Weights & Biases (W&B) tracks and visualizes machine learning work, from datasets through training runs to production models. Its Python library records each training run with its hyperparameters and logged metrics, such as accuracy and loss, and the W&B app shows how those metrics changed over each training step.
W&B integrates with popular ML frameworks and libraries for experiment tracking and data versioning inside existing projects, and a developer guide covers adding W&B to new frameworks. For LLM apps, the separate Weave suite tracks, debugs, evaluates, and monitors GenAI applications.
W&B Server runs in three ways: a multi-tenant managed cloud on Google Cloud, a single-tenant dedicated cloud on AWS, Google Cloud, or Azure with isolated network, compute, and storage, or self-managed on your own cloud account or on-premises infrastructure.
Features
- Experiment tracking: records each run's hyperparameters and logged metrics from a Python script or notebook
- Visualization: dashboards show how metrics change across training steps and runs
- Data versioning: versions data alongside experiment tracking inside existing projects
- Framework integrations: experiment tracking inside popular ML frameworks and libraries
- LLM tooling: the Weave suite for tracking, debugging, evaluating, and monitoring LLM apps
- Hosting options: multi-tenant cloud, dedicated single-tenant cloud, or self-managed on AWS, Google Cloud, Azure, or on-premises
