Release: v4.1.1

Fix TAPAS doc
Put all models in the constants (#9170 )
2020-12-17 11:25:49 -05:00 · 2020-12-17 11:25:05 -05:00 · 2020-12-17 11:23:21 -05:00 · 2020-12-17 10:16:07 -05:00 · 2020-12-17 10:04:55 -05:00 · 2020-12-17 09:47:19 -05:00
1102 changed files with 17486 additions and 47646 deletions
@@ -3,28 +3,6 @@ orbs:
    gcp-gke: circleci/gcp-gke@1.0.4
    go: circleci/go@1.3.0

-commands:
-  skip-job-on-doc-only-changes:
-    description: "Do not continue this job and exit with success for PRs with only doc changes"
-    steps:
-
-      - run:
-          name: docs-only changes skip check
-          command: |
-            # pipeline.git.base_revision is not always defined, so only proceed if all external vars are defined
-            if test -n "<< pipeline.git.base_revision >>" && test -n "<< pipeline.git.revision >>" && test -n "$(git diff --name-only << pipeline.git.base_revision >>...<< pipeline.git.revision >>)"
-            then
-                if git diff --name-only << pipeline.git.base_revision >>...<< pipeline.git.revision >> | egrep -qv '\.(md|rst)$'
-                then
-                    echo "Non-docs were modified in this PR, proceeding normally"
-                else
-                    echo "Only docs were modified in this PR, quitting this job"
-                    # disable skipping for now, as circleCI's base_revision is inconsistent leading to invalid ranges
-                    # circleci step halt
-                fi
-            else
-                echo "Can't perform skipping check w/o base_revision defined, continuing the job"
-            fi

 # TPU REFERENCES
 references:
@@ -95,13 +73,13 @@ jobs:
        parallelism: 1
        steps:
            - checkout
-            - skip-job-on-doc-only-changes
            - restore_cache:
                  keys:
                      - v0.4-torch_and_tf-{{ checksum "setup.py" }}
                      - v0.4-{{ checksum "setup.py" }}
            - run: pip install --upgrade pip
            - run: pip install .[sklearn,tf-cpu,torch,testing,sentencepiece]
+            - run: pip install tapas torch-scatter -f https://pytorch-geometric.com/whl/torch-1.7.0+cpu.html
            - save_cache:
                key: v0.4-{{ checksum "setup.py" }}
                paths:
@@ -122,13 +100,13 @@ jobs:
        parallelism: 1
        steps:
            - checkout
-            - skip-job-on-doc-only-changes
            - restore_cache:
                  keys:
                      - v0.4-torch-{{ checksum "setup.py" }}
                      - v0.4-{{ checksum "setup.py" }}
            - run: pip install --upgrade pip
            - run: pip install .[sklearn,torch,testing,sentencepiece]
+            - run: pip install tapas torch-scatter -f https://pytorch-geometric.com/whl/torch-1.7.0+cpu.html
            - save_cache:
                  key: v0.4-torch-{{ checksum "setup.py" }}
                  paths:
@@ -149,7 +127,6 @@ jobs:
        parallelism: 1
        steps:
            - checkout
-            - skip-job-on-doc-only-changes
            - restore_cache:
                  keys:
                      - v0.4-tf-{{ checksum "setup.py" }}
@@ -176,7 +153,6 @@ jobs:
        parallelism: 1
        steps:
            - checkout
-            - skip-job-on-doc-only-changes
            - restore_cache:
                keys:
                    - v0.4-flax-{{ checksum "setup.py" }}
@@ -203,13 +179,13 @@ jobs:
        parallelism: 1
        steps:
            - checkout
-            - skip-job-on-doc-only-changes
            - restore_cache:
                  keys:
                      - v0.4-torch-{{ checksum "setup.py" }}
                      - v0.4-{{ checksum "setup.py" }}
            - run: pip install --upgrade pip
            - run: pip install .[sklearn,torch,testing,sentencepiece]
+            - run: pip install tapas torch-scatter -f https://pytorch-geometric.com/whl/torch-1.7.0+cpu.html
            - save_cache:
                  key: v0.4-torch-{{ checksum "setup.py" }}
                  paths:
@@ -230,7 +206,6 @@ jobs:
        parallelism: 1
        steps:
            - checkout
-            - skip-job-on-doc-only-changes
            - restore_cache:
                  keys:
                      - v0.4-tf-{{ checksum "setup.py" }}
@@ -255,7 +230,6 @@ jobs:
            RUN_CUSTOM_TOKENIZERS: yes
        steps:
            - checkout
-            - skip-job-on-doc-only-changes
            - restore_cache:
                  keys:
                      - v0.4-custom_tokenizers-{{ checksum "setup.py" }}
@@ -283,14 +257,13 @@ jobs:
        parallelism: 1
        steps:
            - checkout
-            - skip-job-on-doc-only-changes
            - restore_cache:
                  keys:
                      - v0.4-torch_examples-{{ checksum "setup.py" }}
                      - v0.4-{{ checksum "setup.py" }}
            - run: pip install --upgrade pip
            - run: pip install .[sklearn,torch,sentencepiece,testing]
-            - run: pip install -r examples/requirements.txt
+            - run: pip install -r examples/_tests_requirements.txt
            - save_cache:
                  key: v0.4-torch_examples-{{ checksum "setup.py" }}
                  paths:
@@ -53,4 +53,5 @@ deploy_doc "3ebb1b3" v3.2.0
 deploy_doc "0613f05" v3.3.1
 deploy_doc "eb0e0ce" v3.4.0
 deploy_doc "818878d" v3.5.1
-deploy_doc "c781171" # v4.0.0 Latest stable release
+deploy_doc "c781171" v4.0.0
+deploy_doc "f5438ab" # v4.1.0 Latest stable release
@@ -50,6 +50,7 @@ jobs:
          pip install --upgrade pip
          pip install .[torch,sklearn,testing,onnxruntime,sentencepiece]
          pip install git+https://github.com/huggingface/datasets
+          pip install pandas torch-scatter -f https://pytorch-geometric.com/whl/torch-1.7.0+cu102.html

      - name: Are GPUs recognized by our DL frameworks
        run: |
@@ -187,6 +188,7 @@ jobs:
          pip install --upgrade pip
          pip install .[torch,sklearn,testing,onnxruntime,sentencepiece]
          pip install git+https://github.com/huggingface/datasets
+          pip install pandas torch-scatter -f https://pytorch-geometric.com/whl/torch-1.7.0+cu102.html

      - name: Are GPUs recognized by our DL frameworks
        run: |
@@ -222,6 +222,7 @@ Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih.
 ultilingual BERT into [DistilmBERT](https://github.com/huggingface/transformers/tree/master/examples/distillation) and a German version of DistilBERT.
 1. **[SqueezeBert](https://huggingface.co/transformers/model_doc/squeezebert.html)** released with the paper [SqueezeBERT: What can computer vision teach NLP about efficient neural networks?](https://arxiv.org/abs/2006.11316) by Forrest N. Iandola, Albert E. Shaw, Ravi Krishna, and Kurt W. Keutzer.
 1. **[T5](https://huggingface.co/transformers/model_doc/t5.html)** (from Google AI) released with the paper [Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer](https://arxiv.org/abs/1910.10683) by Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
+1. **[TAPAS](https://huggingface.co/transformers/master/model_doc/tapas.html)** (from Google AI) released with the paper [TAPAS: Weakly Supervised Table Parsing via Pre-training](https://arxiv.org/abs/2004.02349) by Jonathan Herzig, Paweł Krzysztof Nowak, Thomas Müller, Francesco Piccinno and Julian Martin Eisenschlos.
 1. **[Transformer-XL](https://huggingface.co/transformers/model_doc/transformerxl.html)** (from Google/CMU) released with the paper [Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context](https://arxiv.org/abs/1901.02860) by Zihang Dai*, Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov.
 1. **[XLM](https://huggingface.co/transformers/model_doc/xlm.html)** (from Facebook) released together with the paper [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) by Guillaume Lample and Alexis Conneau.
 1. **[XLM-ProphetNet](https://huggingface.co/transformers/model_doc/xlmprophetnet.html)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
@@ -1,10 +1,11 @@
 // These two things need to be updated at each release for the version selector.
 // Last stable version
-const stableVersion = "v4.0.0"
+const stableVersion = "v4.1.0"
 // Dictionary doc folder to label. The last stable version should have an empty key.
 const versionMapping = {
    "master": "master",
-    "": "v4.0.0 (stable)",
+    "": "v4.1.0 (stable)",
+    "v4.0.1": "v4.0.0/v4.0.1",
    "v3.5.1": "v3.5.0/v3.5.1",
    "v3.4.0": "v3.4.0",
    "v3.3.1": "v3.3.0/v3.3.1",
@@ -34,5 +34,5 @@ help people access the inner representations, mainly adapted from the great work
  in https://arxiv.org/abs/1905.10650.

 To help you understand and use these features, we have added a specific example script: `bertology.py
-<https://github.com/huggingface/transformers/blob/master/examples/bertology/run_bertology.py>`_ while extract
-information and prune a model pre-trained on GLUE.
+<https://github.com/huggingface/transformers/blob/master/examples/research_projects/bertology/run_bertology.py>`_ while
+extract information and prune a model pre-trained on GLUE.
@@ -26,7 +26,7 @@ author = u'huggingface'
 # The short X.Y version
 version = u''
 # The full version, including alpha/beta/rc tags
-release = u'4.0.0'
+release = u'4.1.1'


 # -- General configuration ---------------------------------------------------
@@ -176,19 +176,22 @@ and conversion utilities for the following models:
 30. :doc:`T5 <model_doc/t5>` (from Google AI) released with the paper `Exploring the Limits of Transfer Learning with a
    Unified Text-to-Text Transformer <https://arxiv.org/abs/1910.10683>`__ by Colin Raffel and Noam Shazeer and Adam
    Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
-31. :doc:`Transformer-XL <model_doc/transformerxl>` (from Google/CMU) released with the paper `Transformer-XL:
+31. `TAPAS <https://huggingface.co/transformers/master/model_doc/tapas.html>`__ (from Google AI) released with the
+    paper `TAPAS: Weakly Supervised Table Parsing via Pre-training <https://arxiv.org/abs/2004.02349>`__ by Jonathan
+    Herzig, Paweł Krzysztof Nowak, Thomas Müller, Francesco Piccinno and Julian Martin Eisenschlos.
+32. :doc:`Transformer-XL <model_doc/transformerxl>` (from Google/CMU) released with the paper `Transformer-XL:
    Attentive Language Models Beyond a Fixed-Length Context <https://arxiv.org/abs/1901.02860>`__ by Zihang Dai*,
    Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov.
-32. :doc:`XLM <model_doc/xlm>` (from Facebook) released together with the paper `Cross-lingual Language Model
+33. :doc:`XLM <model_doc/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.
-33. :doc:`XLM-ProphetNet <model_doc/xlmprophetnet>` (from Microsoft Research) released with the paper `ProphetNet:
+34. :doc:`XLM-ProphetNet <model_doc/xlmprophetnet>` (from Microsoft Research) released with the paper `ProphetNet:
    Predicting Future N-gram for Sequence-to-Sequence Pre-training <https://arxiv.org/abs/2001.04063>`__ by Yu Yan,
    Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
-34. :doc:`XLM-RoBERTa <model_doc/xlmroberta>` (from Facebook AI), released together with the paper `Unsupervised
+35. :doc:`XLM-RoBERTa <model_doc/xlmroberta>` (from Facebook AI), released together with the paper `Unsupervised
    Cross-lingual Representation Learning at Scale <https://arxiv.org/abs/1911.02116>`__ by Alexis Conneau*, Kartikay
    Khandelwal*, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke
    Zettlemoyer and Veselin Stoyanov.
-35. :doc:`XLNet <model_doc/xlnet>` (from Google/CMU) released with the paper `XLNet: Generalized Autoregressive
+36. :doc:`XLNet <model_doc/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.

