Aim

Aim is a free, open source ML experiment tracker for researchers and teams who want to keep training records on their own infrastructure. It runs in your training environment or on a self-hosted server, with Docker and Kubernetes deployment support. Its Apache 2.0 license permits use and modification.

Run comparison is its focus. The interface lets you filter experiments by recorded parameters, group related runs and aggregate metrics, so you can compare training choices without encoding every detail in run names. Tags and experiment groups help organize larger collections of results.

Aim records more than numerical metrics. It tracks images, text, audio and distributions, and provides views for inspecting them alongside training data. System information, resource usage and execution logs help explain a run's behavior; progress alerts and notifications can flag stalled training.

The Python SDK gives you access to recorded metadata for custom analysis and automation, including work in Jupyter notebooks. A central tracking server collects results from experiments running across multiple hosts, keeping the records in a location you control.

Built-in integrations connect Aim to Hugging Face, PyTorch Lightning, Keras and other training frameworks, as well as tuning tools such as Optuna. Converters import existing logs from TensorBoard, MLflow and Weights & Biases. The separate aimlflow tool uses Aim's interface to explore MLflow experiments.