A lightweight, extensible logging framework for PyTorch training loops with automatic Visdom integration and minimal boilerplate.
This project reduces manual metric instrumentation when using Visdom by introducing structured logging, automatic metric grouping, and a clean Trainer abstraction.
Visdom is powerful for visualizing training progress, but it requires explicit manual logging for every metric:
vis.line(Y=[loss], X=[epoch], win='loss', update='append')
vis.line(Y=[acc], X=[epoch], win='acc', update='append')