Open-Source Release of Machine Learning Metrics Tracker "Aspara"
PredNext Inc. released “Aspara,” an open-source (OSS) metrics visualization software that tracks the changes in various metrics obtained during machine learning training, on February 18, 2026.
Aspara is a dashboard designed to quickly grasp changes during training and make it easy to compare and reference experiments later.
Background: Growing Experiment Logs Due to AI Model Scale-Up
With recent advances in AI research, state-of-the-art AI models have grown significantly larger over the past decade. As models scale up, training takes longer, and the logs (experiment records) generated per training run tend to increase.
Additionally, training production models is costly in terms of computational resources and time, and there are cases where more trial and error is needed beforehand to refine training conditions.
Against this backdrop, the importance of visualization and management tools that allow you to quickly survey experiment logs and reach the results you need has been growing.
Features of Aspara
In machine learning experiments, you need to quickly determine the direction of improvement while tracking metrics such as accuracy and loss. At the same time, it’s also important to quickly find and compare the results you need from past experiment logs.
Aspara was developed with the goal of achieving both of the following, in response to growing experiment logs and increasing numbers of experiments:
- Ability to quickly check ongoing training
- Ability to organize past experiment logs for comfortable investigation and comparison
Always Responsive Interface
In Aspara, downsampling is performed on the server side before sending data to the dashboard. This design ensures that display and operations remain responsive even as data volume increases. The LTTB algorithm is used for downsampling, enabling visualization that captures outliers effectively.
Organization and Filtering with Tags
In Aspara, a single experiment is called a “run,” and runs belong to a “project.” Tags can be attached to runs and projects, and filtering by tags is supported.
This allows you to preserve information such as “what perspective do I want to compare from?” and “what was the intent behind this result?” in a form that’s easy to reference later.
Two Dashboards (Web / TUI)
Aspara provides two types of dashboards: a web dashboard and a TUI (terminal-based interface).
- The web dashboard allows you to check metric trends with graphs
- The TUI dashboard enables quick result checking even in environments where web access is inconvenient
How to Use
Aspara is available on GitHub. Anyone can freely download and use it in local or on-premises environments.
GitHub URL: https://github.com/prednext/aspara
Future Plans
We plan to add unimplemented features such as image upload functionality and stabilize the API toward the 1.0 release. At the same time, we will work on feature improvements and stability enhancements while incorporating feedback from the community.
Additionally, we are considering offering Aspara as a SaaS as an option to reduce the operational burden of running tracking servers. Even after launching the SaaS, all basic features, including the server side, will continue to be provided under an OSS license.
About PredNext
PredNext Inc. is an engineering group specializing in AI. Based on an understanding of both AI’s potential and its current limitations, we propose AI solutions tailored to our clients’ challenges. Rather than assuming that “AI can solve all problems,” we propose the most effective ways to leverage AI to address our clients’ challenges.