aim

Why we included this project

Aim is a self-hosted experiment tracker: you point it at your training runs, it logs the metrics and hyperparameters you care about along with any other metadata, and you get a fast web UI plus a Python API to dig back through them. The UI is where it earns its keep compared with a plain logging library: explorers for metrics, images, and text let you filter and group runs with Python-style expressions, which is exactly what you want when debugging what changed between two experiments. It handles thousands of runs, can track system resource usage, and supports real-time alerts, so it holds up as a central dashboard for a solo researcher or a whole team. If you're switching from another tracker, Aim can migrate existing runs over, which lowers the cost of moving. For teams that want the comparison and observation layer living on their own infrastructure instead of a hosted service, Aim is a straightforward fit.