MASE Docker

Docker images for running MASE. Two variants are provided:

• Dockerfile-cpu — CPU-only, used for CI and general development
• Dockerfile-gpu — CUDA-enabled, for GPU-accelerated training and inference

Official Images (DeepWok)

The official MASE Docker images are automatically built and pushed to Docker Hub by GitHub Actions whenever changes are merged into main:

Image	Docker Hub
CPU	deepwok/mase-docker-cpu:latest
GPU	deepwok/mase-docker-gpu:latest

To use the official images:

# CPU
docker pull deepwok/mase-docker-cpu:latest

# GPU
docker pull deepwok/mase-docker-gpu:latest

---

Build Your Own Image

If you need to customise the image (e.g. different CUDA version, private fork), you can build and push it yourself.

Prerequisites:

• Docker Desktop with buildx support
• A Docker Hub account
• For GPU image: check your host driver version with nvidia-smi

Replace <your-dockerhub-username> with your Docker Hub username before running.

CPU image:

DOCKER_USER=<your-dockerhub-username>
TAG=$(date -u +%Y%m%d%H%M%S) && \
docker buildx build --no-cache \
  --platform linux/amd64 \
  -f Dockerfile-cpu \
  --tag docker.io/$DOCKER_USER/mase-cpu:$TAG \
  --tag docker.io/$DOCKER_USER/mase-cpu:latest \
  --push \
  .

GPU image (see Dockerfile-gpu for available CUDA version options):

DOCKER_USER=<your-dockerhub-username>
TAG=$(date -u +%Y%m%d%H%M%S) && \
docker buildx build --no-cache \
  --platform linux/amd64 \
  -f Dockerfile-gpu \
  --tag docker.io/$DOCKER_USER/mase-gpu:$TAG \
  --tag docker.io/$DOCKER_USER/mase-gpu:latest \
  --push \
  .