[AMD] Build Miles nightly ROCm images with docker/build.py and test ROCm 7.2 (#31757)

.github/workflows/nightly-test-amd-miles-rocm720.yml

@@ -1,4 +1,4 @@
-name: Nightly Test Miles (AMD ROCm 7.0)
+name: Nightly Test Miles (AMD ROCm 7.2)
 
 on:
   schedule:
@@ -7,7 +7,7 @@ on:
   workflow_dispatch:
     inputs:
       image_tag:
-        description: "Override miles image tag (e.g. miles-rocm700-mi35x-20260621); empty = newest by date"
+        description: "Override miles image tag (e.g. miles-rocm720-mi35x-20260621); empty = newest by date"
         required: false
         type: string
         default: ""
@@ -46,10 +46,10 @@ jobs:
           username: ${{ secrets.DOCKERHUB_AMD_USERNAME }}
           password: ${{ secrets.DOCKERHUB_AMD_TOKEN }}
 
-      - name: Resolve newest miles ROCm 7.0 image
+      - name: Resolve newest miles ROCm 7.2 image
         run: |
           repo="rocm/sgl-dev"
-          base="miles-rocm700-mi35x"
+          base="miles-rocm720-mi35x"
           tag="${{ inputs.image_tag }}"
           if [ -z "${tag}" ]; then
             # Walk back from today; the build job pushes a dated tag at 12:00 UTC.

.github/workflows/release-docker-amd-miles-rocm700-nightly.yml

@@ -14,17 +14,14 @@ jobs:
     if: github.repository == 'sgl-project/sglang'
     runs-on: amd-docker-scale
     environment: "prod"
+    permissions:
+      contents: read
     strategy:
       fail-fast: false
       matrix:
-        include:
+        variant:
-          - gpu_arch: gfx950
+          - rocm700-mi35x
-            rocm_tag: rocm700-mi35x
+          - rocm700-mi30x
-            sglang_image_repo: rocm/sgl-dev
-            sglang_image_tag: v0.5.14-rocm700-mi35x-20260627
-          - gpu_arch: gfx942
-            rocm_tag: rocm700-mi30x
-            sglang_image_tag: v0.5.10-rocm700-mi30x
     steps:
       - name: Checkout miles repository
         uses: actions/checkout@v4
@@ -32,13 +29,21 @@ jobs:
           repository: radixark/miles
           ref: main
           fetch-depth: 0
+          persist-credentials: false
+
+      - name: Record miles commit
+        run: echo "Building from miles main @ $(git rev-parse HEAD)"
 
       - name: Set date
         run: echo "DATE=$(date +%Y%m%d)" >> $GITHUB_ENV
 
-      - name: Record miles commit
+      - name: Install docker/build.py dependencies
-        id: miles
+        run: python3 -m pip install --quiet typer
-        run: echo "sha=$(git rev-parse HEAD)" >> $GITHUB_OUTPUT
+
+      - name: Build image via miles docker/build.py and tag with date
+        run: |
+          python3 docker/build.py --variant ${{ matrix.variant }} --image-tag custom --custom-tag miles
+          docker tag rocm/sgl-dev:miles-${{ matrix.variant }} rocm/sgl-dev:miles-${{ matrix.variant }}-${DATE}
 
       - name: Login to Docker Hub (AMD)
         uses: docker/login-action@v2
@@ -46,18 +51,5 @@ jobs:
           username: ${{ secrets.DOCKERHUB_AMD_USERNAME }}
           password: ${{ secrets.DOCKERHUB_AMD_TOKEN }}
 
-      - name: Build and push miles image to rocm/sgl-dev
+      - name: Push dated tag
-        run: |
+        run: docker push rocm/sgl-dev:miles-${{ matrix.variant }}-${DATE}
-          base="miles-${{ matrix.rocm_tag }}"
-          dated="${base}-${{ env.DATE }}"
-          echo "IMAGE_TAG=${dated}" >> $GITHUB_ENV
-          echo "Building rocm/sgl-dev:${dated} from miles main @ ${{ steps.miles.outputs.sha }}"
-          docker build . \
-            -f docker/Dockerfile.rocm \
-            --build-arg GPU_ARCH=${{ matrix.gpu_arch }} \
-            --build-arg SGLANG_IMAGE_REPO=${{ matrix.sglang_image_repo || 'lmsysorg/sglang' }} \
-            --build-arg SGLANG_IMAGE_TAG=${{ matrix.sglang_image_tag }} \
-            --build-arg MILES_COMMIT=${{ steps.miles.outputs.sha }} \
-            -t rocm/sgl-dev:${dated} \
-            --no-cache
-          docker push rocm/sgl-dev:${dated}

.github/workflows/release-docker-amd-miles-rocm720-nightly.yml

@@ -14,13 +14,13 @@ jobs:
     if: github.repository == 'sgl-project/sglang'
     runs-on: amd-docker-scale
     environment: "prod"
+    permissions:
+      contents: read
     strategy:
       fail-fast: false
       matrix:
-        include:
+        variant:
-          - gpu_arch: gfx950
+          - rocm720-mi35x
-            rocm_tag: rocm720-mi35x
-            sglang_image_tag: v0.5.10-rocm720-mi35x
     steps:
       - name: Checkout miles repository
         uses: actions/checkout@v4
@@ -28,13 +28,21 @@ jobs:
           repository: radixark/miles
           ref: main
           fetch-depth: 0
+          persist-credentials: false
+
+      - name: Record miles commit
+        run: echo "Building from miles main @ $(git rev-parse HEAD)"
 
       - name: Set date
         run: echo "DATE=$(date +%Y%m%d)" >> $GITHUB_ENV
 
-      - name: Record miles commit
+      - name: Install docker/build.py dependencies
-        id: miles
+        run: python3 -m pip install --quiet typer
-        run: echo "sha=$(git rev-parse HEAD)" >> $GITHUB_OUTPUT
+
+      - name: Build image via miles docker/build.py and tag with date
+        run: |
+          python3 docker/build.py --variant ${{ matrix.variant }} --image-tag custom --custom-tag miles
+          docker tag rocm/sgl-dev:miles-${{ matrix.variant }} rocm/sgl-dev:miles-${{ matrix.variant }}-${DATE}
 
       - name: Login to Docker Hub (AMD)
         uses: docker/login-action@v2
@@ -42,17 +50,5 @@ jobs:
           username: ${{ secrets.DOCKERHUB_AMD_USERNAME }}
           password: ${{ secrets.DOCKERHUB_AMD_TOKEN }}
 
-      - name: Build and push miles image to rocm/sgl-dev
+      - name: Push dated tag
-        run: |
+        run: docker push rocm/sgl-dev:miles-${{ matrix.variant }}-${DATE}
-          base="miles-${{ matrix.rocm_tag }}"
-          dated="${base}-${{ env.DATE }}"
-          echo "IMAGE_TAG=${dated}" >> $GITHUB_ENV
-          echo "Building rocm/sgl-dev:${dated} from miles main @ ${{ steps.miles.outputs.sha }}"
-          docker build . \
-            -f docker/Dockerfile.rocm \
-            --build-arg GPU_ARCH=${{ matrix.gpu_arch }} \
-            --build-arg SGLANG_IMAGE_TAG=${{ matrix.sglang_image_tag }} \
-            --build-arg MILES_COMMIT=${{ steps.miles.outputs.sha }} \
-            -t rocm/sgl-dev:${dated} \
-            --no-cache
-          docker push rocm/sgl-dev:${dated}