3 ms·
What you wrote is somewhat ambiguous, so allow me to rephrase. It is true that most fine-tunes of relatively small (which can mean anything up to 150B params, d
by int_19h 1y ago
What you wrote is somewhat ambiguous, so allow me to rephrase. It is true that most fine-tunes of relatively small (which can mean anything up to 150B params, depending on who you ask!) LLMs are for uncensored roleplay purposes.
- pogue 1y agoSorry for the late reply. I've read a bit about this, but I don't know how it's done. I've seen some uncensored models using a process called "abliteration" [1] but I'm not sure if that's the same thing or not. I've never tried self training an AI model or anything but I'd be very curious about the process. Maybe this is something someone could do as a side hustle? [1] Uncensor any LLM with abliteration by Maxime Labonne https://huggingface.co/blog/mlabonne/abliteration https://huggingface.co/blog/mlabonne/abliteration