Weights & Biases

Weights & Biases (W&B) is the leading MLOps platform for machine learning experiment tracking, hyperparameter optimization, dataset versioning, and model evaluation. Used by over 1 million ML practitioners at organizations including OpenAI, NVIDIA, Stability AI, and Johns Hopkins, W&B is the de facto standard for tracking ML experiments across any framework (PyTorch, TensorFlow, JAX, scikit-learn). W&B's core product is the Runs dashboard — a web interface that automatically captures training metrics (loss, accuracy, learning rate), hardware utilization (GPU/CPU/memory), code diffs, hyperparameters, and output artifacts for every training run. Runs are grouped into Projects, and comparisons across runs (different learning rates, batch sizes, architectures) are done through interactive parallel coordinate plots and scatter charts. W&B Sweeps automates hyperparameter search using Bayesian optimization, grid search, or random search across distributed training runs. W&B Artifacts provides dataset and model versioning with lineage tracking — recording exactly which dataset version and code commit produced each model checkpoint.