I still haven't seen a good evaluation of the differences between different quantized versions of why models to be honest – or even any good anecdotes about things that work and things that don't at different levels So I'm effectively flying blind when it comes to quantization

→ View original post on X — @simonw,