How to beginner · 3 min read

How to use Weights and Biases for experiment tracking

Q: How to use Weights and Biases for experiment tracking

Use Weights and Biases (wandb) by installing the wandb Python package, initializing a run with wandb.init(), and logging metrics or artifacts with wandb.log(). This enables automatic experiment tracking, visualization, and collaboration.

Quick answer

Use Weights and Biases (wandb) by installing the wandb Python package, initializing a run with wandb.init(), and logging metrics or artifacts with wandb.log(). This enables automatic experiment tracking, visualization, and collaboration.

PREREQUISITES

Python 3.8+
pip install wandb
Weights and Biases account (free tier available)
Basic Python knowledge

Setup

Install the wandb package and log in to your W&B account to enable experiment tracking.

bash

pip install wandb
wandb login

output

Enter your API key from https://wandb.ai/authorize:
Successfully logged in to Weights & Biases!

Step by step

This example shows a simple training loop where metrics are logged to W&B for tracking.

python

import wandb

# Initialize a new W&B run
wandb.init(project="my-ml-project", entity="your-username")

for epoch in range(3):
    # Simulate training metrics
    loss = 0.1 * (3 - epoch)
    accuracy = 0.8 + 0.05 * epoch

    # Log metrics to W&B
    wandb.log({"epoch": epoch, "loss": loss, "accuracy": accuracy})

wandb.finish()

output

wandb: Currently logged in as: your-username
wandb: Tracking run with id: <run_id>
wandb: Run summary and metrics logged successfully.

Common variations

Use wandb.init(resume=True) to resume interrupted runs.
Log model checkpoints or datasets as artifacts with wandb.Artifact.
Integrate with frameworks like PyTorch Lightning or TensorFlow using built-in W&B callbacks.
Use asynchronous logging or customize the dashboard with tags and notes.

Troubleshooting

If you see wandb: ERROR API key not found, run wandb login again and verify your API key.
For network issues, check your firewall or proxy settings blocking W&B servers.
If metrics don’t appear, ensure wandb.log() is called inside your training loop.

✅

Key Takeaways

Install and authenticate with wandb login before tracking experiments.
Initialize runs with wandb.init() and log metrics using wandb.log() inside training loops.
Use artifacts to track datasets and model files for reproducibility.
Leverage integrations with popular ML frameworks for streamlined tracking.
Troubleshoot common issues by verifying API keys and network connectivity.

Verified 2026-04

Verify ↗