vboussot/konfai repository overview

⁠KonfAI

GPU-ready Docker image for KonfAI, a YAML-driven PyTorch framework for training, prediction, evaluation, and medical imaging workflows.

⁠What is included

• konfai
• konfai-apps
• konfai-apps-server
• konfai-cluster

The image installs KonfAI from PyPI and is designed to run CLI workflows inside a container with optional NVIDIA GPU support.

⁠Quick start

Run the main CLI:

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

Run KonfAI Apps:

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 tests/assets/apps.json

⁠Notes

• NVIDIA Container Toolkit is required for --gpus all.
• If no command is provided, the container defaults to konfai --help.
• If the first argument is not a known KonfAI executable, it is forwarded to konfai.
• The default image is GPU-oriented because it installs a CUDA-enabled PyTorch wheel.

⁠Source

• GitHub: https://github.com/vboussot/KonfAI⁠
• PyPI: https://pypi.org/project/konfai/⁠