Anyone else not using using "preserve_thinking" ?
retained text
I am running QFN, and I have disabled preserve\_thinking by default on all my tests for C++ backend and front-end programming and I have yet to see any decline in performance or intelligence, my context window stays cleaner as well. Deepseek for what I tested in API it also does not preserve thinking and saves context in general.
document fetched 2026-11-12 00:44 · canonical https://reddit.com/r/LocalLLaMA/comments/1xtf9xp/anyone_else_not_using_using_preserve_thinking