WIP: causal prefix mask with adjusted tests #2
Conversation
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.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.
DO NOT LAND ON MAIN BRANCH
Implementation of causal prefix mask for cross attention (see Dao-AILab#20 (comment) for more info).
The original FlashAttention tests have been partially adjusted to take into account the new causal prefix masking scheme. The modified tests should correctly test that the output of
flash_attn_unpadded_*_func,out, is correct, and the that gradientsdq,dk, anddvare correct.It has not been adjusted to properly test the output
S_dmask(contains information about the attention values and dropout) because doing so requires figuring out the format ofS_dmask(which is a non-standard format, seeconvert_flash_attn_S_to_softmaxin the test file). This means that we cannot be sure about (1) whether the returned attention values are correct, and (2) whether causal prefix masking works with dropout. My guess is it does, assuming one can figure out how the data is formatted, but it hasn't been proven.There may also be an effect on performance. I've seen the backward pass maybe taking longer... but hard to say.