Release: v3.5.0

* Update RST

* Finer details

* Re-organize

* Style

.circleci/config.yml

@@ -77,16 +77,17 @@ jobs:
                       - v0.4-torch_and_tf-{{ checksum "setup.py" }}
                       - v0.4-{{ checksum "setup.py" }}
             - run: pip install --upgrade pip
-            - run: pip install git+https://github.com/huggingface/datasets
             - run: pip install .[sklearn,tf-cpu,torch,testing]
             - save_cache:
                 key: v0.4-{{ checksum "setup.py" }}
                 paths:
                     - '~/.cache/pip'
-            - run: RUN_PT_TF_CROSS_TESTS=1 python -m pytest -n 8 --dist=loadfile -rA -s ./tests/ -m is_pt_tf_cross_test --durations=0 | tee output.txt
+            - run: RUN_PT_TF_CROSS_TESTS=1 python -m pytest -n 8 --dist=loadfile -rA -s --make-reports=tests_torch_and_tf ./tests/ -m is_pt_tf_cross_test --durations=0 | tee tests_output.txt
             - store_artifacts:
-                  path: ~/transformers/output.txt
-                  destination: test_output.txt
+                  path: ~/transformers/tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
+
     run_tests_torch:
         working_directory: ~/transformers
         docker:
@@ -102,13 +103,12 @@ jobs:
                       - v0.4-torch-{{ checksum "setup.py" }}
                       - v0.4-{{ checksum "setup.py" }}
             - run: pip install --upgrade pip
-            - run: pip install git+https://github.com/huggingface/datasets
             - run: pip install .[sklearn,torch,testing]
             - save_cache:
                   key: v0.4-torch-{{ checksum "setup.py" }}
                   paths:
                       - '~/.cache/pip'
-            - run: python -m pytest -n 8 --dist=loadfile -s --make_reports=tests ./tests/ | tee tests_output.txt
+            - run: python -m pytest -n 8 --dist=loadfile -s --make-reports=tests_torch ./tests/ | tee tests_output.txt
             - store_artifacts:
                   path: ~/transformers/tests_output.txt
             - store_artifacts:
@@ -129,16 +129,17 @@ jobs:
                       - v0.4-tf-{{ checksum "setup.py" }}
                       - v0.4-{{ checksum "setup.py" }}
             - run: pip install --upgrade pip
-            - run: pip install git+https://github.com/huggingface/datasets
             - run: pip install .[sklearn,tf-cpu,testing]
             - save_cache:
                   key: v0.4-tf-{{ checksum "setup.py" }}
                   paths:
                       - '~/.cache/pip'
-            - run: python -m pytest -n 8 --dist=loadfile -rA -s ./tests/ | tee output.txt
+            - run: python -m pytest -n 8 --dist=loadfile -rA -s --make-reports=tests_tf ./tests/ | tee tests_output.txt
             - store_artifacts:
-               path: ~/transformers/output.txt
-               destination: test_output.txt
+                  path: ~/transformers/tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
+
     run_tests_flax:
         working_directory: ~/transformers
         docker:
@@ -154,16 +155,17 @@ jobs:
                     - v0.4-flax-{{ checksum "setup.py" }}
                     - v0.4-{{ checksum "setup.py" }}
             - run: pip install --upgrade pip
-            - run: pip install git+https://github.com/huggingface/datasets
             - run: sudo pip install .[flax,sklearn,torch,testing]
             - save_cache:
                   key: v0.4-flax-{{ checksum "setup.py" }}
                   paths:
                       - '~/.cache/pip'
-            - run: python -m pytest -n 8 --dist=loadfile -rA -s ./tests/ | tee output.txt
+            - run: python -m pytest -n 8 --dist=loadfile -rA -s --make-reports=tests_flax ./tests/ | tee tests_output.txt
             - store_artifacts:
-                  path: ~/transformers/output.txt
-                  destination: test_output.txt
+                  path: ~/transformers/tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
+
     run_tests_pipelines_torch:
         working_directory: ~/transformers
         docker:
@@ -179,16 +181,17 @@ jobs:
                       - v0.4-torch-{{ checksum "setup.py" }}
                       - v0.4-{{ checksum "setup.py" }}
             - run: pip install --upgrade pip
-            - run: pip install git+https://github.com/huggingface/datasets
             - run: pip install .[sklearn,torch,testing]
             - save_cache:
                   key: v0.4-torch-{{ checksum "setup.py" }}
                   paths:
                       - '~/.cache/pip'
-            - run: RUN_PIPELINE_TESTS=1 python -m pytest -n 8 --dist=loadfile -rA -s ./tests/ -m is_pipeline_test | tee output.txt
+            - run: RUN_PIPELINE_TESTS=1 python -m pytest -n 8 --dist=loadfile -rA -s --make-reports=tests_pipelines_torch -m is_pipeline_test ./tests/ | tee tests_output.txt
             - store_artifacts:
-                  path: ~/transformers/output.txt
-                  destination: test_output.txt
+                  path: ~/transformers/tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
+
     run_tests_pipelines_tf:
         working_directory: ~/transformers
         docker:
@@ -204,16 +207,17 @@ jobs:
                       - v0.4-tf-{{ checksum "setup.py" }}
                       - v0.4-{{ checksum "setup.py" }}
             - run: pip install --upgrade pip
-            - run: pip install git+https://github.com/huggingface/datasets
             - run: pip install .[sklearn,tf-cpu,testing]
             - save_cache:
                   key: v0.4-tf-{{ checksum "setup.py" }}
                   paths:
                       - '~/.cache/pip'
-            - run: RUN_PIPELINE_TESTS=1 python -m pytest -n 8 --dist=loadfile -rA -s ./tests/ -m is_pipeline_test | tee output.txt
+            - run: RUN_PIPELINE_TESTS=1 python -m pytest -n 8 --dist=loadfile -rA -s --make-reports=tests_pipelines_tf ./tests/ -m is_pipeline_test | tee tests_output.txt
             - store_artifacts:
-               path: ~/transformers/output.txt
-               destination: test_output.txt
+                  path: ~/transformers/tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
+
     run_tests_custom_tokenizers:
         working_directory: ~/transformers
         docker:
@@ -233,10 +237,12 @@ jobs:
                   key: v0.4-custom_tokenizers-{{ checksum "setup.py" }}
                   paths:
                       - '~/.cache/pip'
-            - run: python -m pytest -s ./tests/test_tokenization_bert_japanese.py | tee output.txt
+            - run: python -m pytest -s --make-reports=tests_custom_tokenizers ./tests/test_tokenization_bert_japanese.py | tee tests_output.txt
             - store_artifacts:
-                path: ~/transformers/output.txt
-                destination: test_output.txt
+                  path: ~/transformers/tests_output.txt
+            - store_artifacts:
+                  path: ~/transformers/reports
+
     run_examples_torch:
         working_directory: ~/transformers
         docker:
@@ -258,11 +264,12 @@ jobs:
                   key: v0.4-torch_examples-{{ checksum "setup.py" }}
                   paths:
                       - '~/.cache/pip'
-            - run: python -m pytest -n 8 --dist=loadfile -s --make_reports=examples ./examples/ | tee examples_output.txt
+            - run: python -m pytest -n 8 --dist=loadfile -s --make-reports=examples_torch ./examples/ | tee examples_output.txt
             - store_artifacts:
                   path: ~/transformers/examples_output.txt
             - store_artifacts:
                   path: ~/transformers/reports
+
     build_doc:
         working_directory: ~/transformers
         docker:
@@ -282,6 +289,7 @@ jobs:
             - run: cd docs && make html SPHINXOPTS="-W"
             - store_artifacts:
                 path: ./docs/_build
+
     deploy_doc:
         working_directory: ~/transformers
         docker:
@@ -301,6 +309,7 @@ jobs:
                   paths:
                       - '~/.cache/pip'
             - run: ./.circleci/deploy.sh
+
     check_code_quality:
         working_directory: ~/transformers
         docker:
@@ -327,6 +336,7 @@ jobs:
             - run: python utils/check_copies.py
             - run: python utils/check_dummies.py
             - run: python utils/check_repo.py
+
     check_repository_consistency:
         working_directory: ~/transformers
         docker:
@@ -357,6 +367,7 @@ jobs:
             - setup_remote_docker
             - *build_push_docker
             - *deploy_cluster
+
     cleanup-gke-jobs:
         docker:
             - image: circleci/python:3.6
@@ -366,6 +377,7 @@ jobs:
                   cluster: $GKE_CLUSTER
                   perform-login: true
             - *delete_gke_jobs
+
 workflow_filters: &workflow_filters
     filters:
         branches:

.github/ISSUE_TEMPLATE/bug-report.md

@@ -30,20 +30,22 @@ assignees: ''
  Trainer: @sgugger
  Speed and Memory Benchmarks: @patrickvonplaten
  Model Cards: @julien-c
- Translation: @sshleifer
- Summarization: @sshleifer
  TextGeneration: @TevenLeScao 
  examples/distillation: @VictorSanh
  nlp datasets: [different repo](https://github.com/huggingface/nlp)
  rust tokenizers: [different repo](https://github.com/huggingface/tokenizers)
- Text Generation: @TevenLeScao
- blenderbot: @mariamabarham
- Bart: @sshleifer
- Marian: @sshleifer
+ Text Generation: @patrickvonplaten @TevenLeScao
+ Blenderbot: @patrickvonplaten
+ Bart: @patrickvonplaten
+ Marian: @patrickvonplaten
+ Pegasus: @patrickvonplaten
+ mBART: @patrickvonplaten
  T5: @patrickvonplaten
  Longformer/Reformer: @patrickvonplaten
- TransfoXL/XLNet: @TevenLeScao 
- examples/seq2seq: @sshleifer
+ TransfoXL/XLNet: @TevenLeScao
+ RAG: @patrickvonplaten, @lhoestq
+ FSMT: @stas00
+ examples/seq2seq: @patil-suraj
  examples/bert-loses-patience: @JetRunner
  tensorflow: @jplu
  examples/token-classification: @stefan-it

.github/PULL_REQUEST_TEMPLATE.md

@@ -43,21 +43,20 @@ members/contributors which may be interested in your PR.
  Trainer: @sgugger
  Benchmarks: @patrickvonplaten
  Model Cards: @julien-c
- Translation: @sshleifer
- Summarization: @sshleifer
  examples/distillation: @VictorSanh
  nlp datasets: [different repo](https://github.com/huggingface/nlp)
  rust tokenizers: [different repo](https://github.com/huggingface/tokenizers)
  Text Generation: @patrickvonplaten, @TevenLeScao
- Blenderbot, Bart, Marian, Pegasus: @sshleifer
+ Blenderbot, Bart, Marian, Pegasus: @patrickvonplaten
  T5: @patrickvonplaten
  Rag: @patrickvonplaten, @lhoestq
  EncoderDecoder: @patrickvonplaten
  Longformer, Reformer: @patrickvonplaten
  TransfoXL, XLNet: @TevenLeScao, @patrickvonplaten
- examples/seq2seq: @sshleifer
+ examples/seq2seq: @patil-suraj
  examples/bert-loses-patience: @JetRunner
  tensorflow: @jplu
  examples/token-classification: @stefan-it
  documentation: @sgugger
+ FSTM: @stas00
  -->

.github/workflows/self-push.yml

@@ -13,7 +13,7 @@ on:
 
 
 jobs:
-  run_tests_torch_and_tf_gpu:
+  run_tests_torch_gpu:
     runs-on: [self-hosted, single-gpu]
     steps:
       - uses: actions/checkout@v2
@@ -32,7 +32,7 @@ jobs:
         id: cache
         with:
           path: .env
-          key: v1-tests_tf_torch_gpu-${{ hashFiles('setup.py') }}
+          key: v1.1-tests_torch_gpu-${{ hashFiles('setup.py') }}
 
       - name: Create new python env (on self-hosted runners we have to handle isolation ourselves)
         run: |
@@ -46,8 +46,7 @@ jobs:
         run: |
           source .env/bin/activate
           pip install --upgrade pip
-          pip install torch!=1.6.0
-          pip install .[sklearn,testing,onnxruntime]
+          pip install .[torch,sklearn,testing,onnxruntime]
           pip install git+https://github.com/huggingface/datasets
 
       - name: Are GPUs recognized by our DL frameworks
@@ -58,16 +57,26 @@ jobs:
 
       - name: Run all non-slow tests on GPU
         env:
-          TF_FORCE_GPU_ALLOW_GROWTH: "true"
-          # TF_GPU_MEMORY_LIMIT: 4096
           OMP_NUM_THREADS: 1
+          CUDA_VISIBLE_DEVICES: 0
         run: |
           source .env/bin/activate
-          python -m pytest -n 2 --dist=loadfile -s tests
+          python -m pytest -n 2 --dist=loadfile -s --make-reports=tests_torch_gpu tests
 
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_torch_gpu_failures_short.txt
+        
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: run_all_tests_torch_gpu_test_reports
+          path: reports
+                  
 
-  run_tests_torch_and_tf_multiple_gpu:
-    runs-on: [self-hosted, multi-gpu]
+  run_tests_tf_gpu:
+    runs-on: [self-hosted, single-gpu]
     steps:
       - uses: actions/checkout@v2
       - name: Python version
@@ -84,7 +93,68 @@ jobs:
         id: cache
         with:
           path: .env
-          key: v1-tests_tf_torch_multiple_gpu-${{ hashFiles('setup.py') }}
+          key: v1.1-tests_tf_gpu-${{ hashFiles('setup.py') }}
+
+      - name: Create new python env (on self-hosted runners we have to handle isolation ourselves)
+        run: |
+          python -m venv .env
+          source .env/bin/activate
+          which python
+          python --version
+          pip --version
+
+      - name: Install dependencies
+        run: |
+          source .env/bin/activate
+          pip install --upgrade pip
+          pip install .[tf,sklearn,testing,onnxruntime]
+          pip install git+https://github.com/huggingface/datasets
+
+      - name: Are GPUs recognized by our DL frameworks
+        run: |
+          source .env/bin/activate
+          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: Run all non-slow tests on GPU
+        env:
+          OMP_NUM_THREADS: 1
+          CUDA_VISIBLE_DEVICES: 0
+        run: |
+          source .env/bin/activate
+          python -m pytest -n 2 --dist=loadfile -s --make-reports=tests_tf_gpu tests
+
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_tf_gpu_failures_short.txt
+        
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: run_all_tests_tf_gpu_test_reports
+          path: reports
+
+  run_tests_torch_multiple_gpu:
+    runs-on: [self-hosted, multi-gpu]
+    steps:
+      - uses: actions/checkout@v2
+      - name: Python version
+        run: |
+          which python
+          python --version
+          pip --version
+
+      - name: Current dir
+        run: pwd
+      - run: nvidia-smi
+
+      - name: Loading cache.
+        uses: actions/cache@v2
+        id: cache
+        with:
+          path: .env
+          key: v1.1-tests_torch_multiple_gpu-${{ hashFiles('setup.py') }}
 
       - name: Create new python env (on self-hosted runners we have to handle isolation ourselves)
         run: |
@@ -97,8 +167,7 @@ jobs:
         run: |
           source .env/bin/activate
           pip install --upgrade pip
-          pip install torch!=1.6.0
-          pip install .[sklearn,testing,onnxruntime]
+          pip install .[torch,sklearn,testing,onnxruntime]
           pip install git+https://github.com/huggingface/datasets
 
       - name: Are GPUs recognized by our DL frameworks
@@ -109,9 +178,78 @@ jobs:
 
       - name: Run all non-slow tests on GPU
         env:
-          TF_FORCE_GPU_ALLOW_GROWTH: "true"
-          # TF_GPU_MEMORY_LIMIT: 4096
           OMP_NUM_THREADS: 1
         run: |
           source .env/bin/activate
-          python -m pytest -n 2 --dist=loadfile -s ./tests/
+          python -m pytest -n 2 --dist=loadfile -s --make-reports=tests_torch_multiple_gpu tests
+
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_torch_multiple_gpu_failures_short.txt          
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: run_all_tests_torch_multi_gpu_test_reports
+          path: reports
+
+  run_tests_tf_multiple_gpu:
+    runs-on: [self-hosted, multi-gpu]
+    steps:
+      - uses: actions/checkout@v2
+      - name: Python version
+        run: |
+          which python
+          python --version
+          pip --version
+
+      - name: Current dir
+        run: pwd
+      - run: nvidia-smi
+
+      - name: Loading cache.
+        uses: actions/cache@v2
+        id: cache
+        with:
+          path: .env
+          key: v1.1-tests_tf_multiple_gpu-${{ hashFiles('setup.py') }}
+
+      - name: Create new python env (on self-hosted runners we have to handle isolation ourselves)
+        run: |
+          python -m venv .env
+          source .env/bin/activate
+          which python
+          python --version
+          pip --version
+      - name: Install dependencies
+        run: |
+          source .env/bin/activate
+          pip install --upgrade pip
+          pip install .[tf,sklearn,testing,onnxruntime]
+          pip install git+https://github.com/huggingface/datasets
+
+      - name: Are GPUs recognized by our DL frameworks
+        run: |
+          source .env/bin/activate
+          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: Run all non-slow tests on GPU
+        env:
+          OMP_NUM_THREADS: 1
+        run: |
+          source .env/bin/activate
+          python -m pytest -n 2 --dist=loadfile -s --make-reports=tests_tf_multiple_gpu tests
+
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_tf_multiple_gpu_failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: run_all_tests_tf_multi_gpu_test_reports
+          path: reports
+          

.github/workflows/self-scheduled.yml

@@ -1,3 +1,8 @@
+# configuration notes:
+#
+# - `source .env/bin/activate` is currently needed to be run first thing first in each step. Otherwise
+# the step uses the system-wide python interpreter.
+
 name: Self-hosted runner (scheduled)
 
 on:
@@ -9,7 +14,7 @@ on:
     - cron: "0 0 * * *"
 
 jobs:
-  run_all_tests_torch_and_tf_gpu:
+  run_all_tests_torch_gpu:
     runs-on: [self-hosted, single-gpu]
     steps:
       - uses: actions/checkout@v2
@@ -19,7 +24,7 @@ jobs:
         id: cache
         with:
           path: .env
-          key: v1-slow_tests_tf_torch_gpu-${{ hashFiles('setup.py') }}
+          key: v  1.1-slow_tests_torch_gpu-${{ hashFiles('setup.py') }}
 
       - name: Python version
         run: |
@@ -44,9 +49,9 @@ jobs:
         run: |
           source .env/bin/activate
           pip install --upgrade pip
-          pip install torch!=1.6.0
-          pip install .[sklearn,testing,onnxruntime]
+          pip install .[torch,sklearn,testing,onnxruntime]
           pip install git+https://github.com/huggingface/datasets
+          pip list
 
       - name: Are GPUs recognized by our DL frameworks
         run: |
@@ -56,31 +61,29 @@ jobs:
 
       - name: Run all tests on GPU
         env:
-          TF_FORCE_GPU_ALLOW_GROWTH: "true"
           OMP_NUM_THREADS: 1
           RUN_SLOW: yes
         run: |
           source .env/bin/activate
-          python -m pytest -n 1 --dist=loadfile -s --make_reports=tests tests
+          python -m pytest -n 1 --dist=loadfile -s --make-reports=tests_torch_gpu tests
 
       - name: Failure short reports
         if: ${{ always() }}
-        run: cat reports/report_tests_failures_short.txt
+        run: cat reports/tests_torch_gpu_failures_short.txt
         
       - name: Run examples tests on GPU
         if: ${{ always() }}
         env:
-          TF_FORCE_GPU_ALLOW_GROWTH: "true"
           OMP_NUM_THREADS: 1
           RUN_SLOW: yes
         run: |
           source .env/bin/activate
           pip install -r examples/requirements.txt
-          python -m pytest -n 1 --dist=loadfile -s --make_reports=examples examples
+          python -m pytest -n 1 --dist=loadfile -s --make-reports=examples_torch_gpu examples
 
       - name: Failure short reports
         if: ${{ always() }}
-        run: cat reports/report_examples_failures_short.txt
+        run: cat reports/examples_torch_gpu_failures_short.txt
 
       - name: Run all pipeline tests on GPU
         if: ${{ always() }}
@@ -91,21 +94,100 @@ jobs:
           RUN_PIPELINE_TESTS: yes
         run: |
           source .env/bin/activate
-          python -m pytest -n 1 --dist=loadfile -s -m is_pipeline_test --make_reports=tests_pipeline tests
+          python -m pytest -n 1 --dist=loadfile -s -m is_pipeline_test --make-reports=tests_torch_pipeline_gpu tests
 
       - name: Failure short reports
         if: ${{ always() }}
-        run: cat reports/report_tests_pipeline_failures_short.txt
+        run: cat reports/tests_torch_pipeline_gpu_failures_short.txt
 
       - name: Test suite reports artifacts
         if: ${{ always() }}
         uses: actions/upload-artifact@v2
         with:
-          name: run_all_tests_torch_and_tf_gpu_test_reports
+          name: run_all_tests_torch_gpu_test_reports
           path: reports
 
 
-  run_all_tests_torch_and_tf_multiple_gpu:
+  run_all_tests_tf_gpu:
+    runs-on: [self-hosted, single-gpu]
+    steps:
+      - uses: actions/checkout@v2
+
+      - name: Loading cache.
+        uses: actions/cache@v2
+        id: cache
+        with:
+          path: .env
+          key: v1.1-slow_tests_tf_gpu-${{ hashFiles('setup.py') }}
+
+      - name: Python version
+        run: |
+          which python
+          python --version
+          pip --version
+
+      - name: Current dir
+        run: pwd
+      - run: nvidia-smi
+
+      - name: Create new python env (on self-hosted runners we have to handle isolation ourselves)
+        if: steps.cache.outputs.cache-hit != 'true'
+        run: |
+          python -m venv .env
+          source .env/bin/activate
+          which python
+          python --version
+          pip --version
+
+      - name: Install dependencies
+        run: |
+          source .env/bin/activate
+          pip install --upgrade pip
+          pip install .[tf,sklearn,testing,onnxruntime]
+          pip install git+https://github.com/huggingface/datasets
+          pip list
+
+      - name: Are GPUs recognized by our DL frameworks
+        run: |
+          source .env/bin/activate
+          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: Run all tests on GPU
+        env:
+          OMP_NUM_THREADS: 1
+          RUN_SLOW: yes
+        run: |
+          source .env/bin/activate
+          python -m pytest -n 1 --dist=loadfile -s --make-reports=tests_tf_gpu tests
+          
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_tf_gpu_failures_short.txt
+
+      - name: Run all pipeline tests on GPU
+        if: ${{ always() }}
+        env:
+          TF_FORCE_GPU_ALLOW_GROWTH: "true"
+          OMP_NUM_THREADS: 1
+          RUN_SLOW: yes
+          RUN_PIPELINE_TESTS: yes
+        run: |
+          source .env/bin/activate
+          python -m pytest -n 1 --dist=loadfile -s -m is_pipeline_test --make-reports=tests_tf_pipelines_gpu tests
+
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_tf_pipelines_gpu_failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: run_all_tests_tf_gpu_test_reports
+          path: reports
+          
+  run_all_tests_torch_multiple_gpu:
     runs-on: [self-hosted, multi-gpu]
     steps:
       - uses: actions/checkout@v2
@@ -115,16 +197,18 @@ jobs:
         id: cache
         with:
           path: .env
-          key: v1-slow_tests_tf_torch_multi_gpu-${{ hashFiles('setup.py') }}
+          key: v1.1-slow_tests_torch_multi_gpu-${{ hashFiles('setup.py') }}
 
       - name: Python version
         run: |
           which python
           python --version
           pip --version
+
       - name: Current dir
         run: pwd
       - run: nvidia-smi
+
       - name: Create new python env (on self-hosted runners we have to handle isolation ourselves)
         if: steps.cache.outputs.cache-hit != 'true'
         run: |
@@ -133,13 +217,14 @@ jobs:
           which python
           python --version
           pip --version
+
       - name: Install dependencies
         run: |
           source .env/bin/activate
           pip install --upgrade pip
-          pip install torch!=1.6.0
-          pip install .[sklearn,testing,onnxruntime]
+          pip install .[torch,sklearn,testing,onnxruntime]
           pip install git+https://github.com/huggingface/datasets
+          pip list
 
       - name: Are GPUs recognized by our DL frameworks
         run: |
@@ -147,26 +232,32 @@ jobs:
           python -c "import torch; print('Cuda available:', torch.cuda.is_available())"
           python -c "import torch; print('Number of GPUs available:', torch.cuda.device_count())"
 
-      - name: Run all tests on GPU
+      - name: Run all tests on multi-GPU
         env:
-          TF_FORCE_GPU_ALLOW_GROWTH: "true"
           OMP_NUM_THREADS: 1
           RUN_SLOW: yes
         run: |
           source .env/bin/activate
-          python -m pytest -n 1 --dist=loadfile -s ./tests/ --durations=50
+          python -m pytest -n 1 --dist=loadfile -s --make-reports=tests_torch_multiple_gpu tests
 
-      - name: Run examples tests on GPU
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_torch_multiple_gpu_failures_short.txt
+
+      - name: Run examples tests on multi-GPU
         env:
-          TF_FORCE_GPU_ALLOW_GROWTH: "true"
           OMP_NUM_THREADS: 1
           RUN_SLOW: yes
         run: |
           source .env/bin/activate
-          pip install -r examples/requirements.txt
-          python -m pytest -n 1 --dist=loadfile -s examples --durations=50
+          python -m pytest -n 1 --dist=loadfile -s --make-reports=examples_torch_multiple_gpu examples
 
-      - name: Run all pipeline tests on GPU
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/examples_torch_multiple_gpu_failures_short.txt
+
+      - name: Run all pipeline tests on multi-GPU
+        if: ${{ always() }}
         env:
           TF_FORCE_GPU_ALLOW_GROWTH: "true"
           OMP_NUM_THREADS: 1
@@ -174,4 +265,95 @@ jobs:
           RUN_PIPELINE_TESTS: yes
         run: |
           source .env/bin/activate
-          python -m pytest -n 1 --dist=loadfile -s ./tests/ -m is_pipeline_test --durations=50
+          python -m pytest -n 1 --dist=loadfile -s -m is_pipeline_test --make-reports=tests_torch_pipeline_multiple_gpu tests
+
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_torch_pipeline_multiple_gpu_failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: run_all_tests_torch_multi_gpu_test_reports
+          path: reports
+
+  run_all_tests_tf_multiple_gpu:
+    runs-on: [self-hosted, multi-gpu]
+    steps:
+      - uses: actions/checkout@v2
+
+      - name: Loading cache.
+        uses: actions/cache@v2
+        id: cache
+        with:
+          path: .env
+          key: v1.1-slow_tests_tf_multi_gpu-${{ hashFiles('setup.py') }}
+
+      - name: Python version
+        run: |
+          which python
+          python --version
+          pip --version
+
+      - name: Current dir
+        run: pwd
+      - run: nvidia-smi
+
+      - name: Create new python env (on self-hosted runners we have to handle isolation ourselves)
+        if: steps.cache.outputs.cache-hit != 'true'
+        run: |
+          python -m venv .env
+          source .env/bin/activate
+          which python
+          python --version
+          pip --version
+
+      - name: Install dependencies
+        run: |
+          source .env/bin/activate
+          pip install --upgrade pip
+          pip install .[tf,sklearn,testing,onnxruntime]
+          pip install git+https://github.com/huggingface/datasets
+          pip list
+
+      - name: Are GPUs recognized by our DL frameworks
+        run: |
+          source .env/bin/activate
+          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: Run all tests on multi-GPU
+        env:
+          OMP_NUM_THREADS: 1
+          RUN_SLOW: yes
+        run: |
+          source .env/bin/activate
+          python -m pytest -n 1 --dist=loadfile -s --make-reports=tests_tf_multiple_gpu tests
+
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_tf_multiple_gpu_failures_short.txt
+
+      - name: Run all pipeline tests on multi-GPU
+        if: ${{ always() }}
+        env:
+          TF_FORCE_GPU_ALLOW_GROWTH: "true"
+          OMP_NUM_THREADS: 1
+          RUN_SLOW: yes
+          RUN_PIPELINE_TESTS: yes
+        run: |
+          source .env/bin/activate
+          python -m pytest -n 1 --dist=loadfile -s -m is_pipeline_test --make-reports=tests_tf_pipelines_multiple_gpu tests
+          
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_tf_multiple_gpu_pipelines_failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: run_all_tests_tf_multi_gpu_test_reports
+          path: reports
+          

Makefile

@@ -3,14 +3,9 @@
 
 check_dirs := examples tests src utils
 
-# get modified files since the branch was made
-fork_point_sha := $(shell git merge-base --fork-point master)
-joined_dirs := $(shell echo $(check_dirs) | tr " " "|")
-modified_py_files := $(shell git diff --name-only $(fork_point_sha) | egrep '^($(joined_dirs))' | egrep '\.py$$')
-#$(info modified files are: $(modified_py_files))
-
 modified_only_fixup:
-	@if [ -n "$(modified_py_files)" ]; then \
+	$(eval modified_py_files := $(shell python utils/get_modified_files.py $(check_dirs)))
+	@if test -n "$(modified_py_files)"; then \
 		echo "Checking/fixing $(modified_py_files)"; \
 		black $(modified_py_files); \
 		isort $(modified_py_files); \

docs/source/conf.py

@@ -26,7 +26,7 @@ author = u'huggingface'
 # The short X.Y version
 version = u''
 # The full version, including alpha/beta/rc tags
-release = u'3.4.0'
+release = u'3.5.0'
 
 
 # -- General configuration ---------------------------------------------------

docs/source/index.rst

@@ -272,3 +272,4 @@ conversion utilities for the following models:
     internal/pipelines_utils
     internal/tokenization_utils
     internal/trainer_utils
+    internal/generation_utils

docs/source/internal/generation_utils.rst

@@ -0,0 +1,50 @@
+Utilities for Generation
+-----------------------------------------------------------------------------------------------------------------------
+
+This page lists all the utility functions used by :meth:`~transformers.PretrainedModel.generate`,
+:meth:`~transformers.PretrainedModel.greedy_search`, :meth:`~transformers.PretrainedModel.sample`,
+:meth:`~transformers.PretrainedModel.beam_search`, and :meth:`~transformers.PretrainedModel.beam_sample`.
+
+Most of those are only useful if you are studying the code of the generate methods in the library.
+
+LogitsProcessor
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+A :class:`~transformers.LogitsProcessor` can be used to modify the prediction scores of a language model head for
+generation.
+
+.. autoclass:: transformers.LogitsProcessor
+    :members: __call__
+
+.. autoclass:: transformers.LogitsProcessorList
+    :members: __call__
+
+.. autoclass:: transformers.MinLengthLogitsProcessor
+    :members: __call__
+
+.. autoclass:: transformers.TemperatureLogitsWarper
+    :members: __call__
+
+.. autoclass:: transformers.RepetitionPenaltyLogitsProcessor
+    :members: __call__
+
+.. autoclass:: transformers.TopPLogitsWarper
+    :members: __call__
+
+.. autoclass:: transformers.TopKLogitsWarper
+    :members: __call__
+
+.. autoclass:: transformers.NoRepeatNGramLogitsProcessor
+    :members: __call__
+
+.. autoclass:: transformers.NoBadWordsLogitsProcessor
+    :members: __call__
+
+BeamSearch
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.BeamScorer
+    :members: process, finalize
+
+.. autoclass:: transformers.BeamSearchScorer
+    :members: process, finalize

docs/source/main_classes/model.rst

@@ -45,7 +45,7 @@ TFModelUtilsMixin
     :members:
 
 
-Generative models
+Generation
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 .. autoclass:: transformers.generation_utils.GenerationMixin

docs/source/main_classes/output.rst

@@ -65,12 +65,34 @@ BaseModelOutputWithPooling
     :members:
 
 
+BaseModelOutputWithCrossAttentions
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.modeling_outputs.BaseModelOutputWithCrossAttentions
+    :members:
+
+
+BaseModelOutputWithPoolingAndCrossAttentions
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.modeling_outputs.BaseModelOutputWithPoolingAndCrossAttentions
+    :members:
+
+
 BaseModelOutputWithPast
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 .. autoclass:: transformers.modeling_outputs.BaseModelOutputWithPast
     :members:
 
+
+BaseModelOutputWithPastAndCrossAttentions
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.modeling_outputs.BaseModelOutputWithPastAndCrossAttentions
+    :members:
+
+
 Seq2SeqModelOutput
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
@@ -85,6 +107,20 @@ CausalLMOutput
     :members:
 
 
+CausalLMOutputWithCrossAttentions
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.modeling_outputs.CausalLMOutputWithCrossAttentions
+    :members:
+
+
+CausalLMOutputWithPastAndCrossAttentions
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.modeling_outputs.CausalLMOutputWithPastAndCrossAttentions
+    :members:
+
+
 CausalLMOutputWithPast
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 

docs/source/model_doc/auto.rst

@@ -81,6 +81,13 @@ AutoModelForMultipleChoice
     :members:
 
 
+AutoModelForNextSentencePrediction
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.AutoModelForNextSentencePrediction
+    :members:
+
+
 AutoModelForTokenClassification
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 

docs/source/model_doc/bart.rst

@@ -3,7 +3,7 @@ BART
 
 **DISCLAIMER:** If you see something strange, file a `Github Issue
 <https://github.com/huggingface/transformers/issues/new?assignees=&labels=&template=bug-report.md&title>`__ and assign
-@sshleifer
+@patrickvonplaten
 
 Overview
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
@@ -26,6 +26,16 @@ According to the abstract,
 The Authors' code can be found `here <https://github.com/pytorch/fairseq/tree/master/examples/bart>`__.
 
