Skip to main content

Getting Started

Prerequisites​

  • Docker Desktop or Docker Engine
  • NVIDIA stack for GPU mode
  • Python is not required when using Docker-first workflows

Configure​

Copy .env.example to .env and set:

VAQ_HF_CACHE_HOST_PATH=/absolute/path/to/huggingface/cache

Optional images:

VAQ_VLLM_IMAGE=vllm/vllm-openai:latest
VAQ_VLLM_CPU_IMAGE=vllm/vllm-openai-cpu:latest-x86_64

Build​

docker compose build vaq

Use prebuilt GHCR image​

If you do not want to build locally, use the published image:

docker pull ghcr.io/xschahl/vaquila:v0.1.0-beta.1
docker run --rm ghcr.io/xschahl/vaquila:v0.1.0-beta.1 --help

You can also use it as a base image in your own Dockerfile:

FROM ghcr.io/xschahl/vaquila:v0.1.0-beta.1

Functional example files are available in docs/examples/ghcr/:

  • docs/examples/ghcr/docker-compose.yml
  • docs/examples/ghcr/Dockerfile
  • docs/examples/ghcr/.env.example

Quick test with the example compose file:

cd docs/examples/ghcr
cp .env.example .env
docker compose run --rm vaq --help

Run a model​

GPU mode:

docker compose run --rm vaq run Qwen/Qwen3-0.6B --gpu 0 --port 8000

CPU mode:

docker compose run --rm vaq run openai-community/gpt2 --device cpu --port 8000