Follow up for Group Offload to Disk #11760
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
DN6
requested review from
a-r-r-o-w and
sayakpaul
and removed request for
a-r-r-o-w
|
btw @sayakpaul This line Uses the memory address of the group object to create the filename, but that can change across sessions. So you could have the offload directory growing quite substantially if running an inference script multiple times. Might need to come up with another way to name the files so that new ones aren't created across sessions. |
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
sayakpaul
approved these changes
Jun 20, 2025
|
|
||
| @torch.compiler.disable() | ||
| def onload_(self): | ||
| r"""Onloads the group of modules to the onload_device.""" |
There was a problem hiding this comment.
Any reason for it to et out? 👁️
| for tensor_obj in self.tensor_to_key.keys(): | ||
| tensor_obj.data = torch.empty_like(tensor_obj.data, device=self.offload_device) | ||
|
|
||
| @torch.compiler.disable() |
There was a problem hiding this comment.
I guess we don't need it as it's being called from offload_() which is already decorated with it.
a-r-r-o-w
approved these changes
Jun 25, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What does this PR do?
Since we now have multiple offload options available in group offload (standard, low_cpu_mem_usage, offload_to_disk), it would make sense to create separate utility functions for them and try to reuse common patterns e.g. try to use
cpu_param_dictand the pinned_memory context manager with streams.This PR is just to organize the module a bit so that it's a bit easier to add further functionality, e.g. #11758 without having a lot of conditional paths.
Fixes # (issue)
Before submitting
documentation guidelines, and
here are tips on formatting docstrings.
Who can review?
Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.