Conversation
Contributor
|
Hi 👋, thank you for opening this pull request! The pull request is converted to draft by default. 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. |
ArthurZucker
approved these changes
Mar 17, 2025
shethaadit
approved these changes
Mar 17, 2025
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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?
Profiling of CPU memory usage showed that calling
gc.collect()has no effect (thedelstatement is enough and cleaning will happen automatically as it should), i.e. the peak cpu memory usage is never larger than 1 state dict (even with older.binsharded checkpoints).Removing the statement speeds up loading by 15-20% as
gc.collect()is costly and should in general not be called.As for the caching allocator, profiling shows that having an
allocation_factor > 1does not help at all neither for TP nor usual loading on multiple GPUs. It is then much easier to always use 1, as we will never encounter cases of blowing GPUs. Also, code was not allocating anything in case allocation size was larger than gpu size, which could lead to big slowdown if reaching this state only because of the factor 2, as the model may still fit.Also, improved granularity of the allocator by checking each param dtype, which may be different with composite models/keep_in_fp_32_modules.