kur

Why we included this project

Kur takes a different route than most deep learning tools: you describe the model in a YAML file instead of writing training code, and the tool handles the rest. That makes it a practical on-ramp for teams who want to prototype models quickly without first learning a framework's Python API, while still leaving room for custom architectures through an extensible interface. It comes with working examples for handwriting recognition and speech-to-text, so you can walk through a complete train-and-evaluate workflow. It runs on Theano, TensorFlow, or PyTorch and supports multi-GPU training, which means you can define a model once and run it on whichever backend you have available. For teams weighing declarative, config-driven training, it's a compact and readable reference.