diff --git a/onnxruntime/contrib_ops/cpu/bert/attention_common.h b/onnxruntime/contrib_ops/cpu/bert/attention_common.h
index f8bb820743f87..0fd8790e0d29d 100644
--- a/onnxruntime/contrib_ops/cpu/bert/attention_common.h
+++ b/onnxruntime/contrib_ops/cpu/bert/attention_common.h
@@ -96,7 +96,7 @@ struct GroupQueryAttentionParameters {
   int head_size;
   int kv_hidden_size;
   int kv_num_heads;
-  int num_splits;  // number of splits for splitkv
+  int num_splits;          // number of splits for splitkv
   bool is_unidirectional;  // causal
   float scale;
   AttentionQkvFormat qkv_format;
diff --git a/onnxruntime/contrib_ops/cuda/bert/flash_attention/flash_api.cc b/onnxruntime/contrib_ops/cuda/bert/flash_attention/flash_api.cc
index a7fed3903b78f..89a27c4d2b0d3 100644
--- a/onnxruntime/contrib_ops/cuda/bert/flash_attention/flash_api.cc
+++ b/onnxruntime/contrib_ops/cuda/bert/flash_attention/flash_api.cc
@@ -195,7 +195,7 @@ std::tuple<int, int, int> get_num_splits_and_buffer_sizes(int batch_size, int se
   int max_splits = 128;
   // split kv buffers
   int num_splits = num_splits_heuristic(batch_size, seqlen_q, seqlen_k, num_heads, head_size,
-                                                            num_SMs, max_splits);
+                                        num_SMs, max_splits);
   if (num_splits > 1) {
     // softmax_lse_accum buffer
     int softmax_lse_accum_bytes = get_softmax_lse_accum_size(num_splits, batch_size, num_heads, seqlen_q);