How to beginner · 3 min read

How to use wandb.log

Quick answer

Use wandb.log to record metrics and other data during your training loop by passing a dictionary of key-value pairs. This function sends the logged data to your Weights & Biases project for real-time tracking and visualization.

PREREQUISITES

Python 3.8+
pip install wandb
Weights & Biases account (free tier available)
wandb login or API key configured

Setup

Install the wandb Python package and authenticate your environment to enable logging.

Install with pip install wandb
Run wandb login in your terminal and enter your API key from Weights & Biases

bash

pip install wandb
wandb login

Step by step

Initialize a W&B run, then use wandb.log inside your training loop to log metrics such as loss and accuracy.

python

import wandb
import time

# Initialize a new run
wandb.init(project="my-project")

for step in range(1, 6):
    # Simulate metrics
    loss = 0.1 * step
    accuracy = 0.8 + 0.04 * step

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

    print(f"Logged step {step}: loss={loss}, accuracy={accuracy}")
    time.sleep(1)  # Simulate time delay

output

Logged step 1: loss=0.1, accuracy=0.84
Logged step 2: loss=0.2, accuracy=0.88
Logged step 3: loss=0.30000000000000004, accuracy=0.92
Logged step 4: loss=0.4, accuracy=0.96
Logged step 5: loss=0.5, accuracy=1.0

Common variations

You can log any numeric or media data by passing a dictionary to wandb.log. For example, log images, histograms, or custom objects. You can also log asynchronously or batch multiple metrics in one call.

python

import wandb
import numpy as np
from PIL import Image

wandb.init(project="my-project")

# Log multiple metrics at once
wandb.log({"loss": 0.25, "accuracy": 0.9, "epoch": 1})

# Log an image
image = Image.new("RGB", (64, 64), color="red")
wandb.log({"sample_image": wandb.Image(image)})

# Log a histogram
values = np.random.randn(1000)
wandb.log({"weights_histogram": wandb.Histogram(values)})

Troubleshooting

If wandb.log does not appear to send data, ensure you called wandb.init() before logging.
If you see authentication errors, run wandb login again or set the WANDB_API_KEY environment variable.
For slow logging, consider batching metrics or using commit=False in wandb.log and calling wandb.log with commit=True less frequently.

✅

Key Takeaways

Always call wandb.init() before logging metrics with wandb.log.
wandb.log accepts a dictionary of key-value pairs representing metrics or media.
Use wandb.log inside your training loop to track metrics in real time.
You can log images, histograms, and other media types with wandb.log.
Troubleshoot by verifying authentication and proper initialization.

Verified 2026-04

Verify ↗