Weights & Biases (W&B) is an MLOps platform focused on experiment tracking, model versioning, and collaborative ML development. It has become the de facto standard for research teams and production ML engineering.
**Experiment Tracking**
W&B's core feature is logging ML experiments with a few lines of code. Metrics, hyperparameters, gradients, system stats, and media (images, audio, video) are automatically captured and visualised in an interactive dashboard.
**Sweeps**
The Sweeps feature automates hyperparameter search using Bayesian optimisation, random search, or grid search. Defining a search space and launching parallel runs across machines is straightforward and the results are immediately comparable.
**Artifacts**
W&B Artifacts provide versioned storage for datasets, models, and other pipeline outputs. Lineage tracking shows exactly which dataset version produced which model, enabling reproducible ML workflows.
**Reports**
Collaborative reports allow creating narrative documents that embed live W&B charts — ideal for sharing experiment results with stakeholders or documenting research findings.
**LLM Support**
W&B Weave, launched in 2024, adds LLM-specific tooling: prompt versioning, LLM call tracing, evaluation frameworks, and cost tracking across providers.
**Pricing**
Free for individuals and academic use. Team plans from $50/month. Enterprise pricing available.
**Verdict**
W&B is essential infrastructure for any serious ML team. The experiment tracking alone justifies adoption; the broader MLOps features make it a complete platform.