Patch release: v4.9.2

Tpu tie weights (#13030 )
* Fix tied weights on TPU * Manually tie weights in no trainer examples * Fix for test * One last missing * Gettning owned by my scripts * Address review comments * Fix test * Fix tests * Fix reformer tests
2021-08-09 16:01:36 +02:00 · 2021-08-09 15:53:05 +02:00 · 2021-08-09 15:52:16 +02:00 · 2021-08-09 15:52:07 +02:00 · 2021-08-09 15:51:04 +02:00 · 2021-08-09 15:50:58 +02:00
408 changed files with 8132 additions and 40829 deletions
--- a/.circleci/config.yml
+++ b/.circleci/config.yml
@@ -80,7 +80,7 @@ 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 .[sklearn,tf-cpu,torch,testing,sentencepiece,torch-speech,vision]
+            - 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" }}
@@ -97,37 +97,6 @@ jobs:
                  path: ~/transformers/tests_output.txt
            - store_artifacts:
                  path: ~/transformers/reports
-    
-    run_tests_torch_and_tf_all:
-        working_directory: ~/transformers
-        docker:
-            - 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:
-            - checkout
-            - restore_cache:
-                  keys:
-                      - v0.4-torch_and_tf-{{ checksum "setup.py" }}
-                      - 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,torch-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: |
-                  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:
-                  path: ~/transformers/reports

    run_tests_torch_and_flax:
        working_directory: ~/transformers
@@ -147,7 +116,7 @@ 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 .[sklearn,flax,torch,testing,sentencepiece,torch-speech,vision]
+            - 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" }}
@@ -164,37 +133,6 @@ jobs:
                  path: ~/transformers/tests_output.txt
            - store_artifacts:
                  path: ~/transformers/reports
-    
-    run_tests_torch_and_flax_all:
-        working_directory: ~/transformers
-        docker:
-            - 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:
-            - checkout
-            - restore_cache:
-                  keys:
-                      - v0.4-torch_and_flax-{{ checksum "setup.py" }}
-                      - 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,torch-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: |
-                  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:
-                  path: ~/transformers/reports

    run_tests_torch:
        working_directory: ~/transformers
@@ -213,7 +151,7 @@ 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 .[sklearn,torch,testing,sentencepiece,torch-speech,vision,timm]
+            - 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" }}
@@ -230,36 +168,6 @@ jobs:
                  path: ~/transformers/tests_output.txt
            - store_artifacts:
                  path: ~/transformers/reports
-    
-    run_tests_torch_all:
-        working_directory: ~/transformers
-        docker:
-            - image: circleci/python:3.7
-        environment:
-            OMP_NUM_THREADS: 1
-            TRANSFORMERS_IS_CI: yes
-        resource_class: xlarge
-        parallelism: 1
-        steps:
-            - checkout
-            - restore_cache:
-                  keys:
-                      - v0.4-torch-{{ checksum "setup.py" }}
-                      - 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,torch-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 3 --dist=loadfile -s --make-reports=tests_torch tests | tee tests_output.txt
-            - store_artifacts:
-                  path: ~/transformers/tests_output.txt
-            - store_artifacts:
-                  path: ~/transformers/reports

    run_tests_tf:
        working_directory: ~/transformers
@@ -277,7 +185,7 @@ jobs:
                      - v0.4-tf-{{ checksum "setup.py" }}
                      - v0.4-{{ checksum "setup.py" }}
            - run: pip install --upgrade pip
-            - run: pip install .[sklearn,tf-cpu,testing,sentencepiece,tf-speech]
+            - run: pip install .[sklearn,tf-cpu,testing,sentencepiece]
            - save_cache:
                  key: v0.4-tf-{{ checksum "setup.py" }}
                  paths:
@@ -293,34 +201,6 @@ jobs:
                  path: ~/transformers/tests_output.txt
            - store_artifacts:
                  path: ~/transformers/reports
-    
-    run_tests_tf_all:
-        working_directory: ~/transformers
-        docker:
-            - image: circleci/python:3.7
-        environment:
-            OMP_NUM_THREADS: 1
-            TRANSFORMERS_IS_CI: yes
-        resource_class: xlarge
-        parallelism: 1
-        steps:
-            - checkout
-            - restore_cache:
-                  keys:
-                      - v0.4-tf-{{ checksum "setup.py" }}
-                      - v0.4-{{ checksum "setup.py" }}
-            - run: pip install --upgrade pip
-            - run: pip install .[sklearn,tf-cpu,testing,sentencepiece,tf-speech]
-            - save_cache:
-                  key: v0.4-tf-{{ checksum "setup.py" }}
-                  paths:
-                      - '~/.cache/pip'
-            - run: |
-                  python -m pytest -n 8 --dist=loadfile -rA -s --make-reports=tests_tf tests | tee tests_output.txt
-            - store_artifacts:
-                  path: ~/transformers/tests_output.txt
-            - store_artifacts:
-                  path: ~/transformers/reports

    run_tests_flax:
        working_directory: ~/transformers
@@ -338,7 +218,7 @@ jobs:
                    - v0.4-flax-{{ checksum "setup.py" }}
                    - v0.4-{{ checksum "setup.py" }}
            - run: pip install --upgrade pip
-            - run: sudo pip install .[flax,testing,sentencepiece,flax-speech,vision]
+            - run: sudo pip install .[flax,testing,sentencepiece]
            - save_cache:
                  key: v0.4-flax-{{ checksum "setup.py" }}
                  paths:
@@ -354,34 +234,6 @@ jobs:
                  path: ~/transformers/tests_output.txt
            - store_artifacts:
                  path: ~/transformers/reports
-    
-    run_tests_flax_all:
-        working_directory: ~/transformers
-        docker:
-            - image: circleci/python:3.7
-        environment:
-            OMP_NUM_THREADS: 1
-            TRANSFORMERS_IS_CI: yes
-        resource_class: xlarge
-        parallelism: 1
-        steps:
-            - checkout
-            - restore_cache:
-                keys:
-                    - v0.4-flax-{{ checksum "setup.py" }}
-                    - v0.4-{{ checksum "setup.py" }}
-            - run: pip install --upgrade pip
-            - run: sudo pip install .[flax,testing,sentencepiece,vision,flax-speech]
-            - save_cache:
-                  key: v0.4-flax-{{ checksum "setup.py" }}
-                  paths:
-                      - '~/.cache/pip'
-            - run: |
-                  python -m pytest -n 8 --dist=loadfile -rA -s --make-reports=tests_flax tests | tee tests_output.txt
-            - store_artifacts:
-                  path: ~/transformers/tests_output.txt
-            - store_artifacts:
-                  path: ~/transformers/reports

    run_tests_pipelines_torch:
        working_directory: ~/transformers
@@ -401,7 +253,7 @@ 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 .[sklearn,torch,testing,sentencepiece,torch-speech,vision]
+            - 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" }}
@@ -418,37 +270,6 @@ jobs:
                  path: ~/transformers/tests_output.txt
            - store_artifacts:
                  path: ~/transformers/reports
-    
-    run_tests_pipelines_torch_all:
-        working_directory: ~/transformers
-        docker:
-            - image: circleci/python:3.7
-        environment:
-            OMP_NUM_THREADS: 1
-            RUN_PIPELINE_TESTS: yes
-            TRANSFORMERS_IS_CI: yes
-        resource_class: xlarge
-        parallelism: 1
-        steps:
-            - checkout
-            - restore_cache:
-                  keys:
-                      - v0.4-torch-{{ checksum "setup.py" }}
-                      - 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,torch-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: |
-                  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:
-                  path: ~/transformers/reports

    run_tests_pipelines_tf:
        working_directory: ~/transformers
@@ -484,35 +305,6 @@ jobs:
            - store_artifacts:
                  path: ~/transformers/reports

-    run_tests_pipelines_tf_all:
-        working_directory: ~/transformers
-        docker:
-            - image: circleci/python:3.7
-        environment:
-            OMP_NUM_THREADS: 1
-            RUN_PIPELINE_TESTS: yes
-            TRANSFORMERS_IS_CI: yes
-        resource_class: xlarge
-        parallelism: 1
-        steps:
-            - checkout
-            - restore_cache:
-                  keys:
-                      - v0.4-tf-{{ checksum "setup.py" }}
-                      - v0.4-{{ checksum "setup.py" }}
-            - run: pip install --upgrade pip
-            - run: pip install .[sklearn,tf-cpu,testing,sentencepiece]
-            - save_cache:
-                  key: v0.4-tf-{{ checksum "setup.py" }}
-                  paths:
-                      - '~/.cache/pip'
-            - 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:
-                  path: ~/transformers/reports
-
    run_tests_custom_tokenizers:
        working_directory: ~/transformers
        docker:
@@ -607,45 +399,8 @@ jobs:
                  path: ~/transformers/test_preparation.txt
            - run: |
                  if [ -f test_list.txt ]; then
-                    python -m pytest -sv --make-reports=tests_hub $(cat test_list.txt) -m is_staging_test | tee tests_output.txt
+                    python -m pytest -sv $(cat test_list.txt) -m is_staging_test
                  fi
-            - store_artifacts:
-                  path: ~/transformers/tests_output.txt
-            - store_artifacts:
-                  path: ~/transformers/reports
-    
-    run_tests_hub_all:
-        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 .[torch,sentencepiece,testing]
-            - save_cache:
-                  key: v0.4-hub-{{ checksum "setup.py" }}
-                  paths:
-                      - '~/.cache/pip'
-            - run: |
-                  python -m pytest -sv --make-reports=tests_hub tests -m is_staging_test | tee tests_output.txt
-            - store_artifacts:
-                  path: ~/transformers/tests_output.txt
-            - store_artifacts:
-                  path: ~/transformers/reports

    run_tests_onnxruntime:
        working_directory: ~/transformers
@@ -673,45 +428,16 @@ jobs:
                  path: ~/transformers/test_preparation.txt
            - run: |
                  if [ -f test_list.txt ]; then
-                    python -m pytest -n 1 --dist=loadfile -s --make-reports=tests_onnx $(cat test_list.txt) -k onnx | tee tests_output.txt
+                    python -m pytest -n 1 --dist=loadfile -s --make-reports=tests_torch $(cat test_list.txt) -k onnx | tee tests_output.txt
                  fi
            - store_artifacts:
                  path: ~/transformers/tests_output.txt
            - store_artifacts:
                  path: ~/transformers/reports
-    
-    run_tests_onnxruntime_all:
-        working_directory: ~/transformers
-        docker:
-            - image: circleci/python:3.7
-        environment:
-            OMP_NUM_THREADS: 1
-            TRANSFORMERS_IS_CI: yes
-        resource_class: xlarge
-        parallelism: 1
-        steps:
-            - checkout
-            - restore_cache:
-                  keys:
-                      - v0.4-torch-{{ checksum "setup.py" }}
-                      - v0.4-{{ checksum "setup.py" }}
-            - run: pip install --upgrade pip
-            - run: pip install .[torch,testing,sentencepiece,onnxruntime]
-            - save_cache:
-                  key: v0.4-onnx-{{ checksum "setup.py" }}
-                  paths:
-                      - '~/.cache/pip'
-            - run: |
-                  python -m pytest -n 1 --dist=loadfile -s --make-reports=tests_onnx tests -k onnx | tee tests_output.txt
-            - store_artifacts:
-                  path: ~/transformers/tests_output.txt
-            - store_artifacts:
-                  path: ~/transformers/reports
-
    build_doc:
        working_directory: ~/transformers
        docker:
-            - image: circleci/python:3.7.11
+            - image: circleci/python:3.6
        resource_class: large
        steps:
            - checkout
@@ -733,7 +459,7 @@ jobs:
    deploy_doc:
        working_directory: ~/transformers
        docker:
-            - image: circleci/python:3.7.11
+            - image: circleci/python:3.6
        resource_class: large
        steps:
            - add_ssh_keys:
@@ -798,44 +524,6 @@ jobs:
            - run: pip install requests
            - run: python ./utils/link_tester.py

-    run_tests_layoutlmv2:
-        working_directory: ~/transformers
-        docker:
-            - image: circleci/python:3.7
-        environment:
-            OMP_NUM_THREADS: 1
-            TRANSFORMERS_IS_CI: yes
-        resource_class: xlarge
-        parallelism: 1
-        steps:
-            - checkout
-            - restore_cache:
-                  keys:
-                      - v0.4-torch-{{ checksum "setup.py" }}
-                      - 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 .[torch,testing,vision]
-            - run: pip install torchvision
-            - run: python -m pip install 'git+https://github.com/facebookresearch/detectron2.git'
-            - run: sudo apt install tesseract-ocr
-            - run: pip install pytesseract
-            - save_cache:
-                  key: v0.4-torch-{{ checksum "setup.py" }}
-                  paths:
-                      - '~/.cache/pip'
-            - run: python utils/tests_fetcher.py | tee test_preparation.txt
-            - store_artifacts:
-                  path: ~/transformers/test_preparation.txt
-            - run: |
-                  if [ -f test_list.txt ]; then
-                    python -m pytest -n 1 tests/*layoutlmv2* --dist=loadfile -s --make-reports=tests_layoutlmv2 --durations=100
-                  fi
-            - store_artifacts:
-                  path: ~/transformers/tests_output.txt
-            - store_artifacts:
-                  path: ~/transformers/reports
-
 # TPU JOBS
    run_examples_tpu:
        docker:
@@ -890,27 +578,7 @@ workflows:
            - run_tests_onnxruntime
            - run_tests_hub
            - build_doc
-            - run_tests_layoutlmv2
            - deploy_doc: *workflow_filters
-    nightly:
-        triggers:
-            - schedule:
-                cron: "0 0 * * *"
-                filters:
-                    branches:
-                        only:
-                            - master
-        jobs:
-            - run_tests_torch_and_tf_all
-            - run_tests_torch_and_flax_all
-            - run_tests_torch_all
-            - run_tests_tf_all
-            - run_tests_flax_all
-            - run_tests_pipelines_torch_all
-            - run_tests_pipelines_tf_all
-            - run_tests_onnxruntime_all
-            - run_tests_hub_all
-
 #    tpu_testing_jobs:
 #        triggers:
 #            - schedule:
--- a/.circleci/deploy.sh
+++ b/.circleci/deploy.sh
@@ -68,6 +68,4 @@ deploy_doc "7a6c9fa" v4.7.0
 deploy_doc "9252a51" v4.8.0
 deploy_doc "1366172" v4.8.1
 deploy_doc "96d1cfb" v4.8.2
-deploy_doc "72aee83" v4.9.0
-deploy_doc "bff1c71" v4.9.1
-deploy_doc "41981a2"  # v4.9.2 Latest stable release
+deploy_doc "72aee83" # v4.9.0 Latest stable release
--- a/.github/workflows/doctests.yml
+++ b/.github/workflows/doctests.yml
@@ -1,42 +0,0 @@
-name: Doctests
-
-on:
-  push:
-    branches:
-      - doctest*
-  repository_dispatch:
-  schedule:
-    - cron: "0 0 * * *"
-
-
-env:
-  HF_HOME: /mnt/cache
-  TRANSFORMERS_IS_CI: yes
-  RUN_SLOW: yes
-  OMP_NUM_THREADS: 16
-  MKL_NUM_THREADS: 16
-  PYTEST_TIMEOUT: 600
-
-jobs:
-  run_doctests:
-    runs-on: [self-hosted, docker-gpu, single-gpu]
-    container:
-      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
-        uses: actions/checkout@v2
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Install dependencies
-        run: |
-          apt -y update && apt install -y libsndfile1-dev
-          pip install --upgrade pip
-          pip install .[dev]
-
-      - name: Run doctests
-        run: |
-          pytest --doctest-modules $(cat utils/documentation_tests.txt) -sv --doctest-continue-on-failure
--- a/.github/workflows/model-templates.yml
+++ b/.github/workflows/model-templates.yml
@@ -59,7 +59,7 @@ jobs:
      - name: Run style changes
        run: |
          git fetch origin master:master
-          make style && make quality
+          make fixup

      - name: Failure short reports
        if: ${{ always() }}
--- a/.github/workflows/self-push.yml
+++ b/.github/workflows/self-push.yml
@@ -11,7 +11,6 @@ on:
      - "tests/**"
      - ".github/**"
      - "templates/**"
-      - "utils/**"
  repository_dispatch:

 env:
@@ -28,47 +27,32 @@ jobs:
      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: Install dependencies
-        run: |
-          apt -y update && apt install -y software-properties-common && apt -y update && add-apt-repository -y ppa:git-core/ppa && apt -y update && apt install -y git
-          apt install -y libsndfile1-dev
-          pip install --upgrade pip
-          pip install .[sklearn,testing,onnxruntime,sentencepiece,torch-speech,vision,timm]
-
      - name: Launcher docker
        uses: actions/checkout@v2
-        with:
-          fetch-depth: 2

      - name: NVIDIA-SMI
        run: |
          nvidia-smi

+      - name: Install dependencies
+        run: |
+          apt -y update && apt install -y libsndfile1-dev
+          pip install --upgrade pip
+          pip install .[sklearn,testing,onnxruntime,sentencepiece,speech,vision,timm]
+
      - 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: Fetch the tests to run
-        run: |
-          python utils/tests_fetcher.py --diff_with_last_commit | tee test_preparation.txt
-
-      - name: Report fetched tests
-        uses: actions/upload-artifact@v2
-        with:
-          name: test_fetched
-          path: test_preparation.txt

      - name: Run all non-slow tests on GPU
        run: |
-          if [ -f test_list.txt ]; then
-            python -m pytest -n 2 --dist=loadfile -v --make-reports=tests_torch_gpu $(cat test_list.txt)
-          fi
+          python -m pytest -n 2 --dist=loadfile -v --make-reports=tests_torch_gpu tests

      - name: Failure short reports
-        if: ${{ failure() }}
+        if: ${{ always() }}
        run: cat reports/tests_torch_gpu_failures_short.txt

      - name: Test suite reports artifacts
@@ -78,61 +62,6 @@ jobs:
          name: run_all_tests_torch_gpu_test_reports
          path: reports

-  run_tests_flax_gpu:
-    runs-on: [self-hosted, docker-gpu, single-gpu]
-    container:
-      image: tensorflow/tensorflow:2.4.1-gpu
-      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    steps:
-      - name: Install dependencies
-        run: |
-          apt -y update && apt install -y software-properties-common && apt -y update && add-apt-repository -y ppa:git-core/ppa && apt -y update && apt install -y git
-          pip install --upgrade "jax[cuda111]" -f https://storage.googleapis.com/jax-releases/jax_releases.html
-          pip install --upgrade pip
-          pip install .[sklearn,testing,sentencepiece,flax,flax-speech,vision]
-
-      - name: Launcher docker
-        uses: actions/checkout@v2
-        with:
-          fetch-depth: 2
-
-      - name: NVIDIA-SMI
-        continue-on-error: true
-        run: |
-          nvidia-smi
-
-      - name: Are GPUs recognized by our DL frameworks
-        run: |
-          python -c "from jax.lib import xla_bridge; print('GPU available:', xla_bridge.get_backend().platform)"
-          python -c "import jax; print('Number of GPUs available:', len(jax.local_devices()))"
-      
-      - name: Fetch the tests to run
-        run: |
-          python utils/tests_fetcher.py --diff_with_last_commit | tee test_preparation.txt
-
-      - name: Report fetched tests
-        uses: actions/upload-artifact@v2
-        with:
-          name: test_fetched
-          path: test_preparation.txt
-
-      - name: Run all non-slow tests on GPU
-        run: |
-          if [ -f test_list.txt ]; then
-            python -m pytest -n 2 --dist=loadfile -v --make-reports=tests_flax_gpu $(cat test_list.txt)
-          fi
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        run: cat reports/tests_flax_gpu_failures_short.txt
-
-      - name: Test suite reports artifacts
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v2
-        with:
-          name: run_all_tests_flax_gpu_test_reports
-          path: reports
-
 #  run_tests_tf_gpu:
 #    runs-on: [self-hosted, docker-gpu, single-gpu]
 #    timeout-minutes: 120
@@ -140,47 +69,32 @@ jobs:
 #      image: tensorflow/tensorflow:2.4.1-gpu
 #      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
 #    steps:
-#      - name: Install dependencies
-#        run: |
-#          apt -y update && apt install -y software-properties-common && apt -y update && add-apt-repository -y ppa:git-core/ppa && apt -y update && apt install -y git
-#          pip install --upgrade pip
-#          pip install .[sklearn,testing,onnxruntime,sentencepiece,tf-speech]
-#
 #      - name: Launcher docker
 #        uses: actions/checkout@v2
-#        with:
-#          fetch-depth: 2
 #
 #      - name: NVIDIA-SMI
 #        run: |
 #          nvidia-smi
 #
+#      - name: Install dependencies
+#        run: |
+#          pip install --upgrade pip
+#          pip install .[sklearn,testing,onnxruntime,sentencepiece]
+#
 #      - name: Are GPUs recognized by our DL frameworks
 #        run: |
 #          TF_CPP_MIN_LOG_LEVEL=3 python -c "import tensorflow as tf; print('TF GPUs available:', bool(tf.config.list_physical_devices('GPU')))"
 #          TF_CPP_MIN_LOG_LEVEL=3 python -c "import tensorflow as tf; print('Number of TF GPUs available:', len(tf.config.list_physical_devices('GPU')))"
 #
-#      - name: Fetch the tests to run
-#        run: |
-#          python utils/tests_fetcher.py --diff_with_last_commit | tee test_preparation.txt
-#
-#      - name: Report fetched tests
-#        uses: actions/upload-artifact@v2
-#        with:
-#          name: test_fetched
-#          path: test_preparation.txt
-#
 #      - name: Run all non-slow tests on GPU
 #        env:
 #          TF_NUM_INTRAOP_THREADS: 8
 #          TF_NUM_INTEROP_THREADS: 1
 #        run: |
-#          if [ -f test_list.txt ]; then
-#            python -m pytest -n 1 --dist=loadfile --make-reports=tests_tf_gpu $(cat test_list.txt)
-#          fi
+#          python -m pytest -n 1 --dist=loadfile --make-reports=tests_tf_gpu tests
 #
 #      - name: Failure short reports
-#        if: ${{ failure() }}
+#        if: ${{ always() }}
 #        run: cat reports/tests_tf_gpu_failures_short.txt
 #
 #      - name: Test suite reports artifacts
@@ -197,23 +111,19 @@ jobs:
      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: Install dependencies
-        run: |
-          apt -y update && apt install -y software-properties-common && apt -y update && add-apt-repository -y ppa:git-core/ppa && apt -y update && apt install -y git
-          apt install -y libsndfile1-dev
-          pip install --upgrade pip
-          pip install .[sklearn,testing,onnxruntime,sentencepiece,torch-speech,vision,timm]
-      
      - name: Launcher docker
        uses: actions/checkout@v2
-        with:
-          fetch-depth: 2

      - name: NVIDIA-SMI
-        continue-on-error: true
        run: |
          nvidia-smi

+      - name: Install dependencies
+        run: |
+          apt -y update && apt install -y libsndfile1-dev
+          pip install --upgrade pip
+          pip install .[sklearn,testing,onnxruntime,sentencepiece,speech,vision,timm]
+
      - name: Are GPUs recognized by our DL frameworks
        run: |
          python -c "import torch; print('Cuda available:', torch.cuda.is_available())"
@@ -221,26 +131,14 @@ jobs:
          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: Fetch the tests to run
-        run: |
-          python utils/tests_fetcher.py --diff_with_last_commit | tee test_preparation.txt
-
-      - name: Report fetched tests
-        uses: actions/upload-artifact@v2
-        with:
-          name: test_fetched
-          path: test_preparation.txt
-
      - name: Run all non-slow tests on GPU
        env:
          MKL_SERVICE_FORCE_INTEL: 1
        run: |
-          if [ -f test_list.txt ]; then
-            python -m pytest -n 2 --dist=loadfile -v --make-reports=tests_torch_multi_gpu $(cat test_list.txt)
-          fi
+          python -m pytest -n 2 --dist=loadfile -v --make-reports=tests_torch_multi_gpu tests

      - name: Failure short reports
-        if: ${{ failure() }}
+        if: ${{ always() }}
        run: cat reports/tests_torch_multi_gpu_failures_short.txt

      - name: Test suite reports artifacts
@@ -250,61 +148,6 @@ jobs:
          name: run_all_tests_torch_multi_gpu_test_reports
          path: reports

-#  run_tests_flax_multi_gpu:
-#    runs-on: [self-hosted, docker-gpu, multi-gpu]
-#    container:
-#      image: tensorflow/tensorflow:2.4.1-gpu
-#      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-#    steps:
-#      - name: Install dependencies
-#        run: |
-#          apt -y update && apt install -y software-properties-common && apt -y update && add-apt-repository -y ppa:git-core/ppa && apt -y update && apt install -y git
-#          pip install --upgrade "jax[cuda111]" -f https://storage.googleapis.com/jax-releases/jax_releases.html
-#          pip install --upgrade pip
-#          pip install .[sklearn,testing,sentencepiece,flax,flax-speech,vision]
-#
-#      - name: Launcher docker
-#        uses: actions/checkout@v2
-#        with:
-#          fetch-depth: 2
-#
-#      - name: NVIDIA-SMI
-#        continue-on-error: true
-#        run: |
-#          nvidia-smi
-#
-#      - name: Are GPUs recognized by our DL frameworks
-#        run: |
-#          python -c "from jax.lib import xla_bridge; print('GPU available:', xla_bridge.get_backend().platform)"
-#          python -c "import jax; print('Number of GPUs available:', len(jax.local_devices()))"
-#      
-#      - name: Fetch the tests to run
-#        run: |
-#          python utils/tests_fetcher.py --diff_with_last_commit | tee test_preparation.txt
-#
-#      - name: Report fetched tests
-#        uses: actions/upload-artifact@v2
-#        with:
-#          name: test_fetched
-#          path: test_preparation.txt
-#
-#      - name: Run all non-slow tests on GPU
-#        run: |
-#          if [ -f test_list.txt ]; then
-#            python -m pytest -n 2 --dist=loadfile -v --make-reports=tests_flax_multi_gpu $(cat test_list.txt)
-#          fi
-#
-#      - name: Failure short reports
-#        if: ${{ failure() }}
-#        run: cat reports/tests_flax_multi_gpu_failures_short.txt
-#
-#      - name: Test suite reports artifacts
-#        if: ${{ always() }}
-#        uses: actions/upload-artifact@v2
-#        with:
-#          name: run_all_tests_flax_multi_gpu_test_reports
-#          path: reports
-
 #  run_tests_tf_multi_gpu:
 #    runs-on: [self-hosted, docker-gpu, multi-gpu]
 #    timeout-minutes: 120
@@ -312,47 +155,32 @@ jobs:
 #      image: tensorflow/tensorflow:2.4.1-gpu
 #      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
 #    steps:
-#      - name: Install dependencies
-#        run: |
-#          apt -y update && apt install -y software-properties-common && apt -y update && add-apt-repository -y ppa:git-core/ppa && apt -y update && apt install -y git
-#          pip install --upgrade pip
-#          pip install .[sklearn,testing,onnxruntime,sentencepiece,tf-speech]
-#
 #      - name: Launcher docker
 #        uses: actions/checkout@v2
-#        with:
-#          fetch-depth: 2
 #
 #      - name: NVIDIA-SMI
 #        run: |
 #          nvidia-smi
 #
+#      - name: Install dependencies
+#        run: |
+#          pip install --upgrade pip
+#          pip install .[sklearn,testing,onnxruntime,sentencepiece]
+#
 #      - name: Are GPUs recognized by our DL frameworks
 #        run: |
 #          TF_CPP_MIN_LOG_LEVEL=3 python -c "import tensorflow as tf; print('TF GPUs available:', bool(tf.config.list_physical_devices('GPU')))"
 #          TF_CPP_MIN_LOG_LEVEL=3 python -c "import tensorflow as tf; print('Number of TF GPUs available:', len(tf.config.list_physical_devices('GPU')))"
 #
-#      - name: Fetch the tests to run
-#        run: |
-#          python utils/tests_fetcher.py --diff_with_last_commit | tee test_preparation.txt
-#
-#      - name: Report fetched tests
-#        uses: actions/upload-artifact@v2
-#        with:
-#          name: test_fetched
-#          path: test_preparation.txt
-#
 #      - name: Run all non-slow tests on GPU
 #        env:
 #          TF_NUM_INTRAOP_THREADS: 8
 #          TF_NUM_INTEROP_THREADS: 1
 #        run: |
-#          if [ -f test_list.txt ]; then
-#            python -m pytest -n 1 --dist=loadfile --make-reports=tests_tf_multi_gpu $(cat test_list.txt)
-#          fi
+#          python -m pytest -n 1 --dist=loadfile --make-reports=tests_tf_multi_gpu tests
 #
 #      - name: Failure short reports
-#        if: ${{ failure() }}
+#        if: ${{ always() }}
 #        run: cat reports/tests_tf_multi_gpu_failures_short.txt
 #
 #      - name: Test suite reports artifacts
@@ -370,8 +198,6 @@ jobs:
    steps:
      - name: Launcher docker
        uses: actions/checkout@v2
-        with:
-          fetch-depth: 2

      - name: NVIDIA-SMI
        run: |
@@ -389,25 +215,13 @@ jobs:
          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: Fetch the tests to run
-        run: |
-          python utils/tests_fetcher.py --diff_with_last_commit --filters tests/deepspeed tests/extended | tee test_preparation.txt
-      
-      - name: Report fetched tests
-        uses: actions/upload-artifact@v2
-        with:
-          name: test_fetched
-          path: test_preparation.txt

      - name: Run all tests on GPU
        run: |
-          if [ -f test_list.txt ]; then
-            python -m pytest -n 1 --dist=loadfile -v --make-reports=tests_torch_cuda_extensions_gpu $(cat test_list.txt)
-          fi
+          python -m pytest -n 1 --dist=loadfile -v --make-reports=tests_torch_cuda_extensions_gpu tests/deepspeed tests/extended

      - name: Failure short reports
-        if: ${{ failure() }}
+        if: ${{ always() }}
        run: cat reports/tests_torch_cuda_extensions_gpu_failures_short.txt

      - name: Test suite reports artifacts
@@ -425,11 +239,8 @@ jobs:
    steps:
      - name: Launcher docker
        uses: actions/checkout@v2
-        with:
-          fetch-depth: 2

      - name: NVIDIA-SMI
-        continue-on-error: true
        run: |
          nvidia-smi

@@ -446,24 +257,12 @@ jobs:
          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: Fetch the tests to run
-        run: |
-          python utils/tests_fetcher.py --diff_with_last_commit --filters tests/deepspeed tests/extended | tee test_preparation.txt
-
-      - name: Report fetched tests
-        uses: actions/upload-artifact@v2
-        with:
-          name: test_fetched
-          path: test_preparation.txt
-
      - name: Run all tests on GPU
        run: |
-          if [ -f test_list.txt ]; then
-            python -m pytest -n 1 --dist=loadfile -v --make-reports=tests_torch_cuda_extensions_multi_gpu $(cat test_list.txt)
-          fi
+          python -m pytest -n 1 --dist=loadfile -v --make-reports=tests_torch_cuda_extensions_multi_gpu tests/deepspeed tests/extended

      - name: Failure short reports
-        if: ${{ failure() }}
+        if: ${{ always() }}
        run: cat reports/tests_torch_cuda_extensions_multi_gpu_failures_short.txt

      - name: Test suite reports artifacts
--- a/.github/workflows/self-scheduled.yml
+++ b/.github/workflows/self-scheduled.yml
@@ -32,9 +32,9 @@ jobs:

      - name: Install dependencies
        run: |
-          apt -y update && apt install -y libsndfile1-dev git
+          apt -y update && apt install -y libsndfile1-dev
          pip install --upgrade pip
-          pip install .[integrations,sklearn,testing,onnxruntime,sentencepiece,torch-speech,vision,timm]
+          pip install .[integrations,sklearn,testing,onnxruntime,sentencepiece,speech,vision,timm]

      - name: Are GPUs recognized by our DL frameworks
        run: |
@@ -85,46 +85,6 @@ jobs:
          name: run_all_tests_torch_gpu_test_reports
          path: reports

