[simple_fsdp] Support SimpleFSDP DSv3 + EP full/selective activation checkpointing.#2406
Open
IvanKobzarev wants to merge 1 commit intomainfrom
Open
[simple_fsdp] Support SimpleFSDP DSv3 + EP full/selective activation checkpointing.#2406IvanKobzarev wants to merge 1 commit intomainfrom
IvanKobzarev wants to merge 1 commit intomainfrom
Conversation
meta-cla
bot
added
the
CLA Signed
Contributor
|
cc @yiming0416 |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Before that we required SimpleFSDP DSv3 to not use activation_checkpointing.
It resulted in illegal memory acceses because of differences in recomputed num tokens per expert in fw and bw.
Enabling them via must_save for all_to_all (for full we set it via graph_pass)
For selective it is added explicitly.
Setting
torch._dynamo.config.skip_fwd_side_effects_in_bwd_under_checkpoint = Trueto bypass mutation checks from dynamo (currently in MoE impl there is a field mutation)