Weights & Biases

Weights & Biases (W&B) has become the standard experiment tracking platform in machine learning research and production, used at major AI labs, research universities, and enterprise ML teams worldwide. The platform addresses a practical problem that anyone who has trained machine learning models recognises: keeping track of which hyperparameters, datasets, and code versions produced which results.

The core feature is experiment tracking. By integrating a few lines of Python code into a training script, W&B automatically logs metrics, hyperparameters, model checkpoints, and system metrics throughout the training run. The resulting dashboard allows comparison of multiple experiment runs side by side to identify which configurations produce the best results, visualise training curves, and understand model behaviour.

W&B Weave, added more recently, extends the platform to LLM application development, providing tracing, evaluation, and monitoring for AI applications built with language models. This positions W&B as an evaluation and observability layer for the LLM application development workflow alongside traditional ML experiment tracking.

The collaboration features allow research teams to share experiment results, compare findings, and work together on model development in a way that is substantially better than sharing screenshots of training curves in Slack.

W&B is free for individual use with unlimited projects, which is how it has built a large community of researchers and practitioners. The Teams plan at $50/user/month adds collaboration features suitable for professional ML teams.