-  run_all_tests_flax_gpu:
-    runs-on: [self-hosted, docker-gpu-test, single-gpu]
-    container:
-      image: tensorflow/tensorflow:2.4.1-gpu
-      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
-        continue-on-error: true
-        run: |
-          nvidia-smi
-
-      - name: Install dependencies
-        run: |
-          pip install --upgrade pip
-          pip install --upgrade "jax[cuda111]" -f https://storage.googleapis.com/jax-releases/jax_releases.html
-          pip install .[flax,integrations,sklearn,testing,sentencepiece,flax-speech,vision]
-
-      - name: Are GPUs recognized by our DL frameworks
-        run: |
-          python -c "from jax.lib import xla_bridge; print('GPU available:', xla_bridge.get_backend().platform)"
-          python -c "import jax; print('Number of GPUs available:', len(jax.local_devices()))"
-
-      - name: Run all tests on GPU
-        run: |
-          python -m pytest -n 1 -v --dist=loadfile --make-reports=tests_flax_gpu tests
-
-      - name: Failure short reports
-        if: ${{ always() }}
-        run: cat reports/tests_flax_gpu_failures_short.txt
-
-      - name: Test suite reports artifacts
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v2
-        with:
-          name: run_all_tests_flax_gpu_test_reports
-          path: reports
-
  run_all_tests_tf_gpu:
    runs-on: [self-hosted, docker-gpu, single-gpu]
    container:
@@ -140,9 +100,8 @@ jobs:

      - name: Install dependencies
        run: |
-          apt -y update && apt install -y git
          pip install --upgrade pip
-          pip install .[sklearn,testing,onnx,sentencepiece,tf-speech]
+          pip install .[sklearn,testing,onnx,sentencepiece]

      - name: Are GPUs recognized by our DL frameworks
        run: |
@@ -190,15 +149,14 @@ jobs:
        uses: actions/checkout@v2

      - name: NVIDIA-SMI
-        continue-on-error: true
        run: |
          nvidia-smi

      - name: Install dependencies
        run: |
-          apt -y update && apt install -y libsndfile1-dev git
+          apt -y update && apt install -y libsndfile1-dev
          pip install --upgrade pip
-          pip install .[integrations,sklearn,testing,onnxruntime,sentencepiece,torch-speech,vision,timm]
+          pip install .[integrations,sklearn,testing,onnxruntime,sentencepiece,speech,vision,timm]

      - name: Are GPUs recognized by our DL frameworks
        run: |
@@ -245,15 +203,13 @@ jobs:
        uses: actions/checkout@v2

      - name: NVIDIA-SMI
-        continue-on-error: true
        run: |
          nvidia-smi

      - name: Install dependencies
        run: |
-          apt -y update && apt install -y git
          pip install --upgrade pip
-          pip install .[sklearn,testing,onnx,sentencepiece,tf-speech]
+          pip install .[sklearn,testing,onnx,sentencepiece]

      - name: Are GPUs recognized by our DL frameworks
        run: |
@@ -291,45 +247,6 @@ jobs:
          name: run_all_tests_tf_multi_gpu_test_reports
          path: reports

-#  run_all_tests_flax_multi_gpu:
-#    runs-on: [self-hosted, docker-gpu, multi-gpu]
-#    container:
-#      image: tensorflow/tensorflow:2.4.1-gpu
-#      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: |
-#          pip install --upgrade pip
-#          pip install --upgrade "jax[cuda111]" -f https://storage.googleapis.com/jax-releases/jax_releases.html
-#          pip install .[flax,integrations,sklearn,testing,sentencepiece,flax-speech,vision]
-#
-#      - name: Are GPUs recognized by our DL frameworks
-#        run: |
-#          python -c "from jax.lib import xla_bridge; print('GPU available:', xla_bridge.get_backend().platform)"
-#          python -c "import jax; print('Number of GPUs available:', len(jax.local_devices()))"
-#
-#      - name: Run all tests on GPU
-#        run: |
-#          python -m pytest -n 1 -v --dist=loadfile --make-reports=tests_flax_gpu tests
-#
-#      - name: Failure short reports
-#        if: ${{ always() }}
-#        run: cat reports/tests_flax_gpu_failures_short.txt
-#
-#      - name: Test suite reports artifacts
-#        if: ${{ always() }}
-#        uses: actions/upload-artifact@v2
-#        with:
-#          name: run_all_tests_flax_gpu_test_reports
-#          path: reports
-
  run_all_tests_torch_cuda_extensions_gpu:
    runs-on: [self-hosted, docker-gpu, single-gpu]
    container:
@@ -381,7 +298,6 @@ jobs:
        uses: actions/checkout@v2

      - name: NVIDIA-SMI
-        continue-on-error: true
        run: |
          nvidia-smi

@@ -434,7 +350,6 @@ jobs:
        env:
          CI_SLACK_BOT_TOKEN: ${{ secrets.CI_SLACK_BOT_TOKEN }}
          CI_SLACK_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID }}
-          CI_SLACK_CHANNEL_ID_DAILY: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY }}


        run: |
--- a/CITATION.cff
+++ b/CITATION.cff
@@ -1,82 +0,0 @@
-cff-version: "1.2.0"
-date-released: 2020-10
-message: "If you use this software, please cite it using these metadata."
-title: "Transformers: State-of-the-Art Natural Language Processing"
-url: "https://github.com/huggingface/transformers"
-authors: 
-  - family-names: Wolf
-    given-names: Thomas
-  - family-names: Debut
-    given-names: Lysandre
-  - family-names: Sanh
-    given-names: Victor
-  - family-names: Chaumond
-    given-names: Julien
-  - family-names: Delangue
-    given-names: Clement
-  - family-names: Moi
-    given-names: Anthony
-  - family-names: Cistac
-    given-names: Perric
-  - family-names: Ma
-    given-names: Clara
-  - family-names: Jernite
-    given-names: Yacine
-  - family-names: Plu
-    given-names: Julien
-  - family-names: Xu
-    given-names: Canwen
-  - family-names: "Le Scao"
-    given-names: Teven
-  - family-names: Gugger
-    given-names: Sylvain
-  - family-names: Drame
-    given-names: Mariama
-  - family-names: Lhoest
-    given-names: Quentin
-  - family-names: Rush
-    given-names: "Alexander M."
-preferred-citation:
-  type: inproceedings
-  authors:
-  - family-names: Wolf
-    given-names: Thomas
-  - family-names: Debut
-    given-names: Lysandre
-  - family-names: Sanh
-    given-names: Victor
-  - family-names: Chaumond
-    given-names: Julien
-  - family-names: Delangue
-    given-names: Clement
-  - family-names: Moi
-    given-names: Anthony
-  - family-names: Cistac
-    given-names: Perric
-  - family-names: Ma
-    given-names: Clara
-  - family-names: Jernite
-    given-names: Yacine
-  - family-names: Plu
-    given-names: Julien
-  - family-names: Xu
-    given-names: Canwen
-  - family-names: "Le Scao"
-    given-names: Teven
-  - family-names: Gugger
-    given-names: Sylvain
-  - family-names: Drame
-    given-names: Mariama
-  - family-names: Lhoest
-    given-names: Quentin
-  - family-names: Rush
-    given-names: "Alexander M."
-  booktitle: "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations"
-  month: 10
-  start: 38
-  end: 45
-  title: "Transformers: State-of-the-Art Natural Language Processing"
-  year: 2020
-  publisher: "Association for Computational Linguistics"
-  url: "https://www.aclweb.org/anthology/2020.emnlp-demos.6"
-  address: "Online"
--- a/1
+++ b/1
@@ -30,6 +30,7 @@ deps_table_check_updated:
 # autogenerating code

 autogenerate_code: deps_table_update
+	python utils/class_mapping_update.py

 # Check that source code meets quality standards

--- a/README.md
+++ b/README.md
@@ -211,7 +211,6 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 1. **[ALBERT](https://huggingface.co/transformers/model_doc/albert.html)** (from Google Research and the Toyota Technological Institute at Chicago) released with the paper [ALBERT: A Lite BERT for Self-supervised Learning of Language Representations](https://arxiv.org/abs/1909.11942), by Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, Radu Soricut.
 1. **[BART](https://huggingface.co/transformers/model_doc/bart.html)** (from Facebook) released with the paper [BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension](https://arxiv.org/pdf/1910.13461.pdf) by Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov and Luke Zettlemoyer.
 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. **[BEiT](https://huggingface.co/transformers/model_doc/beit.html)** (from Microsoft) released with the paper [BEiT: BERT Pre-Training of Image Transformers](https://arxiv.org/abs/2106.08254) by Hangbo Bao, Li Dong, Furu Wei.
 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.
@@ -244,8 +243,6 @@ Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih.
 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. **[LayoutLMv2](https://huggingface.co/transformers/model_doc/layoutlmv2.html)** (from Microsoft Research Asia) released with the paper [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) by Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou.
-1. **[LayoutXLM](https://huggingface.co/transformers/model_doc/layoutlmv2.html)** (from Microsoft Research Asia) released with the paper [LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding](https://arxiv.org/abs/2104.08836) by Yiheng Xu, Tengchao Lv, Lei Cui, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Furu Wei.
 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.
@@ -261,11 +258,9 @@ Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih.
 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. **[RemBERT](https://huggingface.co/transformers/model_doc/rembert.html)** (from Google Research) released with the paper [Rethinking embedding coupling in pre-trained language models](https://arxiv.org/pdf/2010.12821.pdf) by Hyung Won Chung, Thibault Févry, Henry Tsai, M. Johnson, Sebastian Ruder.
 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. **[Splinter](https://huggingface.co/transformers/model_doc/splinter.html)** (from Tel Aviv University), released together with the paper [Few-Shot Question Answering by Pretraining Span Selection](https://arxiv.org/abs/2101.00438) by Ori Ram, Yuval Kirstain, Jonathan Berant, Amir Globerson, Omer Levy.
 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.
--- a/docs/source/_static/js/custom.js
+++ b/docs/source/_static/js/custom.js
@@ -1,10 +1,10 @@
 // These two things need to be updated at each release for the version selector.
 // Last stable version
-const stableVersion = "v4.9.2"
+const stableVersion = "v4.9.0"
 // Dictionary doc folder to label. The last stable version should have an empty key.
 const versionMapping = {
    "master": "master",
-    "": "v4.9.0/v4.9.1/v4.9.2 (stable)",
+    "": "v4.9.0 (stable)",
    "v4.8.2": "v4.8.0/v4.8.1/v4.8.2",
    "v4.7.0": "v4.7.0",
    "v4.6.0": "v4.6.0",
--- a/docs/source/community.md
+++ b/docs/source/community.md
@@ -1,4 +1,4 @@
-# Community
+# Community

 This page regroups resources around 🤗 Transformers developed by the community.

@@ -12,7 +12,6 @@ This page regroups resources around 🤗 Transformers developed by the community

 | Notebook     |      Description      |      Author      |      |
 |:----------|:-------------|:-------------|------:|
-| [Fine-tune a pre-trained Transformer to generate lyrics](https://github.com/AlekseyKorshuk/huggingartists) | How to generate lyrics in the style of your favorite artist by fine-tuning a GPT-2 model |  [Aleksey Korshuk](https://github.com/AlekseyKorshuk) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/AlekseyKorshuk/huggingartists/blob/master/huggingartists-demo.ipynb) |
 | [Train T5 in Tensorflow 2 ](https://github.com/snapthat/TF-T5-text-to-text) | How to train T5 for any task using Tensorflow 2. This notebook demonstrates a Question & Answer task implemented in Tensorflow 2 using SQUAD | [Muhammad Harris](https://github.com/HarrisDePerceptron) |[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/snapthat/TF-T5-text-to-text/blob/master/snapthatT5/notebooks/TF-T5-Datasets%20Training.ipynb) |
 | [Train T5 on TPU](https://github.com/patil-suraj/exploring-T5/blob/master/T5_on_TPU.ipynb)  | How to train T5 on SQUAD with Transformers and Nlp | [Suraj Patil](https://github.com/patil-suraj) |[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/patil-suraj/exploring-T5/blob/master/T5_on_TPU.ipynb#scrollTo=QLGiFCDqvuil) |
 | [Fine-tune T5 for Classification and Multiple Choice](https://github.com/patil-suraj/exploring-T5/blob/master/t5_fine_tuning.ipynb)  | How to fine-tune T5 for classification and multiple choice tasks using a text-to-text format with PyTorch Lightning |  [Suraj Patil](https://github.com/patil-suraj) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/patil-suraj/exploring-T5/blob/master/t5_fine_tuning.ipynb) |
--- a/docs/source/conf.py
+++ b/docs/source/conf.py
@@ -27,11 +27,7 @@ author = "huggingface"
 # The short X.Y version
 version = ""
 # The full version, including alpha/beta/rc tags
-release = "4.10.3"
-
-
-
-
+release = u'4.7.0'



@@ -212,9 +208,6 @@ epub_title = project
 # A list of files that should not be packed into the epub file.
 epub_exclude_files = ["search.html"]

-# Localization
-locale_dirs = ['locale/']
-gettext_compact = False

 def setup(app):
    app.add_css_file("css/huggingface.css")
--- a/docs/source/index.rst
+++ b/docs/source/index.rst
@@ -105,202 +105,187 @@ Supported models
 3. :doc:`BARThez <model_doc/barthez>` (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.
-4. :doc:`BEiT <model_doc/beit>` (from Microsoft) released with the paper `BEiT: BERT Pre-Training of Image Transformers
-   <https://arxiv.org/abs/2106.08254>`__ by Hangbo Bao, Li Dong, Furu Wei.
-5. :doc:`BERT <model_doc/bert>` (from Google) released with the paper `BERT: Pre-training of Deep Bidirectional
+4. :doc:`BERT <model_doc/bert>` (from Google) released with the paper `BERT: Pre-training of Deep Bidirectional
   Transformers for Language Understanding <https://arxiv.org/abs/1810.04805>`__ by Jacob Devlin, Ming-Wei Chang,
   Kenton Lee and Kristina Toutanova.
-6. :doc:`BERT For Sequence Generation <model_doc/bertgeneration>` (from Google) released with the paper `Leveraging
+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.
-7. :doc:`BigBird-RoBERTa <model_doc/bigbird>` (from Google Research) released with the paper `Big Bird: Transformers
+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.
-8. :doc:`BigBird-Pegasus <model_doc/bigbird_pegasus>` (from Google Research) released with the paper `Big Bird:
+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.
-9. :doc:`Blenderbot <model_doc/blenderbot>` (from Facebook) released with the paper `Recipes for building an
+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.
-10. :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.
-11. :doc:`BORT <model_doc/bort>` (from Alexa) released with the paper `Optimal Subarchitecture Extraction For BERT
+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.
+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.
-12. :doc:`ByT5 <model_doc/byt5>` (from Google Research) released with the paper `ByT5: Towards a token-free future with
+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.
-13. :doc:`CamemBERT <model_doc/camembert>` (from Inria/Facebook/Sorbonne) released with the paper `CamemBERT: a Tasty
+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.
-14. :doc:`CANINE <model_doc/canine>` (from Google Research) released with the paper `CANINE: Pre-training an Efficient
+13. :doc:`CANINE <model_doc/canine>` (from Google Research) released with the paper `CANINE: Pre-training an Efficient
    Tokenization-Free Encoder for Language Representation <https://arxiv.org/abs/2103.06874>`__ by Jonathan H. Clark,
    Dan Garrette, Iulia Turc, John Wieting.
-15. :doc:`CLIP <model_doc/clip>` (from OpenAI) released with the paper `Learning Transferable Visual Models From
+14. :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.
-16. :doc:`ConvBERT <model_doc/convbert>` (from YituTech) released with the paper `ConvBERT: Improving BERT with
+15. :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.
-17. :doc:`CPM <model_doc/cpm>` (from Tsinghua University) released with the paper `CPM: A Large-scale Generative
+16. :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.
-18. :doc:`CTRL <model_doc/ctrl>` (from Salesforce) released with the paper `CTRL: A Conditional Transformer Language
+17. :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.
-19. :doc:`DeBERTa <model_doc/deberta>` (from Microsoft) released with the paper `DeBERTa: Decoding-enhanced BERT with
+18. :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.
-20. :doc:`DeBERTa-v2 <model_doc/deberta_v2>` (from Microsoft) released with the paper `DeBERTa: Decoding-enhanced BERT
+19. :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.
-21. :doc:`DeiT <model_doc/deit>` (from Facebook) released with the paper `Training data-efficient image transformers &
+20. :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.
-22. :doc:`DETR <model_doc/detr>` (from Facebook) released with the paper `End-to-End Object Detection with Transformers
+21. :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.
-23. :doc:`DialoGPT <model_doc/dialogpt>` (from Microsoft Research) released with the paper `DialoGPT: Large-Scale
+22. :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.
-24. :doc:`DistilBERT <model_doc/distilbert>` (from HuggingFace), released together with the paper `DistilBERT, a
+23. :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.
-25. :doc:`DPR <model_doc/dpr>` (from Facebook) released with the paper `Dense Passage Retrieval for Open-Domain
+24. :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.
-26. :doc:`ELECTRA <model_doc/electra>` (from Google Research/Stanford University) released with the paper `ELECTRA:
+25. :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.
-27. :doc:`FlauBERT <model_doc/flaubert>` (from CNRS) released with the paper `FlauBERT: Unsupervised Language Model
+26. :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.
-28. :doc:`Funnel Transformer <model_doc/funnel>` (from CMU/Google Brain) released with the paper `Funnel-Transformer:
+27. :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.
-29. :doc:`GPT <model_doc/gpt>` (from OpenAI) released with the paper `Improving Language Understanding by Generative
+28. :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.
-30. :doc:`GPT-2 <model_doc/gpt2>` (from OpenAI) released with the paper `Language Models are Unsupervised Multitask
+29. :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**.
-31. :doc:`GPT Neo <model_doc/gpt_neo>` (from EleutherAI) released in the repository `EleutherAI/gpt-neo
+30. :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.
-32. :doc:`Hubert <model_doc/hubert>` (from Facebook) released with the paper `HuBERT: Self-Supervised Speech
+31. :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.
-33. :doc:`I-BERT <model_doc/ibert>` (from Berkeley) released with the paper `I-BERT: Integer-only BERT Quantization
+32. :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
-34. :doc:`LayoutLM <model_doc/layoutlm>` (from Microsoft Research Asia) released with the paper `LayoutLM: Pre-training
+33. :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.
-35. :doc:`LayoutLMv2 <model_doc/layoutlmv2>` (from Microsoft Research Asia) released with the paper `LayoutLMv2:
-    Multi-modal Pre-training for Visually-Rich Document Understanding <https://arxiv.org/abs/2012.14740>`__ by Yang Xu,
-    Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min
-    Zhang, Lidong Zhou.
-36. :doc:`LayoutXLM <model_doc/layoutlmv2>` (from Microsoft Research Asia) released with the paper `LayoutXLM:
-    Multimodal Pre-training for Multilingual Visually-rich Document Understanding <https://arxiv.org/abs/2104.08836>`__
-    by Yiheng Xu, Tengchao Lv, Lei Cui, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Furu Wei.
-37. :doc:`LED <model_doc/led>` (from AllenAI) released with the paper `Longformer: The Long-Document Transformer
+34. :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.
-38. :doc:`Longformer <model_doc/longformer>` (from AllenAI) released with the paper `Longformer: The Long-Document
+35. :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.
-39. :doc:`LUKE <model_doc/luke>` (from Studio Ousia) released with the paper `LUKE: Deep Contextualized Entity
+36. :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.
-40. :doc:`LXMERT <model_doc/lxmert>` (from UNC Chapel Hill) released with the paper `LXMERT: Learning Cross-Modality
+37. :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.
-41. :doc:`M2M100 <model_doc/m2m_100>` (from Facebook) released with the paper `Beyond English-Centric Multilingual
+38. :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.
-42. :doc:`MarianMT <model_doc/marian>` Machine translation models trained using `OPUS <http://opus.nlpl.eu/>`__ data by
+39. :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.
-43. :doc:`MBart <model_doc/mbart>` (from Facebook) released with the paper `Multilingual Denoising Pre-training for
+40. :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.
-44. :doc:`MBart-50 <model_doc/mbart>` (from Facebook) released with the paper `Multilingual Translation with Extensible
+41. :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.
-45. :doc:`Megatron-BERT <model_doc/megatron_bert>` (from NVIDIA) released with the paper `Megatron-LM: Training
+42. :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.
-46. :doc:`Megatron-GPT2 <model_doc/megatron_gpt2>` (from NVIDIA) released with the paper `Megatron-LM: Training
+43. :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.
-47. :doc:`MPNet <model_doc/mpnet>` (from Microsoft Research) released with the paper `MPNet: Masked and Permuted
+44. :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.
-48. :doc:`MT5 <model_doc/mt5>` (from Google AI) released with the paper `mT5: A massively multilingual pre-trained
+45. :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.
-49. :doc:`Pegasus <model_doc/pegasus>` (from Google) released with the paper `PEGASUS: Pre-training with Extracted
+46. :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.
-50. :doc:`ProphetNet <model_doc/prophetnet>` (from Microsoft Research) released with the paper `ProphetNet: Predicting
+47. :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.
-51. :doc:`Reformer <model_doc/reformer>` (from Google Research) released with the paper `Reformer: The Efficient
+48. :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.
-52. :doc:`RemBERT <model_doc/rembert>` (from Google Research) released with the paper `Rethinking embedding coupling in
-    pre-trained language models <https://arxiv.org/pdf/2010.12821.pdf>`__ by Hyung Won Chung, Thibault Févry, Henry
-    Tsai, M. Johnson, Sebastian Ruder.
-53. :doc:`RoBERTa <model_doc/roberta>` (from Facebook), released together with the paper a `Robustly Optimized BERT
+49. :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.
-54. :doc:`RoFormer <model_doc/roformer>` (from ZhuiyiTechnology), released together with the paper a `RoFormer:
+50. :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.
-55. :doc:`SpeechToTextTransformer <model_doc/speech_to_text>` (from Facebook), released together with the paper
+51. :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.
-56. :doc:`Splinter <model_doc/splinter>` (from Tel Aviv University), released together with the paper `Few-Shot
-    Question Answering by Pretraining Span Selection <https://arxiv.org/abs/2101.00438>`__ by Ori Ram, Yuval Kirstain,
-    Jonathan Berant, Amir Globerson, Omer Levy.
-57. :doc:`SqueezeBert <model_doc/squeezebert>` released with the paper `SqueezeBERT: What can computer vision teach NLP
+52. :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.
-58. :doc:`T5 <model_doc/t5>` (from Google AI) released with the paper `Exploring the Limits of Transfer Learning with a
+53. :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.
-59. :doc:`TAPAS <model_doc/tapas>` (from Google AI) released with the paper `TAPAS: Weakly Supervised Table Parsing via
+54. :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.
-60. :doc:`Transformer-XL <model_doc/transformerxl>` (from Google/CMU) released with the paper `Transformer-XL:
+55. :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.
-61. :doc:`Vision Transformer (ViT) <model_doc/vit>` (from Google AI) released with the paper `An Image is Worth 16x16
+56. :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.
-62. :doc:`VisualBERT <model_doc/visual_bert>` (from UCLA NLP) released with the paper `VisualBERT: A Simple and
+57. :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.
-63. :doc:`Wav2Vec2 <model_doc/wav2vec2>` (from Facebook AI) released with the paper `wav2vec 2.0: A Framework for
+58. :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.
-64. :doc:`XLM <model_doc/xlm>` (from Facebook) released together with the paper `Cross-lingual Language Model
+59. :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.
-65. :doc:`XLM-ProphetNet <model_doc/xlmprophetnet>` (from Microsoft Research) released with the paper `ProphetNet:
+60. :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.
-66. :doc:`XLM-RoBERTa <model_doc/xlmroberta>` (from Facebook AI), released together with the paper `Unsupervised
+61. :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.
-67. :doc:`XLNet <model_doc/xlnet>` (from Google/CMU) released with the paper `XLNet: Generalized Autoregressive
+62. :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.
-68. :doc:`XLSR-Wav2Vec2 <model_doc/xlsr_wav2vec2>` (from Facebook AI) released with the paper `Unsupervised
+63. :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.

@@ -320,12 +305,10 @@ Flax), PyTorch, and/or TensorFlow.
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |            Model            | Tokenizer slow | Tokenizer fast | PyTorch support | TensorFlow support | Flax Support |
 +=============================+================+================+=================+====================+==============+
-|           ALBERT            |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
+|           ALBERT            |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |            BART             |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|            BeiT             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |            BERT             |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |       Bert Generation       |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
@@ -338,31 +321,31 @@ Flax), PyTorch, and/or TensorFlow.
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |       BlenderbotSmall       |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
+|            CLIP             |       ✅       |       ✅       |       ✅        |         ❌         |      ✅      |
+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
+|            CTRL             |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |          CamemBERT          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |           Canine            |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|            CLIP             |       ✅       |       ✅       |       ✅        |         ❌         |      ✅      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |          ConvBERT           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|            CTRL             |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|           DeBERTa           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|         DeBERTa-v2          |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|            DeiT             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |            DETR             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|         DistilBERT          |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |             DPR             |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
+|           DeBERTa           |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
+|         DeBERTa-v2          |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
+|            DeiT             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
+|         DistilBERT          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |           ELECTRA           |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|       Encoder decoder       |       ❌       |       ❌       |       ✅        |         ❌         |      ✅      |
+|       Encoder decoder       |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 | FairSeq Machine-Translation |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
@@ -376,32 +359,26 @@ Flax), PyTorch, and/or TensorFlow.
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |           I-BERT            |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|          LayoutLM           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|         LayoutLMv2          |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |             LED             |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|         Longformer          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |            LUKE             |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |           LXMERT            |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
+|          LayoutLM           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
+|         Longformer          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |           M2M100            |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|           Marian            |       ✅       |       ❌       |       ✅        |         ✅         |      ✅      |
+|            MPNet            |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|            mBART            |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
+|           Marian            |       ✅       |       ❌       |       ✅        |         ✅         |      ✅      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |        MegatronBert         |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |         MobileBERT          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|            MPNet            |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|             mT5             |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |         OpenAI GPT          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |        OpenAI GPT-2         |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
@@ -414,8 +391,6 @@ Flax), PyTorch, and/or TensorFlow.
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |          Reformer           |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|           RemBERT           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |          RetriBERT          |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |           RoBERTa           |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
@@ -424,8 +399,6 @@ Flax), PyTorch, and/or TensorFlow.
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |         Speech2Text         |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|          Splinter           |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |         SqueezeBERT         |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |             T5              |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
@@ -434,10 +407,10 @@ Flax), PyTorch, and/or TensorFlow.
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |       Transformer-XL        |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|         VisualBert          |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |             ViT             |       ❌       |       ❌       |       ✅        |         ❌         |      ✅      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
+|         VisualBert          |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |          Wav2Vec2           |       ✅       |       ❌       |       ✅        |         ✅         |      ✅      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |             XLM             |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
@@ -448,6 +421,10 @@ Flax), PyTorch, and/or TensorFlow.
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |            XLNet            |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
+|            mBART            |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
+|             mT5             |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
+-----------------------------+----------------+----------------+-----------------+--------------------+--------------+

 .. toctree::
    :maxdepth: 2
@@ -526,7 +503,6 @@ Flax), PyTorch, and/or TensorFlow.
    model_doc/auto
    model_doc/bart
    model_doc/barthez
-    model_doc/beit
    model_doc/bert
    model_doc/bertweet
    model_doc/bertgeneration
@@ -558,8 +534,6 @@ Flax), PyTorch, and/or TensorFlow.
    model_doc/herbert
    model_doc/ibert
    model_doc/layoutlm
-    model_doc/layoutlmv2
-    model_doc/layoutxlm
    model_doc/led
    model_doc/longformer
    model_doc/luke
@@ -581,12 +555,10 @@ Flax), PyTorch, and/or TensorFlow.
    model_doc/prophetnet
    model_doc/rag
    model_doc/reformer
-    model_doc/rembert
    model_doc/retribert
    model_doc/roberta
    model_doc/roformer
    model_doc/speech_to_text
-    model_doc/splinter
    model_doc/squeezebert
    model_doc/t5
    model_doc/tapas
--- a/docs/source/internal/modeling_utils.rst
+++ b/docs/source/internal/modeling_utils.rst
@@ -63,6 +63,7 @@ TensorFlow custom layers
    :members: call

 .. autoclass:: transformers.modeling_tf_utils.TFSequenceSummary
+    :members: call


 TensorFlow loss functions
--- a/docs/source/main_classes/data_collator.rst
+++ b/docs/source/main_classes/data_collator.rst
@@ -18,7 +18,7 @@ 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)
-on the formed batch.
+oin the formed batch.

 Examples of use can be found in the :doc:`example scripts <../examples>` or :doc:`example notebooks <../notebooks>`.

@@ -54,18 +54,18 @@ DataCollatorForLanguageModeling
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

 .. autoclass:: transformers.data.data_collator.DataCollatorForLanguageModeling
-    :members: numpy_mask_tokens, tf_mask_tokens, torch_mask_tokens
+    :members: mask_tokens


 DataCollatorForWholeWordMask
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

 .. autoclass:: transformers.data.data_collator.DataCollatorForWholeWordMask
-    :members: numpy_mask_tokens, tf_mask_tokens, torch_mask_tokens
+    :members: mask_tokens


 DataCollatorForPermutationLanguageModeling
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

 .. autoclass:: transformers.data.data_collator.DataCollatorForPermutationLanguageModeling
-    :members: numpy_mask_tokens, tf_mask_tokens, torch_mask_tokens
+    :members: mask_tokens
--- a/docs/source/main_classes/output.rst
+++ b/docs/source/main_classes/output.rst
@@ -299,93 +299,3 @@ TFSeq2SeqQuestionAnsweringModelOutput

 .. autoclass:: transformers.modeling_tf_outputs.TFSeq2SeqQuestionAnsweringModelOutput
    :members:
-
-
-FlaxBaseModelOutput
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.modeling_flax_outputs.FlaxBaseModelOutput
-
-
-FlaxBaseModelOutputWithPast
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.modeling_flax_outputs.FlaxBaseModelOutputWithPast
-
-
-FlaxBaseModelOutputWithPooling
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.modeling_flax_outputs.FlaxBaseModelOutputWithPooling
-
-
-FlaxBaseModelOutputWithPastAndCrossAttentions
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.modeling_flax_outputs.FlaxBaseModelOutputWithPastAndCrossAttentions
-
-
-FlaxSeq2SeqModelOutput
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.modeling_flax_outputs.FlaxSeq2SeqModelOutput
-
-
-FlaxCausalLMOutputWithCrossAttentions
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.modeling_flax_outputs.FlaxCausalLMOutputWithCrossAttentions
-
-
-FlaxMaskedLMOutput
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.modeling_flax_outputs.FlaxMaskedLMOutput
-
-
-FlaxSeq2SeqLMOutput
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.modeling_flax_outputs.FlaxSeq2SeqLMOutput
-
-
-FlaxNextSentencePredictorOutput
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.modeling_flax_outputs.FlaxNextSentencePredictorOutput
-
-
-FlaxSequenceClassifierOutput
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.modeling_flax_outputs.FlaxSequenceClassifierOutput
-
-
-FlaxSeq2SeqSequenceClassifierOutput
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.modeling_flax_outputs.FlaxSeq2SeqSequenceClassifierOutput
-
-
-FlaxMultipleChoiceModelOutput
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.modeling_flax_outputs.FlaxMultipleChoiceModelOutput
-
-
-FlaxTokenClassifierOutput
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.modeling_flax_outputs.FlaxTokenClassifierOutput
-
-
-FlaxQuestionAnsweringModelOutput
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.modeling_flax_outputs.FlaxQuestionAnsweringModelOutput
-
-
-FlaxSeq2SeqQuestionAnsweringModelOutput
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.modeling_flax_outputs.FlaxSeq2SeqQuestionAnsweringModelOutput
--- a/docs/source/main_classes/trainer.rst
+++ b/docs/source/main_classes/trainer.rst
@@ -197,7 +197,7 @@ which should make the "stop and resume" style of training as close as possible t
 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 deterministic (.e.g., ``torch.backends.cudnn.deterministic``) may slow things down, therefore this
+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.


--- a/docs/source/model_doc/albert.rst
+++ b/docs/source/model_doc/albert.rst
@@ -43,8 +43,7 @@ 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.

