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

llama_cpp_python server > 0.2.79 breaks the vulkan image #742

Open
lstocchi opened this issue Aug 8, 2024 · 0 comments
Open

llama_cpp_python server > 0.2.79 breaks the vulkan image #742

lstocchi opened this issue Aug 8, 2024 · 0 comments

Comments

@lstocchi
Copy link
Collaborator

lstocchi commented Aug 8, 2024

When you build the vulkan image using llama_cpp_python 0.2.79 you see that it is actually able to detect and use gpu bc in the logs you can find

ggml_vulkan: Found 1 Vulkan devices:
Vulkan0: Virtio-GPU Venus (Apple M2 Pro) (venus) | uma: 1 | fp16: 1 | warp size: 32
llm_load_tensors: ggml ctx size =  0.30 MiB
warning: failed to mlock 73732096-byte buffer (after previously locking 0 bytes): Cannot allocate memory
Try increasing RLIMIT_MEMLOCK ('ulimit -l' as root).
llm_load_tensors: offloading 32 repeating layers to GPU
llm_load_tensors: offloading non-repeating layers to GPU
llm_load_tensors: offloaded 33/33 layers to GPU
llm_load_tensors:    CPU buffer size =  70.31 MiB
llm_load_tensors:  Vulkan0 buffer size = 4095.05 MiB
.................................................................................................

However starting from 0.2.80+ there is something broken and the gpu detection/usage is completely skipped. In the logs you just find

...
llm_load_tensors:    CPU buffer size = 4165.37 MiB
...

I also tested with the latest version 0.2.87 and it is still broken. Now we're using 0.2.85 -> https://github.com/containers/ai-lab-recipes/blob/main/model_servers/llamacpp_python/src/requirements.txt#L1

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant