Skip to main content

quantize_q5_k

Function quantize_q5_k 

Source
pub fn quantize_q5_k(data: &[f32]) -> Vec<u8> 
Expand description

Quantize F32 data to Q5_K format

Q5_K: 256 elements per super-block, 176 bytes per block Layout: d (2B) + dmin (2B) + scales (12B) + qh (32B) + qs (128B)