From bfca775142fef27cb9e703df8287225394e82d31 Mon Sep 17 00:00:00 2001
From: yangxiaoyu14 <yangxiaoyu14@baidu.com>
Date: Wed, 23 Aug 2023 02:43:41 +0000
Subject: [PATCH 01/14] def dtensor_from_fn first edition

---
 build/test/auto_parallel/test_dist_tensor.py | 157 +++++++++++++++++++
 1 file changed, 157 insertions(+)
 create mode 100644 build/test/auto_parallel/test_dist_tensor.py

diff --git a/build/test/auto_parallel/test_dist_tensor.py b/build/test/auto_parallel/test_dist_tensor.py
new file mode 100644
index 00000000000000..a53d32f6cd6620
--- /dev/null
+++ b/build/test/auto_parallel/test_dist_tensor.py
@@ -0,0 +1,157 @@
+# Copyright (c) 2023 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import unittest
+
+import numpy as np
+
+import paddle
+import paddle.distributed as dist
+
+class TestDistTensor(unittest.TestCase):
+    def test_dist_tensor_creation(self):
+        shape = [10, 5]
+        mesh = dist.ProcessMesh([[0, 1], [2, 3]], dim_names=["x", "y"])
+        dist_attr = dist.DistAttr(mesh=mesh, sharding_specs=['x', 'y'])
+
+        # create dist tensor using numpy
+        dist_tensor_with_numpy = dist.shard_tensor(
+            np.ones(shape, dtype=np.float32), dist_attr=dist_attr
+        )
+
+        # create dist tensor using tensor
+        dist_tensor_with_tensor = dist.shard_tensor(
+            paddle.ones(shape), dist_attr=dist_attr
+        )
+
+        # create normal tensor
+        tensor = paddle.ones(shape)
+
+        # test dist tensor properties
+        self.assertEqual(dist_tensor_with_numpy.shape, shape)
+        self.assertEqual(dist_tensor_with_tensor.shape, shape)
+        self.assertEqual(dist_tensor_with_numpy.is_dist(), True)
+        self.assertEqual(dist_tensor_with_tensor.is_dist(), True)
+        self.assertEqual(tensor.is_dist(), False)
+        self.assertEqual(
+            str(dist_tensor_with_numpy), str(dist_tensor_with_tensor)
+        )
+        self.assertEqual(dist_tensor_with_numpy.dist_attr, dist_attr)
+        self.assertEqual(dist_tensor_with_tensor.dist_attr, dist_attr)
+
+        
+class TestDistributedTensor(unittest.TestCase):
+    def test_dtensor_from_fn(self):
+        # Define a function for generating a tensor
+        def generate_tensor_ones():
+            return paddle.ones(shape=[2, 3])
+        
+        def generate_tensor_zeros():
+             return paddle.zeros(shape=[2, 3])
+
+        def generate_tensor_random():
+              return paddle.random(shape=[2, 3])
+
+        def generate_tensor_range():
+             return paddle.range(start=1, end=7).reshape([2, 3])
+        
+
+        # Create a distributed attribute
+        mesh = dist.ProcessMesh([[2, 4, 5], [0, 1, 3]], dim_names=["x", "y"])
+        dist_attr = dist.DistAttr(mesh=mesh, sharding_specs=['x', 'y'])
+        
+        # Test with generate_tensor_ones()
+        # Call the function dtensor_from_fn with dist_attr parameter
+        result = dist.dtensor_from_fn(paddle.ones, dist_attr=dist_attr, shape=[2, 3])
+
+        # Verify the result
+        self.assertIsInstance(result, paddle.Tensor)
+        self.assertEqual(result.shape, [2, 3])
+        self.assertEqual(result.dist_attr, dist_attr)
+
+        # Test with generate_tensor_zeros()
+        result_zeros = dist.dtensor_from_fn(paddle.zeros, dist_attr=dist_attr, shape=[2, 3])
+        self.assertIsInstance(result_zeros, paddle.Tensor)
+        self.assertEqual(result_zeros.shape, [2, 3])
+        self.assertEqual(result_zeros.dist_attr, dist_attr)
+
+        # Test with generate_tensor_random()
+        result_random = dist.dtensor_from_fn(paddle.random, dist_attr=dist_attr, shape=[2, 3])
+        self.assertIsInstance(result_random, paddle.Tensor)
+        self.assertEqual(result_random.shape, [2, 3])
+        self.assertEqual(result_random.dist_attr, dist_attr)
+
+        # Test with generate_tensor_range()
+        result_range = dist.dtensor_from_fn(paddle.range, dist_attr=dist_attr, shape=[2, 3])
+        self.assertIsInstance(result_range, paddle.Tensor)
+        self.assertEqual(result_range.shape, [2, 3])
+        self.assertEqual(result_range.dist_attr, dist_attr)
+
+"""
+         # Additional assertions
+        self.assertTrue((result.numpy() == 1).all())  # Check tensor values
+
+        # Test with another function
+        def generate_tensor_zeros():
+            return paddle.zeros(shape=[4, 2])
+
+        # Call the function dtensor_from_fn with dist_attr parameter and another function
+        result_zeros = dist.dtensor_from_fn(generate_tensor_zeros, dist_attr=dist_attr)
+
+        # Verify the result
+        self.assertIsInstance(result_zeros, paddle.Tensor)
+        self.assertEqual(result_zeros.shape, [4, 2])
+        self.assertEqual(result_zeros.dist_attr, dist_attr)
+
+        # Additional assertions
+        self.assertTrue((result_zeros.numpy() == 0).all())  # Check tensor values
+
+         # Test static mode (NotImplementedError should be raised)
+        with self.assertRaises(NotImplementedError):
+            with paddle.static.program_guard(paddle.static.Program()):
+                # Call the function dtensor_from_fn with dist_attr parameter
+                result = dist.dtensor_from_fn(generate_tensor, dist_attr)
+
+    def test_dtensor_from_fn_additional_args(self):
+        # Define a function for generating a tensor with additional arguments
+        def generate_tensor_with_args(shape, value):
+            return paddle.full(shape=shape, fill_value=value)
+
+        # Create a distributed attribute
+        mesh = dist.ProcessMesh([[2, 4, 5], [0, 1, 3]], dim_names=["x", "y"])
+        dist_attr = dist.DistAttr(mesh=mesh, sharding_specs=['x', 'y'])
+
+
+        # Test dynamic mode
+        with paddle.fluid.dygraph.guard():
+            # Call the function dtensor_from_fn with additional arguments and dist_attr parameter
+            result = dist.dtensor_from_fn(generate_tensor_with_args, dist_attr, shape=[2, 3], value=5)
+
+            # Verify the result
+            self.assertIsInstance(result, paddle.Tensor)
+            self.assertEqual(result.shape, [2, 3])
+            self.assertEqual(result.dist_attr, dist_attr)
+            self.assertTrue((result.numpy() == 5).all())
+
+            # Additional assertions
+            self.assertEqual(result.numpy().sum(), 30)  # Check tensor sum
+
+        # Test static mode (NotImplementedError should be raised)
+        with self.assertRaises(NotImplementedError):
+            with paddle.static.program_guard(paddle.static.Program()):
+                # Call the function dtensor_from_fn with additional arguments and dist_attr parameter
+                result = dist.dtensor_from_fn(generate_tensor_with_args, dist_attr, shape=[2, 3], value=5)        
+"""
+if __name__ == "__main__":
+    unittest.main()
\ No newline at end of file

From d20a0328f8f2746b506798187f2b940870de2dc1 Mon Sep 17 00:00:00 2001
From: yangxiaoyu14 <yangxiaoyu14@baidu.com>
Date: Wed, 23 Aug 2023 02:45:02 +0000
Subject: [PATCH 02/14] dtensor_from_fn first edition

---
 python/paddle/distributed/__init__.py         |  2 +
 .../paddle/distributed/auto_parallel/api.py   | 39 +++++++++++++++++-
 test/auto_parallel/test_dist_tensor.py        | 41 ++++++++++++++++++-
 3 files changed, 79 insertions(+), 3 deletions(-)

diff --git a/python/paddle/distributed/__init__.py b/python/paddle/distributed/__init__.py
index 183f307607c362..fe914bbb3422a5 100644
--- a/python/paddle/distributed/__init__.py
+++ b/python/paddle/distributed/__init__.py
@@ -66,6 +66,7 @@
 
 from .auto_parallel import shard_op  # noqa: F401
 from .auto_parallel.api import shard_tensor  # noqa: F401
+from .auto_parallel.api import dtensor_from_fn  # noqa: F401
 
 from .fleet import BoxPSDataset  # noqa: F401
 
@@ -126,4 +127,5 @@
     "ProcessMesh",
     "DistAttr",
     "shard_tensor",
+    "dtensor_from_fn",
 ]
diff --git a/python/paddle/distributed/auto_parallel/api.py b/python/paddle/distributed/auto_parallel/api.py
index b25799d058ad2f..acbc96a748d7ed 100644
--- a/python/paddle/distributed/auto_parallel/api.py
+++ b/python/paddle/distributed/auto_parallel/api.py
@@ -24,7 +24,7 @@
 class DistAttr(core.TensorDistAttr):
     """
     DistAttr specifies how tensors are distributed or sliced on ProcessMesh.
-
+        
     Args:
         mesh(paddle.distributed.ProcessMesh): The `ProcessMesh` object describes the Cartesian topology of the used processes.
         sharding_specs(list[str|None]): The specification describing how to shard the Tensor.
@@ -124,3 +124,40 @@ def shard_tensor(
         raise NotImplementedError(
             "The `paddle.distributed.shard_tensor` for static mode will be implemented later."
         )
+
+def dtensor_from_fn(
+        fn, dist_attr, *args, **kwargs
+):
+    """
+    Construct a Distributed Tensor from a function of arguments.
+
+    Args:
+        fn (callable): A callable function that takes arguments of Distributed Tensor and returns tensor.
+        dist_attr(paddle.distributed.DistAttr): Specify how tensors are distributed or sliced on ProcessMesh.
+        *args: A list of arguments to be passed to the ``fn`` function.
+        **kwargs: A list of arguments to be passed to the ``fn`` function.
+
+    Retruns:
+        Tensor: A Tensor constructed from ``fn`` with distributed attributes.       
+
+    Examples:
+
+    .. code-block:: python     
+
+        import paddle
+        import paddle.distribute as dist
+
+        def generate_tensor():
+            return paddle.ones(shape=[2, 3])
+
+        # Create a distributed attribute
+        mesh = dist.ProcessMesh([[2, 4, 5], [0, 1, 3]], dim_names=["x", "y"])
+        dist_attr = dist.DistAttr(mesh=mesh, sharding_specs=['x', 'y'])
+
+        # Call the function dtensor_from_fn with dist_attr parameter
+        d_tensor = dist.dtensor_from_fn(paddle.ones, dist_attr=dist_attr, shape=[2, 3])
+
+        print(d_tensor)
+    """ 
+    tensor = fn(*args, **kwargs)
+    return shard_tensor(tensor, dist_attr=dist_attr)
diff --git a/test/auto_parallel/test_dist_tensor.py b/test/auto_parallel/test_dist_tensor.py
index 61705a322e2d60..680f07a90b755a 100644
--- a/test/auto_parallel/test_dist_tensor.py
+++ b/test/auto_parallel/test_dist_tensor.py
@@ -19,7 +19,6 @@
 import paddle
 import paddle.distributed as dist
 
