π“Ÿ host1.bitchan.one

𓀐 @ncXKNLEK

⏱ 1w seenβ›“ 0πŸ‘ 0β§– 1

filter [-]

languages:onother

5090, and yes I paid too much for it. She spills over a bit onto CPU, but it's not too bad. Still gens tokens about as fast as I can read with a 96K context size. If you don't need "realtime" or better, and can leave prompts running for a while in the background, it's worth it even with a much smaller GPU. First local model I've tried that is actually good at code to work on real projects, and I've tried bigger ones.⏱ 1w↩