Скачать или смотреть Streamlined Efficiency: Unshackling Kubernetes Image Volumes for Rapid AI Model... E. Rey & Y. Yuan

Don't miss out! Join us at our next Flagship Conference: KubeCon + CloudNativeCon events in Hong Kong, China (June 10-11); Tokyo, Japan (June 16-17); Hyderabad, India (August 6-7); Atlanta, US (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

Streamlined Efficiency: Unshackling Kubernetes Image Volumes for Rapid AI Model and Dataset Loading - Esteban Rey, Microsoft & Yifan Yuan, AlibabaCloud

In this presentation, we will introduce a novel approach to utilizing Kubernetes’ new Image Volumes for quickly and efficiently loading large language models and extensive datasets. We will explain how streaming loading and open-source technologies speed up mounting Open Container Initiative (OCI) artifacts without packaging existing object storage blobs. This ensures effective usage of storage space and faster loading times.

Packaging large models and petabyte-level datasets into OCI artifacts presents two challenges:

1. Converting existing datasets is time-consuming.
2. Pulling time and disk space usage are unacceptable.

Our approach eliminates the need to convert existing data and uses streaming loading technology to mount image volumes without pulling. It ensures high performance for accessing numerous small files and loading large models, making it practical for new and demanding scenarios.