Eliminate multi-tensor LAMB reduction in favor of applying reduce_square_sum each tensor#6023
Closed
Eliminate multi-tensor LAMB reduction in favor of applying reduce_square_sum each tensor#6023
Conversation
added 8 commits
October 29, 2020 21:15
…lambdeterminism sync with master
wschin
suggested changes
Dec 4, 2020
Contributor
wschin
left a comment
wschin
left a comment
There was a problem hiding this comment.
Let's merge another PR which nicely fixes the randomness of Reduce in Lamb.
Contributor
Author
Agreed. |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Description: Eliminate multi-tensor LAMB reduction in favor of invoking reduce_square_sum individually all tensors. This simplifies the existing code but comes at a 1% perf reduction for BERT-L seqlen 128 bs 64 gradacc 1. It will be less for gradacc > 1, as in a more realistic training scenario.
Motivation and Context
This simplifies the existing code and makes it deterministic. It is also possible to make the existing multi-tensor LAMB reduction kernel deterministic by ordering the reduction across thread blocks. This does not reduce perf.