8000 llama-server not using GPU · Issue #1826 · abetlen/llama-cpp-python · GitHub
[go: up one dir, main page]
Skip to content

llama-server not using GPU #1826

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
Open
RakshitAralimatti opened this issue Nov 9, 2024 · 5 comments
Open

llama-server not using GPU #1826

RakshitAralimatti opened this issue Nov 9, 2024 · 5 comments

Comments

@RakshitAralimatti
Copy link
RakshitAralimatti commented Nov 9, 2024 •
edited
Loading

After I install llama-cpp-python-server with cuda support and run
python3 -m llama_cpp.server --model starcoderbase-3b/starcoderbase-3b.Q4_K_M.gguf --n_gpu_layers 10
The GPU is not getting used its running on the CPU
@pepijndevos
Copy link

I'm seeing the same problem with Vulkan.

I see all layers of the model actually getting loaded on the GPU, and nvtop shows significant memory use.
But then htop shows it's using 100% CPU and only a small blip of GPU.

llama.cpp itself works fine on the same hardware.

@pepijndevos
Copy link

Wait maybe it's actually using the GPU but just insanely bottlenecked on Python CPU performance?

@chozillla
Copy link

I still have this issue! The GPU is being used when I use llama-cli

@pepijndevos
Copy link

Can you post logs? In my case the GPU was being detected and used but bottlenecked by the CPU because the python part is slow

I ended up building https://github.com/sanctuary-systems-com/llama_multiserver

@chozillla
Copy link

Nvm I used the wrong argument, make sure that ngl is not negative! ngl 99999 not ngl -99999 since ngl -1 in theory should use all cores.

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
3 participants
@pepijndevos @chozillla @RakshitAralimatti
0