ELECTRA (#3257)

* Electra wip

* helpers

* Electra wip

* Electra v1

* ELECTRA may be saved/loaded

* Generator & Discriminator

* Embedding size instead of halving the hidden size

* ELECTRA Tokenizer

* Revert BERT helpers

* ELECTRA Conversion script

* Archive maps

* PyTorch tests

* Start fixing tests

* Tests pass

* Same configuration for both models

* Compatible with base + large

* Simplification + weight tying

* Archives

* Auto + Renaming to standard names

* ELECTRA is uncased

* Tests

* Slight API changes

* Update tests

* wip

* ElectraForTokenClassification

* temp

* Simpler arch + tests

Removed ElectraForPreTraining which will be in a script

* Conversion script

* Auto model

* Update links to S3

* Split ElectraForPreTraining and ElectraForTokenClassification

* Actually test PreTraining model

* Remove num_labels from configuration

* wip

* wip

* From discriminator and generator to electra

* Slight API changes

* Better naming

* TensorFlow ELECTRA tests

* Accurate conversion script

* Added to conversion script

* Fast ELECTRA tokenizer

* Style

* Add ELECTRA to README

* Modeling Pytorch Doc + Real style

* TF Docs

* Docs

* Correct links

* Correct model intialized

* random fixes

* style

* Addressing Patrick's and Sam's comments

* Correct links in docs

README.md

@@ -164,8 +164,9 @@ At some point in the future, you'll be able to seamlessly move from pre-training
 14. **[MMBT](https://github.com/facebookresearch/mmbt/)** (from Facebook), released together with the paper a [Supervised Multimodal Bitransformers for Classifying Images and Text](https://arxiv.org/pdf/1909.02950.pdf) by Douwe Kiela, Suvrat Bhooshan, Hamed Firooz, Davide Testuggine.
 15. **[FlauBERT](https://github.com/getalp/Flaubert)** (from CNRS) released with the paper [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) by Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab.
 16. **[BART](https://github.com/pytorch/fairseq/tree/master/examples/bart)** (from Facebook) released with the paper [BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension](https://arxiv.org/pdf/1910.13461.pdf) by Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov and Luke Zettlemoyer.
-17. **[Other community models](https://huggingface.co/models)**, contributed by the [community](https://huggingface.co/users).
-18. Want to contribute a new model? We have added a **detailed guide and templates** to guide you in the process of adding a new model. You can find them in the [`templates`](./templates) folder of the repository. Be sure to check the [contributing guidelines](./CONTRIBUTING.md) and contact the maintainers or open an issue to collect feedbacks before starting your PR.
+17. **[ELECTRA](https://github.com/google-research/electra)** (from Google Research/Stanford University) released with the paper [ELECTRA: Pre-training text encoders as discriminators rather than generators](https://arxiv.org/abs/2003.10555) by Kevin Clark, Minh-Thang Luong, Quoc V. Le, Christopher D. Manning.
+18. **[Other community models](https://huggingface.co/models)**, contributed by the [community](https://huggingface.co/users).
+19. Want to contribute a new model? We have added a **detailed guide and templates** to guide you in the process of adding a new model. You can find them in the [`templates`](./templates) folder of the repository. Be sure to check the [contributing guidelines](./CONTRIBUTING.md) and contact the maintainers or open an issue to collect feedbacks before starting your PR.
 
 These implementations have been tested on several datasets (see the example scripts) and should match the performances of the original implementations (e.g. ~93 F1 on SQuAD for BERT Whole-Word-Masking, ~88 F1 on RocStories for OpenAI GPT, ~18.3 perplexity on WikiText 103 for Transformer-XL, ~0.916 Peason R coefficient on STS-B for XLNet). You can find more details on the performances in the Examples section of the [documentation](https://huggingface.co/transformers/examples.html).
 

docs/source/index.rst

@@ -104,3 +104,4 @@ The library currently contains PyTorch and Tensorflow implementations, pre-train
     model_doc/flaubert
     model_doc/bart
     model_doc/t5
+    model_doc/electra
\ No newline at end of file

docs/source/model_doc/electra.rst

+ELECTRA
+----------------------------------------------------
+
+The ELECTRA model was proposed in the paper.
+`ELECTRA: Pre-training Text Encoders as Discriminators Rather Than Generators <https://openreview.net/pdf?id=r1xMH1BtvB>`__.
+ELECTRA is a new pre-training approach which trains two transformer models: the generator and the discriminator. The
+generator's role is to replace tokens in a sequence, and is therefore trained as a masked language model. The discriminator,
+which is the model we're interested in, tries to identify which tokens were replaced by the generator in the sequence.
+
+The abstract from the paper is the following:
+
+*Masked language modeling (MLM) pre-training methods such as BERT corrupt
+the input by replacing some tokens with [MASK] and then train a model to
+reconstruct the original tokens. While they produce good results when transferred
+to downstream NLP tasks, they generally require large amounts of compute to be
+effective. As an alternative, we propose a more sample-efficient pre-training task
+called replaced token detection. Instead of masking the input, our approach
+corrupts it by replacing some tokens with plausible alternatives sampled from a small
+generator network. Then, instead of training a model that predicts the original
+identities of the corrupted tokens, we train a discriminative model that predicts
+whether each token in the corrupted input was replaced by a generator sample
+or not. Thorough experiments demonstrate this new pre-training task is more
+efficient than MLM because the task is defined over all input tokens rather than
+just the small subset that was masked out. As a result, the contextual representations
+learned by our approach substantially outperform the ones learned by BERT
+given the same model size, data, and compute. The gains are particularly strong
+for small models; for example, we train a model on one GPU for 4 days that
+outperforms GPT (trained using 30x more compute) on the GLUE natural language
+understanding benchmark. Our approach also works well at scale, where it
+performs comparably to RoBERTa and XLNet while using less than 1/4 of their
+compute and outperforms them when using the same amount of compute.*
+
+Tips:
+
+- ELECTRA is the pre-training approach, therefore there is nearly no changes done to the underlying model: BERT. The
+  only change is the separation of the embedding size and the hidden size -> The embedding size is generally smaller,
+  while the hidden size is larger. An additional projection layer (linear) is used to project the embeddings from
+  their embedding size to the hidden size. In the case where the embedding size is the same as the hidden size, no
+  projection layer is used.
+- The ELECTRA checkpoints saved using `Google Research's implementation <https://github.com/google-research/electra>`__
+  contain both the generator and discriminator. The conversion script requires the user to name which model to export
+  into the correct architecture. Once converted to the HuggingFace format, these checkpoints may be loaded into all
+  available ELECTRA models, however. This means that the discriminator may be loaded in the `ElectraForMaskedLM` model,
+  and the generator may be loaded in the `ElectraForPreTraining` model (the classification head will be randomly
+  initialized as it doesn't exist in the generator).
+
+
+ElectraConfig
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.ElectraConfig
+    :members:
+
+
+ElectraTokenizer
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.ElectraTokenizer
+    :members:
+
+
+ElectraModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.ElectraModel
+    :members:
+
+
+ElectraForPreTraining
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.ElectraForPreTraining
+    :members:
+
+
+ElectraForMaskedLM
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.ElectraForMaskedLM
+    :members:
+
+
+ElectraForTokenClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.ElectraForTokenClassification
+    :members:
+
+
+TFElectraModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFElectraModel
+    :members:
+
+
+TFElectraForPreTraining
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFElectraForPreTraining
+    :members:
+
+
+TFElectraForMaskedLM
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFElectraForMaskedLM
+    :members:
+
+
+TFElectraForTokenClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFElectraForTokenClassification
+    :members:

src/transformers/__init__.py

@@ -38,6 +38,7 @@ from .configuration_bert import BERT_PRETRAINED_CONFIG_ARCHIVE_MAP, BertConfig
 from .configuration_camembert import CAMEMBERT_PRETRAINED_CONFIG_ARCHIVE_MAP, CamembertConfig
 from .configuration_ctrl import CTRL_PRETRAINED_CONFIG_ARCHIVE_MAP, CTRLConfig
 from .configuration_distilbert import DISTILBERT_PRETRAINED_CONFIG_ARCHIVE_MAP, DistilBertConfig
+from .configuration_electra import ELECTRA_PRETRAINED_CONFIG_ARCHIVE_MAP, ElectraConfig
 from .configuration_flaubert import FLAUBERT_PRETRAINED_CONFIG_ARCHIVE_MAP, FlaubertConfig
 from .configuration_gpt2 import GPT2_PRETRAINED_CONFIG_ARCHIVE_MAP, GPT2Config
 from .configuration_mmbt import MMBTConfig
@@ -127,6 +128,7 @@ from .tokenization_bert_japanese import BertJapaneseTokenizer, CharacterTokenize
 from .tokenization_camembert import CamembertTokenizer
 from .tokenization_ctrl import CTRLTokenizer
 from .tokenization_distilbert import DistilBertTokenizer, DistilBertTokenizerFast
+from .tokenization_electra import ElectraTokenizer, ElectraTokenizerFast
 from .tokenization_flaubert import FlaubertTokenizer
 from .tokenization_gpt2 import GPT2Tokenizer, GPT2TokenizerFast
 from .tokenization_openai import OpenAIGPTTokenizer, OpenAIGPTTokenizerFast
@@ -297,6 +299,15 @@ if is_torch_available():
         FLAUBERT_PRETRAINED_MODEL_ARCHIVE_MAP,
     )
 
+    from .modeling_electra import (
+        ElectraForPreTraining,
+        ElectraForMaskedLM,
+        ElectraForTokenClassification,
+        ElectraModel,
+        load_tf_weights_in_electra,
+        ELECTRA_PRETRAINED_MODEL_ARCHIVE_MAP,
+    )
+
     # Optimization
     from .optimization import (
         AdamW,
@@ -463,6 +474,15 @@ if is_tf_available():
         TF_T5_PRETRAINED_MODEL_ARCHIVE_MAP,
     )
 
+    from .modeling_tf_electra import (
+        TFElectraPreTrainedModel,
+        TFElectraModel,
+        TFElectraForPreTraining,
+        TFElectraForMaskedLM,
+        TFElectraForTokenClassification,
+        TF_ELECTRA_PRETRAINED_MODEL_ARCHIVE_MAP,
+    )
+
     # Optimization
     from .optimization_tf import WarmUp, create_optimizer, AdamWeightDecay, GradientAccumulator
 

src/transformers/configuration_auto.py

@@ -24,6 +24,7 @@ from .configuration_bert import BERT_PRETRAINED_CONFIG_ARCHIVE_MAP, BertConfig
 from .configuration_camembert import CAMEMBERT_PRETRAINED_CONFIG_ARCHIVE_MAP, CamembertConfig
 from .configuration_ctrl import CTRL_PRETRAINED_CONFIG_ARCHIVE_MAP, CTRLConfig
 from .configuration_distilbert import DISTILBERT_PRETRAINED_CONFIG_ARCHIVE_MAP, DistilBertConfig
+from .configuration_electra import ELECTRA_PRETRAINED_CONFIG_ARCHIVE_MAP, ElectraConfig
 from .configuration_flaubert import FLAUBERT_PRETRAINED_CONFIG_ARCHIVE_MAP, FlaubertConfig
 from .configuration_gpt2 import GPT2_PRETRAINED_CONFIG_ARCHIVE_MAP, GPT2Config
 from .configuration_openai import OPENAI_GPT_PRETRAINED_CONFIG_ARCHIVE_MAP, OpenAIGPTConfig
@@ -57,6 +58,7 @@ ALL_PRETRAINED_CONFIG_ARCHIVE_MAP = dict(
         T5_PRETRAINED_CONFIG_ARCHIVE_MAP,
         XLM_ROBERTA_PRETRAINED_CONFIG_ARCHIVE_MAP,
         FLAUBERT_PRETRAINED_CONFIG_ARCHIVE_MAP,
+        ELECTRA_PRETRAINED_CONFIG_ARCHIVE_MAP,
     ]
     for key, value, in pretrained_map.items()
 )
@@ -79,6 +81,7 @@ CONFIG_MAPPING = OrderedDict(
         ("xlnet", XLNetConfig,),
         ("xlm", XLMConfig,),
         ("ctrl", CTRLConfig,),
+        ("electra", ElectraConfig,),
     ]
 )
 
@@ -133,6 +136,7 @@ class AutoConfig:
             - contains `xlm`: :class:`~transformers.XLMConfig` (XLM model)
             - contains `ctrl` : :class:`~transformers.CTRLConfig` (CTRL model)
             - contains `flaubert` : :class:`~transformers.FlaubertConfig` (Flaubert model)
+            - contains `electra` : :class:`~transformers.ElectraConfig` (ELECTRA model)
 
