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.
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.
Maybe switching effort routes you to a different rack of gpu’s which don’t have the cache
That makes sense to me. The output styles work the same way.
Hmmm... Why wouldn't this be handled like other end of prompt things like the current mode?