SwanLab

Why we included this project

Most teams that train with PyTorch or LLaMA Factory spend a surprising amount of time just reconstructing what a finished run actually did. SwanLab removes that chore with a lightweight Python SDK you drop into a training script; it records metrics, hyperparameters, hardware utilization, and media like images and videos into a searchable dashboard the whole team can browse. The integrations cover more than 50 frameworks, so you usually don't need to change how you train, just add a few lines that surface data you already had. It works as a hosted cloud service or a self-hosted deployment, which means a small lab can start right away and a security-conscious company can keep everything on its own servers. If you've outgrown TensorBoard's static charts and want run comparison, grouping, baseline views, and a modern web UI, this is a practical step up.