Creative
#1
by redaihf - opened
This model is creative but unstable. As with other Llama based models it is quite sensitive to the structure of prompts. It is very uncensored but will occasionally write disclaimers out of thin air.
Not sure what might be triggering disclaimers with the MOE merge. Hopefully v2 is better. It could be something to do with the way experts are routed idk
the way experts are routed
That's my guess too.