build: add new vLLM CUDA image

Signed-off-by: Dorin Geman <[email protected]>

Author: doringeman

.github/workflows/release.yml

@@ -19,6 +19,11 @@ on:
         required: false
         type: string
         default: "latest"
+      vllmVersion:
+        description: 'vLLM version'
+        required: false
+        type: string
+        default: "latest"
 
 jobs:
   test:
@@ -59,6 +64,12 @@ jobs:
             echo "docker/model-runner:latest-cuda" >> "$GITHUB_OUTPUT"
           fi
           echo 'EOF' >> "$GITHUB_OUTPUT"
+          echo "vllm-cuda<<EOF" >> "$GITHUB_OUTPUT"
+          echo "docker/model-runner:${{ inputs.releaseTag }}-vllm-cuda" >> "$GITHUB_OUTPUT"
+          if [ "${{ inputs.pushLatest }}" == "true" ]; then
+            echo "docker/model-runner:latest-vllm-cuda" >> "$GITHUB_OUTPUT"
+          fi
+          echo 'EOF' >> "$GITHUB_OUTPUT"
 
       - name: Log in to DockerHub
         uses: docker/login-action@v3
@@ -99,3 +110,19 @@ jobs:
           sbom: true
           provenance: mode=max
           tags: ${{ steps.tags.outputs.cuda }}
+
+      - name: Build vLLM CUDA image
+        uses: docker/build-push-action@v5
+        with:
+          file: Dockerfile
+          target: vllm
+          platforms: linux/amd64
+          build-args: |
+            "LLAMA_SERVER_VERSION=${{ inputs.llamaServerVersion }}"
+            "LLAMA_SERVER_VARIANT=cuda"
+            "BASE_IMAGE=nvidia/cuda:12.9.0-runtime-ubuntu24.04"
+            "VLLM_VERSION=${{ inputs.vllmVersion }}""
+          push: true
+          sbom: true
+          provenance: mode=max
+          tags: ${{ steps.tags.outputs.vllm-cuda }}

Dockerfile

@@ -77,3 +77,18 @@ ENV LD_LIBRARY_PATH=/app/lib
 LABEL com.docker.desktop.service="model-runner"
 
 ENTRYPOINT ["/app/model-runner"]
+
+# --- vLLM variant ---
+FROM final AS vllm
+
+ARG VLLM_VERSION
+
+USER root
+
+# Install uv
+RUN curl -LsSf https://astral.sh/uv/install.sh | sh
+
+USER modelrunner
+
+# Install vLLM using uv
+RUN uv pip install vllm==${VLLM_VERSION}