Add raise_last_usage memory optimization pass to Inductor #125559
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Outdated suggestions cannot be applied.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 adds a new Inductor memory optimization pass to raise the consumer nodes if moving them up can free more memory (i.e. if its output tensor is smaller than the sum memory of all its last-usage input tensors).
Particularly, for each node in the graph, we raise its consumer node to be right after the current node, if it satisfies the following conditions:
This pass is particularly important for significantly reducing peak memory usage for Compiled FSDP2 on llama-7b model (decreasing mem usage by more than 57GB).
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @peterbell10 @ipiszy @chenyang78 @kadeng @muchulee8 @ColinPeppler @amjames @desertfire @chauhang