Weights & Biases (W&B) is a machine learning operations (MLOps) platform that provides tools for experiment tracking, model evaluation, dataset versioning, and collaborative ML development. Founded in 2017 by Lukas Biewald, Chris Van Pelt, and Shawn Lewis, W&B has become one of the most widely used experiment tracking tools in the ML community, adopted by researchers and engineers at organizations including OpenAI, NVIDIA, Meta, Google DeepMind, and thousands of academic institutions. The core product, W&B Experiments, allows machine learning practitioners to log hyperparameters, metrics, model outputs, system performance, and artifacts from training runs, then visualize and compare results through an interactive web dashboard. This eliminates the need for manual spreadsheets or ad-hoc logging and makes ML experiments reproducible and shareable. W&B Sweeps automates hyperparameter optimization using strategies like Bayesian optimization, grid search, and random search. W&B Artifacts provides version control for datasets and models, tracking lineage and dependencies throughout the ML pipeline. W&B Tables enables interactive exploration and visualization of training data and model predictions, facilitating error analysis and dataset debugging. W&B Reports allows teams to create collaborative documents that combine visualizations, code, and narrative to document and share ML findings. More recently, W&B has expanded into LLM-specific tooling with W&B Weave, a framework for evaluating, monitoring, and debugging LLM applications in production. The platform integrates with virtually all major ML frameworks including PyTorch, TensorFlow, Keras, Hugging Face, scikit-learn, and XGBoost. W&B offers a free tier for individuals and academic users, a Teams plan starting at $50 per user per month, and a custom-priced Enterprise plan with on-premises deployment options and advanced security controls.
wandb.ai →