Merge remote-tracking branch 'refs/remotes/huggingface/master'

.circleci/config.yml

@@ -70,6 +70,27 @@ jobs:
             - run: sudo pip install pytest codecov pytest-cov
             - run: python -m pytest -sv ./transformers/tests/ --cov
             - run: codecov
+    build_py3_custom_tokenizers:
+        working_directory: ~/transformers
+        docker:
+            - image: circleci/python:3.5
+        steps:
+            - checkout
+            - run: sudo pip install --progress-bar off .
+            - run: sudo pip install pytest
+            - run: sudo pip install mecab-python3
+            - run: RUN_CUSTOM_TOKENIZERS=1 python -m pytest -sv ./transformers/tests/tokenization_bert_japanese_test.py
+    build_py2_custom_tokenizers:
+        working_directory: ~/transformers
+        docker:
+            - image: circleci/python:2.7
+        steps:
+            - checkout
+            - run: sudo pip install --progress-bar off .
+            - run: sudo pip install pytest
+            - run: sudo apt-get -y install libmecab-dev mecab mecab-ipadic-utf8 swig
+            - run: sudo pip install mecab-python
+            - run: RUN_CUSTOM_TOKENIZERS=1 python -m pytest -sv ./transformers/tests/tokenization_bert_japanese_test.py
     deploy_doc:
         working_directory: ~/transformers
         docker:
@@ -81,7 +102,17 @@ jobs:
             - checkout
             - run: sudo pip install --progress-bar off -r docs/requirements.txt
             - run: sudo pip install --progress-bar off -r requirements.txt
-            - run: cd docs && make clean && make html && scp -r -oStrictHostKeyChecking=no _build/html/* $doc:$dir
+            - run: ./.circleci/deploy.sh
+    repository_consistency:
+        working_directory: ~/transformers
+        docker:
+            - image: circleci/python:3.5
+        resource_class: small
+        parallelism: 1
+        steps:
+            - checkout
+            - run: sudo pip install requests
+            - run: python ./utils/link_tester.py
 workflow_filters: &workflow_filters
     filters:
         branches:
@@ -91,9 +122,12 @@ workflows:
     version: 2
     build_and_test:
         jobs:
+            - repository_consistency
+            - build_py3_custom_tokenizers
+            - build_py2_custom_tokenizers
             - build_py3_torch_and_tf
             - build_py3_torch
             - build_py3_tf
             - build_py2_torch
             - build_py2_tf
-            - deploy_doc: *workflow_filters
\ No newline at end of file
+            - deploy_doc: *workflow_filters

.circleci/deploy.sh

+cd docs
+
+function deploy_doc(){
+	echo "Creating doc at commit $1 and pushing to folder $2"
+	git checkout $1
+	if [ ! -z "$2" ] 
+	then
+		if [ -d "$dir/$2" ]; then
+			echo "Directory" $2 "already exists"
+		else
+			echo "Pushing version" $2
+			make clean && make html && scp -r -oStrictHostKeyChecking=no _build/html $doc:$dir/$2
+		fi
+	else
+		echo "Pushing master"
+		make clean && make html && scp -r -oStrictHostKeyChecking=no _build/html/* $doc:$dir
+	fi
+}
+
+deploy_doc "master" 
+deploy_doc "b33a385" v1.0.0
+deploy_doc "fe02e45" v1.1.0
+deploy_doc "89fd345" v1.2.0
+deploy_doc "fc9faa8" v2.0.0
+deploy_doc "3ddce1d" v2.1.1
+deploy_doc "3616209" v2.2.0

.github/ISSUE_TEMPLATE/--new-model-addition.md

@@ -17,6 +17,7 @@ assignees: ''
 
 * [ ] the model implementation is available: (give details)
 * [ ] the model weights are available: (give details)
+* [ ] who are the authors: (mention them)
 
 ## Additional context
 

.gitignore

@@ -137,4 +137,5 @@ examples/runs
 serialization_dir
 
 # emacs
-*.*~
\ No newline at end of file
+*.*~
+debug.env

CONTRIBUTING.md

@@ -62,6 +62,8 @@ Awesome! Please provide the following information:
 If you are willing to contribute the model yourself, let us know so we can best
 guide you.
 
+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.
+
 ### Do you want a new feature (that is not a model)?
 
 A world-class feature request addresses the following points:
@@ -81,6 +83,8 @@ A world-class feature request addresses the following points:
 If your issue is well written we're already 80% of the way there by the time you
 post it.
 
+We have added **templates** to guide you in the process of adding a new example script for training or testing the models in the library. You can find them in the [`templates`](./templates) folder.
+
 ## Start contributing! (Pull Requests)
 
 Before writing code, we strongly advise you to search through the exising PRs or
@@ -102,7 +106,7 @@ Follow these steps to start contributing:
    ```bash
    $ git clone git@github.com:<your Github handle>/transformers.git
    $ cd transformers
-   $ git remote add upstream git@github.com:huggingface/transformers.git
+   $ git remote add upstream https://github.com/huggingface/transformers.git
    ```
 
 3. Create a new branch to hold your development changes:

README.md

@@ -39,7 +39,7 @@ State-of-the-art NLP for everyone
 Lower compute costs, smaller carbon footprint
 - Researchers can share trained models instead of always retraining
 - Practitioners can reduce compute time and production costs
-- 8 architectures with over 30 pretrained models, some in more than 100 languages
+- 10 architectures with over 30 pretrained models, some in more than 100 languages
 
