Page MenuHomePhabricator

Add a user-requestable RAM and/or CPU limit modification in PAWS
Open, MediumPublic

Description

Right now, it is very easy to crash your kernel in PAWS by simply running a pytorch or tensorflow example because ML likes RAM. We have xlarge nodes in some cases, and if the container's "request" was high enough, it would not be scheduled where RAM > 3GB would be a problem at this scale. The trick is providing that option to the user so that they can schedule their notebook accordingly.