Refactor attention kernels #53
Merged
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
zhuohan123
reviewed
May 2, 2023
| @@ -0,0 +1,5 @@ | |||
| #pragma once | |||
There was a problem hiding this comment.
Let's use the define guard instead of #pragma once per Google's C++ style guide :)
There was a problem hiding this comment.
Either options have pros and cons. I think it's safe to use #pragma once, because it is commonly used in DL projects such as PyTorch and FasterTransformer.
|
Performance (batch_size=8, context_len=512, num_heads=40, head_size=128): There's slight improvement in the kernel performance due to the use of fp16 in |
This was referenced May 4, 2023
hongxiayang
pushed a commit
to hongxiayang/vllm
that referenced
this pull request
Feb 13, 2024
dtrifiro
pushed a commit
to dtrifiro/vllm
that referenced
this pull request
Jun 18, 2024
…rovements dockerfile improvements
yukavio
pushed a commit
to yukavio/vllm
that referenced
this pull request
Jul 3, 2024
dllehr-amd
pushed a commit
to dllehr-amd/vllm
that referenced
this pull request
Jul 22, 2024
…kar-amd-patch-1 Revert "Revert "Tune fused_moe_kernel for TP 1,2,4,8 and bf16 and fp16, updated moe kern…""
Closed
wuhuikx
pushed a commit
to wuhuikx/vllm
that referenced
this pull request
Mar 27, 2025
### What this PR does / why we need it? 1. Add vllm-ascend tutorial doc for Qwen/Qwen2.5-7B-Instruct model serving doc 2. fix format of files in `docs` dir, e.g. format tables, add underline for links, add line feed... ### Does this PR introduce _any_ user-facing change? <!-- Note that it means *any* user-facing change including all aspects such as API, interface or other behavior changes. Documentation-only updates are not considered user-facing changes. --> no. ### How was this patch tested? doc CI passed --------- Signed-off-by: Shanshan Shen <87969357+shen-shanshan@users.noreply.github.com>
1 task
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 PR refactors attention kernels, making the helper functions more modular and pruning unused code. This PR will make it easier to add support for a new data type such as bfloat16.
In addition, this PR reduces the computation overhead of the attention kernel, by using the reduced precision (i.e., fp16) for
logits * Vinstead of the full precision. This is compatible with the FasterTransformer's implementation.