 Choose the right framework for every part of a model's lifetime
 - Train state-of-the-art models in 3 lines of code
@@ -58,7 +58,7 @@ Choose the right framework for every part of a model's lifetime
 | [Quick tour: Fine-tuning/usage scripts](#quick-tour-of-the-fine-tuningusage-scripts) | Using provided scripts: GLUE, SQuAD and Text generation |
 | [Migrating from pytorch-transformers to transformers](#Migrating-from-pytorch-transformers-to-transformers) | Migrating your code from pytorch-transformers to transformers |
 | [Migrating from pytorch-pretrained-bert to pytorch-transformers](#Migrating-from-pytorch-pretrained-bert-to-transformers) | Migrating your code from pytorch-pretrained-bert to transformers |
-| [Documentation](https://huggingface.co/transformers/) | Full API documentation and more |
+| [Documentation][(v2.2.0/v2.2.1)](https://huggingface.co/transformers/v2.2.0) [(v2.1.1)](https://huggingface.co/transformers/v2.1.1) [(v2.0.0)](https://huggingface.co/transformers/v2.0.0) [(v1.2.0)](https://huggingface.co/transformers/v1.2.0) [(v1.1.0)](https://huggingface.co/transformers/v1.1.0) [(v1.0.0)](https://huggingface.co/transformers/v1.0.0) [(master)](https://huggingface.co/transformers) | Full API documentation and more |
 
 ## Installation
 
@@ -86,21 +86,41 @@ When TensorFlow 2.0 and/or PyTorch has been installed, you can install from sour
 pip install [--editable] .
 ```
 
+### Run the examples
+
+Examples are included in the repository but are not shipped with the library.
+Therefore, in order to run the latest versions of the examples you also need to install from source. To do so, create a new virtual environment and follow these steps:
+
+```bash
+git clone https://github.com/huggingface/transformers
+cd transformers
+pip install [--editable] .
+```
+
 ### Tests
 
 A series of tests are included for the library and the example scripts. Library tests can be found in the [tests folder](https://github.com/huggingface/transformers/tree/master/transformers/tests) and examples tests in the [examples folder](https://github.com/huggingface/transformers/tree/master/examples).
 
-These tests can be run using `pytest` (install pytest if needed with `pip install pytest`).
+These tests can be run using `unittest` or `pytest` (install pytest if needed with `pip install pytest`).
 
 Depending on which framework is installed (TensorFlow 2.0 and/or PyTorch), the irrelevant tests will be skipped. Ensure that both frameworks are installed if you want to execute all tests.
 
 You can run the tests from the root of the cloned repository with the commands:
 
+```bash
+python -m unittest discover -s transformers/tests -p "*test.py" -t .
+python -m unittest discover -s examples -p "*test.py" -t examples
+```
+
+or
+
 ```bash
 python -m pytest -sv ./transformers/tests/
 python -m pytest -sv ./examples/
 ```
 
+By default, slow tests are skipped. Set the `RUN_SLOW` environment variable to `yes` to run them.
+
 ### Do you want to run a Transformer model on a mobile device?
 
 You should check out our [`swift-coreml-transformers`](https://github.com/huggingface/swift-coreml-transformers) repo.
@@ -111,7 +131,7 @@ At some point in the future, you'll be able to seamlessly move from pre-training
 
 ## Model architectures
 
-🤗 Transformers currently provides 8 NLU/NLG architectures:
+🤗 Transformers currently provides 10 NLU/NLG architectures:
 
 1. **[BERT](https://github.com/google-research/bert)** (from Google) released with the paper [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://arxiv.org/abs/1810.04805) by Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova.
 2. **[GPT](https://github.com/openai/finetune-transformer-lm)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
@@ -120,8 +140,11 @@ At some point in the future, you'll be able to seamlessly move from pre-training
 5. **[XLNet](https://github.com/zihangdai/xlnet/)** (from Google/CMU) released with the paper [​XLNet: Generalized Autoregressive Pretraining for Language Understanding](https://arxiv.org/abs/1906.08237) by Zhilin Yang*, Zihang Dai*, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, Quoc V. Le.
 6. **[XLM](https://github.com/facebookresearch/XLM/)** (from Facebook) released together with the paper [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) by Guillaume Lample and Alexis Conneau.
 7. **[RoBERTa](https://github.com/pytorch/fairseq/tree/master/examples/roberta)** (from Facebook), released together with the paper a [Robustly Optimized BERT Pretraining Approach](https://arxiv.org/abs/1907.11692) by Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov.
-8. **[DistilBERT](https://github.com/huggingface/transformers/tree/master/examples/distillation)** (from HuggingFace), released together with the paper [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://arxiv.org/abs/1910.01108) by Victor Sanh, Lysandre Debut and Thomas Wolf. The same method has been applied to compress GPT2 into [DistilGPT2](https://github.com/huggingface/transformers/tree/master/examples/distillation).
+8. **[DistilBERT](https://github.com/huggingface/transformers/tree/master/examples/distillation)** (from HuggingFace), released together with the paper [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://arxiv.org/abs/1910.01108) by Victor Sanh, Lysandre Debut and Thomas Wolf. The same method has been applied to compress GPT2 into [DistilGPT2](https://github.com/huggingface/transformers/tree/master/examples/distillation), RoBERTa into [DistilRoBERTa](https://github.com/huggingface/transformers/tree/master/examples/distillation), Multilingual BERT into [DistilmBERT](https://github.com/huggingface/transformers/tree/master/examples/distillation) and a German version of DistilBERT.
 9. **[CTRL](https://github.com/salesforce/ctrl/)** (from Salesforce) released with the paper [CTRL: A Conditional Transformer Language Model for Controllable Generation](https://arxiv.org/abs/1909.05858) by Nitish Shirish Keskar*, Bryan McCann*, Lav R. Varshney, Caiming Xiong and Richard Socher.
+10. **[CamemBERT](https://camembert-model.fr)** (from Inria/Facebook/Sorbonne) released with the paper [CamemBERT: a Tasty French Language Model](https://arxiv.org/abs/1911.03894) by Louis Martin*, Benjamin Muller*, Pedro Javier Ortiz Suárez*, Yoann Dupont, Laurent Romary, Éric Villemonte de la Clergerie, Djamé Seddah and Benoît Sagot.
+11. **[ALBERT](https://github.com/google-research/ALBERT)** (from Google Research and the Toyota Technological Institute at Chicago) released with the paper [ALBERT: A Lite BERT for Self-supervised Learning of Language Representations](https://arxiv.org/abs/1909.11942), by Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, Radu Soricut.
+11. 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).
 
@@ -170,8 +193,7 @@ for model_class, tokenizer_class, pretrained_weights in MODELS:
 
 # Each architecture is provided with several class for fine-tuning on down-stream tasks, e.g.
 BERT_MODEL_CLASSES = [BertModel, BertForPreTraining, BertForMaskedLM, BertForNextSentencePrediction,
-                      BertForSequenceClassification, BertForMultipleChoice, BertForTokenClassification,
-                      BertForQuestionAnswering]
+                      BertForSequenceClassification, BertForTokenClassification, BertForQuestionAnswering]
 
 # All the classes for an architecture can be initiated from pretrained weights for this architecture
 # Note that additional weights added for fine-tuning are only initialized
@@ -252,6 +274,11 @@ print("sentence_2 is", "a paraphrase" if pred_2 else "not a paraphrase", "of sen
 
 ## Quick tour of the fine-tuning/usage scripts
 
+**Important**  
+Before running the fine-tuning scripts, please read the
+[instructions](#run-the-examples) on how to
+setup your environment to run the examples.
+
 The library comprises several example scripts with SOTA performances for NLU and NLG tasks:
 
 - `run_glue.py`: an example fine-tuning Bert, XLNet and XLM on nine different GLUE tasks (*sequence-level classification*)
@@ -413,7 +440,7 @@ and from the Salesforce CTRL model:
 python ./examples/run_generation.py \
     --model_type=ctrl \
     --length=20 \
-    --model_name_or_path=gpt2 \
+    --model_name_or_path=ctrl \
     --temperature=0 \
     --repetition_penalty=1.2 \
 ```
@@ -520,12 +547,12 @@ Here is a conversion examples from `BertAdam` with a linear warmup and decay sch
 # Parameters:
 lr = 1e-3
 max_grad_norm = 1.0
-num_total_steps = 1000
+num_training_steps = 1000
 num_warmup_steps = 100
-warmup_proportion = float(num_warmup_steps) / float(num_total_steps)  # 0.1
+warmup_proportion = float(num_warmup_steps) / float(num_training_steps)  # 0.1
 
