Skip to content

[training_utils] fix: enforce 1D object array shape for non-tensor data in collate_fn #2741

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 7 commits into from
Jul 30, 2025
Merged

[training_utils] fix: enforce 1D object array shape for non-tensor data in collate_fn #2741

Changes from 1 commit
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
b969ee2
fix: enforce 1D object array shape for non-tensor data in collate_fn
kibitzing Jul 24, 2025
c2a26b2
refactor: replace object array construction with np.fromiter in colla…
kibitzing Jul 24, 2025
a8cbac5
test: add unit test for collate_fn in rl_dataset
kibitzing Jul 25, 2025
ce33651
chore: apply formatting fixes from pre-commit
kibitzing Jul 28, 2025
1108504
fix: remove tolist() usage where np.array was previously assumed
kibitzing Jul 28, 2025
afdd48f
chore: apply formatting fixes from pre-commit
kibitzing Jul 28, 2025
3f42222
refactor: replace type branching with list(obj) and add type guard
kibitzing Jul 29, 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
3 changes: 2 additions & 1 deletion verl/utils/dataset/rl_dataset.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -60,7 +60,8 @@ def collate_fn(data_list: list[dict]) -> dict:
tensors[key] = torch.stack(val, dim=0)

for key, val in non_tensors.items():
non_tensors[key] = np.array(val, dtype=object)
non_tensors[key] = np.empty(len(val), dtype=object)
non_tensors[key][:] = val
Copy link
Contributor

@gemini-code-assist gemini-code-assist bot Jul 24, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

high

The current implementation uses np.empty followed by assignment. A more concise and potentially performant approach is to use np.fromiter to create the 1D object array directly, as collate_fn is on a critical performance path for data loading.

non_tensors[key] = np.fromiter(val, dtype=object, count=len(val))

Copy link
Contributor Author

@kibitzing kibitzing Jul 24, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for the suggestion!
I've updated the code to use np.fromiter(val, dtype=object, count=len(val)) as recommended.


return {**tensors, **non_tensors}

Expand Down