(Re-)Enable Nightly + Past CI (#22393)

* Enable Nightly + Past CI

* put schedule

---------

Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

.github/workflows/build-docker-images.yml

@@ -3,7 +3,7 @@ name: Build docker images (scheduled)
 on:
   push:
     branches:
-      - docker-image*
+      - build_ci_docker_image*
   repository_dispatch:
   workflow_call:
     inputs:
@@ -67,35 +67,6 @@ jobs:
           push: true
           tags: huggingface/transformers-all-latest-gpu-push-ci
 
-  latest-with-torch-nightly-docker:
-    name: "Nightly PyTorch + Stable TensorFlow"
-    # Push CI doesn't need this image
-    if: inputs.image_postfix != '-push-ci'
-    runs-on: ubuntu-latest
-    steps:
-      -
-        name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v2
-      -
-        name: Check out code
-        uses: actions/checkout@v3
-      -
-        name: Login to DockerHub
-        uses: docker/login-action@v2
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_PASSWORD }}
-      -
-        name: Build and push
-        uses: docker/build-push-action@v3
-        with:
-          context: ./docker/transformers-all-latest-gpu
-          build-args: |
-            REF=main
-            PYTORCH=pre
-          push: true
-          tags: huggingface/transformers-all-latest-torch-nightly-gpu
-
   latest-torch-deepspeed-docker:
     name: "Latest PyTorch + DeepSpeed"
     runs-on: ubuntu-latest
@@ -153,34 +124,6 @@ jobs:
           push: true
           tags: huggingface/transformers-pytorch-deepspeed-latest-gpu-push-ci
 
-  nightly-torch-deepspeed-docker:
-    name: "Nightly PyTorch + DeepSpeed"
-    # Push CI doesn't need this image
-    if: inputs.image_postfix != '-push-ci'
-    runs-on: ubuntu-latest
-    steps:
-      -
-        name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v2
-      -
-        name: Check out code
-        uses: actions/checkout@v3
-      -
-        name: Login to DockerHub
-        uses: docker/login-action@v2
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_PASSWORD }}
-      -
-        name: Build and push
-        uses: docker/build-push-action@v3
-        with:
-          context: ./docker/transformers-pytorch-deepspeed-nightly-gpu
-          build-args: |
-            REF=main
-          push: true
-          tags: huggingface/transformers-pytorch-deepspeed-nightly-gpu
-
   doc-builder:
     name: "Doc builder"
     # Push CI doesn't need this image

.github/workflows/build-nightly-ci-docker-images.yml

@@ -0,0 +1,75 @@
+name: Build docker images (Nightly CI)
+
+on:
+  workflow_call:
+  push:
+    branches:
+      - build_nightly_ci_docker_image*
+
+concurrency:
+  group: docker-images-builds
+  cancel-in-progress: false
+
+jobs:
+  latest-with-torch-nightly-docker:
+    name: "Nightly PyTorch + Stable TensorFlow"
+    runs-on: ubuntu-latest
+    steps:
+      - name: Cleanup disk
+        run: |
+          sudo ls -l /usr/local/lib/
+          sudo ls -l /usr/share/
+          sudo du -sh /usr/local/lib/
+          sudo du -sh /usr/share/
+          sudo rm -rf /usr/local/lib/android
+          sudo rm -rf /usr/share/dotnet
+          sudo du -sh /usr/local/lib/
+          sudo du -sh /usr/share/
+      -
+        name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v2
+      -
+        name: Check out code
+        uses: actions/checkout@v3
+      -
+        name: Login to DockerHub
+        uses: docker/login-action@v2
+        with:
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_PASSWORD }}
+      -
+        name: Build and push
+        uses: docker/build-push-action@v3
+        with:
+          context: ./docker/transformers-all-latest-gpu
+          build-args: |
+            REF=main
+            PYTORCH=pre
+          push: true
+          tags: huggingface/transformers-all-latest-torch-nightly-gpu
+
+  nightly-torch-deepspeed-docker:
+    name: "Nightly PyTorch + DeepSpeed"
+    runs-on: ubuntu-latest
+    steps:
+      -
+        name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v2
+      -
+        name: Check out code
+        uses: actions/checkout@v3
+      -
+        name: Login to DockerHub
+        uses: docker/login-action@v2
+        with:
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_PASSWORD }}
+      -
+        name: Build and push
+        uses: docker/build-push-action@v3
+        with:
+          context: ./docker/transformers-pytorch-deepspeed-nightly-gpu
+          build-args: |
+            REF=main
+          push: true
+          tags: huggingface/transformers-pytorch-deepspeed-nightly-gpu

.github/workflows/build-past-ci-docker-images.yml

@@ -3,7 +3,7 @@ name: Build docker images (Past CI)
 on:
   push:
     branches:
-      - past-ci-docker-image*
+      - build_past_ci_docker_image*
 
