Patch release: v4.8.2

* Add option to save on each training node

* Apply suggestions from code review

Co-authored-by: Stas Bekman <stas00@users.noreply.github.com>

* Address review comments

Co-authored-by: Stas Bekman <stas00@users.noreply.github.com>

.circleci/config.yml

@@ -3,7 +3,6 @@ orbs:
     gcp-gke: circleci/gcp-gke@1.0.4
     go: circleci/go@1.3.0
 
-
 # TPU REFERENCES
 references:
     checkout_ml_testing: &checkout_ml_testing
@@ -69,6 +68,8 @@ jobs:
             - image: circleci/python:3.6
         environment:
             OMP_NUM_THREADS: 1
+            RUN_PT_TF_CROSS_TESTS: yes
+            TRANSFORMERS_IS_CI: yes
         resource_class: xlarge
         parallelism: 1
         steps:
@@ -79,13 +80,13 @@ jobs:
                       - v0.4-{{ checksum "setup.py" }}
             - run: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev
             - run: pip install --upgrade pip
-            - run: pip install .[sklearn,tf-cpu,torch,testing,sentencepiece,speech]
-            - run: pip install tapas torch-scatter -f https://pytorch-geometric.com/whl/torch-1.8.0+cpu.html
+            - run: pip install .[sklearn,tf-cpu,torch,testing,sentencepiece,speech,vision]
+            - run: pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-1.9.0+cpu.html
             - save_cache:
                 key: v0.4-{{ checksum "setup.py" }}
                 paths:
                     - '~/.cache/pip'
-            - run: RUN_PT_TF_CROSS_TESTS=1 python -m pytest -n 8 --dist=loadfile -rA -s --make-reports=tests_torch_and_tf ./tests/ -m is_pt_tf_cross_test --durations=0 | tee tests_output.txt
+            - run: python -m pytest -n 8 --dist=loadfile -rA -s --make-reports=tests_torch_and_tf ./tests/ -m is_pt_tf_cross_test --durations=0 | tee tests_output.txt
             - store_artifacts:
                   path: ~/transformers/tests_output.txt
             - store_artifacts:
@@ -97,6 +98,8 @@ jobs:
             - image: circleci/python:3.6
         environment:
             OMP_NUM_THREADS: 1
+            RUN_PT_FLAX_CROSS_TESTS: yes
+            TRANSFORMERS_IS_CI: yes
         resource_class: xlarge
         parallelism: 1
         steps:
@@ -107,13 +110,13 @@ jobs:
                       - v0.4-{{ checksum "setup.py" }}
             - run: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev
             - run: pip install --upgrade pip
-            - run: pip install .[sklearn,flax,torch,testing,sentencepiece,speech]
-            - run: pip install tapas torch-scatter -f https://pytorch-geometric.com/whl/torch-1.8.0+cpu.html
+            - run: pip install .[sklearn,flax,torch,testing,sentencepiece,speech,vision]
+            - run: pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-1.9.0+cpu.html
             - save_cache:
                 key: v0.4-{{ checksum "setup.py" }}
                 paths:
                     - '~/.cache/pip'
-            - run: RUN_PT_FLAX_CROSS_TESTS=1 python -m pytest -n 8 --dist=loadfile -rA -s --make-reports=tests_torch_and_flax ./tests/ -m is_pt_flax_cross_test --durations=0 | tee tests_output.txt
+            - run: python -m pytest -n 8 --dist=loadfile -rA -s --make-reports=tests_torch_and_flax ./tests/ -m is_pt_flax_cross_test --durations=0 | tee tests_output.txt
             - store_artifacts:
                   path: ~/transformers/tests_output.txt
             - store_artifacts:
@@ -125,6 +128,7 @@ jobs:
             - image: circleci/python:3.7
         environment:
             OMP_NUM_THREADS: 1
+            TRANSFORMERS_IS_CI: yes
         resource_class: xlarge
         parallelism: 1
         steps:
@@ -135,13 +139,13 @@ jobs:
                       - v0.4-{{ checksum "setup.py" }}
             - run: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev
             - run: pip install --upgrade pip
-            - run: pip install .[sklearn,torch,testing,sentencepiece,speech]
-            - run: pip install tapas torch-scatter -f https://pytorch-geometric.com/whl/torch-1.8.0+cpu.html
+            - run: pip install .[sklearn,torch,testing,sentencepiece,speech,vision,timm]
+            - run: pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-1.9.0+cpu.html
             - save_cache:
                   key: v0.4-torch-{{ checksum "setup.py" }}
                   paths:
                       - '~/.cache/pip'
-            - run: python -m pytest -n 8 --dist=loadfile -s --make-reports=tests_torch ./tests/ | tee tests_output.txt
+            - run: python -m pytest -n 3 --dist=loadfile -s --make-reports=tests_torch ./tests/ | tee tests_output.txt
             - store_artifacts:
                   path: ~/transformers/tests_output.txt
             - store_artifacts:
@@ -153,6 +157,7 @@ jobs:
             - image: circleci/python:3.7
         environment:
             OMP_NUM_THREADS: 1
+            TRANSFORMERS_IS_CI: yes
         resource_class: xlarge
         parallelism: 1
         steps:
@@ -179,6 +184,7 @@ jobs:
             - image: circleci/python:3.7
         environment:
             OMP_NUM_THREADS: 1
+            TRANSFORMERS_IS_CI: yes
         resource_class: xlarge
         parallelism: 1
         steps:
@@ -205,6 +211,8 @@ jobs:
             - image: circleci/python:3.7
         environment:
             OMP_NUM_THREADS: 1
+            RUN_PIPELINE_TESTS: yes
+            TRANSFORMERS_IS_CI: yes
         resource_class: xlarge
         parallelism: 1
         steps:
@@ -215,13 +223,13 @@ jobs:
                       - v0.4-{{ checksum "setup.py" }}
             - run: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev
             - run: pip install --upgrade pip
-            - run: pip install .[sklearn,torch,testing,sentencepiece,speech]
-            - run: pip install tapas torch-scatter -f https://pytorch-geometric.com/whl/torch-1.8.0+cpu.html
+            - run: pip install .[sklearn,torch,testing,sentencepiece,speech,vision]
+            - run: pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-1.9.0+cpu.html
             - save_cache:
                   key: v0.4-torch-{{ checksum "setup.py" }}
                   paths:
                       - '~/.cache/pip'
-            - run: RUN_PIPELINE_TESTS=1 python -m pytest -n 8 --dist=loadfile -rA -s --make-reports=tests_pipelines_torch -m is_pipeline_test ./tests/ | tee tests_output.txt
+            - run: python -m pytest -n 8 --dist=loadfile -rA -s --make-reports=tests_pipelines_torch -m is_pipeline_test ./tests/ | tee tests_output.txt
             - store_artifacts:
                   path: ~/transformers/tests_output.txt
             - store_artifacts:
@@ -233,6 +241,8 @@ jobs:
             - image: circleci/python:3.7
         environment:
             OMP_NUM_THREADS: 1
+            RUN_PIPELINE_TESTS: yes
+            TRANSFORMERS_IS_CI: yes
         resource_class: xlarge
         parallelism: 1
         steps:
@@ -247,7 +257,7 @@ jobs:
                   key: v0.4-tf-{{ checksum "setup.py" }}
                   paths:
                       - '~/.cache/pip'
-            - run: RUN_PIPELINE_TESTS=1 python -m pytest -n 8 --dist=loadfile -rA -s --make-reports=tests_pipelines_tf ./tests/ -m is_pipeline_test | tee tests_output.txt
+            - run: python -m pytest -n 8 --dist=loadfile -rA -s --make-reports=tests_pipelines_tf ./tests/ -m is_pipeline_test | tee tests_output.txt
             - store_artifacts:
                   path: ~/transformers/tests_output.txt
             - store_artifacts:
@@ -259,6 +269,7 @@ jobs:
             - image: circleci/python:3.7
         environment:
             RUN_CUSTOM_TOKENIZERS: yes
+            TRANSFORMERS_IS_CI: yes
         steps:
             - checkout
             - restore_cache:
@@ -266,7 +277,7 @@ jobs:
                       - v0.4-custom_tokenizers-{{ checksum "setup.py" }}
                       - v0.4-{{ checksum "setup.py" }}
             - run: pip install --upgrade pip
-            - run: pip install .[ja,testing,sentencepiece]
+            - run: pip install .[ja,testing,sentencepiece,jieba]
             - run: python -m unidic download
             - save_cache:
                   key: v0.4-custom_tokenizers-{{ checksum "setup.py" }}
@@ -284,6 +295,7 @@ jobs:
             - image: circleci/python:3.6
         environment:
             OMP_NUM_THREADS: 1
+            TRANSFORMERS_IS_CI: yes
         resource_class: xlarge
         parallelism: 1
         steps:
@@ -294,32 +306,44 @@ jobs:
                       - v0.4-{{ checksum "setup.py" }}
             - run: pip install --upgrade pip
             - run: pip install .[sklearn,torch,sentencepiece,testing]
-            - run: pip install -r examples/_tests_requirements.txt
+            - run: pip install -r examples/pytorch/_tests_requirements.txt
             - save_cache:
                   key: v0.4-torch_examples-{{ checksum "setup.py" }}
                   paths:
                       - '~/.cache/pip'
-            - run: python -m pytest -n 8 --dist=loadfile -s --make-reports=examples_torch ./examples/ | tee examples_output.txt
+            - run: TRANSFORMERS_IS_CI=1 python -m pytest -n 8 --dist=loadfile -s --make-reports=examples_torch ./examples/pytorch/ | tee examples_output.txt
             - store_artifacts:
                   path: ~/transformers/examples_output.txt
             - store_artifacts:
                   path: ~/transformers/reports
 
-    run_tests_git_lfs:
+    run_tests_hub:
         working_directory: ~/transformers
         docker:
             - image: circleci/python:3.7
+        environment:
+            HUGGINGFACE_CO_STAGING: yes
+            RUN_GIT_LFS_TESTS: yes
+            TRANSFORMERS_IS_CI: yes
         resource_class: xlarge
         parallelism: 1
         steps:
             - checkout
+            - restore_cache:
+                  keys:
+                      - v0.4-hub-{{ checksum "setup.py" }}
+                      - v0.4-{{ checksum "setup.py" }}
             - run: sudo apt-get install git-lfs
             - run: |
                 git config --global user.email "ci@dummy.com"
                 git config --global user.name "ci"
             - run: pip install --upgrade pip
-            - run: pip install .[testing]
-            - run: RUN_GIT_LFS_TESTS=1 python -m pytest -sv ./tests/test_hf_api.py -k "HfLargefilesTest"
+            - run: pip install .[torch,sentencepiece,testing]
+            - save_cache:
+                  key: v0.4-hub-{{ checksum "setup.py" }}
+                  paths:
+                      - '~/.cache/pip'
+            - run: python -m pytest -sv ./tests/ -m is_staging_test
 
     build_doc:
         working_directory: ~/transformers
@@ -333,7 +357,7 @@ jobs:
                       - v0.4-{{ checksum "setup.py" }}
             - run: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev
             - run: pip install --upgrade pip
-            - run: pip install ."[all, docs]"
+            - run: pip install ."[docs]"
             - save_cache:
                   key: v0.4-build_doc-{{ checksum "setup.py" }}
                   paths:
@@ -355,7 +379,9 @@ jobs:
                   keys:
                       - v0.4-deploy_doc-{{ checksum "setup.py" }}
                       - v0.4-{{ checksum "setup.py" }}
-            - run: pip install ."[all,docs]"
+            - run: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev
+            - run: pip install --upgrade pip
+            - run: pip install ."[docs]"
             - save_cache:
                   key: v0.4-deploy_doc-{{ checksum "setup.py" }}
                   paths:
@@ -367,6 +393,8 @@ jobs:
         docker:
             - image: circleci/python:3.6
         resource_class: medium
+        environment:
+            TRANSFORMERS_IS_CI: yes
         parallelism: 1
         steps:
             - checkout
@@ -383,12 +411,14 @@ jobs:
                       - '~/.cache/pip'
             - run: black --check examples tests src utils
             - run: isort --check-only examples tests src utils
+            - run: python utils/custom_init_isort.py --check_only
             - run: flake8 examples tests src utils
             - run: python utils/style_doc.py src/transformers docs/source --max_len 119 --check_only
             - run: python utils/check_copies.py
             - run: python utils/check_table.py
             - run: python utils/check_dummies.py
             - run: python utils/check_repo.py
+            - run: python utils/check_inits.py
 
     check_repository_consistency:
         working_directory: ~/transformers
@@ -407,6 +437,7 @@ jobs:
             - image: circleci/python:3.6
         environment:
             OMP_NUM_THREADS: 1
+            TRANSFORMERS_IS_CI: yes
         resource_class: xlarge
         parallelism: 1
         steps:
@@ -451,7 +482,7 @@ workflows:
             - run_tests_flax
             - run_tests_pipelines_torch
             - run_tests_pipelines_tf
-            - run_tests_git_lfs
+            - run_tests_hub
             - build_doc
             - deploy_doc: *workflow_filters
 #    tpu_testing_jobs:

.circleci/deploy.sh

@@ -57,4 +57,11 @@ deploy_doc "818878d" v3.5.1
 deploy_doc "c781171" v4.0.1
 deploy_doc "bfa4ccf" v4.1.1
 deploy_doc "7d9a9d0" v4.2.2
-deploy_doc "bae0c79"  # v4.3.3 Latest stable release
+deploy_doc "bae0c79" v4.3.3
+deploy_doc "c988db5" v4.4.0
+deploy_doc "c5d6a28" v4.4.1
+deploy_doc "6bc89ed" v4.4.2
+deploy_doc "4906a29" v4.5.0
+deploy_doc "4bae96e" v4.5.1
+deploy_doc "25dee4a" v4.6.0
+deploy_doc "7a6c9fa"  # v4.7.0 Latest stable release

.github/ISSUE_TEMPLATE/bug-report.md

@@ -34,7 +34,7 @@ Models:
 - funnel: @sgugger
 - gpt2: @patrickvonplaten, @LysandreJik
 - rag: @patrickvonplaten, @lhoestq
-- tensorflow: @jplu
+- tensorflow: @Rocketknight1
 
 Library:
 
@@ -48,9 +48,13 @@ Library:
 
 Documentation: @sgugger
 
+Model hub:
+
+- for issues with a model report at https://discuss.huggingface.co/ and tag the model's creator.
+
 HF projects:
 
-- nlp datasets: [different repo](https://github.com/huggingface/nlp)
+- datasets: [different repo](https://github.com/huggingface/datasets)
 - rust tokenizers: [different repo](https://github.com/huggingface/tokenizers)
 
 Examples:

.github/PULL_REQUEST_TEMPLATE.md

@@ -30,7 +30,7 @@ Fixes # (issue)
 ## Who can review?
 
 Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
-members/contributors which may be interested in your PR.
+members/contributors who may be interested in your PR.
 
 <!-- Your PR will be replied to more quickly if you can figure out the right person to tag with @
 
@@ -46,7 +46,7 @@ Models:
 - funnel: @sgugger
 - gpt2: @patrickvonplaten, @LysandreJik
 - rag: @patrickvonplaten, @lhoestq
-- tensorflow: @jplu
+- tensorflow: @LysandreJik
 
 Library:
 
@@ -62,7 +62,7 @@ Documentation: @sgugger
 
 HF projects:
 
-- nlp datasets: [different repo](https://github.com/huggingface/nlp)
+- datasets: [different repo](https://github.com/huggingface/datasets)
 - rust tokenizers: [different repo](https://github.com/huggingface/tokenizers)
 
 Examples:

.github/conda/meta.yaml

@@ -16,6 +16,8 @@ requirements:
     - pip
     - numpy >=1.17
     - dataclasses
+    - importlib_metadata
+    - huggingface_hub
     - packaging
     - filelock
     - requests
@@ -24,10 +26,13 @@ requirements:
     - regex !=2019.12.17
     - protobuf
     - tokenizers >=0.10.1,<0.11.0
+    - pyyaml
   run:
     - python
     - numpy >=1.17
     - dataclasses
+    - importlib_metadata
+    - huggingface_hub
     - packaging
     - filelock
     - requests
@@ -36,6 +41,7 @@ requirements:
     - regex !=2019.12.17
     - protobuf
     - tokenizers >=0.10.1,<0.11.0
+    - pyyaml
 
 test:
   imports:

.github/workflows/github-torch-hub.yml

@@ -37,10 +37,10 @@ jobs:
         # no longer needed
         pip uninstall -y transformers
 
-    - name: Torch hub list
-      run: |
-        python -c "import torch; print(torch.hub.list('huggingface/transformers:$BRANCH'))"
+    #- name: Torch hub list
+    #  run: |
+    #    python -c "import torch; print(torch.hub.list('huggingface/transformers:$BRANCH'))"
 
-    - name: Torch hub help
-      run: |
-        python -c "import torch; print(torch.hub.help('huggingface/transformers:$BRANCH', 'modelForSequenceClassification'))"
+    #- name: Torch hub help
+    #  run: |
+    #    python -c "import torch; print(torch.hub.help('huggingface/transformers:$BRANCH', 'modelForSequenceClassification'))"

.github/workflows/model-templates.yml

@@ -1,6 +1,9 @@
 name: Model templates runner
 
 on:
+  push:
+    branches:
+      - master
   pull_request:
     paths:
       - "src/**"
@@ -34,6 +37,7 @@ jobs:
       - name: Install dependencies
         run: |
           pip install --upgrade pip
+          sudo apt -y update && sudo apt install -y libsndfile1-dev
           pip install .[dev]
       - name: Create model files
         run: |
@@ -46,6 +50,7 @@ jobs:
           make style
           python utils/check_table.py --fix_and_overwrite
           python utils/check_dummies.py --fix_and_overwrite
+          python utils/check_copies.py --fix_and_overwrite
 
       - name: Run all non-slow tests
         run: |

.github/workflows/release-conda.yml

@@ -4,6 +4,8 @@ on:
   push:
     tags:
       - v*
+    branches:
+      - conda_*
 
 env:
   ANACONDA_API_TOKEN: ${{ secrets.ANACONDA_API_TOKEN }}
@@ -24,6 +26,7 @@ jobs:
         with:
           auto-update-conda: true
           auto-activate-base: false
+          python-version: 3.8
           activate-environment: "build-transformers"
           channels: huggingface
 

.github/workflows/self-push.yml

@@ -5,6 +5,7 @@ on:
     branches:
       - master
       - ci_*
+      - ci-*
     paths:
       - "src/**"
       - "tests/**"
@@ -12,11 +13,17 @@ on:
       - "templates/**"
   repository_dispatch:
 
+env:
+  HF_HOME: /mnt/cache
+  TRANSFORMERS_IS_CI: yes
+  OMP_NUM_THREADS: 8
+  MKL_NUM_THREADS: 8
+
 jobs:
   run_tests_torch_gpu:
     runs-on: [self-hosted, docker-gpu, single-gpu]
     container:
-      image: pytorch/pytorch:1.8.0-cuda11.1-cudnn8-runtime
+      image: pytorch/pytorch:1.9.0-cuda11.1-cudnn8-runtime
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
     steps:
       - name: Launcher docker
@@ -30,7 +37,7 @@ jobs:
         run: |
           apt -y update && apt install -y libsndfile1-dev
           pip install --upgrade pip
-          pip install .[sklearn,testing,onnxruntime,sentencepiece,speech]
+          pip install .[sklearn,testing,onnxruntime,sentencepiece,speech,vision,timm]
 
       - name: Are GPUs recognized by our DL frameworks
         run: |
@@ -40,10 +47,6 @@ jobs:
           python -c "import torch; print('Number of GPUs available:', torch.cuda.device_count())"
 
       - name: Run all non-slow tests on GPU
-        env:
-          OMP_NUM_THREADS: 8
-          MKL_NUM_THREADS: 8
-          HF_HOME: /mnt/cache
         run: |
           python -m pytest -n 2 --dist=loadfile --make-reports=tests_torch_gpu tests
 
@@ -60,6 +63,7 @@ jobs:
 
   run_tests_tf_gpu:
     runs-on: [self-hosted, docker-gpu, single-gpu]
+    timeout-minutes: 120
     container:
       image: tensorflow/tensorflow:2.4.1-gpu
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -83,13 +87,10 @@ jobs:
 
       - name: Run all non-slow tests on GPU
         env:
-          OMP_NUM_THREADS: 8
-          MKL_NUM_THREADS: 8
           TF_NUM_INTRAOP_THREADS: 8
           TF_NUM_INTEROP_THREADS: 1
-          HF_HOME: /mnt/cache
         run: |
-          python -m pytest -n 2 --dist=loadfile --make-reports=tests_tf_gpu tests
+          python -m pytest -n 1 --dist=loadfile --make-reports=tests_tf_gpu tests
 
       - name: Failure short reports
         if: ${{ always() }}
@@ -106,7 +107,7 @@ jobs:
   run_tests_torch_multi_gpu:
     runs-on: [self-hosted, docker-gpu, multi-gpu]
     container:
-      image: pytorch/pytorch:1.8.0-cuda11.1-cudnn8-runtime
+      image: pytorch/pytorch:1.9.0-cuda11.1-cudnn8-runtime
       options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
     steps:
       - name: Launcher docker
@@ -120,7 +121,7 @@ jobs:
         run: |
           apt -y update && apt install -y libsndfile1-dev
           pip install --upgrade pip
-          pip install .[sklearn,testing,onnxruntime,sentencepiece,speech]
+          pip install .[sklearn,testing,onnxruntime,sentencepiece,speech,vision,timm]
 
       - name: Are GPUs recognized by our DL frameworks
         run: |
@@ -131,10 +132,7 @@ jobs:
 
       - name: Run all non-slow tests on GPU
         env:
-          OMP_NUM_THREADS: 8
-          MKL_NUM_THREADS: 8
           MKL_SERVICE_FORCE_INTEL: 1
-          HF_HOME: /mnt/cache
         run: |
           python -m pytest -n 2 --dist=loadfile --make-reports=tests_torch_multi_gpu tests
 
@@ -151,6 +149,7 @@ jobs:
 
   run_tests_tf_multi_gpu:
     runs-on: [self-hosted, docker-gpu, multi-gpu]
+    timeout-minutes: 120
     container:
       image: tensorflow/tensorflow:2.4.1-gpu
       options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -174,13 +173,10 @@ jobs:
 
       - name: Run all non-slow tests on GPU
         env:
-          OMP_NUM_THREADS: 8
-          MKL_NUM_THREADS: 8
           TF_NUM_INTRAOP_THREADS: 8
           TF_NUM_INTEROP_THREADS: 1
-          HF_HOME: /mnt/cache
         run: |
-          python -m pytest -n 2 --dist=loadfile --make-reports=tests_tf_multi_gpu tests
+          python -m pytest -n 1 --dist=loadfile --make-reports=tests_tf_multi_gpu tests
 
       - name: Failure short reports
         if: ${{ always() }}
@@ -193,11 +189,101 @@ jobs:
           name: run_all_tests_tf_multi_gpu_test_reports
           path: reports
 
+  run_tests_torch_cuda_extensions_gpu:
+    runs-on: [self-hosted, docker-gpu, single-gpu]
+    container:
+      image: nvcr.io/nvidia/pytorch:21.03-py3
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: Launcher docker
+        uses: actions/checkout@v2
+
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+      - name: Install dependencies
+        run: |
+          apt -y update && apt install -y libaio-dev
+          pip install --upgrade pip
+          pip install .[testing,deepspeed]
+
+      - name: Are GPUs recognized by our DL frameworks
+        run: |
+          python -c "import torch; print('Cuda available:', torch.cuda.is_available())"
+          python -c "import torch; print('Cuda version:', torch.version.cuda)"
+          python -c "import torch; print('CuDNN version:', torch.backends.cudnn.version())"
+          python -c "import torch; print('Number of GPUs available:', torch.cuda.device_count())"
+
+      - name: Run all tests on GPU
+        run: |
+          python -m pytest -n 1 --dist=loadfile --make-reports=tests_torch_cuda_extensions_gpu tests/deepspeed tests/extended
+
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_torch_cuda_extensions_gpu_failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: run_tests_torch_cuda_extensions_gpu_test_reports
+          path: reports
+
+  run_tests_torch_cuda_extensions_multi_gpu:
+    runs-on: [self-hosted, docker-gpu, multi-gpu]
+    container:
+      image: nvcr.io/nvidia/pytorch:21.03-py3
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: Launcher docker
+        uses: actions/checkout@v2
+
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+      - name: Install dependencies
+        run: |
+          apt -y update && apt install -y libaio-dev
+          pip install --upgrade pip
+          pip install .[testing,deepspeed,fairscale]
+
+      - name: Are GPUs recognized by our DL frameworks
+        run: |
+          python -c "import torch; print('Cuda available:', torch.cuda.is_available())"
+          python -c "import torch; print('Cuda version:', torch.version.cuda)"
+          python -c "import torch; print('CuDNN version:', torch.backends.cudnn.version())"
+          python -c "import torch; print('Number of GPUs available:', torch.cuda.device_count())"
+
+      - name: Run all tests on GPU
+        run: |
+          python -m pytest -n 1 --dist=loadfile --make-reports=tests_torch_cuda_extensions_multi_gpu tests/deepspeed tests/extended
+
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_torch_cuda_extensions_multi_gpu_failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: run_tests_torch_cuda_extensions_multi_gpu_test_reports
+          path: reports
+
+
   send_results:
     name: Send results to webhook
     runs-on: ubuntu-latest
     if: always()
-    needs: [run_tests_torch_gpu, run_tests_tf_gpu, run_tests_torch_multi_gpu, run_tests_tf_multi_gpu]
+    needs: [
+        run_tests_torch_gpu,
+        run_tests_tf_gpu,
+        run_tests_torch_multi_gpu,
+        run_tests_tf_multi_gpu,
+        run_tests_torch_cuda_extensions_gpu,
+        run_tests_torch_cuda_extensions_multi_gpu
+    ]
     steps:
       - uses: actions/checkout@v2
 
@@ -210,4 +296,4 @@ jobs:
 
         run: |
           pip install slack_sdk
-          python utils/notification_service.py push
+          python utils/notification_service.py push

.github/workflows/self-scheduled.yml

@@ -8,11 +8,18 @@ on:
   schedule:
     - cron: "0 0 * * *"
 
+env:
+  HF_HOME: /mnt/cache
+  TRANSFORMERS_IS_CI: yes
+  RUN_SLOW: yes
+  OMP_NUM_THREADS: 16
+  MKL_NUM_THREADS: 16
+
 jobs:
   run_all_tests_torch_gpu:
     runs-on: [self-hosted, docker-gpu, single-gpu]
     container:
-      image: pytorch/pytorch:1.8.0-cuda11.1-cudnn8-runtime
+      image: pytorch/pytorch:1.9.0-cuda11.1-cudnn8-runtime
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
     steps:
       - name: Launcher docker
@@ -26,7 +33,7 @@ jobs:
         run: |
           apt -y update && apt install -y libsndfile1-dev
           pip install --upgrade pip
-          pip install .[sklearn,testing,onnxruntime,sentencepiece,speech]
+          pip install .[integrations,sklearn,testing,onnxruntime,sentencepiece,speech,vision,timm]
 
       - name: Are GPUs recognized by our DL frameworks
         run: |
@@ -36,11 +43,6 @@ jobs:
           python -c "import torch; print('Number of GPUs available:', torch.cuda.device_count())"
 
       - name: Run all tests on GPU
-        env:
-          OMP_NUM_THREADS: 16
-          MKL_NUM_THREADS: 16
-          RUN_SLOW: yes
-          HF_HOME: /mnt/cache
         run: |
           python -m pytest -n 1 --dist=loadfile --make-reports=tests_torch_gpu tests
 
@@ -55,8 +57,9 @@ jobs:
           MKL_NUM_THREADS: 16
           RUN_SLOW: yes
           HF_HOME: /mnt/cache
+          TRANSFORMERS_IS_CI: yes
         run: |
-          pip install -r examples/_tests_requirements.txt
+          pip install -r examples/pytorch/_tests_requirements.txt
           python -m pytest -n 1 --dist=loadfile --make-reports=examples_torch_gpu examples
 
       - name: Failure short reports
@@ -66,11 +69,7 @@ jobs:
       - name: Run all pipeline tests on GPU
         if: ${{ always() }}
         env:
-          OMP_NUM_THREADS: 16
-          MKL_NUM_THREADS: 16
-          RUN_SLOW: yes
           RUN_PIPELINE_TESTS: yes
-          HF_HOME: /mnt/cache
         run: |
           python -m pytest -n 1 --dist=loadfile -m is_pipeline_test --make-reports=tests_torch_pipeline_gpu tests
 
@@ -110,12 +109,8 @@ jobs:
 
       - name: Run all tests on GPU
         env:
-          RUN_SLOW: yes
-          HF_HOME: /mnt/cache
-          OMP_NUM_THREADS: 16
           TF_NUM_INTEROP_THREADS: 1
           TF_NUM_INTRAOP_THREADS: 16
-          MKL_NUM_THREADS: 16
         run: |
           python -m pytest -n 1 --dist=loadfile --make-reports=tests_tf_gpu tests
 
@@ -126,13 +121,9 @@ jobs:
       - name: Run all pipeline tests on GPU
         if: ${{ always() }}
         env:
-          RUN_SLOW: yes
-          HF_HOME: /mnt/cache
-          OMP_NUM_THREADS: 16
           RUN_PIPELINE_TESTS: yes
           TF_NUM_INTEROP_THREADS: 1
           TF_NUM_INTRAOP_THREADS: 16
-          MKL_NUM_THREADS: 16
         run: |
           python -m pytest -n 1 --dist=loadfile -m is_pipeline_test --make-reports=tests_tf_pipeline_gpu tests
 
@@ -150,7 +141,7 @@ jobs:
   run_all_tests_torch_multi_gpu:
     runs-on: [self-hosted, docker-gpu, multi-gpu]
     container:
-      image: pytorch/pytorch:1.8.0-cuda11.1-cudnn8-runtime
+      image: pytorch/pytorch:1.9.0-cuda11.1-cudnn8-runtime
       options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
     steps:
       - name: Launcher docker
@@ -164,7 +155,7 @@ jobs:
         run: |
           apt -y update && apt install -y libsndfile1-dev
           pip install --upgrade pip
-          pip install .[sklearn,testing,onnxruntime,sentencepiece,speech]
+          pip install .[integrations,sklearn,testing,onnxruntime,sentencepiece,speech,vision,timm]
 
       - name: Are GPUs recognized by our DL frameworks
         run: |
@@ -175,10 +166,6 @@ jobs:
 
       - name: Run all tests on GPU
         env:
-          RUN_SLOW: yes
-          HF_HOME: /mnt/cache
-          OMP_NUM_THREADS: 16
-          MKL_NUM_THREADS: 16
           MKL_SERVICE_FORCE_INTEL: 1
         run: |
           python -m pytest -n 1 --dist=loadfile --make-reports=tests_torch_multi_gpu tests
@@ -190,11 +177,7 @@ jobs:
       - name: Run all pipeline tests on GPU
         if: ${{ always() }}
         env:
-          OMP_NUM_THREADS: 16
-          MKL_NUM_THREADS: 16
-          RUN_SLOW: yes
           RUN_PIPELINE_TESTS: yes
-          HF_HOME: /mnt/cache
         run: |
           python -m pytest -n 1 --dist=loadfile -m is_pipeline_test --make-reports=tests_torch_pipeline_multi_gpu tests
 
@@ -234,12 +217,8 @@ jobs:
 
       - name: Run all tests on GPU
         env:
-          OMP_NUM_THREADS: 16
-          RUN_SLOW: yes
-          MKL_NUM_THREADS: 16
           TF_NUM_INTEROP_THREADS: 1
           TF_NUM_INTRAOP_THREADS: 16
-          HF_HOME: /mnt/cache
         run: |
           python -m pytest -n 1 --dist=loadfile --make-reports=tests_tf_multi_gpu tests
 
@@ -250,13 +229,9 @@ jobs:
       - name: Run all pipeline tests on GPU
         if: ${{ always() }}
         env:
-          OMP_NUM_THREADS: 16
-          RUN_SLOW: yes
           RUN_PIPELINE_TESTS: yes
-          MKL_NUM_THREADS: 16
           TF_NUM_INTEROP_THREADS: 1
           TF_NUM_INTRAOP_THREADS: 16
-          HF_HOME: /mnt/cache
         run: |
           python -m pytest -n 1 --dist=loadfile -m is_pipeline_test --make-reports=tests_tf_pipeline_multi_gpu tests
 
@@ -271,11 +246,100 @@ jobs:
           name: run_all_tests_tf_multi_gpu_test_reports
           path: reports
 
+  run_all_tests_torch_cuda_extensions_gpu:
+    runs-on: [self-hosted, docker-gpu, single-gpu]
+    container:
+      image: nvcr.io/nvidia/pytorch:21.03-py3
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: Launcher docker
+        uses: actions/checkout@v2
+
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+      - name: Install dependencies
+        run: |
+          apt -y update && apt install -y libaio-dev
+          pip install --upgrade pip
+          pip install .[testing,deepspeed]
+
+      - name: Are GPUs recognized by our DL frameworks
+        run: |
+          python -c "import torch; print('Cuda available:', torch.cuda.is_available())"
+          python -c "import torch; print('Cuda version:', torch.version.cuda)"
+          python -c "import torch; print('CuDNN version:', torch.backends.cudnn.version())"
+          python -c "import torch; print('Number of GPUs available:', torch.cuda.device_count())"
+
+      - name: Run all tests on GPU
+        run: |
+          python -m pytest -n 1 --dist=loadfile --make-reports=tests_torch_cuda_extensions_gpu tests/deepspeed tests/extended
+
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_torch_cuda_extensions_gpu_failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: run_tests_torch_cuda_extensions_gpu_test_reports
+          path: reports
+
+  run_all_tests_torch_cuda_extensions_multi_gpu:
+    runs-on: [self-hosted, docker-gpu, multi-gpu]
+    container:
+      image: nvcr.io/nvidia/pytorch:21.03-py3
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: Launcher docker
+        uses: actions/checkout@v2
+
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+      - name: Install dependencies
+        run: |
+          apt -y update && apt install -y libaio-dev
+          pip install --upgrade pip
+          pip install .[testing,deepspeed,fairscale]
+
+      - name: Are GPUs recognized by our DL frameworks
+        run: |
+          python -c "import torch; print('Cuda available:', torch.cuda.is_available())"
+          python -c "import torch; print('Cuda version:', torch.version.cuda)"
+          python -c "import torch; print('CuDNN version:', torch.backends.cudnn.version())"
+          python -c "import torch; print('Number of GPUs available:', torch.cuda.device_count())"
+
+      - name: Run all tests on GPU
+        run: |
+          python -m pytest -n 1 --dist=loadfile --make-reports=tests_torch_cuda_extensions_multi_gpu tests/deepspeed tests/extended
+
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_torch_cuda_extensions_multi_gpu_failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: run_tests_torch_cuda_extensions_multi_gpu_test_reports
+          path: reports
+
   send_results:
     name: Send results to webhook
     runs-on: ubuntu-latest
     if: always()
-    needs: [run_all_tests_torch_gpu, run_all_tests_tf_gpu, run_all_tests_torch_multi_gpu, run_all_tests_tf_multi_gpu]
+    needs: [
+        run_all_tests_torch_gpu,
+        run_all_tests_tf_gpu,
+        run_all_tests_torch_multi_gpu,
+        run_all_tests_tf_multi_gpu,
+        run_all_tests_torch_cuda_extensions_gpu,
+        run_all_tests_torch_cuda_extensions_multi_gpu
+    ]
     steps:
       - uses: actions/checkout@v2
 

.github/workflows/stale.yml

@@ -2,7 +2,7 @@ name: Stale Bot
 
 on:
   schedule:
-    - cron: "0 0 * * *"
+    - cron: "0 15 * * *"
 
 jobs:
   close_stale_issues:

.gitignore

@@ -9,8 +9,7 @@ __pycache__/
 *.so
 
 # tests and logs
-tests/fixtures/*
-!tests/fixtures/sample_text_no_unicode.txt
+tests/fixtures/cached_*_text.txt
 logs/
 lightning_logs/
 lang_code_data/

CONTRIBUTING.md

@@ -36,6 +36,13 @@ There are 4 ways you can contribute to transformers:
 * Contributing to the examples or to the documentation;
 * Submitting issues related to bugs or desired new features.
 
+In particular there is a special [Good First
+Issue](https://github.com/huggingface/transformers/contribute) listing. It will give you a list of
+open Issues that are open to anybody to work on. Just comment in the issue that you'd like to work
+on it. In that same listing you will also find some Issues with `Good Second Issue` label. These are
+typically slightly more complicated than the Issues with just `Good First Issue` label. But if you
+feel you know what you're doing, go for it.
+
 *All are equally valuable to the community.*
 
 ## Submitting a new issue or feature request
@@ -46,7 +53,7 @@ feedback.
 
 ### Did you find a bug?
 
-The transformers are robust and reliable thanks to the users who notify us of
+The 🤗 Transformers library is robust and reliable thanks to the users who notify us of
 the problems they encounter. So thank you for reporting an issue.
 
 First, we would really appreciate it if you could **make sure the bug was not
@@ -285,7 +292,7 @@ $ python -m pytest -n auto --dist=loadfile -s -v ./tests/
 and for the examples:
 
 ```bash
-$ pip install -r examples/requirements.txt  # only needed the first time
+$ pip install -r examples/xxx/requirements.txt  # only needed the first time
 $ python -m pytest -n auto --dist=loadfile -s -v ./examples/
 ```
 In fact, that's how `make test` and `make test-examples` are implemented (sans the `pip install` line)!
@@ -343,7 +350,7 @@ You can now use `make` from any terminal (Powershell, cmd.exe, etc) 🎉
 
 ### Syncing forked master with upstream (HuggingFace) master
 
-To avoid pinging the upstream repository which adds reference notes to each upstream PR and sends unnessary notifications to the developers involved in these PRs, 
+To avoid pinging the upstream repository which adds reference notes to each upstream PR and sends unnessary notifications to the developers involved in these PRs,
 when syncing the master branch of a forked repository, please, follow these steps:
 1. When possible, avoid syncing with the upstream using a branch and PR on the forked repository. Instead merge directly into the forked master.
 2. If a PR is absolutely necessary, use the following steps after checking out your branch:

Makefile

@@ -1,5 +1,7 @@
 .PHONY: deps_table_update modified_only_fixup extra_quality_checks quality style fixup fix-copies test test-examples docs
 
+# make sure to test the local checkout in scripts and not the pre-installed one (don't use quotes!)
+export PYTHONPATH = src
 
 check_dirs := examples tests src utils
 
@@ -19,34 +21,44 @@ modified_only_fixup:
 deps_table_update:
 	@python setup.py deps_table_update
 
+# autogenerating code
+
+autogenerate_code: deps_table_update
+	python utils/class_mapping_update.py
+
 # Check that source code meets quality standards
 
-extra_quality_checks: deps_table_update
+extra_quality_checks:
 	python utils/check_copies.py
 	python utils/check_table.py
 	python utils/check_dummies.py
 	python utils/check_repo.py
-	python utils/style_doc.py src/transformers docs/source --max_len 119
-	python utils/class_mapping_update.py
+	python utils/check_inits.py
 
 # this target runs checks on all files
 quality:
 	black --check $(check_dirs)
 	isort --check-only $(check_dirs)
+	python utils/custom_init_isort.py --check_only
 	flake8 $(check_dirs)
-	python utils/style_doc.py src/transformers docs/source --max_len 119 --check_only
 	${MAKE} extra_quality_checks
 
 # Format source code automatically and check is there are any problems left that need manual fixing
 
-style: deps_table_update
+extra_style_checks:
+	python utils/custom_init_isort.py
+	python utils/style_doc.py src/transformers docs/source --max_len 119
+
+# this target runs checks on all files and potentially modifies some of them
+style:
 	black $(check_dirs)
 	isort $(check_dirs)
-	python utils/style_doc.py src/transformers docs/source --max_len 119
+	${MAKE} autogenerate_code
+	${MAKE} extra_style_checks
 
 # Super fast fix and check target that only works on relevant modified files since the branch was made
 
-fixup: modified_only_fixup extra_quality_checks
+fixup: modified_only_fixup extra_style_checks autogenerate_code extra_quality_checks
 
 # Make marked copies of snippets of codes conform to the original
 
@@ -63,7 +75,13 @@ test:
 # Run tests for examples
 
 test-examples:
-	python -m pytest -n auto --dist=loadfile -s -v ./examples/
+	python -m pytest -n auto --dist=loadfile -s -v ./examples/pytorch/
+
+# Run tests for SageMaker DLC release
+
+test-sagemaker: # install sagemaker dependencies in advance with pip install .[sagemaker]
+	TEST_SAGEMAKER=True python -m pytest -n auto  -s -v ./tests/sagemaker
+
 
 # Check that docs can build
 
@@ -83,4 +101,3 @@ post-release:
 
 post-patch:
 	python utils/release.py --post_release --patch
-

README.md

@@ -35,21 +35,26 @@ limitations under the License.
     <a href="https://github.com/huggingface/transformers/blob/master/CODE_OF_CONDUCT.md">
         <img alt="Contributor Covenant" src="https://img.shields.io/badge/Contributor%20Covenant-v2.0%20adopted-ff69b4.svg">
     </a>
+    <a href="https://zenodo.org/badge/latestdoi/155220641"><img src="https://zenodo.org/badge/155220641.svg" alt="DOI"></a>
 </p>
 
 <h3 align="center">
-<p>State-of-the-art Natural Language Processing for PyTorch and TensorFlow 2.0
+    <p>State-of-the-art Natural Language Processing for Jax, PyTorch and TensorFlow</p>
 </h3>
 
-🤗 Transformers provides thousands of pretrained models to perform tasks on texts such as classification, information extraction, question answering, summarization, translation, text generation, etc in 100+ languages. Its aim is to make cutting-edge NLP easier to use for everyone.
+<h3 align="center">
+    <a href="https://hf.co/course"><img src="https://raw.githubusercontent.com/huggingface/transformers/master/docs/source/imgs/course_banner.png"></a>
+</h3>
 
-🤗 Transformers provides APIs to quickly download and use those pretrained models on a given text, fine-tune them on your own datasets then share them with the community on our [model hub](https://huggingface.co/models). At the same time, each python module defining an architecture can be used as a standalone and modified to enable quick research experiments.
+🤗 Transformers provides thousands of pretrained models to perform tasks on texts such as classification, information extraction, question answering, summarization, translation, text generation and more in over 100 languages. Its aim is to make cutting-edge NLP easier to use for everyone.
 
-🤗 Transformers is backed by the two most popular deep learning libraries, [PyTorch](https://pytorch.org/) and [TensorFlow](https://www.tensorflow.org/), with a seamless integration between them, allowing you to train your models with one then load it for inference with the other.
+🤗 Transformers provides APIs to quickly download and use those pretrained models on a given text, fine-tune them on your own datasets and then share them with the community on our [model hub](https://huggingface.co/models). At the same time, each python module defining an architecture is fully standalone and can be modified to enable quick research experiments.
+
+🤗 Transformers is backed by the three most popular deep learning libraries — [Jax](https://jax.readthedocs.io/en/latest/), [PyTorch](https://pytorch.org/) and [TensorFlow](https://www.tensorflow.org/) — with a seamless integration between them. It's straightforward to train your models with one before loading them for inference with the other.
 
 ## Online demos
 
-You can test most of our models directly on their pages from the [model hub](https://huggingface.co/models). We also offer [private model hosting, versioning, & an inference API](https://huggingface.co/pricing) to use those models.
+You can test most of our models directly on their pages from the [model hub](https://huggingface.co/models). We also offer [private model hosting, versioning, & an inference API](https://huggingface.co/pricing) for public and private models.
 
 Here are a few examples:
 - [Masked word completion with BERT](https://huggingface.co/bert-base-uncased?text=Paris+is+the+%5BMASK%5D+of+France)
@@ -62,22 +67,28 @@ Here are a few examples:
 
 **[Write With Transformer](https://transformer.huggingface.co)**, built by the Hugging Face team, is the official demo of this repo’s text generation capabilities.
 
+## If you are looking for custom support from the Hugging Face team
+
+<a target="_blank" href="https://huggingface.co/support">
+    <img alt="HuggingFace Expert Acceleration Program" src="https://huggingface.co/front/thumbnails/support.png" style="max-width: 600px; border: 1px solid #eee; border-radius: 4px; box-shadow: 0 1px 2px 0 rgba(0, 0, 0, 0.05);">
+</a><br>
+
 ## Quick tour
 
-To immediately use a model on a given text, we provide the `pipeline` API. Pipelines group together a pretrained model with the preprocessing that was used during that model training. Here is how to quickly use a pipeline to classify positive versus negative texts
+To immediately use a model on a given text, we provide the `pipeline` API. Pipelines group together a pretrained model with the preprocessing that was used during that model's training. Here is how to quickly use a pipeline to classify positive versus negative texts:
 
 ```python
 >>> from transformers import pipeline
 
 # Allocate a pipeline for sentiment-analysis
 >>> classifier = pipeline('sentiment-analysis')
->>> classifier('We are very happy to include pipeline into the transformers repository.')
-[{'label': 'POSITIVE', 'score': 0.9978193640708923}]
+>>> classifier('We are very happy to introduce pipeline to the transformers repository.')
+[{'label': 'POSITIVE', 'score': 0.9996980428695679}]
 ```
 
-The second line of code downloads and caches the pretrained model used by the pipeline, the third line evaluates it on the given text. Here the answer is "positive" with a confidence of 99.8%.
+The second line of code downloads and caches the pretrained model used by the pipeline, while the third evaluates it on the given text. Here the answer is "positive" with a confidence of 99.97%.
 
-This is another example of pipeline used for that can extract question answers from some context:
+Many NLP tasks have a pre-trained `pipeline` ready to go. For example, we can easily extract question answers given context:
 
 ``` python
 >>> from transformers import pipeline
@@ -86,15 +97,15 @@ This is another example of pipeline used for that can extract question answers f
 >>> question_answerer = pipeline('question-answering')
 >>> question_answerer({
 ...     'question': 'What is the name of the repository ?',
-...     'context': 'Pipeline have been included in the huggingface/transformers repository'
+...     'context': 'Pipeline has been included in the huggingface/transformers repository'
 ... })
-{'score': 0.5135612454720828, 'start': 35, 'end': 59, 'answer': 'huggingface/transformers'}
+{'score': 0.30970096588134766, 'start': 34, 'end': 58, 'answer': 'huggingface/transformers'}
 
 ```
 
-On top of the answer, the pretrained model used here returned its confidence score, along with the start position and its end position in the tokenized sentence. You can learn more about the tasks supported by the `pipeline` API in [this tutorial](https://huggingface.co/transformers/task_summary.html).
+In addition to the answer, the pretrained model used here returned its confidence score, along with the start position and end position of the answer in the tokenized sentence. You can learn more about the tasks supported by the `pipeline` API in [this tutorial](https://huggingface.co/transformers/task_summary.html).
 
-To download and use any of the pretrained models on your given task, you just need to use those three lines of codes (PyTorch version):
+To download and use any of the pretrained models on your given task, all it takes is three lines of code. Here is the PyTorch version:
 ```python
 >>> from transformers import AutoTokenizer, AutoModel
 
@@ -104,7 +115,7 @@ To download and use any of the pretrained models on your given task, you just ne
 >>> inputs = tokenizer("Hello world!", return_tensors="pt")
 >>> outputs = model(**inputs)
 ```
-or for TensorFlow:
+And here is the equivalent code for TensorFlow:
 ```python
 >>> from transformers import AutoTokenizer, TFAutoModel
 
@@ -115,9 +126,9 @@ or for TensorFlow:
 >>> outputs = model(**inputs)
 ```
 
-The tokenizer is responsible for all the preprocessing the pretrained model expects, and can be called directly on one (or list) of texts (as we can see on the fourth line of both code examples). It will output a dictionary you can directly pass to your model (which is done on the fifth line).
+The tokenizer is responsible for all the preprocessing the pretrained model expects, and can be called directly on a single string (as in the above examples) or a list. It will output a dictionary that you can use in downstream code or simply directly pass to your model using the ** argument unpacking operator.
 
-The model itself is a regular [Pytorch `nn.Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) or a [TensorFlow `tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model) (depending on your backend) which you can use normally. For instance, [this tutorial](https://huggingface.co/transformers/training.html) explains how to integrate such a model in classic PyTorch or TensorFlow training loop, or how to use our `Trainer` API to quickly fine-tune the on a new dataset.
+The model itself is a regular [Pytorch `nn.Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) or a [TensorFlow `tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model) (depending on your backend) which you can use normally. [This tutorial](https://huggingface.co/transformers/training.html) explains how to integrate such a model into a classic PyTorch or TensorFlow training loop, or how to use our `Trainer` API to quickly fine-tune on a new dataset.
 
 ## Why should I use transformers?
 
@@ -135,16 +146,16 @@ The model itself is a regular [Pytorch `nn.Module`](https://pytorch.org/docs/sta
 1. Choose the right framework for every part of a model's lifetime:
     - Train state-of-the-art models in 3 lines of code.
     - Move a single model between TF2.0/PyTorch frameworks at will.
-    - Seamlessly pick the right framework for training, evaluation, production.
+    - Seamlessly pick the right framework for training, evaluation and production.
 
 1. Easily customize a model or an example to your needs:
-    - Examples for each architecture to reproduce the results by the official authors of said architecture.
-    - Expose the models internal as consistently as possible.
+    - We provide examples for each architecture to reproduce the results published by its original authors.
+    - Model internals are exposed as consistently as possible.
     - Model files can be used independently of the library for quick experiments.
 
 ## Why shouldn't I use transformers?
 
-- This library is not a modular toolbox of building blocks for neural nets. The code in the model files is not refactored with additional abstractions on purpose, so that researchers can quickly iterate on each of the models without diving in additional abstractions/files.
+- This library is not a modular toolbox of building blocks for neural nets. The code in the model files is not refactored with additional abstractions on purpose, so that researchers can quickly iterate on each of the models without diving into additional abstractions/files.
 - The training API is not intended to work on any model but is optimized to work with the models provided by the library. For generic machine learning loops, you should use another library.
 - While we strive to present as many use cases as possible, the scripts in our [examples folder](https://github.com/huggingface/transformers/tree/master/examples) are just that: examples. It is expected that they won't work out-of-the box on your specific problem and that you will be required to change a few lines of code to adapt them to your needs.
 
@@ -152,16 +163,16 @@ The model itself is a regular [Pytorch `nn.Module`](https://pytorch.org/docs/sta
 
 ### With pip
 
-This repository is tested on Python 3.6+, PyTorch 1.0.0+ (PyTorch 1.3.1+ for [examples](https://github.com/huggingface/transformers/tree/master/examples)) and TensorFlow 2.0.
+This repository is tested on Python 3.6+, Flax 0.3.2+, PyTorch 1.3.1+ and TensorFlow 2.3+.
 
 You should install 🤗 Transformers in a [virtual environment](https://docs.python.org/3/library/venv.html). If you're unfamiliar with Python virtual environments, check out the [user guide](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/).
 
 First, create a virtual environment with the version of Python you're going to use and activate it.
 
-Then, you will need to install at least one of TensorFlow 2.0, PyTorch or Flax.
-Please refer to [TensorFlow installation page](https://www.tensorflow.org/install/pip#tensorflow-2.0-rc-is-available), [PyTorch installation page](https://pytorch.org/get-started/locally/#start-locally) regarding the specific install command for your platform and/or [Flax installation page](https://github.com/google/flax#quick-install).
+Then, you will need to install at least one of Flax, PyTorch or TensorFlow.
+Please refer to [TensorFlow installation page](https://www.tensorflow.org/install/), [PyTorch installation page](https://pytorch.org/get-started/locally/#start-locally) and/or [Flax installation page](https://github.com/google/flax#quick-install) regarding the specific install command for your platform.
 
-When TensorFlow 2.0 and/or PyTorch has been installed, 🤗 Transformers can be installed using pip as follows:
+When one of those backends has been installed, 🤗 Transformers can be installed using pip as follows:
 
 ```bash
 pip install transformers
@@ -179,9 +190,9 @@ Since Transformers version v4.0.0, we now have a conda channel: `huggingface`.
 conda install -c huggingface transformers
 ```
 
-Follow the installation pages of TensorFlow, PyTorch or Flax to see how to install them with conda.
+Follow the installation pages of Flax, PyTorch or TensorFlow to see how to install them with conda.
 
-## Models architectures
+## Model architectures
 
 **[All the model checkpoints](https://huggingface.co/models)** provided by 🤗 Transformers are seamlessly integrated from the huggingface.co [model hub](https://huggingface.co) where they are uploaded directly by [users](https://huggingface.co/users) and [organizations](https://huggingface.co/organizations).
 
@@ -194,14 +205,21 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 1. **[BARThez](https://huggingface.co/transformers/model_doc/barthez.html)** (from École polytechnique) released with the paper [BARThez: a Skilled Pretrained French Sequence-to-Sequence Model](https://arxiv.org/abs/2010.12321) by Moussa Kamal Eddine, Antoine J.-P. Tixier, Michalis Vazirgiannis.
 1. **[BERT](https://huggingface.co/transformers/model_doc/bert.html)** (from Google) released with the paper [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://arxiv.org/abs/1810.04805) by Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova.
 1. **[BERT For Sequence Generation](https://huggingface.co/transformers/model_doc/bertgeneration.html)** (from Google) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
+1. **[BigBird-RoBERTa](https://huggingface.co/transformers/model_doc/bigbird.html)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
+1. **[BigBird-Pegasus](https://huggingface.co/transformers/model_doc/bigbird_pegasus.html)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
 1. **[Blenderbot](https://huggingface.co/transformers/model_doc/blenderbot.html)** (from Facebook) released with the paper [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
 1. **[BlenderbotSmall](https://huggingface.co/transformers/model_doc/blenderbot_small.html)** (from Facebook) released with the paper [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
 1. **[BORT](https://huggingface.co/transformers/model_doc/bort.html)** (from Alexa) released with the paper [Optimal Subarchitecture Extraction For BERT](https://arxiv.org/abs/2010.10499) by Adrian de Wynter and Daniel J. Perry.
+1. **[ByT5](https://huggingface.co/transformers/model_doc/byt5.html)** (from Google Research) released with the paper [ByT5: Towards a token-free future with pre-trained byte-to-byte models](https://arxiv.org/abs/2105.13626) by Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, Colin Raffel.
 1. **[CamemBERT](https://huggingface.co/transformers/model_doc/camembert.html)** (from Inria/Facebook/Sorbonne) released with the paper [CamemBERT: a Tasty French Language Model](https://arxiv.org/abs/1911.03894) by Louis Martin*, Benjamin Muller*, Pedro Javier Ortiz Suárez*, Yoann Dupont, Laurent Romary, Éric Villemonte de la Clergerie, Djamé Seddah and Benoît Sagot.
+1. **[CLIP](https://huggingface.co/transformers/model_doc/clip.html)** from (OpenAI) released with the paper [Learning Transferable Visual Models From Natural Language Supervision](https://arxiv.org/abs/2103.00020) by Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, Ilya Sutskever.
 1. **[ConvBERT](https://huggingface.co/transformers/model_doc/convbert.html)** (from YituTech) released with the paper [ConvBERT: Improving BERT with Span-based Dynamic Convolution](https://arxiv.org/abs/2008.02496) by Zihang Jiang, Weihao Yu, Daquan Zhou, Yunpeng Chen, Jiashi Feng, Shuicheng Yan.
+1. **[CPM](https://huggingface.co/transformers/model_doc/cpm.html)** (from Tsinghua University) released with the paper [CPM: A Large-scale Generative Chinese Pre-trained Language Model](https://arxiv.org/abs/2012.00413) by Zhengyan Zhang, Xu Han, Hao Zhou, Pei Ke, Yuxian Gu, Deming Ye, Yujia Qin, Yusheng Su, Haozhe Ji, Jian Guan, Fanchao Qi, Xiaozhi Wang, Yanan Zheng, Guoyang Zeng, Huanqi Cao, Shengqi Chen, Daixuan Li, Zhenbo Sun, Zhiyuan Liu, Minlie Huang, Wentao Han, Jie Tang, Juanzi Li, Xiaoyan Zhu, Maosong Sun.
 1. **[CTRL](https://huggingface.co/transformers/model_doc/ctrl.html)** (from Salesforce) released with the paper [CTRL: A Conditional Transformer Language Model for Controllable Generation](https://arxiv.org/abs/1909.05858) by Nitish Shirish Keskar*, Bryan McCann*, Lav R. Varshney, Caiming Xiong and Richard Socher.
 1. **[DeBERTa](https://huggingface.co/transformers/model_doc/deberta.html)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
 1. **[DeBERTa-v2](https://huggingface.co/transformers/model_doc/deberta_v2.html)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
+1. **[DeiT](https://huggingface.co/transformers/model_doc/deit.html)** (from Facebook) released with the paper [Training data-efficient image transformers & distillation through attention](https://arxiv.org/abs/2012.12877) by Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, Hervé Jégou.
+1. **[DETR](https://huggingface.co/transformers/model_doc/detr.html)** (from Facebook) released with the paper [End-to-End Object Detection with Transformers](https://arxiv.org/abs/2005.12872) by Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, Sergey Zagoruyko.
 1. **[DialoGPT](https://huggingface.co/transformers/model_doc/dialogpt.html)** (from Microsoft Research) released with the paper [DialoGPT: Large-Scale Generative Pre-training for Conversational Response Generation](https://arxiv.org/abs/1911.00536) by Yizhe Zhang, Siqi Sun, Michel Galley, Yen-Chun Chen, Chris Brockett, Xiang Gao, Jianfeng Gao, Jingjing Liu, Bill Dolan.
 1. **[DistilBERT](https://huggingface.co/transformers/model_doc/distilbert.html)** (from HuggingFace), released together with the paper [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://arxiv.org/abs/1910.01108) by Victor Sanh, Lysandre Debut and Thomas Wolf. The same method has been applied to compress GPT2 into [DistilGPT2](https://github.com/huggingface/transformers/tree/master/examples/distillation), RoBERTa into [DistilRoBERTa](https://github.com/huggingface/transformers/tree/master/examples/distillation), Multilingual BERT into [DistilmBERT](https://github.com/huggingface/transformers/tree/master/examples/distillation) and a German version of DistilBERT.
 1. **[DPR](https://huggingface.co/transformers/model_doc/dpr.html)** (from Facebook) released with the paper [Dense Passage Retrieval
@@ -212,26 +230,34 @@ Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih.
 1. **[Funnel Transformer](https://huggingface.co/transformers/model_doc/funnel.html)** (from CMU/Google Brain) released with the paper [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) by Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le.
 1. **[GPT](https://huggingface.co/transformers/model_doc/gpt.html)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
 1. **[GPT-2](https://huggingface.co/transformers/model_doc/gpt2.html)** (from OpenAI) released with the paper [Language Models are Unsupervised Multitask Learners](https://blog.openai.com/better-language-models/) by Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever**.
+1. **[GPT Neo](https://huggingface.co/transformers/model_doc/gpt_neo.html)** (from EleutherAI) released in the repository [EleutherAI/gpt-neo](https://github.com/EleutherAI/gpt-neo) by Sid Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy.
+1. **[Hubert](https://huggingface.co/transformers/model_doc/hubert.html)** (from Facebook) released with the paper [HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units](https://arxiv.org/abs/2106.07447) by Wei-Ning Hsu, Benjamin Bolte, Yao-Hung Hubert Tsai, Kushal Lakhotia, Ruslan Salakhutdinov, Abdelrahman Mohamed.
 1. **[I-BERT](https://huggingface.co/transformers/model_doc/ibert.html)** (from Berkeley) released with the paper [I-BERT: Integer-only BERT Quantization](https://arxiv.org/abs/2101.01321) by Sehoon Kim, Amir Gholami, Zhewei Yao, Michael W. Mahoney, Kurt Keutzer
 1. **[LayoutLM](https://huggingface.co/transformers/model_doc/layoutlm.html)** (from Microsoft Research Asia) released with the paper [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
 1. **[LED](https://huggingface.co/transformers/model_doc/led.html)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
 1. **[Longformer](https://huggingface.co/transformers/model_doc/longformer.html)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
+1. **[LUKE](https://huggingface.co/transformers/model_doc/luke.html)** (from Studio Ousia) released with the paper [LUKE: Deep Contextualized Entity Representations with Entity-aware Self-attention](https://arxiv.org/abs/2010.01057) by Ikuya Yamada, Akari Asai, Hiroyuki Shindo, Hideaki Takeda, Yuji Matsumoto.
 1. **[LXMERT](https://huggingface.co/transformers/model_doc/lxmert.html)** (from UNC Chapel Hill) released with the paper [LXMERT: Learning Cross-Modality Encoder Representations from Transformers for Open-Domain Question Answering](https://arxiv.org/abs/1908.07490) by Hao Tan and Mohit Bansal.
 1. **[M2M100](https://huggingface.co/transformers/model_doc/m2m_100.html)** (from Facebook) released with the paper [Beyond English-Centric Multilingual Machine Translation](https://arxiv.org/abs/2010.11125) by by Angela Fan, Shruti Bhosale, Holger Schwenk, Zhiyi Ma, Ahmed El-Kishky, Siddharth Goyal, Mandeep Baines, Onur Celebi, Guillaume Wenzek, Vishrav Chaudhary, Naman Goyal, Tom Birch, Vitaliy Liptchinsky, Sergey Edunov, Edouard Grave, Michael Auli, Armand Joulin.
 1. **[MarianMT](https://huggingface.co/transformers/model_doc/marian.html)** Machine translation models trained using [OPUS](http://opus.nlpl.eu/) data by Jörg Tiedemann. The [Marian Framework](https://marian-nmt.github.io/) is being developed by the Microsoft Translator Team.
 1. **[MBart](https://huggingface.co/transformers/model_doc/mbart.html)** (from Facebook) released with the paper [Multilingual Denoising Pre-training for Neural Machine Translation](https://arxiv.org/abs/2001.08210) by Yinhan Liu, Jiatao Gu, Naman Goyal, Xian Li, Sergey Edunov, Marjan Ghazvininejad, Mike Lewis, Luke Zettlemoyer.
 1. **[MBart-50](https://huggingface.co/transformers/model_doc/mbart.html)** (from Facebook) released with the paper [Multilingual Translation with Extensible Multilingual Pretraining and Finetuning](https://arxiv.org/abs/2008.00401) by Yuqing Tang, Chau Tran, Xian Li, Peng-Jen Chen, Naman Goyal, Vishrav Chaudhary, Jiatao Gu, Angela Fan.
+1. **[Megatron-BERT](https://huggingface.co/transformers/model_doc/megatron_bert.html)** (from NVIDIA) released with the paper [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) by Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
+1. **[Megatron-GPT2](https://huggingface.co/transformers/model_doc/megatron_gpt2.html)** (from NVIDIA) released with the paper [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) by Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
 1. **[MPNet](https://huggingface.co/transformers/model_doc/mpnet.html)** (from Microsoft Research) released with the paper [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297) by Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu.
 1. **[MT5](https://huggingface.co/transformers/model_doc/mt5.html)** (from Google AI) released with the paper [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) by Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel.
 1. **[Pegasus](https://huggingface.co/transformers/model_doc/pegasus.html)** (from Google) released with the paper [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777)> by Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.
 1. **[ProphetNet](https://huggingface.co/transformers/model_doc/prophetnet.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. **[Reformer](https://huggingface.co/transformers/model_doc/reformer.html)** (from Google Research) released with the paper [Reformer: The Efficient Transformer](https://arxiv.org/abs/2001.04451) by Nikita Kitaev, Łukasz Kaiser, Anselm Levskaya.
 1. **[RoBERTa](https://huggingface.co/transformers/model_doc/roberta.html)** (from Facebook), released together with the paper a [Robustly Optimized BERT Pretraining Approach](https://arxiv.org/abs/1907.11692) by Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov.
+1. **[RoFormer](https://huggingface.co/transformers/model_doc/roformer.html)** (from ZhuiyiTechnology), released together with the paper a [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/pdf/2104.09864v1.pdf) by Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu.
 1. **[SpeechToTextTransformer](https://huggingface.co/transformers/model_doc/speech_to_text.html)** (from Facebook), released together with the paper [fairseq S2T: Fast Speech-to-Text Modeling with fairseq](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Dmytro Okhonko, Juan Pino.
 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/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. **[Vision Transformer (ViT)](https://huggingface.co/transformers/model_doc/vit.html)** (from Google AI) released with the paper [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929) by Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby.
+1. **[VisualBERT](https://huggingface.co/transformers/model_doc/visual_bert.html)** (from UCLA NLP) released with the paper [VisualBERT: A Simple and Performant Baseline for Vision and Language](https://arxiv.org/pdf/1908.03557) by Liunian Harold Li, Mark Yatskar, Da Yin, Cho-Jui Hsieh, Kai-Wei Chang.
 1. **[Wav2Vec2](https://huggingface.co/transformers/model_doc/wav2vec2.html)** (from Facebook AI) released with the paper [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://arxiv.org/abs/2006.11477) by Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, Michael Auli.
 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.
@@ -240,9 +266,9 @@ Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih.
 1. **[XLSR-Wav2Vec2](https://huggingface.co/transformers/model_doc/xlsr_wav2vec2.html)** (from Facebook AI) released with the paper [Unsupervised Cross-Lingual Representation Learning For Speech Recognition](https://arxiv.org/abs/2006.13979) by Alexis Conneau, Alexei Baevski, Ronan Collobert, Abdelrahman Mohamed, Michael Auli.
 1. Want to contribute a new model? We have added a **detailed guide and templates** to guide you in the process of adding a new model. You can find them in the [`templates`](./templates) folder of the repository. Be sure to check the [contributing guidelines](./CONTRIBUTING.md) and contact the maintainers or open an issue to collect feedbacks before starting your PR.
 
-To check if each model has an implementation in PyTorch/TensorFlow/Flax or has an associated tokenizer backed by the 🤗 Tokenizers library, refer to [this table](https://huggingface.co/transformers/index.html#bigtable)
+To check if each model has an implementation in Flax, PyTorch or TensorFlow, or has an associated tokenizer backed by the 🤗 Tokenizers library, refer to [this table](https://huggingface.co/transformers/index.html#supported-frameworks).
 
-These implementations have been tested on several datasets (see the example scripts) and should match the performances of the original implementations. You can find more details on the performances in the Examples section of the [documentation](https://huggingface.co/transformers/examples.html).
+These implementations have been tested on several datasets (see the example scripts) and should match the performance of the original implementations. You can find more details on performance in the Examples section of the [documentation](https://huggingface.co/transformers/examples.html).
 
 
 ## Learn more

docker/transformers-pytorch-tpu/Dockerfile

@@ -53,7 +53,7 @@ RUN git clone https://github.com/huggingface/transformers.git && \
     git checkout CI && \
     cd .. && \
     pip install ./transformers && \
-    pip install -r ./transformers/examples/requirements.txt && \
+    pip install -r ./transformers/examples/pytorch/_test_requirements.txt && \
     pip install pytest
 
 RUN python -c "import torch_xla; print(torch_xla.__version__)"

docker/transformers-pytorch-tpu/bert-base-cased.jsonnet

@@ -27,7 +27,7 @@ local bertBaseCased = base.BaseTest {
   },
   command: utils.scriptCommand(
     |||
-      python -m pytest -s transformers/examples/test_xla_examples.py -v
+      python -m pytest -s transformers/examples/pytorch/test_xla_examples.py -v
       test_exit_code=$?
       echo "\nFinished running commands.\n"
       test $test_exit_code -eq 0

docs/source/_static/js/custom.js

@@ -1,10 +1,14 @@
 // These two things need to be updated at each release for the version selector.
 // Last stable version
-const stableVersion = "v4.3.2"
+const stableVersion = "v4.7.0"
 // Dictionary doc folder to label. The last stable version should have an empty key.
 const versionMapping = {
     "master": "master",
-    "": "v4.3.0/v4.3.1/v4.3.2/v4.3.3 (stable)",
+    "": "v4.7.0 (stable)",
+    "v4.6.0": "v4.6.0",
+    "v4.5.1": "v4.5.0/v4.5.1",
+    "v4.4.2": "v4.4.0/v4.4.1/v4.4.2",
+    "v4.3.3": "v4.3.0/v4.3.1/v4.3.2/v4.3.3",
     "v4.2.2": "v4.2.0/v4.2.1/v4.2.2",
     "v4.1.1": "v4.1.0/v4.1.1",
     "v4.0.1": "v4.0.0/v4.0.1",
@@ -61,7 +65,7 @@ function addIcon() {
 function addCustomFooter() {
     const customFooter = document.createElement("div");
     const questionOrIssue = document.createElement("div");
-    questionOrIssue.innerHTML = "Stuck? Read our <a href='https://medium.com/huggingface'>Blog posts</a> or <a href='https://github.com/huggingface/transformers'>Create an issue</a>";
+    questionOrIssue.innerHTML = "Stuck? Read our <a href='https://huggingface.co/blog'>Blog posts</a> or <a href='https://github.com/huggingface/transformers'>Create an issue</a>";
     customFooter.appendChild(questionOrIssue);
     customFooter.classList.add("footer");
 

docs/source/add_new_model.rst

@@ -388,7 +388,7 @@ Next, you can finally start adding new code to 🤗 Transformers. Go into the cl
 
 ::
 
-   cd transformers
+    cd transformers
 
 In the special case that you are adding a model whose architecture exactly matches the model architecture of an
 existing model you only have to add a conversion script as described in `this section <#write-a-conversion-script>`__.
@@ -417,27 +417,27 @@ You should do the following:
 
 ::
 
-   git checkout -b add_brand_new_bert
+    git checkout -b add_brand_new_bert
 
 2. Commit the automatically generated code:
 
 ::
 
-   git add .
-   git commit
+    git add .
+    git commit
 
 3. Fetch and rebase to current master
 
 ::
 
-   git fetch upstream
-   git rebase upstream/master
+    git fetch upstream
+    git rebase upstream/master
 
 4. Push the changes to your account using:
 
 ::
 
-   git push -u origin a-descriptive-name-for-my-changes
+    git push -u origin a-descriptive-name-for-my-changes
 
 5. Once you are satisfied, go to the webpage of your fork on GitHub. Click on “Pull request”. Make sure to add the
    GitHub handle of some members of the Hugging Face team as reviewers, so that the Hugging Face team gets notified for
@@ -451,8 +451,8 @@ time to time by doing:
 
 ::
 
-   git fetch upstream
-   git merge upstream/master
+    git fetch upstream
+    git merge upstream/master
 
 In general, all questions you might have regarding the model or your implementation should be asked in your PR and
 discussed/solved in the PR. This way, the Hugging Face team will always be notified when you are committing new code or
@@ -518,7 +518,7 @@ PyTorch, called ``SimpleModel`` as follows:
 
 .. code:: python
 
-   import torch.nn as nn
+   from torch import nn
 
    class SimpleModel(nn.Module):
        def __init__(self):

docs/source/benchmarks.rst

@@ -65,10 +65,10 @@ respectively.
 .. code-block:: bash
 
     ## PYTORCH CODE
-    python examples/benchmarking/run_benchmark.py --help
+    python examples/pytorch/benchmarking/run_benchmark.py --help
 
     ## TENSORFLOW CODE
-    python examples/benchmarking/run_benchmark_tf.py --help
+    python examples/tensorflow/benchmarking/run_benchmark_tf.py --help
 
 
 An instantiated benchmark object can then simply be run by calling ``benchmark.run()``.
@@ -358,4 +358,6 @@ available `here
 <https://docs.google.com/spreadsheets/d/1sryqufw2D0XlUH4sq3e9Wnxu5EAQkaohzrJbd5HdQ_w/edit?usp=sharing>`__.
 
