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

Fix for incorrect usage of detach(), cpu(), to() #6216

Merged
merged 12 commits into from
Mar 1, 2021
Merged

Fix for incorrect usage of detach(), cpu(), to() #6216

Show file tree
Hide file tree
Changes from 3 commits
Commits
Show all changes
12 commits
Select commit Hold shift + click to select a range
85a2ad1
Fix for incorrect detach/cpu calls (#6214)
dvolgyes Feb 26, 2021
1fe84f4
Fix incorrect use of detach(), to(), and cpu(), #6214
dvolgyes Feb 26, 2021
a77ef95
Fix incorrect use of detach() and cpu(), #6214
dvolgyes Feb 26, 2021
48cfb7f
update pr
tchaton Feb 26, 2021
c170e9d
add typing
tchaton Feb 26, 2021
f069c07
chlog
Borda Feb 26, 2021
393dae1
more...
Borda Feb 26, 2021
fc5dca6
Merge branch 'master' into patch-1
tchaton Mar 1, 2021
0e9f340
revert on module
tchaton Mar 1, 2021
aeb346a
update on comments
tchaton Mar 1, 2021
d6be983
revert changes on model
tchaton Mar 1, 2021
30258e2
Merge branch 'master' into patch-1
tchaton Mar 1, 2021
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
6 changes: 3 additions & 3 deletions pytorch_lightning/trainer/connectors/logger_connector/epoch_result_store.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -281,11 +281,11 @@ def cache_result(self) -> None:
# attach capture batch_size
Result.attach_batch_size(self._batch_size, hook_result)

hook_result.detach()
hook_result = hook_result.detach()
if self.trainer.move_metrics_to_cpu:
hook_result.cpu()
hook_result = hook_result.cpu()
elif self.trainer._distrib_type == DistributedType.DP:
hook_result.to(torch.device("cuda", self.trainer.root_gpu))
hook_result = hook_result.to(torch.device("cuda", self.trainer.root_gpu))

self._internals[fx_name].append(hook_result, info)

Expand Down
4 changes: 2 additions & 2 deletions pytorch_lightning/trainer/trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -736,9 +736,9 @@ def run_evaluation(self, max_batches=None, on_epoch=False):
def track_output_for_epoch_end(self, outputs, output):
if output is not None:
if isinstance(output, Result):
output.detach()
output = output.detach()
if self.move_metrics_to_cpu:
output.cpu()
output = output.cpu()
elif isinstance(output, dict):
output = recursive_detach(output, to_cpu=self.move_metrics_to_cpu)
elif isinstance(output, torch.Tensor) and output.is_cuda and self.move_metrics_to_cpu:
Expand Down
6 changes: 3 additions & 3 deletions pytorch_lightning/trainer/training_loop.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -261,7 +261,7 @@ def on_after_backward(self, training_step_output, batch_idx, untouched_loss):
is_result_obj = isinstance(training_step_output, Result)

if is_result_obj:
training_step_output.detach()
training_step_output = training_step_output.detach()
else:
training_step_output.batch_loss = training_step_output.batch_loss.detach()

Expand Down Expand Up @@ -395,9 +395,9 @@ def _process_training_step_output_1_0(self, training_step_output, split_batch):

# track metrics without grads for epoch reduction
training_step_output_for_epoch_end = copy(result)
training_step_output_for_epoch_end.detach()
training_step_output_for_epoch_end = training_step_output_for_epoch_end.detach()
if self.trainer.move_metrics_to_cpu:
training_step_output_for_epoch_end.cpu()
training_step_output_for_epoch_end = training_step_output_for_epoch_end.cpu()

# what flows back into the system
training_step_output = result
Expand Down