pub fn quantize_to_fp8_blockscaled(
bf16_weight: &DenseWeight,
n: usize,
k: usize,
gpu: &dyn GpuBackend,
quantize_kernel: KernelHandle,
stream: u64,
) -> Result<Fp8Weight>Expand description
2026-09-25: Quantize an [n, k] BF16 dense weight to block-scaled FP8 E4M3
on the GPU, as an Fp8Weight tagged Fp8BlockScaled with FP32 scales
[ceil(n/128), ceil(k/128)].
It does not synchronize stream and does not free the BF16 source; the
caller must synchronize before freeing it.