Experimental tracking is an important part of modern machine learning workflows. Whether you're tweaking hyperparameters, monitoring training metrics, or collaborating with colleagues, it's important to have a robust, flexible tool for your tracking experiments that are open, insightful, and flexible. However, many existing experimental tracking solutions require complex setups. License fees are provided or user data is locked into a unique format, making it difficult for individual researchers and small teams to access it.
Meet Trackio – A new open source experimental tracking library developed by hugging faces and gradients. Trackio is a local first, lightweight, completely free tracker designed for today's fast-paced research environment and open collaboration.
What is Trackio?
Trackio is a Python package designed as Drop-in exchange For widely used libraries like WandB, it is compatible with the basic API calls (wandb.init, wandb.log, wandb.finish). This places Trackio simply imports tracio as a wandb to continue working as before, with little code changes required to switch or run legacy scripts.
Important features
- Local First Design: By default, experiments run and persist locally, providing privacy and fast access. Sharing is optional and not by default.
- Free and open source: There are no paywalls or feature restrictions. Everything, including collaborations and online dashboards, is free and available to everyone.
- Lightweight and expandable: The entire codebase is below 1,000 lines of Python, making it easier to audit, extend, or adapt.
- Ecosystem and integration of embracing faces: Aut-Box Support
Transformers,Sentence TransformersandAccelerateusers can start tracking metrics with minimal setup. - Data Portability: Unlike some established tracking tools, Trackio easily exports all experimental data, making it accessible and enhances seamless integration into custom analysis and research pipelines.
Seamless experimental tracking: local or sharing
One of the outstanding features of Trackio is that Sharingability. Researchers can simply monitor metrics on local gradient-driven dashboards, or hug face spaces, and exchange dashboards online and share them with colleagues (or the public). Space is not the complex authentication or onboarding required by your audience.
For example, to view the experimental dashboard locally:
Or from Python:
import trackio
trackio.show()
To launch a dashboard in a space:
- Synchronize logs with hugging face space Instantly share or embed your experimental dashboard with a simple URL.
Importantly, when running in space, Trackio automatically backs up metrics from Epememeral Sqlite DB to Hugging Face Dataset every 5 minutes (as Parquet Files).
Plug and Play Integration with ML Workflows
Integrating with the hugging face ecosystem is as easy as it gets.
- and
transformers.Traineroraccelerateyou can record and visualize metrics by specifying Trackio as a logger.
For example, using Accelerate:
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 means that anyone using a transformer, sentence transformer, or acceleration can immediately begin their tracking and sharing experiments with zero additional setups.
Transparency, sustainability, and data freedom
Trackio goes further than standard metrics and promotes transparency in the use of computational resources. Supports tracking metrics such as: Using GPU Energy (by reading from nvidia-smi), the function that follows the facial embrace against environmental responsibility and reproducibility of the model card documentation.
Unlike a closed platform, Your data is always accessible: Trackio's logs are stored in standard format, and Dashboard is built using open tools such as Gradio and Hugging Face Datasets to make everything easier to remix, analyze and share.
Quick Start
To get started:
pip install trackio
# or
uv pip install trackio
Alternatively, exchange the import into the codebase.
Conclusion
Trackio Located to give strength Open collaboration with individual researchers In ML by providing a transparent and completely free experimental tracker. Local First Default is firmly integrated with easy-to-share, hugging face tools, bringing the promise of robust tracking without the friction or cost of traditional solutions.
Please check Technical details and github page. Please feel free to check GitHub pages for tutorials, code and notebooks. Also, please feel free to follow us Twitter And don't forget to join us 100k+ ml subreddit And subscribe Our Newsletter.
Asif Razzaq is CEO of Marktechpost Media Inc.. As a visionary entrepreneur and engineer, ASIF is committed to leveraging the possibilities of artificial intelligence for social benefits. His latest efforts are the launch of MarkTechPost, an artificial intelligence media platform. This is distinguished by its detailed coverage of machine learning and deep learning news, and is easy to understand by a technically sound and wide audience. The platform has over 2 million views each month, indicating its popularity among viewers.
