resolved_decode_with_prefix

decode_with_prefix
2020-04-02 14:19:12 -04:00 · 2020-04-02 14:14:11 -04:00 · 2020-04-02 14:09:31 -04:00
57 changed files with 304 additions and 3221 deletions
@@ -164,9 +164,8 @@ 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. **[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.
+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.

 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).

@@ -26,7 +26,7 @@ author = u'huggingface'
 # The short X.Y version
 version = u''
 # The full version, including alpha/beta/rc tags
-release = u'2.8.0'
+release = u'2.7.0'


 # -- General configuration ---------------------------------------------------
@@ -104,4 +104,3 @@ The library currently contains PyTorch and Tensorflow implementations, pre-train
    model_doc/flaubert
    model_doc/bart
    model_doc/t5
-    model_doc/electra
@@ -27,7 +27,7 @@ loss = outputs[0]
 # In transformers you can also have access to the logits:
 loss, logits = outputs[:2]

-# And even the attention weights if you configure the model to output them (and other outputs too, see the docstrings and documentation)
+# And even the attention weigths if you configure the model to output them (and other outputs too, see the docstrings and documentation)
 model = BertForSequenceClassification.from_pretrained('bert-base-uncased', output_attentions=True)
 outputs = model(input_ids, labels=labels)
 loss, logits, attentions = outputs
@@ -1,115 +0,0 @@
-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:
@@ -233,9 +233,6 @@ def train(args, train_dataset, model: PreTrainedModel, tokenizer: PreTrainedToke
    else:
        t_total = len(train_dataloader) // args.gradient_accumulation_steps * args.num_train_epochs

-    model = model.module if hasattr(model, "module") else model  # Take care of distributed/parallel training
-    model.resize_token_embeddings(len(tokenizer))
-
    # Prepare optimizer and schedule (linear warmup and decay)
    no_decay = ["bias", "LayerNorm.weight"]
    optimizer_grouped_parameters = [
@@ -312,6 +309,9 @@ def train(args, train_dataset, model: PreTrainedModel, tokenizer: PreTrainedToke

    tr_loss, logging_loss = 0.0, 0.0

+    model_to_resize = model.module if hasattr(model, "module") else model  # Take care of distributed/parallel training
+    model_to_resize.resize_token_embeddings(len(tokenizer))
+
    model.zero_grad()
    train_iterator = trange(
        epochs_trained, int(args.num_train_epochs), desc="Epoch", disable=args.local_rank not in [-1, 0]
@@ -64,8 +64,27 @@ TensorFlow: 2.1.0
 Python: 3.7.6
 ```

-### Access this albert_xxlargev1_sqd2_512 fine-tuned model with:
+### Inferencing / prediction works with the current Transformers v2.4.1
+
+### Access this albert_xxlargev1_sqd2_512 fine-tuned model with "tried & true" code:

 ```python
-tokenizer = AutoTokenizer.from_pretrained("ahotrod/albert_xxlargev1_squad2_512")
-model = AutoModelForQuestionAnswering.from_pretrained("ahotrod/albert_xxlargev1_squad2_512")
+config_class, model_class, tokenizer_class = \
+        AlbertConfig, AlbertForQuestionAnswering, AlbertTokenizer
+
+model_name_or_path = "ahotrod/albert_xxlargev1_squad2_512"
+config = config_class.from_pretrained(model_name_or_path)
+tokenizer = tokenizer_class.from_pretrained(model_name_or_path, do_lower_case=True)
+model = model_class.from_pretrained(model_name_or_path, config=config)
+```
+
+### or the AutoModels (AutoConfig, AutoTokenizer & AutoModel) should also work, however I have yet to use them in my app & confirm:
+
+```python
+from transformers import AutoConfig, AutoTokenizer, AutoModel
+
+model_name_or_path = "ahotrod/albert_xxlargev1_squad2_512"
+config = AutoConfig.from_pretrained(model_name_or_path)
+tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, do_lower_case=True)
+model = AutoModel.from_pretrained(model_name_or_path, config=config)
+```
@@ -1,68 +0,0 @@
-## RoBERTa-large language model fine-tuned on SQuAD2.0
-
-### with the following results:
-
-```
-  "exact": 84.02257222269014,
-  "f1": 87.47063479332766,
-  "total": 11873,
-  "HasAns_exact": 81.19095816464238,
-  "HasAns_f1": 88.0969714745582,
-  "HasAns_total": 5928,
-  "NoAns_exact": 86.84608915054667,
-  "NoAns_f1": 86.84608915054667,
-  "NoAns_total": 5945,
-  "best_exact": 84.02257222269014,
-  "best_exact_thresh": 0.0,
-  "best_f1": 87.47063479332759,
-  "best_f1_thresh": 0.0
-```
-### from script:
-```
-python -m torch.distributed.launch --nproc_per_node=2 ${RUN_SQUAD_DIR}/run_squad.py \
-  --model_type roberta \
-  --model_name_or_path roberta-large \
-  --do_train \
-  --train_file ${SQUAD_DIR}/train-v2.0.json \
-  --predict_file ${SQUAD_DIR}/dev-v2.0.json \
-  --version_2_with_negative \
-  --num_train_epochs 2 \
-  --warmup_steps 328 \
-  --weight_decay 0.01 \
-  --do_lower_case \
-  --learning_rate 1.5e-5 \
-  --max_seq_length 512 \
-  --doc_stride 128 \
-  --save_steps 1000 \
-  --per_gpu_train_batch_size 1 \
-  --gradient_accumulation_steps 24 \
-  --logging_steps 50 \
-  --threads 10 \
-  --overwrite_cache \
-  --overwrite_output_dir \
-  --output_dir ${MODEL_PATH}
-
-python ${RUN_SQUAD_DIR}/run_squad.py \
-  --model_type roberta \
-  --model_name_or_path ${MODEL_PATH} \
-  --do_eval \
-  --train_file ${SQUAD_DIR}/train-v2.0.json \
-  --predict_file ${SQUAD_DIR}/dev-v2.0.json \
-  --version_2_with_negative \
-  --do_lower_case \
-  --max_seq_length 512 \
-  --per_gpu_eval_batch_size 24 \
-  --eval_all_checkpoints \
-  --overwrite_output_dir \
-  --output_dir ${MODEL_PATH}
-$@
-```
-### using the following system & software:
-```
-OS/Platform: Linux-4.15.0-91-generic-x86_64-with-debian-buster-sid
-GPU/CPU: 2 x NVIDIA 1080Ti / Intel i7-8700
-Transformers: 2.7.0
-PyTorch: 1.4.0
-TensorFlow: 2.1.0
-Python: 3.7.7
-```
@@ -56,8 +56,22 @@ PyTorch: 1.4.0
 TensorFlow: 2.1.0
 Python: 3.7.6
 ```
+### Inferencing / prediction works with Transformers v2.4.1, the latest version tested
+
 ### Utilize this xlnet_large_squad2_512 fine-tuned model with:
 ```python
-tokenizer = AutoTokenizer.from_pretrained("ahotrod/xlnet_large_squad2_512")
-model = AutoModelForQuestionAnswering.from_pretrained("ahotrod/xlnet_large_squad2_512")
+config_class, model_class, tokenizer_class = \
+        XLNetConfig, XLNetforQuestionAnswering, XLNetTokenizer
+model_name_or_path = "ahotrod/xlnet_large_squad2_512"
+config = config_class.from_pretrained(model_name_or_path)
+tokenizer = tokenizer_class.from_pretrained(model_name_or_path, do_lower_case=True)
+model = model_class.from_pretrained(model_name_or_path, config=config)
+```
+### or the AutoModels (AutoConfig, AutoTokenizer & AutoModel) should also work, however I have yet to use them in my apps & confirm:
+```python
+from transformers import AutoConfig, AutoTokenizer, AutoModel
+model_name_or_path = "ahotrod/xlnet_large_squad2_512"
+config = AutoConfig.from_pretrained(model_name_or_path)
+tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, do_lower_case=True)
+model = AutoModel.from_pretrained(model_name_or_path, config=config)
 ```
@@ -1,6 +0,0 @@
---
-tags:
- exbert
---
-
-[![ExBERT](https://img.shields.io/badge/Visualize%20Attentions-ExBERT-green)](https://huggingface.co/exbert/?model=albert-base-v1)
@@ -1,6 +0,0 @@
---
-tags:
- exbert
---
-
-[![ExBERT](https://img.shields.io/badge/Visualize%20Attentions-ExBERT-green)](https://huggingface.co/exbert/?model=albert-xxlarge-v2)
@@ -1,6 +0,0 @@
---
-tags:
- exbert
---
-
-[![ExBERT](https://img.shields.io/badge/Visualize%20Attentions-ExBERT-green)](https://huggingface.co/exbert/?model=bert-base-cased)
@@ -1,12 +1,8 @@
 ---
 language: german
 thumbnail: https://static.tildacdn.com/tild6438-3730-4164-b266-613634323466/german_bert.png
-tags:
- exbert
 ---

-[![ExBERT](https://img.shields.io/badge/Visualize%20Attentions-ExBERT-green)](https://huggingface.co/exbert/?model=bert-base-german-cased)
-
 # German BERT
 ![bert_image](https://static.tildacdn.com/tild6438-3730-4164-b266-613634323466/german_bert.png)
 ## Overview
@@ -22,7 +18,6 @@ tags:
 - We trained 810k steps with a batch size of 1024 for sequence length 128 and 30k steps with sequence length 512. Training took about 9 days.
 - As training data we used the latest German Wikipedia dump (6GB of raw txt files), the OpenLegalData dump (2.4 GB) and news articles (3.6 GB).
 - We cleaned the data dumps with tailored scripts and segmented sentences with spacy v2.1. To create tensorflow records we used the recommended sentencepiece library for creating the word piece vocabulary and tensorflow scripts to convert the text to data usable by BERT.
- Update April 3rd, 2020: updated the vocab file on deepset s3 to adjust tokenization of punctuation.

 See https://deepset.ai/german-bert for more details

@@ -1,6 +0,0 @@
---
-tags:
- exbert
---
-
-[![ExBERT](https://img.shields.io/badge/Visualize%20Attentions-ExBERT-green)](https://huggingface.co/exbert/?model=bert-base-uncased)
@@ -1,7 +1,3 @@
---
-language: french
---
-
 # CamemBERT 

 CamemBERT is a state-of-the-art language model for French based on the RoBERTa architecture pretrained on the French subcorpus of the newly available multilingual corpus OSCAR.  
@@ -1,6 +0,0 @@
---
-tags:
- exbert
---
-
-[![ExBERT](https://img.shields.io/badge/Visualize%20Attentions-ExBERT-green)](https://huggingface.co/exbert/?model=distilbert-base-uncased)
@@ -1,6 +0,0 @@
---
-tags:
- exbert
---
-
-[![ExBERT](https://img.shields.io/badge/Visualize%20Attentions-ExBERT-green)](https://huggingface.co/exbert/?model=distilgpt2)
@@ -1,6 +0,0 @@
---
-tags:
- exbert
---
-
-[![ExBERT](https://img.shields.io/badge/Visualize%20Attentions-ExBERT-green)](https://huggingface.co/exbert/?model=distilroberta-base)
@@ -1,6 +0,0 @@
---
-tags:
- exbert
---
-
-[![ExBERT](https://img.shields.io/badge/Visualize%20Attentions-ExBERT-green)](https://huggingface.co/exbert/?model=gpt2)
@@ -1,94 +0,0 @@
---
-language: polish
---
-
-# Multilingual + Polish SQuAD1.1
-
-This model is the multilingual model provided by the Google research team with a fine-tuned polish Q&A downstream task.
-
-## Details of the language model
-
-Language model ([**bert-base-multilingual-cased**](https://github.com/google-research/bert/blob/master/multilingual.md)):
-12-layer, 768-hidden, 12-heads, 110M parameters.
-Trained on cased text in the top 104 languages with the largest Wikipedias.
-
-## Details of the downstream task
-Using the `mtranslate` Python module, [**SQuAD1.1**](https://rajpurkar.github.io/SQuAD-explorer/) was machine-translated. In order to find the start tokens, the direct translations of the answers were searched in the corresponding paragraphs. Due to the different translations depending on the context (missing context in the pure answer), the answer could not always be found in the text, and thus a loss of question-answer examples occurred. This is a potential problem where errors can occur in the data set.
-
-| Dataset                | # Q&A |
-| ---------------------- | ----- |
-| SQuAD1.1 Train         | 87.7 K |
-| Polish SQuAD1.1 Train   | 39.5 K |
-| SQuAD1.1 Dev           |  10.6 K |
-| Polish SQuAD1.1 Dev     |  2.6 K |
-
-
-## Model benchmark
-
-| Model                | EM | F1 |
-| ---------------------- | ----- | ----- |
-| [SlavicBERT](https://huggingface.co/DeepPavlov/bert-base-bg-cs-pl-ru-cased)   | **60.89** | 71.68 |
-| [polBERT](https://huggingface.co/dkleczek/bert-base-polish-uncased-v1)   | 57.46 | 68.87 |
-| [multiBERT](https://huggingface.co/bert-base-multilingual-cased) | 60.67 | **71.89** |
-| [xlm](https://huggingface.co/xlm-mlm-100-1280)     | 47.98 | 59.42 |
-## Model training
-
-The model was trained on a **Tesla V100** GPU with the following command:
-
-```python
-export SQUAD_DIR=path/to/pl_squad
-
-python run_squad.py 
-  --model_type bert \
-  --model_name_or_path bert-base-multilingual-cased \
-  --do_train \
-  --do_eval \
-  --train_file $SQUAD_DIR/pl_squadv1_train_clean.json \
-  --predict_file $SQUAD_DIR/pl_squadv1_dev_clean.json \
-  --num_train_epochs 2 \
-  --max_seq_length 384 \
-  --doc_stride 128 \
-  --save_steps=8000 \
-  --output_dir ../../output \
-  --overwrite_cache \
-  --overwrite_output_dir
-```
-
-**Results**:
-
-{'exact': 60.670731707317074, 'f1': 71.8952193697293, 'total': 2624, 'HasAns_exact': 60.670731707317074, 'HasAns_f1': 71.8952193697293,
-'HasAns_total': 2624, 'best_exact': 60.670731707317074, 'best_exact_thresh': 0.0, 'best_f1': 71.8952193697293, 'best_f1_thresh': 0.0}
-
-## Model in action
-
-Fast usage with **pipelines**:
-
-```python
-from transformers import pipeline
-
-qa_pipeline = pipeline(
-    "question-answering",
-    model="henryk/bert-base-multilingual-cased-finetuned-polish-squad1",
-    tokenizer="henryk/bert-base-multilingual-cased-finetuned-polish-squad1"
-)
-
-qa_pipeline({
-    'context': "Warszawa jest największym miastem w Polsce pod względem liczby ludności i powierzchni",
-    'question': "Jakie jest największe miasto w Polsce?"})
-
-```
-
-# Output:
-
-```json
-{
-  "score": 0.9988,
-  "start": 0, 
-  "end": 8,
-  "answer": "Warszawa"
-}
-```
-
-## Contact
-
-Please do not hesitate to contact me via [LinkedIn](https://www.linkedin.com/in/henryk-borzymowski-0755a2167/) if you want to discuss or get access to the Polish version of SQuAD.
@@ -1,96 +0,0 @@
---
-language: polish
---
-
-# Multilingual + Polish SQuAD2.0
-
-This model is the multilingual model provided by the Google research team with a fine-tuned polish Q&A downstream task.
-
-## Details of the language model
-
-Language model ([**bert-base-multilingual-cased**](https://github.com/google-research/bert/blob/master/multilingual.md)):
-12-layer, 768-hidden, 12-heads, 110M parameters.
-Trained on cased text in the top 104 languages with the largest Wikipedias.
-
-## Details of the downstream task
-Using the `mtranslate` Python module, [**SQuAD2.0**](https://rajpurkar.github.io/SQuAD-explorer/) was machine-translated. In order to find the start tokens, the direct translations of the answers were searched in the corresponding paragraphs. Due to the different translations depending on the context (missing context in the pure answer), the answer could not always be found in the text, and thus a loss of question-answer examples occurred. This is a potential problem where errors can occur in the data set.
-
-| Dataset                | # Q&A |
-| ---------------------- | ----- |
-| SQuAD2.0 Train         | 130 K |
-| Polish SQuAD2.0 Train   | 83.1 K |
-| SQuAD2.0 Dev           |  12 K |
-| Polish SQuAD2.0 Dev     | 8.5  K |
-
-
-## Model benchmark
-
-| Model                | EM/F1 |HasAns (EM/F1) | NoAns |
-| ---------------------- | ----- | ----- | ----- |
-| [SlavicBERT](https://huggingface.co/DeepPavlov/bert-base-bg-cs-pl-ru-cased)   | 69.35/71.51  | 47.02/54.09 | 79.20 |
-| [polBERT](https://huggingface.co/dkleczek/bert-base-polish-uncased-v1)   | 67.33/69.80| 45.73/53.80  | 76.87 |
-| [multiBERT](https://huggingface.co/bert-base-multilingual-cased) | **70.76**/**72.92**  |45.00/52.04 | 82.13 |
-
-## Model training
-
-The model was trained on a **Tesla V100** GPU with the following command:
-
-```python
-export SQUAD_DIR=path/to/pl_squad
-
-python run_squad.py 
-  --model_type bert \
-  --model_name_or_path bert-base-multilingual-cased \
-  --do_train \
-  --do_eval \
-  --version_2_with_negative \
-  --train_file $SQUAD_DIR/pl_squadv2_train.json \
-  --predict_file $SQUAD_DIR/pl_squadv2_dev.json \
-  --num_train_epochs 2 \
-  --max_seq_length 384 \
-  --doc_stride 128 \
-  --save_steps=8000 \
-  --output_dir ../../output \
-  --overwrite_cache \
-  --overwrite_output_dir
-```
-
-**Results**:
-
-{'exact': 70.76671723655035, 'f1': 72.92156947155917, 'total': 8569, 'HasAns_exact': 45.00762195121951, 'HasAns_f1': 52.04456128116991, 'HasAns_total': 2624, 'NoAns_exact': 82.13624894869638, '
-NoAns_f1': 82.13624894869638, 'NoAns_total': 5945, 'best_exact': 71.72365503559342, 'best_exact_thresh': 0.0, 'best_f1': 73.62662512059369, 'best_f1_thresh': 0.0}
-
-
-## Model in action
-
-Fast usage with **pipelines**:
-
-```python
-from transformers import pipeline
-
-qa_pipeline = pipeline(
-    "question-answering",
-    model="henryk/bert-base-multilingual-cased-finetuned-polish-squad2",
-    tokenizer="henryk/bert-base-multilingual-cased-finetuned-polish-squad2"
-)
-
-qa_pipeline({
-    'context': "Warszawa jest największym miastem w Polsce pod względem liczby ludności i powierzchni",
-    'question': "Jakie jest największe miasto w Polsce?"})
-
-```
-
-# Output:
-
-```json
-{
-  "score": 0.9986,
-  "start": 0, 
-  "end": 8,
-  "answer": "Warszawa"
-}
-```
-
-## Contact
-
-Please do not hesitate to contact me via [LinkedIn](https://www.linkedin.com/in/henryk-borzymowski-0755a2167/) if you want to discuss or get access to the Polish version of SQuAD.
@@ -1,86 +0,0 @@
---
-language: malay
---
-
-# Bahasa Albert Model
-
-Pretrained Albert base language model for Malay and Indonesian. 
-
-## Pretraining Corpus
-
-`albert-base-bahasa-cased` model was pretrained on ~1.8 Billion words. We trained on both standard and social media language structures, and below is list of data we trained on,
-
-1. [dumping wikipedia](https://github.com/huseinzol05/Malaya-Dataset#wikipedia-1).
-2. [local instagram](https://github.com/huseinzol05/Malaya-Dataset#instagram).
-3. [local twitter](https://github.com/huseinzol05/Malaya-Dataset#twitter-1).
-4. [local news](https://github.com/huseinzol05/Malaya-Dataset#public-news).
-5. [local parliament text](https://github.com/huseinzol05/Malaya-Dataset#parliament).
-6. [local singlish/manglish text](https://github.com/huseinzol05/Malaya-Dataset#singlish-text).
-7. [IIUM Confession](https://github.com/huseinzol05/Malaya-Dataset#iium-confession).
-8. [Wattpad](https://github.com/huseinzol05/Malaya-Dataset#wattpad).
-9. [Academia PDF](https://github.com/huseinzol05/Malaya-Dataset#academia-pdf).
-
-Preprocessing steps can reproduce from here, [Malaya/pretrained-model/preprocess](https://github.com/huseinzol05/Malaya/tree/master/pretrained-model/preprocess).
-
-## Pretraining details
-
- This model was trained using Google Albert's github [repository](https://github.com/google-research/ALBERT) on v3-8 TPU.
- All steps can reproduce from here, [Malaya/pretrained-model/albert](https://github.com/huseinzol05/Malaya/tree/master/pretrained-model/albert).
-
-## Load Pretrained Model
-
-You can use this model by installing `torch` or `tensorflow` and Huggingface library `transformers`. And you can use it directly by initializing it like this:  
-
-```python
-from transformers import AlbertTokenizer, AlbertModel
-
-model = BertModel.from_pretrained('huseinzol05/albert-base-bahasa-cased')
-tokenizer = AlbertTokenizer.from_pretrained(
-    'huseinzol05/albert-base-bahasa-cased',
-    do_lower_case = False,
-)
-```
-
-## Example using AutoModelWithLMHead
-
-```python
-from transformers import AlbertTokenizer, AutoModelWithLMHead, pipeline
-
-model = AutoModelWithLMHead.from_pretrained('huseinzol05/albert-base-bahasa-cased')
-tokenizer = AlbertTokenizer.from_pretrained(
-    'huseinzol05/albert-base-bahasa-cased',
-    do_lower_case = False,
-)
-fill_mask = pipeline('fill-mask', model = model, tokenizer = tokenizer)
-print(fill_mask('makan ayam dengan [MASK]'))
-```
-
-Output is,
-
-```text
-[{'sequence': '[CLS] makan ayam dengan ayam[SEP]',
-  'score': 0.044952988624572754,
-  'token': 629},
- {'sequence': '[CLS] makan ayam dengan sayur[SEP]',
-  'score': 0.03621877357363701,
-  'token': 1639},
- {'sequence': '[CLS] makan ayam dengan ikan[SEP]',
-  'score': 0.034429922699928284,
-  'token': 758},
- {'sequence': '[CLS] makan ayam dengan nasi[SEP]',
-  'score': 0.032447945326566696,
-  'token': 453},
- {'sequence': '[CLS] makan ayam dengan rendang[SEP]',
-  'score': 0.028885239735245705,
-  'token': 2451}]
-```
-
-## Results
-
-For further details on the model performance, simply checkout accuracy page from Malaya, https://malaya.readthedocs.io/en/latest/Accuracy.html, we compared with traditional models.
-
-## Acknowledgement
-
-Thanks to [Im Big](https://www.facebook.com/imbigofficial/), [LigBlou](https://www.facebook.com/ligblou), [Mesolitica](https://mesolitica.com/) and [KeyReply](https://www.keyreply.com/) for sponsoring AWS, Google and GPU clouds to train Albert for Bahasa. 
-
-
@@ -4,7 +4,7 @@ language: malay

 # Bahasa Tiny-BERT Model

-General Distilled Tiny BERT language model for Malay and Indonesian. 
+General Distilled Tiny BERT base language model for Malay and Indonesian. 

 ## Pretraining Corpus

@@ -36,7 +36,7 @@ from transformers import AlbertTokenizer, BertModel

 model = BertModel.from_pretrained('huseinzol05/tiny-bert-bahasa-cased')
 tokenizer = AlbertTokenizer.from_pretrained(
-    'huseinzol05/tiny-bert-bahasa-cased',
+    'huseinzol05/tiny-base-bahasa-cased',
    unk_token = '[UNK]',
    pad_token = '[PAD]',
    do_lower_case = False,
@@ -50,9 +50,9 @@ We use [google/sentencepiece](https://github.com/google/sentencepiece) to train
 ```python
 from transformers import AlbertTokenizer, AutoModelWithLMHead, pipeline

