Skip to content

Distributed load padding correction #439

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
oleksost merged 1 commit into from
Jan 7, 2026
+3 −1
Merged

Distributed load padding correction #439

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
4 changes: 3 additions & 1 deletion fast_llm/engine/multi_stage/fsdp.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -270,7 +270,9 @@ def reset_shard_pad(self, shard: torch.Tensor, shard_name: str) -> int:
# Also ensures a correct parameter count in loading context.
shard_meta = self._weight_shard_meta if shard_name == ShardName.weights else self._grad_shard_meta
shard_meta.validate(shard)
if self._shard_pad > 0:
# Only count padding for non-empty shards. Frozen FSDPs have empty optimizer shards
# (numel()==0) but non-zero shard_pad, which would incorrectly inflate the count.
if self._shard_pad > 0 and shard.numel() > 0:
shard[-self._shard_pad :].zero_()
return self._shard_pad
return 0
Expand Down