 ### Previously BertAdam optimizer was instantiated like this:
-optimizer = BertAdam(model.parameters(), lr=lr, schedule='warmup_linear', warmup=warmup_proportion, t_total=num_total_steps)
+optimizer = BertAdam(model.parameters(), lr=lr, schedule='warmup_linear', warmup=warmup_proportion, t_total=num_training_steps)
 ### and used like this:
 for batch in train_data:
     loss = model(batch)
@@ -534,9 +561,10 @@ for batch in train_data:
 
 ### In Transformers, optimizer and schedules are splitted and instantiated like this:
 optimizer = AdamW(model.parameters(), lr=lr, correct_bias=False)  # To reproduce BertAdam specific behavior set correct_bias=False
-scheduler = WarmupLinearSchedule(optimizer, warmup_steps=num_warmup_steps, t_total=num_total_steps)  # PyTorch scheduler
+scheduler = get_linear_schedule_with_warmup(optimizer, num_warmup_steps=num_warmup_steps, num_training_steps=num_training_steps)  # PyTorch scheduler
 ### and used like this:
 for batch in train_data:
+    model.train()
     loss = model(batch)
     loss.backward()
     torch.nn.utils.clip_grad_norm_(model.parameters(), max_grad_norm)  # Gradient clipping is not in AdamW anymore (so you can use amp without issue)
@@ -549,12 +577,11 @@ for batch in train_data:
 
 We now have a paper you can cite for the 🤗 Transformers library:
 ```
-@misc{wolf2019transformers,
-    title={Transformers: State-of-the-art Natural Language Processing},
-    author={Thomas Wolf and Lysandre Debut and Victor Sanh and Julien Chaumond and Clement Delangue and Anthony Moi and Pierric Cistac and Tim Rault and Rémi Louf and Morgan Funtowicz and Jamie Brew},
-    year={2019},
-    eprint={1910.03771},
-    archivePrefix={arXiv},
-    primaryClass={cs.CL}
+@article{Wolf2019HuggingFacesTS,
+  title={HuggingFace's Transformers: State-of-the-art Natural Language Processing},
+  author={Thomas Wolf and Lysandre Debut and Victor Sanh and Julien Chaumond and Clement Delangue and Anthony Moi and Pierric Cistac and Tim Rault and R'emi Louf and Morgan Funtowicz and Jamie Brew},
+  journal={ArXiv},
+  year={2019},
+  volume={abs/1910.03771}
 }
 ```

deploy_multi_version_doc.sh

+cd docs
+
+function deploy_doc(){
+	echo "Creating doc at commit $1 and pushing to folder $2"
+	git checkout $1
+	if [ ! -z "$2" ] 
+	then
+		echo "Pushing version" $2
+		make clean && make html && scp -r -oStrictHostKeyChecking=no _build/html $doc:$dir/$2
+	else
+		echo "Pushing master"
+		make clean && make html && scp -r -oStrictHostKeyChecking=no _build/html/* $doc:$dir
+	fi
+}
+
+deploy_doc "master" 
+deploy_doc "b33a385" v1.0.0
+deploy_doc "fe02e45" v1.1.0
+deploy_doc "89fd345" v1.2.0
+deploy_doc "fc9faa8" v2.0.0
+deploy_doc "3ddce1d" v2.1.1
+deploy_doc "f2f3294" v2.2.0
\ No newline at end of file

docs/source/_static/js/custom.js

 function addIcon() {
-    const huggingFaceLogo = "https://huggingface.co/assets/transformers-docs/huggingface_logo.svg";
+    const huggingFaceLogo = "https://huggingface.co/landing/assets/transformers-docs/huggingface_logo.svg";
     const image = document.createElement("img");
     image.setAttribute("src", huggingFaceLogo);
 