 With the new `benchmark` tools, it is easier than ever to share your benchmark results with the community
-:prefix_link:`here <examples/benchmarking/README.md>`.
+
+- :prefix_link:`PyTorch Benchmarking Results<examples/pytorch/benchmarking/README.md>`.
+- :prefix_link:`TensorFlow Benchmarking Results<examples/tensorflow/benchmarking/README.md>`.

docs/source/community.md

@@ -48,3 +48,16 @@ This page regroups resources around 🤗 Transformers developed by the community
 |[Fine-Tune LED on up to 8K tokens](https://github.com/patrickvonplaten/notebooks/blob/master/Fine_tune_Longformer_Encoder_Decoder_(LED)_for_Summarization_on_pubmed.ipynb) | How to fine-tune LED on pubmed for long-range summarization | [Patrick von Platen](https://github.com/patrickvonplaten) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/patrickvonplaten/notebooks/blob/master/Fine_tune_Longformer_Encoder_Decoder_(LED)_for_Summarization_on_pubmed.ipynb)|
 |[Evaluate LED on Arxiv](https://github.com/patrickvonplaten/notebooks/blob/master/LED_on_Arxiv.ipynb) | How to effectively evaluate LED on long-range summarization | [Patrick von Platen](https://github.com/patrickvonplaten) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/patrickvonplaten/notebooks/blob/master/LED_on_Arxiv.ipynb)|
 |[Fine-tune LayoutLM on RVL-CDIP (a document image classification dataset)](https://github.com/NielsRogge/Transformers-Tutorials/blob/master/LayoutLM/Fine_tuning_LayoutLMForSequenceClassification_on_RVL_CDIP.ipynb) | How to fine-tune *LayoutLMForSequenceClassification* on the RVL-CDIP dataset for scanned document classification | [Niels Rogge](https://github.com/nielsrogge) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/NielsRogge/Transformers-Tutorials/blob/master/LayoutLM/Fine_tuning_LayoutLMForSequenceClassification_on_RVL_CDIP.ipynb)|
+|[Wav2Vec2 CTC decoding with GPT2 adjustment](https://github.com/voidful/huggingface_notebook/blob/main/xlsr_gpt.ipynb) | How to decode CTC sequence with language model adjustment | [Eric Lam](https://github.com/voidful) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1e_z5jQHYbO2YKEaUgzb1ww1WwiAyydAj?usp=sharing)|
+|[Fine-tune BART for summarization in two languages with Trainer class](https://github.com/elsanns/xai-nlp-notebooks/blob/master/fine_tune_bart_summarization_two_langs.ipynb) | How to fine-tune BART for summarization in two languages with Trainer class | [Eliza Szczechla](https://github.com/elsanns) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/elsanns/xai-nlp-notebooks/blob/master/fine_tune_bart_summarization_two_langs.ipynb)|
+|[Evaluate Big Bird on Trivia QA](https://github.com/patrickvonplaten/notebooks/blob/master/Evaluating_Big_Bird_on_TriviaQA.ipynb) | How to evaluate BigBird on long document question answering on Trivia QA | [Patrick von Platen](https://github.com/patrickvonplaten) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/patrickvonplaten/notebooks/blob/master/Evaluating_Big_Bird_on_TriviaQA.ipynb)|
+| [Create video captions using Wav2Vec2](https://github.com/Muennighoff/ytclipcc/blob/main/wav2vec_youtube_captions.ipynb) | How to create YouTube captions from any video by transcribing the audio with Wav2Vec | [Niklas Muennighoff](https://github.com/Muennighoff) |[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/Muennighoff/ytclipcc/blob/main/wav2vec_youtube_captions.ipynb) |
+| [Fine-tune the Vision Transformer on CIFAR-10 using PyTorch Lightning](https://github.com/NielsRogge/Transformers-Tutorials/blob/master/VisionTransformer/Fine_tuning_the_Vision_Transformer_on_CIFAR_10_with_PyTorch_Lightning.ipynb) | How to fine-tune the Vision Transformer (ViT) on CIFAR-10 using HuggingFace Transformers, Datasets and PyTorch Lightning | [Niels Rogge](https://github.com/nielsrogge) |[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/NielsRogge/Transformers-Tutorials/blob/master/VisionTransformer/Fine_tuning_the_Vision_Transformer_on_CIFAR_10_with_PyTorch_Lightning.ipynb) |
+| [Fine-tune the Vision Transformer on CIFAR-10 using the 🤗 Trainer](https://github.com/NielsRogge/Transformers-Tutorials/blob/master/VisionTransformer/Fine_tuning_the_Vision_Transformer_on_CIFAR_10_with_the_%F0%9F%A4%97_Trainer.ipynb) | How to fine-tune the Vision Transformer (ViT) on CIFAR-10 using HuggingFace Transformers, Datasets and the 🤗 Trainer | [Niels Rogge](https://github.com/nielsrogge) |[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/NielsRogge/Transformers-Tutorials/blob/master/VisionTransformer/Fine_tuning_the_Vision_Transformer_on_CIFAR_10_with_the_%F0%9F%A4%97_Trainer.ipynb) |
+| [Evaluate LUKE on Open Entity, an entity typing dataset](https://github.com/studio-ousia/luke/blob/master/notebooks/huggingface_open_entity.ipynb) | How to evaluate *LukeForEntityClassification* on the Open Entity dataset | [Ikuya Yamada](https://github.com/ikuyamada) |[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/studio-ousia/luke/blob/master/notebooks/huggingface_open_entity.ipynb) |
+| [Evaluate LUKE on TACRED, a relation extraction dataset](https://github.com/studio-ousia/luke/blob/master/notebooks/huggingface_tacred.ipynb) | How to evaluate *LukeForEntityPairClassification* on the TACRED dataset | [Ikuya Yamada](https://github.com/ikuyamada) |[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/studio-ousia/luke/blob/master/notebooks/huggingface_tacred.ipynb) |
+| [Evaluate LUKE on CoNLL-2003, an important NER benchmark](https://github.com/studio-ousia/luke/blob/master/notebooks/huggingface_conll_2003.ipynb) | How to evaluate *LukeForEntitySpanClassification* on the CoNLL-2003 dataset | [Ikuya Yamada](https://github.com/ikuyamada) |[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/studio-ousia/luke/blob/master/notebooks/huggingface_conll_2003.ipynb) |
+| [Evaluate BigBird-Pegasus on PubMed dataset](https://github.com/vasudevgupta7/bigbird/blob/main/notebooks/bigbird_pegasus_evaluation.ipynb) | How to evaluate *BigBirdPegasusForConditionalGeneration* on PubMed dataset | [Vasudev Gupta](https://github.com/vasudevgupta7) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/vasudevgupta7/bigbird/blob/main/notebooks/bigbird_pegasus_evaluation.ipynb) |
+| [Speech Emotion Classification with Wav2Vec2](https://github/m3hrdadfi/soxan/blob/main/notebooks/Emotion_recognition_in_Greek_speech_using_Wav2Vec2.ipynb) | How to leverage a pretrained Wav2Vec2 model for Emotion Classification on the MEGA dataset | [Mehrdad Farahani](https://github.com/m3hrdadfi) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/m3hrdadfi/soxan/blob/main/notebooks/Emotion_recognition_in_Greek_speech_using_Wav2Vec2.ipynb) |
+| [Detect objects in an image with DETR](https://github.com/NielsRogge/Transformers-Tutorials/blob/master/DETR/DETR_minimal_example_(with_DetrFeatureExtractor).ipynb) | How to use a trained *DetrForObjectDetection* model to detect objects in an image and visualize attention | [Niels Rogge](https://github.com/NielsRogge) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/NielsRogge/Transformers-Tutorials/blob/master/DETR/DETR_minimal_example_(with_DetrFeatureExtractor).ipynb) |
+| [Fine-tune DETR on a custom object detection dataset](https://github.com/NielsRogge/Transformers-Tutorials/blob/master/DETR/Fine_tuning_DetrForObjectDetection_on_custom_dataset_(balloon).ipynb) | How to fine-tune *DetrForObjectDetection* on a custom object detection dataset | [Niels Rogge](https://github.com/NielsRogge) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/NielsRogge/Transformers-Tutorials/blob/master/DETR/Fine_tuning_DetrForObjectDetection_on_custom_dataset_(balloon).ipynb) |

docs/source/conf.py

@@ -14,24 +14,25 @@
 #
 import os
 import sys
-sys.path.insert(0, os.path.abspath('../../src'))
+
+sys.path.insert(0, os.path.abspath("../../src"))
 
 
 # -- Project information -----------------------------------------------------
 
-project = u'transformers'
-copyright = u'2020, The Hugging Face Team, Licenced under the Apache License, Version 2.0'
-author = u'huggingface'
+project = "transformers"
+copyright = "2020, The Hugging Face Team, Licenced under the Apache License, Version 2.0"
+author = "huggingface"
 
 # The short X.Y version
-version = u''
+version = ""
 # The full version, including alpha/beta/rc tags
-release = u'4.4.2'
+release = u'4.7.0'
 
 
 
 # Prefix link to point to master, comment this during version release and uncomment below line
-extlinks = {'prefix_link': ('https://github.com/huggingface/transformers/blob/master/%s', '')}
+extlinks = {"prefix_link": ("https://github.com/huggingface/transformers/blob/master/%s", "")}
 # Prefix link to always point to corresponding version, uncomment this during version release
 # extlinks = {'prefix_link': ('https://github.com/huggingface/transformers/blob/v'+ release + '/%s', '')}
 
@@ -45,27 +46,28 @@ extlinks = {'prefix_link': ('https://github.com/huggingface/transformers/blob/ma
 # extensions coming with Sphinx (named 'sphinx.ext.*') or your custom
 # ones.
 extensions = [
-    'sphinx.ext.autodoc',
-    'sphinx.ext.extlinks',
-    'sphinx.ext.coverage',
-    'sphinx.ext.napoleon',
-    'recommonmark',
-    'sphinx.ext.viewcode',
-    'sphinx_markdown_tables',
-    'sphinx_copybutton'
+    "sphinx.ext.autodoc",
+    "sphinx.ext.extlinks",
+    "sphinx.ext.coverage",
+    "sphinx.ext.napoleon",
+    "recommonmark",
+    "sphinx.ext.viewcode",
+    "sphinx_markdown_tables",
+    "sphinxext.opengraph",
+    "sphinx_copybutton",
 ]
 
 # Add any paths that contain templates here, relative to this directory.
-templates_path = ['_templates']
+templates_path = ["_templates"]
 
 # The suffix(es) of source filenames.
 # You can specify multiple suffix as a list of string:
 #
-source_suffix = ['.rst', '.md']
+source_suffix = [".rst", ".md"]
 # source_suffix = '.rst'
 
 # The master toctree document.
-master_doc = 'index'
+master_doc = "index"
 
 # The language for content autogenerated by Sphinx. Refer to documentation
 # for a list of supported languages.
@@ -77,7 +79,7 @@ language = None
 # List of patterns, relative to source directory, that match files and
 # directories to ignore when looking for source files.
 # This pattern also affects html_static_path and html_extra_path.
-exclude_patterns = [u'_build', 'Thumbs.db', '.DS_Store']
+exclude_patterns = ["_build", "Thumbs.db", ".DS_Store"]
 
 # The name of the Pygments (syntax highlighting) style to use.
 pygments_style = None
@@ -91,21 +93,30 @@ copybutton_prompt_is_regexp = True
 # The theme to use for HTML and HTML Help pages.  See the documentation for
 # a list of builtin themes.
 #
-html_theme = 'sphinx_rtd_theme'
+html_theme = "sphinx_rtd_theme"
 
 # Theme options are theme-specific and customize the look and feel of a theme
 # further.  For a list of options available for each theme, see the
 # documentation.
 #
-html_theme_options = {
-    'analytics_id': 'UA-83738774-2',
-    'navigation_with_keys': True
-}
+html_theme_options = {"analytics_id": "UA-83738774-2", "navigation_with_keys": True}
+
+#  Configuration for OpenGraph and Twitter Card Tags.
+# These are responsible for creating nice shareable social images https://ahrefs.com/blog/open-graph-meta-tags/
+# https://ogp.me/#type_website
+ogp_image = "https://huggingface.co/front/thumbnails/transformers.png"
+ogp_description = "State-of-the-art Natural Language Processing for PyTorch and TensorFlow 2.0. Transformers provides thousands of pretrained models to perform tasks on texts such as classification, information extraction, question answering, summarization, translation, text generation, etc in 100+ languages. Its aim is to make cutting-edge NLP easier to use for everyone"
+ogp_description_length = 160
+
+ogp_custom_meta_tags = [
+    f'<meta name="twitter:image" content="{ogp_image}">',
+    f'<meta name="twitter:description" content="{ogp_description}">',
+]
 
 # Add any paths that contain custom static files (such as style sheets) here,
 # relative to this directory. They are copied after the builtin static files,
 # so a file named "default.css" will overwrite the builtin "default.css".
-html_static_path = ['_static']
+html_static_path = ["_static"]
 
 # Custom sidebar templates, must be a dictionary that maps document names
 # to template names.
@@ -117,17 +128,17 @@ html_static_path = ['_static']
 #
 # html_sidebars = {}
 
-# This must be the name of an image file (path relative to the configuration 
-# directory) that is the favicon of the docs. Modern browsers use this as 
-# the icon for tabs, windows and bookmarks. It should be a Windows-style 
+# This must be the name of an image file (path relative to the configuration
+# directory) that is the favicon of the docs. Modern browsers use this as
+# the icon for tabs, windows and bookmarks. It should be a Windows-style
 # icon file (.ico).
-html_favicon = 'favicon.ico'
+html_favicon = "favicon.ico"
 
 
 # -- Options for HTMLHelp output ---------------------------------------------
 
 # Output file base name for HTML help builder.
-htmlhelp_basename = 'transformersdoc'
+htmlhelp_basename = "transformersdoc"
 
 
 # -- Options for LaTeX output ------------------------------------------------
@@ -136,15 +147,12 @@ latex_elements = {
     # The paper size ('letterpaper' or 'a4paper').
     #
     # 'papersize': 'letterpaper',
-
     # The font size ('10pt', '11pt' or '12pt').
     #
     # 'pointsize': '10pt',
-
     # Additional stuff for the LaTeX preamble.
     #
     # 'preamble': '',
-
     # Latex figure (float) alignment
     #
     # 'figure_align': 'htbp',
@@ -154,8 +162,7 @@ latex_elements = {
 # (source start file, target name, title,
 #  author, documentclass [howto, manual, or own class]).
 latex_documents = [
-    (master_doc, 'transformers.tex', u'transformers Documentation',
-     u'huggingface', 'manual'),
+    (master_doc, "transformers.tex", "transformers Documentation", "huggingface", "manual"),
 ]
 
 
@@ -163,10 +170,7 @@ latex_documents = [
 
 # One entry per manual page. List of tuples
 # (source start file, name, description, authors, manual section).
-man_pages = [
-    (master_doc, 'transformers', u'transformers Documentation',
-     [author], 1)
-]
+man_pages = [(master_doc, "transformers", "transformers Documentation", [author], 1)]
 
 
 # -- Options for Texinfo output ----------------------------------------------
@@ -175,9 +179,15 @@ man_pages = [
 # (source start file, target name, title, author,
 #  dir menu entry, description, category)
 texinfo_documents = [
-    (master_doc, 'transformers', u'transformers Documentation',
-     author, 'transformers', 'One line description of project.',
-     'Miscellaneous'),
+    (
+        master_doc,
+        "transformers",
+        "transformers Documentation",
+        author,
+        "transformers",
+        "One line description of project.",
+        "Miscellaneous",
+    ),
 ]
 
 
@@ -196,11 +206,13 @@ epub_title = project
 # epub_uid = ''
 
 # A list of files that should not be packed into the epub file.
-epub_exclude_files = ['search.html']
+epub_exclude_files = ["search.html"]
+
 
 def setup(app):
-    app.add_css_file('css/huggingface.css')
-    app.add_css_file('css/code-snippets.css')
-    app.add_js_file('js/custom.js')
+    app.add_css_file("css/huggingface.css")
+    app.add_css_file("css/code-snippets.css")
+    app.add_js_file("js/custom.js")
+
 
 # -- Extension configuration -------------------------------------------------

docs/source/converting_tensorflow_models.rst

@@ -33,8 +33,8 @@ You can convert any TensorFlow checkpoint for BERT (in particular `the pre-train
 This CLI takes as input a TensorFlow checkpoint (three files starting with ``bert_model.ckpt``\ ) and the associated
 configuration file (\ ``bert_config.json``\ ), and creates a PyTorch model for this configuration, loads the weights
 from the TensorFlow checkpoint in the PyTorch model and saves the resulting model in a standard PyTorch save file that
-can be imported using ``from_pretrained()`` (see example in :doc:`quicktour` , `run_glue.py
-<https://github.com/huggingface/transformers/blob/master/examples/text-classification/run_glue.py>`_\ ).
+can be imported using ``from_pretrained()`` (see example in :doc:`quicktour` , :prefix_link:`run_glue.py
+<examples/pytorch/text-classification/run_glue.py>` \ ).
 
 You only need to run this conversion script **once** to get a PyTorch model. You can then disregard the TensorFlow
 checkpoint (the three files starting with ``bert_model.ckpt``\ ) but be sure to keep the configuration file (\
@@ -47,12 +47,12 @@ Here is an example of the conversion process for a pre-trained ``BERT-Base Uncas
 
 .. code-block:: shell
 
-   export BERT_BASE_DIR=/path/to/bert/uncased_L-12_H-768_A-12
+    export BERT_BASE_DIR=/path/to/bert/uncased_L-12_H-768_A-12
 
-   transformers-cli convert --model_type bert \
-     --tf_checkpoint $BERT_BASE_DIR/bert_model.ckpt \
-     --config $BERT_BASE_DIR/bert_config.json \
-     --pytorch_dump_output $BERT_BASE_DIR/pytorch_model.bin
+    transformers-cli convert --model_type bert \
+      --tf_checkpoint $BERT_BASE_DIR/bert_model.ckpt \
+      --config $BERT_BASE_DIR/bert_config.json \
+      --pytorch_dump_output $BERT_BASE_DIR/pytorch_model.bin
 
 You can download Google's pre-trained models for the conversion `here
 <https://github.com/google-research/bert#pre-trained-models>`__.
@@ -72,12 +72,12 @@ Here is an example of the conversion process for the pre-trained ``ALBERT Base``
 
 .. code-block:: shell
 
-   export ALBERT_BASE_DIR=/path/to/albert/albert_base
+    export ALBERT_BASE_DIR=/path/to/albert/albert_base
 
-   transformers-cli convert --model_type albert \
-     --tf_checkpoint $ALBERT_BASE_DIR/model.ckpt-best \
-     --config $ALBERT_BASE_DIR/albert_config.json \
-     --pytorch_dump_output $ALBERT_BASE_DIR/pytorch_model.bin
+    transformers-cli convert --model_type albert \
+      --tf_checkpoint $ALBERT_BASE_DIR/model.ckpt-best \
+      --config $ALBERT_BASE_DIR/albert_config.json \
+      --pytorch_dump_output $ALBERT_BASE_DIR/pytorch_model.bin
 
 You can download Google's pre-trained models for the conversion `here
 <https://github.com/google-research/albert#pre-trained-models>`__.
@@ -91,13 +91,13 @@ save as the same format than OpenAI pretrained model (see `here <https://github.
 
 .. code-block:: shell
 
-   export OPENAI_GPT_CHECKPOINT_FOLDER_PATH=/path/to/openai/pretrained/numpy/weights
+    export OPENAI_GPT_CHECKPOINT_FOLDER_PATH=/path/to/openai/pretrained/numpy/weights
 
-   transformers-cli convert --model_type gpt \
-     --tf_checkpoint $OPENAI_GPT_CHECKPOINT_FOLDER_PATH \
-     --pytorch_dump_output $PYTORCH_DUMP_OUTPUT \
-     [--config OPENAI_GPT_CONFIG] \
-     [--finetuning_task_name OPENAI_GPT_FINETUNED_TASK] \
+    transformers-cli convert --model_type gpt \
+      --tf_checkpoint $OPENAI_GPT_CHECKPOINT_FOLDER_PATH \
+      --pytorch_dump_output $PYTORCH_DUMP_OUTPUT \
+      [--config OPENAI_GPT_CONFIG] \
+      [--finetuning_task_name OPENAI_GPT_FINETUNED_TASK] \
 
 
 OpenAI GPT-2
@@ -108,13 +108,13 @@ Here is an example of the conversion process for a pre-trained OpenAI GPT-2 mode
 
 .. code-block:: shell
 
-   export OPENAI_GPT2_CHECKPOINT_PATH=/path/to/gpt2/pretrained/weights
+    export OPENAI_GPT2_CHECKPOINT_PATH=/path/to/gpt2/pretrained/weights
 
-   transformers-cli convert --model_type gpt2 \
-     --tf_checkpoint $OPENAI_GPT2_CHECKPOINT_PATH \
-     --pytorch_dump_output $PYTORCH_DUMP_OUTPUT \
-     [--config OPENAI_GPT2_CONFIG] \
-     [--finetuning_task_name OPENAI_GPT2_FINETUNED_TASK]
+    transformers-cli convert --model_type gpt2 \
+      --tf_checkpoint $OPENAI_GPT2_CHECKPOINT_PATH \
+      --pytorch_dump_output $PYTORCH_DUMP_OUTPUT \
+      [--config OPENAI_GPT2_CONFIG] \
+      [--finetuning_task_name OPENAI_GPT2_FINETUNED_TASK]
 
 Transformer-XL
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
@@ -124,13 +124,13 @@ Here is an example of the conversion process for a pre-trained Transformer-XL mo
 
 .. code-block:: shell
 
-   export TRANSFO_XL_CHECKPOINT_FOLDER_PATH=/path/to/transfo/xl/checkpoint
+    export TRANSFO_XL_CHECKPOINT_FOLDER_PATH=/path/to/transfo/xl/checkpoint
 
-   transformers-cli convert --model_type transfo_xl \
-     --tf_checkpoint $TRANSFO_XL_CHECKPOINT_FOLDER_PATH \
-     --pytorch_dump_output $PYTORCH_DUMP_OUTPUT \
-     [--config TRANSFO_XL_CONFIG] \
-     [--finetuning_task_name TRANSFO_XL_FINETUNED_TASK]
+    transformers-cli convert --model_type transfo_xl \
+      --tf_checkpoint $TRANSFO_XL_CHECKPOINT_FOLDER_PATH \
+      --pytorch_dump_output $PYTORCH_DUMP_OUTPUT \
+      [--config TRANSFO_XL_CONFIG] \
+      [--finetuning_task_name TRANSFO_XL_FINETUNED_TASK]
 
 
 XLNet
@@ -140,14 +140,14 @@ Here is an example of the conversion process for a pre-trained XLNet model:
 
 .. code-block:: shell
 
-   export TRANSFO_XL_CHECKPOINT_PATH=/path/to/xlnet/checkpoint
-   export TRANSFO_XL_CONFIG_PATH=/path/to/xlnet/config
+    export TRANSFO_XL_CHECKPOINT_PATH=/path/to/xlnet/checkpoint
+    export TRANSFO_XL_CONFIG_PATH=/path/to/xlnet/config
 
-   transformers-cli convert --model_type xlnet \
-     --tf_checkpoint $TRANSFO_XL_CHECKPOINT_PATH \
-     --config $TRANSFO_XL_CONFIG_PATH \
-     --pytorch_dump_output $PYTORCH_DUMP_OUTPUT \
-     [--finetuning_task_name XLNET_FINETUNED_TASK] \
+    transformers-cli convert --model_type xlnet \
+      --tf_checkpoint $TRANSFO_XL_CHECKPOINT_PATH \
+      --config $TRANSFO_XL_CONFIG_PATH \
+      --pytorch_dump_output $PYTORCH_DUMP_OUTPUT \
+      [--finetuning_task_name XLNET_FINETUNED_TASK] \
 
 
 XLM
@@ -157,13 +157,13 @@ Here is an example of the conversion process for a pre-trained XLM model:
 
 .. code-block:: shell
 
-   export XLM_CHECKPOINT_PATH=/path/to/xlm/checkpoint
+    export XLM_CHECKPOINT_PATH=/path/to/xlm/checkpoint
 
-   transformers-cli convert --model_type xlm \
-     --tf_checkpoint $XLM_CHECKPOINT_PATH \
-     --pytorch_dump_output $PYTORCH_DUMP_OUTPUT
-    [--config XML_CONFIG] \
-    [--finetuning_task_name XML_FINETUNED_TASK]
+    transformers-cli convert --model_type xlm \
+      --tf_checkpoint $XLM_CHECKPOINT_PATH \
+      --pytorch_dump_output $PYTORCH_DUMP_OUTPUT
+     [--config XML_CONFIG] \
+     [--finetuning_task_name XML_FINETUNED_TASK]
 
 
 T5
@@ -173,9 +173,9 @@ Here is an example of the conversion process for a pre-trained T5 model:
 
 .. code-block:: shell
 
-   export T5=/path/to/t5/uncased_L-12_H-768_A-12
+    export T5=/path/to/t5/uncased_L-12_H-768_A-12
 
-   transformers-cli convert --model_type t5 \
-     --tf_checkpoint $T5/t5_model.ckpt \
-     --config $T5/t5_config.json \
-     --pytorch_dump_output $T5/pytorch_model.bin
+    transformers-cli convert --model_type t5 \
+      --tf_checkpoint $T5/t5_model.ckpt \
+      --config $T5/t5_config.json \
+      --pytorch_dump_output $T5/pytorch_model.bin

docs/source/custom_datasets.rst

@@ -15,10 +15,10 @@ Fine-tuning with custom datasets
 
 .. note::
 
-    The datasets used in this tutorial are available and can be more easily accessed using the `🤗 NLP library
-    <https://github.com/huggingface/nlp>`_. We do not use this library to access the datasets here since this tutorial
-    meant to illustrate how to work with your own data. A brief of introduction can be found at the end of the tutorial
-    in the section ":ref:`nlplib`".
+    The datasets used in this tutorial are available and can be more easily accessed using the `🤗 Datasets library
+    <https://github.com/huggingface/datasets>`_. We do not use this library to access the datasets here since this
+    tutorial meant to illustrate how to work with your own data. A brief of introduction can be found at the end of the
+    tutorial in the section ":ref:`datasetslib`".
 
 This tutorial will take you through several examples of using 🤗 Transformers models with your own datasets. The guide
 shows one of many valid workflows for using these models and is meant to be illustrative rather than definitive. We
@@ -41,7 +41,7 @@ Sequence Classification with IMDb Reviews
 .. note::
 
     This dataset can be explored in the Hugging Face model hub (`IMDb <https://huggingface.co/datasets/imdb>`_), and
-    can be alternatively downloaded with the 🤗 NLP library with ``load_dataset("imdb")``.
+    can be alternatively downloaded with the 🤗 Datasets library with ``load_dataset("imdb")``.
 
 In this example, we'll show how to download, tokenize, and train a model on the IMDb reviews dataset. This task takes
 the text of a review and requires the model to predict whether the sentiment of the review is positive or negative.
@@ -260,7 +260,7 @@ Token Classification with W-NUT Emerging Entities
 .. note::
 
     This dataset can be explored in the Hugging Face model hub (`WNUT-17 <https://huggingface.co/datasets/wnut_17>`_),
-    and can be alternatively downloaded with the 🤗 NLP library with ``load_dataset("wnut_17")``.
+    and can be alternatively downloaded with the 🤗 Datasets library with ``load_dataset("wnut_17")``.
 
 Next we will look at token classification. Rather than classifying an entire sequence, this task classifies token by
 token. We'll demonstrate how to do this with `Named Entity Recognition
@@ -459,7 +459,7 @@ Question Answering with SQuAD 2.0
 .. note::
 
     This dataset can be explored in the Hugging Face model hub (`SQuAD V2
-    <https://huggingface.co/datasets/squad_v2>`_), and can be alternatively downloaded with the 🤗 NLP library with
+    <https://huggingface.co/datasets/squad_v2>`_), and can be alternatively downloaded with the 🤗 Datasets library with
     ``load_dataset("squad_v2")``.
 
 Question answering comes in many forms. In this example, we'll look at the particular type of extractive QA that
@@ -677,22 +677,23 @@ Additional Resources
   - :doc:`Preprocessing <preprocessing>`. Docs page on data preprocessing.
   - :doc:`Training <training>`. Docs page on training and fine-tuning.
 
-.. _nlplib:
+.. _datasetslib:
 
-Using the 🤗 NLP Datasets & Metrics library
+Using the 🤗 Datasets & Metrics library
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 This tutorial demonstrates how to read in datasets from various raw text formats and prepare them for training with 🤗
 Transformers so that you can do the same thing with your own custom datasets. However, we recommend users use the `🤗
-NLP library <https://github.com/huggingface/nlp>`_ for working with the 150+ datasets included in the `hub
+Datasets library <https://github.com/huggingface/datasets>`_ for working with the 150+ datasets included in the `hub
 <https://huggingface.co/datasets>`_, including the three datasets used in this tutorial. As a very brief overview, we
-will show how to use the NLP library to download and prepare the IMDb dataset from the first example, :ref:`seq_imdb`.
+will show how to use the Datasets library to download and prepare the IMDb dataset from the first example,
+:ref:`seq_imdb`.
 
 Start by downloading the dataset:
 
 .. code-block:: python
 
-    from nlp import load_dataset
+    from datasets import load_dataset
     train = load_dataset("imdb", split="train")
 
 Each dataset has multiple columns corresponding to different features. Let's see what our columns are.
@@ -724,5 +725,5 @@ dataset elements.
     >>> {key: val.shape for key, val in train[0].items()})
     {'labels': TensorShape([]), 'input_ids': TensorShape([512]), 'attention_mask': TensorShape([512])}
 
-We now have a fully-prepared dataset. Check out `the 🤗 NLP docs <https://huggingface.co/nlp/processing.html>`_ for a
-more thorough introduction.
+We now have a fully-prepared dataset. Check out `the 🤗 Datasets docs
+<https://huggingface.co/docs/datasets/processing.html>`_ for a more thorough introduction.

docs/source/debugging.rst

@@ -0,0 +1,295 @@
+..
+    Copyright 2021 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.
+
+
+
+Debugging
+=======================================================================================================================
+
+Underflow and Overflow Detection
+-----------------------------------------------------------------------------------------------------------------------
+
+.. note::
+
+   This feature is currently available for PyTorch-only.
+
+.. note::
+
+   This feature can be used with any ``nn.Module``-based model
+
+If you start getting ``loss=NaN`` or the model inhibits some other abnormal behavior due to ``inf`` or ``nan`` in
+activations or weights one needs to discover where the first underflow or overflow happens and what led to it. Luckily
+you can accomplish that easily by activating a special module that will do the detection automatically.
+
+If you're using :class:`~transformers.Trainer`, you just need to add:
+
+.. code-block:: bash
+
+    --debug underflow_overflow
+
+to the normal command line arguments, or pass ``debug="underflow_overflow"`` when creating the
+:class:`~transformers.TrainingArguments` object.
+
+If you're using your own training loop or another Trainer you can accomplish the same with:
+
+.. code-block:: python
+
+    from .debug_utils import DebugUnderflowOverflow
+    debug_overflow = DebugUnderflowOverflow(model)
+
+:class:`~transformers.debug_utils.DebugUnderflowOverflow` inserts hooks into the model that immediately after each
+forward call will test input and output variables and also the corresponding module's weights. As soon as ``inf`` or
+``nan`` is detected in at least one element of the activations or weights, the program will assert and print a report
+like this (this was caught with ``google/mt5-small`` under fp16 mixed precision):
+
+.. code-block::
+
+    Detected inf/nan during batch_number=0
+    Last 21 forward frames:
+    abs min  abs max  metadata
+                      encoder.block.1.layer.1.DenseReluDense.dropout Dropout
+    0.00e+00 2.57e+02 input[0]
+    0.00e+00 2.85e+02 output
+    [...]
+                      encoder.block.2.layer.0 T5LayerSelfAttention
+    6.78e-04 3.15e+03 input[0]
+    2.65e-04 3.42e+03 output[0]
+                 None output[1]
+    2.25e-01 1.00e+04 output[2]
+                      encoder.block.2.layer.1.layer_norm T5LayerNorm
+    8.69e-02 4.18e-01 weight
+    2.65e-04 3.42e+03 input[0]
+    1.79e-06 4.65e+00 output
+                      encoder.block.2.layer.1.DenseReluDense.wi_0 Linear
+    2.17e-07 4.50e+00 weight
+    1.79e-06 4.65e+00 input[0]
+    2.68e-06 3.70e+01 output
+                      encoder.block.2.layer.1.DenseReluDense.wi_1 Linear
+    8.08e-07 2.66e+01 weight
+    1.79e-06 4.65e+00 input[0]
+    1.27e-04 2.37e+02 output
+                      encoder.block.2.layer.1.DenseReluDense.dropout Dropout
+    0.00e+00 8.76e+03 input[0]
+    0.00e+00 9.74e+03 output
+                      encoder.block.2.layer.1.DenseReluDense.wo Linear
+    1.01e-06 6.44e+00 weight
+    0.00e+00 9.74e+03 input[0]
+    3.18e-04 6.27e+04 output
+                      encoder.block.2.layer.1.DenseReluDense T5DenseGatedGeluDense
+    1.79e-06 4.65e+00 input[0]
+    3.18e-04 6.27e+04 output
+                      encoder.block.2.layer.1.dropout Dropout
+    3.18e-04 6.27e+04 input[0]
+    0.00e+00      inf output
+
+The example output has been trimmed in the middle for brevity.
+
+The second column shows the value of the absolute largest element, so if you have a closer look at the last few frames,
+the inputs and outputs were in the range of ``1e4``. So when this training was done under fp16 mixed precision the very
+last step overflowed (since under ``fp16`` the largest number before ``inf`` is ``64e3``). To avoid overflows under
+``fp16`` the activations must remain way below ``1e4``, because ``1e4 * 1e4 = 1e8`` so any matrix multiplication with
+large activations is going to lead to a numerical overflow condition.
+
+At the very start of the trace you can discover at which batch number the problem occurred (here ``Detected inf/nan
+during batch_number=0`` means the problem occurred on the first batch).
+
+Each reported frame starts by declaring the fully qualified entry for the corresponding module this frame is reporting
+for. If we look just at this frame:
+
+.. code-block::
+
+                      encoder.block.2.layer.1.layer_norm T5LayerNorm
+    8.69e-02 4.18e-01 weight
+    2.65e-04 3.42e+03 input[0]
+    1.79e-06 4.65e+00 output
+
+Here, ``encoder.block.2.layer.1.layer_norm`` indicates that it was a layer norm for the first layer, of the second
+block of the encoder. And the specific calls of the ``forward`` is ``T5LayerNorm``.
+
+Let's look at the last few frames of that report:
+
+.. code-block::
+
+        Detected inf/nan during batch_number=0
+        Last 21 forward frames:
+        abs min  abs max  metadata
+        [...]
+                          encoder.block.2.layer.1.DenseReluDense.wi_0 Linear
+        2.17e-07 4.50e+00 weight
+        1.79e-06 4.65e+00 input[0]
+        2.68e-06 3.70e+01 output
+                          encoder.block.2.layer.1.DenseReluDense.wi_1 Linear
+        8.08e-07 2.66e+01 weight
+        1.79e-06 4.65e+00 input[0]
+        1.27e-04 2.37e+02 output
+                          encoder.block.2.layer.1.DenseReluDense.wo Linear
+        1.01e-06 6.44e+00 weight
+        0.00e+00 9.74e+03 input[0]
+        3.18e-04 6.27e+04 output
+                          encoder.block.2.layer.1.DenseReluDense T5DenseGatedGeluDense
+        1.79e-06 4.65e+00 input[0]
+        3.18e-04 6.27e+04 output
+                          encoder.block.2.layer.1.dropout Dropout
+        3.18e-04 6.27e+04 input[0]
+        0.00e+00      inf output
+
+The last frame reports for ``Dropout.forward`` function with the first entry for the only input and the second for the
+only output. You can see that it was called from an attribute ``dropout`` inside ``DenseReluDense`` class. We can see
+that it happened during the first layer, of the 2nd block, during the very first batch. Finally, the absolute largest
+input elements was ``6.27e+04`` and same for the output was ``inf``.
+
+You can see here, that ``T5DenseGatedGeluDense.forward`` resulted in output activations, whose absolute max value was
+around 62.7K, which is very close to fp16's top limit of 64K. In the next frame we have ``Dropout`` which renormalizes
+the weights, after it zeroed some of the elements, which pushes the absolute max value to more than 64K, and we get an
+overlow (``inf``).
+
+As you can see it's the previous frames that we need to look into when the numbers start going into very large for fp16
+numbers.
+
+Let's match the report to the code from ``models/t5/modeling_t5.py``:
+
+.. code-block:: python
+
+    class T5DenseGatedGeluDense(nn.Module):
+        def __init__(self, config):
+            super().__init__()
+            self.wi_0 = nn.Linear(config.d_model, config.d_ff, bias=False)
+            self.wi_1 = nn.Linear(config.d_model, config.d_ff, bias=False)
+            self.wo = nn.Linear(config.d_ff, config.d_model, bias=False)
+            self.dropout = nn.Dropout(config.dropout_rate)
+            self.gelu_act = ACT2FN["gelu_new"]
+
+        def forward(self, hidden_states):
+            hidden_gelu = self.gelu_act(self.wi_0(hidden_states))
+            hidden_linear = self.wi_1(hidden_states)
+            hidden_states = hidden_gelu * hidden_linear
+            hidden_states = self.dropout(hidden_states)
+            hidden_states = self.wo(hidden_states)
+            return hidden_states
+
+Now it's easy to see the ``dropout`` call, and all the previous calls as well.
+
+Since the detection is happening in a forward hook, these reports are printed immediately after each ``forward``
+returns.
+
+Going back to the full report, to act on it and to fix the problem, we need to go a few frames up where the numbers
+started to go up and most likely switch to the ``fp32`` mode here, so that the numbers don't overflow when multiplied
+or summed up. Of course, there might be other solutions. For example, we could turn off ``amp`` temporarily if it's
+enabled, after moving the original ``forward`` into a helper wrapper, like so:
+
+.. code-block:: python
+
+    def _forward(self, hidden_states):
+        hidden_gelu = self.gelu_act(self.wi_0(hidden_states))
+        hidden_linear = self.wi_1(hidden_states)
+        hidden_states = hidden_gelu * hidden_linear
+        hidden_states = self.dropout(hidden_states)
+        hidden_states = self.wo(hidden_states)
+        return hidden_states
+
+    import torch
+    def forward(self, hidden_states):
+        if torch.is_autocast_enabled():
+             with torch.cuda.amp.autocast(enabled=False):
+                 return self._forward(hidden_states)
+         else:
+             return self._forward(hidden_states)
+
+Since the automatic detector only reports on inputs and outputs of full frames, once you know where to look, you may
+want to analyse the intermediary stages of any specific ``forward`` function as well. In such a case you can use the
+``detect_overflow`` helper function to inject the detector where you want it, for example:
+
+.. code-block:: python
+
+    from debug_utils import detect_overflow
+
+    class T5LayerFF(nn.Module):
+        [...]
+        def forward(self, hidden_states):
+            forwarded_states = self.layer_norm(hidden_states)
+            detect_overflow(forwarded_states, "after layer_norm")
+            forwarded_states = self.DenseReluDense(forwarded_states)
+            detect_overflow(forwarded_states, "after DenseReluDense")
+            return hidden_states + self.dropout(forwarded_states)
+
+You can see that we added 2 of these and now we track if ``inf`` or ``nan`` for ``forwarded_states`` was detected
+somewhere in between.
+
+Actually, the detector already reports these because each of the calls in the example above is a `nn.Module``, but
+let's say if you had some local direct calculations this is how you'd do that.
+
+Additionally, if you're instantiating the debugger in your own code, you can adjust the number of frames printed from
+its default, e.g.:
+
+.. code-block:: python
+
+    from .debug_utils import DebugUnderflowOverflow
+    debug_overflow = DebugUnderflowOverflow(model, max_frames_to_save=100)
+
+Specific batch absolute mix and max value tracing
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The same debugging class can be used for per-batch tracing with the underflow/overflow detection feature turned off.
+
+Let's say you want to watch the absolute min and max values for all the ingredients of each ``forward`` call of a given
+batch, and only do that for batches 1 and 3. Then you instantiate this class as:
+
+.. code-block:: python
+
+    debug_overflow = DebugUnderflowOverflow(model, trace_batch_nums=[1,3])
+
+And now full batches 1 and 3 will be traced using the same format as the underflow/overflow detector does.
+
+Batches are 0-indexed.
+
+This is helpful if you know that the program starts misbehaving after a certain batch number, so you can fast-forward
+right to that area. Here is a sample truncated output for such configuration:
+
+.. code-block::
+
+                      *** Starting batch number=1 ***
+    abs min  abs max  metadata
+                      shared Embedding
+    1.01e-06 7.92e+02 weight
+    0.00e+00 2.47e+04 input[0]
+    5.36e-05 7.92e+02 output
+    [...]
+                      decoder.dropout Dropout
+    1.60e-07 2.27e+01 input[0]
+    0.00e+00 2.52e+01 output
+                      decoder T5Stack
+         not a tensor output
+                      lm_head Linear
+    1.01e-06 7.92e+02 weight
+    0.00e+00 1.11e+00 input[0]
+    6.06e-02 8.39e+01 output
+                       T5ForConditionalGeneration
+         not a tensor output
+
+                      *** Starting batch number=3 ***
+    abs min  abs max  metadata
+                      shared Embedding
+    1.01e-06 7.92e+02 weight
+    0.00e+00 2.78e+04 input[0]
+    5.36e-05 7.92e+02 output
+    [...]
+
+Here you will get a huge number of frames dumped - as many as there were forward calls in your model, so it may or may
+not what you want, but sometimes it can be easier to use for debugging purposes than a normal debugger. For example, if
+a problem starts happening at batch number 150. So you can dump traces for batches 149 and 150 and compare where
+numbers started to diverge.
+
+You can also specify the batch number after which to stop the training, with:
+
+.. code-block:: python
+
+    debug_overflow = DebugUnderflowOverflow(model, trace_batch_nums=[1,3], abort_after_batch_num=3)

docs/source/fast_tokenizers.rst

@@ -0,0 +1,62 @@
+Using tokenizers from 🤗 Tokenizers
+=======================================================================================================================
+
+The :class:`~transformers.PreTrainedTokenizerFast` depends on the `tokenizers
+<https://huggingface.co/docs/tokenizers>`__ library. The tokenizers obtained from the 🤗 Tokenizers library can be
+loaded very simply into 🤗 Transformers.
+
+Before getting in the specifics, let's first start by creating a dummy tokenizer in a few lines:
+
+.. code-block::
+
+    >>> from tokenizers import Tokenizer
+    >>> from tokenizers.models import BPE
+    >>> from tokenizers.trainers import BpeTrainer
+    >>> from tokenizers.pre_tokenizers import Whitespace
+
+    >>> tokenizer = Tokenizer(BPE(unk_token="[UNK]"))
+    >>> trainer = BpeTrainer(special_tokens=["[UNK]", "[CLS]", "[SEP]", "[PAD]", "[MASK]"])
+
+    >>> tokenizer.pre_tokenizer = Whitespace()
+    >>> files = [...]
+    >>> tokenizer.train(files, trainer)
+
+We now have a tokenizer trained on the files we defined. We can either continue using it in that runtime, or save it to
+a JSON file for future re-use.
+
+Loading directly from the tokenizer object
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Let's see how to leverage this tokenizer object in the 🤗 Transformers library. The
+:class:`~transformers.PreTrainedTokenizerFast` class allows for easy instantiation, by accepting the instantiated
+`tokenizer` object as an argument:
+
+.. code-block::
+
+    >>> from transformers import PreTrainedTokenizerFast
+
+    >>> fast_tokenizer = PreTrainedTokenizerFast(tokenizer_object=tokenizer)
+
+This object can now be used with all the methods shared by the 🤗 Transformers tokenizers! Head to :doc:`the tokenizer
+page <main_classes/tokenizer>` for more information.
+
+Loading from a JSON file
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+In order to load a tokenizer from a JSON file, let's first start by saving our tokenizer:
+
+.. code-block::
+
+    >>> tokenizer.save("tokenizer.json")
+
+The path to which we saved this file can be passed to the :class:`~transformers.PreTrainedTokenizerFast` initialization
+method using the :obj:`tokenizer_file` parameter:
+
+.. code-block::
+
+    >>> from transformers import PreTrainedTokenizerFast
+
+    >>> fast_tokenizer = PreTrainedTokenizerFast(tokenizer_file="tokenizer.json")
+
+This object can now be used with all the methods shared by the 🤗 Transformers tokenizers! Head to :doc:`the tokenizer
+page <main_classes/tokenizer>` for more information.

docs/source/glossary.rst

@@ -55,6 +55,12 @@ Input IDs
 The input ids are often the only required parameters to be passed to the model as input. *They are token indices,
 numerical representations of tokens building the sequences that will be used as input by the model*.
 
