Function vec128_b_index 

pub const fn vec128_b_index(m_pad: usize, token: usize, k_group: usize) -> usize

Offset of the VEC128 scale for (token, k_group) in the layout cuBLASLt documents for the B operand: shape N x L, N-major, N = m_pad tokens.

This is the SSOT the CUDA adapter fp8_act_scale_to_kmajor mirrors.