feat(sfa): Integrate fused mla_preprocess operator #4533
+131
−14
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
Replaces several discrete operations in the SFA forward pass with a single call to the fused `mla_preprocess` custom operator. This operator combines Q/K/V projection, RoPE application, and KV cache updates into one kernel. A new weight processing method is added to transform and pre-process weights into the specific layout required by the fused operator. This change aims to improve performance by reducing kernel launch overhead. Additionally, the condition for allocating RoPE caches is relaxed to support MLA in modes other than just full decode. Signed-off-by: Yizhou Liu <[email protected]>
|
👋 Hi! Thank you for contributing to the vLLM Ascend project. The following points will speed up your PR merge:
If CI fails, you can run linting and testing checks locally according Contributing and Testing. |
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Outdated suggestions cannot be applied.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.
What this PR does / why we need it?
Replaces several discrete operations in the SFA forward pass with a single call to the fused
mla_preprocesscustom operator. This operator combines Q/K/V projection, RoPE application, and KV cache updates into one kernel.A new weight processing method is added to transform and pre-process weights into the specific layout required by the fused operator. This change aims to improve performance by reducing kernel launch overhead.
Additionally, the condition for allocating RoPE caches is relaxed to support MLA in modes other than just full decode.
Does this PR introduce any user-facing change?
None
How was this patch tested?
None