experiment-registry

To track ML experiments with a YAML registry, you add a YAML entry describing your experiment and run the `exp register` command. This stores metadata in a human-readable format, making it easy to manage configurations and statuses across runs.

A YAML-based experiment registry is a human-readable system for managing the lifecycle of ML experiments. It stores configurations, results, and statuses in YAML files, enabling simple register and update flows to track benchmarks across different runs.

Yes, you need the `exp` CLI installed to use this YAML registry workflow. The project root must contain an `exp.config.yaml` file along with per-experiment YAML entries to initialize, list, and report on your machine learning runs.

The best way to manage ML benchmarks across datasets is using a registry-driven approach that tracks results and evaluation modes in YAML files. This allows you to update and compare benchmark performance across multiple experiment runs efficiently.

Using YAML files for experiment tracking provides a human-readable, lightweight format for storing metadata without database overhead. It simplifies version control integration and makes reviewing configurations and benchmark results across runs straightforward.

To initialize a project for ML experiment tracking, you set up an `exp.config.yaml` file in your project root and create per-experiment YAML entries. You then use the `exp` CLI to establish the registry and register your configurations.