Spyke

Replies

Comment on

RecursiveSI/flint 40b-a2.5b

Reply in thread

This one is a Mixture of Experts model (40B total weight, 2.5B active during token generation). Inference engines like llama.cpp have flags you can pass to offload the lesser used experts to system RAM.

It's slower than having the entire model resident in VRAM, but generally pretty usable.

mtf

Comment on

Becoming feminine again after a long depression.

Mental health can be difficult to untangle or understand even in ideal conditions, let alone when one's peers make the situation worse. I'm happy you could turn that turmoil into clarity and find some inner peace.

Also, you're gorgeous. People who discourage or bully us aren't worth a shred mental energy, but living our best lives is a nice way of telling them to "get fucked".