 concurrency:
   group: docker-images-builds
@@ -15,7 +15,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        version: ["1.11", "1.10", "1.9", "1.8", "1.7", "1.6", "1.5", "1.4"]
+        version: ["1.13", "1.12", "1.11", "1.10", "1.9"]
     runs-on: ubuntu-latest
     steps:
       -
@@ -24,6 +24,17 @@ jobs:
       -
         name: Check out code
         uses: actions/checkout@v3
+      -
+        id: get-base-image
+        name: Get Base Image
+        env:
+          framework_version: ${{ matrix.version }}
+        run: |
+          echo "base_image=$(python3 -c 'import os; from utils.past_ci_versions import past_versions_testing; base_image = past_versions_testing["pytorch"][os.environ["framework_version"]]["base_image"]; print(base_image)')" >> $GITHUB_OUTPUT
+      -
+        name: Print Base Image
+        run: |
+          echo ${{ steps.get-base-image.outputs.base_image }}
       -
         name: Login to DockerHub
         uses: docker/login-action@v2
@@ -37,6 +48,7 @@ jobs:
           context: ./docker/transformers-past-gpu
           build-args: |
             REF=main
+            BASE_DOCKER_IMAGE=${{ steps.get-base-image.outputs.base_image }}
             FRAMEWORK=pytorch
             VERSION=${{ matrix.version }}
           push: true
@@ -47,7 +59,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        version: ["2.8", "2.7", "2.6", "2.5"]
+        version: ["2.11", "2.10", "2.9", "2.8", "2.7", "2.6", "2.5"]
     runs-on: ubuntu-latest
     steps:
       -
@@ -56,6 +68,17 @@ jobs:
       -
         name: Check out code
         uses: actions/checkout@v3
+      -
+        id: get-base-image
+        name: Get Base Image
+        env:
+          framework_version: ${{ matrix.version }}
+        run: |
+          echo "base_image=$(python3 -c 'import os; from utils.past_ci_versions import past_versions_testing; base_image = past_versions_testing["tensorflow"][os.environ["framework_version"]]["base_image"]; print(base_image)')" >> $GITHUB_OUTPUT
+      -
+        name: Print Base Image
+        run: |
+          echo ${{ steps.get-base-image.outputs.base_image }}
       -
         name: Login to DockerHub
         uses: docker/login-action@v2
@@ -69,39 +92,7 @@ jobs:
           context: ./docker/transformers-past-gpu
           build-args: |
             REF=main
-            FRAMEWORK=tensorflow
+            BASE_DOCKER_IMAGE=${{ steps.get-base-image.outputs.base_image }}
-            VERSION=${{ matrix.version }}
-          push: true
-          tags: huggingface/transformers-tensorflow-past-${{ matrix.version }}-gpu
-
-  past-tensorflow-docker-2-4:
-    name: "Past TensorFlow Docker"
-    strategy:
-      fail-fast: false
-      matrix:
-        version: ["2.4"]
-    runs-on: ubuntu-latest
-    steps:
-      -
-        name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v2
-      -
-        name: Check out code
-        uses: actions/checkout@v3
-      -
-        name: Login to DockerHub
-        uses: docker/login-action@v2
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_PASSWORD }}
-      -
-        name: Build and push
-        uses: docker/build-push-action@v3
-        with:
-          context: ./docker/transformers-past-gpu
-          build-args: |
-            REF=main
-            BASE_DOCKER_IMAGE=nvidia/cuda:11.0.3-cudnn8-devel-ubuntu20.04
             FRAMEWORK=tensorflow
             VERSION=${{ matrix.version }}
           push: true

.github/workflows/self-nightly-past-ci-caller.yml

@@ -0,0 +1,143 @@
+name: Self-hosted runner (nightly-past-ci-caller)
+
+on:
+  schedule:
+    # 2 am on each Sunday and Thursday
+    - cron: "0 2 * * 0,4"
+  push:
+    branches:
+      - run_nightly_ci*
+      - run_past_ci*
+
+jobs:
+  build_nightly_ci_images:
+    name: Build Nightly CI Docker Images
+    if: (github.event_name == 'schedule') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_nightly_ci'))
+    uses: ./.github/workflows/build-nightly-ci-docker-images.yml
+    secrets: inherit
+
+  run_nightly_ci:
+    name: Nightly CI
+    needs: [build_nightly_ci_images]
+    uses: ./.github/workflows/self-nightly-scheduled.yml
+    secrets: inherit
+
+  run_past_ci_pytorch_1-13:
+    name: PyTorch 1.13
+    if: (cancelled() != true) && ((github.event_name == 'schedule') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci')))
+    needs: [run_nightly_ci]
+    uses: ./.github/workflows/self-past.yml
+    with:
+      framework: pytorch
+      version: "1.13"
+    secrets: inherit
+
+  run_past_ci_pytorch_1-12:
+    name: PyTorch 1.12
+    if: (cancelled() != true) && ((github.event_name == 'schedule') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci')))
+    needs: [run_past_ci_pytorch_1-13]
+    uses: ./.github/workflows/self-past.yml
+    with:
+      framework: pytorch
+      version: "1.12"
+    secrets: inherit
+
+  run_past_ci_pytorch_1-11:
+    name: PyTorch 1.11
+    if: (cancelled() != true) && ((github.event_name == 'schedule') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci')))
+    needs: [run_past_ci_pytorch_1-12]
+    uses: ./.github/workflows/self-past.yml
+    with:
+      framework: pytorch
+      version: "1.11"
+    secrets: inherit
+
+  run_past_ci_pytorch_1-10:
+    name: PyTorch 1.10
+    if: (cancelled() != true) && ((github.event_name == 'schedule') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci')))
+    needs: [run_past_ci_pytorch_1-11]
+    uses: ./.github/workflows/self-past.yml
+    with:
+      framework: pytorch
+      version: "1.10"
+    secrets: inherit
+
+  run_past_ci_pytorch_1-9:
+    name: PyTorch 1.9
+    if: (cancelled() != true) && ((github.event_name == 'schedule') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci')))
+    needs: [run_past_ci_pytorch_1-10]
+    uses: ./.github/workflows/self-past.yml
+    with:
+      framework: pytorch
+      version: "1.9"
+    secrets: inherit
+
+  run_past_ci_tensorflow_2-11:
+    name: TensorFlow 2.11
+    if: (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
+    needs: [run_past_ci_pytorch_1-9]
+    uses: ./.github/workflows/self-past.yml
+    with:
+      framework: tensorflow
+      version: "2.11"
+    secrets: inherit
+
+  run_past_ci_tensorflow_2-10:
+    name: TensorFlow 2.10
+    if: (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
+    needs: [run_past_ci_tensorflow_2-11]
+    uses: ./.github/workflows/self-past.yml
+    with:
+      framework: tensorflow
+      version: "2.10"
+    secrets: inherit
+
+  run_past_ci_tensorflow_2-9:
+    name: TensorFlow 2.9
+    if: (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
+    needs: [run_past_ci_tensorflow_2-10]
+    uses: ./.github/workflows/self-past.yml
+    with:
+      framework: tensorflow
+      version: "2.9"
+    secrets: inherit
+
+  run_past_ci_tensorflow_2-8:
+    name: TensorFlow 2.8
+    if: (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
+    needs: [run_past_ci_tensorflow_2-9]
+    uses: ./.github/workflows/self-past.yml
+    with:
+      framework: tensorflow
+      version: "2.8"
+    secrets: inherit
+
+  run_past_ci_tensorflow_2-7:
+    name: TensorFlow 2.7
+    if: (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
+    needs: [run_past_ci_tensorflow_2-8]
+    uses: ./.github/workflows/self-past.yml
+    with:
+      framework: tensorflow
+      version: "2.7"
+    secrets: inherit
+
+  run_past_ci_tensorflow_2-6:
+    name: TensorFlow 2.6
+    if: (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
+    needs: [run_past_ci_tensorflow_2-7]
+    uses: ./.github/workflows/self-past.yml
+    with:
+      framework: tensorflow
+      version: "2.6"
+    secrets: inherit
+
+  run_past_ci_tensorflow_2-5:
+    name: TensorFlow 2.5
+    if: (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
+    needs: [run_past_ci_tensorflow_2-6]
+    uses: ./.github/workflows/self-past.yml
+    with:
+      framework: tensorflow
+      version: "2.5"
+    secrets: inherit

