Skip to main content

dequantize_nvfp4_block

Function dequantize_nvfp4_block 

pub fn dequantize_nvfp4_block(
    packed: &[u8; 8],
    fp8_scale: u8,
    global_scale: f32,
    out: &mut [f32; 16],
)
Expand description

Dequantize a single 16-element block of NVFP4 weights.

packed: 8 bytes containing 16 packed 4-bit nibbles. fp8_scale: FP8 E4M3 scale factor for this block. global_scale: Multiplier for the entire tensor / weight group. out: Output array of 16 f32 values.