Weights & Biases (wandb)

Weights & Biases (wandb)

Weights & Biases is the open-source ML experiment tracking library used by tens of thousands of researchers and ML engineers. It logs training runs to a central dashboard, enabling comparison of experiments, hyperparameter sweeps, model versioning, and dataset versioning — all from a two-line code addition.

Key Features

• Automatic tracking: metrics, hyperparameters, code, system (GPU/CPU/memory)
• Interactive dashboards: compare any runs with custom plots
• Sweeps: distributed hyperparameter search (grid, random, Bayesian)
• Artifacts: version datasets, models, and evaluation tables
• W&B Tables: analyze model predictions with rich media (images, audio, text)
• Self-hosted option (wandb server) for on-prem deployments

Quick Start

import wandb

wandb.init(project="my-project", config={"lr": 0.001, "epochs": 10})

for epoch in range(10):
    loss = train_one_epoch()
    wandb.log({"epoch": epoch, "loss": loss})

wandb.finish()

Install via ai-supply

npx ai-supply add wandb-experiment-tracking

Curated mirror of the open-source Weights & Biases (MIT). Get it from the source.