Delete activation memory buffers proactively in UPipe forward pass#4
Merged
Conversation
ghost
force-pushed
the
fwd_pass_mem_leaks
branch
from
db3ff00 to
5ad64ee
Compare
ghost
changed the title
mryab
approved these changes
Apr 29, 2026
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
2 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
This PR removes potential memory leaks in the forward pass of UPipe attention implementation. While it does not affect the overall memory usage because the peak memory is dictated by the backward pass, it is still useful to proactively clean up unused buffers in the forward pass.
In particular, it is useful for inference scenarios such as RL rollouts.
The memory profile for Llama3-8B on a single 8xH100, when training on 128K context length is attached BEFORE and AFTER the cleanup. It can be seen that PyTorch deallocates buffers only when the same buffers are reused in the next stage, which leads to unnecessary memory pressure during the current stage.
Before cleanup:
After cleanup:
