Enable GitHub CI/CD runners

.github/workflows/self-push-amd.yml

@@ -17,25 +17,12 @@ env:
   RUN_PT_TF_CROSS_TESTS: 1
 
 jobs:
-  check_runner_status:
-    name: Check Runner Status
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout transformers
-        uses: actions/checkout@v3
-        with:
-          fetch-depth: 2
-
-      - name: Check Runner Status
-        run: python utils/check_self_hosted_runner.py --target_runners amd-mi210-single-gpu-ci-runner-docker --token ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
-
   check_runners:
     name: Check Runners
-    needs: check_runner_status
     strategy:
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: [self-hosted, docker-gpu, amd-gpu, '${{ matrix.machine_type }}', '${{ inputs.gpu_flavor }}']
+    runs-on: rocm
     container:
       image: huggingface/transformers-pytorch-amd-gpu-push-ci  # <--- We test only for PyTorch for now
       options: --device /dev/kfd --device /dev/dri --env HIP_VISIBLE_DEVICES --env ROCR_VISIBLE_DEVICES --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -54,14 +41,21 @@ jobs:
     strategy:
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: [self-hosted, docker-gpu, amd-gpu, '${{ matrix.machine_type }}', '${{ inputs.gpu_flavor }}']
+    runs-on: rocm
     container:
       image: huggingface/transformers-pytorch-amd-gpu-push-ci  # <--- We test only for PyTorch for now
       options: --device /dev/kfd --device /dev/dri --env HIP_VISIBLE_DEVICES --env ROCR_VISIBLE_DEVICES --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
     outputs:
       matrix: ${{ steps.set-matrix.outputs.matrix }}
       test_map: ${{ steps.set-matrix.outputs.test_map }}
     steps:
+      - name: Remove transformers repository (installed during docker image build)
+        working-directory: /
+        shell: bash
+        run:  | 
+          rm -r transformers
+          git clone https://github.com/ROCmSoftwarePlatform/transformers.git
+
       # Necessary to get the correct branch name and commit SHA for `workflow_run` event
       # We also take into account the `push` event (we might want to test some changes in a branch)
       - name: Prepare custom environment variables
@@ -152,11 +146,23 @@ jobs:
       matrix:
         folders: ${{ fromJson(needs.setup_gpu.outputs.matrix) }}
         machine_type: [single-gpu, multi-gpu]
-    runs-on: [self-hosted, docker-gpu, amd-gpu, '${{ matrix.machine_type }}', '${{ inputs.gpu_flavor }}']
+    runs-on: rocm
     container:
       image: huggingface/transformers-pytorch-amd-gpu-push-ci  # <--- We test only for PyTorch for now
       options: --device /dev/kfd --device /dev/dri --env HIP_VISIBLE_DEVICES --env ROCR_VISIBLE_DEVICES --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
     steps:
+
+      - name: Remove transformers repository (installed during docker image build)
+        working-directory: /
+        shell: bash
+        run:  | 
+          rm -r transformers
+          git clone https://github.com/ROCmSoftwarePlatform/transformers.git
+
+      - 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 .
+
       # Necessary to get the correct branch name and commit SHA for `workflow_run` event
       # We also take into account the `push` event (we might want to test some changes in a branch)
       - name: Prepare custom environment variables
@@ -189,10 +195,6 @@ jobs:
           git checkout ${{ env.CI_SHA }}
           echo "log = $(git log -n 1)"
 
-      - 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: 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
@@ -244,19 +246,15 @@ jobs:
     runs-on: ubuntu-latest
     if: always()
     needs: [
-        check_runner_status,
         check_runners,
         setup_gpu,
-        run_tests_amdgpu,
-#        run_tests_torch_cuda_extensions_single_gpu,
-#        run_tests_torch_cuda_extensions_multi_gpu
+        run_tests_amdgpu
     ]
     steps:
       - name: Preliminary job status
         shell: bash
         # For the meaning of these environment variables, see the job `Setup`
         run: |
-          echo "Runner availability: ${{ needs.check_runner_status.result }}"
           echo "Setup status: ${{ needs.setup_gpu.result }}"
           echo "Runner status: ${{ needs.check_runners.result }}"
 
