π“Ÿ host1.bitchan.one

𓀐 @7zoKU-ed

⏱ 3d seenβ›“ 0πŸ‘ 0β§– 1

filter [-]

languages:onother

Maybe j-space something or ablit whatever or neuron tickling can still make it usable as long as it's smart. It's thin-and-deep like Gemma and it uses a Gemma 3N-like architecture. It already beating all the bloated sparse MoEs despite not even being finished training, having only half or less of the training data of the others. I will devote the next year to make preaching this model and doing everything I can make this bitch usable for daily driving. Our 70b dense prayers have finally been answered.⏱ 3d↩