Code beginner · 3 min read

How to use wandb.init in Python

Q: How to use wandb.init in Python

Use wandb.init() in Python to initialize a new run for experiment tracking, specifying parameters like project and entity to organize your runs.

Direct answer

Use wandb.init() in Python to initialize a new run for experiment tracking, specifying parameters like project and entity to organize your runs.

Setup

Install

bash

pip install wandb

Env vars

WANDB_API_KEY

Imports

python

import wandb

Examples

inwandb.init(project='my-project', entity='my-team')

outInitialized a new W&B run under project 'my-project' and entity 'my-team'.

inwandb.init(project='image-classification', config={'lr': 0.001, 'epochs': 10})

outStarted a run logging hyperparameters lr=0.001 and epochs=10.

inwandb.init(name='experiment-42', reinit=True)

outCreated a new run named 'experiment-42', allowing multiple runs in the same script.

Integration steps

Install the wandb Python package via pip.
Set your W&B API key in the environment variable WANDB_API_KEY.
Import wandb in your Python script.
Call wandb.init() with parameters like project, entity, and config to start a run.
Log metrics and artifacts during training using wandb.log() and wandb.save().
Finish the run automatically or call wandb.finish() to close it.

Full code

python

import os
import wandb

# Ensure your WANDB_API_KEY is set in environment variables
# os.environ['WANDB_API_KEY'] = 'your_api_key_here'  # Do NOT hardcode in production

# Initialize a new W&B run
run = wandb.init(
    project='my-sample-project',
    entity='my-team',
    config={
        'learning_rate': 0.001,
        'epochs': 5,
        'batch_size': 32
    },
    name='experiment-001',
    reinit=True
)

print(f"W&B run initialized with id: {run.id}")

# Example of logging metrics
for epoch in range(run.config.epochs):
    loss = 0.1 * (run.config.epochs - epoch)  # dummy loss
    wandb.log({'epoch': epoch + 1, 'loss': loss})

# Finish the run explicitly (optional)
wandb.finish()

output

W&B run initialized with id: 123abc456def
# (followed by metrics logged to the W&B dashboard)

API trace

Request

json

{"project": "my-sample-project", "entity": "my-team", "config": {"learning_rate": 0.001, "epochs": 5, "batch_size": 32}, "name": "experiment-001", "reinit": true}

Response

json

{"id": "123abc456def", "name": "experiment-001", "project": "my-sample-project", "entity": "my-team", "state": "running"}

Extractrun.id or run.name for run identification

Variants

Streaming metrics logging ›

Use when you want to log metrics incrementally during training or evaluation.

python

import wandb

run = wandb.init(project='streaming-project')

for step in range(10):
    metric = step * 2
    wandb.log({'step': step, 'metric': metric})

wandb.finish()

Async initialization with callback ›

Use when integrating W&B in asynchronous or multi-threaded environments.

python

import wandb
import threading

def on_run_start():
    print('Run started asynchronously')

run = wandb.init(project='async-project', callbacks=[on_run_start])
wandb.finish()

Initialize without project (default project) ›

Use when you want to log runs under your default project or when project is set in W&B settings.

python

import wandb

run = wandb.init(entity='my-team')
print(f"Run ID: {run.id}")
wandb.finish()

Performance

Latency~100-300ms for wandb.init network handshake

CostFree for basic usage; check https://wandb.ai/pricing for enterprise features

Rate limitsNo strict rate limits for <code>wandb.init()</code>, but excessive logging may be throttled

Log only essential metrics to reduce bandwidth and storage.
Use <code>reinit=True</code> to manage multiple runs in one script efficiently.
Batch metric logs with <code>wandb.log()</code> to minimize API calls.

Approach	Latency	Cost/call	Best for
Basic `wandb.init()`	~100-300ms	Free	Standard experiment tracking
Streaming metrics logging	Low per log (~10-50ms)	Free	Real-time metric updates
Async initialization	~100-300ms + threading overhead	Free	Concurrent or async workflows

✓

Quick tip

Always set your <code>WANDB_API_KEY</code> as an environment variable before calling <code>wandb.init()</code> to authenticate your runs securely.

⚠

Common mistake

Beginners often forget to set the <code>WANDB_API_KEY</code> environment variable, causing authentication failures when calling <code>wandb.init()</code>.

Verified 2026-04

Verify ↗