Watch the talks: Videos from Nebius AI Cloud Unveiled meetup
Watch the talks: Videos from Nebius AI Cloud Unveiled meetup
We wrapped up our first series of technical meetups with stops in Paris, London and San Francisco — each time providing a deep dive into AI Cloud and AI Studio, the work of Nebius’ in-house AI R&D team, as well as our contributions to open source. Adam Grzywaczewski, Senior Deep Learning Data Scientist at NVIDIA, has also joined us in London. Here, you can watch all the talks from the first event in the series.
How we build cloud for AI workloads
Gleb Kholodov, Head of Foundation Services at Nebius, shared insights into our hardware and software challenges, the decision-making process and the architecture behind Nebius AI Cloud.
NVIDIA’s talk: Laying foundations for the future of AI
Adam Grzywaczewski, Senior Deep Learning Data Scientist at NVIDIA, explored key highlights and guidance for choosing the best architecture for your AI projects.
Marrying Slurm and Kubernetes for workload management
Grigorii Rochev, Senior SRE, gave a talk on Soperator, the open-source Kubernetes operator for Slurm that we released six months ago. Soperator addresses the challenge of integrating Slurm with K8s, helping to manage the complexity of Slurm environments and compensate for the lack of native autoscaling. It also introduces additional features not available in either vanilla Slurm or vanilla Kubernetes.
What it takes to win the large-scale training game: we made the mistakes, so you don’t have to
Vasily Pantyukhin, our Head of Customer Experience, shared how to avoid the top mistakes and master best practices for scaling AI models.
Improving agentic systems with test-time computation
Boris Yangel ia a research engineer with more than a decade of experience leading complex AI projects and Head of AI R&D at Nebius. He spoke about his team’s recent research on combining guided search with agent inference, and how these techniques enable us to build better software engineering agents.
Inference, all you need to know about it
Last but not least, Head of Product Nikita Vdovushkin and Product Manager Roman Gaev spoke about their experience building Nebius AI Studio and its core Inference Service, which provides GenAI open-source models on a per-token basis. They also shared tips on choosing inference providers and models for your specific application.
Explore Nebius AI Studio
Videos
- How we build cloud for AI workloads
- NVIDIA’s talk: Laying foundations for the future of AI
- Marrying Slurm and Kubernetes for workload management
- What it takes to win the large-scale training game: we made the mistakes, so you don’t have to
- Improving agentic systems with test-time computation
- Inference, all you need to know about it
See also
Nebius opens pre-orders for NVIDIA Blackwell GPU-powered clusters
Nebius opens pre-orders for NVIDIA Blackwell GPU-powered clusters
We are now accepting pre-orders for NVIDIA GB200 NVL72 and NVIDIA HGX B200 clusters to be deployed in our data centers in the United States and Finland from early 2025. Based on NVIDIA Blackwell, the architecture to power a new industrial revolution of generative AI, these new clusters deliver a massive leap forward over existing solutions.
Kvax: Fast and easy-to-use Flash Attention implementation for JAX
Kvax: Fast and easy-to-use Flash Attention implementation for JAX
Today, we’re open-sourcing Kvax, our Flash Attention implementation based on JAX. 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.