Skip to content

bug fix: RuntimeError when training GRPO with LoRA and PtEngine #5645

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

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

bug fix: RuntimeError when training GRPO with LoRA and PtEngine #5645

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
17f918f
bug fix: RuntimeError when training GRPO with LoRA and PtEngine
chenjianhuii Sep 3, 2025
723af9c
fix pt zero3
hjh0119 Sep 4, 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
2 changes: 1 addition & 1 deletion swift/llm/infer/infer_engine/pt_engine.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -461,7 +461,7 @@ async def _gen_wrapper():
return await queue.get()

# Ensure `template._post_encode` has no gradient.
Copy link
Contributor

Choose a reason for hiding this comment

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

medium

To improve maintainability and prevent future regressions, it would be beneficial to add a comment explaining why @torch.no_grad() is used here instead of the potentially more performant @torch.inference_mode(). This will clarify the reasoning for other developers who might not have the context of the GRPOTrainer RuntimeError.

# Use `torch.no_grad()` instead of `torch.inference_mode()` to allow model
# parameters to have `requires_grad=True` set on them later (e.g., during
# PEFT training). This prevents a RuntimeError when the model is used for
# both inference and training.

@torch.inference_mode()
@torch.no_grad()
def _infer(
self,
infer_requests: List[InferRequest],
Expand Down