 
         Args:

src/transformers/configuration_electra.py

+# coding=utf-8
+# Copyright 2018 The Google AI Language Team Authors and The HuggingFace Inc. team.
+# Copyright (c) 2018, NVIDIA CORPORATION.  All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+""" ELECTRA model configuration """
+
+
+import logging
+
+from .configuration_utils import PretrainedConfig
+
+
+logger = logging.getLogger(__name__)
+
+ELECTRA_PRETRAINED_CONFIG_ARCHIVE_MAP = {
+    "google/electra-small-generator": "https://s3.amazonaws.com/models.huggingface.co/bert/google/electra-small-generator/config.json",
+    "google/electra-base-generator": "https://s3.amazonaws.com/models.huggingface.co/bert/google/electra-base-generator/config.json",
+    "google/electra-large-generator": "https://s3.amazonaws.com/models.huggingface.co/bert/google/electra-large-generator/config.json",
+    "google/electra-small-discriminator": "https://s3.amazonaws.com/models.huggingface.co/bert/google/electra-small-discriminator/config.json",
+    "google/electra-base-discriminator": "https://s3.amazonaws.com/models.huggingface.co/bert/google/electra-base-discriminator/config.json",
+    "google/electra-large-discriminator": "https://s3.amazonaws.com/models.huggingface.co/bert/google/electra-large-discriminator/config.json",
+}
+
+
+class ElectraConfig(PretrainedConfig):
+    r"""
+        This is the configuration class to store the configuration of a :class:`~transformers.ElectraModel`.
+        It is used to instantiate an ELECTRA model according to the specified arguments, defining the model
+        architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of
+        the ELECTRA `google/electra-small-discriminator <https://huggingface.co/google/electra-small-discriminator>`__
+        architecture.
+
+        Configuration objects inherit from  :class:`~transformers.PretrainedConfig` and can be used
+        to control the model outputs. Read the documentation from  :class:`~transformers.PretrainedConfig`
+        for more information.
+
+
+        Args:
+            vocab_size (:obj:`int`, optional, defaults to 30522):
+                Vocabulary size of the ELECTRA model. Defines the different tokens that
+                can be represented by the `inputs_ids` passed to the forward method of :class:`~transformers.ElectraModel`.
+            embedding_size (:obj:`int`, optional, defaults to 128):
+                Dimensionality of the encoder layers and the pooler layer.
+            hidden_size (:obj:`int`, optional, defaults to 256):
+                Dimensionality of the encoder layers and the pooler layer.
+            num_hidden_layers (:obj:`int`, optional, defaults to 12):
+                Number of hidden layers in the Transformer encoder.
+            num_attention_heads (:obj:`int`, optional, defaults to 4):
+                Number of attention heads for each attention layer in the Transformer encoder.
+            intermediate_size (:obj:`int`, optional, defaults to 1024):
+                Dimensionality of the "intermediate" (i.e., feed-forward) layer in the Transformer encoder.
+            hidden_act (:obj:`str` or :obj:`function`, optional, defaults to "gelu"):
+                The non-linear activation function (function or string) in the encoder and pooler.
+                If string, "gelu", "relu", "swish" and "gelu_new" are supported.
+            hidden_dropout_prob (:obj:`float`, optional, defaults to 0.1):
+                The dropout probabilitiy for all fully connected layers in the embeddings, encoder, and pooler.
+            attention_probs_dropout_prob (:obj:`float`, optional, defaults to 0.1):
+                The dropout ratio for the attention probabilities.
+            max_position_embeddings (:obj:`int`, optional, defaults to 512):
+                The maximum sequence length that this model might ever be used with.
+                Typically set this to something large just in case (e.g., 512 or 1024 or 2048).
+            type_vocab_size (:obj:`int`, optional, defaults to 2):
+                The vocabulary size of the `token_type_ids` passed into :class:`~transformers.ElectraModel`.
+            initializer_range (:obj:`float`, optional, defaults to 0.02):
+                The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+            layer_norm_eps (:obj:`float`, optional, defaults to 1e-12):
+                The epsilon used by the layer normalization layers.
+
+        Example::
+
+            from transformers import ElectraModel, ElectraConfig
+
+            # Initializing a ELECTRA electra-base-uncased style configuration
+            configuration = ElectraConfig()
+
+            # Initializing a model from the electra-base-uncased style configuration
+            model = ElectraModel(configuration)
+
+            # Accessing the model configuration
+            configuration = model.config
+
+        Attributes:
+            pretrained_config_archive_map (Dict[str, str]):
+                A dictionary containing all the available pre-trained checkpoints.
+    """
+    pretrained_config_archive_map = ELECTRA_PRETRAINED_CONFIG_ARCHIVE_MAP
+    model_type = "electra"
+
+    def __init__(
+        self,
+        vocab_size=30522,
+        embedding_size=128,
+        hidden_size=256,
+        num_hidden_layers=12,
+        num_attention_heads=4,
+        intermediate_size=1024,
+        hidden_act="gelu",
+        hidden_dropout_prob=0.1,
+        attention_probs_dropout_prob=0.1,
+        max_position_embeddings=512,
+        type_vocab_size=2,
+        initializer_range=0.02,
+        layer_norm_eps=1e-12,
+        pad_token_id=0,
+        **kwargs
+    ):
+        super().__init__(pad_token_id=pad_token_id, **kwargs)
+
+        self.vocab_size = vocab_size
+        self.embedding_size = embedding_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.initializer_range = initializer_range
+        self.layer_norm_eps = layer_norm_eps

src/transformers/convert_electra_original_tf_checkpoint_to_pytorch.py

+# coding=utf-8
+# Copyright 2018 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.
+"""Convert ELECTRA checkpoint."""
+
+
+import argparse
+import logging
+
+import torch
+
+from transformers import ElectraConfig, ElectraForMaskedLM, ElectraForPreTraining, load_tf_weights_in_electra
+
+
+logging.basicConfig(level=logging.INFO)
+
+
+def convert_tf_checkpoint_to_pytorch(tf_checkpoint_path, config_file, pytorch_dump_path, discriminator_or_generator):
+    # Initialise PyTorch model
+    config = ElectraConfig.from_json_file(config_file)
+    print("Building PyTorch model from configuration: {}".format(str(config)))
+
+    if discriminator_or_generator == "discriminator":
+        model = ElectraForPreTraining(config)
+    elif discriminator_or_generator == "generator":
+        model = ElectraForMaskedLM(config)
+    else:
+        raise ValueError("The discriminator_or_generator argument should be either 'discriminator' or 'generator'")
+
+    # Load weights from tf checkpoint
+    load_tf_weights_in_electra(
+        model, config, tf_checkpoint_path, discriminator_or_generator=discriminator_or_generator
+    )
+
+    # Save pytorch-model
+    print("Save PyTorch model to {}".format(pytorch_dump_path))
+    torch.save(model.state_dict(), pytorch_dump_path)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    # Required parameters
+    parser.add_argument(
+        "--tf_checkpoint_path", default=None, type=str, required=True, help="Path to the TensorFlow checkpoint path."
+    )
+    parser.add_argument(
+        "--config_file",
+        default=None,
+        type=str,
+        required=True,
+        help="The config json file corresponding to the pre-trained model. \n"
+        "This specifies the model architecture.",
+    )
+    parser.add_argument(
+        "--pytorch_dump_path", default=None, type=str, required=True, help="Path to the output PyTorch model."
+    )
+    parser.add_argument(
+        "--discriminator_or_generator",
+        default=None,
+        type=str,
+        required=True,
+        help="Whether to export the generator or the discriminator. Should be a string, either 'discriminator' or "
+        "'generator'.",
+    )
+    args = parser.parse_args()
+    convert_tf_checkpoint_to_pytorch(
+        args.tf_checkpoint_path, args.config_file, args.pytorch_dump_path, args.discriminator_or_generator
+    )

