The fold block_sf (float8) * global_sf (float32) -> float8 loses ~25% precision.
Product of ~56-448 block_sf * ~4.65e-05 global_sf lands in float8 low-precision
zone where step size is 25%. This makes model output garbage despite finite values.
Fix: keep block scales as original float8, return global scales separately as
float32 per-expert vectors. Apply global scale as per-expert GEMM alpha in
cutlass_grouped_nvfp4_gemm (already iterates per-expert). For L1 with separate
gate/up global scales, use gate_gs as alpha and apply up_correction ratio to
the up half post-GEMM.
weight_transform.py: no more _fold_global_scale, returns (w, sf, global_sf)
nvfp4_mega_moe.py: per-expert alpha = activation_gs * weight_gs
kernel.py: per_expert_alpha parameter in grouped GEMM
deepseek_v4.py: updated type hints and comments
cute::crd2idx requires hierarchical coordinates matching the layout's
nested shape, which we don't have from flat (m, k_sf). Reverted to
idx2crd dest-first approach. The real bug was cute::size vs
cute::cosize for allocation, not the remap direction.
CuTe Layout objects with hierarchical shapes can't be called directly
with flat (m, k_sf). Use cute::crd2idx(make_coord(m, k_sf), layout_sf)
to convert logical coordinates to physical indices.
