Add TF2 XLM-RoBERTa model

src/transformers/__init__.py

@@ -31,7 +31,6 @@ from .configuration_openai import OPENAI_GPT_PRETRAINED_CONFIG_ARCHIVE_MAP, Open
 from .configuration_roberta import ROBERTA_PRETRAINED_CONFIG_ARCHIVE_MAP, RobertaConfig
 from .configuration_t5 import T5_PRETRAINED_CONFIG_ARCHIVE_MAP, T5Config
 from .configuration_transfo_xl import TRANSFO_XL_PRETRAINED_CONFIG_ARCHIVE_MAP, TransfoXLConfig
-
 # Configurations
 from .configuration_utils import PretrainedConfig
 from .configuration_xlm import XLM_PRETRAINED_CONFIG_ARCHIVE_MAP, XLMConfig
@@ -56,7 +55,6 @@ from .data import (
     xnli_processors,
     xnli_tasks_num_labels,
 )
-
 # Files and general utilities
 from .file_utils import (
     CONFIG_NAME,
@@ -73,10 +71,8 @@ from .file_utils import (
     is_tf_available,
     is_torch_available,
 )
-
 # Model Cards
 from .modelcard import ModelCard
-
 # TF 2.0 <=> PyTorch conversion utilities
 from .modeling_tf_pytorch_utils import (
     convert_tf_weight_name_to_pt_weight_name,
@@ -87,7 +83,6 @@ from .modeling_tf_pytorch_utils import (
     load_tf2_model_in_pytorch_model,
     load_tf2_weights_in_pytorch_model,
 )
-
 # Pipelines
 from .pipelines import (
     CsvPipelineDataFormat,
@@ -113,7 +108,6 @@ from .tokenization_openai import OpenAIGPTTokenizer
 from .tokenization_roberta import RobertaTokenizer
 from .tokenization_t5 import T5Tokenizer
 from .tokenization_transfo_xl import TransfoXLCorpus, TransfoXLTokenizer
-
 # Tokenizers
 from .tokenization_utils import PreTrainedTokenizer
 from .tokenization_xlm import XLMTokenizer
@@ -249,6 +243,7 @@ if is_torch_available():
         XLMRobertaForMultipleChoice,
         XLMRobertaForSequenceClassification,
         XLMRobertaForTokenClassification,
+        XLM_ROBERTA_PRETRAINED_MODEL_ARCHIVE_MAP,
     )
     from .modeling_mmbt import ModalEmbeddings, MMBTModel, MMBTForClassification
 
@@ -338,6 +333,14 @@ if is_tf_available():
         TF_XLM_PRETRAINED_MODEL_ARCHIVE_MAP,
     )
 
+    from .modeling_tf_xlm_roberta import (
+        TFXLMRobertaForMaskedLM,
+        TFXLMRobertaModel,
+        TFXLMRobertaForSequenceClassification,
+        TFXLMRobertaForTokenClassification,
+        TF_XLM_ROBERTA_PRETRAINED_MODEL_ARCHIVE_MAP,
+    )
+
     from .modeling_tf_roberta import (
         TFRobertaPreTrainedModel,
         TFRobertaMainLayer,

src/transformers/convert_pytorch_checkpoint_to_tf2.py

@@ -30,6 +30,7 @@ from transformers import (
     T5_PRETRAINED_CONFIG_ARCHIVE_MAP,
     TRANSFO_XL_PRETRAINED_CONFIG_ARCHIVE_MAP,
     XLM_PRETRAINED_CONFIG_ARCHIVE_MAP,
+    XLM_ROBERTA_PRETRAINED_CONFIG_ARCHIVE_MAP,
     XLNET_PRETRAINED_CONFIG_ARCHIVE_MAP,
     AlbertConfig,
     BertConfig,
@@ -52,10 +53,13 @@ from transformers import (
     TFRobertaForSequenceClassification,
     TFT5WithLMHeadModel,
     TFTransfoXLLMHeadModel,
+    TFXLMRobertaForMaskedLM,
+    TFXLMRobertaForSequenceClassification,
     TFXLMWithLMHeadModel,
     TFXLNetLMHeadModel,
     TransfoXLConfig,
     XLMConfig,
+    XLMRobertaConfig,
     XLNetConfig,
     cached_path,
     is_torch_available,
@@ -77,6 +81,8 @@ if is_torch_available():
         XLNET_PRETRAINED_MODEL_ARCHIVE_MAP,
         XLMWithLMHeadModel,
         XLM_PRETRAINED_MODEL_ARCHIVE_MAP,
+        XLM_ROBERTA_PRETRAINED_MODEL_ARCHIVE_MAP,
+        XLMRobertaForMaskedLM,
         TransfoXLLMHeadModel,
         TRANSFO_XL_PRETRAINED_MODEL_ARCHIVE_MAP,
         OpenAIGPTLMHeadModel,
@@ -107,6 +113,8 @@ else:
         XLNET_PRETRAINED_MODEL_ARCHIVE_MAP,
         XLMWithLMHeadModel,
         XLM_PRETRAINED_MODEL_ARCHIVE_MAP,
+        XLM_ROBERTA_PRETRAINED_MODEL_ARCHIVE_MAP,
+        XLMRobertaForMaskedLM,
         TransfoXLLMHeadModel,
         TRANSFO_XL_PRETRAINED_MODEL_ARCHIVE_MAP,
         OpenAIGPTLMHeadModel,
@@ -152,6 +160,8 @@ else:
         None,
         None,
         None,
+        None,
+        None,
     )
 
 
@@ -207,6 +217,13 @@ MODEL_CLASSES = {
         XLM_PRETRAINED_MODEL_ARCHIVE_MAP,
         XLM_PRETRAINED_CONFIG_ARCHIVE_MAP,
     ),
+    "xlm-roberta": (
+        XLMRobertaConfig,
+        TFXLMRobertaForMaskedLM,
+        XLMRobertaForMaskedLM,
+        XLM_ROBERTA_PRETRAINED_MODEL_ARCHIVE_MAP,
+        XLM_ROBERTA_PRETRAINED_CONFIG_ARCHIVE_MAP,
+    ),
     "transfo-xl": (
         TransfoXLConfig,
         TFTransfoXLLMHeadModel,

src/transformers/modeling_tf_xlm_roberta.py

+# coding=utf-8
+# Copyright 2019 Facebook AI Research 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  XLM-RoBERTa model. """
+
+
+import logging
+
+from .configuration_xlm_roberta import XLMRobertaConfig
+from .file_utils import add_start_docstrings
+from .modeling_tf_roberta import (
+    TFRobertaForMaskedLM,
+    TFRobertaForSequenceClassification,
+    TFRobertaForTokenClassification,
+    TFRobertaModel,
+)
+
+
+logger = logging.getLogger(__name__)
+
+TF_XLM_ROBERTA_PRETRAINED_MODEL_ARCHIVE_MAP = {}
+
+
+XLM_ROBERTA_START_DOCSTRING = r"""    The XLM-RoBERTa model was proposed in
+    `Unsupervised Cross-lingual Representation Learning at Scale`_
+    by Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer and Veselin Stoyanov. It is based on Facebook's RoBERTa model released in 2019.
+
+    It is a large multi-lingual language model, trained on 2.5TB of filtered CommonCrawl data.
+
+    This implementation is the same as RoBERTa.
+
+    This model is a tf.keras.Model `tf.keras.Model`_ sub-class. Use it as a regular TF 2.0 Keras Model and
+    refer to the TF 2.0 documentation for all matter related to general usage and behavior.
+
+    .. _`Unsupervised Cross-lingual Representation Learning at Scale`:
+        https://arxiv.org/abs/1911.02116
+
+    .. _`tf.keras.Model`:
+        https://www.tensorflow.org/versions/r2.0/api_docs/python/tf/keras/Model
+
+    Parameters:
+        config (:class:`~transformers.XLMRobertaConfig`): Model configuration class with all the parameters of the
+            model. Initializing with a config file does not load the weights associated with the model, only the configuration.
+            Check out the :meth:`~transformers.PreTrainedModel.from_pretrained` method to load the model weights.
+"""
+
+XLM_ROBERTA_INPUTS_DOCSTRING = r"""
+    Inputs:
+        **input_ids**: ``Numpy array`` or ``tf.Tensor`` of shape ``(batch_size, sequence_length)``:
+            Indices of input sequence tokens in the vocabulary.
+            To match pre-training, XLM-RoBERTa input sequence should be formatted with <s> and </s> tokens as follows:
+
+            (a) For sequence pairs:
+
+                ``tokens:         <s> Is this Jacksonville ? </s> </s> No it is not . </s>``
+
+            (b) For single sequences:
+
+                ``tokens:         <s> the dog is hairy . </s>``
+
+            Fully encoded sequences or sequence pairs can be obtained using the XLMRobertaTokenizer.encode function with
+            the ``add_special_tokens`` parameter set to ``True``.
+
+            XLM-RoBERTa is a model with absolute position embeddings so it's usually advised to pad the inputs on
+            the right rather than the left.
+
+            See :func:`transformers.PreTrainedTokenizer.encode` and
+            :func:`transformers.PreTrainedTokenizer.convert_tokens_to_ids` for details.
+        **attention_mask**: (`optional`) ``Numpy array`` or ``tf.Tensor`` of shape ``(batch_size, sequence_length)``:
+            Mask to avoid performing attention on padding token indices.
+            Mask values selected in ``[0, 1]``:
+            ``1`` for tokens that are NOT MASKED, ``0`` for MASKED tokens.
+        **token_type_ids**: (`optional` need to be trained) ``Numpy array`` or ``tf.Tensor`` of shape ``(batch_size, sequence_length)``:
+            Optional segment token indices to indicate first and second portions of the inputs.
+            This embedding matrice is not trained (not pretrained during XLM-RoBERTa pretraining), you will have to train it
+            during finetuning.
