restructure workflow files

.github/workflows/build_and_test.yml

@@ -0,0 +1,34 @@
+name: Build images and run tests
+
+on:
+  workflow_call:
+    inputs:
+      dockerfile_path:
+        description: 'Path to the Dockerfile'
+        type: string
+        required: true
+      docker_image_tag: # It should be in the form as mentioned in our internal docs. Just pass insensitive data here like huggingface-text-generation-inference-${accelerator}.${version}:latest
+        description: 'Docker Image Tag'
+        type: string
+        required: true
+
+jobs:
+  build-push-image:
+    uses: ./.github/workflows/build_push_image.yml
+    secrets: inherit
+    with:
+      region: us-central1
+      dockerfile_path: ${{ inputs.dockerfile_path }}
+      docker_image_tag: ${{ inputs.docker_image_tag }}
+      gcp_artifact_registry_repository: deep-learning-images
+      GCP_PROJECT_ID: ${{ vars.GCP_PROJECT_ID }}
+
+  model_test:
+    needs: build-push-image
+    uses: ./.github/workflows/test_model.yml
+    secrets: inherit
+    with:
+      region: us-central1
+      docker_image_tag: ${{ inputs.docker_image_tag }}
+      gcp_artifact_registry_repository: deep-learning-images
+      GCP_PROJECT_ID: ${{ vars.GCP_PROJECT_ID }}

.github/workflows/build-push-image.yml → .github/workflows/build_push_image.yml

@@ -1,4 +1,4 @@
-name: Build and Push to GCP Artifact Registry Resuable Workflow
+name: Build and push images
 
 on:
   workflow_call:
@@ -19,16 +19,13 @@ on:
         description: 'GCP Artifact Registry Repository'
         type: string
         required: true
-    secrets:
-      GCP_SERVICE_ACCOUNT_JSON_KEY:
-        description: 'Service Account'
-        required: true
       GCP_PROJECT_ID:
         description: 'GCP Project ID'
+        type: string
         required: true
 
 jobs:
-    build-push-image-job:
+    build-push-image:
       name: Build and Push Docker Image to GCP Artifact Registry
       runs-on:  [intel-cpu, 8-cpu, ci]
       steps:
@@ -55,7 +52,5 @@ jobs:
           uses: 'docker/[email protected]'
           with:
             file: ${{ inputs.dockerfile_path }} # Read more about it here: https://github.com/marketplace/actions/build-and-push-docker-images#customizing
-            tags: ${{ inputs.region }}-docker.pkg.dev/${{ secrets.GCP_PROJECT_ID }}/${{ inputs.gcp_artifact_registry_repository }}/${{ inputs.docker_image_tag }}
+            tags: ${{ inputs.region }}-docker.pkg.dev/${{ inputs.GCP_PROJECT_ID }}/${{ inputs.gcp_artifact_registry_repository }}/${{ inputs.docker_image_tag }}
             push: true
-
-

.github/workflows/run_build_and_test.yml

@@ -0,0 +1,20 @@
+name: Build images and run tests
+
+on:
+  push:
+    branches:
+      - build_images_and_run_tests_*
+
+jobs:
+  build-push-image:
+    strategy:
+      fail-fast: false
+      matrix:
+        torch: [2.1]
+        transformers: [4.38.1]
+        python: [310]
+    uses: ./.github/workflows/build_and_test.yml
+    secrets: inherit
+    with:
+      dockerfile_path: ./containers/pytorch/training/gpu/${{ matrix.torch }}/transformers/${{ matrix.transformers }}/py${{ matrix.python }}/Dockerfile
+      docker_image_tag: huggingface-pytorch-training-gpu-${{ matrix.torch }}.transformers.${{ matrix.transformers }}.py${{ matrix.python }}:latest

.github/workflows/test_model.yml

@@ -1,9 +1,24 @@
-name: Model Test
+name: Model tests
 
 on:
-  push:
-    branches:
-      - run_tests_against_images_*
+  workflow_call:
+    inputs:
+      region:
+        description: 'Region where the Artifact Registry is located'
+        type: string
+        required: true
+      docker_image_tag: # It should be in the form as mentioned in our internal docs. Just pass insensitive data here like huggingface-text-generation-inference-${accelerator}.${version}:latest
+        description: 'Docker Image Tag'
+        type: string
+        required: true
+      gcp_artifact_registry_repository:
+        description: 'GCP Artifact Registry Repository'
+        type: string
+        required: true
+      GCP_PROJECT_ID:
+        description: 'GCP Project ID'
+        type: string
+        required: true
 
 env:
   HF_HOME: /mnt/cache
@@ -13,16 +28,28 @@ env:
 
 jobs:
   run_test:
-    name: github repo
+    name: Run model tests
     runs-on: [single-gpu, nvidia-gpu, a10, ci]
     container:
-      # TODO: make this $ {{ inputs.image }} to use GCP's DLC images
-      image: huggingface/transformers-all-latest-gpu
+      image: ${{ inputs.region }}-docker.pkg.dev/${{ inputs.GCP_PROJECT_ID }}/${{ inputs.gcp_artifact_registry_repository }}/${{ inputs.docker_image_tag }}
       options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+      credentials:
+        username: _json_key
+        password: ${{ secrets.GCP_SERVICE_ACCOUNT_JSON_KEY }}
     steps:
       - name: Checkout repository
         uses: actions/checkout@v3
 
+      - name: Prepare `transformers` examples/pytorch directory
+        shell: bash
+        run: |
+          git config --global --add safe.directory /__w/Google-Cloud-Containers/Google-Cloud-Containers
+          git clone https://github.com/huggingface/transformers.git
+          cd transformers
+          git checkout tags/v4.38.1
+          git log -n 1
+
       - name: run tests
         run: |
+          export TRANSFORMERS_DIR=/__w/Google-Cloud-Containers/Google-Cloud-Containers/transformers
           python3 -m pytest -v tests/models

tests/models/vit/test_training_vit.py

@@ -4,14 +4,8 @@
 from ..test_model import ModelTrainingTestMixin, TestCasePlus
 
 
-# So far, the image we use has a `/transformers` directory.
-# TODO: Find a way to get the installed `transformers` directory.
-SRC_DIRS = [
-    os.path.join("/transformers/examples/pytorch", dirname)
-    for dirname in [
-        "image-classification",
-    ]
-]
+# `TRANSFORMERS_DIR` is an environment variable pointing to a `transformers` source directory (containing `examples`)
+SRC_DIRS = [os.path.join(os.getenv("TRANSFORMERS_DIR"), "examples/pytorch", dirname) for dirname in ["image-classification"]]
 sys.path.extend(SRC_DIRS)