Engineering Insights

Policy Gradient with PyTorch

Identifying slow Rails queries with sqlcommenter

In a large Rails application, it can be tricky to track down where a slow query is coming from in the app. Solutions to this problem have been steadily improving. First, we had the Marginalia gem, which adds comments to all your queries. This allows you to see which controller or job a query came from by reading your logs. As of Rails 7, Marginalia has become a native feature to Rails. No more gem needed. While this is a great improvement, we can still take our Rails query comments one step furt

Liftoff! How to get started with your first ML project 🚀

Next.js 12.2

Next.js 12.2 introduces stable Middleware and On-Demand ISR, experimental Edge SSR and API Routes, and more!

Announcing Vitess 14

We are pleased to announce the general availability of Vitess 14. In this new release, major improvements have been made in several areas of Vitess, including usability and reliability: Online DDL is now GA Gen4 planner is the new default planner VTAdmin and VTOrc are officially in beta with Vitess 14 Usability Command-line syntax deprecation This release marks the beginning of Vitess standardizing its command-line and flags syntax. Some former syntaxes have been deprecated and will break in the

Vercel Edge Middleware: Dynamic at the speed of static

Since we announced Middleware last October, we’ve seen 80% month-over-month growth and over 30 billion requests routed through Edge Middleware on Vercel during public beta. Customers like Vox Media, Hackernoon, Datastax, and HashiCorp are using Edge Middleware to have complete control over routing requests in their Next.js applications. With the release of , Vercel Edge Middleware for Next.js is now generally available (GA) for all customers. Edge Middleware is also available for frameworks—now

DALL·E 2 pre-training mitigations

In order to share the magic of DALL·E 2 with a broad audience, we needed to reduce the risks associated with powerful image generation models. To this end, we put various guardrails in place to prevent generated images from violating our content policy.

Accelerate Large Model Training using DeepSpeed