Searchlight memory issues

[PuddleJumper2018]

Hi, I'm using this toolbox to perform searchlight:
It seems this line of code:
rsa.fmri.fMRISearchlight(fullBrainVols, binaryMasks_nS, models, betaCorrespondence, userOptions);

is chewing up 30GB of memory per participant (I have total of 34 ppts and one model so requires 1000GB in total) to process. We've already upgraded memory at the computing cluster once, but insufficiently. Can anyone advice on where does this function digest so much memory? What can we do to make it lightweight?