Accelerating ML Initiatives: What’s New at Weights & Biases
Our mission is to make the best tools for machine learning practitioners. Here's what we're building now.
Created on October 18|Last edited on October 25
Comment
Since the company’s inception, our vision at Weights & Biases has been to build the best tools for developers throughout the ML workflow. What started with a focus on experiment tracking quickly diversified into dataset versioning, model evaluation, data visualization, model training, pipeline management and more. From first ideas through model development – and ultimately monitoring and deployment – we’re helping practitioners build the best models and get them to production faster by making their ML workflow more efficient and collaborative than ever before.
While we may have expanded our footprint, we have remained laser-focused on making ML work in the real world, all on a platform that ML practitioners love using. Following our Series C and subsequent investment from NVIDIA, we're excited to unveil the latest in a slew of platform enhancements that will better serve our customers.
Before I talk about what’s new, a little on our investment focus. Our product strategy is guided by three main principles:
- Making delightful products that ML developers love
- Building additional ML workflow capabilities to achieve production ML
- Building an enterprise-grade platform
Each of these pillars is key to expanding the ML workflow and allowing us to better serve our users so they can make an impact with ML.
So what’s new? In short, a lot! Here are a couple of the highlights:
Building Additional ML Workflows
We offer a seamless, end-to-end MLOps experience from model prototyping to production, and these new capabilities provide even more flexibility across the ML workflow. We know ML practitioners face particular challenges when it comes to model management, including siloed projects, missing context, ad-hoc handoff and unclear lineage, as well as manual testing and ad-hoc retraining.
W&B Models is a central repository for your team’s model management, and it’s now available for you to try! Answer questions like “what’s the exact dataset that this model trained on?” and “how does this new candidate compare to the model in production?”
- Lineage: Capture the exact history of where a model came from, including upstream data preprocessing, exact dataset versions, and training code.
- Lifecycle: Coordinate centrally when models move between stages, such as upgrading a new candidate model to production.
The second big product update in our ML workflows pillar is W&B Launch. W&B Launch is the ultimate tool for ML workload orchestration. It allows you to build reproducible ML workflows for agility, team efficiency, and optimized resource utilization. Additionally, you can launch jobs directly within your existing infrastructure. Containerized workflows (Docker and Kubernetes) provide easy reproducibility in any environment, at scale, and rule-based, time-based triggers provide CI/CD automation of the ML workflow. Plus, the simple UI makes it easy to build a bridge between ML Practitioner and MLOps Engineer, inspiring collaboration.
Check out this portion of our recent event here:
Making Delightful Products that ML Developers Love
This pillar is focused on investing in usability, performance and use case coverage to increase your productivity.
To start, we’ve enhanced our Time Series Support. That means you can optimize the accuracy of their time series models, quickly back-test performance, compare with baseline models, and ultimately visualize their results exactly the way they want. You can also flexibly configure any visualization for time series models and share it in a W&B report. Whether you’re a CFO forecasting sales for the rest of the year or a researcher modeling climate change, our time series visualization tools help you evaluate, compare, and improve your time series models to better predict the future.
Building on our best-in-class support of all media types, we enhanced our 3D Bounding Boxes. You can log 3D objects and LiDAR point clouds with 3D bounding box annotations, prediction scores, and camera viewpoint, as well as easily render in W&B to explore your media data interactively. To see what's possible, check out Stacey Svetlichnaya’s 3D Objects in Driving Scenes W&B Report.
Easily visualize and annotate 3D objects and scenes using W&B
We also introduced W&B Profile Pages. These customizable profile pages let you proudly display your ML portfolio to share your hard work with the world. Like a GitHub profile for a code developer, the new profiles allow you allow you to curate and showcase your W&B projects–enabling you to build an “ML Portfolio” that will inspire and facilitate collaboration within the ML community.
You can watch this portion of our recent webinar here:
Building an Enterprise-Grade Platform
We're continuously investing in flexible, easy deployment options and enterprise security for peace of mind. The first in our enterprise-level upgrades is an enhancement to the W&B Dedicated Cloud, which provides a secure, single-tenant managed environment for deploying Weights & Biases. We’re dedicated to meeting customers where they are, and with more and more enterprises relying on Microsoft enterprise applications and services, we added support for Microsoft Azure. The W&B Dedicated Cloud is now available on the three major cloud platforms – Azure, GCP, and AWS. We also added Single Sign On (SSO) authentication with Microsoft to quickly allow Microsoft users to access W&B without sacrificing security.
The W&B Security Portal (security.wandb.ai) provides a centralized hub to access all W&B security frameworks and certifications. Customers around the world entrust sensitive data to W&B, and nothing is more important to us than honoring our custodial commitments to protect this data. For our customers in highly regulated industries, the portal ensures fast compliance with security standards, saving you time and money.
For even more insight into the new features, check out our press release. If you’re an existing Weights & Biases user or want to become one, please visit our website and contact us to learn more.
Add a comment
Tags: Articles
Iterate on AI agents and models faster. Try Weights & Biases today.