logoalt Hacker News

janalsncmyesterday at 6:16 PM3 repliesview on HN

I’m guessing that there’s a system prompt at the top telling the model about its reasoning budget. So when you switch reasoning effort it busts the cache.


Replies

cellularmitosistoday at 9:05 AM

Maybe switching effort routes you to a different rack of gpu’s which don’t have the cache

SoMomentarytoday at 2:11 AM

That makes sense to me. The output styles work the same way.

footayesterday at 10:20 PM

Hmmm... Why wouldn't this be handled like other end of prompt things like the current mode?