Skip to main content

dequant_vbit

cortiq_core::quant

Function dequant_vbit 

Source 
pub fn dequant_vbit(
    bytes: &[u8],
    rows: usize,
    cols: usize,
    dst: &mut [f32],
) -> Result<(), String>
Expand description

Dequantize a full vbit tensor (P13 FIG.3, grouped variant): [u8 bits: rows][f16 scales: rows·cols/32][bit-packed rows MSB-first, each row padded to a whole byte]. w = (u − L)·scale, L = 2^{b−1}−1.