pub fn quantize_q5_k(data: &[f32]) -> Vec<u8> ⓘ
Quantize F32 data to Q5_K format
Q5_K
Q5_K: 256 elements per super-block, 176 bytes per block Layout: d (2B) + dmin (2B) + scales (12B) + qh (32B) + qs (128B)