Skip to content
BENIBIX Learn

Weights & Biases (W&B)

Also known as Weights and Biases, W&B, wandb

A web tool that charts each training run live. A team can then compare them.

Think of it like

It is like a scoreboard at a race. You see each runner, live, side by side.

Example

Lucy trains a model overnight on four GPUs. In the morning she opens W&B and sees that run 7's loss stopped falling after two hours.

Why it matters

Training runs take hours and cost money. Seeing a bad run early means you can stop it and save both.

Definition

Weights & Biases is a paid platform for tracking machine learning experiments. A few lines in a training script send each run's settings and scores to its website. There, a team sees live charts of each run and compares them side by side. It also stores datasets and models, and traces apps built on language models. CoreWeave, a GPU cloud company, bought it in 2025.

Example code

python
import wandb

run = wandb.init(project="house-prices")
run.log({"loss": 0.42})
Start a run in a project, then send a score. It shows up on a live chart on the website.
Weights & Biases sits on the track runs box, where each run is logged and compared.

Common mistake

Logging so many charts that nobody reads them. Pick the few numbers that tell you a run is going well.

Key takeaways

  1. 1
    Live charts. Watch a run while it trains.
  2. 2
    Side by side. Compare runs to find the best one.
  3. 3
    Team view. Everyone sees the same results.