Merge docker changes into main

.github/workflows/docker-image.yml

@@ -2,18 +2,135 @@ name: Docker Image CI
 
 on:
   push:
-    branches: [ "main" ]
+    branches: ["docker"]
+  release:
+    types: [published]
+
+permissions:
+  contents: read
+  packages: write
 
 jobs:
-  build:
+  build-base:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          submodules: recursive
+
+      - name: Log in to GitHub Container Registry
+        uses: docker/login-action@v3
+        with:
+          registry: ghcr.io
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Build and push dev-base
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          file: docker/Dockerfile.dev
+          target: dev-base
+          platforms: linux/amd64
+          push: true
+          tags: ghcr.io/collaborativebioinformatics/oncolearn:dev-base
+          cache-from: type=gha,scope=dev-base
+          cache-to: type=gha,mode=max,scope=dev-base
+
+      - name: Build and push prod-base
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          file: docker/Dockerfile.prod
+          target: pytorch-prod-base
+          platforms: linux/amd64
+          push: true
+          tags: ghcr.io/collaborativebioinformatics/oncolearn:prod-base
+          cache-from: type=gha,scope=prod-base
+          cache-to: type=gha,mode=max,scope=prod-base
+
+  build-gpu:
+    needs: build-base
     runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        include:
+          - stage: dev
+            gpu_extra: cpu
+            image_tag: dev-cpu
+            rocm_wsl: "0"
+          - stage: dev
+            gpu_extra: cu130
+            image_tag: dev-cuda
+            rocm_wsl: "0"
+          - stage: dev
+            gpu_extra: rocm
+            image_tag: dev-rocm
+            rocm_wsl: "0"
+          - stage: prod
+            gpu_extra: cpu
+            image_tag: prod-cpu
+            rocm_wsl: "0"
+          - stage: prod
+            gpu_extra: cu130
+            image_tag: prod-cuda
+            rocm_wsl: "0"
+          - stage: prod
+            gpu_extra: rocm
+            image_tag: prod-rocm
+            rocm_wsl: "0"
+          - stage: prod
+            gpu_extra: rocm
+            image_tag: prod-rocm-wsl
+            rocm_wsl: "1"
 
     steps:
-    - uses: actions/checkout@v4
-
-    - name: Build the Docker image (amd-wsl)
-      run: |
-        docker build . \
-          --file Dockerfile \
-          --build-arg GPU_EXTRA=rocm62 \
-          --tag my-image-name:amd-wsl-$(date +%s)
+      - uses: actions/checkout@v4
+        with:
+          submodules: recursive
+
+      - name: Log in to GitHub Container Registry
+        uses: docker/login-action@v3
+        with:
+          registry: ghcr.io
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Set image variables
+        id: vars
+        run: |
+          if [ "${{ matrix.stage }}" = "dev" ]; then
+            echo "dockerfile=docker/Dockerfile.dev" >> "$GITHUB_OUTPUT"
+            echo "target=dev" >> "$GITHUB_OUTPUT"
+            echo "base_scope=dev-base" >> "$GITHUB_OUTPUT"
+            echo "base_image=ghcr.io/collaborativebioinformatics/oncolearn:dev-base" >> "$GITHUB_OUTPUT"
+          else
+            echo "dockerfile=docker/Dockerfile.prod" >> "$GITHUB_OUTPUT"
+            echo "target=pytorch-prod" >> "$GITHUB_OUTPUT"
+            echo "base_scope=prod-base" >> "$GITHUB_OUTPUT"
+            echo "base_image=ghcr.io/collaborativebioinformatics/oncolearn:prod-base" >> "$GITHUB_OUTPUT"
+          fi
+
+      - name: Build and push GPU variant
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          file: ${{ steps.vars.outputs.dockerfile }}
+          target: ${{ steps.vars.outputs.target }}
+          platforms: linux/amd64
+          push: true
+          tags: ghcr.io/collaborativebioinformatics/oncolearn:${{ matrix.image_tag }}
+          build-args: |
+            GPU_EXTRA=${{ matrix.gpu_extra }}
+            ROCM_WSL=${{ matrix.rocm_wsl }}
+          cache-from: |
+            type=gha,scope=${{ matrix.image_tag }}
+            type=gha,scope=${{ steps.vars.outputs.base_scope }}
+            type=registry,ref=${{ steps.vars.outputs.base_image }}
+          cache-to: type=gha,mode=max,scope=${{ matrix.image_tag }}

docker-compose.yml

@@ -1,6 +1,5 @@
 # Shared configuration using YAML anchors
 x-common-config: &common-config
-  env_file: .docker.env
   volumes:
     - .:/workspace
     - uv-cache:/root/.cache/uv
@@ -21,9 +20,10 @@ x-prod-build: &prod-build
   dockerfile: docker/Dockerfile.prod
 
 services:
-  # Main development environment with Python + R (CPU only)
+  # Development environment — CPU only
   dev-cpu:
     <<: *common-config
+    image: ghcr.io/collaborativebioinformatics/oncolearn:dev-cpu
     profiles: ["dev-cpu"]
     build:
       <<: *dev-build
@@ -32,50 +32,54 @@ services:
       target: dev
     container_name: oncolearn-dev-cpu
 
-  # Main development environment with Python + R (NVIDIA GPU)
-  dev-nvidia:
+  # Development environment — NVIDIA GPU
+  dev-cuda:
     <<: *common-config
-    profiles: ["dev-nvidia"]
+    image: ghcr.io/collaborativebioinformatics/oncolearn:dev-cuda
+    profiles: ["dev-cuda"]
     build:
       <<: *dev-build
       args:
         GPU_EXTRA: cu130
       target: dev
-    container_name: oncolearn-dev-nvidia
+    container_name: oncolearn-dev-cuda
     deploy:
       resources:
         reservations:
           devices:
-            - driver: nvidia
+            - driver: cuda
               count: all
               capabilities: [gpu]
 
-  # Main development environment with Python + R (AMD GPU - Native Linux)
-  dev-amd:
+  # Development environment — AMD GPU (native Linux)
+  dev-rocm:
     <<: *common-config
-    profiles: ["dev-amd"]
+    image: ghcr.io/collaborativebioinformatics/oncolearn:dev-rocm
+    profiles: ["dev-rocm"]
     build:
       <<: *dev-build
       args:
         GPU_EXTRA: rocm
       target: dev
-    container_name: oncolearn-dev-amd
+    container_name: oncolearn-dev-rocm
     devices:
       - /dev/kfd
       - /dev/dri
     group_add:
       - video
 
-  # Main development environment with Python + R (AMD GPU - WSL2)
-  dev-amd-wsl:
+  # Development environment — AMD GPU (WSL2)
+  # Uses the same image as dev-rocm; WSL2 differences are runtime-only.
+  dev-rocm-wsl:
     <<: *common-config
-    profiles: ["dev-amd-wsl"]
+    image: ghcr.io/collaborativebioinformatics/oncolearn:dev-rocm
+    profiles: ["dev-rocm-wsl"]
     build:
       <<: *dev-build
       args:
         GPU_EXTRA: rocm
       target: dev
-    container_name: oncolearn-dev-amd-wsl
+    container_name: oncolearn-dev-rocm-wsl
     volumes:
       - .:/workspace
       - uv-cache:/root/.cache/uv
@@ -99,52 +103,86 @@ services:
     ipc: host  # required for ROCm shared memory
     shm_size: '2gb'
 
-  # Production PyTorch environment (CPU only)
-  pytorch-prod-cpu:
-    env_file: .docker.env
+  # Production image — CPU only
+  prod-cpu:
+    image: ghcr.io/collaborativebioinformatics/oncolearn:prod-cpu
     profiles: ["prod-cpu"]
     build:
       <<: *prod-build
       args:
         GPU_EXTRA: cpu
       target: pytorch-prod
-    container_name: oncolearn-pytorch-prod-cpu
+    container_name: oncolearn-prod-cpu
     volumes:
       - ./data:/workspace/data:ro
       - ./data/configs:/workspace/data/configs
 
-  # Production PyTorch environment (NVIDIA GPU - Optimized Size)
-  pytorch-prod-nvidia:
-    env_file: .docker.env
-    profiles: ["prod-nvidia"]
+  # Production image — NVIDIA GPU
+  prod-cuda:
+    image: ghcr.io/collaborativebioinformatics/oncolearn:prod-cuda
+    profiles: ["prod-cuda"]
     build:
       <<: *prod-build
       args:
         GPU_EXTRA: cu130
       target: pytorch-prod
-    container_name: oncolearn-pytorch-prod-nvidia
+    container_name: oncolearn-prod-cuda
     volumes:
       - ./data:/workspace/data:ro
       - ./data/configs:/workspace/data/configs
     deploy:
       resources:
         reservations:
           devices:
-            - driver: nvidia
+            - driver: cuda
               count: all
               capabilities: [gpu]
 
-  # Production PyTorch environment (AMD GPU - WSL2)
-  pytorch-prod-amd-wsl:
-    env_file: .docker.env
-    profiles: ["prod-amd"]
+  # Production image — AMD GPU (native Linux)
+  prod-rocm:
+    image: ghcr.io/collaborativebioinformatics/oncolearn:prod-rocm
+    profiles: ["prod-rocm"]
+    build:
+      <<: *prod-build
+      args:
+        GPU_EXTRA: rocm
+        ROCM_WSL: "0"
+      target: pytorch-prod
+    container_name: oncolearn-prod-rocm
+    volumes:
+      - ./data:/workspace/data:ro
+      - ./data/configs:/workspace/data/configs
+      - ./models:/workspace/models
+      - ./outputs:/workspace/outputs
+      - ./.hf-cache:/root/.cache/huggingface
+      - ./src:/workspace/src
+      - /opt/rocm/lib:/opt/rocm/lib:ro  # ROCm runtime libs
+    environment:
+      - LD_LIBRARY_PATH=/opt/rocm/lib
+    devices:
+      - /dev/kfd
+      - /dev/dri
+    group_add:
+      - video
+    cap_add:
+      - SYS_PTRACE
+    security_opt:
+      - seccomp:unconfined
+    ipc: host
+    shm_size: '2gb'
+
+  # Production image — AMD GPU (WSL2)
+  # Uses a separate image with ROCm 7.2 LW wheels built for /dev/dxg.
+  prod-rocm-wsl:
+    image: ghcr.io/collaborativebioinformatics/oncolearn:prod-rocm-wsl
+    profiles: ["prod-rocm-wsl"]
     build:
       <<: *prod-build
       args:
         GPU_EXTRA: rocm
         ROCM_WSL: "1"
       target: pytorch-prod
-    container_name: oncolearn-pytorch-prod-amd
+    container_name: oncolearn-prod-rocm-wsl
     mem_limit: 20g        # cap below WSL limit so OOM kills container, not WSL
     memswap_limit: 22g    # allow some swap headroom during model loading
     volumes: