Evolving Deep Learning Platform with Knative
summary
Describes how Baidu used Knative components to rebuild its internal deep learning platform: using Knative Eventing to automate the pipeline from training to inference, Knative Serving to provide smart routing and autoscaling for inference services, and Knative Build to automate creating training job images, achieving a reported 20% reduction in resource consumption.
AI-written summary · about
chapters (19)
- 0:00Intro
- 2:05Why DLP on Kubernetes
- 3:50What we did
- 5:33Why serverless
- 7:53What' s changed in 2018/2019
- 9:49Our current stack
- 12:28Knative components
- 15:25Our old architecture
- 16:17Knative build - DLP example
- 17:09Tektoncd-pipeline with buildkit
- 18:19What is Knative serving
- 19:45Knative serving - DLP example
- 20:57Knative serving - Ingress
- 23:02Knative serving -autoscale old solution
- 23:45Knative serving - custom autoscale class
- 25:10Knative serving - cold start
- 26:20Knative with container instance - old solution
- 30:17Compute and Network on Edge
- 32:32Knative eventing - CloudEvents
related talks
39mEvolving Deep Learning Platform with Knative
22mYi Wang, Tech Lead of AI Platform, Baidu, NYC 2017
18mKeynote: Optimizing AI Workload Scheduling: Bilibili's Journey To an Efficient Cloud Native AI Platform
31mFrom Model Serving To Distributed Inference: How Llm-d Evolves AI Platforms on…
50mBuilding Next-Generation Apps with Knative
10mBuilding Machine Learning Inference Through Knative Serverless...
52mServerless with Knative
01h 00mEventing with Knative and Cloud Run: From Basics to Advanced