aim
Aim 💫 — An easy-to-use & supercharged open-source experiment tracker.
What is aim?
Aim 💫 — An easy-to-use & supercharged open-source experiment tracker.
Overview of aim
What it is
Aim is an open-source, self-hosted ML experiment tracking tool designed to handle 10,000s of training runs. It logs training runs and any AI metadata, provides a UI to compare and observe them, and offers an API to query them programmatically. Aim integrates with popular ML frameworks and includes built-in converters for migrating logs from TensorBoard, MLflow, and Weights & Biases.
Key features
- Logs metadata across the ML pipeline, including hyperparameters, metrics, images, text, and more.
- Provides a performant UI for visualizing and comparing training runs with grouping, aggregation, and Python expression querying.
- Tracks system info and resource usage, with real-time alerting and configurable notifications.
- Supports run grouping with tags and experiments, and a centralized dashboard for a holistic view.
- Includes a remote tracking server for multi-host environments and can be deployed on Kubernetes.
- Offers an SDK for programmatic access to tracked metadata, suitable for automations and Jupyter Notebook analysis.
Use cases
- Tracking and comparing ML experiments across many runs.
- Integrating with frameworks like PyTorch Lightning, Hugging Face, Keras, XGBoost, and others via built-in callbacks.
- Migrating from other experiment trackers (TensorBoard, MLflow, Weights & Biases) using built-in converters.
- Setting up centralized tracking for teams working in multi-host environments.
- Querying and analyzing experiment data programmatically for custom workflows.
Getting started
shell pip3 install aim python from aim import Run run = Run() run['hparams'] = {'learning_rate': 0.001, 'batch_size': 32} for i in range(10): run.track(i, name='loss', step=i, context={'subset':'train'}) run.track(i, name='acc', step=i, context={'subset':'train'}) shell aim up
License and pricing
The license is Apache-2.0. Aim is open-source and free to use.
Based on the project's public README and repository metadata.