Android Support

[TERMSTRAANGE]

The model page on Hugging Face lists benchmark on Samsung S25 Ultra. How was it tested? Was this demo installer used or was llama.cpp fork compiled by itself?

[khosravipasha]

Was using OpecCL backend of llama.cpp, it was a older branch, recently rebased it and put it here:
https://github.com/PrismML-Eng/llama.cpp/tree/prism-android

Someone also has PR with vulkan backend: PrismML-Eng/llama.cpp#9
Seems to be working have not tested it myself.

Vulkan backend might also work.
Ran them through terminal have not tested inside an app.
Don't think the kernels are that optimized so should be able to speed them up significantly.

This demo we did not include android.