Spaces:
Running
Running
[New model] Maybe add some ablitterated models?
#652
by
Infranta
- opened
Let's give a shoutout to Maxime Labonne for proposing the abliteration method. Although abliteration is less effective for newer models like Qwen, Deepseek, and Llama, which likely have keyword filters on their training data for strict censorship, it can still effectively prevent false positives.
Therefore, I suggest adding or replacing some of the current models with their ablitterated versions.