[Zero-Dim] Support output 0D for to_tensor. (#52741)

* test=allcase * test=allcase * test=allcase * test=allcase * test=allcase * test=allcase * test=allcase * test=allcase * test=allcase * test=allcase * test=allcase * test=allcase * test=allcase * test=allcase * test=allcase * test=allcase * test=allcase * test=allcase * test=allcase * test=allcase * test=allcase * test=allcase * test=allcase * test=allcase * test=allcase * test=allcase * fix doc erros, test=allcase
PaddlePaddle · Apr 24, 2023 · 81fb7df · 81fb7df
1 parent 83c2e68
commit 81fb7df
Show file tree

Hide file tree

Showing 29 changed files with 169 additions and 129 deletions.
diff --git a/paddle/fluid/pybind/imperative.cc b/paddle/fluid/pybind/imperative.cc
@@ -1321,7 +1321,7 @@ void BindImperative(py::module *m_ptr) {
 
                 import paddle
 
-                x = paddle.to_tensor(1.0, stop_gradient=False)
+                x = paddle.to_tensor([1.0], stop_gradient=False)
                 detach_x = x.detach()
                 detach_x[:] = 10.0
                 print(x)  # Tensor(shape=[1], dtype=float32, place=CPUPlace, stop_gradient=False,

diff --git a/python/paddle/audio/functional/window.py b/python/paddle/audio/functional/window.py
@@ -13,6 +13,8 @@
 import math
 from typing import List, Tuple, Union
 
+import numpy as np
+
 import paddle
 from paddle import Tensor
 
@@ -38,7 +40,12 @@ def get(self, name):
 
 @window_function_register.register()
 def _cat(x: List[Tensor], data_type: str) -> Tensor:
-    l = [paddle.to_tensor(_, data_type) for _ in x]
+    l = []
+    for t in x:
+        if np.isscalar(t) and not isinstance(t, str):
+            l.append(paddle.to_tensor([t], data_type))
+        else:
+            l.append(paddle.to_tensor(t, data_type))
     return paddle.concat(l)
 
 

diff --git a/python/paddle/distributed/fleet/meta_parallel/pp_utils/p2p_communication.py b/python/paddle/distributed/fleet/meta_parallel/pp_utils/p2p_communication.py
@@ -117,7 +117,7 @@ def recv_meta(self, group):
 
     def _send_dims_shape_dtype(self, tensor, group):
         # send len(shape)
-        dims = paddle.to_tensor(len(tensor.shape))
+        dims = paddle.to_tensor([len(tensor.shape)])
         dst_rank = _hcg._get_p2p_next_rank()
 
         paddle.distributed.send(dims, dst=dst_rank, group=group)
@@ -127,11 +127,11 @@ def _send_dims_shape_dtype(self, tensor, group):
         paddle.distributed.send(shape, dst=dst_rank, group=group)
 
         # send dtype
-        dtype = paddle.to_tensor(paddle_2_number(tensor.dtype))
+        dtype = paddle.to_tensor([paddle_2_number(tensor.dtype)])
         paddle.distributed.send(dtype, dst=dst_rank, group=group)
 
         # send trainable
-        stop_grad = paddle.to_tensor(int(tensor.stop_gradient))
+        stop_grad = paddle.to_tensor([int(tensor.stop_gradient)])
         paddle.distributed.send(stop_grad, dst=dst_rank, group=group)
 
     def send_meta(self, tensor, group):
@@ -148,7 +148,7 @@ def send_meta(self, tensor, group):
             # send tensor type
             paddle.distributed.send(tensor_type, dst=dst_rank, group=group)
 
-            nums = paddle.to_tensor(len(tensor))
+            nums = paddle.to_tensor([len(tensor)])
             paddle.distributed.send(nums, dst=dst_rank, group=group)
 
             for d in tensor:

diff --git a/python/paddle/fluid/tests/unittests/parallel_dygraph_none_var.py b/python/paddle/fluid/tests/unittests/parallel_dygraph_none_var.py
@@ -40,7 +40,7 @@ def __init__(self):
         self.step = 0
 
     def forward(self, x):
-        return paddle.to_tensor(0.0, dtype='float32')
+        return paddle.to_tensor([0.0], dtype='float32')
 
 
 def fake_sample_reader():

diff --git a/python/paddle/fluid/tests/unittests/test_cholesky_op.py b/python/paddle/fluid/tests/unittests/test_cholesky_op.py
@@ -104,7 +104,7 @@ def test_dygraph(self):
         a = np.random.rand(3, 3)
         a_t = np.transpose(a, [1, 0])
         x_data = np.matmul(a, a_t) + 1e-03
-        x = paddle.to_tensor(x_data)
+        x = paddle.to_tensor([x_data])
         out = paddle.cholesky(x, upper=False)
 
