Nebius

Here’s what Matt Weigand, Partner at Accel who led the $700-million investment round we announced yesterday, says about Nebius and our technology. Matt will soon be joining our board. #quotes #AI #investment #funding #raised

Today, we’re excited to share our investment in Nebius, a full-stack AI platform built for intensive AI & ML workloads. As AI reshapes industries and drives innovation, reliable, scalable infrastructure has become paramount. Nebius, with its years-long expertise in building one of the top tech infrastructures, delivers a vertically integrated AI-native solution optimized from data center to cloud platform. We’ve been fortunate to know the Nebius team since its early days, and we’re thrilled to partner with founder Arkady Volozh and the entire team in fueling this era of rapid AI innovation. Read more from our partner, Matt Weigand, on today’s news #AccelFamily https://lnkd.in/gHr52M2t

Nebius raises $700 million from investors including Accel, NVIDIA and Orbis Investments. The investment supports our previously announced plans to further build out our full-stack AI infrastructure – including large-scale GPU clusters, cloud platforms and tools and services for developers – for AI pioneers globally. Read the full story: https://lnkd.in/eydgaqN3 #AI #investment #funding #raised

Register for a hands-on webinar on fine-tuning LLMs with MLflow: https://lnkd.in/emRbbgnP We’ll explore how MLflow can streamline your LLM fine-tuning workflow. You’ll be able to learn best practices for experiment tracking, model management and collaboration in LLM projects with MLflow’s powerful features. For who: - ML engineers and data scientists working with LLMs - MLOps engineers managing LLM deployments - Technical leads evaluating MLOps tools for LLM projects - Teams transitioning from prototype to production LLM applications You will learn: - Practical approaches to organizing LLM fine-tuning experiments. - How to make your fine-tuning process reproducible and traceable. - Best practices for model versioning and management. - Real-world examples of MLflow integration in LLM projects. - Streamlining team collaboration in LLM development. Where: Zoom. You will receive the link after you register on the Nebius website: https://lnkd.in/emRbbgnP #webinar #LLMs #largemodels #finetuning

Over the past two weekends, Nebius supported hackathons organized by Meta and Cerebral Valley in London and Toronto. We provided GPU compute for participants and extended access for the top three winners. In London, 312 hackers participated in the 36-hour event, submitting 59 projects across three key areas: accelerating clean and efficient energy; supporting delivery of public services and healthcare; and breaking down barriers to opportunity. Our own Alexey Tsishevski gave a brief talk and served as a judge in the initial round. The final round was held at Meta's Kings Cross offices, where six finalists presented to a packed audience of participants, government officials and other guests. Karina Zainullina, an ML Engineer from Nebius’ AI R&D team, was our final-round judge. Of the 59 submissions, 19 projects used Nebius resources, including all three winning teams: 🥇 Guardian, saving lives with rapid and accurate A&E triaging. 🥈 Gripmind. In their installation, Llama 3.2 controlled a robotic arm, enabling real-world robotics ops and BCI-based control. 🥉 Pharmallama. A locally-run, privately fine-tuned Llama 3.2 3B quantized model has been trained on a curated pharmacy knowledge dataset to provide medication advice and collaborate with pharmacists securely via iPhone. A week later in Toronto, 230 participants signed up for the second hackathon. The challenges focused on “AI as a public good,” with many submissions addressing accessibility and education. 34 projects out of 40 submitted in total, including the second- and third-place winners, leveraged Nebius resources. Andrei Meganov served as both a speaker and judge. The three winners in Canada were: 🥇 Circuit XR, an educational VR startup that allows you to take a hand-drawn electrical circuit, visualize it in AR using Unity and play with it: unplug a wire, flip a switch, etc. 🥈 EyeSpeak, an original gaze-driven interface for computer control with AI-assisted text entry and text-to-speech to assist paralyzed people. 🥉 Fraudy, a phone call companion that analyzes calls in real-time, alerting users to potential fraud risks. Special thanks to our architects Cyril Kondratenko and Liran Jdanov for providing GPU access and supporting participants throughout the events. And congrats to all the teams! We look forward to partnering with Cerebral Valley again in 2025. Here’s a video created by CV in London to capture the memories.