 
+Examples
+_______________________________________________________________________________________________________________________
+
+- Examples and scripts for fine-tuning BART and other models for sequence to sequence tasks can be found in
+  `examples/seq2seq/ <https://github.com/huggingface/transformers/blob/master/examples/seq2seq/README.md>`__.
+- An example of how to train :class:`~transformers.BartForConditionalGeneration` with a Hugging Face :obj:`datasets`
+  object can be found in this `forum discussion
+  <https://discuss.huggingface.co/t/train-bart-for-conditional-generation-e-g-summarization/1904>`__.
+
+
 Implementation Notes
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 

docs/source/model_doc/blenderbot.rst

@@ -95,3 +95,12 @@ See :obj:`transformers.BartForConditionalGeneration` for arguments to `forward`
 
 .. autoclass:: transformers.BlenderbotForConditionalGeneration
     :members:
+
+
+TFBlenderbotForConditionalGeneration
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+See :obj:`transformers.TFBartForConditionalGeneration` for arguments to `forward` and `generate`
+
+.. autoclass:: transformers.TFBlenderbotForConditionalGeneration
+    :members:

docs/source/model_doc/longformer.rst

@@ -90,6 +90,32 @@ LongformerTokenizerFast
 .. autoclass:: transformers.LongformerTokenizerFast
     :members: 
 
+Longformer specific outputs
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.modeling_longformer.LongformerBaseModelOutput
+    :members: 
+
+.. autoclass:: transformers.modeling_longformer.LongformerBaseModelOutputWithPooling
+    :members: 
+
+.. autoclass:: transformers.modeling_longformer.LongformerMultipleChoiceModelOutput
+    :members: 
+
+.. autoclass:: transformers.modeling_longformer.LongformerQuestionAnsweringModelOutput
+    :members: 
+
+.. autoclass:: transformers.modeling_tf_longformer.TFLongformerBaseModelOutput
+    :members: 
+
+.. autoclass:: transformers.modeling_tf_longformer.TFLongformerBaseModelOutputWithPooling
+    :members: 
+
+.. autoclass:: transformers.modeling_tf_longformer.TFLongformerQuestionAnsweringModelOutput
+    :members: 
+
+LongformerModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 LongformerModel
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

docs/source/model_doc/marian.rst

@@ -3,7 +3,7 @@ MarianMT
 
 **Bugs:** If you see something strange, file a `Github Issue
 <https://github.com/huggingface/transformers/issues/new?assignees=sshleifer&labels=&template=bug-report.md&title>`__
-and assign @sshleifer.
+and assign @patrickvonplaten.
 
 Translations should be similar, but not identical to, output in the test set linked to in each model card.
 
@@ -129,3 +129,9 @@ MarianMTModel
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 .. autoclass:: transformers.MarianMTModel
+
+
+TFMarianMTModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFMarianMTModel

docs/source/model_doc/mbart.rst

@@ -3,7 +3,7 @@ MBart
 
 **DISCLAIMER:** If you see something strange, file a `Github Issue
 <https://github.com/huggingface/transformers/issues/new?assignees=&labels=&template=bug-report.md&title>`__ and assign
-@sshleifer
+@patrickvonplaten
 
 Overview
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
@@ -79,4 +79,11 @@ MBartForConditionalGeneration
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 .. autoclass:: transformers.MBartForConditionalGeneration
-    :members: forward
+    :members:
+
+
+TFMBartForConditionalGeneration
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFMBartForConditionalGeneration
+    :members:

docs/source/model_doc/pegasus.rst

@@ -3,7 +3,7 @@ Pegasus
 
 **DISCLAIMER:** If you see something strange, file a `Github Issue
 <https://github.com/huggingface/transformers/issues/new?assignees=sshleifer&labels=&template=bug-report.md&title>`__
-and assign @sshleifer.
+and assign @patrickvonplaten.
 
 
 Overview
@@ -95,3 +95,9 @@ PegasusForConditionalGeneration
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 .. autoclass:: transformers.PegasusForConditionalGeneration
+
+
+TFPegasusForConditionalGeneration
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TFPegasusForConditionalGeneration

docs/source/model_sharing.rst

@@ -18,39 +18,65 @@ done something similar on your task, either using the model directly in your own
 :class:`~.transformers.Trainer`/:class:`~.transformers.TFTrainer` class. Let's see how you can share the result on the
 `model hub <https://huggingface.co/models>`__.
 
+Model versioning
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Since version v3.5.0, the model hub has built-in model versioning based on git and git-lfs. It is based on the paradigm
+that one model *is* one repo.
+
+This allows:
+
+- built-in versioning
+- access control
+- scalability
+
+This is built around *revisions*, which is a way to pin a specific version of a model, using a commit hash, tag or
+branch.
+
+For instance:
+
+.. code-block::
+
+    >>> tokenizer = AutoTokenizer.from_pretrained(
+    >>>   "julien-c/EsperBERTo-small",
+    >>>   revision="v2.0.1" # tag name, or branch name, or commit hash
+    >>> )
+
 Basic steps
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
-.. 
-    When #5258 is merged, we can remove the need to create the directory.
+In order to upload a model, you'll need to first create a git repo. This repo will live on the model hub, allowing
+users to clone it and you (and your organization members) to push to it. First, you should ensure you are logged in the
+``transformers-cli``:
 
-First, pick a directory with the name you want your model to have on the model hub (its full name will then be
-`username/awesome-name-you-picked` or `organization/awesome-name-you-picked`) and create it with either
+Go in a terminal and run the following command. It should be in the virtual environment where you installed 🤗
+Transformers, since that command :obj:`transformers-cli` comes from the library.
 
 .. code-block::
 
-    mkdir path/to/awesome-name-you-picked
+    transformers-cli login
 
-or in python
+
+Once you are logged in with your model hub credentials, you can start building your repositories. To create a repo:
 
 .. code-block::
 
-    import os
-    os.makedirs("path/to/awesome-name-you-picked")
+    transformers-cli repo create your-model-name
 
-then you can save your model and tokenizer with:
+This creates a repo on the model hub, which can be cloned. You can then add/remove from that repo as you would with any
+other git repo.
 
 .. code-block::
 
-    model.save_pretrained("path/to/awesome-name-you-picked")
-    tokenizer.save_pretrained("path/to/awesome-name-you-picked")
+    git clone https://huggingface.co/username/your-model-name
 
-Or, if you're using the Trainer API
+    # Then commit as usual
+    cd your-model-name
+    echo "hello" >> README.md
+    git add . && git commit -m "Update from $USER"
 
-.. code-block::
+We are intentionally not wrapping git too much, so as to stay intuitive and easy-to-use.
 
-    trainer.save_model("path/to/awesome-name-you-picked")
-    tokenizer.save_pretrained("path/to/awesome-name-you-picked")
 
 Make your model work on all frameworks
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
@@ -71,13 +97,13 @@ or removing TF. For instance, if you trained a :class:`~transformers.DistilBertF
 
 .. code-block::
 
-    from transformers import TFDistilBertForSequenceClassification
+    >>> from transformers import TFDistilBertForSequenceClassification
 
 and if you trained a :class:`~transformers.TFDistilBertForSequenceClassification`, try to type
 
 .. code-block::
 
-    from transformers import DistilBertForSequenceClassification
+    >>> from transformers import DistilBertForSequenceClassification
 
 This will give back an error if your model does not exist in the other framework (something that should be pretty rare
 since we're aiming for full parity between the two frameworks). In this case, skip this and go to the next step.
@@ -87,20 +113,20 @@ model class:
 
 .. code-block::
 
-    tf_model = TFDistilBertForSequenceClassification.from_pretrained("path/to/awesome-name-you-picked", from_pt=True)
-    tf_model.save_pretrained("path/to/awesome-name-you-picked")
+    >>> tf_model = TFDistilBertForSequenceClassification.from_pretrained("path/to/awesome-name-you-picked", from_pt=True)
+    >>> tf_model.save_pretrained("path/to/awesome-name-you-picked")
 
 and if you trained your model in TensorFlow and have to create a PyTorch version, adapt the following code to your
 model class:
 
 .. code-block::
 
-    pt_model = DistilBertForSequenceClassification.from_pretrained("path/to/awesome-name-you-picked", from_tf=True)
-    pt_model.save_pretrained("path/to/awesome-name-you-picked")
+    >>> pt_model = DistilBertForSequenceClassification.from_pretrained("path/to/awesome-name-you-picked", from_tf=True)
+    >>> pt_model.save_pretrained("path/to/awesome-name-you-picked")
 
 That's all there is to it!
 
-Check the directory before uploading
+Check the directory before pushing to the model hub.
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
 Make sure there are no garbage files in the directory you'll upload. It should only have:
@@ -116,62 +142,46 @@ Make sure there are no garbage files in the directory you'll upload. It should o
 
 Other files can safely be deleted.
 
-Upload your model with the CLI
+
+Uploading your files
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
-Now go in a terminal and run the following command. It should be in the virtual environment where you installed 🤗
-Transformers, since that command :obj:`transformers-cli` comes from the library.
+Once the repo is cloned, you can add the model, configuration and tokenizer files. For instance, saving the model and
+tokenizer files:
 
 .. code-block::
 
-    transformers-cli login
+    >>> model.save_pretrained("path/to/repo/clone/your-model-name")
+    >>> tokenizer.save_pretrained("path/to/repo/clone/your-model-name")
 
-Then log in using the same credentials as on huggingface.co. To upload your model, just type
+Or, if you're using the Trainer API
 
 .. code-block::
 
-    transformers-cli upload path/to/awesome-name-you-picked/
+    >>> trainer.save_model("path/to/awesome-name-you-picked")
 
-This will upload the folder containing the weights, tokenizer and configuration we prepared in the previous section.
-
-By default you will be prompted to confirm that you want these files to be uploaded. If you are uploading multiple
-models and need to script that process, you can add `-y` to bypass the prompt. For example:
+You can then add these files to the staging environment and verify that they have been correctly staged with the ``git
+status`` command:
 
 .. code-block::
 
-    transformers-cli upload -y path/to/awesome-name-you-picked/
+    git add --all
+    git status
 
-
-If you want to upload a single file (a new version of your model, or the other framework checkpoint you want to add),
-just type:
+Finally, the files should be comitted:
 
 .. code-block::
 
-    transformers-cli upload path/to/awesome-name-you-picked/that-file 
+    git commit -m "First version of the your-model-name model and tokenizer."
 
-or
+And pushed to the remote:
 
 .. code-block::
 
-   transformers-cli upload path/to/awesome-name-you-picked/that-file --filename awesome-name-you-picked/new_name
+    git push
 
-if you want to change its filename.
+This will upload the folder containing the weights, tokenizer and configuration we have just prepared.
 
-This uploads the model to your personal account. If you want your model to be namespaced by your organization name
-rather than your username, add the following flag to any command:
-
-.. code-block::
-
-    --organization organization_name
-
-so for instance:
-
-.. code-block::
-
-    transformers-cli upload path/to/awesome-name-you-picked/ --organization organization_name
-
-Your model will then be accessible through its identifier, which is, as we saw above,
-`username/awesome-name-you-picked` or `organization/awesome-name-you-picked`.
 
 Add a model card
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
@@ -203,20 +213,15 @@ Anyone can load it from code:
 
 .. code-block::
 
-    tokenizer = AutoTokenizer.from_pretrained("namespace/awesome-name-you-picked")
-    model = AutoModel.from_pretrained("namespace/awesome-name-you-picked")
+    >>> tokenizer = AutoTokenizer.from_pretrained("namespace/awesome-name-you-picked")
+    >>> model = AutoModel.from_pretrained("namespace/awesome-name-you-picked")
 
-Additional commands
-^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
-You can list all the files you uploaded on the hub like this:
+You may specify a revision by using the ``revision`` flag in the ``from_pretrained`` method:
 
 .. code-block::
 
-    transformers-cli s3 ls
-
-You can also delete unneeded files with
-
-.. code-block::
-
-    transformers-cli s3 rm awesome-name-you-picked/filename
+    >>> tokenizer = AutoTokenizer.from_pretrained(
+    >>>   "julien-c/EsperBERTo-small",
+    >>>   revision="v2.0.1" # tag name, or branch name, or commit hash
+    >>> )

docs/source/testing.rst

@@ -409,6 +409,22 @@ decorators are used to set the requirements of tests CPU/GPU/TPU-wise:
 * ``require_torch_non_multigpu`` - as ``require_torch`` plus requires 0 or 1 GPUs
 * ``require_torch_tpu`` - as ``require_torch`` plus requires at least 1 TPU
 
+Let's depict the GPU requirements in the following table:
+
+
++----------+---------------------------------+
+| n gpus   |  decorator                      |
++==========+=================================+
+| ``>= 0`` | ``@require_torch``              |
++----------+---------------------------------+
+| ``>= 1`` | ``@require_torch_gpu``          |
++----------+---------------------------------+
+| ``>= 2`` | ``@require_torch_multigpu``     |
++----------+---------------------------------+
+| ``< 2``  | ``@require_torch_non_multigpu`` |
++----------+---------------------------------+
+
+
 For example, here is a test that must be run only when there are 2 or more GPUs available and pytorch is installed:
 
 .. code-block:: python
@@ -450,7 +466,8 @@ Inside tests:
 
 .. code-block:: bash
 
-   torch.cuda.device_count()
+   from transformers.testing_utils import get_gpu_count
+   n_gpu = get_gpu_count() # works with torch and tf
 
 
 

examples/README.md

@@ -5,7 +5,9 @@ Running the examples requires PyTorch 1.3.1+ or TensorFlow 2.2+.
 
 Here is the list of all our examples:
 - **grouped by task** (all official examples work for multiple models)
-- with information on whether they are **built on top of `Trainer`/`TFTrainer`** (if not, they still work, they might just lack some features),
+- with information on whether they are **built on top of `Trainer`/`TFTrainer`** (if not, they still work, they might
+  just lack some features),
+- whether or not they leverage the [🤗 Datasets](https://github.com/huggingface/datasets) library.
 - links to **Colab notebooks** to walk through the scripts and run them easily,
 - links to **Cloud deployments** to be able to deploy large-scale trainings in the Cloud with little to no setup.
 
@@ -31,19 +33,19 @@ git checkout tags/v3.4.0
 
 ## The Big Table of Tasks
 
-| Task | Example datasets | Trainer support | TFTrainer support | Colab
-|---|---|:---:|:---:|:---:|
-| [**`language-modeling`**](https://github.com/huggingface/transformers/tree/master/examples/language-modeling)       | Raw text        | ✅ | -  | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/blog/blob/master/notebooks/01_how_to_train.ipynb)
-| [**`text-classification`**](https://github.com/huggingface/transformers/tree/master/examples/text-classification)   | GLUE, XNLI      | ✅ | ✅ | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/blog/blob/master/notebooks/trainer/01_text_classification.ipynb)
-| [**`token-classification`**](https://github.com/huggingface/transformers/tree/master/examples/token-classification) | CoNLL NER       | ✅ | ✅ | -
-| [**`multiple-choice`**](https://github.com/huggingface/transformers/tree/master/examples/multiple-choice)           | SWAG, RACE, ARC | ✅ | ✅ | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/ViktorAlm/notebooks/blob/master/MPC_GPU_Demo_for_TF_and_PT.ipynb)
-| [**`question-answering`**](https://github.com/huggingface/transformers/tree/master/examples/question-answering)     | SQuAD           | ✅ | ✅ | -
-| [**`text-generation`**](https://github.com/huggingface/transformers/tree/master/examples/text-generation)           | -               | n/a | n/a | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/blog/blob/master/notebooks/02_how_to_generate.ipynb)
-| [**`distillation`**](https://github.com/huggingface/transformers/tree/master/examples/distillation)                 | All             | - | -  | -
-| [**`summarization`**](https://github.com/huggingface/transformers/tree/master/examples/seq2seq)                     | CNN/Daily Mail  | ✅  | - | -
-| [**`translation`**](https://github.com/huggingface/transformers/tree/master/examples/seq2seq)                       | WMT             | ✅  | - | -
-| [**`bertology`**](https://github.com/huggingface/transformers/tree/master/examples/bertology)                       | -               | - | - | -
-| [**`adversarial`**](https://github.com/huggingface/transformers/tree/master/examples/adversarial)                   | HANS            | ✅ | - | -
+| Task | Example datasets | Trainer support | TFTrainer support | 🤗 Datasets | Colab
+|---|---|:---:|:---:|:---:|:---:|
+| [**`language-modeling`**](https://github.com/huggingface/transformers/tree/master/examples/language-modeling)       | Raw text        | ✅ | -  | ✅ | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/blog/blob/master/notebooks/01_how_to_train.ipynb)
+| [**`text-classification`**](https://github.com/huggingface/transformers/tree/master/examples/text-classification)   | GLUE, XNLI      | ✅ | ✅ | ✅ | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://github.com/huggingface/notebooks/blob/master/examples/text_classification.ipynb)
+| [**`token-classification`**](https://github.com/huggingface/transformers/tree/master/examples/token-classification) | CoNLL NER       | ✅ | ✅ | ✅ | -
+| [**`multiple-choice`**](https://github.com/huggingface/transformers/tree/master/examples/multiple-choice)           | SWAG, RACE, ARC | ✅ | ✅ | - | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/ViktorAlm/notebooks/blob/master/MPC_GPU_Demo_for_TF_and_PT.ipynb)
+| [**`question-answering`**](https://github.com/huggingface/transformers/tree/master/examples/question-answering)     | SQuAD           | ✅ | ✅ | - | -
+| [**`text-generation`**](https://github.com/huggingface/transformers/tree/master/examples/text-generation)           | -               | n/a | n/a | - | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/blog/blob/master/notebooks/02_how_to_generate.ipynb)
+| [**`distillation`**](https://github.com/huggingface/transformers/tree/master/examples/distillation)                 | All             | - | -  | - | -
+| [**`summarization`**](https://github.com/huggingface/transformers/tree/master/examples/seq2seq)                     | CNN/Daily Mail  | ✅  | - | - | -
+| [**`translation`**](https://github.com/huggingface/transformers/tree/master/examples/seq2seq)                       | WMT             | ✅  | - | - | -
+| [**`bertology`**](https://github.com/huggingface/transformers/tree/master/examples/bertology)                       | -               | - | - | - | -
+| [**`adversarial`**](https://github.com/huggingface/transformers/tree/master/examples/adversarial)                   | HANS            | ✅ | - | - | -
 
 
 <br>
@@ -60,13 +62,14 @@ When using PyTorch, we support TPUs thanks to `pytorch/xla`. For more context an
 very detailed [pytorch/xla README](https://github.com/pytorch/xla/blob/master/README.md).
 
 In this repo, we provide a very simple launcher script named [xla_spawn.py](https://github.com/huggingface/transformers/tree/master/examples/xla_spawn.py) that lets you run our example scripts on multiple TPU cores without any boilerplate.
-Just pass a `--num_cores` flag to this script, then your regular training script with its arguments (this is similar to the `torch.distributed.launch` helper for torch.distributed).
+Just pass a `--num_cores` flag to this script, then your regular training script with its arguments (this is similar to the `torch.distributed.launch` helper for torch.distributed). 
+Note that this approach does not work for examples that use `pytorch-lightning`.
 
 For example for `run_glue`:
 
 ```bash
 python examples/xla_spawn.py --num_cores 8 \
-	examples/text-classification/run_glue.py
+	examples/text-classification/run_glue.py \
 	--model_name_or_path bert-base-cased \
 	--task_name mnli \
 	--data_dir ./data/glue_data/MNLI \

examples/bert-loses-patience/test_run_glue_with_pabee.py

@@ -4,7 +4,7 @@ import sys
 from unittest.mock import patch
 
 import run_glue_with_pabee
-from transformers.testing_utils import TestCasePlus
+from transformers.testing_utils import TestCasePlus, require_torch_non_multigpu_but_fix_me
 
 
 logging.basicConfig(level=logging.DEBUG)
@@ -20,6 +20,7 @@ def get_setup_file():
 
 
 class PabeeTests(TestCasePlus):
+    @require_torch_non_multigpu_but_fix_me
     def test_run_glue(self):
         stream_handler = logging.StreamHandler(sys.stdout)
         logger.addHandler(stream_handler)

examples/conftest.py

@@ -17,17 +17,14 @@ warnings.simplefilter(action="ignore", category=FutureWarning)
 
 
 def pytest_addoption(parser):
-    parser.addoption(
-        "--make_reports",
-        action="store",
-        default=False,
-        help="generate report files - the value will be used as a `report_`+val+`reportname.txt`",
-    )
+    from transformers.testing_utils import pytest_addoption_shared
+
+    pytest_addoption_shared(parser)
 
 
 def pytest_terminal_summary(terminalreporter):
     from transformers.testing_utils import pytest_terminal_summary_main
 
-    make_reports = terminalreporter.config.getoption("--make_reports")
+    make_reports = terminalreporter.config.getoption("--make-reports")
     if make_reports:
         pytest_terminal_summary_main(terminalreporter, id=make_reports)

examples/deebert/test_glue_deebert.py

@@ -5,7 +5,7 @@ import unittest
 from unittest.mock import patch
 
 import run_glue_deebert
-from transformers.testing_utils import slow
+from transformers.testing_utils import require_torch_non_multigpu_but_fix_me, slow
 
 
 logging.basicConfig(level=logging.DEBUG)
@@ -26,6 +26,7 @@ class DeeBertTests(unittest.TestCase):
         logger.addHandler(stream_handler)
 
     @slow
+    @require_torch_non_multigpu_but_fix_me
     def test_glue_deebert_train(self):
 
         train_args = """

examples/distillation/README.md

@@ -12,7 +12,7 @@ This folder contains the original code used to train Distil* as well as examples
 
 **October 3, 2019 - Update** We release our [NeurIPS workshop paper](https://arxiv.org/abs/1910.01108) explaining our approach on **DistilBERT**. It includes updated results and further experiments. We applied the same method to GPT2 and release the weights of **DistilGPT2**. DistilGPT2 is two times faster and 33% smaller than GPT2. **The paper supersedes our [previous blogpost](https://medium.com/huggingface/distilbert-8cf3380435b5) with a different distillation loss and better performances. Please use the paper as a reference when comparing/reporting results on DistilBERT.**
 
-**September 19, 2019 - Update:** We fixed bugs in the code and released an upadted version of the weights trained with a modification of the distillation loss. DistilBERT now reaches 99% of `BERT-base`'s performance on GLUE, and 86.9 F1 score on SQuAD v1.1 dev set (compared to 88.5 for `BERT-base`). We will publish a formal write-up of our approach in the near future!
+**September 19, 2019 - Update:** We fixed bugs in the code and released an updated version of the weights trained with a modification of the distillation loss. DistilBERT now reaches 99% of `BERT-base`'s performance on GLUE, and 86.9 F1 score on SQuAD v1.1 dev set (compared to 88.5 for `BERT-base`). We will publish a formal write-up of our approach in the near future!
 
 
 ## What is Distil*

examples/language-modeling/README.md

@@ -1,16 +1,19 @@
-
 ## Language model training
 
-Based on the script [`run_language_modeling.py`](https://github.com/huggingface/transformers/blob/master/examples/language-modeling/run_language_modeling.py).
+Fine-tuning (or training from scratch) the library models for language modeling on a text dataset for GPT, GPT-2,
+ALBERT, BERT, DistilBERT, RoBERTa, XLNet... GPT and GPT-2 are trained or fine-tuned using a causal language modeling
+(CLM) loss while ALBERT, BERT, DistilBERT and RoBERTa are trained or fine-tuned using a masked language modeling (MLM)
+loss. XLNet uses permutation language modeling (PLM), you can find more information about the differences between those
+objectives in our [model summary](https://huggingface.co/transformers/model_summary.html).
 
-Fine-tuning (or training from scratch) the library models for language modeling on a text dataset for GPT, GPT-2, BERT, DistilBERT and RoBERTa. GPT and GPT-2 are fine-tuned using a causal language modeling (CLM) loss while BERT, DistilBERT and RoBERTa
-are fine-tuned using a masked language modeling (MLM) loss.
+These scripts leverage the 🤗 Datasets library and the Trainer API. You can easily customize them to your needs if you
+need extra processing on your datasets.
 
-Before running the following example, you should get a file that contains text on which the language model will be
-trained or fine-tuned. A good example of such text is the [WikiText-2 dataset](https://blog.einstein.ai/the-wikitext-long-term-dependency-language-modeling-dataset/).
+**Note:** The old script `run_language_modeling.py` is still available
+[here](https://github.com/huggingface/transformers/blob/master/examples/contrib/legacy/run_language_modeling.py).
 
-We will refer to two different files: `$TRAIN_FILE`, which contains text for training, and `$TEST_FILE`, which contains
-text that will be used for evaluation.
+The following examples, will run on a datasets hosted on our [hub](https://huggingface.co/datasets) or with your own
+text files for training and validation. We give examples of both below.
 
 ### GPT-2/GPT and causal language modeling
 
@@ -18,66 +21,105 @@ The following example fine-tunes GPT-2 on WikiText-2. We're using the raw WikiTe
 the tokenization). The loss here is that of causal language modeling.
 
 ```bash
-export TRAIN_FILE=/path/to/dataset/wiki.train.raw
-export TEST_FILE=/path/to/dataset/wiki.test.raw
-
-python run_language_modeling.py \
-    --output_dir=output \
-    --model_type=gpt2 \
-    --model_name_or_path=gpt2 \
+python run_clm.py \
+    --model_name_or_path gpt2 \
+    --dataset_name wikitext \
+    --dataset_config_name wikitext-2-raw-v1 \
     --do_train \
-    --train_data_file=$TRAIN_FILE \
     --do_eval \
-    --eval_data_file=$TEST_FILE
+    --output_dir /tmp/test-clm
 ```
 
 This takes about half an hour to train on a single K80 GPU and about one minute for the evaluation to run. It reaches
 a score of ~20 perplexity once fine-tuned on the dataset.
 
+To run on your own training and validation files, use the following command:
+
+```bash
+python run_clm.py \
+    --model_name_or_path gpt2 \
+    --train_file path_to_train_file \
+    --validation_file path_to_validation_file \
+    --do_train \
+    --do_eval \
+    --output_dir /tmp/test-clm
+```
+
+
 ### RoBERTa/BERT/DistilBERT and masked language modeling
 
 The following example fine-tunes RoBERTa on WikiText-2. Here too, we're using the raw WikiText-2. The loss is different
 as BERT/RoBERTa have a bidirectional mechanism; we're therefore using the same loss that was used during their
 pre-training: masked language modeling.
 
-In accordance to the RoBERTa paper, we use dynamic masking rather than static masking. The model may, therefore, converge
-slightly slower (over-fitting takes more epochs).
-
-We use the `--mlm` flag so that the script may change its loss function.
-
-If using whole-word masking, use both the`--mlm` and `--wwm` flags.
+In accordance to the RoBERTa paper, we use dynamic masking rather than static masking. The model may, therefore,
+converge slightly slower (over-fitting takes more epochs).
 
