diff --git a/example/65_gemm_multiply_multiply/moe_gemm1_xdl_fp8_blockscale_splitk.cpp b/example/65_gemm_multiply_multiply/moe_gemm1_xdl_fp8_blockscale_splitk.cpp
index 08a1d6614e..ae707e74a2 100644
--- a/example/65_gemm_multiply_multiply/moe_gemm1_xdl_fp8_blockscale_splitk.cpp
+++ b/example/65_gemm_multiply_multiply/moe_gemm1_xdl_fp8_blockscale_splitk.cpp
@@ -209,11 +209,15 @@ int main(int argc, char* argv[])
     ck::index_t sorted_tile_num = 261;
     ck::index_t valid_tile_num  = 256;
 #endif
-
+    ck::index_t KBatch = 6;
     if(argc == 1)
     {
         // use default case
     }
+    else if(argc == 2)
+    {
+        KBatch = std::stoi(argv[1]);
+    }
     else if(argc == 4)
     {
         // use default case
@@ -267,8 +271,6 @@ int main(int argc, char* argv[])
     ck::index_t Scale_Stride_BN      = (K + Scale_Block_K - 1) / Scale_Block_K;
     ck::index_t Scale_Stride_B       = (N + Scale_Block_N - 1) / Scale_Block_N * 2;
 
-    ck::index_t KBatch = 6;
-
     Tensor<ck::index_t> expert_ids(HostTensorDescriptor({sorted_tile_num}, {1}));
     Tensor<ck::index_t> sorted_token_ids(HostTensorDescriptor({sorted_size}, {1}));
     Tensor<ck::index_t> max_token_id(HostTensorDescriptor({1 + sorted_tile_num}));
@@ -318,6 +320,8 @@ int main(int argc, char* argv[])
     std::cout << "b0_e_n_k: " << b0_e_n_k.mDesc << std::endl;
     std::cout << "b1_e_n_k: " << b1_e_n_k.mDesc << std::endl;
     std::cout << "e_t_n: " << e_t_n_host_result.mDesc << std::endl;
+    std::cout << "k_batch:" << KBatch << std::endl;
+    std::cout << "init_method:" << init_method << std::endl;
 
     switch(init_method)
     {
@@ -326,7 +330,7 @@ int main(int argc, char* argv[])
         a0_t_k.GenerateTensorValue(GeneratorTensor_3<A0DataType>{-1.0, 1.0});
         a1_t_k.GenerateTensorValue(GeneratorTensor_3<A1DataType>{0.0, 1.0});
         b0_e_n_k.GenerateTensorValue(GeneratorTensor_3<B0DataType>{-1.0, 1.0});
-        b1_e_n_k.GenerateTensorValue(GeneratorTensor_3<B1DataType>{0, 1.0});
+        b1_e_n_k.GenerateTensorValue(GeneratorTensor_3<B1DataType>{0.0, 1.0});
         break;
     case 2:
         a0_t_k.GenerateTensorValue(GeneratorTensor_1<A0DataType>{});
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_moe_gemm_blockscale.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_moe_gemm_blockscale.hpp
index 4ba8ab00ea..c556dbec10 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_moe_gemm_blockscale.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_moe_gemm_blockscale.hpp
@@ -765,9 +765,8 @@ struct GridwiseMoeGemmBlockScale
             else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, BLayout>)
             {
                 // KPack * NLane * KLane * K0 * N0
-                b_k_split_offset = k_id * karg.KRead * NLane / BPackedSize;
-                bscale_k_split_offset =
-                    k_id * karg.KRead / ScaleBlockK + k_id * NLane / ScaleBlockN;
+                b_k_split_offset      = k_id * karg.KRead * NLane / BPackedSize;
+                bscale_k_split_offset = k_id * karg.KRead / ScaleBlockK;
             }
 
             // if(k_id < karg.KBatch - 1)