Conceptuncensored fine-tuning / Heretic系列
Uncensored Fine-Tuning
A technical approach in which community developers perform secondary fine-tuning on aligned models using datasets that exclude refusal examples, or apply DPO training methods, with the goal of removing or weakening the model's built-in refusal mechanisms.
Timeline (last 90 days)
Sep 21
The popular 'uncensoring fine-tuning' in the community typically only removes safety restriction layers and has no effect on the implicit biases caused by training data.
Unverified50%