Simplify the cache decoding graph #780
Merged
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
mattdangerw
commented
Feb 25, 2023
mattdangerw
force-pushed
the
simplify-cache
branch
from
8cd63c6 to
ced9248
Compare
mattdangerw
commented
Feb 25, 2023
chenmoneygithub
suggested changes
Feb 28, 2023
mattdangerw
force-pushed
the
simplify-cache
branch
from
ced9248 to
c554b8f
Compare
|
Thanks for the review! Addressed all comments. |
mattdangerw
force-pushed
the
simplify-cache
branch
from
c554b8f to
c2dd044
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Outdated suggestions cannot be applied.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.
This updates our
CachedMultiHeadAttentionlayer to avoid slicing the input to a dynamic length during generative decoding. Instead, our cache is always a fixed length, as are the keys and values after the cache is applied.Overall this ends up being both a nice simplification and speedup. After patching in a fix for #779, generating 25 sequences of length 256 with the base model goes from 52s -> 33s.