[offload] respect max_memory argument when factoring in unused reserved memory
#37982
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
Contributor
|
Hi 👋, thank you for opening this pull request! The pull request is converted to draft by default. The CI will be paused while the PR is in draft mode. When it is ready for review, please click the |
|
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. |
ydshieh
approved these changes
May 6, 2025
Collaborator
ydshieh
left a comment
ydshieh
left a comment
There was a problem hiding this comment.
Looks good to me although not diving into the details 💯 !
gante
changed the title
max_memory argumentmax_memory argument when factoring in unused reserved memory
Cyrilvallez
approved these changes
May 6, 2025
Member
Cyrilvallez
left a comment
Cyrilvallez
left a comment
There was a problem hiding this comment.
Ha, this got through the radar indeed! Thanks! Though I believe we can simply check unused_memory when max_memory is set, to keep the exact same behavior
Comment on lines
-1298
to
+1301
| max_memory[device_name] += unused_memory | ||
| device_map_kwargs["max_memory"] = max_memory | ||
| inferred_max_memory[device_name] += unused_memory | ||
| # respect the `max_memory` passed by the user | ||
| if max_memory is not None and device_name in max_memory: | ||
| inferred_max_memory[device_name] = min(inferred_max_memory[device_name], max_memory[device_name]) |
Member
There was a problem hiding this comment.
Probably easier to check unused_memory only if max_memory is not None and device_name in max_memory no?
zucchini-nlp
pushed a commit
to zucchini-nlp/transformers
that referenced
this pull request
May 14, 2025
BenjaminBossan
added a commit
to BenjaminBossan/peft
that referenced
this pull request
Jun 25, 2025
As discussed internally, since huggingface/transformers#37982, some multi-GPU tests started failing because all parameters are loaded onto a single GPU. This should now be fixed by providing an explicit device_map instead of relying on "auto". Furthermore, for an unknown reason, the HQQ test started failing as the correlation dipped below 0.97 -- to 0.9696 actually. I think this is close enough to not warrant further investigation. Therefore, I only decreased the threshold.
BenjaminBossan
added a commit
to huggingface/peft
that referenced
this pull request
Jul 2, 2025
As discussed internally, since huggingface/transformers#37982, some multi-GPU tests started failing because all parameters are loaded onto a single GPU. This should now be fixed by providing an explicit device_map instead of relying on "auto". Furthermore, for an unknown reason, the HQQ test started failing as the correlation dipped below 0.97 -- to 0.9696 actually. I think this is close enough to not warrant further investigation. Therefore, I only decreased the threshold.
efraimdahl
pushed a commit
to efraimdahl/peft
that referenced
this pull request
Jul 12, 2025
As discussed internally, since huggingface/transformers#37982, some multi-GPU tests started failing because all parameters are loaded onto a single GPU. This should now be fixed by providing an explicit device_map instead of relying on "auto". Furthermore, for an unknown reason, the HQQ test started failing as the correlation dipped below 0.97 -- to 0.9696 actually. I think this is close enough to not warrant further investigation. Therefore, I only decreased the threshold.
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?
Our daily CI had >1000 new failures since #37920, on tests that use the
max_memoryargument infrom_pretrained. This is because we now recapture unused reserved GPU memory, which caused us to go beyond the user-definedmax_memory. This PR fixes it.Affected tests, which set
max_memory:test_cpu_offloadtest_disk_offload_bintest_model_parallelism