-model = AutoModelWithLMHead.from_pretrained('huseinzol05/tiny-bert-bahasa-cased')
+model = AutoModelWithLMHead.from_pretrained('huseinzol05/tiny-base-bahasa-cased')
 tokenizer = AlbertTokenizer.from_pretrained(
-    'huseinzol05/tiny-bert-bahasa-cased',
+    'huseinzol05/tiny-base-bahasa-cased',
    unk_token = '[UNK]',
    pad_token = '[PAD]',
    do_lower_case = False,
@@ -1,61 +0,0 @@
-### Model
-**[`albert-xlarge-v2`](https://huggingface.co/albert-xlarge-v2)** fine-tuned on **[`SQuAD V2`](https://rajpurkar.github.io/SQuAD-explorer/)** using **[`run_squad.py`](https://github.com/huggingface/transformers/blob/master/examples/run_squad.py)**
-
-### Training Parameters
-Trained on 4 NVIDIA GeForce RTX 2080 Ti 11Gb
-```bash
-BASE_MODEL=albert-xlarge-v2
-python run_squad.py \
-  --version_2_with_negative \
-  --model_type albert \
-  --model_name_or_path $BASE_MODEL \
-  --output_dir $OUTPUT_MODEL \
-  --do_eval \
-  --do_lower_case \
-  --train_file $SQUAD_DIR/train-v2.0.json \
-  --predict_file $SQUAD_DIR/dev-v2.0.json \
-  --per_gpu_train_batch_size 3 \
-  --per_gpu_eval_batch_size 64 \
-  --learning_rate 3e-5 \
-  --num_train_epochs 3.0 \
-  --max_seq_length 384 \
-  --doc_stride 128 \
-  --save_steps 2000 \
-  --threads 24 \
-  --warmup_steps 814 \
-  --gradient_accumulation_steps 4 \
-  --fp16 \
-  --do_train
-```
-  
-### Evaluation
-
-Evaluation on the dev set. I did not sweep for best threshold.
-
-|                   | val               |
-|-------------------|-------------------|
-| exact             | 84.41842836688285 |
-| f1                | 87.4628460501696  |
-| total             | 11873.0           |
-| HasAns_exact      | 80.68488529014844 |
-| HasAns_f1         | 86.78245127423482 |
-| HasAns_total      | 5928.0            |
-| NoAns_exact       | 88.1412952060555  |
-| NoAns_f1          | 88.1412952060555  |
-| NoAns_total       | 5945.0            |
-| best_exact        | 84.41842836688285 |
-| best_exact_thresh | 0.0               |
-| best_f1           | 87.46284605016956 |
-| best_f1_thresh    | 0.0               |
-
-
-### Usage
-
-See [huggingface documentation](https://huggingface.co/transformers/model_doc/albert.html#albertforquestionanswering). Training on `SQuAD V2` allows the model to score if a paragraph contains an answer:
-```python
-start_scores, end_scores = model(input_ids) 
-span_scores = start_scores.softmax(dim=1).log()[:,:,None] + end_scores.softmax(dim=1).log()[:,None,:]
-ignore_score = span_scores[:,0,0] #no answer scores
-    
-```
-
@@ -1,61 +0,0 @@
-### Model
-**[`allenai/scibert_scivocab_uncased`](https://huggingface.co/allenai/scibert_scivocab_uncased)** fine-tuned on **[`SQuAD V2`](https://rajpurkar.github.io/SQuAD-explorer/)** using **[`run_squad.py`](https://github.com/huggingface/transformers/blob/master/examples/run_squad.py)**
-
-### Training Parameters
-Trained on 4 NVIDIA GeForce RTX 2080 Ti 11Gb
-```bash
-BASE_MODEL=allenai/scibert_scivocab_uncased
-python run_squad.py \
-  --version_2_with_negative \
-  --model_type albert \
-  --model_name_or_path $BASE_MODEL \
-  --output_dir $OUTPUT_MODEL \
-  --do_eval \
-  --do_lower_case \
-  --train_file $SQUAD_DIR/train-v2.0.json \
-  --predict_file $SQUAD_DIR/dev-v2.0.json \
-  --per_gpu_train_batch_size 18 \
-  --per_gpu_eval_batch_size 64 \
-  --learning_rate 3e-5 \
-  --num_train_epochs 3.0 \
-  --max_seq_length 384 \
-  --doc_stride 128 \
-  --save_steps 2000 \
-  --threads 24 \
-  --warmup_steps 550 \
-  --gradient_accumulation_steps 1 \
-  --fp16 \
-  --logging_steps 50 \
-  --do_train
-```
-  
-### Evaluation
-
-Evaluation on the dev set. I did not sweep for best threshold.
-
-|                   | val               |
-|-------------------|-------------------|
-| exact             | 75.07790785816559 |
-| f1                | 78.47735207283013 |
-| total             | 11873.0           |
-| HasAns_exact      | 70.76585695006747 |
-| HasAns_f1         | 77.57449412292718 |
-| HasAns_total      | 5928.0            |
-| NoAns_exact       | 79.37762825904122 |
-| NoAns_f1          | 79.37762825904122 |
-| NoAns_total       | 5945.0            |
-| best_exact        | 75.08633032931863 |
-| best_exact_thresh | 0.0               |
-| best_f1           | 78.48577454398324 |
-| best_f1_thresh    | 0.0               |
-
-### Usage
-
-See [huggingface documentation](https://huggingface.co/transformers/model_doc/bert.html#bertforquestionanswering). Training on `SQuAD V2` allows the model to score if a paragraph contains an answer:
-```python
-start_scores, end_scores = model(input_ids) 
-span_scores = start_scores.softmax(dim=1).log()[:,:,None] + end_scores.softmax(dim=1).log()[:,None,:]
-ignore_score = span_scores[:,0,0] #no answer scores
-    
-```
-
@@ -1,27 +0,0 @@
-# GPT2-IMDB-neg (LM + RL) 🎞😡✍
-
-All credits to [@lvwerra](https://twitter.com/lvwerra)
-
-## What is it?
-A small GPT2 (`lvwerra/gpt2-imdb`) language model fine-tuned to produce **negative** movie reviews based the [IMDB dataset](https://www.kaggle.com/lakshmi25npathi/imdb-dataset-of-50k-movie-reviews). The model is trained with rewards from a BERT sentiment classifier (`lvwerra/gpt2-imdb`) via **PPO**.
-
-## Why?
-I wanted to reproduce the experiment [lvwerra/gpt2-imdb-pos](https://huggingface.co/lvwerra/gpt2-imdb-pos) but for generating **negative** movie reviews.
-
-## Training setting
-The model was trained for `100` optimisation steps with a batch size of `256` which corresponds to `25600` training samples. The full experiment setup (for positive samples) in [trl repo](https://lvwerra.github.io/trl/04-gpt2-sentiment-ppo-training/).
-
-## Examples
-A few examples of the model response to a query before and after optimisation:
-
-| query | response (before) | response (after) | rewards (before) | rewards (after) |
-|-------|-------------------|------------------|------------------|-----------------|
-|This movie is a fine |	attempt as far as live action is concerned, n...|example of how bad Hollywood in theatrics pla...|	2.118391 |	-3.31625|
-|I have watched 3 episodes |with this guy and he is such a talented actor...|	but the show is just plain awful and there ne...|	2.681171|	-4.512792|
-|We know that firefighters and|	police officers are forced to become populari...|	other chains have going to get this disaster ...|	1.367811|	-3.34017|
-
-
-
-> Created by [Manuel Romero/@mrm8488](https://twitter.com/mrm8488)
-
-> Made with <span style="color: #e25555;">&hearts;</span> in Spain
@@ -1,57 +0,0 @@
---
-language: german
---
-
-# Model description
-## Dataset
-Trained on fictional and non-fictional German texts written between 1840 and 1920:
-* Narrative texts from Digitale Bibliothek (https://textgrid.de/digitale-bibliothek)
-* Fairy tales and sagas from Grimm Korpus (https://www1.ids-mannheim.de/kl/projekte/korpora/archiv/gri.html)
-* Newspaper and magazine article from Mannheimer Korpus Historischer Zeitungen und Zeitschriften (https://repos.ids-mannheim.de/mkhz-beschreibung.html)
-* Magazine article from the journal „Die Grenzboten“ (http://www.deutschestextarchiv.de/doku/textquellen#grenzboten)
-* Fictional and non-fictional texts from Projekt Gutenberg (https://www.projekt-gutenberg.org)
-
-## Hardware used
-1 Tesla P4 GPU
-
-## Hyperparameters
-
-| Parameter                     | Value    |
-|-------------------------------|----------|
-| Epochs                        | 3        |
-| Gradient_accumulation_steps   | 1        |
-| Train_batch_size              | 32       |
-| Learning_rate                 | 0.00003  |
-| Max_seq_len                   | 128      |
-
-## Evaluation results: Automatic tagging of four forms of speech/thought/writing representation in historical fictional and non-fictional German texts
-
-The language model was used in the task to tag direct, indirect, reported and free indirect speech/thought/writing representation in fictional and non-fictional German texts. The tagger is available and described in detail at https://github.com/redewiedergabe/tagger.
-
-The tagging model was trained using the SequenceTagger Class of the Flair framework ([Akbik et al., 2019](https://www.aclweb.org/anthology/N19-4010)) which implements a BiLSTM-CRF architecture on top of a language embedding (as proposed by [Huang et al. (2015)](https://arxiv.org/abs/1508.01991)). 
-
-
-Hyperparameters
-
-| Parameter                     | Value      |
-|-------------------------------|------------|
-| Hidden_size                   | 256        |
-| Learning_rate                 | 0.1        |
-| Mini_batch_size               | 8          |
-| Max_epochs                    | 150        |
-
-Results are reported below in comparison to a custom trained flair embedding, which was stacked onto a custom trained fastText-model. Both models were trained on the same dataset.
-
-|                | BERT       ||| FastText+Flair  |||Test data|
-|----------------|----------|-----------|----------|------|-----------|--------|--------|
-|                | F1       | Precision | Recall   | F1   | Precision | Recall ||
-| Direct         | 0.80     | 0.86      | 0.74     | 0.84 | 0.90      | 0.79   |historical German, fictional & non-fictional|
-| Indirect       | **0.76** | **0.79**  | **0.73** | 0.73 | 0.78      | 0.68   |historical German, fictional & non-fictional|
-| Reported       | **0.58** | **0.69**  | **0.51** | 0.56 | 0.68      | 0.48   |historical German, fictional & non-fictional|
-| Free indirect  | **0.57** | **0.80**  | **0.44** | 0.47 | 0.78      | 0.34   |modern German, fictional|
-
-## Intended use:
-Historical German Texts (1840 to 1920)
-
-(Showed good performance with modern German fictional texts as well)
-
@@ -1,6 +0,0 @@
---
-tags:
- exbert
---
-
-[![ExBERT](https://img.shields.io/badge/Visualize%20Attentions-ExBERT-green)](https://huggingface.co/exbert/?model=roberta-base)
@@ -1,6 +0,0 @@
---
-tags:
- exbert
---
-
-[![ExBERT](https://img.shields.io/badge/Visualize%20Attentions-ExBERT-green)](https://huggingface.co/exbert/?model=xlm-mlm-en-2048)
@@ -1,6 +0,0 @@
---
-tags:
- exbert
---
-
-[![ExBERT](https://img.shields.io/badge/Visualize%20Attentions-ExBERT-green)](https://huggingface.co/exbert/?model=xlm-roberta-base)
@@ -76,14 +76,14 @@ extras["testing"] = ["pytest", "pytest-xdist"]
 extras["docs"] = ["recommonmark", "sphinx", "sphinx-markdown-tables", "sphinx-rtd-theme"]
 extras["quality"] = [
    "black",
-    "isort",
+    "isort @ git+git://github.com/timothycrosley/isort.git@e63ae06ec7d70b06df9e528357650281a3d3ec22#egg=isort",
    "flake8",
 ]
 extras["dev"] = extras["testing"] + extras["quality"] + ["mecab-python3", "scikit-learn", "tensorflow", "torch"]

 setup(
    name="transformers",
-    version="2.8.0",
+    version="2.7.0",
    author="Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Sam Shleifer, Google AI Language Team Authors, Open AI team Authors, Facebook AI Authors, Carnegie Mellon University Authors",
    author_email="thomas@huggingface.co",
    description="State-of-the-art Natural Language Processing for TensorFlow 2.0 and PyTorch",
@@ -2,7 +2,7 @@
 # There's no way to ignore "F401 '...' imported but unused" warnings in this
 # module, but to preserve other warnings. So, don't check this module at all.

-__version__ = "2.8.0"
+__version__ = "2.7.0"

 # Work around to update TensorFlow's absl.logging threshold which alters the
 # default Python logging output behavior when present.
@@ -38,7 +38,6 @@ 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
@@ -128,7 +127,6 @@ 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
@@ -299,15 +297,6 @@ 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,
@@ -474,15 +463,6 @@ 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

@@ -18,6 +18,12 @@ def _gelu_python(x):
    return x * 0.5 * (1.0 + torch.erf(x / math.sqrt(2.0)))


+if torch.__version__ < "1.4.0":
+    gelu = _gelu_python
+else:
+    gelu = F.gelu
+
+
 def gelu_new(x):
    """ Implementation of the gelu activation function currently in Google Bert repo (identical to OpenAI GPT).
        Also see https://arxiv.org/abs/1606.08415
@@ -25,12 +31,6 @@ def gelu_new(x):
    return 0.5 * x * (1 + torch.tanh(math.sqrt(2 / math.pi) * (x + 0.044715 * torch.pow(x, 3))))


-if torch.__version__ < "1.4.0":
-    gelu = _gelu_python
-else:
-    gelu = F.gelu
-    gelu_new = torch.jit.script(gelu_new)
-
 ACT2FN = {
    "relu": F.relu,
    "swish": swish,
@@ -24,7 +24,6 @@ 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
@@ -58,7 +57,6 @@ 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()
 )
@@ -81,7 +79,6 @@ CONFIG_MAPPING = OrderedDict(
        ("xlnet", XLNetConfig,),
        ("xlm", XLMConfig,),
        ("ctrl", CTRLConfig,),
-        ("electra", ElectraConfig,),
    ]
 )

@@ -136,7 +133,6 @@ 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:
@@ -1,132 +0,0 @@
-# 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
@@ -1,79 +0,0 @@
-# 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
-    )
@@ -25,7 +25,6 @@ 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,
@@ -40,7 +39,6 @@ from transformers import (
    CamembertConfig,
    CTRLConfig,
    DistilBertConfig,
-    ElectraConfig,
    FlaubertConfig,
    GPT2Config,
    OpenAIGPTConfig,
@@ -54,7 +52,6 @@ from transformers import (
    TFCTRLLMHeadModel,
    TFDistilBertForMaskedLM,
    TFDistilBertForQuestionAnswering,
-    TFElectraForPreTraining,
    TFFlaubertWithLMHeadModel,
    TFGPT2LMHeadModel,
    TFOpenAIGPTLMHeadModel,
@@ -113,8 +110,6 @@ if is_torch_available():
        ALBERT_PRETRAINED_MODEL_ARCHIVE_MAP,
        T5ForConditionalGeneration,
        T5_PRETRAINED_MODEL_ARCHIVE_MAP,
-        ElectraForPreTraining,
-        ELECTRA_PRETRAINED_MODEL_ARCHIVE_MAP,
    )
 else:
    (
@@ -152,8 +147,6 @@ else:
        ALBERT_PRETRAINED_MODEL_ARCHIVE_MAP,
        T5ForConditionalGeneration,
        T5_PRETRAINED_MODEL_ARCHIVE_MAP,
-        ElectraForPreTraining,
-        ELECTRA_PRETRAINED_MODEL_ARCHIVE_MAP,
    ) = (
        None,
        None,
@@ -189,8 +182,6 @@ else:
        None,
        None,
        None,
-        None,
-        None,
    )


@@ -330,13 +321,6 @@ 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,
-    ),
 }


@@ -26,7 +26,6 @@ from .configuration_auto import (
    CamembertConfig,
    CTRLConfig,
    DistilBertConfig,
-    ElectraConfig,
    FlaubertConfig,
    GPT2Config,
    OpenAIGPTConfig,
@@ -77,13 +76,6 @@ 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,
@@ -149,7 +141,6 @@ 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()
 )
@@ -171,7 +162,6 @@ MODEL_MAPPING = OrderedDict(
        (FlaubertConfig, FlaubertModel),
        (XLMConfig, XLMModel),
        (CTRLConfig, CTRLModel),
-        (ElectraConfig, ElectraModel),
    ]
 )

@@ -192,7 +182,6 @@ MODEL_FOR_PRETRAINING_MAPPING = OrderedDict(
        (FlaubertConfig, FlaubertWithLMHeadModel),
        (XLMConfig, XLMWithLMHeadModel),
        (CTRLConfig, CTRLLMHeadModel),
-        (ElectraConfig, ElectraForPreTraining),
    ]
 )

@@ -213,7 +202,6 @@ MODEL_WITH_LM_HEAD_MAPPING = OrderedDict(
        (FlaubertConfig, FlaubertWithLMHeadModel),
        (XLMConfig, XLMWithLMHeadModel),
        (CTRLConfig, CTRLLMHeadModel),
-        (ElectraConfig, ElectraForMaskedLM),
    ]
 )

@@ -254,7 +242,6 @@ MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING = OrderedDict(
        (BertConfig, BertForTokenClassification),
        (XLNetConfig, XLNetForTokenClassification),
        (AlbertConfig, AlbertForTokenClassification),
-        (ElectraConfig, ElectraForTokenClassification),
    ]
 )

@@ -294,8 +281,7 @@ 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` (Flaubert model)
-                - isInstance of `electra` configuration class: :class:`~transformers.ElectraModel` (Electra model)
+                - isInstance of `flaubert` configuration class: :class:`~transformers.FlaubertModel` (XLM model)

        Examples::

@@ -336,8 +322,7 @@ 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.FlaubertModel` (Flaubert  model)
-            - contains `electra`: :class:`~transformers.ElectraModel` (Electra  model)
+            - contains `flaubert`: :class:`~transformers.Flaubert` (Flaubert  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()`
@@ -445,7 +430,6 @@ 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::

@@ -486,7 +470,6 @@ 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()`
@@ -588,7 +571,6 @@ 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::

@@ -630,7 +612,6 @@ 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()`
@@ -1017,7 +998,6 @@ 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::

@@ -1055,7 +1035,6 @@ 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()`
@@ -440,7 +440,7 @@ class BartDecoder(nn.Module):
        # embed positions
        positions = self.embed_positions(input_ids, generation_mode=generation_mode)

-        if generation_mode:
+        if generation_mode and decoder_cached_states is not None:
            input_ids = input_ids[:, -1:]
            positions = positions[:, -1:]  # happens after we embed them
            assert input_ids.ne(self.padding_idx).any()
@@ -476,7 +476,7 @@ class BartDecoder(nn.Module):
                causal_mask=decoder_causal_mask,
            )

-            if self.output_past:
+            if generation_mode:
                next_decoder_cache.append(layer_past.copy())
            if self.output_hidden_states:
                all_hidden_states += (x,)
@@ -488,7 +488,7 @@ class BartDecoder(nn.Module):
        x = x.transpose(0, 1)
        encoder_hidden_states = encoder_hidden_states.transpose(0, 1)

-        if self.output_past:
+        if generation_mode:
            next_cache = ((encoder_hidden_states, encoder_padding_mask), next_decoder_cache)
        else:
            next_cache = None
@@ -907,23 +907,18 @@ class BartForConditionalGeneration(PretrainedBartModel):
    def prepare_inputs_for_generation(self, decoder_input_ids, past, attention_mask, **kwargs):
        assert past is not None, "past has to be defined for encoder_outputs"

-        # first step, decoder_cached_states are empty
-        if not past[1]:
-            encoder_outputs, decoder_cached_states = past, None
-        else:
-            encoder_outputs, decoder_cached_states = past
+        # first step, decoder_cached_states are empty (None)
+        (encoder_outputs, encoder_attention_mask), decoder_cached_states = past
        return {
            "input_ids": None,  # encoder_outputs is defined. input_ids not needed
            "encoder_outputs": encoder_outputs,
+            "attention_mask": encoder_attention_mask,
            "decoder_cached_states": decoder_cached_states,
            "decoder_input_ids": decoder_input_ids,
-            "attention_mask": attention_mask,
            "generation_mode": True,
        }

    def prepare_scores_for_generation(self, scores, cur_len, max_length):
-        if cur_len == 1:
-            self._force_token_ids_generation(scores, self.config.bos_token_id)
        if cur_len == max_length - 1 and self.config.eos_token_id is not None:
            self._force_token_ids_generation(scores, self.config.eos_token_id)
        return scores
@@ -931,15 +926,19 @@ class BartForConditionalGeneration(PretrainedBartModel):
    @staticmethod
    def _reorder_cache(past, beam_idx):
        ((enc_out, enc_mask), decoder_cached_states) = past
-        reordered_past = []
-        for layer_past in decoder_cached_states:
-            # get the correct batch idx from decoder layer's batch dim for cross and self-attn
-            layer_past_new = {
-                attn_key: _reorder_buffer(attn_cache, beam_idx) for attn_key, attn_cache in layer_past.items()
-            }
-            reordered_past.append(layer_past_new)

-        new_enc_out = enc_out if enc_out is None else enc_out.index_select(0, beam_idx)
+        if decoder_cached_states is not None:
+            reordered_past = []
+            for layer_past in decoder_cached_states:
+                # get the correct batch idx from decoder layer's batch dim for cross and self-attn
+                layer_past_new = {
+                    attn_key: _reorder_buffer(attn_cache, beam_idx) for attn_key, attn_cache in layer_past.items()
+                }
+                reordered_past.append(layer_past_new)
+        else:
+            reordered_past = None
+
+        new_enc_out = enc_out if enc_out is None else (enc_out[0].index_select(1, beam_idx), *enc_out[1:])
        new_enc_mask = enc_mask if enc_mask is None else enc_mask.index_select(0, beam_idx)

