Weights and Biases ML Platform

Weights and Biases (W&B) is a machine learning platform for experiment tracking, dataset versioning, and model management. Founded in 2017 by Lukasz Kaiser, Chris Van Pelt, Shawn Lewis, and Lucas Peterson in San Francisco, California. Raised over $250 million from Insight Partners, Coatue, Felicis Ventures, and Bond. $1 billion valuation. Serves over 700,000 ML practitioners at organizations including OpenAI, Toyota Research, GitHub, Square, and Pfizer. Key features: experiment tracking: automatically log hyperparameters, metrics, system statistics (GPU/CPU/memory usage), gradients, and output files during training runs. Compare runs side by side with interactive charts. Visualize training progress in real-time. W&B Artifacts: version control for datasets and models. Track data lineage from raw data through preprocessing to final model. Reproduce any version of a dataset or model. W&B Models: centralized model registry for managing, evaluating, and deploying ML models. Model evaluation, staged transitions (staging, production, archived), and approval workflows. W&B Tables: interactive tables for exploring, querying, and analyzing structured data including predictions and ground truth. Visualize image classification, object detection, and segmentation results. W&B Sweeps: automated hyperparameter optimization. Grid search, random search, and Bayesian optimization. Run multiple experiments in parallel. Reports: create shareable, interactive reports combining code, visualizations, metrics, and markdown notes. Document experiments and share with stakeholders. Model evaluation: compare model predictions, identify failure cases, and measure model quality. Custom metrics and visualizations. Integrations: PyTorch, TensorFlow, Keras, JAX, Hugging Face, XGBoost, LightGBM, scikit-learn, Spark, Keras. Python/TypeScript. Proprietary with free tier.