-This model was contributed by `lysandre <https://huggingface.co/lysandre>`__. This model jax version was contributed by
-`kamalkraj <https://huggingface.co/kamalkraj>`__. The original code can be found `here
+This model was contributed by `lysandre <https://huggingface.co/lysandre>`__. The original code can be found `here
 <https://github.com/google-research/ALBERT>`__.

 AlbertConfig
@@ -175,52 +174,3 @@ TFAlbertForQuestionAnswering

 .. autoclass:: transformers.TFAlbertForQuestionAnswering
    :members: call
-
-
-FlaxAlbertModel
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxAlbertModel
-    :members: __call__
-
-
-FlaxAlbertForPreTraining
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxAlbertForPreTraining
-    :members: __call__
-
-
-FlaxAlbertForMaskedLM
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxAlbertForMaskedLM
-    :members: __call__
-
-
-FlaxAlbertForSequenceClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxAlbertForSequenceClassification
-    :members: __call__
-
-
-FlaxAlbertForMultipleChoice
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxAlbertForMultipleChoice
-    :members: __call__
-
-
-FlaxAlbertForTokenClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxAlbertForTokenClassification
-    :members: __call__
-
-
-FlaxAlbertForQuestionAnswering
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxAlbertForQuestionAnswering
-    :members: __call__
--- a/docs/source/model_doc/beit.rst
+++ b/docs/source/model_doc/beit.rst
@@ -1,97 +0,0 @@
-.. 
-    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.
-
-BEiT
-----------------------------------------------------------------------------------------------------------------------
-
-Overview
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-The BEiT model was proposed in `BEiT: BERT Pre-Training of Image Transformers <https://arxiv.org/abs/2106.08254>`__ by
-Hangbo Bao, Li Dong and Furu Wei. Inspired by BERT, BEiT is the first paper that makes self-supervised pre-training of
-Vision Transformers (ViTs) outperform supervised pre-training. Rather than pre-training the model to predict the class
-of an image (as done in the `original ViT paper <https://arxiv.org/abs/2010.11929>`__), BEiT models are pre-trained to
-predict visual tokens from the codebook of OpenAI's `DALL-E model <https://arxiv.org/abs/2102.12092>`__ given masked
-patches.
-
-The abstract from the paper is the following:
-
-*We introduce a self-supervised vision representation model BEiT, which stands for Bidirectional Encoder representation
-from Image Transformers. Following BERT developed in the natural language processing area, we propose a masked image
-modeling task to pretrain vision Transformers. Specifically, each image has two views in our pre-training, i.e, image
-patches (such as 16x16 pixels), and visual tokens (i.e., discrete tokens). We first "tokenize" the original image into
-visual tokens. Then we randomly mask some image patches and fed them into the backbone Transformer. The pre-training
-objective is to recover the original visual tokens based on the corrupted image patches. After pre-training BEiT, we
-directly fine-tune the model parameters on downstream tasks by appending task layers upon the pretrained encoder.
-Experimental results on image classification and semantic segmentation show that our model achieves competitive results
-with previous pre-training methods. For example, base-size BEiT achieves 83.2% top-1 accuracy on ImageNet-1K,
-significantly outperforming from-scratch DeiT training (81.8%) with the same setup. Moreover, large-size BEiT obtains
-86.3% only using ImageNet-1K, even outperforming ViT-L with supervised pre-training on ImageNet-22K (85.2%).*
-
-Tips:
-
- BEiT models are regular Vision Transformers, but pre-trained in a self-supervised way rather than supervised. They
-  outperform both the original model (ViT) as well as Data-efficient Image Transformers (DeiT) when fine-tuned on
-  ImageNet-1K and CIFAR-100.
- As the BEiT models expect each image to be of the same size (resolution), one can use
-  :class:`~transformers.BeitFeatureExtractor` to resize (or rescale) and normalize images for the model.
- Both the patch resolution and image resolution used during pre-training or fine-tuning are reflected in the name of
-  each checkpoint. For example, :obj:`microsoft/beit-base-patch16-224` refers to a base-sized architecture with patch
-  resolution of 16x16 and fine-tuning resolution of 224x224. All checkpoints can be found on the `hub
-  <https://huggingface.co/models?search=microsoft/beit>`__.
- The available checkpoints are either (1) pre-trained on `ImageNet-22k <http://www.image-net.org/>`__ (a collection of
-  14 million images and 22k classes) only, (2) also fine-tuned on ImageNet-22k or (3) also fine-tuned on `ImageNet-1k
-  <http://www.image-net.org/challenges/LSVRC/2012/>`__ (also referred to as ILSVRC 2012, a collection of 1.3 million
-  images and 1,000 classes).
- BEiT uses relative position embeddings, inspired by the T5 model. During pre-training, the authors shared the
-  relative position bias among the several self-attention layers. During fine-tuning, each layer's relative position
-  bias is initialized with the shared relative position bias obtained after pre-training. Note that, if one wants to
-  pre-train a model from scratch, one needs to either set the :obj:`use_relative_position_bias` or the
-  :obj:`use_relative_position_bias` attribute of :class:`~transformers.BeitConfig` to :obj:`True` in order to add
-  position embeddings.
-
-This model was contributed by `nielsr <https://huggingface.co/nielsr>`__. The original code can be found `here
-<https://github.com/microsoft/unilm/tree/master/beit>`__.
-
-BeitConfig
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BeitConfig
-    :members:
-
-
-BeitFeatureExtractor
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BeitFeatureExtractor
-    :members: __call__
-
-
-BeitModel
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BeitModel
-    :members: forward
-
-
-BeitForMaskedImageModeling
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BeitForMaskedImageModeling
-    :members: forward
-
-
-BeitForImageClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.BeitForImageClassification
-    :members: forward
--- a/docs/source/model_doc/bert.rst
+++ b/docs/source/model_doc/bert.rst
@@ -76,9 +76,6 @@ Bert specific outputs
 .. autoclass:: transformers.models.bert.modeling_tf_bert.TFBertForPreTrainingOutput
    :members:

-.. autoclass:: transformers.models.bert.modeling_flax_bert.FlaxBertForPreTrainingOutput
-    :members:
-

 BertModel
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
--- a/docs/source/model_doc/deberta.rst
+++ b/docs/source/model_doc/deberta.rst
@@ -38,8 +38,7 @@ 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.*


-This model was contributed by `DeBERTa <https://huggingface.co/DeBERTa>`__. This model TF 2.0 implementation was
-contributed by `kamalkraj <https://huggingface.co/kamalkraj>`__ . The original code can be found `here
+This model was contributed by `DeBERTa <https://huggingface.co/DeBERTa>`__. The original code can be found `here
 <https://github.com/microsoft/DeBERTa>`__.


@@ -104,45 +103,3 @@ DebertaForQuestionAnswering

 .. autoclass:: transformers.DebertaForQuestionAnswering
    :members: forward
-
-
-TFDebertaModel
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFDebertaModel
-    :members: call
-
-
-TFDebertaPreTrainedModel
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFDebertaPreTrainedModel
-    :members: call
-
-
-TFDebertaForMaskedLM
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFDebertaForMaskedLM
-    :members: call
-
-
-TFDebertaForSequenceClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFDebertaForSequenceClassification
-    :members: call
-
-
-TFDebertaForTokenClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFDebertaForTokenClassification
-    :members: call
-
-
-TFDebertaForQuestionAnswering
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFDebertaForQuestionAnswering
-    :members: call
--- a/docs/source/model_doc/deberta_v2.rst
+++ b/docs/source/model_doc/deberta_v2.rst
@@ -53,13 +53,12 @@ New in v2:
  transformer layer to better learn the local dependency of input tokens.
 - **Sharing position projection matrix with content projection matrix in attention layer** Based on previous
  experiments, this can save parameters without affecting the performance.
- **Apply bucket to encode relative positions** The DeBERTa-v2 model uses log bucket to encode relative positions
+- **Apply bucket to encode relative postions** The DeBERTa-v2 model uses log bucket to encode relative positions
  similar to T5.
 - **900M model & 1.5B model** Two additional model sizes are available: 900M and 1.5B, which significantly improves the
  performance of downstream tasks.

-This model was contributed by `DeBERTa <https://huggingface.co/DeBERTa>`__. This model TF 2.0 implementation was
-contributed by `kamalkraj <https://huggingface.co/kamalkraj>`__. The original code can be found `here
+This model was contributed by `DeBERTa <https://huggingface.co/DeBERTa>`__. The original code can be found `here
 <https://github.com/microsoft/DeBERTa>`__.


@@ -118,45 +117,3 @@ DebertaV2ForQuestionAnswering

 .. autoclass:: transformers.DebertaV2ForQuestionAnswering
    :members: forward
-
-
-TFDebertaV2Model
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFDebertaV2Model
-    :members: call
-
-
-TFDebertaV2PreTrainedModel
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFDebertaV2PreTrainedModel
-    :members: call
-
-
-TFDebertaV2ForMaskedLM
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFDebertaV2ForMaskedLM
-    :members: call
-
-
-TFDebertaV2ForSequenceClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFDebertaV2ForSequenceClassification
-    :members: call
-
-
-TFDebertaV2ForTokenClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFDebertaV2ForTokenClassification
-    :members: call
-
-
-TFDebertaV2ForQuestionAnswering
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFDebertaV2ForQuestionAnswering
-    :members: call
--- a/docs/source/model_doc/distilbert.rst
+++ b/docs/source/model_doc/distilbert.rst
@@ -44,9 +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.

-This model was contributed by `victorsanh <https://huggingface.co/victorsanh>`__. This model jax version was
-contributed by `kamalkraj <https://huggingface.co/kamalkraj>`__. The original code can be found :prefix_link:`here
-<examples/research-projects/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
@@ -153,45 +152,3 @@ TFDistilBertForQuestionAnswering

 .. autoclass:: transformers.TFDistilBertForQuestionAnswering
    :members: call
-
-
-FlaxDistilBertModel
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxDistilBertModel
-    :members: __call__
-
-
-FlaxDistilBertForMaskedLM
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxDistilBertForMaskedLM
-    :members: __call__
-
-
-FlaxDistilBertForSequenceClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxDistilBertForSequenceClassification
-    :members: __call__
-
-
-FlaxDistilBertForMultipleChoice
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxDistilBertForMultipleChoice
-    :members: __call__
-
-
-FlaxDistilBertForTokenClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxDistilBertForTokenClassification
-    :members: __call__
-
-
-FlaxDistilBertForQuestionAnswering
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxDistilBertForQuestionAnswering
-    :members: __call__
--- a/docs/source/model_doc/encoderdecoder.rst
+++ b/docs/source/model_doc/encoderdecoder.rst
@@ -40,10 +40,3 @@ EncoderDecoderModel

 .. autoclass:: transformers.EncoderDecoderModel
    :members: forward, from_encoder_decoder_pretrained
-
-
-FlaxEncoderDecoderModel
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxEncoderDecoderModel
-    :members: __call__, from_encoder_decoder_pretrained
--- a/docs/source/model_doc/gpt2.rst
+++ b/docs/source/model_doc/gpt2.rst
@@ -108,13 +108,6 @@ GPT2ForSequenceClassification
    :members: forward


-GPT2ForTokenClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.GPT2ForTokenClassification
-    :members: forward
-
-
 TFGPT2Model
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

--- a/docs/source/model_doc/hubert.rst
+++ b/docs/source/model_doc/hubert.rst
@@ -64,14 +64,6 @@ HubertForCTC
 .. autoclass:: transformers.HubertForCTC
    :members: forward

-
-HubertForSequenceClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.HubertForSequenceClassification
-    :members: forward
-
-
 TFHubertModel
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

--- a/docs/source/model_doc/layoutlmv2.rst
+++ b/docs/source/model_doc/layoutlmv2.rst
@@ -1,314 +0,0 @@
-.. 
-    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.
-
-LayoutLMV2
-----------------------------------------------------------------------------------------------------------------------
-
-Overview
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-The LayoutLMV2 model was proposed in `LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding
-<https://arxiv.org/abs/2012.14740>`__ by Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu,
-Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou. LayoutLMV2 improves `LayoutLM
-<https://huggingface.co/transformers/model_doc/layoutlm.html>`__ to obtain state-of-the-art results across several
-document image understanding benchmarks:
-
- information extraction from scanned documents: the `FUNSD <https://guillaumejaume.github.io/FUNSD/>`__ dataset (a
-  collection of 199 annotated forms comprising more than 30,000 words), the `CORD <https://github.com/clovaai/cord>`__
-  dataset (a collection of 800 receipts for training, 100 for validation and 100 for testing), the `SROIE
-  <https://rrc.cvc.uab.es/?ch=13>`__ dataset (a collection of 626 receipts for training and 347 receipts for testing)
-  and the `Kleister-NDA <https://github.com/applicaai/kleister-nda>`__ dataset (a collection of non-disclosure
-  agreements from the EDGAR database, including 254 documents for training, 83 documents for validation, and 203
-  documents for testing).
- document image classification: the `RVL-CDIP <https://www.cs.cmu.edu/~aharley/rvl-cdip/>`__ dataset (a collection of
-  400,000 images belonging to one of 16 classes).
- document visual question answering: the `DocVQA <https://arxiv.org/abs/2007.00398>`__ dataset (a collection of 50,000
-  questions defined on 12,000+ document images).
-
-The abstract from the paper is the following:
-
-*Pre-training of text and layout has proved effective in a variety of visually-rich document understanding tasks due to
-its effective model architecture and the advantage of large-scale unlabeled scanned/digital-born documents. In this
-paper, we present LayoutLMv2 by pre-training text, layout and image in a multi-modal framework, where new model
-architectures and pre-training tasks are leveraged. Specifically, LayoutLMv2 not only uses the existing masked
-visual-language modeling task but also the new text-image alignment and text-image matching tasks in the pre-training
-stage, where cross-modality interaction is better learned. Meanwhile, it also integrates a spatial-aware self-attention
-mechanism into the Transformer architecture, so that the model can fully understand the relative positional
-relationship among different text blocks. Experiment results show that LayoutLMv2 outperforms strong baselines and
-achieves new state-of-the-art results on a wide variety of downstream visually-rich document understanding tasks,
-including FUNSD (0.7895 -> 0.8420), CORD (0.9493 -> 0.9601), SROIE (0.9524 -> 0.9781), Kleister-NDA (0.834 -> 0.852),
-RVL-CDIP (0.9443 -> 0.9564), and DocVQA (0.7295 -> 0.8672). The pre-trained LayoutLMv2 model is publicly available at
-this https URL.*
-
-Tips:
-
- The main difference between LayoutLMv1 and LayoutLMv2 is that the latter incorporates visual embeddings during
-  pre-training (while LayoutLMv1 only adds visual embeddings during fine-tuning).
- LayoutLMv2 adds both a relative 1D attention bias as well as a spatial 2D attention bias to the attention scores in
-  the self-attention layers. Details can be found on page 5 of the `paper <https://arxiv.org/abs/2012.14740>`__.
- Demo notebooks on how to use the LayoutLMv2 model on RVL-CDIP, FUNSD, DocVQA, CORD can be found `here
-  <https://github.com/NielsRogge/Transformers-Tutorials>`__.
- LayoutLMv2 uses Facebook AI's `Detectron2 <https://github.com/facebookresearch/detectron2/>`__ package for its visual
-  backbone. See `this link <https://detectron2.readthedocs.io/en/latest/tutorials/install.html>`__ for installation
-  instructions.
- In addition to :obj:`input_ids`, :meth:`~transformer.LayoutLMv2Model.forward` expects 2 additional inputs, namely
-  :obj:`image` and :obj:`bbox`. The :obj:`image` input corresponds to the original document image in which the text
-  tokens occur. The model expects each document image to be of size 224x224. This means that if you have a batch of
-  document images, :obj:`image` should be a tensor of shape (batch_size, 3, 224, 224). This can be either a
-  :obj:`torch.Tensor` or a :obj:`Detectron2.structures.ImageList`. You don't need to normalize the channels, as this is
-  done by the model. Important to note is that the visual backbone expects BGR channels instead of RGB, as all models
-  in Detectron2 are pre-trained using the BGR format. The :obj:`bbox` input are the bounding boxes (i.e. 2D-positions)
-  of the input text tokens. This is identical to :class:`~transformer.LayoutLMModel`. These can be obtained using an
-  external OCR engine such as Google's `Tesseract <https://github.com/tesseract-ocr/tesseract>`__ (there's a `Python
-  wrapper <https://pypi.org/project/pytesseract/>`__ available). Each bounding box should be in (x0, y0, x1, y1)
-  format, where (x0, y0) corresponds to the position of the upper left corner in the bounding box, and (x1, y1)
-  represents the position of the lower right corner. Note that one first needs to normalize the bounding boxes to be on
-  a 0-1000 scale. To normalize, you can use the following function:
-
-.. 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)),
-         ]
-
-Here, :obj:`width` and :obj:`height` correspond to the width and height of the original document in which the token
-occurs (before resizing the image). Those can be obtained using the Python Image Library (PIL) library for example, as
-follows:
-
-.. code-block::
-
-    from PIL import Image
-
-    image = Image.open("name_of_your_document - can be a png file, pdf, etc.")
-
-    width, height = image.size
-
-However, this model includes a brand new :class:`~transformer.LayoutLMv2Processor` which can be used to directly
-prepare data for the model (including applying OCR under the hood). More information can be found in the "Usage"
-section below.
-
- Internally, :class:`~transformer.LayoutLMv2Model` will send the :obj:`image` input through its visual backbone to
-  obtain a lower-resolution feature map, whose shape is equal to the :obj:`image_feature_pool_shape` attribute of
-  :class:`~transformer.LayoutLMv2Config`. This feature map is then flattened to obtain a sequence of image tokens. As
-  the size of the feature map is 7x7 by default, one obtains 49 image tokens. These are then concatenated with the text
-  tokens, and send through the Transformer encoder. This means that the last hidden states of the model will have a
-  length of 512 + 49 = 561, if you pad the text tokens up to the max length. More generally, the last hidden states
-  will have a shape of :obj:`seq_length` + :obj:`image_feature_pool_shape[0]` *
-  :obj:`config.image_feature_pool_shape[1]`.
- When calling :meth:`~transformer.LayoutLMv2Model.from_pretrained`, a warning will be printed with a long list of
-  parameter names that are not initialized. This is not a problem, as these parameters are batch normalization
-  statistics, which are going to have values when fine-tuning on a custom dataset.
- If you want to train the model in a distributed environment, make sure to call :meth:`synchronize_batch_norm` on the
-  model in order to properly synchronize the batch normalization layers of the visual backbone.
-
-In addition, there's LayoutXLM, which is a multilingual version of LayoutLMv2. More information can be found on
-:doc:`LayoutXLM's documentation page <layoutxlm>`.
-
-Usage: LayoutLMv2Processor
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-The easiest way to prepare data for the model is to use :class:`~transformer.LayoutLMv2Processor`, which internally
-combines a feature extractor (:class:`~transformer.LayoutLMv2FeatureExtractor`) and a tokenizer
-(:class:`~transformer.LayoutLMv2Tokenizer` or :class:`~transformer.LayoutLMv2TokenizerFast`). The feature extractor
-handles the image modality, while the tokenizer handles the text modality. A processor combines both, which is ideal
-for a multi-modal model like LayoutLMv2. Note that you can still use both separately, if you only want to handle one
-modality.
-
-.. code-block::
-
-    from transformers import LayoutLMv2FeatureExtractor, LayoutLMv2TokenizerFast, LayoutLMv2Processor
-
-    feature_extractor = LayoutLMv2FeatureExtractor() # apply_ocr is set to True by default
-    tokenizer = LayoutLMv2TokenizerFast.from_pretrained("microsoft/layoutlmv2-base-uncased")
-    processor = LayoutLMv2Processor(feature_extractor, tokenizer)
-
-In short, one can provide a document image (and possibly additional data) to :class:`~transformer.LayoutLMv2Processor`,
-and it will create the inputs expected by the model. Internally, the processor first uses
-:class:`~transformer.LayoutLMv2FeatureExtractor` to apply OCR on the image to get a list of words and normalized
-bounding boxes, as well to resize the image to a given size in order to get the :obj:`image` input. The words and
-normalized bounding boxes are then provided to :class:`~transformer.LayoutLMv2Tokenizer` or
-:class:`~transformer.LayoutLMv2TokenizerFast`, which converts them to token-level :obj:`input_ids`,
-:obj:`attention_mask`, :obj:`token_type_ids`, :obj:`bbox`. Optionally, one can provide word labels to the processor,
-which are turned into token-level :obj:`labels`.
-
-:class:`~transformer.LayoutLMv2Processor` uses `PyTesseract <https://pypi.org/project/pytesseract/>`__, a Python
-wrapper around Google's Tesseract OCR engine, under the hood. Note that you can still use your own OCR engine of
-choice, and provide the words and normalized boxes yourself. This requires initializing
-:class:`~transformer.LayoutLMv2FeatureExtractor` with :obj:`apply_ocr` set to :obj:`False`.
-
-In total, there are 5 use cases that are supported by the processor. Below, we list them all. Note that each of these
-use cases work for both batched and non-batched inputs (we illustrate them for non-batched inputs).
-
-**Use case 1: document image classification (training, inference) + token classification (inference), apply_ocr =
-True**
-
-This is the simplest case, in which the processor (actually the feature extractor) will perform OCR on the image to get
-the words and normalized bounding boxes.
-
-.. code-block::
-
-    from transformers import LayoutLMv2Processor
-    from PIL import Image
-
-    processor = LayoutLMv2Processor.from_pretrained("microsoft/layoutlmv2-base-uncased")
-
-    image = Image.open("name_of_your_document - can be a png file, pdf, etc.").convert("RGB")
-    encoding = processor(image, return_tensors="pt") # you can also add all tokenizer parameters here such as padding, truncation
-    print(encoding.keys())
-    # dict_keys(['input_ids', 'token_type_ids', 'attention_mask', 'bbox', 'image'])
-
-**Use case 2: document image classification (training, inference) + token classification (inference), apply_ocr=False**
-
-In case one wants to do OCR themselves, one can initialize the feature extractor with :obj:`apply_ocr` set to
-:obj:`False`. In that case, one should provide the words and corresponding (normalized) bounding boxes themselves to
-the processor.
-
-.. code-block::
-
-    from transformers import LayoutLMv2Processor
-    from PIL import Image
-
-    processor = LayoutLMv2Processor.from_pretrained("microsoft/layoutlmv2-base-uncased", revision="no_ocr")
-
-    image = Image.open("name_of_your_document - can be a png file, pdf, etc.").convert("RGB")
-    words = ["hello", "world"]
-    boxes = [[1, 2, 3, 4], [5, 6, 7, 8]] # make sure to normalize your bounding boxes
-    encoding = processor(image, words, boxes=boxes, return_tensors="pt")
-    print(encoding.keys())
-    # dict_keys(['input_ids', 'token_type_ids', 'attention_mask', 'bbox', 'image'])
-
-**Use case 3: token classification (training), apply_ocr=False**
-
-For token classification tasks (such as FUNSD, CORD, SROIE, Kleister-NDA), one can also provide the corresponding word
-labels in order to train a model. The processor will then convert these into token-level :obj:`labels`. By default, it
-will only label the first wordpiece of a word, and label the remaining wordpieces with -100, which is the
-:obj:`ignore_index` of PyTorch's CrossEntropyLoss. In case you want all wordpieces of a word to be labeled, you can
-initialize the tokenizer with :obj:`only_label_first_subword` set to :obj:`False`.
-
-.. code-block::
-
-    from transformers import LayoutLMv2Processor
-    from PIL import Image
-
-    processor = LayoutLMv2Processor.from_pretrained("microsoft/layoutlmv2-base-uncased", revision="no_ocr")
-
-    image = Image.open("name_of_your_document - can be a png file, pdf, etc.").convert("RGB")
-    words = ["hello", "world"]
-    boxes = [[1, 2, 3, 4], [5, 6, 7, 8]] # make sure to normalize your bounding boxes
-    word_labels = [1, 2]
-    encoding = processor(image, words, boxes=boxes, word_labels=word_labels, return_tensors="pt")
-    print(encoding.keys())
-    # dict_keys(['input_ids', 'token_type_ids', 'attention_mask', 'bbox', 'labels', 'image'])
-
-**Use case 4: visual question answering (inference), apply_ocr=True**
-
-For visual question answering tasks (such as DocVQA), you can provide a question to the processor. By default, the
-processor will apply OCR on the image, and create [CLS] question tokens [SEP] word tokens [SEP].
-
-.. code-block::
-
-    from transformers import LayoutLMv2Processor
-    from PIL import Image
-
-    processor = LayoutLMv2Processor.from_pretrained("microsoft/layoutlmv2-base-uncased")
-
-    image = Image.open("name_of_your_document - can be a png file, pdf, etc.").convert("RGB")
-    question = "What's his name?"
-    encoding = processor(image, question, return_tensors="pt") 
-    print(encoding.keys())
-    # dict_keys(['input_ids', 'token_type_ids', 'attention_mask', 'bbox', 'image'])
-
-**Use case 5: visual question answering (inference), apply_ocr=False**
-
-For visual question answering tasks (such as DocVQA), you can provide a question to the processor. If you want to
-perform OCR yourself, you can provide your own words and (normalized) bounding boxes to the processor.
-
-.. code-block::
-
-    from transformers import LayoutLMv2Processor
-    from PIL import Image
-
-    processor = LayoutLMv2Processor.from_pretrained("microsoft/layoutlmv2-base-uncased", revision="no_ocr")
-
-    image = Image.open("name_of_your_document - can be a png file, pdf, etc.").convert("RGB")
-    question = "What's his name?"
-    words = ["hello", "world"]
-    boxes = [[1, 2, 3, 4], [5, 6, 7, 8]] # make sure to normalize your bounding boxes
-    encoding = processor(image, question, words, boxes=boxes, return_tensors="pt")  
-    print(encoding.keys())
-    # dict_keys(['input_ids', 'token_type_ids', 'attention_mask', 'bbox', 'image'])
-
-LayoutLMv2Config
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.LayoutLMv2Config
-    :members:
-
-
-LayoutLMv2FeatureExtractor
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.LayoutLMv2FeatureExtractor
-    :members: __call__
-
-
-LayoutLMv2Tokenizer
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.LayoutLMv2Tokenizer
-    :members: __call__, save_vocabulary
-
-
-LayoutLMv2TokenizerFast
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.LayoutLMv2TokenizerFast
-    :members: __call__
-
-
-LayoutLMv2Processor
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.LayoutLMv2Processor
-    :members: __call__
-
-
-LayoutLMv2Model
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.LayoutLMv2Model
-    :members: forward
-
-
-LayoutLMv2ForSequenceClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.LayoutLMv2ForSequenceClassification
-    :members:
-
-
-LayoutLMv2ForTokenClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.LayoutLMv2ForTokenClassification
-    :members:
-
-
-LayoutLMv2ForQuestionAnswering
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.LayoutLMv2ForQuestionAnswering
-    :members:
--- a/docs/source/model_doc/layoutxlm.rst
+++ b/docs/source/model_doc/layoutxlm.rst
@@ -1,47 +0,0 @@
-.. 
-    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.
-
-LayoutXLM
-----------------------------------------------------------------------------------------------------------------------
-
-Overview
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-LayoutXLM was proposed in `LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding
-<https://arxiv.org/abs/2104.08836>`__ by Yiheng Xu, Tengchao Lv, Lei Cui, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha
-Zhang, Furu Wei. It's a multilingual extension of the `LayoutLMv2 model <https://arxiv.org/abs/2012.14740>`__ trained
-on 53 languages.
-
-The abstract from the paper is the following:
-
-*Multimodal pre-training with text, layout, and image has achieved SOTA performance for visually-rich document
-understanding tasks recently, which demonstrates the great potential for joint learning across different modalities. In
-this paper, we present LayoutXLM, a multimodal pre-trained model for multilingual document understanding, which aims to
-bridge the language barriers for visually-rich document understanding. To accurately evaluate LayoutXLM, we also
-introduce a multilingual form understanding benchmark dataset named XFUN, which includes form understanding samples in
-7 languages (Chinese, Japanese, Spanish, French, Italian, German, Portuguese), and key-value pairs are manually labeled
-for each language. Experiment results show that the LayoutXLM model has significantly outperformed the existing SOTA
-cross-lingual pre-trained models on the XFUN dataset.*
-
-One can directly plug in the weights of LayoutXLM into a LayoutLMv2 model, like so:
-
-.. code-block::
-
-    from transformers import LayoutLMv2Model
-
-    model = LayoutLMv2Model.from_pretrained('microsoft/layoutxlm-base') 
-
-As LayoutXLM's architecture is equivalent to that of LayoutLMv2, one can refer to :doc:`LayoutLMv2's documentation page
-<layoutlmv2>` for all tips, code examples and notebooks.
-
-This model was contributed by `nielsr <https://huggingface.co/nielsr>`__. The original code can be found `here
-<https://github.com/microsoft/unilm>`__.
--- a/docs/source/model_doc/m2m_100.rst
+++ b/docs/source/model_doc/m2m_100.rst
@@ -58,7 +58,7 @@ examples. To install :obj:`sentencepiece` run ``pip install sentencepiece``.
    tokenizer = M2M100Tokenizer.from_pretrained('facebook/m2m100_418M', src_lang="en", tgt_lang="fr")

    src_text = "Life is like a box of chocolates."
-    tgt_text = "La vie est comme une boîte de chocolat."
+    tgt_lang = "La vie est comme une boîte de chocolat."

    model_inputs = tokenizer(src_text, return_tensors="pt")
    with tokenizer.as_target_tokenizer():
--- a/docs/source/model_doc/marian.rst
+++ b/docs/source/model_doc/marian.rst
@@ -103,8 +103,8 @@ Here is the code to see all available pretrained models on the hub:

 .. code-block:: python

-    from huggingface_hub.hf_api import HfApi
-    model_list = HfApi().list_models()
+    from transformers.hf_api import HfApi
+    model_list = HfApi().model_list()
    org = "Helsinki-NLP"
    model_ids = [x.modelId for x in model_list if x.modelId.startswith(org)]
    suffix = [x.split('/')[1] for x in model_ids]
--- a/docs/source/model_doc/mt5.rst
+++ b/docs/source/model_doc/mt5.rst
@@ -94,17 +94,3 @@ TFMT5EncoderModel

 .. autoclass:: transformers.TFMT5EncoderModel
    :members:
-
-
-FlaxMT5Model
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxMT5Model
-    :members:
-
-
-FlaxMT5ForConditionalGeneration
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.FlaxMT5ForConditionalGeneration
-    :members:
--- a/docs/source/model_doc/rembert.rst
+++ b/docs/source/model_doc/rembert.rst
@@ -1,161 +0,0 @@
-.. 
-    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.
-
-RemBERT
-----------------------------------------------------------------------------------------------------------------------
-
-Overview
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-The RemBERT model was proposed in `Rethinking Embedding Coupling in Pre-trained Language Models
-<https://arxiv.org/abs/2010.12821>`__ by Hyung Won Chung, Thibault Févry, Henry Tsai, Melvin Johnson, Sebastian Ruder.
-
-The abstract from the paper is the following:
-
-*We re-evaluate the standard practice of sharing weights between input and output embeddings in state-of-the-art
-pre-trained language models. We show that decoupled embeddings provide increased modeling flexibility, allowing us to
-significantly improve the efficiency of parameter allocation in the input embedding of multilingual models. By
-reallocating the input embedding parameters in the Transformer layers, we achieve dramatically better performance on
-standard natural language understanding tasks with the same number of parameters during fine-tuning. We also show that
-allocating additional capacity to the output embedding provides benefits to the model that persist through the
-fine-tuning stage even though the output embedding is discarded after pre-training. Our analysis shows that larger
-output embeddings prevent the model's last layers from overspecializing to the pre-training task and encourage
-Transformer representations to be more general and more transferable to other tasks and languages. Harnessing these
-findings, we are able to train models that achieve strong performance on the XTREME benchmark without increasing the
-number of parameters at the fine-tuning stage.*
-
-Tips:
-
-For fine-tuning, RemBERT can be thought of as a bigger version of mBERT with an ALBERT-like factorization of the
-embedding layer. The embeddings are not tied in pre-training, in contrast with BERT, which enables smaller input
-embeddings (preserved during fine-tuning) and bigger output embeddings (discarded at fine-tuning). The tokenizer is
-also similar to the Albert one rather than the BERT one.
-
-RemBertConfig
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.RemBertConfig
-    :members:
-
-
-RemBertTokenizer
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.RemBertTokenizer
-    :members: build_inputs_with_special_tokens, get_special_tokens_mask,
-        create_token_type_ids_from_sequences, save_vocabulary
-
-
-RemBertTokenizerFast
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.RemBertTokenizerFast
-    :members: build_inputs_with_special_tokens, get_special_tokens_mask,
-        create_token_type_ids_from_sequences, save_vocabulary
-
-
-RemBertModel
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.RemBertModel
-    :members: forward
-
-
-RemBertForCausalLM
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.RemBertForCausalLM
-    :members: forward
-
-
-RemBertForMaskedLM
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.RemBertForMaskedLM
-    :members: forward
-
-
-RemBertForSequenceClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.RemBertForSequenceClassification
-    :members: forward
-
-
-RemBertForMultipleChoice
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.RemBertForMultipleChoice
-    :members: forward
-
-
-RemBertForTokenClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.RemBertForTokenClassification
-    :members: forward
-
-
-RemBertForQuestionAnswering
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.RemBertForQuestionAnswering
-    :members: forward
-
-
-TFRemBertModel
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFRemBertModel
-    :members: call
-
-
-TFRemBertForMaskedLM
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFRemBertForMaskedLM
-    :members: call
-
-
-TFRemBertForCausalLM
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFRemBertForCausalLM
-    :members: call
-
-
-TFRemBertForSequenceClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFRemBertForSequenceClassification
-    :members: call
-
-
-TFRemBertForMultipleChoice
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFRemBertForMultipleChoice
-    :members: call
-
-
-TFRemBertForTokenClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFRemBertForTokenClassification
-    :members: call
-
-
-TFRemBertForQuestionAnswering
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.TFRemBertForQuestionAnswering
-    :members: call
--- a/docs/source/model_doc/speech_to_text.rst
+++ b/docs/source/model_doc/speech_to_text.rst
@@ -42,8 +42,8 @@ features. The :class:`~transformers.Speech2TextProcessor` wraps :class:`~transfo
 predicted token ids.

