I’m curious whether that would work and someone might already try. They are both finetunes from mistral, so i would imagine. I have a feeling that this frankenmerge could produce a very good small billion parameter model that might be better than any current <=14b.

  • No-Link-2778B
    link
    fedilink
    English
    arrow-up
    1
    ·
    1 year ago

    Do you think there is any scientific basis for the merge? This is medieval alchemy again.