Switch to pre-built vLLM wheels for CPU builds

.dockerignore

@@ -18,4 +18,4 @@ venv
 __pycache__
 
 # Docker files
-Dockerfile
+Dockerfile

.github/workflows/ci-pr-checks.yaml

@@ -23,6 +23,9 @@ jobs:
             src:
               - '**/*.go'
               - '**/*.py'
+              - Dockerfile.epp
+              - Dockerfile.sidecar
+              - Makefile*
   lint-and-test:
     needs: check-changes
     if: ${{ needs.check-changes.outputs.src == 'true' }}
@@ -62,8 +65,7 @@ jobs:
       - name: Install dependencies
         run: |
           go mod tidy
-          sudo -E env "PATH=$PATH" make install-dependencies
-          sudo -E env "PATH=$PATH" make install-python-deps
+          sudo -E env "PATH=$PATH" make install-dependencies install-python-deps
 
       - name: Run lint checks
         uses: golangci/golangci-lint-action@v9

Dockerfile.epp

@@ -18,17 +18,27 @@ RUN go mod download
 # Extract version dynamically and copy to a known location
 RUN KV_CACHE_PKG=$(go list -m -f '{{.Dir}}' github.com/llm-d/llm-d-kv-cache) && \
     mkdir -p /workspace/kv-cache && \
-    cp -r $KV_CACHE_PKG/* /workspace/kv-cache && \
-    chmod +x /workspace/kv-cache/pkg/preprocessing/chat_completions/setup.sh
+    cp -r $KV_CACHE_PKG/* /workspace/kv-cache
 
 FROM python:3.12-slim AS python-builder
 
-RUN apt-get update && apt-get install -y --no-install-recommends build-essential
+ARG TARGETARCH
 
 COPY --from=go-deps /workspace/kv-cache /workspace/kv-cache
 WORKDIR /workspace/kv-cache
-# llm-d-kv-cache's Makefile. not llm-d-inference-scheduler's
-RUN KV_CACHE_PKG=/workspace/kv-cache make install-python-deps
+
+# Create venv and install vLLM based on architecture using pre-built wheels
+RUN python3.12 -m venv /workspace/kv-cache/build/venv && \
+    . /workspace/kv-cache/build/venv/bin/activate && \
+    pip install --upgrade pip && \
+    VLLM_VERSION="0.14.0" && \
+    if [ "$TARGETARCH" = "arm64" ]; then \
+        pip install https://github.com/vllm-project/vllm/releases/download/v${VLLM_VERSION}/vllm-${VLLM_VERSION}+cpu-cp38-abi3-manylinux_2_35_aarch64.whl; \
+    elif [ "$TARGETARCH" = "amd64" ]; then \
+        pip install https://github.com/vllm-project/vllm/releases/download/v${VLLM_VERSION}/vllm-${VLLM_VERSION}+cpu-cp38-abi3-manylinux_2_35_x86_64.whl --extra-index-url https://download.pytorch.org/whl/cpu; \
+    else \
+        echo "ERROR: Unsupported architecture: $TARGETARCH. Only arm64 and amd64 are supported." && exit 1; \
+    fi
 
 # Go build stage
 FROM quay.io/projectquay/golang:1.24 AS go-builder

Makefile.tools.mk

@@ -18,6 +18,7 @@ GINKGO_VERSION ?= v2.27.2
 GOLANGCI_LINT_VERSION ?= v2.1.6
 KUSTOMIZE_VERSION ?= v5.5.0
 TYPOS_VERSION ?= v1.34.0
+VLLM_VERSION ?= 0.14.0
 
 ## Python Configuration
 PYTHON_VERSION ?= 3.12
@@ -194,14 +195,30 @@ install-python-deps: setup-venv ## installs dependencies.
 	@printf "\033[33;1m==== Setting up Python virtual environment in $(VENV_DIR) ====\033[0m\n"
 	@echo "install vllm..."
 	@KV_CACHE_PKG=$${KV_CACHE_PKG:-$$(go list -m -f '{{.Dir}}' github.com/llm-d/llm-d-kv-cache 2>/dev/null)}; \
-	if [ -n "$$KV_CACHE_PKG" ] && [ -f "$$KV_CACHE_PKG/pkg/preprocessing/chat_completions/setup.sh" ]; then \
-		echo "Running kv-cache setup script..."; \
-		cp "$$KV_CACHE_PKG/pkg/preprocessing/chat_completions/setup.sh" build/kv-cache-setup.sh; \
-		chmod +x build/kv-cache-setup.sh; \
-		cd build && PATH=$(VENV_BIN):$$PATH ./kv-cache-setup.sh && cd ..; \
-	else \
-		echo "ERROR: kv-cache package not found or setup script missing."; \
+	if [ -z "$$KV_CACHE_PKG" ]; then \
+		echo "ERROR: kv-cache package not found."; \
 		exit 1; \
+	fi; \
+	if [ "$(TARGETOS)" = "darwin" ]; then \
+		if [ -f "$$KV_CACHE_PKG/pkg/preprocessing/chat_completions/setup.sh" ]; then \
+			echo "Running kv-cache setup script for macOS..."; \
+			cp "$$KV_CACHE_PKG/pkg/preprocessing/chat_completions/setup.sh" build/kv-cache-setup.sh; \
+			chmod +wx build/kv-cache-setup.sh; \
+			cd build && PATH=$(VENV_BIN):$$PATH ./kv-cache-setup.sh && cd ..; \
+		else \
+			echo "ERROR: setup script not found at $$KV_CACHE_PKG/pkg/preprocessing/chat_completions/setup.sh"; \
+			exit 1; \
+		fi; \
+	else \
+		echo "Installing vLLM for Linux $(TARGETARCH)..."; \
+		if [ "$(TARGETARCH)" = "arm64" ]; then \
+			$(VENV_BIN)/pip install https://github.com/vllm-project/vllm/releases/download/v$(VLLM_VERSION)/vllm-$(VLLM_VERSION)+cpu-cp38-abi3-manylinux_2_35_aarch64.whl; \
+		elif [ "$(TARGETARCH)" = "amd64" ]; then \
+			$(VENV_BIN)/pip install https://github.com/vllm-project/vllm/releases/download/v$(VLLM_VERSION)/vllm-$(VLLM_VERSION)+cpu-cp38-abi3-manylinux_2_35_x86_64.whl --extra-index-url https://download.pytorch.org/whl/cpu; \
+		else \
+			echo "ERROR: Unsupported architecture: $(TARGETARCH). Only arm64 and amd64 are supported."; \
+			exit 1; \
+		fi; \
 	fi
 	@echo "Verifying vllm installation..."
 	@$(VENV_BIN)/python -c "import vllm; print('✅ vllm version ' + vllm.__version__ + ' installed.')" || { \

go.mod

@@ -10,7 +10,7 @@ require (
 	github.com/google/uuid v1.6.0
 	github.com/hashicorp/golang-lru/v2 v2.0.7
 	github.com/jellydator/ttlcache/v3 v3.4.0
-	github.com/llm-d/llm-d-kv-cache v0.4.1-0.20260120091923-2d261e30d383
+	github.com/llm-d/llm-d-kv-cache v0.4.1-0.20260121180456-e3fafddd09f4
 	github.com/onsi/ginkgo/v2 v2.27.5
 	github.com/onsi/gomega v1.39.0
 	github.com/openai/openai-go v1.12.0

go.sum

@@ -183,8 +183,8 @@ github.com/kr/text v0.2.0 h1:5Nx0Ya0ZqY2ygV366QzturHI13Jq95ApcVaJBhpS+AY=
 github.com/kr/text v0.2.0/go.mod h1:eLer722TekiGuMkidMxC/pM04lWEeraHUUmBw8l2grE=
 github.com/kylelemons/godebug v1.1.0 h1:RPNrshWIDI6G2gRW9EHilWtl7Z6Sb1BR0xunSBf0SNc=
 github.com/kylelemons/godebug v1.1.0/go.mod h1:9/0rRGxNHcop5bhtWyNeEfOS8JIWk580+fNqagV/RAw=
-github.com/llm-d/llm-d-kv-cache v0.4.1-0.20260120091923-2d261e30d383 h1:cHA1+Qe27oYDsWTmNToiidMBty6bxfuCUXvyhGPJ2FM=
-github.com/llm-d/llm-d-kv-cache v0.4.1-0.20260120091923-2d261e30d383/go.mod h1:XyhzHBYeOWamBMPkuRySB5nJ0zzQpK/mbuXKqJRFT6A=
+github.com/llm-d/llm-d-kv-cache v0.4.1-0.20260121180456-e3fafddd09f4 h1:3LHSnDQ2tLsSIbh4BgN+7RYz/Wi+KjvIigcxVHb3mkE=
+github.com/llm-d/llm-d-kv-cache v0.4.1-0.20260121180456-e3fafddd09f4/go.mod h1:XyhzHBYeOWamBMPkuRySB5nJ0zzQpK/mbuXKqJRFT6A=
 github.com/mailru/easyjson v0.9.0 h1:PrnmzHw7262yW8sTBwxi1PdJA3Iw/EKBa8psRf7d9a4=
 github.com/mailru/easyjson v0.9.0/go.mod h1:1+xMtQp2MRNVL/V1bOzuP3aP8VNwRW55fQUto+XFtTU=
 github.com/maruel/natural v1.1.1 h1:Hja7XhhmvEFhcByqDoHz9QZbkWey+COd9xWfCfn1ioo=