Revive Nightly/Past CI (#31159)

* build

* build

* build

* build

---------

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

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

@@ -15,16 +15,6 @@ jobs:
     name: "Nightly PyTorch + Stable TensorFlow"
     runs-on: [intel-cpu, 8-cpu, ci]
     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
@@ -52,16 +42,6 @@ jobs:
     name: "Nightly PyTorch + DeepSpeed"
     runs-on: [intel-cpu, 8-cpu, ci]
     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

.github/workflows/model_jobs.yml

@@ -12,6 +12,12 @@ on:
       slice_id:
         required: true
         type: number
+      runner:
+        required: true
+        type: string
+      docker:
+        required: true
+        type: string
 
 env:
   HF_HOME: /mnt/cache
@@ -31,12 +37,13 @@ jobs:
   run_models_gpu:
     name: " "
     strategy:
+      max-parallel: 8
       fail-fast: false
       matrix:
         folders: ${{ fromJson(inputs.folder_slices)[inputs.slice_id] }}
-    runs-on: ['${{ inputs.machine_type }}', nvidia-gpu, t4, daily-ci]
+    runs-on: ['${{ inputs.machine_type }}', nvidia-gpu, t4, '${{ inputs.runner }}']
     container:
-      image: huggingface/transformers-all-latest-gpu
+      image: ${{ inputs.docker }}
       options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
     steps:
       - name: Echo input and matrix info
@@ -65,6 +72,18 @@ jobs:
         working-directory: /transformers
         run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
 
+      - name: Update / Install some packages (for Past CI)
+        if: ${{ contains(inputs.docker, '-past-') }}
+        working-directory: /transformers
+        run: |
+          python3 -m pip install -U datasets
+
+      - name: Update / Install some packages (for Past CI)
+        if: ${{ contains(inputs.docker, '-past-') && contains(inputs.docker, '-pytorch-') }}
+        working-directory: /transformers
+        run: |
+          python3 -m pip install --no-cache-dir git+https://github.com/huggingface/accelerate@main#egg=accelerate
+
       - name: NVIDIA-SMI
         run: |
           nvidia-smi

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

@@ -0,0 +1,43 @@
+name: Self-hosted runner (nightly-ci)
+
+
+on:
+  repository_dispatch:
+  schedule:
+    - cron: "17 2 * * *"
+  push:
+    branches:
+      - run_nightly_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
+
+  model-ci:
+    name: Model CI
+    needs: [build_nightly_ci_images]
+    uses: ./.github/workflows/self-scheduled.yml
+    with:
+      job: run_models_gpu
+      slack_report_channel: "#transformers-ci-past-future"
+      runner: ci
+      docker: huggingface/transformers-all-latest-torch-nightly-gpu
+      ci_event: Nightly CI
+    secrets: inherit
+
+  deepspeed-ci:
+    name: DeepSpeed CI
+    needs: [build_nightly_ci_images]
+    uses: ./.github/workflows/self-scheduled.yml
+    with:
+      job: run_torch_cuda_extensions_gpu
+      slack_report_channel: "#transformers-ci-past-future"
+      runner: ci
+      # test deepspeed nightly build with the latest release torch
+      docker: huggingface/transformers-pytorch-deepspeed-latest-gpu
+      ci_event: Nightly CI
+      working-directory-prefix: /workspace
+    secrets: inherit

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

@@ -2,32 +2,30 @@ name: Self-hosted runner (nightly-past-ci-caller)
 
 on:
   schedule:
-    # 2:17 am on each Sunday and Thursday
+    - cron: "17 2,14 * * *"
-
-    - cron: "17 2 * * 0,4"
   push:
     branches:
-      - run_nightly_ci*
       - run_past_ci*
 
 jobs:
-  build_nightly_ci_images:
+  get_number:
-    name: Build Nightly CI Docker Images
+    name: Get number
-    if: (github.event_name == 'schedule') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_nightly_ci'))
+    runs-on: ubuntu-22.04
-    uses: ./.github/workflows/build-nightly-ci-docker-images.yml
+    outputs:
-    secrets: inherit
+      run_number: ${{ steps.get_number.outputs.run_number }}
-
+    steps:
-  run_nightly_ci:
+      - name: Get number
-    name: Nightly CI
+        id: get_number
-    needs: [build_nightly_ci_images]
+        run: |
-    uses: ./.github/workflows/self-nightly-scheduled.yml
+          echo "${{ github.run_number }}"
-    secrets: inherit
+          echo "$(python3 -c 'print(int(${{ github.run_number }}) % 10)')"
+          echo "run_number=$(python3 -c 'print(int(${{ github.run_number }}) % 10)')" >> $GITHUB_OUTPUT
 
