A self-hosted, open-source application for logging, exploring, and comparing thousands of ML training runs. It provides a web UI and Python SDK without requiring a paid service.
Project overview
A self-hosted, open-source application for logging, exploring, and comparing thousands of ML training runs. It provides a web UI and Python SDK without requiring a paid service.
Project type
Evaluation & Observability · Infrastructure
Use cases
Data Analysis
Deployment
Refer to project documentation
License
Apache-2.0
Best for
AI engineers, developers, and data teams who need to track and compare thousands of training runs and require a self-hosted solution without paid service dependencies.
Key capabilities
Logs ML training runs, metrics, parameters, and any AI metadata.
Provides a performant UI for exploring, visualizing, and comparing metadata, metrics, images, and audio across runs.
Enables programmatic access to tracked metadata for automations and Jupyter Notebook analysis.
Easily migrate logs from other tools like TensorBoard, MLflow, and Weights & Biases.
Allows running experiments in a multi-host environment and collecting tracked data in a centralized location.
Tracks system info, environment variables, CLI arguments, dependencies, and resource usage.
Real-time alerting on training progress and notifications for stuck runs.
Planned UI explorer for visualizing distribution data.
Limitations and risks
The UI may get shaky when exploring 1000s of metrics with 10000s of steps each.
Getting started
Install the tool using pip (e.g., pip3 install aim), integrate it with your existing code, and start the UI using the 'aim up' command.
Evidence and sources
GitHub project description: An easy-to-use & supercharged open-source experiment tracker
README: Aim is an open-source, self-hosted ML experiment tracking tool designed to handle 10,000s of training runs.