+            Indices are selected in ``[0, 1]``: ``0`` corresponds to a `sentence A` token, ``1``
+            corresponds to a `sentence B` token
+            (see `BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding`_ for more details).
+        **position_ids**: (`optional`) ``Numpy array`` or ``tf.Tensor`` of shape ``(batch_size, sequence_length)``:
+            Indices of positions of each input sequence tokens in the position embeddings.
+            Selected in the range ``[0, config.max_position_embeddings - 1[``.
+        **head_mask**: (`optional`) ``Numpy array`` or ``tf.Tensor`` of shape ``(num_heads,)`` or ``(num_layers, num_heads)``:
+            Mask to nullify selected heads of the self-attention modules.
+            Mask values selected in ``[0, 1]``:
+            ``1`` indicates the head is **not masked**, ``0`` indicates the head is **masked**.
+        **inputs_embeds**: (`optional`) ``Numpy array`` or ``tf.Tensor`` of shape ``(batch_size, sequence_length, embedding_dim)``:
+            Optionally, instead of passing ``input_ids`` you can choose to directly pass an embedded representation.
+            This is useful if you want more control over how to convert `input_ids` indices into associated vectors
+            than the model's internal embedding lookup matrix.
+"""
+
+
+@add_start_docstrings(
+    "The bare XLM-RoBERTa Model transformer outputting raw hidden-states without any specific head on top.",
+    XLM_ROBERTA_START_DOCSTRING,
+    XLM_ROBERTA_INPUTS_DOCSTRING,
+)
+class TFXLMRobertaModel(TFRobertaModel):
+    r"""
+    Outputs: `Tuple` comprising various elements depending on the configuration (config) and inputs:
+        **last_hidden_state**: ``Numpy array`` or ``tf.Tensor`` of shape ``(batch_size, sequence_length, hidden_size)``
+            Sequence of hidden-states at the output of the last layer of the model.
+        **pooler_output**: ``Numpy array`` or ``tf.Tensor`` of shape ``(batch_size, hidden_size)``
+            Last layer hidden-state of the first token of the sequence (classification token)
+            further processed by a Linear layer and a Tanh activation function. The Linear
+            layer weights are trained from the next sentence prediction (classification)
+            eo match pre-training, XLM-RoBERTa input sequence should be formatted with <s> and </s> tokens as follows:
+
+            (a) For sequence pairs:
+
+                ``tokens:         <s> is this jack ##son ##ville ? </s> </s> no it is not . </s>``
+
+                ``token_type_ids:   0   0  0    0    0     0       0   0   0     1  1  1  1   1   1``
+
+            (b) For single sequences:
+
+                ``tokens:         <s> the dog is hairy . </s>``
+
+                ``token_type_ids:   0   0   0   0  0     0   0``
+
+            objective during Bert pretraining. This output is usually *not* a good summary
+            of the semantic content of the input, you're often better with averaging or pooling
+            the sequence of hidden-states for the whole input sequence.
+        **hidden_states**: (`optional`, returned when ``config.output_hidden_states=True``)
+            list of ``Numpy array`` or ``tf.Tensor`` (one for the output of each layer + the output of the embeddings)
+            of shape ``(batch_size, sequence_length, hidden_size)``:
+            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
+        **attentions**: (`optional`, returned when ``config.output_attentions=True``)
+            list of ``Numpy array`` or ``tf.Tensor`` (one for each layer) of shape ``(batch_size, num_heads, sequence_length, sequence_length)``:
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
+
+    Examples::
+
+        tokenizer = XLMRobertaTokenizer.from_pretrained('xlm-roberta-large')
+        model = TFXLMRobertaModel.from_pretrained('xlm-roberta-large')
+        input_ids = tf.constant(tokenizer.encode("Schloß Nymphenburg ist sehr schön ."))[None, :]  # Batch size 1
+        outputs = model(input_ids)
+        last_hidden_states = outputs[0]  # The last hidden-state is the first element of the output tuple
