“flash” -> looks inside -> over 200B parameters

i mean i’m sure it’s much smaller than the non flash version, but my hope dies pretty quick when i realize i need a datacenter to run a model

i wonder if it would be better to use bonsai-2 over a regular 2-bit quant 🤔

bookmarked, thanks!

also, anyone here know a good opencode skill to use with it?

that’s nice, but at today’s pace 3.5 seems a bit dated (3.8 available at time of comment, qwen4 supposedly “soon”)

heretic is a good abliteration method from my experience

glad it’s a 9b model