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 9a7cf49f..c5852e5d 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
@@ -136,7 +136,7 @@ __global__ void remap_sf_to_cutlass_kernel(
     // This avoids garbled output from concurrent threads
     if (threadIdx.x == 0 && blockIdx.x == 0) {
         // Print a series of indices to understand the decomposition
-        int debug_indices[] = {0, 1, 2, 3, 4, 15, 16, 31, 32, 63, 64, 127, 128, 255, 256, 511, 512, 512*448-1};
+        int debug_indices[] = {0, 1, 4, 16, 64, 128, 256, 511, 512, 513, 516, 1024, 2048, 4096, 8192, 65536, 131072};
         int n_debug = sizeof(debug_indices) / sizeof(debug_indices[0]);
         for (int di = 0; di < n_debug; di++) {
             int didx = debug_indices[di];