Introducing the new Weights & Biases Workspace API
The new W&B Workspace API offers ML practitioners using Weights & Biases a way to programmatically create and customize saved workspace views
Created on July 15|Last edited on July 15
Comment
Personal workspaces
A Weights & Biases project provides every user with a personal dashboard populated with the tracked output from every experiment run. We call it a workspace. Since relevant metrics vary with each experiment and workspaces are not one-size-fits-all, Weights & Biases allows users to customize their personal workspaces and views, enabling them to visually organize content according to their specific needs. Each personal workspace reflects the priorities of its owner and it's accessible to the entire team, facilitating seamless collaboration and sharing of insights.
The Weights & Biases workspace provides a simple and intuitive way to add, remove, and rearrange sections, panels, charts, and other visualizations to create the perfect layout. Teammates can view and interact with, but not save changes to, other users’ personal workspaces, ensuring that original configurations remain intact while still allowing for collaborative exploration.
However, a user can copy and save another user's personal workspace as their own, allowing them to build on proven setups and streamline their workflow. If a ML engineer joins a project late or discovers that a peer has created an ideal personal workspace, they can replicate it with a single click. This allows users to quickly adopt effective setups, enhancing productivity and streamlining a collaborative workflow.
Saved views
While personal workspaces support knowledge sharing, they don’t facilitate collaborative workspace creation. Saved workspace views, however, enable teammates to collaboratively build, share, and update workspaces.
Experiment evaluation often requires reviewing results from various perspectives, as some models may perform better in certain aspects and worse in others. Project members can create multiple workspace views to focus on specific development methodologies, a particular set of model training parameters, or different evaluation criteria.
When new results are available from each run, any team member can update a workspace view with relevant logging data and new charts, providing fresh insights to the entire team and more streamlined insight sharing.
All updates in workspace views occur in real-time, ensuring everyone remains informed on the latest experiment iterations. After each run, sharing the latest results is effortless. Simply send a teammate the link to the saved workspace URL, allowing for quick and easy sharing for project collaborators.
Workspace API
Saved workspace views often provide value beyond their original projects, making it beneficial to reuse them for similar projects. Copying or migrating a workspace view no longer requires starting from scratch and tediously rearranging panels, sections, and charts manually.
The new Workspace API enables users to programmatically create, duplicate, and update workspace views with just a few lines of code. This API allows users to publish workspace view templates to any project quickly. By adding a code block, favorite customized workspace views can be instantly applied to new or existing projects, saving time and effort. This approach not only enhances efficiency but also fosters collaboration in a familiar, shared environment for the entire team.
Getting Started
It’s easy to get started using the W&B Workspace API. We’ve created a tutorial to help you get started programmatically creating and customizing your own saved workspace views. Just click the colab below and we'll walk you through it.
Add a comment
Tags: Articles, W&B Features
Iterate on AI agents and models faster. Try Weights & Biases today.