wehub-resource-sync
341 lines
10 KiB
Python
341 lines
10 KiB
Python
# Copyright (c) 2025 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 os
|
|
import platform
|
|
import shutil
|
|
import subprocess
|
|
import sys
|
|
import unittest
|
|
from unittest.mock import patch
|
|
|
|
import paddle
|
|
|
|
|
|
# Test the dump_backward_graph_path params in backward
|
|
# Just check whether the debug file is generated
|
|
class TestDumpDebugInfo(unittest.TestCase):
|
|
def test_dump_debug_info(self):
|
|
# windows ci may have some permission issues
|
|
if 'Windows' == platform.system():
|
|
return
|
|
paddle.disable_static()
|
|
self._test_Tensor_backward()
|
|
self._test_paddle_grad()
|
|
self._test_autograd_backward()
|
|
paddle.enable_static()
|
|
|
|
def _test_Tensor_backward(self):
|
|
x = paddle.randn([5, 5], dtype='float32')
|
|
y = paddle.randn([5, 5], dtype='float16')
|
|
x.stop_gradient = False
|
|
y.stop_gradient = False
|
|
z = x + y
|
|
h = z + 1
|
|
h = h * z
|
|
w = h + y
|
|
# test Tensor.backward
|
|
dump_backward_graph_path = "_Tensor_backward/"
|
|
w.backward(dump_backward_graph_path=dump_backward_graph_path)
|
|
self._check_files_in_directory(dump_backward_graph_path)
|
|
shutil.rmtree(dump_backward_graph_path)
|
|
|
|
def _test_paddle_grad(self):
|
|
x = paddle.randn([5, 5], dtype='float32')
|
|
y = paddle.randn([5, 5], dtype='float32')
|
|
x.stop_gradient = False
|
|
y.stop_gradient = False
|
|
z = x + y
|
|
h = x * z
|
|
w = h + y
|
|
# test paddle.grad
|
|
dump_backward_graph_path = "_paddle_grad/"
|
|
grads = paddle.grad(
|
|
[w], [x, y], dump_backward_graph_path=dump_backward_graph_path
|
|
)
|
|
self._check_files_in_directory(dump_backward_graph_path)
|
|
shutil.rmtree(dump_backward_graph_path)
|
|
|
|
def _test_autograd_backward(self):
|
|
x = paddle.randn([5, 5], dtype='float32')
|
|
y = paddle.randn([5, 5], dtype='float32')
|
|
x.stop_gradient = False
|
|
y.stop_gradient = False
|
|
z = x + y
|
|
h = x * z
|
|
w = h + y
|
|
# test paddle.autograd.backward
|
|
dump_backward_graph_path = "_paddle_autograd_backward/"
|
|
grads = paddle.autograd.backward(
|
|
[x, y],
|
|
[None, None],
|
|
dump_backward_graph_path=dump_backward_graph_path,
|
|
)
|
|
self._check_files_in_directory(dump_backward_graph_path)
|
|
shutil.rmtree(dump_backward_graph_path)
|
|
|
|
def _check_files_in_directory(self, directory):
|
|
# Check whether the expected file exists in the directory
|
|
entries = os.listdir(directory)
|
|
files = [
|
|
entry
|
|
for entry in entries
|
|
if os.path.isfile(os.path.join(directory, entry))
|
|
]
|
|
expect_keywords_in_file_name = [
|
|
"backward_graph.dot",
|
|
"ref_forward_graph.dot",
|
|
"call_stack.log",
|
|
]
|
|
for keywords in expect_keywords_in_file_name:
|
|
if not any(keywords in f for f in files):
|
|
raise AssertionError(
|
|
f"Error: File '{keywords}' not found in directory '{directory}'! "
|
|
)
|
|
|
|
# Just execute vlog for the coverage ci
|
|
def test_vlog(self):
|
|
code = """
|
|
import os
|
|
os.environ['GLOG_v'] = '{glog_level}'
|
|
import paddle
|
|
x = paddle.randn([5, 5], dtype='float32')
|
|
y = paddle.randn([5, 5], dtype='float32')
|
|
x.stop_gradient = False
|
|
y.stop_gradient = False
|
|
z = x + y
|
|
h = x * z
|
|
w = h + y
|
|
grads = paddle.autograd.backward(
|
|
[x, y],
|
|
[None, None],
|
|
)
|
|
paddle.base.core.set_vlog_level(4)
|
|
"""
|
|
process = subprocess.run(
|
|
[sys.executable, '-c', code.format(glog_level=4)],
|
|
capture_output=True,
|
|
text=True,
|
|
)
|
|
process = subprocess.run(
|
|
[sys.executable, '-c', code.format(glog_level=5)],
|
|
capture_output=True,
|
|
text=True,
|
|
)
|
|
process = subprocess.run(
|
|
[sys.executable, '-c', code.format(glog_level=6)],
|
|
capture_output=True,
|
|
text=True,
|
|
)
|
|
process = subprocess.run(
|
|
[sys.executable, '-c', code.format(glog_level=11)],
|
|
capture_output=True,
|
|
text=True,
|
|
)
|
|
|
|
def test_dump_call_stack(self):
|
|
code = """
|
|
import os
|
|
os.environ['FLAGS_dump_api_and_gradnode_python_stack_dir']="{dir}"
|
|
import paddle
|
|
x = paddle.randn([5, 5], dtype='float32')
|
|
y = paddle.randn([5, 5], dtype='float32')
|
|
x.stop_gradient = False
|
|
y.stop_gradient = False
|
|
z = x + y
|
|
h = x * z
|
|
w = h + y
|
|
grads = paddle.autograd.backward(
|
|
[x, y],
|
|
[None, None],
|
|
)
|
|
paddle.base.core.set_vlog_level(4)
|
|
"""
|
|
process = subprocess.run(
|
|
[sys.executable, '-c', code.format(dir="./")],
|
|
capture_output=True,
|
|
text=True,
|
|
)
|
|
process = subprocess.run(
|
|
[sys.executable, '-c', code.format(dir=".")],
|
|
capture_output=True,
|
|
text=True,
|
|
)
|
|
|
|
def test_manual_vlog(self):
|
|
if 'Windows' == platform.system():
|
|
return
|
|
code = """
|
|
import os
|
|
os.environ['GLOG_v'] = '6'
|
|
os.environ['FLAGS_dump_grad_node_forward_stack_path']="call_stack.log"
|
|
os.environ['FLAGS_call_stack_level']='3'
|
|
os.environ['FLAGS_dump_api_python_stack_path']="forward_call_stack"
|
|
|
|
import paddle
|
|
import paddle.nn.functional as F
|
|
import paddle.nn as nn
|
|
|
|
# Pylayer indent log
|
|
from paddle.autograd import PyLayer
|
|
class cus_tanh(PyLayer):
|
|
@staticmethod
|
|
def forward(ctx, x):
|
|
y = paddle.tanh(x)
|
|
# Pass tensors to backward.
|
|
ctx.save_for_backward(y)
|
|
return y
|
|
@staticmethod
|
|
def backward(ctx, dy):
|
|
# Get the tensors passed by forward.
|
|
y, = ctx.saved_tensor()
|
|
grad = dy * (1 - paddle.square(y))
|
|
return grad
|
|
|
|
pylayer_input = paddle.rand([3, 4])
|
|
pylayer_input.stop_gradient = False
|
|
custom_tanh = cus_tanh.apply
|
|
pylayer_output = custom_tanh(pylayer_input)
|
|
pylayer_output.mean().backward()
|
|
|
|
paddle.base.core.set_vlog_level({"backward":6, "*": 7})
|
|
|
|
x = paddle.randn([3,3],dtype='float16')
|
|
y = paddle.randn([3,3],dtype='float32')
|
|
z = paddle.randn([3,3],dtype='float64')
|
|
w = paddle.randn([3,3],dtype='float64')
|
|
x.stop_gradient = False
|
|
y.stop_gradient = False
|
|
z.stop_gradient = False
|
|
w.stop_gradient = True
|
|
|
|
conv_x = paddle.randn((2, 3, 8, 8), dtype='float32')
|
|
conv_w = paddle.randn((6, 3, 3, 3), dtype='float16')
|
|
|
|
sync_bn_input = paddle.to_tensor([[[[0.3, 0.4], [0.3, 0.07]], [[0.83, 0.37], [0.18, 0.93]]]]).astype('float32')
|
|
|
|
conv_x.stop_gradient = False
|
|
conv_w.stop_gradient = False
|
|
sync_bn_input.stop_gradient = False
|
|
|
|
with paddle.amp.auto_cast(enable=True):
|
|
out1 = paddle.add_n([x,y])
|
|
out2 = paddle.multiply(x,y)
|
|
out6 = F.conv2d(conv_x,conv_w)
|
|
|
|
out3 = paddle.add_n([out1,y])
|
|
out4 = paddle.multiply(out2,z)
|
|
out5 = paddle.multiply_(w, y)
|
|
if paddle.is_compiled_with_cuda():
|
|
sync_batch_norm = nn.SyncBatchNorm(2)
|
|
hidden1 = sync_batch_norm(sync_bn_input)
|
|
loss = out1 + out2 + out3 + out4 + out5 + out6.sum()+hidden1.sum()
|
|
loss.backward(dump_backward_graph_path="./backward")
|
|
|
|
|
|
"""
|
|
process = subprocess.run(
|
|
[sys.executable, '-c', code],
|
|
capture_output=True,
|
|
text=True,
|
|
)
|
|
|
|
# Test the input path is not valid
|
|
@patch('os.path.exists')
|
|
@patch('os.path.isdir')
|
|
def test_raise_not_a_directory_error(self, mock_isdir, mock_exists):
|
|
# simulate
|
|
mock_exists.return_value = True
|
|
mock_isdir.return_value = False
|
|
paddle.disable_static()
|
|
with self.assertRaises(NotADirectoryError) as context:
|
|
x = paddle.randn([5, 5], dtype='float32')
|
|
y = paddle.randn([5, 5], dtype='float32')
|
|
x.stop_gradient = False
|
|
y.stop_gradient = False
|
|
z = x + y
|
|
h = x * z
|
|
w = h + y
|
|
grads = paddle.autograd.backward(
|
|
[x, y], [None, None], dump_backward_graph_path="/path/to/check"
|
|
)
|
|
|
|
self.assertTrue(
|
|
" path:'/path/to/check' must be directory "
|
|
in str(context.exception)
|
|
)
|
|
|
|
@patch('os.makedirs')
|
|
def test_create_file_error(self, mock_makedirs):
|
|
# simulate os.makedirs throw exception
|
|
mock_makedirs.side_effect = Exception("Mocked exception")
|
|
with self.assertRaises(OSError) as context:
|
|
x = paddle.randn([5, 5], dtype='float32')
|
|
y = paddle.randn([5, 5], dtype='float32')
|
|
x.stop_gradient = False
|
|
y.stop_gradient = False
|
|
z = x + y
|
|
h = x * z
|
|
w = h + y
|
|
grads = paddle.autograd.backward(
|
|
[x, y], [None, None], dump_backward_graph_path='/path/to/create'
|
|
)
|
|
|
|
self.assertTrue(
|
|
"Create '/path/to/create' failed : Mocked exception"
|
|
in str(context.exception)
|
|
)
|
|
|
|
|
|
class TestSetVlogLevelError(unittest.TestCase):
|
|
def test_input_invalid(self):
|
|
with self.assertRaises(ValueError):
|
|
paddle.base.core.set_vlog_level("3")
|
|
|
|
|
|
class TestVlogGuard(unittest.TestCase):
|
|
# Just run it for coverage ci and don't check the res
|
|
def test_guard(self):
|
|
with paddle.base.framework.vlog_guard(0):
|
|
x = paddle.randn([3, 3], dtype='float16')
|
|
with paddle.base.framework.vlog_guard({"api": 0}):
|
|
y = paddle.randn([3, 3], dtype='float16')
|
|
|
|
# Check the invalid input
|
|
def test_error(self):
|
|
def test_invalid_input():
|
|
with paddle.base.framework.vlog_guard("api"):
|
|
x = paddle.randn([3, 3], dtype='float16')
|
|
|
|
self.assertRaises(TypeError, test_invalid_input)
|
|
|
|
|
|
class TestBackwardVlogGuard(unittest.TestCase):
|
|
def test_guard(self):
|
|
x = paddle.randn([3, 3], dtype='float32')
|
|
y = paddle.randn([3, 3], dtype='float32')
|
|
x.stop_gradient = False
|
|
y.stop_gradient = False
|
|
|
|
with paddle.base.framework.backward_vlog_guard(4):
|
|
z = x + y
|
|
h = x * z
|
|
w = h + y
|
|
loss = w.sum()
|
|
loss.backward()
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|