Good points, I didn't mean to come off abrasive but I can see why I would. My attention was to get attention on a thread where my new comment would be buried un

Good points, I didn't mean to come off abrasive but I can see why I would. My attention was to get attention on a thread where my new comment would be buried under the 8 other replies, so I put a big attention grabber at the start. But again good points about the nuances of lower precision. For LLMs at least 'The Case for 4-bit Precision' and 'GPTQ' seem fairly conclusive that over ~10B parameters even 3-bit precision has virtually undetectable loss with the right trircks. Levels which, if they even mattered, can easily be overcome with a little additional training. Newer ongoing research on LLaMA specifically[0] shows we can reduce the model's size around 84% without any meaningful performance loss through a combination of GPTQ, binning, and 3-bit. [0] https://nolanoorg.substack.com/p/int-4-llama-is-not-enough-int-3-and https://nolanoorg.substack.com/p/int-4-llama-is-not-enough-i...