vllm-eks-helm Terraform module

Terraform module which creates a vLLM workload on EKS and manage related Helm releases.

Usage

See the examples directory for working examples to reference:

module "vllm_mistral_7b" {
  source = "drizzle-ai-systems/vllm-eks-helm/aws"

  eks = {
    cluster_name = "eks-cluster" # cluster name needs to be existing
  }

  vllm = {
    enabled      = true
    force_update = true
  }

  k8s = {
    create_namespace          = true
    kubernetes_namespace      = "vllm-mistral-7b"
    service_account_namespace = "vllm-mistral-7b"
    service_account_name      = "vllm-mistral-7b-sa"
    iam_role_enabled          = true
  }

  helm = {
    atomic          = true
    cleanup_on_fail = true
    timeout         = 300
    wait            = true
    values_template_path = "${path.module}/helm/mistral-7b-values.yaml"
  }

  tags = local.tags
}

Dependencies

• helm release module by CloudPosse
• vLLM Helm chart

Configure LLM Model with vLLM

This module is using the vLLM Helm chart to deploy a vLLM workload on EKS. To configure your LLM model with vLLM, you need to create a values template file and pass it to the module using the values_template_path variable. See vLLM Helm chart for more information.

Examples

Examples codified under the examples directory are intended to give users references for how to use the module(s) and for testing/validating changes to the module. If contributing, update the relevant examples so maintainers can test changes and users have up-to-date references.

• Complete example

Requirements

Name	Version
terraform	>= 1.3.0
aws	>= 6.19

Providers

Name	Version
aws	>= 6.19

Modules

Name	Source	Version
vllm	./modules/helm-release	n/a

Resources

Name	Type
aws_eks_cluster.this	data source

Inputs

Name	Description	Type	Default	Required
eks	EKS cluster information. Only cluster_name is required by this module.	object({ cluster_name = string })	n/a	yes
helm	Helm client options	object({ atomic = bool cleanup_on_fail = bool timeout = number wait = bool values_template_path = string })	{ "atomic": true, "cleanup_on_fail": true, "timeout": 300, "wait": true }	no
k8s	Kubernetes-related settings for the helm release	object({ create_namespace = bool kubernetes_namespace = string service_account_namespace = string service_account_name = string iam_role_enabled = bool })	{ "create_namespace": true, "iam_role_enabled": true, "kubernetes_namespace": "vllm", "service_account_name": "vllm-sa", "service_account_namespace": "vllm" }	no
tags	A map of tags to add to all resources	map(string)	{}	no
vllm	vllm helm module settings	object({ enabled = bool force_update = bool })	{ "enabled": true, "force_update": true }	no

Outputs

No outputs.

License

Apache-2.0 Licensed. See LICENSE.