traceml

Why we included this project

TraceML is the tracking and visualization engine that sits behind Polyaxon, and it works as a standalone library too. You log hyperparameters, metrics, data references, and artifacts from plain Python scripts or through callbacks for PyTorch, TensorFlow, Keras, Fastai, and Hugging Face. Beyond experiment logging it also profiles data, explains model behavior, and flags drift, so you can tell whether the data feeding a deployed model has shifted even when the loss curve looks fine. The offline mode lets you initialize tracking without a running API and capture runs locally, which is handy for trying it out before committing to a full Polyaxon setup. If you are already on Polyaxon it is the obvious companion; if you are shopping for an experiment tracker, you can test its logging API in your own training loop without any infrastructure.