Use same sacle in qsigmoid UT (#2094)

Author: ZhiweiYan-96

tests/gpu/examples/test_qsigmoid.py

@@ -6,19 +6,24 @@
 
 
 class TestTorchMethod(TestCase):
-    @pytest.mark.skip(reason="Need create a PR for torch to align output_scale")
     def test_qsigmoid(self, dtype=torch.float):
         dtype = torch.qint8
 
         input0 = torch.randn(1, 1, 5, 5, device="xpu")
         q_input = torch.quantize_per_tensor(input0, 0.4, 0, dtype=dtype)
         q_input_cpu = torch.quantize_per_tensor(input0.to("cpu"), 0.4, 0, dtype=dtype)
+
         result_functional = torch.dequantize(torch.sigmoid(q_input))
         result_inplace = torch.dequantize(torch.sigmoid_(q_input))
         result_output = torch.randn(1, 1, 5, 5, device="xpu")
         result_out = torch.dequantize(torch.sigmoid(q_input, out=result_output))
-        result_cpu = torch.dequantize(torch.sigmoid(q_input_cpu))
 
-        self.assertEqual(result_functional.to("cpu"), result_cpu)
-        self.assertEqual(result_inplace.to("cpu"), result_cpu)
-        self.assertEqual(result_out.to("cpu"), result_cpu)
+        dqX = torch.dequantize(q_input_cpu)
+        Y_ref = torch.sigmoid(dqX)
+        qY_ref = torch.quantize_per_tensor(Y_ref, 1.0/255.0, 0, torch.quint8)
+        dqY_ref = qY_ref.dequantize()
+
+        self.assertEqual(result_functional.to("cpu"), dqY_ref)
+        self.assertEqual(result_inplace.to("cpu"), dqY_ref)
+        self.assertEqual(result_out.to("cpu"), dqY_ref)
+