Unread
Abliterating AI Models
18:13
Richard from the Mr. Radermacher team on Hugging Face joins us to explain quantization and uncensoring of open-weight models.
We talk 4-bit vs. 6-bit quality, VRAM vs. memory bandwidth, why private local models are growing, and how uncensored models help with creative writing, coding, research, and what not.
Plus: a quick etymology of abliteration, that is a blend of ablation (removing) and…