A little error

#1
by lhhvc - opened

I want to use the Sentinel-v2 model to build an LLM input detector, but during testing, I found that the model misclassifies "hello" as a dangerous input

Qualifire AI org
edited Sep 16

Hi!

Thank you for testing! we found out one dataset that really biased against one word prompts, we fixed the issue and pushed a new revision 2 days ago. please pull the latest revision and try again. if the issue persists please LMK
Other then that how's the overall experience with the model?

Cheers.
Dror

Thank you for your reply. The model works very well and I will try the new version. If there are any other questions, I will share them with you in a timely manner😀

Sign up or log in to comment