Initial implemention of disaggregated attention and qkv projection #1433
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
* Add scripts for evaluation * Add absolute request rate value * Fix script for target arrival rate * Fix cpp req rate benchmark * update to use new dataset * Fix infinite loop * update * add data --------- Co-authored-by: Remi Delacourt <[email protected]> Co-authored-by: Gabriele Oliaro <[email protected]>
yingchen21
force-pushed
the
attn-qkv-proj
branch
from
07d8e67 to
94e1563
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Outdated suggestions cannot be applied.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.
Description of changes:
This PR moves the qkv projection (and output projection) from the
attentionoperator to a seperatedenselayer, so we can apply LORA to the projections of attention.The qkv projection (and output projection) is removed from the attention operator, and done by dense instead. The attention operator now no longer has tensor weights. It takes in qkv projection and output the raw attention result (without applying output projection) with dimension vdim * num_kv_heads instead of the hidden demension.
Related Issues:
Linked Issues:
Issues closed by this PR:
This change is