this model thinks a lot
#14
by
shihab456321
- opened
decisively the best model i have ever seen, but becoming unusable for too much thinking
Agreed. It uses the majority of it's tokens on thinking. It really hampers using it in edge scenarios or on mobile. A way to limit thinking would be good.
The model seems really promising. I think it just needs a confidence boost, so it doesn't endlessly worry about how to respond to 'Hello!'
On further review, it appears the thinking tokens are optimized for deep tool usage.
It would be great to have a way to reduce thinking tokens.