[TUTORIAL] non-causal mode in fused-attention backward #5241
+49
−38
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Outdated suggestions cannot be applied.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
This makes some simple additions to add support for non-causal attention in
06-fused-attention.py
. The forward pass already uses thecausal
parameter, so this just changes the backward pass.I think having the non-causal mode supported makes the code a little easier to understand. I.e., I did not find it obvious why the backward pass always masked the diagonal blocks until I realized it was just always computing the causal version.
New contributor declaration
I am not making a trivial change, such as fixing a typo in a comment.
I have written a PR description following these
rules.
I have run
pre-commit run --from-ref origin/main --to-ref HEAD
.Select one of the following.
/test
forlit
tests/unittest
for C++ tests/python/test
for end-to-end testsThe script includes a simple test, so I set it to run the non-causal mode as well
.Select one of the following.
lit
tests.lit
tests I have added follow these best practices,including the "tests should be minimal" section. (Usually running Python code
and using the instructions it generates is not minimal.)