@@ -269,6 +272,8 @@ TensorFlow and/or Flax.
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |             T5              |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
+|            TAPAS            |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |       Transformer-XL        |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |             XLM             |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
@@ -382,6 +387,7 @@ TensorFlow and/or Flax.
    model_doc/roberta
    model_doc/squeezebert
    model_doc/t5
+    model_doc/tapas
    model_doc/transformerxl
    model_doc/xlm
    model_doc/xlmprophetnet
@@ -32,6 +32,9 @@ generation.
 .. autoclass:: transformers.LogitsProcessorList
    :members: __call__

+.. autoclass:: transformers.LogitsWarper
+    :members: __call__
+
 .. autoclass:: transformers.MinLengthLogitsProcessor
    :members: __call__

@@ -67,3 +70,10 @@ BeamSearch

 .. autoclass:: transformers.BeamSearchScorer
    :members: process, finalize
+
+Utilities
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autofunction:: transformers.top_k_top_p_filtering
+
+.. autofunction:: transformers.tf_top_k_top_p_filtering
@@ -91,8 +91,6 @@ TensorFlow loss functions
 TensorFlow Helper Functions
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

-.. autofunction:: transformers.modeling_tf_utils.cast_bool_to_primitive
-
 .. autofunction:: transformers.modeling_tf_utils.get_initializer

 .. autofunction:: transformers.modeling_tf_utils.keras_serializable
@@ -22,6 +22,8 @@ Utilities

 .. autoclass:: transformers.EvalPrediction

+.. autoclass:: transformers.EvaluationStrategy
+
 .. autofunction:: transformers.set_seed

 .. autofunction:: transformers.torch_distributed_zero_first
@@ -32,8 +34,15 @@ Callbacks internals

 .. autoclass:: transformers.trainer_callback.CallbackHandler

+
 Distributed Evaluation
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

 .. autoclass:: transformers.trainer_pt_utils.DistributedTensorGatherer
    :members:
+
+
+Distributed Evaluation
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.HfArgumentParser
@@ -13,9 +13,10 @@
 Models
 -----------------------------------------------------------------------------------------------------------------------

-The base classes :class:`~transformers.PreTrainedModel` and :class:`~transformers.TFPreTrainedModel` implement the
-common methods for loading/saving a model either from a local file or directory, or from a pretrained model
-configuration provided by the library (downloaded from HuggingFace's AWS S3 repository).
+The base classes :class:`~transformers.PreTrainedModel`, :class:`~transformers.TFPreTrainedModel`, and
+:class:`~transformers.FlaxPreTrainedModel` implement the common methods for loading/saving a model either from a local
+file or directory, or from a pretrained model configuration provided by the library (downloaded from HuggingFace's AWS
+S3 repository).

 :class:`~transformers.PreTrainedModel` and :class:`~transformers.TFPreTrainedModel` also implement a few methods which
 are common among all the models to:
@@ -57,6 +58,13 @@ TFModelUtilsMixin
    :members:


+FlaxPreTrainedModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxPreTrainedModel
+    :members:
+
+
 Generation
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

@@ -74,6 +74,10 @@ Learning Rate Schedules (Pytorch)
    :target: /imgs/warmup_linear_schedule.png
    :alt:

+
+.. autofunction:: transformers.get_polynomial_decay_schedule_with_warmup
+
+
 Warmup (TensorFlow)
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

@@ -34,6 +34,7 @@ There are two categories of pipeline abstractions to be aware about:
    - :class:`~transformers.TranslationPipeline`
    - :class:`~transformers.ZeroShotClassificationPipeline`
    - :class:`~transformers.Text2TextGenerationPipeline`
+    - :class:`~transformers.TableQuestionAnsweringPipeline`

 The pipeline abstraction
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
@@ -73,8 +74,9 @@ FillMaskPipeline
 NerPipeline
 =======================================================================================================================

-This class is an alias of the :class:`~transformers.TokenClassificationPipeline` defined below. Please refer to that
-pipeline for documentation and usage examples.
+.. autoclass:: transformers.NerPipeline
+
+See :class:`~transformers.TokenClassificationPipeline` for all details.

 QuestionAnsweringPipeline
 =======================================================================================================================
@@ -90,6 +92,13 @@ SummarizationPipeline
    :special-members: __call__
    :members:

+TableQuestionAnsweringPipeline
+=======================================================================================================================
+
+.. autoclass:: transformers.TableQuestionAnsweringPipeline
+    :special-members: __call__
+
+
 TextClassificationPipeline
 =======================================================================================================================

@@ -118,6 +127,13 @@ TokenClassificationPipeline
    :special-members: __call__
    :members:

+TranslationPipeline
+=======================================================================================================================
+
+.. autoclass:: transformers.TranslationPipeline
+    :special-members: __call__
+    :members:
+
 ZeroShotClassificationPipeline
 =======================================================================================================================

@@ -60,6 +60,13 @@ AlbertTokenizer
        create_token_type_ids_from_sequences, save_vocabulary


+AlbertTokenizerFast
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.AlbertTokenizerFast
+    :members:
+
+
 Albert specific outputs
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

@@ -114,6 +114,13 @@ AutoModelForQuestionAnswering
    :members:


+AutoModelForTableQuestionAnswering
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.AutoModelForTableQuestionAnswering
+    :members:
+
+
 TFAutoModel
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

@@ -175,3 +182,10 @@ TFAutoModelForQuestionAnswering

 .. autoclass:: transformers.TFAutoModelForQuestionAnswering
    :members:
+
+
+FlaxAutoModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxAutoModel
+    :members:
@@ -98,6 +98,12 @@ BartTokenizer
    :members:


+BartTokenizerFast
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.BartTokenizerFast
+    :members:
+

 BartModel
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
@@ -51,3 +51,9 @@ BarthezTokenizer
 .. autoclass:: transformers.BarthezTokenizer
    :members:

+
+BarthezTokenizerFast
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.BarthezTokenizerFast
+    :members:
@@ -207,3 +207,10 @@ FlaxBertModel

 .. autoclass:: transformers.FlaxBertModel
    :members: __call__
+
+
+FlaxBertForMaskedLM
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxBertForMaskedLM
+    :members: __call__
@@ -54,6 +54,13 @@ CamembertTokenizer
        create_token_type_ids_from_sequences, save_vocabulary


+CamembertTokenizerFast
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.CamembertTokenizerFast
+    :members:
+
+
 CamembertModel
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

@@ -138,3 +138,9 @@ TFOpenAIGPTDoubleHeadsModel

 .. autoclass:: transformers.TFOpenAIGPTDoubleHeadsModel
    :members: call
+
+TFOpenAIGPTForSequenceClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFOpenAIGPTForSequenceClassification
+    :members: call
@@ -57,6 +57,13 @@ LayoutLMTokenizer
    :members:


+LayoutLMTokenizerFast
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.LayoutLMTokenizerFast
+    :members:
+
+
 LayoutLMModel
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

@@ -34,6 +34,12 @@ contrast to most prior work, we also pretrain Longformer and finetune it on a va
 pretrained Longformer consistently outperforms RoBERTa on long document tasks and sets new state-of-the-art results on
 WikiHop and TriviaQA.*

+Tips:
+
+- Since the Longformer is based on RoBERTa, it doesn't have :obj:`token_type_ids`. You don't need to indicate which
+  token belongs to which segment. Just separate your segments with the separation token :obj:`tokenizer.sep_token` (or
+  :obj:`</s>`).
+
 The Authors' code can be found `here <https://github.com/allenai/longformer>`__.

 Longformer Self Attention
@@ -90,6 +90,13 @@ MBartTokenizer
    :members: build_inputs_with_special_tokens, prepare_seq2seq_batch


+MBartTokenizerFast
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.MBartTokenizerFast
+    :members:
+
+
 MBartForConditionalGeneration
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

@@ -1,3 +1,15 @@
+.. 
+    Copyright 2020 The HuggingFace Team. 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.
+
 MPNet
 -----------------------------------------------------------------------------------------------------------------------

@@ -37,6 +37,22 @@ MT5Config
    :members:


+MT5Tokenizer
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.MT5Tokenizer
+
+See :class:`~transformers.T5Tokenizer` for all details.
+
+
+MT5TokenizerFast
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.MT5TokenizerFast
+
+See :class:`~transformers.T5TokenizerFast` for all details.
+
+
 MT5Model
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

@@ -112,6 +112,13 @@ warning: ``add_tokens`` does not work at the moment.
    :members: __call__, prepare_seq2seq_batch


+PegasusTokenizerFast
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.PegasusTokenizerFast
+    :members:
+
+
 PegasusForConditionalGeneration
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

@@ -163,6 +163,13 @@ ReformerTokenizer
    :members: save_vocabulary


+ReformerTokenizerFast
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.ReformerTokenizerFast
+    :members:
+
+
 ReformerModel
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

@@ -107,6 +107,13 @@ T5Tokenizer
        create_token_type_ids_from_sequences, prepare_seq2seq_batch, save_vocabulary


+T5TokenizerFast
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.T5TokenizerFast
+    :members:
+
+
 T5Model
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

@@ -124,7 +131,7 @@ T5EncoderModel
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

 .. autoclass:: transformers.T5EncoderModel
-    :members: forward
+    :members: forward, parallelize, deparallelize

 TFT5Model
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
@@ -0,0 +1,434 @@
+TAPAS
+-----------------------------------------------------------------------------------------------------------------------
+
+.. note::
+
+    This is a recently introduced model so the API hasn't been tested extensively. There may be some bugs or slight
+    breaking changes to fix them in the future.
+
+
+
+Overview
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The TAPAS model was proposed in `TAPAS: Weakly Supervised Table Parsing via Pre-training
+<https://www.aclweb.org/anthology/2020.acl-main.398>`__ by Jonathan Herzig, Paweł Krzysztof Nowak, Thomas Müller,
+Francesco Piccinno and Julian Martin Eisenschlos. It's a BERT-based model specifically designed (and pre-trained) for
+answering questions about tabular data. Compared to BERT, TAPAS uses relative position embeddings and has 7 token types
+that encode tabular structure. TAPAS is pre-trained on the masked language modeling (MLM) objective on a large dataset
+comprising millions of tables from English Wikipedia and corresponding texts. For question answering, TAPAS has 2 heads
+on top: a cell selection head and an aggregation head, for (optionally) performing aggregations (such as counting or
+summing) among selected cells. TAPAS has been fine-tuned on several datasets: `SQA
+<https://www.microsoft.com/en-us/download/details.aspx?id=54253>`__ (Sequential Question Answering by Microsoft), `WTQ
+<https://github.com/ppasupat/WikiTableQuestions>`__ (Wiki Table Questions by Stanford University) and `WikiSQL
+<https://github.com/salesforce/WikiSQL>`__ (by Salesforce). It achieves state-of-the-art on both SQA and WTQ, while
+having comparable performance to SOTA on WikiSQL, with a much simpler architecture.
+
+The abstract from the paper is the following:
+
+*Answering natural language questions over tables is usually seen as a semantic parsing task. To alleviate the
+collection cost of full logical forms, one popular approach focuses on weak supervision consisting of denotations
+instead of logical forms. However, training semantic parsers from weak supervision poses difficulties, and in addition,
+the generated logical forms are only used as an intermediate step prior to retrieving the denotation. In this paper, we
+present TAPAS, an approach to question answering over tables without generating logical forms. TAPAS trains from weak
+supervision, and predicts the denotation by selecting table cells and optionally applying a corresponding aggregation
+operator to such selection. TAPAS extends BERT's architecture to encode tables as input, initializes from an effective
+joint pre-training of text segments and tables crawled from Wikipedia, and is trained end-to-end. We experiment with
+three different semantic parsing datasets, and find that TAPAS outperforms or rivals semantic parsing models by
+improving state-of-the-art accuracy on SQA from 55.1 to 67.2 and performing on par with the state-of-the-art on WIKISQL
+and WIKITQ, but with a simpler model architecture. We additionally find that transfer learning, which is trivial in our
+setting, from WIKISQL to WIKITQ, yields 48.7 accuracy, 4.2 points above the state-of-the-art.*
+
+In addition, the authors have further pre-trained TAPAS to recognize **table entailment**, by creating a balanced
+dataset of millions of automatically created training examples which are learned in an intermediate step prior to
+fine-tuning. The authors of TAPAS call this further pre-training intermediate pre-training (since TAPAS is first
+pre-trained on MLM, and then on another dataset). They found that intermediate pre-training further improves
+performance on SQA, achieving a new state-of-the-art as well as state-of-the-art on `TabFact
+<https://github.com/wenhuchen/Table-Fact-Checking>`__, a large-scale dataset with 16k Wikipedia tables for table
+entailment (a binary classification task). For more details, see their follow-up paper: `Understanding tables with
+intermediate pre-training <https://www.aclweb.org/anthology/2020.findings-emnlp.27/>`__ by Julian Martin Eisenschlos,
+Syrine Krichene and Thomas Müller.
+
+The original code can be found `here <https://github.com/google-research/tapas>`__.
+
+Tips:
+
+- TAPAS is a model that uses relative position embeddings by default (restarting the position embeddings at every cell
+  of the table). Note that this is something that was added after the publication of the original TAPAS paper.
+  According to the authors, this usually results in a slightly better performance, and allows you to encode longer
+  sequences without running out of embeddings. This is reflected in the ``reset_position_index_per_cell`` parameter of
+  :class:`~transformers.TapasConfig`, which is set to ``True`` by default. The default versions of the models available
+  in the `model hub <https://huggingface.co/models?search=tapas>`_ all use relative position embeddings. You can still
+  use the ones with absolute position embeddings by passing in an additional argument ``revision="no_reset"`` when
+  calling the ``.from_pretrained()`` method. Note that it's usually advised to pad the inputs on the right rather than
+  the left.
+- TAPAS is based on BERT, so ``TAPAS-base`` for example corresponds to a ``BERT-base`` architecture. Of course,
+  TAPAS-large will result in the best performance (the results reported in the paper are from TAPAS-large). Results of
+  the various sized models are shown on the `original Github repository <https://github.com/google-research/tapas>`_.
+- TAPAS has checkpoints fine-tuned on SQA, which are capable of answering questions related to a table in a
+  conversational set-up. This means that you can ask follow-up questions such as "what is his age?" related to the
+  previous question. Note that the forward pass of TAPAS is a bit different in case of a conversational set-up: in that
+  case, you have to feed every table-question pair one by one to the model, such that the `prev_labels` token type ids
+  can be overwritten by the predicted `labels` of the model to the previous question. See "Usage" section for more
+  info.
+- TAPAS is similar to BERT and therefore relies on the masked language modeling (MLM) objective. It is therefore
+  efficient at predicting masked tokens and at NLU in general, but is not optimal for text generation. Models trained
+  with a causal language modeling (CLM) objective are better in that regard.
+
+
+Usage: fine-tuning
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+Here we explain how you can fine-tune :class:`~transformers.TapasForQuestionAnswering` on your own dataset.
+
+**STEP 1: Choose one of the 3 ways in which you can use TAPAS - or experiment**
+
+Basically, there are 3 different ways in which one can fine-tune :class:`~transformers.TapasForQuestionAnswering`,
+corresponding to the different datasets on which Tapas was fine-tuned:
+
+1. SQA: if you're interested in asking follow-up questions related to a table, in a conversational set-up. For example
+   if you first ask "what's the name of the first actor?" then you can ask a follow-up question such as "how old is
+   he?". Here, questions do not involve any aggregation (all questions are cell selection questions).
+2. WTQ: if you're not interested in asking questions in a conversational set-up, but rather just asking questions
+   related to a table, which might involve aggregation, such as counting a number of rows, summing up cell values or
+   averaging cell values. You can then for example ask "what's the total number of goals Cristiano Ronaldo made in his
+   career?". This case is also called **weak supervision**, since the model itself must learn the appropriate
+   aggregation operator (SUM/COUNT/AVERAGE/NONE) given only the answer to the question as supervision.
+3. WikiSQL-supervised: this dataset is based on WikiSQL with the model being given the ground truth aggregation
+   operator during training. This is also called **strong supervision**. Here, learning the appropriate aggregation
+   operator is much easier.
+
+To summarize:
+
+------------------------------------+----------------------+-------------------------------------------------------------------------------------------------------------------+
+| **Task**                           | **Example dataset**  | **Description**                                                                                                   |
+------------------------------------+----------------------+-------------------------------------------------------------------------------------------------------------------+
+| Conversational                     | SQA                  | Conversational, only cell selection questions                                                                     |
+------------------------------------+----------------------+-------------------------------------------------------------------------------------------------------------------+
+| Weak supervision for aggregation   | WTQ                  | Questions might involve aggregation, and the model must learn this given only the answer as supervision           |
+------------------------------------+----------------------+-------------------------------------------------------------------------------------------------------------------+
+| Strong supervision for aggregation | WikiSQL-supervised   | Questions might involve aggregation, and the model must learn this given the gold aggregation operator            |
+------------------------------------+----------------------+-------------------------------------------------------------------------------------------------------------------+
+
+Initializing a model with a pre-trained base and randomly initialized classification heads from the model hub can be
+done as follows (be sure to have installed the `torch-scatter dependency <https://github.com/rusty1s/pytorch_scatter>`_
+for your environment):
+
+.. code-block::
+
+        >>> from transformers import TapasConfig, TapasForQuestionAnswering
+
+        >>> # for example, the base sized model with default SQA configuration
+        >>> model = TapasForQuestionAnswering.from_pretrained('google/tapas-base')
+
+        >>> # or, the base sized model with WTQ configuration
+        >>> config = TapasConfig.from_pretrained('google/tapas-base-finetuned-wtq')
+        >>> model = TapasForQuestionAnswering.from_pretrained('google/tapas-base', config=config)
+
+        >>> # or, the base sized model with WikiSQL configuration
+        >>> config = TapasConfig('google-base-finetuned-wikisql-supervised')
+        >>> model = TapasForQuestionAnswering.from_pretrained('google/tapas-base', config=config)
+
+
+Of course, you don't necessarily have to follow one of these three ways in which TAPAS was fine-tuned. You can also
+experiment by defining any hyperparameters you want when initializing :class:`~transformers.TapasConfig`, and then
+create a :class:`~transformers.TapasForQuestionAnswering` based on that configuration. For example, if you have a
+dataset that has both conversational questions and questions that might involve aggregation, then you can do it this
+way. Here's an example:
+
+.. code-block::
+
+        >>> from transformers import TapasConfig, TapasForQuestionAnswering
+
+        >>> # you can initialize the classification heads any way you want (see docs of TapasConfig)
+        >>> config = TapasConfig(num_aggregation_labels=3, average_logits_per_cell=True, select_one_column=False)
+        >>> # initializing the pre-trained base sized model with our custom classification heads
+        >>> model = TapasForQuestionAnswering.from_pretrained('google/tapas-base', config=config)
+
+What you can also do is start from an already fine-tuned checkpoint. A note here is that the already fine-tuned
+checkpoint on WTQ has some issues due to the L2-loss which is somewhat brittle. See `here
+<https://github.com/google-research/tapas/issues/91#issuecomment-735719340>`__ for more info.
+
+For a list of all pre-trained and fine-tuned TAPAS checkpoints available in the HuggingFace model hub, see `here
+<https://huggingface.co/models?search=tapas>`__.
+
+**STEP 2: Prepare your data in the SQA format**
+
+Second, no matter what you picked above, you should prepare your dataset in the `SQA format
+<https://www.microsoft.com/en-us/download/details.aspx?id=54253>`__. This format is a TSV/CSV file with the following
+columns:
+
+- ``id``: optional, id of the table-question pair, for bookkeeping purposes.
+- ``annotator``: optional, id of the person who annotated the table-question pair, for bookkeeping purposes.
+- ``position``: integer indicating if the question is the first, second, third,... related to the table. Only required
+  in case of conversational setup (SQA). You don't need this column in case you're going for WTQ/WikiSQL-supervised.
+- ``question``: string
+- ``table_file``: string, name of a csv file containing the tabular data
+- ``answer_coordinates``: list of one or more tuples (each tuple being a cell coordinate, i.e. row, column pair that is
+  part of the answer)
+- ``answer_text``: list of one or more strings (each string being a cell value that is part of the answer)
+- ``aggregation_label``: index of the aggregation operator. Only required in case of strong supervision for aggregation
+  (the WikiSQL-supervised case)
+- ``float_answer``: the float answer to the question, if there is one (np.nan if there isn't). Only required in case of
+  weak supervision for aggregation (such as WTQ and WikiSQL)
+
+The tables themselves should be present in a folder, each table being a separate csv file. Note that the authors of the
+TAPAS algorithm used conversion scripts with some automated logic to convert the other datasets (WTQ, WikiSQL) into the
+SQA format. The author explains this `here
+<https://github.com/google-research/tapas/issues/50#issuecomment-705465960>`__. Interestingly, these conversion scripts
+are not perfect (the ``answer_coordinates`` and ``float_answer`` fields are populated based on the ``answer_text``),
+meaning that WTQ and WikiSQL results could actually be improved.
+
+**STEP 3: Convert your data into PyTorch tensors using TapasTokenizer**
+
+Third, given that you've prepared your data in this TSV/CSV format (and corresponding CSV files containing the tabular
+data), you can then use :class:`~transformers.TapasTokenizer` to convert table-question pairs into :obj:`input_ids`,
+:obj:`attention_mask`, :obj:`token_type_ids` and so on. Again, based on which of the three cases you picked above,
+:class:`~transformers.TapasForQuestionAnswering` requires different inputs to be fine-tuned:
+
+------------------------------------+----------------------------------------------------------------------------------------------+
+| **Task**                           | **Required inputs**                                                                          |
+------------------------------------+----------------------------------------------------------------------------------------------+
+| Conversational                     | ``input_ids``, ``attention_mask``, ``token_type_ids``, ``labels``                            |
+------------------------------------+----------------------------------------------------------------------------------------------+
+| Weak supervision for aggregation   | ``input_ids``, ``attention_mask``, ``token_type_ids``, ``labels``, ``numeric_values``,       |
+|                                    | ``numeric_values_scale``, ``float_answer``                                                   |
+------------------------------------+----------------------------------------------------------------------------------------------+
+| Strong supervision for aggregation | ``input ids``, ``attention mask``, ``token type ids``, ``labels``, ``aggregation_labels``    |
+------------------------------------+----------------------------------------------------------------------------------------------+
+
+:class:`~transformers.TapasTokenizer` creates the ``labels``, ``numeric_values`` and ``numeric_values_scale`` based on
+the ``answer_coordinates`` and ``answer_text`` columns of the TSV file. The ``float_answer`` and ``aggregation_labels``
+are already in the TSV file of step 2. Here's an example:
+
+.. code-block::
+
+        >>> from transformers import TapasTokenizer
+        >>> import pandas as pd
+
+        >>> model_name = 'google/tapas-base'
+        >>> tokenizer = TapasTokenizer.from_pretrained(model_name)
+
+        >>> data = {'Actors': ["Brad Pitt", "Leonardo Di Caprio", "George Clooney"], 'Number of movies': ["87", "53", "69"]}
+        >>> queries = ["What is the name of the first actor?", "How many movies has George Clooney played in?", "What is the total number of movies?"]
+        >>> answer_coordinates = [[(0, 0)], [(2, 1)], [(0, 1), (1, 1), (2, 1)]]
+        >>> answer_text = [["Brad Pitt"], ["69"], ["209"]]
+        >>> table = pd.DataFrame.from_dict(data)
+        >>> inputs = tokenizer(table=table, queries=queries, answer_coordinates=answer_coordinates, answer_text=answer_text, padding='max_length', return_tensors='pt')
+        >>> inputs
+        {'input_ids': tensor([[ ... ]]), 'attention_mask': tensor([[...]]), 'token_type_ids': tensor([[[...]]]),
+        'numeric_values': tensor([[ ... ]]), 'numeric_values_scale: tensor([[ ... ]]), labels: tensor([[ ... ]])}
+
+Note that :class:`~transformers.TapasTokenizer` expects the data of the table to be **text-only**. You can use
+``.astype(str)`` on a dataframe to turn it into text-only data. Of course, this only shows how to encode a single
+training example. It is advised to create a PyTorch dataset and a corresponding dataloader:
+
+.. code-block::
+
+        >>> import torch
+        >>> import pandas as pd
+
+        >>> tsv_path = "your_path_to_the_tsv_file"
+        >>> table_csv_path = "your_path_to_a_directory_containing_all_csv_files"
+
+        >>> class TableDataset(torch.utils.data.Dataset):
+        ...     def __init__(self, data, tokenizer):
+        ...         self.data = data
+        ...         self.tokenizer = tokenizer
+        ...
+        ...     def __getitem__(self, idx):
+        ...         item = data.iloc[idx]
+        ...         table = pd.read_csv(table_csv_path + item.table_file).astype(str) # be sure to make your table data text only
+        ...         encoding = self.tokenizer(table=table, 
+        ...                                   queries=item.question, 
+        ...                                   answer_coordinates=item.answer_coordinates, 
+        ...                                   answer_text=item.answer_text,
+        ...                                   truncation=True,
+        ...                                   padding="max_length",
+        ...                                   return_tensors="pt"
+        ...         )
+        ...         # remove the batch dimension which the tokenizer adds by default
+        ...         encoding = {key: val.squeeze(0) for key, val in encoding.items()}
+        ...         # add the float_answer which is also required (weak supervision for aggregation case)
+        ...         encoding["float_answer"] = torch.tensor(item.float_answer) 
+        ...         return encoding
+        ...
+        ...     def __len__(self):
+        ...        return len(self.data)
+
+        >>> data = pd.read_csv(tsv_path, sep='\t')
+        >>> train_dataset = TableDataset(data, tokenizer)
+        >>> train_dataloader = torch.utils.data.DataLoader(train_dataset, batch_size=32)
+
+Note that here, we encode each table-question pair independently. This is fine as long as your dataset is **not
+conversational**. In case your dataset involves conversational questions (such as in SQA), then you should first group
+together the ``queries``, ``answer_coordinates`` and ``answer_text`` per table (in the order of their ``position``
+index) and batch encode each table with its questions. This will make sure that the ``prev_labels`` token types (see
+docs of :class:`~transformers.TapasTokenizer`) are set correctly. See `this notebook
+<https://github.com/NielsRogge/Transformers-Tutorials/blob/master/Fine_tuning_TapasForQuestionAnswering_on_SQA.ipynb>`__
+for more info.
+
+**STEP 4: Train (fine-tune) TapasForQuestionAnswering**
+
+You can then fine-tune :class:`~transformers.TapasForQuestionAnswering` using native PyTorch as follows (shown here for
+the weak supervision for aggregation case):
+
+.. code-block::
+
+        >>> from transformers import TapasConfig, TapasForQuestionAnswering, AdamW
+
+        >>> # this is the default WTQ configuration
+        >>> config = TapasConfig(
+        ...            num_aggregation_labels = 4,
+        ...            use_answer_as_supervision = True,
+        ...            answer_loss_cutoff = 0.664694,
+        ...            cell_selection_preference = 0.207951,
+        ...            huber_loss_delta = 0.121194,
+        ...            init_cell_selection_weights_to_zero = True,
+        ...            select_one_column = True,
+        ...            allow_empty_column_selection = False,
+        ...            temperature = 0.0352513,
+        ... )
+        >>> model = TapasForQuestionAnswering.from_pretrained("google/tapas-base", config=config)
+
+        >>> optimizer = AdamW(model.parameters(), lr=5e-5)
+
+        >>> for epoch in range(2):  # loop over the dataset multiple times
+        ...    for idx, batch in enumerate(train_dataloader):
+        ...         # get the inputs; 
+        ...         input_ids = batch["input_ids"]
+        ...         attention_mask = batch["attention_mask"]
+        ...         token_type_ids = batch["token_type_ids"]
+        ...         labels = batch["labels"]
+        ...         numeric_values = batch["numeric_values"]
+        ...         numeric_values_scale = batch["numeric_values_scale"]
+        ...         float_answer = batch["float_answer"]
+
+        ...         # zero the parameter gradients
+        ...         optimizer.zero_grad()
+
+        ...         # forward + backward + optimize
+        ...         outputs = model(input_ids=input_ids, attention_mask=attention_mask, token_type_ids=token_type_ids, 
+        ...                        labels=labels, numeric_values=numeric_values, numeric_values_scale=numeric_values_scale, 
+        ...                        float_answer=float_answer)
+        ...         loss = outputs.loss
+        ...         loss.backward()
+        ...         optimizer.step()
+
+Usage: inference
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+Here we explain how you can use :class:`~transformers.TapasForQuestionAnswering` for inference (i.e. making predictions
+on new data). For inference, only ``input_ids``, ``attention_mask`` and ``token_type_ids`` (which you can obtain using
+:class:`~transformers.TapasTokenizer`) have to be provided to the model to obtain the logits. Next, you can use the
+handy ``convert_logits_to_predictions`` method of :class:`~transformers.TapasTokenizer` to convert these into predicted
+coordinates and optional aggregation indices.
+
+However, note that inference is **different** depending on whether or not the setup is conversational. In a
+non-conversational set-up, inference can be done in parallel on all table-question pairs of a batch. Here's an example
+of that:
+
+.. code-block::
+
+        >>> from transformers import TapasTokenizer, TapasForQuestionAnswering
+        >>> import pandas as pd 
+
+        >>> model_name = 'google/tapas-base-finetuned-wtq'
+        >>> model = TapasForQuestionAnswering.from_pretrained(model_name)
+        >>> tokenizer = TapasTokenizer.from_pretrained(model_name)
+
+        >>> data = {'Actors': ["Brad Pitt", "Leonardo Di Caprio", "George Clooney"], 'Number of movies': ["87", "53", "69"]}
+        >>> queries = ["What is the name of the first actor?", "How many movies has George Clooney played in?", "What is the total number of movies?"]
+        >>> table = pd.DataFrame.from_dict(data)
+        >>> inputs = tokenizer(table=table, queries=queries, padding='max_length', return_tensors="pt") 
+        >>> outputs = model(**inputs)
+        >>> predicted_answer_coordinates, predicted_aggregation_indices = tokenizer.convert_logits_to_predictions(
+        ...         inputs, 
+        ...         outputs.logits.detach(), 
+        ...         outputs.logits_aggregation.detach()
+        ... )
+
+        >>> # let's print out the results:
+        >>> id2aggregation = {0: "NONE", 1: "SUM", 2: "AVERAGE", 3:"COUNT"}
+        >>> aggregation_predictions_string = [id2aggregation[x] for x in predicted_aggregation_indices]
+
+        >>> answers = []
+        >>> for coordinates in predicted_answer_coordinates:
+        ...   if len(coordinates) == 1:
+        ...     # only a single cell:
+        ...     answers.append(table.iat[coordinates[0]])
+        ...   else:
+        ...     # multiple cells
+        ...     cell_values = []
+        ...     for coordinate in coordinates:
+        ...        cell_values.append(table.iat[coordinate])
+        ...     answers.append(", ".join(cell_values))
+
+        >>> display(table)
+        >>> print("")
+        >>> for query, answer, predicted_agg in zip(queries, answers, aggregation_predictions_string):
+        ...   print(query)
+        ...   if predicted_agg == "NONE":
+        ...     print("Predicted answer: " + answer)
+        ...   else:
+        ...     print("Predicted answer: " + predicted_agg + " > " + answer)    
+        What is the name of the first actor?
+        Predicted answer: Brad Pitt
+        How many movies has George Clooney played in?
+        Predicted answer: COUNT > 69
+        What is the total number of movies?
+        Predicted answer: SUM > 87, 53, 69
+
+In case of a conversational set-up, then each table-question pair must be provided **sequentially** to the model, such
+that the ``prev_labels`` token types can be overwritten by the predicted ``labels`` of the previous table-question
+pair. Again, more info can be found in `this notebook
+<https://github.com/NielsRogge/Transformers-Tutorials/blob/master/Fine_tuning_TapasForQuestionAnswering_on_SQA.ipynb>`__.
+
+
+Tapas specific outputs
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.models.tapas.modeling_tapas.TableQuestionAnsweringOutput
+    :members:
+
+
+TapasConfig
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TapasConfig
+    :members:
+
+
+TapasTokenizer
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TapasTokenizer
+    :members: __call__, convert_logits_to_predictions, save_vocabulary
+
+
+TapasModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TapasModel
+    :members: forward
+
+
+TapasForMaskedLM
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TapasForMaskedLM
+    :members: forward
+
+
+TapasForSequenceClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TapasForSequenceClassification
+    :members: forward
+
+
+TapasForQuestionAnswering
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TapasForQuestionAnswering
+    :members: forward
@@ -105,3 +105,11 @@ TFTransfoXLLMHeadModel

 .. autoclass:: transformers.TFTransfoXLLMHeadModel
    :members: call
+
+
+Internal Layers
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.AdaptiveEmbedding
+
+.. autoclass:: transformers.TFAdaptiveEmbedding
@@ -62,6 +62,13 @@ XLMRobertaTokenizer
        create_token_type_ids_from_sequences, save_vocabulary


+XLMRobertaTokenizerFast
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.XLMRobertaTokenizerFast
+    :members:
+
+
 XLMRobertaModel
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

@@ -62,6 +62,13 @@ XLNetTokenizer
        create_token_type_ids_from_sequences, save_vocabulary


+XLNetTokenizerFast
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.XLNetTokenizerFast
+    :members:
+
+
 XLNet specific outputs
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

@@ -60,7 +60,7 @@ Basic steps
 In order to upload a model, you'll need to first create a git repo. This repo will live on the model hub, allowing
 users to clone it and you (and your organization members) to push to it.

-You can create a model repo directly from the website, `here <https://huggingface.co/new>`.
+You can create a model repo **directly from `the /new page on the website <https://huggingface.co/new>`__.**

 Alternatively, you can use the ``transformers-cli``. The next steps describe that process:

@@ -82,12 +82,12 @@ This creates a repo on the model hub, which can be cloned.

 .. code-block:: bash

-    git clone https://huggingface.co/username/your-model-name
-
    # Make sure you have git-lfs installed
    # (https://git-lfs.github.com/)
    git lfs install

+    git clone https://huggingface.co/username/your-model-name
+
 When you have your local clone of your repo and lfs installed, you can then add/remove from that clone as you would
 with any other git repo.

@@ -98,8 +98,12 @@ with any other git repo.
    echo "hello" >> README.md
    git add . && git commit -m "Update from $USER"

-We are intentionally not wrapping git too much, so as to stay intuitive and easy-to-use.
+We are intentionally not wrapping git too much, so that you can go on with the workflow you're used to and the tools
+you already know.

+The only learning curve you might have compared to regular git is the one for git-lfs. The documentation at
+`git-lfs.github.com <https://git-lfs.github.com/>`__ is decent, but we'll work on a tutorial with some tips and tricks
+in the coming weeks!

 Make your model work on all frameworks
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
@@ -110,7 +114,7 @@ Make your model work on all frameworks
 You probably have your favorite framework, but so will other users! That's why it's best to upload your model with both
 PyTorch `and` TensorFlow checkpoints to make it easier to use (if you skip this step, users will still be able to load
 your model in another framework, but it will be slower, as it will have to be converted on the fly). Don't worry, it's
-super easy to do (and in a future version, it will all be automatic). You will need to install both PyTorch and
+super easy to do (and in a future version, it might all be automatic). You will need to install both PyTorch and
 TensorFlow for this step, but you don't need to worry about the GPU, so it should be very easy. Check the `TensorFlow
 installation page <https://www.tensorflow.org/install/pip#tensorflow-2.0-rc-is-available>`__ and/or the `PyTorch
 installation page <https://pytorch.org/get-started/locally/#start-locally>`__ to see how.
@@ -192,7 +196,7 @@ status`` command:
    git add --all
    git status

-Finally, the files should be comitted:
+Finally, the files should be committed:

 .. code-block:: bash

@@ -210,23 +214,20 @@ This will upload the folder containing the weights, tokenizer and configuration
 Add a model card
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

-To make sure everyone knows what your model can do, what its limitations and potential bias or ethetical
-considerations, please add a README.md model card to the 🤗 Transformers repo under `model_cards/`. It should then be
-placed in a subfolder with your username or organization, then another subfolder named like your model
-(`awesome-name-you-picked`). Or just click on the "Create a model card on GitHub" button on the model page, it will get
-you directly to the right location. If you need one, `here <https://github.com/huggingface/model_card>`__ is a model
-card template (meta-suggestions are welcome).
+To make sure everyone knows what your model can do, what its limitations, potential bias or ethical considerations are,
+please add a README.md model card to your model repo. You can just create it, or there's also a convenient button
+titled "Add a README.md" on your model page. A model card template can be found `here
+<https://github.com/huggingface/model_card>`__ (meta-suggestions are welcome). model card template (meta-suggestions
+are welcome).
+
+.. note::
+
+    Model cards used to live in the 🤗 Transformers repo under `model_cards/`, but for consistency and scalability we
+    migrated every model card from the repo to its corresponding huggingface.co model repo.

 If your model is fine-tuned from another model coming from the model hub (all 🤗 Transformers pretrained models do),
 don't forget to link to its model card so that people can fully trace how your model was built.

-If you have never made a pull request to the 🤗 Transformers repo, look at the :doc:`contributing guide <contributing>`
-to see the steps to follow.
-
-.. note::
-
-    You can also send your model card in the folder you uploaded with the CLI by placing it in a `README.md` file
-    inside `path/to/awesome-name-you-picked/`.

 Using your model
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
@@ -262,7 +263,8 @@ First you need to install `git-lfs` in the environment used by the notebook:

    sudo apt-get install git-lfs

-Then you can use the :obj:`transformers-cli` to create your new repo:
+Then you can use either create a repo directly from `huggingface.co <https://huggingface.co/>`__ , or use the
+:obj:`transformers-cli` to create it:


 .. code-block:: bash
@@ -274,13 +276,14 @@ Once it's created, you can clone it and configure it (replace username by your u

 .. code-block:: bash

+    git lfs install
+
    git clone https://username:password@huggingface.co/username/your-model-name
    # Alternatively if you have a token,
    # you can use it instead of your password
    git clone https://username:token@huggingface.co/username/your-model-name

    cd your-model-name
-    git lfs install
    git config --global user.email "email@example.com"
    # Tip: using the same email than for your huggingface.co account will link your commits to your profile
    git config --global user.name "Your name"
@@ -1142,3 +1142,66 @@ To start a debugger at the point of the warning, do this:
 .. code-block:: bash

    pytest tests/test_logging.py -W error::UserWarning --pdb
+
+
+
+Testing Experimental CI Features
+-----------------------------------------------------------------------------------------------------------------------
+
+Testing CI features can be potentially problematic as it can interfere with the normal CI functioning. Therefore if a
+new CI feature is to be added, it should be done as following.
+
+1. Create a new dedicated job that tests what needs to be tested
+2. The new job must always succeed so that it gives us a green ✓ (details below).
+3. Let it run for some days to see that a variety of different PR types get to run on it (user fork branches,
+   non-forked branches, branches originating from github.com UI direct file edit, various forced pushes, etc. - there
+   are so many) while monitoring the experimental job's logs (not the overall job green as it's purposefully always
+   green)
+4. When it's clear that everything is solid, then merge the new changes into existing jobs.
+
+That way experiments on CI functionality itself won't interfere with the normal workflow.
+
+Now how can we make the job always succeed while the new CI feature is being developed?
+
+Some CIs, like TravisCI support ignore-step-failure and will report the overall job as successful, but CircleCI and
+Github Actions as of this writing don't support that.
+
+So the following workaround can be used:
+
+1. ``set +euo pipefail`` at the beginning of the run command to suppress most potential failures in the bash script.
+2. the last command must be a success: ``echo "done"`` or just ``true`` will do
+
+Here is an example:
+
+.. code-block:: yaml
+
+    - run:
+        name: run CI experiment
+        command: |
+            set +euo pipefail
+            echo "setting run-all-despite-any-errors-mode"
+            this_command_will_fail
+            echo "but bash continues to run"
+            # emulate another failure
+            false
+            # but the last command must be a success
+            echo "during experiment do not remove: reporting success to CI, even if there were failures"
+
+For simple commands you could also do:
+
+.. code-block:: bash
+
+    cmd_that_may_fail || true
+
+Of course, once satisfied with the results, integrate the experimental step or job with the rest of the normal jobs,
+while removing ``set +euo pipefail`` or any other things you may have added to ensure that the experimental job doesn't
+interfere with the normal CI functioning.
+
+This whole process would have been much easier if we only could set something like ``allow-failure`` for the
+experimental step, and let it fail without impacting the overall status of PRs. But as mentioned earlier CircleCI and
+Github Actions don't support it at the moment.
+
+You can vote for this feature and see where it is at at these CI-specific threads:
+
+* `Github Actions: <https://github.com/actions/toolkit/issues/399>`__
+* `CircleCI: <https://ideas.circleci.com/ideas/CCI-I-344>`__
@@ -16,59 +16,58 @@ limitations under the License.

 # Examples

-Version 2.9 of 🤗 Transformers introduced a new [`Trainer`](https://github.com/huggingface/transformers/blob/master/src/transformers/trainer.py) class for PyTorch, and its equivalent [`TFTrainer`](https://github.com/huggingface/transformers/blob/master/src/transformers/trainer_tf.py) for TF 2.
-Running the examples requires PyTorch 1.3.1+ or TensorFlow 2.2+.
-
-Here is the list of all our examples:
- **grouped by task** (all official examples work for multiple models)
- with information on whether they are **built on top of `Trainer`/`TFTrainer`** (if not, they still work, they might
-  just lack some features),
- whether or not they leverage the [🤗 Datasets](https://github.com/huggingface/datasets) library.
- links to **Colab notebooks** to walk through the scripts and run them easily,
- links to **Cloud deployments** to be able to deploy large-scale trainings in the Cloud with little to no setup.
-
+This folder contains actively maintained examples of use of 🤗 Transformers organized along NLP tasks. If you are looking for an example that used to
+be in this folder, it may have moved to our [research projects](https://github.com/huggingface/transformers/tree/master/examples/research_projects) subfolder (which contains frozen snapshots of research projects).

 ## Important note

 **Important**

-To make sure you can successfully run the latest versions of the example scripts, you have to **install the library from source** and install some example-specific requirements.
-Execute the following steps in a new virtual environment:
-
+To make sure you can successfully run the latest versions of the example scripts, you have to **install the library from source** and install some example-specific requirements. To do this, execute the following steps in a new virtual environment:
 ```bash
 git clone https://github.com/huggingface/transformers
 cd transformers
 pip install .
-pip install -r ./examples/requirements.txt
+```
+Then cd in the example folder of your choice and run
+```bash
+pip install -r requirements.txt
 ```

-Alternatively, you can run the version of the examples as they were for your current version of Transformers via (for instance with v3.4.0):
+Alternatively, you can run the version of the examples as they were for your current version of Transformers via (for instance with v3.5.1):
 ```bash
-git checkout tags/v3.4.0
+git checkout tags/v3.5.1
 ```

 ## The Big Table of Tasks

+Here is the list of all our examples:
+- with information on whether they are **built on top of `Trainer`/`TFTrainer`** (if not, they still work, they might
+  just lack some features),
+- whether or not they leverage the [🤗 Datasets](https://github.com/huggingface/datasets) library.
+- links to **Colab notebooks** to walk through the scripts and run them easily,
+<!--
+Coming soon!
+- links to **Cloud deployments** to be able to deploy large-scale trainings in the Cloud with little to no setup.
+-->
+
 | Task | Example datasets | Trainer support | TFTrainer support | 🤗 Datasets | Colab
 |---|---|:---:|:---:|:---:|:---:|
 | [**`language-modeling`**](https://github.com/huggingface/transformers/tree/master/examples/language-modeling)       | Raw text        | ✅ | -  | ✅ | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/blog/blob/master/notebooks/01_how_to_train.ipynb)
-| [**`text-classification`**](https://github.com/huggingface/transformers/tree/master/examples/text-classification)   | GLUE, XNLI      | ✅ | ✅ | ✅ | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://github.com/huggingface/notebooks/blob/master/examples/text_classification.ipynb)
-| [**`token-classification`**](https://github.com/huggingface/transformers/tree/master/examples/token-classification) | CoNLL NER       | ✅ | ✅ | ✅ | -
 | [**`multiple-choice`**](https://github.com/huggingface/transformers/tree/master/examples/multiple-choice)           | SWAG, RACE, ARC | ✅ | ✅ | - | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/ViktorAlm/notebooks/blob/master/MPC_GPU_Demo_for_TF_and_PT.ipynb)
-| [**`question-answering`**](https://github.com/huggingface/transformers/tree/master/examples/question-answering)     | SQuAD           | ✅ | ✅ | ✅ | -
-| [**`text-generation`**](https://github.com/huggingface/transformers/tree/master/examples/text-generation)           | -               | n/a | n/a | - | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/blog/blob/master/notebooks/02_how_to_generate.ipynb)
-| [**`distillation`**](https://github.com/huggingface/transformers/tree/master/examples/distillation)                 | All             | - | -  | - | -
+| [**`question-answering`**](https://github.com/huggingface/transformers/tree/master/examples/question-answering)     | SQuAD           | ✅ | ✅ | ✅ | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://github.com/huggingface/notebooks/blob/master/examples/question_answering.ipynb)
 | [**`summarization`**](https://github.com/huggingface/transformers/tree/master/examples/seq2seq)                     | CNN/Daily Mail  | ✅  | - | - | -
+| [**`text-classification`**](https://github.com/huggingface/transformers/tree/master/examples/text-classification)   | GLUE, XNLI      | ✅ | ✅ | ✅ | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://github.com/huggingface/notebooks/blob/master/examples/text_classification.ipynb)
+| [**`text-generation`**](https://github.com/huggingface/transformers/tree/master/examples/text-generation)           | -               | n/a | n/a | - | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/blog/blob/master/notebooks/02_how_to_generate.ipynb)
+| [**`token-classification`**](https://github.com/huggingface/transformers/tree/master/examples/token-classification) | CoNLL NER       | ✅ | ✅ | ✅ | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://github.com/huggingface/notebooks/blob/master/examples/token_classification.ipynb)
 | [**`translation`**](https://github.com/huggingface/transformers/tree/master/examples/seq2seq)                       | WMT             | ✅  | - | - | -
-| [**`bertology`**](https://github.com/huggingface/transformers/tree/master/examples/bertology)                       | -               | - | - | - | -
-| [**`adversarial`**](https://github.com/huggingface/transformers/tree/master/examples/adversarial)                   | HANS            | ✅ | - | - | -


-<br>
-
+<!--
 ## One-click Deploy to Cloud (wip)

 **Coming soon!**
+-->

 ## Running on TPUs

@@ -0,0 +1,20 @@
+tensorboard
+scikit-learn
+seqeval
+psutil
+sacrebleu
+rouge-score
+tensorflow_datasets
+matplotlib
+git-python==1.0.3
+faiss-cpu
+streamlit
+elasticsearch
+nltk
+pandas
+datasets >= 1.1.3
+fire
+pytest
+conllu
+sentencepiece != 0.1.92
+protobuf
@@ -1,3 +1,19 @@
+<!---
+Copyright 2020 The HuggingFace Team. 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.
+-->
+
 # 🤗 Benchmark results

 Here, you can find a list of the different benchmark results created by the community.
@@ -1,3 +1,17 @@
+# Copyright 2020 The HuggingFace Team. 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.
+
 import csv
 from collections import defaultdict
 from dataclasses import dataclass, field
@@ -1,3 +1,17 @@
+# Copyright 2020 The HuggingFace Team. 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.
+
 # tests directory-specific settings - this file is run automatically
 # by pytest before any tests are run

@@ -1,5 +0,0 @@
-# Community contributed examples
-
-This folder contains examples which are not actively maintained (mostly contributed by the community).
-
-Using these examples together with a recent version of the library usually requires to make small (sometimes big) adaptations to get the scripts working.
@@ -1,3 +1,19 @@
+<!---
+Copyright 2020 The HuggingFace Team. 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.
+-->
+
 ## Language model training

 Fine-tuning (or training from scratch) the library models for language modeling on a text dataset for GPT, GPT-2,
@@ -0,0 +1,3 @@
+datasets >= 1.1.3
+sentencepiece != 0.1.92
+protobuf
@@ -113,6 +113,12 @@ class DataTrainingArguments:
    overwrite_cache: bool = field(
        default=False, metadata={"help": "Overwrite the cached training and evaluation sets"}
    )
+    validation_split_percentage: Optional[int] = field(
+        default=5,
+        metadata={
+            "help": "The percentage of the train set used as validation set in case there's no validation split"
+        },
+    )
    preprocessing_num_workers: Optional[int] = field(
        default=None,
        metadata={"help": "The number of processes to use for the preprocessing."},
@@ -188,6 +194,17 @@ def main():
    if data_args.dataset_name is not None:
        # Downloading and loading a dataset from the hub.
        datasets = load_dataset(data_args.dataset_name, data_args.dataset_config_name)
+        if "validation" not in datasets.keys():
+            datasets["validation"] = load_dataset(
+                data_args.dataset_name,
+                data_args.dataset_config_name,
+                split=f"train[:{data_args.validation_split_percentage}%]",
+            )
+            datasets["train"] = load_dataset(
+                data_args.dataset_name,
+                data_args.dataset_config_name,
+                split=f"train[{data_args.validation_split_percentage}%:]",
+            )
    else:
        data_files = {}
        if data_args.train_file is not None:
@@ -103,6 +103,12 @@ class DataTrainingArguments:
    overwrite_cache: bool = field(
        default=False, metadata={"help": "Overwrite the cached training and evaluation sets"}
    )
+    validation_split_percentage: Optional[int] = field(
+        default=5,
+        metadata={
+            "help": "The percentage of the train set used as validation set in case there's no validation split"
+        },
+    )
    max_seq_length: Optional[int] = field(
        default=None,
        metadata={
@@ -199,6 +205,17 @@ def main():
    if data_args.dataset_name is not None:
        # Downloading and loading a dataset from the hub.
        datasets = load_dataset(data_args.dataset_name, data_args.dataset_config_name)
+        if "validation" not in datasets.keys():
+            datasets["validation"] = load_dataset(
+                data_args.dataset_name,
+                data_args.dataset_config_name,
+                split=f"train[:{data_args.validation_split_percentage}%]",
+            )
+            datasets["train"] = load_dataset(
+                data_args.dataset_name,
+                data_args.dataset_config_name,
+                split=f"train[{data_args.validation_split_percentage}%:]",
+            )
    else:
        data_files = {}
        if data_args.train_file is not None:
@@ -134,6 +134,12 @@ class DataTrainingArguments:
    overwrite_cache: bool = field(
        default=False, metadata={"help": "Overwrite the cached training and evaluation sets"}
    )
+    validation_split_percentage: Optional[int] = field(
+        default=5,
+        metadata={
+            "help": "The percentage of the train set used as validation set in case there's no validation split"
+        },
+    )
    max_seq_length: Optional[int] = field(
        default=None,
        metadata={
@@ -379,7 +385,7 @@ def training_step(optimizer, batch, dropout_rng):
        # Hide away tokens which doesn't participate in the optimization
        token_mask = jnp.where(targets > 0, 1.0, 0.0)

-        pooled, logits = model(**batch, params=params, dropout_rng=dropout_rng, train=True)
+        logits = model(**batch, params=params, dropout_rng=dropout_rng, train=True)[0]
        loss, weight_sum = cross_entropy(logits, targets, token_mask)
        return loss / weight_sum

@@ -401,7 +407,7 @@ def eval_step(params, batch):

    # Hide away tokens which doesn't participate in the optimization
    token_mask = jnp.where(targets > 0, 1.0, 0.0)
-    _, logits = model(**batch, params=params, train=False)
+    logits = model(**batch, params=params, train=False)[0]

    return compute_metrics(logits, targets, token_mask)

@@ -413,7 +419,7 @@ def generate_batch_splits(samples_idx: jnp.ndarray, batch_size: int) -> jnp.ndar
    if samples_to_remove != 0:
        samples_idx = samples_idx[:-samples_to_remove]
    sections_split = nb_samples // batch_size
-    batch_idx = jnp.split(samples_idx, sections_split)
+    batch_idx = np.split(samples_idx, sections_split)
    return batch_idx


@@ -473,6 +479,17 @@ if __name__ == "__main__":
    if data_args.dataset_name is not None:
        # Downloading and loading a dataset from the hub.
        datasets = load_dataset(data_args.dataset_name, data_args.dataset_config_name)
+        if "validation" not in datasets.keys():
+            datasets["validation"] = load_dataset(
+                data_args.dataset_name,
+                data_args.dataset_config_name,
+                split=f"train[:{data_args.validation_split_percentage}%]",
+            )
+            datasets["train"] = load_dataset(
+                data_args.dataset_name,
+                data_args.dataset_config_name,
+                split=f"train[{data_args.validation_split_percentage}%:]",
+            )
    else:
        data_files = {}
        if data_args.train_file is not None:
@@ -525,9 +542,9 @@ if __name__ == "__main__":

    def tokenize_function(examples):
        # Remove empty lines
-        examples["text"] = [line for line in examples["text"] if len(line) > 0 and not line.isspace()]
+        examples = [line for line in examples if len(line) > 0 and not line.isspace()]
        return tokenizer(
-            examples["text"],
+            examples,
            return_special_tokens_mask=True,
            padding=padding,
            truncation=True,
@@ -536,9 +553,10 @@ if __name__ == "__main__":

    tokenized_datasets = datasets.map(
        tokenize_function,
+        input_columns=[text_column_name],
        batched=True,
        num_proc=data_args.preprocessing_num_workers,
-        remove_columns=[text_column_name],
+        remove_columns=column_names,
        load_from_cache_file=not data_args.overwrite_cache,
    )

@@ -554,8 +572,13 @@ if __name__ == "__main__":
    rng = jax.random.PRNGKey(training_args.seed)
    dropout_rngs = jax.random.split(rng, jax.local_device_count())

-    model = FlaxBertForMaskedLM.from_pretrained("bert-base-cased", dtype=jnp.float32, dropout_rate=0.1)
-    model.init(jax.random.PRNGKey(training_args.seed), (training_args.train_batch_size, model.config.max_length))
+    model = FlaxBertForMaskedLM.from_pretrained(
+        "bert-base-cased",
+        dtype=jnp.float32,
+        input_shape=(training_args.train_batch_size, config.max_position_embeddings),
+        seed=training_args.seed,
+        dropout_rate=0.1,
+    )

    # Setup optimizer
    optimizer = Adam(
@@ -566,8 +589,9 @@ if __name__ == "__main__":
    ).create(model.params)

    # Create learning rate scheduler
+    # warmup_steps = 0 causes the Flax optimizer to return NaNs; warmup_steps = 1 is functionally equivalent.
    lr_scheduler_fn = create_learning_rate_scheduler(
-        base_learning_rate=training_args.learning_rate, warmup_steps=training_args.warmup_steps
+        base_learning_rate=training_args.learning_rate, warmup_steps=min(training_args.warmup_steps, 1)
    )

    # Create parallel version of the training and evaluation steps
@@ -606,13 +630,13 @@ if __name__ == "__main__":
        epochs.write(f"Loss: {loss}")

        # ======================== Evaluating ==============================
-        nb_eval_samples = len(tokenized_datasets["test"])
+        nb_eval_samples = len(tokenized_datasets["validation"])
        eval_samples_idx = jnp.arange(nb_eval_samples)
        eval_batch_idx = generate_batch_splits(eval_samples_idx, eval_batch_size)

        eval_metrics = []
        for i, batch_idx in enumerate(tqdm(eval_batch_idx, desc="Evaluating ...", position=2)):
-            samples = [tokenized_datasets["test"][int(idx)] for idx in batch_idx]
+            samples = [tokenized_datasets["validation"][int(idx)] for idx in batch_idx]
            model_inputs = data_collator(samples, pad_to_multiple_of=16)

            # Model forward
@@ -91,6 +91,12 @@ class DataTrainingArguments:
    Arguments pertaining to what data we are going to input our model for training and eval.
    """

+    dataset_name: Optional[str] = field(
+        default=None, metadata={"help": "The name of the dataset to use (via the datasets library)."}
+    )
+    dataset_config_name: Optional[str] = field(
+        default=None, metadata={"help": "The configuration name of the dataset to use (via the datasets library)."}
+    )
    train_file: Optional[str] = field(default=None, metadata={"help": "The input training data file (a text file)."})
    validation_file: Optional[str] = field(
        default=None,
@@ -107,6 +113,12 @@ class DataTrainingArguments:
    overwrite_cache: bool = field(
        default=False, metadata={"help": "Overwrite the cached training and evaluation sets"}
    )
+    validation_split_percentage: Optional[int] = field(
+        default=5,
+        metadata={
+            "help": "The percentage of the train set used as validation set in case there's no validation split"
+        },
+    )
    max_seq_length: Optional[int] = field(
        default=None,
        metadata={
@@ -203,15 +215,30 @@ def main():
    #
    # In distributed training, the load_dataset function guarantee that only one local process can concurrently
    # download the dataset.
-    data_files = {}
-    if data_args.train_file is not None:
-        data_files["train"] = data_args.train_file
-    if data_args.validation_file is not None:
-        data_files["validation"] = data_args.validation_file
-    extension = data_args.train_file.split(".")[-1]
-    if extension == "txt":
-        extension = "text"
-    datasets = load_dataset(extension, data_files=data_files)
+    if data_args.dataset_name is not None:
+        # Downloading and loading a dataset from the hub.
+        datasets = load_dataset(data_args.dataset_name, data_args.dataset_config_name)
+        if "validation" not in datasets.keys():
+            datasets["validation"] = load_dataset(
+                data_args.dataset_name,
+                data_args.dataset_config_name,
+                split=f"train[:{data_args.validation_split_percentage}%]",
+            )
+            datasets["train"] = load_dataset(
+                data_args.dataset_name,
+                data_args.dataset_config_name,
+                split=f"train[{data_args.validation_split_percentage}%:]",
+            )
+    else:
+        data_files = {}
+        if data_args.train_file is not None:
+            data_files["train"] = data_args.train_file
+        if data_args.validation_file is not None:
+            data_files["validation"] = data_args.validation_file
+        extension = data_args.train_file.split(".")[-1]
+        if extension == "txt":
+            extension = "text"
+        datasets = load_dataset(extension, data_files=data_files)
    # See more about loading any type of standard or custom dataset (from files, python dict, pandas DataFrame, etc) at
    # https://huggingface.co/docs/datasets/loading_datasets.html.

@@ -281,7 +308,7 @@ def main():
    # Add the chinese references if provided
    if data_args.train_ref_file is not None:
        tokenized_datasets["train"] = add_chinese_references(tokenized_datasets["train"], data_args.train_ref_file)
-    if data_args.valid_ref_file is not None:
+    if data_args.validation_ref_file is not None:
        tokenized_datasets["validation"] = add_chinese_references(
            tokenized_datasets["validation"], data_args.validation_ref_file
        )
@@ -93,6 +93,12 @@ class DataTrainingArguments:
    overwrite_cache: bool = field(
        default=False, metadata={"help": "Overwrite the cached training and evaluation sets"}
    )
+    validation_split_percentage: Optional[int] = field(
+        default=5,
+        metadata={
+            "help": "The percentage of the train set used as validation set in case there's no validation split"
+        },
+    )
    max_seq_length: int = field(
        default=512,
        metadata={
@@ -196,6 +202,17 @@ def main():
    if data_args.dataset_name is not None:
        # Downloading and loading a dataset from the hub.
        datasets = load_dataset(data_args.dataset_name, data_args.dataset_config_name)
+        if "validation" not in datasets.keys():
+            datasets["validation"] = load_dataset(
+                data_args.dataset_name,
+                data_args.dataset_config_name,
+                split=f"train[:{data_args.validation_split_percentage}%]",
+            )
+            datasets["train"] = load_dataset(
+                data_args.dataset_name,
+                data_args.dataset_config_name,
+                split=f"train[{data_args.validation_split_percentage}%:]",
+            )
    else:
        data_files = {}
        if data_args.train_file is not None:
@@ -0,0 +1,21 @@
+<!---
+Copyright 2020 The HuggingFace Team. 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.
+-->
+
+# Legacy examples
+
+This folder contains examples which are not actively maintained (mostly contributed by the community).
+
+Using these examples together with a recent version of the library usually requires to make small (sometimes big) adaptations to get the scripts working.
@@ -21,7 +21,7 @@ mkdir -p $OUTPUT_DIR
 # Add parent directory to python path to access lightning_base.py
 export PYTHONPATH="../":"${PYTHONPATH}"

-python3 run_pl_glue.py --gpus 1 --data_dir $DATA_DIR \
+python3 run_glue.py --gpus 1 --data_dir $DATA_DIR \
 --task $TASK \
 --model_name_or_path $BERT_MODEL \
 --output_dir $OUTPUT_DIR \
@@ -31,7 +31,7 @@ mkdir -p $OUTPUT_DIR
 # Add parent directory to python path to access lightning_base.py
 export PYTHONPATH="../":"${PYTHONPATH}"

-python3 run_pl_ner.py --data_dir ./ \
+python3 run_ner.py --data_dir ./ \
 --labels ./labels.txt \
 --model_name_or_path $BERT_MODEL \
 --output_dir $OUTPUT_DIR \
@@ -26,7 +26,7 @@ export SEED=1
 # Add parent directory to python path to access lightning_base.py
 export PYTHONPATH="../":"${PYTHONPATH}"

-python3 run_pl_ner.py --data_dir ./ \
+python3 run_ner.py --data_dir ./ \
 --task_type POS \
 --model_name_or_path $BERT_MODEL \
 --output_dir $OUTPUT_DIR \
@@ -0,0 +1,229 @@
+## Token classification
+
+Based on the scripts [`run_ner.py`](https://github.com/huggingface/transformers/blob/master/examples/contrib/legacy/token-classification/run_ner.py).
+
+The following examples are covered in this section:
+
+* NER on the GermEval 2014 (German NER) dataset
+* Emerging and Rare Entities task: WNUT’17 (English NER) dataset
+
+Details and results for the fine-tuning provided by @stefan-it.
+
+### GermEval 2014 (German NER) dataset
+
+#### Data (Download and pre-processing steps)
+
+Data can be obtained from the [GermEval 2014](https://sites.google.com/site/germeval2014ner/data) shared task page.
+
+Here are the commands for downloading and pre-processing train, dev and test datasets. The original data format has four (tab-separated) columns, in a pre-processing step only the two relevant columns (token and outer span NER annotation) are extracted:
+
+```bash
+curl -L 'https://drive.google.com/uc?export=download&id=1Jjhbal535VVz2ap4v4r_rN1UEHTdLK5P' \
+| grep -v "^#" | cut -f 2,3 | tr '\t' ' ' > train.txt.tmp
+curl -L 'https://drive.google.com/uc?export=download&id=1ZfRcQThdtAR5PPRjIDtrVP7BtXSCUBbm' \
+| grep -v "^#" | cut -f 2,3 | tr '\t' ' ' > dev.txt.tmp
+curl -L 'https://drive.google.com/uc?export=download&id=1u9mb7kNJHWQCWyweMDRMuTFoOHOfeBTH' \
+| grep -v "^#" | cut -f 2,3 | tr '\t' ' ' > test.txt.tmp
+```
+
+The GermEval 2014 dataset contains some strange "control character" tokens like `'\x96', '\u200e', '\x95', '\xad' or '\x80'`.
+One problem with these tokens is, that `BertTokenizer` returns an empty token for them, resulting in misaligned `InputExample`s.
+The `preprocess.py` script located in the `scripts` folder a) filters these tokens and b) splits longer sentences into smaller ones (once the max. subtoken length is reached).
+
+Let's define some variables that we need for further pre-processing steps and training the model:
+
+```bash
+export MAX_LENGTH=128
+export BERT_MODEL=bert-base-multilingual-cased
+```
+
+Run the pre-processing script on training, dev and test datasets:
+
+```bash
+python3 scripts/preprocess.py train.txt.tmp $BERT_MODEL $MAX_LENGTH > train.txt
+python3 scripts/preprocess.py dev.txt.tmp $BERT_MODEL $MAX_LENGTH > dev.txt
+python3 scripts/preprocess.py test.txt.tmp $BERT_MODEL $MAX_LENGTH > test.txt
+```
+
+The GermEval 2014 dataset has much more labels than CoNLL-2002/2003 datasets, so an own set of labels must be used:
+
+```bash
+cat train.txt dev.txt test.txt | cut -d " " -f 2 | grep -v "^$"| sort | uniq > labels.txt
+```
+
+#### Prepare the run
+
+Additional environment variables must be set:
+
+```bash
+export OUTPUT_DIR=germeval-model
+export BATCH_SIZE=32
+export NUM_EPOCHS=3
+export SAVE_STEPS=750
+export SEED=1
+```
+
+#### Run the Pytorch version
+
+To start training, just run:
+
+```bash
+python3 run_ner.py --data_dir ./ \
+--labels ./labels.txt \
+--model_name_or_path $BERT_MODEL \
+--output_dir $OUTPUT_DIR \
+--max_seq_length  $MAX_LENGTH \
+--num_train_epochs $NUM_EPOCHS \
+--per_device_train_batch_size $BATCH_SIZE \
+--save_steps $SAVE_STEPS \
+--seed $SEED \
+--do_train \
+--do_eval \
+--do_predict
+```
+
+If your GPU supports half-precision training, just add the `--fp16` flag. After training, the model will be both evaluated on development and test datasets.
+
+#### JSON-based configuration file
+
+Instead of passing all parameters via commandline arguments, the `run_ner.py` script also supports reading parameters from a json-based configuration file:
+
+```json
+{
+    "data_dir": ".",
+    "labels": "./labels.txt",
+    "model_name_or_path": "bert-base-multilingual-cased",
+    "output_dir": "germeval-model",
+    "max_seq_length": 128,
+    "num_train_epochs": 3,
+    "per_device_train_batch_size": 32,
+    "save_steps": 750,
+    "seed": 1,
+    "do_train": true,
+    "do_eval": true,
+    "do_predict": true
+}
+```
+
+It must be saved with a `.json` extension and can be used by running `python3 run_ner.py config.json`.
+
+#### Evaluation
+
+Evaluation on development dataset outputs the following for our example:
+
+```bash
+10/04/2019 00:42:06 - INFO - __main__ -   ***** Eval results  *****
+10/04/2019 00:42:06 - INFO - __main__ -     f1 = 0.8623348017621146
+10/04/2019 00:42:06 - INFO - __main__ -     loss = 0.07183869666975543
+10/04/2019 00:42:06 - INFO - __main__ -     precision = 0.8467916366258111
+10/04/2019 00:42:06 - INFO - __main__ -     recall = 0.8784592370979806
+```
+
+On the test dataset the following results could be achieved:
+
+```bash
+10/04/2019 00:42:42 - INFO - __main__ -   ***** Eval results  *****
+10/04/2019 00:42:42 - INFO - __main__ -     f1 = 0.8614389652384803
+10/04/2019 00:42:42 - INFO - __main__ -     loss = 0.07064602487454782
+10/04/2019 00:42:42 - INFO - __main__ -     precision = 0.8604651162790697
+10/04/2019 00:42:42 - INFO - __main__ -     recall = 0.8624150210424085
+```
+
+### Emerging and Rare Entities task: WNUT’17 (English NER) dataset
+
+Description of the WNUT’17 task from the [shared task website](http://noisy-text.github.io/2017/index.html):
+
+> The WNUT’17 shared task focuses on identifying unusual, previously-unseen entities in the context of emerging discussions.
+> Named entities form the basis of many modern approaches to other tasks (like event clustering and summarization), but recall on
+> them is a real problem in noisy text - even among annotators. This drop tends to be due to novel entities and surface forms.
+
+Six labels are available in the dataset. An overview can be found on this [page](http://noisy-text.github.io/2017/files/).
+
+#### Data (Download and pre-processing steps)
+
+The dataset can be downloaded from the [official GitHub](https://github.com/leondz/emerging_entities_17) repository.
+
+The following commands show how to prepare the dataset for fine-tuning:
+
+```bash
+mkdir -p data_wnut_17
+
+curl -L 'https://github.com/leondz/emerging_entities_17/raw/master/wnut17train.conll'  | tr '\t' ' ' > data_wnut_17/train.txt.tmp
+curl -L 'https://github.com/leondz/emerging_entities_17/raw/master/emerging.dev.conll' | tr '\t' ' ' > data_wnut_17/dev.txt.tmp
+curl -L 'https://raw.githubusercontent.com/leondz/emerging_entities_17/master/emerging.test.annotated' | tr '\t' ' ' > data_wnut_17/test.txt.tmp
+```
+
+Let's define some variables that we need for further pre-processing steps:
+
+```bash
+export MAX_LENGTH=128
+export BERT_MODEL=bert-large-cased
+```
+
+Here we use the English BERT large model for fine-tuning.
+The `preprocess.py` scripts splits longer sentences into smaller ones (once the max. subtoken length is reached):
+
+```bash
+python3 scripts/preprocess.py data_wnut_17/train.txt.tmp $BERT_MODEL $MAX_LENGTH > data_wnut_17/train.txt
+python3 scripts/preprocess.py data_wnut_17/dev.txt.tmp $BERT_MODEL $MAX_LENGTH > data_wnut_17/dev.txt
+python3 scripts/preprocess.py data_wnut_17/test.txt.tmp $BERT_MODEL $MAX_LENGTH > data_wnut_17/test.txt
+```
+
+In the last pre-processing step, the `labels.txt` file needs to be generated. This file contains all available labels:
+
+```bash
+cat data_wnut_17/train.txt data_wnut_17/dev.txt data_wnut_17/test.txt | cut -d " " -f 2 | grep -v "^$"| sort | uniq > data_wnut_17/labels.txt
+```
+
+#### Run the Pytorch version
+
+Fine-tuning with the PyTorch version can be started using the `run_ner.py` script. In this example we use a JSON-based configuration file.
+
+This configuration file looks like:
+
+```json
+{
+    "data_dir": "./data_wnut_17",
+    "labels": "./data_wnut_17/labels.txt",
+    "model_name_or_path": "bert-large-cased",
+    "output_dir": "wnut-17-model-1",
+    "max_seq_length": 128,
+    "num_train_epochs": 3,
+    "per_device_train_batch_size": 32,
+    "save_steps": 425,
+    "seed": 1,
+    "do_train": true,
+    "do_eval": true,
+    "do_predict": true,
+    "fp16": false
+}
+```
+
+If your GPU supports half-precision training, please set `fp16` to `true`.
+
+Save this JSON-based configuration under `wnut_17.json`. The fine-tuning can be started with `python3 run_ner_old.py wnut_17.json`.
+
+#### Evaluation
+
+Evaluation on development dataset outputs the following:
+
+```bash
+05/29/2020 23:33:44 - INFO - __main__ -   ***** Eval results *****
+05/29/2020 23:33:44 - INFO - __main__ -     eval_loss = 0.26505235286212275
+05/29/2020 23:33:44 - INFO - __main__ -     eval_precision = 0.7008264462809918
+05/29/2020 23:33:44 - INFO - __main__ -     eval_recall = 0.507177033492823
+05/29/2020 23:33:44 - INFO - __main__ -     eval_f1 = 0.5884802220680084
+05/29/2020 23:33:44 - INFO - __main__ -     epoch = 3.0
+```
+
+On the test dataset the following results could be achieved:
+
+```bash
+05/29/2020 23:33:44 - INFO - transformers.trainer -   ***** Running Prediction *****
+05/29/2020 23:34:02 - INFO - __main__ -     eval_loss = 0.30948806500973547
+05/29/2020 23:34:02 - INFO - __main__ -     eval_precision = 0.5840108401084011
+05/29/2020 23:34:02 - INFO - __main__ -     eval_recall = 0.3994439295644115
+05/29/2020 23:34:02 - INFO - __main__ -     eval_f1 = 0.47440836543753434
+```
+
+WNUT’17 is a very difficult task. Current state-of-the-art results on this dataset can be found [here](http://nlpprogress.com/english/named_entity_recognition.html).
@@ -20,7 +20,7 @@ export NUM_EPOCHS=3
 export SAVE_STEPS=750
 export SEED=1

-python3 run_ner_old.py \
+python3 run_ner.py \
 --task_type NER \
 --data_dir . \
 --labels ./labels.txt \
@@ -21,7 +21,7 @@ export NUM_EPOCHS=3
 export SAVE_STEPS=750
 export SEED=1

-python3 run_ner_old.py \
+python3 run_ner.py \
 --task_type Chunk \
 --data_dir . \
 --model_name_or_path $BERT_MODEL \
@@ -21,7 +21,7 @@ export NUM_EPOCHS=3
 export SAVE_STEPS=750
 export SEED=1

-python3 run_ner_old.py \
+python3 run_ner.py \
 --task_type POS \
 --data_dir . \
 --model_name_or_path $BERT_MODEL \
@@ -1,3 +1,19 @@
+<!---
+Copyright 2020 The HuggingFace Team. 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.
+-->
+
 ## Multiple Choice

 Based on the script [`run_multiple_choice.py`]().
@@ -0,0 +1,2 @@
+sentencepiece != 0.1.92
+protobuf
@@ -1,8 +1,29 @@
+<!---
+Copyright 2020 The HuggingFace Team. 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.
+-->

 ## SQuAD

-Based on the script [`run_squad.py`](https://github.com/huggingface/transformers/blob/master/examples/question-answering/run_squad.py).
+Based on the script [`run_qa.py`](https://github.com/huggingface/transformers/blob/master/examples/question-answering/run_qa.py).
+
+**Note:** This script only works with models that have a fast tokenizer (backed by the 🤗 Tokenizers library) as it
+uses special features of those tokenizers. You can check if your favorite model has a fast tokenizer in
+[this table](https://huggingface.co/transformers/index.html#bigtable), if it doesn't you can still use the old version
+of the script.
+
+The old version of this script can be found [here](https://github.com/huggingface/transformers/blob/master/examples/contrib/legacy/question-answering/run_squad.py).

 #### Fine-tuning BERT on SQuAD1.0

@@ -0,0 +1 @@
+datasets >= 1.1.3
@@ -76,9 +76,7 @@ def postprocess_qa_predictions(
    assert len(predictions) == 2, "`predictions` should be a tuple with two elements (start_logits, end_logits)."
    all_start_logits, all_end_logits = predictions

-    assert len(predictions[0]) == len(
-        features
-    ), f"Got {len(predictions[0])} predicitions and {len(features)} features."
+    assert len(predictions[0]) == len(features), f"Got {len(predictions[0])} predictions and {len(features)} features."

    # Build a map example to its corresponding features.
    example_id_to_index = {k: i for i, k in enumerate(examples["id"])}
@@ -118,7 +116,7 @@ def postprocess_qa_predictions(

            # Update minimum null prediction.
            feature_null_score = start_logits[0] + end_logits[0]
-            if min_null_prediction is None or min_null_prediction["score"] < feature_null_score:
+            if min_null_prediction is None or min_null_prediction["score"] > feature_null_score:
                min_null_prediction = {
                    "offsets": (0, 0),
                    "score": feature_null_score,
@@ -0,0 +1,28 @@
+<!---
+Copyright 2020 The HuggingFace Team. 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.
+-->
+
+# Research projects
+
+This folder contains various research projects using 🤗 Transformers. They are not maintained and require a specific
+version of 🤗 Transformers that is indicated in the requirements file of each folder. Updating them to the most recent version of the library will require some work.
+
+To use any of them, just run the command
+```
+pip install -r requirements.txt
+```
+inside the folder of your choice.
+
+If you need help with any of those, contact the author(s), indicated at the top of the `README` of each folder.
@@ -0,0 +1 @@
+transformers == 3.5.1
@@ -0,0 +1 @@
+transformers == 3.5.1
@@ -1,4 +1,4 @@
-transformers
+transformers == 3.5.1

 # For ROUGE
 nltk
@@ -0,0 +1 @@
+transformers == 3.5.1
--- a/Show More
+++ b/Show More