diff --git a/tests/python/unittest/test_operator.py b/tests/python/unittest/test_operator.py
index 7db07596d7f8..9f27ddedc8fe 100644
--- a/tests/python/unittest/test_operator.py
+++ b/tests/python/unittest/test_operator.py
@@ -7331,16 +7331,16 @@ def check_bilinear_resize_modes_op(shape, scale_height=None, scale_width=None, s
         assert_almost_equal(y.asnumpy(), expected, 1e-3, 0)
         if mode != 'like':
             resize_sym = mx.sym.contrib.BilinearResize2D(data_sym, None, scale_height=scale_height, scale_width=scale_width, mode=mode)
-            check_symbolic_forward(resize_sym, [data_np], [expected], rtol=1e-3)
-            check_symbolic_backward(resize_sym, [data_np], [out_grads], expected_backward, rtol=1e-3)
-            check_numeric_gradient(resize_sym, [data_np])
+            check_symbolic_forward(resize_sym, [data_np], [expected], rtol=1e-3, atol=1e-5)
+            check_symbolic_backward(resize_sym, [data_np], [out_grads], expected_backward, rtol=1e-3, atol=1e-5)
+            check_numeric_gradient(resize_sym, [data_np], rtol=1e-2, atol=1e-5)
         else:
             data_sym_like = mx.sym.var('data_like')
             resize_sym = mx.sym.contrib.BilinearResize2D(data_sym, data_sym_like, mode=mode)
             date_np_like = x_1.asnumpy()
-            check_symbolic_forward(resize_sym, [data_np, date_np_like], [expected], rtol=1e-3)
-            check_symbolic_backward(resize_sym, [data_np, date_np_like], [out_grads], expected_backward, rtol=1e-3)
-            check_numeric_gradient(resize_sym, [data_np, date_np_like])
+            check_symbolic_forward(resize_sym, [data_np, date_np_like], [expected], rtol=1e-3, atol=1e-5)
+            check_symbolic_backward(resize_sym, [data_np, date_np_like], [out_grads], expected_backward, rtol=1e-3, atol=1e-5)
+            check_numeric_gradient(resize_sym, [data_np, date_np_like], rtol=1e-2, atol=1e-5)
 
     shape = (2, 2, 10, 10)
     check_bilinear_resize_op(shape, 5, 5)