Every story and comment from the Hacker News API, in a layout borrowed from 2001.
A comment on In Defense of Pure 16-Bit Floating-Point Neural Networks. Parent comment
https://www.microsoft.com/en-us/research/publication/1-bit-s...
int4 (fixed point) has already been popular for inference https://developer.nvidia.com/blog/int4-for-ai-inference/ and int3 has seem some use for LLaMA-at-home
There are no comments on this comment yet.
https://www.microsoft.com/en-us/research/publication/1-bit-s...
int4 (fixed point) has already been popular for inference https://developer.nvidia.com/blog/int4-for-ai-inference/ and int3 has seem some use for LLaMA-at-home