Linear with DID loop split #3650
Open
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
|
!build |
Priya2698
requested review from
wujingyue,
samnordmann and
cowanmeg
and removed request for
cowanmeg and
samnordmann
wujingyue
approved these changes
Dec 29, 2024
Comment on lines
+127
to
+128
| self.weight = self.define_tensor([d * e, e], contiguity=[True, True]) | ||
| self.bias = self.define_tensor([d * e], contiguity=[True]) |
There was a problem hiding this comment.
Suggested change
| self.weight = self.define_tensor([d * e, e], contiguity=[True, True]) | |
| self.bias = self.define_tensor([d * e], contiguity=[True]) | |
| self.weight = self.define_tensor([d * e, e], contiguity=True) | |
| self.bias = self.define_tensor([d * e], contiguity=True) |
Also, a question for you: is contiguity=True necessary here? Is there a problem with non-contiguous inputs?
Comment on lines
+169
to
+171
| expected_out_tensor = unsharded_out_tensor.view([b, s, d, e]).permute(2, 0, 1, 3)[ | ||
| rank : rank + 1 | ||
| ] |
There was a problem hiding this comment.
Can we use shard_tensor here?
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Outdated suggestions cannot be applied.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.
This PR adds a test case demonstrating DID loop split for linear.
The test does not require any changes to the
LinearOp::evaluatemethod and works out of the box. DID split on logical domain continues to work for linear using the additional WARs present that squeeze/unsqueeze the DID dimension. Those WARs can be removed once we completely switch to representing device parallelism using allocation and loop domain.