-
 class TestDistTensor(unittest.TestCase):
     def test_dist_tensor_creation(self):
         shape = [10, 5]
@@ -51,6 +50,44 @@ def test_dist_tensor_creation(self):
         self.assertEqual(dist_tensor_with_numpy.dist_attr, dist_attr)
         self.assertEqual(dist_tensor_with_tensor.dist_attr, dist_attr)
 
+        
+class TestDistributedTensor(unittest.TestCase):
+    def test_dtensor_from_fn(self):
+        # Define a function for generating a tensor
+        def generate_tensor_ones():
+            return paddle.ones(shape=[2, 3])
+        
+        def generate_tensor_zeros():
+             return paddle.zeros(shape=[2, 3])
+
+        def generate_tensor_random():
+              return paddle.rand(shape=[2, 3])
+        
+
+        # Create a distributed attribute
+        mesh = dist.ProcessMesh([[2, 4, 5], [0, 1, 3]], dim_names=["x", "y"])
+        dist_attr = dist.DistAttr(mesh=mesh, sharding_specs=['x', 'y'])
+        
+        # Test with generate_tensor_ones()
+        # Call the function dtensor_from_fn with dist_attr parameter
+        result = dist.dtensor_from_fn(paddle.ones, dist_attr=dist_attr, shape=[2, 3])
+
+        # Verify the result
+        self.assertIsInstance(result, paddle.Tensor)
+        self.assertEqual(result.shape, [2, 3])
+        self.assertEqual(result.dist_attr, dist_attr)
+
+        # Test with generate_tensor_zeros()
+        result_zeros = dist.dtensor_from_fn(paddle.zeros, dist_attr=dist_attr, shape=[2, 3])
+        self.assertIsInstance(result_zeros, paddle.Tensor)
+        self.assertEqual(result_zeros.shape, [2, 3])
+        self.assertEqual(result_zeros.dist_attr, dist_attr)
+
+        # Test with generate_tensor_random()
+        result_random = dist.dtensor_from_fn(paddle.rand, dist_attr=dist_attr, shape=[2, 3])
+        self.assertIsInstance(result_random, paddle.Tensor)
+        self.assertEqual(result_random.shape, [2, 3])
+        self.assertEqual(result_random.dist_attr, dist_attr)
 
 if __name__ == "__main__":
-    unittest.main()
+    unittest.main()
\ No newline at end of file

From 3902f7d1e3cea95aeff7449344af34ca8afc0283 Mon Sep 17 00:00:00 2001
From: yangxiaoyu14 <yangxiaoyu14@baidu.com>
Date: Thu, 24 Aug 2023 02:28:12 +0000
Subject: [PATCH 03/14] Delete file
 /home/Paddle/build/test/auto_parallel/test_dist_tensor.py

---
 build/test/auto_parallel/test_dist_tensor.py | 157 -------------------
 1 file changed, 157 deletions(-)
 delete mode 100644 build/test/auto_parallel/test_dist_tensor.py

diff --git a/build/test/auto_parallel/test_dist_tensor.py b/build/test/auto_parallel/test_dist_tensor.py
deleted file mode 100644
index a53d32f6cd6620..00000000000000
--- a/build/test/auto_parallel/test_dist_tensor.py
+++ /dev/null
@@ -1,157 +0,0 @@
-# Copyright (c) 2023 PaddlePaddle Authors. All Rights Reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-import unittest
-
-import numpy as np
-
-import paddle
-import paddle.distributed as dist
-
-class TestDistTensor(unittest.TestCase):
-    def test_dist_tensor_creation(self):
-        shape = [10, 5]
-        mesh = dist.ProcessMesh([[0, 1], [2, 3]], dim_names=["x", "y"])
-        dist_attr = dist.DistAttr(mesh=mesh, sharding_specs=['x', 'y'])
-
-        # create dist tensor using numpy
-        dist_tensor_with_numpy = dist.shard_tensor(
-            np.ones(shape, dtype=np.float32), dist_attr=dist_attr
-        )
-
-        # create dist tensor using tensor
-        dist_tensor_with_tensor = dist.shard_tensor(
-            paddle.ones(shape), dist_attr=dist_attr
-        )
-
-        # create normal tensor
-        tensor = paddle.ones(shape)
-
-        # test dist tensor properties
-        self.assertEqual(dist_tensor_with_numpy.shape, shape)
-        self.assertEqual(dist_tensor_with_tensor.shape, shape)
-        self.assertEqual(dist_tensor_with_numpy.is_dist(), True)
-        self.assertEqual(dist_tensor_with_tensor.is_dist(), True)
-        self.assertEqual(tensor.is_dist(), False)
-        self.assertEqual(
-            str(dist_tensor_with_numpy), str(dist_tensor_with_tensor)
-        )
-        self.assertEqual(dist_tensor_with_numpy.dist_attr, dist_attr)
-        self.assertEqual(dist_tensor_with_tensor.dist_attr, dist_attr)
-
-        
-class TestDistributedTensor(unittest.TestCase):
-    def test_dtensor_from_fn(self):
-        # Define a function for generating a tensor
-        def generate_tensor_ones():
-            return paddle.ones(shape=[2, 3])
-        
-        def generate_tensor_zeros():
-             return paddle.zeros(shape=[2, 3])
-
-        def generate_tensor_random():
-              return paddle.random(shape=[2, 3])
-
-        def generate_tensor_range():
-             return paddle.range(start=1, end=7).reshape([2, 3])
-        
-
-        # Create a distributed attribute
-        mesh = dist.ProcessMesh([[2, 4, 5], [0, 1, 3]], dim_names=["x", "y"])
-        dist_attr = dist.DistAttr(mesh=mesh, sharding_specs=['x', 'y'])
-        
-        # Test with generate_tensor_ones()
-        # Call the function dtensor_from_fn with dist_attr parameter
-        result = dist.dtensor_from_fn(paddle.ones, dist_attr=dist_attr, shape=[2, 3])
-
-        # Verify the result
-        self.assertIsInstance(result, paddle.Tensor)
-        self.assertEqual(result.shape, [2, 3])
-        self.assertEqual(result.dist_attr, dist_attr)
-
-        # Test with generate_tensor_zeros()
-        result_zeros = dist.dtensor_from_fn(paddle.zeros, dist_attr=dist_attr, shape=[2, 3])
-        self.assertIsInstance(result_zeros, paddle.Tensor)
-        self.assertEqual(result_zeros.shape, [2, 3])
-        self.assertEqual(result_zeros.dist_attr, dist_attr)
-
-        # Test with generate_tensor_random()
-        result_random = dist.dtensor_from_fn(paddle.random, dist_attr=dist_attr, shape=[2, 3])
-        self.assertIsInstance(result_random, paddle.Tensor)
-        self.assertEqual(result_random.shape, [2, 3])
-        self.assertEqual(result_random.dist_attr, dist_attr)
-
-        # Test with generate_tensor_range()
-        result_range = dist.dtensor_from_fn(paddle.range, dist_attr=dist_attr, shape=[2, 3])
-        self.assertIsInstance(result_range, paddle.Tensor)
-        self.assertEqual(result_range.shape, [2, 3])
-        self.assertEqual(result_range.dist_attr, dist_attr)
-
-"""
-         # Additional assertions
-        self.assertTrue((result.numpy() == 1).all())  # Check tensor values
-
-        # Test with another function
-        def generate_tensor_zeros():
-            return paddle.zeros(shape=[4, 2])
-
-        # Call the function dtensor_from_fn with dist_attr parameter and another function
-        result_zeros = dist.dtensor_from_fn(generate_tensor_zeros, dist_attr=dist_attr)
-
-        # Verify the result
-        self.assertIsInstance(result_zeros, paddle.Tensor)
-        self.assertEqual(result_zeros.shape, [4, 2])
-        self.assertEqual(result_zeros.dist_attr, dist_attr)
-
-        # Additional assertions
-        self.assertTrue((result_zeros.numpy() == 0).all())  # Check tensor values
-
-         # Test static mode (NotImplementedError should be raised)
-        with self.assertRaises(NotImplementedError):
-            with paddle.static.program_guard(paddle.static.Program()):
-                # Call the function dtensor_from_fn with dist_attr parameter
-                result = dist.dtensor_from_fn(generate_tensor, dist_attr)
-
-    def test_dtensor_from_fn_additional_args(self):
-        # Define a function for generating a tensor with additional arguments
-        def generate_tensor_with_args(shape, value):
-            return paddle.full(shape=shape, fill_value=value)
-
-        # Create a distributed attribute
-        mesh = dist.ProcessMesh([[2, 4, 5], [0, 1, 3]], dim_names=["x", "y"])
-        dist_attr = dist.DistAttr(mesh=mesh, sharding_specs=['x', 'y'])
-
-
-        # Test dynamic mode
-        with paddle.fluid.dygraph.guard():
-            # Call the function dtensor_from_fn with additional arguments and dist_attr parameter
-            result = dist.dtensor_from_fn(generate_tensor_with_args, dist_attr, shape=[2, 3], value=5)
-
-            # Verify the result
-            self.assertIsInstance(result, paddle.Tensor)
-            self.assertEqual(result.shape, [2, 3])
-            self.assertEqual(result.dist_attr, dist_attr)
-            self.assertTrue((result.numpy() == 5).all())
-
-            # Additional assertions
-            self.assertEqual(result.numpy().sum(), 30)  # Check tensor sum
-
-        # Test static mode (NotImplementedError should be raised)
-        with self.assertRaises(NotImplementedError):
-            with paddle.static.program_guard(paddle.static.Program()):
-                # Call the function dtensor_from_fn with additional arguments and dist_attr parameter
-                result = dist.dtensor_from_fn(generate_tensor_with_args, dist_attr, shape=[2, 3], value=5)        
-"""
-if __name__ == "__main__":
-    unittest.main()
\ No newline at end of file

From 50fd5a951ac0432722d3547026bf18ce64716511 Mon Sep 17 00:00:00 2001
From: yangxiaoyu14 <yangxiaoyu14@baidu.com>
Date: Mon, 28 Aug 2023 02:40:20 +0000
Subject: [PATCH 04/14] polish code format