src/transformers/convert_pytorch_checkpoint_to_tf2.py

@@ -25,6 +25,7 @@ from transformers import (
     CAMEMBERT_PRETRAINED_CONFIG_ARCHIVE_MAP,
     CTRL_PRETRAINED_CONFIG_ARCHIVE_MAP,
     DISTILBERT_PRETRAINED_CONFIG_ARCHIVE_MAP,
+    ELECTRA_PRETRAINED_CONFIG_ARCHIVE_MAP,
     FLAUBERT_PRETRAINED_CONFIG_ARCHIVE_MAP,
     GPT2_PRETRAINED_CONFIG_ARCHIVE_MAP,
     OPENAI_GPT_PRETRAINED_CONFIG_ARCHIVE_MAP,
@@ -39,6 +40,7 @@ from transformers import (
     CamembertConfig,
     CTRLConfig,
     DistilBertConfig,
+    ElectraConfig,
     FlaubertConfig,
     GPT2Config,
     OpenAIGPTConfig,
@@ -52,6 +54,7 @@ from transformers import (
     TFCTRLLMHeadModel,
     TFDistilBertForMaskedLM,
     TFDistilBertForQuestionAnswering,
+    TFElectraForPreTraining,
     TFFlaubertWithLMHeadModel,
     TFGPT2LMHeadModel,
     TFOpenAIGPTLMHeadModel,
@@ -110,6 +113,8 @@ if is_torch_available():
         ALBERT_PRETRAINED_MODEL_ARCHIVE_MAP,
         T5ForConditionalGeneration,
         T5_PRETRAINED_MODEL_ARCHIVE_MAP,
+        ElectraForPreTraining,
+        ELECTRA_PRETRAINED_MODEL_ARCHIVE_MAP,
     )
 else:
     (
@@ -147,6 +152,8 @@ else:
         ALBERT_PRETRAINED_MODEL_ARCHIVE_MAP,
         T5ForConditionalGeneration,
         T5_PRETRAINED_MODEL_ARCHIVE_MAP,
+        ElectraForPreTraining,
+        ELECTRA_PRETRAINED_MODEL_ARCHIVE_MAP,
     ) = (
         None,
         None,
@@ -182,6 +189,8 @@ else:
         None,
         None,
         None,
+        None,
+        None,
     )
 
 
@@ -321,6 +330,13 @@ MODEL_CLASSES = {
         T5_PRETRAINED_MODEL_ARCHIVE_MAP,
         T5_PRETRAINED_CONFIG_ARCHIVE_MAP,
     ),
+    "electra": (
+        ElectraConfig,
+        TFElectraForPreTraining,
+        ElectraForPreTraining,
+        ELECTRA_PRETRAINED_MODEL_ARCHIVE_MAP,
+        ELECTRA_PRETRAINED_CONFIG_ARCHIVE_MAP,
+    ),
 }
 
 

src/transformers/modeling_auto.py

@@ -26,6 +26,7 @@ from .configuration_auto import (
     CamembertConfig,
     CTRLConfig,
     DistilBertConfig,
+    ElectraConfig,
     FlaubertConfig,
     GPT2Config,
     OpenAIGPTConfig,
@@ -76,6 +77,13 @@ from .modeling_distilbert import (
     DistilBertForTokenClassification,
     DistilBertModel,
 )
+from .modeling_electra import (
+    ELECTRA_PRETRAINED_MODEL_ARCHIVE_MAP,
+    ElectraForMaskedLM,
+    ElectraForPreTraining,
+    ElectraForTokenClassification,
+    ElectraModel,
+)
 from .modeling_flaubert import (
     FLAUBERT_PRETRAINED_MODEL_ARCHIVE_MAP,
     FlaubertForQuestionAnsweringSimple,
@@ -141,6 +149,7 @@ ALL_PRETRAINED_MODEL_ARCHIVE_MAP = dict(
         T5_PRETRAINED_MODEL_ARCHIVE_MAP,
         FLAUBERT_PRETRAINED_MODEL_ARCHIVE_MAP,
         XLM_ROBERTA_PRETRAINED_MODEL_ARCHIVE_MAP,
+        ELECTRA_PRETRAINED_MODEL_ARCHIVE_MAP,
     ]
     for key, value, in pretrained_map.items()
 )
@@ -162,6 +171,7 @@ MODEL_MAPPING = OrderedDict(
         (FlaubertConfig, FlaubertModel),
         (XLMConfig, XLMModel),
         (CTRLConfig, CTRLModel),
+        (ElectraConfig, ElectraModel),
     ]
 )
 
@@ -182,6 +192,7 @@ MODEL_FOR_PRETRAINING_MAPPING = OrderedDict(
         (FlaubertConfig, FlaubertWithLMHeadModel),
         (XLMConfig, XLMWithLMHeadModel),
         (CTRLConfig, CTRLLMHeadModel),
+        (ElectraConfig, ElectraForPreTraining),
     ]
 )
 
@@ -202,6 +213,7 @@ MODEL_WITH_LM_HEAD_MAPPING = OrderedDict(
         (FlaubertConfig, FlaubertWithLMHeadModel),
         (XLMConfig, XLMWithLMHeadModel),
         (CTRLConfig, CTRLLMHeadModel),
+        (ElectraConfig, ElectraForMaskedLM),
     ]
 )
 
@@ -242,6 +254,7 @@ MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING = OrderedDict(
         (BertConfig, BertForTokenClassification),
         (XLNetConfig, XLNetForTokenClassification),
         (AlbertConfig, AlbertForTokenClassification),
+        (ElectraConfig, ElectraForTokenClassification),
     ]
 )
 