.github/workflows/self-nightly-scheduled.yml

@@ -1,4 +1,4 @@
-name: Self-hosted runner (nightly)
+name: Self-hosted runner (nightly-ci)
 
 # Note that each job's dependencies go into a corresponding docker file.
 #
@@ -8,9 +8,7 @@ name: Self-hosted runner (nightly)
 
 on:
   repository_dispatch:
-# Disable temporarily until the test suite can be run under 12 hours.
+  workflow_call:
-#  schedule:
-#    - cron: "0 16 * * *"
 
 env:
   HF_HOME: /mnt/cache
@@ -33,7 +31,7 @@ jobs:
           fetch-depth: 2
 
       - name: Check Runner Status
-        run: python utils/check_self_hosted_runner.py --target_runners single-gpu-scheduled-ci-runner-docker,multi-gpu-scheduled-ci-runner-docker --token ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
+        run: python utils/check_self_hosted_runner.py --target_runners single-gpu-past-ci-runner-docker,multi-gpu-past-ci-runner-docker --token ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
 
   check_runners:
     name: Check Runners
@@ -41,7 +39,7 @@ jobs:
     strategy:
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
+    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker-past-ci') }}
     container:
       image: huggingface/transformers-all-latest-torch-nightly-gpu
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -56,7 +54,7 @@ jobs:
     strategy:
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
+    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker-past-ci') }}
     container:
       image: huggingface/transformers-all-latest-torch-nightly-gpu
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -96,7 +94,7 @@ jobs:
       matrix:
         folders: ${{ fromJson(needs.setup.outputs.matrix) }}
         machine_type: [single-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
+    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker-past-ci') }}
     container:
       image: huggingface/transformers-all-latest-torch-nightly-gpu
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -143,7 +141,7 @@ jobs:
         if: ${{ always() }}
         uses: actions/upload-artifact@v3
         with:
-          name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports
+          name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports_postfix_nightly
           path: /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}
 
   run_tests_multi_gpu:
@@ -153,7 +151,7 @@ jobs:
       matrix:
         folders: ${{ fromJson(needs.setup.outputs.matrix) }}
         machine_type: [multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
+    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker-past-ci') }}
     container:
       image: huggingface/transformers-all-latest-torch-nightly-gpu
       options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -200,7 +198,7 @@ jobs:
         if: ${{ always() }}
         uses: actions/upload-artifact@v3
         with:
-          name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports
+          name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports_postfix_nightly
           path: /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}
 
   run_all_tests_torch_cuda_extensions_gpu:
@@ -209,7 +207,7 @@ jobs:
       fail-fast: false
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
+    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker-past-ci') }}
     needs: setup
     container:
       image: huggingface/transformers-pytorch-deepspeed-nightly-gpu
@@ -258,7 +256,7 @@ jobs:
         if: ${{ always() }}
         uses: actions/upload-artifact@v3
         with:
-          name: ${{ matrix.machine_type }}_run_tests_torch_cuda_extensions_gpu_test_reports
+          name: ${{ matrix.machine_type }}_run_tests_torch_cuda_extensions_gpu_test_reports_postfix_nightly
           path: /workspace/transformers/reports/${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu
 
