Summary

Trackio is a lightweight, free, and local-first experiment tracking library designed for machine learning practitioners and LLM-driven autonomous experiments. It offers a drop-in replacement for wandb's API, enabling users to log metrics locally via SQLite and visualize them with an integrated Gradio dashboard. The library facilitates seamless deployment to Hugging Face Spaces for collaboration and sharing, all at no cost.

How It Works

Trackio provides API compatibility with wandb.init, wandb.log, and wandb.finish, allowing existing logging code to be used with minimal changes by simply aliasing trackio as wandb. Its core design prioritizes a local-first approach, persisting experiment data in a local SQLite database by default. For cloud-based visualization and collaboration, users can specify a space_id in trackio.init(), which deploys or utilizes an existing Hugging Face Space, storing data in a private Hugging Face Dataset. The library features an LLM-friendly CLI and Python API for programmatic experiment management and data querying, alongside an embedded Gradio dashboard for interactive visualization.

Quick Start & Requirements

• Installation: pip install trackio or uv pip install trackio.
• Prerequisites: Python 3.10 or higher.
• Running the Dashboard: Execute trackio show in the terminal or trackio.show() in Python.
• Documentation: https://huggingface.co/docs/trackio/index
• Examples: fake-training.py, persist-dataset.py, deploy-on-spaces.py

Highlighted Details

• Drop-in API replacement for wandb.
• Local-first design with SQLite persistence.
• Free hosting and deployment on Hugging Face Spaces.
• LLM-friendly CLI and Python API for programmatic experiment management.
• Embeddable Gradio dashboard with customizable query parameters.

Maintenance & Community

• Follow updates on Twitter: https://twitter.com/trackioapp

Licensing & Compatibility

• License: MIT License.
• Compatibility: Permissive MIT license allows for commercial use and integration with closed-source projects.

Limitations & Caveats

Trackio is currently in beta, with potential for breaking changes, particularly concerning its SQLite database schema. Users may need to migrate or delete existing local database files (~/.cache/huggingface/trackio). While designed to be extensible, it is not a fully-featured replacement for more mature experiment tracking tools.