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.

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

Clip model from LLaVA is not using GPU, but LLaVA model is using GPU in the same piece of code. #650

Closed
dcostea opened this issue Apr 3, 2024 · 4 comments
Labels
duplicate This issue or pull request already exists

Comments

@dcostea
Copy link

dcostea commented Apr 3, 2024

Congratulations for adding LLaVA support starting with version 0.11.1.
It works very well!

Still, clip model is very slow.
I have Cuda12 and llava-mistral model is using GPU, as expected, but the clip model is not, instead is using the CPU.

image

image

NativeLibraryConfig.Instance
.WithCuda(true)
.WithLogs(true)
.WithAvx(NativeLibraryConfig.AvxLevel.Avx512);

Should the clip model run on GPU as well?

@AsakusaRinne
Copy link
Collaborator

Thank you for reporting us for this problem! It's the same with #639, which is supposed to be fixed by #647. We will publish a patch release 0.11.2 soon.

@AsakusaRinne AsakusaRinne added bug Something isn't working llava backend duplicate This issue or pull request already exists and removed bug Something isn't working llava backend labels Apr 3, 2024
@AsakusaRinne
Copy link
Collaborator

v0.11.2 has been released in #651, please take a try to see if this issue has been fixed. Thank you!

@dcostea
Copy link
Author

dcostea commented Apr 6, 2024

wow, look here, less than half of a second!
image
isn't it beautiful?

Thank you!

@dcostea dcostea closed this as completed Apr 6, 2024
@dcostea
Copy link
Author

dcostea commented Apr 6, 2024

issue is fixed!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
duplicate This issue or pull request already exists
Projects
None yet
Development

No branches or pull requests

2 participants