diff --git a/python/cudnn/deepseek_sparse_attention/sparse_attention_backward/dsa_bwd_sm100.py b/python/cudnn/deepseek_sparse_attention/sparse_attention_backward/dsa_bwd_sm100.py
index 0b4d3c320..7f23c5ef5 100644
--- a/python/cudnn/deepseek_sparse_attention/sparse_attention_backward/dsa_bwd_sm100.py
+++ b/python/cudnn/deepseek_sparse_attention/sparse_attention_backward/dsa_bwd_sm100.py
@@ -572,9 +572,9 @@ class SharedStorage:
 
         convert_grid_x = (mKV.shape[0] + self.block_seq - 1) // self.block_seq
         convert_grid = [
-            1,
             convert_grid_x,
             1,
+            1,
         ]
         convert_block = [self.num_threads_D_convert, self.num_threads_seq, 1]
         self.convert(
@@ -616,8 +616,8 @@ def convert(
     ):
         tidx, tidy, _ = cute.arch.thread_idx()
         (
-            _,
             seq_block_idx,
+            _,
             batch_idx,
         ) = cute.arch.block_idx()