+.. raw:: html
+
+   <iframe width="560" height="315" src="https://www.youtube.com/embed/VFp38yj8h3A" title="YouTube video player"
+   frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope;
+   picture-in-picture" allowfullscreen></iframe>
+
 Each tokenizer works differently but the underlying mechanism remains the same. Here's an example using the BERT
 tokenizer, which is a `WordPiece <https://arxiv.org/pdf/1609.08144.pdf>`__ tokenizer:
 
@@ -120,8 +126,15 @@ because this is the way a :class:`~transformers.BertModel` is going to expect it
 Attention mask
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
-The attention mask is an optional argument used when batching sequences together. This argument indicates to the model
-which tokens should be attended to, and which should not.
+The attention mask is an optional argument used when batching sequences together.
+
+.. raw:: html
+
+   <iframe width="560" height="315" src="https://www.youtube.com/embed/M6adb1j2jPI" title="YouTube video player"
+   frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope;
+   picture-in-picture" allowfullscreen></iframe>
+
+This argument indicates to the model which tokens should be attended to, and which should not.
 
 For example, consider these two sequences:
 
@@ -175,14 +188,21 @@ in the dictionary returned by the tokenizer under the key "attention_mask":
 Token Type IDs
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
-Some models' purpose is to do sequence classification or question answering. These require two different sequences to
-be joined in a single "input_ids" entry, which usually is performed with the help of special tokens, such as the
-classifier (``[CLS]``) and separator (``[SEP]``) tokens. For example, the BERT model builds its two sequence input as
-such:
+Some models' purpose is to do classification on pairs of sentences or question answering.
+
+.. raw:: html
+
+   <iframe width="560" height="315" src="https://www.youtube.com/embed/0u3ioSwev3s" title="YouTube video player"
+   frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope;
+   picture-in-picture" allowfullscreen></iframe>
+
+These require two different sequences to be joined in a single "input_ids" entry, which usually is performed with the
+help of special tokens, such as the classifier (``[CLS]``) and separator (``[SEP]``) tokens. For example, the BERT
+model builds its two sequence input as such:
 
 .. code-block::
 
-   >>> # [CLS] SEQUENCE_A [SEP] SEQUENCE_B [SEP]
+    >>> # [CLS] SEQUENCE_A [SEP] SEQUENCE_B [SEP]
 
 We can use our tokenizer to automatically generate such a sentence by passing the two sequences to ``tokenizer`` as two
 arguments (and not a list, like before) like this:

docs/source/index.rst

@@ -1,14 +1,25 @@
 Transformers
 =======================================================================================================================
 
-State-of-the-art Natural Language Processing for Pytorch and TensorFlow 2.0.
+State-of-the-art Natural Language Processing for Jax, Pytorch and TensorFlow
 
 🤗 Transformers (formerly known as `pytorch-transformers` and `pytorch-pretrained-bert`) provides general-purpose
 architectures (BERT, GPT-2, RoBERTa, XLM, DistilBert, XLNet...) for Natural Language Understanding (NLU) and Natural
-Language Generation (NLG) with over 32+ pretrained models in 100+ languages and deep interoperability between
-TensorFlow 2.0 and PyTorch.
+Language Generation (NLG) with over 32+ pretrained models in 100+ languages and deep interoperability between Jax,
+PyTorch and TensorFlow.
 
-This is the documentation of our repository `transformers <https://github.com/huggingface/transformers>`_.
+This is the documentation of our repository `transformers <https://github.com/huggingface/transformers>`__. You can
+also follow our `online course <https://huggingface.co/course>`__ that teaches how to use this library, as well as the
+other libraries developed by Hugging Face and the Hub.
+
+If you are looking for custom support from the Hugging Face team
+-----------------------------------------------------------------------------------------------------------------------
+
+.. raw:: html
+
+    <a target="_blank" href="https://huggingface.co/support">
+        <img alt="HuggingFace Expert Acceleration Program" src="https://huggingface.co/front/thumbnails/support.png" style="max-width: 600px; border: 1px solid #eee; border-radius: 4px; box-shadow: 0 1px 2px 0 rgba(0, 0, 0, 0.05);">
+    </a><br>
 
 Features
 -----------------------------------------------------------------------------------------------------------------------
@@ -22,7 +33,7 @@ State-of-the-art NLP for everyone:
 - Hands-on practitioners
 - AI/ML/NLP teachers and educators
 
-.. 
+..
     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
@@ -43,11 +54,11 @@ Lower compute costs, smaller carbon footprint:
 Choose the right framework for every part of a model's lifetime:
 
 - Train state-of-the-art models in 3 lines of code
-- Deep interoperability between TensorFlow 2.0 and PyTorch models
-- Move a single model between TF2.0/PyTorch frameworks at will
+- Deep interoperability between Jax, Pytorch and TensorFlow models
+- Move a single model between Jax/PyTorch/TensorFlow frameworks at will
 - Seamlessly pick the right framework for training, evaluation, production
 
-Experimental support for Flax with a few models right now, expected to grow in the coming months.
+The support for Jax is still experimental (with a few models right now), expect to see it grow in the coming months!
 
 `All the model checkpoints <https://huggingface.co/models>`__ are seamlessly integrated from the huggingface.co `model
 hub <https://huggingface.co>`__ where they are uploaded directly by `users <https://huggingface.co/users>`__ and
@@ -74,8 +85,11 @@ The documentation is organized in five parts:
     - **MODELS** for the classes and functions related to each model implemented in the library.
     - **INTERNAL HELPERS** for the classes and functions we use internally.
 
-The library currently contains PyTorch, Tensorflow and Flax implementations, pretrained model weights, usage scripts
-and conversion utilities for the following models:
+The library currently contains Jax, PyTorch and Tensorflow implementations, pretrained model weights, usage scripts and
+conversion utilities for the following models.
+
+Supported models
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 ..
     This list is updated automatically from the README with `make fix-copies`. Do not update manually!
@@ -97,139 +111,188 @@ and conversion utilities for the following models:
 5. :doc:`BERT For Sequence Generation <model_doc/bertgeneration>` (from Google) released with the paper `Leveraging
    Pre-trained Checkpoints for Sequence Generation Tasks <https://arxiv.org/abs/1907.12461>`__ by Sascha Rothe, Shashi
    Narayan, Aliaksei Severyn.
-6. :doc:`Blenderbot <model_doc/blenderbot>` (from Facebook) released with the paper `Recipes for building an
+6. :doc:`BigBird-RoBERTa <model_doc/bigbird>` (from Google Research) released with the paper `Big Bird: Transformers
+   for Longer Sequences <https://arxiv.org/abs/2007.14062>`__ by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua
+   Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
+7. :doc:`BigBird-Pegasus <model_doc/bigbird_pegasus>` (from Google Research) released with the paper `Big Bird:
+   Transformers for Longer Sequences <https://arxiv.org/abs/2007.14062>`__ by Manzil Zaheer, Guru Guruganesh, Avinava
+   Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
+8. :doc:`Blenderbot <model_doc/blenderbot>` (from Facebook) released with the paper `Recipes for building an
    open-domain chatbot <https://arxiv.org/abs/2004.13637>`__ by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary
    Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
-7. :doc:`BlenderbotSmall <model_doc/blenderbot_small>` (from Facebook) released with the paper `Recipes for building an
+9. :doc:`BlenderbotSmall <model_doc/blenderbot_small>` (from Facebook) released with the paper `Recipes for building an
    open-domain chatbot <https://arxiv.org/abs/2004.13637>`__ by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary
    Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
-8. :doc:`BORT <model_doc/bort>` (from Alexa) released with the paper `Optimal Subarchitecture Extraction For BERT
-   <https://arxiv.org/abs/2010.10499>`__ by Adrian de Wynter and Daniel J. Perry.
-9. :doc:`CamemBERT <model_doc/camembert>` (from Inria/Facebook/Sorbonne) released with the paper `CamemBERT: a Tasty
-   French Language Model <https://arxiv.org/abs/1911.03894>`__ by Louis Martin*, Benjamin Muller*, Pedro Javier Ortiz
-   Suárez*, Yoann Dupont, Laurent Romary, Éric Villemonte de la Clergerie, Djamé Seddah and Benoît Sagot.
-10. :doc:`ConvBERT <model_doc/convbert>` (from YituTech) released with the paper `ConvBERT: Improving BERT with
+10. :doc:`BORT <model_doc/bort>` (from Alexa) released with the paper `Optimal Subarchitecture Extraction For BERT
+    <https://arxiv.org/abs/2010.10499>`__ by Adrian de Wynter and Daniel J. Perry.
+11. :doc:`ByT5 <model_doc/byt5>` (from Google Research) released with the paper `ByT5: Towards a token-free future with
+    pre-trained byte-to-byte models <https://arxiv.org/abs/2105.13626>`__ by Linting Xue, Aditya Barua, Noah Constant,
+    Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, Colin Raffel.
+12. :doc:`CamemBERT <model_doc/camembert>` (from Inria/Facebook/Sorbonne) released with the paper `CamemBERT: a Tasty
+    French Language Model <https://arxiv.org/abs/1911.03894>`__ by Louis Martin*, Benjamin Muller*, Pedro Javier Ortiz
+    Suárez*, Yoann Dupont, Laurent Romary, Éric Villemonte de la Clergerie, Djamé Seddah and Benoît Sagot.
+13. :doc:`CLIP <model_doc/clip>` from (OpenAI) released with the paper `Learning Transferable Visual Models From
+    Natural Language Supervision <https://arxiv.org/abs/2103.00020>`__ by Alec Radford, Jong Wook Kim, Chris Hallacy,
+    Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen
+    Krueger, Ilya Sutskever.
+14. :doc:`ConvBERT <model_doc/convbert>` (from YituTech) released with the paper `ConvBERT: Improving BERT with
     Span-based Dynamic Convolution <https://arxiv.org/abs/2008.02496>`__ by Zihang Jiang, Weihao Yu, Daquan Zhou,
     Yunpeng Chen, Jiashi Feng, Shuicheng Yan.
-11. :doc:`CTRL <model_doc/ctrl>` (from Salesforce) released with the paper `CTRL: A Conditional Transformer Language
+15. :doc:`CPM <model_doc/cpm>` (from Tsinghua University) released with the paper `CPM: A Large-scale Generative
+    Chinese Pre-trained Language Model <https://arxiv.org/abs/2012.00413>`__ by Zhengyan Zhang, Xu Han, Hao Zhou, Pei
+    Ke, Yuxian Gu, Deming Ye, Yujia Qin, Yusheng Su, Haozhe Ji, Jian Guan, Fanchao Qi, Xiaozhi Wang, Yanan Zheng,
+    Guoyang Zeng, Huanqi Cao, Shengqi Chen, Daixuan Li, Zhenbo Sun, Zhiyuan Liu, Minlie Huang, Wentao Han, Jie Tang,
+    Juanzi Li, Xiaoyan Zhu, Maosong Sun.
+16. :doc:`CTRL <model_doc/ctrl>` (from Salesforce) released with the paper `CTRL: A Conditional Transformer Language
     Model for Controllable Generation <https://arxiv.org/abs/1909.05858>`__ by Nitish Shirish Keskar*, Bryan McCann*,
     Lav R. Varshney, Caiming Xiong and Richard Socher.
-12. :doc:`DeBERTa <model_doc/deberta>` (from Microsoft) released with the paper `DeBERTa: Decoding-enhanced BERT with
+17. :doc:`DeBERTa <model_doc/deberta>` (from Microsoft) released with the paper `DeBERTa: Decoding-enhanced BERT with
     Disentangled Attention <https://arxiv.org/abs/2006.03654>`__ by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu
     Chen.
-13. :doc:`DeBERTa-v2 <model_doc/deberta_v2>` (from Microsoft) released with the paper `DeBERTa: Decoding-enhanced BERT
+18. :doc:`DeBERTa-v2 <model_doc/deberta_v2>` (from Microsoft) released with the paper `DeBERTa: Decoding-enhanced BERT
     with Disentangled Attention <https://arxiv.org/abs/2006.03654>`__ by Pengcheng He, Xiaodong Liu, Jianfeng Gao,
     Weizhu Chen.
-14. :doc:`DialoGPT <model_doc/dialogpt>` (from Microsoft Research) released with the paper `DialoGPT: Large-Scale
+19. :doc:`DeiT <model_doc/deit>` (from Facebook) released with the paper `Training data-efficient image transformers &
+    distillation through attention <https://arxiv.org/abs/2012.12877>`__ by Hugo Touvron, Matthieu Cord, Matthijs
+    Douze, Francisco Massa, Alexandre Sablayrolles, Hervé Jégou.
+20. :doc:`DETR <model_doc/detr>` (from Facebook) released with the paper `End-to-End Object Detection with Transformers
+    <https://arxiv.org/abs/2005.12872>`__ by Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier,
+    Alexander Kirillov, Sergey Zagoruyko.
+21. :doc:`DialoGPT <model_doc/dialogpt>` (from Microsoft Research) released with the paper `DialoGPT: Large-Scale
     Generative Pre-training for Conversational Response Generation <https://arxiv.org/abs/1911.00536>`__ by Yizhe
     Zhang, Siqi Sun, Michel Galley, Yen-Chun Chen, Chris Brockett, Xiang Gao, Jianfeng Gao, Jingjing Liu, Bill Dolan.
-15. :doc:`DistilBERT <model_doc/distilbert>` (from HuggingFace), released together with the paper `DistilBERT, a
+22. :doc:`DistilBERT <model_doc/distilbert>` (from HuggingFace), released together with the paper `DistilBERT, a
     distilled version of BERT: smaller, faster, cheaper and lighter <https://arxiv.org/abs/1910.01108>`__ by Victor
     Sanh, Lysandre Debut and Thomas Wolf. The same method has been applied to compress GPT2 into `DistilGPT2
     <https://github.com/huggingface/transformers/tree/master/examples/distillation>`__, RoBERTa into `DistilRoBERTa
     <https://github.com/huggingface/transformers/tree/master/examples/distillation>`__, Multilingual BERT into
     `DistilmBERT <https://github.com/huggingface/transformers/tree/master/examples/distillation>`__ and a German
     version of DistilBERT.
-16. :doc:`DPR <model_doc/dpr>` (from Facebook) released with the paper `Dense Passage Retrieval for Open-Domain
+23. :doc:`DPR <model_doc/dpr>` (from Facebook) released with the paper `Dense Passage Retrieval for Open-Domain
     Question Answering <https://arxiv.org/abs/2004.04906>`__ by Vladimir Karpukhin, Barlas Oğuz, Sewon Min, Patrick
     Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih.
-17. :doc:`ELECTRA <model_doc/electra>` (from Google Research/Stanford University) released with the paper `ELECTRA:
+24. :doc:`ELECTRA <model_doc/electra>` (from Google Research/Stanford University) released with the paper `ELECTRA:
     Pre-training text encoders as discriminators rather than generators <https://arxiv.org/abs/2003.10555>`__ by Kevin
     Clark, Minh-Thang Luong, Quoc V. Le, Christopher D. Manning.
-18. :doc:`FlauBERT <model_doc/flaubert>` (from CNRS) released with the paper `FlauBERT: Unsupervised Language Model
+25. :doc:`FlauBERT <model_doc/flaubert>` (from CNRS) released with the paper `FlauBERT: Unsupervised Language Model
     Pre-training for French <https://arxiv.org/abs/1912.05372>`__ by Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne,
     Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab.
-19. :doc:`Funnel Transformer <model_doc/funnel>` (from CMU/Google Brain) released with the paper `Funnel-Transformer:
+26. :doc:`Funnel Transformer <model_doc/funnel>` (from CMU/Google Brain) released with the paper `Funnel-Transformer:
     Filtering out Sequential Redundancy for Efficient Language Processing <https://arxiv.org/abs/2006.03236>`__ by
     Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le.
-20. :doc:`GPT <model_doc/gpt>` (from OpenAI) released with the paper `Improving Language Understanding by Generative
+27. :doc:`GPT <model_doc/gpt>` (from OpenAI) released with the paper `Improving Language Understanding by Generative
     Pre-Training <https://blog.openai.com/language-unsupervised/>`__ by Alec Radford, Karthik Narasimhan, Tim Salimans
     and Ilya Sutskever.
-21. :doc:`GPT-2 <model_doc/gpt2>` (from OpenAI) released with the paper `Language Models are Unsupervised Multitask
+28. :doc:`GPT-2 <model_doc/gpt2>` (from OpenAI) released with the paper `Language Models are Unsupervised Multitask
     Learners <https://blog.openai.com/better-language-models/>`__ by Alec Radford*, Jeffrey Wu*, Rewon Child, David
     Luan, Dario Amodei** and Ilya Sutskever**.
-22. :doc:`I-BERT <model_doc/ibert>` (from Berkeley) released with the paper `I-BERT: Integer-only BERT Quantization
+29. :doc:`GPT Neo <model_doc/gpt_neo>` (from EleutherAI) released in the repository `EleutherAI/gpt-neo
+    <https://github.com/EleutherAI/gpt-neo>`__ by Sid Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy.
+30. :doc:`Hubert <model_doc/hubert>` (from Facebook) released with the paper `HuBERT: Self-Supervised Speech
+    Representation Learning by Masked Prediction of Hidden Units <https://arxiv.org/abs/2106.07447>`__ by Wei-Ning Hsu,
+    Benjamin Bolte, Yao-Hung Hubert Tsai, Kushal Lakhotia, Ruslan Salakhutdinov, Abdelrahman Mohamed.
+31. :doc:`I-BERT <model_doc/ibert>` (from Berkeley) released with the paper `I-BERT: Integer-only BERT Quantization
     <https://arxiv.org/abs/2101.01321>`__ by Sehoon Kim, Amir Gholami, Zhewei Yao, Michael W. Mahoney, Kurt Keutzer
-23. :doc:`LayoutLM <model_doc/layoutlm>` (from Microsoft Research Asia) released with the paper `LayoutLM: Pre-training
+32. :doc:`LayoutLM <model_doc/layoutlm>` (from Microsoft Research Asia) released with the paper `LayoutLM: Pre-training
     of Text and Layout for Document Image Understanding <https://arxiv.org/abs/1912.13318>`__ by Yiheng Xu, Minghao Li,
     Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
-24. :doc:`LED <model_doc/led>` (from AllenAI) released with the paper `Longformer: The Long-Document Transformer
+33. :doc:`LED <model_doc/led>` (from AllenAI) released with the paper `Longformer: The Long-Document Transformer
     <https://arxiv.org/abs/2004.05150>`__ by Iz Beltagy, Matthew E. Peters, Arman Cohan.
-25. :doc:`Longformer <model_doc/longformer>` (from AllenAI) released with the paper `Longformer: The Long-Document
+34. :doc:`Longformer <model_doc/longformer>` (from AllenAI) released with the paper `Longformer: The Long-Document
     Transformer <https://arxiv.org/abs/2004.05150>`__ by Iz Beltagy, Matthew E. Peters, Arman Cohan.
-26. :doc:`LXMERT <model_doc/lxmert>` (from UNC Chapel Hill) released with the paper `LXMERT: Learning Cross-Modality
+35. :doc:`LUKE <model_doc/luke>` (from Studio Ousia) released with the paper `LUKE: Deep Contextualized Entity
+    Representations with Entity-aware Self-attention <https://arxiv.org/abs/2010.01057>`__ by Ikuya Yamada, Akari Asai,
+    Hiroyuki Shindo, Hideaki Takeda, Yuji Matsumoto.
+36. :doc:`LXMERT <model_doc/lxmert>` (from UNC Chapel Hill) released with the paper `LXMERT: Learning Cross-Modality
     Encoder Representations from Transformers for Open-Domain Question Answering <https://arxiv.org/abs/1908.07490>`__
     by Hao Tan and Mohit Bansal.
-27. :doc:`M2M100 <model_doc/m2m_100>` (from Facebook) released with the paper `Beyond English-Centric Multilingual
+37. :doc:`M2M100 <model_doc/m2m_100>` (from Facebook) released with the paper `Beyond English-Centric Multilingual
     Machine Translation <https://arxiv.org/abs/2010.11125>`__ by by Angela Fan, Shruti Bhosale, Holger Schwenk, Zhiyi
     Ma, Ahmed El-Kishky, Siddharth Goyal, Mandeep Baines, Onur Celebi, Guillaume Wenzek, Vishrav Chaudhary, Naman
     Goyal, Tom Birch, Vitaliy Liptchinsky, Sergey Edunov, Edouard Grave, Michael Auli, Armand Joulin.
-28. :doc:`MarianMT <model_doc/marian>` Machine translation models trained using `OPUS <http://opus.nlpl.eu/>`__ data by
+38. :doc:`MarianMT <model_doc/marian>` Machine translation models trained using `OPUS <http://opus.nlpl.eu/>`__ data by
     Jörg Tiedemann. The `Marian Framework <https://marian-nmt.github.io/>`__ is being developed by the Microsoft
     Translator Team.
-29. :doc:`MBart <model_doc/mbart>` (from Facebook) released with the paper `Multilingual Denoising Pre-training for
+39. :doc:`MBart <model_doc/mbart>` (from Facebook) released with the paper `Multilingual Denoising Pre-training for
     Neural Machine Translation <https://arxiv.org/abs/2001.08210>`__ by Yinhan Liu, Jiatao Gu, Naman Goyal, Xian Li,
     Sergey Edunov, Marjan Ghazvininejad, Mike Lewis, Luke Zettlemoyer.
-30. :doc:`MBart-50 <model_doc/mbart>` (from Facebook) released with the paper `Multilingual Translation with Extensible
+40. :doc:`MBart-50 <model_doc/mbart>` (from Facebook) released with the paper `Multilingual Translation with Extensible
     Multilingual Pretraining and Finetuning <https://arxiv.org/abs/2008.00401>`__ by Yuqing Tang, Chau Tran, Xian Li,
     Peng-Jen Chen, Naman Goyal, Vishrav Chaudhary, Jiatao Gu, Angela Fan.
-31. :doc:`MPNet <model_doc/mpnet>` (from Microsoft Research) released with the paper `MPNet: Masked and Permuted
+41. :doc:`Megatron-BERT <model_doc/megatron_bert>` (from NVIDIA) released with the paper `Megatron-LM: Training
+    Multi-Billion Parameter Language Models Using Model Parallelism <https://arxiv.org/abs/1909.08053>`__ by Mohammad
+    Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
+42. :doc:`Megatron-GPT2 <model_doc/megatron_gpt2>` (from NVIDIA) released with the paper `Megatron-LM: Training
+    Multi-Billion Parameter Language Models Using Model Parallelism <https://arxiv.org/abs/1909.08053>`__ by Mohammad
+    Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
+43. :doc:`MPNet <model_doc/mpnet>` (from Microsoft Research) released with the paper `MPNet: Masked and Permuted
     Pre-training for Language Understanding <https://arxiv.org/abs/2004.09297>`__ by Kaitao Song, Xu Tan, Tao Qin,
     Jianfeng Lu, Tie-Yan Liu.
-32. :doc:`MT5 <model_doc/mt5>` (from Google AI) released with the paper `mT5: A massively multilingual pre-trained
+44. :doc:`MT5 <model_doc/mt5>` (from Google AI) released with the paper `mT5: A massively multilingual pre-trained
     text-to-text transformer <https://arxiv.org/abs/2010.11934>`__ by Linting Xue, Noah Constant, Adam Roberts, Mihir
     Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel.
-33. :doc:`Pegasus <model_doc/pegasus>` (from Google) released with the paper `PEGASUS: Pre-training with Extracted
+45. :doc:`Pegasus <model_doc/pegasus>` (from Google) released with the paper `PEGASUS: Pre-training with Extracted
     Gap-sentences for Abstractive Summarization <https://arxiv.org/abs/1912.08777>`__> by Jingqing Zhang, Yao Zhao,
     Mohammad Saleh and Peter J. Liu.
-34. :doc:`ProphetNet <model_doc/prophetnet>` (from Microsoft Research) released with the paper `ProphetNet: Predicting
+46. :doc:`ProphetNet <model_doc/prophetnet>` (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.
-35. :doc:`Reformer <model_doc/reformer>` (from Google Research) released with the paper `Reformer: The Efficient
+47. :doc:`Reformer <model_doc/reformer>` (from Google Research) released with the paper `Reformer: The Efficient
     Transformer <https://arxiv.org/abs/2001.04451>`__ by Nikita Kitaev, Łukasz Kaiser, Anselm Levskaya.
-36. :doc:`RoBERTa <model_doc/roberta>` (from Facebook), released together with the paper a `Robustly Optimized BERT
+48. :doc:`RoBERTa <model_doc/roberta>` (from Facebook), released together with the paper a `Robustly Optimized BERT
     Pretraining Approach <https://arxiv.org/abs/1907.11692>`__ by Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar
     Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov.
-37. :doc:`SpeechToTextTransformer <model_doc/speech_to_text>` (from Facebook), released together with the paper
+49. :doc:`RoFormer <model_doc/roformer>` (from ZhuiyiTechnology), released together with the paper a `RoFormer:
+    Enhanced Transformer with Rotary Position Embedding <https://arxiv.org/pdf/2104.09864v1.pdf>`__ by Jianlin Su and
+    Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu.
+50. :doc:`SpeechToTextTransformer <model_doc/speech_to_text>` (from Facebook), released together with the paper
     `fairseq S2T: Fast Speech-to-Text Modeling with fairseq <https://arxiv.org/abs/2010.05171>`__ by Changhan Wang, Yun
     Tang, Xutai Ma, Anne Wu, Dmytro Okhonko, Juan Pino.
-38. :doc:`SqueezeBert <model_doc/squeezebert>` released with the paper `SqueezeBERT: What can computer vision teach NLP
+51. :doc:`SqueezeBert <model_doc/squeezebert>` 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.
-39. :doc:`T5 <model_doc/t5>` (from Google AI) released with the paper `Exploring the Limits of Transfer Learning with a
+52. :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.
-40. :doc:`TAPAS <model_doc/tapas>` (from Google AI) released with the paper `TAPAS: Weakly Supervised Table Parsing via
+53. :doc:`TAPAS <model_doc/tapas>` (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.
-41. :doc:`Transformer-XL <model_doc/transformerxl>` (from Google/CMU) released with the paper `Transformer-XL:
+54. :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.
-42. :doc:`Wav2Vec2 <model_doc/wav2vec2>` (from Facebook AI) released with the paper `wav2vec 2.0: A Framework for
+55. :doc:`Vision Transformer (ViT) <model_doc/vit>` (from Google AI) released with the paper `An Image is Worth 16x16
+    Words: Transformers for Image Recognition at Scale <https://arxiv.org/abs/2010.11929>`__ by Alexey Dosovitskiy,
+    Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias
+    Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby.
+56. :doc:`VisualBERT <model_doc/visual_bert>` (from UCLA NLP) released with the paper `VisualBERT: A Simple and
+    Performant Baseline for Vision and Language <https://arxiv.org/pdf/1908.03557>`__ by Liunian Harold Li, Mark
+    Yatskar, Da Yin, Cho-Jui Hsieh, Kai-Wei Chang.
+57. :doc:`Wav2Vec2 <model_doc/wav2vec2>` (from Facebook AI) released with the paper `wav2vec 2.0: A Framework for
     Self-Supervised Learning of Speech Representations <https://arxiv.org/abs/2006.11477>`__ by Alexei Baevski, Henry
     Zhou, Abdelrahman Mohamed, Michael Auli.
-43. :doc:`XLM <model_doc/xlm>` (from Facebook) released together with the paper `Cross-lingual Language Model
+58. :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.
-44. :doc:`XLM-ProphetNet <model_doc/xlmprophetnet>` (from Microsoft Research) released with the paper `ProphetNet:
+59. :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.
-45. :doc:`XLM-RoBERTa <model_doc/xlmroberta>` (from Facebook AI), released together with the paper `Unsupervised
+60. :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.
-46. :doc:`XLNet <model_doc/xlnet>` (from Google/CMU) released with the paper `​XLNet: Generalized Autoregressive
+61. :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.
-47. :doc:`XLSR-Wav2Vec2 <model_doc/xlsr_wav2vec2>` (from Facebook AI) released with the paper `Unsupervised
+62. :doc:`XLSR-Wav2Vec2 <model_doc/xlsr_wav2vec2>` (from Facebook AI) released with the paper `Unsupervised
     Cross-Lingual Representation Learning For Speech Recognition <https://arxiv.org/abs/2006.13979>`__ by Alexis
     Conneau, Alexei Baevski, Ronan Collobert, Abdelrahman Mohamed, Michael Auli.
 
 
-.. _bigtable:
+Supported frameworks
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 The table below represents the current support in the library for each of those models, whether they have a Python
-tokenizer (called "slow"). A "fast" tokenizer backed by the 🤗 Tokenizers library, whether they have support in PyTorch,
-TensorFlow and/or Flax.
+tokenizer (called "slow"). A "fast" tokenizer backed by the 🤗 Tokenizers library, whether they have support in Jax (via
+Flax), PyTorch, and/or TensorFlow.
 
 ..
     This table is updated automatically from the auto modules with `make fix-copies`. Do not update manually!
@@ -241,31 +304,41 @@ TensorFlow and/or Flax.
 +=============================+================+================+=================+====================+==============+
 |           ALBERT            |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|            BART             |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
+|            BART             |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |            BERT             |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |       Bert Generation       |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
+|           BigBird           |       ✅       |       ✅       |       ✅        |         ❌         |      ✅      |
++-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
+|       BigBirdPegasus        |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
++-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |         Blenderbot          |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |       BlenderbotSmall       |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
+|            CLIP             |       ✅       |       ✅       |       ✅        |         ❌         |      ✅      |
++-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |            CTRL             |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |          CamemBERT          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |          ConvBERT           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
+|            DETR             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
++-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |             DPR             |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|           DeBERTa           |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
+|           DeBERTa           |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |         DeBERTa-v2          |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
+|            DeiT             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
++-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |         DistilBERT          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|           ELECTRA           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
+|           ELECTRA           |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |       Encoder decoder       |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
@@ -275,13 +348,19 @@ TensorFlow and/or Flax.
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |     Funnel Transformer      |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
+|           GPT Neo           |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
++-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
+|           Hubert            |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
++-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |           I-BERT            |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |             LED             |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
+|            LUKE             |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
++-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |           LXMERT            |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|          LayoutLM           |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
+|          LayoutLM           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |         Longformer          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
@@ -291,11 +370,13 @@ TensorFlow and/or Flax.
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |           Marian            |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
+|        MegatronBert         |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
++-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |         MobileBERT          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |         OpenAI GPT          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|        OpenAI GPT-2         |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
+|        OpenAI GPT-2         |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |           Pegasus           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
@@ -309,17 +390,23 @@ TensorFlow and/or Flax.
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |           RoBERTa           |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
+|          RoFormer           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
++-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |         Speech2Text         |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |         SqueezeBERT         |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|             T5              |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
+|             T5              |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |            TAPAS            |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |       Transformer-XL        |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|          Wav2Vec2           |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
+|             ViT             |       ❌       |       ❌       |       ✅        |         ❌         |      ✅      |
++-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
+|         VisualBert          |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
++-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
+|          Wav2Vec2           |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |             XLM             |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
@@ -361,14 +448,19 @@ TensorFlow and/or Flax.
 
     pretrained_models
     examples
+    troubleshooting
     custom_datasets
     notebooks
+    sagemaker
     community
     converting_tensorflow_models
     migration
     contributing
     add_new_model
+    fast_tokenizers
+    performance
     testing
+    debugging
     serialization
 
 .. toctree::
@@ -385,6 +477,7 @@ TensorFlow and/or Flax.
 
     main_classes/callback
     main_classes/configuration
+    main_classes/data_collator
     main_classes/logging
     main_classes/model
     main_classes/optimizer_schedules
@@ -393,6 +486,7 @@ TensorFlow and/or Flax.
     main_classes/processors
     main_classes/tokenizer
     main_classes/trainer
+    main_classes/deepspeed
     main_classes/feature_extractor
 
 .. toctree::
@@ -406,14 +500,22 @@ TensorFlow and/or Flax.
     model_doc/bert
     model_doc/bertweet
     model_doc/bertgeneration
+    model_doc/bert_japanese
+    model_doc/bigbird
+    model_doc/bigbird_pegasus
     model_doc/blenderbot
     model_doc/blenderbot_small
     model_doc/bort
+    model_doc/byt5
     model_doc/camembert
+    model_doc/clip
     model_doc/convbert
+    model_doc/cpm
     model_doc/ctrl
     model_doc/deberta
     model_doc/deberta_v2
+    model_doc/deit
+    model_doc/detr
     model_doc/dialogpt
     model_doc/distilbert
     model_doc/dpr
@@ -427,15 +529,20 @@ TensorFlow and/or Flax.
     model_doc/layoutlm
     model_doc/led
     model_doc/longformer
+    model_doc/luke
     model_doc/lxmert
     model_doc/marian
     model_doc/m2m_100
     model_doc/mbart
+    model_doc/megatron_bert
+    model_doc/megatron_gpt2
     model_doc/mobilebert
     model_doc/mpnet
     model_doc/mt5
     model_doc/gpt
     model_doc/gpt2
+    model_doc/gpt_neo
+    model_doc/hubert
     model_doc/pegasus
     model_doc/phobert
     model_doc/prophetnet
@@ -443,11 +550,14 @@ TensorFlow and/or Flax.
     model_doc/reformer
     model_doc/retribert
     model_doc/roberta
+    model_doc/roformer
     model_doc/speech_to_text
     model_doc/squeezebert
     model_doc/t5
     model_doc/tapas
     model_doc/transformerxl
+    model_doc/vit
+    model_doc/visual_bert
     model_doc/wav2vec2
     model_doc/xlm
     model_doc/xlmprophetnet

docs/source/installation.md

@@ -107,7 +107,7 @@ This command performs a magical link between the folder you cloned the repositor
 ```
 now this editable install will reside where you clone the folder to, e.g. `~/transformers/` and python will search it too.
 
-Do note that you have to keep that `transformers` folder around and not delete it to continue using the  `transfomers` library.
+Do note that you have to keep that `transformers` folder around and not delete it to continue using the  `transformers` library.
 
 Now, let's get to the real benefit of this installation approach. Say, you saw some new feature has been just committed into `master`. If you have already performed all the steps above, to update your transformers to include all the latest commits, all you need to do is to `cd` into that cloned repository folder and update the clone to the latest version:
 
@@ -149,12 +149,6 @@ So if you don't have any specific environment variable set, the cache directory
 (``PYTORCH_TRANSFORMERS_CACHE`` or ``PYTORCH_PRETRAINED_BERT_CACHE``), those will be used if there is no shell
 environment variable for ``TRANSFORMERS_CACHE``.
 
-### Note on model downloads (Continuous Integration or large-scale deployments)
-
-If you expect to be downloading large volumes of models (more than 1,000) from our hosted bucket (for instance through
-your CI setup, or a large-scale production deployment), please cache the model files on your end. It will be way
-faster, and cheaper. Feel free to contact us privately if you need any help.
-
 ### Offline mode
 
 It's possible to run 🤗 Transformers in a firewalled or a no-network environment.
@@ -168,17 +162,29 @@ Here is an example of how this can be used on a filesystem that is shared betwee
 On the instance with the normal network run your program which will download and cache models (and optionally datasets if you use 🤗 Datasets). For example:
 
