diff --git a/debug/accuracy_tools/api_accuracy_checker/compare/algorithm.py b/debug/accuracy_tools/api_accuracy_checker/compare/algorithm.py
index a79125d832b39e9074c8dff9d3eb5c24f2003c4d..2ab8735a3128d004ecfd44597b240ac125b16b1b 100644
--- a/debug/accuracy_tools/api_accuracy_checker/compare/algorithm.py
+++ b/debug/accuracy_tools/api_accuracy_checker/compare/algorithm.py
@@ -30,7 +30,7 @@ def get_max_rel_err(n_value, b_value):
         print_warn_log("Max rel err only support numpy array!")
         raise ValueError("Max rel err only support numpy array!")
     if n_value.dtype != b_value.dtype:
-        raise ValueError("npu and bench value dtype is different.")
+        return CompareConst.NA, False
     if n_value.dtype in Const.FLOAT_TYPE:
         rel_err = np.abs((n_value - b_value) / (b_value + np.finfo(b_value.dtype).eps)).max()
         return rel_err, rel_err < 0.001
diff --git a/debug/accuracy_tools/api_accuracy_checker/compare/compare.py b/debug/accuracy_tools/api_accuracy_checker/compare/compare.py
index b877bc50d4d71ed8d9b5550bab245ae6f5c87bb6..5cb0777e43f0d51d2c3864b55ee887e927750f13 100644
--- a/debug/accuracy_tools/api_accuracy_checker/compare/compare.py
+++ b/debug/accuracy_tools/api_accuracy_checker/compare/compare.py
@@ -23,7 +23,8 @@ class Comparator:
         self.compare_alg_names = [] 
         self.register_compare_algorithm("Cosine Similarity", cosine_sim, cosine_standard) 
         self.test_results = []
-        self.test_result_cnt = {"forward_fail_num":0, "backward_fail_num":0, "forward_and_backward_fail_num":0, "success_num":0}
+        self.test_result_cnt = {"forward_fail_num": 0, "backward_fail_num": 0, "forward_and_backward_fail_num": 0,
+                                "success_num": 0}
 
     def print_pretest_result(self):
         res_dict = {
@@ -34,7 +35,7 @@ class Comparator:
         }    
         tb = PrettyTable()
         tb.add_column("Category", list(res_dict.keys()))
-        tb.add_column("statistics",list(res_dict.values()))
+        tb.add_column("statistics", list(res_dict.values()))
         info_tb = str(tb)
         print_info_log(info_tb)
 
@@ -62,9 +63,15 @@ class Comparator:
         self.compare_alg_names.append(name)
 
     def compare_output(self, api_name, bench_out, npu_out, bench_grad=None, npu_grad=None):
-        is_fwd_success, fwd_compare_alg_results = self._compare_core_wrapper(bench_out, npu_out)
+        if "dropout" in api_name:
+            is_fwd_success, fwd_compare_alg_results = self._compare_dropout(bench_out, npu_out)
+        else:
+            is_fwd_success, fwd_compare_alg_results = self._compare_core_wrapper(bench_out, npu_out)
         if bench_grad and npu_grad:
-            is_bwd_success, bwd_compare_alg_results = self._compare_core_wrapper(bench_grad, npu_grad)
+            if "dropout" in api_name:
+                is_bwd_success, bwd_compare_alg_results = self._compare_dropout(bench_grad[0], npu_grad[0])
+            else:
+                is_bwd_success, bwd_compare_alg_results = self._compare_core_wrapper(bench_grad, npu_grad)
         else:
             is_bwd_success, bwd_compare_alg_results = CompareConst.NA, None 
         self.record_results(api_name, is_fwd_success, is_bwd_success, fwd_compare_alg_results, bwd_compare_alg_results)
@@ -80,4 +87,15 @@ class Comparator:
     def _compare_core_wrapper(self, bench_out, npu_out):
         name = self.compare_alg_names[0]
         detailed_result, test_success = compare_core(bench_out, npu_out, self.compare_alg[name][0])
-        return test_success, detailed_result 
\ No newline at end of file
+        return test_success, detailed_result
+
+    @staticmethod
+    def _compare_dropout(bench_out, npu_out):
+        tensor_num = bench_out.numel()
+        if tensor_num >= 100:
+            if abs((bench_out == 0).sum() - (npu_out == 0).sum()) / tensor_num < 0.1:
+                return True, 1
+            else:
+                return False, 0
+        else:
+            return True, 1