ml-tooling
Support GPUs on multiple machines (via docker-swarm or kubernetes)?
Feature description:
Support docker-swarm (with GPUs support) out-of-the-box.
Problem and motivation:
As here describes, CURRENTLY it is not possible to run ml-hub with GPU support across multiple machines (while every machine may have one or more GPU cards). Since it is not easy to build a kubernetes cluster with GPU support and management (and I'm not farmiliar with kubernetes), maybe a more lightweight solution (like docker-swarm?) would support it more seamlessly (via nvidia-docker).
Is this something you're interested in working on?
Yes
By the way, kubernetes seems to support GPU management via Device plugins https://kubernetes.io/docs/tasks/manage-gpus/scheduling-gpus/#using-device-plugins. So why the gpu mode is not supported in kubernetes? Is it due to lack of standards, historical reasons, or just waiting someone to implement?