 ```
-python examples/seq2seq/run_translation.py --model_name_or_path t5-small --dataset_name wmt16 --dataset_config ro-en ...
+python examples/pytorch/translation/run_translation.py --model_name_or_path t5-small --dataset_name wmt16 --dataset_config ro-en ...
 ```
 
 and then with the same filesystem you can now run the same program on a firewalled instance:
 ```
 HF_DATASETS_OFFLINE=1 TRANSFORMERS_OFFLINE=1 \
-python examples/seq2seq/run_translation.py --model_name_or_path t5-small --dataset_name wmt16 --dataset_config ro-en ...
+python examples/pytorch/translation/run_translation.py --model_name_or_path t5-small --dataset_name wmt16 --dataset_config ro-en ...
 ```
 and it should succeed without any hanging waiting to timeout.
 
+#### Fetching models and tokenizers to use offline
 
+When running a script the first time like mentioned above, the downloaded files will be cached for future reuse. 
+However, it is also possible to download files and point to their local path instead.
+
+Downloading files can be done through the Web Interface by clicking on the "Download" button, but it can also be handled
+programmatically using the `huggingface_hub` library that is a dependency to `transformers`:
+
+- Using `snapshot_download` to download an entire repository
+- Using `hf_hub_download` to download a specific file
+
+See the reference for these methods in the huggingface_hub
+[documentation](https://github.com/huggingface/huggingface_hub/tree/main/src/huggingface_hub).
 
 ## Do you want to run a Transformer model on a mobile device?
 

docs/source/internal/generation_utils.rst

@@ -13,19 +13,21 @@
 Utilities for Generation
 -----------------------------------------------------------------------------------------------------------------------
 
-This page lists all the utility functions used by :meth:`~transformers.PreTrainedModel.generate`,
-:meth:`~transformers.PreTrainedModel.greedy_search`, :meth:`~transformers.PreTrainedModel.sample`,
-:meth:`~transformers.PreTrainedModel.beam_search`, :meth:`~transformers.PreTrainedModel.beam_sample`, and
-:meth:`~transformers.PreTrainedModel.group_beam_search`.
+This page lists all the utility functions used by :meth:`~transformers.generation_utils.GenerationMixin.generate`,
+:meth:`~transformers.generation_utils.GenerationMixin.greedy_search`,
+:meth:`~transformers.generation_utils.GenerationMixin.sample`,
+:meth:`~transformers.generation_utils.GenerationMixin.beam_search`,
+:meth:`~transformers.generation_utils.GenerationMixin.beam_sample`, and
+:meth:`~transformers.generation_utils.GenerationMixin.group_beam_search`.
 
 Most of those are only useful if you are studying the code of the generate methods in the library.
 
 Generate Outputs
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
-The output of :meth:`~transformers.PreTrainedModel.generate` is an instance of a subclass of
+The output of :meth:`~transformers.generation_utils.GenerationMixin.generate` is an instance of a subclass of
 :class:`~transformers.file_utils.ModelOutput`. This output is a data structure containing all the information returned
-by :meth:`~transformers.PreTrainedModel.generate`, but that can also be used as tuple or dictionary.
+by :meth:`~transformers.generation_utils.GenerationMixin.generate`, but that can also be used as tuple or dictionary.
 
 Here's an example:
 
@@ -78,6 +80,9 @@ GreedySearchOutput
 .. autoclass:: transformers.generation_utils.GreedySearchEncoderDecoderOutput
     :members:
 
+.. autoclass:: transformers.generation_flax_utils.FlaxGreedySearchOutput
+    :members:
+
 
 SampleOutput
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
@@ -88,6 +93,9 @@ SampleOutput
 .. autoclass:: transformers.generation_utils.SampleEncoderDecoderOutput
     :members:
 
+.. autoclass:: transformers.generation_flax_utils.FlaxSampleOutput
+    :members:
+
 
 BeamSearchOutput
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
@@ -151,6 +159,43 @@ generation.
 .. autoclass:: transformers.HammingDiversityLogitsProcessor
     :members: __call__
 
+.. autoclass:: transformers.ForcedBOSTokenLogitsProcessor
+    :members: __call__
+
+.. autoclass:: transformers.ForcedEOSTokenLogitsProcessor
+    :members: __call__
+
+.. autoclass:: transformers.InfNanRemoveLogitsProcessor
+    :members: __call__
+
+.. autoclass:: transformers.FlaxLogitsProcessor
+    :members: __call__
+
+.. autoclass:: transformers.FlaxLogitsProcessorList
+    :members: __call__
+
+.. autoclass:: transformers.FlaxLogitsWarper
+    :members: __call__
+
+.. autoclass:: transformers.FlaxTemperatureLogitsWarper
+    :members: __call__
+
+.. autoclass:: transformers.FlaxTopPLogitsWarper
+    :members: __call__
+
+.. autoclass:: transformers.FlaxTopKLogitsWarper
+    :members: __call__
+
+.. autoclass:: transformers.FlaxForcedBOSTokenLogitsProcessor
+    :members: __call__
+
+.. autoclass:: transformers.FlaxForcedEOSTokenLogitsProcessor
+    :members: __call__
+
+.. autoclass:: transformers.FlaxMinLengthLogitsProcessor
+    :members: __call__
+
+
 StoppingCriteria
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 

docs/source/internal/pipelines_utils.rst

@@ -47,6 +47,4 @@ Data format
 Utilities
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
-.. autofunction:: transformers.pipelines.get_framework
-
 .. autoclass:: transformers.pipelines.PipelineException

docs/source/internal/trainer_utils.rst

@@ -1,4 +1,4 @@
-.. 
+..
     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
@@ -46,3 +46,9 @@ Distributed Evaluation
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 .. autoclass:: transformers.HfArgumentParser
+
+
+Debug Utilities
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.debug_utils.DebugUnderflowOverflow

docs/source/main_classes/callback.rst

@@ -74,6 +74,32 @@ TrainerCallback
 .. autoclass:: transformers.TrainerCallback
     :members:
 
+Here is an example of how to register a custom callback with the PyTorch :class:`~transformers.Trainer`:
+
+.. code-block:: python
+
+    class MyCallback(TrainerCallback):
+        "A callback that prints a message at the beginning of training"
+
+        def on_train_begin(self, args, state, control, **kwargs):
+            print("Starting training")
+
+    trainer = Trainer(
+        model,
+        args,
+        train_dataset=train_dataset,
+        eval_dataset=eval_dataset,
+        callbacks=[MyCallback]  # We can either pass the callback class this way or an instance of it (MyCallback())
+    )
+
+Another way to register a callback is to call ``trainer.add_callback()`` as follows:
+
+.. code-block:: python
+
+    trainer = Trainer(...)
+    trainer.add_callback(MyCallback)
+    # Alternatively, we can pass an instance of the callback class
+    trainer.add_callback(MyCallback())
 
 TrainerState
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

docs/source/main_classes/data_collator.rst

@@ -0,0 +1,71 @@
+.. 
+    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.
+
+Data Collator
+-----------------------------------------------------------------------------------------------------------------------
+
+Data collators are objects that will form a batch by using a list of dataset elements as input. These elements are of
+the same type as the elements of :obj:`train_dataset` or :obj:`eval_dataset`.
+
+To be able to build batches, data collators may apply some processing (like padding). Some of them (like
+:class:`~transformers.DataCollatorForLanguageModeling`) also apply some random data augmentation (like random masking)
+oin the formed batch.
+
+Examples of use can be found in the :doc:`example scripts <../examples>` or :doc:`example notebooks <../notebooks>`.
+
+
+Default data collator
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autofunction:: transformers.data.data_collator.default_data_collator
+
+
+DataCollatorWithPadding
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.data.data_collator.DataCollatorWithPadding
+    :members:
+
+
+DataCollatorForTokenClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.data.data_collator.DataCollatorForTokenClassification
+    :members:
+
+
+DataCollatorForSeq2Seq
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.data.data_collator.DataCollatorForSeq2Seq
+    :members:
+
+
+DataCollatorForLanguageModeling
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.data.data_collator.DataCollatorForLanguageModeling
+    :members: mask_tokens
+
+
+DataCollatorForWholeWordMask
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.data.data_collator.DataCollatorForWholeWordMask
+    :members: mask_tokens
+
+
+DataCollatorForPermutationLanguageModeling
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.data.data_collator.DataCollatorForPermutationLanguageModeling
+    :members: mask_tokens

docs/source/main_classes/feature_extractor.rst

@@ -39,3 +39,10 @@ BatchFeature
 
 .. autoclass:: transformers.BatchFeature
     :members:
+
+
+ImageFeatureExtractionMixin
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.image_utils.ImageFeatureExtractionMixin
+    :members:

docs/source/main_classes/model.rst

@@ -26,8 +26,9 @@ are common among all the models to:
 
 The other methods that are common to each model are defined in :class:`~transformers.modeling_utils.ModuleUtilsMixin`
 (for the PyTorch models) and :class:`~transformers.modeling_tf_utils.TFModuleUtilsMixin` (for the TensorFlow models) or
-for text generation, :class:`~transformers.generation_utils.GenerationMixin` (for the PyTorch models) and
-:class:`~transformers.generation_tf_utils.TFGenerationMixin` (for the TensorFlow models)
+for text generation, :class:`~transformers.generation_utils.GenerationMixin` (for the PyTorch models),
+:class:`~transformers.generation_tf_utils.TFGenerationMixin` (for the TensorFlow models) and
+:class:`~transformers.generation_flax_utils.FlaxGenerationMixin` (for the Flax/JAX models).
 
 
 PreTrainedModel
@@ -73,3 +74,13 @@ Generation
 
 .. autoclass:: transformers.generation_tf_utils.TFGenerationMixin
     :members:
+
+.. autoclass:: transformers.generation_flax_utils.FlaxGenerationMixin
+    :members:
+
+
+Pushing to the Hub
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.file_utils.PushToHubMixin
+    :members:

docs/source/main_classes/output.rst

@@ -13,8 +13,8 @@
 Model outputs
 -----------------------------------------------------------------------------------------------------------------------
 
-PyTorch models have outputs that are instances of subclasses of :class:`~transformers.file_utils.ModelOutput`. Those
-are data structures containing all the information returned by the model, but that can also be used as tuples or
+All models have outputs that are instances of subclasses of :class:`~transformers.file_utils.ModelOutput`. Those are
+data structures containing all the information returned by the model, but that can also be used as tuples or
 dictionaries.
 
 Let's see of this looks on an example:

docs/source/main_classes/pipelines.rst

@@ -23,9 +23,11 @@ There are two categories of pipeline abstractions to be aware about:
 - The :func:`~transformers.pipeline` which is the most powerful object encapsulating all other pipelines.
 - The other task-specific pipelines:
 
+    - :class:`~transformers.AutomaticSpeechRecognitionPipeline`
     - :class:`~transformers.ConversationalPipeline`
     - :class:`~transformers.FeatureExtractionPipeline`
     - :class:`~transformers.FillMaskPipeline`
+    - :class:`~transformers.ImageClassificationPipeline`
     - :class:`~transformers.QuestionAnsweringPipeline`
     - :class:`~transformers.SummarizationPipeline`
     - :class:`~transformers.TextClassificationPipeline`
@@ -48,6 +50,13 @@ pipeline but requires an additional argument which is the `task`.
 The task specific pipelines
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
+AutomaticSpeechRecognitionPipeline
+=======================================================================================================================
+
+.. autoclass:: transformers.AutomaticSpeechRecognitionPipeline
+    :special-members: __call__
+    :members:
+
 ConversationalPipeline
 =======================================================================================================================
 
@@ -71,6 +80,13 @@ FillMaskPipeline
     :special-members: __call__
     :members:
 
+ImageClassificationPipeline
+=======================================================================================================================
+
+.. autoclass:: transformers.ImageClassificationPipeline
+    :special-members: __call__
+    :members:
+
 NerPipeline
 =======================================================================================================================
 

docs/source/main_classes/processors.rst

@@ -68,8 +68,8 @@ Additionally, the following method can be used to load values from a data file a
 Example usage
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
-An example using these processors is given in the `run_glue.py
-<https://github.com/huggingface/pytorch-transformers/blob/master/examples/text-classification/run_glue.py>`__ script.
+An example using these processors is given in the :prefix_link:`run_glue.py
+<examples/legacy/text-classification/run_glue.py>` script.
 
 
 XNLI
@@ -89,8 +89,8 @@ This library hosts the processor to load the XNLI data:
 
 Please note that since the gold labels are available on the test set, evaluation is performed on the test set.
 
-An example using these processors is given in the `run_xnli.py
-<https://github.com/huggingface/pytorch-transformers/blob/master/examples/text-classification/run_xnli.py>`__ script.
+An example using these processors is given in the :prefix_link:`run_xnli.py
+<examples/legacy/text-classification/run_xnli.py>` script.
 
 
 SQuAD
@@ -169,4 +169,4 @@ Using `tensorflow_datasets` is as easy as using a data file:
 
 
 Another example using these processors is given in the :prefix_link:`run_squad.py
-<examples/question-answering/run_squad.py>` script.
+<examples/legacy/question-answering/run_squad.py>` script.

docs/source/main_classes/tokenizer.rst

@@ -62,6 +62,11 @@ PreTrainedTokenizer
 PreTrainedTokenizerFast
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
+The :class:`~transformers.PreTrainedTokenizerFast` depend on the `tokenizers
+<https://huggingface.co/docs/tokenizers>`__ library. The tokenizers obtained from the 🤗 tokenizers library can be
+loaded very simply into 🤗 transformers. Take a look at the :doc:`Using tokenizers from 🤗 tokenizers
+<../fast_tokenizers>` page to understand how this is done.
+
 .. autoclass:: transformers.PreTrainedTokenizerFast
     :special-members: __call__
     :members: batch_decode, convert_ids_to_tokens, convert_tokens_to_ids, convert_tokens_to_string, decode, encode, 

docs/source/main_classes/trainer.rst

@@ -31,7 +31,10 @@ the above features. To inject custom behavior you can subclass them and override
 - **get_test_dataloader**/**get_test_tfdataset** -- Creates the test DataLoader (PyTorch) or TF Dataset.
 - **log** -- Logs information on the various objects watching training.
 - **create_optimizer_and_scheduler** -- Sets up the optimizer and learning rate scheduler if they were not passed at
-  init.
+  init. Note, that you can also subclass or override the ``create_optimizer`` and ``create_scheduler`` methods
+  separately.
+- **create_optimizer** -- Sets up the optimizer if it wasn't passed at init.
+- **create_scheduler** -- Sets up the learning rate scheduler if it wasn't passed at init.
 - **compute_loss** - Computes the loss on a batch of training inputs.
 - **training_step** -- Performs a training step.
 - **prediction_step** -- Performs an evaluation/test step.
@@ -56,7 +59,7 @@ classification:
 
 .. code-block:: python
 
-    import torch
+    from torch import nn
     from transformers import Trainer
 
     class MultilabelTrainer(Trainer):
@@ -64,7 +67,7 @@ classification:
             labels = inputs.pop("labels")
             outputs = model(**inputs)
             logits = outputs.logits
-            loss_fct = torch.nn.BCEWithLogitsLoss()
+            loss_fct = nn.BCEWithLogitsLoss()
             loss = loss_fct(logits.view(-1, self.model.config.num_labels),
                             labels.float().view(-1, self.model.config.num_labels))
             return (loss, outputs) if return_outputs else loss
@@ -116,6 +119,88 @@ TFTrainingArguments
     :members:
 
 
+Logging
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+By default :class:`~transformers.Trainer` will use ``logging.INFO`` for the main process and ``logging.WARNING`` for
+the replicas if any.
+
+These defaults can be overridden to use any of the 5 ``logging`` levels with :class:`~transformers.TrainingArguments`'s
+arguments:
+
+- ``log_level`` - for the main process
+- ``log_level_replica`` - for the replicas
+
+Further, if :class:`~transformers.TrainingArguments`'s ``log_on_each_node`` is set to ``False`` only the main node will
+use the log level settings for its main process, all other nodes will use the log level settings for replicas.
+
+Note that :class:`~transformers.Trainer` is going to set ``transformers``'s log level separately for each node in its
+:meth:`~transformers.Trainer.__init__`. So you may want to set this sooner (see the next example) if you tap into other
+``transformers`` functionality before creating the :class:`~transformers.Trainer` object.
+
+Here is an example of how this can be used in an application:
+
+.. code-block:: python
+
+    [...]
+    logger = logging.getLogger(__name__)
+
+    # Setup logging
+    logging.basicConfig(
+        format="%(asctime)s - %(levelname)s - %(name)s -   %(message)s",
+        datefmt="%m/%d/%Y %H:%M:%S",
+        handlers=[logging.StreamHandler(sys.stdout)],
+    )
+
+    # set the main code and the modules it uses to the same log-level according to the node
+    log_level = training_args.get_process_log_level()
+    logger.setLevel(log_level)
+    datasets.utils.logging.set_verbosity(log_level)
+    transformers.utils.logging.set_verbosity(log_level)
+
+    trainer = Trainer(...)
+
+And then if you only want to see warnings on the main node and all other nodes to not print any most likely duplicated
+warnings you could run it as:
+
+.. code-block:: bash
+
+    my_app.py ... --log_level warning --log_level_replica error
+
+In the multi-node environment if you also don't want the logs to repeat for each node's main process, you will want to
+change the above to:
+
+.. code-block:: bash
+
+    my_app.py ... --log_level warning --log_level_replica error --log_on_each_node 0
+
+and then only the main process of the first node will log at the "warning" level, and all other processes on the main
+node and all processes on other nodes will log at the "error" level.
+
+If you need your application to be as quiet as possible you could do:
+
+.. code-block:: bash
+
+    my_app.py ... --log_level error --log_level_replica error --log_on_each_node 0
+
+(add ``--log_on_each_node 0`` if on multi-node environment)
+
+
+
+Randomness
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+When resuming from a checkpoint generated by :class:`~transformers.Trainer` all efforts are made to restore the
+`python`, `numpy` and `pytorch` RNG states to the same states as they were at the moment of saving that checkpoint,
+which should make the "stop and resume" style of training as close as possible to non-stop training.
+
+However, due to various default non-deterministic pytorch settings this might not fully work. If you want full
+determinism please refer to `Controlling sources of randomness
+<https://pytorch.org/docs/stable/notes/randomness.html>`__. As explained in the document, that some of those settings
+that make things determinstic (.e.g., ``torch.backends.cudnn.deterministic``) may slow things down, therefore this
+can't be done by default, but you can enable those yourself if needed.
+
+
 Trainer Integrations
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
@@ -131,7 +216,9 @@ Toward Training Trillion Parameter Models, by Samyam Rajbhandari, Jeff Rasley, O
 
 This provided support is new and experimental as of this writing.
 
-Installation Notes
+.. _zero-install-notes:
+
+CUDA Extension Installation Notes
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
 As of this writing, both FairScale and Deepspeed require compilation of CUDA C++ code, before they can be used.
@@ -153,7 +240,8 @@ please, read the following notes first.
 In these notes we give examples for what to do when ``pytorch`` has been built with CUDA ``10.2``. If your situation is
 different remember to adjust the version number to the one you are after.
 
-**Possible problem #1:**
+Possible problem #1
+=======================================================================================================================
 
 While, Pytorch comes with its own CUDA toolkit, to build these two projects you must have an identical version of CUDA
 installed system-wide.
@@ -173,7 +261,8 @@ If you don't have CUDA installed system-wide, install it first. You will find th
 search engine. For example, if you're on Ubuntu you may want to search for: `ubuntu cuda 10.2 install
 <https://www.google.com/search?q=ubuntu+cuda+10.2+install>`__.
 
-**Possible problem #2:**
+Possible problem #2
+=======================================================================================================================
 
 Another possible common problem is that you may have more than one CUDA toolkit installed system-wide. For example you
 may have:
@@ -219,7 +308,8 @@ exist. ``lib64`` sub-directory is where the various CUDA ``.so`` objects, like `
 that your system will have it named differently, but if it is adjust it to reflect your reality.
 
 
-**Possible problem #3:**
+Possible problem #3
+=======================================================================================================================
 
 Some older CUDA versions may refuse to build with newer compilers. For example, you my have ``gcc-9`` but it wants
 ``gcc-7``.
@@ -244,13 +334,6 @@ should find ``gcc-7`` (and ``g++7``) and then the build will succeed.
 
 As always make sure to edit the paths in the example to match your situation.
 
-**If still unsuccessful:**
-
-If after addressing these you still encounter build issues, please, proceed with the GitHub Issue of `FairScale
-<https://github.com/facebookresearch/fairscale/issues>`__ and `Deepspeed
-<https://github.com/microsoft/DeepSpeed/issues>`__, depending on the project you have the problem with.
-
-
 FairScale
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
@@ -264,26 +347,80 @@ provides support for the following features from `the ZeRO paper <https://arxiv.
 
 You will need at least two GPUs to use this feature.
 
-To deploy this feature:
 
-1. Install the library via pypi:
+**Installation**:
 
-   .. code-block:: bash
+Install the library via pypi:
 
-       pip install fairscale
+.. code-block:: bash
 
-   or find more details on `the FairScale's GitHub page
-   <https://github.com/facebookresearch/fairscale/#installation>`__.
+    pip install fairscale
 
-2. To use the first version of Sharded data-parallelism, add ``--sharded_ddp simple`` to the command line arguments,
-   and make sure you have added the distributed launcher ``-m torch.distributed.launch
-   --nproc_per_node=NUMBER_OF_GPUS_YOU_HAVE`` if you haven't been using it already.
+or via ``transformers``' ``extras``:
+
+.. code-block:: bash
+
+    pip install transformers[fairscale]
+
+(will become available starting from ``transformers==4.6.0``)
+
+or find more details on `the FairScale's GitHub page <https://github.com/facebookresearch/fairscale/#installation>`__.
+
+If you're still struggling with the build, first make sure to read :ref:`zero-install-notes`.
+
+If it's still not resolved the build issue, here are a few more ideas.
+
+``fairscale`` seems to have an issue with the recently introduced by pip build isolation feature. If you have a problem
+with it, you may want to try one of:
+
+.. code-block:: bash
+
+    pip install fairscale --no-build-isolation .
+
+or:
+
+.. code-block:: bash
+
+    git clone https://github.com/facebookresearch/fairscale/
+    cd fairscale
+    rm -r dist build
+    python setup.py bdist_wheel
+    pip uninstall -y fairscale
+    pip install dist/fairscale-*.whl
+
+``fairscale`` also has issues with building against pytorch-nightly, so if you use it you may have to try one of:
+
+.. code-block:: bash
+
+    pip uninstall -y fairscale; pip install fairscale --pre \
+    -f https://download.pytorch.org/whl/nightly/cu110/torch_nightly.html \
+    --no-cache --no-build-isolation
+
+or:
+
+.. code-block:: bash
+
+    pip install -v --disable-pip-version-check . \
+    -f https://download.pytorch.org/whl/nightly/cu110/torch_nightly.html --pre
+
+Of course, adjust the urls to match the cuda version you use.
+
+If after trying everything suggested you still encounter build issues, please, proceed with the GitHub Issue of
+`FairScale <https://github.com/facebookresearch/fairscale/issues>`__.
+
+
+
+**Usage**:
+
+To use the first version of Sharded data-parallelism, add ``--sharded_ddp simple`` to the command line arguments, and
+make sure you have added the distributed launcher ``-m torch.distributed.launch
+--nproc_per_node=NUMBER_OF_GPUS_YOU_HAVE`` if you haven't been using it already.
 
 For example here is how you could use it for ``run_translation.py`` with 2 GPUs:
 
 .. code-block:: bash
 
-    python -m torch.distributed.launch --nproc_per_node=2 examples/seq2seq/run_translation.py \
+    python -m torch.distributed.launch --nproc_per_node=2 examples/pytorch/translation/run_translation.py \
     --model_name_or_path t5-small --per_device_train_batch_size 1   \
     --output_dir output_dir --overwrite_output_dir \
     --do_train --max_train_samples 500 --num_train_epochs 1 \
@@ -300,15 +437,15 @@ Notes:
   able to use significantly larger batch sizes using the same hardware (e.g. 3x and even bigger) which should lead to
   significantly shorter training time.
 
-3. To use the second version of Sharded data-parallelism, add ``--sharded_ddp zero_dp_2`` or ``--sharded_ddp zero_dp_3`
-   to the command line arguments, and make sure you have added the distributed launcher ``-m torch.distributed.launch
-   --nproc_per_node=NUMBER_OF_GPUS_YOU_HAVE`` if you haven't been using it already.
+3. To use the second version of Sharded data-parallelism, add ``--sharded_ddp zero_dp_2`` or ``--sharded_ddp
+   zero_dp_3`` to the command line arguments, and make sure you have added the distributed launcher ``-m
+   torch.distributed.launch --nproc_per_node=NUMBER_OF_GPUS_YOU_HAVE`` if you haven't been using it already.
 
 For example here is how you could use it for ``run_translation.py`` with 2 GPUs:
 
 .. code-block:: bash
 
-    python -m torch.distributed.launch --nproc_per_node=2 examples/seq2seq/run_translation.py \
+    python -m torch.distributed.launch --nproc_per_node=2 examples/pytorch/translation/run_translation.py \
     --model_name_or_path t5-small --per_device_train_batch_size 1   \
     --output_dir output_dir --overwrite_output_dir \
     --do_train --max_train_samples 500 --num_train_epochs 1 \
@@ -342,614 +479,131 @@ Known caveats:
 DeepSpeed
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
-`DeepSpeed <https://github.com/microsoft/DeepSpeed>`__ implements everything described in the `ZeRO paper
-<https://arxiv.org/abs/1910.02054>`__, except ZeRO's stage 3. "Parameter Partitioning (Pos+g+p)". Currently it provides
-full support for:
-
-1. Optimizer State Partitioning (ZeRO stage 1)
-2. Add Gradient Partitioning (ZeRO stage 2)
-3. Custom fp16 handling
-4. A range of fast Cuda-extension-based Optimizers
-5. ZeRO-Offload
-
-ZeRO-Offload has its own dedicated paper: `ZeRO-Offload: Democratizing Billion-Scale Model Training
-<https://arxiv.org/abs/2101.06840>`__.
-
-DeepSpeed is currently used only for training, as all the currently available features are of no use to inference.
 
+Moved to :ref:`deepspeed-trainer-integration`.
 
 
 Installation
 =======================================================================================================================
 
-Install the library via pypi:
+Moved to :ref:`deepspeed-installation`.
 
-.. code-block:: bash
-
-    pip install deepspeed
-
-or find more details on `the DeepSpeed's GitHub page <https://github.com/microsoft/deepspeed#installation>`__.
 
 Deployment with multiple GPUs
 =======================================================================================================================
 
-To deploy this feature with multiple GPUs adjust the :class:`~transformers.Trainer` command line arguments as
-following:
-
-1. replace ``python -m torch.distributed.launch`` with ``deepspeed``.
-2. add a new argument ``--deepspeed ds_config.json``, where ``ds_config.json`` is the DeepSpeed configuration file as
-   documented `here <https://www.deepspeed.ai/docs/config-json/>`__. The file naming is up to you.
-
-Therefore, if your original command line looked as following:
-
-.. code-block:: bash
-
-    python -m torch.distributed.launch --nproc_per_node=2 your_program.py <normal cl args>
-
-Now it should be:
-
-.. code-block:: bash
-
-    deepspeed --num_gpus=2 your_program.py <normal cl args> --deepspeed ds_config.json
-
-Unlike, ``torch.distributed.launch`` where you have to specify how many GPUs to use with ``--nproc_per_node``, with the
-``deepspeed`` launcher you don't have to use the corresponding ``--num_gpus`` if you want all of your GPUs used. The
-full details on how to configure various nodes and GPUs can be found `here
-<https://www.deepspeed.ai/getting-started/#resource-configuration-multi-node>`__.
-
-In fact, you can continue using ``-m torch.distributed.launch`` with DeepSpeed as long as you don't need to use
-``deepspeed`` launcher-specific arguments. Typically if you don't need a multi-node setup you're not required to use
-the ``deepspeed`` launcher. But since in the DeepSpeed documentation it'll be used everywhere, for consistency we will
-use it here as well.
-
-Here is an example of running ``run_translation.py`` under DeepSpeed deploying all available GPUs:
-
-.. code-block:: bash
-
-    deepspeed examples/seq2seq/run_translation.py \
-    --deepspeed examples/tests/deepspeed/ds_config.json \
-    --model_name_or_path t5-small --per_device_train_batch_size 1   \
-    --output_dir output_dir --overwrite_output_dir --fp16 \
-    --do_train --max_train_samples 500 --num_train_epochs 1 \
-    --dataset_name wmt16 --dataset_config "ro-en" \
-    --source_lang en --target_lang ro
-
-
-Note that in the DeepSpeed documentation you are likely to see ``--deepspeed --deepspeed_config ds_config.json`` - i.e.
-two DeepSpeed-related arguments, but for the sake of simplicity, and since there are already so many arguments to deal
-with, we combined the two into a single argument.
-
-For some practical usage examples, please, see this `post
-<https://github.com/huggingface/transformers/issues/8771#issuecomment-759248400>`__.
-
+Moved to :ref:`deepspeed-multi-gpu`.
 
 
 Deployment with one GPU
 =======================================================================================================================
 
-To deploy DeepSpeed with one GPU adjust the :class:`~transformers.Trainer` command line arguments as following:
-
-.. code-block:: bash
-
-    deepspeed --num_gpus=1 examples/seq2seq/run_translation.py \
-    --deepspeed examples/tests/deepspeed/ds_config.json \
-    --model_name_or_path t5-small --per_device_train_batch_size 1   \
-    --output_dir output_dir --overwrite_output_dir --fp16 \
-    --do_train --max_train_samples 500 --num_train_epochs 1 \
-    --dataset_name wmt16 --dataset_config "ro-en" \
-    --source_lang en --target_lang ro
-
-This is almost the same as with multiple-GPUs, but here we tell DeepSpeed explicitly to use just one GPU. By default,
-DeepSpeed deploys all GPUs it can see. If you have only 1 GPU to start with, then you don't need this argument. The
-following `documentation <https://www.deepspeed.ai/getting-started/#resource-configuration-multi-node>`__ discusses the
-launcher options.
-
-Why would you want to use DeepSpeed with just one GPU?
-
-1. It has a ZeRO-offload feature which can delegate some computations and memory to the host's CPU and RAM, and thus
-   leave more GPU resources for model's needs - e.g. larger batch size, or enabling a fitting of a very big model which
-   normally won't fit.
-2. It provides a smart GPU memory management system, that minimizes memory fragmentation, which again allows you to fit
-   bigger models and data batches.
-
-While we are going to discuss the configuration in details next, the key to getting a huge improvement on a single GPU
-with DeepSpeed is to have at least the following configuration in the configuration file:
-
-.. code-block:: json
-
-  {
-    "zero_optimization": {
-       "stage": 2,
-       "allgather_partitions": true,
-       "allgather_bucket_size": 2e8,
-       "reduce_scatter": true,
-       "reduce_bucket_size": 2e8,
-       "overlap_comm": true,
-       "contiguous_gradients": true,
-       "cpu_offload": true
-    },
-  }
-
-which enables ``cpu_offload`` and some other important features. You may experiment with the buffer sizes, you will
-find more details in the discussion below.
-
-For a practical usage example of this type of deployment, please, see this `post
-<https://github.com/huggingface/transformers/issues/8771#issuecomment-759176685>`__.
-
-Notes:
-
-- if you need to run on a specific GPU, which is different from GPU 0, you can't use ``CUDA_VISIBLE_DEVICES`` to limit
-  the visible scope of available GPUs. Instead, you have to use the following syntax:
-
-   .. code-block:: bash
-
-       deepspeed --include localhost:1 examples/seq2seq/run_translation.py ...
-
-   In this example, we tell DeepSpeed to use GPU 1 (second gpu).
-
+Moved to :ref:`deepspeed-one-gpu`.
 
 
 Deployment in Notebooks
 =======================================================================================================================
 
-The problem with running notebook cells as a script is that there is no normal ``deepspeed`` launcher to rely on, so
-under certain setups we have to emulate it.
-
-Here is how you'd have to adjust your training code in the notebook to use DeepSpeed.
-
-.. code-block:: python
-
-    # DeepSpeed requires a distributed environment even when only one process is used.
-    # This emulates a launcher in the notebook
-    import os
-    os.environ['MASTER_ADDR'] = 'localhost'
-    os.environ['MASTER_PORT'] = '9994' # modify if RuntimeError: Address already in use
-    os.environ['RANK'] = "0"
-    os.environ['LOCAL_RANK'] = "0"
-    os.environ['WORLD_SIZE'] = "1"
-
-    # Now proceed as normal, plus pass the deepspeed config file
-    training_args = TrainingArguments(..., deepspeed="ds_config.json")
-    trainer = Trainer(...)
-    trainer.train()
-
-Note: `...` stands for the normal arguments that you'd pass to the functions.
-
-If you want to create the config file on the fly in the notebook in the current directory, you could have a dedicated
-cell with:
-
-.. code-block:: python
-
-    %%bash
-    cat <<'EOT' > ds_config.json
-    {
-        "fp16": {
-            "enabled": true,
-            "loss_scale": 0,
-            "loss_scale_window": 1000,
-            "hysteresis": 2,
-            "min_loss_scale": 1
-        },
-
-        "zero_optimization": {
-            "stage": 2,
-            "allgather_partitions": true,
-            "allgather_bucket_size": 2e8,
-            "overlap_comm": true,
-            "reduce_scatter": true,
-            "reduce_bucket_size": 2e8,
-            "contiguous_gradients": true,
-            "cpu_offload": true
-        },
-
-        "zero_allow_untested_optimizer": true,
-
-        "optimizer": {
-            "type": "AdamW",
-            "params": {
-                "lr": 3e-5,
-                "betas": [0.8, 0.999],
-                "eps": 1e-8,
-                "weight_decay": 3e-7
-            }
-        },
-
-        "scheduler": {
-            "type": "WarmupLR",
-            "params": {
-                "warmup_min_lr": 0,
-                "warmup_max_lr": 3e-5,
-                "warmup_num_steps": 500
-            }
-        },
-
-        "steps_per_print": 2000,
-        "wall_clock_breakdown": false
-    }
-    EOT
-
-
-That's said if the script is not in the notebook cells, you can launch ``deepspeed`` normally via shell from a cell
-with:
-
-.. code-block::
-
-   !deepspeed examples/seq2seq/run_translation.py ...
-
-or with bash magic, where you can write a multi-line code for the shell to run:
-
-.. code-block::
-
-   %%bash
-
-   cd /somewhere
-   deepspeed examples/seq2seq/run_translation.py ...
-
-
+Moved to :ref:`deepspeed-notebook`.
 
 
 Configuration
 =======================================================================================================================
 
-For the complete guide to the DeepSpeed configuration options that can be used in its configuration file please refer
-to the `following documentation <https://www.deepspeed.ai/docs/config-json/>`__.
+Moved to :ref:`deepspeed-config`.
 
-You can find dozens of DeepSpeed configuration examples that address various practical needs in `the DeepSpeedExamples
-repo <https://github.com/microsoft/DeepSpeedExamples>`__:
 
-.. code-block:: bash
+Passing Configuration
+=======================================================================================================================
 
-  git clone https://github.com/microsoft/DeepSpeedExamples
-  cd DeepSpeedExamples
-  find . -name '*json'
+Moved to :ref:`deepspeed-config-passing`.
 
-Continuing the code from above, let's say you're looking to configure the Lamb optimizer. So you can search through the
-example ``.json`` files with:
-
-.. code-block:: bash
-
-  grep -i Lamb $(find . -name '*json')
-
-Some more examples are to be found in the `main repo <https://github.com/microsoft/DeepSpeed>`__ as well.
-
-While you always have to supply the DeepSpeed configuration file, you can configure the DeepSpeed integration in
-several ways:
-
-1. Supply most of the configuration inside the file, and just use a few required command line arguments. This is the
-   recommended way as it puts most of the configuration params in one place.
-2. Supply just the ZeRO configuration params inside the file, and configure the rest using the normal
-   :class:`~transformers.Trainer` command line arguments.
-3. Any variation of the first two ways.
-
-To get an idea of what DeepSpeed configuration file looks like, here is one that activates ZeRO stage 2 features,
-enables FP16, uses AdamW optimizer and WarmupLR scheduler:
-
-.. code-block:: json
-
-    {
-        "fp16": {
-            "enabled": true,
-            "loss_scale": 0,
-            "loss_scale_window": 1000,
-            "hysteresis": 2,
-            "min_loss_scale": 1
-        },
-
-       "zero_optimization": {
-           "stage": 2,
-           "allgather_partitions": true,
-           "allgather_bucket_size": 5e8,
-           "overlap_comm": true,
-           "reduce_scatter": true,
-           "reduce_bucket_size": 5e8,
-           "contiguous_gradients": true,
-           "cpu_offload": true
-       },
-
-       "optimizer": {
-         "type": "AdamW",
-         "params": {
-           "lr": 3e-5,
-           "betas": [ 0.8, 0.999 ],
-           "eps": 1e-8,
-           "weight_decay": 3e-7
-         }
-       },
-
-       "scheduler": {
-         "type": "WarmupLR",
-         "params": {
-           "warmup_min_lr": 0,
-           "warmup_max_lr": 3e-5,
-           "warmup_num_steps": 500
-         }
-       }
-    }
-
-If you already have a command line that you have been using with :class:`transformers.Trainer` args, you can continue
-using those and the :class:`~transformers.Trainer` will automatically convert them into the corresponding DeepSpeed
-configuration at run time. For example, you could use the following configuration file:
-
-.. code-block:: json
-
-    {
-       "zero_optimization": {
-           "stage": 2,
-           "allgather_partitions": true,
-           "allgather_bucket_size": 5e8,
-           "overlap_comm": true,
-           "reduce_scatter": true,
-           "reduce_bucket_size": 5e8,
-           "contiguous_gradients": true,
-           "cpu_offload": true
-       }
-    }
-
-and the following command line arguments:
-
-.. code-block:: bash
-
-    --learning_rate 3e-5 --warmup_steps 500 --adam_beta1 0.8 --adam_beta2 0.999 --adam_epsilon 1e-8 \
-    --weight_decay 3e-7 --lr_scheduler_type constant_with_warmup --fp16 --fp16_backend amp
-
-to achieve the same configuration as provided by the longer json file in the first example.
-
-When you execute the program, DeepSpeed will log the configuration it received from the :class:`~transformers.Trainer`
-to the console, so you can see exactly what the final configuration was passed to it.
 
 Shared Configuration
 =======================================================================================================================
 
-Some configuration information is required by both the :class:`~transformers.Trainer` and DeepSpeed to function
-correctly, therefore, to prevent conflicting definitions, which could lead to hard to detect errors, we chose to
-configure those via the :class:`~transformers.Trainer` command line arguments.
-
-Therefore, the following DeepSpeed configuration params shouldn't be used with the :class:`~transformers.Trainer`:
-
-* ``train_batch_size``
-* ``train_micro_batch_size_per_gpu``
-* ``gradient_accumulation_steps``
-
-as these will be automatically derived from the run time environment and the following 2 command line arguments:
-
-.. code-block:: bash
-
-    --per_device_train_batch_size 8 --gradient_accumulation_steps 2
-
-which are always required to be supplied.
-
-Of course, you will need to adjust the values in this example to your situation.
-
-
+Moved to :ref:`deepspeed-config-shared`.
 
 ZeRO
 =======================================================================================================================
 
-The ``zero_optimization`` section of the configuration file is the most important part (`docs
-<https://www.deepspeed.ai/docs/config-json/#zero-optimizations-for-fp16-training>`__), since that is where you define
-which ZeRO stages you want to enable and how to configure them.
+Moved to :ref:`deepspeed-zero`.
 
-.. code-block:: json
+ZeRO-2 Config
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 
-    {
-       "zero_optimization": {
-           "stage": 2,
-           "allgather_partitions": true,
-           "allgather_bucket_size": 5e8,
-           "overlap_comm": true,
-           "reduce_scatter": true,
-           "reduce_bucket_size": 5e8,
-           "contiguous_gradients": true,
-           "cpu_offload": true
-       }
-    }
+Moved to :ref:`deepspeed-zero2-config`.
 
-Notes:
+ZeRO-3 Config
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 
-- enabling ``cpu_offload`` should reduce GPU RAM usage (it requires ``"stage": 2``)
-- ``"overlap_comm": true`` trades off increased GPU RAM usage to lower all-reduce latency. ``overlap_comm`` uses 4.5x
-  the ``allgather_bucket_size`` and ``reduce_bucket_size`` values. So if they are set to 5e8, this requires a 9GB
-  footprint (``5e8 x 2Bytes x 2 x 4.5``). Therefore, if you have a GPU with 8GB or less RAM, to avoid getting
-  OOM-errors you will need to reduce those parameters to about ``2e8``, which would require 3.6GB. You will want to do
-  the same on larger capacity GPU as well, if you're starting to hit OOM.
-- when reducing these buffers you're trading communication speed to avail more GPU RAM. The smaller the buffer size,
-  the slower the communication, and the more GPU RAM will be available to other tasks. So if a bigger batch size is
-  important, getting a slightly slower training time could be a good trade.
+Moved to :ref:`deepspeed-zero3-config`.
 
-This section has to be configured exclusively via DeepSpeed configuration - the :class:`~transformers.Trainer` provides
-no equivalent command line arguments.
+
+NVMe Support
+=======================================================================================================================
+
+Moved to :ref:`deepspeed-nvme`.
+
+ZeRO-2 vs ZeRO-3 Performance
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
+
+Moved to :ref:`deepspeed-zero2-zero3-performance`.
+
+ZeRO-2 Example
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
+
+Moved to :ref:`deepspeed-zero2-example`.
+
+ZeRO-3 Example
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
+
+Moved to :ref:`deepspeed-zero3-example`.
+
+Optimizer and Scheduler
+=======================================================================================================================
 
 
 
 Optimizer
-=======================================================================================================================
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 
-
-DeepSpeed's main optimizers are Adam, AdamW, OneBitAdam, and Lamb. These have been thoroughly tested with ZeRO and are
-thus recommended to be used. It, however, can import other optimizers from ``torch``. The full documentation is `here
-<https://www.deepspeed.ai/docs/config-json/#optimizer-parameters>`__.
-
-If you don't configure the ``optimizer`` entry in the configuration file, the :class:`~transformers.Trainer` will
-automatically set it to ``AdamW`` and will use the supplied values or the defaults for the following command line
-arguments: ``--learning_rate``, ``--adam_beta1``, ``--adam_beta2``, ``--adam_epsilon`` and ``--weight_decay``.
-
-Here is an example of the pre-configured ``optimizer`` entry for AdamW:
-
-.. code-block:: json
-
-    {
-       "optimizer": {
-           "type": "AdamW",
-           "params": {
-             "lr": 0.001,
-             "betas": [0.8, 0.999],
-             "eps": 1e-8,
-             "weight_decay": 3e-7
-           }
-         }
-    }
-
-If you want to use another optimizer which is not listed above, you will have to add ``"zero_allow_untested_optimizer":
-true`` to the top level configuration.
-
-If you want to use one of the officially supported optimizers, configure them explicitly in the configuration file, and
-make sure to adjust the values. e.g. if use Adam you will want ``weight_decay`` around ``0.01``.
+Moved to :ref:`deepspeed-optimizer`.
 
 
 Scheduler
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
+
+Moved to :ref:`deepspeed-scheduler`.
+
+fp32 Precision
 =======================================================================================================================
 
-DeepSpeed supports LRRangeTest, OneCycle, WarmupLR and WarmupDecayLR LR schedulers. The full documentation is `here
-<https://www.deepspeed.ai/docs/config-json/#scheduler-parameters>`__.
-
-If you don't configure the ``scheduler`` entry in the configuration file, the :class:`~transformers.Trainer` will use
-the value of ``--lr_scheduler_type`` to configure it. Currently the :class:`~transformers.Trainer` supports only 2 LR
-schedulers that are also supported by DeepSpeed:
-
-* ``WarmupLR`` via ``--lr_scheduler_type constant_with_warmup``
-* ``WarmupDecayLR`` via ``--lr_scheduler_type linear``. This is also the default value for ``--lr_scheduler_type``,
-  therefore, if you don't configure the scheduler this is scheduler that will get configured by default.
-
-In either case, the values of ``--learning_rate`` and ``--warmup_steps`` will be used for the configuration.
-
-In other words, if you don't use the configuration file to set the ``scheduler`` entry, provide either:
-
-.. code-block:: bash
-
-    --lr_scheduler_type constant_with_warmup --learning_rate 3e-5 --warmup_steps 500
-
-or
-
-.. code-block:: bash
-
-    --lr_scheduler_type linear --learning_rate 3e-5 --warmup_steps 500
-
-with the desired values. If you don't pass these arguments, reasonable default values will be used instead.
-
-In the case of WarmupDecayLR ``total_num_steps`` gets set either via the ``--max_steps`` command line argument, or if
-it is not provided, derived automatically at run time based on the environment and the size of the dataset and other
-command line arguments.
-
-Here is an example of the pre-configured ``scheduler`` entry for WarmupLR (``constant_with_warmup`` in the
-:class:`~transformers.Trainer` API):
-
-.. code-block:: json
-
-    {
-       "scheduler": {
-             "type": "WarmupLR",
-             "params": {
-                 "warmup_min_lr": 0,
-                 "warmup_max_lr": 0.001,
-                 "warmup_num_steps": 1000
-             }
-         }
-    }
+Moved to :ref:`deepspeed-fp32`.
 
 Automatic Mixed Precision
 =======================================================================================================================
 
-You can work with FP16 in one of the following ways:
+Moved to :ref:`deepspeed-amp`.
 
-1. Pytorch native amp, as documented `here <https://www.deepspeed.ai/docs/config-json/#fp16-training-options>`__.
-2. NVIDIA's apex, as documented `here
-   <https://www.deepspeed.ai/docs/config-json/#automatic-mixed-precision-amp-training-options>`__.
-
-If you want to use an equivalent of the Pytorch native amp, you can either configure the ``fp16`` entry in the
-configuration file, or use the following command line arguments: ``--fp16 --fp16_backend amp``.
-
-Here is an example of the ``fp16`` configuration:
-
-.. code-block:: json
-
-    {
-        "fp16": {
-            "enabled": true,
-            "loss_scale": 0,
-            "loss_scale_window": 1000,
-            "hysteresis": 2,
-            "min_loss_scale": 1
-        },
-    }
-
-If you want to use NVIDIA's apex instead, you can can either configure the ``amp`` entry in the configuration file, or
-use the following command line arguments: ``--fp16 --fp16_backend apex --fp16_opt_level 01``.
-
-Here is an example of the ``amp`` configuration:
-
-.. code-block:: json
-
-    {
-        "amp": {
-            "enabled": true,
-            "opt_level": "O1"
-        }
-    }
+Batch Size
+=======================================================================================================================
 
+Moved to :ref:`deepspeed-bs`.
 
 Gradient Accumulation
 =======================================================================================================================
 
-While normally DeepSpeed gets gradient accumulation configured with:
-
-.. code-block:: json
-
-    {
-        "gradient_accumulation_steps": 3,
-    }
-
-in this case, to enable gradient accumulation, pass the command line `--gradient_accumulation_steps` argument as normal
-and it will get injected into the DeepSpeed configuration.
-
-If you try to add it directly to the configuration file, you will receive an error from the Trainer - this is because
-this setting is needed by the Trainer too, and so this approach ensures that there is a single way of setting this
-value and thus avoid potential subtle errors.
-
-
-
-
+Moved to :ref:`deepspeed-grad-acc`.
 
 
 Gradient Clipping
 =======================================================================================================================
 
-If you don't configure the ``gradient_clipping`` entry in the configuration file, the :class:`~transformers.Trainer`
-will use the value of the ``--max_grad_norm`` command line argument to set it.
-
-Here is an example of the ``gradient_clipping`` configuration:
-
-.. code-block:: json
-
-    {
-        "gradient_clipping": 1.0,
-    }
+Moved to :ref:`deepspeed-grad-clip`.
 
 
-
-Notes
+Getting The Model Weights Out
 =======================================================================================================================
 
-* DeepSpeed works with the PyTorch :class:`~transformers.Trainer` but not TF :class:`~transformers.TFTrainer`.
-* While DeepSpeed has a pip installable PyPI package, it is highly recommended that it gets installed from `source
-  <https://github.com/microsoft/deepspeed#installation>`__ to best match your hardware and also if you need to enable
-  certain features, like 1-bit Adam, which aren't available in the pypi distribution.
-* You don't have to use the :class:`~transformers.Trainer` to use DeepSpeed with HuggingFace ``transformers`` - you can
-  use any model with your own trainer, and you will have to adapt the latter according to `the DeepSpeed integration
-  instructions <https://www.deepspeed.ai/getting-started/#writing-deepspeed-models>`__.
-
-Main DeepSpeed Resources
-=======================================================================================================================
-
-- `Project's github <https://github.com/microsoft/deepspeed>`__
-- `Usage docs <https://www.deepspeed.ai/getting-started/>`__
-- `API docs <https://deepspeed.readthedocs.io/en/latest/index.html>`__
-- `Blog posts <https://www.microsoft.com/en-us/research/search/?q=deepspeed>`__
-
-Papers:
-
-- `ZeRO: Memory Optimizations Toward Training Trillion Parameter Models <https://arxiv.org/abs/1910.02054>`__
-- `ZeRO-Offload: Democratizing Billion-Scale Model Training <https://arxiv.org/abs/2101.06840>`__
-
-Finally, please, remember that, HuggingFace :class:`~transformers.Trainer` only integrates DeepSpeed, therefore if you
-have any problems or questions with regards to DeepSpeed usage, please, file an issue with `DeepSpeed GitHub
-<https://github.com/microsoft/DeepSpeed/issues>`__.
+Moved to :ref:`deepspeed-weight-extraction`.

docs/source/migration.md

@@ -23,7 +23,7 @@ expected changes:
 
 #### 1. AutoTokenizers and pipelines now use fast (rust) tokenizers by default.
 
-The python and rust tokenizers have roughly the same API, but the rust tokenizers have a more complete feature set. 
+The python and rust tokenizers have roughly the same API, but the rust tokenizers have a more complete feature set.
 
 This introduces two breaking changes:
 - The handling of overflowing tokens between the python and rust tokenizers is different.
@@ -85,7 +85,7 @@ This is a breaking change as importing intermediary layers using a model's modul
 
 ##### How to obtain the same behavior as v3.x in v4.x
 
-In order to obtain the same behavior as version `v3.x`, you should update the path used to access the layers. 
+In order to obtain the same behavior as version `v3.x`, you should update the path used to access the layers.
 
 In version `v3.x`:
 ```bash
@@ -169,8 +169,8 @@ Regarding the `TFTrainer` class:
 - The `TFTrainer` method `_setup_wandb` is deprecated in favor of `setup_wandb`.
 - The `TFTrainer` method `_run_model` is deprecated in favor of `run_model`.
 