        past = ((new_enc_out, new_enc_mask), reordered_past)
@@ -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 and not hasattr(config, "embedding_size"):
+        if config.hidden_size % config.num_attention_heads != 0:
            raise ValueError(
                "The hidden size (%d) is not a multiple of the number of attention "
                "heads (%d)" % (config.hidden_size, config.num_attention_heads)
@@ -1,671 +0,0 @@
-import logging
-import os
-
-import torch
-import torch.nn as nn
-
-from transformers import ElectraConfig, add_start_docstrings
-from transformers.activations import get_activation
-
-from .file_utils import add_start_docstrings_to_callable
-from .modeling_bert import BertEmbeddings, BertEncoder, BertLayerNorm, BertPreTrainedModel
-
-
-logger = logging.getLogger(__name__)
-
-
-ELECTRA_PRETRAINED_MODEL_ARCHIVE_MAP = {
-    "google/electra-small-generator": "https://s3.amazonaws.com/models.huggingface.co/bert/google/electra-small-generator/pytorch_model.bin",
-    "google/electra-base-generator": "https://s3.amazonaws.com/models.huggingface.co/bert/google/electra-base-generator/pytorch_model.bin",
-    "google/electra-large-generator": "https://s3.amazonaws.com/models.huggingface.co/bert/google/electra-large-generator/pytorch_model.bin",
-    "google/electra-small-discriminator": "https://s3.amazonaws.com/models.huggingface.co/bert/google/electra-small-discriminator/pytorch_model.bin",
-    "google/electra-base-discriminator": "https://s3.amazonaws.com/models.huggingface.co/bert/google/electra-base-discriminator/pytorch_model.bin",
-    "google/electra-large-discriminator": "https://s3.amazonaws.com/models.huggingface.co/bert/google/electra-large-discriminator/pytorch_model.bin",
-}
-
-
-def load_tf_weights_in_electra(model, config, tf_checkpoint_path, discriminator_or_generator="discriminator"):
-    """ Load tf checkpoints in a pytorch model.
-    """
-    try:
-        import re
-        import numpy as np
-        import tensorflow as tf
-    except ImportError:
-        logger.error(
-            "Loading a TensorFlow model in PyTorch, requires TensorFlow to be installed. Please see "
-            "https://www.tensorflow.org/install/ for installation instructions."
-        )
-        raise
-    tf_path = os.path.abspath(tf_checkpoint_path)
-    logger.info("Converting TensorFlow checkpoint from {}".format(tf_path))
-    # Load weights from TF model
-    init_vars = tf.train.list_variables(tf_path)
-    names = []
-    arrays = []
-    for name, shape in init_vars:
-        logger.info("Loading TF weight {} with shape {}".format(name, shape))
-        array = tf.train.load_variable(tf_path, name)
-        names.append(name)
-        arrays.append(array)
-    for name, array in zip(names, arrays):
-        original_name: str = name
-
-        try:
-            if isinstance(model, ElectraForMaskedLM):
-                name = name.replace("electra/embeddings/", "generator/embeddings/")
-
-            if discriminator_or_generator == "generator":
-                name = name.replace("electra/", "discriminator/")
-                name = name.replace("generator/", "electra/")
-
-            name = name.replace("dense_1", "dense_prediction")
-            name = name.replace("generator_predictions/output_bias", "generator_lm_head/bias")
-
-            name = name.split("/")
-            # print(original_name, name)
-            # adam_v and adam_m are variables used in AdamWeightDecayOptimizer to calculated m and v
-            # which are not required for using pretrained model
-            if any(n in ["global_step", "temperature"] for n in name):
-                logger.info("Skipping {}".format(original_name))
-                continue
-            pointer = model
-            for m_name in name:
-                if re.fullmatch(r"[A-Za-z]+_\d+", m_name):
-                    scope_names = re.split(r"_(\d+)", m_name)
-                else:
-                    scope_names = [m_name]
-                if scope_names[0] == "kernel" or scope_names[0] == "gamma":
-                    pointer = getattr(pointer, "weight")
-                elif scope_names[0] == "output_bias" or scope_names[0] == "beta":
-                    pointer = getattr(pointer, "bias")
-                elif scope_names[0] == "output_weights":
-                    pointer = getattr(pointer, "weight")
-                elif scope_names[0] == "squad":
-                    pointer = getattr(pointer, "classifier")
-                else:
-                    pointer = getattr(pointer, scope_names[0])
-                if len(scope_names) >= 2:
-                    num = int(scope_names[1])
-                    pointer = pointer[num]
-            if m_name.endswith("_embeddings"):
-                pointer = getattr(pointer, "weight")
-            elif m_name == "kernel":
-                array = np.transpose(array)
-            try:
-                assert pointer.shape == array.shape, original_name
-            except AssertionError as e:
-                e.args += (pointer.shape, array.shape)
-                raise
-            print("Initialize PyTorch weight {}".format(name), original_name)
-            pointer.data = torch.from_numpy(array)
-        except AttributeError as e:
-            print("Skipping {}".format(original_name), name, e)
-            continue
-    return model
-
-
-class ElectraEmbeddings(BertEmbeddings):
-    """Construct the embeddings from word, position and token_type embeddings."""
-
-    def __init__(self, config):
-        super().__init__(config)
-        self.word_embeddings = nn.Embedding(config.vocab_size, config.embedding_size, padding_idx=config.pad_token_id)
-        self.position_embeddings = nn.Embedding(config.max_position_embeddings, config.embedding_size)
-        self.token_type_embeddings = nn.Embedding(config.type_vocab_size, config.embedding_size)
-
-        # self.LayerNorm is not snake-cased to stick with TensorFlow model variable name and be able to load
-        # any TensorFlow checkpoint file
-        self.LayerNorm = BertLayerNorm(config.embedding_size, eps=config.layer_norm_eps)
-
-
-class ElectraDiscriminatorPredictions(nn.Module):
-    """Prediction module for the discriminator, made up of two dense layers."""
-
-    def __init__(self, config):
-        super().__init__()
-
-        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
-        self.dense_prediction = nn.Linear(config.hidden_size, 1)
-        self.config = config
-
-    def forward(self, discriminator_hidden_states, attention_mask):
-        hidden_states = self.dense(discriminator_hidden_states)
-        hidden_states = get_activation(self.config.hidden_act)(hidden_states)
-        logits = self.dense_prediction(hidden_states).squeeze()
-
-        return logits
-
-
-class ElectraGeneratorPredictions(nn.Module):
-    """Prediction module for the generator, made up of two dense layers."""
-
-    def __init__(self, config):
-        super().__init__()
-
-        self.LayerNorm = BertLayerNorm(config.embedding_size)
-        self.dense = nn.Linear(config.hidden_size, config.embedding_size)
-
-    def forward(self, generator_hidden_states):
-        hidden_states = self.dense(generator_hidden_states)
-        hidden_states = get_activation("gelu")(hidden_states)
-        hidden_states = self.LayerNorm(hidden_states)
-
-        return hidden_states
-
-
-class ElectraPreTrainedModel(BertPreTrainedModel):
-    """ An abstract class to handle weights initialization and
-        a simple interface for downloading and loading pretrained models.
-    """
-
-    config_class = ElectraConfig
-    pretrained_model_archive_map = ELECTRA_PRETRAINED_MODEL_ARCHIVE_MAP
-    load_tf_weights = load_tf_weights_in_electra
-    base_model_prefix = "electra"
-
-    def get_extended_attention_mask(self, attention_mask, input_shape, device):
-        # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
-        # ourselves in which case we just need to make it broadcastable to all heads.
-        if attention_mask.dim() == 3:
-            extended_attention_mask = attention_mask[:, None, :, :]
-        elif attention_mask.dim() == 2:
-            # Provided a padding mask of dimensions [batch_size, seq_length]
-            # - if the model is a decoder, apply a causal mask in addition to the padding mask
-            # - if the model is an encoder, make the mask broadcastable to [batch_size, num_heads, seq_length, seq_length]
-            if self.config.is_decoder:
-                batch_size, seq_length = input_shape
-                seq_ids = torch.arange(seq_length, device=device)
-                causal_mask = seq_ids[None, None, :].repeat(batch_size, seq_length, 1) <= seq_ids[None, :, None]
-                causal_mask = causal_mask.to(
-                    attention_mask.dtype
-                )  # causal and attention masks must have same type with pytorch version < 1.3
-                extended_attention_mask = causal_mask[:, None, :, :] * attention_mask[:, None, None, :]
-            else:
-                extended_attention_mask = attention_mask[:, None, None, :]
-        else:
-            raise ValueError(
-                "Wrong shape for input_ids (shape {}) or attention_mask (shape {})".format(
-                    input_shape, attention_mask.shape
-                )
-            )
-
-        # Since attention_mask is 1.0 for positions we want to attend and 0.0 for
-        # masked positions, this operation will create a tensor which is 0.0 for
-        # positions we want to attend and -10000.0 for masked positions.
-        # Since we are adding it to the raw scores before the softmax, this is
-        # effectively the same as removing these entirely.
-        extended_attention_mask = extended_attention_mask.to(dtype=next(self.parameters()).dtype)  # fp16 compatibility
-        extended_attention_mask = (1.0 - extended_attention_mask) * -10000.0
-
-        return extended_attention_mask
-
-    def get_head_mask(self, head_mask):
-        # Prepare head mask if needed
-        # 1.0 in head_mask indicate we keep the head
-        # attention_probs has shape bsz x n_heads x N x N
-        # input head_mask has shape [num_heads] or [num_hidden_layers x num_heads]
-        # and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length]
-        num_hidden_layers = self.config.num_hidden_layers
-        if head_mask is not None:
-            if head_mask.dim() == 1:
-                head_mask = head_mask.unsqueeze(0).unsqueeze(0).unsqueeze(-1).unsqueeze(-1)
-                head_mask = head_mask.expand(num_hidden_layers, -1, -1, -1, -1)
-            elif head_mask.dim() == 2:
-                head_mask = (
-                    head_mask.unsqueeze(1).unsqueeze(-1).unsqueeze(-1)
-                )  # We can specify head_mask for each layer
-            head_mask = head_mask.to(
-                dtype=next(self.parameters()).dtype
-            )  # switch to fload if need + fp16 compatibility
-        else:
-            head_mask = [None] * num_hidden_layers
-
-        return head_mask
-
-
-ELECTRA_START_DOCSTRING = r"""
-    This model is a PyTorch `torch.nn.Module <https://pytorch.org/docs/stable/nn.html#torch.nn.Module>`_ sub-class.
-    Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general
-    usage and behavior.
-
-    Parameters:
-        config (:class:`~transformers.ElectraConfig`): 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.
-"""
-
-ELECTRA_INPUTS_DOCSTRING = r"""
-    Args:
-        input_ids (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`):
-            Indices of input sequence tokens in the vocabulary.
-
-            Indices can be obtained using :class:`transformers.ElectraTokenizer`.
-            See :func:`transformers.PreTrainedTokenizer.encode` and
-            :func:`transformers.PreTrainedTokenizer.encode_plus` for details.
-
-            `What are input IDs? <../glossary.html#input-ids>`__
-        attention_mask (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`):
-            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.
-
-            `What are attention masks? <../glossary.html#attention-mask>`__
-        token_type_ids (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`):
-            Segment token indices to indicate first and second portions of the inputs.
-            Indices are selected in ``[0, 1]``: ``0`` corresponds to a `sentence A` token, ``1``
-            corresponds to a `sentence B` token
-
-            `What are token type IDs? <../glossary.html#token-type-ids>`_
-        position_ids (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`):
-            Indices of positions of each input sequence tokens in the position embeddings.
-            Selected in the range ``[0, config.max_position_embeddings - 1]``.
-
-            `What are position IDs? <../glossary.html#position-ids>`_
-        head_mask (:obj:`torch.FloatTensor` of shape :obj:`(num_heads,)` or :obj:`(num_layers, num_heads)`, `optional`, defaults to :obj:`None`):
-            Mask to nullify selected heads of the self-attention modules.
-            Mask values selected in ``[0, 1]``:
-            :obj:`1` indicates the head is **not masked**, :obj:`0` indicates the head is **masked**.
-        inputs_embeds (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`, defaults to :obj:`None`):
-            Optionally, instead of passing :obj:`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.
-        encoder_hidden_states  (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`, defaults to :obj:`None`):
-            Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention
-            if the model is configured as a decoder.
-        encoder_attention_mask (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`):
-            Mask to avoid performing attention on the padding token indices of the encoder input. This mask
-            is used in the cross-attention if the model is configured as a decoder.
-            Mask values selected in ``[0, 1]``:
-            ``1`` for tokens that are NOT MASKED, ``0`` for MASKED tokens.
-"""
-
-
-@add_start_docstrings(
-    "The bare Electra Model transformer outputting raw hidden-states without any specific head on top. Identical to "
-    "the BERT model except that it uses an additional linear layer between the embedding layer and the encoder if the "
-    "hidden size and embedding size are different."
-    ""
-    "Both the generator and discriminator checkpoints may be loaded into this model.",
-    ELECTRA_START_DOCSTRING,
-)
-class ElectraModel(ElectraPreTrainedModel):
-
-    config_class = ElectraConfig
-
-    def __init__(self, config):
-        super().__init__(config)
-        self.embeddings = ElectraEmbeddings(config)
-
-        if config.embedding_size != config.hidden_size:
-            self.embeddings_project = nn.Linear(config.embedding_size, config.hidden_size)
-
-        self.encoder = BertEncoder(config)
-        self.config = config
-        self.init_weights()
-
-    def get_input_embeddings(self):
-        return self.embeddings.word_embeddings
-
-    def set_input_embeddings(self, value):
-        self.embeddings.word_embeddings = value
-
-    def _prune_heads(self, heads_to_prune):
-        """ Prunes heads of the model.
-            heads_to_prune: dict of {layer_num: list of heads to prune in this layer}
-            See base class PreTrainedModel
-        """
-        for layer, heads in heads_to_prune.items():
-            self.encoder.layer[layer].attention.prune_heads(heads)
-
-    @add_start_docstrings_to_callable(ELECTRA_INPUTS_DOCSTRING)
-    def forward(
-        self,
-        input_ids=None,
-        attention_mask=None,
-        token_type_ids=None,
-        position_ids=None,
-        head_mask=None,
-        inputs_embeds=None,
-    ):
-        r"""
-    Return:
-        :obj:`tuple(torch.FloatTensor)` comprising various elements depending on the configuration (:class:`~transformers.ElectraConfig`) and inputs:
-        last_hidden_state (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`):
-            Sequence of hidden-states at the output of the last layer of the model.
-        hidden_states (:obj:`tuple(torch.FloatTensor)`, `optional`, returned when ``config.output_hidden_states=True``):
-            Tuple of :obj:`torch.FloatTensor` (one for the output of the embeddings + one for the output of each layer)
-            of shape :obj:`(batch_size, sequence_length, hidden_size)`.
-
-            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
-        attentions (:obj:`tuple(torch.FloatTensor)`, `optional`, returned when ``config.output_attentions=True``):
-            Tuple of :obj:`torch.FloatTensor` (one for each layer) of shape
-            :obj:`(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::
-
-        from transformers import ElectraModel, ElectraTokenizer
-        import torch
-
-        tokenizer = ElectraTokenizer.from_pretrained('google/electra-small-discriminator')
-        model = ElectraModel.from_pretrained('google/electra-small-discriminator')
-
-        input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute", add_special_tokens=True)).unsqueeze(0)  # Batch size 1
-        outputs = model(input_ids)
-
-        last_hidden_states = outputs[0]  # The last hidden-state is the first element of the output tuple
-
-        """
-        if input_ids is not None and inputs_embeds is not None:
-            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
-        elif input_ids is not None:
-            input_shape = input_ids.size()
-        elif inputs_embeds is not None:
-            input_shape = inputs_embeds.size()[:-1]
-        else:
-            raise ValueError("You have to specify either input_ids or inputs_embeds")
-
-        device = input_ids.device if input_ids is not None else inputs_embeds.device
-
-        if attention_mask is None:
-            attention_mask = torch.ones(input_shape, device=device)
-        if token_type_ids is None:
-            token_type_ids = torch.zeros(input_shape, dtype=torch.long, device=device)
-
-        extended_attention_mask = self.get_extended_attention_mask(attention_mask, input_shape, device)
-        head_mask = self.get_head_mask(head_mask)
-
-        hidden_states = self.embeddings(
-            input_ids=input_ids, position_ids=position_ids, token_type_ids=token_type_ids, inputs_embeds=inputs_embeds
-        )
-
-        if hasattr(self, "embeddings_project"):
-            hidden_states = self.embeddings_project(hidden_states)
-
-        hidden_states = self.encoder(hidden_states, attention_mask=extended_attention_mask, head_mask=head_mask)
-
-        return hidden_states
-
-
-@add_start_docstrings(
-    """
-    Electra model with a binary classification head on top as used during pre-training for identifying generated
-    tokens.
-
-    It is recommended to load the discriminator checkpoint into that model.""",
-    ELECTRA_START_DOCSTRING,
-)
-class ElectraForPreTraining(ElectraPreTrainedModel):
-    def __init__(self, config):
-        super().__init__(config)
-
-        self.electra = ElectraModel(config)
-        self.discriminator_predictions = ElectraDiscriminatorPredictions(config)
-        self.init_weights()
-
-    @add_start_docstrings_to_callable(ELECTRA_INPUTS_DOCSTRING)
-    def forward(
-        self,
-        input_ids=None,
-        attention_mask=None,
-        token_type_ids=None,
-        position_ids=None,
-        head_mask=None,
-        inputs_embeds=None,
-        labels=None,
-    ):
-        r"""
-        labels (``torch.LongTensor`` of shape ``(batch_size, sequence_length)``, `optional`, defaults to :obj:`None`):
-            Labels for computing the ELECTRA loss. Input should be a sequence of tokens (see :obj:`input_ids` docstring)
-            Indices should be in ``[0, 1]``.
-            ``0`` indicates the token is an original token,
-            ``1`` indicates the token was replaced.
-
-    Returns:
-        :obj:`tuple(torch.FloatTensor)` comprising various elements depending on the configuration (:class:`~transformers.ElectraConfig`) and inputs:
-        loss (`optional`, returned when ``labels`` is provided) ``torch.FloatTensor`` of shape ``(1,)``:
-            Total loss of the ELECTRA objective.
-        scores (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length)`)
-            Prediction scores of the head (scores for each token before SoftMax).
-        hidden_states (:obj:`tuple(torch.FloatTensor)`, `optional`, returned when :obj:`config.output_hidden_states=True`):
-            Tuple of :obj:`torch.FloatTensor` (one for the output of the embeddings + one for the output of each layer)
-            of shape :obj:`(batch_size, sequence_length, hidden_size)`.
-
-            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
-        attentions (:obj:`tuple(torch.FloatTensor)`, `optional`, returned when ``config.output_attentions=True``):
-            Tuple of :obj:`torch.FloatTensor` (one for each layer) of shape
-            :obj:`(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::
-
-        from transformers import ElectraTokenizer, ElectraForPreTraining
-        import torch
-
-        tokenizer = ElectraTokenizer.from_pretrained('google/electra-small-discriminator')
-        model = ElectraForPreTraining.from_pretrained('google/electra-small-discriminator')
-
-        input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute", add_special_tokens=True)).unsqueeze(0)  # Batch size 1
-        outputs = model(input_ids)
-
-        prediction_scores, seq_relationship_scores = outputs[:2]
-
-        """
-
-        discriminator_hidden_states = self.electra(
-            input_ids, attention_mask, token_type_ids, position_ids, head_mask, inputs_embeds
-        )
-        discriminator_sequence_output = discriminator_hidden_states[0]
-
-        logits = self.discriminator_predictions(discriminator_sequence_output, attention_mask)
-
-        output = (logits,)
-
-        if labels is not None:
-            loss_fct = nn.BCEWithLogitsLoss()
-            if attention_mask is not None:
-                active_loss = attention_mask.view(-1, discriminator_sequence_output.shape[1]) == 1
-                active_logits = logits.view(-1, discriminator_sequence_output.shape[1])[active_loss]
-                active_labels = labels[active_loss]
-                loss = loss_fct(active_logits, active_labels.float())
-            else:
-                loss = loss_fct(logits.view(-1, discriminator_sequence_output.shape[1]), labels.float())
-
-            output = (loss,) + output
-
-        output += discriminator_hidden_states[1:]
-
-        return output  # (loss), scores, (hidden_states), (attentions)
-
-
-@add_start_docstrings(
-    """
-    Electra model with a language modeling head on top.
-
-    Even though both the discriminator and generator may be loaded into this model, the generator is
-    the only model of the two to have been trained for the masked language modeling task.""",
-    ELECTRA_START_DOCSTRING,
-)
-class ElectraForMaskedLM(ElectraPreTrainedModel):
-    def __init__(self, config):
-        super().__init__(config)
-
-        self.electra = ElectraModel(config)
-        self.generator_predictions = ElectraGeneratorPredictions(config)
-
-        self.generator_lm_head = nn.Linear(config.embedding_size, config.vocab_size)
-        self.init_weights()
-
-    def get_output_embeddings(self):
-        return self.generator_lm_head
-
-    @add_start_docstrings_to_callable(ELECTRA_INPUTS_DOCSTRING)
-    def forward(
-        self,
-        input_ids=None,
-        attention_mask=None,
-        token_type_ids=None,
-        position_ids=None,
-        head_mask=None,
-        inputs_embeds=None,
-        masked_lm_labels=None,
-    ):
-        r"""
-        masked_lm_labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`):
-            Labels for computing the masked language modeling loss.
-            Indices should be in ``[-100, 0, ..., config.vocab_size]`` (see ``input_ids`` docstring)
-            Tokens with indices set to ``-100`` are ignored (masked), the loss is only computed for the tokens with labels
-            in ``[0, ..., config.vocab_size]``
-
-    Returns:
-        :obj:`tuple(torch.FloatTensor)` comprising various elements depending on the configuration (:class:`~transformers.ElectraConfig`) and inputs:
-        masked_lm_loss (`optional`, returned when ``masked_lm_labels`` is provided) ``torch.FloatTensor`` of shape ``(1,)``:
-            Masked language modeling loss.
-        prediction_scores (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, config.vocab_size)`)
-            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
-        hidden_states (:obj:`tuple(torch.FloatTensor)`, `optional`, returned when ``config.output_hidden_states=True``):
-            Tuple of :obj:`torch.FloatTensor` (one for the output of the embeddings + one for the output of each layer)
-            of shape :obj:`(batch_size, sequence_length, hidden_size)`.
-
-            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
-        attentions (:obj:`tuple(torch.FloatTensor)`, `optional`, returned when ``config.output_attentions=True``):
-            Tuple of :obj:`torch.FloatTensor` (one for each layer) of shape
-            :obj:`(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::
-
-            from transformers import ElectraTokenizer, ElectraForMaskedLM
-            import torch
-
-            tokenizer = ElectraTokenizer.from_pretrained('google/electra-small-generator')
-            model = ElectraForMaskedLM.from_pretrained('google/electra-small-generator')
-
-            input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute", add_special_tokens=True)).unsqueeze(0)  # Batch size 1
-            outputs = model(input_ids, masked_lm_labels=input_ids)
-
-            loss, prediction_scores = outputs[:2]
-
-        """
-
-        generator_hidden_states = self.electra(
-            input_ids, attention_mask, token_type_ids, position_ids, head_mask, inputs_embeds
-        )
-        generator_sequence_output = generator_hidden_states[0]
-
-        prediction_scores = self.generator_predictions(generator_sequence_output)
-        prediction_scores = self.generator_lm_head(prediction_scores)
-
-        output = (prediction_scores,)
-
-        # Masked language modeling softmax layer
-        if masked_lm_labels is not None:
-            loss_fct = nn.CrossEntropyLoss()  # -100 index = padding token
-            loss = loss_fct(prediction_scores.view(-1, self.config.vocab_size), masked_lm_labels.view(-1))
-            output = (loss,) + output
-
-        output += generator_hidden_states[1:]
-
-        return output  # (masked_lm_loss), prediction_scores, (hidden_states), (attentions)
-
-
-@add_start_docstrings(
-    """
-    Electra model with a token classification head on top.
-
-    Both the discriminator and generator may be loaded into this model.""",
-    ELECTRA_START_DOCSTRING,
-)
-class ElectraForTokenClassification(ElectraPreTrainedModel):
-    def __init__(self, config):
-        super().__init__(config)
-
-        self.electra = ElectraModel(config)
-        self.dropout = nn.Dropout(config.hidden_dropout_prob)
-        self.classifier = nn.Linear(config.hidden_size, config.num_labels)
-        self.init_weights()
-
-    @add_start_docstrings_to_callable(ELECTRA_INPUTS_DOCSTRING)
-    def forward(
-        self,
-        input_ids=None,
-        attention_mask=None,
-        token_type_ids=None,
-        position_ids=None,
-        head_mask=None,
-        inputs_embeds=None,
-        labels=None,
-    ):
-        r"""
-        labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`):
-            Labels for computing the token classification loss.
-            Indices should be in ``[0, ..., config.num_labels - 1]``.
-
-    Returns:
-        :obj:`tuple(torch.FloatTensor)` comprising various elements depending on the configuration (:class:`~transformers.ElectraConfig`) and inputs:
-        loss (:obj:`torch.FloatTensor` of shape :obj:`(1,)`, `optional`, returned when ``labels`` is provided) :
-            Classification loss.
-        scores (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, config.num_labels)`)
-            Classification scores (before SoftMax).
-        hidden_states (:obj:`tuple(torch.FloatTensor)`, `optional`, returned when ``config.output_hidden_states=True``):
-            Tuple of :obj:`torch.FloatTensor` (one for the output of the embeddings + one for the output of each layer)
-            of shape :obj:`(batch_size, sequence_length, hidden_size)`.
-
-            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
-        attentions (:obj:`tuple(torch.FloatTensor)`, `optional`, returned when ``config.output_attentions=True``):
-            Tuple of :obj:`torch.FloatTensor` (one for each layer) of shape
-            :obj:`(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::
-
-        from transformers import ElectraTokenizer, ElectraForTokenClassification
-        import torch
-
-        tokenizer = ElectraTokenizer.from_pretrained('google/electra-small-discriminator')
-        model = ElectraForTokenClassification.from_pretrained('google/electra-small-discriminator')
-
