Skip to content
New issue

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

#

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

Already on GitHub? # to your account

Relieve memory usage criteria on batch size 2 during adaptive_bs #4009

Merged
merged 3 commits into from
Oct 10, 2024
Merged

Relieve memory usage criteria on batch size 2 during adaptive_bs #4009

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
360bd23
release memory usage cirteria on batch size 2 during adpative_bs
eunwoosh Oct 10, 2024
6b9153a
update unit test
eunwoosh Oct 10, 2024
47d80ee
update unit test
eunwoosh Oct 10, 2024
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
20 changes: 16 additions & 4 deletions src/otx/engine/adaptive_bs/bs_search_algo.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -112,8 +112,14 @@
break

if available_bs == 0:
msg = "Current device can't train model even with 2."
raise RuntimeError(msg)
if oom:
msg = "Current device can't train model even with 2."
raise RuntimeError(msg)
logger.warning(
"Even with a batch size of 2, most of the memory is used, "
"which could cause the training to fail midway.",
)
available_bs = 2

return available_bs

Expand Down Expand Up @@ -141,8 +147,14 @@
if oom or bs_mem_usage > self._mem_upper_bound:
self._default_bs -= 2
if self._default_bs <= 0:
msg = "Current device can't train model even with 2."
raise RuntimeError(msg)
if oom:
msg = "Current device can't train model even with 2."
raise RuntimeError(msg)

Check warning on line 152 in src/otx/engine/adaptive_bs/bs_search_algo.py

View check run for this annotation

Codecov / codecov/patch

src/otx/engine/adaptive_bs/bs_search_algo.py#L151-L152 

Added lines #L151 - L152 were not covered by tests
logger.warning(
"Even with a batch size of 2, most of the memory is used, "
"which could cause the training to fail midway.",
)
return 2

return self.auto_decrease_batch_size()

Expand Down
18 changes: 16 additions & 2 deletions tests/unit/engine/adaptive_bs/test_bs_search_algo.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -99,12 +99,19 @@ def test_auto_decrease_batch_size(self):
assert adapted_bs == 80

def test_find_max_usable_bs_gpu_memory_too_small(self):
mock_train_func = self.get_mock_train_func(cuda_oom_bound=4, max_runnable_bs=1)
mock_train_func = self.get_mock_train_func(cuda_oom_bound=1, max_runnable_bs=1)

bs_search_algo = BsSearchAlgo(mock_train_func, 128, 1000)
with pytest.raises(RuntimeError):
bs_search_algo.auto_decrease_batch_size()

def test_auto_decrease_batch_size_bs2_not_oom_but_most_mem(self):
"""Batch size 2 doesn't make oom but use most of memory."""
mock_train_func = self.get_mock_train_func(cuda_oom_bound=2, max_runnable_bs=1)

bs_search_algo = BsSearchAlgo(mock_train_func, 128, 1000)
assert bs_search_algo.auto_decrease_batch_size() == 2

@pytest.mark.parametrize(
("max_runnable_bs", "max_bs", "expected_bs"),
[
Expand All @@ -126,12 +133,19 @@ def test_find_big_enough_batch_size(self, max_runnable_bs, max_bs, expected_bs):
assert adapted_bs == expected_bs

def test_find_big_enough_batch_size_gpu_memory_too_small(self):
mock_train_func = self.get_mock_train_func(cuda_oom_bound=4, max_runnable_bs=1)
mock_train_func = self.get_mock_train_func(cuda_oom_bound=1, max_runnable_bs=1)

bs_search_algo = BsSearchAlgo(mock_train_func, 128, 1000)
with pytest.raises(RuntimeError):
bs_search_algo.find_big_enough_batch_size()

def test_find_big_enough_batch_size_bs2_not_oom_but_most_mem(self):
"""Batch size 2 doesn't make oom but use most of memory."""
mock_train_func = self.get_mock_train_func(cuda_oom_bound=2, max_runnable_bs=1)

bs_search_algo = BsSearchAlgo(mock_train_func, 2, 1000)
assert bs_search_algo.find_big_enough_batch_size() == 2

def test_find_big_enough_batch_size_gradient_zero(self):
def mock_train_func(batch_size) -> int:
if batch_size > 1000:
Expand Down
Loading