 The feature extractor depends on :obj:`torchaudio` and the tokenizer depends on :obj:`sentencepiece` so be sure to
-install those packages before running the examples. You could either install those as extra speech dependencies with
-``pip install transformers"[speech, sentencepiece]"`` or install the packages seperately with ``pip install torchaudio
+install those packages before running the examples. You could either install those as extra speech dependancies with
+``pip install transformers"[speech, sentencepiece]"`` or install the packages seperatly with ``pip install torchaudio
 sentencepiece``. Also ``torchaudio`` requires the development version of the `libsndfile
 <http://www.mega-nerd.com/libsndfile/>`__ package which can be installed via a system package manager. On Ubuntu it can
 be installed as follows: ``apt install libsndfile1-dev``
--- a/docs/source/model_doc/splinter.rst
+++ b/docs/source/model_doc/splinter.rst
@@ -1,87 +0,0 @@
-.. 
-    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.
-
-Splinter
-----------------------------------------------------------------------------------------------------------------------
-
-Overview
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-The Splinter model was proposed in `Few-Shot Question Answering by Pretraining Span Selection
-<https://arxiv.org/abs/2101.00438>`__ by Ori Ram, Yuval Kirstain, Jonathan Berant, Amir Globerson, Omer Levy. Splinter
-is an encoder-only transformer (similar to BERT) pretrained using the recurring span selection task on a large corpus
-comprising Wikipedia and the Toronto Book Corpus.
-
-The abstract from the paper is the following:
-
-In several question answering benchmarks, pretrained models have reached human parity through fine-tuning on an order
-of 100,000 annotated questions and answers. We explore the more realistic few-shot setting, where only a few hundred
-training examples are available, and observe that standard models perform poorly, highlighting the discrepancy between
-current pretraining objectives and question answering. We propose a new pretraining scheme tailored for question
-answering: recurring span selection. Given a passage with multiple sets of recurring spans, we mask in each set all
-recurring spans but one, and ask the model to select the correct span in the passage for each masked span. Masked spans
-are replaced with a special token, viewed as a question representation, that is later used during fine-tuning to select
-the answer span. The resulting model obtains surprisingly good results on multiple benchmarks (e.g., 72.7 F1 on SQuAD
-with only 128 training examples), while maintaining competitive performance in the high-resource setting.
-
-Tips:
-
- Splinter was trained to predict answers spans conditioned on a special [QUESTION] token. These tokens contextualize
-  to question representations which are used to predict the answers. This layer is called QASS, and is the default
-  behaviour in the :class:`~transformers.SplinterForQuestionAnswering` class. Therefore:
- Use :class:`~transformers.SplinterTokenizer` (rather than :class:`~transformers.BertTokenizer`), as it already
-  contains this special token. Also, its default behavior is to use this token when two sequences are given (for
-  example, in the `run_qa.py` script).
- If you plan on using Splinter outside `run_qa.py`, please keep in mind the question token - it might be important for
-  the success of your model, especially in a few-shot setting.
- Please note there are two different checkpoints for each size of Splinter. Both are basically the same, except that
-  one also has the pretrained wights of the QASS layer (`tau/splinter-base-qass` and `tau/splinter-large-qass`) and one
-  doesn't (`tau/splinter-base` and `tau/splinter-large`). This is done to support randomly initializing this layer at
-  fine-tuning, as it is shown to yield better results for some cases in the paper.
-
-This model was contributed by `yuvalkirstain <https://huggingface.co/yuvalkirstain>`__ and `oriram
-<https://huggingface.co/oriram>`__. The original code can be found `here <https://github.com/oriram/splinter>`__.
-
-SplinterConfig
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.SplinterConfig
-    :members:
-
-
-SplinterTokenizer
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.SplinterTokenizer
-    :members: build_inputs_with_special_tokens, get_special_tokens_mask,
-        create_token_type_ids_from_sequences, save_vocabulary
-
-
-SplinterTokenizerFast
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.SplinterTokenizerFast
-    :members:
-
-
-SplinterModel
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.SplinterModel
-    :members: forward
-
-
-SplinterForQuestionAnswering
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.SplinterForQuestionAnswering
-    :members: forward
--- a/docs/source/model_doc/visual_bert.rst
+++ b/docs/source/model_doc/visual_bert.rst
@@ -58,17 +58,9 @@ layer, and is expected to be bound by [CLS] and a [SEP] tokens, as in BERT. The
 appropriately for the textual and visual parts.

 The :class:`~transformers.BertTokenizer` is used to encode the text. A custom detector/feature extractor must be used
-to get the visual embeddings. The following example notebooks show how to use VisualBERT with Detectron-like models:
-
-* `VisualBERT VQA demo notebook
-  <https://github.com/huggingface/transformers/tree/master/examples/research_projects/visual_bert>`__ : This notebook
-  contains an example on VisualBERT VQA.
-
-* `Generate Embeddings for VisualBERT (Colab Notebook)
-  <https://colab.research.google.com/drive/1bLGxKdldwqnMVA5x4neY7-l_8fKGWQYI?usp=sharing>`__ : This notebook contains
-  an example on how to generate visual embeddings.
-
-The following example shows how to get the last hidden state using :class:`~transformers.VisualBertModel`:
+to get the visual embeddings. For an example on how to generate visual embeddings, see the `colab notebook
+<https://colab.research.google.com/drive/1bLGxKdldwqnMVA5x4neY7-l_8fKGWQYI?usp=sharing>`__. The following example shows
+how to get the last hidden state using :class:`~transformers.VisualBertModel`:

 .. code-block::

@@ -82,13 +74,6 @@ The following example shows how to get the last hidden state using :class:`~tran
        >>> # this is a custom function that returns the visual embeddings given the image path
        >>> visual_embeds = get_visual_embeddings(image_path)

-        >>> visual_token_type_ids = torch.ones(visual_embeds.shape[:-1], dtype=torch.long)
-        >>> visual_attention_mask = torch.ones(visual_embeds.shape[:-1], dtype=torch.float)
-        >>> inputs.update({
-        ...     "visual_embeds": visual_embeds,
-        ...     "visual_token_type_ids": visual_token_type_ids,
-        ...     "visual_attention_mask": visual_attention_mask
-        ... })
        >>> outputs = model(**inputs)
        >>> last_hidden_state = outputs.last_hidden_state

--- a/docs/source/model_doc/vit.rst
+++ b/docs/source/model_doc/vit.rst
@@ -66,23 +66,6 @@ Tips:
  language modeling). With this approach, the smaller ViT-B/16 model achieves 79.9% accuracy on ImageNet, a significant
  improvement of 2% to training from scratch, but still 4% behind supervised pre-training.

-Following the original Vision Transformer, some follow-up works have been made:
-
- DeiT (Data-efficient Image Transformers) by Facebook AI. DeiT models are distilled vision transformers. Refer to
-  :doc:`DeiT's documentation page <deit>`. The authors of DeiT also released more efficiently trained ViT models, which
-  you can directly plug into :class:`~transformers.ViTModel` or :class:`~transformers.ViTForImageClassification`. 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.
-
- BEiT (BERT pre-training of Image Transformers) by Microsoft Research. BEiT models outperform supervised pre-trained
-  vision transformers using a self-supervised method inspired by BERT (masked image modeling) and based on a VQ-VAE.
-  Refer to :doc:`BEiT's documentation page <beit>`.
-
- DINO (a method for self-supervised training of Vision Transformers) by Facebook AI. Vision Transformers trained using
-  the DINO method show very interesting properties not seen with convolutional models. They are capable of segmenting
-  objects, without having ever been trained to do so. DINO checkpoints can be found on the `hub
-  <https://huggingface.co/models?other=dino>`__.

 This model was contributed by `nielsr <https://huggingface.co/nielsr>`__. The original code (written in JAX) can be
 found `here <https://github.com/google-research/vision_transformer>`__.
--- a/docs/source/model_doc/wav2vec2.rst
+++ b/docs/source/model_doc/wav2vec2.rst
@@ -67,22 +67,6 @@ Wav2Vec2Processor
    :members: __call__, pad, from_pretrained, save_pretrained, batch_decode, decode, as_target_processor


-Wav2Vec2 specific outputs
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.models.wav2vec2.modeling_wav2vec2.Wav2Vec2BaseModelOutput
-    :members: 
-
-.. autoclass:: transformers.models.wav2vec2.modeling_wav2vec2.Wav2Vec2ForPreTrainingOutput
-    :members: 
-
-.. autoclass:: transformers.models.wav2vec2.modeling_flax_wav2vec2.FlaxWav2Vec2BaseModelOutput
-    :members: 
-
-.. autoclass:: transformers.models.wav2vec2.modeling_flax_wav2vec2.FlaxWav2Vec2ForPreTrainingOutput
-    :members: 
-
-
 Wav2Vec2Model
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

@@ -96,14 +80,6 @@ Wav2Vec2ForCTC
 .. autoclass:: transformers.Wav2Vec2ForCTC
    :members: forward

-
-Wav2Vec2ForSequenceClassification
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. autoclass:: transformers.Wav2Vec2ForSequenceClassification
-    :members: forward
-
-
 Wav2Vec2ForPreTraining
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

--- a/docs/source/parallelism.md
+++ b/docs/source/parallelism.md
@@ -58,7 +58,7 @@ a0 | b0 | c0
 a1 | b1 | c1
 a2 | b2 | c2
 ```
-Layer La has weights a0, a1 and a2.
+Layer La has weights a0, at and a2.

 If we have 3 GPUs, the Sharded DDP (= Zero-DP) splits the model onto 3 GPUs like so:

@@ -220,12 +220,9 @@ Special considerations: TP requires very fast network, and therefore it's not ad
 This section is based on the original much more [detailed TP overview](https://github.com/huggingface/transformers/issues/10321#issuecomment-783543530).
 by [@anton-l](https://github.com/anton-l).

-Alternative names:
- DeepSpeed calls it [tensor slicing](https://www.deepspeed.ai/features/#model-parallelism)
-
 Implementations:
- [Megatron-LM](https://github.com/NVIDIA/Megatron-LM) has an internal implementation, as it's very model-specific
- [parallelformers](https://github.com/tunib-ai/parallelformers) (only inference at the moment)
+- DeepSpeed calls it [tensor slicing](https://www.deepspeed.ai/features/#model-parallelism)
+- [Megatron-LM](https://github.com/NVIDIA/Megatron-LM) has an internal implementation.

 🤗 Transformers status:
 - core: not yet implemented in the core
--- a/docs/source/perplexity.rst
+++ b/docs/source/perplexity.rst
@@ -96,7 +96,7 @@ dataset in memory.

 .. code-block:: python

-    from datasets import load_dataset
+    from nlp import load_dataset
    test = load_dataset('wikitext', 'wikitext-2-raw-v1', split='test')
    encodings = tokenizer('\n\n'.join(test['text']), return_tensors='pt')

--- a/docs/source/preprocessing.rst
+++ b/docs/source/preprocessing.rst
@@ -243,16 +243,15 @@ three arguments you need to know for this are :obj:`padding`, :obj:`truncation`

 - :obj:`truncation` controls the truncation. It can be a boolean or a string which should be:

-    - :obj:`True` or :obj:`'longest_first'` truncate to a maximum length specified by the :obj:`max_length` argument or
+    - :obj:`True` or :obj:`'only_first'` truncate to a maximum length specified by the :obj:`max_length` argument or
      the maximum length accepted by the model if no :obj:`max_length` is provided (``max_length=None``). This will
-      truncate token by token, removing a token from the longest sequence in the pair until the proper length is
-      reached.
+      only truncate the first sentence of a pair if a pair of sequence (or a batch of pairs of sequences) is provided.
    - :obj:`'only_second'` truncate to a maximum length specified by the :obj:`max_length` argument or the maximum
      length accepted by the model if no :obj:`max_length` is provided (``max_length=None``). This will only truncate
      the second sentence of a pair if a pair of sequence (or a batch of pairs of sequences) is provided.
-    - :obj:`'only_first'` truncate to a maximum length specified by the :obj:`max_length` argument or the maximum
-      length accepted by the model if no :obj:`max_length` is provided (``max_length=None``). This will only truncate
-      the first sentence of a pair if a pair of sequence (or a batch of pairs of sequences) is provided.
+    - :obj:`'longest_first'` truncate to a maximum length specified by the :obj:`max_length` argument or the maximum
+      length accepted by the model if no :obj:`max_length` is provided (``max_length=None``). This will truncate token
+      by token, removing a token from the longest sequence in the pair until the proper length is reached.
    - :obj:`False` or :obj:`'do_not_truncate'` to not truncate the sequences. As we have seen before, this is the
      default behavior.

--- a/docs/source/quicktour.rst
+++ b/docs/source/quicktour.rst
@@ -65,7 +65,7 @@ make them readable. For instance:
 .. code-block::

    >>> classifier('We are very happy to show you the 🤗 Transformers library.')
-    [{'label': 'POSITIVE', 'score': 0.9998}]
+    [{'label': 'POSITIVE', 'score': 0.9997795224189758}]

 That's encouraging! You can use it on a list of sentences, which will be preprocessed then fed to the model as a
 `batch`, returning a list of dictionaries like this one:
@@ -195,8 +195,7 @@ sequence:
 .. code-block::

    >>> print(inputs)
-    {'input_ids': [101, 2057, 2024, 2200, 3407, 2000, 2265, 2017, 1996, 100, 19081, 3075, 1012, 102],
-     'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]}
+    {'input_ids': [101, 2057, 2024, 2200, 3407, 2000, 2265, 2017, 1996, 100, 19081, 3075, 1012, 102], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]}

 You can pass a list of sentences directly to your tokenizer. If your goal is to send them through your model as a
 batch, you probably want to pad them all to the same length, truncate them to the maximum length the model can accept
@@ -261,12 +260,12 @@ objects are described in greater detail :doc:`here <main_classes/output>`. For n
    >>> ## PYTORCH CODE
    >>> print(pt_outputs)
    SequenceClassifierOutput(loss=None, logits=tensor([[-4.0833,  4.3364],
-            [ 0.0818, -0.0418]], grad_fn=<AddmmBackward>), hidden_states=None, attentions=None)
+        [ 0.0818, -0.0418]], grad_fn=<AddmmBackward>), hidden_states=None, attentions=None)
    >>> ## TENSORFLOW CODE
    >>> print(tf_outputs)
    TFSequenceClassifierOutput(loss=None, logits=<tf.Tensor: shape=(2, 2), dtype=float32, numpy=
-    array([[-4.0833 ,  4.3364  ],
-           [ 0.0818, -0.0418]], dtype=float32)>, hidden_states=None, attentions=None)
+    array([[-4.0832963 ,  4.3364143 ],
+           [ 0.081807  , -0.04178282]], dtype=float32)>, hidden_states=None, attentions=None)

 Notice how the output object has a ``logits`` attribute. You can use this to access the model's final activations.

@@ -284,7 +283,7 @@ Let's apply the SoftMax activation to get predictions.
    >>> pt_predictions = nn.functional.softmax(pt_outputs.logits, dim=-1)
    >>> ## TENSORFLOW CODE
    >>> import tensorflow as tf
-    >>> tf_predictions = tf.nn.softmax(tf_outputs.logits, axis=-1)
+    >>> tf.nn.softmax(tf_outputs.logits, axis=-1)

 We can see we get the numbers from before:

@@ -293,8 +292,8 @@ We can see we get the numbers from before:
    >>> ## TENSORFLOW CODE
    >>> print(tf_predictions)
    tf.Tensor(
-    [[2.2043e-04 9.9978e-01]
-     [5.3086e-01 4.6914e-01]], shape=(2, 2), dtype=float32)
+    [[2.2042994e-04 9.9977952e-01]
+     [5.3086340e-01 4.6913657e-01]], shape=(2, 2), dtype=float32)
    >>> ## PYTORCH CODE
    >>> print(pt_predictions)
    tensor([[2.2043e-04, 9.9978e-01],
@@ -310,14 +309,14 @@ attribute:
    >>> pt_outputs = pt_model(**pt_batch, labels = torch.tensor([1, 0]))
    >>> print(pt_outputs)
    SequenceClassifierOutput(loss=tensor(0.3167, grad_fn=<NllLossBackward>), logits=tensor([[-4.0833,  4.3364],
-            [ 0.0818, -0.0418]], grad_fn=<AddmmBackward>), hidden_states=None, attentions=None)
+    [ 0.0818, -0.0418]], grad_fn=<AddmmBackward>), hidden_states=None, attentions=None)
    >>> ## TENSORFLOW CODE
    >>> import tensorflow as tf
    >>> tf_outputs = tf_model(tf_batch, labels = tf.constant([1, 0]))
    >>> print(tf_outputs)
-    TFSequenceClassifierOutput(loss=<tf.Tensor: shape=(2,), dtype=float32, numpy=array([2.2051e-04, 6.3326e-01], dtype=float32)>, logits=<tf.Tensor: shape=(2, 2), dtype=float32, numpy=
-    array([[-4.0833 ,  4.3364  ],
-           [ 0.0818, -0.0418]], dtype=float32)>, hidden_states=None, attentions=None)
+    TFSequenceClassifierOutput(loss=<tf.Tensor: shape=(2,), dtype=float32, numpy=array([2.2051287e-04, 6.3326043e-01], dtype=float32)>, logits=<tf.Tensor: shape=(2, 2), dtype=float32, numpy=
+    array([[-4.0832963 ,  4.3364143 ],
+           [ 0.081807  , -0.04178282]], dtype=float32)>, hidden_states=None, attentions=None)

 Models are standard `torch.nn.Module <https://pytorch.org/docs/stable/nn.html#torch.nn.Module>`__ or `tf.keras.Model
 <https://www.tensorflow.org/api_docs/python/tf/keras/Model>`__ so you can use them in your usual training loop. 🤗
--- a/docs/source/task_summary.rst
+++ b/docs/source/task_summary.rst
@@ -107,8 +107,7 @@ each other. The process is the following:
    >>> sequence_1 = "Apples are especially bad for your health"
    >>> sequence_2 = "HuggingFace's headquarters are situated in Manhattan"

-    >>> # The tokenizer will automatically add any model specific separators (i.e. <CLS> and <SEP>) and tokens to
-    >>> # the sequence, as well as compute the attention masks.
+    >>> # The tokekenizer will automatically add any model specific separators (i.e. <CLS> and <SEP>) and tokens to the sequence, as well as compute the attention masks.
    >>> paraphrase = tokenizer(sequence_0, sequence_2, return_tensors="pt")
    >>> not_paraphrase = tokenizer(sequence_0, sequence_1, return_tensors="pt")

@@ -142,13 +141,12 @@ each other. The process is the following:
    >>> sequence_1 = "Apples are especially bad for your health"
    >>> sequence_2 = "HuggingFace's headquarters are situated in Manhattan"

-    >>> # The tokenizer will automatically add any model specific separators (i.e. <CLS> and <SEP>) and tokens to
-    >>> # the sequence, as well as compute the attention masks.
+    >>> # The tokekenizer will automatically add any model specific separators (i.e. <CLS> and <SEP>) and tokens to the sequence, as well as compute the attention masks.
    >>> paraphrase = tokenizer(sequence_0, sequence_2, return_tensors="tf")
    >>> not_paraphrase = tokenizer(sequence_0, sequence_1, return_tensors="tf")

-    >>> paraphrase_classification_logits = model(paraphrase).logits
-    >>> not_paraphrase_classification_logits = model(not_paraphrase).logits
+    >>> paraphrase_classification_logits = model(paraphrase)[0]
+    >>> not_paraphrase_classification_logits = model(not_paraphrase)[0]

    >>> paraphrase_results = tf.nn.softmax(paraphrase_classification_logits, axis=1).numpy()[0]
    >>> not_paraphrase_results = tf.nn.softmax(not_paraphrase_classification_logits, axis=1).numpy()[0]
@@ -199,11 +197,11 @@ positions of the extracted answer in the text.

    >>> result = question_answerer(question="What is extractive question answering?", context=context)
    >>> print(f"Answer: '{result['answer']}', score: {round(result['score'], 4)}, start: {result['start']}, end: {result['end']}")
-    Answer: 'the task of extracting an answer from a text given a question', score: 0.6177, start: 34, end: 95
+    Answer: 'the task of extracting an answer from a text given a question.', score: 0.6226, start: 34, end: 96

    >>> result = question_answerer(question="What is a good example of a question answering dataset?", context=context)
    >>> print(f"Answer: '{result['answer']}', score: {round(result['score'], 4)}, start: {result['start']}, end: {result['end']}")
-    Answer: 'SQuAD dataset', score: 0.5152, start: 147, end: 160
+    Answer: 'SQuAD dataset,', score: 0.5053, start: 147, end: 161


 Here is an example of question answering using a model and a tokenizer. The process is the following:
@@ -249,10 +247,10 @@ Here is an example of question answering using a model and a tokenizer. The proc
    ...     answer_start_scores = outputs.start_logits
    ...     answer_end_scores = outputs.end_logits
    ...
-    ...     # Get the most likely beginning of answer with the argmax of the score
-    ...     answer_start = torch.argmax(answer_start_scores)
-    ...     # Get the most likely end of answer with the argmax of the score 
-    ...     answer_end = torch.argmax(answer_end_scores) + 1
+    ...     answer_start = torch.argmax(
+    ...         answer_start_scores
+    ...     )  # Get the most likely beginning of answer with the argmax of the score
+    ...     answer_end = torch.argmax(answer_end_scores) + 1  # Get the most likely end of answer with the argmax of the score
    ...
    ...     answer = tokenizer.convert_tokens_to_string(tokenizer.convert_ids_to_tokens(input_ids[answer_start:answer_end]))
    ...
@@ -263,7 +261,7 @@ Here is an example of question answering using a model and a tokenizer. The proc
    Question: What does 🤗 Transformers provide?
    Answer: general - purpose architectures
    Question: 🤗 Transformers provides interoperability between which frameworks?
-    Answer: tensorflow 2. 0 and pytorch
+    Answer: tensorflow 2 . 0 and pytorch
    >>> ## TENSORFLOW CODE
    >>> from transformers import AutoTokenizer, TFAutoModelForQuestionAnswering
    >>> import tensorflow as tf
@@ -292,11 +290,12 @@ Here is an example of question answering using a model and a tokenizer. The proc
    ...     answer_start_scores = outputs.start_logits
    ...     answer_end_scores = outputs.end_logits
    ...
-    ...     # Get the most likely beginning of answer with the argmax of the score
-    ...     answer_start = tf.argmax(answer_start_scores, axis=1).numpy()[0]
-    ...     # Get the most likely end of answer with the argmax of the score
-    ...     answer_end = tf.argmax(answer_end_scores, axis=1).numpy()[0] + 1
-    ...
+    ...     answer_start = tf.argmax(
+    ...         answer_start_scores, axis=1
+    ...     ).numpy()[0]  # Get the most likely beginning of answer with the argmax of the score
+    ...     answer_end = (
+    ...         tf.argmax(answer_end_scores, axis=1) + 1
+    ...     ).numpy()[0]  # Get the most likely end of answer with the argmax of the score
    ...     answer = tokenizer.convert_tokens_to_string(tokenizer.convert_ids_to_tokens(input_ids[answer_start:answer_end]))
    ...
    ...     print(f"Question: {question}")
@@ -306,7 +305,7 @@ Here is an example of question answering using a model and a tokenizer. The proc
    Question: What does 🤗 Transformers provide?
    Answer: general - purpose architectures
    Question: 🤗 Transformers provides interoperability between which frameworks?
-    Answer: tensorflow 2. 0 and pytorch
+    Answer: tensorflow 2 . 0 and pytorch



@@ -345,31 +344,31 @@ This outputs the sequences with the mask filled, the confidence score, and the t

    >>> from pprint import pprint
    >>> pprint(unmasker(f"HuggingFace is creating a {unmasker.tokenizer.mask_token} that the community uses to solve NLP tasks."))
-    [{'score': 0.1793,
-      'sequence': 'HuggingFace is creating a tool that the community uses to solve '
-                  'NLP tasks.',
+    [{'score': 0.1792745739221573,
+      'sequence': '<s>HuggingFace is creating a tool that the community uses to '
+                  'solve NLP tasks.</s>',
      'token': 3944,
-      'token_str': ' tool'},
-     {'score': 0.1135,
-      'sequence': 'HuggingFace is creating a framework that the community uses to '
-                  'solve NLP tasks.',
+      'token_str': 'Ġtool'},
+     {'score': 0.11349421739578247,
+      'sequence': '<s>HuggingFace is creating a framework that the community uses '
+                  'to solve NLP tasks.</s>',
      'token': 7208,
-      'token_str': ' framework'},
-     {'score': 0.0524,
-      'sequence': 'HuggingFace is creating a library that the community uses to '
-                  'solve NLP tasks.',
+      'token_str': 'Ġframework'},
+     {'score': 0.05243554711341858,
+      'sequence': '<s>HuggingFace is creating a library that the community uses to '
+                  'solve NLP tasks.</s>',
      'token': 5560,
-      'token_str': ' library'},
-     {'score': 0.0349,
-      'sequence': 'HuggingFace is creating a database that the community uses to '
-                  'solve NLP tasks.',
+      'token_str': 'Ġlibrary'},
+     {'score': 0.03493533283472061,
+      'sequence': '<s>HuggingFace is creating a database that the community uses '
+                  'to solve NLP tasks.</s>',
      'token': 8503,
-      'token_str': ' database'},
-     {'score': 0.0286,
-      'sequence': 'HuggingFace is creating a prototype that the community uses to '
-                  'solve NLP tasks.',
+      'token_str': 'Ġdatabase'},
+     {'score': 0.02860250137746334,
+      'sequence': '<s>HuggingFace is creating a prototype that the community uses '
+                  'to solve NLP tasks.</s>',
      'token': 17715,
-      'token_str': ' prototype'}]
+      'token_str': 'Ġprototype'}]

 Here is an example of doing masked language modeling using a model and a tokenizer. The process is the following:

@@ -386,48 +385,43 @@ Here is an example of doing masked language modeling using a model and a tokeniz
 .. code-block::

    >>> ## PYTORCH CODE
-    >>> from transformers import AutoModelForMaskedLM, AutoTokenizer
+    >>> from transformers import AutoModelWithLMHead, AutoTokenizer
    >>> import torch

    >>> tokenizer = AutoTokenizer.from_pretrained("distilbert-base-cased")
-    >>> model = AutoModelForMaskedLM.from_pretrained("distilbert-base-cased")
+    >>> model = AutoModelWithLMHead.from_pretrained("distilbert-base-cased")

-    >>> sequence = "Distilled models are smaller than the models they mimic. Using them instead of the large " \
-    ...     f"versions would help {tokenizer.mask_token} our carbon footprint."
+    >>> sequence = f"Distilled models are smaller than the models they mimic. Using them instead of the large versions would help {tokenizer.mask_token} our carbon footprint."

-    >>> inputs = tokenizer(sequence, return_tensors="pt")
-    >>> mask_token_index = torch.where(inputs["input_ids"] == tokenizer.mask_token_id)[1]
+    >>> input = tokenizer.encode(sequence, return_tensors="pt")
+    >>> mask_token_index = torch.where(input == tokenizer.mask_token_id)[1]

-    >>> token_logits = model(**inputs).logits
+    >>> token_logits = model(input).logits
    >>> mask_token_logits = token_logits[0, mask_token_index, :]

    >>> top_5_tokens = torch.topk(mask_token_logits, 5, dim=1).indices[0].tolist()
-
-    >>> for token in top_5_tokens:
-    ...     print(sequence.replace(tokenizer.mask_token, tokenizer.decode([token])))
-    Distilled models are smaller than the models they mimic. Using them instead of the large versions would help reduce our carbon footprint.
-    Distilled models are smaller than the models they mimic. Using them instead of the large versions would help increase our carbon footprint.
-    Distilled models are smaller than the models they mimic. Using them instead of the large versions would help decrease our carbon footprint.
-    Distilled models are smaller than the models they mimic. Using them instead of the large versions would help offset our carbon footprint.
-    Distilled models are smaller than the models they mimic. Using them instead of the large versions would help improve our carbon footprint.
    >>> ## TENSORFLOW CODE
-    >>> from transformers import TFAutoModelForMaskedLM, AutoTokenizer
+    >>> from transformers import TFAutoModelWithLMHead, AutoTokenizer
    >>> import tensorflow as tf

    >>> tokenizer = AutoTokenizer.from_pretrained("distilbert-base-cased")
-    >>> model = TFAutoModelForMaskedLM.from_pretrained("distilbert-base-cased")
+    >>> model = TFAutoModelWithLMHead.from_pretrained("distilbert-base-cased")

-    >>> sequence = "Distilled models are smaller than the models they mimic. Using them instead of the large " \
-    ...     f"versions would help {tokenizer.mask_token} our carbon footprint."
+    >>> sequence = f"Distilled models are smaller than the models they mimic. Using them instead of the large versions would help {tokenizer.mask_token} our carbon footprint."

-    >>> inputs = tokenizer(sequence, return_tensors="tf")
-    >>> mask_token_index = tf.where(inputs["input_ids"] == tokenizer.mask_token_id)[0, 1]
+    >>> input = tokenizer.encode(sequence, return_tensors="tf")
+    >>> mask_token_index = tf.where(input == tokenizer.mask_token_id)[0, 1]

-    >>> token_logits = model(**inputs).logits
+    >>> token_logits = model(input)[0]
    >>> mask_token_logits = token_logits[0, mask_token_index, :]

    >>> top_5_tokens = tf.math.top_k(mask_token_logits, 5).indices.numpy()

+
+This prints five sequences, with the top 5 tokens predicted by the model:
+
+.. code-block::
+
    >>> for token in top_5_tokens:
    ...     print(sequence.replace(tokenizer.mask_token, tokenizer.decode([token])))
    Distilled models are smaller than the models they mimic. Using them instead of the large versions would help reduce our carbon footprint.
@@ -437,9 +431,6 @@ Here is an example of doing masked language modeling using a model and a tokeniz
    Distilled models are smaller than the models they mimic. Using them instead of the large versions would help improve our carbon footprint.


-This prints five sequences, with the top 5 tokens predicted by the model.
-
-
 Causal Language Modeling
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

@@ -458,20 +449,19 @@ of tokens.
 .. code-block::

    >>> ## PYTORCH CODE
-    >>> from transformers import AutoModelForCausalLM, AutoTokenizer, top_k_top_p_filtering
+    >>> from transformers import AutoModelWithLMHead, AutoTokenizer, top_k_top_p_filtering
    >>> import torch
    >>> from torch import nn

    >>> tokenizer = AutoTokenizer.from_pretrained("gpt2")
-    >>> model = AutoModelForCausalLM.from_pretrained("gpt2")
+    >>> model = AutoModelWithLMHead.from_pretrained("gpt2")

    >>> sequence = f"Hugging Face is based in DUMBO, New York City, and"

-    >>> inputs = tokenizer(sequence, return_tensors="pt")
-    >>> input_ids = inputs["input_ids"]
+    >>> input_ids = tokenizer.encode(sequence, return_tensors="pt")

    >>> # get logits of last hidden state
-    >>> next_token_logits = model(**inputs).logits[:, -1, :]
+    >>> next_token_logits = model(input_ids).logits[:, -1, :]

    >>> # filter
    >>> filtered_next_token_logits = top_k_top_p_filtering(next_token_logits, top_k=50, top_p=1.0)
@@ -483,22 +473,19 @@ of tokens.
    >>> generated = torch.cat([input_ids, next_token], dim=-1)

    >>> resulting_string = tokenizer.decode(generated.tolist()[0])
-    >>> print(resulting_string)
-    Hugging Face is based in DUMBO, New York City, and ...
    >>> ## TENSORFLOW CODE
-    >>> from transformers import TFAutoModelForCausalLM, AutoTokenizer, tf_top_k_top_p_filtering
+    >>> from transformers import TFAutoModelWithLMHead, AutoTokenizer, tf_top_k_top_p_filtering
    >>> import tensorflow as tf

    >>> tokenizer = AutoTokenizer.from_pretrained("gpt2")
-    >>> model = TFAutoModelForCausalLM.from_pretrained("gpt2")
+    >>> model = TFAutoModelWithLMHead.from_pretrained("gpt2")

-    >>> sequence = f"Hugging Face is based in DUMBO, New York City, and"
+    >>> sequence = f"Hugging Face is based in DUMBO, New York City, and "

-    >>> inputs = tokenizer(sequence, return_tensors="tf")
-    >>> input_ids = inputs["input_ids"]
+    >>> input_ids = tokenizer.encode(sequence, return_tensors="tf")

    >>> # get logits of last hidden state
-    >>> next_token_logits = model(**inputs).logits[:, -1, :]
+    >>> next_token_logits = model(input_ids)[0][:, -1, :]

    >>> # filter
    >>> filtered_next_token_logits = tf_top_k_top_p_filtering(next_token_logits, top_k=50, top_p=1.0)
@@ -509,11 +496,14 @@ of tokens.
    >>> generated = tf.concat([input_ids, next_token], axis=1)

    >>> resulting_string = tokenizer.decode(generated.numpy().tolist()[0])
-    >>> print(resulting_string)
-    Hugging Face is based in DUMBO, New York City, and ...

-This outputs a (hopefully) coherent next token following the original sequence, which in our case is the word *is* or
-*features*.
+
+This outputs a (hopefully) coherent next token following the original sequence, which in our case is the word *has*:
+
+.. code-block::
+
+    >>> print(resulting_string)
+    Hugging Face is based in DUMBO, New York City, and has

 In the next section, we show how :func:`~transformers.generation_utils.GenerationMixin.generate` can be used to
 generate multiple tokens up to a specified length instead of one token at a time.
@@ -532,8 +522,7 @@ As a default all models apply *Top-K* sampling when used in pipelines, as config

    >>> text_generator = pipeline("text-generation")
    >>> print(text_generator("As far as I am concerned, I will", max_length=50, do_sample=False))
-    [{'generated_text': 'As far as I am concerned, I will be the first to admit that I am not a fan of the idea of a
-    "free market." I think that the idea of a free market is a bit of a stretch. I think that the idea'}]
+    [{'generated_text': 'As far as I am concerned, I will be the first to admit that I am not a fan of the idea of a "free market." I think that the idea of a free market is a bit of a stretch. I think that the idea'}]



@@ -547,9 +536,9 @@ Below is an example of text generation using ``XLNet`` and its tokenizer, which
 .. code-block::

    >>> ## PYTORCH CODE
-    >>> from transformers import AutoModelForCausalLM, AutoTokenizer
+    >>> from transformers import AutoModelWithLMHead, AutoTokenizer

-    >>> model = AutoModelForCausalLM.from_pretrained("xlnet-base-cased")
+    >>> model = AutoModelWithLMHead.from_pretrained("xlnet-base-cased")
    >>> tokenizer = AutoTokenizer.from_pretrained("xlnet-base-cased")

    >>> # Padding text helps XLNet with short prompts - proposed by Aman Rusia in https://github.com/rusiaaman/XLNet-gen#methodology
@@ -565,18 +554,16 @@ Below is an example of text generation using ``XLNet`` and its tokenizer, which
    ... with people, even a bishop, begging for his blessing. <eod> </s> <eos>"""

    >>> prompt = "Today the weather is really nice and I am planning on "
-    >>> inputs = tokenizer(PADDING_TEXT + prompt, add_special_tokens=False, return_tensors="pt")["input_ids"]
+    >>> inputs = tokenizer.encode(PADDING_TEXT + prompt, add_special_tokens=False, return_tensors="pt")

-    >>> prompt_length = len(tokenizer.decode(inputs[0]))
+    >>> prompt_length = len(tokenizer.decode(inputs[0], skip_special_tokens=True, clean_up_tokenization_spaces=True))
    >>> outputs = model.generate(inputs, max_length=250, do_sample=True, top_p=0.95, top_k=60)
-    >>> generated = prompt + tokenizer.decode(outputs[0])[prompt_length+1:]
+    >>> generated = prompt + tokenizer.decode(outputs[0])[prompt_length:]

-    >>> print(generated)
-    Today the weather is really nice and I am planning ...
    >>> ## TENSORFLOW CODE
-    >>> from transformers import TFAutoModelForCausalLM, AutoTokenizer
+    >>> from transformers import TFAutoModelWithLMHead, AutoTokenizer

-    >>> model = TFAutoModelForCausalLM.from_pretrained("xlnet-base-cased")
+    >>> model = TFAutoModelWithLMHead.from_pretrained("xlnet-base-cased")
    >>> tokenizer = AutoTokenizer.from_pretrained("xlnet-base-cased")

    >>> # Padding text helps XLNet with short prompts - proposed by Aman Rusia in https://github.com/rusiaaman/XLNet-gen#methodology
@@ -592,15 +579,16 @@ Below is an example of text generation using ``XLNet`` and its tokenizer, which
    ... with people, even a bishop, begging for his blessing. <eod> </s> <eos>"""

    >>> prompt = "Today the weather is really nice and I am planning on "
-    >>> inputs = tokenizer(PADDING_TEXT + prompt, add_special_tokens=False, return_tensors="tf")["input_ids"]
+    >>> inputs = tokenizer.encode(PADDING_TEXT + prompt, add_special_tokens=False, return_tensors="tf")

-    >>> prompt_length = len(tokenizer.decode(inputs[0]))
+    >>> prompt_length = len(tokenizer.decode(inputs[0], skip_special_tokens=True, clean_up_tokenization_spaces=True))
    >>> outputs = model.generate(inputs, max_length=250, do_sample=True, top_p=0.95, top_k=60)
-    >>> generated = prompt + tokenizer.decode(outputs[0])[prompt_length+1:]
+    >>> generated = prompt + tokenizer.decode(outputs[0])[prompt_length:]
+
+.. code-block::

    >>> print(generated)
-    Today the weather is really nice and I am planning ...
-
+    Today the weather is really nice and I am planning on anning on taking a nice...... of a great time!<eop>...............

 Text generation is currently possible with *GPT-2*, *OpenAi-GPT*, *CTRL*, *XLNet*, *Transfo-XL* and *Reformer* in
 PyTorch and for most models in Tensorflow as well. As can be seen in the example above *XLNet* and *Transfo-XL* often
@@ -650,20 +638,21 @@ Here are the expected results:

 .. code-block::

-    >>> for entity in ner_pipe(sequence):
-    ...     print(entity)
-    {'entity': 'I-ORG', 'score': 0.9996, 'index': 1, 'word': 'Hu', 'start': 0, 'end': 2}
-    {'entity': 'I-ORG', 'score': 0.9910, 'index': 2, 'word': '##gging', 'start': 2, 'end': 7}
-    {'entity': 'I-ORG', 'score': 0.9982, 'index': 3, 'word': 'Face', 'start': 8, 'end': 12}
-    {'entity': 'I-ORG', 'score': 0.9995, 'index': 4, 'word': 'Inc', 'start': 13, 'end': 16}
-    {'entity': 'I-LOC', 'score': 0.9994, 'index': 11, 'word': 'New', 'start': 40, 'end': 43}
-    {'entity': 'I-LOC', 'score': 0.9993, 'index': 12, 'word': 'York', 'start': 44, 'end': 48}
-    {'entity': 'I-LOC', 'score': 0.9994, 'index': 13, 'word': 'City', 'start': 49, 'end': 53}
-    {'entity': 'I-LOC', 'score': 0.9863, 'index': 19, 'word': 'D', 'start': 79, 'end': 80}
-    {'entity': 'I-LOC', 'score': 0.9514, 'index': 20, 'word': '##UM', 'start': 80, 'end': 82}
-    {'entity': 'I-LOC', 'score': 0.9337, 'index': 21, 'word': '##BO', 'start': 82, 'end': 84}
-    {'entity': 'I-LOC', 'score': 0.9762, 'index': 28, 'word': 'Manhattan', 'start': 114, 'end': 123}
-    {'entity': 'I-LOC', 'score': 0.9915, 'index': 29, 'word': 'Bridge', 'start': 124, 'end': 130}
+    >>> print(ner_pipe(sequence))
+    [
+        {'word': 'Hu', 'score': 0.9995632767677307, 'entity': 'I-ORG'},
+        {'word': '##gging', 'score': 0.9915938973426819, 'entity': 'I-ORG'},
+        {'word': 'Face', 'score': 0.9982671737670898, 'entity': 'I-ORG'},
+        {'word': 'Inc', 'score': 0.9994403719902039, 'entity': 'I-ORG'},
+        {'word': 'New', 'score': 0.9994346499443054, 'entity': 'I-LOC'},
+        {'word': 'York', 'score': 0.9993270635604858, 'entity': 'I-LOC'},
+        {'word': 'City', 'score': 0.9993864893913269, 'entity': 'I-LOC'},
+        {'word': 'D', 'score': 0.9825621843338013, 'entity': 'I-LOC'},
+        {'word': '##UM', 'score': 0.936983048915863, 'entity': 'I-LOC'},
+        {'word': '##BO', 'score': 0.8987102508544922, 'entity': 'I-LOC'},
+        {'word': 'Manhattan', 'score': 0.9758241176605225, 'entity': 'I-LOC'},
+        {'word': 'Bridge', 'score': 0.990249514579773, 'entity': 'I-LOC'}
+    ]

 Note how the tokens of the sequence "Hugging Face" have been identified as an organisation, and "New York City",
 "DUMBO" and "Manhattan Bridge" have been identified as locations.
@@ -690,13 +679,26 @@ Here is an example of doing named entity recognition, using a model and a tokeni
    >>> model = AutoModelForTokenClassification.from_pretrained("dbmdz/bert-large-cased-finetuned-conll03-english")
    >>> tokenizer = AutoTokenizer.from_pretrained("bert-base-cased")

-    >>> sequence = "Hugging Face Inc. is a company based in New York City. Its headquarters are in DUMBO, " \
-    ...            "therefore very close to the Manhattan Bridge."
+    >>> label_list = [
+    ...     "O",       # Outside of a named entity
+    ...     "B-MISC",  # Beginning of a miscellaneous entity right after another miscellaneous entity
+    ...     "I-MISC",  # Miscellaneous entity
+    ...     "B-PER",   # Beginning of a person's name right after another person's name
+    ...     "I-PER",   # Person's name
+    ...     "B-ORG",   # Beginning of an organisation right after another organisation
+    ...     "I-ORG",   # Organisation
+    ...     "B-LOC",   # Beginning of a location right after another location
+    ...     "I-LOC"    # Location
+    ... ]

-    >>> inputs = tokenizer(sequence, return_tensors="pt")
-    >>> tokens = inputs.tokens()
+    >>> sequence = "Hugging Face Inc. is a company based in New York City. Its headquarters are in DUMBO, therefore very" \
+    ...            "close to the Manhattan Bridge."

-    >>> outputs = model(**inputs).logits
+    >>> # Bit of a hack to get the tokens with the special tokens
+    >>> tokens = tokenizer.tokenize(tokenizer.decode(tokenizer.encode(sequence)))
+    >>> inputs = tokenizer.encode(sequence, return_tensors="pt")
+
+    >>> outputs = model(inputs).logits
    >>> predictions = torch.argmax(outputs, dim=2)
    >>> ## TENSORFLOW CODE
    >>> from transformers import TFAutoModelForTokenClassification, AutoTokenizer
@@ -705,13 +707,14 @@ Here is an example of doing named entity recognition, using a model and a tokeni
    >>> model = TFAutoModelForTokenClassification.from_pretrained("dbmdz/bert-large-cased-finetuned-conll03-english")
    >>> tokenizer = AutoTokenizer.from_pretrained("bert-base-cased")

-    >>> sequence = "Hugging Face Inc. is a company based in New York City. Its headquarters are in DUMBO, " \
-    ...            "therefore very close to the Manhattan Bridge."
+    >>> sequence = "Hugging Face Inc. is a company based in New York City. Its headquarters are in DUMBO, therefore very" \
+    ...            "close to the Manhattan Bridge."

-    >>> inputs = tokenizer(sequence, return_tensors="tf")
-    >>> tokens = inputs.tokens()
+    >>> # Bit of a hack to get the tokens with the special tokens
+    >>> tokens = tokenizer.tokenize(tokenizer.decode(tokenizer.encode(sequence)))
+    >>> inputs = tokenizer.encode(sequence, return_tensors="tf")

-    >>> outputs = model(**inputs)[0]
+    >>> outputs = model(inputs)[0]
    >>> predictions = tf.argmax(outputs, axis=2)


@@ -752,7 +755,8 @@ illustrated below:
    (',', 'O')
    ('therefore', 'O')
    ('very', 'O')
-    ('close', 'O')
+    ('##c', 'O')
+    ('##lose', 'O')
    ('to', 'O')
    ('the', 'O')
    ('Manhattan', 'I-LOC')
@@ -760,7 +764,6 @@ illustrated below:
    ('.', 'O')
    ('[SEP]', 'O')

-
 Summarization
 -----------------------------------------------------------------------------------------------------------------------

@@ -808,9 +811,7 @@ below. This outputs the following summary:
 .. code-block::

    >>> print(summarizer(ARTICLE, max_length=130, min_length=30, do_sample=False))
-    [{'summary_text': ' Liana Barrientos, 39, is charged with two counts of "offering a false instrument for filing in
-    the first degree" In total, she has been married 10 times, with nine of her marriages occurring between 1999 and
-    2002 . At one time, she was married to eight men at once, prosecutors say .'}]
+    [{'summary_text': 'Liana Barrientos, 39, is charged with two counts of "offering a false instrument for filing in the first degree" In total, she has been married 10 times, with nine of her marriages occurring between 1999 and 2002. She is believed to still be married to four men.'}]

 Here is an example of doing summarization using a model and a tokenizer. The process is the following:

@@ -832,15 +833,8 @@ CNN / Daily Mail), it yields very good results.
    >>> tokenizer = AutoTokenizer.from_pretrained("t5-base")

    >>> # T5 uses a max_length of 512 so we cut the article to 512 tokens.
-    >>> inputs = tokenizer("summarize: " + ARTICLE, return_tensors="pt", max_length=512, truncation=True)
-    >>> outputs = model.generate(
-    ...     inputs["input_ids"], max_length=150, min_length=40, length_penalty=2.0, num_beams=4, early_stopping=True
-    ... )
-
-    >>> print(tokenizer.decode(outputs[0]))
-    <pad> prosecutors say the marriages were part of an immigration scam. if convicted, barrientos faces two criminal
-    counts of "offering a false instrument for filing in the first degree" she has been married 10 times, nine of them
-    between 1999 and 2002.</s>
+    >>> inputs = tokenizer.encode("summarize: " + ARTICLE, return_tensors="pt", max_length=512, truncation=True)
+    >>> outputs = model.generate(inputs, max_length=150, min_length=40, length_penalty=2.0, num_beams=4, early_stopping=True)
    >>> ## TENSORFLOW CODE
    >>> from transformers import TFAutoModelForSeq2SeqLM, AutoTokenizer

@@ -848,15 +842,13 @@ CNN / Daily Mail), it yields very good results.
    >>> tokenizer = AutoTokenizer.from_pretrained("t5-base")

    >>> # T5 uses a max_length of 512 so we cut the article to 512 tokens.
-    >>> inputs = tokenizer("summarize: " + ARTICLE, return_tensors="tf", max_length=512)
-    >>> outputs = model.generate(
-    ...     inputs["input_ids"], max_length=150, min_length=40, length_penalty=2.0, num_beams=4, early_stopping=True
-    ... )
+    >>> inputs = tokenizer.encode("summarize: " + ARTICLE, return_tensors="tf", max_length=512)
+    >>> outputs = model.generate(inputs, max_length=150, min_length=40, length_penalty=2.0, num_beams=4, early_stopping=True)
+
+.. code-block::

    >>> print(tokenizer.decode(outputs[0]))
-    <pad> prosecutors say the marriages were part of an immigration scam. if convicted, barrientos faces two criminal
-    counts of "offering a false instrument for filing in the first degree" she has been married 10 times, nine of them
-    between 1999 and 2002.
+    <pad> prosecutors say the marriages were part of an immigration scam. if convicted, barrientos faces two criminal counts of "offering a false instrument for filing in the first degree" she has been married 10 times, nine of them between 1999 and 2002.</s>


 Translation
@@ -896,32 +888,25 @@ Here is an example of doing translation using a model and a tokenizer. The proce
 .. code-block::

    >>> ## PYTORCH CODE
-    >>> from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
+    >>> from transformers import AutoModelWithLMHead, AutoTokenizer

-    >>> model = AutoModelForSeq2SeqLM.from_pretrained("t5-base")
+    >>> model = AutoModelWithLMHead.from_pretrained("t5-base")
    >>> tokenizer = AutoTokenizer.from_pretrained("t5-base")

-    >>> inputs = tokenizer(
-    ...     "translate English to German: Hugging Face is a technology company based in New York and Paris",
-    ...     return_tensors="pt"
-    ... )
-    >>> outputs = model.generate(inputs["input_ids"], max_length=40, num_beams=4, early_stopping=True)
-
-    >>> print(tokenizer.decode(outputs[0]))
-    <pad> Hugging Face ist ein Technologieunternehmen mit Sitz in New York und Paris.</s>
+    >>> inputs = tokenizer.encode("translate English to German: Hugging Face is a technology company based in New York and Paris", return_tensors="pt")
+    >>> outputs = model.generate(inputs, max_length=40, num_beams=4, early_stopping=True)
    >>> ## TENSORFLOW CODE
-    >>> from transformers import TFAutoModelForSeq2SeqLM, AutoTokenizer
+    >>> from transformers import TFAutoModelWithLMHead, AutoTokenizer

-    >>> model = TFAutoModelForSeq2SeqLM.from_pretrained("t5-base")
+    >>> model = TFAutoModelWithLMHead.from_pretrained("t5-base")
    >>> tokenizer = AutoTokenizer.from_pretrained("t5-base")

-    >>> inputs = tokenizer(
-    ...     "translate English to German: Hugging Face is a technology company based in New York and Paris",
-    ...     return_tensors="tf"
-    ... )
-    >>> outputs = model.generate(inputs["input_ids"], max_length=40, num_beams=4, early_stopping=True)
+    >>> inputs = tokenizer.encode("translate English to German: Hugging Face is a technology company based in New York and Paris", return_tensors="tf")
+    >>> outputs = model.generate(inputs, max_length=40, num_beams=4, early_stopping=True)
+
+As with the pipeline example, we get the same translation:
+
+.. code-block::

    >>> print(tokenizer.decode(outputs[0]))
-    <pad> Hugging Face ist ein Technologieunternehmen mit Sitz in New York und Paris.
-
-We get the same translation as with the pipeline example.
+    Hugging Face ist ein Technologieunternehmen mit Sitz in New York und Paris.
--- a/docs/source/training.rst
+++ b/docs/source/training.rst
@@ -281,7 +281,7 @@ Fine-tuning in native PyTorch
   frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope;
   picture-in-picture" allowfullscreen></iframe>

-You might need to restart your notebook at this stage to free some memory, or execute the following code:
+You might need to restart your notebook at this stage to free some memory, or excute the following code:

 .. code-block:: python

--- a/examples/flax/language-modeling/README.md
+++ b/examples/flax/language-modeling/README.md
@@ -49,15 +49,21 @@ Next we clone the model repository to add the tokenizer and model files.
 git clone https://huggingface.co/<your-username>/norwegian-roberta-base
 ```

