Skip to main content

repack_q4_k

Function repack_q4_k 

Source
pub fn repack_q4_k(data: &[u8]) -> Result<(Vec<u32>, Vec<f32>, Vec<u32>)>
Expand description

Layout step for Q4_K: split each 144-byte superblock into SoA device arrays — (qs words, [d, dmin] f32 pairs, scale words). 148 B per 256 weights = 4.63 bits/weight (GGUF native is 4.5).