fix typo

Author: titaiwangms

onnxruntime/core/providers/cuda/llm/attention.cc

@@ -4,7 +4,6 @@
 #include "core/providers/cuda/cuda_common.h"
 #include "core/providers/cpu/llm/attention_helper.h"
 #include "core/providers/cuda/llm/attention.h"
-#include "core/providers/cuda/llm/attention_naive.h"
 #include "contrib_ops/cuda/bert/attention_data.h"
 #include "contrib_ops/cuda/bert/attention_impl.h"
 

onnxruntime/core/providers/cuda/llm/attention.h

@@ -9,10 +9,9 @@ namespace onnxruntime {
 namespace cuda {
 
 using namespace onnxruntime::cuda;
-1Code has alerts.Press enter to view.
 
-    template <typename T>
-    class Attention final : public CudaKernel {
+template <typename T>
+class Attention final : public CudaKernel {
  public:
   Attention(const OpKernelInfo& info);
   Status ComputeInternal(OpKernelContext* context) const override;