Evolving Deep Learning Platform with Knative
The emergence of Knative has enhanced the capabilities of Baidu internal deep learning platform. The workflow between training and inference services can be implemented based on Knative eventing. The …
| Talk Title | Evolving Deep Learning Platform with Knative |
| Speakers | Ti Zhou (Architect, Baidu) |
| Conference | KubeCon + CloudNativeCon |
| Conf Tag | |
| Location | Shanghai, China |
| Date | Jun 23-26, 2019 |
| URL | Talk Page |
| Slides | Talk Slides |
| Video | |
The emergence of Knative has enhanced the capabilities of Baidu internal deep learning platform. The workflow between training and inference services can be implemented based on Knative eventing. The training and inference services can also implement smart routing and auto-scaling by leveraging Knative serving. And the training job image can also be built by Knative build framework. After using Knative, Baidu internal deep learning platform has reduced resource consumption by 20%. In this speech, we will show how to use the capabilities of Knative to reconstruct deep learning platform, including: Expanding eventing to achieve training to inference pipeline automation Use serving to implement smart routing and auto-scaling of inference services Use build to automate the process of generating the training images
