diff --git a/deep_gemm/mega/__init__.py b/deep_gemm/mega/__init__.py
index 9bbc65f..7375625 100644
--- a/deep_gemm/mega/__init__.py
+++ b/deep_gemm/mega/__init__.py
@@ -182,18 +182,17 @@ def transform_nvfp4_weights_for_mega_moe(
     l1_sf_packed = pack_ue4m3_to_int32(l1_sf)
     l2_sf_packed = pack_ue4m3_to_int32(l2_sf)
     
-    # Reshape to 2D for transform_sf_into_required_layout
-    # (experts, mn, K//64) → (experts * mn, K//64)
-    # The C++ function expects 2D or properly-strided 3D tensors
-    l1_sf_2d = l1_sf_packed.reshape(-1, l1_sf_packed.shape[-1])
-    l2_sf_2d = l2_sf_packed.reshape(-1, l2_sf_packed.shape[-1])
+    # Transpose to MN-major layout (stride(-2)=1) and make contiguous
+    # transform_sf_into_required_layout expects MN-major input for TMA stride checks
+    l1_sf_mn = l1_sf_packed.transpose(-2, -1).contiguous().transpose(-2, -1)
+    l2_sf_mn = l2_sf_packed.transpose(-2, -1).contiguous().transpose(-2, -1)
     
     # Transform SF into TMA-aligned UTCCP layout using DeepGEMM's C++ function
     # recipe (1, 16): gran_mn=1, gran_k=16
     l1_sf_transformed = transform_sf_into_required_layout(
-        l1_sf_2d, l1_n, l1_k, (1, 16), num_experts)
+        l1_sf_mn, l1_n, l1_k, (1, 16), num_experts)
     l2_sf_transformed = transform_sf_into_required_layout(
-        l2_sf_2d, l2_n, l2_k, (1, 16), num_experts)
+        l2_sf_mn, l2_n, l2_k, (1, 16), num_experts)
     
     # L1: interleave gate/up
     l1_interleaved = _interleave_l1_weights((l1_weights[0], l1_sf_packed))