Skip to main content

dequantize_q4_0_gpu

combs_models

Function dequantize_q4_0_gpu 

Source 
pub fn dequantize_q4_0_gpu<R: Runtime>(
    client: &ComputeClient<R>,
    data: &[u8],
) -> Result<Vec<f32>>
Expand description

Runs the dequant-only kernel over a raw Q4_0 block stream. Exists for validation (bit-exact vs the CPU reference) and debugging, not the hot path.