-
Notifications
You must be signed in to change notification settings - Fork 3.6k
[webgpu] Support Split-K in more situations
#26806
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[webgpu] Support Split-K in more situations
#26806
Conversation
This patch adds the support of more `dim_inner` (up to 3072) for Split-K to optimize more models.
dim_inner <= 3072 for Split-KSplit-K in more situations
|
/azp run Linux QNN CI Pipeline,Win_TRT_Minimal_CUDA_Test_CI,Windows ARM64 QNN CI Pipeline,Windows GPU Doc Gen CI Pipeline |
|
Azure Pipelines successfully started running 4 pipeline(s). |
|
/azp run Linux QNN CI Pipeline,Win_TRT_Minimal_CUDA_Test_CI,Windows ARM64 QNN CI Pipeline,Windows GPU Doc Gen CI Pipeline |
|
Azure Pipelines successfully started running 4 pipeline(s). |
Head branch was pushed to by a user without write access
|
Hi @guschmue, sorry about that! I've fixed this failure, PTAL, thanks! |
|
/azp run Linux QNN CI Pipeline,Win_TRT_Minimal_CUDA_Test_CI,Windows ARM64 QNN CI Pipeline,Windows GPU Doc Gen CI Pipeline |
|
Azure Pipelines successfully started running 4 pipeline(s). |


Description
This patch supports more
dim_inner(up to 4096) forSplit-Kto optimize more models.This patch also enables
Split-Kongen-12lp.Motivation and Context
With this PR we can achieve about 30% improvement on
jina-clip-v1-text-fp16and 20% improvement onjina-embeddings-v2-base-code-fp16on Lunar Lake iGPUs.