Announcing new capabilities in Vertex AI Training for large-scale training

Building and scaling generative AI models requires significant resources, often leading to cumbersome infrastructure management. Developers encounter challenges like managing job queues and provisioning clusters, which can slow down model development. Vertex AI Training offers expanded capabilities to streamline the process of building large, differentiated models. The new managed training features leverage Google Cloud's AI infrastructure, including Cluster Director for a managed Slurm environment. These features include pre-built data science tools and optimized recipes for specialized model building using frameworks like NVIDIA NeMo. Vertex AI Training offers various customization options, from cost-effective tunings like LoRA to large-scale training on custom clusters. It is organized around flexible, self-healing infrastructure, comprehensive data science tooling, and integrated recipes and frameworks. Customers like Salesforce and AI Singapore are already using Vertex AI Training to improve model performance and efficiency. Vertex AI Training also provides options for those needing more control, available through Google Compute Engine or Google Kubernetes Engine. Vertex AI Training offers the infrastructure and expertise to make AI a powerful competitive asset.