Llama-cpp-python integration in android project made with briefcase

[Dodo509]

Hello everyone,

I’m working on an Android application written in Python and packaged with Briefcase that uses the llama-cpp-python library for AI model inference. I’ve already compiled the necessary native libraries (e.g. libllama.so) and placed them in the jniLibs directory, but at runtime I still get an error saying that the native library “libllama.so” can’t be loaded—even though I updated my build.gradle to include it .

I would be very grateful to anyone who can help me resolve this issue.

Thank you in advance to everyone who takes the time to assist

[freakboy3742]

If you're looking to use llama-cpp-python as a Python module, it's not a matter of just copying the .so files; the project needs to be packaged as a wheel because the location of the .so files is significant. Once you've got a wheel, you don't need to do any manipulation of the gradle files; you can just add the Python dependency to your Briefcase requires definition.

Instructions for building Python wheels for Android can be found here; this will hopefully become a lot easier in the near future with a modification to cibuildwheel

[Dodo509]

Thank you for taking the time to respond to me. I'm struggling to build a wheel because I'm on Windows, and I saw that it requires Linux. I was wondering if there is another solution or a prebuilt wheel that could help me.

[freakboy3742]

There isn't a pre-build wheel; as noted in the Android platform guide, the list of available wheels, and requests for new wheels, can be submitted through the same Chaquopy repository that I linked previously.