   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: get_number
-    needs: [run_nightly_ci]
+    if: needs.get_number.outputs.run_number == 0 && (cancelled() != true) && ((github.event_name == 'schedule') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci')))
-    uses: ./.github/workflows/self-past.yml
+    uses: ./.github/workflows/self-past-caller.yml
     with:
       framework: pytorch
       version: "1.13"
@@ -36,9 +34,9 @@ jobs:
 
   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: get_number
-    needs: [run_past_ci_pytorch_1-13]
+    if: needs.get_number.outputs.run_number == 1 && (cancelled() != true) && ((github.event_name == 'schedule') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci')))
-    uses: ./.github/workflows/self-past.yml
+    uses: ./.github/workflows/self-past-caller.yml
     with:
       framework: pytorch
       version: "1.12"
@@ -47,9 +45,9 @@ jobs:
 
   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: get_number
-    needs: [run_past_ci_pytorch_1-12]
+    if: needs.get_number.outputs.run_number == 2 && (cancelled() != true) && ((github.event_name == 'schedule') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci')))
-    uses: ./.github/workflows/self-past.yml
+    uses: ./.github/workflows/self-past-caller.yml
     with:
       framework: pytorch
       version: "1.11"
@@ -58,9 +56,9 @@ jobs:
 
   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: get_number
-    needs: [run_past_ci_pytorch_1-11]
+    if: needs.get_number.outputs.run_number == 3 && (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
-    uses: ./.github/workflows/self-past.yml
+    uses: ./.github/workflows/self-past-caller.yml
     with:
       framework: tensorflow
       version: "2.11"
@@ -69,9 +67,9 @@ jobs:
 
   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: get_number
-    needs: [run_past_ci_tensorflow_2-11]
+    if: needs.get_number.outputs.run_number == 4 && (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
-    uses: ./.github/workflows/self-past.yml
+    uses: ./.github/workflows/self-past-caller.yml
     with:
       framework: tensorflow
       version: "2.10"
@@ -80,9 +78,9 @@ jobs:
 
   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: get_number
-    needs: [run_past_ci_tensorflow_2-10]
+    if: needs.get_number.outputs.run_number == 5 && (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
-    uses: ./.github/workflows/self-past.yml
+    uses: ./.github/workflows/self-past-caller.yml
     with:
       framework: tensorflow
       version: "2.9"
@@ -91,9 +89,9 @@ jobs:
 
   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: get_number
-    needs: [run_past_ci_tensorflow_2-9]
+    if: needs.get_number.outputs.run_number == 6 && (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
-    uses: ./.github/workflows/self-past.yml
+    uses: ./.github/workflows/self-past-caller.yml
     with:
       framework: tensorflow
       version: "2.8"
@@ -102,9 +100,9 @@ jobs:
 
   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: get_number
-    needs: [run_past_ci_tensorflow_2-8]
+    if: needs.get_number.outputs.run_number == 7 && (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
-    uses: ./.github/workflows/self-past.yml
+    uses: ./.github/workflows/self-past-caller.yml
     with:
       framework: tensorflow
       version: "2.7"
@@ -113,9 +111,9 @@ jobs:
 
   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: get_number
-    needs: [run_past_ci_tensorflow_2-7]
+    if: needs.get_number.outputs.run_number == 8 && (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
-    uses: ./.github/workflows/self-past.yml
+    uses: ./.github/workflows/self-past-caller.yml
     with:
       framework: tensorflow
       version: "2.6"
@@ -124,9 +122,9 @@ jobs:
 
   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: get_number
-    needs: [run_past_ci_tensorflow_2-6]
+    if: needs.get_number.outputs.run_number == 9 &&  (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
-    uses: ./.github/workflows/self-past.yml
+    uses: ./.github/workflows/self-past-caller.yml
     with:
       framework: tensorflow
       version: "2.5"