-To setup all relevant files for training, let's go into the cloned model directory.
+To ensure that all tensorboard traces will be uploaded correctly, we need to 
+track them. You can run the following command inside your model repo to do so.

-```bash
-cd norwegian-roberta-base
 ```
+cd norwegian-roberta-base
+git lfs track "*tfevents*"
+```
+
+Great, we have set up our model repository. During training, we will automatically
+push the training logs and model weights to the repo.

 Next, let's add a symbolic link to the `run_mlm_flax.py`.

 ```bash
+export MODEL_DIR="./norwegian-roberta-base"
 ln -s ~/transformers/examples/flax/language-modeling/run_mlm_flax.py run_mlm_flax.py
 ```

@@ -65,13 +71,15 @@ ln -s ~/transformers/examples/flax/language-modeling/run_mlm_flax.py run_mlm_fla

 In the first step, we train a tokenizer to efficiently process the text input for the model. Similar to how it is shown in [How to train a new language model from scratch using Transformers and Tokenizers](https://huggingface.co/blog/how-to-train), we use a **`ByteLevelBPETokenizer`**.
 The tokenizer is trained on the complete Norwegian dataset of OSCAR
-and consequently saved in the cloned model directory.
+and consequently saved in `${MODEL_DIR}`
 This can take up to 10 minutes depending on your hardware ☕.

 ```python
 from datasets import load_dataset
 from tokenizers import trainers, Tokenizer, normalizers, ByteLevelBPETokenizer

+model_dir = "./norwegian-roberta-base"  # ${MODEL_DIR}
+
 # load dataset
 dataset = load_dataset("oscar", "unshuffled_deduplicated_no", split="train")

@@ -92,7 +100,7 @@ tokenizer.train_from_iterator(batch_iterator(), vocab_size=50265, min_frequency=
 ])

 # Save files to disk
-tokenizer.save("./")
+tokenizer.save(f"{model_dir}/tokenizer.json")
 ```

 ### Create configuration
@@ -104,12 +112,11 @@ in the local model folder:
 ```python
 from transformers import RobertaConfig

-config = RobertaConfig.from_pretrained("roberta-base", vocab_size=50265)
-config.save_pretrained("./")
-```
+model_dir = "./norwegian-roberta-base"  # ${MODEL_DIR}

-Great, we have set up our model repository. During training, we will automatically
-push the training logs and model weights to the repo.
+config = RobertaConfig.from_pretrained("roberta-base", vocab_size=tokenizer.vocab_size)
+config.save_pretrained(model_dir)
+```

 ### Train model

@@ -117,10 +124,10 @@ Next we can run the example script to pretrain the model:

 ```bash
 ./run_mlm_flax.py \
-    --output_dir="./" \
+    --output_dir="${MODEL_DIR}" \
    --model_type="roberta" \
-    --config_name="./" \
-    --tokenizer_name="./" \
+    --config_name="${MODEL_DIR}" \
+    --tokenizer_name="${MODEL_DIR}" \
    --dataset_name="oscar" \
    --dataset_config_name="unshuffled_deduplicated_no" \
    --max_seq_length="128" \
@@ -142,7 +149,7 @@ Next we can run the example script to pretrain the model:
 Training should converge at a loss and accuracy 
 of 1.78 and 0.64 respectively after 18 epochs on a single TPUv3-8.
 This should take less than 18 hours.
-Training statistics can be accessed on [tfhub.dev](https://tensorboard.dev/experiment/GdYmdak2TWeVz0DDRYOrrg).
+Training statistics can be accessed on [tfhub.de](https://tensorboard.dev/experiment/GdYmdak2TWeVz0DDRYOrrg).

 For a step-by-step walkthrough of how to do masked language modeling in Flax, please have a 
 look at [this](https://colab.research.google.com/github/huggingface/notebooks/blob/master/examples/masked_language_modeling_flax.ipynb) google colab.
@@ -173,51 +180,25 @@ Next we clone the model repository to add the tokenizer and model files.
 git clone https://huggingface.co/<your-username>/norwegian-gpt2
 ```

-To setup all relevant files for training, let's go into the cloned model directory.
+To ensure that all tensorboard traces will be uploaded correctly, we need to 
+track them. You can run the following command inside your model repo to do so.

-```bash
+```
 cd norwegian-gpt2
+git lfs track "*tfevents*"
 ```

-Next, let's add a symbolic link to the training script `run_clm_flax.py`.
+Great, we have set up our model repository. During training, we will automatically
+push the training logs and model weights to the repo.
+
+Next, let's add a symbolic link to the `run_clm_flax.py`.

 ```bash
+export MODEL_DIR="./norwegian-gpt2"
 ln -s ~/transformers/examples/flax/language-modeling/run_clm_flax.py run_clm_flax.py
 ```

-### Train tokenizer
-
-In the first step, we train a tokenizer to efficiently process the text input for the model. Similar to how it is shown in [How to train a new language model from scratch using Transformers and Tokenizers](https://huggingface.co/blog/how-to-train), we use a **`ByteLevelBPETokenizer`**.
-The tokenizer is trained on the complete Norwegian dataset of OSCAR
-and consequently saved in the cloned model directory.
-This can take up to 10 minutes depending on your hardware ☕.
-
-```python
-from datasets import load_dataset
-from tokenizers import trainers, Tokenizer, normalizers, ByteLevelBPETokenizer
-
-# load dataset
-dataset = load_dataset("oscar", "unshuffled_deduplicated_no", split="train")
-
-# Instantiate tokenizer
-tokenizer = ByteLevelBPETokenizer()
-
-def batch_iterator(batch_size=1000):
-    for i in range(0, len(dataset), batch_size):
-        yield dataset[i: i + batch_size]["text"]
-
-# Customized training
-tokenizer.train_from_iterator(batch_iterator(), vocab_size=50257, min_frequency=2, special_tokens=[
-    "<s>",
-    "<pad>",
-    "</s>",
-    "<unk>",
-    "<mask>",
-])
-
-# Save files to disk
-tokenizer.save("./tokenizer.json")
-```
+Next, we'll follow the same steps as above in [Train tokenizer](#train-tokenizer) to train the tokenizer.

 ### Create configuration

@@ -228,23 +209,22 @@ in the local model folder:
 ```python
 from transformers import GPT2Config

-config = GPT2Config.from_pretrained("gpt2", resid_pdrop=0.0, embd_pdrop=0.0, attn_pdrop=0.0, vocab_size=50257)
-config.save_pretrained("./")
-```
+model_dir = "./norwegian-gpt2"  # ${MODEL_DIR}

-Great, we have set up our model repository. During training, we will now automatically
-push the training logs and model weights to the repo.
+config = GPT2Config.from_pretrained("gpt2", resid_pdrop=0.0, embd_pdrop=0.0, attn_pdrop=0.0, vocab_size=tokenizer.vocab_size)
+config.save_pretrained(model_dir)
+```

 ### Train model

-Finally, we can run the example script to pretrain the model:
+Next we can run the example script to pretrain the model:

 ```bash
 ./run_clm_flax.py \
-    --output_dir="./l" \
+    --output_dir="${MODEL_DIR}" \
    --model_type="gpt2" \
-    --config_name="./" \
-    --tokenizer_name="./" \
+    --config_name="${MODEL_DIR}" \
+    --tokenizer_name="${MODEL_DIR}" \
    --dataset_name="oscar" \
    --dataset_config_name="unshuffled_deduplicated_no" \
    --do_train --do_eval \
@@ -266,9 +246,6 @@ of 3.24 and 25.72 respectively after 20 epochs on a single TPUv3-8.
 This should take less than ~21 hours.
 Training statistics can be accessed on [tfhub.de](https://tensorboard.dev/experiment/2zEhLwJ0Qp2FAkI3WVH9qA).

-For a step-by-step walkthrough of how to do causal language modeling in Flax, please have a 
-look at [this](https://colab.research.google.com/github/huggingface/notebooks/blob/master/examples/causal_language_modeling_flax.ipynb) google colab.
-
 ## T5-like span-masked language modeling

 In the following, we demonstrate how to train a T5 model using the span-masked language model 
@@ -295,15 +272,21 @@ Next we clone the model repository to add the tokenizer and model files.
 git clone https://huggingface.co/<your-username>/norwegian-t5-base
 ```

-To setup all relevant files for trairing, let's go into the cloned model directory.
+To ensure that all tensorboard traces will be uploaded correctly, we need to 
+track them. You can run the following command inside your model repo to do so.

-```bash
-cd norwegian-t5-base
 ```
+cd norwegian-t5-base
+git lfs track "*tfevents*"
+```
+
+Great, we have set up our model repository. During training, we will automatically
+push the training logs and model weights to the repo.

 Next, let's add a symbolic link to the `run_t5_mlm_flax.py` and `t5_tokenizer_model` scripts.

 ```bash
+export MODEL_DIR="./norwegian-t5-base"
 ln -s ~/transformers/examples/flax/language-modeling/run_t5_mlm_flax.py run_t5_mlm_flax.py
 ln -s ~/transformers/examples/flax/language-modeling/t5_tokenizer_model.py t5_tokenizer_model.py
 ```
@@ -316,7 +299,7 @@ a sentencepiece unigram tokenizer as shown in [t5_tokenizer_model.py](https://gi
 which is heavily inspired from [yandex-research/DeDLOC's tokenizer model](https://github.com/yandex-research/DeDLOC/blob/5c994bc64e573702a9a79add3ecd68b38f14b548/sahajbert/tokenizer/tokenizer_model.py) .

 The tokenizer is trained on the complete Norwegian dataset of OSCAR
-and consequently saved in the cloned model directory.
+and consequently saved in `${MODEL_DIR}`
 This can take up to 120 minutes depending on your hardware ☕☕☕ .

 ```python
@@ -327,6 +310,7 @@ from t5_tokenizer_model import SentencePieceUnigramTokenizer

 vocab_size = 32_000
 input_sentence_size = None
+model_dir = "./norwegian-t5-base"  # ${MODEL_DIR}

 # Initialize a dataset
 dataset = datasets.load_dataset("oscar", name="unshuffled_deduplicated_no", split="train")
@@ -351,7 +335,7 @@ tokenizer.train_from_iterator(
 )

 # Save files to disk
-tokenizer.save("./tokenizer.json")
+tokenizer.save(f"{model_dir}/tokenizer.json")
 ```

 ### Create configuration
@@ -363,12 +347,11 @@ in the local model folder:
 ```python
 from transformers import T5Config

-config = T5Config.from_pretrained("google/t5-v1_1-base", vocab_size=tokenizer.get_vocab_size())
-config.save_pretrained("./")
-```
+model_dir = "./norwegian-t5-base"  # ${MODEL_DIR}

-Great, we have set up our model repository. During training, we will automatically
-push the training logs and model weights to the repo.
+config = T5Config.from_pretrained("google/t5-v1_1-base", vocab_size=tokenizer.vocab_size)
+config.save_pretrained(model_dir)
+```

 ### Train model

@@ -390,15 +373,15 @@ Next we can run the example script to pretrain the model:
 	--weight_decay="0.001" \
 	--warmup_steps="2000" \
 	--overwrite_output_dir \
-	--logging_steps="500" \
-	--save_steps="10000" \
-	--eval_steps="2500" \
+	--logging_steps="100" \
+	--save_steps="1000" \
+	--eval_steps="1000" \
 	--push_to_hub
 ```

 Training should converge at a loss and accuracy 
-of 2.36 and 57.0 respectively after 3 epochs on a single TPUv3-8.
-This should take around 4.5 hours.
+of 2.2 and 58.0 respectively after 2 epochs on a single TPUv3-8.
+This should take around 24 hours.
 Training statistics can be accessed on directly on the 🤗 [hub](https://huggingface.co/patrickvonplaten/t5-base-norwegian/tensorboard)

 ## Runtime evaluation
--- a/examples/flax/language-modeling/run_clm_flax.py
+++ b/examples/flax/language-modeling/run_clm_flax.py
@@ -31,7 +31,6 @@ from pathlib import Path
 from typing import Callable, Optional

 import datasets
-import numpy as np
 from datasets import Dataset, load_dataset
 from tqdm import tqdm

@@ -52,7 +51,6 @@ from transformers import (
    HfArgumentParser,
    TrainingArguments,
    is_tensorboard_available,
-    set_seed,
 )
 from transformers.testing_utils import CaptureLogger

@@ -156,9 +154,6 @@ class DataTrainingArguments:
        default=None,
        metadata={"help": "The number of processes to use for the preprocessing."},
    )
-    keep_linebreaks: bool = field(
-        default=True, metadata={"help": "Whether to keep line breaks when using TXT files or not."}
-    )

    def __post_init__(self):
        if self.dataset_name is None and self.train_file is None and self.validation_file is None:
@@ -187,16 +182,18 @@ def data_loader(rng: jax.random.PRNGKey, dataset: Dataset, batch_size: int, shuf
    steps_per_epoch = len(dataset) // batch_size

    if shuffle:
-        batch_idx = np.random.permutation(len(dataset))
+        batch_idx = jax.random.permutation(rng, len(dataset))
    else:
-        batch_idx = np.arange(len(dataset))
+        batch_idx = jnp.arange(len(dataset))

    batch_idx = batch_idx[: steps_per_epoch * batch_size]  # Skip incomplete batch.
    batch_idx = batch_idx.reshape((steps_per_epoch, batch_size))

    for idx in batch_idx:
        batch = dataset[idx]
-        batch = {k: np.array(v) for k, v in batch.items()}
+        batch = {k: jnp.array(v) for k, v in batch.items()}
+
+        batch = shard(batch)

        yield batch

@@ -272,9 +269,6 @@ def main():
    # Set the verbosity to info of the Transformers logger (on main process only):
    logger.info(f"Training/evaluation parameters {training_args}")

-    # Set seed before initializing model.
-    set_seed(training_args.seed)
-
    #  Get the datasets: you can either provide your own CSV/JSON/TXT training and evaluation files (see below)
    # or just provide the name of one of the public datasets available on the hub at https://huggingface.co/datasets/
    # (the dataset will be downloaded automatically from the datasets Hub).
@@ -305,7 +299,6 @@ def main():
            )
    else:
        data_files = {}
-        dataset_args = {}
        if data_args.train_file is not None:
            data_files["train"] = data_args.train_file
        if data_args.validation_file is not None:
@@ -313,23 +306,20 @@ def main():
        extension = data_args.train_file.split(".")[-1]
        if extension == "txt":
            extension = "text"
-            dataset_args["keep_linebreaks"] = data_args.keep_linebreaks
-        dataset = load_dataset(extension, data_files=data_files, cache_dir=model_args.cache_dir, **dataset_args)
+        dataset = load_dataset(extension, data_files=data_files, cache_dir=model_args.cache_dir)

-        if "validation" not in dataset.keys():
-            dataset["validation"] = load_dataset(
+        if "validation" not in datasets.keys():
+            datasets["validation"] = load_dataset(
                extension,
                data_files=data_files,
                split=f"train[:{data_args.validation_split_percentage}%]",
                cache_dir=model_args.cache_dir,
-                **dataset_args,
            )
-            dataset["train"] = load_dataset(
+            datasets["train"] = load_dataset(
                extension,
                data_files=data_files,
                split=f"train[{data_args.validation_split_percentage}%:]",
                cache_dir=model_args.cache_dir,
-                **dataset_args,
            )
    # See more about loading any type of standard or custom dataset (from files, python dict, pandas DataFrame, etc) at
    # https://huggingface.co/docs/datasets/loading_datasets.html.
@@ -587,7 +577,7 @@ def main():

    train_time = 0
    train_metrics = []
-    epochs = tqdm(range(num_epochs), desc="Epoch ... ", position=0)
+    epochs = tqdm(range(num_epochs), desc=f"Epoch ... (1/{num_epochs})", position=0)
    for epoch in epochs:
        # ======================== Training ================================
        train_start = time.time()
@@ -601,7 +591,6 @@ def main():
        # train
        for step in tqdm(range(steps_per_epoch), desc="Training...", position=1, leave=False):
            batch = next(train_loader)
-            batch = shard(batch)
            state, train_metric = p_train_step(state, batch)
            train_metrics.append(train_metric)

@@ -628,7 +617,6 @@ def main():
                for _ in tqdm(range(eval_steps), desc="Evaluating...", position=2, leave=False):
                    # Model forward
                    batch = next(eval_loader)
-                    batch = shard(batch)
                    metrics = p_eval_step(state.params, batch)
                    eval_metrics.append(metrics)

--- a/examples/flax/language-modeling/run_mlm_flax.py
+++ b/examples/flax/language-modeling/run_mlm_flax.py
@@ -214,7 +214,7 @@ class FlaxDataCollatorForLanguageModeling:

    def mask_tokens(
        self, inputs: np.ndarray, special_tokens_mask: Optional[np.ndarray]
-    ) -> Tuple[np.ndarray, np.ndarray]:
+    ) -> Tuple[jnp.ndarray, jnp.ndarray]:
        """
        Prepare masked tokens inputs/labels for masked language modeling: 80% MASK, 10% random, 10% original.
        """
--- a/examples/flax/language-modeling/run_t5_mlm_flax.py
+++ b/examples/flax/language-modeling/run_t5_mlm_flax.py
@@ -353,8 +353,7 @@ class FlaxDataCollatorForT5MLM:
            np.random.shuffle(mask_indices)
            first_in_segment = np.pad(mask_indices, [[1, 0]])
            segment_id = np.cumsum(first_in_segment)
-            # count length of sub segments assuming that list is sorted
-            _, segment_length = np.unique(segment_id, return_counts=True)
+            segment_length = np.asarray(jax.ops.segment_sum(np.ones_like(segment_id), segment_id))
            return segment_length

        noise_span_lengths = _random_segmentation(num_noise_tokens, num_noise_spans)
@@ -721,7 +720,7 @@ if __name__ == "__main__":
    state = jax_utils.replicate(state)

    train_time = 0
-    epochs = tqdm(range(num_epochs), desc="Epoch ... ", position=0)
+    epochs = tqdm(range(num_epochs), desc=f"Epoch ... (1/{num_epochs})", position=0)
    for epoch in epochs:
        # ======================== Training ================================
        train_start = time.time()
--- a/examples/flax/text-classification/README.md
+++ b/examples/flax/text-classification/README.md
@@ -100,7 +100,7 @@ In the Tensorboard results linked below, the random seed of each model is equal

 | Task  | Metric                       | Acc (best run) | Acc (avg/5runs) | Stdev     | Metrics                                                                  |
 |-------|------------------------------|----------------|-----------------|-----------|--------------------------------------------------------------------------|
-| CoLA  | Matthews corr                | 60.57          | 59.04           | 1.06      | [tfhub.dev](https://tensorboard.dev/experiment/lfr2adVpRtmLDALKrElkzg/)  |
+| CoLA  | Matthew's corr               | 60.57          | 59.04           | 1.06      | [tfhub.dev](https://tensorboard.dev/experiment/lfr2adVpRtmLDALKrElkzg/)  |
 | SST-2 | Accuracy                     | 92.66          | 92.23           | 0.57      | [tfhub.dev](https://tensorboard.dev/experiment/jYvfv2trRHKMjoWnXVwrZA/)  |
 | MRPC  | F1/Accuracy                  | 89.90/85.78    | 88.97/84.36     | 0.72/1.09 | [tfhub.dev](https://tensorboard.dev/experiment/bo3W3DEoRw2Q7YXjWrJkfg/)  |
 | STS-B | Pearson/Spearman corr.       | 89.04/88.70    | 88.94/88.63     | 0.07/0.07 | [tfhub.dev](https://tensorboard.dev/experiment/fxVwbLD7QpKhbot0r9rn2w/)  |
--- a/examples/legacy/multiple_choice/utils_multiple_choice.py
+++ b/examples/legacy/multiple_choice/utils_multiple_choice.py
@@ -77,7 +77,7 @@ class Split(Enum):

 if is_torch_available():
    import torch
-    from torch.utils.data import Dataset
+    from torch.utils.data.dataset import Dataset

    class MultipleChoiceDataset(Dataset):
        """
--- a/examples/legacy/seq2seq/seq2seq_trainer.py
+++ b/examples/legacy/seq2seq/seq2seq_trainer.py
@@ -141,7 +141,7 @@ class Seq2SeqTrainer(Trainer):
            )
        return scheduler

-    def _get_train_sampler(self) -> Optional[torch.utils.data.Sampler]:
+    def _get_train_sampler(self) -> Optional[torch.utils.data.sampler.Sampler]:
        if isinstance(self.train_dataset, torch.utils.data.IterableDataset):
            return None
        elif is_torch_tpu_available():
--- a/examples/legacy/token-classification/utils_ner.py
+++ b/examples/legacy/token-classification/utils_ner.py
@@ -206,7 +206,7 @@ class TokenClassificationTask:
 if is_torch_available():
    import torch
    from torch import nn
-    from torch.utils.data import Dataset
+    from torch.utils.data.dataset import Dataset

    class TokenClassificationDataset(Dataset):
        """
--- a/examples/pytorch/language-modeling/requirements.txt
+++ b/examples/pytorch/language-modeling/requirements.txt
@@ -1,4 +1,3 @@
-accelerate
 torch >= 1.3
 datasets >= 1.8.0
 sentencepiece != 0.1.92
--- a/examples/pytorch/language-modeling/run_clm.py
+++ b/examples/pytorch/language-modeling/run_clm.py
@@ -51,7 +51,7 @@ from transformers.utils.versions import require_version


 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.10.0")
+check_min_version("4.9.0")

 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/language-modeling/requirements.txt")

@@ -172,9 +172,6 @@ class DataTrainingArguments:
        default=None,
        metadata={"help": "The number of processes to use for the preprocessing."},
    )
-    keep_linebreaks: bool = field(
-        default=True, metadata={"help": "Whether to keep line breaks when using TXT files or not."}
-    )

    def __post_init__(self):
        if self.dataset_name is None and self.train_file is None and self.validation_file is None:
@@ -269,7 +266,6 @@ def main():
            )
    else:
        data_files = {}
-        dataset_args = {}
        if data_args.train_file is not None:
            data_files["train"] = data_args.train_file
        if data_args.validation_file is not None:
@@ -281,8 +277,7 @@ def main():
        )
        if extension == "txt":
            extension = "text"
-            dataset_args["keep_linebreaks"] = data_args.keep_linebreaks
-        raw_datasets = load_dataset(extension, data_files=data_files, cache_dir=model_args.cache_dir, **dataset_args)
+        raw_datasets = load_dataset(extension, data_files=data_files, cache_dir=model_args.cache_dir)
        # If no validation data is there, validation_split_percentage will be used to divide the dataset.
        if "validation" not in raw_datasets.keys():
            raw_datasets["validation"] = load_dataset(
@@ -290,14 +285,12 @@ def main():
                data_files=data_files,
                split=f"train[:{data_args.validation_split_percentage}%]",
                cache_dir=model_args.cache_dir,
-                **dataset_args,
            )
            raw_datasets["train"] = load_dataset(
                extension,
                data_files=data_files,
                split=f"train[{data_args.validation_split_percentage}%:]",
                cache_dir=model_args.cache_dir,
-                **dataset_args,
            )

    # See more about loading any type of standard or custom dataset (from files, python dict, pandas DataFrame, etc) at
--- a/examples/pytorch/language-modeling/run_clm_no_trainer.py
+++ b/examples/pytorch/language-modeling/run_clm_no_trainer.py
@@ -31,7 +31,7 @@ import random
 import datasets
 import torch
 from datasets import load_dataset
-from torch.utils.data import DataLoader
+from torch.utils.data.dataloader import DataLoader
 from tqdm.auto import tqdm

 import transformers
@@ -173,9 +173,6 @@ def parse_args():
    parser.add_argument(
        "--overwrite_cache", type=bool, default=False, help="Overwrite the cached training and evaluation sets"
    )
-    parser.add_argument(
-        "--no_keep_linebreaks", action="store_true", help="Do not keep line breaks when using TXT files."
-    )

    args = parser.parse_args()

@@ -248,7 +245,6 @@ def main():
            )
    else:
        data_files = {}
-        dataset_args = {}
        if args.train_file is not None:
            data_files["train"] = args.train_file
        if args.validation_file is not None:
@@ -256,21 +252,18 @@ def main():
        extension = args.train_file.split(".")[-1]
        if extension == "txt":
            extension = "text"
-            dataset_args["keep_linebreaks"] = not args.no_keep_linebreaks
-        raw_datasets = load_dataset(extension, data_files=data_files, **dataset_args)
+        raw_datasets = load_dataset(extension, data_files=data_files)
        # If no validation data is there, validation_split_percentage will be used to divide the dataset.
        if "validation" not in raw_datasets.keys():
            raw_datasets["validation"] = load_dataset(
                extension,
                data_files=data_files,
                split=f"train[:{args.validation_split_percentage}%]",
-                **dataset_args,
            )
            raw_datasets["train"] = load_dataset(
                extension,
                data_files=data_files,
                split=f"train[{args.validation_split_percentage}%:]",
-                **dataset_args,
            )

    # See more about loading any type of standard or custom dataset (from files, python dict, pandas DataFrame, etc) at
--- a/examples/pytorch/language-modeling/run_mlm.py
+++ b/examples/pytorch/language-modeling/run_mlm.py
@@ -50,7 +50,7 @@ from transformers.utils.versions import require_version


 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.10.0")
+check_min_version("4.9.0")

 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/language-modeling/requirements.txt")

--- a/examples/pytorch/language-modeling/run_mlm_no_trainer.py
+++ b/examples/pytorch/language-modeling/run_mlm_no_trainer.py
@@ -31,7 +31,7 @@ import random
 import datasets
 import torch
 from datasets import load_dataset
-from torch.utils.data import DataLoader
+from torch.utils.data.dataloader import DataLoader
 from tqdm.auto import tqdm

 import transformers
--- a/examples/pytorch/language-modeling/run_plm.py
+++ b/examples/pytorch/language-modeling/run_plm.py
@@ -46,7 +46,7 @@ from transformers.utils.versions import require_version


 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.10.0")
+check_min_version("4.9.0")

 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/language-modeling/requirements.txt")

--- a/examples/pytorch/multiple-choice/requirements.txt
+++ b/examples/pytorch/multiple-choice/requirements.txt
@@ -1,4 +1,3 @@
-accelerate
 sentencepiece != 0.1.92
 protobuf
 torch >= 1.3
--- a/examples/pytorch/multiple-choice/run_swag.py
+++ b/examples/pytorch/multiple-choice/run_swag.py
@@ -47,7 +47,7 @@ from transformers.utils import check_min_version


 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.10.0")
+check_min_version("4.9.0")

 logger = logging.getLogger(__name__)

--- a/examples/pytorch/multiple-choice/run_swag_no_trainer.py
+++ b/examples/pytorch/multiple-choice/run_swag_no_trainer.py
@@ -29,7 +29,7 @@ from typing import Optional, Union
 import datasets
 import torch
 from datasets import load_dataset, load_metric
-from torch.utils.data import DataLoader
+from torch.utils.data.dataloader import DataLoader
 from tqdm.auto import tqdm

 import transformers
--- a/examples/pytorch/question-answering/requirements.txt
+++ b/examples/pytorch/question-answering/requirements.txt
@@ -1,3 +1,2 @@
-accelerate
 datasets >= 1.8.0
 torch >= 1.3.0
--- a/examples/pytorch/question-answering/run_qa.py
+++ b/examples/pytorch/question-answering/run_qa.py
@@ -48,7 +48,7 @@ from utils_qa import postprocess_qa_predictions


 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.10.0")
+check_min_version("4.9.0")

 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/question-answering/requirements.txt")

@@ -339,11 +339,6 @@ def main():

    # Training preprocessing
    def prepare_train_features(examples):
-        # Some of the questions have lots of whitespace on the left, which is not useful and will make the
-        # truncation of the context fail (the tokenized question will take a lots of space). So we remove that
-        # left whitespace
-        examples[question_column_name] = [q.lstrip() for q in examples[question_column_name]]
-
        # Tokenize our examples with truncation and maybe padding, but keep the overflows using a stride. This results
        # in one example possible giving several features when a context is long, each of those features having a
        # context that overlaps a bit the context of the previous feature.
@@ -438,11 +433,6 @@ def main():

    # Validation preprocessing
    def prepare_validation_features(examples):
-        # Some of the questions have lots of whitespace on the left, which is not useful and will make the
-        # truncation of the context fail (the tokenized question will take a lots of space). So we remove that
-        # left whitespace
-        examples[question_column_name] = [q.lstrip() for q in examples[question_column_name]]
-
        # Tokenize our examples with truncation and maybe padding, but keep the overflows using a stride. This results
        # in one example possible giving several features when a context is long, each of those features having a
        # context that overlaps a bit the context of the previous feature.
--- a/examples/pytorch/question-answering/run_qa_beam_search.py
+++ b/examples/pytorch/question-answering/run_qa_beam_search.py
@@ -47,7 +47,7 @@ from utils_qa import postprocess_qa_predictions_with_beam_search


 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.10.0")
+check_min_version("4.9.0")

 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/question-answering/requirements.txt")

@@ -327,11 +327,6 @@ def main():

    # Training preprocessing
    def prepare_train_features(examples):
-        # Some of the questions have lots of whitespace on the left, which is not useful and will make the
-        # truncation of the context fail (the tokenized question will take a lots of space). So we remove that
-        # left whitespace
-        examples[question_column_name] = [q.lstrip() for q in examples[question_column_name]]
-
        # Tokenize our examples with truncation and maybe padding, but keep the overflows using a stride. This results
        # in one example possible giving several features when a context is long, each of those features having a
        # context that overlaps a bit the context of the previous feature.
--- a/examples/pytorch/question-answering/run_qa_beam_search_no_trainer.py
+++ b/examples/pytorch/question-answering/run_qa_beam_search_no_trainer.py
@@ -28,7 +28,7 @@ import datasets
 import numpy as np
 import torch
 from datasets import load_dataset, load_metric
-from torch.utils.data import DataLoader
+from torch.utils.data.dataloader import DataLoader
 from tqdm.auto import tqdm

 import transformers
@@ -51,7 +51,7 @@ from utils_qa import postprocess_qa_predictions_with_beam_search


 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.10.0")
+check_min_version("4.9.0")

 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/question-answering/requirements.txt")

@@ -315,11 +315,6 @@ def main():

    # Training preprocessing
    def prepare_train_features(examples):
-        # Some of the questions have lots of whitespace on the left, which is not useful and will make the
-        # truncation of the context fail (the tokenized question will take a lots of space). So we remove that
-        # left whitespace
-        examples[question_column_name] = [q.lstrip() for q in examples[question_column_name]]
-
        # Tokenize our examples with truncation and maybe padding, but keep the overflows using a stride. This results
        # in one example possible giving several features when a context is long, each of those features having a
        # context that overlaps a bit the context of the previous feature.
@@ -435,11 +430,6 @@ def main():

    # Validation preprocessing
    def prepare_validation_features(examples):
-        # Some of the questions have lots of whitespace on the left, which is not useful and will make the
-        # truncation of the context fail (the tokenized question will take a lots of space). So we remove that
-        # left whitespace
-        examples[question_column_name] = [q.lstrip() for q in examples[question_column_name]]
-
        # Tokenize our examples with truncation and maybe padding, but keep the overflows using a stride. This results
        # in one example possible giving several features when a context is long, each of those features having a
        # context that overlaps a bit the context of the previous feature.
--- a/examples/pytorch/question-answering/run_qa_no_trainer.py
+++ b/examples/pytorch/question-answering/run_qa_no_trainer.py
@@ -28,7 +28,7 @@ import datasets
 import numpy as np
 import torch
 from datasets import load_dataset, load_metric
-from torch.utils.data import DataLoader
+from torch.utils.data.dataloader import DataLoader
 from tqdm.auto import tqdm

 import transformers
@@ -53,7 +53,7 @@ from utils_qa import postprocess_qa_predictions


 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.10.0")
+check_min_version("4.9.0")

 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/question-answering/requirements.txt")

@@ -367,11 +367,6 @@ def main():

    # Training preprocessing
    def prepare_train_features(examples):
-        # Some of the questions have lots of whitespace on the left, which is not useful and will make the
-        # truncation of the context fail (the tokenized question will take a lots of space). So we remove that
-        # left whitespace
-        examples[question_column_name] = [q.lstrip() for q in examples[question_column_name]]
-
        # Tokenize our examples with truncation and maybe padding, but keep the overflows using a stride. This results
        # in one example possible giving several features when a context is long, each of those features having a
        # context that overlaps a bit the context of the previous feature.
@@ -464,11 +459,6 @@ def main():

    # Validation preprocessing
    def prepare_validation_features(examples):
-        # Some of the questions have lots of whitespace on the left, which is not useful and will make the
-        # truncation of the context fail (the tokenized question will take a lots of space). So we remove that
-        # left whitespace
-        examples[question_column_name] = [q.lstrip() for q in examples[question_column_name]]
-
        # Tokenize our examples with truncation and maybe padding, but keep the overflows using a stride. This results
        # in one example possible giving several features when a context is long, each of those features having a
        # context that overlaps a bit the context of the previous feature.
--- a/examples/pytorch/summarization/requirements.txt
+++ b/examples/pytorch/summarization/requirements.txt
@@ -1,4 +1,3 @@
-accelerate
 datasets >= 1.8.0
 sentencepiece != 0.1.92
 protobuf
--- a/examples/pytorch/summarization/run_summarization.py
+++ b/examples/pytorch/summarization/run_summarization.py
@@ -48,7 +48,7 @@ from transformers.utils.versions import require_version


 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.10.0")
+check_min_version("4.9.0")

 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/summarization/requirements.txt")

@@ -556,15 +556,12 @@ def main():

    # Evaluation
    results = {}
-    max_length = (
-        training_args.generation_max_length
-        if training_args.generation_max_length is not None
-        else data_args.val_max_target_length
-    )
-    num_beams = data_args.num_beams if data_args.num_beams is not None else training_args.generation_num_beams
    if training_args.do_eval:
        logger.info("*** Evaluate ***")
-        metrics = trainer.evaluate(max_length=max_length, num_beams=num_beams, metric_key_prefix="eval")
+
+        metrics = trainer.evaluate(
+            max_length=data_args.val_max_target_length, num_beams=data_args.num_beams, metric_key_prefix="eval"
+        )
        max_eval_samples = data_args.max_eval_samples if data_args.max_eval_samples is not None else len(eval_dataset)
        metrics["eval_samples"] = min(max_eval_samples, len(eval_dataset))

@@ -575,7 +572,10 @@ def main():
        logger.info("*** Predict ***")

        predict_results = trainer.predict(
-            predict_dataset, metric_key_prefix="predict", max_length=max_length, num_beams=num_beams
+            predict_dataset,
+            metric_key_prefix="predict",
+            max_length=data_args.val_max_target_length,
+            num_beams=data_args.num_beams,
        )
        metrics = predict_results.metrics
        max_predict_samples = (
--- a/examples/pytorch/summarization/run_summarization_no_trainer.py
+++ b/examples/pytorch/summarization/run_summarization_no_trainer.py
@@ -29,7 +29,7 @@ import nltk
 import numpy as np
 import torch
 from datasets import load_dataset, load_metric
-from torch.utils.data import DataLoader
+from torch.utils.data.dataloader import DataLoader
 from tqdm.auto import tqdm

 import transformers
--- a/examples/pytorch/text-classification/README.md
+++ b/examples/pytorch/text-classification/README.md
@@ -51,10 +51,10 @@ single Titan RTX was used):

 | Task  | Metric                       | Result      | Training time |
 |-------|------------------------------|-------------|---------------|
-| CoLA  | Matthews corr                | 56.53       | 3:17          |
+| CoLA  | Matthew's corr               | 56.53       | 3:17          |
 | SST-2 | Accuracy                     | 92.32       | 26:06         |
 | MRPC  | F1/Accuracy                  | 88.85/84.07 | 2:21          |
-| STS-B | Pearson/Spearman corr.       | 88.64/88.48 | 2:13          |
+| STS-B | Person/Spearman corr.        | 88.64/88.48 | 2:13          |
 | QQP   | Accuracy/F1                  | 90.71/87.49 | 2:22:26       |
 | MNLI  | Matched acc./Mismatched acc. | 83.91/84.10 | 2:35:23       |
 | QNLI  | Accuracy                     | 90.66       | 40:57         |
@@ -90,10 +90,10 @@ Using mixed precision training usually results in 2x-speedup for training with t

 | Task  | Metric                       | Result      | Training time | Result (FP16) | Training time (FP16) |
 |-------|------------------------------|-------------|---------------|---------------|----------------------|
-| CoLA  | Matthews corr                | 56.53       | 3:17          | 56.78         | 1:41                 |
+| CoLA  | Matthew's corr               | 56.53       | 3:17          | 56.78         | 1:41                 |
 | SST-2 | Accuracy                     | 92.32       | 26:06         | 91.74         | 13:11                |
 | MRPC  | F1/Accuracy                  | 88.85/84.07 | 2:21          | 88.12/83.58   | 1:10                 |
-| STS-B | Pearson/Spearman corr.       | 88.64/88.48 | 2:13          | 88.71/88.55   | 1:08                 |
+| STS-B | Person/Spearman corr.        | 88.64/88.48 | 2:13          | 88.71/88.55   | 1:08                 |
 | QQP   | Accuracy/F1                  | 90.71/87.49 | 2:22:26       | 90.67/87.43   | 1:11:54              |
 | MNLI  | Matched acc./Mismatched acc. | 83.91/84.10 | 2:35:23       | 84.04/84.06   | 1:17:06              |
 | QNLI  | Accuracy                     | 90.66       | 40:57         | 90.96         | 20:16                |
--- a/examples/pytorch/text-classification/run_glue.py
+++ b/examples/pytorch/text-classification/run_glue.py
@@ -47,7 +47,7 @@ from transformers.utils.versions import require_version


 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.10.0")
+check_min_version("4.9.0")

 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/text-classification/requirements.txt")

@@ -380,9 +380,6 @@ def main():
    if label_to_id is not None:
        model.config.label2id = label_to_id
        model.config.id2label = {id: label for label, id in config.label2id.items()}
-    elif data_args.task_name is not None and not is_regression:
-        model.config.label2id = {l: i for i, l in enumerate(label_list)}
-        model.config.id2label = {id: label for label, id in config.label2id.items()}

    if data_args.max_seq_length > tokenizer.model_max_length:
        logger.warning(
--- a/examples/pytorch/text-classification/run_glue_no_trainer.py
+++ b/examples/pytorch/text-classification/run_glue_no_trainer.py
@@ -21,7 +21,7 @@ import random

 import datasets
 from datasets import load_dataset, load_metric
-from torch.utils.data import DataLoader
+from torch.utils.data.dataloader import DataLoader
 from tqdm.auto import tqdm

 import transformers
@@ -288,9 +288,6 @@ def main():
    if label_to_id is not None:
        model.config.label2id = label_to_id
        model.config.id2label = {id: label for label, id in config.label2id.items()}
-    elif args.task_name is not None and not is_regression:
-        model.config.label2id = {l: i for i, l in enumerate(label_list)}
-        model.config.id2label = {id: label for label, id in config.label2id.items()}

    padding = "max_length" if args.pad_to_max_length else False

--- a/examples/pytorch/text-classification/run_xnli.py
+++ b/examples/pytorch/text-classification/run_xnli.py
@@ -47,7 +47,7 @@ from transformers.utils.versions import require_version


 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.10.0")
+check_min_version("4.9.0")

 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/text-classification/requirements.txt")

--- a/examples/pytorch/token-classification/requirements.txt
+++ b/examples/pytorch/token-classification/requirements.txt
@@ -1,4 +1,3 @@
-accelerate
 seqeval
 datasets >= 1.8.0
 torch >= 1.3
--- a/examples/pytorch/token-classification/run_ner.py
+++ b/examples/pytorch/token-classification/run_ner.py
@@ -47,7 +47,7 @@ from transformers.utils.versions import require_version


 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.10.0")
+check_min_version("4.9.0")

 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/token-classification/requirements.txt")

@@ -123,13 +123,6 @@ class DataTrainingArguments:
        default=None,
        metadata={"help": "The number of processes to use for the preprocessing."},
    )
-    max_seq_length: int = field(
-        default=None,
-        metadata={
-            "help": "The maximum total input sequence length after tokenization. If set, sequences longer "
-            "than this will be truncated, sequences shorter will be padded."
-        },
-    )
    pad_to_max_length: bool = field(
        default=False,
        metadata={
@@ -365,7 +358,6 @@ def main():
            examples[text_column_name],
            padding=padding,
            truncation=True,
-            max_length=data_args.max_seq_length,
            # We use this argument because the texts in our dataset are lists of words (with a label for each word).
            is_split_into_words=True,
        )
--- a/examples/pytorch/token-classification/run_ner_no_trainer.py
+++ b/examples/pytorch/token-classification/run_ner_no_trainer.py
@@ -27,7 +27,7 @@ import random
 import datasets
 import torch
 from datasets import ClassLabel, load_dataset, load_metric
-from torch.utils.data import DataLoader
+from torch.utils.data.dataloader import DataLoader
 from tqdm.auto import tqdm

 import transformers
--- a/examples/pytorch/translation/requirements.txt
+++ b/examples/pytorch/translation/requirements.txt
@@ -1,4 +1,3 @@
-accelerate
 datasets >= 1.8.0
 sentencepiece != 0.1.92
 protobuf
--- a/examples/pytorch/translation/run_translation.py
+++ b/examples/pytorch/translation/run_translation.py
@@ -51,7 +51,7 @@ from transformers.utils.versions import require_version


 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.10.0")
+check_min_version("4.9.0")

 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/translation/requirements.txt")

@@ -549,16 +549,12 @@ def main():

    # Evaluation
    results = {}
-    max_length = (
-        training_args.generation_max_length
-        if training_args.generation_max_length is not None
-        else data_args.val_max_target_length
-    )
-    num_beams = data_args.num_beams if data_args.num_beams is not None else training_args.generation_num_beams
    if training_args.do_eval:
        logger.info("*** Evaluate ***")

-        metrics = trainer.evaluate(max_length=max_length, num_beams=num_beams, metric_key_prefix="eval")
+        metrics = trainer.evaluate(
+            max_length=data_args.val_max_target_length, num_beams=data_args.num_beams, metric_key_prefix="eval"
+        )
        max_eval_samples = data_args.max_eval_samples if data_args.max_eval_samples is not None else len(eval_dataset)
        metrics["eval_samples"] = min(max_eval_samples, len(eval_dataset))

@@ -569,7 +565,10 @@ def main():
        logger.info("*** Predict ***")

        predict_results = trainer.predict(
-            predict_dataset, metric_key_prefix="predict", max_length=max_length, num_beams=num_beams
+            predict_dataset,
+            metric_key_prefix="predict",
+            max_length=data_args.val_max_target_length,
+            num_beams=data_args.num_beams,
        )
        metrics = predict_results.metrics
        max_predict_samples = (
--- a/examples/pytorch/translation/run_translation_no_trainer.py
+++ b/examples/pytorch/translation/run_translation_no_trainer.py
@@ -28,7 +28,7 @@ import datasets
 import numpy as np
 import torch
 from datasets import load_dataset, load_metric
-from torch.utils.data import DataLoader
+from torch.utils.data.dataloader import DataLoader
 from tqdm.auto import tqdm

 import transformers
--- a/examples/research_projects/adversarial/utils_hans.py
+++ b/examples/research_projects/adversarial/utils_hans.py
@@ -88,7 +88,7 @@ class InputFeatures:

 if is_torch_available():
    import torch
-    from torch.utils.data import Dataset
+    from torch.utils.data.dataset import Dataset

    class HansDataset(Dataset):
        """
--- a/examples/research_projects/distillation/distiller.py
+++ b/examples/research_projects/distillation/distiller.py
@@ -380,19 +380,21 @@ class Distiller:
        lm_labels: `torch.tensor(bs, seq_length)` - The language modeling labels (mlm labels for MLM and clm labels for CLM).
        """
        if self.mlm:
-            student_outputs = self.student(
+            s_logits, s_hidden_states = self.student(
                input_ids=input_ids, attention_mask=attention_mask
            )  # (bs, seq_length, voc_size)
            with torch.no_grad():
-                teacher_outputs = self.teacher(
+                t_logits, t_hidden_states = self.teacher(
                    input_ids=input_ids, attention_mask=attention_mask
                )  # (bs, seq_length, voc_size)
        else:
-            student_outputs = self.student(input_ids=input_ids, attention_mask=None)  # (bs, seq_length, voc_size)
+            s_logits, _, s_hidden_states = self.student(
+                input_ids=input_ids, attention_mask=None
+            )  # (bs, seq_length, voc_size)
            with torch.no_grad():
-                teacher_outputs = self.teacher(input_ids=input_ids, attention_mask=None)  # (bs, seq_length, voc_size)
-        s_logits, s_hidden_states = student_outputs["logits"], student_outputs["hidden_states"]
-        t_logits, t_hidden_states = teacher_outputs["logits"], teacher_outputs["hidden_states"]
+                t_logits, _, t_hidden_states = self.teacher(
+                    input_ids=input_ids, attention_mask=None
+                )  # (bs, seq_length, voc_size)
        assert s_logits.size() == t_logits.size()

        # https://github.com/peterliht/knowledge-distillation-pytorch/blob/master/model/net.py#L100
--- a/examples/research_projects/distillation/grouped_batch_sampler.py
+++ b/examples/research_projects/distillation/grouped_batch_sampler.py
@@ -19,7 +19,7 @@ import copy
 from collections import defaultdict

 import numpy as np
-from torch.utils.data import BatchSampler, Sampler
+from torch.utils.data.sampler import BatchSampler, Sampler

 from utils import logger

--- a/examples/research_projects/jax-projects/big_bird/evaluate.py
+++ b/examples/research_projects/jax-projects/big_bird/evaluate.py
@@ -110,7 +110,7 @@ def main():
        inputs = tokenizer(
            example["question"],
            example["context"],
-            return_tensors="np",
+            return_tensors="jax",
            max_length=4096,
            padding="max_length",
            truncation=True,
--- a/examples/research_projects/jax-projects/hybrid_clip/modeling_hybrid_clip.py
+++ b/examples/research_projects/jax-projects/hybrid_clip/modeling_hybrid_clip.py
@@ -208,7 +208,6 @@ class FlaxHybridCLIP(FlaxPreTrainedModel):
        attention_mask=None,
        position_ids=None,
        token_type_ids=None,
-        params: dict = None,
        dropout_rng: jax.random.PRNGKey = None,
        train=False,
    ):
@@ -225,7 +224,7 @@ class FlaxHybridCLIP(FlaxPreTrainedModel):
                `What are input IDs? <../glossary.html#input-ids>`__

        Returns:
-            text_features (:obj:`jnp.ndarray` of shape :obj:`(batch_size, output_dim`): The text embeddings
+            text_features (:obj:`jax_xla.DeviceArray` of shape :obj:`(batch_size, output_dim`): The text embeddings
            obtained by applying the projection layer to the pooled output of text model.
        """
        if position_ids is None:
@@ -255,7 +254,7 @@ class FlaxHybridCLIP(FlaxPreTrainedModel):
            return text_features

        return self.module.apply(
-            {"params": params or self.params},
+            {"params": self.params},
            jnp.array(input_ids, dtype="i4"),
            jnp.array(attention_mask, dtype="i4"),
            jnp.array(position_ids, dtype="i4"),
@@ -265,9 +264,7 @@ class FlaxHybridCLIP(FlaxPreTrainedModel):
            rngs=rngs,
        )

-    def get_image_features(
-        self, pixel_values, params: dict = None, dropout_rng: jax.random.PRNGKey = None, train=False
-    ):
+    def get_image_features(self, pixel_values, dropout_rng: jax.random.PRNGKey = None, train=False):
        r"""
        Args:
            pixel_values (:obj:`numpy.ndarray` of shape :obj:`(batch_size, num_channels, height, width)`):
@@ -276,7 +273,7 @@ class FlaxHybridCLIP(FlaxPreTrainedModel):
                :meth:`transformers.ImageFeatureExtractionMixin.__call__` for details.

        Returns:
-            image_features (:obj:`jnp.ndarray` of shape :obj:`(batch_size, output_dim`): The image embeddings
+            image_features (:obj:`jax_xla.DeviceArray` of shape :obj:`(batch_size, output_dim`): The image embeddings
            obtained by applying the projection layer to the pooled output of vision model.
        """

@@ -292,7 +289,7 @@ class FlaxHybridCLIP(FlaxPreTrainedModel):
            return image_features

        return self.module.apply(
-            {"params": params or self.params},
+            {"params": self.params},
            jnp.array(pixel_values, dtype=jnp.float32),
            not train,
            method=_get_features,
--- a/examples/research_projects/lxmert/requirements.txt
+++ b/examples/research_projects/lxmert/requirements.txt
@@ -46,7 +46,7 @@ nbclient==0.5.0
 nbconvert==6.0.1
 nbformat==5.0.7
 nest-asyncio==1.4.0
-notebook==6.4.1
+notebook==6.1.5
 numpy==1.19.2
 opencv-python==4.4.0.42
 packaging==20.3
--- a/examples/research_projects/seq2seq-distillation/_test_seq2seq_examples.py
+++ b/examples/research_projects/seq2seq-distillation/_test_seq2seq_examples.py
@@ -14,10 +14,10 @@ import lightning_base
 from convert_pl_checkpoint_to_hf import convert_pl_to_hf
 from distillation import distill_main
 from finetune import SummarizationModule, main
-from huggingface_hub.hf_api import HfApi
 from parameterized import parameterized
 from run_eval import generate_summaries_or_translations
 from transformers import AutoConfig, AutoModelForSeq2SeqLM
+from transformers.hf_api import HfApi
 from transformers.testing_utils import CaptureStderr, CaptureStdout, TestCasePlus, require_torch_gpu, slow
 from utils import label_smoothed_nll_loss, lmap, load_json

@@ -130,7 +130,7 @@ class TestSummarizationDistiller(TestCasePlus):
    def test_hub_configs(self):
        """I put require_torch_gpu cause I only want this to run with self-scheduled."""

-        model_list = HfApi().list_models()
+        model_list = HfApi().model_list()
        org = "sshleifer"
        model_ids = [x.modelId for x in model_list if x.modelId.startswith(org)]
        allowed_to_be_broken = ["sshleifer/blenderbot-3B", "sshleifer/blenderbot-90M"]
--- a/examples/research_projects/visual_bert/README.md
+++ b/examples/research_projects/visual_bert/README.md
@@ -1,6 +0,0 @@
-# VisualBERT Demo
-
-This demo shows usage of VisualBERT VQA model and is adapted from LXMERT demo present [here](https://github.com/huggingface/transformers/blob/master/examples/research_projects/lxmert/demo.ipynb).
-1. make a virtualenv: ``virtualenv venv`` and activate ``source venv/bin/activate``
-2. install reqs: ``pip install -r ./requirements.txt``
-3. usage is as shown in demo.ipynb
--- a/examples/research_projects/visual_bert/demo.ipynb
+++ b/examples/research_projects/visual_bert/demo.ipynb
--- a/examples/research_projects/visual_bert/extracting_data.py
+++ b/examples/research_projects/visual_bert/extracting_data.py
@@ -1,149 +0,0 @@
-import getopt
-import json
-import os
-
-# import numpy as np
-import sys
-from collections import OrderedDict
-
-import datasets
-import numpy as np
-import torch
-
-from modeling_frcnn import GeneralizedRCNN
-from processing_image import Preprocess
-from utils import Config
-
-
-"""
-USAGE:
-``python extracting_data.py -i <img_dir> -o <dataset_file>.datasets <batch_size>``
-"""
-
-
-TEST = False
-CONFIG = Config.from_pretrained("unc-nlp/frcnn-vg-finetuned")
-DEFAULT_SCHEMA = datasets.Features(
-    OrderedDict(
-        {
-            "attr_ids": datasets.Sequence(length=CONFIG.MAX_DETECTIONS, feature=datasets.Value("float32")),
-            "attr_probs": datasets.Sequence(length=CONFIG.MAX_DETECTIONS, feature=datasets.Value("float32")),
-            "boxes": datasets.Array2D((CONFIG.MAX_DETECTIONS, 4), dtype="float32"),
-            "img_id": datasets.Value("int32"),
-            "obj_ids": datasets.Sequence(length=CONFIG.MAX_DETECTIONS, feature=datasets.Value("float32")),
-            "obj_probs": datasets.Sequence(length=CONFIG.MAX_DETECTIONS, feature=datasets.Value("float32")),
-            "roi_features": datasets.Array2D((CONFIG.MAX_DETECTIONS, 2048), dtype="float32"),
-            "sizes": datasets.Sequence(length=2, feature=datasets.Value("float32")),
-            "preds_per_image": datasets.Value(dtype="int32"),
-        }
-    )
-)
-
-
-class Extract:
-    def __init__(self, argv=sys.argv[1:]):
-        inputdir = None
