Skip to content
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

[Question]Computation and register/shared memory wasted during decoding phase? #1224

Open
sleepwalker2017 opened this issue Sep 13, 2024 · 0 comments
Open

[Question]Computation and register/shared memory wasted during decoding phase? #1224

sleepwalker2017 opened this issue Sep 13, 2024 · 0 comments

Comments

@sleepwalker2017
Copy link

sleepwalker2017 commented Sep 13, 2024 •
edited
Loading

I notice the kBlockM is set for 64 or 128, which is fine for the prefill phase.

For decoding phase, the query length is 1, is 64 or 128 too large?

If kBlockM is set to 64, it means there are 63/64 computations are useless, and 63/64 shared memory and registers for query are wasted.

why not set the kBlockM smaller? ie 16 or 8, then we can waste less resources and enable more parallelism.

Is that the fact? Thank you!
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
1 participant
@sleepwalker2017