[refactor] condense group offloading #11990
Open
+74
−102
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
|
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. |
a-r-r-o-w
requested review from
sayakpaul and
DN6
and removed request for
sayakpaul
sayakpaul
approved these changes
Aug 6, 2025
There was a problem hiding this comment.
Looks super clean!
I just ran some benchmarks to confirm if the changes don't have any detrimental effect on the speed-memory trade-off (code) and they look alright to me.
Would maybe also run all the group offloading tests on the GPU, too.
| finally: | ||
| pinned_dict = None | ||
|
|
||
| def _transfer_tensor_to_device(self, tensor, source_tensor, current_stream=None): |
|
|
||
| if self.offload_to_disk_path: |
| with context: | ||
| # Load to CPU (if using streams) or directly to target device, pin, and async copy to device | ||
| device = self.onload_device if self.stream is None else "cpu" | ||
| loaded_tensors = safetensors.torch.load_file(self.safetensors_file_path, device=device) |
There was a problem hiding this comment.
device when supplied as torch.device("cuda") to the onload_device argument, it would fail here from safetensors complaining invalid device cuda. Simply wrapping up it within str() solves the issue.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
The current implementation, after many of the recent updates, is hard to understand or debug/reason through. There is also one code path that seems completely unused (see the removed
self.offload_to_disk_pathrelated changes and the call intoself._onload_from_disk).This PR tries to refactor and clean up some of the implementation so that implementing new changes is easier in the future.
Related to comment which I'm trying to debug through