to_remove]
+        elif truncation_strategy == 'only_second':
+            assert pair_ids is not None and len(pair_ids) > num_tokens_to_remove
+            window_len = min(len(pair_ids), stride + num_tokens_to_remove)
+            overflowing_tokens = pair_ids[-window_len:]
+            pair_ids = pair_ids[:-num_tokens_to_remove]
+        elif truncation_strategy == 'do_not_truncate':
+            raise ValueError("Input sequence are too long for max_length. Please select a truncation strategy.")
+        else:
+            raise ValueError("Truncation_strategy should be selected in ['longest_first', 'only_first', 'only_second', 'do_not_truncate']")
+        return (ids, pair_ids, overflowing_tokens)
+
+    def create_token_type_ids_from_sequences(self, token_ids_0, token_ids_1=None):
         logger.warning("This tokenizer does not make use of special tokens.")
+        if token_ids_1 is None:
+            return len(token_ids_0) * [0]
         return [0] * len(token_ids_0) + [1] * len(token_ids_1)
 
-    def add_special_tokens_single_sequence(self, token_ids):
-        logger.warning("This tokenizer does not make use of special tokens. The sequence has been returned with no modification.")
-        return token_ids
-
-    def add_special_tokens_sequence_pair(self, token_ids_0, token_ids_1):
-        logger.warning("This tokenizer does not make use of special tokens. The two sequences have been concatenated.")
+    def build_inputs_with_special_tokens(self, token_ids_0, token_ids_1=None):
+        """
+        Build model inputs from a sequence or a pair of sequence for sequence classification tasks
+        by concatenating and adding special tokens.
+        A RoBERTa sequence has the following format:
+            single sequence: <s> X </s>
+            pair of sequences: <s> A </s></s> B </s>
+        """
+        logger.warning("This tokenizer does not make use of special tokens. Input is returned with no modification.")
+        if token_ids_1 is None:
+            return token_ids_0
         return token_ids_0 + token_ids_1
 
+    def get_special_tokens_mask(self, token_ids_0, token_ids_1=None, already_has_special_tokens=False):
+        """
+        Retrieves sequence ids from a token list that has no special tokens added. This method is called when adding
+        special tokens using the tokenizer ``prepare_for_model`` or ``encode_plus`` methods.
+
+        Args:
+            token_ids_0: list of ids (must not contain special tokens)
+            token_ids_1: Optional list of ids (must not contain special tokens), necessary when fetching sequence ids
+                for sequence pairs
+            already_has_special_tokens: (default False) Set to True if the token list is already formated with
+                special tokens for the model
+
+        Returns:
+            A list of integers in the range [0, 1]: 0 for a special token, 1 for a sequence token.
+        """
+        return [0] * ((len(token_ids_1) if token_ids_1 else 0) + len(token_ids_0))
+
     def convert_ids_to_tokens(self, ids, skip_special_tokens=False):
         """ Converts a single index or a sequence of indices (integers) in a token "
             (resp.) a sequence of tokens (str/unicode), using the vocabulary and added tokens.
@@ -911,6 +992,11 @@ class PreTrainedTokenizer(object):
         Converts a sequence of ids (integer) in a string, using the tokenizer and vocabulary
         with options to remove special tokens and clean up tokenization spaces.
         Similar to doing ``self.convert_tokens_to_string(self.convert_ids_to_tokens(token_ids))``.
+
+        Args:
+            token_ids: list of tokenized input ids. Can be obtained using the `encode` or `encode_plus` methods.
+            skip_special_tokens: if set to True, will replace special tokens.
+            clean_up_tokenization_spaces: if set to True, will clean up the tokenization spaces.
         """
         filtered_tokens = self.convert_ids_to_tokens(token_ids, skip_special_tokens=skip_special_tokens)