>>47302014Just a quick note: This model hash cannot be reproduced even if you used the following weights in Supermerger:
0.0,1.0,0.717151771531863,1.0,0.2816244896130977,0.2010302033400634,0.0,0.18092130456854016,0.48057951248856945,0.740215570710713,0.179118258208746,0.8186229820836797,0.3891107064707853,0.17874629316268234,0.38155965368926403,0.730589117141002,0.7212453752136743,0.2643296695195444,0.0,0.18087508357504728,0.00895768002906372,0.45750954096377483,0.42712680180461904,1.0,0.10625951309740733,0.2695064305905318
This model was merged with a special fork of bayesian merger that did diff add by filtering the diff with a median and gaussian filter before adding it to model a. This was done in order to try and eliminate some of the issues that can be caused by repeatedly stacking diff merges ontop of each other, and in theory should allow this model to be diff merged using the standard checkpoint merger at least once more because of it.
