[API Compatibility] enhance test for modified smooth_l1_loss -part

paconvert/api_mapping.json

@@ -7829,21 +7829,15 @@
     "Matcher": "ChangePrefixMatcher"
   },
   "torch.nn.SmoothL1Loss": {
-    "Matcher": "SizeAverageMatcher",
-    "paddle_api": "paddle.nn.SmoothL1Loss",
+    "Matcher": "ChangeAPIMatcher",
+    "paddle_api": "paddle.compat.nn.SmoothL1Loss",
     "min_input_args": 0,
     "args_list": [
       "size_average",
       "reduce",
       "reduction",
       "beta"
-    ],
-    "kwargs_change": {
-      "beta": "delta"
-    },
-    "paddle_default_kwargs": {
-      "is_huber": "False"
-    }
+    ]
   },
   "torch.nn.SoftMarginLoss": {
     "Matcher": "SizeAverageMatcher",
@@ -9014,8 +9008,8 @@
     "Matcher": "ChangePrefixMatcher"
   },
   "torch.nn.functional.smooth_l1_loss": {
-    "Matcher": "SizeAverageMatcher",
-    "paddle_api": "paddle.nn.functional.smooth_l1_loss",
+    "Matcher": "ChangeAPIMatcher",
+    "paddle_api": "paddle.compat.nn.functional.smooth_l1_loss",
     "args_list": [
       "input",
       "target",
@@ -9024,13 +9018,6 @@
       "reduction",
       "beta"
     ],
-    "kwargs_change": {
-      "target": "label",
-      "beta": "delta"
-    },
-    "paddle_default_kwargs": {
-      "is_huber": "False"
-    },
     "min_input_args": 2
   },
   "torch.nn.functional.soft_margin_loss": {

tests/test_nn_SmoothL1Loss.py

@@ -86,7 +86,6 @@ def test_case_5():
     obj.run(pytorch_code, ["result"])
 
 
-# paddle result has diff with pytorch result
 def test_case_6():
     pytorch_code = textwrap.dedent(
         """
@@ -100,7 +99,6 @@ def test_case_6():
     obj.run(pytorch_code, ["result"])
 
 
-# paddle result has diff with pytorch result
 def test_case_7():
     pytorch_code = textwrap.dedent(
         """
@@ -113,3 +111,44 @@ def test_case_7():
         """
     )
     obj.run(pytorch_code, ["result"])
+
+
+# beta=0 degrades to the L1 loss, aligned with PyTorch via paddle.compat
+def test_case_8():
+    pytorch_code = textwrap.dedent(
+        """
+        import torch
+        loss = torch.nn.SmoothL1Loss(beta=0.0)
+        input = torch.tensor([[1., 2., 3.], [4., 5., 6.]])
+        label = torch.tensor([[1.5, 0.5, 5.], [4., 2., 9.]])
+        result = loss(input, label)
+        """
+    )
+    obj.run(pytorch_code, ["result"])
+
+
+def test_case_9():
+    pytorch_code = textwrap.dedent(
+        """
+        import torch
+        loss = torch.nn.SmoothL1Loss(beta=0.0, reduction='none')
+        input = torch.tensor([[1., 2., 3.], [4., 5., 6.]])
+        label = torch.tensor([[1.5, 0.5, 5.], [4., 2., 9.]])
+        result = loss(input, label)
+        """
+    )
+    obj.run(pytorch_code, ["result"])
+
+
+# kwargs out of order
+def test_case_10():
+    pytorch_code = textwrap.dedent(
+        """
+        import torch
+        loss = torch.nn.SmoothL1Loss(reduction='sum', beta=0.5)
+        input = torch.ones([3, 3]).to(dtype=torch.float32)
+        label = torch.full([3, 3], 2).to(dtype=torch.float32)
+        result = loss(input, label)
+        """
+    )
+    obj.run(pytorch_code, ["result"])

tests/test_nn_functional_smooth_l1_loss.py

@@ -156,3 +156,40 @@ def test_case_11():
         """
     )
     obj.run(pytorch_code, ["result"])
+
+
+# beta=0 degrades to the L1 loss, aligned with PyTorch via paddle.compat
+def test_case_12():
+    pytorch_code = textwrap.dedent(
+        """
+        import torch
+        input = torch.tensor([[1., 2., 3.], [4., 5., 6.]])
+        label = torch.tensor([[1.5, 0.5, 5.], [4., 2., 9.]])
+        result = torch.nn.functional.smooth_l1_loss(input, label, beta=0.0)
+        """
+    )
+    obj.run(pytorch_code, ["result"])
+
+
+def test_case_13():
+    pytorch_code = textwrap.dedent(
+        """
+        import torch
+        input = torch.tensor([[1., 2., 3.], [4., 5., 6.]])
+        label = torch.tensor([[1.5, 0.5, 5.], [4., 2., 9.]])
+        result = torch.nn.functional.smooth_l1_loss(input, label, beta=0.0, reduction='sum')
+        """
+    )
+    obj.run(pytorch_code, ["result"])
+
+
+def test_case_14():
+    pytorch_code = textwrap.dedent(
+        """
+        import torch
+        input = torch.tensor([[1., 2., 3.], [4., 5., 6.]])
+        label = torch.tensor([[1.5, 0.5, 5.], [4., 2., 9.]])
+        result = torch.nn.functional.smooth_l1_loss(input, label, beta=2.0, reduction='none')
+        """
+    )
+    obj.run(pytorch_code, ["result"])