   send_results:
@@ -292,7 +290,7 @@ jobs:
           CI_SLACK_CHANNEL_DUMMY_TESTS: ${{ secrets.CI_SLACK_CHANNEL_DUMMY_TESTS }}
           CI_SLACK_REPORT_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID_PAST_FUTURE }}
           ACCESS_REPO_INFO_TOKEN: ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
-          CI_EVENT: nightly-build
+          CI_EVENT: Nightly CI
           RUNNER_STATUS: ${{ needs.check_runner_status.result }}
           RUNNER_ENV_STATUS: ${{ needs.check_runners.result }}
           SETUP_STATUS: ${{ needs.setup.result }}
@@ -302,3 +300,11 @@ jobs:
           pip install slack_sdk
           pip show slack_sdk
           python utils/notification_service.py "${{ needs.setup.outputs.matrix }}"
+
+
+      # delete-artifact
+      - uses: geekyeggo/delete-artifact@v2
+        with:
+          name: |
+              single-*
+              multi-*

.github/workflows/self-past-caller.yml

@@ -1,136 +0,0 @@
-name: Self-hosted runner (past-ci-caller)
-
-on:
-  push:
-    branches:
-      - run-past-ci*
-
-jobs:
-  run_past_ci_pytorch_1-11:
-    name: PyTorch 1.11
-    if: always()
-    uses: ./.github/workflows/self-past.yml
-    with:
-      framework: pytorch
-      version: "1.11"
-    secrets: inherit
-
-  run_past_ci_pytorch_1-10:
-    name: PyTorch 1.10
-    if: always()
-    needs: [run_past_ci_pytorch_1-11]
-    uses: ./.github/workflows/self-past.yml
-    with:
-      framework: pytorch
-      version: "1.10"
-    secrets: inherit
-
-  run_past_ci_pytorch_1-9:
-    name: PyTorch 1.9
-    if: always()
-    needs: [run_past_ci_pytorch_1-10]
-    uses: ./.github/workflows/self-past.yml
-    with:
-      framework: pytorch
-      version: "1.9"
-    secrets: inherit
-
-  run_past_ci_pytorch_1-8:
-    name: PyTorch 1.8
-    if: always()
-    needs: [run_past_ci_pytorch_1-9]
-    uses: ./.github/workflows/self-past.yml
-    with:
-      framework: pytorch
-      version: "1.8"
-    secrets: inherit
-
-  run_past_ci_pytorch_1-7:
-    name: PyTorch 1.7
-    if: always()
-    needs: [run_past_ci_pytorch_1-8]
-    uses: ./.github/workflows/self-past.yml
-    with:
-      framework: pytorch
-      version: "1.7"
-    secrets: inherit
-
-  run_past_ci_pytorch_1-6:
-    name: PyTorch 1.6
-    if: always()
-    needs: [run_past_ci_pytorch_1-7]
-    uses: ./.github/workflows/self-past.yml
-    with:
-      framework: pytorch
-      version: "1.6"
-    secrets: inherit
-
-  run_past_ci_pytorch_1-5:
-    name: PyTorch 1.5
-    if: always()
-    needs: [run_past_ci_pytorch_1-6]
-    uses: ./.github/workflows/self-past.yml
-    with:
-      framework: pytorch
-      version: "1.5"
-    secrets: inherit
-
-  run_past_ci_pytorch_1-4:
-    name: PyTorch 1.4
-    if: always()
-    needs: [run_past_ci_pytorch_1-5]
-    uses: ./.github/workflows/self-past.yml
-    with:
-      framework: pytorch
-      version: "1.4"
-    secrets: inherit
-
-  run_past_ci_tensorflow_2-8:
-    name: TensorFlow 2.8
-    if: always()
-    needs: [run_past_ci_pytorch_1-4]
-    uses: ./.github/workflows/self-past.yml
-    with:
-      framework: tensorflow
-      version: "2.8"
-    secrets: inherit
-
-  run_past_ci_tensorflow_2-7:
-    name: TensorFlow 2.7
-    if: always()
-    needs: [run_past_ci_tensorflow_2-8]
-    uses: ./.github/workflows/self-past.yml
-    with:
-      framework: tensorflow
-      version: "2.7"
-    secrets: inherit
-
-  run_past_ci_tensorflow_2-6:
-    name: TensorFlow 2.6
-    if: always()
-    needs: [run_past_ci_tensorflow_2-7]
-    uses: ./.github/workflows/self-past.yml
-    with:
-      framework: tensorflow
-      version: "2.6"
-    secrets: inherit
-
-  run_past_ci_tensorflow_2-5:
-    name: TensorFlow 2.5
-    if: always()
-    needs: [run_past_ci_tensorflow_2-6]
-    uses: ./.github/workflows/self-past.yml
-    with:
-      framework: tensorflow
-      version: "2.5"
-    secrets: inherit
-
-  run_past_ci_tensorflow_2-4:
-    name: TensorFlow 2.4
-    if: always()
-    needs: [run_past_ci_tensorflow_2-5]
-    uses: ./.github/workflows/self-past.yml
-    with:
-      framework: tensorflow
-      version: "2.4"
-    secrets: inherit

.github/workflows/self-past.yml

@@ -1,4 +1,4 @@
-name: Self-hosted runner (past)
+name: Self-hosted runner (past-ci)
 
 # Note that each job's dependencies go into a corresponding docker file.
 #
@@ -157,7 +157,7 @@ jobs:
         if: ${{ always() }}
         uses: actions/upload-artifact@v3
         with:
-          name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports
+          name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports_postfix_${{ inputs.framework }}-${{ inputs.version }}
           path: /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}
 
   run_tests_multi_gpu:
@@ -223,14 +223,80 @@ jobs:
         if: ${{ always() }}
         uses: actions/upload-artifact@v3
         with:
-          name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports
+          name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports_postfix_${{ inputs.framework }}-${{ inputs.version }}
           path: /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}
 
+  run_all_tests_torch_cuda_extensions_gpu:
+    name: Torch CUDA extension tests
+    if: inputs.framework == 'pytorch'
+    strategy:
+      fail-fast: false
+      matrix:
+        machine_type: [single-gpu, multi-gpu]
+    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker-past-ci') }}
+    needs: setup
+    container:
+      image: huggingface/transformers-${{ inputs.framework }}-past-${{ inputs.version }}-gpu
+      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: Update clone
+        working-directory: /transformers
+        run: git fetch && git checkout ${{ github.sha }}
+
+      - name: Remove cached torch extensions
+        run: rm -rf /github/home/.cache/torch_extensions/
+
+      # To avoid unknown test failures
+      - name: Pre build DeepSpeed *again*
+        working-directory: /
+        run: |
+          python3 -m pip uninstall -y deepspeed
+          rm -rf DeepSpeed
+          git clone https://github.com/microsoft/DeepSpeed && cd DeepSpeed && rm -rf build
+          DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 DS_BUILD_AIO=1 DS_BUILD_UTILS=1 python3 -m pip install . --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check
+
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+      - name: Environment
+        working-directory: /transformers
+        run: |
+          python3 utils/print_env.py
+
+      - name: Show installed libraries and their versions
+        working-directory: /transformers
+        run: pip freeze
+
+      - name: Run all tests on GPU
+        working-directory: /transformers
+        run: |
+          python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu tests/deepspeed tests/extended
+
+      - name: Failure short reports
+        if: ${{ failure() }}
+        continue-on-error: true
+        run: cat /transformers/reports/${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu/failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v3
+        with:
+          name: ${{ matrix.machine_type }}_run_tests_torch_cuda_extensions_gpu_test_reports_postfix_${{ inputs.framework }}-${{ inputs.version }}
+          path: /transformers/reports/${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu
+
   send_results:
     name: Send results to webhook
     runs-on: ubuntu-latest
     if: always()
-    needs: [check_runner_status, check_runners, setup, run_tests_single_gpu, run_tests_multi_gpu]
+    needs: [
+      check_runner_status,
+      check_runners,
+      setup,
+      run_tests_single_gpu,
+      run_tests_multi_gpu,
+      run_all_tests_torch_cuda_extensions_gpu
+    ]
     steps:
       - name: Preliminary job status
         shell: bash
@@ -273,3 +339,10 @@ jobs:
         with:
           name: test_failure_tables_${{ inputs.framework }}-${{ inputs.version }}
           path: test_failure_tables
+
+      # delete-artifact
+      - uses: geekyeggo/delete-artifact@v2
+        with:
+          name: |
+              single-*
+              multi-*

.github/workflows/self-scheduled.yml

@@ -10,6 +10,9 @@ on:
   repository_dispatch:
   schedule:
     - cron: "0 2 * * *"
+  push:
+    branches:
+      - run_scheduled_ci*
 
 env:
   HF_HOME: /mnt/cache

docker/transformers-past-gpu/Dockerfile

@@ -1,4 +1,4 @@
-ARG BASE_DOCKER_IMAGE="nvidia/cuda:11.2.2-cudnn8-devel-ubuntu20.04"
+ARG BASE_DOCKER_IMAGE
 FROM $BASE_DOCKER_IMAGE
 LABEL maintainer="Hugging Face"
 
@@ -8,7 +8,7 @@ ARG DEBIAN_FRONTEND=noninteractive
 SHELL ["sh", "-lc"]
 
 RUN apt update
-RUN apt install -y git libsndfile1-dev tesseract-ocr espeak-ng python3 python3-pip ffmpeg git-lfs
+RUN apt install -y git libsndfile1-dev tesseract-ocr espeak-ng python3 python3-pip ffmpeg git-lfs libaio-dev
 RUN git lfs install
 RUN python3 -m pip install --no-cache-dir --upgrade pip
 
@@ -23,6 +23,9 @@ RUN cd transformers && python3 setup.py develop
 ARG FRAMEWORK
 ARG VERSION
 
+# Control `setuptools` version to avoid some issues
+RUN [ "$VERSION" != "1.9" -a "$VERSION" != "1.10" ] && python3 -m pip install -U setuptools || python3 -m pip install -U "setuptools<=59.5"
+
 # Remove all frameworks
 # (`accelerate` requires `torch`, and this causes import issues for TF-only testing)
 RUN python3 -m pip uninstall -y torch torchvision torchaudio accelerate tensorflow jax flax
@@ -34,4 +37,20 @@ RUN python3 ./transformers/utils/past_ci_versions.py --framework $FRAMEWORK --ve
 RUN echo "INSTALL_CMD = $INSTALL_CMD"
 RUN $INSTALL_CMD
 
