Ignite

This library provides a high level layer on top of PyTorch for training and evaluating neural networks. It provides an engine that runs training and evaluation loops, and an event system that can trigger handlers at any built in or custom event. Handlers cover checkpointing, early stopping, parameter scheduling, profiling and a learning rate finder. The library is aimed at machine learning engineers and researchers who want flexible, transparent training code. Its capabilities include more than 50 metrics that are ready for distributed use, distributed training across CPUs, GPUs and TPUs with helpers for models, optimizers and data loaders, and integrations with TensorBoard, MLflow, Weights and Biases and Neptune. The library is an open source community project, and the project publishes no paid plans or prices. The vendor site lists contribution guidelines, a code of conduct and public channels for bug reports, feature requests and discussion. The project is a NumFOCUS affiliated project and lists MONAI as an ecosystem project.