-        input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute", add_special_tokens=True)).unsqueeze(0)  # Batch size 1
-        labels = torch.tensor([1] * input_ids.size(1)).unsqueeze(0)  # Batch size 1
-        outputs = model(input_ids, labels=labels)
-
-        loss, scores = outputs[:2]
-
-        """
-
-        discriminator_hidden_states = self.electra(
-            input_ids, attention_mask, token_type_ids, position_ids, head_mask, inputs_embeds
-        )
-        discriminator_sequence_output = discriminator_hidden_states[0]
-
-        discriminator_sequence_output = self.dropout(discriminator_sequence_output)
-        logits = self.classifier(discriminator_sequence_output)
-
-        output = (logits,)
-
-        if labels is not None:
-            loss_fct = nn.CrossEntropyLoss()
-            # Only keep active parts of the loss
-            if attention_mask is not None:
-                active_loss = attention_mask.view(-1) == 1
-                active_logits = logits.view(-1, self.config.num_labels)[active_loss]
-                active_labels = labels.view(-1)[active_loss]
-                loss = loss_fct(active_logits, active_labels)
-            else:
-                loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
-
-            output = (loss,) + output
-
-        output += discriminator_hidden_states[1:]
-
-        return output  # (loss), scores, (hidden_states), (attentions)
@@ -1,615 +0,0 @@
-import logging
-
-import tensorflow as tf
-
-from transformers import ElectraConfig
-
-from .file_utils import add_start_docstrings, add_start_docstrings_to_callable
-from .modeling_tf_bert import ACT2FN, TFBertEncoder, TFBertPreTrainedModel
-from .modeling_tf_utils import get_initializer, shape_list
-
-
-logger = logging.getLogger(__name__)
-
-
-TF_ELECTRA_PRETRAINED_MODEL_ARCHIVE_MAP = {
-    "google/electra-small-generator": "https://s3.amazonaws.com/models.huggingface.co/bert/google/electra-small-generator/tf_model.h5",
-    "google/electra-base-generator": "https://s3.amazonaws.com/models.huggingface.co/bert/google/electra-base-generator/tf_model.h5",
-    "google/electra-large-generator": "https://s3.amazonaws.com/models.huggingface.co/bert/google/electra-large-generator/tf_model.h5",
-    "google/electra-small-discriminator": "https://s3.amazonaws.com/models.huggingface.co/bert/google/electra-small-discriminator/tf_model.h5",
-    "google/electra-base-discriminator": "https://s3.amazonaws.com/models.huggingface.co/bert/google/electra-base-discriminator/tf_model.h5",
-    "google/electra-large-discriminator": "https://s3.amazonaws.com/models.huggingface.co/bert/google/electra-large-discriminator/tf_model.h5",
-}
-
-
-class TFElectraEmbeddings(tf.keras.layers.Layer):
-    """Construct the embeddings from word, position and token_type embeddings.
-    """
-
-    def __init__(self, config, **kwargs):
-        super().__init__(**kwargs)
-        self.vocab_size = config.vocab_size
-        self.embedding_size = config.embedding_size
-        self.initializer_range = config.initializer_range
-
-        self.position_embeddings = tf.keras.layers.Embedding(
-            config.max_position_embeddings,
-            config.embedding_size,
-            embeddings_initializer=get_initializer(self.initializer_range),
-            name="position_embeddings",
-        )
-        self.token_type_embeddings = tf.keras.layers.Embedding(
-            config.type_vocab_size,
-            config.embedding_size,
-            embeddings_initializer=get_initializer(self.initializer_range),
-            name="token_type_embeddings",
-        )
-
-        # self.LayerNorm is not snake-cased to stick with TensorFlow model variable name and be able to load
-        # any TensorFlow checkpoint file
-        self.LayerNorm = tf.keras.layers.LayerNormalization(epsilon=config.layer_norm_eps, name="LayerNorm")
-        self.dropout = tf.keras.layers.Dropout(config.hidden_dropout_prob)
-
-    def build(self, input_shape):
-        """Build shared word embedding layer """
-        with tf.name_scope("word_embeddings"):
-            # Create and initialize weights. The random normal initializer was chosen
-            # arbitrarily, and works well.
-            self.word_embeddings = self.add_weight(
-                "weight",
-                shape=[self.vocab_size, self.embedding_size],
-                initializer=get_initializer(self.initializer_range),
-            )
-        super().build(input_shape)
-
-    def call(self, inputs, mode="embedding", training=False):
-        """Get token embeddings of inputs.
-        Args:
-            inputs: list of three int64 tensors with shape [batch_size, length]: (input_ids, position_ids, token_type_ids)
-            mode: string, a valid value is one of "embedding" and "linear".
-        Returns:
-            outputs: (1) If mode == "embedding", output embedding tensor, float32 with
-                shape [batch_size, length, embedding_size]; (2) mode == "linear", output
-                linear tensor, float32 with shape [batch_size, length, vocab_size].
-        Raises:
-            ValueError: if mode is not valid.
-
-        Shared weights logic adapted from
-            https://github.com/tensorflow/models/blob/a009f4fb9d2fc4949e32192a944688925ef78659/official/transformer/v2/embedding_layer.py#L24
-        """
-        if mode == "embedding":
-            return self._embedding(inputs, training=training)
-        elif mode == "linear":
-            return self._linear(inputs)
-        else:
-            raise ValueError("mode {} is not valid.".format(mode))
-
-    def _embedding(self, inputs, training=False):
-        """Applies embedding based on inputs tensor."""
-        input_ids, position_ids, token_type_ids, inputs_embeds = inputs
-
-        if input_ids is not None:
-            input_shape = shape_list(input_ids)
-        else:
-            input_shape = shape_list(inputs_embeds)[:-1]
-
-        seq_length = input_shape[1]
-        if position_ids is None:
-            position_ids = tf.range(seq_length, dtype=tf.int32)[tf.newaxis, :]
-        if token_type_ids is None:
-            token_type_ids = tf.fill(input_shape, 0)
-
-        if inputs_embeds is None:
-            inputs_embeds = tf.gather(self.word_embeddings, input_ids)
-        position_embeddings = self.position_embeddings(position_ids)
-        token_type_embeddings = self.token_type_embeddings(token_type_ids)
-
-        embeddings = inputs_embeds + position_embeddings + token_type_embeddings
-        embeddings = self.LayerNorm(embeddings)
-        embeddings = self.dropout(embeddings, training=training)
-        return embeddings
-
-    def _linear(self, inputs):
-        """Computes logits by running inputs through a linear layer.
-            Args:
-                inputs: A float32 tensor with shape [batch_size, length, hidden_size]
-            Returns:
-                float32 tensor with shape [batch_size, length, vocab_size].
-        """
-        batch_size = shape_list(inputs)[0]
-        length = shape_list(inputs)[1]
-
-        x = tf.reshape(inputs, [-1, self.embedding_size])
-        logits = tf.matmul(x, self.word_embeddings, transpose_b=True)
-
-        return tf.reshape(logits, [batch_size, length, self.vocab_size])
-
-
-class TFElectraDiscriminatorPredictions(tf.keras.layers.Layer):
-    def __init__(self, config, **kwargs):
-        super().__init__(**kwargs)
-
-        self.dense = tf.keras.layers.Dense(config.hidden_size, name="dense")
-        self.dense_prediction = tf.keras.layers.Dense(1, name="dense_prediction")
-        self.config = config
-
-    def call(self, discriminator_hidden_states, training=False):
-        hidden_states = self.dense(discriminator_hidden_states)
-        hidden_states = ACT2FN[self.config.hidden_act](hidden_states)
-        logits = tf.squeeze(self.dense_prediction(hidden_states))
-
-        return logits
-
-
-class TFElectraGeneratorPredictions(tf.keras.layers.Layer):
-    def __init__(self, config, **kwargs):
-        super().__init__(**kwargs)
-
-        self.LayerNorm = tf.keras.layers.LayerNormalization(epsilon=config.layer_norm_eps, name="LayerNorm")
-        self.dense = tf.keras.layers.Dense(config.embedding_size, name="dense")
-
-    def call(self, generator_hidden_states, training=False):
-        hidden_states = self.dense(generator_hidden_states)
-        hidden_states = ACT2FN["gelu"](hidden_states)
-        hidden_states = self.LayerNorm(hidden_states)
-
-        return hidden_states
-
-
-class TFElectraPreTrainedModel(TFBertPreTrainedModel):
-
-    config_class = ElectraConfig
-    pretrained_model_archive_map = TF_ELECTRA_PRETRAINED_MODEL_ARCHIVE_MAP
-    base_model_prefix = "electra"
-
-    def get_extended_attention_mask(self, attention_mask, input_shape):
-        if attention_mask is None:
-            attention_mask = tf.fill(input_shape, 1)
-
-        # We create a 3D attention mask from a 2D tensor mask.
-        # Sizes are [batch_size, 1, 1, to_seq_length]
-        # So we can broadcast to [batch_size, num_heads, from_seq_length, to_seq_length]
-        # this attention mask is more simple than the triangular masking of causal attention
-        # used in OpenAI GPT, we just need to prepare the broadcast dimension here.
-        extended_attention_mask = attention_mask[:, tf.newaxis, tf.newaxis, :]
-
-        # Since attention_mask is 1.0 for positions we want to attend and 0.0 for
-        # masked positions, this operation will create a tensor which is 0.0 for
-        # positions we want to attend and -10000.0 for masked positions.
-        # Since we are adding it to the raw scores before the softmax, this is
-        # effectively the same as removing these entirely.
-
-        extended_attention_mask = tf.cast(extended_attention_mask, tf.float32)
-        extended_attention_mask = (1.0 - extended_attention_mask) * -10000.0
-
-        return extended_attention_mask
-
-    def get_head_mask(self, head_mask):
-        if head_mask is not None:
-            raise NotImplementedError
-        else:
-            head_mask = [None] * self.config.num_hidden_layers
-
-        return head_mask
-
-
-class TFElectraMainLayer(TFElectraPreTrainedModel):
-
-    config_class = ElectraConfig
-
-    def __init__(self, config, **kwargs):
-        super().__init__(config, **kwargs)
-        self.embeddings = TFElectraEmbeddings(config, name="embeddings")
-
-        if config.embedding_size != config.hidden_size:
-            self.embeddings_project = tf.keras.layers.Dense(config.hidden_size, name="embeddings_project")
-        self.encoder = TFBertEncoder(config, name="encoder")
-        self.config = config
-
-    def get_input_embeddings(self):
-        return self.embeddings
-
-    def _resize_token_embeddings(self, new_num_tokens):
-        raise NotImplementedError
-
-    def _prune_heads(self, heads_to_prune):
-        """ Prunes heads of the model.
-            heads_to_prune: dict of {layer_num: list of heads to prune in this layer}
-            See base class PreTrainedModel
-        """
-        raise NotImplementedError
-
-    def call(
-        self,
-        inputs,
-        attention_mask=None,
-        token_type_ids=None,
-        position_ids=None,
-        head_mask=None,
-        inputs_embeds=None,
-        training=False,
-    ):
-        if isinstance(inputs, (tuple, list)):
-            input_ids = inputs[0]
-            attention_mask = inputs[1] if len(inputs) > 1 else attention_mask
-            token_type_ids = inputs[2] if len(inputs) > 2 else token_type_ids
-            position_ids = inputs[3] if len(inputs) > 3 else position_ids
-            head_mask = inputs[4] if len(inputs) > 4 else head_mask
-            inputs_embeds = inputs[5] if len(inputs) > 5 else inputs_embeds
-            assert len(inputs) <= 6, "Too many inputs."
-        elif isinstance(inputs, dict):
-            input_ids = inputs.get("input_ids")
-            attention_mask = inputs.get("attention_mask", attention_mask)
-            token_type_ids = inputs.get("token_type_ids", token_type_ids)
-            position_ids = inputs.get("position_ids", position_ids)
-            head_mask = inputs.get("head_mask", head_mask)
-            inputs_embeds = inputs.get("inputs_embeds", inputs_embeds)
-            assert len(inputs) <= 6, "Too many inputs."
-        else:
-            input_ids = inputs
-
-        if input_ids is not None and inputs_embeds is not None:
-            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
-        elif input_ids is not None:
-            input_shape = shape_list(input_ids)
-        elif inputs_embeds is not None:
-            input_shape = shape_list(inputs_embeds)[:-1]
-        else:
-            raise ValueError("You have to specify either input_ids or inputs_embeds")
-
-        if attention_mask is None:
-            attention_mask = tf.fill(input_shape, 1)
-        if token_type_ids is None:
-            token_type_ids = tf.fill(input_shape, 0)
-
-        extended_attention_mask = self.get_extended_attention_mask(attention_mask, input_shape)
-        head_mask = self.get_head_mask(head_mask)
-
-        hidden_states = self.embeddings([input_ids, position_ids, token_type_ids, inputs_embeds], training=training)
-
-        if hasattr(self, "embeddings_project"):
-            hidden_states = self.embeddings_project(hidden_states, training=training)
-
-        hidden_states = self.encoder([hidden_states, extended_attention_mask, head_mask], training=training)
-
-        return hidden_states
-
-
-ELECTRA_START_DOCSTRING = r"""
-    This model is a `tf.keras.Model <https://www.tensorflow.org/api_docs/python/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.
-
-    .. note::
-
-        TF 2.0 models accepts two formats as inputs:
-
-            - having all inputs as keyword arguments (like PyTorch models), or
-            - having all inputs as a list, tuple or dict in the first positional arguments.
-
-        This second option is useful when using :obj:`tf.keras.Model.fit()` method which currently requires having
-        all the tensors in the first argument of the model call function: :obj:`model(inputs)`.
-
-        If you choose this second option, there are three possibilities you can use to gather all the input Tensors
-        in the first positional argument :
-
-        - a single Tensor with input_ids only and nothing else: :obj:`model(inputs_ids)`
-        - a list of varying length with one or several input Tensors IN THE ORDER given in the docstring:
-          :obj:`model([input_ids, attention_mask])` or :obj:`model([input_ids, attention_mask, token_type_ids])`
-        - a dictionary with one or several input Tensors associated to the input names given in the docstring:
-          :obj:`model({'input_ids': input_ids, 'token_type_ids': token_type_ids})`
-
-    Parameters:
-        config (:class:`~transformers.ElectraConfig`): 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.
-"""
-
-ELECTRA_INPUTS_DOCSTRING = r"""
-    Args:
-        input_ids (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length)`):
-            Indices of input sequence tokens in the vocabulary.
-
-            Indices can be obtained using :class:`transformers.ElectraTokenizer`.
-            See :func:`transformers.PreTrainedTokenizer.encode` and
-            :func:`transformers.PreTrainedTokenizer.encode_plus` for details.
-
-            `What are input IDs? <../glossary.html#input-ids>`__
-        attention_mask (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`):
-            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.
-
-            `What are attention masks? <../glossary.html#attention-mask>`__
-        head_mask (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(num_heads,)` or :obj:`(num_layers, num_heads)`, `optional`, defaults to :obj:`None`):
-            Mask to nullify selected heads of the self-attention modules.
-            Mask values selected in ``[0, 1]``:
-            :obj:`1` indicates the head is **not masked**, :obj:`0` indicates the head is **masked**.
-        inputs_embeds (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length, embedding_dim)`, `optional`, defaults to :obj:`None`):
-            Optionally, instead of passing :obj:`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.
-        training (:obj:`boolean`, `optional`, defaults to :obj:`False`):
-            Whether to activate dropout modules (if set to :obj:`True`) during training or to de-activate them
-            (if set to :obj:`False`) for evaluation.
-
-"""
-
-
-@add_start_docstrings(
-    "The bare Electra Model transformer outputting raw hidden-states without any specific head on top. Identical to "
-    "the BERT model except that it uses an additional linear layer between the embedding layer and the encoder if the "
-    "hidden size and embedding size are different."
-    ""
-    "Both the generator and discriminator checkpoints may be loaded into this model.",
-    ELECTRA_START_DOCSTRING,
-)
-class TFElectraModel(TFElectraPreTrainedModel):
-    def __init__(self, config, *inputs, **kwargs):
-        super().__init__(config, *inputs, **kwargs)
-        self.electra = TFElectraMainLayer(config, name="electra")
-
-    def get_input_embeddings(self):
-        return self.electra.embeddings
-
-    @add_start_docstrings_to_callable(ELECTRA_INPUTS_DOCSTRING)
-    def call(self, inputs, **kwargs):
-        r"""
-    Returns:
-        :obj:`tuple(torch.FloatTensor)` comprising various elements depending on the configuration (:class:`~transformers.ElectraConfig`) and inputs:
-        last_hidden_state (:obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`):
-            Sequence of hidden-states at the output of the last layer of the model.
-        hidden_states (:obj:`tuple(tf.Tensor)`, `optional`, returned when :obj:`config.output_hidden_states=True`):
-            tuple of :obj:`tf.Tensor` (one for the output of the embeddings + one for the output of each layer)
-            of shape :obj:`(batch_size, sequence_length, hidden_size)`.
-
-            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
-        attentions (:obj:`tuple(tf.Tensor)`, `optional`, returned when ``config.output_attentions=True``):
-            tuple of :obj:`tf.Tensor` (one for each layer) of shape
-            :obj:`(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::
-
-        import tensorflow as tf
-        from transformers import ElectraTokenizer, TFElectraModel
-
-        tokenizer = ElectraTokenizer.from_pretrained('google/electra-small-discriminator')
-        model = TFElectraModel.from_pretrained('google/electra-small-discriminator')
-        input_ids = tf.constant(tokenizer.encode("Hello, my dog is cute"))[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
-        """
-        outputs = self.electra(inputs, **kwargs)
-        return outputs
-
-
-@add_start_docstrings(
-    """
-Electra model with a binary classification head on top as used during pre-training for identifying generated
-tokens.
-
-Even though both the discriminator and generator may be loaded into this model, the discriminator is
-the only model of the two to have the correct classification head to be used for this model.""",
-    ELECTRA_START_DOCSTRING,
-)
-class TFElectraForPreTraining(TFElectraPreTrainedModel):
-    def __init__(self, config, **kwargs):
-        super().__init__(config, **kwargs)
-
-        self.electra = TFElectraMainLayer(config, name="electra")
-        self.discriminator_predictions = TFElectraDiscriminatorPredictions(config, name="discriminator_predictions")
-
-    def get_input_embeddings(self):
-        return self.electra.embeddings
-
-    @add_start_docstrings_to_callable(ELECTRA_INPUTS_DOCSTRING)
-    def call(
-        self,
-        input_ids=None,
-        attention_mask=None,
-        token_type_ids=None,
-        position_ids=None,
-        head_mask=None,
-        inputs_embeds=None,
-        training=False,
-    ):
-        r"""
-    Returns:
-        :obj:`tuple(torch.FloatTensor)` comprising various elements depending on the configuration (:class:`~transformers.ElectraConfig`) and inputs:
-        scores (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length, config.num_labels)`):
-            Prediction scores of the head (scores for each token before SoftMax).
-        hidden_states (:obj:`tuple(tf.Tensor)`, `optional`, returned when :obj:`config.output_hidden_states=True`):
-            tuple of :obj:`tf.Tensor` (one for the output of the embeddings + one for the output of each layer)
-            of shape :obj:`(batch_size, sequence_length, hidden_size)`.
-
-            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
-        attentions (:obj:`tuple(tf.Tensor)`, `optional`, returned when ``config.output_attentions=True``):
-            tuple of :obj:`tf.Tensor` (one for each layer) of shape
-            :obj:`(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::
-
-        import tensorflow as tf
-        from transformers import ElectraTokenizer, TFElectraForPreTraining
-
-        tokenizer = ElectraTokenizer.from_pretrained('google/electra-small-discriminator')
-        model = TFElectraForPreTraining.from_pretrained('google/electra-small-discriminator')
-        input_ids = tf.constant(tokenizer.encode("Hello, my dog is cute"))[None, :]  # Batch size 1
-        outputs = model(input_ids)
-        scores = outputs[0]
-        """
-
-        discriminator_hidden_states = self.electra(
-            input_ids, attention_mask, token_type_ids, position_ids, head_mask, inputs_embeds, training=training
-        )
-        discriminator_sequence_output = discriminator_hidden_states[0]
-        logits = self.discriminator_predictions(discriminator_sequence_output)
-        output = (logits,)
-        output += discriminator_hidden_states[1:]
-
-        return output  # (loss), scores, (hidden_states), (attentions)
-
-
-class TFElectraMaskedLMHead(tf.keras.layers.Layer):
-    def __init__(self, config, input_embeddings, **kwargs):
-        super().__init__(**kwargs)
-        self.vocab_size = config.vocab_size
-        self.input_embeddings = input_embeddings
-
-    def build(self, input_shape):
-        self.bias = self.add_weight(shape=(self.vocab_size,), initializer="zeros", trainable=True, name="bias")
-        super().build(input_shape)
-
-    def call(self, hidden_states, training=False):
-        hidden_states = self.input_embeddings(hidden_states, mode="linear")
-        hidden_states = hidden_states + self.bias
-        return hidden_states
-
-
-@add_start_docstrings(
-    """
-Electra model with a language modeling head on top.
-
-Even though both the discriminator and generator may be loaded into this model, the generator is
-the only model of the two to have been trained for the masked language modeling task.""",
-    ELECTRA_START_DOCSTRING,
-)
-class TFElectraForMaskedLM(TFElectraPreTrainedModel):
-    def __init__(self, config, **kwargs):
-        super().__init__(config, **kwargs)
-
-        self.vocab_size = config.vocab_size
-        self.electra = TFElectraMainLayer(config, name="electra")
-        self.generator_predictions = TFElectraGeneratorPredictions(config, name="generator_predictions")
-        if isinstance(config.hidden_act, str):
-            self.activation = ACT2FN[config.hidden_act]
-        else:
-            self.activation = config.hidden_act
-        self.generator_lm_head = TFElectraMaskedLMHead(config, self.electra.embeddings, name="generator_lm_head")
-
-    def get_input_embeddings(self):
-        return self.electra.embeddings
-
-    def get_output_embeddings(self):
-        return self.generator_lm_head
-
-    @add_start_docstrings_to_callable(ELECTRA_INPUTS_DOCSTRING)
-    def call(
-        self,
-        input_ids=None,
-        attention_mask=None,
-        token_type_ids=None,
-        position_ids=None,
-        head_mask=None,
-        inputs_embeds=None,
-        training=False,
-    ):
-        r"""
-    Returns:
-        :obj:`tuple(torch.FloatTensor)` comprising various elements depending on the configuration (:class:`~transformers.ElectraConfig`) and inputs:
-        prediction_scores (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length, config.vocab_size)`):
-            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
-        hidden_states (:obj:`tuple(tf.Tensor)`, `optional`, returned when :obj:`config.output_hidden_states=True`):
-            tuple of :obj:`tf.Tensor` (one for the output of the embeddings + one for the output of each layer)
-            of shape :obj:`(batch_size, sequence_length, hidden_size)`.
-
-            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
-        attentions (:obj:`tuple(tf.Tensor)`, `optional`, returned when ``config.output_attentions=True``):
-            tuple of :obj:`tf.Tensor` (one for each layer) of shape
-            :obj:`(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::
-
-        import tensorflow as tf
-        from transformers import ElectraTokenizer, TFElectraForMaskedLM
-
-        tokenizer = ElectraTokenizer.from_pretrained('google/electra-small-generator')
-        model = TFElectraForMaskedLM.from_pretrained('google/electra-small-generator')
-        input_ids = tf.constant(tokenizer.encode("Hello, my dog is cute"))[None, :]  # Batch size 1
-        outputs = model(input_ids)
-        prediction_scores = outputs[0]
-
-        """
-
-        generator_hidden_states = self.electra(
-            input_ids, attention_mask, token_type_ids, position_ids, head_mask, inputs_embeds, training=training
-        )
-        generator_sequence_output = generator_hidden_states[0]
-        prediction_scores = self.generator_predictions(generator_sequence_output, training=training)
-        prediction_scores = self.generator_lm_head(prediction_scores, training=training)
-        output = (prediction_scores,)
-        output += generator_hidden_states[1:]
-
-        return output  # (masked_lm_loss), prediction_scores, (hidden_states), (attentions)
-
-
-@add_start_docstrings(
-    """
-Electra model with a token classification head on top.
-
-Both the discriminator and generator may be loaded into this model.""",
-    ELECTRA_START_DOCSTRING,
-)
-class TFElectraForTokenClassification(TFElectraPreTrainedModel):
-    def __init__(self, config, **kwargs):
-        super().__init__(config, **kwargs)
-
-        self.electra = TFElectraMainLayer(config, name="electra")
-        self.dropout = tf.keras.layers.Dropout(config.hidden_dropout_prob)
-        self.classifier = tf.keras.layers.Dense(config.num_labels, name="classifier")
-
-    @add_start_docstrings_to_callable(ELECTRA_INPUTS_DOCSTRING)
-    def call(
-        self,
-        input_ids=None,
-        attention_mask=None,
-        token_type_ids=None,
-        position_ids=None,
-        head_mask=None,
-        inputs_embeds=None,
-        training=False,
-    ):
-        r"""
-    Returns:
-        :obj:`tuple(torch.FloatTensor)` comprising various elements depending on the configuration (:class:`~transformers.ElectraConfig`) and inputs:
-        scores (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length, config.num_labels)`):
-            Classification scores (before SoftMax).
-        hidden_states (:obj:`tuple(tf.Tensor)`, `optional`, returned when :obj:`config.output_hidden_states=True`):
-            tuple of :obj:`tf.Tensor` (one for the output of the embeddings + one for the output of each layer)
-            of shape :obj:`(batch_size, sequence_length, hidden_size)`.
-
-            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
-        attentions (:obj:`tuple(tf.Tensor)`, `optional`, returned when ``config.output_attentions=True``):
-            tuple of :obj:`tf.Tensor` (one for each layer) of shape
-            :obj:`(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::
-
-        import tensorflow as tf
-        from transformers import ElectraTokenizer, TFElectraForTokenClassification
-
-        tokenizer = ElectraTokenizer.from_pretrained('google/electra-small-discriminator')
-        model = TFElectraForTokenClassification.from_pretrained('google/electra-small-discriminator')
-        input_ids = tf.constant(tokenizer.encode("Hello, my dog is cute"))[None, :]  # Batch size 1
-        outputs = model(input_ids)
-        scores = outputs[0]
-        """
-
-        discriminator_hidden_states = self.electra(
-            input_ids, attention_mask, token_type_ids, position_ids, head_mask, inputs_embeds, training=training
-        )
-        discriminator_sequence_output = discriminator_hidden_states[0]
-        discriminator_sequence_output = self.dropout(discriminator_sequence_output)
-        logits = self.classifier(discriminator_sequence_output)
-        output = (logits,)
-        output += discriminator_hidden_states[1:]
-
-        return output  # (loss), scores, (hidden_states), (attentions)
@@ -731,7 +731,7 @@ class TFT5Model(TFT5PreTrainedModel):
        tokenizer = T5Tokenizer.from_pretrained('t5-small')
        model = TFT5Model.from_pretrained('t5-small')
        input_ids = tokenizer.encode("Hello, my dog is cute", return_tensors="tf")  # Batch size 1
-        outputs = model(input_ids, decoder_input_ids=input_ids)
+        outputs = model(input_ids, input_ids=input_ids)
        last_hidden_states = outputs[0]  # The last hidden-state is the first element of the output tuple

        """
@@ -829,7 +829,7 @@ class TFT5ForConditionalGeneration(TFT5PreTrainedModel):
        tokenizer = T5Tokenizer.from_pretrained('t5-small')
        model = TFT5ForConditionalGeneration.from_pretrained('t5-small')
        input_ids = tokenizer.encode("Hello, my dog is cute", return_tensors="tf")  # Batch size 1
-        outputs = model(input_ids, decoder_input_ids=input_ids)
+        outputs = model(input_ids, input_ids=input_ids)
        prediction_scores = outputs[0]

        tokenizer = T5Tokenizer.from_pretrained('t5-small')
@@ -488,7 +488,7 @@ class TFTransfoXLMainLayer(tf.keras.layers.Layer):
        else:
            return None

-    def _update_mems(self, hids, mems, mlen, qlen):
+    def _update_mems(self, hids, mems, qlen, mlen):
        # does not deal with None
        if mems is None:
            return None
@@ -1177,6 +1177,8 @@ class TFPreTrainedModel(tf.keras.Model, TFModelUtilsMixin):
            beam_tokens = tf.convert_to_tensor([x[1] for x in next_batch_beam], dtype=tf.int32)
            beam_idx = tf.convert_to_tensor([x[2] for x in next_batch_beam], dtype=tf.int32)

+            print("Scores: {}-{}".format(cur_len, beam_scores.numpy()))
+
            # re-order batch
            input_ids = tf.stack([tf.identity(input_ids[x, :]) for x in beam_idx])
            input_ids = tf.concat([input_ids, tf.expand_dims(beam_tokens, 1)], axis=-1)
@@ -136,7 +136,7 @@ def load_tf_weights_in_transfo_xl(model, config, tf_path):
        if "kernel" in name or "proj" in name:
            array = np.transpose(array)
        if ("r_r_bias" in name or "r_w_bias" in name) and len(pointer) > 1:
-            # Here we will split the TF weights
+            # Here we will split the TF weigths
            assert len(pointer) == array.shape[0]
            for i, p_i in enumerate(pointer):
                arr_i = array[i, ...]
@@ -658,24 +658,26 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin):
    def generate(
        self,
        input_ids=None,
-        max_length=None,
+        attention_mask=None,
+        encoder_input_ids=None,
+        decoder_input_ids=None,
        min_length=None,
-        do_sample=None,
+        max_length=None,
        early_stopping=None,
+        num_return_sequences=None,
        num_beams=None,
+        do_sample=None,
        temperature=None,
        top_k=None,
        top_p=None,
-        repetition_penalty=None,
        bad_words_ids=None,
        bos_token_id=None,
        pad_token_id=None,
        eos_token_id=None,
        length_penalty=None,
        no_repeat_ngram_size=None,
-        num_return_sequences=None,
-        attention_mask=None,
-        decoder_start_token_id=None,
+        repetition_penalty=None,
+        use_cache=None,
    ):
        r""" Generates sequences for models with a LM head. The method currently supports greedy decoding, beam-search decoding, sampling with temperature, sampling with top-k or nucleus sampling.

@@ -688,24 +690,42 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin):
        Parameters:

            input_ids: (`optional`) `torch.LongTensor` of shape `(batch_size, sequence_length)`
-                The sequence used as a prompt for the generation. If `None` the method initializes
-                it as an empty `torch.LongTensor` of shape `(1,)`.
+                Short-hand for either encoder_input_ids in seq2seq models or decoder_input_ids for language models.

-            max_length: (`optional`) int
-                The max length of the sequence to be generated.  Between `min_length` and infinity. Default to 20.
+            attention_mask (`optional`) obj: `torch.LongTensor` of same shape as `input_ids`
+                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.
+                Defaults to `None`.
+
+            `What are attention masks? <../glossary.html#attention-mask>`__
+
+            encoder_input_ids: (`optional`) `torch.LongTensor` of shape `(batch_size, sequence_length)`
+                encoder inputs in the encoder-decoder setting (uses input_ids if None)
+
+            decoder_input_ids: (`optional`) `torch.LongTensor` of shape `(batch_size, sequence_length)`
+                The sequence used as a prompt for the generation in the encoder-decoder setting.
+                If `None` in the seq2seq setting, the method initializes it as a `torch.LongTensor` of shape `(batch_size, 1)` filled with bos_token_id.
+                If `None` in language modeling setting, uses input_ids if available, or initializes with bos_token_id otherwise.

            min_length: (`optional`) int
                The min length of the sequence to be generated.  Between 0 and infinity. Default to 0.

-            do_sample: (`optional`) bool
-                If set to `False` greedy decoding is used. Otherwise sampling is used. Defaults to `False` as defined in `configuration_utils.PretrainedConfig`.
+            max_length: (`optional`) int
+                The max length of the sequence to be generated.  Between `min_length` and infinity. Default to 20.

            early_stopping: (`optional`) bool
                if set to `True` beam search is stopped when at least `num_beams` sentences finished per batch. Defaults to `False` as defined in `configuration_utils.PretrainedConfig`.

+            num_return_sequences: (`optional`) int
+                The number of independently computed returned sequences for each element in the batch. Default to 1.
+
            num_beams: (`optional`) int
                Number of beams for beam search. Must be between 1 and infinity. 1 means no beam search. Default to 1.

+            do_sample: (`optional`) bool
+                If set to `False` greedy decoding is used. Otherwise sampling is used. Defaults to `False` as defined in `configuration_utils.PretrainedConfig`.
+
            temperature: (`optional`) float
                The value used to module the next token probabilities. Must be strictly positive. Default to 1.0.

@@ -715,9 +735,6 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin):
            top_p: (`optional`) float
                The cumulative probability of parameter highest probability vocabulary tokens to keep for nucleus sampling. Must be between 0 and 1. Default to 1.

-            repetition_penalty: (`optional`) float
-                The parameter for repetition penalty. Between 1.0 and infinity. 1.0 means no penalty. Default to 1.0.
-
            pad_token_id: (`optional`) int
                Padding token. Default to specicic model pad_token_id or None if it does not exist.

@@ -732,23 +749,16 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin):

            no_repeat_ngram_size: (`optional`) int
                If set to int > 0, all ngrams of size `no_repeat_ngram_size` can only occur once.
+
+            repetition_penalty: (`optional`) float
+                The parameter for repetition penalty. Between 1.0 and infinity. 1.0 means no penalty. Default to 1.0.
+
            bad_words_ids: (`optional`) list of lists of int
                `bad_words_ids` contains tokens that are not allowed to be generated. In order to get the tokens of the words that should not appear in the generated text, use `tokenizer.encode(bad_word, add_prefix_space=True)`.

-            num_return_sequences: (`optional`) int
-                The number of independently computed returned sequences for each element in the batch. Default to 1.
+            use_cache: (`optional`) bool
+                If set to `True` the model re-uses pre-computed decoder hidden states from one time step to the next.

-            attention_mask (`optional`) obj: `torch.LongTensor` of same shape as `input_ids`
-                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.
-                Defaults to `None`.
-
-            `What are attention masks? <../glossary.html#attention-mask>`__
-
-            decoder_start_token_id=None: (`optional`) int
-                If an encoder-decoder model starts decoding with a different token than BOS.
-                Defaults to `None` and is changed to `BOS` later.

        Return:

@@ -800,16 +810,18 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin):
                "Please use another model class (e.g. `OpenAIGPTLMHeadModel`, `XLNetLMHeadModel`, `GPT2LMHeadModel`, `CTRLLMHeadModel`, `T5WithLMHeadModel`, `TransfoXLLMHeadModel`, `XLMWithLMHeadModel`, `BartForConditionalGeneration` )"
            )

-        max_length = max_length if max_length is not None else self.config.max_length
        min_length = min_length if min_length is not None else self.config.min_length
-        do_sample = do_sample if do_sample is not None else self.config.do_sample
+        max_length = max_length if max_length is not None else self.config.max_length
        early_stopping = early_stopping if early_stopping is not None else self.config.early_stopping
+        num_return_sequences = (
+            num_return_sequences if num_return_sequences is not None else self.config.num_return_sequences
+        )
        num_beams = num_beams if num_beams is not None else self.config.num_beams
+        do_sample = do_sample if do_sample is not None else self.config.do_sample
        temperature = temperature if temperature is not None else self.config.temperature
        top_k = top_k if top_k is not None else self.config.top_k
        top_p = top_p if top_p is not None else self.config.top_p
        repetition_penalty = repetition_penalty if repetition_penalty is not None else self.config.repetition_penalty
-        bos_token_id = bos_token_id if bos_token_id is not None else self.config.bos_token_id
        pad_token_id = pad_token_id if pad_token_id is not None else self.config.pad_token_id
        eos_token_id = eos_token_id if eos_token_id is not None else self.config.eos_token_id
        length_penalty = length_penalty if length_penalty is not None else self.config.length_penalty
@@ -817,15 +829,12 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin):
            no_repeat_ngram_size if no_repeat_ngram_size is not None else self.config.no_repeat_ngram_size
        )
        bad_words_ids = bad_words_ids if bad_words_ids is not None else self.config.bad_words_ids
-        num_return_sequences = (
-            num_return_sequences if num_return_sequences is not None else self.config.num_return_sequences
-        )
-        decoder_start_token_id = (
-            decoder_start_token_id if decoder_start_token_id is not None else self.config.decoder_start_token_id
-        )
+

        if input_ids is not None:
            batch_size = input_ids.shape[0]  # overriden by the input batch_size
+        elif encoder_input_ids is not None:
+            batch_size = encoder_input_ids.shape[0]
        else:
            batch_size = 1

@@ -838,9 +847,6 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin):
        assert isinstance(top_k, int) and top_k >= 0, "`top_k` should be a positive integer."
        assert 0 <= top_p <= 1, "`top_p` should be between 0 and 1."
        assert repetition_penalty >= 1.0, "`repetition_penalty` should be >= 1."
-        assert input_ids is not None or (
-            isinstance(bos_token_id, int) and bos_token_id >= 0
-        ), "If input_ids is not defined, `bos_token_id` should be a positive integer."
        assert pad_token_id is None or (
            isinstance(pad_token_id, int) and (pad_token_id >= 0)
        ), "`pad_token_id` should be a positive integer."
@@ -858,16 +864,32 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin):
            bad_words_ids is None or isinstance(bad_words_ids, list) and isinstance(bad_words_ids[0], list)
        ), "`bad_words_ids` is either `None` or a list of lists of tokens that should not be generated"

-        if input_ids is None:
+        # different requirements fod decoder-only and encoder-decoder
+        if self.config.is_encoder_decoder:
+            assert hasattr(self, "get_encoder"), "{} should have a 'get_encoder' function defined".format(self)
+            assert callable(self.get_encoder), "{} should be a method".format(self.get_encoder)
+            assert input_ids is not None or encoder_input_ids is not None, "the encoder inputs should to be provided"
+            bos_token_id = bos_token_id if bos_token_id is not None else self.config.decoder_start_token_id
+            encoder_ids = encoder_input_ids if encoder_input_ids is not None else input_ids # for back-compatibility, esp summarization examples
+            decoder_ids = decoder_input_ids
+        else:
+            decoder_ids = decoder_input_ids if decoder_input_ids is not None else input_ids
+
+        bos_token_id = bos_token_id if bos_token_id is not None else self.config.bos_token_id
+        
+        assert decoder_ids is not None or (
+            isinstance(bos_token_id, int) and bos_token_id >= 0
+        ), "If input_ids is not defined, `bos_token_id` should be a positive integer."
+        if decoder_ids is None:
            assert isinstance(bos_token_id, int) and bos_token_id >= 0, (
                "you should either supply a context to complete as `input_ids` input "
                "or a `bos_token_id` (integer >= 0) as a first token to start the generation."
            )
-            input_ids = torch.full(
+            decoder_ids = torch.full(
                (batch_size, 1), bos_token_id, dtype=torch.long, device=next(self.parameters()).device,
            )
        else:
-            assert input_ids.dim() == 2, "Input prompt should be of shape (batch_size, sequence length)."
+            assert decoder_ids.dim() == 2, "Input prompt should be of shape (batch_size, sequence length)."

        # not allow to duplicate outputs when greedy decoding
        if do_sample is False:
@@ -876,7 +898,6 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin):
                assert (
                    num_return_sequences == 1
                ), "Greedy decoding will always produce the same output for num_beams == 1 and num_return_sequences > 1. Please set num_return_sequences = 1"
-
            else:
                # beam_search greedy generation conditions
                assert (
@@ -885,10 +906,23 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin):

        # create attention mask if necessary
        # TODO (PVP): this should later be handled by the forward fn() in each model in the future see PR 3140
-        if (attention_mask is None) and (pad_token_id is not None) and (pad_token_id in input_ids):
-            attention_mask = input_ids.ne(pad_token_id).long()
-        elif attention_mask is None:
-            attention_mask = input_ids.new_ones(input_ids.shape)
+        if self.config.is_encoder_decoder:
+            if (attention_mask is None) and (pad_token_id is not None) and (pad_token_id in encoder_ids):
+                enc_attention_mask = encoder_ids.ne(pad_token_id).long()
+            elif attention_mask is None:
+                enc_attention_mask = encoder_ids.new_ones(decoder_ids.shape)
+            else:
+                enc_attention_mask = attention_mask
+            # TODO (Yacine): implement behavior for decoder side in encoder-decoder
+            dec_attention_mask = None
+        else:
+            enc_attention_mask = None
+            if (attention_mask is None) and (pad_token_id is not None) and (pad_token_id in decoder_ids):
+                dec_attention_mask = decoder_ids.ne(pad_token_id).long()
+            elif attention_mask is None:
+                dec_attention_mask = decoder_ids.new_ones(decoder_ids.shape)
+            else:
+                dec_attention_mask = attention_mask

        # set pad_token_id to eos_token_id if not set. Important that this is done after
        # attention_mask is created
@@ -909,45 +943,19 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin):
            effective_batch_size = batch_size
            effective_batch_mult = 1

-        if self.config.is_encoder_decoder:
-            if decoder_start_token_id is None:
-                decoder_start_token_id = bos_token_id
-
-            assert (
-                decoder_start_token_id is not None
-            ), "decoder_start_token_id or bos_token_id has to be defined for encoder-decoder generation"
-            assert hasattr(self, "get_encoder"), "{} should have a 'get_encoder' function defined".format(self)
-            assert callable(self.get_encoder), "{} should be a method".format(self.get_encoder)
-
-            # get encoder and store encoder outputs
-            encoder = self.get_encoder()
-
-            encoder_outputs = encoder(input_ids, attention_mask=attention_mask)
-
-        # Expand input ids if num_beams > 1 or num_return_sequences > 1
+        # Expand decoder_ids if num_beams > 1 or num_return_sequences > 1
        if num_return_sequences > 1 or num_beams > 1:
-            input_ids_len = input_ids.shape[-1]
-            input_ids = input_ids.unsqueeze(1).expand(batch_size, effective_batch_mult * num_beams, input_ids_len)
-            attention_mask = attention_mask.unsqueeze(1).expand(
-                batch_size, effective_batch_mult * num_beams, input_ids_len
-            )
-
-            input_ids = input_ids.contiguous().view(
-                effective_batch_size * num_beams, input_ids_len
-            )  # shape: (batch_size * num_return_sequences * num_beams, cur_len)
-            attention_mask = attention_mask.contiguous().view(
-                effective_batch_size * num_beams, input_ids_len
+            decoder_ids_len = decoder_ids.shape[-1]
+            decoder_ids = decoder_ids.unsqueeze(1).expand(batch_size, effective_batch_mult * num_beams, decoder_ids_len)
+            decoder_ids = decoder_ids.contiguous().view(
+                effective_batch_size * num_beams, decoder_ids_len
            )  # shape: (batch_size * num_return_sequences * num_beams, cur_len)

+
+        # compute encoder outputs once if necessary, and expand to match number of generated sequences
        if self.config.is_encoder_decoder:
-            # create empty decoder_input_ids
-            input_ids = torch.full(
-                (effective_batch_size * num_beams, 1),
-                decoder_start_token_id,
-                dtype=torch.long,
-                device=next(self.parameters()).device,
-            )
-            cur_len = 1
+            encoder = self.get_encoder()
+            encoder_outputs = encoder(encoder_ids, attention_mask=enc_attention_mask)

            assert (
                batch_size == encoder_outputs[0].shape[0]
@@ -959,19 +967,30 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin):
                .view(-1, 1)
                .repeat(1, num_beams * effective_batch_mult)
                .view(-1)
-                .to(input_ids.device)
+                .to(decoder_ids.device)
            )
-            # expand encoder_outputs
-            encoder_outputs = (encoder_outputs[0].index_select(0, expanded_batch_idxs), *encoder_outputs[1:])
+            encoder_outputs = (encoder_outputs[0].index_select(0, expanded_batch_idxs), *encoder_outputs[1:]) # (x, encoder_states, all_attentions)
+            enc_attention_mask = enc_attention_mask.index_select(0, expanded_batch_idxs)
+            # TODO (Yacine): attention_mask was prepared for the encoder, not the decoder

        else:
            encoder_outputs = None
-            cur_len = input_ids.shape[-1]
+            dec_attention_mask = dec_attention_mask.unsqueeze(1).expand(
+                batch_size, effective_batch_mult * num_beams, decoder_ids_len
+            )
+            dec_attention_mask = dec_attention_mask.contiguous().view(
+                effective_batch_size * num_beams, decoder_ids_len
+            )  # shape: (batch_size * num_return_sequences * num_beams, cur_len)
+        
+        cur_len = decoder_ids.shape[-1]

        if num_beams > 1:
            output = self._generate_beam_search(
-                input_ids,
+                input_ids=decoder_ids,
                cur_len=cur_len,
+                encoder_outputs=encoder_outputs,
+                encoder_attention_mask=enc_attention_mask,
+                decoder_attention_mask=dec_attention_mask,
                max_length=max_length,
                min_length=min_length,
                do_sample=do_sample,
@@ -984,20 +1003,21 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin):
                bad_words_ids=bad_words_ids,
                bos_token_id=bos_token_id,
                pad_token_id=pad_token_id,
-                decoder_start_token_id=decoder_start_token_id,
                eos_token_id=eos_token_id,
                batch_size=effective_batch_size,
                num_return_sequences=num_return_sequences,
                length_penalty=length_penalty,
                num_beams=num_beams,
                vocab_size=vocab_size,
-                encoder_outputs=encoder_outputs,
-                attention_mask=attention_mask,
+                use_cache=use_cache,
            )
        else:
            output = self._generate_no_beam_search(
-                input_ids,
+                input_ids=decoder_ids,
                cur_len=cur_len,
+                encoder_outputs=encoder_outputs,
+                encoder_attention_mask=enc_attention_mask,
+                decoder_attention_mask=dec_attention_mask,
                max_length=max_length,
                min_length=min_length,
                do_sample=do_sample,
@@ -1009,11 +1029,9 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin):
                bad_words_ids=bad_words_ids,
                bos_token_id=bos_token_id,
                pad_token_id=pad_token_id,
-                decoder_start_token_id=decoder_start_token_id,
                eos_token_id=eos_token_id,
                batch_size=effective_batch_size,
-                encoder_outputs=encoder_outputs,
-                attention_mask=attention_mask,
+                use_cache=use_cache,
            )

        return output
@@ -1022,6 +1040,9 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin):
        self,
        input_ids,
        cur_len,
+        encoder_outputs,
+        encoder_attention_mask,
+        decoder_attention_mask,
        max_length,
        min_length,
        do_sample,
@@ -1034,10 +1055,8 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin):
        bos_token_id,
        pad_token_id,
        eos_token_id,
-        decoder_start_token_id,
        batch_size,
-        encoder_outputs,
-        attention_mask,
+        use_cache,
    ):
        """ Generate sequences for each example without beam search (num_beams == 1).
            All returned sequence are generated independantly.
@@ -1046,17 +1065,18 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin):
        unfinished_sents = input_ids.new(batch_size).fill_(1)
        sent_lengths = input_ids.new(batch_size).fill_(max_length)

-        past = encoder_outputs  # defined for encoder-decoder models, None for decoder-only models
+        past = None if encoder_outputs is None else ((encoder_outputs, encoder_attention_mask), None) # defined for encoder-decoder models, None for decoder-only models

        while cur_len < max_length:
-            model_inputs = self.prepare_inputs_for_generation(input_ids, past=past, attention_mask=attention_mask)
+            model_inputs = self.prepare_inputs_for_generation(input_ids, past=past, attention_mask=decoder_attention_mask)

            outputs = self(**model_inputs)
            next_token_logits = outputs[0][:, -1, :]

            # if model has past, then set the past variable to speed up decoding
-            if self._do_output_past(outputs):
-                past = outputs[1]
+            if use_cache:
+                _, decoder_cache = outputs[1]
+                past = ((encoder_outputs, encoder_attention_mask), decoder_cache)

            # repetition penalty from CTRL paper (https://arxiv.org/abs/1909.05858)
            if repetition_penalty != 1.0:
@@ -1139,6 +1159,9 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin):
        self,
        input_ids,
        cur_len,
+        encoder_outputs,
+        encoder_attention_mask,
+        decoder_attention_mask,
        max_length,
        min_length,
        do_sample,
@@ -1152,14 +1175,12 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin):
        bos_token_id,
        pad_token_id,
        eos_token_id,
-        decoder_start_token_id,
        batch_size,
        num_return_sequences,
        length_penalty,
        num_beams,
        vocab_size,
-        encoder_outputs,
-        attention_mask,
+        use_cache,
    ):
        """ Generate sequences for each example with beam search.
        """
@@ -1179,19 +1200,20 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin):
        beam_scores = beam_scores.view(-1)  # shape (batch_size * num_beams,)

        # cache compute states
-        past = encoder_outputs  # defined for encoder-decoder models, None for decoder-only models
+        past = None if encoder_outputs is None else ((encoder_outputs, encoder_attention_mask), None) # defined for encoder-decoder models, None for decoder-only models

        # done sentences
        done = [False for _ in range(batch_size)]

        while cur_len < max_length:
-            model_inputs = self.prepare_inputs_for_generation(input_ids, past=past, attention_mask=attention_mask)
+            model_inputs = self.prepare_inputs_for_generation(input_ids, past=past, attention_mask=decoder_attention_mask)
            outputs = self(**model_inputs)  # (batch_size * num_beams, cur_len, vocab_size)
            next_token_logits = outputs[0][:, -1, :]  # (batch_size * num_beams, vocab_size)

            # if model has past, then set the past variable to speed up decoding
-            if self._do_output_past(outputs):
-                past = outputs[1]
+            if use_cache:
+                _, decoder_cache = outputs[1]
+                past = ((encoder_outputs, encoder_attention_mask), decoder_cache)

            # repetition penalty (from CTRL paper https://arxiv.org/abs/1909.05858)
            if repetition_penalty != 1.0:
@@ -156,7 +156,7 @@ def load_tf_weights_in_xlnet(model, config, tf_path):
            logger.info("Transposing")
            array = np.transpose(array)
        if isinstance(pointer, list):
-            # Here we will split the TF weights
+            # Here we will split the TF weigths
            assert len(pointer) == array.shape[0]
            for i, p_i in enumerate(pointer):
                arr_i = array[i, ...]
@@ -26,7 +26,6 @@ from .configuration_auto import (
    CamembertConfig,
    CTRLConfig,
    DistilBertConfig,
-    ElectraConfig,
    FlaubertConfig,
    GPT2Config,
    OpenAIGPTConfig,
@@ -45,7 +44,6 @@ 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
@@ -69,7 +67,6 @@ TOKENIZER_MAPPING = OrderedDict(
        (XLMRobertaConfig, (XLMRobertaTokenizer, None)),
        (BartConfig, (BartTokenizer, None)),
        (RobertaConfig, (RobertaTokenizer, RobertaTokenizerFast)),
-        (ElectraConfig, (ElectraTokenizer, ElectraTokenizerFast)),
        (BertConfig, (BertTokenizer, BertTokenizerFast)),
        (OpenAIGPTConfig, (OpenAIGPTTokenizer, OpenAIGPTTokenizerFast)),
        (GPT2Config, (GPT2Tokenizer, GPT2TokenizerFast)),
@@ -107,7 +104,6 @@ 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).
    """
@@ -139,7 +135,6 @@ 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:
@@ -19,7 +19,6 @@ import collections
 import logging
 import os
 import unicodedata
-from typing import Optional

 from .tokenization_bert import BasicTokenizer, BertTokenizer, WordpieceTokenizer, load_vocab

@@ -90,7 +89,6 @@ class BertJapaneseTokenizer(BertTokenizer):
        pad_token="[PAD]",
        cls_token="[CLS]",
        mask_token="[MASK]",
-        mecab_kwargs=None,
        **kwargs
    ):
        """Constructs a MecabBertTokenizer.
@@ -108,7 +106,6 @@ class BertJapaneseTokenizer(BertTokenizer):
                Type of word tokenizer.
            **subword_tokenizer_type**: (`optional`) string (default "wordpiece")
                Type of subword tokenizer.
-            **mecab_kwargs**: (`optional`) dict passed to `MecabTokenizer` constructor (default None)
        """
        super(BertTokenizer, self).__init__(
            unk_token=unk_token,
@@ -137,9 +134,7 @@ class BertJapaneseTokenizer(BertTokenizer):
                    do_lower_case=do_lower_case, never_split=never_split, tokenize_chinese_chars=False
                )
            elif word_tokenizer_type == "mecab":
-                self.word_tokenizer = MecabTokenizer(
-                    do_lower_case=do_lower_case, never_split=never_split, **(mecab_kwargs or {})
-                )
+                self.word_tokenizer = MecabTokenizer(do_lower_case=do_lower_case, never_split=never_split)
            else:
                raise ValueError("Invalid word_tokenizer_type '{}' is specified.".format(word_tokenizer_type))

@@ -166,10 +161,10 @@ class BertJapaneseTokenizer(BertTokenizer):
        return split_tokens


-class MecabTokenizer:
+class MecabTokenizer(object):
    """Runs basic tokenization with MeCab morphological parser."""

-    def __init__(self, do_lower_case=False, never_split=None, normalize_text=True, mecab_option: Optional[str] = None):
+    def __init__(self, do_lower_case=False, never_split=None, normalize_text=True):
        """Constructs a MecabTokenizer.

        Args:
@@ -181,7 +176,6 @@ class MecabTokenizer:
                List of token not to split.
            **normalize_text**: (`optional`) boolean (default True)
                Whether to apply unicode normalization to text before tokenization.
-            **mecab_option**: (`optional`) string passed to `MeCab.Tagger` constructor (default "")
        """
        self.do_lower_case = do_lower_case
        self.never_split = never_split if never_split is not None else []
@@ -189,7 +183,7 @@ class MecabTokenizer:

        import MeCab

-        self.mecab = MeCab.Tagger(mecab_option) if mecab_option is not None else MeCab.Tagger()
+        self.mecab = MeCab.Tagger()

    def tokenize(self, text, never_split=None, **kwargs):
        """Tokenizes a piece of text."""
@@ -1,80 +0,0 @@
-# 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
@@ -59,4 +59,4 @@ You can then finish the addition step by adding imports for your classes in the
 - [ ] add a link to your conversion script in the main conversion utility (in `commands/convert.py`)
 - [ ] edit the PyTorch to TF 2.0 conversion script to add your model in the `convert_pytorch_checkpoint_to_tf2.py` file
 - [ ] add a mention of your model in the doc: `README.md` and the documentation itself at `docs/source/pretrained_models.rst`.
- [ ] upload the pretrained weights, configurations and vocabulary files.
+- [ ] upload the pretrained weigths, configurations and vocabulary files.
@@ -624,96 +624,70 @@ class ModelTesterMixin:
            with torch.no_grad():
                model(**inputs_dict)

-    def test_lm_head_model_random_no_beam_search_generate(self):
+    def test_lm_head_model_random_generate(self):
+
        config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
-        input_ids = inputs_dict["input_ids"] if "input_ids" in inputs_dict else inputs_dict["inputs"]
-
-        # iterate over all generative models
-        for model_class in self.all_generative_model_classes:
-            model = model_class(config)
-
-            if config.bos_token_id is None:
-                # if bos token id is not defined mobel needs input_ids
-                with self.assertRaises(AssertionError):
-                    model.generate(do_sample=True, max_length=5)
-                # num_return_sequences = 1
-                self._check_generated_ids(model.generate(input_ids, do_sample=True))
-            else:
-                # num_return_sequences = 1
-                self._check_generated_ids(model.generate(do_sample=True, max_length=5))
-
-            with self.assertRaises(AssertionError):
-                # generating multiple sequences when no beam search generation
-                # is not allowed as it would always generate the same sequences
-                model.generate(input_ids, do_sample=False, num_return_sequences=2)
-
-            # num_return_sequences > 1, sample
-            self._check_generated_ids(model.generate(input_ids, do_sample=True, num_return_sequences=2))
-
-            # check bad words tokens language generation
-            # create list of 1-seq bad token and list of 2-seq of bad tokens
-            bad_words_ids = [self._generate_random_bad_tokens(1, model), self._generate_random_bad_tokens(2, model)]
-            output_tokens = model.generate(
-                input_ids, do_sample=True, bad_words_ids=bad_words_ids, num_return_sequences=2
-            )
-            # only count generated tokens
-            generated_ids = output_tokens[:, input_ids.shape[-1] :]
-            self.assertFalse(self._check_match_tokens(generated_ids.tolist(), bad_words_ids))
-
-    def test_lm_head_model_random_beam_search_generate(self):
-        config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
-        input_ids = inputs_dict["input_ids"] if "input_ids" in inputs_dict else inputs_dict["inputs"]
+        input_ids = inputs_dict.get("input_ids")

        if self.is_encoder_decoder:
-            # needed for Bart beam search
-            config.output_past = True
+            config.output_past = True  # needed for Bart TODO: might have to update for other encoder-decoder models

        for model_class in self.all_generative_model_classes:
            model = model_class(config)
+            model.to(torch_device)
+            model.eval()

            if config.bos_token_id is None:
-                # if bos token id is not defined mobel needs input_ids, num_return_sequences = 1
-                self._check_generated_ids(model.generate(input_ids, do_sample=True, num_beams=2))
+                with self.assertRaises(AssertionError):
+                    model.generate(do_sample=True, max_length=5)
+                # batch_size = 1
+                self._check_generated_ids(model.generate(input_ids, do_sample=True))
+                # batch_size = 1, num_beams > 1
+                self._check_generated_ids(model.generate(input_ids, do_sample=True, num_beams=3))
            else:
-                # num_return_sequences = 1
-                self._check_generated_ids(model.generate(do_sample=True, max_length=5, num_beams=2))
+                # batch_size = 1
+                self._check_generated_ids(model.generate(do_sample=True, max_length=5))
+                # batch_size = 1, num_beams > 1
+                self._check_generated_ids(model.generate(do_sample=True, max_length=5, num_beams=3))
+
+            with self.assertRaises(AssertionError):
+                # generating multiple sequences when greedy no beam generation
+                # is not allowed as it would always generate the same sequences
+                model.generate(input_ids, do_sample=False, num_return_sequences=2)

            with self.assertRaises(AssertionError):
                # generating more sequences than having beams leads is not possible
                model.generate(input_ids, do_sample=False, num_return_sequences=3, num_beams=2)

-            # num_return_sequences > 1, sample
-            self._check_generated_ids(model.generate(input_ids, do_sample=True, num_beams=2, num_return_sequences=2,))
-            # num_return_sequences > 1, greedy
-            self._check_generated_ids(model.generate(input_ids, do_sample=False, num_beams=2, num_return_sequences=2))
+            # batch_size > 1, sample
+            self._check_generated_ids(model.generate(input_ids, do_sample=True, num_return_sequences=3))
+            # batch_size > 1, greedy
+            self._check_generated_ids(model.generate(input_ids, do_sample=False))
+
+            # batch_size > 1, num_beams > 1, sample
+            self._check_generated_ids(model.generate(input_ids, do_sample=True, num_beams=3, num_return_sequences=3,))
+            # batch_size > 1, num_beams > 1, greedy
+            self._check_generated_ids(model.generate(input_ids, do_sample=False, num_beams=3, num_return_sequences=3))

            # check bad words tokens language generation
-            # create list of 1-seq bad token and list of 2-seq of bad tokens
-            bad_words_ids = [self._generate_random_bad_tokens(1, model), self._generate_random_bad_tokens(2, model)]
+            bad_words_ids = [
+                ids_tensor((1, 1), self.model_tester.vocab_size).squeeze(-1).tolist(),
+                ids_tensor((2, 1), self.model_tester.vocab_size).squeeze(-1).tolist(),
+            ]
+
+            # sampling
            output_tokens = model.generate(
-                input_ids, do_sample=False, bad_words_ids=bad_words_ids, num_beams=2, num_return_sequences=2
+                input_ids, do_sample=True, bad_words_ids=bad_words_ids, num_return_sequences=3
            )
-            # only count generated tokens
            generated_ids = output_tokens[:, input_ids.shape[-1] :]
            self.assertFalse(self._check_match_tokens(generated_ids.tolist(), bad_words_ids))

-    def _generate_random_bad_tokens(self, num_bad_tokens, model):
-        # special tokens cannot be bad tokens
-        special_tokens = []
-        if model.config.bos_token_id is not None:
-            special_tokens.append(model.config.bos_token_id)
-        if model.config.pad_token_id is not None:
-            special_tokens.append(model.config.pad_token_id)
-        if model.config.eos_token_id is not None:
-            special_tokens.append(model.config.eos_token_id)
-
-        # create random bad tokens that are not special tokens
-        bad_tokens = []
-        while len(bad_tokens) < num_bad_tokens:
-            token = ids_tensor((1, 1), self.model_tester.vocab_size).squeeze(0).numpy()[0]
-            if token not in special_tokens:
-                bad_tokens.append(token)
-        return bad_tokens
+            # beam search
+            output_tokens = model.generate(
+                input_ids, do_sample=False, bad_words_ids=bad_words_ids, num_beams=3, num_return_sequences=3
+            )
+            generated_ids = output_tokens[:, input_ids.shape[-1] :]
+            self.assertFalse(self._check_match_tokens(generated_ids.tolist(), bad_words_ids))

    def _check_generated_ids(self, output_ids):
        for token_id in output_ids[0].tolist():
@@ -1,287 +0,0 @@
-# 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)
@@ -420,96 +420,68 @@ class TFModelTesterMixin:

            model(inputs_dict)

-    def test_lm_head_model_random_no_beam_search_generate(self):
-        config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
-        input_ids = inputs_dict["input_ids"] if "input_ids" in inputs_dict else inputs_dict["inputs"]
+    def test_lm_head_model_random_generate(self):

-        # iterate over all generative models
-        for model_class in self.all_generative_model_classes:
-            model = model_class(config)
-
-            if config.bos_token_id is None:
-                # if bos token id is not defined mobel needs input_ids
-                with self.assertRaises(AssertionError):
-                    model.generate(do_sample=True, max_length=5)
-                # num_return_sequences = 1
-                self._check_generated_ids(model.generate(input_ids, do_sample=True))
-            else:
-                # num_return_sequences = 1
-                self._check_generated_ids(model.generate(do_sample=True, max_length=5))
-
-            with self.assertRaises(AssertionError):
-                # generating multiple sequences when no beam search generation
-                # is not allowed as it would always generate the same sequences
-                model.generate(input_ids, do_sample=False, num_return_sequences=2)
-
-            # num_return_sequences > 1, sample
-            self._check_generated_ids(model.generate(input_ids, do_sample=True, num_return_sequences=2))
-
-            # check bad words tokens language generation
-            # create list of 1-seq bad token and list of 2-seq of bad tokens
-            bad_words_ids = [self._generate_random_bad_tokens(1, model), self._generate_random_bad_tokens(2, model)]
-            output_tokens = model.generate(
-                input_ids, do_sample=True, bad_words_ids=bad_words_ids, num_return_sequences=2
-            )
-            # only count generated tokens
-            generated_ids = output_tokens[:, input_ids.shape[-1] :]
-            self.assertFalse(self._check_match_tokens(generated_ids.numpy().tolist(), bad_words_ids))
-
-    def test_lm_head_model_random_beam_search_generate(self):
        config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
        input_ids = inputs_dict["input_ids"] if "input_ids" in inputs_dict else inputs_dict["inputs"]

        if self.is_encoder_decoder:
-            # needed for Bart beam search
-            config.output_past = True
+            config.output_past = True  # needed for Bart TODO: might have to update for other encoder-decoder models

        for model_class in self.all_generative_model_classes:
            model = model_class(config)

            if config.bos_token_id is None:
-                # if bos token id is not defined mobel needs input_ids, num_return_sequences = 1
-                self._check_generated_ids(model.generate(input_ids, do_sample=True, num_beams=2))
+                with self.assertRaises(AssertionError):
+                    model.generate(do_sample=True, max_length=5)
+                # batch_size = 1
+                self._check_generated_ids(model.generate(input_ids, do_sample=True))
+                # batch_size = 1, num_beams > 1
+                self._check_generated_ids(model.generate(input_ids, do_sample=True, num_beams=3))
            else:
-                # num_return_sequences = 1
-                self._check_generated_ids(model.generate(do_sample=True, max_length=5, num_beams=2))
+                # batch_size = 1
+                self._check_generated_ids(model.generate(do_sample=True, max_length=5))
+                # batch_size = 1, num_beams > 1
+                self._check_generated_ids(model.generate(do_sample=True, max_length=5, num_beams=3))
+
+            with self.assertRaises(AssertionError):
+                # generating multiple sequences when greedy no beam generation
+                # is not allowed as it would always generate the same sequences
+                model.generate(input_ids, do_sample=False, num_return_sequences=2)

            with self.assertRaises(AssertionError):
                # generating more sequences than having beams leads is not possible
                model.generate(input_ids, do_sample=False, num_return_sequences=3, num_beams=2)

-            # num_return_sequences > 1, sample
-            self._check_generated_ids(model.generate(input_ids, do_sample=True, num_beams=2, num_return_sequences=2,))
-            # num_return_sequences > 1, greedy
-            self._check_generated_ids(model.generate(input_ids, do_sample=False, num_beams=2, num_return_sequences=2))
+            # batch_size > 1, sample
+            self._check_generated_ids(model.generate(input_ids, do_sample=True, num_return_sequences=3))
+            # batch_size > 1, greedy
+            self._check_generated_ids(model.generate(input_ids, do_sample=False))
+
+            # batch_size > 1, num_beams > 1, sample
+            self._check_generated_ids(model.generate(input_ids, do_sample=True, num_beams=3, num_return_sequences=3,))
+            # batch_size > 1, num_beams > 1, greedy
+            self._check_generated_ids(model.generate(input_ids, do_sample=False, num_beams=3, num_return_sequences=3))

            # check bad words tokens language generation
-            # create list of 1-seq bad token and list of 2-seq of bad tokens
-            bad_words_ids = [self._generate_random_bad_tokens(1, model), self._generate_random_bad_tokens(2, model)]
+            bad_words_ids = [
+                tf.squeeze(ids_tensor((1, 1), self.model_tester.vocab_size), -1).numpy().tolist(),
+                tf.squeeze(ids_tensor((2, 1), self.model_tester.vocab_size), -1).numpy().tolist(),
+            ]
+
+            # sampling
            output_tokens = model.generate(
-                input_ids, do_sample=False, bad_words_ids=bad_words_ids, num_beams=2, num_return_sequences=2
+                input_ids, do_sample=True, bad_words_ids=bad_words_ids, num_return_sequences=3
            )
-            # only count generated tokens
            generated_ids = output_tokens[:, input_ids.shape[-1] :]
            self.assertFalse(self._check_match_tokens(generated_ids.numpy().tolist(), bad_words_ids))

-    def _generate_random_bad_tokens(self, num_bad_tokens, model):
-        # special tokens cannot be bad tokens
-        special_tokens = []
-        if model.config.bos_token_id is not None:
-            special_tokens.append(model.config.bos_token_id)
-        if model.config.pad_token_id is not None:
-            special_tokens.append(model.config.pad_token_id)
-        if model.config.eos_token_id is not None:
-            special_tokens.append(model.config.eos_token_id)
-
-        # create random bad tokens that are not special tokens
-        bad_tokens = []
-        while len(bad_tokens) < num_bad_tokens:
-            token = tf.squeeze(ids_tensor((1, 1), self.model_tester.vocab_size), 0).numpy()[0]
-            if token not in special_tokens:
-                bad_tokens.append(token)
-        return bad_tokens
+            # beam search
+            output_tokens = model.generate(
+                input_ids, do_sample=False, bad_words_ids=bad_words_ids, num_beams=3, num_return_sequences=3
+            )
+            generated_ids = output_tokens[:, input_ids.shape[-1] :]
+            self.assertFalse(self._check_match_tokens(generated_ids.numpy().tolist(), bad_words_ids))

    def _check_generated_ids(self, output_ids):
        for token_id in output_ids[0].numpy().tolist():
@@ -1,227 +0,0 @@
-# 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)
@@ -91,20 +91,6 @@ class BertJapaneseTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
            ["アップルストア", "で", "iphone", "8", "が", "発売", "さ", "れ", "た", "。"],
        )

-    def test_mecab_tokenizer_with_option(self):
-        try:
-            tokenizer = MecabTokenizer(
-                do_lower_case=True, normalize_text=False, mecab_option="-d /usr/local/lib/mecab/dic/jumandic"
-            )
-        except RuntimeError:
-            # if dict doesn't exist in the system, previous code raises this error.
-            return
-
-        self.assertListEqual(
-            tokenizer.tokenize(" \tｱｯﾌﾟﾙストアでiPhone８ が  \n 発売された　。  "),
-            ["ｱｯﾌﾟﾙストア", "で", "iPhone", "８", "が", "発売", "さ", "れた", "\u3000", "。"],
-        )
-
    def test_mecab_tokenizer_no_normalize(self):
        tokenizer = MecabTokenizer(normalize_text=False)
Author	SHA1	Message	Date
yjernite	d1406e9a63	resolved_decode_with_prefix	2020-04-02 14:19:12 -04:00
yjernite	76f5cd8b7c	resolved_decode_with_prefix	2020-04-02 14:14:11 -04:00
yjernite	ae98046330	decode_with_prefix	2020-04-02 14:09:31 -04:00