tnt

Why we included this project

TorchTNT comes from the PyTorch team and is aimed at people who are tired of rewriting the same training-loop scaffolding for every experiment. It bundles a fit and eval loop framework with a hook-based callback system, along with data-loading helpers and utilities for checkpointing, learning-rate schedules, logging, and profiling. Instead of forcing a whole new framework on you, it lets you keep your own model and training code and pull in just the pieces you need. Teams that want a consistent training setup across projects, and researchers building custom pipelines, can both adopt it piece by piece, since it composes with plain PyTorch rather than replacing it.