Use Local to host the W&B App on your own servers, either private cloud or on-prem. To quickly test Local, you can even run the app from your laptop. Local gives you the power of W&B tracking and visualization without sending data to the W&B cloud servers.

We also offer W&B Enterprise Cloud, which runs a completely scalable infrastructure within your company's AWS or GCP account. This system is a good choice for massively scalable experiment tracking.

Features

• Keep your data safe on your own company's network

• Integrate with your company's authentication system

• Premier support by the W&B engineering team

Quickstart

1. Run everything locally with W&B Docker

On any machine with Docker and Python installed, run

wandb local

to pull the latest version of our Docker image from Docker Hub and run the container on your system. The full command, in case you want to modify the port or other details, is

docker run --rm -d -v wandb:/vol -p 8080:8080 --name wandb-local wandb/local

This runs a local instance of wandb on your machine: all your training logs and metrics will be stored locally inside /vol and visible at http://localhost:8080 on your machine.

Once the container spins up, you will be able to create a new user account and will be prompted to login to wandb by fetching the API key for this new account from http://localhost:8080/authorize. To start logging locally, install the wandb Python package (e.g. from the command line, pip install wandb). Then run any script like the toy example below (or any of these).

import wandb
​
# 1. Start a new W&B experiment run
wandb.init(project="my_project")
​
# 2. Save any model configuration
wandb.config.learning_rate = 0.01
​
# [train your model]
​
# 3. Log metrics to visualize model performance
for i in range(10):
	wandb.log({"loss" : 10-i, "acc" : float(i)/10.0})

​

2. Create and scale a shared instance

This private instance of wandb is excellent for initial testing. To enjoy the powerful collaborative features of wandb, you will need a shared instance on a central server, which you can set up on AWS, GCP, Azure, Kubernetes, or Docker.

​

3. Set up a new user account

For the individual local instance, you can create a new user account when you first spin up a local instance. You can see and edit account settings at http://localhost:8080/settings.

For the shared local instance, the default user is local and the default password is perceptron. Follow the prompts to reset your password.

To change your login, you can always run

wandb login --relogin

​

4. Control where to log: local or wandb cloud

If you're running wandb on multiple machines or switching between a local instance and the wandb cloud, there are several ways to control where your runs are logged. If you want to send metrics to the shared local instance and you've configured DNS, you can

• set the host flag to the address of the local instance whenever you login:

 wandb login --host=http://wandb.your-shared-local-host.com

• set the environment variable WANDB_BASE_URL to the address of the local instance:

export WANDB_BASE_URL = "http://wandb.your-shared-local-host.com"

In an automated environment, you can set the WANDB_API_KEY which is accessible at wandb.your-shared-local-host.com/authorize.

To switch to logging to the public cloud instance of wandb, set the host to api.wandb.ai:

wandb login --host=https://api.wandb.ai

or

export WANDB_BASE_URL = "https://api.wandb.ai"

You can also switch to your cloud API key, available at https://wandb.ai/settings when you're logged in to your cloud-hosted wandb account in your browser.

​

5. Request a team license

Contact us at [email protected] to request an upgrade to your license. We can give you a free trial of up to 3 users in your local install. We also provide free teams for academics and open source project teams. Reach out to discuss options.