修改COPY-FROM No. 3 autograd (#54921)

* [autograd] add copy-from; test=document_fix

* [autograd] add copy-from; test=document_fix

* fix

python/paddle/autograd/py_layer.py

@@ -29,6 +29,27 @@ def with_mateclass(meta, *bases):
 class PyLayerContext:
     """
     ``PyLayerContext`` can assist the :ref:`api_paddle_autograd_PyLayer` in implementing certain functionalities.
+
+    Examples:
+        .. code-block:: python
+
+            import paddle
+            from paddle.autograd import PyLayer
+
+            class cus_tanh(PyLayer):
+                @staticmethod
+                def forward(ctx, x):
+                    # ctx is a object of PyLayerContext.
+                    y = paddle.tanh(x)
+                    ctx.save_for_backward(y)
+                    return y
+
+                @staticmethod
+                def backward(ctx, dy):
+                    # ctx is a object of PyLayerContext.
+                    y, = ctx.saved_tensor()
+                    grad = dy * (1 - paddle.square(y))
+                    return grad
     """
 
     def save_for_backward(self, *tensors):
@@ -266,6 +287,33 @@ class PyLayer(with_mateclass(PyLayerMeta, core.eager.PyLayer, PyLayerContext)):
 
     After building the custom operator, apply it by running the ``apply`` method.
 
+    Examples:
+        .. code-block:: python
+
+            import paddle
+            from paddle.autograd import PyLayer
+
+            class cus_tanh(PyLayer):
+                @staticmethod
+                def forward(ctx, x):
+                    y = paddle.tanh(x)
+                    # Pass tensors to backward.
+                    ctx.save_for_backward(y)
+                    return y
+
+                @staticmethod
+                def backward(ctx, dy):
+                    # Get the tensors passed by forward.
+                    y, = ctx.saved_tensor()
+                    grad = dy * (1 - paddle.square(y))
+                    return grad
+
+            data = paddle.randn([2, 3], dtype="float64")
+            data.stop_gradient = False
+            z = cus_tanh.apply(data)
+            z.mean().backward()
+
+            print(data.grad)
     """
 
     @staticmethod