Skip to content

[megatron] make empty_cuda_cache default true for megatron #1731

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
erictang000 merged 1 commit into from
May 29, 2026
Merged

[megatron] make empty_cuda_cache default true for megatron #1731

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
2 changes: 1 addition & 1 deletion skyrl/train/config/config.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -188,7 +188,7 @@ class MegatronConfig(BaseConfig):
transformer_config_kwargs: Dict[str, Any] = field(
default_factory=lambda: copy.deepcopy(DEFAULT_TRANSFORMER_CONFIG_KWARGS)
)
empty_cuda_cache: Optional[bool] = None
empty_cuda_cache: Optional[bool] = True
Copy link
Copy Markdown
Contributor

@gemini-code-assist gemini-code-assist Bot May 29, 2026

Choose a reason for hiding this comment

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

medium

While setting empty_cuda_cache to True by default helps prevent Out-Of-Memory (OOM) issues during the optimization step, calling torch.cuda.empty_cache() after every forward and forward-backward pass (per mini-batch) introduces significant CUDA synchronization overhead, which can drastically degrade training throughput.

Additionally, since None is no longer the default and is functionally treated as False in the worker implementation (if self.empty_cuda_cache:), we can simplify the type annotation from Optional[bool] to bool.

Suggested change
empty_cuda_cache: Optional[bool] = True
empty_cuda_cache: bool = True

model_config_kwargs: dict = field(default_factory=dict)
dist_ckpt_optim_fully_reshardable: bool = False
freeze_moe_router: bool = False
Expand Down
Loading