Nebius monthly digest, February 2025

We are proud to be the Platinum sponsor of NVIDIA GTC 2025 — the premier developer conference at the heart of AI. Don’t miss your chance to meet our leaders, architects and tech experts. Visit our booth #809 and get NVIDIA GPU credits for your projects.

Nebius’ AI, hardware and development leads will also give two tech talks:

• From zero to scale: how to build an efficient AI cloud platform from scratch
• Enabling the agent-first future: advancing test-time search and compute infrastructure for agentic systems

And don’t forget to join us for an amazing AI Kingdom Afterparty.

AI Studio launches fine-tuning and showcases customer successAI Studio launches fine-tuning and showcases customer success

• A newly launched fine-tuning on Nebius AI Studio allows you to transform leading open-source models into specialized solutions tailored to your needs.
• Read how Chatfuel used Nebius AI Studio to leverage a cascade of Llama-405B models along with a custom SDK and achieved significant efficiency.

Managed Service for PostgreSQL® is now in GAManaged Service for PostgreSQL® is now in GA

• Managed PostgreSQL is now generally available to all Nebius users. We’ve been improving the service to empower AI innovators with a reliable tool for storing structured data in the cloud. As the service transitions to GA, we are introducing pricing and SLAs.

Kvax: Fast and easy-to-use Flash Attention implementation for JAXKvax: Fast and easy-to-use Flash Attention implementation for JAX

• We’re open-sourcing Kvax, our Flash Attention implementation. Designed for efficient training with long sequences, Kvax supports context parallelism and optimized computation of document masks. It outperforms many other Flash Attention implementations in long-context training with dense packing, achieving state-of-the-art performance.

Supporting research: award and creditsSupporting research: award and credits

• AI Discovery is our brand new annual award for startups that are using AI in drug discovery, biotech, genomics and healthtech. Apply until April 30, 2025, and compete for $100,000 in AI Cloud credits.

• Introducing Nebius research credits program. Eligible specialists can now gain access to AI Cloud or Studio inference tokens, helping in modern studies and tackling scientific challenges.

Sharing our own and our customers’ experiencesSharing our own and our customers’ experiences

• The final stop in our ‘Nebius AI Cloud Unveiled’ series will be in San Francisco! Join us on March 13 for a deep dive into our AI Cloud Accelerated by NVIDIA. Our AI developers will share insights on how we build the cloud and contribute to the AI field and open source.

• Read how Positronic Robotics trains ML models on Nebius AI Cloud, developing tools that allow robots to handle cleaning tasks more effectively than humans.

And as always, updates in our tech docsAnd as always, updates in our tech docs

• For GA, Managed PostgreSQL got a big documentation overhaul. With the new articles, you can learn how to manage and connect to databases, work with users and extensions, and migrate or replicate your data from external clusters to Nebius AI Cloud.

• Two new tutorials added to Compute. A quick-start tutorial focuses on hosting your first LLM on Compute virtual machines. And if you need to secure network access to your VMs, deploy a jump server with WireGuard.

• Choosing the right storage for your VMs is now easier with new sections dedicated to performance of disks and shared filesystems. TL;DR: larger disks and filesystems perform better — and our guide on resizing storage volumes can help you get that extra edge.

• You have to see performance to believe it. In Nebius AI Cloud, you can see performance on monitoring dashboards in the console, which we’ve started to cover in the documentation. Learn about dashboards and metrics for VMs and storage volumes in Compute, Managed Kubernetes clusters and nodes and Managed PostgreSQL clusters.