 ```bash
-export TRAIN_FILE=/path/to/dataset/wiki.train.raw
-export TEST_FILE=/path/to/dataset/wiki.test.raw
-
-python run_language_modeling.py \
-    --output_dir=output \
-    --model_type=roberta \
-    --model_name_or_path=roberta-base \
+python run_mlm.py \
+    --model_name_or_path roberta-base \
+    --dataset_name wikitext \
+    --dataset_config_name wikitext-2-raw-v1 \
     --do_train \
-    --train_data_file=$TRAIN_FILE \
     --do_eval \
-    --eval_data_file=$TEST_FILE \
-    --mlm \
-    --whole_word_mask
+    --output_dir /tmp/test-mlm
 ```
 
-For Chinese models, it's same with English model with only `--mlm`. If using whole-word masking, we need to generate a reference files, because it's char level.
+To run on your own training and validation files, use the following command:
 
-**Q :** Why ref file ?
+```bash
+python run_mlm.py \
+    --model_name_or_path roberta-base \
+    --train_file path_to_train_file \
+    --validation_file path_to_validation_file \
+    --do_train \
+    --do_eval \
+    --output_dir /tmp/test-mlm
+```
 
-**A :** Suppose we have a Chinese sentence like : `我喜欢你` The original Chinese-BERT will tokenize it as `['我','喜','欢','你']` in char level.
-Actually, `喜欢` is a whole word. For whole word mask proxy, We need res like `['我','喜','##欢','你']`.
-So we need a ref file to tell model which pos of BERT original token should be added `##`.
+If your dataset is organized with one sample per line, you can use the `--line_by_line` flag (otherwise the script
+concatenates all texts and then splits them in blocks of the same length).
+
+**Note:** On TPU, you should use the flag `--pad_to_max_length` in conjunction with the `--line_by_line` flag to make
+sure all your batches have the same length.
+
+### Whole word masking
+
+The BERT authors released a new version of BERT using Whole Word Masking in May 2019. Instead of masking randomly
+selected tokens (which may be part of words), they mask randomly selected words (masking all the tokens corresponding
+to that word). This technique has been refined for Chinese in [this paper](https://arxiv.org/abs/1906.08101).
+
+To fine-tune a model using whole word masking, use the following script:
+
+python run_mlm_wwm.py \
+    --model_name_or_path roberta-base \
+    --dataset_name wikitext \
+    --dataset_config_name wikitext-2-raw-v1 \
+    --do_train \
+    --do_eval \
+    --output_dir /tmp/test-mlm-wwm
+```
+
+For Chinese models, we need to generate a reference files (which requires the ltp library), because it's tokenized at
+the character level.
+
+**Q :** Why a reference file?
+
+**A :** Suppose we have a Chinese sentence like: `我喜欢你` The original Chinese-BERT will tokenize it as
+`['我','喜','欢','你']` (character level). But `喜欢` is a whole word. For whole word masking proxy, we need a result
+like `['我','喜','##欢','你']`, so we need a reference file to tell the model which position of the BERT original token
+should be added `##`.
 
 **Q :** Why LTP ?
 
-**A :** Cause the best known Chinese WWM BERT is [Chinese-BERT-wwm](https://github.com/ymcui/Chinese-BERT-wwm) by HIT. It works well on so many Chines Task like CLUE (Chinese GLUE).
-They use LTP, so if we want to fine-tune their model, we need LTP.
+**A :** Cause the best known Chinese WWM BERT is [Chinese-BERT-wwm](https://github.com/ymcui/Chinese-BERT-wwm) by HIT.
+It works well on so many Chines Task like CLUE (Chinese GLUE). They use LTP, so if we want to fine-tune their model,
+we need LTP.
 
 Now LTP only only works well on `transformers==3.2.0`. So we don't add it to requirements.txt.
-You need to check to `3.2.0` for `run_chinese_ref.py`. And the code  could be found in `examples/contrib`.
+You need to create a separate environment with this version of Transformers to run the `run_chinese_ref.py` script that
+will create the reference files. The script is in `examples/contrib`. Once in the proper environment, run the
+following:
 
 
 ```bash
@@ -87,33 +129,29 @@ export BERT_RESOURCE=/path/to/bert/tokenizer
 export SAVE_PATH=/path/to/data/ref.txt
 
 python examples/contrib/run_chinese_ref.py \
-    --file_name=$TRAIN_FILE \
-    --ltp=$LTP_RESOURCE \
-    --bert=$BERT_RESOURCE \
-    --save_path=$SAVE_PATH 
+    --file_name=path_to_train_or_eval_file \
+    --ltp=path_to_ltp_tokenizer \
+    --bert=path_to_bert_tokenizer \
+    --save_path=path_to_reference_file
 ```
-Now Chinese Ref is only supported by `LineByLineWithRefDataset` Class, so we need add `line_by_line` flag: 
+
+Then you can run the script like this: 
 
 
 ```bash
-export TRAIN_FILE=/path/to/dataset/wiki.train.raw
-export TEST_FILE=/path/to/dataset/wiki.test.raw
-export REF_FILE=/path/to/ref.txt
-
-python run_language_modeling.py \
-    --output_dir=output \
-    --model_type=roberta \
-    --model_name_or_path=roberta-base \
+python run_mlm_wwm.py \
+    --model_name_or_path roberta-base \
+    --train_file path_to_train_file \
+    --validation_file path_to_validation_file \
+    --train_ref_file path_to_train_chinese_ref_file \
+    --validation_ref_file path_to_validation_chinese_ref_file \
     --do_train \
-    --train_data_file=$TRAIN_FILE \
-    --chinese_ref_file=$REF_FILE \
     --do_eval \
-    --eval_data_file=$TEST_FILE \
-    --mlm \
-    --line_by_line \
-    --whole_word_mask
+    --output_dir /tmp/test-mlm-wwm
 ```
 
+**Note:** On TPU, you should the flag `--pad_to_max_length` to make sure all your batches have the same length.
+
 ### XLNet and permutation language modeling
 
 XLNet uses a different training objective, which is permutation language modeling. It is an autoregressive method 
@@ -126,15 +164,32 @@ context length for permutation language modeling.
 The `--max_span_length` flag may also be used to limit the length of a span of masked tokens used 
 for permutation language modeling.
 
-```bash
-export TRAIN_FILE=/path/to/dataset/wiki.train.raw
-export TEST_FILE=/path/to/dataset/wiki.test.raw
+Here is how to fine-tun XLNet on wikitext-2:
 
-python run_language_modeling.py \
-    --output_dir=output \
+```bash
+python run_plm.py \
     --model_name_or_path=xlnet-base-cased \
+    --dataset_name wikitext \
+    --dataset_config_name wikitext-2-raw-v1 \
     --do_train \
-    --train_data_file=$TRAIN_FILE \
     --do_eval \
-    --eval_data_file=$TEST_FILE \
+    --output_dir /tmp/test-plm
 ```
+
+To fine-tune it on your own training and validation file, run:
+
+```bash
+python run_plm.py \
+    --model_name_or_path=xlnet-base-cased \
+    --train_file path_to_train_file \
+    --validation_file path_to_validation_file \
+    --do_train \
+    --do_eval \
+    --output_dir /tmp/test-plm
+```
+
+If your dataset is organized with one sample per line, you can use the `--line_by_line` flag (otherwise the script
+concatenates all texts and then splits them in blocks of the same length).
+
+**Note:** On TPU, you should use the flag `--pad_to_max_length` in conjunction with the `--line_by_line` flag to make
+sure all your batches have the same length.

examples/language-modeling/run_clm.py

@@ -175,10 +175,10 @@ def main():
 
     # 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
+    # (the dataset will be downloaded automatically from the datasets Hub).
     #
-    # For CSV/JSON files, this script will use the column called 'text' or the first column. You can easily tweak this
-    # behavior (see below)
+    # For CSV/JSON files, this script will use the column called 'text' or the first column if no column called
+    # 'text' is found. You can easily tweak this behavior (see below).
     #
     # In distributed training, the load_dataset function guarantee that only one local process can concurrently
     # download the dataset.
@@ -190,7 +190,7 @@ def main():
         if data_args.train_file is not None:
             data_files["train"] = data_args.train_file
         if data_args.validation_file is not None:
-            data_files["validation"] = data_args.train_file
+            data_files["validation"] = data_args.validation_file
         extension = data_args.train_file.split(".")[-1]
         if extension == "txt":
             extension = "text"
@@ -259,14 +259,14 @@ def main():
     )
 
     if data_args.block_size <= 0:
-        block_size = tokenizer.max_len
+        block_size = tokenizer.model_max_length
     else:
-        if data_args.block_size > tokenizer.max_len:
+        if data_args.block_size > tokenizer.model_max_length:
             logger.warn(
                 f"The block_size passed ({data_args.block_size}) is larger than the maximum length for the model"
-                f"({tokenizer.max_len}). Using block_size={tokenizer.max_len}."
+                f"({tokenizer.model_max_length}). Using block_size={tokenizer.model_max_length}."
             )
-        block_size = min(data_args.block_size, tokenizer.max_len)
+        block_size = min(data_args.block_size, tokenizer.model_max_length)
 
     # Main data processing function that will concatenate all texts from our dataset and generate chunks of block_size.
     def group_texts(examples):

examples/language-modeling/run_mlm.py

@@ -116,6 +116,17 @@ class DataTrainingArguments:
     mlm_probability: float = field(
         default=0.15, metadata={"help": "Ratio of tokens to mask for masked language modeling loss"}
     )
+    line_by_line: bool = field(
+        default=False,
+        metadata={"help": "Whether distinct lines of text in the dataset are to be handled as distinct sequences."},
+    )
+    pad_to_max_length: bool = field(
+        default=False,
+        metadata={
+            "help": "Whether to pad all samples to `max_seq_length`. "
+            "If False, will pad the samples dynamically when batching to the maximum length in the batch."
+        },
+    )
 
     def __post_init__(self):
         if self.dataset_name is None and self.train_file is None and self.validation_file is None:
@@ -190,7 +201,7 @@ def main():
         if data_args.train_file is not None:
             data_files["train"] = data_args.train_file
         if data_args.validation_file is not None:
-            data_files["validation"] = data_args.train_file
+            data_files["validation"] = data_args.validation_file
         extension = data_args.train_file.split(".")[-1]
         if extension == "txt":
             extension = "text"
@@ -246,18 +257,83 @@ def main():
         column_names = datasets["validation"].column_names
     text_column_name = "text" if "text" in column_names else column_names[0]
 
-    def tokenize_function(examples):
-        # Remove empty lines
-        examples["text"] = [line for line in examples["text"] if len(line) > 0 and not line.isspace()]
-        return tokenizer(examples["text"], truncation=True, max_length=data_args.max_seq_length)
+    if data_args.line_by_line:
+        # When using line_by_line, we just tokenize each nonempty line.
+        padding = "max_length" if data_args.pad_to_max_length else False
 
-    tokenized_datasets = datasets.map(
-        tokenize_function,
-        batched=True,
-        num_proc=data_args.preprocessing_num_workers,
-        remove_columns=[text_column_name],
-        load_from_cache_file=not data_args.overwrite_cache,
-    )
+        def tokenize_function(examples):
+            # Remove empty lines
+            examples["text"] = [line for line in examples["text"] if len(line) > 0 and not line.isspace()]
+            return tokenizer(
+                examples["text"],
+                padding=padding,
+                truncation=True,
+                max_length=data_args.max_seq_length,
+                # We use this option because DataCollatorForLanguageModeling (see below) is more efficient when it
+                # receives the `special_tokens_mask`.
+                return_special_tokens_mask=True,
+            )
+
+        tokenized_datasets = datasets.map(
+            tokenize_function,
+            batched=True,
+            num_proc=data_args.preprocessing_num_workers,
+            remove_columns=[text_column_name],
+            load_from_cache_file=not data_args.overwrite_cache,
+        )
+    else:
+        # Otherwise, we tokenize every text, then concatenate them together before splitting them in smaller parts.
+        # We use `return_special_tokens_mask=True` because DataCollatorForLanguageModeling (see below) is more
+        # efficient when it receives the `special_tokens_mask`.
+        def tokenize_function(examples):
+            return tokenizer(examples[text_column_name], return_special_tokens_mask=True)
+
+        tokenized_datasets = datasets.map(
+            tokenize_function,
+            batched=True,
+            num_proc=data_args.preprocessing_num_workers,
+            remove_columns=[text_column_name],
+            load_from_cache_file=not data_args.overwrite_cache,
+        )
+
+        if data_args.max_seq_length is None:
+            max_seq_length = tokenizer.model_max_length
+        else:
+            if data_args.max_seq_length > tokenizer.model_max_length:
+                logger.warn(
+                    f"The max_seq_length passed ({data_args.max_seq_length}) is larger than the maximum length for the"
+                    f"model ({tokenizer.model_max_length}). Using max_seq_length={tokenizer.model_max_length}."
+                )
+            max_seq_length = min(data_args.max_seq_length, tokenizer.model_max_length)
+
+        # Main data processing function that will concatenate all texts from our dataset and generate chunks of
+        # max_seq_length.
+        def group_texts(examples):
+            # Concatenate all texts.
+            concatenated_examples = {k: sum(examples[k], []) for k in examples.keys()}
+            total_length = len(concatenated_examples[list(examples.keys())[0]])
+            # We drop the small remainder, we could add padding if the model supported it instead of this drop, you can
+            # customize this part to your needs.
+            total_length = (total_length // max_seq_length) * max_seq_length
+            # Split by chunks of max_len.
+            result = {
+                k: [t[i : i + max_seq_length] for i in range(0, total_length, max_seq_length)]
+                for k, t in concatenated_examples.items()
+            }
+            return result
+
+        # Note that with `batched=True`, this map processes 1,000 texts together, so group_texts throws away a
+        # remainder for each of those groups of 1,000 texts. You can adjust that batch_size here but a higher value
+        # might be slower to preprocess.
+        #
+        # To speed up this part, we use multiprocessing. See the documentation of the map method for more information:
+        # https://huggingface.co/docs/datasets/package_reference/main_classes.html#datasets.Dataset.map
+        tokenized_datasets = tokenized_datasets.map(
+            group_texts,
+            batched=True,
+            num_proc=data_args.preprocessing_num_workers,
+            load_from_cache_file=not data_args.overwrite_cache,
+        )
 
     # Data collator
     # This one will take care of randomly masking the tokens.

examples/language-modeling/run_mlm_wwm.py

@@ -0,0 +1,334 @@
+# coding=utf-8
+# 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.
+"""
+Fine-tuning the library models for masked language modeling (BERT, ALBERT, RoBERTa...) with whole word masking on a
+text file or a dataset.
+
+Here is the full list of checkpoints on the hub that can be fine-tuned by this script:
+https://huggingface.co/models?filter=masked-lm
+"""
+# You can also adapt this script on your own masked language modeling task. Pointers for this are left as comments.
+
+import json
+import logging
+import math
+import os
+import sys
+from dataclasses import dataclass, field
+from typing import Optional
+
+from datasets import Dataset, load_dataset
+
+import transformers
+from transformers import (
+    CONFIG_MAPPING,
+    MODEL_FOR_MASKED_LM_MAPPING,
+    AutoConfig,
+    AutoModelForMaskedLM,
+    AutoTokenizer,
+    DataCollatorForWholeWordMask,
+    HfArgumentParser,
+    Trainer,
+    TrainingArguments,
+    set_seed,
+)
+from transformers.trainer_utils import is_main_process
+
+
+logger = logging.getLogger(__name__)
+MODEL_CONFIG_CLASSES = list(MODEL_FOR_MASKED_LM_MAPPING.keys())
+MODEL_TYPES = tuple(conf.model_type for conf in MODEL_CONFIG_CLASSES)
+
+
+@dataclass
+class ModelArguments:
+    """
+    Arguments pertaining to which model/config/tokenizer we are going to fine-tune, or train from scratch.
+    """
+
+    model_name_or_path: Optional[str] = field(
+        default=None,
+        metadata={
+            "help": "The model checkpoint for weights initialization."
+            "Don't set if you want to train a model from scratch."
+        },
+    )
+    model_type: Optional[str] = field(
+        default=None,
+        metadata={"help": "If training from scratch, pass a model type from the list: " + ", ".join(MODEL_TYPES)},
+    )
+    config_name: Optional[str] = field(
+        default=None, metadata={"help": "Pretrained config name or path if not the same as model_name"}
+    )
+    tokenizer_name: Optional[str] = field(
+        default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
+    )
+    cache_dir: Optional[str] = field(
+        default=None, metadata={"help": "Where do you want to store the pretrained models downloaded from s3"}
+    )
+    use_fast_tokenizer: bool = field(
+        default=True,
+        metadata={"help": "Whether to use one of the fast tokenizer (backed by the tokenizers library) or not."},
+    )
+
+
+@dataclass
+class DataTrainingArguments:
+    """
+    Arguments pertaining to what data we are going to input our model for training and eval.
+    """
+
+    train_file: Optional[str] = field(default=None, metadata={"help": "The input training data file (a text file)."})
+    validation_file: Optional[str] = field(
+        default=None,
+        metadata={"help": "An optional input evaluation data file to evaluate the perplexity on (a text file)."},
+    )
+    train_ref_file: Optional[str] = field(
+        default=None,
+        metadata={"help": "An optional input train ref data file for whole word masking in Chinese."},
+    )
+    validation_ref_file: Optional[str] = field(
+        default=None,
+        metadata={"help": "An optional input validation ref data file for whole word masking in Chinese."},
+    )
+    overwrite_cache: bool = field(
+        default=False, metadata={"help": "Overwrite the cached training and evaluation sets"}
+    )
+    max_seq_length: Optional[int] = field(
+        default=None,
+        metadata={
+            "help": "The maximum total input sequence length after tokenization. Sequences longer "
+            "than this will be truncated. Default to the max input length of the model."
+        },
+    )
+    preprocessing_num_workers: Optional[int] = field(
+        default=None,
+        metadata={"help": "The number of processes to use for the preprocessing."},
+    )
+    mlm_probability: float = field(
+        default=0.15, metadata={"help": "Ratio of tokens to mask for masked language modeling loss"}
+    )
+    pad_to_max_length: bool = field(
+        default=False,
+        metadata={
+            "help": "Whether to pad all samples to `max_seq_length`. "
+            "If False, will pad the samples dynamically when batching to the maximum length in the batch."
+        },
+    )
+
+    def __post_init__(self):
+        if self.train_file is not None:
+            extension = self.train_file.split(".")[-1]
+            assert extension in ["csv", "json", "txt"], "`train_file` should be a csv, a json or a txt file."
+        if self.validation_file is not None:
+            extension = self.validation_file.split(".")[-1]
+            assert extension in ["csv", "json", "txt"], "`validation_file` should be a csv, a json or a txt file."
+
+
+def add_chinese_references(dataset, ref_file):
+    with open(ref_file, "r", encoding="utf-8") as f:
+        refs = [json.loads(line) for line in f.read().splitlines() if (len(line) > 0 and not line.isspace())]
+    assert len(dataset) == len(refs)
+
+    dataset_dict = {c: dataset[c] for c in dataset.column_names}
+    dataset_dict["chinese_ref"] = refs
+    return Dataset.from_dict(dataset_dict)
+
+
+def main():
+    # See all possible arguments in src/transformers/training_args.py
+    # or by passing the --help flag to this script.
+    # We now keep distinct sets of args, for a cleaner separation of concerns.
+
+    parser = HfArgumentParser((ModelArguments, DataTrainingArguments, TrainingArguments))
+    if len(sys.argv) == 2 and sys.argv[1].endswith(".json"):
+        # If we pass only one argument to the script and it's the path to a json file,
+        # let's parse it to get our arguments.
+        model_args, data_args, training_args = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1]))
+    else:
+        model_args, data_args, training_args = parser.parse_args_into_dataclasses()
+
+    if (
+        os.path.exists(training_args.output_dir)
+        and os.listdir(training_args.output_dir)
+        and training_args.do_train
+        and not training_args.overwrite_output_dir
+    ):
+        raise ValueError(
+            f"Output directory ({training_args.output_dir}) already exists and is not empty."
+            "Use --overwrite_output_dir to overcome."
+        )
+
+    # Setup logging
+    logging.basicConfig(
+        format="%(asctime)s - %(levelname)s - %(name)s -   %(message)s",
+        datefmt="%m/%d/%Y %H:%M:%S",
+        level=logging.INFO if is_main_process(training_args.local_rank) else logging.WARN,
+    )
+
+    # Log on each process the small summary:
+    logger.warning(
+        f"Process rank: {training_args.local_rank}, device: {training_args.device}, n_gpu: {training_args.n_gpu}"
+        + f"distributed training: {bool(training_args.local_rank != -1)}, 16-bits training: {training_args.fp16}"
+    )
+    # Set the verbosity to info of the Transformers logger (on main process only):
+    if is_main_process(training_args.local_rank):
+        transformers.utils.logging.set_verbosity_info()
+    logger.info("Training/evaluation parameters %s", 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).
+    #
+    # For CSV/JSON files, this script will use the column called 'text' or the first column if no column called
+    # 'text' is found. You can easily tweak this behavior (see below).
+    #
+    # In distributed training, the load_dataset function guarantee that only one local process can concurrently
+    # download the dataset.
+    data_files = {}
+    if data_args.train_file is not None:
+        data_files["train"] = data_args.train_file
+    if data_args.validation_file is not None:
+        data_files["validation"] = data_args.validation_file
+    extension = data_args.train_file.split(".")[-1]
+    if extension == "txt":
+        extension = "text"
+    datasets = load_dataset(extension, data_files=data_files)
+    # 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.
+
+    # Load pretrained model and tokenizer
+    #
+    # Distributed training:
+    # The .from_pretrained methods guarantee that only one local process can concurrently
+    # download model & vocab.
+    if model_args.config_name:
+        config = AutoConfig.from_pretrained(model_args.config_name, cache_dir=model_args.cache_dir)
+    elif model_args.model_name_or_path:
+        config = AutoConfig.from_pretrained(model_args.model_name_or_path, cache_dir=model_args.cache_dir)
+    else:
+        config = CONFIG_MAPPING[model_args.model_type]()
+        logger.warning("You are instantiating a new config instance from scratch.")
+
+    if model_args.tokenizer_name:
+        tokenizer = AutoTokenizer.from_pretrained(
+            model_args.tokenizer_name, cache_dir=model_args.cache_dir, use_fast=model_args.use_fast_tokenizer
+        )
+    elif model_args.model_name_or_path:
+        tokenizer = AutoTokenizer.from_pretrained(
+            model_args.model_name_or_path, cache_dir=model_args.cache_dir, use_fast=model_args.use_fast_tokenizer
+        )
+    else:
+        raise ValueError(
+            "You are instantiating a new tokenizer from scratch. This is not supported by this script."
+            "You can do it from another script, save it, and load it from here, using --tokenizer_name."
+        )
+
+    if model_args.model_name_or_path:
+        model = AutoModelForMaskedLM.from_pretrained(
+            model_args.model_name_or_path,
+            from_tf=bool(".ckpt" in model_args.model_name_or_path),
+            config=config,
+            cache_dir=model_args.cache_dir,
+        )
+    else:
+        logger.info("Training new model from scratch")
+        model = AutoModelForMaskedLM.from_config(config)
+
+    model.resize_token_embeddings(len(tokenizer))
+
+    # Preprocessing the datasets.
+    # First we tokenize all the texts.
+    if training_args.do_train:
+        column_names = datasets["train"].column_names
+    else:
+        column_names = datasets["validation"].column_names
+    text_column_name = "text" if "text" in column_names else column_names[0]
+
+    padding = "max_length" if data_args.pad_to_max_length else False
+
+    def tokenize_function(examples):
+        # Remove empty lines
+        examples["text"] = [line for line in examples["text"] if len(line) > 0 and not line.isspace()]
+        return tokenizer(examples["text"], padding=padding, truncation=True, max_length=data_args.max_seq_length)
+
+    tokenized_datasets = datasets.map(
+        tokenize_function,
+        batched=True,
+        num_proc=data_args.preprocessing_num_workers,
+        remove_columns=[text_column_name],
+        load_from_cache_file=not data_args.overwrite_cache,
+    )
+
+    # Add the chinese references if provided
+    if data_args.train_ref_file is not None:
+        tokenized_datasets["train"] = add_chinese_references(tokenized_datasets["train"], data_args.train_ref_file)
+    if data_args.valid_ref_file is not None:
+        tokenized_datasets["validation"] = add_chinese_references(
+            tokenized_datasets["validation"], data_args.validation_ref_file
+        )
+
+    # Data collator
+    # This one will take care of randomly masking the tokens.
+    data_collator = DataCollatorForWholeWordMask(tokenizer=tokenizer, mlm_probability=data_args.mlm_probability)
+
+    # Initialize our Trainer
+    trainer = Trainer(
+        model=model,
+        args=training_args,
+        train_dataset=tokenized_datasets["train"] if training_args.do_train else None,
+        eval_dataset=tokenized_datasets["validation"] if training_args.do_eval else None,
+        tokenizer=tokenizer,
+        data_collator=data_collator,
+    )
+
+    # Training
+    if training_args.do_train:
+        trainer.train(
+            model_path=model_args.model_name_or_path if os.path.isdir(model_args.model_name_or_path) else None
+        )
+        trainer.save_model()  # Saves the tokenizer too for easy upload
+
+    # Evaluation
+    results = {}
+    if training_args.do_eval:
+        logger.info("*** Evaluate ***")
+
+        eval_output = trainer.evaluate()
+
+        perplexity = math.exp(eval_output["eval_loss"])
+        results["perplexity"] = perplexity
+
+        output_eval_file = os.path.join(training_args.output_dir, "eval_results_mlm_wwm.txt")
+        if trainer.is_world_process_zero():
+            with open(output_eval_file, "w") as writer:
+                logger.info("***** Eval results *****")
+                for key, value in results.items():
+                    logger.info(f"  {key} = {value}")
+                    writer.write(f"{key} = {value}\n")
+
+    return results
+
+
+def _mp_fn(index):
+    # For xla_spawn (TPUs)
+    main()
+
+
+if __name__ == "__main__":
+    main()

examples/language-modeling/run_plm.py

@@ -96,7 +96,7 @@ class DataTrainingArguments:
         default=None,
         metadata={
             "help": "The maximum total input sequence length after tokenization. Sequences longer "
-            "than this will be truncated."
+            "than this will be truncated. Default to the max input length of the model."
         },
     )
     preprocessing_num_workers: Optional[int] = field(
@@ -113,6 +113,17 @@ class DataTrainingArguments:
     max_span_length: int = field(
         default=5, metadata={"help": "Maximum length of a span of masked tokens for permutation language modeling."}
     )
+    line_by_line: bool = field(
+        default=False,
+        metadata={"help": "Whether distinct lines of text in the dataset are to be handled as distinct sequences."},
+    )
+    pad_to_max_length: bool = field(
+        default=False,
+        metadata={
+            "help": "Whether to pad all samples to `max_seq_length`. "
+            "If False, will pad the samples dynamically when batching to the maximum length in the batch."
+        },
+    )
 
     def __post_init__(self):
         if self.dataset_name is None and self.train_file is None and self.validation_file is None:
@@ -172,10 +183,10 @@ def main():
 
     # 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
+    # (the dataset will be downloaded automatically from the datasets Hub).
     #
-    # For CSV/JSON files, this script will use the column called 'text' or the first column. You can easily tweak this
-    # behavior (see below)
+    # For CSV/JSON files, this script will use the column called 'text' or the first column if no column called
+    # 'text' is found. You can easily tweak this behavior (see below).
     #
     # In distributed training, the load_dataset function guarantee that only one local process can concurrently
     # download the dataset.
@@ -187,7 +198,7 @@ def main():
         if data_args.train_file is not None:
             data_files["train"] = data_args.train_file
         if data_args.validation_file is not None:
-            data_files["validation"] = data_args.train_file
+            data_files["validation"] = data_args.validation_file
         extension = data_args.train_file.split(".")[-1]
         if extension == "txt":
             extension = "text"
@@ -243,18 +254,73 @@ def main():
         column_names = datasets["validation"].column_names
     text_column_name = "text" if "text" in column_names else column_names[0]
 
-    def tokenize_function(examples):
-        # Remove empty lines
-        examples["text"] = [line for line in examples["text"] if len(line) > 0 and not line.isspace()]
-        return tokenizer(examples["text"], truncation=True, max_length=data_args.max_seq_length)
+    if data_args.line_by_line:
+        # When using line_by_line, we just tokenize each nonempty line.
+        padding = "max_length" if data_args.pad_to_max_length else False
 
-    tokenized_datasets = datasets.map(
-        tokenize_function,
-        batched=True,
-        num_proc=data_args.preprocessing_num_workers,
-        remove_columns=[text_column_name],
-        load_from_cache_file=not data_args.overwrite_cache,
-    )
+        def tokenize_function(examples):
+            # Remove empty lines
+            examples["text"] = [line for line in examples["text"] if len(line) > 0 and not line.isspace()]
+            return tokenizer(examples["text"], padding=padding, truncation=True, max_length=data_args.max_seq_length)
+
+        tokenized_datasets = datasets.map(
+            tokenize_function,
+            batched=True,
+            num_proc=data_args.preprocessing_num_workers,
+            remove_columns=[text_column_name],
+            load_from_cache_file=not data_args.overwrite_cache,
+        )
+    else:
+        # Otherwise, we tokenize every text, then concatenate them together before splitting them in smaller parts.
+        def tokenize_function(examples):
+            return tokenizer(examples[text_column_name])
+
+        tokenized_datasets = datasets.map(
+            tokenize_function,
+            batched=True,
+            num_proc=data_args.preprocessing_num_workers,
+            remove_columns=[text_column_name],
+            load_from_cache_file=not data_args.overwrite_cache,
+        )
+
+        if data_args.max_seq_length is None:
+            max_seq_length = tokenizer.model_max_length
+        else:
+            if data_args.max_seq_length > tokenizer.model_max_length:
+                logger.warn(
+                    f"The max_seq_length passed ({data_args.max_seq_length}) is larger than the maximum length for the"
+                    f"model ({tokenizer.model_max_length}). Using max_seq_length={tokenizer.model_max_length}."
+                )
+            max_seq_length = min(data_args.max_seq_length, tokenizer.model_max_length)
+
+        # Main data processing function that will concatenate all texts from our dataset and generate chunks of
+        # max_seq_length.
+        def group_texts(examples):
+            # Concatenate all texts.
+            concatenated_examples = {k: sum(examples[k], []) for k in examples.keys()}
+            total_length = len(concatenated_examples[list(examples.keys())[0]])
+            # We drop the small remainder, we could add padding if the model supported it instead of this drop, you can
+            # customize this part to your needs.
+            total_length = (total_length // max_seq_length) * max_seq_length
+            # Split by chunks of max_len.
+            result = {
+                k: [t[i : i + max_seq_length] for i in range(0, total_length, max_seq_length)]
+                for k, t in concatenated_examples.items()
+            }
+            return result
+
+        # Note that with `batched=True`, this map processes 1,000 texts together, so group_texts throws away a
+        # remainder for each of those groups of 1,000 texts. You can adjust that batch_size here but a higher value
+        # might be slower to preprocess.
+        #
+        # To speed up this part, we use multiprocessing. See the documentation of the map method for more information:
+        # https://huggingface.co/docs/datasets/package_reference/main_classes.html#datasets.Dataset.map
+        tokenized_datasets = tokenized_datasets.map(
+            group_texts,
+            batched=True,
+            num_proc=data_args.preprocessing_num_workers,
+            load_from_cache_file=not data_args.overwrite_cache,
+        )
 
     # Data collator
     data_collator = DataCollatorForPermutationLanguageModeling(

examples/lightning_base.py

@@ -7,6 +7,7 @@ from typing import Any, Dict
 import pytorch_lightning as pl
 from pytorch_lightning.utilities import rank_zero_info
 
+import pkg_resources
 from transformers import (
     AdamW,
     AutoConfig,
@@ -32,6 +33,15 @@ from transformers.optimization import (
 
 logger = logging.getLogger(__name__)
 
+try:
+    pkg = "pytorch_lightning"
+    min_ver = "1.0.4"
+    pkg_resources.require(f"{pkg}>={min_ver}")
+except pkg_resources.VersionConflict:
+    logger.warning(
+        f"{pkg}>={min_ver} is required for a normal functioning of this module, but found {pkg}=={pkg_resources.get_distribution(pkg).version}. Try pip install -r examples/requirements.txt"
+    )
+
 
 MODEL_MODES = {
     "base": AutoModel,

examples/lxmert/modeling_frcnn.py

@@ -1801,7 +1801,7 @@ class GeneralizedRCNN(nn.Module):
                 f"Some weights of the model checkpoint at {pretrained_model_name_or_path} were not used when "
                 f"initializing {model.__class__.__name__}: {unexpected_keys}\n"
                 f"- This IS expected if you are initializing {model.__class__.__name__} from the checkpoint of a model trained on another task "
-                f"or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPretraining model).\n"
+                f"or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).\n"
                 f"- This IS NOT expected if you are initializing {model.__class__.__name__} from the checkpoint of a model that you expect "
                 f"to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model)."
             )

examples/rag/test_distributed_retriever.py

@@ -16,6 +16,7 @@ from transformers.configuration_dpr import DPRConfig
 from transformers.configuration_rag import RagConfig
 from transformers.file_utils import is_datasets_available, is_faiss_available, is_psutil_available, is_torch_available
 from transformers.retrieval_rag import CustomHFIndex
+from transformers.testing_utils import require_torch_non_multigpu_but_fix_me
 from transformers.tokenization_bart import BartTokenizer
 from transformers.tokenization_bert import VOCAB_FILES_NAMES as DPR_VOCAB_FILES_NAMES
 from transformers.tokenization_dpr import DPRQuestionEncoderTokenizer
@@ -178,6 +179,7 @@ class RagRetrieverTest(TestCase):
             retriever.init_retrieval(port)
         return retriever
 
+    @require_torch_non_multigpu_but_fix_me
     def test_pytorch_distributed_retriever_retrieve(self):
         n_docs = 1
         retriever = self.get_dummy_pytorch_distributed_retriever(init_retrieval=True)
@@ -193,6 +195,7 @@ class RagRetrieverTest(TestCase):
         self.assertEqual(doc_dicts[1]["id"][0], "0")  # max inner product is reached with first doc
         self.assertListEqual(doc_ids.tolist(), [[1], [0]])
 
+    @require_torch_non_multigpu_but_fix_me
     def test_custom_hf_index_retriever_retrieve(self):
         n_docs = 1
         retriever = self.get_dummy_custom_hf_index_retriever(init_retrieval=True, from_disk=False)
@@ -208,6 +211,7 @@ class RagRetrieverTest(TestCase):
         self.assertEqual(doc_dicts[1]["id"][0], "0")  # max inner product is reached with first doc
         self.assertListEqual(doc_ids.tolist(), [[1], [0]])
 
+    @require_torch_non_multigpu_but_fix_me
     def test_custom_pytorch_distributed_retriever_retrieve_from_disk(self):
         n_docs = 1
         retriever = self.get_dummy_custom_hf_index_retriever(init_retrieval=True, from_disk=True)

examples/seq2seq/README.md

@@ -1,7 +1,7 @@
 ## Sequence to Sequence Training and Evaluation
 
 This directory contains examples for finetuning and evaluating transformers on summarization and translation tasks.
-Please tag @sshleifer with any issues/unexpected behaviors, or send a PR!
+Please tag @patil-suraj with any issues/unexpected behaviors, or send a PR!
 For deprecated `bertabs` instructions, see [`bertabs/README.md`](bertabs/README.md).
 
