✨ feat: add model-downloader service with init container integration

Signed-off-by: samzong <[email protected]>

Author: samzong

.github/workflows/docker-publish.yml

@@ -39,7 +39,7 @@ jobs:
       packages: write
     strategy:
       matrix:
-        image: [extproc, llm-katan, dashboard]
+        image: [extproc, llm-katan, dashboard, model-downloader]
         # Multi-architecture build strategy:
         # - AMD64: Native build on ubuntu-latest (fast)
         # - ARM64: Cross-compilation on ubuntu-latest (faster than emulation)
@@ -171,6 +171,10 @@ jobs:
             echo "context=." >> $GITHUB_OUTPUT
             echo "dockerfile=./dashboard/backend/Dockerfile" >> $GITHUB_OUTPUT
             echo "platform=linux/${{ matrix.arch }}" >> $GITHUB_OUTPUT
+          elif [ "${{ matrix.image }}" = "model-downloader" ]; then
+            echo "context=." >> $GITHUB_OUTPUT
+            echo "dockerfile=./Dockerfile.model-downloader" >> $GITHUB_OUTPUT
+            echo "platform=linux/${{ matrix.arch }}" >> $GITHUB_OUTPUT
           fi
 
       - name: Generate tags
@@ -236,7 +240,7 @@ jobs:
       packages: write
     strategy:
       matrix:
-        image: [extproc, llm-katan, dashboard]
+        image: [extproc, llm-katan, dashboard, model-downloader]
 
     steps:
       - name: Log in to GitHub Container Registry

Dockerfile.model-downloader

@@ -0,0 +1,24 @@
+FROM python:3.11-slim
+
+# Set environment variables
+ENV PIP_DISABLE_PIP_VERSION_CHECK=1 \
+    PIP_NO_CACHE_DIR=1 \
+    PYTHONUNBUFFERED=1
+
+# Add OCI labels
+LABEL org.opencontainers.image.source="https://github.com/vllm-project/semantic-router" \
+      org.opencontainers.image.description="Lightweight model downloader for Semantic Router with huggingface_hub pre-installed"
+
+WORKDIR /app
+
+# Install minimal dependencies in a single layer
+RUN apt-get update && \
+    apt-get install -y --no-install-recommends \
+        ca-certificates \
+    && rm -rf /var/lib/apt/lists/* /var/cache/apt/archives/* \
+    && pip install --no-cache-dir "huggingface_hub>=0.19.0" \
+    && find /usr/local -type d -name '__pycache__' -exec rm -rf {} + 2>/dev/null || true \
+    && find /usr/local -type f -name '*.pyc' -delete
+
+# Set default entrypoint to bash to allow script execution from arguments
+ENTRYPOINT ["/bin/bash", "-c"]

deploy/helm/semantic-router/templates/deployment.yaml

@@ -32,15 +32,14 @@ spec:
       {{- if .Values.initContainer.enabled }}
       initContainers:
       - name: model-downloader
-        image: {{ .Values.initContainer.image }}
+        image: "{{ .Values.initContainer.image.repository }}:{{ .Values.initContainer.image.tag }}"
+        imagePullPolicy: {{ .Values.initContainer.image.pullPolicy | default "IfNotPresent" }}
         securityContext:
           {{- toYaml .Values.securityContext | nindent 10 }}
         command: ["/bin/bash", "-c"]
         args:
         - |
           set -e
-          echo "Installing Hugging Face Hub..."
-          pip install -U --no-cache-dir "huggingface_hub>=0.19.0"
 
           echo "Downloading models to persistent volume..."
           cd /app/models

deploy/helm/semantic-router/values.yaml

@@ -131,7 +131,10 @@ initContainer:
   # -- Enable init container
   enabled: true
   # -- Init container image
-  image: python:3.11-slim
+  image:
+    repository: ghcr.io/vllm-project/semantic-router/model-downloader
+    tag: latest
+    pullPolicy: IfNotPresent
   # -- Resource limits for init container
   resources:
     limits: