kdeps-recipes-vllm

id: vllm
name: vLLM
description: High-throughput OpenAI-compatible server (vLLM) for GPU inference
version: 1.0.0

api:
  port: 8000
  base_path: /v1
  chat_path: /v1/chat/completions
  health:
    method: GET
    path: /v1/models
  auth:
    mode: none
    env: LLM_API_KEY

engine:
  kind: vllm
  base_image: "vllm/vllm-openai:latest"
  packages: []
  install: |
    true
  command:
    - sh
    - -c
    - exec python3 -m vllm.entrypoints.openai.api_server --host 0.0.0.0 --port 8000 --model "${LLM_MODEL}"
  env: {}
  internal_port: 8000
  openai_bridge: false

models:
  strategy: pull
  default: ""

resources:
  gpu: required
  memory_hint: "16Gi"

capabilities:
  - chat
  - tools