diff --git a/dsv4/kernels/router/nvfp4_fused_router_kernel.py b/dsv4/kernels/router/nvfp4_fused_router_kernel.py
index eaa06fe6..920600ac 100644
--- a/dsv4/kernels/router/nvfp4_fused_router_kernel.py
+++ b/dsv4/kernels/router/nvfp4_fused_router_kernel.py
@@ -73,7 +73,7 @@ class Nvfp4FusedRouterKernel:
 
         self.use_2cta_instrs = mma_tiler_mn[0] == 256
         self.cta_group = tcgen05.CtaGroup.TWO if self.use_2cta_instrs else tcgen05.CtaGroup.ONE
-        self.mma_kind = tcgen05.mma.Kind.BLOCK_SCALE
+        # No mma_kind needed — make_blockscaled_trivial_tiled_mma infers from dtypes
 
         # Warp layout (6 warps: 4 epi + 1 MMA + 1 TMA)
         self.epilog_warp_id = (0, 1, 2, 3)
@@ -95,10 +95,10 @@ class Nvfp4FusedRouterKernel:
     # ----------------------------------------------------------------
     def _create_tiled_mma(self):
         """Create the tiled MMA for NVFP4 block-scaled GEMM."""
-        return utils.sm100.make_trivial_tiled_mma(
+        return sm100_utils.make_blockscaled_trivial_tiled_mma(
             self.a_dtype, self.a_major_mode, self.b_major_mode,
-            self.acc_dtype, self.cta_group, self.mma_tiler_mn,
-            self.mma_kind, self.sf_dtype,
+            self.sf_dtype, self.cta_group, self.mma_tiler_mn,
+            self.sf_vec_size,
         )
 
     def _setup_attributes(self):