 ### Supported Architectures

examples/seq2seq/bertabs/configuration_bertabs.py

@@ -23,7 +23,7 @@ logger = logging.getLogger(__name__)
 
 
 BERTABS_FINETUNED_CONFIG_MAP = {
-    "bertabs-finetuned-cnndm": "https://s3.amazonaws.com/models.huggingface.co/bert/remi/bertabs-finetuned-cnndm-extractive-abstractive-summarization/config.json",
+    "bertabs-finetuned-cnndm": "https://huggingface.co/remi/bertabs-finetuned-cnndm-extractive-abstractive-summarization/resolve/main/config.json",
 }
 
 

examples/seq2seq/finetune_trainer.py

@@ -4,16 +4,9 @@ import sys
 from dataclasses import dataclass, field
 from typing import Optional
 
-from seq2seq_trainer import Seq2SeqTrainer, arg_to_scheduler_choices
-from transformers import (
-    AutoConfig,
-    AutoModelForSeq2SeqLM,
-    AutoTokenizer,
-    HfArgumentParser,
-    MBartTokenizer,
-    TrainingArguments,
-    set_seed,
-)
+from seq2seq_trainer import Seq2SeqTrainer
+from seq2seq_training_args import Seq2SeqTrainingArguments
+from transformers import AutoConfig, AutoModelForSeq2SeqLM, AutoTokenizer, HfArgumentParser, MBartTokenizer, set_seed
 from transformers.trainer_utils import EvaluationStrategy
 from utils import (
     Seq2SeqDataCollator,
@@ -33,41 +26,6 @@ from utils import (
 logger = logging.getLogger(__name__)
 
 
-@dataclass
-class Seq2SeqTrainingArguments(TrainingArguments):
-    """
-    Parameters:
-        label_smoothing (:obj:`float`, `optional`, defaults to 0):
-            The label smoothing epsilon to apply (if not zero).
-        sortish_sampler (:obj:`bool`, `optional`, defaults to :obj:`False`):
-            Whether to SortishSamler or not. It sorts the inputs according to lenghts in-order to minimizing the padding size.
-        predict_with_generate (:obj:`bool`, `optional`, defaults to :obj:`False`):
-            Whether to use generate to calculate generative metrics (ROUGE, BLEU).
-    """
-
-    label_smoothing: Optional[float] = field(
-        default=0.0, metadata={"help": "The label smoothing epsilon to apply (if not zero)."}
-    )
-    sortish_sampler: bool = field(default=False, metadata={"help": "Whether to SortishSamler or not."})
-    predict_with_generate: bool = field(
-        default=False, metadata={"help": "Whether to use generate to calculate generative metrics (ROUGE, BLEU)."}
-    )
-    adafactor: bool = field(default=False, metadata={"help": "whether to use adafactor"})
-    encoder_layerdrop: Optional[float] = field(
-        default=None, metadata={"help": "Encoder layer dropout probability. Goes into model.config."}
-    )
-    decoder_layerdrop: Optional[float] = field(
-        default=None, metadata={"help": "Decoder layer dropout probability. Goes into model.config."}
-    )
-    dropout: Optional[float] = field(default=None, metadata={"help": "Dropout probability. Goes into model.config."})
-    attention_dropout: Optional[float] = field(
-        default=None, metadata={"help": "Attention dropout probability. Goes into model.config."}
-    )
-    lr_scheduler: Optional[str] = field(
-        default="linear", metadata={"help": f"Which lr scheduler to use. Selected in {arg_to_scheduler_choices}"}
-    )
-
-
 @dataclass
 class ModelArguments:
     """

examples/seq2seq/run_distiller.sh

@@ -1,10 +0,0 @@
-# Add parent directory to python path to access lightning_base.py
-export PYTHONPATH="../":"${PYTHONPATH}"
-
-python distillation.py \
---learning_rate=3e-4 \
---do_train \
---do_predict \
---fp16 \
---val_check_interval 0.1 \
-"$@"

examples/seq2seq/seq2seq_trainer.py

@@ -20,12 +20,6 @@ from transformers.optimization import (
 from transformers.trainer_pt_utils import get_tpu_sampler
 
 
-try:
-    from .utils import label_smoothed_nll_loss
-except ImportError:
-    from utils import label_smoothed_nll_loss
-
-
 logger = logging.get_logger(__name__)
 
 arg_to_scheduler = {
@@ -36,7 +30,6 @@ arg_to_scheduler = {
     "constant": get_constant_schedule,
     "constant_w_warmup": get_constant_schedule_with_warmup,
 }
-arg_to_scheduler_choices = sorted(arg_to_scheduler.keys())
 
 
 class Seq2SeqTrainer(Trainer):
@@ -64,6 +57,14 @@ class Seq2SeqTrainer(Trainer):
                 f"The `config.pad_token_id` is `None`. Using `config.eos_token_id` = {self.config.eos_token_id} for padding.."
             )
 
+        if self.args.label_smoothing == 0:
+            self.loss_fn = torch.nn.CrossEntropyLoss(ignore_index=self.config.pad_token_id)
+        else:
+            # dynamically import label_smoothed_nll_loss
+            from utils import label_smoothed_nll_loss
+
+            self.loss_fn = label_smoothed_nll_loss
+
     def create_optimizer_and_scheduler(self, num_training_steps: int):
         """
         Setup the optimizer and the learning rate scheduler.
@@ -135,9 +136,7 @@ class Seq2SeqTrainer(Trainer):
             if self.data_args is not None and self.data_args.ignore_pad_token_for_loss:
                 # force training to ignore pad token
                 logits = model(**inputs, use_cache=False)[0]
-
-                loss_fct = torch.nn.CrossEntropyLoss(ignore_index=self.config.pad_token_id)
-                loss = loss_fct(logits.view(-1, logits.shape[-1]), labels.view(-1))
+                loss = self.loss_fn(logits.view(-1, logits.shape[-1]), labels.view(-1))
             else:
                 # compute usual loss via models
                 loss, logits = model(**inputs, labels=labels, use_cache=False)[:2]
@@ -145,9 +144,7 @@ class Seq2SeqTrainer(Trainer):
             # compute label smoothed loss
             logits = model(**inputs, use_cache=False)[0]
             lprobs = torch.nn.functional.log_softmax(logits, dim=-1)
-            loss, _ = label_smoothed_nll_loss(
-                lprobs, labels, self.args.label_smoothing, ignore_index=self.config.pad_token_id
-            )
+            loss, _ = self.loss_fn(lprobs, labels, self.args.label_smoothing, ignore_index=self.config.pad_token_id)
         return loss, logits
 
     def compute_loss(self, model, inputs):

examples/seq2seq/seq2seq_training_args.py

@@ -0,0 +1,45 @@
+import logging
+from dataclasses import dataclass, field
+from typing import Optional
+
+from seq2seq_trainer import arg_to_scheduler
+from transformers import TrainingArguments
+
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass
+class Seq2SeqTrainingArguments(TrainingArguments):
+    """
+    Parameters:
+        label_smoothing (:obj:`float`, `optional`, defaults to 0):
+            The label smoothing epsilon to apply (if not zero).
+        sortish_sampler (:obj:`bool`, `optional`, defaults to :obj:`False`):
+            Whether to SortishSamler or not. It sorts the inputs according to lenghts in-order to minimizing the padding size.
+        predict_with_generate (:obj:`bool`, `optional`, defaults to :obj:`False`):
+            Whether to use generate to calculate generative metrics (ROUGE, BLEU).
+    """
+
+    label_smoothing: Optional[float] = field(
+        default=0.0, metadata={"help": "The label smoothing epsilon to apply (if not zero)."}
+    )
+    sortish_sampler: bool = field(default=False, metadata={"help": "Whether to SortishSamler or not."})
+    predict_with_generate: bool = field(
+        default=False, metadata={"help": "Whether to use generate to calculate generative metrics (ROUGE, BLEU)."}
+    )
+    adafactor: bool = field(default=False, metadata={"help": "whether to use adafactor"})
+    encoder_layerdrop: Optional[float] = field(
+        default=None, metadata={"help": "Encoder layer dropout probability. Goes into model.config."}
+    )
+    decoder_layerdrop: Optional[float] = field(
+        default=None, metadata={"help": "Decoder layer dropout probability. Goes into model.config."}
+    )
+    dropout: Optional[float] = field(default=None, metadata={"help": "Dropout probability. Goes into model.config."})
+    attention_dropout: Optional[float] = field(
+        default=None, metadata={"help": "Attention dropout probability. Goes into model.config."}
+    )
+    lr_scheduler: Optional[str] = field(
+        default="linear",
+        metadata={"help": f"Which lr scheduler to use. Selected in {sorted(arg_to_scheduler.keys())}"},
+    )

examples/seq2seq/test_bash_script.py

@@ -3,93 +3,110 @@
 import argparse
 import os
 import sys
-from pathlib import Path
 from unittest.mock import patch
 
-import pytest
 import pytorch_lightning as pl
 import timeout_decorator
 import torch
 
 from distillation import BartSummarizationDistiller, distill_main
 from finetune import SummarizationModule, main
-from test_seq2seq_examples import CUDA_AVAILABLE, MBART_TINY
-from transformers import BartForConditionalGeneration, MarianMTModel
-from transformers.testing_utils import TestCasePlus, slow
+from transformers import MarianMTModel
+from transformers.file_utils import cached_path
+from transformers.testing_utils import TestCasePlus, require_torch_gpu, require_torch_non_multigpu_but_fix_me, slow
 from utils import load_json
 
 
-MODEL_NAME = MBART_TINY
-MARIAN_MODEL = "sshleifer/student_marian_en_ro_6_1"
+MARIAN_MODEL = "sshleifer/mar_enro_6_3_student"
 
 
-class TestAll(TestCasePlus):
+class TestMbartCc25Enro(TestCasePlus):
+    def setUp(self):
+        super().setUp()
+
+        data_cached = cached_path(
+            "https://cdn-datasets.huggingface.co/translation/wmt_en_ro-tr40k-va0.5k-te0.5k.tar.gz",
+            extract_compressed_file=True,
+        )
+        self.data_dir = f"{data_cached}/wmt_en_ro-tr40k-va0.5k-te0.5k"
+
     @slow
-    @pytest.mark.skipif(not CUDA_AVAILABLE, reason="too slow to run on CPU")
+    @require_torch_gpu
+    @require_torch_non_multigpu_but_fix_me
     def test_model_download(self):
         """This warms up the cache so that we can time the next test without including download time, which varies between machines."""
-        BartForConditionalGeneration.from_pretrained(MODEL_NAME)
         MarianMTModel.from_pretrained(MARIAN_MODEL)
 
-    @timeout_decorator.timeout(120)
+    # @timeout_decorator.timeout(1200)
     @slow
-    @pytest.mark.skipif(not CUDA_AVAILABLE, reason="too slow to run on CPU")
+    @require_torch_gpu
+    @require_torch_non_multigpu_but_fix_me
     def test_train_mbart_cc25_enro_script(self):
-        data_dir = "examples/seq2seq/test_data/wmt_en_ro"
         env_vars_to_replace = {
-            "--fp16_opt_level=O1": "",
-            "$MAX_LEN": 128,
-            "$BS": 4,
+            "$MAX_LEN": 64,
+            "$BS": 64,
             "$GAS": 1,
-            "$ENRO_DIR": data_dir,
-            "facebook/mbart-large-cc25": MODEL_NAME,
-            # Download is 120MB in previous test.
-            "val_check_interval=0.25": "val_check_interval=1.0",
+            "$ENRO_DIR": self.data_dir,
+            "facebook/mbart-large-cc25": MARIAN_MODEL,
+            # "val_check_interval=0.25": "val_check_interval=1.0",
+            "--learning_rate=3e-5": "--learning_rate 3e-4",
+            "--num_train_epochs 6": "--num_train_epochs 1",
         }
 
         # Clean up bash script
-        bash_script = Path("examples/seq2seq/train_mbart_cc25_enro.sh").open().read().split("finetune.py")[1].strip()
+        bash_script = (self.test_file_dir / "train_mbart_cc25_enro.sh").open().read().split("finetune.py")[1].strip()
         bash_script = bash_script.replace("\\\n", "").strip().replace('"$@"', "")
         for k, v in env_vars_to_replace.items():
             bash_script = bash_script.replace(k, str(v))
         output_dir = self.get_auto_remove_tmp_dir()
 
-        bash_script = bash_script.replace("--fp16 ", "")
-        testargs = (
-            ["finetune.py"]
-            + bash_script.split()
-            + [
-                f"--output_dir={output_dir}",
-                "--gpus=1",
-                "--learning_rate=3e-1",
-                "--warmup_steps=0",
-                "--val_check_interval=1.0",
-                "--tokenizer_name=facebook/mbart-large-en-ro",
-            ]
-        )
+        # bash_script = bash_script.replace("--fp16 ", "")
+        args = f"""
+            --output_dir {output_dir}
+            --tokenizer_name Helsinki-NLP/opus-mt-en-ro
+            --sortish_sampler
+            --do_predict
+            --gpus 1
+            --freeze_encoder
+            --n_train 40000
+            --n_val 500
+            --n_test 500
+            --fp16_opt_level O1
+            --num_sanity_val_steps 0
+            --eval_beams 2
+        """.split()
+        # XXX: args.gpus > 1 : handle multigpu in the future
+
+        testargs = ["finetune.py"] + bash_script.split() + args
         with patch.object(sys, "argv", testargs):
             parser = argparse.ArgumentParser()
             parser = pl.Trainer.add_argparse_args(parser)
             parser = SummarizationModule.add_model_specific_args(parser, os.getcwd())
             args = parser.parse_args()
-            args.do_predict = False
-            # assert args.gpus == gpus THIS BREAKS for multigpu
             model = main(args)
 
         # Check metrics
         metrics = load_json(model.metrics_save_path)
         first_step_stats = metrics["val"][0]
         last_step_stats = metrics["val"][-1]
-        assert (
-            len(metrics["val"]) == (args.max_epochs / args.val_check_interval) + 1
-        )  # +1 accounts for val_sanity_check
-
-        assert last_step_stats["val_avg_gen_time"] >= 0.01
-
-        assert first_step_stats["val_avg_bleu"] < last_step_stats["val_avg_bleu"]  # model learned nothing
-        assert 1.0 >= last_step_stats["val_avg_gen_time"]  # model hanging on generate. Maybe bad config was saved.
+        self.assertEqual(len(metrics["val"]), (args.max_epochs / args.val_check_interval))
         assert isinstance(last_step_stats[f"val_avg_{model.val_metric}"], float)
 
+        self.assertGreater(last_step_stats["val_avg_gen_time"], 0.01)
+        # model hanging on generate. Maybe bad config was saved. (XXX: old comment/assert?)
+        self.assertLessEqual(last_step_stats["val_avg_gen_time"], 1.0)
+
+        # test learning requirements:
+
+        # 1. BLEU improves over the course of training by more than 2 pts
+        self.assertGreater(last_step_stats["val_avg_bleu"] - first_step_stats["val_avg_bleu"], 2)
+
+        # 2. BLEU finishes above 17
+        self.assertGreater(last_step_stats["val_avg_bleu"], 17)
+
+        # 3. test BLEU and val BLEU within ~1.1 pt.
+        self.assertLess(abs(metrics["val"][-1]["val_avg_bleu"] - metrics["test"][-1]["test_avg_bleu"]), 1.1)
+
         # check lightning ckpt can be loaded and has a reasonable statedict
         contents = os.listdir(output_dir)
         ckpt_path = [x for x in contents if x.endswith(".ckpt")][0]
@@ -107,11 +124,14 @@ class TestAll(TestCasePlus):
             # assert len(metrics["val"]) ==  desired_n_evals
             assert len(metrics["test"]) == 1
 
+
+class TestDistilMarianNoTeacher(TestCasePlus):
     @timeout_decorator.timeout(600)
     @slow
-    @pytest.mark.skipif(not CUDA_AVAILABLE, reason="too slow to run on CPU")
+    @require_torch_gpu
+    @require_torch_non_multigpu_but_fix_me
     def test_opus_mt_distill_script(self):
-        data_dir = "examples/seq2seq/test_data/wmt_en_ro"
+        data_dir = f"{self.test_file_dir_str}/test_data/wmt_en_ro"
         env_vars_to_replace = {
             "--fp16_opt_level=O1": "",
             "$MAX_LEN": 128,
@@ -124,7 +144,7 @@ class TestAll(TestCasePlus):
 
         # Clean up bash script
         bash_script = (
-            Path("examples/seq2seq/distil_marian_no_teacher.sh").open().read().split("distillation.py")[1].strip()
+            (self.test_file_dir / "distil_marian_no_teacher.sh").open().read().split("distillation.py")[1].strip()
         )
         bash_script = bash_script.replace("\\\n", "").strip().replace('"$@"', "")
         bash_script = bash_script.replace("--fp16 ", " ")

examples/seq2seq/test_datasets.py

@@ -11,7 +11,7 @@ from save_len_file import save_len_file
 from test_seq2seq_examples import ARTICLES, BART_TINY, MARIAN_TINY, MBART_TINY, SUMMARIES, T5_TINY, make_test_data_dir
 from transformers import AutoTokenizer
 from transformers.modeling_bart import shift_tokens_right
-from transformers.testing_utils import TestCasePlus, slow
+from transformers.testing_utils import TestCasePlus, require_torch_non_multigpu_but_fix_me, slow
 from utils import FAIRSEQ_AVAILABLE, DistributedSortishSampler, LegacySeq2SeqDataset, Seq2SeqDataset
 
 
@@ -30,6 +30,7 @@ class TestAll(TestCasePlus):
         ],
     )
     @slow
+    @require_torch_non_multigpu_but_fix_me
     def test_seq2seq_dataset_truncation(self, tok_name):
         tokenizer = AutoTokenizer.from_pretrained(tok_name)
         tmp_dir = make_test_data_dir(tmp_dir=self.get_auto_remove_tmp_dir())
@@ -69,6 +70,7 @@ class TestAll(TestCasePlus):
             break  # No need to test every batch
 
     @parameterized.expand([BART_TINY, BERT_BASE_CASED])
+    @require_torch_non_multigpu_but_fix_me
     def test_legacy_dataset_truncation(self, tok):
         tokenizer = AutoTokenizer.from_pretrained(tok)
         tmp_dir = make_test_data_dir(tmp_dir=self.get_auto_remove_tmp_dir())
@@ -93,6 +95,7 @@ class TestAll(TestCasePlus):
             assert max_len_target > trunc_target  # Truncated
             break  # No need to test every batch
 
+    @require_torch_non_multigpu_but_fix_me
     def test_pack_dataset(self):
         tokenizer = AutoTokenizer.from_pretrained("facebook/mbart-large-cc25")
 
@@ -111,6 +114,7 @@ class TestAll(TestCasePlus):
         assert orig_paths == new_paths
 
     @pytest.mark.skipif(not FAIRSEQ_AVAILABLE, reason="This test requires fairseq")
+    @require_torch_non_multigpu_but_fix_me
     def test_dynamic_batch_size(self):
         if not FAIRSEQ_AVAILABLE:
             return
@@ -135,6 +139,7 @@ class TestAll(TestCasePlus):
         if failures:
             raise AssertionError(f"too many tokens in {len(failures)} batches")
 
+    @require_torch_non_multigpu_but_fix_me
     def test_sortish_sampler_reduces_padding(self):
         ds, _, tokenizer = self._get_dataset(max_len=512)
         bs = 2
@@ -174,6 +179,7 @@ class TestAll(TestCasePlus):
         )
         return ds, max_tokens, tokenizer
 
+    @require_torch_non_multigpu_but_fix_me
     def test_distributed_sortish_sampler_splits_indices_between_procs(self):
         ds, max_tokens, tokenizer = self._get_dataset()
         ids1 = set(DistributedSortishSampler(ds, 256, num_replicas=2, rank=0, add_extra_examples=False))
@@ -189,6 +195,7 @@ class TestAll(TestCasePlus):
             PEGASUS_XSUM,
         ],
     )
+    @require_torch_non_multigpu_but_fix_me
     def test_dataset_kwargs(self, tok_name):
         tokenizer = AutoTokenizer.from_pretrained(tok_name)
         if tok_name == MBART_TINY:

examples/seq2seq/test_finetune_trainer.py

@@ -2,9 +2,9 @@ import os
 import sys
 from unittest.mock import patch
 
-from transformers import BertTokenizer, EncoderDecoderModel, is_torch_available
+from transformers import BertTokenizer, EncoderDecoderModel
 from transformers.file_utils import is_datasets_available
-from transformers.testing_utils import TestCasePlus, execute_subprocess_async, slow
+from transformers.testing_utils import TestCasePlus, execute_subprocess_async, get_gpu_count, slow
 from transformers.trainer_callback import TrainerState
 from transformers.trainer_utils import set_seed
 
@@ -13,9 +13,6 @@ from .seq2seq_trainer import Seq2SeqTrainer
 from .test_seq2seq_examples import MBART_TINY
 
 
-if is_torch_available():
-    import torch
-
 set_seed(42)
 MARIAN_MODEL = "sshleifer/student_marian_en_ro_6_1"
 
@@ -180,7 +177,7 @@ class TestFinetuneTrainer(TestCasePlus):
             --num_train_epochs {str(num_train_epochs)}
             --per_device_train_batch_size 4
             --per_device_eval_batch_size 4
-            --learning_rate 3e-4
+            --learning_rate 3e-3
             --warmup_steps 8
             --evaluate_during_training
             --predict_with_generate
@@ -196,7 +193,7 @@ class TestFinetuneTrainer(TestCasePlus):
         """.split()
         # --eval_beams  2
 
-        n_gpu = torch.cuda.device_count()
+        n_gpu = get_gpu_count()
         if n_gpu > 1:
             distributed_args = f"""
                 -m torch.distributed.launch

examples/seq2seq/test_fsmt_bleu_score.py

@@ -19,7 +19,13 @@ import unittest
 
 from parameterized import parameterized
 from transformers import FSMTForConditionalGeneration, FSMTTokenizer
-from transformers.testing_utils import get_tests_dir, require_torch, slow, torch_device
+from transformers.testing_utils import (
+    get_tests_dir,
+    require_torch,
+    require_torch_non_multigpu_but_fix_me,
+    slow,
+    torch_device,
+)
 from utils import calculate_bleu
 
 
@@ -48,6 +54,7 @@ class ModelEvalTester(unittest.TestCase):
         ]
     )
     @slow
+    @require_torch_non_multigpu_but_fix_me
     def test_bleu_scores(self, pair, min_bleu_score):
         # note: this test is not testing the best performance since it only evals a small batch
         # but it should be enough to detect a regression in the output quality

examples/seq2seq/test_make_student.py

@@ -4,7 +4,7 @@ import unittest
 from make_student import create_student_by_copying_alternating_layers
 from transformers import AutoConfig
 from transformers.file_utils import cached_property
-from transformers.testing_utils import require_torch
+from transformers.testing_utils import require_torch, require_torch_non_multigpu_but_fix_me
 
 
 TINY_BART = "sshleifer/bart-tiny-random"
@@ -17,23 +17,28 @@ class MakeStudentTester(unittest.TestCase):
     def teacher_config(self):
         return AutoConfig.from_pretrained(TINY_BART)
 
+    @require_torch_non_multigpu_but_fix_me
     def test_valid_t5(self):
         student, *_ = create_student_by_copying_alternating_layers(TINY_T5, tempfile.mkdtemp(), e=1, d=1)
         self.assertEqual(student.config.num_hidden_layers, 1)
 
+    @require_torch_non_multigpu_but_fix_me
     def test_asymmetric_t5(self):
         student, *_ = create_student_by_copying_alternating_layers(TINY_T5, tempfile.mkdtemp(), e=1, d=None)
 
+    @require_torch_non_multigpu_but_fix_me
     def test_same_decoder_small_encoder(self):
         student, *_ = create_student_by_copying_alternating_layers(TINY_BART, tempfile.mkdtemp(), e=1, d=None)
         self.assertEqual(student.config.encoder_layers, 1)
         self.assertEqual(student.config.decoder_layers, self.teacher_config.encoder_layers)
 
+    @require_torch_non_multigpu_but_fix_me
     def test_small_enc_small_dec(self):
         student, *_ = create_student_by_copying_alternating_layers(TINY_BART, tempfile.mkdtemp(), e=1, d=1)
         self.assertEqual(student.config.encoder_layers, 1)
         self.assertEqual(student.config.decoder_layers, 1)
 
+    @require_torch_non_multigpu_but_fix_me
     def test_raises_assert(self):
         with self.assertRaises(AssertionError):
             create_student_by_copying_alternating_layers(TINY_BART, tempfile.mkdtemp(), e=None, d=None)

examples/seq2seq/test_seq2seq_examples.py

@@ -19,7 +19,14 @@ from run_eval import generate_summaries_or_translations, run_generate
 from run_eval_search import run_search
 from transformers import AutoConfig, AutoModelForSeq2SeqLM
 from transformers.hf_api import HfApi
-from transformers.testing_utils import CaptureStderr, CaptureStdout, TestCasePlus, require_torch_gpu, slow
+from transformers.testing_utils import (
+    CaptureStderr,
+    CaptureStdout,
+    TestCasePlus,
+    require_torch_gpu,
+    require_torch_non_multigpu_but_fix_me,
+    slow,
+)
 from utils import ROUGE_KEYS, label_smoothed_nll_loss, lmap, load_json
 
 
@@ -126,6 +133,7 @@ class TestSummarizationDistiller(TestCasePlus):
 
     @slow
     @require_torch_gpu
+    @require_torch_non_multigpu_but_fix_me
     def test_hub_configs(self):
         """I put require_torch_gpu cause I only want this to run with self-scheduled."""
 