@@ -281,7 +294,8 @@ class AutoModel(object):
                 - isInstance of `transfo-xl` configuration class: :class:`~transformers.TransfoXLModel` (Transformer-XL model)
                 - isInstance of `xlnet` configuration class: :class:`~transformers.XLNetModel` (XLNet model)
                 - isInstance of `xlm` configuration class: :class:`~transformers.XLMModel` (XLM model)
-                - isInstance of `flaubert` configuration class: :class:`~transformers.FlaubertModel` (XLM model)
+                - isInstance of `flaubert` configuration class: :class:`~transformers.FlaubertModel` (Flaubert model)
+                - isInstance of `electra` configuration class: :class:`~transformers.ElectraModel` (Electra model)
 
         Examples::
 
@@ -322,7 +336,8 @@ class AutoModel(object):
             - contains `xlnet`: :class:`~transformers.XLNetModel` (XLNet model)
             - contains `xlm`: :class:`~transformers.XLMModel` (XLM model)
             - contains `ctrl`: :class:`~transformers.CTRLModel` (Salesforce CTRL  model)
-            - contains `flaubert`: :class:`~transformers.Flaubert` (Flaubert  model)
+            - contains `flaubert`: :class:`~transformers.FlaubertModel` (Flaubert  model)
+            - contains `electra`: :class:`~transformers.ElectraModel` (Electra  model)
 
             The model is set in evaluation mode by default using `model.eval()` (Dropout modules are deactivated)
             To train the model, you should first set it back in training mode with `model.train()`
@@ -430,6 +445,7 @@ class AutoModelForPreTraining(object):
                 - isInstance of `xlnet` configuration class: :class:`~transformers.XLNetLMHeadModel` (XLNet model)
                 - isInstance of `xlm` configuration class: :class:`~transformers.XLMWithLMHeadModel` (XLM model)
                 - isInstance of `flaubert` configuration class: :class:`~transformers.FlaubertWithLMHeadModel` (Flaubert model)
+                - isInstance of `electra` configuration class: :class:`~transformers.ElectraForPreTraining` (Electra model)
 
         Examples::
 
@@ -470,6 +486,7 @@ class AutoModelForPreTraining(object):
             - contains `xlm`: :class:`~transformers.XLMWithLMHeadModel` (XLM model)
             - contains `ctrl`: :class:`~transformers.CTRLLMHeadModel` (Salesforce CTRL model)
             - contains `flaubert`: :class:`~transformers.FlaubertWithLMHeadModel` (Flaubert model)
+            - contains `electra`: :class:`~transformers.ElectraForPreTraining` (Electra model)
 
         The model is set in evaluation mode by default using `model.eval()` (Dropout modules are deactivated)
         To train the model, you should first set it back in training mode with `model.train()`
@@ -571,6 +588,7 @@ class AutoModelWithLMHead(object):
                 - isInstance of `xlnet` configuration class: :class:`~transformers.XLNetLMHeadModel` (XLNet model)
                 - isInstance of `xlm` configuration class: :class:`~transformers.XLMWithLMHeadModel` (XLM model)
                 - isInstance of `flaubert` configuration class: :class:`~transformers.FlaubertWithLMHeadModel` (Flaubert model)
+                - isInstance of `electra` configuration class: :class:`~transformers.ElectraForMaskedLM` (Electra model)
 
         Examples::
 
@@ -612,6 +630,7 @@ class AutoModelWithLMHead(object):
             - contains `xlm`: :class:`~transformers.XLMWithLMHeadModel` (XLM model)
             - contains `ctrl`: :class:`~transformers.CTRLLMHeadModel` (Salesforce CTRL model)
             - contains `flaubert`: :class:`~transformers.FlaubertWithLMHeadModel` (Flaubert model)
+            - contains `electra`: :class:`~transformers.ElectraForMaskedLM` (Electra model)
 
         The model is set in evaluation mode by default using `model.eval()` (Dropout modules are deactivated)
         To train the model, you should first set it back in training mode with `model.train()`
@@ -998,6 +1017,7 @@ class AutoModelForTokenClassification:
                 - isInstance of `xlnet` configuration class: :class:`~transformers.XLNetModelForTokenClassification` (XLNet model)
                 - isInstance of `camembert` configuration class: :class:`~transformers.CamembertModelForTokenClassification` (Camembert model)
                 - isInstance of `roberta` configuration class: :class:`~transformers.RobertaModelForTokenClassification` (Roberta model)
+                - isInstance of `electra` configuration class: :class:`~transformers.ElectraForTokenClassification` (Electra model)
 
         Examples::
 
@@ -1035,6 +1055,7 @@ class AutoModelForTokenClassification:
             - contains `bert`: :class:`~transformers.BertForTokenClassification` (Bert model)
             - contains `xlnet`: :class:`~transformers.XLNetForTokenClassification` (XLNet model)
             - contains `roberta`: :class:`~transformers.RobertaForTokenClassification` (Roberta model)
+            - contains `electra`: :class:`~transformers.ElectraForTokenClassification` (Electra model)
 
         The model is set in evaluation mode by default using `model.eval()` (Dropout modules are deactivated)
         To train the model, you should first set it back in training mode with `model.train()`

src/transformers/modeling_bert.py

@@ -183,7 +183,7 @@ class BertEmbeddings(nn.Module):
 class BertSelfAttention(nn.Module):
     def __init__(self, config):
         super().__init__()
-        if config.hidden_size % config.num_attention_heads != 0:
+        if config.hidden_size % config.num_attention_heads != 0 and not hasattr(config, "embedding_size"):
             raise ValueError(
                 "The hidden size (%d) is not a multiple of the number of attention "
                 "heads (%d)" % (config.hidden_size, config.num_attention_heads)

src/transformers/tokenization_auto.py

@@ -26,6 +26,7 @@ from .configuration_auto import (
     CamembertConfig,
     CTRLConfig,
     DistilBertConfig,
+    ElectraConfig,
     FlaubertConfig,
     GPT2Config,
     OpenAIGPTConfig,
@@ -44,6 +45,7 @@ from .tokenization_bert_japanese import BertJapaneseTokenizer
 from .tokenization_camembert import CamembertTokenizer
 from .tokenization_ctrl import CTRLTokenizer
 from .tokenization_distilbert import DistilBertTokenizer, DistilBertTokenizerFast
+from .tokenization_electra import ElectraTokenizer, ElectraTokenizerFast
 from .tokenization_flaubert import FlaubertTokenizer
 from .tokenization_gpt2 import GPT2Tokenizer, GPT2TokenizerFast
 from .tokenization_openai import OpenAIGPTTokenizer, OpenAIGPTTokenizerFast
@@ -67,6 +69,7 @@ TOKENIZER_MAPPING = OrderedDict(
         (XLMRobertaConfig, (XLMRobertaTokenizer, None)),
         (BartConfig, (BartTokenizer, None)),
         (RobertaConfig, (RobertaTokenizer, RobertaTokenizerFast)),
+        (ElectraConfig, (ElectraTokenizer, ElectraTokenizerFast)),
         (BertConfig, (BertTokenizer, BertTokenizerFast)),
         (OpenAIGPTConfig, (OpenAIGPTTokenizer, OpenAIGPTTokenizerFast)),
         (GPT2Config, (GPT2Tokenizer, GPT2TokenizerFast)),
@@ -104,6 +107,7 @@ class AutoTokenizer:
             - contains `xlnet`: XLNetTokenizer (XLNet model)
             - contains `xlm`: XLMTokenizer (XLM model)
             - contains `ctrl`: CTRLTokenizer (Salesforce CTRL model)
+            - contains `electra`: ElectraTokenizer (Google ELECTRA model)
 
         This class cannot be instantiated using `__init__()` (throw an error).
     """
@@ -135,6 +139,7 @@ class AutoTokenizer:
             - contains `xlnet`: XLNetTokenizer (XLNet model)
             - contains `xlm`: XLMTokenizer (XLM model)
             - contains `ctrl`: CTRLTokenizer (Salesforce CTRL model)
+            - contains `electra`: ElectraTokenizer (Google ELECTRA model)
 
         Params:
             pretrained_model_name_or_path: either:

src/transformers/tokenization_electra.py

+# coding=utf-8
+# Copyright 2020 The Google AI Team, Stanford University 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.
+
+from .tokenization_bert import BertTokenizer, BertTokenizerFast
+
+
+VOCAB_FILES_NAMES = {"vocab_file": "vocab.txt"}
+
+PRETRAINED_VOCAB_FILES_MAP = {
+    "vocab_file": {
+        "google/electra-small-generator": "https://s3.amazonaws.com/models.huggingface.co/bert/google/electra-small-generator/vocab.txt",
+        "google/electra-base-generator": "https://s3.amazonaws.com/models.huggingface.co/bert/google/electra-base-generator/vocab.txt",
+        "google/electra-large-generator": "https://s3.amazonaws.com/models.huggingface.co/bert/google/electra-large-generator/vocab.txt",
+        "google/electra-small-discriminator": "https://s3.amazonaws.com/models.huggingface.co/bert/google/electra-small-discriminator/vocab.txt",
+        "google/electra-base-discriminator": "https://s3.amazonaws.com/models.huggingface.co/bert/google/electra-base-discriminator/vocab.txt",
+        "google/electra-large-discriminator": "https://s3.amazonaws.com/models.huggingface.co/bert/google/electra-large-discriminator/vocab.txt",
+    }
+}
+
+PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
+    "google/electra-small-generator": 512,
+    "google/electra-base-generator": 512,
+    "google/electra-large-generator": 512,
+    "google/electra-small-discriminator": 512,
+    "google/electra-base-discriminator": 512,
+    "google/electra-large-discriminator": 512,
+}
+
+
+PRETRAINED_INIT_CONFIGURATION = {
+    "google/electra-small-generator": {"do_lower_case": True},
+    "google/electra-base-generator": {"do_lower_case": True},
+    "google/electra-large-generator": {"do_lower_case": True},
+    "google/electra-small-discriminator": {"do_lower_case": True},
+    "google/electra-base-discriminator": {"do_lower_case": True},
+    "google/electra-large-discriminator": {"do_lower_case": True},
+}
+
+
+class ElectraTokenizer(BertTokenizer):
+    r"""
+    Constructs an Electra tokenizer.
+    :class:`~transformers.ElectraTokenizer` is identical to :class:`~transformers.BertTokenizer` and runs end-to-end
+    tokenization: punctuation splitting + wordpiece.
+
+    Refer to superclass :class:`~transformers.BertTokenizer` for usage examples and documentation concerning
+    parameters.
+    """
+
+    vocab_files_names = VOCAB_FILES_NAMES
+    pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
+    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
+    pretrained_init_configuration = PRETRAINED_INIT_CONFIGURATION
+
+
+class ElectraTokenizerFast(BertTokenizerFast):
+    r"""
+    Constructs an Electra Fast tokenizer.
+    :class:`~transformers.ElectraTokenizerFast` is identical to :class:`~transformers.BertTokenizerFast` and runs end-to-end
+    tokenization: punctuation splitting + wordpiece.
+
+    Refer to superclass :class:`~transformers.BertTokenizerFast` for usage examples and documentation concerning
+    parameters.
+    """
+    vocab_files_names = VOCAB_FILES_NAMES
+    pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
+    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
+    pretrained_init_configuration = PRETRAINED_INIT_CONFIGURATION

