model nor reasoning
#1
by
NEWWWWWbie
- opened
Can I know why the model is not reasoning, even though it's supposed to be? Or is it my issue that it's not reasoning properly?
Hi
@NEWWWWWbie
,
We’ve made a similar observation. I suspect it's because our distillation setup is heavily focused on CTI-Bench, which limits its scope. Outside of cybersecurity, reasoning is rarely triggered, and even within cybersecurity, it only activates partially — typically when the question resembles CTI-Bench.
A simple fix is to include "Please reason through the problem before generating the output." in the system prompt, which significantly increases the likelihood of reasoning. 😊