runyard

runyard shows your machine learning training runs as they happen, from your own runyard server. Step away from the desk and still see the loss curve, the GPU load and the last lines of the log, and know the moment a run dies. RUNS - Every running run on one screen, with its latest numbers and how far along it is. - For each run: charts of every metric, system load, sample images, logs, config and files. - A run that crashes, fails or stops sending stands out the moment it happens. MACHINES - Every machine running the runyard agent, with every process on its GPUs and who started it. ALERTS - The alerts your runs and your server raise: a NaN loss, a stalled run, a message from your training script. YOUR SERVER, YOUR DATA runyard is a free, open-source run tracker you host yourself: github.com/cobanov/runyard. Your training scripts log to it with a few lines of Python. Point the app at your server's address, on your network, a tailnet or the internet. The app talks only to your server and collects nothing. No server yet? Tap Try the demo on the first screen: the whole app on made-up runs that keep training.