pub fn dequant_turbo8_block_to_bf16(
raw: &[u8],
bs: usize,
nkv: usize,
hd: usize,
out: &mut [bf16],
)Expand description
Dequant a Turbo8 (FP8 E4M3 data + per-group BF16 scales) KV block to BF16.
Layout per block (post 2026-04-28 BF16-scale upgrade):
data: bs * nkv * hd bytes (1 FP8 byte per element).
scales: bs * nkv * (hd / NVFP4_GROUP_SIZE) * 2 bytes (BF16 = 2 bytes/scale).
Mirrors kernels/gb10/common/paged_decode_attn_turbo8*.cu post-upgrade.