Check whether there is any inplace operation affecting gradient calculation. (#27901)
* Add a class TensorInplaceVersion to count the inplace version and put it in framework::Tensor instead of Allocation or Variable. * Add a new attribute `_inplace_version` for VarBase. * Raise exception if an inplace operation can result in incorrect gradient computation. * Add a new interface _bump_inplace_version() for VarBase to bump the version whenever the Tensor is modified through an inplace operation. * For api assign, call _bump_inplace_version() when it's an inplace operation inn dynamic mode. * Use original var_wrapper if the inplace_version is not changed. * Replace SnapshotVarWrapperList with SnapshotVarWrapper to optimize performane.release/2.0-rc1
liym27
5 years ago
committed by
GitHub
parent
c21a979790
commit
865a45984f
@ -0,0 +1,99 @@
|
||||
# Copyright (c) 2020 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.
|
||||
|
||||
from __future__ import print_function
|
||||
|
||||
import unittest
|
||||
import numpy as np
|
||||
|
||||
import paddle
|
||||
import paddle.fluid.core as core
|
||||
|
||||
|
||||
class TestInplace(unittest.TestCase):
|
||||
def test_forward_version(self):
|
||||
with paddle.fluid.dygraph.guard():
|
||||
var = paddle.to_tensor(np.ones((4, 2, 3)).astype(np.float32))
|
||||
self.assertEqual(var.inplace_version, 0)
|
||||
|
||||
var[0] = 1.1
|
||||
self.assertEqual(var.inplace_version, 1)
|
||||
|
||||
paddle.nn.functional.assign(paddle.ones(shape=[3]), var)
|
||||
|
||||
# NOTE(liym27): assign(input, output) is an inplace operation for output.
|
||||
# There is inplace-related processing for api assign, var.inplace_version should be 2 not 1.
|
||||
self.assertEqual(var.inplace_version, 2)
|
||||
|
||||
var[2] = 3
|
||||
self.assertEqual(var.inplace_version, 3)
|
||||
|
||||
def test_backward_error(self):
|
||||
# It raises an error because the inplace operator will result
|
||||
# in incorrect gradient computation.
|
||||
with paddle.fluid.dygraph.guard():
|
||||
var_a = paddle.ones(shape=[4, 2, 3], dtype="float32")
|
||||
var_a.stop_gradient = False
|
||||
|
||||
var_b = var_a**2
|
||||
|
||||
# Here, the gradient computation will use the value of var_b
|
||||
var_c = var_b**2
|
||||
var_b[1:2] = 3.3 # var_b is modified inplace after using it
|
||||
|
||||
var_d = var_b**2
|
||||
|
||||
loss = paddle.nn.functional.relu(var_c + var_d)
|
||||
with self.assertRaisesRegexp(
|
||||
RuntimeError,
|
||||
"received tensor_version:{} != wrapper_version_snapshot:{}".
|
||||
format(1, 0)):
|
||||
loss.backward()
|
||||
|
||||
def test_backward_success_1(self):
|
||||
# var_b is modified inplace before using it, the inplace operator doesn't result
|
||||
# in incorrect gradient computation.
|
||||
with paddle.fluid.dygraph.guard():
|
||||
var_a = paddle.ones(shape=[4, 2, 3], dtype="float32")
|
||||
var_a.stop_gradient = False
|
||||
|
||||
var_b = var_a**2
|
||||
var_b[1:2] = 3 # var_b is modified inplace before using it
|
||||
|
||||
# Here, the gradient computation will use the value of var_b
|
||||
var_c = var_b**2
|
||||
loss = var_c.sum()
|
||||
loss.backward()
|
||||
|
||||
def test_backward_success_2(self):
|
||||
# Although var_b is modified inplace after using it, it does not used in gradient computation.
|
||||
# The inplace operator doesn't result in incorrect gradient computation.
|
||||
with paddle.fluid.dygraph.guard():
|
||||
var_a = paddle.ones(shape=[4, 2, 3], dtype="float32")
|
||||
var_a.stop_gradient = False
|
||||
|
||||
var_b = var_a**2
|
||||
|
||||
var_b[1:2] = 3 # var_b is modified inplace before using it
|
||||
|
||||
var_c = var_b + var_b # Here, the grad op of sum doesn't use the value of var_b
|
||||
loss = var_c.sum()
|
||||
|
||||
var_b[1:2] = 3 # var_b is modified inplace after using it
|
||||
|
||||
loss.backward()
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
unittest.main()
|
||||
Loading…
Reference in new issue