Improve aggressive factor propagation strategy in Shardy. There are two main differences from BasicFactorPropagation.
#28
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
![copybara-service[bot]](https://avatars.githubusercontent.com/in/44061?s=60&v=4){kind=small}
copybara-service
bot
force-pushed
the
test_655675663
branch
from
c7c5d33
to
7a3f6c1
Compare
…wo main differences from `BasicFactorPropagation`.
### Difference 1
`BasicFactorPropagation` propagates the same sharding axes to all the tensors
along a factor. This strategy can propagate different sharding axes to
different tensors. For example, Tensors T0, T1, T2 contains Factor F0. T0/F0
is already sharded along ["a", "b"], and "b" is already used by T2 ("b" can be
explicitly replicated, or it is used to shard another factor).
`BasicFactorPropagation` propagates ["a"] to both T1/F0 and T2/F0, while this
strategy propagates ["a", "b"] to T1/F0 and ["a"] to T2/F0, respectively.
### Difference 2
`BasicFactorPropagation` is conservative in terms of conflicts across
factors. The overlapped axis between factors cannot be propagated. This
strategy is more aggressive by allowing the overlapped axis being propagated
along different factors if there is no overlapped axis in the result
shardings.
PiperOrigin-RevId: 657641564
copybara-service
bot
force-pushed
the
test_655675663
branch
from
7a3f6c1
to
2eeee20
Compare
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Outdated suggestions cannot be applied.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.
Improve aggressive factor propagation strategy in Shardy. There are two main differences from
BasicFactorPropagation.Difference 1
BasicFactorPropagationpropagates the same sharding axes to all the tensorsalong a factor. This strategy can propagate different sharding axes to
different tensors. For example, Tensors T0, T1, T2 contains Factor F0. T0/F0
is already sharded along ["a", "b"], and "b" is already used by T2 ("b" can be
explicitly replicated, or it is used to shard another factor).
BasicFactorPropagationpropagates ["a"] to both T1/F0 and T2/F0, while thisstrategy propagates ["a", "b"] to T1/F0 and ["a"] to T2/F0, respectively.
Difference 2
BasicFactorPropagationis conservative in terms of conflicts acrossfactors. The overlapped axis between factors cannot be propagated. This
strategy is more aggressive by allowing the overlapped axis being propagated
along different factors if there is no overlapped axis in the result
shardings.