Skip to content

[hardware] fix: Workaround for torch-npu's lack of support for creating nested tensors from NPU tensors. #4309

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
wuxibin89 merged 1 commit into from
Nov 27, 2025
+20 −0
Merged

[hardware] fix: Workaround for torch-npu's lack of support for creating nested tensors from NPU tensors. #4309

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
20 changes: 20 additions & 0 deletions verl/__init__.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -52,6 +52,26 @@
patch_hub()

if is_npu_available:
# Workaround for torch-npu's lack of support for creating nested tensors from NPU tensors.
#
# ```
# >>> a, b = torch.arange(3).npu(), torch.arange(5).npu() + 3
# >>> nt = torch.nested.nested_tensor([a, b], layout=torch.jagged)
# ```
# throws "not supported in npu" on Ascend NPU.
# See https://github.com/Ascend/pytorch/blob/294cdf5335439b359991cecc042957458a8d38ae/torch_npu/utils/npu_intercept.py#L109
# for details.

import torch

try:
if hasattr(torch.nested.nested_tensor, "__wrapped__"):
torch.nested.nested_tensor = torch.nested.nested_tensor.__wrapped__
if hasattr(torch.nested.as_nested_tensor, "__wrapped__"):
torch.nested.as_nested_tensor = torch.nested.as_nested_tensor.__wrapped__
except AttributeError:
pass

from .models.transformers import npu_patch as npu_patch

package_name = "transformers"
Expand Down
Loading