Merge pull request #2924 from AI-Hypercomputer:xibin/ci

PiperOrigin-RevId: 868178412

Author: Google-ML-Automation

.github/workflows/UploadDockerImages.yml

@@ -15,7 +15,7 @@
 # This workflow builds and pushes MaxText images for both TPU and GPU devices.
 # It runs automatically daily at 12am UTC, on Pull Requests, or manually via Workflow Dispatch.
 
-name: Build Images
+name: Build and Test Images
 
 on:
   schedule:
@@ -32,6 +32,11 @@ on:
           - all
           - tpu
           - gpu
+      for_dev_test:
+        description: 'For development test purpose. All images will be added a -test suffix'
+        required: false
+        type: boolean
+        default: false
 
 permissions:
   contents: read
@@ -42,6 +47,7 @@ jobs:
     outputs:
       maxtext_sha: ${{ steps.vars.outputs.maxtext_sha }}
       image_date: ${{ steps.vars.outputs.image_date }}
+      image_suffix: ${{ steps.vars.outputs.image_suffix }}
     steps:
      - name: Checkout MaxText
        uses: actions/checkout@v5
@@ -55,6 +61,13 @@ jobs:
         # Image date
         echo "image_date=$(date +%Y-%m-%d)" >> $GITHUB_OUTPUT
 
+        # If for_dev_test is true, set suffix to -test, otherwise empty
+        if [[ "${{ github.event.inputs.for_dev_test }}" == "true" ]]; then
+          echo "image_suffix=-test" >> $GITHUB_OUTPUT
+        else
+          echo "image_suffix=" >> $GITHUB_OUTPUT
+        fi
+
   tpu-pre-training:
     name: ${{ matrix.image_name }}
     needs: setup
@@ -72,7 +85,7 @@ jobs:
             dockerfile: ./dependencies/dockerfiles/maxtext_tpu_dependencies.Dockerfile
     uses: ./.github/workflows/build_and_push_docker_image.yml
     with:
-      image_name: ${{ matrix.image_name }}
+      image_name: ${{ matrix.image_name }}${{ needs.setup.outputs.image_suffix  }}
       device: ${{ matrix.device }}
       build_mode: ${{ matrix.build_mode }}
       dockerfile: ${{ matrix.dockerfile }}
@@ -96,14 +109,13 @@ jobs:
             dockerfile: ./dependencies/dockerfiles/maxtext_post_training_local_dependencies.Dockerfile
     uses: ./.github/workflows/build_and_push_docker_image.yml
     with:
-      image_name: ${{ matrix.image_name }}
+      image_name: ${{ matrix.image_name }}${{ needs.setup.outputs.image_suffix  }}
       device: ${{ matrix.device }}
       build_mode: ${{ matrix.build_mode }}
       dockerfile: ${{ matrix.dockerfile }}
       maxtext_sha: ${{ needs.setup.outputs.maxtext_sha }}
       image_date: ${{ needs.setup.outputs.image_date }}
       base_image: gcr.io/tpu-prod-env-multipod/maxtext_jax_stable:${{ needs.setup.outputs.image_date }}
-      is_post_training: true
 
   gpu-pre-training:
     name: ${{ matrix.image_name }}
@@ -122,9 +134,48 @@ jobs:
             dockerfile: ./dependencies/dockerfiles/maxtext_gpu_dependencies.Dockerfile
     uses: ./.github/workflows/build_and_push_docker_image.yml
     with:
-      image_name: ${{ matrix.image_name }}
+      image_name: ${{ matrix.image_name }}${{ needs.setup.outputs.image_suffix  }}
       device: ${{ matrix.device }}
       build_mode: ${{ matrix.build_mode }}
       dockerfile: ${{ matrix.dockerfile }}
       maxtext_sha: ${{ needs.setup.outputs.maxtext_sha }}
       image_date: ${{ needs.setup.outputs.image_date }}
+
+  # TEST JOBS
+  pre-training-tpu-tests:
+    needs: [setup, tpu-pre-training]
+    strategy:
+      fail-fast: false
+      matrix:
+        image: [maxtext_jax_stable, maxtext_jax_nightly]
+    uses: ./.github/workflows/test_and_tag_docker_image.yml
+    with:
+      image_name: ${{ matrix.image }}${{ needs.setup.outputs.image_suffix  }}
+      image_date: ${{ needs.setup.outputs.image_date }}
+      test_mode: tpu-pre-training
+
+  post-training-tpu-tests:
+    needs: [setup, tpu-post-training]
+    strategy:
+      fail-fast: false
+      matrix:
+        image: [maxtext_post_training_stable, maxtext_post_training_nightly]
+    uses: ./.github/workflows/test_and_tag_docker_image.yml
+    with:
+      image_name: ${{ matrix.image }}${{ needs.setup.outputs.image_suffix  }}
+      image_date: ${{ needs.setup.outputs.image_date }}
+      test_mode: tpu-post-training
+
+
+  pre-training-gpu-tests:
+    needs: [setup, gpu-pre-training]
+    strategy:
+      fail-fast: false
+      matrix:
+        image: [maxtext_gpu_jax_stable, maxtext_gpu_jax_nightly]
+    uses: ./.github/workflows/test_and_tag_docker_image.yml
+    with:
+      image_name: ${{ matrix.image }}${{ needs.setup.outputs.image_suffix  }}
+      image_date: ${{ needs.setup.outputs.image_date }}
+      test_mode: gpu-pre-training
+

.github/workflows/build_and_push_docker_image.yml

@@ -41,10 +41,6 @@ on:
         required: false
         type: string
         default: ''
-      is_post_training:
-        required: false
-        type: boolean
-        default: false
 
 permissions:
   contents: read
@@ -82,7 +78,9 @@ jobs:
           ref: ${{ inputs.maxtext_sha }}
 
       - name: Checkout post-training dependencies
-        if: steps.check.outputs.should_run == 'true' && inputs.image_name == 'maxtext_post_training_nightly'
+        if: |
+          steps.check.outputs.should_run == 'true' &&
+          contains(inputs.image_name, 'post_training_nightly')
         run: |
           git clone https://github.com/google/tunix.git ./tunix
           git clone https://github.com/vllm-project/vllm.git ./vllm
@@ -110,8 +108,7 @@ jobs:
           push: true
           context: .
           file: ${{ inputs.dockerfile }}
-          tags: gcr.io/tpu-prod-env-multipod/${{ inputs.image_name }}:latest
-          cache-from: type=gha
+          tags: gcr.io/tpu-prod-env-multipod/${{ inputs.image_name }}:${{ inputs.image_date }}-build-${{ github.run_id }}
           outputs: type=image,compression=zstd,force-compression=true
           build-args: |
             DEVICE=${{ inputs.device }}
@@ -126,23 +123,19 @@ jobs:
         shell: bash
         run: |
           SOURCE_IMAGE="gcr.io/tpu-prod-env-multipod/${{ inputs.image_name }}"
-
-          # Add date tag
-          gcloud container images add-tag "$SOURCE_IMAGE:latest" "$SOURCE_IMAGE:${{ inputs.image_date }}" --quiet
+          TEMP_IMG="$SOURCE_IMAGE:${{ inputs.image_date }}-build-${{ github.run_id }}"
 
           # Convert date to YYYYMMDD format
           clean_date=$(echo "${{ inputs.image_date }}" | sed 's/[-:]//g' | cut -c1-8)
 
           # Add MaxText tag
           maxtext_hash=$(git rev-parse --short HEAD)
-          gcloud container images add-tag "$SOURCE_IMAGE:latest" "$SOURCE_IMAGE:maxtext_${maxtext_hash}_${clean_date}" --quiet
+          gcloud container images add-tag "$TEMP_IMG" "$SOURCE_IMAGE:maxtext_${maxtext_hash}_${clean_date}" --quiet
 
           # Add post-training dependencies tags
-          if [ "${{ inputs.is_post_training }}" == "true" ]; then
-            for dir in tunix vllm tpu-inference; do
-              if [ -d "./$dir" ]; then
-                dir_hash=$(git -C "$dir" rev-parse --short HEAD)
-                gcloud container images add-tag "$SOURCE_IMAGE:latest" "$SOURCE_IMAGE:${dir}_${dir_hash}_${clean_date}" --quiet
-              fi
-            done
-          fi
+          for dir in tunix vllm tpu-inference; do
+            if [ -d "./$dir" ]; then
+              dir_hash=$(git -C "$dir" rev-parse --short HEAD)
+              gcloud container images add-tag "$TEMP_IMG" "$SOURCE_IMAGE:${dir}_${dir_hash}_${clean_date}" --quiet
+            fi
+          done