diff --git a/tester/api_config/5_accuracy/accuracy_gpu_error_grads_diff.txt b/tester/api_config/5_accuracy/accuracy_gpu_error_grads_diff.txt
index af3b5af23..35009e91a 100644
--- a/tester/api_config/5_accuracy/accuracy_gpu_error_grads_diff.txt
+++ b/tester/api_config/5_accuracy/accuracy_gpu_error_grads_diff.txt
@@ -15945,3 +15945,17 @@ paddle.nn.functional.binary_cross_entropy_with_logits(logit=Tensor([256],"float3
 paddle.nn.functional.binary_cross_entropy_with_logits(logit=Tensor([512],"float32"), label=Tensor([512],"float32"), reduction="sum", )
 paddle.nn.functional.binary_cross_entropy_with_logits(logit=Tensor([64, 19],"float32"), label=Tensor([64, 19],"float32"), reduction="none", )
 paddle.nn.functional.binary_cross_entropy_with_logits(logit=Tensor([64, 26],"float32"), label=Tensor([64, 26],"float32"), reduction="none", )
+paddle.nn.functional.kl_div(Tensor([40, 20, 50],"float32"), Tensor([40, 20, 50],"float32"), "batchmean", False, )
+paddle.nn.functional.kl_div(Tensor([40, 20, 50],"float32"), Tensor([40, 20, 50],"float32"), "batchmean", True, )
+paddle.nn.functional.kl_div(Tensor([40, 20, 50],"float32"), Tensor([40, 20, 50],"float32"), "none", False, )
+paddle.nn.functional.kl_div(Tensor([5, 20],"float32"), Tensor([5, 20],"float64"), "mean", False, )
+paddle.nn.functional.kl_div(Tensor([5, 20],"float64"), Tensor([5, 20],"float32"), )
+paddle.nn.functional.kl_div(Tensor([5, 20],"float64"), Tensor([5, 20],"float64"), "batchmean", False, )
+paddle.nn.functional.kl_div(Tensor([5, 20],"float64"), Tensor([5, 20],"float64"), "mean", False, )
+paddle.nn.functional.kl_div(Tensor([5, 20],"float64"), Tensor([5, 20],"float64"), "mean", True, )
+paddle.nn.functional.kl_div(Tensor([5, 20],"float64"), Tensor([5, 20],"float64"), "none", False, )
+paddle.nn.functional.kl_div(Tensor([5, 20],"float64"), Tensor([5, 20],"float64"), "sum", False, )
+paddle.nn.functional.kl_div(Tensor([5, 2],"float32"), label=Tensor([5, 2],"float32"), reduction="mean", name=None, )
+paddle.nn.functional.kl_div(Tensor([5, 2],"float64"), Tensor([5, 2],"float64"), "mean", False, )
+paddle.nn.functional.kl_div(Tensor([],"float64"), Tensor([],"float64"), "batchmean", False, )
+paddle.nn.functional.kl_div(input=Tensor([32, 128, 128],"float32"), label=Tensor([32, 128, 128],"float32"), reduction="batchmean", )