@@ -297,28 +295,3 @@ jobs:
           echo "updated branch = $(git branch --show-current)"
           git checkout ${{ env.CI_SHA }}
           echo "log = $(git log -n 1)"
-
-      - uses: actions/download-artifact@v3
-      - 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_ID_AMD: ${{ secrets.CI_SLACK_CHANNEL_ID_AMD }}
-          CI_SLACK_CHANNEL_DUMMY_TESTS: ${{ secrets.CI_SLACK_CHANNEL_DUMMY_TESTS }}
-          CI_SLACK_REPORT_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID_AMD }}
-          ACCESS_REPO_INFO_TOKEN: ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
-          CI_EVENT: Push CI (AMD) - ${{ inputs.gpu_flavor }}
-          CI_TITLE_PUSH: ${{ github.event.head_commit.message }}
-          CI_TITLE_WORKFLOW_RUN: ${{ github.event.workflow_run.head_commit.message }}
-          CI_SHA: ${{ env.CI_SHA }}
-          RUNNER_STATUS: ${{ needs.check_runner_status.result }}
-          RUNNER_ENV_STATUS: ${{ needs.check_runners.result }}
-          SETUP_STATUS: ${{ needs.setup_gpu.result }}
-
-        # We pass `needs.setup_gpu.outputs.matrix` as the argument. A processing in `notification_service.py` to change
-        # `models/bert` to `models_bert` is required, as the artifact names use `_` instead of `/`.
-        run: |
-          pip install slack_sdk
-          pip show slack_sdk
-          python utils/notification_service.py "${{ needs.setup_gpu.outputs.matrix }}"

utils/tests_fetcher.py

@@ -19,7 +19,7 @@
 This util is designed to fetch tests to run on a PR so that only the tests impacted by the modifications are run, and
 when too many models are being impacted, only run the tests of a subset of core models. It works like this.
 
-Stage 1: Identify the modified files. For jobs that run on the main branch, it's just the diff with the last commit.
+Stage 1: Identify the modified files. For jobs that run on the master branch, it's just the diff with the last commit.
 On a PR, this takes all the files from the branching point to the current commit (so all modifications in a PR, not
 just the last commit) but excludes modifications that are on docstrings or comments only.
 
@@ -42,7 +42,7 @@
 python utils/tests_fetcher.py
 ```
 
-Base use to fetch the tests on a the main branch (with diff from the last commit):
+Base use to fetch the tests on a the master branch (with diff from the last commit):
 
 ```bash
 python utils/tests_fetcher.py --diff_with_last_commit
@@ -300,7 +300,7 @@ def get_modified_python_files(diff_with_last_commit: bool = False) -> List[str]:
     """
     Return a list of python files that have been modified between:
 
-    - the current head and the main branch if `diff_with_last_commit=False` (default)
+    - the current head and the master branch if `diff_with_last_commit=False` (default)
     - the current head and its parent commit otherwise.
 
     Returns:
@@ -311,15 +311,15 @@ def get_modified_python_files(diff_with_last_commit: bool = False) -> List[str]:
     repo = Repo(PATH_TO_REPO)
 
     if not diff_with_last_commit:
-        print(f"main is at {repo.refs.main.commit}")
+        print(f"master is at {repo.refs.master.commit}")
         print(f"Current head is at {repo.head.commit}")
 
-        branching_commits = repo.merge_base(repo.refs.main, repo.head)
+        branching_commits = repo.merge_base(repo.refs.master, repo.head)
         for commit in branching_commits:
             print(f"Branching commit: {commit}")
         return get_diff(repo, repo.head.commit, branching_commits)
     else:
-        print(f"main is at {repo.head.commit}")
+        print(f"master is at {repo.head.commit}")
         parent_commits = repo.head.commit.parents
         for commit in parent_commits:
             print(f"Parent commit: {commit}")
@@ -424,7 +424,7 @@ def get_doctest_files(diff_with_last_commit: bool = False) -> List[str]:
     """
     Return a list of python and Markdown files where doc example have been modified between:
 
-    - the current head and the main branch if `diff_with_last_commit=False` (default)
+    - the current head and the master branch if `diff_with_last_commit=False` (default)
     - the current head and its parent commit otherwise.
 
     Returns:
@@ -435,15 +435,15 @@ def get_doctest_files(diff_with_last_commit: bool = False) -> List[str]:
 
     test_files_to_run = []  # noqa
     if not diff_with_last_commit:
-        print(f"main is at {repo.refs.main.commit}")
+        print(f"master is at {repo.refs.master.commit}")
         print(f"Current head is at {repo.head.commit}")
 
-        branching_commits = repo.merge_base(repo.refs.main, repo.head)
+        branching_commits = repo.merge_base(repo.refs.master, repo.head)
         for commit in branching_commits:
             print(f"Branching commit: {commit}")
         test_files_to_run = get_diff_for_doctesting(repo, repo.head.commit, branching_commits)
     else:
-        print(f"main is at {repo.head.commit}")
+        print(f"master is at {repo.head.commit}")
         parent_commits = repo.head.commit.parents
         for commit in parent_commits:
             print(f"Parent commit: {commit}")
@@ -452,7 +452,7 @@ def get_doctest_files(diff_with_last_commit: bool = False) -> List[str]:
     all_test_files_to_run = get_all_doctest_files()
 
     # Add to the test files to run any removed entry from "utils/not_doctested.txt".
-    new_test_files = get_new_doctest_files(repo, repo.head.commit, repo.refs.main.commit)
+    new_test_files = get_new_doctest_files(repo, repo.head.commit, repo.refs.master.commit)
     test_files_to_run = list(set(test_files_to_run + new_test_files))
 
     # Do not run slow doctest tests on CircleCI
@@ -766,8 +766,8 @@ def create_reverse_dependency_map() -> Dict[str, List[str]]:
         something_changed = False
         for m in all_modules:
             for d in direct_deps[m]:
-                # We stop recursing at an init (cause we always end up in the main init and we don't want to add all
-                # files which the main init imports)
+                # We stop recursing at an init (cause we always end up in the master init and we don't want to add all
+                # files which the master init imports)
                 if d.endswith("__init__.py"):
                     continue
                 if d not in direct_deps:
@@ -910,7 +910,7 @@ def infer_tests_to_run(
     json_output_file: Optional[str] = None,
 ):
     """
-    The main function called by the test fetcher. Determines the tests to run from the diff.
+    The master function called by the test fetcher. Determines the tests to run from the diff.
 
     Args:
         output_file (`str`):
@@ -922,8 +922,8 @@ def infer_tests_to_run(
             - doctest_list.txt: The list of doctests to run.
 
         diff_with_last_commit (`bool`, *optional*, defaults to `False`):
-            Whether to analyze the diff with the last commit (for use on the main branch after a PR is merged) or with
-            the branching point from main (for use on each PR).
+            Whether to analyze the diff with the last commit (for use on the master branch after a PR is merged) or with
+            the branching point from master (for use on each PR).
         filter_models (`bool`, *optional*, defaults to `True`):
             Whether or not to filter the tests to core models only, when a file modified results in a lot of model
             tests.
@@ -1112,8 +1112,8 @@ def parse_commit_message(commit_message: str) -> Dict[str, bool]:
             print("Force-launching all tests")
 
         diff_with_last_commit = args.diff_with_last_commit
-        if not diff_with_last_commit and not repo.head.is_detached and repo.head.ref == repo.refs.main:
-            print("main branch detected, fetching tests against last commit.")
+        if not diff_with_last_commit and not repo.head.is_detached and repo.head.ref == repo.refs.master:
+            print("master branch detected, fetching tests against last commit.")
             diff_with_last_commit = True
 
         if not commit_flags["test_all"]: