Garbage output on Intel Mac

[arun-uride]

Hi, I'm getting garbage outputs via Open Web UI, llama.cpp and the terminal. Is this due to lack of support of Intel Macs?

[khosravipasha]

Thanks for marking this, we are aware of the issue now on x86 CPUs, will be pushing a fix today.
The AVX, simd implementation has issues, for the moment will make it fallback to generic implemenations.

ggml-org/llama.cpp@master...PrismML-Eng:llama.cpp:q1-cpu

[khosravipasha]

PR here: PrismML-Eng/llama.cpp#8
will merge after a bit more testing

[lewis-rescuing-lovable]

I can test on an i9 intel macbook if it will help

[khosravipasha]

@lewis-rescuing-lovable That would be great thanks. The fix for x86 should be in prism branch now.
There is few PRs for faster version with AVX as well if you want to check them out.
Currently x86 is falling back to generic implemenation which is slow but now should be correct

https://github.com/PrismML-Eng/llama.cpp/pulls

[khosravipasha]

I will close this, feel free to reopn if there is still any issues.