Weights & Biases

Weights & Biases is the experiment tracking system most ML teams reach for. Add a few lines to a training script and every run's metrics, hyperparameters, code version, dataset, and system utilisation are logged, versioned, and comparable.

Core platform

• Experiments. Live metric dashboards, run comparison, and the reproducibility record that makes a result defensible six months later.
• Sweeps. Hyperparameter search with Bayesian, grid, or random strategies, distributed across your machines.
• Artifacts. Versioned datasets and model checkpoints with full lineage — which data produced which model.
• Model registry for promoting checkpoints through staging to production.
• Reports. Shareable documents combining live charts and narrative, which is how most teams communicate results.

Weave

W&B's LLM-focused product covers what training tools do not: tracing agent and application calls, evaluating outputs with scorers, comparing prompt and model versions, and human feedback collection.

Access

Free for personal and academic use with generous limits; paid team plans and enterprise deployment including self-hosting. It integrates with essentially every ML framework, usually in one line.