---
 .../paddle/distributed/auto_parallel/api.py   | 34 ++++----------
 test/auto_parallel/test_dist_tensor.py        | 47 ++++++++++++-------
 2 files changed, 39 insertions(+), 42 deletions(-)

diff --git a/python/paddle/distributed/auto_parallel/api.py b/python/paddle/distributed/auto_parallel/api.py
index 1bcce2f5a882d7..7769eac884dc27 100644
--- a/python/paddle/distributed/auto_parallel/api.py
+++ b/python/paddle/distributed/auto_parallel/api.py
@@ -12,6 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+
 import paddle
 from paddle.distributed.auto_parallel.interface import (
     shard_tensor as shard_tensor_static,
@@ -26,7 +27,7 @@
 class DistAttr(core.TensorDistAttr):
     """
     DistAttr specifies how tensors are distributed or sliced on ProcessMesh.
-        
+
     Args:
         mesh(paddle.distributed.ProcessMesh): The `ProcessMesh` object describes the Cartesian topology of the used processes.
         sharding_specs(list[str|None]): The specification describing how to shard the Tensor.
@@ -69,19 +70,6 @@ def __init__(self, mesh, sharding_specs):
         self.process_mesh = mesh
         self.dims_mapping = dims_mapping
 
-        self.mark_annotated("process_mesh")
-        self.mark_annotated("dims_mapping")
-
-    @property
-    def sharding_specs(self):
-        """
-        Get sharding_specs of the dist_attr
-
-        Returns:
-            list[str]: sharding_specs
-        """
-        return self._sharding_specs
-
 
 def shard_tensor(
     data, dtype=None, place=None, stop_gradient=True, dist_attr=None
@@ -143,31 +131,27 @@ def shard_tensor(
             data, dist_attr.process_mesh, dist_attr.sharding_specs
         )
 
-def dtensor_from_fn(
-        fn, dist_attr, *args, **kwargs
-):
+
+def dtensor_from_fn(fn, dist_attr, *args, **kwargs):
     """
-    Construct a Distributed Tensor from a function of arguments.
+    Construct a Distributed Tensor from a paddle api function of arguments.
 
     Args:
-        fn (callable): A callable function that takes arguments of Distributed Tensor and returns tensor.
+        fn (callable): A paddle api function that takes arguments of *args, **kwargs and returns tensor.
         dist_attr(paddle.distributed.DistAttr): Specify how tensors are distributed or sliced on ProcessMesh.
         *args: A list of arguments to be passed to the ``fn`` function.
         **kwargs: A list of arguments to be passed to the ``fn`` function.
 
     Retruns:
-        Tensor: A Tensor constructed from ``fn`` with distributed attributes.       
+        Tensor: A Tensor constructed from ``fn`` with distributed attributes.
 
     Examples:
 
-    .. code-block:: python     
+    .. code-block:: python
 
         import paddle
         import paddle.distribute as dist
 
-        def generate_tensor():
-            return paddle.ones(shape=[2, 3])
-
         # Create a distributed attribute
         mesh = dist.ProcessMesh([[2, 4, 5], [0, 1, 3]], dim_names=["x", "y"])
         dist_attr = dist.DistAttr(mesh=mesh, sharding_specs=['x', 'y'])
@@ -176,6 +160,6 @@ def generate_tensor():
         d_tensor = dist.dtensor_from_fn(paddle.ones, dist_attr=dist_attr, shape=[2, 3])
 
         print(d_tensor)
-    """ 
+    """
     tensor = fn(*args, **kwargs)
     return shard_tensor(tensor, dist_attr=dist_attr)
diff --git a/test/auto_parallel/test_dist_tensor.py b/test/auto_parallel/test_dist_tensor.py
index aae178b5d4e7d6..c6660c3ae63b66 100644
--- a/test/auto_parallel/test_dist_tensor.py
+++ b/test/auto_parallel/test_dist_tensor.py
@@ -20,6 +20,7 @@
 import paddle.distributed as dist
 import paddle.nn.functional as F
 
+
 class TestDistTensor(unittest.TestCase):
     def test_dist_tensor_creation(self):
         shape = [10, 5]
@@ -51,27 +52,18 @@ def test_dist_tensor_creation(self):
         self.assertEqual(dist_tensor_with_numpy.dist_attr, dist_attr)
         self.assertEqual(dist_tensor_with_tensor.dist_attr, dist_attr)
 
-        
+
 class TestDistributedTensor(unittest.TestCase):
     def test_dtensor_from_fn(self):
-        # Define a function for generating a tensor
-        def generate_tensor_ones():
-            return paddle.ones(shape=[2, 3])
-        
-        def generate_tensor_zeros():
-             return paddle.zeros(shape=[2, 3])
-
-        def generate_tensor_random():
-              return paddle.rand(shape=[2, 3])
-        
-
         # Create a distributed attribute
         mesh = dist.ProcessMesh([[2, 4, 5], [0, 1, 3]], dim_names=["x", "y"])
         dist_attr = dist.DistAttr(mesh=mesh, sharding_specs=['x', 'y'])
-        
+
         # Test with generate_tensor_ones()
         # Call the function dtensor_from_fn with dist_attr parameter
-        result = dist.dtensor_from_fn(paddle.ones, dist_attr=dist_attr, shape=[2, 3])
+        result = dist.dtensor_from_fn(
+            paddle.ones, dist_attr=dist_attr, shape=[2, 3]
+        )
 
         # Verify the result
         self.assertIsInstance(result, paddle.Tensor)
@@ -79,17 +71,38 @@ def generate_tensor_random():
         self.assertEqual(result.dist_attr, dist_attr)
 
         # Test with generate_tensor_zeros()
-        result_zeros = dist.dtensor_from_fn(paddle.zeros, dist_attr=dist_attr, shape=[2, 3])
+        result_zeros = dist.dtensor_from_fn(
+            paddle.zeros, dist_attr=dist_attr, shape=[2, 3]
+        )
         self.assertIsInstance(result_zeros, paddle.Tensor)
         self.assertEqual(result_zeros.shape, [2, 3])
         self.assertEqual(result_zeros.dist_attr, dist_attr)
 
         # Test with generate_tensor_random()
-        result_random = dist.dtensor_from_fn(paddle.rand, dist_attr=dist_attr, shape=[2, 3])
+        result_random = dist.dtensor_from_fn(
+            paddle.rand, dist_attr=dist_attr, shape=[2, 3]
+        )
         self.assertIsInstance(result_random, paddle.Tensor)
         self.assertEqual(result_random.shape, [2, 3])
         self.assertEqual(result_random.dist_attr, dist_attr)
 
+        # Test with invalid sharding_specs length
+        with self.assertRaises(AssertionError):
+            invalid_dist_attr = dist.DistAttr(mesh=mesh, sharding_specs=['x'])
+            dist.dtensor_from_fn(
+                paddle.ones, dist_attr=invalid_dist_attr, shape=[2, 3]
+            )
+
+        # Test exceptions when running in static mode
+        paddle.enable_static()
+        with self.assertRaises(NotImplementedError):
+            with paddle.static.program_guard(paddle.static.Program()):
+                dist.dtensor_from_fn(
+                    paddle.ones, dist_attr=dist_attr, shape=[2, 3]
+                )
+        paddle.disable_static()
+
+
 class TestDistTensorForDygraphAPI(unittest.TestCase):
     def check_tensor_eq(self, a, b):
         np1 = a.numpy()
@@ -122,4 +135,4 @@ def test_relu_api_for_dist_tensor(self):
 
 
 if __name__ == "__main__":
-    unittest.main()
\ No newline at end of file
+    unittest.main()

From 3953a5e02341cce1014f06e7d5669e7ddc6e73b2 Mon Sep 17 00:00:00 2001
From: yangxiaoyu14 <yangxiaoyu14@baidu.com>
Date: Tue, 29 Aug 2023 08:20:40 +0000
Subject: [PATCH 05/14] fix sample code formatting issues

---
 python/paddle/distributed/auto_parallel/api.py | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)

diff --git a/python/paddle/distributed/auto_parallel/api.py b/python/paddle/distributed/auto_parallel/api.py
index 7769eac884dc27..09a8bb00747536 100644
--- a/python/paddle/distributed/auto_parallel/api.py
+++ b/python/paddle/distributed/auto_parallel/api.py
@@ -149,17 +149,17 @@ def dtensor_from_fn(fn, dist_attr, *args, **kwargs):
 
     .. code-block:: python
 
-        import paddle
-        import paddle.distribute as dist
+     >>>import paddle
+     >>>import paddle.distribute as dist
 
-        # Create a distributed attribute
-        mesh = dist.ProcessMesh([[2, 4, 5], [0, 1, 3]], dim_names=["x", "y"])
-        dist_attr = dist.DistAttr(mesh=mesh, sharding_specs=['x', 'y'])
+     >>># Create a distributed attribute
+     >>>mesh = dist.ProcessMesh([[2, 4, 5], [0, 1, 3]], dim_names=["x", "y"])
+     >>>dist_attr = dist.DistAttr(mesh=mesh, sharding_specs=['x', 'y'])
 
-        # Call the function dtensor_from_fn with dist_attr parameter
-        d_tensor = dist.dtensor_from_fn(paddle.ones, dist_attr=dist_attr, shape=[2, 3])
+     >>># Call the function dtensor_from_fn with dist_attr parameter
+     >>>d_tensor = dist.dtensor_from_fn(paddle.ones, dist_attr=dist_attr, shape=[2, 3])
 
-        print(d_tensor)
+     >>>print(d_tensor)
     """
     tensor = fn(*args, **kwargs)
     return shard_tensor(tensor, dist_attr=dist_attr)

From 1d510ea743c1c8c93e9b9cecb56e6feedb694281 Mon Sep 17 00:00:00 2001
From: yangxiaoyu14 <yangxiaoyu14@baidu.com>
Date: Wed, 30 Aug 2023 02:22:37 +0000
Subject: [PATCH 06/14] change sample codes ' >>>' to '>>> '

---
 python/paddle/distributed/auto_parallel/api.py | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)

diff --git a/python/paddle/distributed/auto_parallel/api.py b/python/paddle/distributed/auto_parallel/api.py
index 09a8bb00747536..3ae0db7226c424 100644
--- a/python/paddle/distributed/auto_parallel/api.py
+++ b/python/paddle/distributed/auto_parallel/api.py
@@ -149,17 +149,17 @@ def dtensor_from_fn(fn, dist_attr, *args, **kwargs):
 
     .. code-block:: python
 
-     >>>import paddle
-     >>>import paddle.distribute as dist
+    >>> import paddle
+    >>> import paddle.distribute as dist
 
-     >>># Create a distributed attribute
-     >>>mesh = dist.ProcessMesh([[2, 4, 5], [0, 1, 3]], dim_names=["x", "y"])
-     >>>dist_attr = dist.DistAttr(mesh=mesh, sharding_specs=['x', 'y'])
+    >>> # Create a distributed attribute
+    >>> mesh = dist.ProcessMesh([[2, 4, 5], [0, 1, 3]], dim_names=["x", "y"])
+    >>> dist_attr = dist.DistAttr(mesh=mesh, sharding_specs=['x', 'y'])
 
-     >>># Call the function dtensor_from_fn with dist_attr parameter
-     >>>d_tensor = dist.dtensor_from_fn(paddle.ones, dist_attr=dist_attr, shape=[2, 3])
+    >>> # Call the function dtensor_from_fn with dist_attr parameter
+    >>> d_tensor = dist.dtensor_from_fn(paddle.ones, dist_attr=dist_attr, shape=[2, 3])
 
-     >>>print(d_tensor)
+    >>> print(d_tensor)
     """
     tensor = fn(*args, **kwargs)
     return shard_tensor(tensor, dist_attr=dist_attr)

From 47564d2d164e46d08cbe93ff150498ee4d6d2e3c Mon Sep 17 00:00:00 2001
From: yangxiaoyu14 <yangxiaoyu14@baidu.com>
Date: Thu, 31 Aug 2023 05:30:35 +0000
Subject: [PATCH 07/14] Add static image single measurement

---
 .../paddle/distributed/auto_parallel/api.py   | 26 +++++----
 test/auto_parallel/test_dist_tensor.py        | 54 +++++++++++--------
 2 files changed, 48 insertions(+), 32 deletions(-)

diff --git a/python/paddle/distributed/auto_parallel/api.py b/python/paddle/distributed/auto_parallel/api.py
index 3ae0db7226c424..88733a5266e06a 100644
--- a/python/paddle/distributed/auto_parallel/api.py
+++ b/python/paddle/distributed/auto_parallel/api.py
@@ -69,6 +69,17 @@ def __init__(self, mesh, sharding_specs):
 
         self.process_mesh = mesh
         self.dims_mapping = dims_mapping
+        self.mark_annotated("process_mesh")
+        self.mark_annotated("dims_mapping")
+
+    @property
+    def sharding_specs(self):
+        """
+        Get sharding_specs of the dist_attr
+        Returns:
+            list[str]: sharding_specs
+        """
+        return self._sharding_specs
 
 
 def shard_tensor(
@@ -134,13 +145,13 @@ def shard_tensor(
 
 def dtensor_from_fn(fn, dist_attr, *args, **kwargs):
     """
-    Construct a Distributed Tensor from a paddle api function of arguments.
+    Construct a Distributed Tensor from a function of arguments.
 
     Args:
-        fn (callable): A paddle api function that takes arguments of *args, **kwargs and returns tensor.
-        dist_attr(paddle.distributed.DistAttr): Specify how tensors are distributed or sliced on ProcessMesh.
-        *args: A list of arguments to be passed to the ``fn`` function.
-        **kwargs: A list of arguments to be passed to the ``fn`` function.
+        fn (callable): A callable function that takes arguments of Distributed Tensor and returns tensor.
+        dist_attr (paddle.distributed.DistAttr): Specify how tensors are distributed or sliced on ProcessMesh.
+        *args (tuple): A tuple of arguments to be passed to the ``fn`` function.
+        **kwargs (dict): A dict of arguments to be passed to the ``fn`` function.
 
     Retruns:
         Tensor: A Tensor constructed from ``fn`` with distributed attributes.
@@ -150,15 +161,12 @@ def dtensor_from_fn(fn, dist_attr, *args, **kwargs):
     .. code-block:: python
 
     >>> import paddle
-    >>> import paddle.distribute as dist
-
+    >>> import paddle.distributed as dist
     >>> # Create a distributed attribute
     >>> mesh = dist.ProcessMesh([[2, 4, 5], [0, 1, 3]], dim_names=["x", "y"])
     >>> dist_attr = dist.DistAttr(mesh=mesh, sharding_specs=['x', 'y'])
-
     >>> # Call the function dtensor_from_fn with dist_attr parameter
     >>> d_tensor = dist.dtensor_from_fn(paddle.ones, dist_attr=dist_attr, shape=[2, 3])
-
     >>> print(d_tensor)
     """
     tensor = fn(*args, **kwargs)
diff --git a/test/auto_parallel/test_dist_tensor.py b/test/auto_parallel/test_dist_tensor.py
index c6660c3ae63b66..40615e56b8c052 100644
--- a/test/auto_parallel/test_dist_tensor.py
+++ b/test/auto_parallel/test_dist_tensor.py
@@ -53,38 +53,46 @@ def test_dist_tensor_creation(self):
         self.assertEqual(dist_tensor_with_tensor.dist_attr, dist_attr)
 
 
-class TestDistributedTensor(unittest.TestCase):
-    def test_dtensor_from_fn(self):
+class TestDistTensorFromFn(unittest.TestCase):
+    def run_dtensor_from_fn(self):
         # Create a distributed attribute
-        mesh = dist.ProcessMesh([[2, 4, 5], [0, 1, 3]], dim_names=["x", "y"])
-        dist_attr = dist.DistAttr(mesh=mesh, sharding_specs=['x', 'y'])
+        mesh = dist.ProcessMesh([0, 1], dim_names=["x"])
+        dist_attr = dist.DistAttr(mesh=mesh, sharding_specs=[None])
 
-        # Test with generate_tensor_ones()
         # Call the function dtensor_from_fn with dist_attr parameter
         result = dist.dtensor_from_fn(
-            paddle.ones, dist_attr=dist_attr, shape=[2, 3]
+            paddle.ones, dist_attr=dist_attr, shape=[1]
         )
 
         # Verify the result
-        self.assertIsInstance(result, paddle.Tensor)
-        self.assertEqual(result.shape, [2, 3])
-        self.assertEqual(result.dist_attr, dist_attr)
+        if paddle.in_dynamic_mode():
+            self.assertIsInstance(result, paddle.Tensor)
+            self.assertEqual(result.shape, [1])
+        else:
+            self.assertIsInstance(result, paddle.fluid.framework.Variable)
+            self.assertEqual(result.shape, (1,))
 
         # Test with generate_tensor_zeros()
         result_zeros = dist.dtensor_from_fn(
-            paddle.zeros, dist_attr=dist_attr, shape=[2, 3]
+            paddle.zeros, dist_attr=dist_attr, shape=[1]
         )
-        self.assertIsInstance(result_zeros, paddle.Tensor)
-        self.assertEqual(result_zeros.shape, [2, 3])
-        self.assertEqual(result_zeros.dist_attr, dist_attr)
+        if paddle.in_dynamic_mode():
+            self.assertIsInstance(result, paddle.Tensor)
+            self.assertEqual(result.shape, [1])
+        else:
+            self.assertIsInstance(result, paddle.fluid.framework.Variable)
+            self.assertEqual(result.shape, (1,))
 
         # Test with generate_tensor_random()
         result_random = dist.dtensor_from_fn(
-            paddle.rand, dist_attr=dist_attr, shape=[2, 3]
+            paddle.rand, dist_attr=dist_attr, shape=[1]
         )
-        self.assertIsInstance(result_random, paddle.Tensor)
-        self.assertEqual(result_random.shape, [2, 3])
-        self.assertEqual(result_random.dist_attr, dist_attr)
+        if paddle.in_dynamic_mode():
+            self.assertIsInstance(result, paddle.Tensor)
+            self.assertEqual(result.shape, [1])
+        else:
+            self.assertIsInstance(result, paddle.fluid.framework.Variable)
+            self.assertEqual(result.shape, (1,))
 
         # Test with invalid sharding_specs length
         with self.assertRaises(AssertionError):
@@ -93,13 +101,13 @@ def test_dtensor_from_fn(self):
                 paddle.ones, dist_attr=invalid_dist_attr, shape=[2, 3]
             )
 
-        # Test exceptions when running in static mode
+    def test_dynamic_mode(self):
+        self.run_dtensor_from_fn()
+
+    # Test exceptions when running in static mode
+    def test_static_mode(self):
         paddle.enable_static()
-        with self.assertRaises(NotImplementedError):
-            with paddle.static.program_guard(paddle.static.Program()):
-                dist.dtensor_from_fn(
-                    paddle.ones, dist_attr=dist_attr, shape=[2, 3]
-                )
+        self.run_dtensor_from_fn()
         paddle.disable_static()
 
 

From dd6c04f28a30bdbd91a9772ea607d71ef0120037 Mon Sep 17 00:00:00 2001
From: yangxiaoyu14 <yangxiaoyu14@baidu.com>
Date: Fri, 1 Sep 2023 03:24:20 +0000
Subject: [PATCH 08/14] modify the Indent of Sample Code

---
 python/paddle/distributed/auto_parallel/api.py | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)

diff --git a/python/paddle/distributed/auto_parallel/api.py b/python/paddle/distributed/auto_parallel/api.py
index 88733a5266e06a..0a061fb6b35a81 100644
--- a/python/paddle/distributed/auto_parallel/api.py
+++ b/python/paddle/distributed/auto_parallel/api.py
@@ -160,14 +160,14 @@ def dtensor_from_fn(fn, dist_attr, *args, **kwargs):
 
     .. code-block:: python
 
-    >>> import paddle
-    >>> import paddle.distributed as dist
-    >>> # Create a distributed attribute
-    >>> mesh = dist.ProcessMesh([[2, 4, 5], [0, 1, 3]], dim_names=["x", "y"])
-    >>> dist_attr = dist.DistAttr(mesh=mesh, sharding_specs=['x', 'y'])
-    >>> # Call the function dtensor_from_fn with dist_attr parameter
-    >>> d_tensor = dist.dtensor_from_fn(paddle.ones, dist_attr=dist_attr, shape=[2, 3])
-    >>> print(d_tensor)
+        >>> import paddle
+        >>> import paddle.distribute as dist
+        >>> # Create a distributed attribute
+        >>> mesh = dist.ProcessMesh([0, 1], dim_names=["x"])
+        >>> dist_attr = dist.DistAttr(mesh=mesh, sharding_specs=[None])
+        >>> # Call the function dtensor_from_fn with dist_attr parameter
+        >>> d_tensor = dist.dtensor_from_fn(paddle.ones, dist_attr=dist_attr, shape=[1])
+        >>> print(d_tensor)
     """
     tensor = fn(*args, **kwargs)
     return shard_tensor(tensor, dist_attr=dist_attr)

From ff9928df2d26b8b048d156d1d231c25ea323381d Mon Sep 17 00:00:00 2001
From: yangxiaoyu14 <yangxiaoyu14@baidu.com>
Date: Fri, 1 Sep 2023 10:30:21 +0000
Subject: [PATCH 09/14] complete the sample code modification according to
 ZhongKai's suggestion

---
 .../paddle/distributed/auto_parallel/api.py   | 19 +++++++++----------
 1 file changed, 9 insertions(+), 10 deletions(-)

diff --git a/python/paddle/distributed/auto_parallel/api.py b/python/paddle/distributed/auto_parallel/api.py
index 0a061fb6b35a81..a8bb5b7ec44f5b 100644
--- a/python/paddle/distributed/auto_parallel/api.py
+++ b/python/paddle/distributed/auto_parallel/api.py
@@ -158,16 +158,15 @@ def dtensor_from_fn(fn, dist_attr, *args, **kwargs):
 
     Examples:
 
-    .. code-block:: python
-
-        >>> import paddle
-        >>> import paddle.distribute as dist
-        >>> # Create a distributed attribute
-        >>> mesh = dist.ProcessMesh([0, 1], dim_names=["x"])
-        >>> dist_attr = dist.DistAttr(mesh=mesh, sharding_specs=[None])
-        >>> # Call the function dtensor_from_fn with dist_attr parameter
-        >>> d_tensor = dist.dtensor_from_fn(paddle.ones, dist_attr=dist_attr, shape=[1])
-        >>> print(d_tensor)
+        .. code-block:: python
+            >>> import paddle
+            >>> import paddle.distributed as dist
+            >>> # Create a distributed attribute
+            >>> mesh = dist.ProcessMesh([0, 1], dim_names=["x"])
+            >>> dist_attr = dist.DistAttr(mesh=mesh, sharding_specs=[None])
+            >>> # Call the function dtensor_from_fn with dist_attr parameter
+            >>> d_tensor = dist.dtensor_from_fn(paddle.ones, dist_attr=dist_attr, shape=[1])
+            >>> print(d_tensor)
     """
     tensor = fn(*args, **kwargs)
     return shard_tensor(tensor, dist_attr=dist_attr)

From a9dd254ca490d6e60ccfb0fbc2378a22eb3bf005 Mon Sep 17 00:00:00 2001
From: yangxiaoyu14 <yangxiaoyu14@baidu.com>
Date: Mon, 4 Sep 2023 11:32:50 +0000
Subject: [PATCH 10/14] modify according to the review

---
 test/auto_parallel/test_dist_tensor.py | 33 ++++++++++++++++----------
 1 file changed, 21 insertions(+), 12 deletions(-)

diff --git a/test/auto_parallel/test_dist_tensor.py b/test/auto_parallel/test_dist_tensor.py
index 40615e56b8c052..87ee854bca9c4e 100644
--- a/test/auto_parallel/test_dist_tensor.py
+++ b/test/auto_parallel/test_dist_tensor.py
@@ -61,38 +61,47 @@ def run_dtensor_from_fn(self):
 
         # Call the function dtensor_from_fn with dist_attr parameter
         result = dist.dtensor_from_fn(
-            paddle.ones, dist_attr=dist_attr, shape=[1]
+            paddle.ones, dist_attr=dist_attr, shape=[16]
         )
-
         # Verify the result
         if paddle.in_dynamic_mode():
+            dist_attr.dynamic_dims = []
             self.assertIsInstance(result, paddle.Tensor)
-            self.assertEqual(result.shape, [1])
+            self.assertEqual(result.shape, [16])
+            self.assertEqual(result.dist_attr, dist_attr)
         else:
+            dist_attr.dynamic_dims = [0]
             self.assertIsInstance(result, paddle.fluid.framework.Variable)
-            self.assertEqual(result.shape, (1,))
+            self.assertEqual(result.shape, (16,))
+            self.assertEqual(result.dist_attr, dist_attr)
 
-        # Test with generate_tensor_zeros()
         result_zeros = dist.dtensor_from_fn(
-            paddle.zeros, dist_attr=dist_attr, shape=[1]
+            paddle.zeros, dist_attr=dist_attr, shape=[16]
         )
         if paddle.in_dynamic_mode():
+            dist_attr.dynamic_dims = []
             self.assertIsInstance(result, paddle.Tensor)
-            self.assertEqual(result.shape, [1])
+            self.assertEqual(result.shape, [16])
+            self.assertEqual(result.dist_attr, dist_attr)
         else:
+            dist_attr.dynamic_dims = [0]
             self.assertIsInstance(result, paddle.fluid.framework.Variable)
-            self.assertEqual(result.shape, (1,))
+            self.assertEqual(result.shape, (16,))
+            self.assertEqual(result.dist_attr, dist_attr)
 
-        # Test with generate_tensor_random()
         result_random = dist.dtensor_from_fn(
-            paddle.rand, dist_attr=dist_attr, shape=[1]
+            paddle.rand, dist_attr=dist_attr, shape=[16]
         )
         if paddle.in_dynamic_mode():
+            dist_attr.dynamic_dims = []
             self.assertIsInstance(result, paddle.Tensor)
-            self.assertEqual(result.shape, [1])
+            self.assertEqual(result.shape, [16])
+            self.assertEqual(result.dist_attr, dist_attr)
         else:
+            dist_attr.dynamic_dims = [0]
             self.assertIsInstance(result, paddle.fluid.framework.Variable)
-            self.assertEqual(result.shape, (1,))
+            self.assertEqual(result.shape, (16,))
+            self.assertEqual(result.dist_attr, dist_attr)
 
         # Test with invalid sharding_specs length
         with self.assertRaises(AssertionError):

From 832d36b8e40fa31898b3a1969df54631d903dc6a Mon Sep 17 00:00:00 2001
From: yangxiaoyu14 <yangxiaoyu14@baidu.com>
Date: Wed, 6 Sep 2023 02:42:30 +0000
Subject: [PATCH 11/14] change fluid.Variable to static.Variable

---
 test/auto_parallel/test_dist_tensor.py | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/test/auto_parallel/test_dist_tensor.py b/test/auto_parallel/test_dist_tensor.py
index 87ee854bca9c4e..669aa1219352a1 100644
--- a/test/auto_parallel/test_dist_tensor.py
+++ b/test/auto_parallel/test_dist_tensor.py
@@ -71,7 +71,7 @@ def run_dtensor_from_fn(self):
             self.assertEqual(result.dist_attr, dist_attr)
         else:
             dist_attr.dynamic_dims = [0]
-            self.assertIsInstance(result, paddle.fluid.framework.Variable)
+            self.assertIsInstance(result, paddle.static.Variable)
             self.assertEqual(result.shape, (16,))
             self.assertEqual(result.dist_attr, dist_attr)
 
@@ -85,7 +85,7 @@ def run_dtensor_from_fn(self):
             self.assertEqual(result.dist_attr, dist_attr)
         else:
             dist_attr.dynamic_dims = [0]
-            self.assertIsInstance(result, paddle.fluid.framework.Variable)
+            self.assertIsInstance(result, paddle.static.Variable)
             self.assertEqual(result.shape, (16,))
             self.assertEqual(result.dist_attr, dist_attr)
 
@@ -99,7 +99,7 @@ def run_dtensor_from_fn(self):
             self.assertEqual(result.dist_attr, dist_attr)
         else:
             dist_attr.dynamic_dims = [0]
-            self.assertIsInstance(result, paddle.fluid.framework.Variable)
+            self.assertIsInstance(result, paddle.static.Variable)
             self.assertEqual(result.shape, (16,))
             self.assertEqual(result.dist_attr, dist_attr)
 

From bf0598bbaedea99f5bdcf44cc9aaff041b2aecd6 Mon Sep 17 00:00:00 2001
From: yangxiaoyu14 <yangxiaoyu14@baidu.com>
Date: Thu, 7 Sep 2023 07:03:26 +0000
Subject: [PATCH 12/14] modify according to zhongkai's review

---
 .../ir/generated/pd_ops.parsed.yaml           | 487 ++++++++++++++++++
 .../ir/generated/pd_ops_backward.parsed.yaml  |  39 ++
 .../paddle/distributed/auto_parallel/api.py   |   1 +
 3 files changed, 527 insertions(+)
 create mode 100644 paddle/fluid/ir/dialect/paddle_dialect/ir/generated/pd_ops.parsed.yaml
 create mode 100644 paddle/fluid/ir/dialect/paddle_dialect/ir/generated/pd_ops_backward.parsed.yaml

diff --git a/paddle/fluid/ir/dialect/paddle_dialect/ir/generated/pd_ops.parsed.yaml b/paddle/fluid/ir/dialect/paddle_dialect/ir/generated/pd_ops.parsed.yaml
new file mode 100644
index 00000000000000..b95740f24c1715
--- /dev/null
+++ b/paddle/fluid/ir/dialect/paddle_dialect/ir/generated/pd_ops.parsed.yaml
@@ -0,0 +1,487 @@
+- name: add_n_
+  inputs:
+  - typename: Tensor[]
+    name: inputs
+    optional: false
+    no_need_buffer: false
+    data_transform: {}
+  attrs: []
+  outputs:
+  - {typename: Tensor, name: out, optional: false, intermediate: false}
+  no_need_buffer: null
+  data_transform: null
+  kernel:
+    func: [add_n]
+    param: [inputs]
+    backend: null
+    layout: null
+    data_type: null
+    dispatch: {add_n: null}
+    force_backend: null
+  infer_meta:
+    func: AddNInferMeta
+    param: [inputs]
+  inplace: null
+  view: null
+  backward: add_n_grad
+- name: add_n_with_kernel
+  inputs:
+  - typename: Tensor[]
+    name: inputs
+    optional: false
+    no_need_buffer: false
+    data_transform: {}
+  attrs: []
+  outputs:
+  - {typename: Tensor, name: out, optional: false, intermediate: false}
+  no_need_buffer: null
+  data_transform: null
+  kernel:
+    func: [add_n]
+    param: [inputs]
+    backend: null
+    layout: null
+    data_type: null
+    dispatch: {add_n: null}
+    force_backend: null
+  infer_meta:
+    func: AddNInferMeta
+    param: [inputs]
+  inplace: null
+  view: null
+  backward: add_n_grad
+- name: assert
+  inputs:
+  - typename: Tensor
+    name: cond
+    optional: false
+    no_need_buffer: false
+    data_transform: {}
+  - typename: Tensor[]
+    name: data
+    optional: false
+    no_need_buffer: false
+    data_transform: {}
+  attrs:
+  - {typename: int64_t, name: summarize, default_value: '-1'}
+  outputs: []
+  no_need_buffer: null
+  data_transform: null
+  kernel:
+    func: [assert]
+    param: [cond, data, summarize]
+    backend: null
+    layout: null
+    data_type:
+      ordered: false
+      candidates: [cond]
+      to_complex_flag: [false]
+    dispatch: {assert: null}
+    force_backend: null
+  inplace: null
+  view: null
+  backward: null
+- name: assign_value
+  inputs: []
+  attrs:
+  - {typename: 'int[]', name: shape}
+  - {typename: DataType, name: dtype}
+  - {typename: 'Scalar[]', name: values, data_type: 'std::vector<Scalar>'}
+  - {typename: Place, name: place, default_value: '{}'}
+  outputs:
+  - {typename: Tensor, name: out, optional: false, intermediate: false}
+  no_need_buffer: null
+  data_transform: null
+  kernel:
+    func: [assign_value]
+    param: [shape, dtype, values]
+    backend:
+      ordered: true
+      candidates: [place]
+    layout: null
+    data_type:
+      ordered: false
+      candidates: [dtype]
+      to_complex_flag: [false]
+    dispatch: {assign_value: null}
+    force_backend: null
+  infer_meta:
+    func: AssignValueInferMeta
+    param: [shape, dtype]
+  inplace: null
+  view: null
+  backward: null
+- name: embedding_grad_sparse
+  inputs:
+  - typename: Tensor
+    name: x
+    optional: false
+    no_need_buffer: false
+    data_transform: {}
+  - typename: Tensor
+    name: weight
+    optional: false
+    no_need_buffer: false
+    data_transform: {}
+  - typename: Tensor
+    name: out_grad
+    optional: false
+    no_need_buffer: false
+    data_transform: {}
+  attrs:
+  - {typename: int64_t, name: padding_idx, default_value: '-1'}
+  - {typename: bool, name: sparse, default_value: 'false'}
+  outputs:
+  - {typename: SelectedRows, name: weight_grad, optional: false, intermediate: false}
+  no_need_buffer: null
+  data_transform: null
+  kernel:
+    func: [embedding_sparse_grad]
+    param: [x, weight, out_grad, padding_idx, sparse]
+    backend: null
+    layout: null
+    data_type:
+      ordered: false
+      candidates: [weight]
+      to_complex_flag: [false]
+    dispatch: {embedding_sparse_grad: null}
+    force_backend: null
+  infer_meta:
+    func: EmbeddingGradSparseInferMeta
+    param: [weight]
+  inplace: null
+  view: null
+  backward: null
+- name: feed
+  inputs: []
+  attrs:
+  - {typename: str, name: name}
+  - {typename: int, name: col}
+  outputs:
+  - {typename: Tensor, name: out, optional: false, intermediate: false}
+  no_need_buffer: null
+  data_transform: null
+  inplace: null
+  view: null
+  backward: null
+- name: fetch
+  inputs:
+  - typename: Tensor
+    name: x
+    optional: false
+    no_need_buffer: false
+    data_transform: {}
+  attrs:
+  - {typename: str, name: name}
+  - {typename: int, name: col}
+  outputs:
+  - {typename: Tensor, name: out, optional: false, intermediate: false}
+  no_need_buffer: null
+  data_transform: null
+  kernel:
+    func: [fetch]
+    param: [x]
+    backend: null
+    layout: null
+    data_type: null
+    dispatch: {fetch: null}
+    force_backend: null
+  infer_meta:
+    func: UnchangedInferMeta
+    param: [x]
+  inplace: null
+  view: null
+  backward: null
+- name: load_combine
+  inputs: []
+  attrs:
+  - {typename: str, name: file_path}
+  - {typename: bool, name: load_as_fp16}
+  - {typename: bool, name: model_from_memory}
+  outputs:
+  - {typename: 'Tensor[]', name: Out, optional: true, intermediate: false}
+  no_need_buffer: null
+  data_transform: null
+  kernel:
+    func: [load_combine]
+    param: [file_path, load_as_fp16, model_from_memory]
+    backend: null
+    layout: null
+    data_type: null
+    dispatch: {load_combine: null}
+    force_backend: null
+  inplace: null
+  view: null
+  backward: null
+- name: lod_array_length
+  inputs:
+  - typename: Tensor[]
+    name: x
+    optional: false
+    no_need_buffer: false
+    data_transform: {}
+  attrs: []
+  outputs:
+  - {typename: Tensor, name: out, optional: false, intermediate: false}
+  no_need_buffer: null
+  data_transform: null
+  inplace: null
+  view: null
+  backward: null
+- name: print
+  inputs:
+  - typename: Tensor
+    name: in
+    optional: false
+    no_need_buffer: false
+    data_transform: {}
+  attrs:
+  - {typename: int, name: first_n}
+  - {typename: str, name: message}
+  - {typename: int, name: summarize}
+  - {typename: bool, name: print_tensor_name, default_value: 'true'}
+  - {typename: bool, name: print_tensor_type, default_value: 'true'}
+  - {typename: bool, name: print_tensor_shape, default_value: 'true'}
+  - {typename: bool, name: print_tensor_layout, default_value: 'true'}
+  - {typename: bool, name: print_tensor_lod, default_value: 'true'}
+  - {typename: str, name: print_phase, default_value: '"BOTH"'}
+  - {typename: bool, name: is_forward, default_value: 'true'}
+  outputs:
+  - {typename: Tensor, name: out, optional: false, intermediate: false}
+  no_need_buffer: null
+  data_transform: null
+  kernel:
+    func: [print_kernel]
+    param: [in, first_n, message, summarize, print_tensor_name, print_tensor_type,
+      print_tensor_shape, print_tensor_layout, print_tensor_lod, print_phase, is_forward]
+    backend: null
+    layout: null
+    data_type: null
+    dispatch: {print_kernel: null}
+    force_backend: null
+  infer_meta:
+    func: UnchangedInferMeta
+    param: [in]
+  inplace: null
+  view: null
+  backward: null
+- name: recv_v2
+  inputs: []
+  attrs:
+  - {typename: 'int[]', name: out_shape, default_value: '{}'}
+  - {typename: DataType, name: dtype, default_value: 'DataType::FLOAT32'}
+  - {typename: int, name: peer, default_value: '0'}
+  - {typename: int, name: ring_id, default_value: '0'}
+  - {typename: bool, name: use_calc_stream, default_value: 'false'}
+  - {typename: bool, name: dynamic_shape, default_value: 'false'}
+  outputs:
+  - {typename: Tensor, name: out, optional: false, intermediate: false}
+  no_need_buffer: null
+  data_transform: null
+  kernel:
+    func: [recv_v2]
+    param: [ring_id, dynamic_shape, peer, out_shape, dtype, use_calc_stream]
+    backend: null
+    layout: null
+    data_type:
+      ordered: false
+      candidates: [dtype]
+      to_complex_flag: [false]
+    dispatch: {recv_v2: null}
+    force_backend: null
+  infer_meta:
+    func: RecvV2InferMeta
+    param: [peer, dtype, out_shape]
+  inplace: null
+  view: null
+  backward: null
+- name: save_combine
+  inputs:
+  - typename: Tensor[]
+    name: x
+    optional: false
+    no_need_buffer: false
+    data_transform: {}
+  attrs:
+  - {typename: str, name: file_path}
+  - {typename: bool, name: overwrite}
+  - {typename: bool, name: save_as_fp16}
+  - {typename: bool, name: save_to_memory}
+  outputs:
+  - {typename: Tensor, name: out, optional: true, intermediate: false}
+  no_need_buffer: null
+  data_transform: null
+  kernel:
+    func: [save_combine_tensor]
+    param: [x, file_path, overwrite, save_as_fp16, save_to_memory]
+    backend: null
+    layout: null
+    data_type: null
+    dispatch: {save_combine_tensor: null}
+    force_backend: null
+  inplace: null
+  view: null
+  backward: null
+- name: send_v2
+  inputs:
+  - typename: Tensor
+    name: x
+    optional: false
+    no_need_buffer: false
+    data_transform: {}
+  attrs:
+  - {typename: int, name: ring_id, default_value: '0'}
+  - {typename: int, name: peer, default_value: '0'}
+  - {typename: bool, name: use_calc_stream, default_value: 'false'}
+  - {typename: bool, name: dynamic_shape, default_value: 'false'}
+  outputs: []
+  no_need_buffer: null
+  data_transform: null
+  kernel:
+    func: [send_v2]
+    param: [x, ring_id, dynamic_shape, peer, use_calc_stream]
+    backend: null
+    layout: null
+    data_type: null
+    dispatch: {send_v2: null}
+    force_backend: null
+  infer_meta:
+    func: SendV2InferMeta
+    param: [peer, ring_id]
+  inplace: null
+  view: null
+  backward: null
+- name: set_value
+  inputs:
+  - typename: Tensor
+    name: x
+    optional: false
+    no_need_buffer: false
+    data_transform: {}
+  attrs:
+  - {typename: 'int64_t[]', name: starts}
+  - {typename: 'int64_t[]', name: ends}
+  - {typename: 'int64_t[]', name: steps}
+  - {typename: 'int64_t[]', name: axes}
+  - {typename: 'int64_t[]', name: decrease_axes}
+  - {typename: 'int64_t[]', name: none_axes}
+  - {typename: 'int64_t[]', name: shape}
+  - {typename: 'Scalar[]', name: values, data_type: 'std::vector<Scalar>'}
+  outputs:
+  - {typename: Tensor, name: out, optional: false, intermediate: false}
+  no_need_buffer: null
+  data_transform: null
+  kernel:
+    func: [set_value]
+    param: [x, starts, ends, steps, axes, decrease_axes, none_axes, shape, values]
+    backend: null
+    layout: null
+    data_type: null
+    dispatch: {set_value: null}
+    force_backend: null
+  infer_meta:
+    func: SetValueInferMeta
+    param: [x]
+  inplace: {out: x}
+  view: null
+  backward: set_value_grad
+- name: set_value_with_tensor
+  inputs:
+  - typename: Tensor
+    name: x
+    optional: false
+    no_need_buffer: false
+    data_transform: {}
+  - typename: Tensor
+    name: values
+    optional: false
+    no_need_buffer: false
+    data_transform: {}
+  attrs:
+  - {typename: 'int64_t[]', name: starts}
+  - {typename: 'int64_t[]', name: ends}
+  - {typename: 'int64_t[]', name: steps}
+  - {typename: 'int64_t[]', name: axes}
+  - {typename: 'int64_t[]', name: decrease_axes}
+  - {typename: 'int64_t[]', name: none_axes}
+  outputs:
+  - {typename: Tensor, name: out, optional: false, intermediate: false}
+  no_need_buffer: null
+  data_transform: null
+  kernel:
+    func: [set_value_with_tensor]
+    param: [x, values, starts, ends, steps, axes, decrease_axes, none_axes]
+    backend: null
+    layout: null
+    data_type: null
+    dispatch: {set_value_with_tensor: null}
+    force_backend: null
+  infer_meta:
+    func: SetValueInferMeta
+    param: [x]
+  inplace: {out: x}
+  view: null
+  backward: set_value_grad
+- name: shadow_feed
+  inputs:
+  - typename: Tensor
+    name: x
+    optional: false
+    no_need_buffer: false
+    data_transform: {}
+  attrs: []
+  outputs:
+  - {typename: Tensor, name: out, optional: false, intermediate: false}
+  no_need_buffer: null
+  data_transform: null
+  kernel:
+    func: [shadow_feed]
+    param: [x]
+    backend: null
+    layout: null
+    data_type: null
+    dispatch: {shadow_feed: null}
+    force_backend: null
+  infer_meta:
+    func: UnchangedInferMeta
+    param: [x]
+  inplace: null
+  view: null
+  backward: null
+- name: share_buffer_
+  inputs:
+  - typename: Tensor[]
+    name: x
+    optional: false
+    no_need_buffer: false
+    data_transform: {}
+  attrs:
+  - {typename: 'bool[]', name: share_dims_and_dtype, default_value: '{}'}
+  outputs:
+  - {typename: 'Tensor[]', name: out, size: x.size(), optional: false, intermediate: false}
+  - {typename: 'Tensor[]', name: xout, size: x.size(), optional: false, intermediate: false}
+  no_need_buffer: null
+  data_transform: null
+  inplace: null
+  view: null
+  backward: null
+- name: write_to_array
+  inputs:
+  - typename: Tensor
+    name: i
+    optional: false
+    no_need_buffer: false
+    data_transform: {}
+  - typename: Tensor
+    name: x
+    optional: false
+    no_need_buffer: false
+    data_transform: {}
+  attrs: []
+  outputs:
+  - {typename: 'Tensor[]', name: out, optional: false, intermediate: false}
+  no_need_buffer: null
+  data_transform: null
+  inplace: null
+  view: null
+  backward: write_to_array_grad
diff --git a/paddle/fluid/ir/dialect/paddle_dialect/ir/generated/pd_ops_backward.parsed.yaml b/paddle/fluid/ir/dialect/paddle_dialect/ir/generated/pd_ops_backward.parsed.yaml
new file mode 100644
index 00000000000000..fe24d850a534d0
--- /dev/null
+++ b/paddle/fluid/ir/dialect/paddle_dialect/ir/generated/pd_ops_backward.parsed.yaml
@@ -0,0 +1,39 @@
+- name: set_value_grad
+  inputs:
+  - typename: Tensor
+    name: out_grad
+    optional: false
+    no_need_buffer: false
+    data_transform: {}
+  - typename: Tensor
+    name: values
+    optional: false
+    no_need_buffer: false
+    data_transform: {}
+  attrs:
+  - {typename: 'int64_t[]', name: starts}
+  - {typename: 'int64_t[]', name: ends}
+  - {typename: 'int64_t[]', name: steps}
+  - {typename: 'int64_t[]', name: axes}
+  - {typename: 'int64_t[]', name: decrease_axes}
+  - {typename: 'int64_t[]', name: none_axes}
+  outputs:
+  - {typename: Tensor, name: x_grad, optional: false, intermediate: false}
+  - {typename: Tensor, name: values_grad, optional: false, intermediate: false}
+  no_need_buffer: null
+  data_transform: null
+  kernel:
+    func: [set_value_grad]
+    param: [out_grad, starts, ends, steps, axes, decrease_axes, none_axes]
+    backend: null
+    layout: null
+    data_type: null
+    dispatch: {set_value_grad: null}
+    force_backend: null
+  infer_meta:
+    func: SetValueGradInferMeta
+    param: [out_grad, values]
+  inplace: null
+  view: null
+  backward: null
+  forward: null
diff --git a/python/paddle/distributed/auto_parallel/api.py b/python/paddle/distributed/auto_parallel/api.py
index a8bb5b7ec44f5b..680b9cc95bc2b0 100644
--- a/python/paddle/distributed/auto_parallel/api.py
+++ b/python/paddle/distributed/auto_parallel/api.py
@@ -159,6 +159,7 @@ def dtensor_from_fn(fn, dist_attr, *args, **kwargs):
     Examples:
 
         .. code-block:: python
+
             >>> import paddle
             >>> import paddle.distributed as dist
             >>> # Create a distributed attribute

From d545aed26d73d0e818133fdc76b4a86137ac3249 Mon Sep 17 00:00:00 2001
From: yangxiaoyu14 <yangxiaoyu14@baidu.com>
Date: Mon, 11 Sep 2023 03:10:27 +0000
Subject: [PATCH 13/14] According to Yifan's suggestion, pull the latest code
 to resolve conflicts

---
 test/auto_parallel/test_dist_tensor.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/test/auto_parallel/test_dist_tensor.py b/test/auto_parallel/test_dist_tensor.py
index 17d3b7a46be1e0..0e4ab8e57d9006 100644
--- a/test/auto_parallel/test_dist_tensor.py
+++ b/test/auto_parallel/test_dist_tensor.py
@@ -55,7 +55,7 @@ def test_dist_tensor_creation(self):
 
 class TestDistTensorFromFn(unittest.TestCase):
     def run_dtensor_from_fn(self):
-        # Create a distributed attribute
+        # Create a dist_attr
         mesh = dist.ProcessMesh([0, 1], dim_names=["x"])
         dist_attr = dist.DistAttr(mesh=mesh, sharding_specs=[None])
 

From f064d24226d7e28c1c79cf18727eaef8c80f7e7d Mon Sep 17 00:00:00 2001
From: yangxiaoyu14 <yangxiaoyu14@baidu.com>
Date: Mon, 11 Sep 2023 08:41:08 +0000
Subject: [PATCH 14/14] remove
 paddle/fluid/ir/dialect/paddle_dialect/ir/generated/pd_ops_backward.parsed.yaml
 and paddle/fluid/ir/dialect/paddle_dialect/ir/generated/pd_ops.parsed.yaml

---
 .../ir/generated/pd_ops.parsed.yaml           | 487 ------------------
 .../ir/generated/pd_ops_backward.parsed.yaml  |  39 --
 2 files changed, 526 deletions(-)
 delete mode 100644 paddle/fluid/ir/dialect/paddle_dialect/ir/generated/pd_ops.parsed.yaml
 delete mode 100644 paddle/fluid/ir/dialect/paddle_dialect/ir/generated/pd_ops_backward.parsed.yaml

diff --git a/paddle/fluid/ir/dialect/paddle_dialect/ir/generated/pd_ops.parsed.yaml b/paddle/fluid/ir/dialect/paddle_dialect/ir/generated/pd_ops.parsed.yaml
deleted file mode 100644
index b95740f24c1715..00000000000000
--- a/paddle/fluid/ir/dialect/paddle_dialect/ir/generated/pd_ops.parsed.yaml
+++ /dev/null
@@ -1,487 +0,0 @@
-- name: add_n_
-  inputs:
-  - typename: Tensor[]
-    name: inputs
-    optional: false
-    no_need_buffer: false
-    data_transform: {}
-  attrs: []
-  outputs:
-  - {typename: Tensor, name: out, optional: false, intermediate: false}
-  no_need_buffer: null
-  data_transform: null
-  kernel:
-    func: [add_n]
-    param: [inputs]
-    backend: null
-    layout: null
-    data_type: null
-    dispatch: {add_n: null}
-    force_backend: null
-  infer_meta:
-    func: AddNInferMeta
-    param: [inputs]
-  inplace: null
-  view: null
-  backward: add_n_grad
-- name: add_n_with_kernel
-  inputs:
-  - typename: Tensor[]
-    name: inputs
-    optional: false
-    no_need_buffer: false
-    data_transform: {}
-  attrs: []
-  outputs:
-  - {typename: Tensor, name: out, optional: false, intermediate: false}
-  no_need_buffer: null
-  data_transform: null
-  kernel:
-    func: [add_n]
-    param: [inputs]
-    backend: null
-    layout: null
-    data_type: null
-    dispatch: {add_n: null}
-    force_backend: null
-  infer_meta:
-    func: AddNInferMeta
-    param: [inputs]
-  inplace: null
-  view: null
-  backward: add_n_grad
-- name: assert
-  inputs:
-  - typename: Tensor
-    name: cond
-    optional: false
-    no_need_buffer: false
-    data_transform: {}
-  - typename: Tensor[]
-    name: data
-    optional: false
-    no_need_buffer: false
-    data_transform: {}
-  attrs:
-  - {typename: int64_t, name: summarize, default_value: '-1'}
-  outputs: []
-  no_need_buffer: null
-  data_transform: null
-  kernel:
-    func: [assert]
-    param: [cond, data, summarize]
-    backend: null
-    layout: null
-    data_type:
-      ordered: false
-      candidates: [cond]
-      to_complex_flag: [false]
-    dispatch: {assert: null}
-    force_backend: null
-  inplace: null
-  view: null
-  backward: null
-- name: assign_value
-  inputs: []
-  attrs:
-  - {typename: 'int[]', name: shape}
-  - {typename: DataType, name: dtype}
-  - {typename: 'Scalar[]', name: values, data_type: 'std::vector<Scalar>'}
-  - {typename: Place, name: place, default_value: '{}'}
-  outputs:
-  - {typename: Tensor, name: out, optional: false, intermediate: false}
-  no_need_buffer: null
-  data_transform: null
-  kernel:
-    func: [assign_value]
-    param: [shape, dtype, values]
-    backend:
-      ordered: true
-      candidates: [place]
-    layout: null
-    data_type:
-      ordered: false
-      candidates: [dtype]
-      to_complex_flag: [false]
-    dispatch: {assign_value: null}
-    force_backend: null
-  infer_meta:
-    func: AssignValueInferMeta
-    param: [shape, dtype]
-  inplace: null
-  view: null
-  backward: null
-- name: embedding_grad_sparse
-  inputs:
-  - typename: Tensor
-    name: x
-    optional: false
-    no_need_buffer: false
-    data_transform: {}
-  - typename: Tensor
-    name: weight
-    optional: false
-    no_need_buffer: false
-    data_transform: {}
-  - typename: Tensor
-    name: out_grad
-    optional: false
-    no_need_buffer: false
-    data_transform: {}
-  attrs:
-  - {typename: int64_t, name: padding_idx, default_value: '-1'}
-  - {typename: bool, name: sparse, default_value: 'false'}
-  outputs:
-  - {typename: SelectedRows, name: weight_grad, optional: false, intermediate: false}
-  no_need_buffer: null
-  data_transform: null
-  kernel:
-    func: [embedding_sparse_grad]
-    param: [x, weight, out_grad, padding_idx, sparse]
-    backend: null
-    layout: null
-    data_type:
-      ordered: false
-      candidates: [weight]
-      to_complex_flag: [false]
-    dispatch: {embedding_sparse_grad: null}
-    force_backend: null
-  infer_meta:
-    func: EmbeddingGradSparseInferMeta
-    param: [weight]
-  inplace: null
-  view: null
-  backward: null
-- name: feed
-  inputs: []
-  attrs:
-  - {typename: str, name: name}
-  - {typename: int, name: col}
-  outputs:
-  - {typename: Tensor, name: out, optional: false, intermediate: false}
-  no_need_buffer: null
-  data_transform: null
-  inplace: null
-  view: null
-  backward: null
-- name: fetch
-  inputs:
-  - typename: Tensor
-    name: x
-    optional: false
-    no_need_buffer: false
-    data_transform: {}
-  attrs:
-  - {typename: str, name: name}
-  - {typename: int, name: col}
-  outputs:
-  - {typename: Tensor, name: out, optional: false, intermediate: false}
-  no_need_buffer: null
-  data_transform: null
-  kernel:
-    func: [fetch]
-    param: [x]
-    backend: null
-    layout: null
-    data_type: null
-    dispatch: {fetch: null}
-    force_backend: null
-  infer_meta:
-    func: UnchangedInferMeta
-    param: [x]
-  inplace: null
-  view: null
-  backward: null
-- name: load_combine
-  inputs: []
-  attrs:
-  - {typename: str, name: file_path}
-  - {typename: bool, name: load_as_fp16}
-  - {typename: bool, name: model_from_memory}
-  outputs:
-  - {typename: 'Tensor[]', name: Out, optional: true, intermediate: false}
-  no_need_buffer: null
-  data_transform: null
-  kernel:
-    func: [load_combine]
-    param: [file_path, load_as_fp16, model_from_memory]
-    backend: null
-    layout: null
-    data_type: null
-    dispatch: {load_combine: null}
-    force_backend: null
-  inplace: null
-  view: null
-  backward: null
-- name: lod_array_length
-  inputs:
-  - typename: Tensor[]
-    name: x
-    optional: false
-    no_need_buffer: false
-    data_transform: {}
-  attrs: []
-  outputs:
-  - {typename: Tensor, name: out, optional: false, intermediate: false}
-  no_need_buffer: null
-  data_transform: null
-  inplace: null
-  view: null
-  backward: null
-- name: print
-  inputs:
-  - typename: Tensor
-    name: in
-    optional: false
-    no_need_buffer: false
-    data_transform: {}
-  attrs:
-  - {typename: int, name: first_n}
-  - {typename: str, name: message}
-  - {typename: int, name: summarize}
-  - {typename: bool, name: print_tensor_name, default_value: 'true'}
-  - {typename: bool, name: print_tensor_type, default_value: 'true'}
-  - {typename: bool, name: print_tensor_shape, default_value: 'true'}
-  - {typename: bool, name: print_tensor_layout, default_value: 'true'}
-  - {typename: bool, name: print_tensor_lod, default_value: 'true'}
-  - {typename: str, name: print_phase, default_value: '"BOTH"'}
-  - {typename: bool, name: is_forward, default_value: 'true'}
-  outputs:
-  - {typename: Tensor, name: out, optional: false, intermediate: false}
-  no_need_buffer: null
-  data_transform: null
-  kernel:
-    func: [print_kernel]
-    param: [in, first_n, message, summarize, print_tensor_name, print_tensor_type,
-      print_tensor_shape, print_tensor_layout, print_tensor_lod, print_phase, is_forward]
-    backend: null
-    layout: null
-    data_type: null
-    dispatch: {print_kernel: null}
-    force_backend: null
-  infer_meta:
-    func: UnchangedInferMeta
-    param: [in]
-  inplace: null
-  view: null
-  backward: null
-- name: recv_v2
-  inputs: []
-  attrs:
-  - {typename: 'int[]', name: out_shape, default_value: '{}'}
-  - {typename: DataType, name: dtype, default_value: 'DataType::FLOAT32'}
-  - {typename: int, name: peer, default_value: '0'}
-  - {typename: int, name: ring_id, default_value: '0'}
-  - {typename: bool, name: use_calc_stream, default_value: 'false'}
-  - {typename: bool, name: dynamic_shape, default_value: 'false'}
-  outputs:
-  - {typename: Tensor, name: out, optional: false, intermediate: false}
-  no_need_buffer: null
-  data_transform: null
-  kernel:
-    func: [recv_v2]
-    param: [ring_id, dynamic_shape, peer, out_shape, dtype, use_calc_stream]
-    backend: null
-    layout: null
-    data_type:
-      ordered: false
-      candidates: [dtype]
-      to_complex_flag: [false]
-    dispatch: {recv_v2: null}
-    force_backend: null
-  infer_meta:
-    func: RecvV2InferMeta
-    param: [peer, dtype, out_shape]
-  inplace: null
-  view: null
-  backward: null
-- name: save_combine
-  inputs:
-  - typename: Tensor[]
-    name: x
-    optional: false
-    no_need_buffer: false
-    data_transform: {}
-  attrs:
-  - {typename: str, name: file_path}
-  - {typename: bool, name: overwrite}
-  - {typename: bool, name: save_as_fp16}
-  - {typename: bool, name: save_to_memory}
-  outputs:
-  - {typename: Tensor, name: out, optional: true, intermediate: false}
-  no_need_buffer: null
-  data_transform: null
-  kernel:
-    func: [save_combine_tensor]
-    param: [x, file_path, overwrite, save_as_fp16, save_to_memory]
-    backend: null
-    layout: null
-    data_type: null
-    dispatch: {save_combine_tensor: null}
-    force_backend: null
-  inplace: null
-  view: null
-  backward: null
-- name: send_v2
-  inputs:
-  - typename: Tensor
-    name: x
-    optional: false
-    no_need_buffer: false
-    data_transform: {}
-  attrs:
-  - {typename: int, name: ring_id, default_value: '0'}
-  - {typename: int, name: peer, default_value: '0'}
-  - {typename: bool, name: use_calc_stream, default_value: 'false'}
-  - {typename: bool, name: dynamic_shape, default_value: 'false'}
-  outputs: []
-  no_need_buffer: null
-  data_transform: null
-  kernel:
-    func: [send_v2]
-    param: [x, ring_id, dynamic_shape, peer, use_calc_stream]
-    backend: null
-    layout: null
-    data_type: null
-    dispatch: {send_v2: null}
-    force_backend: null
-  infer_meta:
-    func: SendV2InferMeta
-    param: [peer, ring_id]
-  inplace: null
-  view: null
-  backward: null
-- name: set_value
-  inputs:
-  - typename: Tensor
-    name: x
-    optional: false
-    no_need_buffer: false
-    data_transform: {}
-  attrs:
-  - {typename: 'int64_t[]', name: starts}
-  - {typename: 'int64_t[]', name: ends}
-  - {typename: 'int64_t[]', name: steps}
-  - {typename: 'int64_t[]', name: axes}
-  - {typename: 'int64_t[]', name: decrease_axes}
-  - {typename: 'int64_t[]', name: none_axes}
-  - {typename: 'int64_t[]', name: shape}
-  - {typename: 'Scalar[]', name: values, data_type: 'std::vector<Scalar>'}
-  outputs:
-  - {typename: Tensor, name: out, optional: false, intermediate: false}
-  no_need_buffer: null
-  data_transform: null
-  kernel:
-    func: [set_value]
-    param: [x, starts, ends, steps, axes, decrease_axes, none_axes, shape, values]
-    backend: null
-    layout: null
-    data_type: null
-    dispatch: {set_value: null}
-    force_backend: null
-  infer_meta:
-    func: SetValueInferMeta
-    param: [x]
-  inplace: {out: x}
-  view: null
-  backward: set_value_grad
-- name: set_value_with_tensor
-  inputs:
-  - typename: Tensor
-    name: x
-    optional: false
-    no_need_buffer: false
-    data_transform: {}
-  - typename: Tensor
-    name: values
-    optional: false
-    no_need_buffer: false
-    data_transform: {}
-  attrs:
-  - {typename: 'int64_t[]', name: starts}
-  - {typename: 'int64_t[]', name: ends}
-  - {typename: 'int64_t[]', name: steps}
-  - {typename: 'int64_t[]', name: axes}
-  - {typename: 'int64_t[]', name: decrease_axes}
-  - {typename: 'int64_t[]', name: none_axes}
-  outputs:
-  - {typename: Tensor, name: out, optional: false, intermediate: false}
-  no_need_buffer: null
-  data_transform: null
-  kernel:
-    func: [set_value_with_tensor]
-    param: [x, values, starts, ends, steps, axes, decrease_axes, none_axes]
-    backend: null
-    layout: null
-    data_type: null
-    dispatch: {set_value_with_tensor: null}
-    force_backend: null
-  infer_meta:
-    func: SetValueInferMeta
-    param: [x]
-  inplace: {out: x}
-  view: null
-  backward: set_value_grad
-- name: shadow_feed
-  inputs:
-  - typename: Tensor
-    name: x
-    optional: false
-    no_need_buffer: false
-    data_transform: {}
-  attrs: []
-  outputs:
-  - {typename: Tensor, name: out, optional: false, intermediate: false}
-  no_need_buffer: null
-  data_transform: null
-  kernel:
-    func: [shadow_feed]
-    param: [x]
-    backend: null
-    layout: null
-    data_type: null
-    dispatch: {shadow_feed: null}
-    force_backend: null
-  infer_meta:
-    func: UnchangedInferMeta
-    param: [x]
-  inplace: null
-  view: null
-  backward: null
-- name: share_buffer_
-  inputs:
-  - typename: Tensor[]
-    name: x
-    optional: false
-    no_need_buffer: false
-    data_transform: {}
-  attrs:
-  - {typename: 'bool[]', name: share_dims_and_dtype, default_value: '{}'}
-  outputs:
-  - {typename: 'Tensor[]', name: out, size: x.size(), optional: false, intermediate: false}
-  - {typename: 'Tensor[]', name: xout, size: x.size(), optional: false, intermediate: false}
-  no_need_buffer: null
-  data_transform: null
-  inplace: null
-  view: null
-  backward: null
-- name: write_to_array
-  inputs:
-  - typename: Tensor
-    name: i
-    optional: false
-    no_need_buffer: false
-    data_transform: {}
-  - typename: Tensor
-    name: x
-    optional: false
-    no_need_buffer: false
-    data_transform: {}
-  attrs: []
-  outputs:
-  - {typename: 'Tensor[]', name: out, optional: false, intermediate: false}
-  no_need_buffer: null
-  data_transform: null
-  inplace: null
-  view: null
-  backward: write_to_array_grad
diff --git a/paddle/fluid/ir/dialect/paddle_dialect/ir/generated/pd_ops_backward.parsed.yaml b/paddle/fluid/ir/dialect/paddle_dialect/ir/generated/pd_ops_backward.parsed.yaml
deleted file mode 100644
index fe24d850a534d0..00000000000000
--- a/paddle/fluid/ir/dialect/paddle_dialect/ir/generated/pd_ops_backward.parsed.yaml
+++ /dev/null
@@ -1,39 +0,0 @@
-- name: set_value_grad
-  inputs:
-  - typename: Tensor
-    name: out_grad
-    optional: false
-    no_need_buffer: false
-    data_transform: {}
-  - typename: Tensor
-    name: values
-    optional: false
-    no_need_buffer: false
-    data_transform: {}
-  attrs:
-  - {typename: 'int64_t[]', name: starts}
-  - {typename: 'int64_t[]', name: ends}
-  - {typename: 'int64_t[]', name: steps}
-  - {typename: 'int64_t[]', name: axes}
-  - {typename: 'int64_t[]', name: decrease_axes}
-  - {typename: 'int64_t[]', name: none_axes}
-  outputs:
-  - {typename: Tensor, name: x_grad, optional: false, intermediate: false}
-  - {typename: Tensor, name: values_grad, optional: false, intermediate: false}
-  no_need_buffer: null
-  data_transform: null
-  kernel:
-    func: [set_value_grad]
-    param: [out_grad, starts, ends, steps, axes, decrease_axes, none_axes]
-    backend: null
-    layout: null
-    data_type: null
-    dispatch: {set_value_grad: null}
-    force_backend: null
-  infer_meta:
-    func: SetValueGradInferMeta
-    param: [out_grad, values]
-  inplace: null
-  view: null
-  backward: null
-  forward: null