Skip to content

Conversation

@mmerecki
Copy link

Reduce the number of fa rows for Intel to reduce registers usage.

@mmerecki mmerecki changed the title Reduce the number of fa rows for Intel [WIP] Reduce the number of fa rows for Intel Dec 17, 2025
@jeffbolznv
Copy link
Collaborator

Should this depend on head size? Some models have small head sizes like 64 or even 40, 2 rows seems pretty small for that. But if 2 is best, I don't object.

@github-actions github-actions bot added Vulkan Issues specific to the Vulkan backend ggml changes relating to the ggml tensor library for machine learning labels Dec 17, 2025
@mmerecki
Copy link
Author

Thanks Jeff. I will verify this change with more models and potentially update the value for small head sizes.
I will also add information about the test results before I make the PR ready for review.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

ggml changes relating to the ggml tensor library for machine learning Vulkan Issues specific to the Vulkan backend

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants