Skip to content

fix(accelerate/dataset): change batch handling to prevent memory errors #284

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
l45k merged 1 commit into from
Jan 3, 2026
Merged

fix(accelerate/dataset): change batch handling to prevent memory errors #284

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
7 changes: 4 additions & 3 deletions executors/accelerate/src/hypha/accelerate_executor/dataset.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -97,9 +97,10 @@ def __iter__(self): # type: ignore[no-untyped-def]
while cursor + self.batch_size <= num_new:
end = cursor + self.batch_size

# Yield a CLEAN, CONTIGUOUS copy
# This prevents sending a "View" of the whole file to the worker queue
batch = {k: v[cursor:end].contiguous() for k, v in processed.items()}
# Yield a clean, INDEPENDENT copy using .clone()
# This prevents sending a "View" of the underlying raw_bytes buffer
# which might be gc'ed when the loop iterates.
batch = {k: v[cursor:end].clone() for k, v in processed.items()}
yield batch

cursor = end
Expand Down