+
+    """
+    config_class = XLMRobertaConfig
+    pretrained_model_archive_map = TF_XLM_ROBERTA_PRETRAINED_MODEL_ARCHIVE_MAP
+
+
+@add_start_docstrings(
+    """XLM-RoBERTa Model with a `language modeling` head on top. """,
+    XLM_ROBERTA_START_DOCSTRING,
+    XLM_ROBERTA_INPUTS_DOCSTRING,
+)
+class TFXLMRobertaForMaskedLM(TFRobertaForMaskedLM):
+    r"""
+        **masked_lm_labels**: (`optional`) ``Numpy array`` or ``tf.Tensor`` of shape ``(batch_size, sequence_length)``:
+            Labels for computing the masked language modeling loss.
+            Indices should be in ``[-1, 0, ..., config.vocab_size]`` (see ``input_ids`` docstring)
+            Tokens with indices set to ``-1`` are ignored (masked), the loss is only computed for the tokens with labels
+            in ``[0, ..., config.vocab_size]``
+
+    Outputs: `Tuple` comprising various elements depending on the configuration (config) and inputs:
+        **loss**: (`optional`, returned when ``masked_lm_labels`` is provided) ``Numpy array`` or ``tf.Tensor`` of shape ``(1,)``:
+            Masked language modeling loss.
+        **prediction_scores**: ``Numpy array`` or ``tf.Tensor`` of shape ``(batch_size, sequence_length, config.vocab_size)``
+            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
+        **hidden_states**: (`optional`, returned when ``config.output_hidden_states=True``)
+            list of ``Numpy array`` or ``tf.Tensor`` (one for the output of each layer + the output of the embeddings)
+            of shape ``(batch_size, sequence_length, hidden_size)``:
+            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
+        **attentions**: (`optional`, returned when ``config.output_attentions=True``)
+            list of ``Numpy array`` or ``tf.Tensor`` (one for each layer) of shape ``(batch_size, num_heads, sequence_length, sequence_length)``:
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
+
+    Examples::
+
+        tokenizer = XLMRobertaTokenizer.from_pretrained('xlm-roberta-large')
+        model = TFXLMRobertaForMaskedLM.from_pretrained('xlm-roberta-large')
+        input_ids = tf.constant(tokenizer.encode("Schloß Nymphenburg ist sehr schön ."))[None, :]  # Batch size 1
+        outputs = model(input_ids)
+        loss, prediction_scores = outputs[:2]
+
+    """
+    config_class = XLMRobertaConfig
+    pretrained_model_archive_map = TF_XLM_ROBERTA_PRETRAINED_MODEL_ARCHIVE_MAP
+
+
+@add_start_docstrings(
+    """XLM-RoBERTa Model transformer with a sequence classification/regression head on top (a linear layer
+    on top of the pooled output) e.g. for GLUE tasks. """,
+    XLM_ROBERTA_START_DOCSTRING,
+    XLM_ROBERTA_INPUTS_DOCSTRING,
+)
+class TFXLMRobertaForSequenceClassification(TFRobertaForSequenceClassification):
+    r"""
+        **labels**: (`optional`) ``Numpy array`` or ``tf.Tensor`` of shape ``(batch_size,)``:
+            Labels for computing the sequence classification/regression loss.
+            Indices should be in ``[0, ..., config.num_labels]``.
+            If ``config.num_labels == 1`` a regression loss is computed (Mean-Square loss),
+            If ``config.num_labels > 1`` a classification loss is computed (Cross-Entropy).
+
+    Outputs: `Tuple` comprising various elements depending on the configuration (config) and inputs:
+        **loss**: (`optional`, returned when ``labels`` is provided) ``Numpy array`` or ``tf.Tensor`` of shape ``(1,)``:
+            Classification (or regression if config.num_labels==1) loss.
+        **logits**: ``Numpy array`` or ``tf.Tensor`` of shape ``(batch_size, config.num_labels)``
+            Classification (or regression if config.num_labels==1) scores (before SoftMax).
+        **hidden_states**: (`optional`, returned when ``config.output_hidden_states=True``)
+            list of ``Numpy array`` or ``tf.Tensor`` (one for the output of each layer + the output of the embeddings)
+            of shape ``(batch_size, sequence_length, hidden_size)``:
+            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
