You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi, I have been testing 2 wrappers for llama.cpp. One is from you with unity bindings and one is from nerve sparks Though both of you are using different commit of llama.cpp but as far as ggml_vec_dot_q4_K_q8_K function is concerned, I do not see any difference between your commit of llama. cpp and their commit
But when i run their kotlin based android app, it is a bit slow in response i.e. it takes 4 to 5 seconds in response or more but it does not crash. But when I try to run your library for unity, it crashes for certain cpu types. I'm just wondering why it is happening and is there a way we can fix it?
Hi @amakropoulos we spent more time on it and as per our undersranding, your library is optimized for cpu having support for dot product i.e. asimddp support and that is why it just crashes for the mobile cpu's those does not have this support. So for now, we are just not using your library for unsupported mobile cpus.
Describe the bug
Hi, I have been testing 2 wrappers for llama.cpp. One is from you with unity bindings and one is from nerve sparks Though both of you are using different commit of llama.cpp but as far as ggml_vec_dot_q4_K_q8_K function is concerned, I do not see any difference between your commit of llama. cpp and their commit
But when i run their kotlin based android app, it is a bit slow in response i.e. it takes 4 to 5 seconds in response or more but it does not crash. But when I try to run your library for unity, it crashes for certain cpu types. I'm just wondering why it is happening and is there a way we can fix it?
Here are the crash logs
and here I am testing it on GOOGLE PIXEL 3 device.
Steps to reproduce
Just make a simple unity build with Llama 3.2 1B model (i think it will fail with other models too) and it happens as soon as app is launched
LLMUnity version
Unity 2021.3.14f1
Operating System
Android
The text was updated successfully, but these errors were encountered: