diff --git a/.github/workflows/auto-label.yml b/.github/workflows/auto-label.yml
deleted file mode 100644
index 5feff289..00000000
--- a/.github/workflows/auto-label.yml
+++ /dev/null
@@ -1,53 +0,0 @@
-name: Auto labeling
-on:
-  issues:
-    types: [opened]
-  pull_request:
-    types: [opened]
-
-permissions:
-  contents: read
-  issues: write
-  pull-requests: write
-
-jobs:
-  # Label ISSUES using Renato66/auto-label
-  label-issues:
-    if: github.event_name == 'issues'
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v4
-        with:
-          sparse-checkout: |
-            .github/workflows/auto-label.json5
-          sparse-checkout-cone-mode: false
-      - uses: Renato66/auto-label@v3
-        with:
-          repo-token: ${{ secrets.GITHUB_TOKEN }}
-
-  # Add ISSUES to ROCm Project #91 so they land in Todo
-  add-issues-to-project:
-    if: github.event_name == 'issues'
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/add-to-project@v1.0.2
-        with:
-          project-url: https://github.com/orgs/ROCm/projects/91
-          github-token: ${{ secrets.ADD_TO_PROJECT_PAT }}
-
-  # PRs: label so the project rule moves them to In Progress
-  label-prs:
-    if: github.event_name == 'pull_request' && !github.event.pull_request.head.repo.fork
-    runs-on: ubuntu-latest
-    steps:
-      - name: Add iris + in-progress labels to PR
-        uses: actions/github-script@v7
-        with:
-          script: |
-            await github.rest.issues.addLabels({
-              owner: context.repo.owner,
-              repo: context.repo.repo,
-              issue_number: context.payload.pull_request.number,
-              labels: ["iris", "in-progress"]
-            })
-
diff --git a/.github/workflows/docker-build-push.yml b/.github/workflows/docker-build-push.yml
new file mode 100644
index 00000000..87e84d75
--- /dev/null
+++ b/.github/workflows/docker-build-push.yml
@@ -0,0 +1,98 @@
+name: Build and Push Docker Image
+
+on:
+  push:
+    branches:
+      - muhaawad/docker-images
+  workflow_dispatch:
+    inputs:
+      triton_commit:
+        description: 'Triton commit SHA to use'
+        required: false
+        default: 'aafec417bded34db6308f5b3d6023daefae43905'
+        type: string
+
+# This workflow always runs, even with [skip ci] or [ci skip] in commit message
+
+env:
+  DOCKERHUB_USERNAME: muhaawad
+  IMAGE_NAME: iris-dev
+
+jobs:
+  build-test:
+    runs-on: [self-hosted, mi3008x]
+    permissions:
+      contents: read
+      packages: write
+    timeout-minutes: 60
+
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - dockerfile: ./docker/Dockerfile.rocm6.4
+            base-name: "rocm6.4.4_ubuntu24.04_py3.12_pytorch_release_2.7.1"
+          - dockerfile: ./docker/Dockerfile.rocm7.0
+            base-name: "rocm7.0.2_ubuntu24.04_py3.12_pytorch_release_2.8.0"
+          - dockerfile: ./docker/Dockerfile.rocm7.1
+            base-name: "rocm7.1_ubuntu24.04_py3.12_pytorch_release_2.8.0"
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Set Triton SHA
+        id: triton
+        run: |
+          TRITON_COMMIT="${{ inputs.triton_commit || 'aafec417bded34db6308f5b3d6023daefae43905' }}"
+          TRITON_SHORT="${TRITON_COMMIT:0:7}"
+          echo "short_sha=${TRITON_SHORT}" >> $GITHUB_OUTPUT
+          echo "full_sha=${TRITON_COMMIT}" >> $GITHUB_OUTPUT
+
+      - name: Build Docker image
+        run: |
+          IMAGE_TAG="${{ env.DOCKERHUB_USERNAME }}/${{ env.IMAGE_NAME }}:${{ matrix.base-name }}_triton_${{ steps.triton.outputs.short_sha }}"
+          echo "Building ${IMAGE_TAG}..."
+          docker build \
+            -f ${{ matrix.dockerfile }} \
+            -t ${IMAGE_TAG} \
+            --build-arg TRITON_COMMIT=${{ steps.triton.outputs.full_sha }} \
+            ./docker
+          echo "✅ Build complete!"
+
+#      - name: Run validation tests
+#        run: |
+#          set -e
+#          
+#          IMAGE_TAG="${{ env.DOCKERHUB_USERNAME }}/${{ env.IMAGE_NAME }}:${{ matrix.base-name }}_triton_${{ steps.triton.outputs.short_sha }}"
+#          
+#          echo "::group::Running validation tests"
+#          bash .github/scripts/container_exec.sh --image "${IMAGE_TAG}" "
+#            set -e
+#            pip install -e .
+#            
+#            echo '=== Running external validation test ==='
+#            wget -O test_iris_distributed.py https://gist.githubusercontent.com/mawad-amd/6375dc078e39e256828f379e03310ec7/raw/a527c3192bee4615292769e340b1c73676f6945a/test_iris_distributed.py
+#            python test_iris_distributed.py
+#            
+#            echo '=== Running external gluon validation test ==='
+#            wget -O test_iris_gluon_distributed.py https://gist.githubusercontent.com/mawad-amd/2666dde8ebe2755eb0c4f2108709fcd5/raw/aa567ef3185c37a80d25bc9724ae9589548261b4/test_iris_gluon_distributed.py
+#            python test_iris_gluon_distributed.py
+#          "
+#          echo "::endgroup::"
+#          
+#          echo "✅ All validation tests passed for ${{ matrix.base-name }}!"
+
+      - name: Log in to Docker Hub
+        uses: docker/login-action@v3
+        with:
+          username: ${{ env.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.IRIS_DOCKERHUB_TOKEN }}
+
+      - name: Push Docker image
+        run: |
+          IMAGE_TAG="${{ env.DOCKERHUB_USERNAME }}/${{ env.IMAGE_NAME }}:${{ matrix.base-name }}_triton_${{ steps.triton.outputs.short_sha }}"
+          echo "Pushing ${IMAGE_TAG} to Docker Hub..."
+          docker push ${IMAGE_TAG}
+          echo "✅ Successfully pushed ${{ matrix.base-name }}!"
+
diff --git a/.github/workflows/docs.yml b/.github/workflows/docs.yml
deleted file mode 100644
index 4e7dd3da..00000000
--- a/.github/workflows/docs.yml
+++ /dev/null
@@ -1,80 +0,0 @@
-name: Build and Deploy Documentation
-
-on:
-  push:
-    branches: [ main ]
-    paths:
-      - 'docs/**'
-      - 'iris/**'
-      - 'examples/**'
-      - '.github/workflows/docs.yml'
-  pull_request:
-    branches: [ main ]
-    paths:
-      - 'docs/**'
-      - 'iris/**'
-      - 'examples/**'
-      - '.github/workflows/docs.yml'
-
-permissions:
-  contents: read
-  pages: write
-  id-token: write
-
-concurrency:
-  group: "pages-${{ github.ref }}"
-  cancel-in-progress: false
-
-jobs:
-  build:
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-        with:
-          ref: ${{ github.ref }}
-        
-      - name: Set up Python
-        uses: actions/setup-python@v4
-        with:
-          python-version: '3.11'
-          
-      - name: Install system dependencies
-        run: |
-          sudo apt-get update
-          sudo apt-get install -y build-essential
-          
-      - name: Install Python dependencies
-        run: |
-          python -m pip install --upgrade pip
-          pip install -r docs/sphinx/requirements.txt
-          
-      - name: Build documentation
-        run: |
-          cd docs
-          chmod +x build_docs.sh
-          ./build_docs.sh
-          
-      - name: Upload documentation artifact
-        uses: actions/upload-artifact@v4
-        with:
-          name: documentation
-          path: docs/_build/html
-          retention-days: 30
-          
-      - name: Upload artifact for GitHub Pages
-        uses: actions/upload-pages-artifact@v3
-        with:
-          path: docs/_build/html
-
-  deploy:
-    environment:
-      name: github-pages
-      url: ${{ steps.deployment.outputs.page_url }}
-    runs-on: ubuntu-latest
-    needs: build
-    if: github.ref == 'refs/heads/main' && github.event_name == 'push'
-    steps:
-      - name: Deploy to GitHub Pages
-        id: deployment
-        uses: actions/deploy-pages@v4
diff --git a/.github/workflows/iris-external-validation-test.yml b/.github/workflows/iris-external-validation-test.yml
deleted file mode 100644
index 2cac214c..00000000
--- a/.github/workflows/iris-external-validation-test.yml
+++ /dev/null
@@ -1,95 +0,0 @@
-name: Iris External Validation Test
-
-on:
-  push:
-    branches: [ main ]
-  pull_request:
-    branches: [ main ]
-  workflow_dispatch:
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref || github.ref }}
-  cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
-
-jobs:
-  build-container-image:
-    runs-on: [self-hosted, mi3008x]
-    timeout-minutes: 90
-
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v4
-
-      - name: Setup Apptainer (if not available)
-        run: |
-          if ! command -v apptainer &> /dev/null && ! command -v docker &> /dev/null; then
-            echo "Neither Apptainer nor Docker found, installing Apptainer..."
-            apt-get update && apt-get install -y software-properties-common
-            add-apt-repository -y ppa:apptainer/ppa
-            apt-get update && apt-get install -y apptainer
-          else
-            echo "Container runtime already available"
-          fi
-
-      - name: Build Iris container
-        run: |
-          # Use the universal container build script
-          bash .github/scripts/container_build.sh
-
-  external-validation-test:
-    name: External Validation Test
-    needs: build-container-image
-    runs-on: [self-hosted, mi3008x]
-    timeout-minutes: 30
-
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v4
-
-      - name: Cleanup lingering ports before tests
-        run: |
-          bash .github/scripts/cleanup_ports.sh
-
-      - name: Run External Validation Test
-        run: |
-          set -e
-
-          echo "::group::Running external validation test"
-          bash .github/scripts/container_exec.sh "
-            set -e
-            pip install git+https://github.com/${{ github.repository }}.git@${{ github.sha }}
-            wget -O test_iris_distributed.py https://gist.githubusercontent.com/mawad-amd/6375dc078e39e256828f379e03310ec7/raw/a527c3192bee4615292769e340b1c73676f6945a/test_iris_distributed.py
-            python test_iris_distributed.py
-          "
-          echo "::endgroup::"
-
-          echo "✅ External validation test passed!"
-
-  external-gluon-validation-test:
-    name: External Gluon Validation Test
-    needs: build-container-image
-    runs-on: [self-hosted, mi3008x]
-    timeout-minutes: 30
-
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v4
-
-      - name: Cleanup lingering ports before tests
-        run: |
-          bash .github/scripts/cleanup_ports.sh
-
-      - name: Run External Gluon Validation Test
-        run: |
-          set -e
-
-          echo "::group::Running external gluon validation test"
-          bash .github/scripts/container_exec.sh --gpus "0,1" "
-            set -e
-            pip install git+https://github.com/${{ github.repository }}.git@${{ github.sha }}
-            wget -O test_iris_gluon_distributed.py https://gist.githubusercontent.com/mawad-amd/2666dde8ebe2755eb0c4f2108709fcd5/raw/aa567ef3185c37a80d25bc9724ae9589548261b4/test_iris_gluon_distributed.py
-            python test_iris_gluon_distributed.py
-          "
-          echo "::endgroup::"
-
-          echo "✅ External gluon validation test passed!"
\ No newline at end of file
diff --git a/.github/workflows/iris-performance-regression-test.yml b/.github/workflows/iris-performance-regression-test.yml
deleted file mode 100644
index fa017886..00000000
--- a/.github/workflows/iris-performance-regression-test.yml
+++ /dev/null
@@ -1,88 +0,0 @@
-name: Iris Performance Regression Test
-
-on:
-  push:
-    branches: [ main ]
-  pull_request:
-    branches: [ main ]
-  workflow_dispatch:
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref || github.ref }}
-  cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
-
-jobs:
-  build-container-image:
-    runs-on: [self-hosted, mi3008x]
-    timeout-minutes: 20
-
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v4
-
-      - name: Setup Apptainer (if not available)
-        run: |
-          if ! command -v apptainer &> /dev/null && ! command -v docker &> /dev/null; then
-            echo "Neither Apptainer nor Docker found, installing Apptainer..."
-            apt-get update && apt-get install -y software-properties-common
-            add-apt-repository -y ppa:apptainer/ppa
-            apt-get update && apt-get install -y apptainer
-          else
-            echo "Container runtime already available"
-          fi
-
-      - name: Build Iris container
-        run: |
-          # Use the universal container build script
-          bash .github/scripts/container_build.sh
-
-  performance-test:
-    name: ${{ matrix.example_name }}
-    needs: build-container-image
-    runs-on: [self-hosted, mi3008x]
-    timeout-minutes: 30
-    strategy:
-      fail-fast: false
-      matrix:
-        # Performance baselines measured on AMD Instinct MI325X (8 GPUs)
-        include:
-          # Disabled https://github.com/ROCm/iris/issues/238
-          #- example_name: "GEMM All-Scatter WG Specialization"
-          #  example_path: "10_gemm_all_scatter_wg_specialization"
-          #  tflops_threshold: 1600  # Actual: ~2182 TFLOPs
-          #  benchmark_args: "-m 16384 -n 16384 -k 16384 --BLK_M 128 --BLK_N 128 --BLK_K 64 --gsize_m 6 --gemm_sms 256"
-
-          - example_name: "GEMM All-Scatter"
-            example_path: "07_gemm_all_scatter"
-            tflops_threshold: 1000  # Actual: ~1407 TFLOPs
-            benchmark_args: "-m 16384 -n 16384 -k 16384 --BLK_M 256 --BLK_N 64 --BLK_K 64 --gsize_m 6 --gemm_sms 256"
-
-          - example_name: "GEMM All-Scatter Producer-Consumer"
-            example_path: "11_gemm_all_scatter_producer_consumer"
-            tflops_threshold: 1600  # Actual: ~2190 TFLOPs
-            benchmark_args: "-m 16384 -n 16384 -k 16384 --BLK_M 128 --BLK_N 128 --BLK_K 64 --gsize_m 6 --gemm_sms 256 --comm_sms 48"
-
-          - example_name: "GEMM All-Scatter Bulk Synchronous"
-            example_path: "12_gemm_all_scatter_bulk_synchronous"
-            tflops_threshold: 900  # Actual: ~1262 TFLOPs
-            benchmark_args: "-m 16384 -n 16384 -k 16384 --BLK_M 128 --BLK_N 128 --BLK_K 64 --gsize_m 6 --gemm_sms 256"
-
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v4
-
-      - name: Cleanup lingering ports before tests
-        run: |
-          bash .github/scripts/cleanup_ports.sh
-
-      - name: Run ${{ matrix.example_name }} Benchmark (8 ranks)
-        run: |
-          set -e
-          
-          echo "::group::Running performance benchmark"
-          bash .github/scripts/run_perf_benchmark.sh \
-            "${{ matrix.example_path }}" \
-            "${{ matrix.tflops_threshold }}" \
-            ${{ matrix.benchmark_args }}
-          echo "::endgroup::"
-
diff --git a/.github/workflows/iris-pip-install-test.yml b/.github/workflows/iris-pip-install-test.yml
deleted file mode 100644
index 739af005..00000000
--- a/.github/workflows/iris-pip-install-test.yml
+++ /dev/null
@@ -1,167 +0,0 @@
-name: Iris Pip Install Test
-
-on:
-  push:
-    branches: [ main ]
-  pull_request:
-    branches: [ main ]
-  workflow_dispatch:
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref || github.ref }}
-  cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
-
-jobs:
-  build-container-image:
-    runs-on: [self-hosted, mi3008x]
-    timeout-minutes: 90
-
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v4
-
-      - name: Setup Apptainer (if not available)
-        run: |
-          if ! command -v apptainer &> /dev/null && ! command -v docker &> /dev/null; then
-            echo "Neither Apptainer nor Docker found, installing Apptainer..."
-            apt-get update && apt-get install -y software-properties-common
-            add-apt-repository -y ppa:apptainer/ppa
-            apt-get update && apt-get install -y apptainer
-          else
-            echo "Container runtime already available"
-          fi
-
-      - name: Build Iris container
-        run: |
-          # Use the universal container build script
-          bash .github/scripts/container_build.sh
-
-  test-1-2-4-ranks:
-    name: Pip Install Test 1/2/4 Ranks (Parallel)
-    needs: build-container-image
-    runs-on: [self-hosted, mi3008x]
-    timeout-minutes: 30
-
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-
-      - name: Cleanup lingering ports before tests
-        run: |
-          bash .github/scripts/cleanup_ports.sh
-
-      - name: Run pip install tests for 1, 2, 4 ranks in parallel
-        run: |
-          set -e
-
-          # Run tests in parallel with different GPU assignments
-          # Note: Each test gets 2+ GPUs even if it only uses some of them.
-          # This allows tests like test_empty_device_handling to verify that
-          # allocating on a different device correctly raises an error.
-
-          echo "::group::Starting parallel tests"
-          echo "Starting 1-rank test on GPUs 0,1..."
-          bash .github/scripts/container_exec.sh --gpus "0,1" "
-            set -e
-            pip install git+https://github.com/${{ github.repository }}.git@${{ github.sha }}
-            pip install -e .
-            for test_file in tests/examples/test_*.py; do
-              echo \"Testing: \$test_file with 1 ranks\"
-              python tests/run_tests_distributed.py --num_ranks 1 \"\$test_file\" -v --tb=short --durations=10
-            done
-            for test_file in tests/unittests/test_*.py; do
-              echo \"Testing: \$test_file with 1 ranks\"
-              python tests/run_tests_distributed.py --num_ranks 1 \"\$test_file\" -v --tb=short --durations=10
-            done
-          " &
-          PID1=$!
-
-          echo "Starting 2-rank test on GPUs 2,3..."
-          bash .github/scripts/container_exec.sh --gpus "2,3" "
-            set -e
-            pip install git+https://github.com/${{ github.repository }}.git@${{ github.sha }}
-            pip install -e .
-            for test_file in tests/examples/test_*.py; do
-              echo \"Testing: \$test_file with 2 ranks\"
-              python tests/run_tests_distributed.py --num_ranks 2 \"\$test_file\" -v --tb=short --durations=10
-            done
-            for test_file in tests/unittests/test_*.py; do
-              echo \"Testing: \$test_file with 2 ranks\"
-              python tests/run_tests_distributed.py --num_ranks 2 \"\$test_file\" -v --tb=short --durations=10
-            done
-          " &
-          PID2=$!
-
-          echo "Starting 4-rank test on GPUs 4,5,6,7..."
-          bash .github/scripts/container_exec.sh --gpus "4,5,6,7" "
-            set -e
-            pip install git+https://github.com/${{ github.repository }}.git@${{ github.sha }}
-            pip install -e .
-            for test_file in tests/examples/test_*.py; do
-              echo \"Testing: \$test_file with 4 ranks\"
-              python tests/run_tests_distributed.py --num_ranks 4 \"\$test_file\" -v --tb=short --durations=10
-            done
-            for test_file in tests/unittests/test_*.py; do
-              echo \"Testing: \$test_file with 4 ranks\"
-              python tests/run_tests_distributed.py --num_ranks 4 \"\$test_file\" -v --tb=short --durations=10
-            done
-          " &
-          PID4=$!
-          echo "::endgroup::"
-
-          # Wait for all parallel tests and track failures
-          echo "::group::Waiting for parallel tests to complete"
-          FAIL=0
-          FAILED_TESTS=""
-
-          wait $PID1 || { echo "::error::1-rank test FAILED"; FAILED_TESTS="$FAILED_TESTS 1-rank"; FAIL=1; }
-          wait $PID2 || { echo "::error::2-rank test FAILED"; FAILED_TESTS="$FAILED_TESTS 2-rank"; FAIL=1; }
-          wait $PID4 || { echo "::error::4-rank test FAILED"; FAILED_TESTS="$FAILED_TESTS 4-rank"; FAIL=1; }
-          echo "::endgroup::"
-
-          if [ $FAIL -eq 1 ]; then
-            echo "::error::Parallel tests failed:$FAILED_TESTS"
-            exit 1
-          fi
-
-          echo "✅ All parallel tests (1, 2, 4 ranks) passed!"
-
-  test-8-ranks:
-    name: Pip Install Test 8 Ranks
-    needs: build-container-image
-    runs-on: [self-hosted, mi3008x]
-    timeout-minutes: 30
-
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-
-      - name: Cleanup lingering ports before tests
-        run: |
-          bash .github/scripts/cleanup_ports.sh
-
-      - name: Run 8-rank pip install test
-        run: |
-          set -e
-
-          echo "::group::Running 8-rank test on all GPUs"
-          bash .github/scripts/container_exec.sh --gpus "0,1,2,3,4,5,6,7" "
-            set -e
-            pip install git+https://github.com/${{ github.repository }}.git@${{ github.sha }}
-            pip install -e .
-            for test_file in tests/examples/test_*.py; do
-              echo \"Testing: \$test_file with 8 ranks\"
-              python tests/run_tests_distributed.py --num_ranks 8 \"\$test_file\" -v --tb=short --durations=10
-            done
-            for test_file in tests/unittests/test_*.py; do
-              echo \"Testing: \$test_file with 8 ranks\"
-              python tests/run_tests_distributed.py --num_ranks 8 \"\$test_file\" -v --tb=short --durations=10
-            done
-          "
-          echo "::endgroup::"
-
-          echo "✅ 8-rank test passed!"
diff --git a/.github/workflows/iris-tests-apptainer.yml b/.github/workflows/iris-tests-apptainer.yml
deleted file mode 100644
index 63950f95..00000000
--- a/.github/workflows/iris-tests-apptainer.yml
+++ /dev/null
@@ -1,114 +0,0 @@
-name: Iris Tests
-
-on:
-  push:
-    branches: [ main ]
-  pull_request:
-    branches: [ main ]
-  workflow_dispatch:
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref || github.ref }}
-  cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
-
-jobs:
-  build-container-image:
-    runs-on: [self-hosted, mi3008x]
-    timeout-minutes: 90
-
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v4
-
-      - name: Setup Apptainer (if not available)
-        run: |
-          if ! command -v apptainer &> /dev/null && ! command -v docker &> /dev/null; then
-            echo "Neither Apptainer nor Docker found, installing Apptainer..."
-            apt-get update && apt-get install -y software-properties-common
-            add-apt-repository -y ppa:apptainer/ppa
-            apt-get update && apt-get install -y apptainer
-          else
-            echo "Container runtime already available"
-          fi
-
-      - name: Build Iris container
-        run: |
-          # Use the universal container build script
-          bash .github/scripts/container_build.sh
-  test-1-2-4-ranks:
-    name: Test 1/2/4 Ranks (Parallel)
-    needs: build-container-image
-    runs-on: [self-hosted, mi3008x]
-    timeout-minutes: 20
-
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v4
-
-      - name: Cleanup lingering ports before tests
-        run: |
-          bash .github/scripts/cleanup_ports.sh
-
-      - name: Run 1, 2, 4 rank tests in parallel
-        run: |
-          set -e
-
-          # Run tests in parallel with different GPU assignments
-          # Note: Each test gets 2+ GPUs even if it only uses some of them.
-          # This allows tests like test_empty_device_handling to verify that
-          # allocating on a different device correctly raises an error.
-
-          echo "::group::Starting parallel tests"
-          echo "Starting 1-rank test on GPUs 0,1..."
-          bash .github/scripts/run_tests.sh 1 "0,1" &
-          PID1=$!
-
-          echo "Starting 2-rank test on GPUs 2,3..."
-          bash .github/scripts/run_tests.sh 2 "2,3" &
-          PID2=$!
-
-          echo "Starting 4-rank test on GPUs 4,5,6,7..."
-          bash .github/scripts/run_tests.sh 4 "4,5,6,7" &
-          PID4=$!
-          echo "::endgroup::"
-
-          # Wait for all parallel tests and track failures
-          echo "::group::Waiting for parallel tests to complete"
-          FAIL=0
-          FAILED_TESTS=""
-
-          wait $PID1 || { echo "::error::1-rank test FAILED"; FAILED_TESTS="$FAILED_TESTS 1-rank"; FAIL=1; }
-          wait $PID2 || { echo "::error::2-rank test FAILED"; FAILED_TESTS="$FAILED_TESTS 2-rank"; FAIL=1; }
-          wait $PID4 || { echo "::error::4-rank test FAILED"; FAILED_TESTS="$FAILED_TESTS 4-rank"; FAIL=1; }
-          echo "::endgroup::"
-
-          if [ $FAIL -eq 1 ]; then
-            echo "::error::Parallel tests failed:$FAILED_TESTS"
-            exit 1
-          fi
-
-          echo "✅ All parallel tests (1, 2, 4 ranks) passed!"
-
-  test-8-ranks:
-    name: Test 8 Ranks
-    needs: build-container-image
-    runs-on: [self-hosted, mi3008x]
-    timeout-minutes: 15
-
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v4
-
-      - name: Cleanup lingering ports before tests
-        run: |
-          bash .github/scripts/cleanup_ports.sh
-
-      - name: Run 8-rank test
-        run: |
-          set -e
-
-          echo "::group::Running 8-rank test on all GPUs"
-          bash .github/scripts/run_tests.sh 8 "0,1,2,3,4,5,6,7"
-          echo "::endgroup::"
-
-          echo "✅ 8-rank test passed!"
diff --git a/.github/workflows/lint.yml b/.github/workflows/lint.yml
deleted file mode 100644
index 2b6d1dbb..00000000
--- a/.github/workflows/lint.yml
+++ /dev/null
@@ -1,37 +0,0 @@
-# SPDX-License-Identifier: MIT
-# Copyright (c) 2025 Advanced Micro Devices, Inc. All rights reserved.
-
-name: Lint & Auto-Fix
-
-on: [push, pull_request]
-
-jobs:
-  lint:
-    runs-on: ubuntu-latest
-
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v3
-
-      - name: Set up Python
-        uses: actions/setup-python@v4
-        with:
-          python-version: "3.10"
-
-      - name: Install Ruff
-        run: pip install ruff
-
-      - name: Run Ruff and auto-fix issues
-        run: ruff check . --fix
-
-      - name: Run Ruff formatter
-        run: ruff format .
-
-      - name: Commit and push fixes
-        run: |
-          git config --global user.name "github-actions[bot]"
-          git config --global user.email "github-actions[bot]@users.noreply.github.com"
-          git add .
-          git commit -m "Apply Ruff auto-fixes" || echo "No changes to commit"
-          git push
-        continue-on-error: true
\ No newline at end of file
diff --git a/docker/Dockerfile.rocm6.4 b/docker/Dockerfile.rocm6.4
new file mode 100644
index 00000000..bf201dcc
--- /dev/null
+++ b/docker/Dockerfile.rocm6.4
@@ -0,0 +1,40 @@
+# SPDX-License-Identifier: MIT
+# Copyright (c) 2025 Advanced Micro Devices, Inc. All rights reserved.
+
+FROM rocm/pytorch:rocm6.4.4_ubuntu24.04_py3.12_pytorch_release_2.7.1
+
+# Use bash shell for RUN commands
+SHELL ["/bin/bash", "-c"]
+
+# Set environment variables
+ENV TRITON_PATH=/opt/triton \
+    ROCM_PATH=/opt/rocm \
+    OMPI_MCA_mtl="^ofi" \
+    OMPI_MCA_pml="ob1"
+
+ENV LD_LIBRARY_PATH=$ROCM_PATH/lib:$LD_LIBRARY_PATH \
+    PATH="$ROCM_PATH/bin:$PATH"
+
+ENV OMPI_ALLOW_RUN_AS_ROOT_CONFIRM=1 \
+    OMPI_ALLOW_RUN_AS_ROOT=1
+
+# Install system packages
+RUN apt-get update && \
+    DEBIAN_FRONTEND=noninteractive apt-get install -y \
+    git wget ninja-build cmake python3-pip python3-dev build-essential && \
+    rm -rf /var/lib/apt/lists/*
+
+# Install Python packages with pip
+RUN pip3 install --upgrade pip && \
+    pip3 install wheel jupyter
+
+# Clone and install Triton
+ARG TRITON_COMMIT=aafec417bded34db6308f5b3d6023daefae43905
+WORKDIR $TRITON_PATH
+RUN git clone https://github.com/triton-lang/triton.git $TRITON_PATH
+RUN git checkout ${TRITON_COMMIT}
+RUN pip3 install -e .
+ENV PYTHONPATH=$TRITON_PATH
+
+WORKDIR /workspace
+
diff --git a/docker/Dockerfile.rocm7.0 b/docker/Dockerfile.rocm7.0
new file mode 100644
index 00000000..b679d14c
--- /dev/null
+++ b/docker/Dockerfile.rocm7.0
@@ -0,0 +1,43 @@
+# SPDX-License-Identifier: MIT
+# Copyright (c) 2025 Advanced Micro Devices, Inc. All rights reserved.
+
+FROM rocm/pytorch:rocm7.0.2_ubuntu24.04_py3.12_pytorch_release_2.8.0
+
+# Use bash shell for RUN commands
+SHELL ["/bin/bash", "-c"]
+
+# Set environment variables
+ENV TRITON_PATH=/opt/triton \
+    ROCM_PATH=/opt/rocm \
+    OMPI_MCA_mtl="^ofi" \
+    OMPI_MCA_pml="ob1"
+
+ENV LD_LIBRARY_PATH=$ROCM_PATH/lib:$LD_LIBRARY_PATH \
+    PATH="$ROCM_PATH/bin:$PATH"
+
+ENV OMPI_ALLOW_RUN_AS_ROOT_CONFIRM=1 \
+    OMPI_ALLOW_RUN_AS_ROOT=1
+
+# Install system packages
+RUN apt-get update && \
+    DEBIAN_FRONTEND=noninteractive apt-get install -y \
+    git wget ninja-build cmake python3-pip python3-dev build-essential && \
+    rm -rf /var/lib/apt/lists/*
+
+# Install Python packages with pip
+RUN pip3 install --upgrade pip && \
+    pip3 install wheel jupyter
+
+# Clone and install Triton
+ARG TRITON_COMMIT=aafec417bded34db6308f5b3d6023daefae43905
+WORKDIR $TRITON_PATH
+RUN git clone https://github.com/triton-lang/triton.git $TRITON_PATH
+RUN git checkout ${TRITON_COMMIT}
+RUN pip3 install -e .
+ENV PYTHONPATH=$TRITON_PATH
+
+WORKDIR /workspace
+
+# Make Python site-packages writable by all users for development
+RUN chmod -R a+w /opt/venv/lib/python3.12/site-packages/
+
diff --git a/docker/Dockerfile.rocm7.1 b/docker/Dockerfile.rocm7.1
new file mode 100644
index 00000000..9c6f7666
--- /dev/null
+++ b/docker/Dockerfile.rocm7.1
@@ -0,0 +1,43 @@
+# SPDX-License-Identifier: MIT
+# Copyright (c) 2025 Advanced Micro Devices, Inc. All rights reserved.
+
+FROM rocm/pytorch:rocm7.1_ubuntu24.04_py3.12_pytorch_release_2.8.0
+
+# Use bash shell for RUN commands
+SHELL ["/bin/bash", "-c"]
+
+# Set environment variables
+ENV TRITON_PATH=/opt/triton \
+    ROCM_PATH=/opt/rocm \
+    OMPI_MCA_mtl="^ofi" \
+    OMPI_MCA_pml="ob1"
+
+ENV LD_LIBRARY_PATH=$ROCM_PATH/lib:$LD_LIBRARY_PATH \
+    PATH="$ROCM_PATH/bin:$PATH"
+
+ENV OMPI_ALLOW_RUN_AS_ROOT_CONFIRM=1 \
+    OMPI_ALLOW_RUN_AS_ROOT=1
+
+# Install system packages
+RUN apt-get update && \
+    DEBIAN_FRONTEND=noninteractive apt-get install -y \
+    git wget ninja-build cmake python3-pip python3-dev build-essential && \
+    rm -rf /var/lib/apt/lists/*
+
+# Install Python packages with pip
+RUN pip3 install --upgrade pip && \
+    pip3 install wheel jupyter
+
+# Clone and install Triton
+ARG TRITON_COMMIT=aafec417bded34db6308f5b3d6023daefae43905
+WORKDIR $TRITON_PATH
+RUN git clone https://github.com/triton-lang/triton.git $TRITON_PATH
+RUN git checkout ${TRITON_COMMIT}
+RUN pip3 install -e .
+ENV PYTHONPATH=$TRITON_PATH
+
+WORKDIR /workspace
+
+# Make Python site-packages writable by all users for development
+RUN chmod -R a+w /opt/venv/lib/python3.12/site-packages/
+
diff --git a/pyproject.toml b/pyproject.toml
index cc8757f5..33e7f6b6 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -20,7 +20,6 @@ requires-python = ">=3.8"
 dependencies = [
     "numpy",
     "requests",
-    "ruff",
 ]
 
 [project.urls]
@@ -33,6 +32,7 @@ dev = [
     "pytest",
     "black",
     "mypy",
+    "ruff",
 ]
 
 [tool.setuptools]