@@ -143,10 +151,12 @@ class TestSummarizationDistiller(TestCasePlus):
                 failures.append(m)
         assert not failures, f"The following models could not be loaded through AutoConfig: {failures}"
 
+    @require_torch_non_multigpu_but_fix_me
     def test_distill_no_teacher(self):
         updates = dict(student_encoder_layers=2, student_decoder_layers=1, no_teacher=True)
         self._test_distiller_cli(updates)
 
+    @require_torch_non_multigpu_but_fix_me
     def test_distill_checkpointing_with_teacher(self):
         updates = dict(
             student_encoder_layers=2,
@@ -171,6 +181,7 @@ class TestSummarizationDistiller(TestCasePlus):
         convert_pl_to_hf(ckpts[0], transformer_ckpts[0].parent, out_path_new)
         assert os.path.exists(os.path.join(out_path_new, "pytorch_model.bin"))
 
+    @require_torch_non_multigpu_but_fix_me
     def test_loss_fn(self):
         model = AutoModelForSeq2SeqLM.from_pretrained(BART_TINY, return_dict=True)
         input_ids, mask = model.dummy_inputs["input_ids"], model.dummy_inputs["attention_mask"]
@@ -191,6 +202,7 @@ class TestSummarizationDistiller(TestCasePlus):
             # TODO: understand why this breaks
             self.assertEqual(nll_loss, model_computed_loss)
 
+    @require_torch_non_multigpu_but_fix_me
     def test_distill_mbart(self):
         updates = dict(
             student_encoder_layers=2,
@@ -215,6 +227,7 @@ class TestSummarizationDistiller(TestCasePlus):
         assert len(all_files) > 2
         self.assertEqual(len(transformer_ckpts), 2)
 
+    @require_torch_non_multigpu_but_fix_me
     def test_distill_t5(self):
         updates = dict(
             student_encoder_layers=1,
@@ -296,18 +309,21 @@ class TestTheRest(TestCasePlus):
 
     # test one model to quickly (no-@slow) catch simple problems and do an
     # extensive testing of functionality with multiple models as @slow separately
+    @require_torch_non_multigpu_but_fix_me
     def test_run_eval(self):
         self.run_eval_tester(T5_TINY)
 
     # any extra models should go into the list here - can be slow
     @parameterized.expand([BART_TINY, MBART_TINY])
     @slow
+    @require_torch_non_multigpu_but_fix_me
     def test_run_eval_slow(self, model):
         self.run_eval_tester(model)
 
     # testing with 2 models to validate: 1. translation (t5) 2. summarization (mbart)
     @parameterized.expand([T5_TINY, MBART_TINY])
     @slow
+    @require_torch_non_multigpu_but_fix_me
     def test_run_eval_search(self, model):
         input_file_name = Path(self.get_auto_remove_tmp_dir()) / "utest_input.source"
         output_file_name = input_file_name.parent / "utest_output.txt"
@@ -358,6 +374,7 @@ class TestTheRest(TestCasePlus):
     @parameterized.expand(
         [T5_TINY, BART_TINY, MBART_TINY, MARIAN_TINY, FSMT_TINY],
     )
+    @require_torch_non_multigpu_but_fix_me
     def test_finetune(self, model):
         args_d: dict = CHEAP_ARGS.copy()
         task = "translation" if model in [MBART_TINY, MARIAN_TINY, FSMT_TINY] else "summarization"
@@ -409,6 +426,7 @@ class TestTheRest(TestCasePlus):
         assert isinstance(example_batch, dict)
         assert len(example_batch) >= 4
 
+    @require_torch_non_multigpu_but_fix_me
     def test_finetune_extra_model_args(self):
         args_d: dict = CHEAP_ARGS.copy()
 
@@ -459,6 +477,7 @@ class TestTheRest(TestCasePlus):
             model = main(args)
         assert str(excinfo.value) == f"model config doesn't have a `{unsupported_param}` attribute"
 
+    @require_torch_non_multigpu_but_fix_me
     def test_finetune_lr_schedulers(self):
         args_d: dict = CHEAP_ARGS.copy()
 

examples/seq2seq/test_seq2seq_examples_multi_gpu.py

@@ -3,7 +3,14 @@
 import os
 import sys
 
-from transformers.testing_utils import TestCasePlus, execute_subprocess_async, require_torch_multigpu
+from transformers.testing_utils import (
+    TestCasePlus,
+    execute_subprocess_async,
+    get_gpu_count,
+    require_torch_gpu,
+    require_torch_multigpu,
+    slow,
+)
 
 from .test_seq2seq_examples import CHEAP_ARGS, make_test_data_dir
 from .utils import load_json
@@ -80,3 +87,30 @@ class TestSummarizationDistillerMultiGPU(TestCasePlus):
         self.assertEqual(len(metrics["test"]), 1)
         desired_n_evals = int(args_d["max_epochs"] * (1 / args_d["val_check_interval"]) / 2 + 1)
         self.assertEqual(len(metrics["val"]), desired_n_evals)
+
+    @slow
+    @require_torch_gpu
+    def test_distributed_eval(self):
+        output_dir = self.get_auto_remove_tmp_dir()
+        args = f"""
+            --model_name Helsinki-NLP/opus-mt-en-ro
+            --save_dir {output_dir}
+            --data_dir {self.test_file_dir_str}/test_data/wmt_en_ro
+            --num_beams 2
+            --task translation
+        """.split()
+
+        # we want this test to run even if there is only one GPU, but if there are more we use them all
+        n_gpu = get_gpu_count()
+        distributed_args = f"""
+            -m torch.distributed.launch
+            --nproc_per_node={n_gpu}
+            {self.test_file_dir}/run_distributed_eval.py
+        """.split()
+        cmd = [sys.executable] + distributed_args + args
+        execute_subprocess_async(cmd, env=self.get_env())
+
+        metrics_save_path = os.path.join(output_dir, "test_bleu.json")
+        metrics = load_json(metrics_save_path)
+        # print(metrics)
+        self.assertGreaterEqual(metrics["bleu"], 25)

examples/seq2seq/test_tatoeba_conversion.py

@@ -1,11 +1,13 @@
+import os
 import tempfile
 import unittest
 
-from transformers.convert_marian_tatoeba_to_pytorch import TatoebaConverter
+from transformers.convert_marian_tatoeba_to_pytorch import DEFAULT_REPO, TatoebaConverter
 from transformers.file_utils import cached_property
-from transformers.testing_utils import slow
+from transformers.testing_utils import require_torch_non_multigpu_but_fix_me, slow
 
 
+@unittest.skipUnless(os.path.exists(DEFAULT_REPO), "Tatoeba directory does not exist.")
 class TatoebaConversionTester(unittest.TestCase):
     @cached_property
     def resolver(self):
@@ -13,10 +15,12 @@ class TatoebaConversionTester(unittest.TestCase):
         return TatoebaConverter(save_dir=tmp_dir)
 
     @slow
+    @require_torch_non_multigpu_but_fix_me
     def test_resolver(self):
         self.resolver.convert_models(["heb-eng"])
 
     @slow
+    @require_torch_non_multigpu_but_fix_me
     def test_model_card(self):
         content, mmeta = self.resolver.write_model_card("opus-mt-he-en", dry_run=True)
         assert mmeta["long_pair"] == "heb-eng"

examples/seq2seq/train_distilbart_xsum.sh

@@ -2,6 +2,7 @@
 export PYTHONPATH="../":"${PYTHONPATH}"
 python distillation.py \
   --teacher facebook/bart-large-xsum --data_dir xsum \
+  --tokenizer_name facebook/bart-large-xsum \
   --student_decoder_layers 6 --student_encoder_layers 12 \
   --freeze_encoder --freeze_embeds \
   --learning_rate=3e-4 \

examples/seq2seq/utils.py

@@ -269,7 +269,7 @@ class Seq2SeqDataCollator:
         ), f"pad_token_id is not defined for ({self.tokenizer.__class__.__name__}), it must be defined."
         self.data_args = data_args
         self.tpu_num_cores = tpu_num_cores
-        self.dataset_kwargs = {"add_prefix_space": isinstance(tokenizer, BartTokenizer)}
+        self.dataset_kwargs = {"add_prefix_space": True} if isinstance(tokenizer, BartTokenizer) else {}
         if data_args.src_lang is not None:
             self.dataset_kwargs["src_lang"] = data_args.src_lang
         if data_args.tgt_lang is not None:

examples/test_examples.py

@@ -23,12 +23,18 @@ from unittest.mock import patch
 import torch
 
 from transformers.file_utils import is_apex_available
-from transformers.testing_utils import TestCasePlus, torch_device
+from transformers.testing_utils import TestCasePlus, require_torch_non_multigpu_but_fix_me, torch_device
 
 
 SRC_DIRS = [
     os.path.join(os.path.dirname(__file__), dirname)
-    for dirname in ["text-generation", "text-classification", "language-modeling", "question-answering"]
+    for dirname in [
+        "text-generation",
+        "text-classification",
+        "token-classification",
+        "language-modeling",
+        "question-answering",
+    ]
 ]
 sys.path.extend(SRC_DIRS)
 
@@ -38,6 +44,7 @@ if SRC_DIRS is not None:
     import run_generation
     import run_glue
     import run_mlm
+    import run_ner
     import run_pl_glue
     import run_squad
 
@@ -60,6 +67,7 @@ def is_cuda_and_apex_available():
 
 
 class ExamplesTests(TestCasePlus):
+    @require_torch_non_multigpu_but_fix_me
     def test_run_glue(self):
         stream_handler = logging.StreamHandler(sys.stdout)
         logger.addHandler(stream_handler)
@@ -92,6 +100,7 @@ class ExamplesTests(TestCasePlus):
             for value in result.values():
                 self.assertGreaterEqual(value, 0.75)
 
+    @require_torch_non_multigpu_but_fix_me
     def test_run_pl_glue(self):
         stream_handler = logging.StreamHandler(sys.stdout)
         logger.addHandler(stream_handler)
@@ -129,6 +138,7 @@ class ExamplesTests(TestCasePlus):
             #         self.assertGreaterEqual(v, 0.75, f"({k})")
             #
 
+    @require_torch_non_multigpu_but_fix_me
     def test_run_clm(self):
         stream_handler = logging.StreamHandler(sys.stdout)
         logger.addHandler(stream_handler)
@@ -160,6 +170,7 @@ class ExamplesTests(TestCasePlus):
             result = run_clm.main()
             self.assertLess(result["perplexity"], 100)
 
+    @require_torch_non_multigpu_but_fix_me
     def test_run_mlm(self):
         stream_handler = logging.StreamHandler(sys.stdout)
         logger.addHandler(stream_handler)
@@ -185,6 +196,38 @@ class ExamplesTests(TestCasePlus):
             result = run_mlm.main()
             self.assertLess(result["perplexity"], 42)
 
+    @require_torch_non_multigpu_but_fix_me
+    def test_run_ner(self):
+        stream_handler = logging.StreamHandler(sys.stdout)
+        logger.addHandler(stream_handler)
+
+        tmp_dir = self.get_auto_remove_tmp_dir()
+        testargs = f"""
+            run_ner.py
+            --model_name_or_path bert-base-uncased
+            --train_file tests/fixtures/tests_samples/conll/sample.json
+            --validation_file tests/fixtures/tests_samples/conll/sample.json
+            --output_dir {tmp_dir}
+            --overwrite_output_dir
+            --do_train
+            --do_eval
+            --warmup_steps=2
+            --learning_rate=2e-4
+            --per_gpu_train_batch_size=2
+            --per_gpu_eval_batch_size=2
+            --num_train_epochs=2
+        """.split()
+
+        if torch_device != "cuda":
+            testargs.append("--no_cuda")
+
+        with patch.object(sys, "argv", testargs):
+            result = run_ner.main()
+            self.assertGreaterEqual(result["eval_accuracy_score"], 0.75)
+            self.assertGreaterEqual(result["eval_precision"], 0.75)
+            self.assertLess(result["eval_loss"], 0.5)
+
+    @require_torch_non_multigpu_but_fix_me
     def test_run_squad(self):
         stream_handler = logging.StreamHandler(sys.stdout)
         logger.addHandler(stream_handler)
@@ -213,6 +256,7 @@ class ExamplesTests(TestCasePlus):
             self.assertGreaterEqual(result["f1"], 25)
             self.assertGreaterEqual(result["exact"], 21)
 
+    @require_torch_non_multigpu_but_fix_me
     def test_generation(self):
         stream_handler = logging.StreamHandler(sys.stdout)
         logger.addHandler(stream_handler)

examples/test_xla_examples.py

@@ -20,7 +20,7 @@ import unittest
 from time import time
 from unittest.mock import patch
 
-from transformers.testing_utils import require_torch_tpu
+from transformers.testing_utils import require_torch_non_multigpu_but_fix_me, require_torch_tpu
 
 
 logging.basicConfig(level=logging.DEBUG)
@@ -30,6 +30,7 @@ logger = logging.getLogger()
 
 @require_torch_tpu
 class TorchXLAExamplesTests(unittest.TestCase):
+    @require_torch_non_multigpu_but_fix_me
     def test_run_glue(self):
         import xla_spawn
 
@@ -81,6 +82,7 @@ class TorchXLAExamplesTests(unittest.TestCase):
             # Assert that the script takes less than 300 seconds to make sure it doesn't hang.
             self.assertLess(end - start, 500)
 
+    @require_torch_non_multigpu_but_fix_me
     def test_trainer_tpu(self):
         import xla_spawn
 

examples/text-classification/run_glue.py

@@ -177,7 +177,7 @@ def main():
     set_seed(training_args.seed)
 
     # Get the datasets: you can either provide your own CSV/JSON training and evaluation files (see below)
-    # or specify a GLUE benchmark task (the dataset will be downloaded automatically from the datasets Hub
+    # or specify a GLUE benchmark task (the dataset will be downloaded automatically from the datasets Hub).
     #
     # For CSV/JSON files, this script will use as labels the column called 'label' and as pair of sentences the
     # sentences in columns called 'sentence1' and 'sentence2' if such column exists or the first two columns not named

examples/token-classification/README.md

@@ -1,6 +1,40 @@
-## Named Entity Recognition
+## Token classification
 
-Based on the scripts [`run_ner.py`](https://github.com/huggingface/transformers/blob/master/examples/token-classification/run_ner.py) for Pytorch and
+Fine-tuning the library models for token classification task such as Named Entity Recognition (NER) or Parts-of-speech
+tagging (POS). The main scrip `run_ner.py` leverages the 🤗 Datasets library and the Trainer API. You can easily
+customize it to your needs if you need extra processing on your datasets.
+
+It will either run on a datasets hosted on our [hub](https://huggingface.co/datasets) or with your own text files for
+training and validation.
+
+The following example fine-tunes BERT on CoNLL-2003:
+
+```bash
+python run_ner.py \
+  --model_name_or_path bert-base-uncased \
+  --dataset_name conll2003 \
+  --output_dir /tmp/test-ner \
+  --do_train \
+  --do_eval
+```
+
+or just can just run the bash script `run.sh`.
+
+To run on your own training and validation files, use the following command:
+
+```bash
+python run_ner.py \
+  --model_name_or_path bert-base-uncased \
+  --train_file path_to_train_file \
+  --validation_file path_to_validation_file \
+  --output_dir /tmp/test-ner \
+  --do_train \
+  --do_eval
+```
+
+## Old version of the script
+
+Based on the scripts [`run_ner_old.py`](https://github.com/huggingface/transformers/blob/master/examples/token-classification/run_ner_old.py) for Pytorch and
 [`run_tf_ner.py`](https://github.com/huggingface/transformers/blob/master/examples/token-classification/run_tf_ner.py) for Tensorflow 2.
 
 The following examples are covered in this section:
@@ -69,7 +103,7 @@ export SEED=1
 To start training, just run:
 
 ```bash
-python3 run_ner.py --data_dir ./ \
+python3 run_ner_old.py --data_dir ./ \
 --labels ./labels.txt \
 --model_name_or_path $BERT_MODEL \
 --output_dir $OUTPUT_DIR \
@@ -87,7 +121,7 @@ If your GPU supports half-precision training, just add the `--fp16` flag. After
 
 #### JSON-based configuration file
 
-Instead of passing all parameters via commandline arguments, the `run_ner.py` script also supports reading parameters from a json-based configuration file:
+Instead of passing all parameters via commandline arguments, the `run_ner_old.py` script also supports reading parameters from a json-based configuration file:
 
