Aliterated variant of MiniCPM5-2B-SFT tends to rationalize harmful questions rather than reject them, with a 1/20 chance; details on HuggingFace links

An abliterated variant of the original MiniCPM5-2B-SFT One notable behavioral change is its tendency to rationalize potentially harmful or restricted questions rather then reject but thats a 1/20 chance...