@@ -24,10 +24,10 @@ function addCustomFooter() {
     social.classList.add("footer__Social");
 
     const imageDetails = [
-        { link: "https://huggingface.co", imageLink: "https://huggingface.co/assets/transformers-docs/website.svg" },
-        { link: "https://twitter.com/huggingface", imageLink: "https://huggingface.co/assets/transformers-docs/twitter.svg" },
-        { link: "https://github.com/huggingface", imageLink: "https://huggingface.co/assets/transformers-docs/github.svg" },
-        { link: "https://www.linkedin.com/company/huggingface/", imageLink: "https://huggingface.co/assets/transformers-docs/linkedin.svg" }
+        { link: "https://huggingface.co", imageLink: "https://huggingface.co/landing/assets/transformers-docs/website.svg" },
+        { link: "https://twitter.com/huggingface", imageLink: "https://huggingface.co/landing/assets/transformers-docs/twitter.svg" },
+        { link: "https://github.com/huggingface", imageLink: "https://huggingface.co/landing/assets/transformers-docs/github.svg" },
+        { link: "https://www.linkedin.com/company/huggingface/", imageLink: "https://huggingface.co/landing/assets/transformers-docs/linkedin.svg" }
     ];
 
     imageDetails.forEach(imageLinks => {

docs/source/conf.py

@@ -26,7 +26,7 @@ author = u'huggingface'
 # The short X.Y version
 version = u''
 # The full version, including alpha/beta/rc tags
-release = u'2.1.1'
+release = u'2.2.1'
 
 
 # -- General configuration ---------------------------------------------------

docs/source/index.rst

@@ -47,6 +47,9 @@ The library currently contains PyTorch and Tensorflow implementations, pre-train
 6. `XLM <https://github.com/facebookresearch/XLM>`_ (from Facebook) released together with the paper `Cross-lingual Language Model Pretraining <https://arxiv.org/abs/1901.07291>`_ by Guillaume Lample and Alexis Conneau.
 7. `RoBERTa <https://github.com/pytorch/fairseq/tree/master/examples/roberta>`_ (from Facebook), released together with the paper a `Robustly Optimized BERT Pretraining Approach <https://arxiv.org/abs/1907.11692>`_ by Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov.
 8. `DistilBERT <https://huggingface.co/transformers/model_doc/distilbert.html>`_ (from HuggingFace) released together with the paper `DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter <https://arxiv.org/abs/1910.01108>`_ by Victor Sanh, Lysandre Debut and Thomas Wolf. The same method has been applied to compress GPT2 into `DistilGPT2 <https://github.com/huggingface/transformers/tree/master/examples/distillation>`_.
+9. `CTRL <https://github.com/pytorch/fairseq/tree/master/examples/ctrl>`_ (from Salesforce), released together with the paper `CTRL: A Conditional Transformer Language Model for Controllable Generation <https://www.github.com/salesforce/ctrl>`_ by Nitish Shirish Keskar*, Bryan McCann*, Lav R. Varshney, Caiming Xiong and Richard Socher.
+10. `CamemBERT <https://huggingface.co/transformers/model_doc/camembert.html>`_ (from FAIR, Inria, Sorbonne Université) released together with the paper `CamemBERT: a Tasty French Language Model <https://arxiv.org/abs/1911.03894>`_ by Louis Martin, Benjamin Muller, Pedro Javier Ortiz Suarez, Yoann Dupont, Laurent Romary, Eric Villemonte de la Clergerie, Djame Seddah, and Benoît Sagot.
+11. `ALBERT <https://github.com/google-research/ALBERT>`_ (from Google Research), released together with the paper a `ALBERT: A Lite BERT for Self-supervised Learning of Language Representations <https://arxiv.org/abs/1909.11942>`_ by Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, Radu Soricut.
 
 .. toctree::
     :maxdepth: 2
@@ -89,3 +92,5 @@ The library currently contains PyTorch and Tensorflow implementations, pre-train
     model_doc/roberta
     model_doc/distilbert
     model_doc/ctrl
+    model_doc/camembert
+    model_doc/albert

docs/source/installation.md

@@ -24,15 +24,24 @@ pip install [--editable] .
 
 An extensive test suite is included to test the library behavior and several examples. Library tests can be found in the [tests folder](https://github.com/huggingface/transformers/tree/master/transformers/tests) and examples tests in the [examples folder](https://github.com/huggingface/transformers/tree/master/examples).
 
-Tests can be run using `pytest` (install pytest if needed with `pip install pytest`).
+Tests can be run using `unittest` or `pytest` (install pytest if needed with `pip install pytest`).
 
 Run all the tests from the root of the cloned repository with the commands:
 
+```bash
+python -m unittest discover -s transformers/tests -p "*test.py" -t .
+python -m unittest discover -s examples -p "*test.py" -t examples
+```
+
+or
+
 ``` bash
 python -m pytest -sv ./transformers/tests/
 python -m pytest -sv ./examples/
 ```
 
+By default, slow tests are skipped. Set the `RUN_SLOW` environment variable to `yes` to run them.
+
 ## OpenAI GPT original tokenization workflow
 
 If you want to reproduce the original tokenization process of the `OpenAI GPT` paper, you will need to install `ftfy` (use version 4.4.3 if you are using Python 2) and `SpaCy`:

docs/source/main_classes/optimizer_schedules.rst

@@ -5,6 +5,7 @@ The ``.optimization`` module provides:
 
 - an optimizer with weight decay fixed that can be used to fine-tuned models, and
 - several schedules in the form of schedule objects that inherit from ``_LRSchedule``:
+- a gradient accumulation class to accumulate the gradients of multiple batches
 
