[Kernel] Add flash-attn back#4907
Merged
WoosukKwon merged 6 commits intomainfrom May 20, 2024
Merged
Conversation
Yard1
approved these changes
May 19, 2024
1 task
dtrifiro
pushed a commit
to dtrifiro/vllm
that referenced
this pull request
May 21, 2024
robertgshaw2-redhat
pushed a commit
to neuralmagic/nm-vllm
that referenced
this pull request
Jun 8, 2024
robertgshaw2-redhat
pushed a commit
to neuralmagic/nm-vllm
that referenced
this pull request
Jul 14, 2024
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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 reverts #4820 by adding back
flash-attn. Previously, usingflash-attnfor decoding caused errors when using small models (like the Llama 68M model inlora/test_layer_variation.py). This was because the index calculation for paged KV cache was done inintinstead ofint64_t, leading to integer overflow whennum_blocksis large. Invllm-flash-attn==2.5.8.post2, the overflow bug was fixed.