Best for MLOps, LLMOps, and developing models or work with weights locally.

Container recipes that back the machine-learning images published under ghcr.io/nekomeowww. The repository is organised around reusable base images and higher level tooling built on top of them.

• ghcr.io/nekomeowww/oci-ml-python — source
• ghcr.io/nekomeowww/oci-ml-nvidia-cuda — source
• ghcr.io/nekomeowww/oci-ml-pytorch — source
• ghcr.io/nekomeowww/oci-ml-onnxruntime — source

See source for available tags and dependency graph.

Use Docker Buildx (or regular docker build) from the repository root. The helper script below rebuilds the current stack in dependency order:

./dockerfiles/clis/optimum-cli/0.0.3/build.sh

Each image can also be built individually. For example, to rebuild the PyTorch base image:

docker buildx build . \
  -f dockerfiles/bases/torch/py3.13-cu13.0-2.9.0/Dockerfile \
  --load -t ghcr.io/nekomeowww/oci-ml-pytorch:py3.13-cu13.0-2.9.0