-Regarding the `TrainerArgument` class:
-- The `TrainerArgument` argument `evaluate_during_training` is deprecated in favor of `evaluation_strategy`.
+Regarding the `TrainingArguments` class:
+- The `TrainingArguments` argument `evaluate_during_training` is deprecated in favor of `evaluation_strategy`.
 
 Regarding the Transfo-XL model:
 - The Transfo-XL configuration attribute `tie_weight` becomes `tie_words_embeddings`.

docs/source/model_doc/albert.rst

@@ -43,7 +43,8 @@ Tips:
   similar to a BERT-like architecture with the same number of hidden layers as it has to iterate through the same
   number of (repeating) layers.
 
-The original code can be found `here <https://github.com/google-research/ALBERT>`__.
+This model was contributed by `lysandre <https://huggingface.co/lysandre>`__. The original code can be found `here
+<https://github.com/google-research/ALBERT>`__.
 
 AlbertConfig
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

docs/source/model_doc/auto.rst

@@ -44,6 +44,13 @@ AutoTokenizer
     :members:
 
 
+AutoFeatureExtractor
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.AutoFeatureExtractor
+    :members:
+
+
 AutoModel
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
@@ -121,6 +128,13 @@ AutoModelForTableQuestionAnswering
     :members:
 
 
+AutoModelForImageClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.AutoModelForImageClassification
+    :members:
+
+
 TFAutoModel
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
@@ -189,3 +203,73 @@ FlaxAutoModel
 
 .. autoclass:: transformers.FlaxAutoModel
     :members:
+
+
+FlaxAutoModelForCausalLM
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxAutoModelForCausalLM
+    :members:
+
+
+FlaxAutoModelForPreTraining
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxAutoModelForPreTraining
+    :members:
+
+
+FlaxAutoModelForMaskedLM
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxAutoModelForMaskedLM
+    :members:
+
+
+FlaxAutoModelForSeq2SeqLM
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxAutoModelForSeq2SeqLM
+    :members:
+
+
+FlaxAutoModelForSequenceClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxAutoModelForSequenceClassification
+    :members:
+
+
+FlaxAutoModelForQuestionAnswering
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxAutoModelForQuestionAnswering
+    :members:
+
+
+FlaxAutoModelForTokenClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxAutoModelForTokenClassification
+    :members:
+
+
+FlaxAutoModelForMultipleChoice
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxAutoModelForMultipleChoice
+    :members:
+
+
+FlaxAutoModelForNextSentencePrediction
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxAutoModelForNextSentencePrediction
+    :members:
+
+
+FlaxAutoModelForImageClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxAutoModelForImageClassification
+    :members:

docs/source/model_doc/bart.rst

@@ -35,14 +35,15 @@ According to the abstract,
   state-of-the-art results on a range of abstractive dialogue, question answering, and summarization tasks, with gains
   of up to 6 ROUGE.
 
-The Authors' code can be found `here <https://github.com/pytorch/fairseq/tree/master/examples/bart>`__.
+This model was contributed by `sshleifer <https://huggingface.co/sshleifer>`__. The Authors' code can be found `here
+<https://github.com/pytorch/fairseq/tree/master/examples/bart>`__.
 
 
 Examples
 _______________________________________________________________________________________________________________________
 
 - Examples and scripts for fine-tuning BART and other models for sequence to sequence tasks can be found in
-  :prefix_link:`examples/seq2seq/ <examples/seq2seq/README.md>`.
+  :prefix_link:`examples/pytorch/summarization/ <examples/pytorch/summarization/README.md>`.
 - An example of how to train :class:`~transformers.BartForConditionalGeneration` with a Hugging Face :obj:`datasets`
   object can be found in this `forum discussion
   <https://discuss.huggingface.co/t/train-bart-for-conditional-generation-e-g-summarization/1904>`__.
@@ -60,7 +61,7 @@ Implementation Notes
 - Model predictions are intended to be identical to the original implementation when
   :obj:`force_bos_token_to_be_generated=True`. This only works, however, if the string you pass to
   :func:`fairseq.encode` starts with a space.
-- :meth:`~transformers.BartForConditionalGeneration.generate` should be used for conditional generation tasks like
+- :meth:`~transformers.generation_utils.GenerationMixin.generate` should be used for conditional generation tasks like
   summarization, see the example in that docstrings.
 - Models that load the `facebook/bart-large-cnn` weights will not have a :obj:`mask_token_id`, or be able to perform
   mask-filling tasks.
@@ -130,6 +131,7 @@ BartForQuestionAnswering
 .. autoclass:: transformers.BartForQuestionAnswering
     :members: forward
 
+
 BartForCausalLM
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
@@ -137,7 +139,6 @@ BartForCausalLM
     :members: forward
 
 
-
 TFBartModel
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
@@ -150,3 +151,32 @@ TFBartForConditionalGeneration
 
 .. autoclass:: transformers.TFBartForConditionalGeneration
     :members: call
+
+
+FlaxBartModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxBartModel
+    :members: __call__, encode, decode
+
+
+FlaxBartForConditionalGeneration
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxBartForConditionalGeneration
+    :members: __call__, encode, decode
+
+
+FlaxBartForSequenceClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxBartForSequenceClassification
+    :members: __call__, encode, decode
+
+
+FlaxBartForQuestionAnswering
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxBartForQuestionAnswering
+    :members: __call__, encode, decode
+

docs/source/model_doc/barthez.rst

@@ -16,7 +16,7 @@ BARThez
 Overview
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
-The BARThez model was proposed in `BARThez: a Skilled Pretrained French Sequence-to-Sequence Model`
+The BARThez model was proposed in `BARThez: a Skilled Pretrained French Sequence-to-Sequence Model
 <https://arxiv.org/abs/2010.12321>`__ by Moussa Kamal Eddine, Antoine J.-P. Tixier, Michalis Vazirgiannis on 23 Oct,
 2020.
 
@@ -35,14 +35,15 @@ summarization dataset, OrangeSum, that we release with this paper. We also conti
 pretrained multilingual BART on BARThez's corpus, and we show that the resulting model, which we call mBARTHez,
 provides a significant boost over vanilla BARThez, and is on par with or outperforms CamemBERT and FlauBERT.*
 
-The Authors' code can be found `here <https://github.com/moussaKam/BARThez>`__.
+This model was contributed by `moussakam <https://huggingface.co/moussakam>`__. The Authors' code can be found `here
+<https://github.com/moussaKam/BARThez>`__.
 
 
 Examples
 _______________________________________________________________________________________________________________________
 
 - BARThez can be fine-tuned on sequence-to-sequence tasks in a similar way as BART, check:
-  :prefix_link:`examples/seq2seq/ <examples/seq2seq/README.md>`.
+  :prefix_link:`examples/pytorch/summarization/ <examples/pytorch/summarization/README.md>`.
 
 
 BarthezTokenizer

docs/source/model_doc/bert.rst

@@ -42,7 +42,8 @@ Tips:
 - BERT was trained with the masked language modeling (MLM) and next sentence prediction (NSP) objectives. It is
   efficient at predicting masked tokens and at NLU in general, but is not optimal for text generation.
 
-The original code can be found `here <https://github.com/google-research/bert>`__.
+This model was contributed by `thomwolf <https://huggingface.co/thomwolf>`__. The original code can be found `here
+<https://github.com/google-research/bert>`__.
 
 BertConfig
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
@@ -90,7 +91,7 @@ BertForPreTraining
     :members: forward
 
 
-BertModelLMHeadModel
+BertLMHeadModel
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 .. autoclass:: transformers.BertLMHeadModel
@@ -209,8 +210,50 @@ FlaxBertModel
     :members: __call__
 
 
+FlaxBertForPreTraining
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxBertForPreTraining
+    :members: __call__
+
+
 FlaxBertForMaskedLM
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 .. autoclass:: transformers.FlaxBertForMaskedLM
     :members: __call__
+
+
+FlaxBertForNextSentencePrediction
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxBertForNextSentencePrediction
+    :members: __call__
+
+
+FlaxBertForSequenceClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxBertForSequenceClassification
+    :members: __call__
+
+
+FlaxBertForMultipleChoice
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxBertForMultipleChoice
+    :members: __call__
+
+
+FlaxBertForTokenClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxBertForTokenClassification
+    :members: __call__
+
+
+FlaxBertForQuestionAnswering
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxBertForQuestionAnswering
+    :members: __call__

docs/source/model_doc/bert_japanese.rst

@@ -0,0 +1,80 @@
+.. 
+    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.
+
+BertJapanese
+-----------------------------------------------------------------------------------------------------------------------
+
+Overview
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The BERT models trained on Japanese text.
+
+There are models with two different tokenization methods:
+
+- Tokenize with MeCab and WordPiece. This requires some extra dependencies, `fugashi
+  <https://github.com/polm/fugashi>`__ which is a wrapper around `MeCab <https://taku910.github.io/mecab/>`__.
+- Tokenize into characters.
+
+To use `MecabTokenizer`, you should ``pip install transformers["ja"]`` (or ``pip install -e .["ja"]`` if you install
+from source) to install dependencies.
+
+See `details on cl-tohoku repository <https://github.com/cl-tohoku/bert-japanese>`__.
+
+Example of using a model with MeCab and WordPiece tokenization:
+
+.. code-block::
+
+    >>> import torch
+    >>> from transformers import AutoModel, AutoTokenizer 
+
+    >>> bertjapanese = AutoModel.from_pretrained("cl-tohoku/bert-base-japanese")
+    >>> tokenizer = AutoTokenizer.from_pretrained("cl-tohoku/bert-base-japanese")
+
+    >>> ## Input Japanese Text
+    >>> line = "吾輩は猫である。"
+
+    >>> inputs = tokenizer(line, return_tensors="pt")
+
+    >>> print(tokenizer.decode(inputs['input_ids'][0]))
+    [CLS] 吾輩 は 猫 で ある 。 [SEP]
+
+    >>> outputs = bertjapanese(**inputs)
+
+Example of using a model with Character tokenization:
+
+.. code-block::
+
+    >>> bertjapanese = AutoModel.from_pretrained("cl-tohoku/bert-base-japanese-char")
+    >>> tokenizer = AutoTokenizer.from_pretrained("cl-tohoku/bert-base-japanese-char")
+
+    >>> ## Input Japanese Text
+    >>> line = "吾輩は猫である。"
+
+    >>> inputs = tokenizer(line, return_tensors="pt")
+
+    >>> print(tokenizer.decode(inputs['input_ids'][0]))
+    [CLS] 吾 輩 は 猫 で あ る 。 [SEP]
+
+    >>> outputs = bertjapanese(**inputs)
+
+Tips:
+
+- This implementation is the same as BERT, except for tokenization method. Refer to the :doc:`documentation of BERT
+  <bert>` for more usage examples.
+
+This model was contributed by `cl-tohoku <https://huggingface.co/cl-tohoku>`__.
+
+BertJapaneseTokenizer
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.BertJapaneseTokenizer
+    :members: 

docs/source/model_doc/bertgeneration.rst

@@ -38,22 +38,22 @@ Usage:
 
 .. code-block::
 
-  # leverage checkpoints for Bert2Bert model...
-  # use BERT's cls token as BOS token and sep token as EOS token
-  encoder = BertGenerationEncoder.from_pretrained("bert-large-uncased", bos_token_id=101, eos_token_id=102)
-  # add cross attention layers and use BERT's cls token as BOS token and sep token as EOS token
-  decoder = BertGenerationDecoder.from_pretrained("bert-large-uncased", add_cross_attention=True, is_decoder=True, bos_token_id=101, eos_token_id=102)
-  bert2bert = EncoderDecoderModel(encoder=encoder, decoder=decoder)
+    >>> # leverage checkpoints for Bert2Bert model...
+    >>> # use BERT's cls token as BOS token and sep token as EOS token
+    >>> encoder = BertGenerationEncoder.from_pretrained("bert-large-uncased", bos_token_id=101, eos_token_id=102)
+    >>> # add cross attention layers and use BERT's cls token as BOS token and sep token as EOS token
+    >>> decoder = BertGenerationDecoder.from_pretrained("bert-large-uncased", add_cross_attention=True, is_decoder=True, bos_token_id=101, eos_token_id=102)
+    >>> bert2bert = EncoderDecoderModel(encoder=encoder, decoder=decoder)
 
-  # create tokenizer...
-  tokenizer = BertTokenizer.from_pretrained("bert-large-uncased")
+    >>> # create tokenizer...
+    >>> tokenizer = BertTokenizer.from_pretrained("bert-large-uncased")
 
-  input_ids = tokenizer('This is a long article to summarize', add_special_tokens=False, return_tensors="pt").input_ids
-  labels = tokenizer('This is a short summary', return_tensors="pt").input_ids
+    >>> input_ids = tokenizer('This is a long article to summarize', add_special_tokens=False, return_tensors="pt").input_ids
+    >>> labels = tokenizer('This is a short summary', return_tensors="pt").input_ids
 
-  # train...
-  loss = bert2bert(input_ids=input_ids, decoder_input_ids=labels, labels=labels).loss
-  loss.backward()
+    >>> # train...
+    >>> loss = bert2bert(input_ids=input_ids, decoder_input_ids=labels, labels=labels).loss
+    >>> loss.backward()
 
 
 - Pretrained :class:`~transformers.EncoderDecoderModel` are also directly available in the model hub, e.g.,
@@ -61,15 +61,15 @@ Usage:
 
 .. code-block::
 
-  # instantiate sentence fusion model
-  sentence_fuser = EncoderDecoderModel.from_pretrained("google/roberta2roberta_L-24_discofuse")
-  tokenizer = AutoTokenizer.from_pretrained("google/roberta2roberta_L-24_discofuse")
+    >>> # instantiate sentence fusion model
+    >>> sentence_fuser = EncoderDecoderModel.from_pretrained("google/roberta2roberta_L-24_discofuse")
+    >>> tokenizer = AutoTokenizer.from_pretrained("google/roberta2roberta_L-24_discofuse")
 
-  input_ids = tokenizer('This is the first sentence. This is the second sentence.', add_special_tokens=False, return_tensors="pt").input_ids
+    >>> input_ids = tokenizer('This is the first sentence. This is the second sentence.', add_special_tokens=False, return_tensors="pt").input_ids
 
-  outputs = sentence_fuser.generate(input_ids)
+    >>> outputs = sentence_fuser.generate(input_ids)
 
-  print(tokenizer.decode(outputs[0]))
+    >>> print(tokenizer.decode(outputs[0]))
 
 
 Tips:
@@ -79,7 +79,8 @@ Tips:
 - For summarization, sentence splitting, sentence fusion and translation, no special tokens are required for the input.
   Therefore, no EOS token should be added to the end of the input.
 
-The original code can be found `here <https://tfhub.dev/s?module-type=text-generation&subtype=module,placeholder>`__.
+This model was contributed by `patrickvonplaten <https://huggingface.co/patrickvonplaten>`__. The original code can be
+found `here <https://tfhub.dev/s?module-type=text-generation&subtype=module,placeholder>`__.
 
 BertGenerationConfig
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

docs/source/model_doc/bertweet.rst

@@ -31,31 +31,31 @@ Example of use:
 
 .. code-block::
 
-  import torch
-  from transformers import AutoModel, AutoTokenizer 
+    >>> import torch
+    >>> from transformers import AutoModel, AutoTokenizer 
 
-  bertweet = AutoModel.from_pretrained("vinai/bertweet-base")
+    >>> bertweet = AutoModel.from_pretrained("vinai/bertweet-base")
 
-  # For transformers v4.x+: 
-  tokenizer = AutoTokenizer.from_pretrained("vinai/bertweet-base", use_fast=False)
+    >>> # For transformers v4.x+: 
+    >>> tokenizer = AutoTokenizer.from_pretrained("vinai/bertweet-base", use_fast=False)
 
-  # For transformers v3.x: 
-  # tokenizer = AutoTokenizer.from_pretrained("vinai/bertweet-base")
+    >>> # For transformers v3.x: 
+    >>> # tokenizer = AutoTokenizer.from_pretrained("vinai/bertweet-base")
 
-  # INPUT TWEET IS ALREADY NORMALIZED!
-  line = "SC has first two presumptive cases of coronavirus , DHEC confirms HTTPURL via @USER :cry:"
+    >>> # INPUT TWEET IS ALREADY NORMALIZED!
+    >>> line = "SC has first two presumptive cases of coronavirus , DHEC confirms HTTPURL via @USER :cry:"
 
-  input_ids = torch.tensor([tokenizer.encode(line)])
+    >>> input_ids = torch.tensor([tokenizer.encode(line)])
 
-  with torch.no_grad():
-      features = bertweet(input_ids)  # Models outputs are now tuples
+    >>> with torch.no_grad():
+    ...     features = bertweet(input_ids)  # Models outputs are now tuples
 
-  ## With TensorFlow 2.0+:
-  # from transformers import TFAutoModel
-  # bertweet = TFAutoModel.from_pretrained("vinai/bertweet-base")
+    >>> # With TensorFlow 2.0+:
+    >>> # from transformers import TFAutoModel
+    >>> # bertweet = TFAutoModel.from_pretrained("vinai/bertweet-base")
 
-
-The original code can be found `here <https://github.com/VinAIResearch/BERTweet>`__.
+This model was contributed by `dqnguyen <https://huggingface.co/dqnguyen>`__. The original code can be found `here
+<https://github.com/VinAIResearch/BERTweet>`__.
 
 BertweetTokenizer
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

docs/source/model_doc/bigbird.rst

@@ -0,0 +1,185 @@
+.. 
+    Copyright 2021 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.
+
+BigBird
+-----------------------------------------------------------------------------------------------------------------------
+
+Overview
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The BigBird model was proposed in `Big Bird: Transformers for Longer Sequences <https://arxiv.org/abs/2007.14062>`__ by
+Zaheer, Manzil and Guruganesh, Guru and Dubey, Kumar Avinava and Ainslie, Joshua and Alberti, Chris and Ontanon,
+Santiago and Pham, Philip and Ravula, Anirudh and Wang, Qifan and Yang, Li and others. BigBird, is a sparse-attention
+based transformer which extends Transformer based models, such as BERT to much longer sequences. In addition to sparse
+attention, BigBird also applies global attention as well as random attention to the input sequence. Theoretically, it
+has been shown that applying sparse, global, and random attention approximates full attention, while being
+computationally much more efficient for longer sequences. As a consequence of the capability to handle longer context,
+BigBird has shown improved performance on various long document NLP tasks, such as question answering and
+summarization, compared to BERT or RoBERTa.
+
+The abstract from the paper is the following:
+
+*Transformers-based models, such as BERT, have been one of the most successful deep learning models for NLP.
+Unfortunately, one of their core limitations is the quadratic dependency (mainly in terms of memory) on the sequence
+length due to their full attention mechanism. To remedy this, we propose, BigBird, a sparse attention mechanism that
+reduces this quadratic dependency to linear. We show that BigBird is a universal approximator of sequence functions and
+is Turing complete, thereby preserving these properties of the quadratic, full attention model. Along the way, our
+theoretical analysis reveals some of the benefits of having O(1) global tokens (such as CLS), that attend to the entire
+sequence as part of the sparse attention mechanism. The proposed sparse attention can handle sequences of length up to
+8x of what was previously possible using similar hardware. As a consequence of the capability to handle longer context,
+BigBird drastically improves performance on various NLP tasks such as question answering and summarization. We also
+propose novel applications to genomics data.*
+
+Tips:
+
+- For an in-detail explanation on how BigBird's attention works, see `this blog post
+  <https://huggingface.co/blog/big-bird>`__.
+- BigBird comes with 2 implementations: **original_full** & **block_sparse**. For the sequence length < 1024, using
+  **original_full** is advised as there is no benefit in using **block_sparse** attention.
+- The code currently uses window size of 3 blocks and 2 global blocks.
+- Sequence length must be divisible by block size.
+- Current implementation supports only **ITC**.
+- Current implementation doesn't support **num_random_blocks = 0**
+
+This model was contributed by `vasudevgupta <https://huggingface.co/vasudevgupta>`__. The original code can be found
+`here <https://github.com/google-research/bigbird>`__.
+
+BigBirdConfig
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.BigBirdConfig
+    :members:
+
+
+BigBirdTokenizer
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.BigBirdTokenizer
+    :members: build_inputs_with_special_tokens, get_special_tokens_mask,
+        create_token_type_ids_from_sequences, save_vocabulary
+
+BigBirdTokenizerFast
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.BigBirdTokenizerFast
+    :members:
+
+BigBird specific outputs
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.models.big_bird.modeling_big_bird.BigBirdForPreTrainingOutput
+    :members:
+
+
+BigBirdModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.BigBirdModel
+    :members: forward
+
+
+BigBirdForPreTraining
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.BigBirdForPreTraining
+    :members: forward
+
+
+BigBirdForCausalLM
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.BigBirdForCausalLM
+    :members: forward
+
+
+BigBirdForMaskedLM
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.BigBirdForMaskedLM
+    :members: forward
+
+
+BigBirdForSequenceClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.BigBirdForSequenceClassification
+    :members: forward
+
+
+BigBirdForMultipleChoice
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.BigBirdForMultipleChoice
+    :members: forward
+
+
+BigBirdForTokenClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.BigBirdForTokenClassification
+    :members: forward
+
+
+BigBirdForQuestionAnswering
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.BigBirdForQuestionAnswering
+    :members: forward
+
+
+FlaxBigBirdModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxBigBirdModel
+    :members: __call__
+
+
+FlaxBigBirdForPreTraining
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxBigBirdForPreTraining
+    :members: __call__
+
+
+FlaxBigBirdForMaskedLM
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxBigBirdForMaskedLM
+    :members: __call__
+
+
+FlaxBigBirdForSequenceClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxBigBirdForSequenceClassification
+    :members: __call__
+
+
+FlaxBigBirdForMultipleChoice
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxBigBirdForMultipleChoice
+    :members: __call__
+
+
+FlaxBigBirdForTokenClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxBigBirdForTokenClassification
+    :members: __call__
+
+
+FlaxBigBirdForQuestionAnswering
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxBigBirdForQuestionAnswering
+    :members: __call__

docs/source/model_doc/bigbird_pegasus.rst

@@ -0,0 +1,98 @@
+.. 
+    Copyright 2021 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.
+
+BigBirdPegasus
+-----------------------------------------------------------------------------------------------------------------------
+
+Overview
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The BigBird model was proposed in `Big Bird: Transformers for Longer Sequences <https://arxiv.org/abs/2007.14062>`__ by
+Zaheer, Manzil and Guruganesh, Guru and Dubey, Kumar Avinava and Ainslie, Joshua and Alberti, Chris and Ontanon,
+Santiago and Pham, Philip and Ravula, Anirudh and Wang, Qifan and Yang, Li and others. BigBird, is a sparse-attention
+based transformer which extends Transformer based models, such as BERT to much longer sequences. In addition to sparse
+attention, BigBird also applies global attention as well as random attention to the input sequence. Theoretically, it
+has been shown that applying sparse, global, and random attention approximates full attention, while being
+computationally much more efficient for longer sequences. As a consequence of the capability to handle longer context,
+BigBird has shown improved performance on various long document NLP tasks, such as question answering and
+summarization, compared to BERT or RoBERTa.
+
+The abstract from the paper is the following:
+
+*Transformers-based models, such as BERT, have been one of the most successful deep learning models for NLP.
+Unfortunately, one of their core limitations is the quadratic dependency (mainly in terms of memory) on the sequence
+length due to their full attention mechanism. To remedy this, we propose, BigBird, a sparse attention mechanism that
+reduces this quadratic dependency to linear. We show that BigBird is a universal approximator of sequence functions and
+is Turing complete, thereby preserving these properties of the quadratic, full attention model. Along the way, our
+theoretical analysis reveals some of the benefits of having O(1) global tokens (such as CLS), that attend to the entire
+sequence as part of the sparse attention mechanism. The proposed sparse attention can handle sequences of length up to
+8x of what was previously possible using similar hardware. As a consequence of the capability to handle longer context,
+BigBird drastically improves performance on various NLP tasks such as question answering and summarization. We also
+propose novel applications to genomics data.*
+
+Tips:
+
+- For an in-detail explanation on how BigBird's attention works, see `this blog post
+  <https://huggingface.co/blog/big-bird>`__.
+- BigBird comes with 2 implementations: **original_full** & **block_sparse**. For the sequence length < 1024, using
+  **original_full** is advised as there is no benefit in using **block_sparse** attention.
+- The code currently uses window size of 3 blocks and 2 global blocks.
+- Sequence length must be divisible by block size.
+- Current implementation supports only **ITC**.
+- Current implementation doesn't support **num_random_blocks = 0**.
+- BigBirdPegasus uses the `PegasusTokenizer
+  <https://github.com/huggingface/transformers/blob/master/src/transformers/models/pegasus/tokenization_pegasus.py>`__.
+
+The original code can be found `here <https://github.com/google-research/bigbird>`__.
+
+BigBirdPegasusConfig
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.BigBirdPegasusConfig
+    :members:
+
+
+BigBirdPegasusModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.BigBirdPegasusModel
+    :members: forward
+
+
+BigBirdPegasusForConditionalGeneration
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.BigBirdPegasusForConditionalGeneration
+    :members: forward
+
+
+BigBirdPegasusForSequenceClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.BigBirdPegasusForSequenceClassification
+    :members: forward
+
+
+BigBirdPegasusForQuestionAnswering
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.BigBirdPegasusForQuestionAnswering
+    :members: forward
+
+
+BigBirdPegasusForCausalLM
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.BigBirdPegasusForCausalLM
+    :members: forward
+
+

docs/source/model_doc/blenderbot.rst

@@ -36,7 +36,8 @@ and code publicly available. Human evaluations show our best models are superior
 dialogue in terms of engagingness and humanness measurements. We then discuss the limitations of this work by analyzing
 failure cases of our models.*
 
-The authors' code can be found `here <https://github.com/facebookresearch/ParlAI>`__ .
+This model was contributed by `sshleifer <https://huggingface.co/sshleifer>`__. The authors' code can be found `here
+<https://github.com/facebookresearch/ParlAI>`__ .
 
 
 Implementation Notes

docs/source/model_doc/blenderbot_small.rst

@@ -39,7 +39,8 @@ and code publicly available. Human evaluations show our best models are superior
 dialogue in terms of engagingness and humanness measurements. We then discuss the limitations of this work by analyzing
 failure cases of our models.*
 
-The authors' code can be found `here <https://github.com/facebookresearch/ParlAI>`__ .
+This model was contributed by `patrickvonplaten <https://huggingface.co/patrickvonplaten>`__. The authors' code can be
+found `here <https://github.com/facebookresearch/ParlAI>`__ .
 
 BlenderbotSmallConfig
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

docs/source/model_doc/bort.rst

@@ -43,4 +43,5 @@ Tips:
   that is sadly not open-sourced yet. It would be very useful for the community, if someone tries to implement the
   algorithm to make BORT fine-tuning work.
 
-The original code can be found `here <https://github.com/alexa/bort/>`__.
+This model was contributed by `stefan-it <https://huggingface.co/stefan-it>`__. The original code can be found `here
+<https://github.com/alexa/bort/>`__.

docs/source/model_doc/byt5.rst

@@ -0,0 +1,83 @@
+.. 
+    Copyright 2021 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.
+
+ByT5
+-----------------------------------------------------------------------------------------------------------------------
+
+Overview
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The ByT5 model was presented in `ByT5: Towards a token-free future with pre-trained byte-to-byte models
+<https://arxiv.org/abs/2105.13626>`_ by Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir
+Kale, Adam Roberts, Colin Raffel.
+
+The abstract from the paper is the following:
+
+*Most widely-used pre-trained language models operate on sequences of tokens corresponding to word or subword units.
+Encoding text as a sequence of tokens requires a tokenizer, which is typically created as an independent artifact from
+the model. Token-free models that instead operate directly on raw text (bytes or characters) have many benefits: they
+can process text in any language out of the box, they are more robust to noise, and they minimize technical debt by
+removing complex and error-prone text preprocessing pipelines. Since byte or character sequences are longer than token
+sequences, past work on token-free models has often introduced new model architectures designed to amortize the cost of
+operating directly on raw text. In this paper, we show that a standard Transformer architecture can be used with
+minimal modifications to process byte sequences. We carefully characterize the trade-offs in terms of parameter count,
+training FLOPs, and inference speed, and show that byte-level models are competitive with their token-level
+counterparts. We also demonstrate that byte-level models are significantly more robust to noise and perform better on
+tasks that are sensitive to spelling and pronunciation. As part of our contribution, we release a new set of
+pre-trained byte-level Transformer models based on the T5 architecture, as well as all code and data used in our
+experiments.*
+
+This model was contributed by `patrickvonplaten <https://huggingface.co/patrickvonplaten>`__. The original code can be
+found `here <https://github.com/google-research/byt5>`__.
+
+
+ByT5's architecture is based on the T5 model, so one can refer to :doc:`T5's documentation page <t5>`.
+
+
+Example
+_______________________________________________________________________________________________________________________
+
+ByT5 works on raw UTF-8 bytes, so it can be used without a tokenizer:
+
+.. code-block::
+
+    from transformers import T5ForConditionalGeneration
+    import torch
+
+    model = T5ForConditionalGeneration.from_pretrained('google/byt5-small')
+
+    input_ids = torch.tensor([list("Life is like a box of chocolates.".encode("utf-8"))]) + 3  # add 3 for special tokens
+    labels = torch.tensor([list("La vie est comme une boîte de chocolat.".encode("utf-8"))]) + 3  # add 3 for special tokens
+
+    loss = model(input_ids, labels=labels).loss # forward pass
+
+
+For batched inference and training it is however recommended to make use of the tokenizer:
+
+.. code-block::
+
+    from transformers import T5ForConditionalGeneration, AutoTokenizer
+
+    model = T5ForConditionalGeneration.from_pretrained('google/byt5-small')
+    tokenizer = AutoTokenizer.from_pretrained('google/byt5-small')
+
+    model_inputs = tokenizer(["Life is like a box of chocolates.", "Today is Monday."], padding="longest", return_tensors="pt")
+    labels = tokenizer(["La vie est comme une boîte de chocolat.", "Aujourd'hui c'est lundi."], padding="longest", return_tensors="pt").input_ids
+
+    loss = model(**model_inputs, labels=labels).loss # forward pass
+
+ByT5Tokenizer
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.ByT5Tokenizer
+
+See :class:`~transformers.ByT5Tokenizer` for all details.

docs/source/model_doc/camembert.rst

@@ -37,7 +37,8 @@ Tips:
 - This implementation is the same as RoBERTa. Refer to the :doc:`documentation of RoBERTa <roberta>` for usage examples
   as well as the information relative to the inputs and outputs.
 
