How to beginner · 3 min read

How to log metrics with Weights and Biases

Q: How to log metrics with Weights and Biases

Use the wandb Python package to log metrics by initializing a run with wandb.init() and then calling wandb.log() with a dictionary of metric names and values. This enables real-time tracking and visualization of metrics like loss and accuracy during model training.

Quick answer

Use the wandb Python package to log metrics by initializing a run with wandb.init() and then calling wandb.log() with a dictionary of metric names and values. This enables real-time tracking and visualization of metrics like loss and accuracy during model training.

PREREQUISITES

Python 3.8+
pip install wandb
Weights and Biases account (free tier available)
Set environment variable <code>WANDB_API_KEY</code> with your API key

Setup

Install the wandb package and configure your API key as an environment variable to authenticate your runs.

bash

pip install wandb

Step by step

This example shows how to initialize a Weights and Biases run, log metrics in a loop, and finish the run cleanly.

python

import os
import wandb

# Ensure your API key is set in the environment
# export WANDB_API_KEY=os.environ["WANDB_API_KEY"]

# Initialize a new W&B run
wandb.init(project="my-project", entity="my-team", config={"epochs": 5, "learning_rate": 0.001})

for epoch in range(5):
    # Simulate training metrics
    loss = 0.1 * (5 - epoch)
    accuracy = 0.2 * epoch

    # Log metrics to W&B
    wandb.log({"epoch": epoch + 1, "loss": loss, "accuracy": accuracy})

# Finish the run
wandb.finish()

output

wandb: Currently logged in as: your-username (use `wandb login --relogin` to force relogin)
wandb: Tracking run with wandb version 0.15.0
wandb: Run URL: https://wandb.ai/my-team/my-project/runs/xyz123
# Metrics logged for epochs 1 to 5 with loss and accuracy values

Common variations

Use wandb.init(mode="offline") to log metrics locally without syncing to the cloud.
Log additional data types like images or model checkpoints with wandb.log().
Integrate wandb logging inside training loops of frameworks like PyTorch or TensorFlow.

Troubleshooting

If you see wandb: ERROR API key not found, ensure WANDB_API_KEY is set correctly in your environment.
For network issues, check your firewall or proxy settings blocking wandb from syncing.
Use wandb.init(mode="offline") to debug without internet connectivity.

✅

Key Takeaways

Initialize a W&B run with wandb.init() before logging metrics.
Use wandb.log() to send metric dictionaries for real-time tracking.
Always set your WANDB_API_KEY environment variable for authentication.
Call wandb.finish() to properly close the run and upload data.
You can log various data types beyond scalars, including images and model files.

Verified 2026-04

Verify ↗