helm-vllm

A Crossplane Configuration package that installs the vLLM production stack Helm chart with a minimal, stable interface.

Overview

helm-vllm renders a single Helm release for vLLM production stack. It exposes only the inputs needed for chart values, namespace, and release name, keeping the interface stable while allowing full Helm overrides.

Features

• Minimal Helm interface: values and overrideAllValues with stable defaults
• Predictable naming: defaults to <clusterName>-vllm in the vllm namespace
• GitOps friendly: ships a .gitops/ deploy chart

Prerequisites

• Crossplane installed in the cluster
• Crossplane providers:
  • provider-helm (>=v1.0.6)
• Crossplane function:
  • function-auto-ready (>=v0.6.0)
• GPU resources available in the cluster (NVIDIA recommended)
• NVIDIA Kubernetes Device Plugin

Quick Start

apiVersion: pkg.crossplane.io/v1
kind: Configuration
metadata:
  name: helm-vllm
spec:
  package: ghcr.io/hops-ops/helm-vllm:latest

apiVersion: helm.hops.ops.com.ai/v1alpha1
kind: VLLM
metadata:
  name: vllm
  namespace: example-env
spec:
  clusterName: example-cluster
  values:
    servingEngineSpec:
      runtimeClassName: nvidia
      modelSpec:
        - name: llama-3
          repository: meta-llama/Llama-3.2-1B-Instruct
          gpu: 1

Development

make render
make validate
make test