What’s New With Weights & Biases
A glimpse into what we’ve been building
Created on May 23|Last edited on May 25
Comment
Security with deployment flexibility
At Weights & Biases, we know every tech stack is different, so we’ve built an extensible platform with integrations from all the major cloud providers and ML frameworks. That’s why we are thrilled to announce our new managed deployment: W&B Dedicated Cloud. W&B Dedicated Cloud is a Single-Tenant instance that W&B’s expert team provisions and manages on your behalf. You can securely connect your data to ensure it is properly isolated, while W&B maintains the infrastructure to modern high-security enterprise standards. With the added choice of cloud provider and region, deploying W&B has never been easier.
Security is a core aspect of our identity at Weights & Biases, and it affects every aspect of our business: from how we hire and onboard new employees, how we review, update, and test our code, to deploying our infrastructure. To understand the depth of our security processes take a look at our recently released Security Portal.
Our customers are at the forefront of each decision we make, and their data security is something we take seriously. From vulnerability management and data encryption to audits and SOC2 Type II compliance, the Weights & Biases platform is secure enough for the enterprise but flexible for the scale-up. The interoperable platform supports all major cloud, on-premises, and hybrid deployment models so you can meet enterprise security standards without compromising on velocity.
Learn more about our flexible deployment options or enterprise-grade MLOps at scale with Weights & Biases here.
What’s Coming
A Central Hub for Model Management and Collaboration
We are proud of the single system of record we’ve built for tracking and versioning every stage of the machine learning development cycle. As the ML workflow evolves, so too must the tools that support it. Weights & Biases is improving our support for model management with a new first-class product: Model Registry.
The W&B Model Registry provides centralized tooling for ML teams to:
- Version Models: Easily track every version of a model through training and development, browse model versions, and compare versions apples to apples on the same evaluation metrics and datasets
- Track Lineage: Document and reproduce the complete pipeline of model training and evaluation, from upstream dataset versions and data cleaning code to model training code and hyperparameters
- Manage Lifecycle: Move models through staging to production, with an easy central interface to communicate changes across the team
The Latest & Greatest of W&B
Dynamic visualization, precise comparison, and reproducibility
We’re constantly adding new features and support to the Weights & Biases platform, and we’re excited to showcase some of these awesome new additions that are live and ready for use—check out a few of our favorites below.
Weights & Biases Tables let you interactively query datasets and model predictions, customizing metrics and plots without needing to rerun code. A variety of data types—from audio to text to 3D molecular structures to any pandas DataFrame—can be visualized, sorted, filtered, grouped, and otherwise explored dynamically with Tables.
As the file system underlying Tables, Artifacts support automatic versioning and workflow traceability, updating the version with a smart diff of the folder contents whenever you log to the same Artifact name (which ultimately conserves storage space).
Whether you’re changing the preprocessing approach, sampling strategy, or contents of a dataset, or fine-tuning a new variant of a base architecture, Artifacts can structure your model development cycles, analysis, and workflows for precisely-aligned comparison and painless reproducibility, whether by your future self or collaborators. You can flexibly slice and evaluate your experiments and share custom views through Workspaces and beautiful Reports, where every visualization is deeply linked to the underlying code or metrics logged to W&B.
Dive into more details in this collection of Reports, or try a quick interactive notebook to log and explore your own Artifacts, Tables, or semantic segmentation models.
Add a comment
Iterate on AI agents and models faster. Try Weights & Biases today.