 ```json
 {
@@ -106,7 +140,7 @@ Instead of passing all parameters via commandline arguments, the `run_ner.py` sc
 }
 ```
 
-It must be saved with a `.json` extension and can be used by running `python3 run_ner.py config.json`.
+It must be saved with a `.json` extension and can be used by running `python3 run_ner_old.py config.json`.
 
 #### Evaluation
 
@@ -250,7 +284,7 @@ cat data_wnut_17/train.txt data_wnut_17/dev.txt data_wnut_17/test.txt | cut -d "
 
 #### Run the Pytorch version
 
-Fine-tuning with the PyTorch version can be started using the `run_ner.py` script. In this example we use a JSON-based configuration file.
+Fine-tuning with the PyTorch version can be started using the `run_ner_old.py` script. In this example we use a JSON-based configuration file.
 
 This configuration file looks like:
 
@@ -274,7 +308,7 @@ This configuration file looks like:
 
 If your GPU supports half-precision training, please set `fp16` to `true`.
 
-Save this JSON-based configuration under `wnut_17.json`. The fine-tuning can be started with `python3 run_ner.py wnut_17.json`.
+Save this JSON-based configuration under `wnut_17.json`. The fine-tuning can be started with `python3 run_ner_old.py wnut_17.json`.
 
 #### Evaluation
 

examples/token-classification/run.sh

@@ -1,36 +1,6 @@
-## The relevant files are currently on a shared Google
-## drive at https://drive.google.com/drive/folders/1kC0I2UGl2ltrluI9NqDjaQJGw5iliw_J
-## Monitor for changes and eventually migrate to nlp dataset
-curl -L 'https://drive.google.com/uc?export=download&id=1Jjhbal535VVz2ap4v4r_rN1UEHTdLK5P' \
-| grep -v "^#" | cut -f 2,3 | tr '\t' ' ' > train.txt.tmp
-curl -L 'https://drive.google.com/uc?export=download&id=1ZfRcQThdtAR5PPRjIDtrVP7BtXSCUBbm' \
-| grep -v "^#" | cut -f 2,3 | tr '\t' ' ' > dev.txt.tmp
-curl -L 'https://drive.google.com/uc?export=download&id=1u9mb7kNJHWQCWyweMDRMuTFoOHOfeBTH' \
-| grep -v "^#" | cut -f 2,3 | tr '\t' ' ' > test.txt.tmp
-
-export MAX_LENGTH=128
-export BERT_MODEL=bert-base-multilingual-cased
-python3 scripts/preprocess.py train.txt.tmp $BERT_MODEL $MAX_LENGTH > train.txt
-python3 scripts/preprocess.py dev.txt.tmp $BERT_MODEL $MAX_LENGTH > dev.txt
-python3 scripts/preprocess.py test.txt.tmp $BERT_MODEL $MAX_LENGTH > test.txt
-cat train.txt dev.txt test.txt | cut -d " " -f 2 | grep -v "^$"| sort | uniq > labels.txt
-export OUTPUT_DIR=germeval-model
-export BATCH_SIZE=32
-export NUM_EPOCHS=3
-export SAVE_STEPS=750
-export SEED=1
-
 python3 run_ner.py \
---task_type NER \
---data_dir . \
---labels ./labels.txt \
---model_name_or_path $BERT_MODEL \
---output_dir $OUTPUT_DIR \
---max_seq_length  $MAX_LENGTH \
---num_train_epochs $NUM_EPOCHS \
---per_gpu_train_batch_size $BATCH_SIZE \
---save_steps $SAVE_STEPS \
---seed $SEED \
---do_train \
---do_eval \
---do_predict
+  --model_name_or_path bert-base-uncased \
+  --dataset_name conll2003 \
+  --output_dir /tmp/test-ner \
+  --do_train \
+  --do_eval

examples/token-classification/run_chunk.sh

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

examples/token-classification/run_ner.py

@@ -1,6 +1,5 @@
 # coding=utf-8
-# Copyright 2018 The Google AI Language Team Authors and The HuggingFace Inc. team.
-# Copyright (c) 2018, NVIDIA CORPORATION.  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.
@@ -13,29 +12,33 @@
 # 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.
-""" Fine-tuning the library models for named entity recognition on CoNLL-2003. """
+"""
+Fine-tuning the library models for token classification.
+"""
+# You can also adapt this script on your own token classification task and datasets. Pointers for this are left as comments.
+
 import logging
 import os
 import sys
 from dataclasses import dataclass, field
-from importlib import import_module
-from typing import Dict, List, Optional, Tuple
+from typing import Optional
 
 import numpy as np
+from datasets import load_dataset
 from seqeval.metrics import accuracy_score, f1_score, precision_score, recall_score
-from torch import nn
 
+import transformers
 from transformers import (
     AutoConfig,
     AutoModelForTokenClassification,
     AutoTokenizer,
-    EvalPrediction,
+    DataCollatorForTokenClassification,
     HfArgumentParser,
     Trainer,
     TrainingArguments,
     set_seed,
 )
-from utils_ner import Split, TokenClassificationDataset, TokenClassificationTask
+from transformers.trainer_utils import is_main_process
 
 
 logger = logging.getLogger(__name__)
@@ -53,15 +56,9 @@ class ModelArguments:
     config_name: Optional[str] = field(
         default=None, metadata={"help": "Pretrained config name or path if not the same as model_name"}
     )
-    task_type: Optional[str] = field(
-        default="NER", metadata={"help": "Task type to fine tune in training (e.g. NER, POS, etc)"}
-    )
     tokenizer_name: Optional[str] = field(
         default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
     )
-    use_fast: bool = field(default=False, metadata={"help": "Set this flag to use fast tokenization."})
-    # If you want to tweak more attributes on your tokenizer, you should do it in a distinct script,
-    # or just modify its tokenizer_config.json.
     cache_dir: Optional[str] = field(
         default=None, metadata={"help": "Where do you want to store the pretrained models downloaded from s3"}
     )
@@ -73,23 +70,58 @@ class DataTrainingArguments:
     Arguments pertaining to what data we are going to input our model for training and eval.
     """
 
-    data_dir: str = field(
-        metadata={"help": "The input data dir. Should contain the .txt files for a CoNLL-2003-formatted task."}
+    task_name: Optional[str] = field(default="ner", metadata={"help": "The name of the task (ner, pos...)."})
+    dataset_name: Optional[str] = field(
+        default=None, metadata={"help": "The name of the dataset to use (via the datasets library)."}
     )
-    labels: Optional[str] = field(
+    dataset_config_name: Optional[str] = field(
+        default=None, metadata={"help": "The configuration name of the dataset to use (via the datasets library)."}
+    )
+    train_file: Optional[str] = field(
+        default=None, metadata={"help": "The input training data file (a csv or JSON file)."}
+    )
+    validation_file: Optional[str] = field(
         default=None,
-        metadata={"help": "Path to a file containing all labels. If not specified, CoNLL-2003 labels are used."},
+        metadata={"help": "An optional input evaluation data file to evaluate on (a csv or JSON file)."},
     )
-    max_seq_length: int = field(
-        default=128,
-        metadata={
-            "help": "The maximum total input sequence length after tokenization. Sequences longer "
-            "than this will be truncated, sequences shorter will be padded."
-        },
+    test_file: Optional[str] = field(
+        default=None,
+        metadata={"help": "An optional input test data file to predict on (a csv or JSON file)."},
     )
     overwrite_cache: bool = field(
         default=False, metadata={"help": "Overwrite the cached training and evaluation sets"}
     )
+    preprocessing_num_workers: Optional[int] = field(
+        default=None,
+        metadata={"help": "The number of processes to use for the preprocessing."},
+    )
+    pad_to_max_length: bool = field(
+        default=False,
+        metadata={
+            "help": "Whether to pad all samples to model maximum sentence length. "
+            "If False, will pad the samples dynamically when batching to the maximum length in the batch. More "
+            "efficient on GPU but very bad for TPU."
+        },
+    )
+    label_all_tokens: bool = field(
+        default=False,
+        metadata={
+            "help": "Whether to put the label for one word on all tokens of generated by that word or just on the "
+            "one (in which case the other tokens will have a padding index)."
+        },
+    )
+
+    def __post_init__(self):
+        if self.dataset_name is None and self.train_file is None and self.validation_file is None:
+            raise ValueError("Need either a dataset name or a training/validation file.")
+        else:
+            if self.train_file is not None:
+                extension = self.train_file.split(".")[-1]
+                assert extension in ["csv", "json"], "`train_file` should be a csv or a json file."
+            if self.validation_file is not None:
+                extension = self.validation_file.split(".")[-1]
+                assert extension in ["csv", "json"], "`validation_file` should be a csv or a json file."
+        self.task_name = self.task_name.lower()
 
 
 def main():
@@ -112,60 +144,90 @@ def main():
         and not training_args.overwrite_output_dir
     ):
         raise ValueError(
-            f"Output directory ({training_args.output_dir}) already exists and is not empty. Use --overwrite_output_dir to overcome."
-        )
-
-    module = import_module("tasks")
-    try:
-        token_classification_task_clazz = getattr(module, model_args.task_type)
-        token_classification_task: TokenClassificationTask = token_classification_task_clazz()
-    except AttributeError:
-        raise ValueError(
-            f"Task {model_args.task_type} needs to be defined as a TokenClassificationTask subclass in {module}. "
-            f"Available tasks classes are: {TokenClassificationTask.__subclasses__()}"
+            f"Output directory ({training_args.output_dir}) already exists and is not empty."
+            "Use --overwrite_output_dir to overcome."
         )
 
     # Setup logging
     logging.basicConfig(
         format="%(asctime)s - %(levelname)s - %(name)s -   %(message)s",
         datefmt="%m/%d/%Y %H:%M:%S",
-        level=logging.INFO if training_args.local_rank in [-1, 0] else logging.WARN,
+        level=logging.INFO if is_main_process(training_args.local_rank) else logging.WARN,
     )
+
+    # Log on each process the small summary:
     logger.warning(
-        "Process rank: %s, device: %s, n_gpu: %s, distributed training: %s, 16-bits training: %s",
-        training_args.local_rank,
-        training_args.device,
-        training_args.n_gpu,
-        bool(training_args.local_rank != -1),
-        training_args.fp16,
+        f"Process rank: {training_args.local_rank}, device: {training_args.device}, n_gpu: {training_args.n_gpu}"
+        + f"distributed training: {bool(training_args.local_rank != -1)}, 16-bits training: {training_args.fp16}"
     )
+    # Set the verbosity to info of the Transformers logger (on main process only):
+    if is_main_process(training_args.local_rank):
+        transformers.utils.logging.set_verbosity_info()
     logger.info("Training/evaluation parameters %s", training_args)
 
-    # Set seed
+    # Set seed before initializing model.
     set_seed(training_args.seed)
 
-    # Prepare CONLL-2003 task
-    labels = token_classification_task.get_labels(data_args.labels)
-    label_map: Dict[int, str] = {i: label for i, label in enumerate(labels)}
-    num_labels = len(labels)
+    # 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).
+    #
+    # For CSV/JSON files, this script will use the column called 'text' or the first column if no column called
+    # 'text' is found. You can easily tweak this behavior (see below).
+    #
+    # In distributed training, the load_dataset function guarantee that only one local process can concurrently
+    # download the dataset.
+    if data_args.dataset_name is not None:
+        # Downloading and loading a dataset from the hub.
+        datasets = load_dataset(data_args.dataset_name, data_args.dataset_config_name)
+    else:
+        data_files = {}
+        if data_args.train_file is not None:
+            data_files["train"] = data_args.train_file
+        if data_args.validation_file is not None:
+            data_files["validation"] = data_args.validation_file
+        if data_args.test_file is not None:
+            data_files["test"] = data_args.test_file
+        extension = data_args.train_file.split(".")[-1]
+        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.
+
+    if training_args.do_train:
+        column_names = datasets["train"].column_names
+    else:
+        column_names = datasets["validation"].column_names
+    text_column_name = "words" if "words" in column_names else column_names[0]
+    label_column_name = data_args.task_name if data_args.task_name in column_names else column_names[1]
+
+    # Labeling (this part will be easier when https://github.com/huggingface/datasets/issues/797 is solved)
+    def get_label_list(labels):
+        unique_labels = set()
+        for label in labels:
+            unique_labels = unique_labels | set(label)
+        label_list = list(unique_labels)
+        label_list.sort()
+        return label_list
+
+    label_list = get_label_list(datasets["train"][label_column_name])
+    label_to_id = {l: i for i, l in enumerate(label_list)}
+    num_labels = len(label_list)
 
     # Load pretrained model and tokenizer
     #
     # Distributed training:
     # The .from_pretrained methods guarantee that only one local process can concurrently
     # download model & vocab.
-
     config = AutoConfig.from_pretrained(
         model_args.config_name if model_args.config_name else model_args.model_name_or_path,
         num_labels=num_labels,
-        id2label=label_map,
-        label2id={label: i for i, label in enumerate(labels)},
+        finetuning_task=data_args.task_name,
         cache_dir=model_args.cache_dir,
     )
     tokenizer = AutoTokenizer.from_pretrained(
         model_args.tokenizer_name if model_args.tokenizer_name else model_args.model_name_or_path,
         cache_dir=model_args.cache_dir,
-        use_fast=model_args.use_fast,
+        use_fast=True,
     )
     model = AutoModelForTokenClassification.from_pretrained(
         model_args.model_name_or_path,
@@ -174,67 +236,85 @@ def main():
         cache_dir=model_args.cache_dir,
     )
 
-    # Get datasets
-    train_dataset = (
-        TokenClassificationDataset(
-            token_classification_task=token_classification_task,
-            data_dir=data_args.data_dir,
-            tokenizer=tokenizer,
-            labels=labels,
-            model_type=config.model_type,
-            max_seq_length=data_args.max_seq_length,
-            overwrite_cache=data_args.overwrite_cache,
-            mode=Split.train,
+    # Preprocessing the dataset
+    # Padding strategy
+    padding = "max_length" if data_args.pad_to_max_length else False
+
+    # Tokenize all texts and align the labels with them.
+    def tokenize_and_align_labels(examples):
+        tokenized_inputs = tokenizer(
+            examples[text_column_name],
+            padding=padding,
+            truncation=True,
+            # 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,
+            return_offsets_mapping=True,
         )
-        if training_args.do_train
-        else None
-    )
-    eval_dataset = (
-        TokenClassificationDataset(
-            token_classification_task=token_classification_task,
-            data_dir=data_args.data_dir,
-            tokenizer=tokenizer,
-            labels=labels,
-            model_type=config.model_type,
-            max_seq_length=data_args.max_seq_length,
-            overwrite_cache=data_args.overwrite_cache,
-            mode=Split.dev,
-        )
-        if training_args.do_eval
-        else None
+        offset_mappings = tokenized_inputs.pop("offset_mapping")
+        labels = []
+        for label, offset_mapping in zip(examples[label_column_name], offset_mappings):
+            label_index = 0
+            current_label = -100
+            label_ids = []
+            for offset in offset_mapping:
+                # We set the label for the first token of each word. Special characters will have an offset of (0, 0)
+                # so the test ignores them.
+                if offset[0] == 0 and offset[1] != 0:
+                    current_label = label_to_id[label[label_index]]
+                    label_index += 1
+                    label_ids.append(current_label)
+                # For special tokens, we set the label to -100 so it's automatically ignored in the loss function.
+                elif offset[0] == 0 and offset[1] == 0:
+                    label_ids.append(-100)
+                # For the other tokens in a word, we set the label to either the current label or -100, depending on
+                # the label_all_tokens flag.
+                else:
+                    label_ids.append(current_label if data_args.label_all_tokens else -100)
+
+            labels.append(label_ids)
+        tokenized_inputs["labels"] = labels
+        return tokenized_inputs
+
+    tokenized_datasets = datasets.map(
+        tokenize_and_align_labels,
+        batched=True,
+        num_proc=data_args.preprocessing_num_workers,
+        load_from_cache_file=not data_args.overwrite_cache,
     )
 
-    def align_predictions(predictions: np.ndarray, label_ids: np.ndarray) -> Tuple[List[int], List[int]]:
-        preds = np.argmax(predictions, axis=2)
+    # Data collator
+    data_collator = DataCollatorForTokenClassification(tokenizer)
 
-        batch_size, seq_len = preds.shape
+    # Metrics
+    def compute_metrics(p):
+        predictions, labels = p
+        predictions = np.argmax(predictions, axis=2)
 
-        out_label_list = [[] for _ in range(batch_size)]
-        preds_list = [[] for _ in range(batch_size)]
+        # Remove ignored index (special tokens)
+        true_predictions = [
+            [label_list[p] for (p, l) in zip(prediction, label) if l != -100]
+            for prediction, label in zip(predictions, labels)
+        ]
+        true_labels = [
+            [label_list[l] for (p, l) in zip(prediction, label) if l != -100]
+            for prediction, label in zip(predictions, labels)
+        ]
 
-        for i in range(batch_size):
-            for j in range(seq_len):
-                if label_ids[i, j] != nn.CrossEntropyLoss().ignore_index:
-                    out_label_list[i].append(label_map[label_ids[i][j]])
-                    preds_list[i].append(label_map[preds[i][j]])
-
-        return preds_list, out_label_list
-
-    def compute_metrics(p: EvalPrediction) -> Dict:
-        preds_list, out_label_list = align_predictions(p.predictions, p.label_ids)
         return {
-            "accuracy_score": accuracy_score(out_label_list, preds_list),
-            "precision": precision_score(out_label_list, preds_list),
-            "recall": recall_score(out_label_list, preds_list),
-            "f1": f1_score(out_label_list, preds_list),
+            "accuracy_score": accuracy_score(true_labels, true_predictions),
+            "precision": precision_score(true_labels, true_predictions),
+            "recall": recall_score(true_labels, true_predictions),
+            "f1": f1_score(true_labels, true_predictions),
         }
 
     # Initialize our Trainer
     trainer = Trainer(
         model=model,
         args=training_args,
-        train_dataset=train_dataset,
-        eval_dataset=eval_dataset,
+        train_dataset=tokenized_datasets["train"] if training_args.do_train else None,
+        eval_dataset=tokenized_datasets["validation"] if training_args.do_eval else None,
+        tokenizer=tokenizer,
+        data_collator=data_collator,
         compute_metrics=compute_metrics,
     )
 
@@ -243,58 +323,50 @@ def main():
         trainer.train(
             model_path=model_args.model_name_or_path if os.path.isdir(model_args.model_name_or_path) else None
         )
-        trainer.save_model()
-        # For convenience, we also re-save the tokenizer to the same directory,
-        # so that you can share your model easily on huggingface.co/models =)
-        if trainer.is_world_master():
-            tokenizer.save_pretrained(training_args.output_dir)
+        trainer.save_model()  # Saves the tokenizer too for easy upload
 
     # Evaluation
     results = {}
     if training_args.do_eval:
         logger.info("*** Evaluate ***")
 
-        result = trainer.evaluate()
+        results = trainer.evaluate()
 
-        output_eval_file = os.path.join(training_args.output_dir, "eval_results.txt")
-        if trainer.is_world_master():
+        output_eval_file = os.path.join(training_args.output_dir, "eval_results_ner.txt")
+        if trainer.is_world_process_zero():
             with open(output_eval_file, "w") as writer:
                 logger.info("***** Eval results *****")
-                for key, value in result.items():
-                    logger.info("  %s = %s", key, value)
-                    writer.write("%s = %s\n" % (key, value))
-
-            results.update(result)
+                for key, value in results.items():
+                    logger.info(f"  {key} = {value}")
+                    writer.write(f"{key} = {value}\n")
 
     # Predict
     if training_args.do_predict:
-        test_dataset = TokenClassificationDataset(
-            token_classification_task=token_classification_task,
-            data_dir=data_args.data_dir,
-            tokenizer=tokenizer,
-            labels=labels,
-            model_type=config.model_type,
-            max_seq_length=data_args.max_seq_length,
-            overwrite_cache=data_args.overwrite_cache,
-            mode=Split.test,
-        )
+        logger.info("*** Predict ***")
 
-        predictions, label_ids, metrics = trainer.predict(test_dataset)
-        preds_list, _ = align_predictions(predictions, label_ids)
+        test_dataset = datasets["test"]
+        predictions, labels, metrics = trainer.predict(test_dataset)
+        predictions = np.argmax(predictions, axis=2)
+
+        # Remove ignored index (special tokens)
+        true_predictions = [
+            [label_list[p] for (p, l) in zip(prediction, label) if l != -100]
+            for prediction, label in zip(predictions, labels)
+        ]
 
         output_test_results_file = os.path.join(training_args.output_dir, "test_results.txt")
         if trainer.is_world_master():
             with open(output_test_results_file, "w") as writer:
                 for key, value in metrics.items():
-                    logger.info("  %s = %s", key, value)
-                    writer.write("%s = %s\n" % (key, value))
+                    logger.info(f"  {key} = {value}")
+                    writer.write(f"{key} = {value}\n")
 
         # Save predictions
         output_test_predictions_file = os.path.join(training_args.output_dir, "test_predictions.txt")
         if trainer.is_world_master():
             with open(output_test_predictions_file, "w") as writer:
-                with open(os.path.join(data_args.data_dir, "test.txt"), "r") as f:
-                    token_classification_task.write_predictions_to_file(writer, f, preds_list)
+                for prediction in true_predictions:
+                    writer.write(" ".join(prediction) + "\n")
 
     return results
 

examples/token-classification/run_ner_old.py

@@ -0,0 +1,308 @@
+# coding=utf-8
+# Copyright 2018 The Google AI Language Team Authors and The HuggingFace Inc. team.
+# Copyright (c) 2018, NVIDIA CORPORATION.  All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+""" Fine-tuning the library models for named entity recognition on CoNLL-2003. """
+import logging
+import os
+import sys
+from dataclasses import dataclass, field
+from importlib import import_module
+from typing import Dict, List, Optional, Tuple
+
+import numpy as np
+from seqeval.metrics import accuracy_score, f1_score, precision_score, recall_score
+from torch import nn
+
+from transformers import (
+    AutoConfig,
+    AutoModelForTokenClassification,
+    AutoTokenizer,
+    EvalPrediction,
+    HfArgumentParser,
+    Trainer,
+    TrainingArguments,
+    set_seed,
+)
+from utils_ner import Split, TokenClassificationDataset, TokenClassificationTask
+
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass
+class ModelArguments:
+    """
+    Arguments pertaining to which model/config/tokenizer we are going to fine-tune from.
+    """
+
+    model_name_or_path: str = field(
+        metadata={"help": "Path to pretrained model or model identifier from huggingface.co/models"}
+    )
+    config_name: Optional[str] = field(
+        default=None, metadata={"help": "Pretrained config name or path if not the same as model_name"}
+    )
+    task_type: Optional[str] = field(
+        default="NER", metadata={"help": "Task type to fine tune in training (e.g. NER, POS, etc)"}
+    )
+    tokenizer_name: Optional[str] = field(
+        default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
+    )
+    use_fast: bool = field(default=False, metadata={"help": "Set this flag to use fast tokenization."})
+    # If you want to tweak more attributes on your tokenizer, you should do it in a distinct script,
+    # or just modify its tokenizer_config.json.
+    cache_dir: Optional[str] = field(
+        default=None, metadata={"help": "Where do you want to store the pretrained models downloaded from s3"}
+    )
+
+
+@dataclass
+class DataTrainingArguments:
+    """
+    Arguments pertaining to what data we are going to input our model for training and eval.
+    """
+
+    data_dir: str = field(
+        metadata={"help": "The input data dir. Should contain the .txt files for a CoNLL-2003-formatted task."}
+    )
+    labels: Optional[str] = field(
+        default=None,
+        metadata={"help": "Path to a file containing all labels. If not specified, CoNLL-2003 labels are used."},
+    )
+    max_seq_length: int = field(
+        default=128,
+        metadata={
+            "help": "The maximum total input sequence length after tokenization. Sequences longer "
+            "than this will be truncated, sequences shorter will be padded."
+        },
+    )
+    overwrite_cache: bool = field(
+        default=False, metadata={"help": "Overwrite the cached training and evaluation sets"}
+    )
+
+
+def main():
+    # See all possible arguments in src/transformers/training_args.py
+    # or by passing the --help flag to this script.
+    # We now keep distinct sets of args, for a cleaner separation of concerns.
+
+    parser = HfArgumentParser((ModelArguments, DataTrainingArguments, TrainingArguments))
+    if len(sys.argv) == 2 and sys.argv[1].endswith(".json"):
+        # If we pass only one argument to the script and it's the path to a json file,
+        # let's parse it to get our arguments.
+        model_args, data_args, training_args = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1]))
+    else:
+        model_args, data_args, training_args = parser.parse_args_into_dataclasses()
+
+    if (
+        os.path.exists(training_args.output_dir)
+        and os.listdir(training_args.output_dir)
+        and training_args.do_train
+        and not training_args.overwrite_output_dir
+    ):
+        raise ValueError(
+            f"Output directory ({training_args.output_dir}) already exists and is not empty. Use --overwrite_output_dir to overcome."
+        )
+
+    module = import_module("tasks")
+    try:
+        token_classification_task_clazz = getattr(module, model_args.task_type)
+        token_classification_task: TokenClassificationTask = token_classification_task_clazz()
+    except AttributeError:
+        raise ValueError(
+            f"Task {model_args.task_type} needs to be defined as a TokenClassificationTask subclass in {module}. "
+            f"Available tasks classes are: {TokenClassificationTask.__subclasses__()}"
+        )
+
+    # Setup logging
+    logging.basicConfig(
+        format="%(asctime)s - %(levelname)s - %(name)s -   %(message)s",
+        datefmt="%m/%d/%Y %H:%M:%S",
+        level=logging.INFO if training_args.local_rank in [-1, 0] else logging.WARN,
+    )
+    logger.warning(
+        "Process rank: %s, device: %s, n_gpu: %s, distributed training: %s, 16-bits training: %s",
+        training_args.local_rank,
+        training_args.device,
+        training_args.n_gpu,
+        bool(training_args.local_rank != -1),
+        training_args.fp16,
+    )
+    logger.info("Training/evaluation parameters %s", training_args)
+
+    # Set seed
+    set_seed(training_args.seed)
+
+    # Prepare CONLL-2003 task
+    labels = token_classification_task.get_labels(data_args.labels)
+    label_map: Dict[int, str] = {i: label for i, label in enumerate(labels)}
+    num_labels = len(labels)
+
+    # Load pretrained model and tokenizer
+    #
+    # Distributed training:
+    # The .from_pretrained methods guarantee that only one local process can concurrently
+    # download model & vocab.
+
+    config = AutoConfig.from_pretrained(
+        model_args.config_name if model_args.config_name else model_args.model_name_or_path,
+        num_labels=num_labels,
+        id2label=label_map,
+        label2id={label: i for i, label in enumerate(labels)},
+        cache_dir=model_args.cache_dir,
+    )
+    tokenizer = AutoTokenizer.from_pretrained(
+        model_args.tokenizer_name if model_args.tokenizer_name else model_args.model_name_or_path,
+        cache_dir=model_args.cache_dir,
+        use_fast=model_args.use_fast,
+    )
+    model = AutoModelForTokenClassification.from_pretrained(
+        model_args.model_name_or_path,
+        from_tf=bool(".ckpt" in model_args.model_name_or_path),
+        config=config,
+        cache_dir=model_args.cache_dir,
+    )
+
+    # Get datasets
+    train_dataset = (
+        TokenClassificationDataset(
+            token_classification_task=token_classification_task,
+            data_dir=data_args.data_dir,
+            tokenizer=tokenizer,
+            labels=labels,
+            model_type=config.model_type,
+            max_seq_length=data_args.max_seq_length,
+            overwrite_cache=data_args.overwrite_cache,
+            mode=Split.train,
+        )
+        if training_args.do_train
+        else None
+    )
+    eval_dataset = (
+        TokenClassificationDataset(
+            token_classification_task=token_classification_task,
+            data_dir=data_args.data_dir,
+            tokenizer=tokenizer,
+            labels=labels,
+            model_type=config.model_type,
+            max_seq_length=data_args.max_seq_length,
+            overwrite_cache=data_args.overwrite_cache,
+            mode=Split.dev,
+        )
+        if training_args.do_eval
+        else None
+    )
+
+    def align_predictions(predictions: np.ndarray, label_ids: np.ndarray) -> Tuple[List[int], List[int]]:
+        preds = np.argmax(predictions, axis=2)
+
+        batch_size, seq_len = preds.shape
+
+        out_label_list = [[] for _ in range(batch_size)]
+        preds_list = [[] for _ in range(batch_size)]
+
+        for i in range(batch_size):
+            for j in range(seq_len):
+                if label_ids[i, j] != nn.CrossEntropyLoss().ignore_index:
+                    out_label_list[i].append(label_map[label_ids[i][j]])
+                    preds_list[i].append(label_map[preds[i][j]])
+
+        return preds_list, out_label_list
+
+    def compute_metrics(p: EvalPrediction) -> Dict:
+        preds_list, out_label_list = align_predictions(p.predictions, p.label_ids)
+        return {
+            "accuracy_score": accuracy_score(out_label_list, preds_list),
+            "precision": precision_score(out_label_list, preds_list),
+            "recall": recall_score(out_label_list, preds_list),
+            "f1": f1_score(out_label_list, preds_list),
+        }
+
+    # Initialize our Trainer
+    trainer = Trainer(
+        model=model,
+        args=training_args,
+        train_dataset=train_dataset,
+        eval_dataset=eval_dataset,
+        compute_metrics=compute_metrics,
+    )
+
+    # Training
+    if training_args.do_train:
+        trainer.train(
+            model_path=model_args.model_name_or_path if os.path.isdir(model_args.model_name_or_path) else None
+        )
+        trainer.save_model()
+        # For convenience, we also re-save the tokenizer to the same directory,
+        # so that you can share your model easily on huggingface.co/models =)
+        if trainer.is_world_master():
+            tokenizer.save_pretrained(training_args.output_dir)
+
+    # Evaluation
+    results = {}
+    if training_args.do_eval:
+        logger.info("*** Evaluate ***")
+
+        result = trainer.evaluate()
+
+        output_eval_file = os.path.join(training_args.output_dir, "eval_results.txt")
+        if trainer.is_world_master():
+            with open(output_eval_file, "w") as writer:
+                logger.info("***** Eval results *****")
+                for key, value in result.items():
+                    logger.info("  %s = %s", key, value)
+                    writer.write("%s = %s\n" % (key, value))
+
+            results.update(result)
+
+    # Predict
+    if training_args.do_predict:
+        test_dataset = TokenClassificationDataset(
+            token_classification_task=token_classification_task,
+            data_dir=data_args.data_dir,
+            tokenizer=tokenizer,
+            labels=labels,
+            model_type=config.model_type,
+            max_seq_length=data_args.max_seq_length,
+            overwrite_cache=data_args.overwrite_cache,
+            mode=Split.test,
+        )
+
+        predictions, label_ids, metrics = trainer.predict(test_dataset)
+        preds_list, _ = align_predictions(predictions, label_ids)
+
+        output_test_results_file = os.path.join(training_args.output_dir, "test_results.txt")
+        if trainer.is_world_master():
+            with open(output_test_results_file, "w") as writer:
+                for key, value in metrics.items():
+                    logger.info("  %s = %s", key, value)
+                    writer.write("%s = %s\n" % (key, value))
+
+        # Save predictions
+        output_test_predictions_file = os.path.join(training_args.output_dir, "test_predictions.txt")
+        if trainer.is_world_master():
+            with open(output_test_predictions_file, "w") as writer:
+                with open(os.path.join(data_args.data_dir, "test.txt"), "r") as f:
+                    token_classification_task.write_predictions_to_file(writer, f, preds_list)
+
+    return results
+
+
+def _mp_fn(index):
+    # For xla_spawn (TPUs)
+    main()
+
+
+if __name__ == "__main__":
+    main()

examples/token-classification/run_old.sh

@@ -0,0 +1,36 @@
+## The relevant files are currently on a shared Google
+## drive at https://drive.google.com/drive/folders/1kC0I2UGl2ltrluI9NqDjaQJGw5iliw_J
+## Monitor for changes and eventually migrate to nlp dataset
+curl -L 'https://drive.google.com/uc?export=download&id=1Jjhbal535VVz2ap4v4r_rN1UEHTdLK5P' \
+| grep -v "^#" | cut -f 2,3 | tr '\t' ' ' > train.txt.tmp
+curl -L 'https://drive.google.com/uc?export=download&id=1ZfRcQThdtAR5PPRjIDtrVP7BtXSCUBbm' \
+| grep -v "^#" | cut -f 2,3 | tr '\t' ' ' > dev.txt.tmp
+curl -L 'https://drive.google.com/uc?export=download&id=1u9mb7kNJHWQCWyweMDRMuTFoOHOfeBTH' \
+| grep -v "^#" | cut -f 2,3 | tr '\t' ' ' > test.txt.tmp
+
+export MAX_LENGTH=128
+export BERT_MODEL=bert-base-multilingual-cased
+python3 scripts/preprocess.py train.txt.tmp $BERT_MODEL $MAX_LENGTH > train.txt
+python3 scripts/preprocess.py dev.txt.tmp $BERT_MODEL $MAX_LENGTH > dev.txt
+python3 scripts/preprocess.py test.txt.tmp $BERT_MODEL $MAX_LENGTH > test.txt
+cat train.txt dev.txt test.txt | cut -d " " -f 2 | grep -v "^$"| sort | uniq > labels.txt
+export OUTPUT_DIR=germeval-model
+export BATCH_SIZE=32
+export NUM_EPOCHS=3
+export SAVE_STEPS=750
+export SEED=1
+
+python3 run_ner_old.py \
+--task_type NER \
+--data_dir . \
+--labels ./labels.txt \
+--model_name_or_path $BERT_MODEL \
+--output_dir $OUTPUT_DIR \
+--max_seq_length  $MAX_LENGTH \
+--num_train_epochs $NUM_EPOCHS \
+--per_gpu_train_batch_size $BATCH_SIZE \
+--save_steps $SAVE_STEPS \
+--seed $SEED \
+--do_train \
+--do_eval \
+--do_predict

examples/token-classification/run_pos.sh

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

examples/token-classification/test_ner_examples.py

@@ -3,8 +3,8 @@ import sys
 import unittest
 from unittest.mock import patch
 
-import run_ner
-from transformers.testing_utils import slow
+import run_ner_old as run_ner
+from transformers.testing_utils import require_torch_non_multigpu_but_fix_me, slow
 
 
 logging.basicConfig(level=logging.INFO)
@@ -14,6 +14,7 @@ logger = logging.getLogger()
 
 class ExamplesTests(unittest.TestCase):
     @slow
+    @require_torch_non_multigpu_but_fix_me
     def test_run_ner(self):
         stream_handler = logging.StreamHandler(sys.stdout)
         logger.addHandler(stream_handler)
@@ -34,6 +35,7 @@ class ExamplesTests(unittest.TestCase):
             result = run_ner.main()
             self.assertLess(result["eval_loss"], 1.5)
 
+    @require_torch_non_multigpu_but_fix_me
     def test_run_ner_pl(self):
         stream_handler = logging.StreamHandler(sys.stdout)
         logger.addHandler(stream_handler)

examples/token-classification/utils_ner.py

@@ -66,14 +66,16 @@ class Split(Enum):
 
 
 class TokenClassificationTask:
-    def read_examples_from_file(self, data_dir, mode: Union[Split, str]) -> List[InputExample]:
+    @staticmethod
+    def read_examples_from_file(data_dir, mode: Union[Split, str]) -> List[InputExample]:
         raise NotImplementedError
 
-    def get_labels(self, path: str) -> List[str]:
+    @staticmethod
+    def get_labels(path: str) -> List[str]:
         raise NotImplementedError
 
+    @staticmethod
     def convert_examples_to_features(
-        self,
         examples: List[InputExample],
         label_list: List[str],
         max_seq_length: int,

model_cards/TypicaAI/magbert-ner/README.md

@@ -10,11 +10,12 @@ widget:
 
 [MagBERT-NER] is a state-of-the-art NER model for Moroccan French language (Maghreb). The MagBERT-NER model was fine-tuned for NER Task based the language model for French Camembert (based on the RoBERTa architecture).
 
-For further information or requests, please go to [Typica.AI Website](https://typicasoft.io/)
+For further information or requests, please visite our website at [typica.ai Website](https://typica.ai/) or send us an email at contactus@typica.ai
 
 ## How to use MagBERT-NER with HuggingFace
 
 ##### Load MagBERT-NER and its sub-word tokenizer :
+
 ```python
 from transformers import AutoTokenizer, AutoModelForTokenClassification
 
@@ -45,11 +46,14 @@ nlp("Saad Dine El Otmani, né le 16 janvier 1956 à Inezgane, est un homme d'Ét
 
 ```
 
-```
-
 
 ## Authors 
 
-MagBert-NER was trained and evaluated by Hicham Assoudi, Ph.D.
+MagBert-NER Model was trained by Hicham Assoudi, Ph.D. 
+For any questions, comments you can contact me at assoudi@typica.ai
 
 
+## Citation
+
+If you use our work, please cite:
+Hicham Assoudi, Ph.D., MagBERT-NER: a state-of-the-art NER model for Moroccan French language (Maghreb), (2020)

model_cards/allenai/wmt19-de-en-6-6-base/README.md

@@ -61,7 +61,7 @@ Pretrained weights were left identical to the original model released by allenai
 Here are the BLEU scores:
 
 model   |  transformers
--------|---------|----------
+-------|---------
 wmt19-de-en-6-6-base  |  38.37
 
 The score was calculated using this code:

model_cards/allenai/wmt19-de-en-6-6-big/README.md

@@ -61,7 +61,7 @@ Pretrained weights were left identical to the original model released by allenai
 Here are the BLEU scores:
 
 model   |  transformers
--------|---------|----------
+-------|---------
 wmt19-de-en-6-6-big  |  39.9
 
 The score was calculated using this code:

model_cards/bionlp/bluebert_pubmed_mimic_uncased_L-12_H-768_A-12/README.md

@@ -0,0 +1,80 @@
+---
+language: 
+- en
+tags:
+- bert
+- bluebert
+license: 
+- PUBLIC DOMAIN NOTICE
+datasets:
+- PubMed
+- MIMIC-III
+
+---
+
+# BlueBert-Base, Uncased, PubMed and MIMIC-III
+
+## Model description
+
+A BERT model pre-trained on PubMed abstracts and clinical notes ([MIMIC-III](https://mimic.physionet.org/)).
+
+## Intended uses & limitations
+
+#### How to use
+
+Please see https://github.com/ncbi-nlp/bluebert
+
+## Training data
+
+We provide [preprocessed PubMed texts](https://ftp.ncbi.nlm.nih.gov/pub/lu/Suppl/NCBI-BERT/pubmed_uncased_sentence_nltk.txt.tar.gz) that were used to pre-train the BlueBERT models. 
+The corpus contains ~4000M words extracted from the [PubMed ASCII code version](https://www.ncbi.nlm.nih.gov/research/bionlp/APIs/BioC-PubMed/). 
+
+Pre-trained model: https://huggingface.co/bert-base-uncased
+
+## Training procedure
+
+*  lowercasing the text
+*  removing speical chars `\x00`-`\x7F`
+*  tokenizing the text using the [NLTK Treebank tokenizer](https://www.nltk.org/_modules/nltk/tokenize/treebank.html)
+
+Below is a code snippet for more details.
+
+```python
+value = value.lower()
+value = re.sub(r'[\r\n]+', ' ', value)
+value = re.sub(r'[^\x00-\x7F]+', ' ', value)
+
+tokenized = TreebankWordTokenizer().tokenize(value)
+sentence = ' '.join(tokenized)
+sentence = re.sub(r"\s's\b", "'s", sentence)
+```
+
+### BibTeX entry and citation info
+
+```bibtex
+@InProceedings{peng2019transfer,
+  author    = {Yifan Peng and Shankai Yan and Zhiyong Lu},
+  title     = {Transfer Learning in Biomedical Natural Language Processing: An Evaluation of BERT and ELMo on Ten Benchmarking Datasets},
+  booktitle = {Proceedings of the 2019 Workshop on Biomedical Natural Language Processing (BioNLP 2019)},
+  year      = {2019},
+  pages     = {58--65},
+}
+```
+
+### Acknowledgments
+
+This work was supported by the Intramural Research Programs of the National Institutes of Health, National Library of
+Medicine and Clinical Center. This work was supported by the National Library of Medicine of the National Institutes of Health under award number 4R00LM013001-01.
+
+We are also grateful to the authors of BERT and ELMo to make the data and codes publicly available.
+
+We would like to thank Dr Sun Kim for processing the PubMed texts.
+
+### Disclaimer
+
+This tool shows the results of research conducted in the Computational Biology Branch, NCBI. The information produced
+on this website is not intended for direct diagnostic use or medical decision-making without review and oversight
+by a clinical professional. Individuals should not change their health behavior solely on the basis of information
+produced on this website. NIH does not independently verify the validity or utility of the information produced
+by this tool. If you have questions about the information produced on this website, please see a health care
+professional. More information about NCBI's disclaimer policy is available.

model_cards/bionlp/bluebert_pubmed_uncased_L-12_H-768_A-12/README.md

@@ -0,0 +1,60 @@
+---
+language: 
+- en
+tags:
+- bluebert
+license: 
+- PUBLIC DOMAIN NOTICE
+datasets:
+- pubmed
+
+---
+
+# BlueBert-Base, Uncased, PubMed
+
+## Model description
+
+A BERT model pre-trained on PubMed abstracts
+
+## Intended uses & limitations
+
+#### How to use
+
+Please see https://github.com/ncbi-nlp/bluebert
+
+## Training data
+
+We provide [preprocessed PubMed texts](https://ftp.ncbi.nlm.nih.gov/pub/lu/Suppl/NCBI-BERT/pubmed_uncased_sentence_nltk.txt.tar.gz) that were used to pre-train the BlueBERT models. 
+The corpus contains ~4000M words extracted from the [PubMed ASCII code version](https://www.ncbi.nlm.nih.gov/research/bionlp/APIs/BioC-PubMed/). 
+
+Pre-trained model: https://huggingface.co/bert-base-uncased
+
+## Training procedure
+
+*  lowercasing the text
+*  removing speical chars `\x00`-`\x7F`
+*  tokenizing the text using the [NLTK Treebank tokenizer](https://www.nltk.org/_modules/nltk/tokenize/treebank.html)
+
+Below is a code snippet for more details.
+
+```python
+value = value.lower()
+value = re.sub(r'[\r\n]+', ' ', value)
+value = re.sub(r'[^\x00-\x7F]+', ' ', value)
+
+tokenized = TreebankWordTokenizer().tokenize(value)
+sentence = ' '.join(tokenized)
+sentence = re.sub(r"\s's\b", "'s", sentence)
+```
+
+### BibTeX entry and citation info
+
+```bibtex
+@InProceedings{peng2019transfer,
+  author    = {Yifan Peng and Shankai Yan and Zhiyong Lu},
+  title     = {Transfer Learning in Biomedical Natural Language Processing: An Evaluation of BERT and ELMo on Ten Benchmarking Datasets},
+  booktitle = {Proceedings of the 2019 Workshop on Biomedical Natural Language Processing (BioNLP 2019)},
+  year      = {2019},
+  pages     = {58--65},
+}
+```

model_cards/cimm-kzn/enrudr-bert/README.md

@@ -3,9 +3,9 @@ language:
 - ru
 - en
 ---
-## RuDR-BERT
+## EnRuDR-BERT
 
-EnRuDR-BERT - Multilingual, Cased, which pretrained on the raw part of the RuDReC corpus (1.4M reviews) and collecting of consumer comments on drug administration from [2]. Pre-training was based on the [original BERT code](https://github.com/google-research/bert) provided by Google. In particular, Multi-BERT was for used for initialization; vocabulary of Russian subtokens and parameters are the same as in Multi-BERT. Training details are described in our paper. \
+EnRuDR-BERT - Multilingual, Cased, which pretrained on the raw part of the RuDReC corpus (1.4M reviews) and english collection of consumer comments on drug administration from [2]. Pre-training was based on the [original BERT code](https://github.com/google-research/bert) provided by Google. In particular, Multi-BERT was for used for initialization; vocabulary of Russian subtokens and parameters are the same as in Multi-BERT. Training details are described in our paper. \
    link: https://yadi.sk/d/-PTn0xhk1PqvgQ
    
 

model_cards/dbmdz/bert-base-italian-cased/README.md

@@ -1,12 +1,14 @@
 ---
 language: it
 license: mit
+datasets:
+- wikipedia
 ---
 
-# 🤗 + 📚 dbmdz BERT models
+# 🤗 + 📚 dbmdz BERT and ELECTRA models
 
 In this repository the MDZ Digital Library team (dbmdz) at the Bavarian State
-Library open sources Italian BERT models 🎉
+Library open sources Italian BERT and ELECTRA models 🎉
 
 # Italian BERT
 
@@ -22,23 +24,35 @@ For the XXL Italian models, we use the same training data from OPUS and extend
 it with data from the Italian part of the [OSCAR corpus](https://traces1.inria.fr/oscar/).
 Thus, the final training corpus has a size of 81GB and 13,138,379,147 tokens.
 
+Note: Unfortunately, a wrong vocab size was used when training the XXL models.
+This explains the mismatch of the "real" vocab size of 31102, compared to the
+vocab size specified in `config.json`. However, the model is working and all
+evaluations were done under those circumstances.
+See [this issue](https://github.com/dbmdz/berts/issues/7) for more information.
+
+The Italian ELECTRA model was trained on the "XXL" corpus for 1M steps in total using a batch
+size of 128. We pretty much following the ELECTRA training procedure as used for
+[BERTurk](https://github.com/stefan-it/turkish-bert/tree/master/electra).
+
 ## Model weights
 
 Currently only PyTorch-[Transformers](https://github.com/huggingface/transformers)
 compatible weights are available. If you need access to TensorFlow checkpoints,
 please raise an issue!
 
-| Model                                   | Downloads
-| --------------------------------------- | ---------------------------------------------------------------------------------------------------------------
-| `dbmdz/bert-base-italian-cased`         | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/config.json)       • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/pytorch_model.bin)       • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/vocab.txt)
-| `dbmdz/bert-base-italian-uncased`       | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/config.json)     • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/pytorch_model.bin)     • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/vocab.txt)
-| `dbmdz/bert-base-italian-xxl-cased`     | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/config.json)   • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/pytorch_model.bin)   • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/vocab.txt)
-| `dbmdz/bert-base-italian-xxl-uncased`   | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/config.json) • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/pytorch_model.bin) • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/vocab.txt)
+| Model                                                | Downloads
+| ---------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------
+| `dbmdz/bert-base-italian-cased`                      | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/config.json)                                               • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/pytorch_model.bin)                      • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/vocab.txt)
+| `dbmdz/bert-base-italian-uncased`                    | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/config.json)                                             • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/pytorch_model.bin)                    • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/vocab.txt)
+| `dbmdz/bert-base-italian-xxl-cased`                  | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/config.json)                                           • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/pytorch_model.bin)                  • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/vocab.txt)
+| `dbmdz/bert-base-italian-xxl-uncased`                | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/config.json)                                         • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/pytorch_model.bin)                • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/vocab.txt)
+| `dbmdz/electra-base-italian-xxl-cased-discriminator` | [`config.json`](https://s3.amazonaws.com/models.huggingface.co/bert/dbmdz/electra-base-italian-xxl-cased-discriminator/config.json) • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/electra-base-italian-xxl-cased-discriminator/pytorch_model.bin) • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/electra-base-italian-xxl-cased-discriminator/vocab.txt)
+| `dbmdz/electra-base-italian-xxl-cased-generator`     | [`config.json`](https://s3.amazonaws.com/models.huggingface.co/bert/dbmdz/electra-base-italian-xxl-cased-generator/config.json)     • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/electra-base-italian-xxl-cased-generator/pytorch_model.bin)     • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/electra-base-italian-xxl-cased-generator/vocab.txt)
 
 ## Results
 
 For results on downstream tasks like NER or PoS tagging, please refer to
-[this repository](https://github.com/stefan-it/fine-tuned-berts-seq).
+[this repository](https://github.com/stefan-it/italian-bertelectra).
 
 ## Usage
 
@@ -47,8 +61,11 @@ With Transformers >= 2.3 our Italian BERT models can be loaded like:
 ```python
 from transformers import AutoModel, AutoTokenizer
 
