Skip to content

Faster Q3_K implementation on Metal #2307

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.

Sign up for GitHub

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

Jump to bottom
Merged
merged 4 commits into from
Jul 21, 2023
Merged

Faster Q3_K implementation on Metal #2307

merged 4 commits into from
Jul 21, 2023

Conversation

ikawrakow
Copy link
Contributor

Similar to #2290 but for Q3_K, which is notoriously difficult to implement efficiently.

The table gives TG-128 time for Q3_K_S in ms/t on M2 Max with 30-core GPU:

Model Master This PR Speedup
7B 28.1 22.0 27.7%
13B 50.3 37.5 34.1%
33B 121.3 88.8 36.6%
65B 230.3 167.8 37.2%

@ikawrakow ikawrakow requested a review from ggerganov July 21, 2023 07:56
ggerganov
ggerganov approved these changes Jul 21, 2023
View reviewed changes
Copy link
Member

@ggerganov ggerganov left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

M1 Pro

Model Master This PR
7B 43.6 35.4
13B 78.6 62.9

@ikawrakow ikawrakow merged commit 4d76a5f into master Jul 21, 2023
@ikawrakow ikawrakow deleted the ik/metal_faster_q3k branch July 21, 2023 14:05
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@ggerganov ggerganov ggerganov approved these changes

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
3 participants
@ikawrakow @ggerganov @Kawrakow