diff --git a/test/CMakeLists.txt b/test/CMakeLists.txt
index 39a2340..97d65dc 100644
--- a/test/CMakeLists.txt
+++ b/test/CMakeLists.txt
@@ -2,7 +2,7 @@ find_package(GTest REQUIRED)
 find_package(glog REQUIRED)
 
 set(link_ext_lib glog::glog GTest::gtest)
-add_executable(test_llm test_main.cpp test_buffer.cpp test_math.cpp)
+add_executable(test_llm test_main.cpp test_buffer.cpp test_add.cpp test_math.cpp)
 
 target_link_libraries(test_llm ${link_ext_lib})
 target_include_directories(test_llm PUBLIC ${glog_INCLUDE_DIR})
diff --git a/test/test_add.cpp b/test/test_add.cpp
new file mode 100644
index 0000000..6630c27
--- /dev/null
+++ b/test/test_add.cpp
@@ -0,0 +1,25 @@
+#include <glog/logging.h>
+#include <gtest/gtest.h>
+#include <tensor/tensor.h>
+#include "base/buffer.h"
+#include "../kuiper/source/op/kernels/add_kernel.h"
+
+TEST(test_op, add) {
+  using namespace base;
+  auto alloc_cu = base::CPUDeviceAllocatorFactory::get_instance();
+
+  int32_t size = 32 * 151;
+
+  tensor::Tensor t1(base::DataType::kDataTypeFp32, size, true, alloc_cu);
+  tensor::Tensor t2(base::DataType::kDataTypeFp32, size, true, alloc_cu);
+  tensor::Tensor out(base::DataType::kDataTypeFp32, size, true, alloc_cu);
+  for (int i = 0; i < t1.size(); ++i) {
+    t1.index<float>(i) = 1;
+    t2.index<float>(i) = 2;
+  }
+
+  kernel::get_add_kernel(base::DeviceType::kDeviceCPU)(t1, t2, out);
+  for (int i = 0; i < size; ++i) {
+    ASSERT_EQ(out.index<float>(i), 3.f);
+  }
+}