Скачать или смотреть Simplifying Generative AI Model Training on Kubernetes using Helm Charts - Ajay Vohra & Omri Shiv

Don't miss out! Join us at our next Flagship Conference: KubeCon + CloudNativeCon India in Hyderabad (August 6-7), and KubeCon + CloudNativeCon North America in Atlanta (November 10-13). Connect with our current graduated, incubating, and sandbox projects as the community gathers to further the education and advancement of cloud native computing. Learn more at https://kubecon.io

Simplifying Generative AI Model Training on Kubernetes using Helm Charts - Ajay Vohra & Omri Shiv, AWS

Training generative AI models on Kubernetes offers a wide range of frameworks, tools, and orchestration options. While this diversity fuels innovation, it also introduces significant complexity. In this talk, we present a Helm-based approach that simplifies AI model training using Kubeflow Training Operators. This method abstracts much of the underlying complexity while preserving flexibility in choosing training technologies. Our solution is accelerator-agnostic and provides a consistent YAML interface across various training frameworks. We’ll also introduce a new Kubeflow Pipeline component that enables the construction of complex, end-to-end training workflows using Helm charts. Through real-world examples, we’ll showcase training pipelines using Accelerate, Ray Train + Lightning, and NVIDIA’s NeMo-Megatron libraries. We’ll also demonstrate automatic scaling of accelerator infrastructure using Karpenter.