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

@@ -1,290 +0,0 @@
-name: Self-hosted runner (nightly-ci)
-
-# Note that each job's dependencies go into a corresponding docker file.
-#
-# For example for `run_torch_cuda_extensions_gpu` the docker image is
-# `huggingface/transformers-pytorch-deepspeed-latest-gpu`, which can be found at
-# `docker/transformers-pytorch-deepspeed-latest-gpu/Dockerfile`
-
-on:
-  repository_dispatch:
-  workflow_call:
-
-env:
-  HF_HOME: /mnt/cache
-  TRANSFORMERS_IS_CI: yes
-  OMP_NUM_THREADS: 8
-  MKL_NUM_THREADS: 8
-  RUN_SLOW: yes
-  HF_HUB_READ_TOKEN: ${{ secrets.HF_HUB_READ_TOKEN }}
-  SIGOPT_API_TOKEN: ${{ secrets.SIGOPT_API_TOKEN }}
-  TF_FORCE_GPU_ALLOW_GROWTH: true
-  RUN_PT_TF_CROSS_TESTS: 1
-  CUDA_VISIBLE_DEVICES: 0,1
-
-jobs:
-  setup:
-    name: Setup
-    strategy:
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, 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/
-    outputs:
-      matrix: ${{ steps.set-matrix.outputs.matrix }}
-    steps:
-      - name: Update clone
-        working-directory: /transformers
-        run: |
-          git fetch && git checkout ${{ github.sha }}
-
-      - name: Cleanup
-        working-directory: /transformers
-        run: |
-          rm -rf tests/__pycache__
-          rm -rf tests/models/__pycache__
-          rm -rf reports
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - id: set-matrix
-        name: Identify models to test
-        working-directory: /transformers/tests
-        run: |
-          echo "matrix=$(python3 -c 'import os; tests = os.getcwd(); model_tests = os.listdir(os.path.join(tests, "models")); d1 = sorted(list(filter(os.path.isdir, os.listdir(tests)))); d2 = sorted(list(filter(os.path.isdir, [f"models/{x}" for x in model_tests]))); d1.remove("models"); d = d2 + d1; print(d)')" >> $GITHUB_OUTPUT
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-  run_tests_single_gpu:
-    name: Model tests
-    strategy:
-      fail-fast: false
-      matrix:
-        folders: ${{ fromJson(needs.setup.outputs.matrix) }}
-        machine_type: [single-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, 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/
-    needs: setup
-    steps:
-      - name: Echo folder ${{ matrix.folders }}
-        shell: bash
-        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
-        # set the artifact folder names (because the character `/` is not allowed).
-        run: |
-          echo "${{ matrix.folders }}"
-          matrix_folders=${{ matrix.folders }}
-          matrix_folders=${matrix_folders/'models/'/'models_'}
-          echo "$matrix_folders"
-          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
-
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ github.sha }}
-
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - name: Run all tests on GPU
-        working-directory: /transformers
-        run: python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }} tests/${{ matrix.folders }}
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/failures_short.txt
-
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports_postfix_nightly"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          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:
-    name: Model tests
-    strategy:
-      fail-fast: false
-      matrix:
-        folders: ${{ fromJson(needs.setup.outputs.matrix) }}
-        machine_type: [multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, 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/
-    needs: setup
-    steps:
-      - name: Echo folder ${{ matrix.folders }}
-        shell: bash
-        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
-        # set the artifact folder names (because the character `/` is not allowed).
-        run: |
-          echo "${{ matrix.folders }}"
-          matrix_folders=${{ matrix.folders }}
-          matrix_folders=${matrix_folders/'models/'/'models_'}
-          echo "$matrix_folders"
-          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
-
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ github.sha }}
-
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - name: Run all tests on GPU
-        working-directory: /transformers
-        run: python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }} tests/${{ matrix.folders }}
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/failures_short.txt
-
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports_postfix_nightly"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          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_torch_cuda_extensions_gpu:
-    name: Torch CUDA extension tests
-    strategy:
-      fail-fast: false
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, past-ci]
-    needs: setup
-    container:
-      image: huggingface/transformers-pytorch-deepspeed-nightly-gpu
-      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    steps:
-      - name: Update clone
-        working-directory: /workspace/transformers
-        run: git fetch && git checkout ${{ github.sha }}
-
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /workspace/transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
-      - name: Remove cached torch extensions
-        run: rm -rf /github/home/.cache/torch_extensions/
-
-      # To avoid unknown test failures
-      - name: Pre build DeepSpeed *again*
-        working-directory: /workspace
-        run: |
-          python3 -m pip uninstall -y deepspeed
-          rm -rf DeepSpeed
-          git clone https://github.com/microsoft/DeepSpeed && cd DeepSpeed && rm -rf build
-          DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 python3 -m pip install . --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Environment
-        working-directory: /workspace/transformers
-        run: |
-          python utils/print_env.py
-
-      - name: Show installed libraries and their versions
-        working-directory: /workspace/transformers
-        run: pip freeze
-
-      - name: Run all tests on GPU
-        working-directory: /workspace/transformers
-        run: |
-          python -m pytest -v --make-reports=${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports tests/deepspeed tests/extended
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /workspace/transformers/reports/${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports/failures_short.txt
-
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports_postfix_nightly"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports_postfix_nightly
-          path: /workspace/transformers/reports/${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports
-
-  send_results:
-    name: Send results to webhook
-    runs-on: ubuntu-22.04
-    if: always()
-    needs: [
-      setup,
-      run_tests_single_gpu,
-      run_tests_multi_gpu,
-      run_torch_cuda_extensions_gpu
-    ]
-    steps:
-      - name: Preliminary job status
-        shell: bash
-        # For the meaning of these environment variables, see the job `Setup`
-        run: |
-          echo "Setup status: ${{ needs.setup.result }}"
-
-      - uses: actions/checkout@v4
-      - uses: actions/download-artifact@v4
-      - name: Send message to Slack
-        env:
-          CI_SLACK_BOT_TOKEN: ${{ secrets.CI_SLACK_BOT_TOKEN }}
-          CI_SLACK_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID }}
-          CI_SLACK_CHANNEL_ID_DAILY: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY }}
-          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 CI
-          SETUP_STATUS: ${{ needs.setup.result }}
-        # We pass `needs.setup.outputs.matrix` as the argument. A processing in `notification_service.py` to change
-        # `models/bert` to `models_bert` is required, as the artifact names use `_` instead of `/`.
-        run: |
-          pip install slack_sdk
-          pip show slack_sdk
-          python utils/notification_service.py "${{ needs.setup.outputs.matrix }}"
-
-
-      # delete-artifact
-      - uses: geekyeggo/delete-artifact@v2
-        with:
-          name: |
-              single-*
-              multi-*

