Fix ZeRO 2 + Pipelining #677
Merged
Conversation
This file contains 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
leogao2
requested review from
arashashari,
awan-10,
cli99,
conglongli,
eltonzheng,
jeffra,
minjiaz,
niumanar,
RezaYazdaniAminabadi,
samyam,
ShadenSmith and
tjruwase
as code owners
|
This is great, thank you! |
ShadenSmith
approved these changes
Jan 20, 2021
|
Hi @leogao2 , looks like our auto-formatter wants to make some changes. Can you setup pre-commit with updated formatting? Thanks! |
|
Merged, thanks @leogao2 ! |
Closed
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
ZeRO 2 + Pipelining is broken. After hunting down the issue here (EleutherAI/gpt-neox#62), I've made a patch that seems to solve the issue but might create bad inefficiencies elsewhere.
Since this class turns
enable_backward_allreduceoff,self.overlapping_partition_gradients_reduce_epilogue()defined in theDeepSpeedEngineclass never actually runs. I suspect this is because of efficiency problems;get_flat_partitionin stage2.py might do something expensive memorywise (a quick glance at the code certainly gives off that impression); someone will have to look into that later. But in the meantime, this fixes ZeRO2 + Pipelining enough to run a demo.