 ``AdamW``
 ~~~~~~~~~~~~~~~~
@@ -12,25 +13,32 @@ The ``.optimization`` module provides:
 .. autoclass:: transformers.AdamW
     :members:
 
+``AdamWeightDecay``
+~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.AdamWeightDecay
+    :members:
+
+.. autofunction:: transformers.create_optimizer
+    :members:
+
 Schedules
 ----------------------------------------------------
 
 Learning Rate Schedules
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
-.. autoclass:: transformers.ConstantLRSchedule
-    :members:
+.. autofunction:: transformers.get_constant_schedule
 
 
-.. autoclass:: transformers.WarmupConstantSchedule
-    :members:
+.. autofunction:: transformers.get_constant_schedule_with_warmup
 
 .. image:: /imgs/warmup_constant_schedule.png
     :target: /imgs/warmup_constant_schedule.png
     :alt:
 
 
-.. autoclass:: transformers.WarmupCosineSchedule
+.. autofunction:: transformers.get_cosine_schedule_with_warmup
     :members:
 
 .. image:: /imgs/warmup_cosine_schedule.png
@@ -38,8 +46,7 @@ Learning Rate Schedules
     :alt:
 
 
-.. autoclass:: transformers.WarmupCosineWithHardRestartsSchedule
-    :members:
+.. autofunction:: transformers.get_cosine_with_hard_restarts_schedule_with_warmup
 
 .. image:: /imgs/warmup_cosine_hard_restarts_schedule.png
     :target: /imgs/warmup_cosine_hard_restarts_schedule.png
@@ -47,9 +54,22 @@ Learning Rate Schedules
 
 
 
-.. autoclass:: transformers.WarmupLinearSchedule
-    :members:
+.. autofunction:: transformers.get_linear_schedule_with_warmup
 
 .. image:: /imgs/warmup_linear_schedule.png
     :target: /imgs/warmup_linear_schedule.png
     :alt:
+
+``Warmup``
+~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.Warmup
+    :members:
+
+Gradient Strategies
+----------------------------------------------------
+
+``GradientAccumulator``
+~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.GradientAccumulator

docs/source/main_classes/processors.rst

@@ -54,5 +54,100 @@ Additionally, the following method  can be used to load values from a data file
 Example usage
 ^^^^^^^^^^^^^^^^^^^^^^^^^
 
+An example using these processors is given in the `run_glue.py <https://github.com/huggingface/pytorch-transformers/blob/master/examples/run_glue.py>`__ script.
+
+
+XNLI
+~~~~~~~~~~~~~~~~~~~~~
+
+`The Cross-Lingual NLI Corpus (XNLI) <https://www.nyu.edu/projects/bowman/xnli/>`__ is a benchmark that evaluates
+the quality of cross-lingual text representations. 
+XNLI is crowd-sourced dataset based on `MultiNLI <http://www.nyu.edu/projects/bowman/multinli/>`: pairs of text are labeled with textual entailment 
+annotations for 15 different languages (including both high-ressource language such as English and low-ressource languages such as Swahili).
+
+It was released together with the paper
+`XNLI: Evaluating Cross-lingual Sentence Representations <https://arxiv.org/abs/1809.05053>`__
+
+This library hosts the processor to load the XNLI data:
+    - :class:`~transformers.data.processors.utils.XnliProcessor`
+
+Please note that since the gold labels are available on the test set, evaluation is performed on the test set.
+
 An example using these processors is given in the
-`run_glue.py <https://github.com/huggingface/pytorch-transformers/blob/master/examples/run_glue.py>`__ script.
\ No newline at end of file
+`run_xnli.py <https://github.com/huggingface/pytorch-transformers/blob/master/examples/run_xnli.py>`__ script.
+
+
+SQuAD
+~~~~~~~~~~~~~~~~~~~~~
+
+`The Stanford Question Answering Dataset (SQuAD) <https://rajpurkar.github.io/SQuAD-explorer//>`__ is a benchmark that evaluates
+the performance of models on question answering. Two versions are available, v1.1 and v2.0. The first version (v1.1) was released together with the paper
+`SQuAD: 100,000+ Questions for Machine Comprehension of Text <https://arxiv.org/abs/1606.05250>`__. The second version (v2.0) was released alongside 
+the paper `Know What You Don't Know: Unanswerable Questions for SQuAD <https://arxiv.org/abs/1806.03822>`__.
+
+This library hosts a processor for each of the two versions:
+
+Processors
+^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Those processors are:
+    - :class:`~transformers.data.processors.utils.SquadV1Processor`
+    - :class:`~transformers.data.processors.utils.SquadV2Processor`
+
+They both inherit from the abstract class :class:`~transformers.data.processors.utils.SquadProcessor`
+
+.. autoclass:: transformers.data.processors.squad.SquadProcessor
+    :members:
+
+Additionally, the following method can be used to convert SQuAD examples into :class:`~transformers.data.processors.utils.SquadFeatures`
+that can be used as model inputs.
+
+.. automethod:: transformers.data.processors.squad.squad_convert_examples_to_features
+
+These processors as well as the aforementionned method can be used with files containing the data as well as with the `tensorflow_datasets` package.
+Examples are given below.
+
+
+Example usage
+^^^^^^^^^^^^^^^^^^^^^^^^^
+Here is an example using the processors as well as the conversion method using data files:
+
+Example::
+
+    # Loading a V2 processor
+    processor = SquadV2Processor()
+    examples = processor.get_dev_examples(squad_v2_data_dir)
+
+    # Loading a V1 processor
+    processor = SquadV1Processor()
+    examples = processor.get_dev_examples(squad_v1_data_dir)
+
+    features = squad_convert_examples_to_features( 
+        examples=examples,
+        tokenizer=tokenizer,
+        max_seq_length=max_seq_length,
+        doc_stride=args.doc_stride,
+        max_query_length=max_query_length,
+        is_training=not evaluate,
+    )
+
+Using `tensorflow_datasets` is as easy as using a data file:
+
+Example::
+
+    # tensorflow_datasets only handle Squad V1.
+    tfds_examples = tfds.load("squad")
+    examples = SquadV1Processor().get_examples_from_dataset(tfds_examples, evaluate=evaluate)
+
+    features = squad_convert_examples_to_features( 
+        examples=examples,
+        tokenizer=tokenizer,
+        max_seq_length=max_seq_length,
+        doc_stride=args.doc_stride,
+        max_query_length=max_query_length,
+        is_training=not evaluate,
+    )
+
+
+Another example using these processors is given in the
+`run_squad.py <https://github.com/huggingface/transformers/blob/master/examples/run_squad.py>`__ script.

docs/source/migration.md

@@ -84,12 +84,12 @@ Here is a conversion examples from `BertAdam` with a linear warmup and decay sch
 # Parameters:
 lr = 1e-3
 max_grad_norm = 1.0
-num_total_steps = 1000
+num_training_steps = 1000
 num_warmup_steps = 100
-warmup_proportion = float(num_warmup_steps) / float(num_total_steps)  # 0.1
+warmup_proportion = float(num_warmup_steps) / float(num_training_steps)  # 0.1
 
 ### Previously BertAdam optimizer was instantiated like this:
-optimizer = BertAdam(model.parameters(), lr=lr, schedule='warmup_linear', warmup=warmup_proportion, t_total=num_total_steps)
+optimizer = BertAdam(model.parameters(), lr=lr, schedule='warmup_linear', warmup=warmup_proportion, num_training_steps=num_training_steps)
 ### and used like this:
 for batch in train_data:
     loss = model(batch)
@@ -98,12 +98,12 @@ for batch in train_data:
 
 ### In Transformers, optimizer and schedules are splitted and instantiated like this:
 optimizer = AdamW(model.parameters(), lr=lr, correct_bias=False)  # To reproduce BertAdam specific behavior set correct_bias=False
-scheduler = WarmupLinearSchedule(optimizer, warmup_steps=num_warmup_steps, t_total=num_total_steps)  # PyTorch scheduler
+scheduler = get_linear_schedule_with_warmup(optimizer, num_warmup_steps=num_warmup_steps, num_training_steps=num_training_steps)  # PyTorch scheduler
 ### and used like this:
 for batch in train_data:
     loss = model(batch)
     loss.backward()
     torch.nn.utils.clip_grad_norm_(model.parameters(), max_grad_norm)  # Gradient clipping is not in AdamW anymore (so you can use amp without issue)
-    scheduler.step()
     optimizer.step()
+    scheduler.step()
 ```

docs/source/model_doc/albert.rst

+ALBERT
+----------------------------------------------------
+
+``AlbrtConfig``
+~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.AlbertConfig
+    :members:
+
+
+``AlbertTokenizer``
+~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.AlbertTokenizer
+    :members:
+
+
+``AlbertModel``
+~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.AlbertModel
+    :members:
+
+
+``AlbertForMaskedLM``
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.AlbertForMaskedLM
+    :members:
+
+
+``AlbertForSequenceClassification``
+~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.AlbertForSequenceClassification
+    :members:
+
+
+``AlbertForQuestionAnswering``
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.AlbertForQuestionAnswering
+    :members:
+
+
+``TFAlbertModel``
+~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFAlbertModel
+    :members:
+
+
+``TFAlbertForMaskedLM``
+~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFAlbertForMaskedLM
+    :members:
+
+
+``TFAlbertForSequenceClassification``
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFAlbertForSequenceClassification
+    :members:

docs/source/model_doc/camembert.rst

+CamemBERT
+----------------------------------------------------
+
+``CamembertConfig``
+~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.CamembertConfig
+    :members:
+
+
+``CamembertTokenizer``
+~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.CamembertTokenizer
+    :members:
+
+
+``CamembertModel``
+~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.CamembertModel
+    :members:
+
+
+``CamembertForMaskedLM``
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.CamembertForMaskedLM
+    :members:
+
+
+``CamembertForSequenceClassification``
+~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.CamembertForSequenceClassification
+    :members:
+
+
+``CamembertForMultipleChoice``
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.CamembertForMultipleChoice
+    :members:
+
+
+``CamembertForTokenClassification``
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.CamembertForTokenClassification
+    :members:

docs/source/model_doc/ctrl.rst

 CTRL
 ----------------------------------------------------
 
+Note: if you fine-tune a CTRL model using the Salesforce code (https://github.com/salesforce/ctrl),
+you'll be able to convert from TF to our HuggingFace/Transformers format using the 
+``convert_tf_to_huggingface_pytorch.py`` script (see `issue #1654 <https://github.com/huggingface/transformers/issues/1654>`_).
+
+
 ``CTRLConfig``
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 

docs/source/pretrained_models.rst

@@ -61,6 +61,24 @@ Here is the full list of the currently provided pretrained models together with
 |                   | ``bert-base-german-dbmdz-uncased``                         | | 12-layer, 768-hidden, 12-heads, 110M parameters.                                                                                    |
 |                   |                                                            | | Trained on uncased German text by DBMDZ                                                                                             |
 |                   |                                                            | (see `details on dbmdz repository <https://github.com/dbmdz/german-bert>`__).                                                         |
+|                   +------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
+|                   | ``bert-base-japanese``                                     | | 12-layer, 768-hidden, 12-heads, 110M parameters.                                                                                    |
+|                   |                                                            | | Trained on Japanese text. Text is tokenized with MeCab and WordPiece.                                                               |
+|                   |                                                            | | `MeCab <https://taku910.github.io/mecab/>`__ is required for tokenization.                                                          |
+|                   |                                                            | (see `details on cl-tohoku repository <https://github.com/cl-tohoku/bert-japanese>`__).                                               |
+|                   +------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
+|                   | ``bert-base-japanese-whole-word-masking``                  | | 12-layer, 768-hidden, 12-heads, 110M parameters.                                                                                    |
+|                   |                                                            | | Trained on Japanese text using Whole-Word-Masking. Text is tokenized with MeCab and WordPiece.                                      |
+|                   |                                                            | | `MeCab <https://taku910.github.io/mecab/>`__ is required for tokenization.                                                          |
+|                   |                                                            | (see `details on cl-tohoku repository <https://github.com/cl-tohoku/bert-japanese>`__).                                               |
+|                   +------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
+|                   | ``bert-base-japanese-char``                                | | 12-layer, 768-hidden, 12-heads, 110M parameters.                                                                                    |
+|                   |                                                            | | Trained on Japanese text. Text is tokenized into characters.                                                                        |
+|                   |                                                            | (see `details on cl-tohoku repository <https://github.com/cl-tohoku/bert-japanese>`__).                                               |
+|                   +------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
+|                   | ``bert-base-japanese-char-whole-word-masking``             | | 12-layer, 768-hidden, 12-heads, 110M parameters.                                                                                    |
+|                   |                                                            | | Trained on Japanese text using Whole-Word-Masking. Text is tokenized into characters.                                               |
+|                   |                                                            | (see `details on cl-tohoku repository <https://github.com/cl-tohoku/bert-japanese>`__).                                               |
 +-------------------+------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
 | GPT               | ``openai-gpt``                                             | | 12-layer, 768-hidden, 12-heads, 110M parameters.                                                                                    |
 |                   |                                                            | | OpenAI GPT English model                                                                                                            |
@@ -73,6 +91,9 @@ Here is the full list of the currently provided pretrained models together with
 |                   +------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
 |                   | ``gpt2-large``                                             | | 36-layer, 1280-hidden, 20-heads, 774M parameters.                                                                                   |
 |                   |                                                            | | OpenAI's Large-sized GPT-2 English model                                                                                            |
+|                   +------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
+|                   | ``gpt2-xl``                                                | | 48-layer, 1600-hidden, 25-heads, 1558M parameters.                                                                                  |
+|                   |                                                            | | OpenAI's XL-sized GPT-2 English model                                                                                               |
 +-------------------+------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
 | Transformer-XL    | ``transfo-xl-wt103``                                       | | 18-layer, 1024-hidden, 16-heads, 257M parameters.                                                                                   |
 |                   |                                                            | | English model trained on wikitext-103                                                                                               |
@@ -124,6 +145,14 @@ Here is the full list of the currently provided pretrained models together with
 |                   | ``roberta-large-mnli``                                     | | 24-layer, 1024-hidden, 16-heads, 355M parameters                                                                                    |
 |                   |                                                            | | ``roberta-large`` fine-tuned on `MNLI <http://www.nyu.edu/projects/bowman/multinli/>`__.                                            |
 |                   |                                                            | (see `details <https://github.com/pytorch/fairseq/tree/master/examples/roberta>`__)                                                   |
+|                   +------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
+|                   | ``roberta-base-openai-detector``                           | | 12-layer, 768-hidden, 12-heads, 125M parameters                                                                                     |
+|                   |                                                            | | ``roberta-base`` fine-tuned by OpenAI on the outputs of the 1.5B-parameter GPT-2 model.                                             |
+|                   |                                                            | (see `details <https://github.com/openai/gpt-2-output-dataset/tree/master/detector>`__)                                               |
+|                   +------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
+|                   | ``roberta-large-openai-detector``                          | | 24-layer, 1024-hidden, 16-heads, 355M parameters                                                                                    |
+|                   |                                                            | | ``roberta-large`` fine-tuned by OpenAI on the outputs of the 1.5B-parameter GPT-2 model.                                            |
+|                   |                                                            | (see `details <https://github.com/openai/gpt-2-output-dataset/tree/master/detector>`__)                                               |
 +-------------------+------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
 | DistilBERT        | ``distilbert-base-uncased``                                | | 6-layer, 768-hidden, 12-heads, 66M parameters                                                                                       |
 |                   |                                                            | | The DistilBERT model distilled from the BERT model `bert-base-uncased` checkpoint                                                   |
@@ -136,9 +165,58 @@ Here is the full list of the currently provided pretrained models together with
 |                   | ``distilgpt2``                                             | | 6-layer, 768-hidden, 12-heads, 82M parameters                                                                                       |
 |                   |                                                            | | The DistilGPT2 model distilled from the GPT2 model `gpt2` checkpoint.                                                               |
 |                   |                                                            | (see `details <https://github.com/huggingface/transformers/tree/master/examples/distillation>`__)                                     |
+|                   +------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
+|                   | ``distilroberta-base``                                     | | 6-layer, 768-hidden, 12-heads, 82M parameters                                                                                       |
+|                   |                                                            | | The DistilRoBERTa model distilled from the RoBERTa model `roberta-base` checkpoint.                                                 |
+|                   |                                                            | (see `details <https://github.com/huggingface/transformers/tree/master/examples/distillation>`__)                                     |
+|                   +------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
+|                   | ``distilbert-base-german-cased``                           | | 6-layer, 768-hidden, 12-heads, 66M parameters                                                                                       |
+|                   |                                                            | | The German DistilBERT model distilled from the German DBMDZ BERT model `bert-base-german-dbmdz-cased` checkpoint.                   |
+|                   |                                                            | (see `details <https://github.com/huggingface/transformers/tree/master/examples/distillation>`__)                                     |
+|                   +------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
+|                   | ``distilbert-base-multilingual-cased``                     | | 6-layer, 768-hidden, 12-heads, 134M parameters                                                                                      |
+|                   |                                                            | | The multilingual DistilBERT model distilled from the Multilingual BERT model `bert-base-multilingual-cased` checkpoint.             |
+|                   |                                                            | (see `details <https://github.com/huggingface/transformers/tree/master/examples/distillation>`__)                                     |
 +-------------------+------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
 | CTRL              | ``ctrl``                                                   | | 48-layer, 1280-hidden, 16-heads, 1.6B parameters                                                                                    |
 |                   |                                                            | | Salesforce's Large-sized CTRL English model                                                                                         |
 +-------------------+------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
+| CamemBERT         | ``camembert-base``                                         | | 12-layer, 768-hidden, 12-heads, 110M parameters                                                                                     |
+|                   |                                                            | | CamemBERT using the BERT-base architecture                                                                                          |
+|                   |                                                            | (see `details <https://github.com/pytorch/fairseq/tree/master/examples/camembert>`__)                                                 |
++-------------------+------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
+| ALBERT            | ``albert-base-v1``                                         | | 12 repeating layers, 128 embedding, 768-hidden, 12-heads, 11M parameters                                                            |
+|                   |                                                            | | ALBERT base model                                                                                                                   |
+|                   |                                                            | (see `details <https://github.com/google-research/ALBERT>`__)                                                                         |
+|                   +------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
+|                   | ``albert-large-v1``                                        | | 24 repeating layers, 128 embedding, 1024-hidden, 16-heads, 17M parameters                                                           |
+|                   |                                                            | | ALBERT large model                                                                                                                  |
+|                   |                                                            | (see `details <https://github.com/google-research/ALBERT>`__)                                                                         |
+|                   +------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
+|                   | ``albert-xlarge-v1``                                       | | 24 repeating layers, 128 embedding, 2048-hidden, 16-heads, 58M parameters                                                           |
+|                   |                                                            | | ALBERT xlarge model                                                                                                                 |
+|                   |                                                            | (see `details <https://github.com/google-research/ALBERT>`__)                                                                         |
+|                   +------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
+|                   | ``albert-xxlarge-v1``                                      | | 12 repeating layer, 128 embedding, 4096-hidden, 64-heads, 223M parameters                                                           |
+|                   |                                                            | | ALBERT xxlarge model                                                                                                                |
+|                   |                                                            | (see `details <https://github.com/google-research/ALBERT>`__)                                                                         |
+|                   +------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
+|                   | ``albert-base-v2``                                         | | 12 repeating layers, 128 embedding, 768-hidden, 12-heads, 11M parameters                                                            |
+|                   |                                                            | | ALBERT base model with no dropout, additional training data and longer training                                                     |
+|                   |                                                            | (see `details <https://github.com/google-research/ALBERT>`__)                                                                         |
+|                   +------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
+|                   | ``albert-large-v2``                                        | | 24 repeating layers, 128 embedding, 1024-hidden, 16-heads, 17M parameters                                                           |
+|                   |                                                            | | ALBERT large model with no dropout, additional training data and longer training                                                    |
+|                   |                                                            | (see `details <https://github.com/google-research/ALBERT>`__)                                                                         |
+|                   +------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
+|                   | ``albert-xlarge-v2``                                       | | 24 repeating layers, 128 embedding, 2048-hidden, 16-heads, 58M parameters                                                           |
+|                   |                                                            | | ALBERT xlarge model with no dropout, additional training data and longer training                                                   |
+|                   |                                                            | (see `details <https://github.com/google-research/ALBERT>`__)                                                                         |
+|                   +------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
+|                   | ``albert-xxlarge-v2``                                      | | 12 repeating layer, 128 embedding, 4096-hidden, 64-heads, 223M parameters                                                           |
+|                   |                                                            | | ALBERT xxlarge model with no dropout, additional training data and longer training                                                  |
+|                   |                                                            | (see `details <https://github.com/google-research/ALBERT>`__)                                                                         |
++-------------------+------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
+
 