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

@@ -0,0 +1,40 @@
+name: Self-hosted runner (past-ci)
+
+
+on:
+  workflow_call:
+    inputs:
+      framework:
+        required: true
+        type: string
+      version:
+        required: true
+        type: string
+      # Use this to control the commit to test against
+      sha:
+        default: 'main'
+        required: false
+        type: string
+
+jobs:
+  model-ci:
+    name: Model CI
+    uses: ./.github/workflows/self-scheduled.yml
+    with:
+      job: run_models_gpu
+      slack_report_channel: "#transformers-ci-past-future"
+      runner: past-ci
+      docker: huggingface/transformers-${{ inputs.framework }}-past-${{ inputs.version }}-gpu
+      ci_event: Past CI - ${{ inputs.framework }}-${{ inputs.version }}
+    secrets: inherit
+
+  deepspeed-ci:
+    name: DeepSpeed CI
+    uses: ./.github/workflows/self-scheduled.yml
+    with:
+      job: run_torch_cuda_extensions_gpu
+      slack_report_channel: "#transformers-ci-past-future"
+      runner: past-ci
+      docker: huggingface/transformers-${{ inputs.framework }}-past-${{ inputs.version }}-gpu
+      ci_event: Past CI - ${{ inputs.framework }}-${{ inputs.version }}
+    secrets: inherit