-tokenizer = AutoTokenizer.from_pretrained("dbmdz/bert-base-italian-cased")
-model = AutoModel.from_pretrained("dbmdz/bert-base-italian-cased")
+model_name = "dbmdz/bert-base-italian-cased"
+
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+model = AutoModel.from_pretrained(model_name)
 ```
 
 To load the (recommended) Italian XXL BERT models, just use:
@@ -56,8 +73,23 @@ To load the (recommended) Italian XXL BERT models, just use:
 ```python
 from transformers import AutoModel, AutoTokenizer
 
-tokenizer = AutoTokenizer.from_pretrained("dbmdz/bert-base-italian-xxl-cased")
-model = AutoModel.from_pretrained("dbmdz/bert-base-italian-xxl-cased")
+model_name = "dbmdz/bert-base-italian-xxl-cased"
+
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+model = AutoModel.from_pretrained(model_name)
+```
+
+To load the Italian XXL ELECTRA model (discriminator), just use:
+
+```python
+from transformers import AutoModel, AutoTokenizer
+
+model_name = "dbmdz/electra-base-italian-xxl-cased-discriminator"
+
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+model = AutoModelWithLMHead.from_pretrained(model_name)
 ```
 
 # Huggingface model hub
@@ -66,7 +98,7 @@ All models are available on the [Huggingface model hub](https://huggingface.co/d
 
 # Contact (Bugs, Feedback, Contribution and more)
 
-For questions about our BERT models just open an issue
+For questions about our BERT/ELECTRA models just open an issue
 [here](https://github.com/dbmdz/berts/issues/new) 🤗
 
 # Acknowledgments

model_cards/dbmdz/bert-base-italian-uncased/README.md

@@ -1,12 +1,14 @@
 ---
 language: it
 license: mit
+datasets:
+- wikipedia
 ---
 
-# 🤗 + 📚 dbmdz BERT models
+# 🤗 + 📚 dbmdz BERT and ELECTRA models
 
 In this repository the MDZ Digital Library team (dbmdz) at the Bavarian State
-Library open sources Italian BERT models 🎉
+Library open sources Italian BERT and ELECTRA models 🎉
 
 # Italian BERT
 
@@ -22,23 +24,35 @@ For the XXL Italian models, we use the same training data from OPUS and extend
 it with data from the Italian part of the [OSCAR corpus](https://traces1.inria.fr/oscar/).
 Thus, the final training corpus has a size of 81GB and 13,138,379,147 tokens.
 
+Note: Unfortunately, a wrong vocab size was used when training the XXL models.
+This explains the mismatch of the "real" vocab size of 31102, compared to the
+vocab size specified in `config.json`. However, the model is working and all
+evaluations were done under those circumstances.
+See [this issue](https://github.com/dbmdz/berts/issues/7) for more information.
+
+The Italian ELECTRA model was trained on the "XXL" corpus for 1M steps in total using a batch
+size of 128. We pretty much following the ELECTRA training procedure as used for
+[BERTurk](https://github.com/stefan-it/turkish-bert/tree/master/electra).
+
 ## Model weights
 
 Currently only PyTorch-[Transformers](https://github.com/huggingface/transformers)
 compatible weights are available. If you need access to TensorFlow checkpoints,
 please raise an issue!
 
-| Model                                   | Downloads
-| --------------------------------------- | ---------------------------------------------------------------------------------------------------------------
-| `dbmdz/bert-base-italian-cased`         | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/config.json)       • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/pytorch_model.bin)       • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/vocab.txt)
-| `dbmdz/bert-base-italian-uncased`       | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/config.json)     • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/pytorch_model.bin)     • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/vocab.txt)
-| `dbmdz/bert-base-italian-xxl-cased`     | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/config.json)   • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/pytorch_model.bin)   • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/vocab.txt)
-| `dbmdz/bert-base-italian-xxl-uncased`   | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/config.json) • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/pytorch_model.bin) • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/vocab.txt)
+| Model                                                | Downloads
+| ---------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------
+| `dbmdz/bert-base-italian-cased`                      | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/config.json)                                               • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/pytorch_model.bin)                      • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/vocab.txt)
+| `dbmdz/bert-base-italian-uncased`                    | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/config.json)                                             • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/pytorch_model.bin)                    • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/vocab.txt)
+| `dbmdz/bert-base-italian-xxl-cased`                  | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/config.json)                                           • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/pytorch_model.bin)                  • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/vocab.txt)
+| `dbmdz/bert-base-italian-xxl-uncased`                | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/config.json)                                         • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/pytorch_model.bin)                • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/vocab.txt)
+| `dbmdz/electra-base-italian-xxl-cased-discriminator` | [`config.json`](https://s3.amazonaws.com/models.huggingface.co/bert/dbmdz/electra-base-italian-xxl-cased-discriminator/config.json) • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/electra-base-italian-xxl-cased-discriminator/pytorch_model.bin) • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/electra-base-italian-xxl-cased-discriminator/vocab.txt)
+| `dbmdz/electra-base-italian-xxl-cased-generator`     | [`config.json`](https://s3.amazonaws.com/models.huggingface.co/bert/dbmdz/electra-base-italian-xxl-cased-generator/config.json)     • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/electra-base-italian-xxl-cased-generator/pytorch_model.bin)     • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/electra-base-italian-xxl-cased-generator/vocab.txt)
 
 ## Results
 
 For results on downstream tasks like NER or PoS tagging, please refer to
-[this repository](https://github.com/stefan-it/fine-tuned-berts-seq).
+[this repository](https://github.com/stefan-it/italian-bertelectra).
 
 ## Usage
 
@@ -47,8 +61,11 @@ With Transformers >= 2.3 our Italian BERT models can be loaded like:
 ```python
 from transformers import AutoModel, AutoTokenizer
 
-tokenizer = AutoTokenizer.from_pretrained("dbmdz/bert-base-italian-cased")
-model = AutoModel.from_pretrained("dbmdz/bert-base-italian-cased")
+model_name = "dbmdz/bert-base-italian-cased"
+
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+model = AutoModel.from_pretrained(model_name)
 ```
 
 To load the (recommended) Italian XXL BERT models, just use:
@@ -56,8 +73,23 @@ To load the (recommended) Italian XXL BERT models, just use:
 ```python
 from transformers import AutoModel, AutoTokenizer
 
-tokenizer = AutoTokenizer.from_pretrained("dbmdz/bert-base-italian-xxl-cased")
-model = AutoModel.from_pretrained("dbmdz/bert-base-italian-xxl-cased")
+model_name = "dbmdz/bert-base-italian-xxl-cased"
+
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+model = AutoModel.from_pretrained(model_name)
+```
+
+To load the Italian XXL ELECTRA model (discriminator), just use:
+
+```python
+from transformers import AutoModel, AutoTokenizer
+
+model_name = "dbmdz/electra-base-italian-xxl-cased-discriminator"
+
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+model = AutoModelWithLMHead.from_pretrained(model_name)
 ```
 
 # Huggingface model hub
@@ -66,7 +98,7 @@ All models are available on the [Huggingface model hub](https://huggingface.co/d
 
 # Contact (Bugs, Feedback, Contribution and more)
 
-For questions about our BERT models just open an issue
+For questions about our BERT/ELECTRA models just open an issue
 [here](https://github.com/dbmdz/berts/issues/new) 🤗
 
 # Acknowledgments

model_cards/dbmdz/bert-base-italian-xxl-cased/README.md

@@ -1,12 +1,14 @@
 ---
 language: it
 license: mit
+datasets:
+- wikipedia
 ---
 
-# 🤗 + 📚 dbmdz BERT models
+# 🤗 + 📚 dbmdz BERT and ELECTRA models
 
 In this repository the MDZ Digital Library team (dbmdz) at the Bavarian State
-Library open sources Italian BERT models 🎉
+Library open sources Italian BERT and ELECTRA models 🎉
 
 # Italian BERT
 
@@ -22,23 +24,35 @@ For the XXL Italian models, we use the same training data from OPUS and extend
 it with data from the Italian part of the [OSCAR corpus](https://traces1.inria.fr/oscar/).
 Thus, the final training corpus has a size of 81GB and 13,138,379,147 tokens.
 
+Note: Unfortunately, a wrong vocab size was used when training the XXL models.
+This explains the mismatch of the "real" vocab size of 31102, compared to the
+vocab size specified in `config.json`. However, the model is working and all
+evaluations were done under those circumstances.
+See [this issue](https://github.com/dbmdz/berts/issues/7) for more information.
+
+The Italian ELECTRA model was trained on the "XXL" corpus for 1M steps in total using a batch
+size of 128. We pretty much following the ELECTRA training procedure as used for
+[BERTurk](https://github.com/stefan-it/turkish-bert/tree/master/electra).
+
 ## Model weights
 
 Currently only PyTorch-[Transformers](https://github.com/huggingface/transformers)
 compatible weights are available. If you need access to TensorFlow checkpoints,
 please raise an issue!
 
-| Model                                   | Downloads
-| --------------------------------------- | ---------------------------------------------------------------------------------------------------------------
-| `dbmdz/bert-base-italian-cased`         | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/config.json)       • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/pytorch_model.bin)       • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/vocab.txt)
-| `dbmdz/bert-base-italian-uncased`       | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/config.json)     • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/pytorch_model.bin)     • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/vocab.txt)
-| `dbmdz/bert-base-italian-xxl-cased`     | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/config.json)   • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/pytorch_model.bin)   • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/vocab.txt)
-| `dbmdz/bert-base-italian-xxl-uncased`   | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/config.json) • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/pytorch_model.bin) • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/vocab.txt)
+| Model                                                | Downloads
+| ---------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------
+| `dbmdz/bert-base-italian-cased`                      | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/config.json)                                               • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/pytorch_model.bin)                      • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/vocab.txt)
+| `dbmdz/bert-base-italian-uncased`                    | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/config.json)                                             • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/pytorch_model.bin)                    • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/vocab.txt)
+| `dbmdz/bert-base-italian-xxl-cased`                  | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/config.json)                                           • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/pytorch_model.bin)                  • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/vocab.txt)
+| `dbmdz/bert-base-italian-xxl-uncased`                | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/config.json)                                         • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/pytorch_model.bin)                • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/vocab.txt)
+| `dbmdz/electra-base-italian-xxl-cased-discriminator` | [`config.json`](https://s3.amazonaws.com/models.huggingface.co/bert/dbmdz/electra-base-italian-xxl-cased-discriminator/config.json) • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/electra-base-italian-xxl-cased-discriminator/pytorch_model.bin) • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/electra-base-italian-xxl-cased-discriminator/vocab.txt)
+| `dbmdz/electra-base-italian-xxl-cased-generator`     | [`config.json`](https://s3.amazonaws.com/models.huggingface.co/bert/dbmdz/electra-base-italian-xxl-cased-generator/config.json)     • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/electra-base-italian-xxl-cased-generator/pytorch_model.bin)     • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/electra-base-italian-xxl-cased-generator/vocab.txt)
 
 ## Results
 
 For results on downstream tasks like NER or PoS tagging, please refer to
-[this repository](https://github.com/stefan-it/fine-tuned-berts-seq).
+[this repository](https://github.com/stefan-it/italian-bertelectra).
 
 ## Usage
 
@@ -47,8 +61,11 @@ With Transformers >= 2.3 our Italian BERT models can be loaded like:
 ```python
 from transformers import AutoModel, AutoTokenizer
 
-tokenizer = AutoTokenizer.from_pretrained("dbmdz/bert-base-italian-cased")
-model = AutoModel.from_pretrained("dbmdz/bert-base-italian-cased")
+model_name = "dbmdz/bert-base-italian-cased"
+
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+model = AutoModel.from_pretrained(model_name)
 ```
 
 To load the (recommended) Italian XXL BERT models, just use:
@@ -56,8 +73,23 @@ To load the (recommended) Italian XXL BERT models, just use:
 ```python
 from transformers import AutoModel, AutoTokenizer
 
-tokenizer = AutoTokenizer.from_pretrained("dbmdz/bert-base-italian-xxl-cased")
-model = AutoModel.from_pretrained("dbmdz/bert-base-italian-xxl-cased")
+model_name = "dbmdz/bert-base-italian-xxl-cased"
+
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+model = AutoModel.from_pretrained(model_name)
+```
+
+To load the Italian XXL ELECTRA model (discriminator), just use:
+
+```python
+from transformers import AutoModel, AutoTokenizer
+
+model_name = "dbmdz/electra-base-italian-xxl-cased-discriminator"
+
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+model = AutoModelWithLMHead.from_pretrained(model_name)
 ```
 
 # Huggingface model hub
@@ -66,7 +98,7 @@ All models are available on the [Huggingface model hub](https://huggingface.co/d
 
 # Contact (Bugs, Feedback, Contribution and more)
 
-For questions about our BERT models just open an issue
+For questions about our BERT/ELECTRA models just open an issue
 [here](https://github.com/dbmdz/berts/issues/new) 🤗
 
 # Acknowledgments

model_cards/dbmdz/bert-base-italian-xxl-uncased/README.md

@@ -1,12 +1,14 @@
 ---
 language: it
 license: mit
+datasets:
+- wikipedia
 ---
 
-# 🤗 + 📚 dbmdz BERT models
+# 🤗 + 📚 dbmdz BERT and ELECTRA models
 
 In this repository the MDZ Digital Library team (dbmdz) at the Bavarian State
-Library open sources Italian BERT models 🎉
+Library open sources Italian BERT and ELECTRA models 🎉
 
 # Italian BERT
 
@@ -22,23 +24,35 @@ For the XXL Italian models, we use the same training data from OPUS and extend
 it with data from the Italian part of the [OSCAR corpus](https://traces1.inria.fr/oscar/).
 Thus, the final training corpus has a size of 81GB and 13,138,379,147 tokens.
 
+Note: Unfortunately, a wrong vocab size was used when training the XXL models.
+This explains the mismatch of the "real" vocab size of 31102, compared to the
+vocab size specified in `config.json`. However, the model is working and all
+evaluations were done under those circumstances.
+See [this issue](https://github.com/dbmdz/berts/issues/7) for more information.
+
+The Italian ELECTRA model was trained on the "XXL" corpus for 1M steps in total using a batch
+size of 128. We pretty much following the ELECTRA training procedure as used for
+[BERTurk](https://github.com/stefan-it/turkish-bert/tree/master/electra).
+
 ## Model weights
 
 Currently only PyTorch-[Transformers](https://github.com/huggingface/transformers)
 compatible weights are available. If you need access to TensorFlow checkpoints,
 please raise an issue!
 
-| Model                                   | Downloads
-| --------------------------------------- | ---------------------------------------------------------------------------------------------------------------
-| `dbmdz/bert-base-italian-cased`         | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/config.json)       • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/pytorch_model.bin)       • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/vocab.txt)
-| `dbmdz/bert-base-italian-uncased`       | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/config.json)     • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/pytorch_model.bin)     • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/vocab.txt)
-| `dbmdz/bert-base-italian-xxl-cased`     | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/config.json)   • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/pytorch_model.bin)   • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/vocab.txt)
-| `dbmdz/bert-base-italian-xxl-uncased`   | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/config.json) • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/pytorch_model.bin) • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/vocab.txt)
+| Model                                                | Downloads
+| ---------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------
+| `dbmdz/bert-base-italian-cased`                      | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/config.json)                                               • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/pytorch_model.bin)                      • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/vocab.txt)
+| `dbmdz/bert-base-italian-uncased`                    | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/config.json)                                             • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/pytorch_model.bin)                    • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/vocab.txt)
+| `dbmdz/bert-base-italian-xxl-cased`                  | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/config.json)                                           • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/pytorch_model.bin)                  • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/vocab.txt)
+| `dbmdz/bert-base-italian-xxl-uncased`                | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/config.json)                                         • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/pytorch_model.bin)                • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/vocab.txt)
+| `dbmdz/electra-base-italian-xxl-cased-discriminator` | [`config.json`](https://s3.amazonaws.com/models.huggingface.co/bert/dbmdz/electra-base-italian-xxl-cased-discriminator/config.json) • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/electra-base-italian-xxl-cased-discriminator/pytorch_model.bin) • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/electra-base-italian-xxl-cased-discriminator/vocab.txt)
+| `dbmdz/electra-base-italian-xxl-cased-generator`     | [`config.json`](https://s3.amazonaws.com/models.huggingface.co/bert/dbmdz/electra-base-italian-xxl-cased-generator/config.json)     • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/electra-base-italian-xxl-cased-generator/pytorch_model.bin)     • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/electra-base-italian-xxl-cased-generator/vocab.txt)
 
 ## Results
 
 For results on downstream tasks like NER or PoS tagging, please refer to
-[this repository](https://github.com/stefan-it/fine-tuned-berts-seq).
+[this repository](https://github.com/stefan-it/italian-bertelectra).
 
 ## Usage
 
@@ -47,8 +61,11 @@ With Transformers >= 2.3 our Italian BERT models can be loaded like:
 ```python
 from transformers import AutoModel, AutoTokenizer
 
-tokenizer = AutoTokenizer.from_pretrained("dbmdz/bert-base-italian-cased")
-model = AutoModel.from_pretrained("dbmdz/bert-base-italian-cased")
+model_name = "dbmdz/bert-base-italian-cased"
+
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+model = AutoModel.from_pretrained(model_name)
 ```
 
 To load the (recommended) Italian XXL BERT models, just use:
@@ -56,8 +73,23 @@ To load the (recommended) Italian XXL BERT models, just use:
 ```python
 from transformers import AutoModel, AutoTokenizer
 
-tokenizer = AutoTokenizer.from_pretrained("dbmdz/bert-base-italian-xxl-cased")
-model = AutoModel.from_pretrained("dbmdz/bert-base-italian-xxl-cased")
+model_name = "dbmdz/bert-base-italian-xxl-cased"
+
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+model = AutoModel.from_pretrained(model_name)
+```
+
+To load the Italian XXL ELECTRA model (discriminator), just use:
+
+```python
+from transformers import AutoModel, AutoTokenizer
+
+model_name = "dbmdz/electra-base-italian-xxl-cased-discriminator"
+
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+model = AutoModelWithLMHead.from_pretrained(model_name)
 ```
 
 # Huggingface model hub
@@ -66,7 +98,7 @@ All models are available on the [Huggingface model hub](https://huggingface.co/d
 
 # Contact (Bugs, Feedback, Contribution and more)
 
-For questions about our BERT models just open an issue
+For questions about our BERT/ELECTRA models just open an issue
 [here](https://github.com/dbmdz/berts/issues/new) 🤗
 
 # Acknowledgments

model_cards/dbmdz/electra-base-italian-xxl-cased-discriminator/README.md

@@ -0,0 +1,110 @@
+---
+language: it
+license: mit
+datasets:
+- wikipedia
+---
+
+# 🤗 + 📚 dbmdz BERT and ELECTRA models
+
+In this repository the MDZ Digital Library team (dbmdz) at the Bavarian State
+Library open sources Italian BERT and ELECTRA models 🎉
+
+# Italian BERT
+
+The source data for the Italian BERT model consists of a recent Wikipedia dump and
+various texts from the [OPUS corpora](http://opus.nlpl.eu/) collection. The final
+training corpus has a size of 13GB and 2,050,057,573 tokens.
+
+For sentence splitting, we use NLTK (faster compared to spacy).
+Our cased and uncased models are training with an initial sequence length of 512
+subwords for ~2-3M steps.
+
+For the XXL Italian models, we use the same training data from OPUS and extend
+it with data from the Italian part of the [OSCAR corpus](https://traces1.inria.fr/oscar/).
+Thus, the final training corpus has a size of 81GB and 13,138,379,147 tokens.
+
+Note: Unfortunately, a wrong vocab size was used when training the XXL models.
+This explains the mismatch of the "real" vocab size of 31102, compared to the
+vocab size specified in `config.json`. However, the model is working and all
+evaluations were done under those circumstances.
+See [this issue](https://github.com/dbmdz/berts/issues/7) for more information.
+
+The Italian ELECTRA model was trained on the "XXL" corpus for 1M steps in total using a batch
+size of 128. We pretty much following the ELECTRA training procedure as used for
+[BERTurk](https://github.com/stefan-it/turkish-bert/tree/master/electra).
+
+## Model weights
+
+Currently only PyTorch-[Transformers](https://github.com/huggingface/transformers)
+compatible weights are available. If you need access to TensorFlow checkpoints,
+please raise an issue!
+
+| Model                                                | Downloads
+| ---------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------
+| `dbmdz/bert-base-italian-cased`                      | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/config.json)                                               • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/pytorch_model.bin)                      • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/vocab.txt)
+| `dbmdz/bert-base-italian-uncased`                    | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/config.json)                                             • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/pytorch_model.bin)                    • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/vocab.txt)
+| `dbmdz/bert-base-italian-xxl-cased`                  | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/config.json)                                           • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/pytorch_model.bin)                  • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/vocab.txt)
+| `dbmdz/bert-base-italian-xxl-uncased`                | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/config.json)                                         • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/pytorch_model.bin)                • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/vocab.txt)
+| `dbmdz/electra-base-italian-xxl-cased-discriminator` | [`config.json`](https://s3.amazonaws.com/models.huggingface.co/bert/dbmdz/electra-base-italian-xxl-cased-discriminator/config.json) • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/electra-base-italian-xxl-cased-discriminator/pytorch_model.bin) • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/electra-base-italian-xxl-cased-discriminator/vocab.txt)
+| `dbmdz/electra-base-italian-xxl-cased-generator`     | [`config.json`](https://s3.amazonaws.com/models.huggingface.co/bert/dbmdz/electra-base-italian-xxl-cased-generator/config.json)     • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/electra-base-italian-xxl-cased-generator/pytorch_model.bin)     • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/electra-base-italian-xxl-cased-generator/vocab.txt)
+
+## Results
+
+For results on downstream tasks like NER or PoS tagging, please refer to
+[this repository](https://github.com/stefan-it/italian-bertelectra).
+
+## Usage
+
+With Transformers >= 2.3 our Italian BERT models can be loaded like:
+
+```python
+from transformers import AutoModel, AutoTokenizer
+
+model_name = "dbmdz/bert-base-italian-cased"
+
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+model = AutoModel.from_pretrained(model_name)
+```
+
+To load the (recommended) Italian XXL BERT models, just use:
+
+```python
+from transformers import AutoModel, AutoTokenizer
+
+model_name = "dbmdz/bert-base-italian-xxl-cased"
+
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+model = AutoModel.from_pretrained(model_name)
+```
+
+To load the Italian XXL ELECTRA model (discriminator), just use:
+
+```python
+from transformers import AutoModel, AutoTokenizer
+
+model_name = "dbmdz/electra-base-italian-xxl-cased-discriminator"
+
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+model = AutoModelWithLMHead.from_pretrained(model_name)
+```
+
+# Huggingface model hub
+
+All models are available on the [Huggingface model hub](https://huggingface.co/dbmdz).
+
+# Contact (Bugs, Feedback, Contribution and more)
+
+For questions about our BERT/ELECTRA models just open an issue
+[here](https://github.com/dbmdz/berts/issues/new) 🤗
+
+# Acknowledgments
+
+Research supported with Cloud TPUs from Google's TensorFlow Research Cloud (TFRC).
+Thanks for providing access to the TFRC ❤️
+
+Thanks to the generous support from the [Hugging Face](https://huggingface.co/) team,
+it is possible to download both cased and uncased models from their S3 storage 🤗

model_cards/dbmdz/electra-base-italian-xxl-cased-generator/README.md

@@ -0,0 +1,110 @@
+---
+language: it
+license: mit
+datasets:
+- wikipedia
+---
+
+# 🤗 + 📚 dbmdz BERT and ELECTRA models
+
+In this repository the MDZ Digital Library team (dbmdz) at the Bavarian State
+Library open sources Italian BERT and ELECTRA models 🎉
+
+# Italian BERT
+
+The source data for the Italian BERT model consists of a recent Wikipedia dump and
+various texts from the [OPUS corpora](http://opus.nlpl.eu/) collection. The final
+training corpus has a size of 13GB and 2,050,057,573 tokens.
+
+For sentence splitting, we use NLTK (faster compared to spacy).
+Our cased and uncased models are training with an initial sequence length of 512
+subwords for ~2-3M steps.
+
+For the XXL Italian models, we use the same training data from OPUS and extend
+it with data from the Italian part of the [OSCAR corpus](https://traces1.inria.fr/oscar/).
+Thus, the final training corpus has a size of 81GB and 13,138,379,147 tokens.
+
+Note: Unfortunately, a wrong vocab size was used when training the XXL models.
+This explains the mismatch of the "real" vocab size of 31102, compared to the
+vocab size specified in `config.json`. However, the model is working and all
+evaluations were done under those circumstances.
+See [this issue](https://github.com/dbmdz/berts/issues/7) for more information.
+
+The Italian ELECTRA model was trained on the "XXL" corpus for 1M steps in total using a batch
+size of 128. We pretty much following the ELECTRA training procedure as used for
+[BERTurk](https://github.com/stefan-it/turkish-bert/tree/master/electra).
+
+## Model weights
+
+Currently only PyTorch-[Transformers](https://github.com/huggingface/transformers)
+compatible weights are available. If you need access to TensorFlow checkpoints,
+please raise an issue!
+
+| Model                                                | Downloads
+| ---------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------
+| `dbmdz/bert-base-italian-cased`                      | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/config.json)                                               • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/pytorch_model.bin)                      • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-cased/vocab.txt)
+| `dbmdz/bert-base-italian-uncased`                    | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/config.json)                                             • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/pytorch_model.bin)                    • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-uncased/vocab.txt)
+| `dbmdz/bert-base-italian-xxl-cased`                  | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/config.json)                                           • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/pytorch_model.bin)                  • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-cased/vocab.txt)
+| `dbmdz/bert-base-italian-xxl-uncased`                | [`config.json`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/config.json)                                         • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/pytorch_model.bin)                • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/bert-base-italian-xxl-uncased/vocab.txt)
+| `dbmdz/electra-base-italian-xxl-cased-discriminator` | [`config.json`](https://s3.amazonaws.com/models.huggingface.co/bert/dbmdz/electra-base-italian-xxl-cased-discriminator/config.json) • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/electra-base-italian-xxl-cased-discriminator/pytorch_model.bin) • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/electra-base-italian-xxl-cased-discriminator/vocab.txt)
+| `dbmdz/electra-base-italian-xxl-cased-generator`     | [`config.json`](https://s3.amazonaws.com/models.huggingface.co/bert/dbmdz/electra-base-italian-xxl-cased-generator/config.json)     • [`pytorch_model.bin`](https://cdn.huggingface.co/dbmdz/electra-base-italian-xxl-cased-generator/pytorch_model.bin)     • [`vocab.txt`](https://cdn.huggingface.co/dbmdz/electra-base-italian-xxl-cased-generator/vocab.txt)
+
+## Results
+
+For results on downstream tasks like NER or PoS tagging, please refer to
+[this repository](https://github.com/stefan-it/italian-bertelectra).
+
+## Usage
+
+With Transformers >= 2.3 our Italian BERT models can be loaded like:
+
+```python
+from transformers import AutoModel, AutoTokenizer
+
+model_name = "dbmdz/bert-base-italian-cased"
+
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+model = AutoModel.from_pretrained(model_name)
+```
+
+To load the (recommended) Italian XXL BERT models, just use:
+
+```python
+from transformers import AutoModel, AutoTokenizer
+
+model_name = "dbmdz/bert-base-italian-xxl-cased"
+
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+model = AutoModel.from_pretrained(model_name)
+```
+
+To load the Italian XXL ELECTRA model (discriminator), just use:
+
+```python
+from transformers import AutoModel, AutoTokenizer
+
+model_name = "dbmdz/electra-base-italian-xxl-cased-discriminator"
+
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+model = AutoModelWithLMHead.from_pretrained(model_name)
+```
+
+# Huggingface model hub
+
+All models are available on the [Huggingface model hub](https://huggingface.co/dbmdz).
+
+# Contact (Bugs, Feedback, Contribution and more)
+
+For questions about our BERT/ELECTRA models just open an issue
+[here](https://github.com/dbmdz/berts/issues/new) 🤗
+
+# Acknowledgments
+
+Research supported with Cloud TPUs from Google's TensorFlow Research Cloud (TFRC).
+Thanks for providing access to the TFRC ❤️
+
+Thanks to the generous support from the [Hugging Face](https://huggingface.co/) team,
+it is possible to download both cased and uncased models from their S3 storage 🤗

model_cards/deepset/roberta-base-squad2-v2/README.md

@@ -0,0 +1,117 @@
+---
+datasets:
+- squad_v2
+---
+
+# roberta-base for QA 
+
+## Overview
+**Language model:** roberta-base  
+**Language:** English  
+**Downstream-task:** Extractive QA  
+**Training data:** SQuAD 2.0  
+**Eval data:** SQuAD 2.0  
+**Code:**  See [example](https://github.com/deepset-ai/FARM/blob/master/examples/question_answering.py) in [FARM](https://github.com/deepset-ai/FARM/blob/master/examples/question_answering.py)  
+**Infrastructure**: 4x Tesla v100
+
+## Hyperparameters
+
+```
+batch_size = 96
+n_epochs = 2
+base_LM_model = "roberta-base"
+max_seq_len = 386
+learning_rate = 3e-5
+lr_schedule = LinearWarmup
+warmup_proportion = 0.2
+doc_stride=128
+max_query_length=64
+``` 
+
+## Performance
+Evaluated on the SQuAD 2.0 dev set with the [official eval script](https://worksheets.codalab.org/rest/bundles/0x6b567e1cf2e041ec80d7098f031c5c9e/contents/blob/).
+
+```
+"exact": 79.97136359807968
+"f1": 83.00449234495325
+
+"total": 11873
+"HasAns_exact": 78.03643724696356
+"HasAns_f1": 84.11139298441825
+"HasAns_total": 5928
+"NoAns_exact": 81.90075693860386
+"NoAns_f1": 81.90075693860386
+"NoAns_total": 5945
+```
+
+## Usage
+
+### In Transformers
+```python
+from transformers.pipelines import pipeline
+from transformers.modeling_auto import AutoModelForQuestionAnswering
+from transformers.tokenization_auto import AutoTokenizer
+
+model_name = "deepset/roberta-base-squad2-v2"
+
+# a) Get predictions
+nlp = pipeline('question-answering', model=model_name, tokenizer=model_name)
+QA_input = {
+    'question': 'Why is model conversion important?',
+    'context': 'The option to convert models between FARM and transformers gives freedom to the user and let people easily switch between frameworks.'
+}
+res = nlp(QA_input)
+
+# b) Load model & tokenizer
+model = AutoModelForQuestionAnswering.from_pretrained(model_name)
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+```
+
+### In FARM
+
+```python
+from farm.modeling.adaptive_model import AdaptiveModel
+from farm.modeling.tokenization import Tokenizer
+from farm.infer import Inferencer
+
+model_name = "deepset/roberta-base-squad2-v2"
+
+# a) Get predictions
+nlp = Inferencer.load(model_name, task_type="question_answering")
+QA_input = [{"questions": ["Why is model conversion important?"],
+             "text": "The option to convert models between FARM and transformers gives freedom to the user and let people easily switch between frameworks."}]
+res = nlp.inference_from_dicts(dicts=QA_input, rest_api_schema=True)
+
+# b) Load model & tokenizer
+model = AdaptiveModel.convert_from_transformers(model_name, device="cpu", task_type="question_answering")
+tokenizer = Tokenizer.load(model_name)
+```
+
+### In haystack
+For doing QA at scale (i.e. many docs instead of single paragraph), you can load the model also in [haystack](https://github.com/deepset-ai/haystack/):
+```python
+reader = FARMReader(model_name_or_path="deepset/roberta-base-squad2")
+# or 
+reader = TransformersReader(model="deepset/roberta-base-squad2",tokenizer="deepset/roberta-base-squad2")
+```
+
+
+## Authors
+Branden Chan: `branden.chan [at] deepset.ai`
+Timo Möller: `timo.moeller [at] deepset.ai`
+Malte Pietsch: `malte.pietsch [at] deepset.ai`
+Tanay Soni: `tanay.soni [at] deepset.ai`
+
+## About us
+![deepset logo](https://raw.githubusercontent.com/deepset-ai/FARM/master/docs/img/deepset_logo.png)
+
+We bring NLP to the industry via open source!  
+Our focus: Industry specific language models & large scale QA systems.  
+  
+Some of our work: 
+- [German BERT (aka "bert-base-german-cased")](https://deepset.ai/german-bert)
+- [FARM](https://github.com/deepset-ai/FARM)
+- [Haystack](https://github.com/deepset-ai/haystack/)
+
+Get in touch:
+[Twitter](https://twitter.com/deepset_ai) | [LinkedIn](https://www.linkedin.com/company/deepset-ai/) | [Website](https://deepset.ai)

model_cards/deepset/roberta-base-squad2/README.md

@@ -5,6 +5,8 @@ datasets:
 
 # roberta-base for QA 
 
+NOTE: This model has been superseded by deepset/roberta-base-squad2-v2. For an explanation of why, see [this github issue](https://github.com/deepset-ai/FARM/issues/552) from the FARM repository.
+
 ## Overview
 **Language model:** roberta-base  
 **Language:** English  
@@ -106,4 +108,3 @@ Some of our work:
 
 Get in touch:
 [Twitter](https://twitter.com/deepset_ai) | [LinkedIn](https://www.linkedin.com/company/deepset-ai/) | [Website](https://deepset.ai)
-

model_cards/distilbert-base-cased-README.md

@@ -1,3 +1,40 @@
 ---
+language: en
 license: apache-2.0
+datasets:
+- bookcorpus
+- wikipedia
 ---
+
+# DistilBERT base model (cased)
+
+This model is a distilled version of the [BERT base model](https://huggingface.co/bert-base-cased).
+It was introduced in [this paper](https://arxiv.org/abs/1910.01108).
+The code for the distillation process can be found
+[here](https://github.com/huggingface/transformers/tree/master/examples/distillation).
+This model is cased: it does make a difference between english and English.
+
+All the training details on the pre-training, the uses, limitations and potential biases are the same as for [DistilBERT-base-uncased](https://huggingface.co/distilbert-base-uncased).
+We highly encourage to check it if you want to know more.
+
+## Evaluation results
+
+When fine-tuned on downstream tasks, this model achieves the following results:
+
+Glue test results:
+
+| Task | MNLI | QQP  | QNLI | SST-2 | CoLA | STS-B | MRPC | RTE  |
+|:----:|:----:|:----:|:----:|:-----:|:----:|:-----:|:----:|:----:|
+|      | 81.5 | 87.8 | 88.2 | 90.4  | 47.2 | 85.5  | 85.6 | 60.6 |
+
+### BibTeX entry and citation info
+
+```bibtex
+@article{Sanh2019DistilBERTAD,
+  title={DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter},
+  author={Victor Sanh and Lysandre Debut and Julien Chaumond and Thomas Wolf},
+  journal={ArXiv},
+  year={2019},
+  volume={abs/1910.01108}
+}
+```

model_cards/distilbert-base-cased-distilled-squad-README.md

@@ -6,3 +6,8 @@ metrics:
 - squad
 license: apache-2.0
 ---
+
+# DistilBERT base cased distilled SQuAD
+
+This model is a fine-tune checkpoint of [DistilBERT-base-cased](https://huggingface.co/distilbert-base-cased), fine-tuned using (a second step of) knowledge distillation on SQuAD v1.1.
+This model reaches a F1 score of 87.1 on the dev set (for comparison, BERT bert-base-cased version reaches a F1 score of 88.7).

model_cards/distilbert-base-multilingual-cased-README.md

@@ -1,4 +1,35 @@
 ---
 language: multilingual
 license: apache-2.0
+datasets:
+- wikipedia
 ---
+
+# DistilBERT base multilingual model (cased)
+
+This model is a distilled version of the [BERT base multilingual model](bert-base-multilingual-cased). The code for the distillation process can be found
+[here](https://github.com/huggingface/transformers/tree/master/examples/distillation). This model is cased: it does make a difference between english and English.
+
+The model is trained on the concatenation of Wikipedia in 104 different languages listed [here](https://github.com/google-research/bert/blob/master/multilingual.md#list-of-languages).
+The model has 6 layers, 768 dimension and 12 heads, totalizing 134M parameters (compared to 177M parameters for mBERT-base).
+On average DistilmBERT is twice as fast as mBERT-base.
+
+We encourage to check [BERT base multilingual model](bert-base-multilingual-cased) to know more about usage, limitations and potential biases.
+
+| Model                        | English | Spanish | Chinese | German | Arabic  | Urdu |
+| :---:                        | :---:   | :---:   | :---:   | :---:  | :---:   | :---:|
+| mBERT base cased (computed)  | 82.1    | 74.6    | 69.1    | 72.3   | 66.4    | 58.5 |
+| mBERT base uncased (reported)| 81.4    | 74.3    | 63.8    | 70.5   | 62.1    | 58.3 |
+| DistilmBERT                  | 78.2    | 69.1    | 64.0    | 66.3   | 59.1    | 54.7 |
+
+### BibTeX entry and citation info
+
+```bibtex
+@article{Sanh2019DistilBERTAD,
+  title={DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter},
+  author={Victor Sanh and Lysandre Debut and Julien Chaumond and Thomas Wolf},
+  journal={ArXiv},
+  year={2019},
+  volume={abs/1910.01108}
+}
+```

model_cards/distilbert-base-uncased-README.md

@@ -10,7 +10,7 @@ datasets:
 
 # DistilBERT base model (uncased)
 
-This model is a distilled version of the [BERT base mode](https://huggingface.co/distilbert-base-uncased). It was
+This model is a distilled version of the [BERT base model](https://huggingface.co/bert-base-uncased). It was
 introduced in [this paper](https://arxiv.org/abs/1910.01108). The code for the distillation process can be found
 [here](https://github.com/huggingface/transformers/tree/master/examples/distillation). This model is uncased: it does
 not make a difference between english and English.
@@ -102,7 +102,7 @@ output = model(encoded_input)
 
 Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
 predictions. It also inherits some of
-[the bias of its teacher model](https://huggingface.co/bert-base-uncased#limitations-and-bias). 
+[the bias of its teacher model](https://huggingface.co/bert-base-uncased#limitations-and-bias).
 
 ```python
 >>> from transformers import pipeline
@@ -196,9 +196,9 @@ When fine-tuned on downstream tasks, this model achieves the following results:
 
 Glue test results:
 
-| Task | MNLI | QQP  | QNLI | SST-2 | CoLA | STS-B | MRPC | RTE  | Average |
-|:----:|:----:|:----:|:----:|:-----:|:----:|:-----:|:----:|:----:|:-------:|
-|      | 82.2 | 88.5 | 89.2 | 91.3  | 51.3 | 85.8  | 87.5 | 59.9 | 77.0    |
+| Task | MNLI | QQP  | QNLI | SST-2 | CoLA | STS-B | MRPC | RTE  |
+|:----:|:----:|:----:|:----:|:-----:|:----:|:-----:|:----:|:----:|
+|      | 82.2 | 88.5 | 89.2 | 91.3  | 51.3 | 85.8  | 87.5 | 59.9 |
 
 
 ### BibTeX entry and citation info

model_cards/distilbert-base-uncased-distilled-squad-README.md

@@ -1,4 +1,5 @@
 ---
+language: en
 datasets:
 - squad
 widget:
@@ -8,3 +9,8 @@ widget:
   context: "The Amazon rainforest (Portuguese: Floresta Amazônica or Amazônia; Spanish: Selva Amazónica, Amazonía or usually Amazonia; French: Forêt amazonienne; Dutch: Amazoneregenwoud), also known in English as Amazonia or the Amazon Jungle, is a moist broadleaf forest that covers most of the Amazon basin of South America. This basin encompasses 7,000,000 square kilometres (2,700,000 sq mi), of which 5,500,000 square kilometres (2,100,000 sq mi) are covered by the rainforest. This region includes territory belonging to nine nations. The majority of the forest is contained within Brazil, with 60% of the rainforest, followed by Peru with 13%, Colombia with 10%, and with minor amounts in Venezuela, Ecuador, Bolivia, Guyana, Suriname and French Guiana. States or departments in four nations contain \"Amazonas\" in their names. The Amazon represents over half of the planet's remaining rainforests, and comprises the largest and most biodiverse tract of tropical rainforest in the world, with an estimated 390 billion individual trees divided into 16,000 species."
 license: apache-2.0
 ---
+
+# DistilBERT base uncased distilled SQuAD
+
+This model is a fine-tune checkpoint of [DistilBERT-base-uncased](https://huggingface.co/distilbert-base-uncased), fine-tuned using (a second step of) knowledge distillation on SQuAD v1.1.
+This model reaches a F1 score of 86.9 on the dev set (for comparison, Bert bert-base-uncased version reaches a F1 score of 88.5).

model_cards/distilbert-base-uncased-finetuned-sst-2-english-README.md

@@ -1,3 +1,19 @@
 ---
+language: en
 license: apache-2.0
+datasets:
+- sst-2
 ---
+
+# DistilBERT base uncased finetuned SST-2
+
+This model is a fine-tune checkpoint of [DistilBERT-base-uncased](https://huggingface.co/distilbert-base-uncased), fine-tuned on SST-2.
+This model reaches an accuracy of 91.3 on the dev set (for comparison, Bert bert-base-uncased version reaches an accuracy of 92.7).
+
+# Fine-tuning hyper-parameters
+
+- learning_rate = 1e-5
+- batch_size = 32
+- warmup = 600
+- max_seq_length = 128
+- num_train_epochs = 3.0

model_cards/distilgpt2-README.md

@@ -1,10 +1,21 @@
 ---
+language: en
 tags:
 - exbert
 
 license: apache-2.0
+datasets:
+- openwebtext
 ---
 
+# DistilGPT2
+
+DistilGPT2 English language model pretrained with the supervision of [GPT2](https://huggingface.co/gpt2) (the smallest version of GPT2) on [OpenWebTextCorpus](https://skylion007.github.io/OpenWebTextCorpus/), a reproduction of OpenAI's WebText dataset. The model has 6 layers, 768 dimension and 12 heads, totalizing 82M parameters (compared to 124M parameters for GPT2). On average, DistilGPT2 is two times faster than GPT2.
+
+On the [WikiText-103](https://blog.einstein.ai/the-wikitext-long-term-dependency-language-modeling-dataset/) benchmark, GPT2 reaches a perplexity on the test set of 16.3 compared to 21.1 for DistilGPT2 (after fine-tuning on the train set).
+
+We encourage to check [GPT2](https://huggingface.co/gpt2) to know more about usage, limitations and potential biases.
+
 <a href="https://huggingface.co/exbert/?model=distilgpt2">
 	<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
 </a>

model_cards/distilroberta-base-README.md

@@ -1,10 +1,50 @@
 ---
+language: en
 tags:
 - exbert
 
 license: apache-2.0
+datasets:
+- openwebtext
 ---
 
+# DistilRoBERTa base model
+
+This model is a distilled version of the [RoBERTa-base model](https://huggingface.co/roberta-base). It follows the same training procedure as [DistilBERT](https://huggingface.co/distilbert-base-uncased).
+The code for the distillation process can be found [here](https://github.com/huggingface/transformers/tree/master/examples/distillation).
+This model is case-sensitive: it makes a difference between english and English.
+
+The model has 6 layers, 768 dimension and 12 heads, totalizing 82M parameters (compared to 125M parameters for RoBERTa-base).
+On average DistilRoBERTa is twice as fast as Roberta-base.
+
+We encourage to check [RoBERTa-base model](https://huggingface.co/roberta-base) to know more about usage, limitations and potential biases.
+
+## Training data
+
+DistilRoBERTa was pre-trained on [OpenWebTextCorpus](https://skylion007.github.io/OpenWebTextCorpus/), a reproduction of OpenAI's WebText dataset (it is ~4 times less training data than the teacher RoBERTa).
+
+## Evaluation results
+
+When fine-tuned on downstream tasks, this model achieves the following results:
+
+Glue test results:
+
+| Task | MNLI | QQP  | QNLI | SST-2 | CoLA | STS-B | MRPC | RTE  |
+|:----:|:----:|:----:|:----:|:-----:|:----:|:-----:|:----:|:----:|
+|      | 84.0 | 89.4 | 90.8 | 92.5  | 59.3 | 88.3  | 86.6 | 67.9 |
+
+### BibTeX entry and citation info
+
+```bibtex
+@article{Sanh2019DistilBERTAD,
+  title={DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter},
+  author={Victor Sanh and Lysandre Debut and Julien Chaumond and Thomas Wolf},
+  journal={ArXiv},
+  year={2019},
+  volume={abs/1910.01108}
+}
+```
+
 <a href="https://huggingface.co/exbert/?model=distilroberta-base">
 	<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
 </a>

model_cards/huawei-noah/DynaBERT_MNLI/README.md

@@ -8,13 +8,13 @@ allowing both adaptive width and depth using knowledge distillation.
 * This code is modified based on the repository developed by Hugging Face: [Transformers v2.1.1](https://github.com/huggingface/transformers/tree/v2.1.1), and is released in [GitHub](https://github.com/huawei-noah/Pretrained-Language-Model/tree/master/DynaBERT).
 
 ### Reference
-Lu Hou, Zhiqi Huang, Lifeng Shang, Xin Jiang, Qun Liu.
+Lu Hou, Zhiqi Huang, Lifeng Shang, Xin Jiang, Xiao Chen, Qun Liu.
 [DynaBERT: Dynamic BERT with Adaptive Width and Depth](https://arxiv.org/abs/2004.04037).
 ```
 @inproceedings{hou2020dynabert,
   title = {DynaBERT: Dynamic BERT with Adaptive Width and Depth},
-  author = {Lu Hou, Zhiqi Huang, Lifeng Shang, Xin Jiang, Qun Liu},  
-  booktitle = {NeurIPS},
+  author = {Lu Hou, Zhiqi Huang, Lifeng Shang, Xin Jiang, Xiao Chen, Qun Liu},  
+  booktitle = {Advances in Neural Information Processing Systems},
   year = {2020}
 }
 ```

model_cards/huawei-noah/DynaBERT_SST-2/README.md

@@ -8,13 +8,13 @@ allowing both adaptive width and depth using knowledge distillation.
 * This code is modified based on the repository developed by Hugging Face: [Transformers v2.1.1](https://github.com/huggingface/transformers/tree/v2.1.1), and is released in [GitHub](https://github.com/huawei-noah/Pretrained-Language-Model/tree/master/DynaBERT).
 
 ### Reference
-Lu Hou, Zhiqi Huang, Lifeng Shang, Xin Jiang, Qun Liu.
+Lu Hou, Zhiqi Huang, Lifeng Shang, Xin Jiang, Xiao Chen, Qun Liu.
 [DynaBERT: Dynamic BERT with Adaptive Width and Depth](https://arxiv.org/abs/2004.04037).
 ```
 @inproceedings{hou2020dynabert,
   title = {DynaBERT: Dynamic BERT with Adaptive Width and Depth},
-  author = {Lu Hou, Zhiqi Huang, Lifeng Shang, Xin Jiang, Qun Liu},  
-  booktitle = {NeurIPS},
+  author = {Lu Hou, Zhiqi Huang, Lifeng Shang, Xin Jiang, Xiao Chen, Qun Liu},  
+  booktitle = {Advances in Neural Information Processing Systems},
   year = {2020}
 }
 ```

model_cards/ktrapeznikov/gpt2-medium-topic-news/README.md

@@ -0,0 +1,41 @@
+---
+language: 
+- en
+thumbnail:
+widget:
+ - text: "topic: climate article:"
+---
+
+# GPT2-medium-topic-news
+
+## Model description
+
+GPT2-medium fine tuned on a large news corpus conditioned on a topic
+
+## Intended uses & limitations
+
+#### How to use
+
+To generate a news article text conditioned on a topic, prompt model with: 
+`topic: climate article:`
+
+The following tags were used during training:
+`arts law international science business politics disaster world conflict football sport sports artanddesign environment music film lifeandstyle business health commentisfree books technology media education politics travel stage uk society us money culture religion science news tv fashion uk australia cities global childrens sustainable global voluntary housing law local healthcare theguardian`
+
+Zero shot generation works pretty well as long as `topic` is a single word and not too specific.
+
+```python
+device = "cuda:0"
+tokenizer = AutoTokenizer.from_pretrained("ktrapeznikov/gpt2-medium-topic-news")
+model = AutoModelWithLMHead.from_pretrained("ktrapeznikov/gpt2-medium-topic-news")
+model.to(device)
+topic = "climate"
+prompt = tokenizer(f"topic: {topic} article:", return_tensors="pt")
+out = model.generate(prompt["input_ids"].to(device), do_sample=True,max_length=500, early_stopping=True, top_p=.9)
+print(tokenizer.decode(list(out.cpu()[0])))
+```
+
+## Training data
+
+
+## Training procedure

model_cards/kuppuluri/telugu_bertu/README.md

@@ -0,0 +1,24 @@
+---
+language: te
+---
+# telugu_bertu
+
+## Model description
+
+This model is a BERT MLM model trained on Telugu.
+
+## Intended uses & limitations
+
+#### How to use
+
+```python
+from transformers import AutoModelWithLMHead, AutoTokenizer, pipeline
+tokenizer = AutoTokenizer.from_pretrained("kuppuluri/telugu_bertu",
+                                          clean_text=False,
+                                          handle_chinese_chars=False,
+                                          strip_accents=False,
+                                          wordpieces_prefix='##')
+model = AutoModelWithLMHead.from_pretrained("kuppuluri/telugu_bertu")
+fill_mask = pipeline("fill-mask", model=model, tokenizer=tokenizer)
+results = fill_mask("మక్దూంపల్లి పేరుతో చాలా [MASK] ఉన్నాయి.")
+```

model_cards/kuppuluri/telugu_bertu_ner/README.md

@@ -0,0 +1,35 @@
+# Named Entity Recognition Model for Telugu
+
+#### How to use
+
+```python
+from simpletransformers.ner import NERModel
+model = NERModel('bert',
+                 'kuppuluri/telugu_bertu_ner',
+                 labels=[
+                     'B-PERSON', 'I-ORG', 'B-ORG', 'I-LOC', 'B-MISC',
+                     'I-MISC', 'I-PERSON', 'B-LOC', 'O'
+                 ],
+                 use_cuda=False,
+                 args={"use_multiprocessing": False})
+
+text = "విరాట్ కోహ్లీ కూడా అదే నిర్లక్ష్యాన్ని ప్రదర్శించి కేవలం ఒక పరుగుకే రనౌటై పెవిలియన్ చేరాడు ."
+results = model.predict([text])
+```
+
+## Training data
+
+Training data is from https://github.com/anikethjr/NER_Telugu
+
+## Eval results
+
+On the test set my results were
+
+eval_loss = 0.0004407190410447974
+
+f1_score = 0.999519076627124
+
+precision = 0.9994389677005691
+
+recall = 0.9995991983967936
+

model_cards/kuppuluri/telugu_bertu_pos/README.md

@@ -0,0 +1,36 @@
+# Part of Speech tagging Model for Telugu
+
+#### How to use
+
+```python
+from simpletransformers.ner import NERModel
+model = NERModel('bert',
+                 'kuppuluri/telugu_bertu_pos',
+                 args={"use_multiprocessing": False},
+                 labels=[
+                     'QC', 'JJ', 'NN', 'QF', 'RDP', 'O',
+                     'NNO', 'PRP', 'RP', 'VM', 'WQ',
+                     'PSP', 'UT', 'CC', 'INTF', 'SYMP',
+                     'NNP', 'INJ', 'SYM', 'CL', 'QO',
+                     'DEM', 'RB', 'NST', ],
+                 use_cuda=False)
+
+text = "విరాట్ కోహ్లీ కూడా అదే నిర్లక్ష్యాన్ని ప్రదర్శించి కేవలం ఒక పరుగుకే రనౌటై పెవిలియన్ చేరాడు ."
+results = model.predict([text])
+```
+
+## Training data
+
+Training data is from https://github.com/anikethjr/NER_Telugu
+
+## Eval results
+
+On the test set my results were
+
+eval_loss = 0.0036797842364565416
+
+f1_score = 0.9983795127912227
+
+precision = 0.9984325602401637
+
+recall = 0.9983264709788816

model_cards/kuppuluri/telugu_bertu_tydiqa/README.md

@@ -0,0 +1,18 @@
+# Telugu Question-Answering model trained on Tydiqa dataset from Google
+
+#### How to use
+
+```python
+from transformers.pipelines import pipeline, AutoModelForQuestionAnswering, AutoTokenizer
+model = AutoModelForQuestionAnswering.from_pretrained(model_name)
+tokenizer = AutoTokenizer.from_pretrained("kuppuluri/telugu_bertu_tydiqa",
+                                          clean_text=False,
+                                          handle_chinese_chars=False,
+                                          strip_accents=False,
+                                          wordpieces_prefix='##')
+nlp = pipeline('question-answering', model=model, tokenizer=tokenizer)
+result = nlp({'question': question, 'context': context})
+```
+
+## Training data
+I used Tydiqa Telugu data from Google https://github.com/google-research-datasets/tydiqa

model_cards/mrm8488/GPT-2-finetuned-common_gen/README.md

@@ -0,0 +1,63 @@
+---
+language: en
+datasets:
+- common_gen
+widget:
+- text: "<|endoftext|> apple, tree, pick:"
+---
+
+# GPT-2 fine-tuned on CommonGen
+
+[GPT-2](https://huggingface.co/gpt2) fine-tuned on [CommonGen](https://inklab.usc.edu/CommonGen/index.html) for *Generative Commonsense Reasoning*.
+
+## Details of GPT-2
+
+GPT-2 is a transformers model pretrained on a very large corpus of English data in a self-supervised fashion. This
+means it was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots
+of publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely,
+it was trained to guess the next word in sentences.
+
+More precisely, inputs are sequences of continuous text of a certain length and the targets are the same sequence,
+shifted one token (word or piece of word) to the right. The model uses internally a mask-mechanism to make sure the
+predictions for the token `i` only uses the inputs from `1` to `i` but not the future tokens.
+
+This way, the model learns an inner representation of the English language that can then be used to extract features
+useful for downstream tasks. The model is best at what it was pretrained for however, which is generating texts from a
+prompt.
+
+
+## Details of the dataset 📚 
+
+CommonGen is a constrained text generation task, associated with a benchmark dataset, to explicitly test machines for the ability of generative commonsense reasoning. Given a set of common concepts; the task is to generate a coherent sentence describing an everyday scenario using these concepts.
+
+CommonGen is challenging because it inherently requires 1) relational reasoning using background commonsense knowledge, and 2) compositional generalization ability to work on unseen concept combinations. Our dataset, constructed through a combination of crowd-sourcing from AMT and existing caption corpora, consists of 30k concept-sets and 50k sentences in total.
+
+
+| Dataset  | Split | # samples |
+| -------- | ----- | --------- |
+| common_gen | train  | 67389   |
+| common_gen | valid  | 4018    |
+| common_gen | test   | 1497    |
+
+
+## Model fine-tuning 🏋️‍
+
+You can find the fine-tuning script [here](https://github.com/huggingface/transformers/tree/master/examples/language-modeling)
+
+## Model in Action 🚀
+
+```bash
+python ./transformers/examples/text-generation/run_generation.py \
+    --model_type=gpt2 \
+    --model_name_or_path="mrm8488/GPT-2-finetuned-common_gen" \
+    --num_return_sequences 1 \
+    --prompt "<|endoftext|> kid, room, dance:" \
+    --stop_token "."
+```
+
+> Created by [Manuel Romero/@mrm8488](https://twitter.com/mrm8488) | [LinkedIn](https://www.linkedin.com/in/manuel-romero-cs/)
+
+> Made with <span style="color: #e25555;">&hearts;</span> in Spain
+
+
+

model_cards/mrm8488/RuPERTa-base-finetuned-ner/README.md

@@ -17,7 +17,7 @@ This model is a fine-tuned on [NER-C](https://www.kaggle.com/nltkdata/conll-corp
 | Dev                    | 40 K |
 
 
-- [Fine-tune on NER script provided by Huggingface](https://github.com/huggingface/transformers/blob/master/examples/token-classification/run_ner.py)
+- [Fine-tune on NER script provided by Huggingface](https://github.com/huggingface/transformers/blob/master/examples/token-classification/run_ner_old.py)
 
 - Labels covered:
 

model_cards/mrm8488/RuPERTa-base-finetuned-pos/README.md

@@ -16,7 +16,7 @@ This model is a fine-tuned on [CONLL CORPORA](https://www.kaggle.com/nltkdata/co
 | Train                  | 445 K |
 | Dev                    | 55 K |
 
-- [Fine-tune on NER script provided by Huggingface](https://github.com/huggingface/transformers/blob/master/examples/token-classification/run_ner.py)
+- [Fine-tune on NER script provided by Huggingface](https://github.com/huggingface/transformers/blob/master/examples/token-classification/run_ner_old.py)
 
 - Labels covered:
 

model_cards/mrm8488/TinyBERT-spanish-uncased-finetuned-ner/README.md

@@ -19,7 +19,7 @@ I preprocessed the dataset and split it as train / dev (80/20)
 | Dev                    | 2.2 K |
 
 
-- [Fine-tune on NER script provided by Huggingface](https://github.com/huggingface/transformers/blob/master/examples/token-classification/run_ner.py)
+- [Fine-tune on NER script provided by Huggingface](https://github.com/huggingface/transformers/blob/master/examples/token-classification/run_ner_old.py)
 
 - Labels covered:
 

model_cards/mrm8488/bert-base-german-finetuned-ler/README.md

@@ -18,7 +18,7 @@ Court decisions from 2017 and 2018 were selected for the dataset, published onli
 | Train                  | 1657048 |
 | Eval                    | 500000 |
 
-- Training script: [Fine-tuning script for NER provided by Huggingface](https://github.com/huggingface/transformers/blob/master/examples/token-classification/run_ner.py)
+- Training script: [Fine-tuning script for NER provided by Huggingface](https://github.com/huggingface/transformers/blob/master/examples/token-classification/run_ner_old.py)
 Colab: [How to fine-tune a model for NER using HF scripts](https://colab.research.google.com/drive/156Qrd7NsUHwA3nmQ6gXdZY0NzOvqk9AT?usp=sharing)
 
 - Labels covered (and its distribution):

model_cards/mrm8488/bert-small-finetuned-typo-detection/README.md

@@ -11,7 +11,7 @@ thumbnail:
 
 - Dataset: [GitHub Typo Corpus](https://github.com/mhagiwara/github-typo-corpus) 📚
 
-- [Fine-tune script on NER dataset provided by Huggingface](https://github.com/huggingface/transformers/blob/master/examples/token-classification/run_ner.py) 🏋️‍♂️
+- [Fine-tune script on NER dataset provided by Huggingface](https://github.com/huggingface/transformers/blob/master/examples/token-classification/run_ner_old.py) 🏋️‍♂️
 
 ## Metrics on test set 📋
 

model_cards/mrm8488/bert-spanish-cased-finetuned-ner/README.md

@@ -19,7 +19,7 @@ I preprocessed the dataset and split it as train / dev (80/20)
 | Dev                    | 2.2 K |
 
 
-- [Fine-tune on NER script provided by Huggingface](https://github.com/huggingface/transformers/blob/master/examples/token-classification/run_ner.py)
+- [Fine-tune on NER script provided by Huggingface](https://github.com/huggingface/transformers/blob/master/examples/token-classification/run_ner_old.py)
 
 - Labels covered:
 

model_cards/mrm8488/bert-spanish-cased-finetuned-pos-syntax/README.md

@@ -11,7 +11,7 @@ This model is a fine-tuned version of the Spanish BERT [(BETO)](https://github.c
 
 - [Dataset: CONLL Corpora ES](https://www.kaggle.com/nltkdata/conll-corpora)
 
-#### [Fine-tune script on NER dataset provided by Huggingface](https://github.com/huggingface/transformers/blob/master/examples/token-classification/run_ner.py)
+#### [Fine-tune script on NER dataset provided by Huggingface](https://github.com/huggingface/transformers/blob/master/examples/token-classification/run_ner_old.py)
 
 #### 21 Syntax annotations (Labels) covered:
 

model_cards/mrm8488/bert-spanish-cased-finetuned-pos/README.md

@@ -19,7 +19,7 @@ I preprocessed the dataset and split it as train / dev (80/20)
 | Dev                    | 50 K |
 
 
-- [Fine-tune on NER script provided by Huggingface](https://github.com/huggingface/transformers/blob/master/examples/token-classification/run_ner.py)
+- [Fine-tune on NER script provided by Huggingface](https://github.com/huggingface/transformers/blob/master/examples/token-classification/run_ner_old.py)
 
 - **60** Labels covered: