Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[FixUnitTest] Fix a Batch Size Random Unit Test Failure #24718

Merged
merged 7 commits into from
Jun 1, 2020
Merged

[FixUnitTest] Fix a Batch Size Random Unit Test Failure #24718

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
937e61c
Decrease the batch size of se-resnext in unittest.
zhhsplendid May 20, 2020
13f3268
Decrease the Batch Size of SE-Resnext in Unittest
zhhsplendid May 25, 2020
1a88fcc
Merge branch 'fix_unit_test' of https://github.com/zhhsplendid/Paddle…
zhhsplendid May 25, 2020
4197467
Decrease batch size again to match gpu size.
zhhsplendid May 25, 2020
99348b5
Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into…
zhhsplendid Jun 1, 2020
1daedc7
Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into…
zhhsplendid Jun 1, 2020
0e051aa
Found a precision problem that small batch size can cause CPU test fa…
zhhsplendid Jun 1, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
Found a precision problem that small batch size can cause CPU test fa… 
…ilure.

test=develop
  • Loading branch information
@zhhsplendid
zhhsplendid committed Jun 1, 2020
commit 0e051aa2c55bc9bbb8a45e2015987ed9c9c36a14
11 changes: 7 additions & 4 deletions python/paddle/fluid/tests/unittests/seresnext_net.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -172,8 +172,11 @@ def optimizer(learning_rate=0.01):
model = SE_ResNeXt50Small


def batch_size():
return 4
def batch_size(use_cuda):
if use_cuda:
# Paddle uses 8GB P4 GPU for unittest so we decreased the batch size.
return 8
return 12


def iter(use_cuda):
Expand All @@ -183,9 +186,9 @@ def iter(use_cuda):


gpu_img, gpu_label = init_data(
batch_size=batch_size(), img_shape=img_shape, label_range=999)
batch_size=batch_size(use_cuda=True), img_shape=img_shape, label_range=999)
cpu_img, cpu_label = init_data(
batch_size=batch_size(), img_shape=img_shape, label_range=999)
batch_size=batch_size(use_cuda=False), img_shape=img_shape, label_range=999)
feed_dict_gpu = {"image": gpu_img, "label": gpu_label}
feed_dict_cpu = {"image": cpu_img, "label": cpu_label}

Expand Down
4 changes: 2 additions & 2 deletions python/paddle/fluid/tests/unittests/seresnext_test_base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,7 +32,7 @@ def _compare_result_with_origin_model(self,
seresnext_net.model,
feed_dict=seresnext_net.feed_dict(use_cuda),
iter=seresnext_net.iter(use_cuda),
batch_size=seresnext_net.batch_size(),
batch_size=seresnext_net.batch_size(use_cuda),
use_cuda=use_cuda,
use_reduce=False,
optimizer=seresnext_net.optimizer)
Expand All @@ -41,7 +41,7 @@ def _compare_result_with_origin_model(self,
seresnext_net.model,
feed_dict=seresnext_net.feed_dict(use_cuda),
iter=seresnext_net.iter(use_cuda),
batch_size=seresnext_net.batch_size(),
batch_size=seresnext_net.batch_size(use_cuda),
use_cuda=use_cuda)

if compare_seperately:
Expand Down
8 changes: 4 additions & 4 deletions python/paddle/fluid/tests/unittests/test_parallel_executor_seresnext_with_reduce_cpu.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,15 +28,15 @@ def _compare_reduce_and_allreduce(self, use_cuda, delta2=1e-5):
seresnext_net.model,
feed_dict=seresnext_net.feed_dict(use_cuda),
iter=seresnext_net.iter(use_cuda),
batch_size=seresnext_net.batch_size(),
batch_size=seresnext_net.batch_size(use_cuda),
use_cuda=use_cuda,
use_reduce=False,
optimizer=seresnext_net.optimizer)
reduce_first_loss, reduce_last_loss = self.check_network_convergence(
seresnext_net.model,
feed_dict=seresnext_net.feed_dict(use_cuda),
iter=seresnext_net.iter(use_cuda),
batch_size=seresnext_net.batch_size(),
batch_size=seresnext_net.batch_size(use_cuda),
use_cuda=use_cuda,
use_reduce=True,
optimizer=seresnext_net.optimizer)
Expand All @@ -53,7 +53,7 @@ def _compare_reduce_and_allreduce(self, use_cuda, delta2=1e-5):
seresnext_net.model,
feed_dict=seresnext_net.feed_dict(use_cuda),
iter=seresnext_net.iter(use_cuda),
batch_size=seresnext_net.batch_size(),
batch_size=seresnext_net.batch_size(use_cuda),
use_cuda=use_cuda,
use_reduce=False,
optimizer=seresnext_net.optimizer,
Expand All @@ -63,7 +63,7 @@ def _compare_reduce_and_allreduce(self, use_cuda, delta2=1e-5):
seresnext_net.model,
feed_dict=seresnext_net.feed_dict(use_cuda),
iter=seresnext_net.iter(use_cuda),
batch_size=seresnext_net.batch_size(),
batch_size=seresnext_net.batch_size(use_cuda),
use_cuda=use_cuda,
use_reduce=True,
optimizer=seresnext_net.optimizer,
Expand Down