diff --git a/src/nvfp4_megamoe_kernel/cutlass_nvfp4_gemm/cutlass_nvfp4_gemm.cu b/src/nvfp4_megamoe_kernel/cutlass_nvfp4_gemm/cutlass_nvfp4_gemm.cu index fcfc28dd..88bfd550 100644 --- a/src/nvfp4_megamoe_kernel/cutlass_nvfp4_gemm/cutlass_nvfp4_gemm.cu +++ b/src/nvfp4_megamoe_kernel/cutlass_nvfp4_gemm/cutlass_nvfp4_gemm.cu @@ -38,7 +38,7 @@ using LayoutATag = cutlass::layout::RowMajor; constexpr int AlignmentA = 32; using ElementB = cutlass::nv_float4_t; -using LayoutBTag = cutlass::layout::ColumnMajor; +using LayoutBTag = cutlass::layout::RowMajor; constexpr int AlignmentB = 32; using ElementD = cutlass::bfloat16_t;