Efficient drug discovery is all about achieving those “eureka!” moments. For many laboratories in recent years, a reliable AI infrastructure has become the catalyst behind such moments. AI-powered platforms can generate thousands of molecular structures and screen their potential as drug candidates much faster than traditional methods. As we advance more and more in this field, we’ve noticed several cutting-edge organizations pushing the limits of R&D: Atomwise leverages AI to predict compounds’ binding affinity and interaction with disease-related proteins, focusing on structure-based drug design. If they fit together, it’s a hit! By optimizing key physicochemical properties such as stability and molecular interactions, Chemistry42 by Insilico Medicine ensures AI-generated molecules are both effective and viable for further development. Our client Genesis Therapeutics accelerates drug discovery by nailing down the most promising compound candidates for synthesis and testing, running repeated cycles of AI-powered iterations. Patients are at the heart of Exscientia’s methods to streamline drug development. Patient tissue data is key to define optimal profiles for AI-designed drugs, leading to better clinical conditions in cancer patients. #drugdiscovery #AI #biotech #RnD #healthcare

🎯 Excited to share insights from my talk at DSC Europe 2024 on "Structuring Unstructured Data to Boost Computer Vision and GenAI Applications at Scale"! 🔍 We dove deep into unstructured data management and how it powers AI applications. 🚀 Key highlights: • AI and Data Trends - Unstructured Data is a new gold for better AI • Toolset to enrich, transform, and analyze unstructured data - requires scaling and distributed processing • DataChain is an open-source tool to enrich, transform and analyze unstructured data • Use case: Streamlining PDF processing and LLM evaluation • Use case: Enhancing Computer Vision in Fashion • Use case: Managing complex Video Datasets with Frame-Level Annotations for Sport & Fitness applications 🙏 Thanks to everyone who joined and engaged in the discussion! Your questions and insights made the session even more valuable. Many thanks to the DataChain team, Dmitry Petrov, Ivan Shcheklein, David Berenbaum, and Tibor Mach for the opportunity to work together and for use case examples. Good luck with the DataChain tool! Looking for more starts ⭐ on GitHub: https://lnkd.in/dDxYN8xe 🙌 #AI #DataChain #ComputerVision #GenerativeAI #MachineLearning #DataEngineering

In today’s guide on our blog, we’ll build an AI-powered financial planner using the Meta Llama 3.1 70B model via Nebius AI Studio and a full-stack Next.js app: https://lnkd.in/etmCeb8v This solution works seamlessly whether you’re using the Pages Router or the App Router, as much of the customization relies on setting up a custom Next.js server. #demo #opensource #API #guides #appbuilders

Throwback to this fall’s OCP Global Summit, where our hardware R&D team leaders Igor Z. and Oleg Fedorov gave a talk on Nebius’ in-house server and rack design. Hit play to see the full presentation. Our hardware designs are based on several concepts and principles from the Open Compute Project Foundation, so this year, we came full circle by bringing our own developments to the OCP community. Igor and Oleg shared ideas and specs that help us deliver efficient, robust GPU compute, while Anna Amelechkina supported the guys from backstage. This was the first time we revealed what’s inside our server to such a wide audience of hardware experts. Thanks to everyone who attended and provided feedback afterward. #OCP #hardware #servers #RnD

Before any training comes data preparation. We’ve outlined a high-level overview of the kind of pipeline you can build on Nebius to collect and prepare data for ML models. The datasets can then, of course, be used for pre-training and fine-tuning on modern AI-tailored GPUs, all in one cloud. The software stack shown in the image highlights the tools we provide. With these, you can process unstructured data for multi-modal training or manage your structured data in one of our databases. Learn more about our compute environment, storage types and applications: https://lnkd.in/dzBagyXJ. Scroll down to explore resources and articles we’ve created based on our in-house AI R&D expertise. Even if you’re preparing datasets using a different stack, our experience can still prove valuable. #datapreparation #datacollection #databases #storage