DoK #61 Perfecting Machine Learning Workloads on Kubernetes // Lars Suanet


Manage episode 296540611 series 2865115
โดย Bart Farrell และถูกค้นพบโดย Player FM และชุมชนของเรา -- ลิขสิทธิ์นี้เป็นของผู้เผยแพร่ ไม่ใช่ Player FM โดยมีการสตรีมเสียงโดยตรงจากเซิร์ฟเวอร์ผู้เผยแพร่ กดปุ่มติดตามเพื่อติดตามการอัพเดทใน Player FM หรือวาง URL ฟีดนี้ไปยังแอพพอดคาสท์อื่น

Abstract of the talk…

More and more applications are powered by Machine Learning (ML) models. Where the gap between Software Engineers and a Production environment on Kubernetes is already big, the gap between Data Scientists and that same production environment is enormous. In this talk, we will provide you with a framework for translating ML requirements into infrastructural requirements and concrete Kubernetes resources. In the first half of this talk, we will discuss how ML applications are different from most other applications, how ML workloads are structured and how ML requirements translate into Kubernetes resource configurations. In the second half of the talk, we will put this theory into practice. We will do a live demonstration of an ML Deployment on Kubernetes using Istio, Knative and Kubeflow Serving.


Lars Suanet is a Software Engineer at Deeploy. With his background in Computer Science and his interest in AI, he tries to bridge the gap between Data Scientists and DevOps. His personal interests are Chinese culture, Distributed systems, Meditation and Plants.

123 ตอน