-The original code can be found `here <https://camembert-model.fr/>`__.
+This model was contributed by `camembert <https://huggingface.co/camembert>`__. The original code can be found `here
+<https://camembert-model.fr/>`__.
 
 CamembertConfig
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

docs/source/model_doc/clip.rst

@@ -0,0 +1,175 @@
+.. 
+    Copyright 2021 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.
+
+CLIP
+-----------------------------------------------------------------------------------------------------------------------
+
+Overview
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The CLIP model was proposed in `Learning Transferable Visual Models From Natural Language Supervision
+<https://arxiv.org/abs/2103.00020>`__ by Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh,
+Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, Ilya Sutskever. CLIP
+(Contrastive Language-Image Pre-Training) is a neural network trained on a variety of (image, text) pairs. It can be
+instructed in natural language to predict the most relevant text snippet, given an image, without directly optimizing
+for the task, similarly to the zero-shot capabilities of GPT-2 and 3.
+
+The abstract from the paper is the following:
+
+*State-of-the-art computer vision systems are trained to predict a fixed set of predetermined object categories. This
+restricted form of supervision limits their generality and usability since additional labeled data is needed to specify
+any other visual concept. Learning directly from raw text about images is a promising alternative which leverages a
+much broader source of supervision. We demonstrate that the simple pre-training task of predicting which caption goes
+with which image is an efficient and scalable way to learn SOTA image representations from scratch on a dataset of 400
+million (image, text) pairs collected from the internet. After pre-training, natural language is used to reference
+learned visual concepts (or describe new ones) enabling zero-shot transfer of the model to downstream tasks. We study
+the performance of this approach by benchmarking on over 30 different existing computer vision datasets, spanning tasks
+such as OCR, action recognition in videos, geo-localization, and many types of fine-grained object classification. The
+model transfers non-trivially to most tasks and is often competitive with a fully supervised baseline without the need
+for any dataset specific training. For instance, we match the accuracy of the original ResNet-50 on ImageNet zero-shot
+without needing to use any of the 1.28 million training examples it was trained on. We release our code and pre-trained
+model weights at this https URL.*
+
+Usage
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+CLIP is a multi-modal vision and language model. It can be used for image-text similarity and for zero-shot image
+classification. CLIP uses a ViT like transformer to get visual features and a causal language model to get the text
+features. Both the text and visual features are then projected to a latent space with identical dimension. The dot
+product between the projected image and text features is then used as a similar score.
+
+To feed images to the Transformer encoder, each image is split into a sequence of fixed-size non-overlapping patches,
+which are then linearly embedded. A [CLS] token is added to serve as representation of an entire image. The authors
+also add absolute position embeddings, and feed the resulting sequence of vectors to a standard Transformer encoder.
+The :class:`~transformers.CLIPFeatureExtractor` can be used to resize (or rescale) and normalize images for the model.
+
+The :class:`~transformers.CLIPTokenizer` is used to encode the text. The :class:`~transformers.CLIPProcessor` wraps
+:class:`~transformers.CLIPFeatureExtractor` and :class:`~transformers.CLIPTokenizer` into a single instance to both
+encode the text and prepare the images. The following example shows how to get the image-text similarity scores using
+:class:`~transformers.CLIPProcessor` and :class:`~transformers.CLIPModel`.
+
+
+.. code-block::
+
+        >>> import torch
+        >>> from PIL import Image
+        >>> import requests
+
+        >>> from transformers import CLIPProcessor, CLIPModel
+
+        >>> model = CLIPModel.from_pretrained("openai/clip-vit-base-patch32")
+        >>> processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")
+
+        >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
+        >>> image = Image.open(requests.get(url, stream=True).raw)
+
+        >>> inputs = processor(text=["a photo of a cat", "a photo of a dog"], images=image, return_tensors="pt", padding=True)
+
+        >>> outputs = model(**inputs)
+        >>> logits_per_image = outputs.logits_per_image # this is the image-text similarity score
+        >>> probs = logits_per_image.softmax(dim=1) # we can take the softmax to get the label probabilities
+
+
+This model was contributed by `valhalla <https://huggingface.co/valhalla>`__. The original code can be found `here
+<https://github.com/openai/CLIP>`__.
+
+CLIPConfig
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.CLIPConfig
+    :members: from_text_vision_configs
+
+
+CLIPTextConfig
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.CLIPTextConfig
+    :members:
+
+
+CLIPVisionConfig
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.CLIPVisionConfig
+    :members:
+
+
+
+CLIPTokenizer
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.CLIPTokenizer
+    :members: build_inputs_with_special_tokens, get_special_tokens_mask,
+        create_token_type_ids_from_sequences, save_vocabulary
+
+CLIPTokenizerFast
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.CLIPTokenizerFast
+    :members:
+
+
+CLIPFeatureExtractor
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.CLIPFeatureExtractor
+    :members:
+
+
+CLIPProcessor
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.CLIPProcessor
+    :members:
+
+
+
+CLIPModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.CLIPModel
+    :members: forward, get_text_features, get_image_features
+
+
+CLIPTextModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.CLIPTextModel
+    :members: forward
+
+
+CLIPVisionModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.CLIPVisionModel
+    :members: forward
+
+
+FlaxCLIPModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxCLIPModel
+    :members: __call__, get_text_features, get_image_features
+
+
+FlaxCLIPTextModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxCLIPTextModel
+    :members: __call__
+
+
+FlaxCLIPVisionModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxCLIPVisionModel
+    :members: __call__

docs/source/model_doc/convbert.rst

@@ -34,8 +34,10 @@ ConvBERT significantly outperforms BERT and its variants in various downstream t
 fewer model parameters. Remarkably, ConvBERTbase model achieves 86.4 GLUE score, 0.7 higher than ELECTRAbase, while
 using less than 1/4 training cost. Code and pre-trained models will be released.*
 
-ConvBERT training tips are similar to those of BERT. The original implementation can be found here:
-https://github.com/yitu-opensource/ConvBert
+ConvBERT training tips are similar to those of BERT.
+
+This model was contributed by `abhishek <https://huggingface.co/abhishek>`__. The original implementation can be found
+here: https://github.com/yitu-opensource/ConvBert
 
 ConvBertConfig
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
@@ -56,8 +58,7 @@ ConvBertTokenizerFast
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 .. autoclass:: transformers.ConvBertTokenizerFast
-    :members: build_inputs_with_special_tokens, get_special_tokens_mask,
-        create_token_type_ids_from_sequences, save_vocabulary
+    :members:
 
 
 ConvBertModel

docs/source/model_doc/cpm.rst

@@ -0,0 +1,45 @@
+..
+    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.
+
+CPM
+-----------------------------------------------------------------------------------------------------------------------
+
+Overview
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The CPM model was proposed in `CPM: A Large-scale Generative Chinese Pre-trained Language Model
+<https://arxiv.org/abs/2012.00413>`__ by Zhengyan Zhang, Xu Han, Hao Zhou, Pei Ke, Yuxian Gu, Deming Ye, Yujia Qin,
+Yusheng Su, Haozhe Ji, Jian Guan, Fanchao Qi, Xiaozhi Wang, Yanan Zheng, Guoyang Zeng, Huanqi Cao, Shengqi Chen,
+Daixuan Li, Zhenbo Sun, Zhiyuan Liu, Minlie Huang, Wentao Han, Jie Tang, Juanzi Li, Xiaoyan Zhu, Maosong Sun.
+
+The abstract from the paper is the following:
+
+*Pre-trained Language Models (PLMs) have proven to be beneficial for various downstream NLP tasks. Recently, GPT-3,
+with 175 billion parameters and 570GB training data, drew a lot of attention due to the capacity of few-shot (even
+zero-shot) learning. However, applying GPT-3 to address Chinese NLP tasks is still challenging, as the training corpus
+of GPT-3 is primarily English, and the parameters are not publicly available. In this technical report, we release the
+Chinese Pre-trained Language Model (CPM) with generative pre-training on large-scale Chinese training data. To the best
+of our knowledge, CPM, with 2.6 billion parameters and 100GB Chinese training data, is the largest Chinese pre-trained
+language model, which could facilitate several downstream Chinese NLP tasks, such as conversation, essay generation,
+cloze test, and language understanding. Extensive experiments demonstrate that CPM achieves strong performance on many
+NLP tasks in the settings of few-shot (even zero-shot) learning.*
+
+This model was contributed by `canwenxu <https://huggingface.co/canwenxu>`__. The original implementation can be found
+here: https://github.com/TsinghuaAI/CPM-Generate
+
+Note: We only have a tokenizer here, since the model architecture is the same as GPT-2.
+
+CpmTokenizer
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.CpmTokenizer
+    :members:

docs/source/model_doc/ctrl.rst

@@ -46,7 +46,8 @@ Tips:
   `reusing the past in generative models <../quickstart.html#using-the-past>`__ for more information on the usage of
   this argument.
 
-The original code can be found `here <https://github.com/salesforce/ctrl>`__.
+This model was contributed by `keskarnitishr <https://huggingface.co/keskarnitishr>`__. The original code can be found
+`here <https://github.com/salesforce/ctrl>`__.
 
 
 CTRLConfig

docs/source/model_doc/deberta.rst

@@ -38,7 +38,8 @@ the training data performs consistently better on a wide range of NLP tasks, ach
 pre-trained models will be made publicly available at https://github.com/microsoft/DeBERTa.*
 
 
-The original code can be found `here <https://github.com/microsoft/DeBERTa>`__.
+This model was contributed by `DeBERTa <https://huggingface.co/DeBERTa>`__. The original code can be found `here
+<https://github.com/microsoft/DeBERTa>`__.
 
 
 DebertaConfig
@@ -55,6 +56,12 @@ DebertaTokenizer
     :members: build_inputs_with_special_tokens, get_special_tokens_mask,
         create_token_type_ids_from_sequences, save_vocabulary
 
+DebertaTokenizerFast
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.DebertaTokenizerFast
+    :members: build_inputs_with_special_tokens, create_token_type_ids_from_sequences
+
 
 DebertaModel
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

docs/source/model_doc/deberta_v2.rst

@@ -58,7 +58,8 @@ New in v2:
 - **900M model & 1.5B model** Two additional model sizes are available: 900M and 1.5B, which significantly improves the
   performance of downstream tasks.
 
-The original code can be found `here <https://github.com/microsoft/DeBERTa>`__.
+This model was contributed by `DeBERTa <https://huggingface.co/DeBERTa>`__. The original code can be found `here
+<https://github.com/microsoft/DeBERTa>`__.
 
 
 DebertaV2Config

docs/source/model_doc/deit.rst

@@ -0,0 +1,111 @@
+.. 
+    Copyright 2021 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.
+
+DeiT
+-----------------------------------------------------------------------------------------------------------------------
+
+.. 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 it in the future. If you see something strange, file a `Github Issue
+    <https://github.com/huggingface/transformers/issues/new?assignees=&labels=&template=bug-report.md&title>`__.
+
+
+Overview
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The DeiT model was proposed in `Training data-efficient image transformers & distillation through attention
+<https://arxiv.org/abs/2012.12877>`__ by Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre
+Sablayrolles, Hervé Jégou. The `Vision Transformer (ViT) <https://huggingface.co/transformers/model_doc/vit.html>`__
+introduced in `Dosovitskiy et al., 2020 <https://arxiv.org/abs/2010.11929>`__ has shown that one can match or even
+outperform existing convolutional neural networks using a Transformer encoder (BERT-like). However, the ViT models
+introduced in that paper required training on expensive infrastructure for multiple weeks, using external data. DeiT
+(data-efficient image transformers) are more efficiently trained transformers for image classification, requiring far
+less data and far less computing resources compared to the original ViT models.
+
+The abstract from the paper is the following:
+
+*Recently, neural networks purely based on attention were shown to address image understanding tasks such as image
+classification. However, these visual transformers are pre-trained with hundreds of millions of images using an
+expensive infrastructure, thereby limiting their adoption. In this work, we produce a competitive convolution-free
+transformer by training on Imagenet only. We train them on a single computer in less than 3 days. Our reference vision
+transformer (86M parameters) achieves top-1 accuracy of 83.1% (single-crop evaluation) on ImageNet with no external
+data. More importantly, we introduce a teacher-student strategy specific to transformers. It relies on a distillation
+token ensuring that the student learns from the teacher through attention. We show the interest of this token-based
+distillation, especially when using a convnet as a teacher. This leads us to report results competitive with convnets
+for both Imagenet (where we obtain up to 85.2% accuracy) and when transferring to other tasks. We share our code and
+models.*
+
+Tips:
+
+- Compared to ViT, DeiT models use a so-called distillation token to effectively learn from a teacher (which, in the
+  DeiT paper, is a ResNet like-model). The distillation token is learned through backpropagation, by interacting with
+  the class ([CLS]) and patch tokens through the self-attention layers.
+- There are 2 ways to fine-tune distilled models, either (1) in a classic way, by only placing a prediction head on top
+  of the final hidden state of the class token and not using the distillation signal, or (2) by placing both a
+  prediction head on top of the class token and on top of the distillation token. In that case, the [CLS] prediction
+  head is trained using regular cross-entropy between the prediction of the head and the ground-truth label, while the
+  distillation prediction head is trained using hard distillation (cross-entropy between the prediction of the
+  distillation head and the label predicted by the teacher). At inference time, one takes the average prediction
+  between both heads as final prediction. (2) is also called "fine-tuning with distillation", because one relies on a
+  teacher that has already been fine-tuned on the downstream dataset. In terms of models, (1) corresponds to
+  :class:`~transformers.DeiTForImageClassification` and (2) corresponds to
+  :class:`~transformers.DeiTForImageClassificationWithTeacher`.
+- Note that the authors also did try soft distillation for (2) (in which case the distillation prediction head is
+  trained using KL divergence to match the softmax output of the teacher), but hard distillation gave the best results.
+- All released checkpoints were pre-trained and fine-tuned on ImageNet-1k only. No external data was used. This is in
+  contrast with the original ViT model, which used external data like the JFT-300M dataset/Imagenet-21k for
+  pre-training.
+- The authors of DeiT also released more efficiently trained ViT models, which you can directly plug into
+  :class:`~transformers.ViTModel` or :class:`~transformers.ViTForImageClassification`. Techniques like data
+  augmentation, optimization, and regularization were used in order to simulate training on a much larger dataset
+  (while only using ImageNet-1k for pre-training). There are 4 variants available (in 3 different sizes):
+  `facebook/deit-tiny-patch16-224`, `facebook/deit-small-patch16-224`, `facebook/deit-base-patch16-224` and
+  `facebook/deit-base-patch16-384`. Note that one should use :class:`~transformers.DeiTFeatureExtractor` in order to
+  prepare images for the model.
+
+This model was contributed by `nielsr <https://huggingface.co/nielsr>`__.
+
+
+DeiTConfig
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.DeiTConfig
+    :members:
+
+
+DeiTFeatureExtractor
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.DeiTFeatureExtractor
+    :members: __call__
+
+
+DeiTModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.DeiTModel
+    :members: forward
+
+
+DeiTForImageClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.DeiTForImageClassification
+    :members: forward
+
+
+DeiTForImageClassificationWithTeacher
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.DeiTForImageClassificationWithTeacher
+    :members: forward

docs/source/model_doc/detr.rst

@@ -0,0 +1,207 @@
+.. 
+    Copyright 2021 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.
+
+DETR
+-----------------------------------------------------------------------------------------------------------------------
+
+Overview
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The DETR model was proposed in `End-to-End Object Detection with Transformers <https://arxiv.org/abs/2005.12872>`__ by
+Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov and Sergey Zagoruyko. DETR
+consists of a convolutional backbone followed by an encoder-decoder Transformer which can be trained end-to-end for
+object detection. It greatly simplifies a lot of the complexity of models like Faster-R-CNN and Mask-R-CNN, which use
+things like region proposals, non-maximum suppression procedure and anchor generation. Moreover, DETR can also be
+naturally extended to perform panoptic segmentation, by simply adding a mask head on top of the decoder outputs.
+
+The abstract from the paper is the following:
+
+*We present a new method that views object detection as a direct set prediction problem. Our approach streamlines the
+detection pipeline, effectively removing the need for many hand-designed components like a non-maximum suppression
+procedure or anchor generation that explicitly encode our prior knowledge about the task. The main ingredients of the
+new framework, called DEtection TRansformer or DETR, are a set-based global loss that forces unique predictions via
+bipartite matching, and a transformer encoder-decoder architecture. Given a fixed small set of learned object queries,
+DETR reasons about the relations of the objects and the global image context to directly output the final set of
+predictions in parallel. The new model is conceptually simple and does not require a specialized library, unlike many
+other modern detectors. DETR demonstrates accuracy and run-time performance on par with the well-established and
+highly-optimized Faster RCNN baseline on the challenging COCO object detection dataset. Moreover, DETR can be easily
+generalized to produce panoptic segmentation in a unified manner. We show that it significantly outperforms competitive
+baselines.*
+
+This model was contributed by `nielsr <https://huggingface.co/nielsr>`__. The original code can be found `here
+<https://github.com/facebookresearch/detr>`__.
+
+The quickest way to get started with DETR is by checking the `example notebooks
+<https://github.com/NielsRogge/Transformers-Tutorials/tree/master/DETR>`__ (which showcase both inference and
+fine-tuning on custom data).
+
+Here's a TLDR explaining how :class:`~transformers.DetrForObjectDetection` works:
+
+First, an image is sent through a pre-trained convolutional backbone (in the paper, the authors use
+ResNet-50/ResNet-101). Let's assume we also add a batch dimension. This means that the input to the backbone is a
+tensor of shape :obj:`(batch_size, 3, height, width)`, assuming the image has 3 color channels (RGB). The CNN backbone
+outputs a new lower-resolution feature map, typically of shape :obj:`(batch_size, 2048, height/32, width/32)`. This is
+then projected to match the hidden dimension of the Transformer of DETR, which is :obj:`256` by default, using a
+:obj:`nn.Conv2D` layer. So now, we have a tensor of shape :obj:`(batch_size, 256, height/32, width/32).` Next, the
+feature map is flattened and transposed to obtain a tensor of shape :obj:`(batch_size, seq_len, d_model)` =
+:obj:`(batch_size, width/32*height/32, 256)`. So a difference with NLP models is that the sequence length is actually
+longer than usual, but with a smaller :obj:`d_model` (which in NLP is typically 768 or higher).
+
+Next, this is sent through the encoder, outputting :obj:`encoder_hidden_states` of the same shape (you can consider
+these as image features). Next, so-called **object queries** are sent through the decoder. This is a tensor of shape
+:obj:`(batch_size, num_queries, d_model)`, with :obj:`num_queries` typically set to 100 and initialized with zeros.
+These input embeddings are learnt positional encodings that the authors refer to as object queries, and similarly to
+the encoder, they are added to the input of each attention layer. Each object query will look for a particular object
+in the image. The decoder updates these embeddings through multiple self-attention and encoder-decoder attention layers
+to output :obj:`decoder_hidden_states` of the same shape: :obj:`(batch_size, num_queries, d_model)`. Next, two heads
+are added on top for object detection: a linear layer for classifying each object query into one of the objects or "no
+object", and a MLP to predict bounding boxes for each query.
+
+The model is trained using a **bipartite matching loss**: so what we actually do is compare the predicted classes +
+bounding boxes of each of the N = 100 object queries to the ground truth annotations, padded up to the same length N
+(so if an image only contains 4 objects, 96 annotations will just have a "no object" as class and "no bounding box" as
+bounding box). The `Hungarian matching algorithm <https://en.wikipedia.org/wiki/Hungarian_algorithm>`__ is used to find
+an optimal one-to-one mapping of each of the N queries to each of the N annotations. Next, standard cross-entropy (for
+the classes) and a linear combination of the L1 and `generalized IoU loss <https://giou.stanford.edu/>`__ (for the
+bounding boxes) are used to optimize the parameters of the model.
+
+DETR can be naturally extended to perform panoptic segmentation (which unifies semantic segmentation and instance
+segmentation). :class:`~transformers.DetrForSegmentation` adds a segmentation mask head on top of
+:class:`~transformers.DetrForObjectDetection`. The mask head can be trained either jointly, or in a two steps process,
+where one first trains a :class:`~transformers.DetrForObjectDetection` model to detect bounding boxes around both
+"things" (instances) and "stuff" (background things like trees, roads, sky), then freeze all the weights and train only
+the mask head for 25 epochs. Experimentally, these two approaches give similar results. Note that predicting boxes is
+required for the training to be possible, since the Hungarian matching is computed using distances between boxes.
+
+Tips:
+
+- DETR uses so-called **object queries** to detect objects in an image. The number of queries determines the maximum
+  number of objects that can be detected in a single image, and is set to 100 by default (see parameter
+  :obj:`num_queries` of :class:`~transformers.DetrConfig`). Note that it's good to have some slack (in COCO, the
+  authors used 100, while the maximum number of objects in a COCO image is ~70).
+- The decoder of DETR updates the query embeddings in parallel. This is different from language models like GPT-2,
+  which use autoregressive decoding instead of parallel. Hence, no causal attention mask is used.
+- DETR adds position embeddings to the hidden states at each self-attention and cross-attention layer before projecting
+  to queries and keys. For the position embeddings of the image, one can choose between fixed sinusoidal or learned
+  absolute position embeddings. By default, the parameter :obj:`position_embedding_type` of
+  :class:`~transformers.DetrConfig` is set to :obj:`"sine"`.
+- During training, the authors of DETR did find it helpful to use auxiliary losses in the decoder, especially to help
+  the model output the correct number of objects of each class. If you set the parameter :obj:`auxiliary_loss` of
+  :class:`~transformers.DetrConfig` to :obj:`True`, then prediction feedforward neural networks and Hungarian losses
+  are added after each decoder layer (with the FFNs sharing parameters).
+- If you want to train the model in a distributed environment across multiple nodes, then one should update the
+  `num_boxes` variable in the `DetrLoss` class of `modeling_detr.py`. When training on multiple nodes, this should be
+  set to the average number of target boxes across all nodes, as can be seen in the original implementation `here
+  <https://github.com/facebookresearch/detr/blob/a54b77800eb8e64e3ad0d8237789fcbf2f8350c5/models/detr.py#L227-L232>`__.
+- :class:`~transformers.DetrForObjectDetection` and :class:`~transformers.DetrForSegmentation` can be initialized with
+  any convolutional backbone available in the `timm library <https://github.com/rwightman/pytorch-image-models>`__.
+  Initializing with a MobileNet backbone for example can be done by setting the :obj:`backbone` attribute of
+  :class:`~transformers.DetrConfig` to :obj:`"tf_mobilenetv3_small_075"`, and then initializing the model with that
+  config.
+- DETR resizes the input images such that the shortest side is at least a certain amount of pixels while the longest is
+  at most 1333 pixels. At training time, scale augmentation is used such that the shortest side is randomly set to at
+  least 480 and at most 800 pixels. At inference time, the shortest side is set to 800. One can use
+  :class:`~transformers.DetrFeatureExtractor` to prepare images (and optional annotations in COCO format) for the
+  model. Due to this resizing, images in a batch can have different sizes. DETR solves this by padding images up to the
+  largest size in a batch, and by creating a pixel mask that indicates which pixels are real/which are padding.
+  Alternatively, one can also define a custom :obj:`collate_fn` in order to batch images together, using
+  :meth:`~transformers.DetrFeatureExtractor.pad_and_create_pixel_mask`.
+- The size of the images will determine the amount of memory being used, and will thus determine the :obj:`batch_size`.
+  It is advised to use a batch size of 2 per GPU. See `this Github thread
+  <https://github.com/facebookresearch/detr/issues/150>`__ for more info.
+
+As a summary, consider the following table:
+
++---------------------------------------------+---------------------------------------------------------+----------------------------------------------------------------------+------------------------------------------------------------------------+
+| **Task**                                    | **Object detection**                                    | **Instance segmentation**                                            | **Panoptic segmentation**                                              |
++---------------------------------------------+---------------------------------------------------------+----------------------------------------------------------------------+------------------------------------------------------------------------+
+| **Description**                             | Predicting bounding boxes and class labels around       | Predicting masks around objects (i.e. instances) in an image         | Predicting masks around both objects (i.e. instances) as well as       |
+|                                             | objects in an image                                     |                                                                      | "stuff" (i.e. background things like trees and roads) in an image      |
++---------------------------------------------+---------------------------------------------------------+----------------------------------------------------------------------+------------------------------------------------------------------------+
+| **Model**                                   | :class:`~transformers.DetrForObjectDetection`           | :class:`~transformers.DetrForSegmentation`                           | :class:`~transformers.DetrForSegmentation`                             |
++---------------------------------------------+---------------------------------------------------------+----------------------------------------------------------------------+------------------------------------------------------------------------+
+| **Example dataset**                         | COCO detection                                          | COCO detection,                                                      | COCO panoptic                                                          |
+|                                             |                                                         | COCO panoptic                                                        |                                                                        |
++---------------------------------------------+---------------------------------------------------------+----------------------------------------------------------------------+------------------------------------------------------------------------+
+| **Format of annotations to provide to**     | {‘image_id’: int,                                       | {‘image_id’: int,                                                    | {‘file_name: str,                                                      |
+| :class:`~transformers.DetrFeatureExtractor` | ‘annotations’: List[Dict]}, each Dict being a COCO      | ‘annotations’: [List[Dict]] } (in case of COCO detection)            | ‘image_id: int,                                                        |
+|                                             | object annotation                                       |                                                                      | ‘segments_info’: List[Dict] }                                          |
+|                                             |                                                         | or                                                                   |                                                                        |
+|                                             |                                                         |                                                                      | and masks_path (path to directory containing PNG files of the masks)   |
+|                                             |                                                         | {‘file_name’: str,                                                   |                                                                        |
+|                                             |                                                         | ‘image_id’: int,                                                     |                                                                        |
+|                                             |                                                         | ‘segments_info’: List[Dict]} (in case of COCO panoptic)              |                                                                        |
++---------------------------------------------+---------------------------------------------------------+----------------------------------------------------------------------+------------------------------------------------------------------------+
+| **Postprocessing** (i.e. converting the     | :meth:`~transformers.DetrFeatureExtractor.post_process` | :meth:`~transformers.DetrFeatureExtractor.post_process_segmentation` | :meth:`~transformers.DetrFeatureExtractor.post_process_segmentation`,  |
+| output of the model to COCO API)            |                                                         |                                                                      | :meth:`~transformers.DetrFeatureExtractor.post_process_panoptic`       |
++---------------------------------------------+---------------------------------------------------------+----------------------------------------------------------------------+------------------------------------------------------------------------+
+| **evaluators**                              | :obj:`CocoEvaluator` with iou_types = “bbox”            | :obj:`CocoEvaluator` with iou_types = “bbox”, “segm”                 | :obj:`CocoEvaluator` with iou_tupes = “bbox, “segm”                    |
+|                                             |                                                         |                                                                      |                                                                        |
+|                                             |                                                         |                                                                      | :obj:`PanopticEvaluator`                                               |
++---------------------------------------------+---------------------------------------------------------+----------------------------------------------------------------------+------------------------------------------------------------------------+
+
+In short, one should prepare the data either in COCO detection or COCO panoptic format, then use
+:class:`~transformers.DetrFeatureExtractor` to create :obj:`pixel_values`, :obj:`pixel_mask` and optional
+:obj:`labels`, which can then be used to train (or fine-tune) a model. For evaluation, one should first convert the
+outputs of the model using one of the postprocessing methods of :class:`~transformers.DetrFeatureExtractor`. These can
+be be provided to either :obj:`CocoEvaluator` or :obj:`PanopticEvaluator`, which allow you to calculate metrics like
+mean Average Precision (mAP) and Panoptic Quality (PQ). The latter objects are implemented in the `original repository
+<https://github.com/facebookresearch/detr>`__. See the `example notebooks
+<https://github.com/NielsRogge/Transformers-Tutorials/tree/master/DETR>`__ for more info regarding evaluation.
+
+
+DETR specific outputs
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.models.detr.modeling_detr.DetrModelOutput
+    :members:
+
+.. autoclass:: transformers.models.detr.modeling_detr.DetrObjectDetectionOutput
+    :members:
+
+.. autoclass:: transformers.models.detr.modeling_detr.DetrSegmentationOutput
+    :members:
+
+
+DetrConfig
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.DetrConfig
+    :members:
+
+
+DetrFeatureExtractor
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.DetrFeatureExtractor
+    :members: __call__, pad_and_create_pixel_mask, post_process, post_process_segmentation, post_process_panoptic
+
+
+DetrModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.DetrModel
+    :members: forward
+
+
+DetrForObjectDetection
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.DetrForObjectDetection
+    :members: forward
+
+
+DetrForSegmentation
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.DetrForSegmentation
+    :members: forward

docs/source/model_doc/distilbert.rst

@@ -44,8 +44,8 @@ Tips:
 - DistilBERT doesn't have options to select the input positions (:obj:`position_ids` input). This could be added if
   necessary though, just let us know if you need this option.
 
-The original code can be found `here
-<https://github.com/huggingface/transformers/tree/master/examples/distillation>`__.
+This model was contributed by `victorsanh <https://huggingface.co/victorsanh>`__. The original code can be found
+:prefix_link:`here <examples/research-projects/distillation>`.
 
 
 DistilBertConfig

docs/source/model_doc/dpr.rst

@@ -30,7 +30,8 @@ our dense retriever outperforms a strong Lucene-BM25 system largely by 9%-19% ab
 retrieval accuracy, and helps our end-to-end QA system establish new state-of-the-art on multiple open-domain QA
 benchmarks.*
 
-The original code can be found `here <https://github.com/facebookresearch/DPR>`__.
+This model was contributed by `lhoestq <https://huggingface.co/lhoestq>`__. The original code can be found `here
+<https://github.com/facebookresearch/DPR>`__.
 
 
 DPRConfig

docs/source/model_doc/electra.rst

@@ -54,7 +54,8 @@ Tips:
   :class:`~transformers.ElectraForPreTraining` model (the classification head will be randomly initialized as it
   doesn't exist in the generator).
 
-The original code can be found `here <https://github.com/google-research/electra>`__.
+This model was contributed by `lysandre <https://huggingface.co/lysandre>`__. The original code can be found `here
+<https://github.com/google-research/electra>`__.
 
 
 ElectraConfig
@@ -184,3 +185,52 @@ TFElectraForQuestionAnswering
 
 .. autoclass:: transformers.TFElectraForQuestionAnswering
     :members: call
+
+
+FlaxElectraModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxElectraModel
+    :members: __call__
+
+
+FlaxElectraForPreTraining
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxElectraForPreTraining
+    :members: __call__
+
+
+FlaxElectraForMaskedLM
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxElectraForMaskedLM
+    :members: __call__
+
+
+FlaxElectraForSequenceClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxElectraForSequenceClassification
+    :members: __call__
+
+
+FlaxElectraForMultipleChoice
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxElectraForMultipleChoice
+    :members: __call__
+
+
+FlaxElectraForTokenClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxElectraForTokenClassification
+    :members: __call__
+
+
+FlaxElectraForQuestionAnswering
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxElectraForQuestionAnswering
+    :members: __call__

docs/source/model_doc/flaubert.rst

@@ -35,7 +35,8 @@ time they outperform other pretraining approaches. Different versions of FlauBER
 protocol for the downstream tasks, called FLUE (French Language Understanding Evaluation), are shared to the research
 community for further reproducible experiments in French NLP.*
 
-The original code can be found `here <https://github.com/getalp/Flaubert>`__.
+This model was contributed by `formiel <https://huggingface.co/formiel>`__. The original code can be found `here
+<https://github.com/getalp/Flaubert>`__.
 
 
 FlaubertConfig

docs/source/model_doc/fsmt.rst

@@ -34,7 +34,8 @@ data, then decode using noisy channel model reranking. Our submissions are ranke
 human evaluation campaign. On En->De, our system significantly outperforms other systems as well as human translations.
 This system improves upon our WMT'18 submission by 4.5 BLEU points.*
 
-The original code can be found here <https://github.com/pytorch/fairseq/tree/master/examples/wmt19>__.
+This model was contributed by `stas <https://huggingface.co/stas>`__. The original code can be found here
+<https://github.com/pytorch/fairseq/tree/master/examples/wmt19>__.
 
 Implementation Notes
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

docs/source/model_doc/funnel.rst

@@ -49,7 +49,8 @@ Tips:
   :class:`~transformers.FunnelBaseModel`, :class:`~transformers.FunnelForSequenceClassification` and
   :class:`~transformers.FunnelForMultipleChoice`.
 
-The original code can be found `here <https://github.com/laiguokun/Funnel-Transformer>`__.
+This model was contributed by `sgugger <https://huggingface.co/sgugger>`__. The original code can be found `here
+<https://github.com/laiguokun/Funnel-Transformer>`__.
 
 
 FunnelConfig

docs/source/model_doc/gpt.rst

@@ -45,12 +45,13 @@ Tips:
 `Write With Transformer <https://transformer.huggingface.co/doc/gpt>`__ is a webapp created and hosted by Hugging Face
 showcasing the generative capabilities of several models. GPT is one of them.
 
-The original code can be found `here <https://github.com/openai/finetune-transformer-lm>`__.
+This model was contributed by `thomwolf <https://huggingface.co/thomwolf>`__. The original code can be found `here
+<https://github.com/openai/finetune-transformer-lm>`__.
 
 Note:
 
 If you want to reproduce the original tokenization process of the `OpenAI GPT` paper, you will need to install ``ftfy``
-and ``SpaCy``::
+and ``SpaCy``:
 
 .. code-block:: bash
 

docs/source/model_doc/gpt2.rst

@@ -45,7 +45,8 @@ Tips:
 Hugging Face showcasing the generative capabilities of several models. GPT-2 is one of them and is available in five
 different sizes: small, medium, large, xl and a distilled version of the small checkpoint: `distilgpt-2`.
 
-The original code can be found `here <https://openai.com/blog/better-language-models/>`__.
+This model was contributed by `thomwolf <https://huggingface.co/thomwolf>`__. The original code can be found `here
+<https://openai.com/blog/better-language-models/>`__.
 
 
 GPT2Config
@@ -138,3 +139,17 @@ TFSequenceClassifierOutputWithPast
 
 .. autoclass:: transformers.modeling_tf_outputs.TFSequenceClassifierOutputWithPast
     :members:
+
+
+FlaxGPT2Model
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxGPT2Model
+    :members: __call__
+
+
+FlaxGPT2LMHeadModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxGPT2LMHeadModel
+    :members: __call__

docs/source/model_doc/gpt_neo.rst

@@ -0,0 +1,73 @@
+.. 
+    Copyright 2021 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.
+
+GPT Neo
+-----------------------------------------------------------------------------------------------------------------------
+
+Overview
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The GPTNeo model was released in the `EleutherAI/gpt-neo <https://github.com/EleutherAI/gpt-neo>`__ repository by Sid
+Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy. It is a GPT2 like causal language model trained on the
+`Pile <https://pile.eleuther.ai/>`__ dataset.
+
+The architecture is similar to GPT2 except that GPT Neo uses local attention in every other layer with a window size of
+256 tokens.
+
+This model was contributed by `valhalla <https://huggingface.co/valhalla>`__.
+
+Generation
+_______________________________________________________________________________________________________________________
+
+The :obj:`generate()` method can be used to generate text using GPT Neo model.
+
+.. code-block::
+
+    >>> from transformers import GPTNeoForCausalLM, GPT2Tokenizer
+    >>> model = GPTNeoForCausalLM.from_pretrained("EleutherAI/gpt-neo-1.3B")
+    >>> tokenizer = GPT2Tokenizer.from_pretrained("EleutherAI/gpt-neo-1.3B")
+
+    >>> prompt = "In a shocking finding, scientists discovered a herd of unicorns living in a remote, " \
+    ...          "previously unexplored valley, in the Andes Mountains. Even more surprising to the " \
+    ...          "researchers was the fact that the unicorns spoke perfect English."
+
+    >>> input_ids = tokenizer(prompt, return_tensors="pt").input_ids
+
+    >>> gen_tokens = model.generate(input_ids, do_sample=True, temperature=0.9, max_length=100,)
+    >>> gen_text = tokenizer.batch_decode(gen_tokens)[0]
+
+
+GPTNeoConfig
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.GPTNeoConfig
+    :members:
+
+
+GPTNeoModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.GPTNeoModel
+    :members: forward
+
+
+GPTNeoForCausalLM
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.GPTNeoForCausalLM
+    :members: forward
+
+GPTNeoForSequenceClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.GPTNeoForSequenceClassification
+    :members: forward

docs/source/model_doc/herbert.rst

@@ -40,23 +40,25 @@ Examples of use:
 
 .. code-block::
 
-  from transformers import HerbertTokenizer, RobertaModel
+    >>> from transformers import HerbertTokenizer, RobertaModel
 
-  tokenizer = HerbertTokenizer.from_pretrained("allegro/herbert-klej-cased-tokenizer-v1")
-  model = RobertaModel.from_pretrained("allegro/herbert-klej-cased-v1")
+    >>> tokenizer = HerbertTokenizer.from_pretrained("allegro/herbert-klej-cased-tokenizer-v1")
+    >>> model = RobertaModel.from_pretrained("allegro/herbert-klej-cased-v1")
 
-  encoded_input = tokenizer.encode("Kto ma lepszą sztukę, ma lepszy rząd – to jasne.", return_tensors='pt')
-  outputs = model(encoded_input)
+    >>> encoded_input = tokenizer.encode("Kto ma lepszą sztukę, ma lepszy rząd – to jasne.", return_tensors='pt')
+    >>> outputs = model(encoded_input)
 
-  # HerBERT can also be loaded using AutoTokenizer and AutoModel:
-  import torch
-  from transformers import AutoModel, AutoTokenizer
+    >>> # HerBERT can also be loaded using AutoTokenizer and AutoModel:
+    >>> import torch
+    >>> from transformers import AutoModel, AutoTokenizer
 
-  tokenizer = AutoTokenizer.from_pretrained("allegro/herbert-klej-cased-tokenizer-v1")
-  model = AutoModel.from_pretrained("allegro/herbert-klej-cased-v1")
+    >>> tokenizer = AutoTokenizer.from_pretrained("allegro/herbert-klej-cased-tokenizer-v1")
+    >>> model = AutoModel.from_pretrained("allegro/herbert-klej-cased-v1")
 
 
-The original code can be found `here <https://github.com/allegro/HerBERT>`__.
+This model was contributed by `rmroczkowski <https://huggingface.co/rmroczkowski>`__. The original code can be found
+`here <https://github.com/allegro/HerBERT>`__.
+
 
 HerbertTokenizer
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

docs/source/model_doc/hubert.rst

@@ -0,0 +1,65 @@
+.. 
+    Copyright 2021 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.
+
+Hubert
+-----------------------------------------------------------------------------------------------------------------------
+
+Overview
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+Hubert was proposed in `HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units
+<https://arxiv.org/abs/2106.07447>`__ by Wei-Ning Hsu, Benjamin Bolte, Yao-Hung Hubert Tsai, Kushal Lakhotia, Ruslan
+Salakhutdinov, Abdelrahman Mohamed.
+
+The abstract from the paper is the following:
+
+*Self-supervised approaches for speech representation learning are challenged by three unique problems: (1) there are
+multiple sound units in each input utterance, (2) there is no lexicon of input sound units during the pre-training
+phase, and (3) sound units have variable lengths with no explicit segmentation. To deal with these three problems, we
+propose the Hidden-Unit BERT (HuBERT) approach for self-supervised speech representation learning, which utilizes an
+offline clustering step to provide aligned target labels for a BERT-like prediction loss. A key ingredient of our
+approach is applying the prediction loss over the masked regions only, which forces the model to learn a combined
+acoustic and language model over the continuous inputs. HuBERT relies primarily on the consistency of the unsupervised
+clustering step rather than the intrinsic quality of the assigned cluster labels. Starting with a simple k-means
+teacher of 100 clusters, and using two iterations of clustering, the HuBERT model either matches or improves upon the
+state-of-the-art wav2vec 2.0 performance on the Librispeech (960h) and Libri-light (60,000h) benchmarks with 10min, 1h,
+10h, 100h, and 960h fine-tuning subsets. Using a 1B parameter model, HuBERT shows up to 19% and 13% relative WER
+reduction on the more challenging dev-other and test-other evaluation subsets.*
+
+Tips:
+
+- Hubert is a speech model that accepts a float array corresponding to the raw waveform of the speech signal.
+- Hubert model was fine-tuned using connectionist temporal classification (CTC) so the model output has to be decoded
+  using :class:`~transformers.Wav2Vec2CTCTokenizer`.
+
+This model was contributed by `patrickvonplaten <https://huggingface.co/patrickvonplaten>`__.
+
+
+HubertConfig
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.HubertConfig
+    :members:
+
+
+HubertModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.HubertModel
+    :members: forward
+
+
+HubertForCTC
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.HubertForCTC
+    :members: forward

docs/source/model_doc/ibert.rst

@@ -36,8 +36,9 @@ the full-precision baseline. Furthermore, our preliminary implementation of I-BE
 INT8 inference on a T4 GPU system as compared to FP32 inference. The framework has been developed in PyTorch and has
 been open-sourced.*
 
+This model was contributed by `kssteven <https://huggingface.co/kssteven>`__. The original code can be found `here
+<https://github.com/kssteven418/I-BERT>`__.
 
-The original code can be found `here <https://github.com/kssteven418/I-BERT>`__.
 
 IBertConfig
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

docs/source/model_doc/layoutlm.rst

@@ -56,31 +56,32 @@ Tips:
 
 .. code-block::
 
-   def normalize_bbox(bbox, width, height):
-        return [
-            int(1000 * (bbox[0] / width)),
-            int(1000 * (bbox[1] / height)),
-            int(1000 * (bbox[2] / width)),
-            int(1000 * (bbox[3] / height)),
-        ]
+    def normalize_bbox(bbox, width, height):
+         return [
+             int(1000 * (bbox[0] / width)),
+             int(1000 * (bbox[1] / height)),
+             int(1000 * (bbox[2] / width)),
+             int(1000 * (bbox[3] / height)),
+         ]
 
 Here, :obj:`width` and :obj:`height` correspond to the width and height of the original document in which the token
 occurs. Those can be obtained using the Python Image Library (PIL) library for example, as follows:
 
 .. code-block::
 
-   from PIL import Image
+    from PIL import Image
 
-   image = Image.open("name_of_your_document - can be a png file, pdf, etc.")
+    image = Image.open("name_of_your_document - can be a png file, pdf, etc.")
 
-   width, height = image.size
+    width, height = image.size
 
 - For a demo which shows how to fine-tune :class:`LayoutLMForTokenClassification` on the `FUNSD dataset
   <https://guillaumejaume.github.io/FUNSD/>`__ (a collection of annotated forms), see `this notebook
   <https://github.com/NielsRogge/Transformers-Tutorials/blob/master/LayoutLM/Fine_tuning_LayoutLMForTokenClassification_on_FUNSD.ipynb>`__.
   It includes an inference part, which shows how to use Google's Tesseract on a new document.
 
-The original code can be found `here <https://github.com/microsoft/unilm/tree/master/layoutlm>`_.
+This model was contributed by `liminghao1630 <https://huggingface.co/liminghao1630>`__. The original code can be found
+`here <https://github.com/microsoft/unilm/tree/master/layoutlm>`_.
 
 
 LayoutLMConfig
@@ -130,3 +131,31 @@ LayoutLMForTokenClassification
 
 .. autoclass:: transformers.LayoutLMForTokenClassification
     :members:
+
+
+TFLayoutLMModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFLayoutLMModel
+    :members:
+
+
+TFLayoutLMForMaskedLM
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFLayoutLMForMaskedLM
+    :members:
+
+
+TFLayoutLMForSequenceClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFLayoutLMForSequenceClassification
+    :members:
+
+
+TFLayoutLMForTokenClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFLayoutLMForTokenClassification
+    :members:

docs/source/model_doc/led.rst

@@ -53,6 +53,8 @@ Tips:
 - A notebook showing how to fine-tune LED, can be accessed `here
   <https://colab.research.google.com/drive/12LjJazBl7Gam0XBPy_y0CTOJZeZ34c2v?usp=sharing>`__.
 
+This model was contributed by `patrickvonplaten <https://huggingface.co/patrickvonplaten>`__.
+
 
 LEDConfig
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
@@ -73,8 +75,7 @@ LEDTokenizerFast
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 .. autoclass:: transformers.LEDTokenizerFast
-    :members: build_inputs_with_special_tokens, get_special_tokens_mask,
-        create_token_type_ids_from_sequences, save_vocabulary
+    :members:
 
 
 LED specific outputs

docs/source/model_doc/longformer.rst

@@ -40,7 +40,8 @@ Tips:
   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>`__.
+This model was contributed by `beltagy <https://huggingface.co/beltagy>`__. The Authors' code can be found `here
+<https://github.com/allenai/longformer>`__.
 
 Longformer Self Attention
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

docs/source/model_doc/luke.rst

@@ -0,0 +1,159 @@
+..
+    Copyright 2021 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.
+
+LUKE
+-----------------------------------------------------------------------------------------------------------------------
+
+Overview
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The LUKE model was proposed in `LUKE: Deep Contextualized Entity Representations with Entity-aware Self-attention
+<https://arxiv.org/abs/2010.01057>`_ by Ikuya Yamada, Akari Asai, Hiroyuki Shindo, Hideaki Takeda and Yuji Matsumoto.
+It is based on RoBERTa and adds entity embeddings as well as an entity-aware self-attention mechanism, which helps
+improve performance on various downstream tasks involving reasoning about entities such as named entity recognition,
+extractive and cloze-style question answering, entity typing, and relation classification.
+
+The abstract from the paper is the following:
+
+*Entity representations are useful in natural language tasks involving entities. In this paper, we propose new
+pretrained contextualized representations of words and entities based on the bidirectional transformer. The proposed
+model treats words and entities in a given text as independent tokens, and outputs contextualized representations of
+them. Our model is trained using a new pretraining task based on the masked language model of BERT. The task involves
+predicting randomly masked words and entities in a large entity-annotated corpus retrieved from Wikipedia. We also
+propose an entity-aware self-attention mechanism that is an extension of the self-attention mechanism of the
+transformer, and considers the types of tokens (words or entities) when computing attention scores. The proposed model
+achieves impressive empirical performance on a wide range of entity-related tasks. In particular, it obtains
+state-of-the-art results on five well-known datasets: Open Entity (entity typing), TACRED (relation classification),
+CoNLL-2003 (named entity recognition), ReCoRD (cloze-style question answering), and SQuAD 1.1 (extractive question
+answering).*
+
+Tips:
+
+- This implementation is the same as :class:`~transformers.RobertaModel` with the addition of entity embeddings as well
+  as an entity-aware self-attention mechanism, which improves performance on tasks involving reasoning about entities.
+- LUKE treats entities as input tokens; therefore, it takes :obj:`entity_ids`, :obj:`entity_attention_mask`,
+  :obj:`entity_token_type_ids` and :obj:`entity_position_ids` as extra input. You can obtain those using
+  :class:`~transformers.LukeTokenizer`.
+- :class:`~transformers.LukeTokenizer` takes :obj:`entities` and :obj:`entity_spans` (character-based start and end
+  positions of the entities in the input text) as extra input. :obj:`entities` typically consist of [MASK] entities or
+  Wikipedia entities. The brief description when inputting these entities are as follows:
+
+  - *Inputting [MASK] entities to compute entity representations*: The [MASK] entity is used to mask entities to be
+    predicted during pretraining. When LUKE receives the [MASK] entity, it tries to predict the original entity by
+    gathering the information about the entity from the input text. Therefore, the [MASK] entity can be used to address
+    downstream tasks requiring the information of entities in text such as entity typing, relation classification, and
+    named entity recognition.
+  - *Inputting Wikipedia entities to compute knowledge-enhanced token representations*: LUKE learns rich information
+    (or knowledge) about Wikipedia entities during pretraining and stores the information in its entity embedding. By
+    using Wikipedia entities as input tokens, LUKE outputs token representations enriched by the information stored in
+    the embeddings of these entities. This is particularly effective for tasks requiring real-world knowledge, such as
+    question answering.
+
+- There are three head models for the former use case:
+
+  - :class:`~transformers.LukeForEntityClassification`, for tasks to classify a single entity in an input text such as
+    entity typing, e.g. the `Open Entity dataset <https://www.cs.utexas.edu/~eunsol/html_pages/open_entity.html>`__.
+    This model places a linear head on top of the output entity representation.
+  - :class:`~transformers.LukeForEntityPairClassification`, for tasks to classify the relationship between two entities
+    such as relation classification, e.g. the `TACRED dataset <https://nlp.stanford.edu/projects/tacred/>`__. This
+    model places a linear head on top of the concatenated output representation of the pair of given entities.
+  - :class:`~transformers.LukeForEntitySpanClassification`, for tasks to classify the sequence of entity spans, such as
+    named entity recognition (NER). This model places a linear head on top of the output entity representations. You
+    can address NER using this model by inputting all possible entity spans in the text to the model.
+
+  :class:`~transformers.LukeTokenizer` has a ``task`` argument, which enables you to easily create an input to these
+  head models by specifying ``task="entity_classification"``, ``task="entity_pair_classification"``, or
+  ``task="entity_span_classification"``. Please refer to the example code of each head models.
+
+  There are also 3 notebooks available, which showcase how you can reproduce the results as reported in the paper with
+  the HuggingFace implementation of LUKE. They can be found `here
+  <https://github.com/studio-ousia/luke/tree/master/notebooks>`__.
+
+Example:
+
+.. code-block::
+
+    >>> from transformers import LukeTokenizer, LukeModel, LukeForEntityPairClassification
+
+    >>> model = LukeModel.from_pretrained("studio-ousia/luke-base")
+    >>> tokenizer = LukeTokenizer.from_pretrained("studio-ousia/luke-base")
+
+    # Example 1: Computing the contextualized entity representation corresponding to the entity mention "Beyoncé"
+    >>> text = "Beyoncé lives in Los Angeles."
+    >>> entity_spans = [(0, 7)]  # character-based entity span corresponding to "Beyoncé"
+    >>> inputs = tokenizer(text, entity_spans=entity_spans, add_prefix_space=True, return_tensors="pt")
+    >>> outputs = model(**inputs)
+    >>> word_last_hidden_state = outputs.last_hidden_state
+    >>> entity_last_hidden_state = outputs.entity_last_hidden_state
+
+    # Example 2: Inputting Wikipedia entities to obtain enriched contextualized representations
+    >>> entities = ["Beyoncé", "Los Angeles"]  # Wikipedia entity titles corresponding to the entity mentions "Beyoncé" and "Los Angeles"
+    >>> entity_spans = [(0, 7), (17, 28)]  # character-based entity spans corresponding to "Beyoncé" and "Los Angeles"
+    >>> inputs = tokenizer(text, entities=entities, entity_spans=entity_spans, add_prefix_space=True, return_tensors="pt")
+    >>> outputs = model(**inputs)
+    >>> word_last_hidden_state = outputs.last_hidden_state
+    >>> entity_last_hidden_state = outputs.entity_last_hidden_state
+
+    # Example 3: Classifying the relationship between two entities using LukeForEntityPairClassification head model
+    >>> model = LukeForEntityPairClassification.from_pretrained("studio-ousia/luke-large-finetuned-tacred")
+    >>> tokenizer = LukeTokenizer.from_pretrained("studio-ousia/luke-large-finetuned-tacred")
+    >>> entity_spans = [(0, 7), (17, 28)]  # character-based entity spans corresponding to "Beyoncé" and "Los Angeles"
+    >>> inputs = tokenizer(text, entity_spans=entity_spans, return_tensors="pt")
+    >>> outputs = model(**inputs)
+    >>> logits = outputs.logits
+    >>> predicted_class_idx = int(logits[0].argmax())
+    >>> print("Predicted class:", model.config.id2label[predicted_class_idx])
+
+This model was contributed by `ikuyamada <https://huggingface.co/ikuyamada>`__ and `nielsr
+<https://huggingface.co/nielsr>`__. The original code can be found `here <https://github.com/studio-ousia/luke>`__.
+
+
+LukeConfig
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.LukeConfig
+    :members:
+
+
+LukeTokenizer
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.LukeTokenizer
+    :members: __call__, save_vocabulary
+
+
+LukeModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.LukeModel
+    :members: forward
+
+
+LukeForEntityClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.LukeForEntityClassification
+    :members: forward
+
+
+LukeForEntityPairClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.LukeForEntityPairClassification
+    :members: forward
+
+
+LukeForEntitySpanClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.LukeForEntitySpanClassification
+    :members: forward

docs/source/model_doc/lxmert.rst

@@ -52,7 +52,8 @@ Tips:
   contains self-attention for each respective modality and cross-attention, only the cross attention is returned and
   both self attention outputs are disregarded.
 
-The original code can be found `here <https://github.com/airsplay/lxmert>`__.
+This model was contributed by `eltoto1219 <https://huggingface.co/eltoto1219>`__. The original code can be found `here
+<https://github.com/airsplay/lxmert>`__.
 
 
 LxmertConfig

docs/source/model_doc/m2m_100.rst

@@ -34,6 +34,8 @@ to create high quality models. Our focus on non-English-Centric models brings ga
 translating between non-English directions while performing competitively to the best single systems of WMT. We
 open-source our scripts so that others may reproduce the data, evaluation, and final M2M-100 model.*
 
+This model was contributed by `valhalla <https://huggingface.co/valhalla>`__.
+
 
 Training and Generation
 _______________________________________________________________________________________________________________________

docs/source/model_doc/marian.rst

@@ -37,6 +37,7 @@ Implementation Notes
     - the model starts generating with :obj:`pad_token_id` (which has 0 as a token_embedding) as the prefix (Bart uses
       :obj:`<s/>`),
 - Code to bulk convert models can be found in ``convert_marian_to_pytorch.py``.
+- This model was contributed by `sshleifer <https://huggingface.co/sshleifer>`__.
 
 Naming
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

docs/source/model_doc/mbart.rst

@@ -29,7 +29,8 @@ corpora in many languages using the BART objective. mBART is one of the first me
 sequence-to-sequence model by denoising full texts in multiple languages, while previous approaches have focused only
 on the encoder, decoder, or reconstructing parts of the text.
 
-The Authors' code can be found `here <https://github.com/pytorch/fairseq/tree/master/examples/mbart>`__
+This model was contributed by `valhalla <https://huggingface.co/valhalla>`__. The Authors' code can be found `here
+<https://github.com/pytorch/fairseq/tree/master/examples/mbart>`__
 
 Training of MBart
 _______________________________________________________________________________________________________________________

docs/source/model_doc/megatron_bert.rst

@@ -0,0 +1,154 @@
+.. 
+    Copyright 2021 NVIDIA Corporation and 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.
+
+MegatronBERT
+-----------------------------------------------------------------------------------------------------------------------
+
+Overview
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The MegatronBERT model was proposed in `Megatron-LM: Training Multi-Billion Parameter Language Models Using Model
+Parallelism <https://arxiv.org/abs/1909.08053>`__ by Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley,
+Jared Casper and Bryan Catanzaro.
+
+The abstract from the paper is the following:
+
+*Recent work in language modeling demonstrates that training large transformer models advances the state of the art in
+Natural Language Processing applications. However, very large models can be quite difficult to train due to memory
+constraints. In this work, we present our techniques for training very large transformer models and implement a simple,
+efficient intra-layer model parallel approach that enables training transformer models with billions of parameters. Our
+approach does not require a new compiler or library changes, is orthogonal and complimentary to pipeline model
+parallelism, and can be fully implemented with the insertion of a few communication operations in native PyTorch. We
+illustrate this approach by converging transformer based models up to 8.3 billion parameters using 512 GPUs. We sustain
+15.1 PetaFLOPs across the entire application with 76% scaling efficiency when compared to a strong single GPU baseline
+that sustains 39 TeraFLOPs, which is 30% of peak FLOPs. To demonstrate that large language models can further advance
+the state of the art (SOTA), we train an 8.3 billion parameter transformer language model similar to GPT-2 and a 3.9
+billion parameter model similar to BERT. We show that careful attention to the placement of layer normalization in
+BERT-like models is critical to achieving increased performance as the model size grows. Using the GPT-2 model we
+achieve SOTA results on the WikiText103 (10.8 compared to SOTA perplexity of 15.8) and LAMBADA (66.5% compared to SOTA
+accuracy of 63.2%) datasets. Our BERT model achieves SOTA results on the RACE dataset (90.9% compared to SOTA accuracy
+of 89.4%).*
+
+Tips:
+
+We have provided pretrained `BERT-345M <https://ngc.nvidia.com/catalog/models/nvidia:megatron_bert_345m>`__ checkpoints
+for use to evaluate or finetuning downstream tasks.
+
+To access these checkpoints, first `sign up <https://ngc.nvidia.com/signup>`__ for and setup the NVIDIA GPU Cloud (NGC)
+Registry CLI. Further documentation for downloading models can be found in the `NGC documentation
+<https://docs.nvidia.com/dgx/ngc-registry-cli-user-guide/index.html#topic_6_4_1>`__.
+
+Alternatively, you can directly download the checkpoints using:
+
+BERT-345M-uncased::
+
+.. code-block:: bash
+
+    wget --content-disposition https://api.ngc.nvidia.com/v2/models/nvidia/megatron_bert_345m/versions/v0.1_uncased/zip
+    -O megatron_bert_345m_v0_1_uncased.zip
+
+BERT-345M-cased::
+
+.. code-block:: bash
+
+    wget --content-disposition https://api.ngc.nvidia.com/v2/models/nvidia/megatron_bert_345m/versions/v0.1_cased/zip -O
+    megatron_bert_345m_v0_1_cased.zip
+
+Once you have obtained the checkpoints from NVIDIA GPU Cloud (NGC), you have to convert them to a format that will
+easily be loaded by Hugging Face Transformers and our port of the BERT code.
+
+The following commands allow you to do the conversion. We assume that the folder ``models/megatron_bert`` contains
+``megatron_bert_345m_v0_1_{cased, uncased}.zip`` and that the commands are run from inside that folder::
+
+.. code-block:: bash
+
+    python3 $PATH_TO_TRANSFORMERS/models/megatron_bert/convert_megatron_bert_checkpoint.py megatron_bert_345m_v0_1_uncased.zip 
+
+.. code-block:: bash
+
+    python3 $PATH_TO_TRANSFORMERS/models/megatron_bert/convert_megatron_bert_checkpoint.py megatron_bert_345m_v0_1_cased.zip
+
+This model was contributed by `jdemouth <https://huggingface.co/jdemouth>`__. The original code can be found `here
+<https://github.com/NVIDIA/Megatron-LM>`__. That repository contains a multi-GPU and multi-node implementation of the
+Megatron Language models. In particular, it contains a hybrid model parallel approach using "tensor parallel" and
+"pipeline parallel" techniques.
+
+MegatronBertConfig
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.MegatronBertConfig
+    :members:
+
+
+MegatronBertModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.MegatronBertModel
+    :members: forward
+
+
+MegatronBertForMaskedLM
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.MegatronBertForMaskedLM
+    :members: forward
+
+
+MegatronBertForCausalLM
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.MegatronBertForCausalLM
+    :members: forward
+
+
+MegatronBertForNextSentencePrediction
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.MegatronBertForNextSentencePrediction
+    :members: forward
+
+
+MegatronBertForPreTraining
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.MegatronBertForPreTraining
+    :members: forward
+
+
+MegatronBertForSequenceClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.MegatronBertForSequenceClassification
+    :members: forward
+
+
+MegatronBertForMultipleChoice
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.MegatronBertForMultipleChoice
+    :members: forward
+
+
+MegatronBertForTokenClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.MegatronBertForTokenClassification
+    :members: forward
+
+
+MegatronBertForQuestionAnswering
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.MegatronBertForQuestionAnswering
+    :members: forward
+
+

docs/source/model_doc/megatron_gpt2.rst

@@ -0,0 +1,71 @@
+.. 
+    Copyright 2021 NVIDIA Corporation and 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.
+
+MegatronGPT2
+-----------------------------------------------------------------------------------------------------------------------
+
+Overview
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The MegatronGPT2 model was proposed in `Megatron-LM: Training Multi-Billion Parameter Language Models Using Model
+Parallelism <https://arxiv.org/abs/1909.08053>`__ by Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley,
+Jared Casper and Bryan Catanzaro.
+
+The abstract from the paper is the following:
+
+*Recent work in language modeling demonstrates that training large transformer models advances the state of the art in
+Natural Language Processing applications. However, very large models can be quite difficult to train due to memory
+constraints. In this work, we present our techniques for training very large transformer models and implement a simple,
+efficient intra-layer model parallel approach that enables training transformer models with billions of parameters. Our
+approach does not require a new compiler or library changes, is orthogonal and complimentary to pipeline model
+parallelism, and can be fully implemented with the insertion of a few communication operations in native PyTorch. We
+illustrate this approach by converging transformer based models up to 8.3 billion parameters using 512 GPUs. We sustain
+15.1 PetaFLOPs across the entire application with 76% scaling efficiency when compared to a strong single GPU baseline
+that sustains 39 TeraFLOPs, which is 30% of peak FLOPs. To demonstrate that large language models can further advance
+the state of the art (SOTA), we train an 8.3 billion parameter transformer language model similar to GPT-2 and a 3.9
+billion parameter model similar to BERT. We show that careful attention to the placement of layer normalization in
+BERT-like models is critical to achieving increased performance as the model size grows. Using the GPT-2 model we
+achieve SOTA results on the WikiText103 (10.8 compared to SOTA perplexity of 15.8) and LAMBADA (66.5% compared to SOTA
+accuracy of 63.2%) datasets. Our BERT model achieves SOTA results on the RACE dataset (90.9% compared to SOTA accuracy
+of 89.4%).*
+
+Tips:
+
+We have provided pretrained `GPT2-345M <https://ngc.nvidia.com/catalog/models/nvidia:megatron_lm_345m>`__ checkpoints
+for use to evaluate or finetuning downstream tasks.
+
+To access these checkpoints, first `sign up <https://ngc.nvidia.com/signup>`__ for and setup the NVIDIA GPU Cloud (NGC)
+Registry CLI. Further documentation for downloading models can be found in the `NGC documentation
+<https://docs.nvidia.com/dgx/ngc-registry-cli-user-guide/index.html#topic_6_4_1>`__.
+
+Alternatively, you can directly download the checkpoints using::
+
+.. code-block:: bash
+
+    wget --content-disposition https://api.ngc.nvidia.com/v2/models/nvidia/megatron_lm_345m/versions/v0.0/zip -O
+    megatron_gpt2_345m_v0_0.zip
+
+Once you have obtained the checkpoint from NVIDIA GPU Cloud (NGC), you have to convert it to a format that will easily
+be loaded by Hugging Face Transformers GPT2 implementation.
+
+The following command allows you to do the conversion. We assume that the folder ``models/megatron_gpt2`` contains
+``megatron_gpt2_345m_v0_0.zip`` and that the command is run from that folder::
+
+.. code-block:: bash
+
+    python3 $PATH_TO_TRANSFORMERS/models/megatron_gpt2/convert_megatron_gpt2_checkpoint.py megatron_gpt2_345m_v0_0.zip
+
+This model was contributed by `jdemouth <https://huggingface.co/jdemouth>`__. The original code can be found `here
+<https://github.com/NVIDIA/Megatron-LM>`__. That repository contains a multi-GPU and multi-node implementation of the
+Megatron Language models. In particular, it contains a hybrid model parallel approach using "tensor parallel" and
+"pipeline parallel" techniques.
+

docs/source/model_doc/mobilebert.rst

@@ -44,7 +44,8 @@ Tips:
   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.
 
-The original code can be found `here <https://github.com/google-research/mobilebert>`__.
+This model was contributed by `vshampor <https://huggingface.co/vshampor>`__. The original code can be found `here
+<https://github.com/google-research/mobilebert>`__.
 
 MobileBertConfig
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

docs/source/model_doc/mt5.rst

@@ -28,7 +28,8 @@ multilingual variant of T5 that was pre-trained on a new Common Crawl-based data
 the design and modified training of mT5 and demonstrate its state-of-the-art performance on many multilingual
 benchmarks. All of the code and model checkpoints*
 
-The original code can be found `here <https://github.com/google-research/multilingual-t5>`__.
+This model was contributed by `patrickvonplaten <https://huggingface.co/patrickvonplaten>`__. The original code can be
+found `here <https://github.com/google-research/multilingual-t5>`__.
 
 MT5Config
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

docs/source/model_doc/pegasus.rst

@@ -31,7 +31,8 @@ According to the abstract,
   extractive summary.
 - Pegasus achieves SOTA summarization performance on all 12 downstream tasks, as measured by ROUGE and human eval.
 
-The Authors' code can be found `here <https://github.com/google-research/pegasus>`__.
+This model was contributed by `sshleifer <https://huggingface.co/sshleifer>`__. The Authors' code can be found `here
+<https://github.com/google-research/pegasus>`__.
 
 
 Checkpoints
@@ -52,7 +53,8 @@ Examples
 _______________________________________________________________________________________________________________________
 
 - :prefix_link:`Script <examples/research_projects/seq2seq-distillation/finetune_pegasus_xsum.sh>` to fine-tune pegasus
-  on the XSUM dataset. Data download instructions at :prefix_link:`examples/seq2seq/ <examples/seq2seq/README.md>`.
+  on the XSUM dataset. Data download instructions at :prefix_link:`examples/pytorch/summarization/
+  <examples/pytorch/summarization/README.md>`.
 - FP16 is not supported (help/ideas on this appreciated!).
 - The adafactor optimizer is recommended for pegasus fine-tuning.
 
@@ -88,7 +90,7 @@ Usage Example
     >>> device = 'cuda' if torch.cuda.is_available() else 'cpu'
     >>> tokenizer = PegasusTokenizer.from_pretrained(model_name)
     >>> model = PegasusForConditionalGeneration.from_pretrained(model_name).to(device)
-    >>> batch = tokenizer(src_text, truncation=True, padding='longest', return_tensors="pt").to(torch_device)
+    >>> batch = tokenizer(src_text, truncation=True, padding='longest', return_tensors="pt").to(device)
     >>> translated = model.generate(**batch)
     >>> tgt_text = tokenizer.batch_decode(translated, skip_special_tokens=True)
     >>> assert tgt_text[0] == "California's largest electricity provider has turned off power to hundreds of thousands of customers."

docs/source/model_doc/phobert.rst

@@ -31,26 +31,26 @@ Example of use:
 
 .. code-block::
 
-  import torch
-  from transformers import AutoModel, AutoTokenizer
+    >>> import torch
+    >>> from transformers import AutoModel, AutoTokenizer
 
-  phobert = AutoModel.from_pretrained("vinai/phobert-base")
-  tokenizer = AutoTokenizer.from_pretrained("vinai/phobert-base")
+    >>> phobert = AutoModel.from_pretrained("vinai/phobert-base")
+    >>> tokenizer = AutoTokenizer.from_pretrained("vinai/phobert-base")
 
-  # INPUT TEXT MUST BE ALREADY WORD-SEGMENTED!
-  line = "Tôi là sinh_viên trường đại_học Công_nghệ ."
+    >>> # INPUT TEXT MUST BE ALREADY WORD-SEGMENTED!
+    >>> line = "Tôi là sinh_viên trường đại_học Công_nghệ ."
 
-  input_ids = torch.tensor([tokenizer.encode(line)])
+    >>> input_ids = torch.tensor([tokenizer.encode(line)])
 
-  with torch.no_grad():
-      features = phobert(input_ids)  # Models outputs are now tuples
+    >>> with torch.no_grad():
+    ...     features = phobert(input_ids)  # Models outputs are now tuples
 
-  ## With TensorFlow 2.0+:
-  # from transformers import TFAutoModel
-  # phobert = TFAutoModel.from_pretrained("vinai/phobert-base")
+    >>> # With TensorFlow 2.0+:
+    >>> # from transformers import TFAutoModel
+    >>> # phobert = TFAutoModel.from_pretrained("vinai/phobert-base")
 
 
-The original code can be found `here <https://github.com/VinAIResearch/PhoBERT>`__.
+    This model was contributed by `dqnguyen <https://huggingface.co/dqnguyen>`__. The original code can be found `here <https://github.com/VinAIResearch/PhoBERT>`__.
 
 PhobertTokenizer
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

docs/source/model_doc/rag.rst

@@ -43,6 +43,7 @@ outperforming parametric seq2seq models and task-specific retrieve-and-extract a
 tasks, we find that RAG models generate more specific, diverse and factual language than a state-of-the-art
 parametric-only seq2seq baseline.*
 
+This model was contributed by `ola13 <https://huggingface.co/ola13>`__.
 
 
 RagConfig

docs/source/model_doc/reformer.rst

@@ -32,7 +32,8 @@ layers instead of the standard residuals, which allows storing activations only
 N times, where N is the number of layers. The resulting model, the Reformer, performs on par with Transformer models
 while being much more memory-efficient and much faster on long sequences.*
 
-The Authors' code can be found `here <https://github.com/google/trax/tree/master/trax/models/reformer>`__.
+This model was contributed by `patrickvonplaten <https://huggingface.co/patrickvonplaten>`__. The Authors' code can be
+found `here <https://github.com/google/trax/tree/master/trax/models/reformer>`__.
 
 Axial Positional Encodings
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
@@ -145,8 +146,8 @@ For training, the :class:`~transformers.ReformerModelWithLMHead` should be used
 
 .. code-block::
 
-  input_ids = tokenizer.encode('This is a sentence from the training data', return_tensors='pt')
-  loss = model(input_ids, labels=input_ids)[0]
+    input_ids = tokenizer.encode('This is a sentence from the training data', return_tensors='pt')
+    loss = model(input_ids, labels=input_ids)[0]
 
 
 ReformerConfig

docs/source/model_doc/retribert.rst

@@ -20,8 +20,8 @@ The RetriBERT model was proposed in the blog post `Explain Anything Like I'm Fiv
 Question Answering <https://yjernite.github.io/lfqa.html>`__. RetriBERT is a small model that uses either a single or
 pair of BERT encoders with lower-dimension projection for dense semantic indexing of text.
 
-Code to train and use the model can be found `here
-<https://github.com/huggingface/transformers/tree/master/examples/distillation>`__.
+This model was contributed by `yjernite <https://huggingface.co/yjernite>`__. Code to train and use the model can be
+found :prefix_link:`here <examples/research-projects/distillation>`.
 
 
 RetriBertConfig

docs/source/model_doc/roberta.rst

@@ -44,7 +44,8 @@ Tips:
   separate your segments with the separation token :obj:`tokenizer.sep_token` (or :obj:`</s>`)
 - :doc:`CamemBERT <camembert>` is a wrapper around RoBERTa. Refer to this page for usage examples.
 
-The original code can be found `here <https://github.com/pytorch/fairseq/tree/master/examples/roberta>`_.
+This model was contributed by `julien-c <https://huggingface.co/julien-c>`__. The original code can be found `here
+<https://github.com/pytorch/fairseq/tree/master/examples/roberta>`_.
 
 
 RobertaConfig
@@ -165,3 +166,38 @@ FlaxRobertaModel
 
 .. autoclass:: transformers.FlaxRobertaModel
     :members: __call__
+
+
+FlaxRobertaForMaskedLM
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxRobertaForMaskedLM
+    :members: __call__
+
+
+FlaxRobertaForSequenceClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxRobertaForSequenceClassification
+    :members: __call__
+
+
+FlaxRobertaForMultipleChoice
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxRobertaForMultipleChoice
+    :members: __call__
+
+
+FlaxRobertaForTokenClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxRobertaForTokenClassification
+    :members: __call__
+
+
+FlaxRobertaForQuestionAnswering
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.FlaxRobertaForQuestionAnswering
+    :members: __call__

docs/source/model_doc/roformer.rst

@@ -0,0 +1,161 @@
+.. 
+    Copyright 2021 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.
+
+RoFormer
+-----------------------------------------------------------------------------------------------------------------------
+
+Overview
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The RoFormer model was proposed in `RoFormer: Enhanced Transformer with Rotary Position Embedding
+<https://arxiv.org/pdf/2104.09864v1.pdf>`__ by Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu.
+
+The abstract from the paper is the following:
+
+*Position encoding in transformer architecture provides supervision for dependency modeling between elements at
+different positions in the sequence. We investigate various methods to encode positional information in
+transformer-based language models and propose a novel implementation named Rotary Position Embedding(RoPE). The
+proposed RoPE encodes absolute positional information with rotation matrix and naturally incorporates explicit relative
+position dependency in self-attention formulation. Notably, RoPE comes with valuable properties such as flexibility of
+being expand to any sequence lengths, decaying inter-token dependency with increasing relative distances, and
+capability of equipping the linear self-attention with relative position encoding. As a result, the enhanced
+transformer with rotary position embedding, or RoFormer, achieves superior performance in tasks with long texts. We
+release the theoretical analysis along with some preliminary experiment results on Chinese data. The undergoing
+experiment for English benchmark will soon be updated.*
+
+Tips:
+
+- RoFormer is a BERT-like autoencoding model with rotary position embeddings. Rotary position embeddings have shown
+  improved performance on classification tasks with long texts.
+
+
+This model was contributed by `junnyu <https://huggingface.co/junnyu>`__. The original code can be found `here
+<https://github.com/ZhuiyiTechnology/roformer>`__.
+
+RoFormerConfig
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.RoFormerConfig
+    :members:
+
+
+RoFormerTokenizer
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.RoFormerTokenizer
+    :members: build_inputs_with_special_tokens, get_special_tokens_mask,
+        create_token_type_ids_from_sequences, save_vocabulary
+
+
+RobertaTokenizerFast
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.RoFormerTokenizerFast
+    :members: build_inputs_with_special_tokens
+
+
+RoFormerModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.RoFormerModel
+    :members: forward
+
+
+RoFormerForCausalLM
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.RoFormerForCausalLM
+    :members: forward
+
+
+RoFormerForMaskedLM
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.RoFormerForMaskedLM
+    :members: forward
+
+
+RoFormerForSequenceClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.RoFormerForSequenceClassification
+    :members: forward
+
+
+RoFormerForMultipleChoice
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.RoFormerForMultipleChoice
+    :members: forward
+
+
+RoFormerForTokenClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.RoFormerForTokenClassification
+    :members: forward
+
+
+RoFormerForQuestionAnswering
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.RoFormerForQuestionAnswering
+    :members: forward
+
+
+TFRoFormerModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFRoFormerModel
+    :members: call
+
+
+TFRoFormerForMaskedLM
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFRoFormerForMaskedLM
+    :members: call
+
+
+TFRoFormerForCausalLM
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFRoFormerForCausalLM
+    :members: call
+
+
+TFRoFormerForSequenceClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFRoFormerForSequenceClassification
+    :members: call
+
+
+TFRoFormerForMultipleChoice
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFRoFormerForMultipleChoice
+    :members: call
+
+
+TFRoFormerForTokenClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFRoFormerForTokenClassification
+    :members: call
+
+
+TFRoFormerForQuestionAnswering
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFRoFormerForQuestionAnswering
+    :members: call

docs/source/model_doc/speech_to_text.rst

@@ -25,7 +25,8 @@ transcripts/translations autoregressively. Speech2Text has been fine-tuned on se
 `LibriSpeech <http://www.openslr.org/12>`__, `CoVoST 2 <https://github.com/facebookresearch/covost>`__, `MuST-C
 <https://ict.fbk.eu/must-c/>`__.
 
-The original code can be found `here <https://github.com/pytorch/fairseq/tree/master/examples/speech_to_text>`__.
+This model was contributed by `valhalla <https://huggingface.co/valhalla>`__. The original code can be found `here
+<https://github.com/pytorch/fairseq/tree/master/examples/speech_to_text>`__.
 
 
 Inference

docs/source/model_doc/squeezebert.rst

@@ -47,6 +47,9 @@ Tips:
 - For best results when finetuning on sequence classification tasks, it is recommended to start with the
   `squeezebert/squeezebert-mnli-headless` checkpoint.
 
+This model was contributed by `forresti <https://huggingface.co/forresti>`__.
+
+
 SqueezeBertConfig
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~