Skip to content

Fixes #12673. record_stream in group offloading is not working properly #12721

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

Open
wants to merge 4 commits into
base: main
Choose a base branch
from
+10 −11
Open

Fixes #12673. record_stream in group offloading is not working properly #12721

Changes from 3 commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
9159ed7
Fixes #12673.
KimbingNg Nov 27, 2025
4d39e8c
update
KimbingNg Nov 27, 2025
e5533c1
Merge branch 'main' into record_stream_bug
sayakpaul Nov 27, 2025
e56f511
Update test
KimbingNg Nov 27, 2025
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
18 changes: 10 additions & 8 deletions src/diffusers/hooks/group_offloading.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -153,27 +153,27 @@ def _pinned_memory_tensors(self):
finally:
pinned_dict = None

def _transfer_tensor_to_device(self, tensor, source_tensor):
def _transfer_tensor_to_device(self, tensor, source_tensor, default_stream):
tensor.data = source_tensor.to(self.onload_device, non_blocking=self.non_blocking)
if self.record_stream:
tensor.data.record_stream(self._torch_accelerator_module.current_stream())
tensor.data.record_stream(default_stream)

def _process_tensors_from_modules(self, pinned_memory=None):
def _process_tensors_from_modules(self, pinned_memory=None, default_stream=None):
for group_module in self.modules:
for param in group_module.parameters():
source = pinned_memory[param] if pinned_memory else param.data
self._transfer_tensor_to_device(param, source)
self._transfer_tensor_to_device(param, source, default_stream)
for buffer in group_module.buffers():
source = pinned_memory[buffer] if pinned_memory else buffer.data
self._transfer_tensor_to_device(buffer, source)
self._transfer_tensor_to_device(buffer, source, default_stream)

for param in self.parameters:
source = pinned_memory[param] if pinned_memory else param.data
self._transfer_tensor_to_device(param, source)
self._transfer_tensor_to_device(param, source, default_stream)

for buffer in self.buffers:
source = pinned_memory[buffer] if pinned_memory else buffer.data
self._transfer_tensor_to_device(buffer, source)
self._transfer_tensor_to_device(buffer, source, default_stream)

def _onload_from_disk(self):
if self.stream is not None:
Expand Down Expand Up @@ -208,10 +208,12 @@ def _onload_from_memory(self):
self.stream.synchronize()

context = nullcontext() if self.stream is None else self._torch_accelerator_module.stream(self.stream)
default_stream = self._torch_accelerator_module.current_stream() if self.stream is not None else None

with context:
if self.stream is not None:
with self._pinned_memory_tensors() as pinned_memory:
self._process_tensors_from_modules(pinned_memory)
self._process_tensors_from_modules(pinned_memory, default_stream=default_stream)
else:
self._process_tensors_from_modules(None)

Expand Down
Loading