Setup & Installation
npx skills add https://github.com/huggingface/skills --skill hugging-face-trackio
https://github.com/huggingface/skills/tree/main/skills/hugging-face-trackio
View on GitHubWhat This Skill Does
Trackio is an experiment tracking library for ML training. It logs metrics via a Python API, fires structured alerts for diagnostic conditions, and exposes a CLI for querying logged data. Metrics sync to a Hugging Face Space for real-time dashboard monitoring.
Unlike manually printing metrics to stdout or parsing log files, Trackio stores metrics in a queryable database and exposes them via CLI with JSON output, so automated agents and scripts can read training state without parsing unstructured text.
When to use it
- Logging loss and accuracy metrics during a PyTorch training loop
- Polling for loss divergence alerts from a background training job
- Querying training run metrics as JSON to feed into an automated hyperparameter search
- Setting up Slack or Discord webhooks to receive training error notifications
- Syncing local experiment data to a Hugging Face Space so metrics persist after a cloud instance shuts down