How to obtain unormalized memory counts for cluster-trace-v2017 and cluster-tracev2018

Hi, I am trying to figure out how to un-normalize memory counts so I can translate them to pod specs, however its unclear what the divisor of these values is. what should I multiply them by to get the unnormalized values?