@Thornbloom4Real@KnuteAutomata Ai models can use any storage (disk, ram, etc.) for their weights during inference. But if that storage is slow, then their speed tanks exponentially e.g. 50 tok/s on GPU mem, 5 tok/s on RAM, 0.005 tok/s on SSD
Some fanart of Leila and Erika, both characters of Karen-SCM. You can check their stories on their different websites: https://t.co/td5Z9e0ICa
Additional drawings of Celtie and Karen (World Flipper) on fanbox.
#催眠術#催眠
Have you ever heard about this neat hypnosis technique called "Fractionation"? If not, here is a live demonstration performed on none other than Byleth, because Fire Emblem can never get enough hypno representation! 🔥