[autoparallel] added bias comm spec to matmul strategy #1664
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Outdated suggestions cannot be applied.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.
Problem
The previous PR #1616 considers all matmul sharding strategy generation, however, it failed to consider the the communication action for the bias term.
Solution
This bias term can only have one action which is
IDENTITY_FWD_ALLREDUCE_BWD
to make sure the gradient is synchronized. The mesh dimensions for all reduce depends on the sharding spec of the output. For example:This PR implemented these strategies for the bias term.
Note
The
bias
term in ops liketorch.addmm
may not be 1D tensor, I will make the strategy broadcastable in a separate PR.