+RUN [ "$FRAMEWORK" != "pytorch" ] && echo "`deepspeed-testing` installation is skipped" || python3 -m pip install --no-cache-dir ./transformers[deepspeed-testing]
+
+# Uninstall `torch-tensorrt` and `apex` shipped with the base image
+RUN python3 -m pip uninstall -y torch-tensorrt apex
+
+# Pre-build **nightly** release of DeepSpeed, so it would be ready for testing (otherwise, the 1st deepspeed test will timeout)
+RUN python3 -m pip uninstall -y deepspeed
+# This has to be run inside the GPU VMs running the tests. (So far, it fails here due to GPU checks during compilation.)
+# Issue: https://github.com/microsoft/DeepSpeed/issues/2010
+# RUN git clone https://github.com/microsoft/DeepSpeed && cd DeepSpeed && rm -rf build && \
+#    DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 DS_BUILD_AIO=1 DS_BUILD_UTILS=1 python3 -m pip install . --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check 2>&1
+
 RUN python3 -m pip install -U "itsdangerous<2.1.0"
+
+# When installing in editable mode, `transformers` is not recognized as a package.
+# this line must be added in order for python to be aware of transformers.
+RUN cd transformers && python3 setup.py develop

docker/transformers-pytorch-deepspeed-nightly-gpu/Dockerfile

@@ -1,10 +1,11 @@
-FROM nvcr.io/nvidia/pytorch:21.03-py3
+# https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/rel_22-08.html#rel_22-08
+FROM nvcr.io/nvidia/pytorch:22.08-py3
 LABEL maintainer="Hugging Face"
 
 ARG DEBIAN_FRONTEND=noninteractive
 
 # Example: `cu102`, `cu113`, etc.
-ARG CUDA='cu113'
+ARG CUDA='cu117'
 
 RUN apt -y update
 RUN apt install -y libaio-dev
@@ -20,6 +21,9 @@ RUN python3 -m pip install --no-cache-dir -U --pre torch torchvision torchaudio
 
 RUN python3 -m pip install --no-cache-dir ./transformers[deepspeed-testing]
 
+# Uninstall `torch-tensorrt` and `apex` shipped with the base image
+RUN python3 -m pip uninstall -y torch-tensorrt apex
+
 # Pre-build **nightly** release of DeepSpeed, so it would be ready for testing (otherwise, the 1st deepspeed test will timeout)
 RUN python3 -m pip uninstall -y deepspeed
 # This has to be run inside the GPU VMs running the tests. (So far, it fails here due to GPU checks during compilation.)
@@ -27,23 +31,23 @@ RUN python3 -m pip uninstall -y deepspeed
 # RUN git clone https://github.com/microsoft/DeepSpeed && cd DeepSpeed && rm -rf build && \
 #    DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 DS_BUILD_AIO=1 DS_BUILD_UTILS=1 python3 -m pip install . --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check 2>&1
 
-# For `torchdynamo` tests
+## For `torchdynamo` tests
-# (see https://github.com/huggingface/transformers/pull/17765)
+## (see https://github.com/huggingface/transformers/pull/17765)
-RUN git clone https://github.com/pytorch/functorch
+#RUN git clone https://github.com/pytorch/functorch
-RUN python3 -m pip install --no-cache-dir ./functorch[aot]
+#RUN python3 -m pip install --no-cache-dir ./functorch[aot]
-RUN cd functorch && python3 setup.py develop
+#RUN cd functorch && python3 setup.py develop
-
+#
-RUN git clone https://github.com/pytorch/torchdynamo
+#RUN git clone https://github.com/pytorch/torchdynamo
-RUN python3 -m pip install -r ./torchdynamo/requirements.txt
+#RUN python3 -m pip install -r ./torchdynamo/requirements.txt
-RUN cd torchdynamo && python3 setup.py develop
+#RUN cd torchdynamo && python3 setup.py develop
-
+#
-# install TensorRT
+## install TensorRT
-RUN python3 -m pip install --no-cache-dir -U nvidia-pyindex
+#RUN python3 -m pip install --no-cache-dir -U nvidia-pyindex
-RUN python3 -m pip install --no-cache-dir -U nvidia-tensorrt==8.2.4.2
+#RUN python3 -m pip install --no-cache-dir -U nvidia-tensorrt==8.2.4.2
-
+#
-# install torch_tensorrt (fx path)
+## install torch_tensorrt (fx path)
-RUN git clone https://github.com/pytorch/TensorRT.git
+#RUN git clone https://github.com/pytorch/TensorRT.git
-RUN cd TensorRT/py && python3 setup.py install --fx-only
+#RUN cd TensorRT/py && python3 setup.py install --fx-only
 
 # When installing in editable mode, `transformers` is not recognized as a package.
 # this line must be added in order for python to be aware of transformers.

utils/get_ci_error_statistics.py

@@ -66,12 +66,12 @@ def get_artifacts_links(worflow_run_id, token=None):
 def download_artifact(artifact_name, artifact_url, output_dir, token):
     """Download a GitHub Action artifact from a URL.
 
-    The URL is of the from `https://api.github.com/repos/huggingface/transformers/actions/artifacts/{ARTIFACT_ID}/zip`,
+    The URL is of the form `https://api.github.com/repos/huggingface/transformers/actions/artifacts/{ARTIFACT_ID}/zip`,
     but it can't be used to download directly. We need to get a redirect URL first.
     See https://docs.github.com/en/rest/actions/artifacts#download-an-artifact
     """
     # Get the redirect URL first
-    cmd = f'curl -v -H "Accept: application/vnd.github+json" -H "Authorization: token {token}" {artifact_url}'
+    cmd = f'curl -v -H "Accept: application/vnd.github+json" -H "Authorization: Bearer {token}" {artifact_url}'
     output = subprocess.run(cmd, shell=True, stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
     o = output.stdout.decode("utf-8")
     lines = o.splitlines()

utils/notification_service.py

@@ -590,23 +590,20 @@ class Message:
                     time.sleep(1)
 
 
-def retrieve_artifact(name: str, gpu: Optional[str]):
+def retrieve_artifact(artifact_path: str, gpu: Optional[str]):
     if gpu not in [None, "single", "multi"]:
         raise ValueError(f"Invalid GPU for artifact. Passed GPU: `{gpu}`.")
 
