Docker#

This directory contains the Docker assets used to package KonfAI from the PyPI release.

Included files:

  • Dockerfile: build a KonfAI image with PyTorch preinstalled

  • entrypoint.sh: forwards commands to konfai and related CLIs

Docker Hub#

The official image is available on Docker Hub as vboussot/konfai.

Pull the published image:

docker pull vboussot/konfai

If you build the image locally instead, replace vboussot/konfai with your local tag such as konfai in the examples below.

Run the default help command:

docker run --rm vboussot/konfai

Local Build#

Build the default image from this repository:

docker build -f docker/Dockerfile -t konfai .

Build a specific KonfAI version from PyPI:

docker build -f docker/Dockerfile \
  --build-arg KONFAI_PYPI_VERSION=1.5.3 \
  -t konfai .

Build with additional optional dependencies:

docker build -f docker/Dockerfile \
  --build-arg KONFAI_EXTRAS=server,cluster \
  -t konfai .

Build a CPU-only image:

docker build -f docker/Dockerfile \
  --build-arg TORCH_INDEX_URL=https://download.pytorch.org/whl/cpu \
  -t konfai-cpu .

What The Image Provides#

The image exposes the KonfAI CLI entrypoints:

  • konfai

  • konfai-apps

  • konfai-apps-server

  • konfai-cluster

If no command is provided, the container runs konfai --help. If the first argument is not one of the known executables, it is forwarded to konfai.

GPU Runtime#

The default image installs a CUDA-enabled PyTorch wheel. GPU access still depends on the host runtime.

Quick CUDA check:

docker run --rm -it --gpus all vboussot/konfai \
  python -c "import torch; print(torch.__version__); print(torch.cuda.is_available()); print(torch.cuda.device_count())"

If CUDA is not visible inside the container, check:

  • NVIDIA drivers on the host with nvidia-smi

  • Docker GPU support with docker run --rm --gpus all nvidia/cuda:12.8.1-runtime-ubuntu24.04 nvidia-smi

  • nvidia-container-toolkit installation on the host

Run KonfAI#

Train from the repository root:

docker run --rm -it \
  --gpus all \
  -v "$(pwd):/workspace" \
  -w /workspace \
  vboussot/konfai TRAIN --gpu 0 -c examples/Synthesis/Config.yml

Run prediction or evaluation:

docker run --rm -it \
  --gpus all \
  -v "$(pwd):/workspace" \
  -w /workspace \
  vboussot/konfai PREDICTION --models checkpoint.pt --gpu 0 -c examples/Synthesis/Prediction.yml

docker run --rm -it \
  -v "$(pwd):/workspace" \
  -w /workspace \
  vboussot/konfai EVALUATION -c examples/Synthesis/Evaluation.yml

Run KonfAI Apps#

Run an app command:

docker run --rm -it \
  --gpus all \
  -v "$(pwd):/workspace" \
  -w /workspace \
  vboussot/konfai konfai-apps infer my_app -i input.mha -o ./Output

Run the apps server:

docker run --rm -it -p 8000:8000 \
  --gpus all \
  -v "$(pwd):/workspace" \
  -w /workspace \
  -e KONFAI_API_TOKEN=my-token \
  vboussot/konfai konfai-apps-server --host 0.0.0.0 --port 8000 --apps konfai-apps/tests/assets/apps.json

Notes#

  • The image is intended for CLI workflows executed from a mounted workspace.

  • The default image is GPU-oriented; use a custom TORCH_INDEX_URL if you want a CPU-only variant.

  • For reproducibility, pin the PyPI version with KONFAI_PYPI_VERSION when rebuilding locally.

In the docs#

Docs notes. The docker run examples above mount the current directory as /workspace — run them from the directory that contains your configs, data, and checkpoints.

Next steps:

  • CLI reference — the flags used in the container commands above

  • Installation — install KonfAI directly on the host instead