Skip to content

Add explicit async_fn_kwargs. #204

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 4 commits into from
Oct 23, 2025
Merged

Add explicit async_fn_kwargs. #204

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
0dab279
Add explicit async_fn_kwargs.
herman-ai Oct 21, 2025
d0fff56
Fix linting errors.
herman-ai Oct 21, 2025
10ba9d9
Add defensive check for kwargs being None.
herman-ai Oct 21, 2025
076f2d0
Merge branch 'main' into herman/nvbugs_5596118
herman-ai Oct 23, 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 src/nvidia_resiliency_ext/checkpointing/async_ckpt/torch_ckpt.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,7 +49,7 @@ def async_save(self, state_dict, *args, **kwargs):
preloaded_sd = preload_tensors(state_dict)
torch.cuda.synchronize()
async_request = AsyncRequest(
TorchAsyncCheckpoint.async_fn, (preloaded_sd, *args), [], kwargs
TorchAsyncCheckpoint.async_fn, (preloaded_sd, *args), [], kwargs or {}
)
self._async_calls_queue.schedule_async_request(async_request)

Expand Down
2 changes: 1 addition & 1 deletion src/nvidia_resiliency_ext/checkpointing/local/ckpt_managers/base_manager.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -307,7 +307,7 @@ def finalize_fn():
# we must wait for D2H to complete before returning control to the training
with debug_time("ckpt_D2H_synchronize", logger):
torch.cuda.synchronize()
return AsyncRequest(save_fn, save_args, [finalize_fn])
return AsyncRequest(save_fn, save_args, [finalize_fn], async_fn_kwargs={})

assert not is_async
Copy link
Contributor

@aartibasant aartibasant Oct 21, 2025

Choose a reason for hiding this comment

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

Can you check if this path does not trigger the error: https://github.com/NVIDIA/nvidia-resiliency-ext/blob/main/src/nvidia_resiliency_ext/checkpointing/async_ckpt/torch_ckpt.py#L51

Copy link
Contributor Author

@herman-ai herman-ai Oct 21, 2025

Choose a reason for hiding this comment

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

kwargs inside the function def async_save(self, state_dict, *args, **kwargs): should be {} if nothing is passed to async_save so I don't think this should cause any error.

save_fn(*save_args)
Expand Down
4 changes: 3 additions & 1 deletion tests/checkpointing/unit/test_async_writer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -65,7 +65,9 @@ def finalize_fn():
"""Finalizes async checkpointing and synchronizes processes."""
save_state_dict_async_finalize(*save_state_dict_ret)

return AsyncRequest(save_fn, save_args, [finalize_fn], preload_fn=preload_fn)
return AsyncRequest(
save_fn, save_args, [finalize_fn], preload_fn=preload_fn, async_fn_kwargs={}
)

def async_save_checkpoint(
self,
Expand Down
4 changes: 3 additions & 1 deletion tests/checkpointing/unit/test_async_writer_msc.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,7 +46,9 @@ def finalize_fn():
save_state_dict_async_finalize(*save_state_dict_ret)
torch.distributed.barrier()

return AsyncRequest(save_fn, save_args, [finalize_fn], preload_fn=preload_fn)
return AsyncRequest(
save_fn, save_args, [finalize_fn], preload_fn=preload_fn, async_fn_kwargs={}
)

def async_save_checkpoint(
self, checkpoint_dir, state_dict, planner, async_queue, thread_count=1
Expand Down