-    if gpu is not None:
-        name = f"{gpu}-gpu_{name}"
-
     _artifact = {}
 
-    if os.path.exists(name):
+    if os.path.exists(artifact_path):
-        files = os.listdir(name)
+        files = os.listdir(artifact_path)
         for file in files:
             try:
-                with open(os.path.join(name, file)) as f:
+                with open(os.path.join(artifact_path, file)) as f:
                     _artifact[file.split(".")[0]] = f.read()
             except UnicodeDecodeError as e:
-                raise ValueError(f"Could not open {os.path.join(name, file)}.") from e
+                raise ValueError(f"Could not open {os.path.join(artifact_path, file)}.") from e
 
     return _artifact
 
@@ -629,8 +626,14 @@ def retrieve_available_artifacts():
 
     directories = filter(os.path.isdir, os.listdir())
     for directory in directories:
-        if directory.startswith("single-gpu"):
+        artifact_name = directory
-            artifact_name = directory[len("single-gpu") + 1 :]
+
+        name_parts = artifact_name.split("_postfix_")
+        if len(name_parts) > 1:
+            artifact_name = name_parts[0]
+
+        if artifact_name.startswith("single-gpu"):
+            artifact_name = artifact_name[len("single-gpu") + 1 :]
 
             if artifact_name in _available_artifacts:
                 _available_artifacts[artifact_name].single_gpu = True
@@ -639,7 +642,7 @@ def retrieve_available_artifacts():
 
             _available_artifacts[artifact_name].add_path(directory, gpu="single")
 
-        elif directory.startswith("multi-gpu"):
+        elif artifact_name.startswith("multi-gpu"):
             artifact_name = directory[len("multi-gpu") + 1 :]
 
             if artifact_name in _available_artifacts:
@@ -649,7 +652,6 @@ def retrieve_available_artifacts():
 
             _available_artifacts[artifact_name].add_path(directory, gpu="multi")
         else:
-            artifact_name = directory
             if artifact_name not in _available_artifacts:
                 _available_artifacts[artifact_name] = Artifact(artifact_name)
 
@@ -805,10 +807,12 @@ if __name__ == "__main__":
         framework, version = ci_event.replace("Past CI - ", "").split("-")
         framework = "PyTorch" if framework == "pytorch" else "TensorFlow"
         job_name_prefix = f"{framework} {version}"
+    elif ci_event.startswith("Nightly CI"):
+        job_name_prefix = "Nightly CI"
 
     for model in model_results.keys():
         for artifact_path in available_artifacts[f"run_all_tests_gpu_{model}_test_reports"].paths:
-            artifact = retrieve_artifact(artifact_path["name"], artifact_path["gpu"])
+            artifact = retrieve_artifact(artifact_path["path"], artifact_path["gpu"])
             if "stats" in artifact:
                 # Link to the GitHub Action job
                 # The job names use `matrix.folder` which contain things like `models/bert` instead of `models_bert`
@@ -901,7 +905,7 @@ if __name__ == "__main__":
             else:
                 additional_results[key]["job_link"][artifact_path["gpu"]] = github_actions_job_links.get(key)
 
-            artifact = retrieve_artifact(artifact_path["name"], artifact_path["gpu"])
+            artifact = retrieve_artifact(artifact_path["path"], artifact_path["gpu"])
             stacktraces = handle_stacktraces(artifact["failures_line"])
 
             failed, success, time_spent = handle_test_results(artifact["stats"])

utils/past_ci_versions.py

@@ -4,6 +4,18 @@ import os
 
 past_versions_testing = {
     "pytorch": {
+        "1.13": {
+            "torch": "1.13.1",
+            "torchvision": "0.14.1",
+            "torchaudio": "0.13.1",
+            "python": 3.9,
+            "cuda": "cu116",
+            "install": (
+                "python3 -m pip install --no-cache-dir -U torch==1.13.1 torchvision==0.14.1 torchaudio==0.13.1"
+                " --extra-index-url https://download.pytorch.org/whl/cu116"
+            ),
+            "base_image": "nvidia/cuda:11.6.2-cudnn8-devel-ubuntu20.04",
+        },
         "1.12": {
             "torch": "1.12.1",
             "torchvision": "0.13.1",
@@ -14,6 +26,7 @@ past_versions_testing = {
                 "python3 -m pip install --no-cache-dir -U torch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1"
                 " --extra-index-url https://download.pytorch.org/whl/cu113"
             ),
+            "base_image": "nvidia/cuda:11.2.2-cudnn8-devel-ubuntu20.04",
         },
         "1.11": {
             "torch": "1.11.0",
@@ -25,6 +38,7 @@ past_versions_testing = {
                 "python3 -m pip install --no-cache-dir -U torch==1.11.0 torchvision==0.12.0 torchaudio==0.11.0"
                 " --extra-index-url https://download.pytorch.org/whl/cu113"
             ),
+            "base_image": "nvidia/cuda:11.2.2-cudnn8-devel-ubuntu20.04",
         },
         "1.10": {
             "torch": "1.10.2",
@@ -36,6 +50,7 @@ past_versions_testing = {
                 "python3 -m pip install --no-cache-dir -U torch==1.10.2 torchvision==0.11.3 torchaudio==0.10.2"
                 " --extra-index-url https://download.pytorch.org/whl/cu113"
             ),
+            "base_image": "nvidia/cuda:11.2.2-cudnn8-devel-ubuntu20.04",
         },
         # torchaudio < 0.10 has no CUDA-enabled binary distributions
         "1.9": {
@@ -48,87 +63,44 @@ past_versions_testing = {
                 "python3 -m pip install --no-cache-dir -U torch==1.9.1 torchvision==0.10.1 torchaudio==0.9.1"
                 " --extra-index-url https://download.pytorch.org/whl/cu111"
             ),
-        },
+            "base_image": "nvidia/cuda:11.2.2-cudnn8-devel-ubuntu20.04",
-        "1.8": {
-            "torch": "1.8.1",
-            "torchvision": "0.9.1",
-            "torchaudio": "0.8.1",
-            "python": 3.9,
-            "cuda": "cu111",
-            "install": (
-                "python3 -m pip install --no-cache-dir -U torch==1.8.1 torchvision==0.9.1 torchaudio==0.8.1"
-                " --extra-index-url https://download.pytorch.org/whl/cu111"
-            ),
-        },
-        "1.7": {
-            "torch": "1.7.1",
-            "torchvision": "0.8.2",
-            "torchaudio": "0.7.2",
-            "python": 3.9,
-            "cuda": "cu110",
-            "install": (
-                "python3 -m pip install --no-cache-dir -U torch==1.7.1 torchvision==0.8.2 torchaudio==0.7.2"
-                " --extra-index-url https://download.pytorch.org/whl/cu110"
-            ),
-        },
-        "1.6": {
-            "torch": "1.6.0",
-            "torchvision": "0.7.0",
-            "torchaudio": "0.6.0",
-            "python": 3.8,
-            "cuda": "cu101",
-            "install": (
-                "python3 -m pip install --no-cache-dir -U torch==1.6.0 torchvision==0.7.0 torchaudio==0.6.0"
-                " --extra-index-url https://download.pytorch.org/whl/cu101"
-            ),
-        },
-        "1.5": {
-            "torch": "1.5.1",
-            "torchvision": "0.6.1",
-            "torchaudio": "0.5.1",
-            "python": 3.8,
-            "cuda": "cu101",
-            "install": (
-                "python3 -m pip install --no-cache-dir -U torch==1.5.1 torchvision==0.6.1 torchaudio==0.5.1"
-                " --extra-index-url https://download.pytorch.org/whl/cu101"
-            ),
-        },
-        "1.4": {
-            "torch": "1.4.0",
-            "torchvision": "0.5.0",
-            "torchaudio": "0.4.0",
-            "python": 3.8,
-            "cuda": "cu100",
-            "install": (
-                "python3 -m pip install --no-cache-dir -U torch==1.4.0 torchvision==0.5.0 torchaudio==0.4.0"
-                " --extra-index-url https://download.pytorch.org/whl/cu100"
-            ),
         },
     },
     "tensorflow": {
+        "2.11": {
+            "tensorflow": "2.11.1",
+            "install": "python3 -m pip install --no-cache-dir -U tensorflow==2.11.1",
+            "base_image": "nvidia/cuda:11.2.2-cudnn8-devel-ubuntu20.04",
+        },
+        "2.10": {
+            "tensorflow": "2.10.1",
+            "install": "python3 -m pip install --no-cache-dir -U tensorflow==2.10.1",
+            "base_image": "nvidia/cuda:11.2.2-cudnn8-devel-ubuntu20.04",
+        },
+        "2.9": {
+            "tensorflow": "2.9.3",
+            "install": "python3 -m pip install --no-cache-dir -U tensorflow==2.9.3",
+            "base_image": "nvidia/cuda:11.2.2-cudnn8-devel-ubuntu20.04",
+        },
         "2.8": {
             "tensorflow": "2.8.2",
             "install": "python3 -m pip install --no-cache-dir -U tensorflow==2.8.2",
+            "base_image": "nvidia/cuda:11.2.2-cudnn8-devel-ubuntu20.04",
         },
         "2.7": {
             "tensorflow": "2.7.3",
             "install": "python3 -m pip install --no-cache-dir -U tensorflow==2.7.3",
+            "base_image": "nvidia/cuda:11.2.2-cudnn8-devel-ubuntu20.04",
         },
         "2.6": {
             "tensorflow": "2.6.5",
             "install": "python3 -m pip install --no-cache-dir -U tensorflow==2.6.5",
+            "base_image": "nvidia/cuda:11.2.2-cudnn8-devel-ubuntu20.04",
         },
         "2.5": {
             "tensorflow": "2.5.3",
             "install": "python3 -m pip install --no-cache-dir -U tensorflow==2.5.3",
-        },
+            "base_image": "nvidia/cuda:11.2.2-cudnn8-devel-ubuntu20.04",
-        # need another `nvidia:cuda` docker image, otherwise GPU not working
-        "2.4": {
-            "tensorflow": "2.4.4",
-            "install": "python3 -m pip install --no-cache-dir -U tensorflow==2.4.4",
-            # This should be specified as a docker build argument.
-            # We keep the information here for reference only.
-            "base_docker": "nvidia/cuda:11.0.3-cudnn8-devel-ubuntu20.04",
         },
     },
 }