Trackio: Free and Open-Source Experiment Tracker Revolutionizing Machine Learning Workflows

What Is Trackio?

Trackio is an open-source Python library designed to simplify experiment tracking in machine learning workflows. Developed by Hugging Face and Gradio, it serves as a drop-in replacement for popular tools like wandb, supporting foundational API calls such as wandb.init, wandb.log, and wandb.finish. This enables users to switch or run existing scripts with minimal or no code changes—just import Trackio as wandb.

Core Features

• Local-First Design: Experiments run and store data locally by default, ensuring privacy and fast access. Sharing data is optional.
• Free and Open Source: No licensing fees or feature restrictions. All features, including collaboration and online dashboards, are available at no cost.
• Lightweight and Extensible: The entire codebase is under 1,000 lines of Python, making it easy to audit, modify, or extend.
• Integration with Hugging Face Ecosystem: Supports Transformers, Sentence Transformers, and Accelerate out of the box, enabling quick setup for metric tracking.
• Data Portability: Experiment data is stored in accessible formats, allowing easy export and integration with custom analytics or pipelines.

Seamless Experiment Tracking: Local and Shared Dashboards

Trackio offers flexible dashboard options. Researchers can monitor experiments locally via a Gradio-powered dashboard with commands like:

trackio show

Or in Python:

import trackio
trackio.show()

To share dashboards online, logs can be synced with Hugging Face Spaces, enabling instant sharing or embedding via URLs. Spaces support private or public settings without complex authentication. Trackio automatically backs up data every 5 minutes to Hugging Face Datasets in Parquet format to prevent data loss.

Easy Integration with Machine Learning Workflows

Integration is straightforward, especially for users of the Hugging Face ecosystem. For instance, with Accelerate, simply specify Trackio as the logger:

from accelerate import Accelerator
accelerator = Accelerator(log_with="trackio")
accelerator.init_trackers("my-experiment")
...
accelerator.log({"training_loss": loss}, step=step)

This low-friction approach allows immediate experiment tracking and sharing without extra setup.

Transparency, Sustainability, and Data Freedom

Trackio supports tracking advanced metrics like GPU energy consumption (via nvidia-smi), promoting environmental responsibility and reproducibility. Unlike proprietary platforms, all data is stored in open formats and dashboards use open tools such as Gradio and Hugging Face Datasets, encouraging transparency and ease of analysis.

Getting Started

Install Trackio via pip:

pip install trackio
# or
uv pip install trackio

Then, update your imports to use Trackio:

import trackio as wandb

Trackio empowers researchers and teams with a free, transparent, and local-first experiment tracker, fully integrated with the Hugging Face ecosystem and built for modern machine learning research.

For more resources, visit the GitHub Page and follow updates on Twitter and community forums.