.github/workflows/self-past.yml

@@ -1,357 +0,0 @@
-name: Self-hosted runner (past-ci)
-
-# Note that each job's dependencies go into a corresponding docker file.
-#
-# For example for `run_torch_cuda_extensions_gpu` the docker image is
-# `huggingface/transformers-pytorch-deepspeed-latest-gpu`, which can be found at
-# `docker/transformers-pytorch-deepspeed-latest-gpu/Dockerfile`
-
-on:
-  workflow_call:
-    inputs:
-      framework:
-        required: true
-        type: string
-      version:
-        required: true
-        type: string
-      # Use this to control the commit to test against
-      sha:
-        default: 'main'
-        required: false
-        type: string
-
-env:
-  HF_HOME: /mnt/cache
-  TRANSFORMERS_IS_CI: yes
-  OMP_NUM_THREADS: 8
-  MKL_NUM_THREADS: 8
-  RUN_SLOW: yes
-  HF_HUB_READ_TOKEN: ${{ secrets.HF_HUB_READ_TOKEN }}
-  SIGOPT_API_TOKEN: ${{ secrets.SIGOPT_API_TOKEN }}
-  TF_FORCE_GPU_ALLOW_GROWTH: true
-  RUN_PT_TF_CROSS_TESTS: 1
-  CUDA_VISIBLE_DEVICES: 0,1
-
-jobs:
-  setup:
-    name: Setup
-    strategy:
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, past-ci]
-    container:
-      image: huggingface/transformers-${{ inputs.framework }}-past-${{ inputs.version }}-gpu
-      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    outputs:
-      matrix: ${{ steps.set-matrix.outputs.matrix }}
-    steps:
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ inputs.sha }}
-
-      - name: Cleanup
-        working-directory: /transformers
-        run: |
-          rm -rf tests/__pycache__
-          rm -rf tests/models/__pycache__
-          rm -rf reports
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - id: set-matrix
-        working-directory: /transformers
-        name: Identify models to test
-        run: |
-          cd tests
-          echo "matrix=$(python3 -c 'import os; tests = os.getcwd(); model_tests = os.listdir(os.path.join(tests, "models")); d1 = sorted(list(filter(os.path.isdir, os.listdir(tests)))); d2 = sorted(list(filter(os.path.isdir, [f"models/{x}" for x in model_tests]))); d1.remove("models"); d = d2 + d1; print(d)')" >> $GITHUB_OUTPUT
-
-  run_tests_single_gpu:
-    name: Model tests
-    strategy:
-      fail-fast: false
-      matrix:
-        folders: ${{ fromJson(needs.setup.outputs.matrix) }}
-        machine_type: [single-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, past-ci]
-    container:
-      image: huggingface/transformers-${{ inputs.framework }}-past-${{ inputs.version }}-gpu
-      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    needs: setup
-    steps:
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ inputs.sha }}
-
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
-      - name: Update some packages
-        working-directory: /transformers
-        run: python3 -m pip install -U datasets
-
-      - name: Echo folder ${{ matrix.folders }}
-        shell: bash
-        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
-        # set the artifact folder names (because the character `/` is not allowed).
-        run: |
-          echo "${{ matrix.folders }}"
-          matrix_folders=${{ matrix.folders }}
-          matrix_folders=${matrix_folders/'models/'/'models_'}
-          echo "$matrix_folders"
-          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Install
-        if: inputs.framework == 'pytorch'
-        working-directory: /transformers
-        run: |
-          python3 -m pip install --no-cache-dir git+https://github.com/huggingface/accelerate@main#egg=accelerate
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - name: Run all tests on GPU
-        working-directory: /transformers
-        run: python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }} tests/${{ matrix.folders }}
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/failures_short.txt
-
-      - name: Save job name
-        if: ${{ always() }}
-        shell: bash
-        run: |
-          matrix_folders=${matrix_folders/'models_'/'models/'}
-          job_name="Model tests ($matrix_folders, ${{ matrix.machine_type }})"
-          echo "$job_name"
-          echo "$job_name" > /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/job_name.txt
-
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports_postfix_${{ inputs.framework }}-${{ inputs.version }}"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          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:
-    name: Model tests
-    strategy:
-      fail-fast: false
-      matrix:
-        folders: ${{ fromJson(needs.setup.outputs.matrix) }}
-        machine_type: [multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, past-ci]
-    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/
-    needs: setup
-    steps:
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ inputs.sha }}
-
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
-      - name: Update some packages
-        working-directory: /transformers
-        run: python3 -m pip install -U datasets
-
-      - name: Echo folder ${{ matrix.folders }}
-        shell: bash
-        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
-        # set the artifact folder names (because the character `/` is not allowed).
-        run: |
-          echo "${{ matrix.folders }}"
-          matrix_folders=${{ matrix.folders }}
-          matrix_folders=${matrix_folders/'models/'/'models_'}
-          echo "$matrix_folders"
-          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Install
-        if: inputs.framework == 'pytorch'
-        working-directory: /transformers
-        run: |
-          python3 -m pip install --no-cache-dir git+https://github.com/huggingface/accelerate@main#egg=accelerate
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - name: Run all tests on GPU
-        working-directory: /transformers
-        run: python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }} tests/${{ matrix.folders }}
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/failures_short.txt
-
-      - name: Save job name
-        if: ${{ always() }}
-        shell: bash
-        run: |
-          matrix_folders=${matrix_folders/'models_'/'models/'}
-          job_name="Model tests ($matrix_folders, ${{ matrix.machine_type }})"
-          echo "$job_name"
-          echo "$job_name" > /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/job_name.txt
-
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports_postfix_${{ inputs.framework }}-${{ inputs.version }}"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          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_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: ['${{ matrix.machine_type }}', nvidia-gpu, t4, past-ci]
-    needs: setup
-    container:
-      image: huggingface/transformers-${{ 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 ${{ github.sha }}
-
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
-      - name: Update some packages
-        working-directory: /transformers
-        run: python3 -m pip install -U datasets
-
-      - name: Install
-        working-directory: /transformers
-        run: |
-          python3 -m pip install --no-cache-dir git+https://github.com/huggingface/accelerate@main#egg=accelerate
-
-      - name: Remove cached torch extensions
-        run: rm -rf /github/home/.cache/torch_extensions/
-
-      # To avoid unknown test failures
-      - name: Pre build DeepSpeed *again*
-        working-directory: /
-        run: |
-          python3 -m pip uninstall -y deepspeed
-          rm -rf DeepSpeed
-          git clone https://github.com/microsoft/DeepSpeed && cd DeepSpeed && rm -rf build
-          DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 python3 -m pip install . --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - name: Run all tests on GPU
-        working-directory: /transformers
-        run: |
-          python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports tests/deepspeed tests/extended
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports/failures_short.txt
-
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports_postfix_${{ inputs.framework }}-${{ inputs.version }}"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports_postfix_${{ inputs.framework }}-${{ inputs.version }}
-          path: /transformers/reports/${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports
-
-  send_results:
-    name: Send results to webhook
-    runs-on: ubuntu-22.04
-    if: always()
-    needs: [
-      setup,
-      run_tests_single_gpu,
-      run_tests_multi_gpu,
-      run_torch_cuda_extensions_gpu
-    ]
-    steps:
-      - name: Preliminary job status
-        shell: bash
-        # For the meaning of these environment variables, see the job `Setup`
-        run: |
-          echo "Setup status: ${{ needs.setup.result }}"
-
-      - uses: actions/checkout@v4
-      - uses: actions/download-artifact@v4
-
-      # Create a directory to store test failure tables in the next step
-      - name: Create directory
-        run: mkdir test_failure_tables
-
-      - name: Send message to Slack
-        env:
-          CI_SLACK_BOT_TOKEN: ${{ secrets.CI_SLACK_BOT_TOKEN }}
-          CI_SLACK_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID }}
-          CI_SLACK_CHANNEL_ID_DAILY: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY }}
-          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: Past CI - ${{ inputs.framework }}-${{ inputs.version }}
-          SETUP_STATUS: ${{ needs.setup.result }}
-        # We pass `needs.setup.outputs.matrix` as the argument. A processing in `notification_service.py` to change
-        # `models/bert` to `models_bert` is required, as the artifact names use `_` instead of `/`.
-        run: |
-          pip install slack_sdk
-          pip show slack_sdk
-          python utils/notification_service.py "${{ needs.setup.outputs.matrix }}"
-
-      # Upload complete failure tables, as they might be big and only truncated versions could be sent to Slack.
-      - name: Failure table artifacts
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: test_failure_tables_${{ inputs.framework }}-${{ inputs.version }}
-          path: test_failure_tables
-
-      # delete-artifact
-      - uses: geekyeggo/delete-artifact@v2
-        with:
-          name: |
-              single-*
-              multi-*

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

@@ -16,6 +16,9 @@ jobs:
     with:
       job: run_models_gpu
       slack_report_channel: "#transformers-ci-daily-models"
+      runner: daily-ci
+      docker: huggingface/transformers-all-latest-gpu
+      ci_event: Daily CI
     secrets: inherit
 
   torch-pipeline:
@@ -24,6 +27,9 @@ jobs:
     with:
       job: run_pipelines_torch_gpu
       slack_report_channel: "#transformers-ci-daily-pipeline-torch"
+      runner: daily-ci
+      docker: huggingface/transformers-pytorch-gpu
+      ci_event: Daily CI
     secrets: inherit
 
   tf-pipeline:
@@ -32,6 +38,9 @@ jobs:
     with:
       job: run_pipelines_tf_gpu
       slack_report_channel: "#transformers-ci-daily-pipeline-tf"
+      runner: daily-ci
+      docker: huggingface/transformers-tensorflow-gpu
+      ci_event: Daily CI
     secrets: inherit
 
   example-ci:
@@ -40,6 +49,9 @@ jobs:
     with:
       job: run_examples_gpu
       slack_report_channel: "#transformers-ci-daily-examples"
+      runner: daily-ci
+      docker: huggingface/transformers-all-latest-gpu
+      ci_event: Daily CI
     secrets: inherit
 
   deepspeed-ci:
@@ -48,6 +60,10 @@ jobs:
     with:
       job: run_torch_cuda_extensions_gpu
       slack_report_channel: "#transformers-ci-daily-deepspeed"
+      runner: daily-ci
+      docker: huggingface/transformers-pytorch-deepspeed-latest-gpu
+      ci_event: Daily CI
+      working-directory-prefix: /workspace
     secrets: inherit
 
   quantization-ci:
@@ -56,4 +72,7 @@ jobs:
     with:
       job: run_quantization_torch_gpu
       slack_report_channel: "#transformers-ci-daily-quantization"
+      runner: daily-ci
+      docker: huggingface/transformers-quantization-latest-gpu
+      ci_event: Daily CI
     secrets: inherit

.github/workflows/self-scheduled.yml

@@ -15,6 +15,19 @@ on:
       slack_report_channel:
         required: true
         type: string
+      runner:
+        required: true
+        type: string
+      docker:
+        required: true
+        type: string
+      ci_event:
+        required: true
+        type: string
+      working-directory-prefix:
+        default: ''
+        required: false
+        type: string
 
 env:
   HF_HOME: /mnt/cache
@@ -38,7 +51,7 @@ jobs:
     strategy:
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, daily-ci]
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, '${{ inputs.runner }}']
     container:
       image: huggingface/transformers-all-latest-gpu
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -96,6 +109,8 @@ jobs:
       folder_slices: ${{ needs.setup.outputs.folder_slices }}
       machine_type: ${{ matrix.machine_type }}
       slice_id: ${{ matrix.slice_id }}
+      runner: ${{ inputs.runner }}
+      docker: ${{ inputs.docker }}
     secrets: inherit
 
   run_pipelines_torch_gpu:
@@ -105,7 +120,7 @@ jobs:
       fail-fast: false
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, daily-ci]
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, '${{ inputs.runner }}']
     container:
       image: huggingface/transformers-pytorch-gpu
       options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -155,7 +170,7 @@ jobs:
       fail-fast: false
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, daily-ci]
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, '${{ inputs.runner }}']
     container:
       image: huggingface/transformers-tensorflow-gpu
       options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -206,7 +221,7 @@ jobs:
       fail-fast: false
       matrix:
         machine_type: [single-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, daily-ci]
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, '${{ inputs.runner }}']
     container:
       image: huggingface/transformers-all-latest-gpu
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -257,69 +272,88 @@ jobs:
       fail-fast: false
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, daily-ci]
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, '${{ inputs.runner }}']
     container:
-      image: huggingface/transformers-pytorch-deepspeed-latest-gpu
+      image: ${{ inputs.docker }}
       options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
     steps:
       - name: Update clone
-        working-directory: /workspace/transformers
+        working-directory: ${{ inputs.working-directory-prefix }}/transformers
         run: git fetch && git checkout ${{ github.sha }}
 
       - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /workspace/transformers
+        working-directory: ${{ inputs.working-directory-prefix }}/transformers
         run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
 
+      - name: Update / Install some packages (for Past CI)
+        if: ${{ contains(inputs.docker, '-past-') && contains(inputs.docker, '-pytorch-') }}
+        working-directory: ${{ inputs.working-directory-prefix }}/transformers
+        run: |
+          python3 -m pip install -U datasets
+          python3 -m pip install --no-cache-dir git+https://github.com/huggingface/accelerate@main#egg=accelerate
+
       - name: Remove cached torch extensions
         run: rm -rf /github/home/.cache/torch_extensions/
 
       # To avoid unknown test failures
-      - name: Pre build DeepSpeed *again*
+      - name: Pre build DeepSpeed *again* (for daily CI)
-        working-directory: /workspace
+        if: ${{ contains(inputs.ci_event, 'Daily CI') }}
+        working-directory: ${{ inputs.working-directory-prefix }}/
         run: |
           python3 -m pip uninstall -y deepspeed
           DS_DISABLE_NINJA=1 DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 python3 -m pip install deepspeed --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check
 
+      # To avoid unknown test failures
+      - name: Pre build DeepSpeed *again* (for nightly & Past CI)
+        if: ${{ contains(inputs.ci_event, 'Nightly CI') || contains(inputs.ci_event, 'Past CI') }}
+        working-directory: ${{ inputs.working-directory-prefix }}/
+        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 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: /workspace/transformers
+        working-directory: ${{ inputs.working-directory-prefix }}/transformers
         run: |
-          python utils/print_env.py
+          python3 utils/print_env.py
 
       - name: Show installed libraries and their versions
-        working-directory: /workspace/transformers
+        working-directory: ${{ inputs.working-directory-prefix }}/transformers
         run: pip freeze
 
       - name: Run all tests on GPU
-        working-directory: /workspace/transformers
+        working-directory: ${{ inputs.working-directory-prefix }}/transformers
         run: |
-          python -m pytest -v --make-reports=${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports tests/deepspeed tests/extended
+          python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports tests/deepspeed tests/extended
 
       - name: Failure short reports
         if: ${{ failure() }}
         continue-on-error: true
-        run: cat /workspace/transformers/reports/${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports/failures_short.txt
+        run: cat ${{ inputs.working-directory-prefix }}/transformers/reports/${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports/failures_short.txt
 
       - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports"
         if: ${{ always() }}
         uses: actions/upload-artifact@v4
         with:
           name: ${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports
-          path: /workspace/transformers/reports/${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports
+          path: ${{ inputs.working-directory-prefix }}/transformers/reports/${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports
 
   run_quantization_torch_gpu:
     if: ${{ inputs.job == 'run_quantization_torch_gpu' }}
     name: " "
     needs: setup
     strategy:
+      max-parallel: 4
       fail-fast: false
       matrix:
         folders: ${{ fromJson(needs.setup.outputs.quantization_matrix) }}
         machine_type: [single-gpu, multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, daily-ci]
+    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, '${{ inputs.runner }}']
     container:
       image: huggingface/transformers-quantization-latest-gpu
       options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -434,5 +468,6 @@ jobs:
       # This would be an empty string if `setup` is skipped.
       folder_slices: ${{ needs.setup.outputs.folder_slices }}
       quantization_matrix: ${{ needs.setup.outputs.quantization_matrix }}
-      
+      ci_event: ${{ inputs.ci_event }}
+
     secrets: inherit

.github/workflows/slack-report.yml

@@ -18,6 +18,9 @@ on:
       quantization_matrix:
         required: true
         type: string
+      ci_event:
+        required: true
+        type: string
 
 env:
   TRANSFORMERS_CI_RESULTS_UPLOAD_TOKEN: ${{ secrets.TRANSFORMERS_CI_RESULTS_UPLOAD_TOKEN }}
@@ -45,7 +48,7 @@ jobs:
           CI_SLACK_CHANNEL_DUMMY_TESTS: ${{ secrets.CI_SLACK_CHANNEL_DUMMY_TESTS }}
           SLACK_REPORT_CHANNEL: ${{ inputs.slack_report_channel }}
           ACCESS_REPO_INFO_TOKEN: ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
-          CI_EVENT: scheduled
+          CI_EVENT: ${{ inputs.ci_event }}
           CI_SHA: ${{ github.sha }}
           CI_WORKFLOW_REF: ${{ github.workflow_ref }}
           CI_TEST_JOB: ${{ inputs.job }}
@@ -76,7 +79,7 @@ jobs:
           CI_SLACK_BOT_TOKEN: ${{ secrets.CI_SLACK_BOT_TOKEN }}
           ACCESS_REPO_INFO_TOKEN: ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
           SLACK_REPORT_CHANNEL: ${{ inputs.slack_report_channel }}
-          CI_EVENT: scheduled
+          CI_EVENT: ${{ inputs.ci_event }}
           CI_SHA: ${{ github.sha }}
           CI_TEST_JOB: ${{ inputs.job }}
           SETUP_STATUS: ${{ inputs.setup_status }}

utils/notification_service.py

@@ -641,7 +641,7 @@ class Message:
 
     def get_new_model_failure_blocks(self, with_header=True, to_truncate=True):
         if self.prev_ci_artifacts is None:
-            return {}
+            return []
 
         sorted_dict = sorted(self.model_results.items(), key=lambda t: t[0])
 
@@ -767,10 +767,11 @@ class Message:
 
         # To save the list of new model failures
         blocks = self.get_new_model_failure_blocks(to_truncate=False)
-        failure_text = blocks[-1]["text"]["text"]
+        if blocks:
-        file_path = os.path.join(os.getcwd(), f"ci_results_{job_name}/new_model_failures.txt")
+            failure_text = blocks[-1]["text"]["text"]
-        with open(file_path, "w", encoding="UTF-8") as fp:
+            file_path = os.path.join(os.getcwd(), f"ci_results_{job_name}/new_model_failures.txt")
-            fp.write(failure_text)
+            with open(file_path, "w", encoding="UTF-8") as fp:
+                fp.write(failure_text)
 
 
 def retrieve_artifact(artifact_path: str, gpu: Optional[str]):
@@ -891,7 +892,7 @@ if __name__ == "__main__":
     # To find the PR number in a commit title, for example, `Add AwesomeFormer model (#99999)`
     pr_number_re = re.compile(r"\(#(\d+)\)$")
 
-    title = f"🤗 Results of the {ci_event} tests."
+    title = f"🤗 Results of {ci_event} - {os.getenv('CI_TEST_JOB')}."
     # Add Commit/PR title with a link for push CI
     # (check the title in 2 env. variables - depending on the CI is triggered via `push` or `workflow_run` event)
     ci_title_push = os.environ.get("CI_TITLE_PUSH")

utils/notification_service_quantization.py

@@ -175,7 +175,7 @@ if __name__ == "__main__":
     # This env. variable is set in workflow file (under the job `send_results`).
     ci_event = os.environ["CI_EVENT"]
 
-    title = f"🤗 Results of the {ci_event} tests."
+    title = f"🤗 Results of the {ci_event} - {os.getenv('CI_TEST_JOB')}."
 
     if setup_failed:
         Message.error_out(