Skip to content

fix: fix clone in group_tensors_by_shape_and_dtype #98

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
chaokunyang merged 1 commit into from
Dec 22, 2025
Merged

fix: fix clone in group_tensors_by_shape_and_dtype #98

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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: 5 additions & 3 deletions awex/util/tensor_util.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -211,7 +211,7 @@ def group_tensors_by_shape_and_dtype(
final_tensor_groups = []
metadata = []

for (_shape, _dtype), group_tensors in tensor_groups.items():
for _, group_tensors in tensor_groups.items():
# Sort by original index to maintain order
group_tensors.sort(key=lambda x: x[0])
# Split into multiple groups - try to group tensors efficiently
Expand All @@ -227,7 +227,8 @@ def group_tensors_by_shape_and_dtype(
# Check if this tensor can fit in current group
if current_group_size > max_tensor_size:
# Finalize current group and start new one
concatenated = torch.cat(current_group, dim=0).contiguous()
# Use clone() to ensure a copy so caller can safely release original tensors
concatenated = torch.cat(current_group, dim=0).clone()
final_tensor_groups.append(concatenated)
# Record metadata for tensors in this group
offset_elements = 0
Expand All @@ -251,7 +252,8 @@ def group_tensors_by_shape_and_dtype(

# Finalize any remaining group
if current_group:
concatenated = torch.cat(current_group, dim=0).contiguous()
# Use clone() to ensure a copy so caller can safely release original tensors
concatenated = torch.cat(current_group, dim=0).clone()
final_tensor_groups.append(concatenated)
# Record metadata for tensors in this group
offset_elements = 0
Expand Down