-        outputfile = None
-        subset_list = None
-        batch_size = 1
-        opts, args = getopt.getopt(argv, "i:o:b:s", ["inputdir=", "outfile=", "batch_size=", "subset_list="])
-        for opt, arg in opts:
-            if opt in ("-i", "--inputdir"):
-                inputdir = arg
-            elif opt in ("-o", "--outfile"):
-                outputfile = arg
-            elif opt in ("-b", "--batch_size"):
-                batch_size = int(arg)
-            elif opt in ("-s", "--subset_list"):
-                subset_list = arg
-
-        assert inputdir is not None  # and os.path.isdir(inputdir), f"{inputdir}"
-        assert outputfile is not None and not os.path.isfile(outputfile), f"{outputfile}"
-        if subset_list is not None:
-            with open(os.path.realpath(subset_list)) as f:
-                self.subset_list = set(map(lambda x: self._vqa_file_split()[0], tryload(f)))
-        else:
-            self.subset_list = None
-
-        self.config = CONFIG
-        if torch.cuda.is_available():
-            self.config.model.device = "cuda"
-        self.inputdir = os.path.realpath(inputdir)
-        self.outputfile = os.path.realpath(outputfile)
-        self.preprocess = Preprocess(self.config)
-        self.model = GeneralizedRCNN.from_pretrained("unc-nlp/frcnn-vg-finetuned", config=self.config)
-        self.batch = batch_size if batch_size != 0 else 1
-        self.schema = DEFAULT_SCHEMA
-
-    def _vqa_file_split(self, file):
-        img_id = int(file.split(".")[0].split("_")[-1])
-        filepath = os.path.join(self.inputdir, file)
-        return (img_id, filepath)
-
-    @property
-    def file_generator(self):
-        batch = []
-        for i, file in enumerate(os.listdir(self.inputdir)):
-            if self.subset_list is not None and i not in self.subset_list:
-                continue
-            batch.append(self._vqa_file_split(file))
-            if len(batch) == self.batch:
-                temp = batch
-                batch = []
-                yield list(map(list, zip(*temp)))
-
-        for i in range(1):
-            yield list(map(list, zip(*batch)))
-
-    def __call__(self):
-        # make writer
-        if not TEST:
-            writer = datasets.ArrowWriter(features=self.schema, path=self.outputfile)
-        # do file generator
-        for i, (img_ids, filepaths) in enumerate(self.file_generator):
-            images, sizes, scales_yx = self.preprocess(filepaths)
-            output_dict = self.model(
-                images,
-                sizes,
-                scales_yx=scales_yx,
-                padding="max_detections",
-                max_detections=self.config.MAX_DETECTIONS,
-                pad_value=0,
-                return_tensors="np",
-                location="cpu",
-            )
-            output_dict["boxes"] = output_dict.pop("normalized_boxes")
-            if not TEST:
-                output_dict["img_id"] = np.array(img_ids)
-                batch = self.schema.encode_batch(output_dict)
-                writer.write_batch(batch)
-            if TEST:
-                break
-            # finalizer the writer
-        if not TEST:
-            num_examples, num_bytes = writer.finalize()
-            print(f"Success! You wrote {num_examples} entry(s) and {num_bytes >> 20} mb")
-
-
-def tryload(stream):
-    try:
-        data = json.load(stream)
-        try:
-            data = list(data.keys())
-        except Exception:
-            data = [d["img_id"] for d in data]
-    except Exception:
-        try:
-            data = eval(stream.read())
-        except Exception:
-            data = stream.read().split("\n")
-    return data
-
-
-if __name__ == "__main__":
-    extract = Extract(sys.argv[1:])
-    extract()
-    if not TEST:
-        dataset = datasets.Dataset.from_file(extract.outputfile)
-        # wala!
-        # print(np.array(dataset[0:2]["roi_features"]).shape)
--- a/examples/research_projects/visual_bert/modeling_frcnn.py
+++ b/examples/research_projects/visual_bert/modeling_frcnn.py
--- a/examples/research_projects/visual_bert/processing_image.py
+++ b/examples/research_projects/visual_bert/processing_image.py
@@ -1,149 +0,0 @@
-"""
- coding=utf-8
- Copyright 2018, Antonio Mendoza Hao Tan, Mohit Bansal
- Adapted From Facebook Inc, Detectron2
-
- Licensed under the Apache License, Version 2.0 (the "License");
- you may not use this file except in compliance with the License.
- You may obtain a copy of the License at
-
-     http://www.apache.org/licenses/LICENSE-2.0
-
- Unless required by applicable law or agreed to in writing, software
- distributed under the License is distributed on an "AS IS" BASIS,
- WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- See the License for the specific language governing permissions and
- limitations under the License.import copy
- """
-import sys
-from typing import Tuple
-
-import numpy as np
-import torch
-from PIL import Image
-from torch import nn
-
-from utils import img_tensorize
-
-
-class ResizeShortestEdge:
-    def __init__(self, short_edge_length, max_size=sys.maxsize):
-        """
-        Args:
-            short_edge_length (list[min, max])
-            max_size (int): maximum allowed longest edge length.
-        """
-        self.interp_method = "bilinear"
-        self.max_size = max_size
-        self.short_edge_length = short_edge_length
-
-    def __call__(self, imgs):
-        img_augs = []
-        for img in imgs:
-            h, w = img.shape[:2]
-            # later: provide list and randomly choose index for resize
-            size = np.random.randint(self.short_edge_length[0], self.short_edge_length[1] + 1)
-            if size == 0:
-                return img
-            scale = size * 1.0 / min(h, w)
-            if h < w:
-                newh, neww = size, scale * w
-            else:
-                newh, neww = scale * h, size
-            if max(newh, neww) > self.max_size:
-                scale = self.max_size * 1.0 / max(newh, neww)
-                newh = newh * scale
-                neww = neww * scale
-            neww = int(neww + 0.5)
-            newh = int(newh + 0.5)
-
-            if img.dtype == np.uint8:
-                pil_image = Image.fromarray(img)
-                pil_image = pil_image.resize((neww, newh), Image.BILINEAR)
-                img = np.asarray(pil_image)
-            else:
-                img = img.permute(2, 0, 1).unsqueeze(0)  # 3, 0, 1)  # hw(c) -> nchw
-                img = nn.functional.interpolate(
-                    img, (newh, neww), mode=self.interp_method, align_corners=False
-                ).squeeze(0)
-            img_augs.append(img)
-
-        return img_augs
-
-
-class Preprocess:
-    def __init__(self, cfg):
-        self.aug = ResizeShortestEdge([cfg.INPUT.MIN_SIZE_TEST, cfg.INPUT.MIN_SIZE_TEST], cfg.INPUT.MAX_SIZE_TEST)
-        self.input_format = cfg.INPUT.FORMAT
-        self.size_divisibility = cfg.SIZE_DIVISIBILITY
-        self.pad_value = cfg.PAD_VALUE
-        self.max_image_size = cfg.INPUT.MAX_SIZE_TEST
-        self.device = cfg.MODEL.DEVICE
-        self.pixel_std = torch.tensor(cfg.MODEL.PIXEL_STD).to(self.device).view(len(cfg.MODEL.PIXEL_STD), 1, 1)
-        self.pixel_mean = torch.tensor(cfg.MODEL.PIXEL_MEAN).to(self.device).view(len(cfg.MODEL.PIXEL_STD), 1, 1)
-        self.normalizer = lambda x: (x - self.pixel_mean) / self.pixel_std
-
-    def pad(self, images):
-        max_size = tuple(max(s) for s in zip(*[img.shape for img in images]))
-        image_sizes = [im.shape[-2:] for im in images]
-        images = [
-            nn.functional.pad(
-                im,
-                [0, max_size[-1] - size[1], 0, max_size[-2] - size[0]],
-                value=self.pad_value,
-            )
-            for size, im in zip(image_sizes, images)
-        ]
-
-        return torch.stack(images), torch.tensor(image_sizes)
-
-    def __call__(self, images, single_image=False):
-        with torch.no_grad():
-            if not isinstance(images, list):
-                images = [images]
-            if single_image:
-                assert len(images) == 1
-            for i in range(len(images)):
-                if isinstance(images[i], torch.Tensor):
-                    images.insert(i, images.pop(i).to(self.device).float())
-                elif not isinstance(images[i], torch.Tensor):
-                    images.insert(
-                        i,
-                        torch.as_tensor(img_tensorize(images.pop(i), input_format=self.input_format))
-                        .to(self.device)
-                        .float(),
-                    )
-            # resize smallest edge
-            raw_sizes = torch.tensor([im.shape[:2] for im in images])
-            images = self.aug(images)
-            # transpose images and convert to torch tensors
-            # images = [torch.as_tensor(i.astype("float32")).permute(2, 0, 1).to(self.device) for i in images]
-            # now normalize before pad to avoid useless arithmetic
-            images = [self.normalizer(x) for x in images]
-            # now pad them to do the following operations
-            images, sizes = self.pad(images)
-            # Normalize
-
-            if self.size_divisibility > 0:
-                raise NotImplementedError()
-            # pad
-            scales_yx = torch.true_divide(raw_sizes, sizes)
-            if single_image:
-                return images[0], sizes[0], scales_yx[0]
-            else:
-                return images, sizes, scales_yx
-
-
-def _scale_box(boxes, scale_yx):
-    boxes[:, 0::2] *= scale_yx[:, 1]
-    boxes[:, 1::2] *= scale_yx[:, 0]
-    return boxes
-
-
-def _clip_box(tensor, box_size: Tuple[int, int]):
-    assert torch.isfinite(tensor).all(), "Box tensor contains infinite or NaN!"
-    h, w = box_size
-    tensor[:, 0].clamp_(min=0, max=w)
-    tensor[:, 1].clamp_(min=0, max=h)
-    tensor[:, 2].clamp_(min=0, max=w)
-    tensor[:, 3].clamp_(min=0, max=h)
--- a/examples/research_projects/visual_bert/requirements.txt
+++ b/examples/research_projects/visual_bert/requirements.txt
@@ -1,98 +0,0 @@
-appdirs==1.4.3
-argon2-cffi==20.1.0
-async-generator==1.10
-attrs==20.2.0
-backcall==0.2.0
-CacheControl==0.12.6
-certifi==2020.6.20
-cffi==1.14.2
-chardet==3.0.4
-click==7.1.2
-colorama==0.4.3
-contextlib2==0.6.0
-cycler==0.10.0
-datasets==1.0.0
-decorator==4.4.2
-defusedxml==0.6.0
-dill==0.3.2
-distlib==0.3.0
-distro==1.4.0
-entrypoints==0.3
-filelock==3.0.12
-future==0.18.2
-html5lib==1.0.1
-idna==2.8
-ipaddr==2.2.0
-ipykernel==5.3.4
-ipython
-ipython-genutils==0.2.0
-ipywidgets==7.5.1
-jedi==0.17.2
-Jinja2>=2.11.3
-joblib==0.16.0
-jsonschema==3.2.0
-jupyter==1.0.0
-jupyter-client==6.1.7
-jupyter-console==6.2.0
-jupyter-core==4.6.3
-jupyterlab-pygments==0.1.1
-kiwisolver==1.2.0
-lockfile==0.12.2
-MarkupSafe==1.1.1
-matplotlib==3.3.1
-mistune==0.8.4
-msgpack==0.6.2
-nbclient==0.5.0
-nbconvert==6.0.1
-nbformat==5.0.7
-nest-asyncio==1.4.0
-notebook==6.1.5
-numpy==1.19.2
-opencv-python==4.4.0.42
-packaging==20.3
-pandas==1.1.2
-pandocfilters==1.4.2
-parso==0.7.1
-pep517==0.8.2
-pexpect==4.8.0
-pickleshare==0.7.5
-Pillow>=8.1.1
-progress==1.5
-prometheus-client==0.8.0
-prompt-toolkit==3.0.7
-ptyprocess==0.6.0
-pyaml==20.4.0
-pyarrow==1.0.1
-pycparser==2.20
-Pygments>=2.7.4
-pyparsing==2.4.6
-pyrsistent==0.16.0
-python-dateutil==2.8.1
-pytoml==0.1.21
-pytz==2020.1
-PyYAML>=5.4
-pyzmq==19.0.2
-qtconsole==4.7.7
-QtPy==1.9.0
-regex==2020.7.14
-requests==2.22.0
-retrying==1.3.3
-sacremoses==0.0.43
-Send2Trash==1.5.0
-sentencepiece==0.1.91
-six==1.14.0
-terminado==0.8.3
-testpath==0.4.4
-tokenizers==0.8.1rc2
-torch==1.6.0
-torchvision==0.7.0
-tornado==6.0.4
-tqdm==4.48.2
-traitlets
-git+https://github.com/huggingface/transformers.git
-urllib3==1.26.5
-wcwidth==0.2.5
-webencodings==0.5.1
-wget==3.2
-widgetsnbextension==3.5.1
-xxhash==2.0.0
--- a/examples/research_projects/visual_bert/utils.py
+++ b/examples/research_projects/visual_bert/utils.py
@@ -1,559 +0,0 @@
-"""
- coding=utf-8
- Copyright 2018, Antonio Mendoza Hao Tan, Mohit Bansal, Huggingface team :)
- Adapted From Facebook Inc, Detectron2
-
- Licensed under the Apache License, Version 2.0 (the "License");
- you may not use this file except in compliance with the License.
- You may obtain a copy of the License at
-
-     http://www.apache.org/licenses/LICENSE-2.0
-
- Unless required by applicable law or agreed to in writing, software
- distributed under the License is distributed on an "AS IS" BASIS,
- WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- See the License for the specific language governing permissions and
- limitations under the License.import copy
- """
-
-import copy
-import fnmatch
-import json
-import os
-import pickle as pkl
-import shutil
-import sys
-import tarfile
-import tempfile
-from collections import OrderedDict
-from contextlib import contextmanager
-from functools import partial
-from hashlib import sha256
-from io import BytesIO
-from pathlib import Path
-from urllib.parse import urlparse
-from zipfile import ZipFile, is_zipfile
-
-import numpy as np
-from PIL import Image
-from tqdm.auto import tqdm
-
-import cv2
-import requests
-import wget
-from filelock import FileLock
-from yaml import Loader, dump, load
-
-
-try:
-    import torch
-
-    _torch_available = True
-except ImportError:
-    _torch_available = False
-
-
-try:
-    from torch.hub import _get_torch_home
-
-    torch_cache_home = _get_torch_home()
-except ImportError:
-    torch_cache_home = os.path.expanduser(
-        os.getenv("TORCH_HOME", os.path.join(os.getenv("XDG_CACHE_HOME", "~/.cache"), "torch"))
-    )
-
-default_cache_path = os.path.join(torch_cache_home, "transformers")
-
-CLOUDFRONT_DISTRIB_PREFIX = "https://cdn.huggingface.co"
-S3_BUCKET_PREFIX = "https://s3.amazonaws.com/models.huggingface.co/bert"
-PATH = "/".join(str(Path(__file__).resolve()).split("/")[:-1])
-CONFIG = os.path.join(PATH, "config.yaml")
-ATTRIBUTES = os.path.join(PATH, "attributes.txt")
-OBJECTS = os.path.join(PATH, "objects.txt")
-PYTORCH_PRETRAINED_BERT_CACHE = os.getenv("PYTORCH_PRETRAINED_BERT_CACHE", default_cache_path)
-PYTORCH_TRANSFORMERS_CACHE = os.getenv("PYTORCH_TRANSFORMERS_CACHE", PYTORCH_PRETRAINED_BERT_CACHE)
-TRANSFORMERS_CACHE = os.getenv("TRANSFORMERS_CACHE", PYTORCH_TRANSFORMERS_CACHE)
-WEIGHTS_NAME = "pytorch_model.bin"
-CONFIG_NAME = "config.yaml"
-
-
-def load_labels(objs=OBJECTS, attrs=ATTRIBUTES):
-    vg_classes = []
-    with open(objs) as f:
-        for object in f.readlines():
-            vg_classes.append(object.split(",")[0].lower().strip())
-
-    vg_attrs = []
-    with open(attrs) as f:
-        for object in f.readlines():
-            vg_attrs.append(object.split(",")[0].lower().strip())
-    return vg_classes, vg_attrs
-
-
-def load_checkpoint(ckp):
-    r = OrderedDict()
-    with open(ckp, "rb") as f:
-        ckp = pkl.load(f)["model"]
-    for k in copy.deepcopy(list(ckp.keys())):
-        v = ckp.pop(k)
-        if isinstance(v, np.ndarray):
-            v = torch.tensor(v)
-        else:
-            assert isinstance(v, torch.tensor), type(v)
-        r[k] = v
-    return r
-
-
-class Config:
-    _pointer = {}
-
-    def __init__(self, dictionary: dict, name: str = "root", level=0):
-        self._name = name
-        self._level = level
-        d = {}
-        for k, v in dictionary.items():
-            if v is None:
-                raise ValueError()
-            k = copy.deepcopy(k)
-            v = copy.deepcopy(v)
-            if isinstance(v, dict):
-                v = Config(v, name=k, level=level + 1)
-            d[k] = v
-            setattr(self, k, v)
-
-        self._pointer = d
-
-    def __repr__(self):
-        return str(list((self._pointer.keys())))
-
-    def __setattr__(self, key, val):
-        self.__dict__[key] = val
-        self.__dict__[key.upper()] = val
-        levels = key.split(".")
-        last_level = len(levels) - 1
-        pointer = self._pointer
-        if len(levels) > 1:
-            for i, l in enumerate(levels):
-                if hasattr(self, l) and isinstance(getattr(self, l), Config):
-                    setattr(getattr(self, l), ".".join(levels[i:]), val)
-                if l == last_level:
-                    pointer[l] = val
-                else:
-                    pointer = pointer[l]
-
-    def to_dict(self):
-        return self._pointer
-
-    def dump_yaml(self, data, file_name):
-        with open(f"{file_name}", "w") as stream:
-            dump(data, stream)
-
-    def dump_json(self, data, file_name):
-        with open(f"{file_name}", "w") as stream:
-            json.dump(data, stream)
-
-    @staticmethod
-    def load_yaml(config):
-        with open(config) as stream:
-            data = load(stream, Loader=Loader)
-        return data
-
-    def __str__(self):
-        t = "    "
-        if self._name != "root":
-            r = f"{t * (self._level-1)}{self._name}:\n"
-        else:
-            r = ""
-        level = self._level
-        for i, (k, v) in enumerate(self._pointer.items()):
-            if isinstance(v, Config):
-                r += f"{t * (self._level)}{v}\n"
-                self._level += 1
-            else:
-                r += f"{t * (self._level)}{k}: {v} ({type(v).__name__})\n"
-            self._level = level
-        return r[:-1]
-
-    @classmethod
-    def from_pretrained(cls, pretrained_model_name_or_path: str, **kwargs):
-        config_dict, kwargs = cls.get_config_dict(pretrained_model_name_or_path, **kwargs)
-        return cls(config_dict)
-
-    @classmethod
-    def get_config_dict(cls, pretrained_model_name_or_path: str, **kwargs):
-
-        cache_dir = kwargs.pop("cache_dir", None)
-        force_download = kwargs.pop("force_download", False)
-        resume_download = kwargs.pop("resume_download", False)
-        proxies = kwargs.pop("proxies", None)
-        local_files_only = kwargs.pop("local_files_only", False)
-
-        if os.path.isdir(pretrained_model_name_or_path):
-            config_file = os.path.join(pretrained_model_name_or_path, CONFIG_NAME)
-        elif os.path.isfile(pretrained_model_name_or_path) or is_remote_url(pretrained_model_name_or_path):
-            config_file = pretrained_model_name_or_path
-        else:
-            config_file = hf_bucket_url(pretrained_model_name_or_path, filename=CONFIG_NAME, use_cdn=False)
-
-        try:
-            # Load from URL or cache if already cached
-            resolved_config_file = cached_path(
-                config_file,
-                cache_dir=cache_dir,
-                force_download=force_download,
-                proxies=proxies,
-                resume_download=resume_download,
-                local_files_only=local_files_only,
-            )
-            # Load config dict
-            if resolved_config_file is None:
-                raise EnvironmentError
-
-            config_file = Config.load_yaml(resolved_config_file)
-
-        except EnvironmentError:
-            msg = "Can't load config for"
-            raise EnvironmentError(msg)
-
-        if resolved_config_file == config_file:
-            print("loading configuration file from path")
-        else:
-            print("loading configuration file cache")
-
-        return Config.load_yaml(resolved_config_file), kwargs
-
-
-# quick compare tensors
-def compare(in_tensor):
-
-    out_tensor = torch.load("dump.pt", map_location=in_tensor.device)
-    n1 = in_tensor.numpy()
-    n2 = out_tensor.numpy()[0]
-    print(n1.shape, n1[0, 0, :5])
-    print(n2.shape, n2[0, 0, :5])
-    assert np.allclose(
-        n1, n2, rtol=0.01, atol=0.1
-    ), f"{sum([1 for x in np.isclose(n1, n2, rtol=0.01, atol=0.1).flatten() if x == False])/len(n1.flatten())*100:.4f} % element-wise mismatch"
-    raise Exception("tensors are all good")
-
-    # Hugging face functions below
-
-
-def is_remote_url(url_or_filename):
-    parsed = urlparse(url_or_filename)
-    return parsed.scheme in ("http", "https")
-
-
-def hf_bucket_url(model_id: str, filename: str, use_cdn=True) -> str:
-    endpoint = CLOUDFRONT_DISTRIB_PREFIX if use_cdn else S3_BUCKET_PREFIX
-    legacy_format = "/" not in model_id
-    if legacy_format:
-        return f"{endpoint}/{model_id}-{filename}"
-    else:
-        return f"{endpoint}/{model_id}/{filename}"
-
-
-def http_get(
-    url,
-    temp_file,
-    proxies=None,
-    resume_size=0,
-    user_agent=None,
-):
-    ua = "python/{}".format(sys.version.split()[0])
-    if _torch_available:
-        ua += "; torch/{}".format(torch.__version__)
-    if isinstance(user_agent, dict):
-        ua += "; " + "; ".join("{}/{}".format(k, v) for k, v in user_agent.items())
-    elif isinstance(user_agent, str):
-        ua += "; " + user_agent
-    headers = {"user-agent": ua}
-    if resume_size > 0:
-        headers["Range"] = "bytes=%d-" % (resume_size,)
-    response = requests.get(url, stream=True, proxies=proxies, headers=headers)
-    if response.status_code == 416:  # Range not satisfiable
-        return
-    content_length = response.headers.get("Content-Length")
-    total = resume_size + int(content_length) if content_length is not None else None
-    progress = tqdm(
-        unit="B",
-        unit_scale=True,
-        total=total,
-        initial=resume_size,
-        desc="Downloading",
-    )
-    for chunk in response.iter_content(chunk_size=1024):
-        if chunk:  # filter out keep-alive new chunks
-            progress.update(len(chunk))
-            temp_file.write(chunk)
-    progress.close()
-
-
-def get_from_cache(
-    url,
-    cache_dir=None,
-    force_download=False,
-    proxies=None,
-    etag_timeout=10,
-    resume_download=False,
-    user_agent=None,
-    local_files_only=False,
-):
-
-    if cache_dir is None:
-        cache_dir = TRANSFORMERS_CACHE
-    if isinstance(cache_dir, Path):
-        cache_dir = str(cache_dir)
-
-    os.makedirs(cache_dir, exist_ok=True)
-
-    etag = None
-    if not local_files_only:
-        try:
-            response = requests.head(url, allow_redirects=True, proxies=proxies, timeout=etag_timeout)
-            if response.status_code == 200:
-                etag = response.headers.get("ETag")
-        except (EnvironmentError, requests.exceptions.Timeout):
-            # etag is already None
-            pass
-
-    filename = url_to_filename(url, etag)
-
-    # get cache path to put the file
-    cache_path = os.path.join(cache_dir, filename)
-
-    # etag is None = we don't have a connection, or url doesn't exist, or is otherwise inaccessible.
-    # try to get the last downloaded one
-    if etag is None:
-        if os.path.exists(cache_path):
-            return cache_path
-        else:
-            matching_files = [
-                file
-                for file in fnmatch.filter(os.listdir(cache_dir), filename + ".*")
-                if not file.endswith(".json") and not file.endswith(".lock")
-            ]
-            if len(matching_files) > 0:
-                return os.path.join(cache_dir, matching_files[-1])
-            else:
-                # If files cannot be found and local_files_only=True,
-                # the models might've been found if local_files_only=False
-                # Notify the user about that
-                if local_files_only:
-                    raise ValueError(
-                        "Cannot find the requested files in the cached path and outgoing traffic has been"
-                        " disabled. To enable model look-ups and downloads online, set 'local_files_only'"
-                        " to False."
-                    )
-                return None
-
-    # From now on, etag is not None.
-    if os.path.exists(cache_path) and not force_download:
-        return cache_path
-
-    # Prevent parallel downloads of the same file with a lock.
-    lock_path = cache_path + ".lock"
-    with FileLock(lock_path):
-
-        # If the download just completed while the lock was activated.
-        if os.path.exists(cache_path) and not force_download:
-            # Even if returning early like here, the lock will be released.
-            return cache_path
-
-        if resume_download:
-            incomplete_path = cache_path + ".incomplete"
-
-            @contextmanager
-            def _resumable_file_manager():
-                with open(incomplete_path, "a+b") as f:
-                    yield f
-
-            temp_file_manager = _resumable_file_manager
-            if os.path.exists(incomplete_path):
-                resume_size = os.stat(incomplete_path).st_size
-            else:
-                resume_size = 0
-        else:
-            temp_file_manager = partial(tempfile.NamedTemporaryFile, dir=cache_dir, delete=False)
-            resume_size = 0
-
-        # Download to temporary file, then copy to cache dir once finished.
-        # Otherwise you get corrupt cache entries if the download gets interrupted.
-        with temp_file_manager() as temp_file:
-            print(
-                "%s not found in cache or force_download set to True, downloading to %s",
-                url,
-                temp_file.name,
-            )
-
-            http_get(
-                url,
-                temp_file,
-                proxies=proxies,
-                resume_size=resume_size,
-                user_agent=user_agent,
-            )
-
-        os.replace(temp_file.name, cache_path)
-
-        meta = {"url": url, "etag": etag}
-        meta_path = cache_path + ".json"
-        with open(meta_path, "w") as meta_file:
-            json.dump(meta, meta_file)
-
-    return cache_path
-
-
-def url_to_filename(url, etag=None):
-
-    url_bytes = url.encode("utf-8")
-    url_hash = sha256(url_bytes)
-    filename = url_hash.hexdigest()
-
-    if etag:
-        etag_bytes = etag.encode("utf-8")
-        etag_hash = sha256(etag_bytes)
-        filename += "." + etag_hash.hexdigest()
-
-    if url.endswith(".h5"):
-        filename += ".h5"
-
-    return filename
-
-
-def cached_path(
-    url_or_filename,
-    cache_dir=None,
-    force_download=False,
-    proxies=None,
-    resume_download=False,
-    user_agent=None,
-    extract_compressed_file=False,
-    force_extract=False,
-    local_files_only=False,
-):
-    if cache_dir is None:
-        cache_dir = TRANSFORMERS_CACHE
-    if isinstance(url_or_filename, Path):
-        url_or_filename = str(url_or_filename)
-    if isinstance(cache_dir, Path):
-        cache_dir = str(cache_dir)
-
-    if is_remote_url(url_or_filename):
-        # URL, so get it from the cache (downloading if necessary)
-        output_path = get_from_cache(
-            url_or_filename,
-            cache_dir=cache_dir,
-            force_download=force_download,
-            proxies=proxies,
-            resume_download=resume_download,
-            user_agent=user_agent,
-            local_files_only=local_files_only,
-        )
-    elif os.path.exists(url_or_filename):
-        # File, and it exists.
-        output_path = url_or_filename
-    elif urlparse(url_or_filename).scheme == "":
-        # File, but it doesn't exist.
-        raise EnvironmentError("file {} not found".format(url_or_filename))
-    else:
-        # Something unknown
-        raise ValueError("unable to parse {} as a URL or as a local path".format(url_or_filename))
-
-    if extract_compressed_file:
-        if not is_zipfile(output_path) and not tarfile.is_tarfile(output_path):
-            return output_path
-
-        # Path where we extract compressed archives
-        # We avoid '.' in dir name and add "-extracted" at the end: "./model.zip" => "./model-zip-extracted/"
-        output_dir, output_file = os.path.split(output_path)
-        output_extract_dir_name = output_file.replace(".", "-") + "-extracted"
-        output_path_extracted = os.path.join(output_dir, output_extract_dir_name)
-
-        if os.path.isdir(output_path_extracted) and os.listdir(output_path_extracted) and not force_extract:
-            return output_path_extracted
-
-        # Prevent parallel extractions
-        lock_path = output_path + ".lock"
-        with FileLock(lock_path):
-            shutil.rmtree(output_path_extracted, ignore_errors=True)
-            os.makedirs(output_path_extracted)
-            if is_zipfile(output_path):
-                with ZipFile(output_path, "r") as zip_file:
-                    zip_file.extractall(output_path_extracted)
-                    zip_file.close()
-            elif tarfile.is_tarfile(output_path):
-                tar_file = tarfile.open(output_path)
-                tar_file.extractall(output_path_extracted)
-                tar_file.close()
-            else:
-                raise EnvironmentError("Archive format of {} could not be identified".format(output_path))
-
-        return output_path_extracted
-
-    return output_path
-
-
-def get_data(query, delim=","):
-    assert isinstance(query, str)
-    if os.path.isfile(query):
-        with open(query) as f:
-            data = eval(f.read())
-    else:
-        req = requests.get(query)
-        try:
-            data = requests.json()
-        except Exception:
-            data = req.content.decode()
-            assert data is not None, "could not connect"
-            try:
-                data = eval(data)
-            except Exception:
-                data = data.split("\n")
-        req.close()
-    return data
-
-
-def get_image_from_url(url):
-    response = requests.get(url)
-    img = np.array(Image.open(BytesIO(response.content)))
-    return img
-
-
-# to load legacy frcnn checkpoint from detectron
-def load_frcnn_pkl_from_url(url):
-    fn = url.split("/")[-1]
-    if fn not in os.listdir(os.getcwd()):
-        wget.download(url)
-    with open(fn, "rb") as stream:
-        weights = pkl.load(stream)
-    model = weights.pop("model")
-    new = {}
-    for k, v in model.items():
-        new[k] = torch.from_numpy(v)
-        if "running_var" in k:
-            zero = torch.tensor([0])
-            k2 = k.replace("running_var", "num_batches_tracked")
-            new[k2] = zero
-    return new
-
-
-def get_demo_path():
-    print(f"{os.path.abspath(os.path.join(PATH, os.pardir))}/demo.ipynb")
-
-
-def img_tensorize(im, input_format="RGB"):
-    assert isinstance(im, str)
-    if os.path.isfile(im):
-        img = cv2.imread(im)
-    else:
-        img = get_image_from_url(im)
-        assert img is not None, f"could not connect to: {im}"
-    img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
-    if input_format == "RGB":
-        img = img[:, :, ::-1]
-    return img
-
-
-def chunk(images, batch=1):
-    return (images[i : i + batch] for i in range(0, len(images), batch))
--- a/examples/research_projects/visual_bert/visualizing_image.py
+++ b/examples/research_projects/visual_bert/visualizing_image.py
@@ -1,499 +0,0 @@
-"""
- coding=utf-8
- Copyright 2018, Antonio Mendoza Hao Tan, Mohit Bansal
- Adapted From Facebook Inc, Detectron2
-
- Licensed under the Apache License, Version 2.0 (the "License");
- you may not use this file except in compliance with the License.
- You may obtain a copy of the License at
-
-     http://www.apache.org/licenses/LICENSE-2.0
-
- Unless required by applicable law or agreed to in writing, software
- distributed under the License is distributed on an "AS IS" BASIS,
- WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- See the License for the specific language governing permissions and
- limitations under the License.import copy
- """
-import colorsys
-import io
-
-import matplotlib as mpl
-import matplotlib.colors as mplc
-import matplotlib.figure as mplfigure
-import numpy as np
-import torch
-from matplotlib.backends.backend_agg import FigureCanvasAgg
-
-import cv2
-from utils import img_tensorize
-
-
-_SMALL_OBJ = 1000
-
-
-class SingleImageViz:
-    def __init__(
-        self,
-        img,
-        scale=1.2,
-        edgecolor="g",
-        alpha=0.5,
-        linestyle="-",
-        saveas="test_out.jpg",
-        rgb=True,
-        pynb=False,
-        id2obj=None,
-        id2attr=None,
-        pad=0.7,
-    ):
-        """
-        img: an RGB image of shape (H, W, 3).
-        """
-        if isinstance(img, torch.Tensor):
-            img = img.numpy().astype("np.uint8")
-        if isinstance(img, str):
-            img = img_tensorize(img)
-        assert isinstance(img, np.ndarray)
-
-        width, height = img.shape[1], img.shape[0]
-        fig = mplfigure.Figure(frameon=False)
-        dpi = fig.get_dpi()
-        width_in = (width * scale + 1e-2) / dpi
-        height_in = (height * scale + 1e-2) / dpi
-        fig.set_size_inches(width_in, height_in)
-        ax = fig.add_axes([0.0, 0.0, 1.0, 1.0])
-        ax.axis("off")
-        ax.set_xlim(0.0, width)
-        ax.set_ylim(height)
-
-        self.saveas = saveas
-        self.rgb = rgb
-        self.pynb = pynb
-        self.img = img
-        self.edgecolor = edgecolor
-        self.alpha = 0.5
-        self.linestyle = linestyle
-        self.font_size = int(np.sqrt(min(height, width)) * scale // 3)
-        self.width = width
-        self.height = height
-        self.scale = scale
-        self.fig = fig
-        self.ax = ax
-        self.pad = pad
-        self.id2obj = id2obj
-        self.id2attr = id2attr
-        self.canvas = FigureCanvasAgg(fig)
-
-    def add_box(self, box, color=None):
-        if color is None:
-            color = self.edgecolor
-        (x0, y0, x1, y1) = box
-        width = x1 - x0
-        height = y1 - y0
-        self.ax.add_patch(
-            mpl.patches.Rectangle(
-                (x0, y0),
-                width,
-                height,
-                fill=False,
-                edgecolor=color,
-                linewidth=self.font_size // 3,
-                alpha=self.alpha,
-                linestyle=self.linestyle,
-            )
-        )
-
-    def draw_boxes(self, boxes, obj_ids=None, obj_scores=None, attr_ids=None, attr_scores=None):
-        if len(boxes.shape) > 2:
-            boxes = boxes[0]
-        if len(obj_ids.shape) > 1:
-            obj_ids = obj_ids[0]
-        if len(obj_scores.shape) > 1:
-            obj_scores = obj_scores[0]
-        if len(attr_ids.shape) > 1:
-            attr_ids = attr_ids[0]
-        if len(attr_scores.shape) > 1:
-            attr_scores = attr_scores[0]
-        if isinstance(boxes, torch.Tensor):
-            boxes = boxes.numpy()
-        if isinstance(boxes, list):
-            boxes = np.array(boxes)
-        assert isinstance(boxes, np.ndarray)
-        areas = np.prod(boxes[:, 2:] - boxes[:, :2], axis=1)
-        sorted_idxs = np.argsort(-areas).tolist()
-        boxes = boxes[sorted_idxs] if boxes is not None else None
-        obj_ids = obj_ids[sorted_idxs] if obj_ids is not None else None
-        obj_scores = obj_scores[sorted_idxs] if obj_scores is not None else None
-        attr_ids = attr_ids[sorted_idxs] if attr_ids is not None else None
-        attr_scores = attr_scores[sorted_idxs] if attr_scores is not None else None
-
-        assigned_colors = [self._random_color(maximum=1) for _ in range(len(boxes))]
-        assigned_colors = [assigned_colors[idx] for idx in sorted_idxs]
-        if obj_ids is not None:
-            labels = self._create_text_labels_attr(obj_ids, obj_scores, attr_ids, attr_scores)
-            for i in range(len(boxes)):
-                color = assigned_colors[i]
-                self.add_box(boxes[i], color)
-                self.draw_labels(labels[i], boxes[i], color)
-
-    def draw_labels(self, label, box, color):
-        x0, y0, x1, y1 = box
-        text_pos = (x0, y0)
-        instance_area = (y1 - y0) * (x1 - x0)
-        small = _SMALL_OBJ * self.scale
-        if instance_area < small or y1 - y0 < 40 * self.scale:
-            if y1 >= self.height - 5:
-                text_pos = (x1, y0)
-            else:
-                text_pos = (x0, y1)
-
-        height_ratio = (y1 - y0) / np.sqrt(self.height * self.width)
-        lighter_color = self._change_color_brightness(color, brightness_factor=0.7)
-        font_size = np.clip((height_ratio - 0.02) / 0.08 + 1, 1.2, 2)
-        font_size *= 0.75 * self.font_size
-
-        self.draw_text(
-            text=label,
-            position=text_pos,
-            color=lighter_color,
-        )
-
-    def draw_text(
-        self,
-        text,
-        position,
-        color="g",
-        ha="left",
-    ):
-        rotation = 0
-        font_size = self.font_size
-        color = np.maximum(list(mplc.to_rgb(color)), 0.2)
-        color[np.argmax(color)] = max(0.8, np.max(color))
-        bbox = {
-            "facecolor": "black",
-            "alpha": self.alpha,
-            "pad": self.pad,
-            "edgecolor": "none",
-        }
-        x, y = position
-        self.ax.text(
-            x,
-            y,
-            text,
-            size=font_size * self.scale,
-            family="sans-serif",
-            bbox=bbox,
-            verticalalignment="top",
-            horizontalalignment=ha,
-            color=color,
-            zorder=10,
-            rotation=rotation,
-        )
-
-    def save(self, saveas=None):
-        if saveas is None:
-            saveas = self.saveas
-        if saveas.lower().endswith(".jpg") or saveas.lower().endswith(".png"):
-            cv2.imwrite(
-                saveas,
-                self._get_buffer()[:, :, ::-1],
-            )
-        else:
-            self.fig.savefig(saveas)
-
-    def _create_text_labels_attr(self, classes, scores, attr_classes, attr_scores):
-        labels = [self.id2obj[i] for i in classes]
-        attr_labels = [self.id2attr[i] for i in attr_classes]
-        labels = [
-            f"{label} {score:.2f} {attr} {attr_score:.2f}"
-            for label, score, attr, attr_score in zip(labels, scores, attr_labels, attr_scores)
-        ]
-        return labels
-
-    def _create_text_labels(self, classes, scores):
-        labels = [self.id2obj[i] for i in classes]
-        if scores is not None:
-            if labels is None:
-                labels = ["{:.0f}%".format(s * 100) for s in scores]
-            else:
-                labels = ["{} {:.0f}%".format(li, s * 100) for li, s in zip(labels, scores)]
-        return labels
-
-    def _random_color(self, maximum=255):
-        idx = np.random.randint(0, len(_COLORS))
-        ret = _COLORS[idx] * maximum
-        if not self.rgb:
-            ret = ret[::-1]
-        return ret
-
-    def _get_buffer(self):
-        if not self.pynb:
-            s, (width, height) = self.canvas.print_to_buffer()
-            if (width, height) != (self.width, self.height):
-                img = cv2.resize(self.img, (width, height))
-            else:
-                img = self.img
-        else:
-            buf = io.BytesIO()  # works for cairo backend
-            self.canvas.print_rgba(buf)
-            width, height = self.width, self.height
-            s = buf.getvalue()
-            img = self.img
-
-        buffer = np.frombuffer(s, dtype="uint8")
-        img_rgba = buffer.reshape(height, width, 4)
-        rgb, alpha = np.split(img_rgba, [3], axis=2)
-
-        try:
-            import numexpr as ne  # fuse them with numexpr
-
-            visualized_image = ne.evaluate("img * (1 - alpha / 255.0) + rgb * (alpha / 255.0)")
-        except ImportError:
-            alpha = alpha.astype("float32") / 255.0
-            visualized_image = img * (1 - alpha) + rgb * alpha
-
-        return visualized_image.astype("uint8")
-
-    def _change_color_brightness(self, color, brightness_factor):
-        assert brightness_factor >= -1.0 and brightness_factor <= 1.0
-        color = mplc.to_rgb(color)
-        polygon_color = colorsys.rgb_to_hls(*mplc.to_rgb(color))
-        modified_lightness = polygon_color[1] + (brightness_factor * polygon_color[1])
-        modified_lightness = 0.0 if modified_lightness < 0.0 else modified_lightness
-        modified_lightness = 1.0 if modified_lightness > 1.0 else modified_lightness
-        modified_color = colorsys.hls_to_rgb(polygon_color[0], modified_lightness, polygon_color[2])
-        return modified_color
-
-
-# Color map
-_COLORS = (
-    np.array(
-        [
-            0.000,
-            0.447,
-            0.741,
-            0.850,
-            0.325,
-            0.098,
-            0.929,
-            0.694,
-            0.125,
-            0.494,
-            0.184,
-            0.556,
-            0.466,
-            0.674,
-            0.188,
-            0.301,
-            0.745,
-            0.933,
-            0.635,
-            0.078,
-            0.184,
-            0.300,
-            0.300,
-            0.300,
-            0.600,
-            0.600,
-            0.600,
-            1.000,
-            0.000,
-            0.000,
-            1.000,
-            0.500,
-            0.000,
-            0.749,
-            0.749,
-            0.000,
-            0.000,
-            1.000,
-            0.000,
-            0.000,
-            0.000,
-            1.000,
-            0.667,
-            0.000,
-            1.000,
-            0.333,
-            0.333,
-            0.000,
-            0.333,
-            0.667,
-            0.000,
-            0.333,
-            1.000,
-            0.000,
-            0.667,
-            0.333,
-            0.000,
-            0.667,
-            0.667,
-            0.000,
-            0.667,
-            1.000,
-            0.000,
-            1.000,
-            0.333,
-            0.000,
-            1.000,
-            0.667,
-            0.000,
-            1.000,
-            1.000,
-            0.000,
-            0.000,
-            0.333,
-            0.500,
-            0.000,
-            0.667,
-            0.500,
-            0.000,
-            1.000,
-            0.500,
-            0.333,
-            0.000,
-            0.500,
-            0.333,
-            0.333,
-            0.500,
-            0.333,
-            0.667,
-            0.500,
-            0.333,
-            1.000,
-            0.500,
-            0.667,
-            0.000,
-            0.500,
-            0.667,
-            0.333,
-            0.500,
-            0.667,
-            0.667,
-            0.500,
-            0.667,
-            1.000,
-            0.500,
-            1.000,
-            0.000,
-            0.500,
-            1.000,
-            0.333,
-            0.500,
-            1.000,
-            0.667,
-            0.500,
-            1.000,
-            1.000,
-            0.500,
-            0.000,
-            0.333,
-            1.000,
-            0.000,
-            0.667,
-            1.000,
-            0.000,
-            1.000,
-            1.000,
-            0.333,
-            0.000,
-            1.000,
-            0.333,
-            0.333,
-            1.000,
-            0.333,
-            0.667,
-            1.000,
-            0.333,
-            1.000,
-            1.000,
-            0.667,
-            0.000,
-            1.000,
-            0.667,
-            0.333,
-            1.000,
-            0.667,
-            0.667,
-            1.000,
-            0.667,
-            1.000,
-            1.000,
-            1.000,
-            0.000,
-            1.000,
-            1.000,
-            0.333,
-            1.000,
-            1.000,
-            0.667,
-            1.000,
-            0.333,
-            0.000,
-            0.000,
-            0.500,
-            0.000,
-            0.000,
-            0.667,
-            0.000,
-            0.000,
-            0.833,
-            0.000,
-            0.000,
-            1.000,
-            0.000,
-            0.000,
-            0.000,
-            0.167,
-            0.000,
-            0.000,
-            0.333,
-            0.000,
-            0.000,
-            0.500,
-            0.000,
-            0.000,
-            0.667,
-            0.000,
-            0.000,
-            0.833,
-            0.000,
-            0.000,
-            1.000,
-            0.000,
-            0.000,
-            0.000,
-            0.167,
-            0.000,
-            0.000,
-            0.333,
-            0.000,
-            0.000,
-            0.500,
-            0.000,
-            0.000,
-            0.667,
-            0.000,
-            0.000,
-            0.833,
-            0.000,
-            0.000,
-            1.000,
-            0.000,
-            0.000,
-            0.000,
-            0.143,
-            0.143,
-            0.143,
-            0.857,
-            0.857,
-            0.857,
-            1.000,
-            1.000,
-            1.000,
-        ]
-    )
-    .astype(np.float32)
-    .reshape(-1, 3)
-)
--- a/examples/research_projects/wav2vec2/README.md
+++ b/examples/research_projects/wav2vec2/README.md
@@ -193,7 +193,7 @@ It is recommended to pre-train Wav2Vec2 with Trainer + Deepspeed (please refer t
 Here is an example of how you can use DeepSpeed ZeRO-2 to pretrain a small Wav2Vec2 model:

 ```
-PYTHONPATH=../../../src deepspeed --num_gpus 4 run_pretrain.py \
+PYTHONPATH=../../../src deepspeed --num_gpus 2 run_pretrain.py \
 --output_dir="./wav2vec2-base-libri-100h" \
 --num_train_epochs="3" \
 --per_device_train_batch_size="32" \
--- a/examples/research_projects/wav2vec2/run_asr.py
+++ b/examples/research_projects/wav2vec2/run_asr.py
@@ -55,10 +55,7 @@ class ModelArguments:
        default=True, metadata={"help": "Whether to freeze the feature extractor layers of the model."}
    )
    gradient_checkpointing: Optional[bool] = field(
-        default=False,
-        metadata={
-            "help": "If True, use gradient checkpointing to save memory at the expense of slower backward pass."
-        },
+        default=False, metadata={"help": "Whether to freeze the feature extractor layers of the model."}
    )
    verbose_logging: Optional[bool] = field(
        default=False,
--- a/examples/research_projects/wav2vec2/run_pretrain.py
+++ b/examples/research_projects/wav2vec2/run_pretrain.py
@@ -51,10 +51,7 @@ class ModelArguments:
        default=True, metadata={"help": "Whether to freeze the feature extractor layers of the model."}
    )
    gradient_checkpointing: Optional[bool] = field(
-        default=False,
-        metadata={
-            "help": "If True, use gradient checkpointing to save memory at the expense of slower backward pass."
-        },
+        default=False, metadata={"help": "Whether to freeze the feature extractor layers of the model."}
    )
    verbose_logging: Optional[bool] = field(
        default=False,
--- a/examples/research_projects/wav2vec2/test_wav2vec2_deepspeed.py
+++ b/examples/research_projects/wav2vec2/test_wav2vec2_deepspeed.py
@@ -171,6 +171,7 @@ class TestDeepSpeedWav2Vec2(TestCasePlus):
            --group_by_length
            --freeze_feature_extractor
            --report_to none
+            --logging_steps 0
            --save_steps 0
            --eval_steps {eval_steps}
            --report_to none
--- a/examples/tensorflow/language-modeling/run_clm.py
+++ b/examples/tensorflow/language-modeling/run_clm.py
@@ -43,7 +43,7 @@ import transformers
 from transformers import (
    CONFIG_MAPPING,
    CONFIG_NAME,
-    MODEL_FOR_CAUSAL_LM_MAPPING,
+    MODEL_FOR_MASKED_LM_MAPPING,
    TF2_WEIGHTS_NAME,
    AutoConfig,
    AutoTokenizer,
@@ -58,7 +58,7 @@ from transformers.utils.versions import require_version

 logger = logging.getLogger(__name__)
 require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/language-modeling/requirements.txt")
-MODEL_CONFIG_CLASSES = list(MODEL_FOR_CAUSAL_LM_MAPPING.keys())
+MODEL_CONFIG_CLASSES = list(MODEL_FOR_MASKED_LM_MAPPING.keys())
 MODEL_TYPES = tuple(conf.model_type for conf in MODEL_CONFIG_CLASSES)
 # endregion

@@ -186,9 +186,6 @@ class DataTrainingArguments:
            "value if set."
        },
    )
-    keep_linebreaks: bool = field(
-        default=True, metadata={"help": "Whether to keep line breaks when using TXT files or not."}
-    )

    def __post_init__(self):
        if self.dataset_name is None and self.train_file is None and self.validation_file is None:
@@ -321,7 +318,6 @@ def main():
            )
    else:
        data_files = {}
-        dataset_args = {}
        if data_args.train_file is not None:
            data_files["train"] = data_args.train_file
        if data_args.validation_file is not None:
@@ -329,8 +325,7 @@ def main():
        extension = data_args.train_file.split(".")[-1]
        if extension == "txt":
            extension = "text"
-            dataset_args["keep_linebreaks"] = data_args.keep_linebreaks
-        raw_datasets = load_dataset(extension, data_files=data_files, **dataset_args)
+        raw_datasets = load_dataset(extension, data_files=data_files)
    # See more about loading any type of standard or custom dataset (from files, python dict, pandas DataFrame, etc) at
    # https://huggingface.co/docs/datasets/loading_datasets.html.
    # endregion
@@ -443,7 +438,7 @@ def main():
            f"Validation file not found: using {data_args.validation_split_percentage}% of the dataset as validation as provided in data_args"
        )
        train_indices, val_indices = train_test_split(
-            list(range(len(train_dataset))), test_size=data_args.validation_split_percentage / 100
+            list(range(len(train_dataset))), test_size=data_args.validation_split_percentage
        )

        eval_dataset = train_dataset.select(val_indices)
--- a/examples/tensorflow/language-modeling/run_mlm.py
+++ b/examples/tensorflow/language-modeling/run_mlm.py
@@ -499,7 +499,7 @@ def main():
            f"Validation file not found: using {data_args.validation_split_percentage}% of the dataset as validation as provided in data_args"
        )
        train_indices, val_indices = train_test_split(
-            list(range(len(train_dataset))), test_size=data_args.validation_split_percentage / 100
+            list(range(len(train_dataset))), test_size=data_args.validation_split_percentage
        )

        eval_dataset = train_dataset.select(val_indices)
--- a/examples/tensorflow/multiple-choice/README.md
+++ b/examples/tensorflow/multiple-choice/README.md
@@ -1,5 +1,5 @@
 <!---
-Copyright 2021 The HuggingFace Team. All rights reserved.
+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.
@@ -13,31 +13,26 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License.
 -->
-# Multiple-choice training (e.g. SWAG)

-This folder contains the `run_swag.py` script, showing an examples of *multiple-choice answering* with the 
-🤗 Transformers library. For straightforward use-cases you may be able to use these scripts without modification, 
-although we have also included comments in the code to indicate areas that you may need to adapt to your own projects.
+# Multiple Choice

-### Multi-GPU and TPU usage
+## Fine-tuning on SWAG

-By default, the script uses a `MirroredStrategy` and will use multiple GPUs effectively if they are available. TPUs
-can also be used by passing the name of the TPU resource with the `--tpu` argument.
-
-### Memory usage and data loading
-
-One thing to note is that all data is loaded into memory in this script. Most multiple-choice datasets are small
-enough that this is not an issue, but if you have a very large dataset you will need to modify the script to handle
-data streaming. This is particularly challenging for TPUs, given the stricter requirements and the sheer volume of data
-required to keep them fed. A full explanation of all the possible pitfalls is a bit beyond this example script and 
-README, but for more information you can see the 'Input Datasets' section of 
-[this document](https://www.tensorflow.org/guide/tpu).
-
-### Example command
 ```bash
-python run_swag.py \
- --model_name_or_path distilbert-base-cased \
- --output_dir output \
- --do_eval \
- --do_train
+export SWAG_DIR=/path/to/swag_data_dir
+python ./examples/multiple-choice/run_tf_multiple_choice.py \
+--task_name swag \
+--model_name_or_path bert-base-cased \
+--do_train \
+--do_eval \
+--data_dir $SWAG_DIR \
+--learning_rate 5e-5 \
+--num_train_epochs 3 \
+--max_seq_length 80 \
+--output_dir models_bert/swag_base \
+--per_gpu_eval_batch_size=16 \
+--per_device_train_batch_size=16 \
+--logging-dir logs \
+--gradient_accumulation_steps 2 \
+--overwrite_output
 ```
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Lysandre	41981a25cd	Patch release: v4.9.2 Some checks failed Release - Conda / build_and_package (push) Has been cancelled Details	2021-08-09 16:01:36 +02:00
Sylvain Gugger	ec784223ea	Tpu tie weights (#13030 ) * Fix tied weights on TPU * Manually tie weights in no trainer examples * Fix for test * One last missing * Gettning owned by my scripts * Address review comments * Fix test * Fix tests * Fix reformer tests	2021-08-09 15:53:05 +02:00
Lysandre Debut	bfd53549b0	Add to ONNX docs (#13048 ) * Add to ONNX docs * Add MBART example * Update docs/source/serialization.rst Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com> Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>	2021-08-09 15:52:16 +02:00
Lysandre Debut	226763a262	Add MBART to models exportable with ONNX (#13049 ) * Add MBART to models exportable with ONNX * unittest mock * Add tests * Misc fixes	2021-08-09 15:52:07 +02:00
Lysandre Debut	f595ea33d9	Put smaller ALBERT model (#13028 )	2021-08-09 15:51:04 +02:00
Michael Benayoun	a12fa50693	T5 with past ONNX export (#13014 ) T5 with past ONNX export, and more explicit past_key_values inputs and outputs names for ONNX model Authored-by: Michael Benayoun <michael@huggingface.co>	2021-08-09 15:50:58 +02:00
Michael Benayoun	94b7db97bf	GPT-Neo ONNX export (#12911 ) GPT-Neo ONNX export and task / feature refactoring Authored-by: Michael Benayoun <michael@huggingface.co>	2021-08-09 15:50:44 +02:00
Sylvain Gugger	2c255a2e0c	Fix push_to_hub for TPUs (#12895 )	2021-08-09 15:47:29 +02:00
Funtowicz Morgan	ca272fc523	ONNX v2 raises an Exception when using PyTorch < 1.8.0 (#12933 ) * Raise an issue if the pytorch version is < 1.8.0 * Attempt to add a test to ensure it correctly raises. * Missing docstring. * Second attempt, patch with string absolute import. * Let's do the call before checking it was called ... * use the correct function ... 🤦 * Raise ImportError and AssertionError respectively when unable to find torch and torch version is not sufficient. * Correct path mock patching * relax constraint for torch_onnx_dict_inputs to ge instead of eq. * Style. * Split each version requirements for torch. * Let's compare version directly. * Import torch_version after checking pytorch is installed. * @require_torch	2021-08-09 15:43:31 +02:00
Sylvain Gugger	bff1c71e84	Release: v4.9.1 Some checks failed Release - Conda / build_and_package (push) Has been cancelled Details	2021-07-26 10:21:55 -04:00
Sylvain Gugger	8ee16d84ce	Fix barrier for SM distributed (#12853 )	2021-07-26 10:21:07 -04:00
Sylvain Gugger	6cab8b32e3	Add doc for v4.9.0	2021-07-26 10:20:51 -04:00