-.. <https://huggingface.co/transformers/examples.html>`__
\ No newline at end of file
+.. <https://huggingface.co/transformers/examples.html>`__

docs/source/quickstart.md

@@ -188,3 +188,35 @@ assert predicted_text == 'Who was Jim Henson? Jim Henson was a man'
 ```
 
 Examples for each model class of each model architecture (Bert, GPT, GPT-2, Transformer-XL, XLNet and XLM) can be found in the [documentation](#documentation).
+
+#### Using the past
+
+GPT-2 as well as some other models (GPT, XLNet, Transfo-XL, CTRL) make use of a `past` or `mems` attribute which can be used to prevent re-computing the key/value pairs when using sequential decoding. It is useful when generating sequences as a big part of the attention mechanism benefits from previous computations.
+
+Here is a fully-working example using the `past` with `GPT2LMHeadModel` and argmax decoding (which should only be used as an example, as argmax decoding introduces a lot of repetition):
+
+```python
+from transformers import GPT2LMHeadModel, GPT2Tokenizer
+import torch
+
+tokenizer = GPT2Tokenizer.from_pretrained("gpt2")
+model = GPT2LMHeadModel.from_pretrained('gpt2')
+
+generated = tokenizer.encode("The Manhattan bridge")
+context = torch.tensor([generated])
+past = None
+
+for i in range(100):
+    print(i)
+    output, past = model(context, past=past)
+    token = torch.argmax(output[0, :])
+
+    generated += [token.tolist()]
+    context = token.unsqueeze(0)
+
+sequence = tokenizer.decode(generated)
+
+print(sequence)
+```
+
+The model only requires a single token as input as all the previous tokens' key/value pairs are contained in the `past`.
\ No newline at end of file

docs/source/serialization.rst

@@ -106,7 +106,7 @@ This section explain how you can save and re-load a fine-tuned model (BERT, GPT,
 There are three types of files you need to save to be able to reload a fine-tuned model:
 
 
-* the model it-self which should be saved following PyTorch serialization `best practices <https://pytorch.org/docs/stable/notes/serialization.html#best-practices>`__\ ,
+* the model itself which should be saved following PyTorch serialization `best practices <https://pytorch.org/docs/stable/notes/serialization.html#best-practices>`__\ ,
 * the configuration file of the model which is saved as a JSON file, and
 * the vocabulary (and the merges for the BPE-based models GPT and GPT-2).