Software
Luz is a higher-level API for torch in R that reduces the verbosity of deep learning training loops. It lets you take a torch nn_module definition and fit it to a dataloader with a concise setup() %>% fit() interface, handling device placement, weight updates, progress tracking, and metrics automatically.
Luz draws on ideas from FastAI, Keras, PyTorch Lightning, and HuggingFace Accelerate. It provides a callbacks API for customizing training behavior and treats luz_module as a subclass of nn_module, so existing torch models work with minimal changes. It currently supports CPU and single-GPU training.
