diff --git a/debug/accuracy_tools/api_accuracy_checker/README.md b/debug/accuracy_tools/api_accuracy_checker/README.md
index 5ff4b26868afa0319dcb9ef723b5081411bb372a..1260a58334968865e0cf38e3a66a6f3c45426344 100644
--- a/debug/accuracy_tools/api_accuracy_checker/README.md
+++ b/debug/accuracy_tools/api_accuracy_checker/README.md
@@ -51,7 +51,7 @@ Ascend模型精度预检工具能在昇腾NPU上扫描用户训练模型中所
 
    forward和backward两个命令行参数根据实际存盘的json文件名配置。比对结果存盘路径默认是'./'（相对于run_ut的路径），可以在运行run_ut.py时通过 --out_path命令行参数配置。结果包括pretest_result.csv和pretest_details.csv两个文件。前者是api粒度的，标明每个api是否通过测试。建议用户先查看前者，对于其中没有通过测试的或者特定感兴趣的api，根据其API name字段在pretest_details.csv中查询其各个输出的达标情况以及比较指标。
 
-   注意：目前API通过测试的标准是每个输出与标杆比对的余弦相似度大于0.99，pretest_details.csv中的相对误差供用户分析时使用。
+   注意：目前API通过测试的标准是每个输出与标杆比对的余弦相似度大于0.99，并且float16数据要通过双千分之一标准，float32数据要通过双万分之一标准，pretest_details.csv中的相对误差供用户分析时使用。
 
 
 ## FAQ 
diff --git a/debug/accuracy_tools/api_accuracy_checker/compare/algorithm.py b/debug/accuracy_tools/api_accuracy_checker/compare/algorithm.py
index c8ecea17b01d62de6d3ceef7c47ce339861a87e3..29b10608e496a7808a13dcb8656521e6bcc8cba8 100644
--- a/debug/accuracy_tools/api_accuracy_checker/compare/algorithm.py
+++ b/debug/accuracy_tools/api_accuracy_checker/compare/algorithm.py
@@ -70,7 +70,7 @@ def get_rel_err_ratio_ten_thousandth(n_value, b_value):
     ratio, bool_result, msg = get_rel_err_ratio(n_value, b_value, 0.0001)
     if b_value.dtype == np.float16:
         msg = f"This indicator is not used to evaluate {b_value.dtype} data"
-        return ratio, CompareConst.NA, msg
+        return ratio, True, msg
     return ratio, bool_result, msg
 
 def get_rel_err_ratio(n_value, b_value, thresholding):
@@ -187,7 +187,7 @@ def compare_core(bench_out, npu_out, alg):
     if isinstance(compare_result, list):
         compare_result = flatten_compare_result(compare_result)
     else:
-        compare_result = [(compare_result, str(test_success), msg)]
+        compare_result = [(compare_result, msg)]
     if isinstance(bench_dtype, list):
         bench_dtype = flatten_compare_result(bench_dtype)
         npu_dtype = flatten_compare_result(npu_dtype)
diff --git a/debug/accuracy_tools/api_accuracy_checker/compare/compare.py b/debug/accuracy_tools/api_accuracy_checker/compare/compare.py
index dd2d5f3b0f56c7d169897e61f2024b879ad919ba..c5ae3729947ed75eaefe439b3d9a2fd6c243ef1b 100644
--- a/debug/accuracy_tools/api_accuracy_checker/compare/compare.py
+++ b/debug/accuracy_tools/api_accuracy_checker/compare/compare.py
@@ -68,12 +68,12 @@ class Comparator:
     def write_detail_csv(self):
         test_rows = [[
             "Subject", "Bench Dtype", "NPU Dtype",
-            "Cosine Similarity", "Cosine Similarity Pass", "Cosine Similarity Message",
-            "Max Rel Error", "Max Rel Err Pass", "Max Rel Err Message",
-            "Thousandth Rel Error Ratio", "Thousandth Rel Error Ratio Pass", "Thousandth Rel Error Ratio Message",
-            "Ten Thousandth Rel Error Ratio", "Ten Thousandth Rel Error Ratio Pass", "Ten Thousandth Rel Error Ratio Message",
-            "Compare Builtin Type", "Builtin Type Pass",
-            "Builtin Type Message"
+            "Cosine Similarity", "Cosine Similarity Message",
+            "Max Rel Error", "Max Rel Err Message",
+            "Thousandth Rel Error Ratio", "Thousandth Rel Error Ratio Message",
+            "Ten Thousandth Rel Error Ratio", "Ten Thousandth Rel Error Ratio Message",
+            "Compare Builtin Type", "Builtin Type Message",
+            "Pass"
         ]]  
         for test_result in self.test_results:
             subject_prefix = test_result[0]
@@ -129,18 +129,19 @@ class Comparator:
             detailed_result, test_success, bench_dtype, npu_dtype = compare_core(bench_out, npu_out, alg)
             bench_dtype_total = bench_dtype
             npu_dtype_total = npu_dtype
-            if name != "Max Relative Error" and test_success != CompareConst.NA:
+            if name != "Max Relative Error":
                 test_success_total = test_success_total and test_success
             if detailed_result_total:
                 for i in range(len(detailed_result_total)):
                     detailed_result_total[i] += detailed_result[i]
             else:
                 detailed_result_total = detailed_result
-        # dtype加到所有指标的前面
+        # dtype加到所有指标的前面, 是否pass放到所有指标的后面
         for i in range(len(detailed_result_total)):
             detailed_result = list(detailed_result_total[i])
             detailed_result.insert(0, bench_dtype_total[i])
             detailed_result.insert(1, npu_dtype_total[i])
+            detailed_result.append(str(test_success_total))
             detailed_result_total[i] = tuple(detailed_result)
         return test_success_total, detailed_result_total