Skip to content

Fix CI slow test ValueError: Backward pass should have cleared tracker of all tensors #4236

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

Merged
merged 4 commits into from
Oct 10, 2025
Merged

Fix CI slow test ValueError: Backward pass should have cleared tracker of all tensors #4236

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
b1468c7
fix error of
sywangyi Oct 9, 2025
d42bf31
Merge branch 'main' into offload_test_fix
sywangyi Oct 9, 2025
b3a681b
Apply suggestion from @kashif
kashif Oct 10, 2025
20cc0a1
Apply suggestion from @kashif
kashif Oct 10, 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
8 changes: 6 additions & 2 deletions trl/models/activation_offloading.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -219,7 +219,6 @@ def unpack_tensor_single_stream(unpack_tensor_id: int) -> torch.Tensor:
verify_sufficient_virtual_memory()

self.is_first_backward_call = False
self.is_first_forward_call = True

if unpack_tensor_id not in self.tracker:
raise ValueError(f"Untracked tensor with id {unpack_tensor_id}")
Expand All @@ -231,6 +230,9 @@ def unpack_tensor_single_stream(unpack_tensor_id: int) -> torch.Tensor:

# clear tensor from tracking
del self.tracker[unpack_tensor_id]
# Only set is_first_forward_call to True when all tensors have been unpacked
if len(self.tracker) == 0:
self.is_first_forward_call = True
return maybe_accelerator_tensor

def unpack_tensor_with_streams(unpack_tensor_id: int) -> torch.Tensor:
Expand All @@ -254,7 +256,6 @@ def wait_and_del_remaining_references() -> None:
verify_sufficient_virtual_memory()

self.is_first_backward_call = False
self.is_first_forward_call = True

if unpack_tensor_id not in self.tracker:
raise ValueError(f"untracked tensor with id {unpack_tensor_id}")
Expand Down Expand Up @@ -359,6 +360,9 @@ def hook(outputs, inputs):

# clear tensor from tracking
del self.tracker[unpack_tensor_id]
# Only set is_first_forward_call to True when all tensors have been unpacked
if len(self.tracker) == 0:
self.is_first_forward_call = True
return maybe_accelerator_tensor

unpack_tensor = unpack_tensor_with_streams if self.use_streams else unpack_tensor_single_stream
Expand Down
Loading