Weights & Biases

Weights & Biases (W&B) is a developer-first MLOps platform headquartered in San Francisco, California, specializing in machine learning experiment tracking, model visualization, and collaborative development. Founded in 2017, W&B raised 50 million in total funding across 5 rounds before being acquired by CoreWeave in March 2025 for .7 billion, though it continues to operate independently. The platform is used by over 700,000 ML practitioners and leading AI organizations including OpenAI, NVIDIA, Meta, Toyota, and Hugging Face for training and deploying production models. W&B's core capabilities include experiment tracking with automatic versioning of code, data, and hyperparameters, interactive visualization dashboards for model performance metrics, collaborative workspaces for team knowledge sharing, and model registry for versioning and deployment management. The platform excels at monitoring large-scale model training runs, providing real-time metrics, GPU utilization tracking, and system performance monitoring critical for debugging deep learning experiments. W&B integrates seamlessly with popular frameworks including PyTorch, TensorFlow, Keras, Hugging Face Transformers, and JAX, requiring just a few lines of code to instrument existing training scripts. The platform supports both cloud-hosted and self-hosted deployments for enterprises requiring on-premises data residency. W&B has become essential infrastructure for the generative AI era, used to train many leading foundation models and LLMs, with particular strength in tracking long-running distributed training jobs across hundreds of GPUs.