+        **attentions**: (`optional`, returned when ``config.output_attentions=True``)
+            list of ``Numpy array`` or ``tf.Tensor`` (one for each layer) of shape ``(batch_size, num_heads, sequence_length, sequence_length)``:
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
+
+    Examples::
+
+        tokenizer = XLMRobertaTokenizer.from_pretrained('xlm-roberta-large')
+        model = TFXLMRobertaForSequenceClassification.from_pretrained('xlm-roberta-large')
+        input_ids = tf.constant(tokenizer.encode("Schloß Nymphenburg ist sehr schön ."))[None, :]  # Batch size 1
+        outputs = model(input_ids)
+        loss, logits = outputs[:2]
+
+    """
+    config_class = XLMRobertaConfig
+    pretrained_model_archive_map = TF_XLM_ROBERTA_PRETRAINED_MODEL_ARCHIVE_MAP
+
+
+@add_start_docstrings(
+    """XLM-RoBERTa Model with a multiple choice classification head on top (a linear layer on top of
+    the pooled output and a softmax) e.g. for RocStories/SWAG tasks. """,
+    XLM_ROBERTA_START_DOCSTRING,
+    XLM_ROBERTA_INPUTS_DOCSTRING,
+)
+@add_start_docstrings(
+    """XLM-RoBERTa Model with a token classification head on top (a linear layer on top of
+    the hidden-states output) e.g. for Named-Entity-Recognition (NER) tasks. """,
+    XLM_ROBERTA_START_DOCSTRING,
+    XLM_ROBERTA_INPUTS_DOCSTRING,
+)
+class TFXLMRobertaForTokenClassification(TFRobertaForTokenClassification):
+    r"""
+        **labels**: (`optional`) ``Numpy array`` or ``tf.Tensor`` of shape ``(batch_size, sequence_length)``:
+            Labels for computing the token classification loss.
+            Indices should be in ``[0, ..., config.num_labels - 1]``.
+
+    Outputs: `Tuple` comprising various elements depending on the configuration (config) and inputs:
+        **loss**: (`optional`, returned when ``labels`` is provided) ``Numpy array`` or ``tf.Tensor`` of shape ``(1,)``:
+            Classification loss.
+        **scores**: ``Numpy array`` or ``tf.Tensor`` of shape ``(batch_size, sequence_length, config.num_labels)``
+            Classification scores (before SoftMax).
+        **hidden_states**: (`optional`, returned when ``config.output_hidden_states=True``)
+            list of ``Numpy array`` or ``tf.Tensor`` (one for the output of each layer + the output of the embeddings)
+            of shape ``(batch_size, sequence_length, hidden_size)``:
+            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
+        **attentions**: (`optional`, returned when ``config.output_attentions=True``)
+            list of ``Numpy array`` or ``tf.Tensor`` (one for each layer) of shape ``(batch_size, num_heads, sequence_length, sequence_length)``:
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
+
+    Examples::
+
+        tokenizer = XLMRobertaTokenizer.from_pretrained('xlm-roberta-large')
+        model = TFXLMRobertaForTokenClassification.from_pretrained('xlm-roberta-large')
+        input_ids = tf.constant(tokenizer.encode("Schloß Nymphenburg ist sehr schön .", add_special_tokens=True))[None, :]  # Batch size 1
+        outputs = model(input_ids)
+        loss, scores = outputs[:2]
+
+    """
+    config_class = XLMRobertaConfig
+    pretrained_model_archive_map = TF_XLM_ROBERTA_PRETRAINED_MODEL_ARCHIVE_MAP

templates/adding_a_new_example_script/run_xxx.py

@@ -52,7 +52,6 @@ from utils_squad import (
     write_predictions,
     write_predictions_extended,
 )
-
 # The follwing import is the official SQuAD evaluation script (2.0).
 # You can remove it from the dependencies if you are using this script outside of the library
 # We've added it here for automated tests (see examples/test_examples.py file)

templates/adding_a_new_example_script/utils_xxx.py

@@ -21,7 +21,6 @@ import logging
 import math
 
 from transformers.tokenization_bert import BasicTokenizer, whitespace_tokenize
-
 # Required by XLNet evaluation method to compute optimal threshold (see write_predictions_extended() method)
 from utils_squad_evaluate import find_all_best_thresh_v2, get_raw_scores, make_qid_to_has_ans