>>103629708
Lorebooks can be slightly wordier and more token heavy and cards work better at lower token counts. Even then, what gets retained really does depend on token limits and i'd say anything below 64k these days tends to be incredibly lobotimized solely because most of anything like 8k or even 16k gets easily clogged up by just prompts, lorebooks and card tokens. Add to this if you're using any form of reasoning and you're on fast track for your responses turning into slop. Most of the 128k and 256k models tend to be able to retain quite a reasonable amount of data without suffering aneurysm some 3000-4000 messages in.