tests/test_modeling_electra.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.
+
+
+import unittest
+
+from transformers import is_torch_available
+
+from .test_configuration_common import ConfigTester
+from .test_modeling_common import ModelTesterMixin, ids_tensor
+from .utils import CACHE_DIR, require_torch, slow, torch_device
+
+
+if is_torch_available():
+    from transformers import (
+        ElectraConfig,
+        ElectraModel,
+        ElectraForMaskedLM,
+        ElectraForTokenClassification,
+        ElectraForPreTraining,
+    )
+    from transformers.modeling_electra import ELECTRA_PRETRAINED_MODEL_ARCHIVE_MAP
+
+
+@require_torch
+class ElectraModelTest(ModelTesterMixin, unittest.TestCase):
+
+    all_model_classes = (
+        (ElectraModel, ElectraForMaskedLM, ElectraForTokenClassification,) if is_torch_available() else ()
+    )
+
+    class ElectraModelTester(object):
+        def __init__(
+            self,
+            parent,
+            batch_size=13,
+            seq_length=7,
+            is_training=True,
+            use_input_mask=True,
+            use_token_type_ids=True,
+            use_labels=True,
+            vocab_size=99,
+            hidden_size=32,
+            num_hidden_layers=5,
+            num_attention_heads=4,
+            intermediate_size=37,
+            hidden_act="gelu",
+            hidden_dropout_prob=0.1,
+            attention_probs_dropout_prob=0.1,
+            max_position_embeddings=512,
+            type_vocab_size=16,
+            type_sequence_label_size=2,
+            initializer_range=0.02,
+            num_labels=3,
+            num_choices=4,
+            scope=None,
+        ):
+            self.parent = parent
+            self.batch_size = batch_size
+            self.seq_length = seq_length
+            self.is_training = is_training
+            self.use_input_mask = use_input_mask
+            self.use_token_type_ids = use_token_type_ids
+            self.use_labels = use_labels
+            self.vocab_size = vocab_size
+            self.hidden_size = hidden_size
+            self.num_hidden_layers = num_hidden_layers
+            self.num_attention_heads = num_attention_heads
+            self.intermediate_size = intermediate_size
+            self.hidden_act = hidden_act
+            self.hidden_dropout_prob = hidden_dropout_prob
+            self.attention_probs_dropout_prob = attention_probs_dropout_prob
+            self.max_position_embeddings = max_position_embeddings
+            self.type_vocab_size = type_vocab_size
+            self.type_sequence_label_size = type_sequence_label_size
+            self.initializer_range = initializer_range
+            self.num_labels = num_labels
+            self.num_choices = num_choices
+            self.scope = scope
+
+        def prepare_config_and_inputs(self):
+            input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size)
+
+            input_mask = None
+            if self.use_input_mask:
+                input_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
+
+            token_type_ids = None
+            if self.use_token_type_ids:
+                token_type_ids = ids_tensor([self.batch_size, self.seq_length], self.type_vocab_size)
+
+            sequence_labels = None
+            token_labels = None
+            choice_labels = None
+            if self.use_labels:
+                sequence_labels = ids_tensor([self.batch_size], self.type_sequence_label_size)
+                token_labels = ids_tensor([self.batch_size, self.seq_length], self.num_labels)
+                choice_labels = ids_tensor([self.batch_size], self.num_choices)
+                fake_token_labels = ids_tensor([self.batch_size, self.seq_length], 1)
+
+            config = ElectraConfig(
+                vocab_size=self.vocab_size,
+                hidden_size=self.hidden_size,
+                num_hidden_layers=self.num_hidden_layers,
+                num_attention_heads=self.num_attention_heads,
+                intermediate_size=self.intermediate_size,
+                hidden_act=self.hidden_act,
+                hidden_dropout_prob=self.hidden_dropout_prob,
+                attention_probs_dropout_prob=self.attention_probs_dropout_prob,
+                max_position_embeddings=self.max_position_embeddings,
+                type_vocab_size=self.type_vocab_size,
+                is_decoder=False,
+                initializer_range=self.initializer_range,
+            )
+
+            return (
+                config,
+                input_ids,
+                token_type_ids,
+                input_mask,
+                sequence_labels,
+                token_labels,
+                choice_labels,
+                fake_token_labels,
+            )
+
+        def check_loss_output(self, result):
+            self.parent.assertListEqual(list(result["loss"].size()), [])
+
+        def create_and_check_electra_model(
+            self,
+            config,
+            input_ids,
+            token_type_ids,
+            input_mask,
+            sequence_labels,
+            token_labels,
+            choice_labels,
+            fake_token_labels,
+        ):
+            model = ElectraModel(config=config)
+            model.to(torch_device)
+            model.eval()
+            (sequence_output,) = model(input_ids, attention_mask=input_mask, token_type_ids=token_type_ids)
+            (sequence_output,) = model(input_ids, token_type_ids=token_type_ids)
+            (sequence_output,) = model(input_ids)
+
+            result = {
+                "sequence_output": sequence_output,
+            }
+            self.parent.assertListEqual(
+                list(result["sequence_output"].size()), [self.batch_size, self.seq_length, self.hidden_size]
+            )
+
+        def create_and_check_electra_for_masked_lm(
+            self,
+            config,
+            input_ids,
+            token_type_ids,
+            input_mask,
+            sequence_labels,
+            token_labels,
+            choice_labels,
+            fake_token_labels,
+        ):
+            model = ElectraForMaskedLM(config=config)
+            model.to(torch_device)
+            model.eval()
+            loss, prediction_scores = model(
+                input_ids, attention_mask=input_mask, token_type_ids=token_type_ids, masked_lm_labels=token_labels
+            )
+            result = {
+                "loss": loss,
+                "prediction_scores": prediction_scores,
+            }
+            self.parent.assertListEqual(
+                list(result["prediction_scores"].size()), [self.batch_size, self.seq_length, self.vocab_size]
+            )
+            self.check_loss_output(result)
+
+        def create_and_check_electra_for_token_classification(
+            self,
+            config,
+            input_ids,
+            token_type_ids,
+            input_mask,
+            sequence_labels,
+            token_labels,
+            choice_labels,
+            fake_token_labels,
+        ):
+            config.num_labels = self.num_labels
+            model = ElectraForTokenClassification(config=config)
+            model.to(torch_device)
+            model.eval()
+            loss, logits = model(
+                input_ids, attention_mask=input_mask, token_type_ids=token_type_ids, labels=token_labels
+            )
+            result = {
+                "loss": loss,
+                "logits": logits,
+            }
+            self.parent.assertListEqual(
+                list(result["logits"].size()), [self.batch_size, self.seq_length, self.num_labels]
+            )
+            self.check_loss_output(result)
+
+        def create_and_check_electra_for_pretraining(
+            self,
+            config,
+            input_ids,
+            token_type_ids,
+            input_mask,
+            sequence_labels,
+            token_labels,
+            choice_labels,
+            fake_token_labels,
+        ):
+            config.num_labels = self.num_labels
+            model = ElectraForPreTraining(config=config)
+            model.to(torch_device)
+            model.eval()
+            loss, logits = model(
+                input_ids, attention_mask=input_mask, token_type_ids=token_type_ids, labels=fake_token_labels
+            )
+            result = {
+                "loss": loss,
+                "logits": logits,
+            }
+            self.parent.assertListEqual(list(result["logits"].size()), [self.batch_size, self.seq_length])
+            self.check_loss_output(result)
+
+        def prepare_config_and_inputs_for_common(self):
+            config_and_inputs = self.prepare_config_and_inputs()
+            (
+                config,
+                input_ids,
+                token_type_ids,
+                input_mask,
+                sequence_labels,
+                token_labels,
+                choice_labels,
+                fake_token_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 = ElectraModelTest.ElectraModelTester(self)
+        self.config_tester = ConfigTester(self, config_class=ElectraConfig, hidden_size=37)
+
+    def test_config(self):
+        self.config_tester.run_common_tests()
+
+    def test_electra_model(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_electra_model(*config_and_inputs)
+
+    def test_for_masked_lm(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_electra_for_masked_lm(*config_and_inputs)
+
+    def test_for_token_classification(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_electra_for_token_classification(*config_and_inputs)
+
+    def test_for_pre_training(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_electra_for_pretraining(*config_and_inputs)
+
+    @slow
+    def test_model_from_pretrained(self):
+        for model_name in list(ELECTRA_PRETRAINED_MODEL_ARCHIVE_MAP.keys())[:1]:
+            model = ElectraModel.from_pretrained(model_name, cache_dir=CACHE_DIR)
+            self.assertIsNotNone(model)

tests/test_modeling_tf_electra.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.
+
+
+import unittest
+
+from transformers import ElectraConfig, is_tf_available
+
+from .test_configuration_common import ConfigTester
+from .test_modeling_tf_common import TFModelTesterMixin, ids_tensor
+from .utils import CACHE_DIR, require_tf, slow
+
+
+if is_tf_available():
+    from transformers.modeling_tf_electra import (
+        TFElectraModel,
+        TFElectraForMaskedLM,
+        TFElectraForPreTraining,
+        TFElectraForTokenClassification,
+    )
+
+
+@require_tf
+class TFElectraModelTest(TFModelTesterMixin, unittest.TestCase):
+
+    all_model_classes = (
+        (TFElectraModel, TFElectraForMaskedLM, TFElectraForPreTraining, TFElectraForTokenClassification,)
+        if is_tf_available()
+        else ()
+    )
+
+    class TFElectraModelTester(object):
+        def __init__(
+            self,
+            parent,
+            batch_size=13,
+            seq_length=7,
+            is_training=True,
+            use_input_mask=True,
+            use_token_type_ids=True,
+            use_labels=True,
+            vocab_size=99,
+            hidden_size=32,
+            num_hidden_layers=5,
+            num_attention_heads=4,
+            intermediate_size=37,
+            hidden_act="gelu",
+            hidden_dropout_prob=0.1,
+            attention_probs_dropout_prob=0.1,
+            max_position_embeddings=512,
+            type_vocab_size=16,
+            type_sequence_label_size=2,
+            initializer_range=0.02,
+            num_labels=3,
+            num_choices=4,
+            scope=None,
+        ):
+            self.parent = parent
+            self.batch_size = batch_size
+            self.seq_length = seq_length
+            self.is_training = is_training
+            self.use_input_mask = use_input_mask
+            self.use_token_type_ids = use_token_type_ids
+            self.use_labels = use_labels
+            self.vocab_size = vocab_size
+            self.hidden_size = hidden_size
+            self.num_hidden_layers = num_hidden_layers
+            self.num_attention_heads = num_attention_heads
+            self.intermediate_size = intermediate_size
+            self.hidden_act = hidden_act
+            self.hidden_dropout_prob = hidden_dropout_prob
+            self.attention_probs_dropout_prob = attention_probs_dropout_prob
+            self.max_position_embeddings = max_position_embeddings
+            self.type_vocab_size = type_vocab_size
+            self.type_sequence_label_size = type_sequence_label_size
+            self.initializer_range = initializer_range
+            self.num_labels = num_labels
+            self.num_choices = num_choices
+            self.scope = scope
+
+        def prepare_config_and_inputs(self):
+            input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size)
+
+            input_mask = None
+            if self.use_input_mask:
+                input_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
+
+            token_type_ids = None
+            if self.use_token_type_ids:
+                token_type_ids = ids_tensor([self.batch_size, self.seq_length], self.type_vocab_size)
+
+            sequence_labels = None
+            token_labels = None
+            choice_labels = None
+            if self.use_labels:
+                sequence_labels = ids_tensor([self.batch_size], self.type_sequence_label_size)
+                token_labels = ids_tensor([self.batch_size, self.seq_length], self.num_labels)
+                choice_labels = ids_tensor([self.batch_size], self.num_choices)
+
+            config = ElectraConfig(
+                vocab_size=self.vocab_size,
+                hidden_size=self.hidden_size,
+                num_hidden_layers=self.num_hidden_layers,
+                num_attention_heads=self.num_attention_heads,
+                intermediate_size=self.intermediate_size,
+                hidden_act=self.hidden_act,
+                hidden_dropout_prob=self.hidden_dropout_prob,
+                attention_probs_dropout_prob=self.attention_probs_dropout_prob,
+                max_position_embeddings=self.max_position_embeddings,
+                type_vocab_size=self.type_vocab_size,
+                initializer_range=self.initializer_range,
+            )
+
+            return config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+
+        def create_and_check_electra_model(
+            self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+        ):
+            model = TFElectraModel(config=config)
+            inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
+            (sequence_output,) = model(inputs)
+
+            inputs = [input_ids, input_mask]
+            (sequence_output,) = model(inputs)
+
+            (sequence_output,) = model(input_ids)
+
+            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_electra_for_masked_lm(
+            self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+        ):
+            model = TFElectraForMaskedLM(config=config)
+            inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
+            (prediction_scores,) = model(inputs)
+            result = {
+                "prediction_scores": prediction_scores.numpy(),
+            }
+            self.parent.assertListEqual(
+                list(result["prediction_scores"].shape), [self.batch_size, self.seq_length, self.vocab_size]
+            )
+
+        def create_and_check_electra_for_pretraining(
+            self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+        ):
+            model = TFElectraForPreTraining(config=config)
+            inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
+            (prediction_scores,) = model(inputs)
+            result = {
+                "prediction_scores": prediction_scores.numpy(),
+            }
+            self.parent.assertListEqual(list(result["prediction_scores"].shape), [self.batch_size, self.seq_length])
+
+        def create_and_check_electra_for_token_classification(
+            self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+        ):
+            config.num_labels = self.num_labels
+            model = TFElectraForTokenClassification(config=config)
+            inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
+            (logits,) = model(inputs)
+            result = {
+                "logits": logits.numpy(),
+            }
+            self.parent.assertListEqual(
+                list(result["logits"].shape), [self.batch_size, self.seq_length, self.num_labels]
+            )
+
+        def prepare_config_and_inputs_for_common(self):
+            config_and_inputs = self.prepare_config_and_inputs()
+            (
+                config,
+                input_ids,
+                token_type_ids,
+                input_mask,
+                sequence_labels,
+                token_labels,
+                choice_labels,
+            ) = config_and_inputs
+            inputs_dict = {"input_ids": input_ids, "token_type_ids": token_type_ids, "attention_mask": input_mask}
+            return config, inputs_dict
+
+    def setUp(self):
+        self.model_tester = TFElectraModelTest.TFElectraModelTester(self)
+        self.config_tester = ConfigTester(self, config_class=ElectraConfig, hidden_size=37)
+
+    def test_config(self):
+        self.config_tester.run_common_tests()
+
+    def test_electra_model(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_electra_model(*config_and_inputs)
+
+    def test_for_masked_lm(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_electra_for_masked_lm(*config_and_inputs)
+
+    def test_for_pretraining(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_electra_for_pretraining(*config_and_inputs)
+
+    def test_for_token_classification(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_electra_for_token_classification(*config_and_inputs)
+
+    @slow
+    def test_model_from_pretrained(self):
+        # for model_name in list(TF_ELECTRA_PRETRAINED_MODEL_ARCHIVE_MAP.keys())[:1]:
+        for model_name in ["electra-small-discriminator"]:
+            model = TFElectraModel.from_pretrained(model_name, cache_dir=CACHE_DIR)
+            self.assertIsNotNone(model)