 

diff --git a/python/paddle/fluid/tests/unittests/test_deg2rad.py b/python/paddle/fluid/tests/unittests/test_deg2rad.py
@@ -74,7 +74,7 @@ def setUp(self):
     def test_dygraph(self):
         paddle.disable_static()
 
-        x2 = paddle.to_tensor(180)
+        x2 = paddle.to_tensor([180])
         result2 = paddle.deg2rad(x2)
         np.testing.assert_allclose(np.pi, result2.numpy(), rtol=1e-05)
 

diff --git a/python/paddle/fluid/tests/unittests/test_deprecated_decorator.py b/python/paddle/fluid/tests/unittests/test_deprecated_decorator.py
@@ -111,7 +111,7 @@ def test_ops_elementwise_mul(self):
     def test_tensor_gradient(self):
         paddle.__version__ = '2.1.0'
 
-        x = paddle.to_tensor(5.0, stop_gradient=False)
+        x = paddle.to_tensor([5.0], stop_gradient=False)
         y = paddle.pow(x, 4.0)
         y.backward()
 

diff --git a/python/paddle/fluid/tests/unittests/test_einsum_v2.py b/python/paddle/fluid/tests/unittests/test_einsum_v2.py
@@ -553,7 +553,7 @@ def test_shape(self):
             B = paddle.to_tensor(np.array([2.0, 3.0])).astype(paddle.bfloat16)
             B = B.cuda()
             C = paddle.einsum('i,i->', A, B)
-            D = paddle.to_tensor(8.0).astype(paddle.bfloat16)
+            D = paddle.to_tensor([8.0]).astype(paddle.bfloat16)
             self.assertEqual(C.item(), D.item())
 
 

diff --git a/python/paddle/fluid/tests/unittests/test_inplace.py b/python/paddle/fluid/tests/unittests/test_inplace.py
@@ -259,11 +259,11 @@ def inplace_api_processing(self, var):
 
 class TestDygraphInplaceReshapeTensor(TestDygraphInplace):
     def non_inplace_api_processing(self, var):
-        shape = paddle.to_tensor(-1)
+        shape = paddle.to_tensor([-1])
         return paddle.reshape(var, shape)
 
     def inplace_api_processing(self, var):
-        shape = paddle.to_tensor(-1)
+        shape = paddle.to_tensor([-1])
         return paddle.reshape_(var, shape)
 
 

diff --git a/python/paddle/fluid/tests/unittests/test_jit_save_load.py b/python/paddle/fluid/tests/unittests/test_jit_save_load.py
@@ -283,7 +283,7 @@ def __init__(self, in_size, out_size):
         super().__init__()
         self._linear_0 = Linear(in_size, out_size)
         self._linear_1 = Linear(in_size, out_size)
-        self._scale = paddle.to_tensor(9.9)
+        self._scale = paddle.to_tensor([9.9])
 
     @paddle.jit.to_static
     def forward(self, x):
@@ -1196,7 +1196,7 @@ def __init__(self, in_size, out_size):
         self._linear_1_0 = Linear(self.hidden, self.hidden)
         self._linear_1_1 = Linear(self.hidden, self.hidden)
         self._linear_2 = Linear(self.hidden, out_size)
-        self._scale = paddle.to_tensor(9.9)
+        self._scale = paddle.to_tensor([9.9])
 
     @paddle.jit.to_static
     def forward(self, x):
@@ -1319,7 +1319,7 @@ def __init__(self, in_size, out_size, load_path):
         self._linear_1_0 = Linear(out_size, in_size)
         self._linear_1_1 = Linear(out_size, in_size)
         self._linear_2 = Linear(out_size, out_size)
-        self._scale = paddle.to_tensor(9.9)
+        self._scale = paddle.to_tensor([9.9])
 
         # Load multiple times
         self._load_l1 = paddle.jit.load(load_path)

diff --git a/python/paddle/fluid/tests/unittests/test_lbfgs_v2.py b/python/paddle/fluid/tests/unittests/test_lbfgs_v2.py
@@ -208,64 +208,64 @@ def error_func():
 
     def test_line_search(self):
         def func1(x, alpha, d):
-            return paddle.to_tensor(x + alpha * d), paddle.to_tensor(0.0)
+            return paddle.to_tensor(x + alpha * d), paddle.to_tensor([0.0])
 
         def func2(x, alpha, d):
-            return paddle.to_tensor(x + alpha * d), paddle.to_tensor(1.0)
+            return paddle.to_tensor(x + alpha * d), paddle.to_tensor([1.0])
 
         def func3(x, alpha, d):
-            return paddle.to_tensor(x + alpha * d), paddle.to_tensor(-1.0)
+            return paddle.to_tensor(x + alpha * d), paddle.to_tensor([-1.0])
 
         _strong_wolfe(
             func1,
-            paddle.to_tensor(1.0),
-            paddle.to_tensor(0.001),
-            paddle.to_tensor(0.0),
-            paddle.to_tensor(1.0),
-            paddle.to_tensor(0.0),
-            paddle.to_tensor(0.0),
+            paddle.to_tensor([1.0]),
+            paddle.to_tensor([0.001]),
+            paddle.to_tensor([0.0]),
+            paddle.to_tensor([1.0]),
+            paddle.to_tensor([0.0]),
+            paddle.to_tensor([0.0]),
             max_ls=0,
         )
 
         _strong_wolfe(
             func2,
-            paddle.to_tensor(1.0),
-            paddle.to_tensor(-0.001),
-            paddle.to_tensor(1.0),
-            paddle.to_tensor(1.0),
-            paddle.to_tensor(1.0),
-            paddle.to_tensor(1.0),
+            paddle.to_tensor([1.0]),
+            paddle.to_tensor([-0.001]),
+            paddle.to_tensor([1.0]),
+            paddle.to_tensor([1.0]),
+            paddle.to_tensor([1.0]),
+            paddle.to_tensor([1.0]),
             max_ls=1,
         )
 
         _strong_wolfe(
             func3,
-            paddle.to_tensor(1.0),
-            paddle.to_tensor(-0.001),
-            paddle.to_tensor(1.0),
-            paddle.to_tensor(1.0),
-            paddle.to_tensor(1.0),
-            paddle.to_tensor(1.0),
+            paddle.to_tensor([1.0]),
+            paddle.to_tensor([-0.001]),
+            paddle.to_tensor([1.0]),
+            paddle.to_tensor([1.0]),
+            paddle.to_tensor([1.0]),
+            paddle.to_tensor([1.0]),
             max_ls=1,
         )
 
         _cubic_interpolate(
-            paddle.to_tensor(2.0),
-            paddle.to_tensor(1.0),
-            paddle.to_tensor(0.0),
-            paddle.to_tensor(1.0),
-            paddle.to_tensor(2.0),
-            paddle.to_tensor(0.0),
+            paddle.to_tensor([2.0]),
+            paddle.to_tensor([1.0]),
+            paddle.to_tensor([0.0]),
+            paddle.to_tensor([1.0]),
+            paddle.to_tensor([2.0]),
+            paddle.to_tensor([0.0]),
             [0.1, 0.5],
         )
 
         _cubic_interpolate(
-            paddle.to_tensor(2.0),
-            paddle.to_tensor(0.0),
-            paddle.to_tensor(-3.0),
-            paddle.to_tensor(1.0),
-            paddle.to_tensor(1.0),
-            paddle.to_tensor(-0.1),
+            paddle.to_tensor([2.0]),
+            paddle.to_tensor([0.0]),
+            paddle.to_tensor([-3.0]),
+            paddle.to_tensor([1.0]),
+            paddle.to_tensor([1.0]),
+            paddle.to_tensor([-0.1]),
             [0.1, 0.5],
         )
 

diff --git a/python/paddle/fluid/tests/unittests/test_rad2deg.py b/python/paddle/fluid/tests/unittests/test_rad2deg.py
@@ -73,7 +73,7 @@ def setUp(self):
     def test_dygraph(self):
         paddle.disable_static()
 
-        x2 = paddle.to_tensor(np.pi / 2)
+        x2 = paddle.to_tensor([np.pi / 2])
         result2 = paddle.rad2deg(x2)
         np.testing.assert_allclose(90, result2.numpy(), rtol=1e-05)
 
@@ -91,7 +91,7 @@ def setUp(self):
     def test_dygraph(self):
         paddle.disable_static()
 
-        x2 = paddle.to_tensor(1)
+        x2 = paddle.to_tensor([1])
         result2 = paddle.rad2deg(x2)
         np.testing.assert_allclose(180 / np.pi, result2.numpy(), rtol=1e-05)
 

diff --git a/python/paddle/fluid/tests/unittests/test_state_dict_convert.py b/python/paddle/fluid/tests/unittests/test_state_dict_convert.py
@@ -88,7 +88,7 @@ class TestStateDictReturn(unittest.TestCase):
     def test_missing_keys_and_unexpected_keys(self):
         model1 = MyModel2()
         tmp_dict = {}
-        tmp_dict["unexpected_keys"] = paddle.to_tensor(1)
+        tmp_dict["unexpected_keys"] = paddle.to_tensor([1])
         missing_keys, unexpected_keys = model1.set_state_dict(tmp_dict)
         self.assertEqual(len(missing_keys), 2)
         self.assertEqual(missing_keys[0], "linear.weight")

diff --git a/python/paddle/fluid/tests/unittests/test_tensor_register_hook.py b/python/paddle/fluid/tests/unittests/test_tensor_register_hook.py
@@ -553,7 +553,7 @@ def test_register_backward_hook(self):
         global HOOK_INIT_VALUE
         global HOOK_IS_CALLED
         for device in self.devices:
-            x = paddle.to_tensor(5.0, stop_gradient=False)
+            x = paddle.to_tensor([5.0], stop_gradient=False)
             x._register_backward_hook(global_void_hook)
             for i in range(5):
                 y = paddle.pow(x, 4.0)
@@ -567,14 +567,14 @@ def test_register_backward_hook(self):
             HOOK_IS_CALLED = False
 
     def test_register_backward_hook_for_interior_var(self):
-        x = paddle.to_tensor(5.0, stop_gradient=False)
+        x = paddle.to_tensor([5.0], stop_gradient=False)
         y = paddle.pow(x, 4.0)
 
         with self.assertRaises(ValueError):
             y._register_backward_hook(global_void_hook)
 
     def test_register_backward_hook_for_var_without_gradient(self):
-        x = paddle.to_tensor(5.0)
+        x = paddle.to_tensor([5.0])
         y = paddle.pow(x, 4.0)
 
         with self.assertRaises(ValueError):

diff --git a/python/paddle/fluid/tests/unittests/test_var_base.py b/python/paddle/fluid/tests/unittests/test_var_base.py
@@ -114,7 +114,7 @@ def check_with_place(place):
                 )
                 np.testing.assert_array_equal(x.numpy(), [1.0])
                 self.assertEqual(x.dtype, core.VarDesc.VarType.FP32)
-                self.assertEqual(x.shape, [1])
+                self.assertEqual(x.shape, [])
                 self.assertEqual(x.stop_gradient, False)
                 self.assertEqual(x.type, core.VarDesc.VarType.LOD_TENSOR)
 
@@ -407,7 +407,7 @@ def test_leaf_tensor(self):
 
     def test_detach(self):
         with fluid.dygraph.guard():
-            x = paddle.to_tensor(1.0, dtype="float64", stop_gradient=False)
+            x = paddle.to_tensor([1.0], dtype="float64", stop_gradient=False)
             detach_x = x.detach()
             self.assertTrue(detach_x.stop_gradient, True)
 

diff --git a/python/paddle/fluid/tests/unittests/test_zero_dim_tensor.py b/python/paddle/fluid/tests/unittests/test_zero_dim_tensor.py
@@ -2385,6 +2385,20 @@ def body(i, x):
         self.assertEqual(x.grad.shape, [])
         np.testing.assert_allclose(x.grad, np.array(1.0))
 
+    def test_to_tensor(self):
+        out1 = paddle.to_tensor(1)
+        out2 = paddle.to_tensor(2.5)
+
+        out1.retain_grads()
+        out1.backward()
+        out2.retain_grads()
+        out2.backward()
+
+        self.assertEqual(out1.shape, [])
+        self.assertEqual(out1, 1)
+        self.assertEqual(out2.shape, [])
+        self.assertEqual(out2, 2.5)
+
     def test_linalg_slogdet(self):
         # 2-D input
         x = paddle.randn([3, 3])
@@ -4355,6 +4369,19 @@ def test_broadcast_tensors(self):
         self.assertEqual(out1.shape, (2, 3))
         self.assertEqual(out2.shape, (2, 3))
 
+    @prog_scope()
+    def test_to_tensor(self):
+        out1 = paddle.to_tensor(1)
+        out2 = paddle.to_tensor(2.5)
+
+        prog = paddle.static.default_main_program()
+        res = self.exe.run(prog, fetch_list=[out1, out2])
+
+        self.assertEqual(res[0].shape, ())
+        self.assertEqual(res[0], 1)
+        self.assertEqual(res[1].shape, ())
+        self.assertEqual(res[1], 2.5)
+
     @prog_scope()
     def test_linalg_slogdet(self):
         # 2-D input