Looking to know something about model merge
ive been testing a bunch of models and i found [matricide-12B-Unslop-Unleashed-v2.fp16] to be very good in terms of personality and speech pattern retention but 12B is a smaller model, i was wondering if it could be merged into a bigger model that could use the matricide's features such as
[MS-Schisandra-22B-v0.3-Q8_0] or [magnum-v4-22b-Q8_0], i like these models they are good at long form and description / knowledges of stuff but they're lacking in term of personality, which i found the one in matricide to be perfect. could it be possible to transpose the knowledge of these models with the way matricide respond? and will it keep these models within 22~26 gb for the Q8? or will they combine both sizes.
We are just a small cog in the machinery that takes hf models and converts them to ggufs, so... this is unfortunately not a good place to find answers to your questions... I would tell you if I knew :) Your best bet would be to ask somebody who made such a model. And while there are doubtless some good rules on what to do, it's trial and error, to a large part, I suspect.