DenisovAV/flutter_gemma

<uses-native-library> missing in Android LLM Inference Guide

Closed this issue · 4 comments

ahndwon commented

I was not able to run the example after I added the model.bin to my device.

I followed the details on the fix and added
in AndroidManifest.xml.

And then I was able to run the app and use gemma.

It should be mentioned in the Readme.md and added to the example.

DenisovAV commented

Thank you for your comment! Interestingly, on my device, everything works without this fix. I'll try to reproduce it too, could you tell, which device and which Android version you have, and with which type of build you reproduce it?

ahndwon commented

Okay below are my device details.

Google Pixel 7
Android 14
BuildNumber : AP1A.240405.002
model : Tensorflow lite gemma-2b-it-gpu-int4

I was able to reproduce by removing tags.
iOS and web works fine.

DenisovAV commented

I reproduced it. Yes, you're right, this needs to be added to the manifest if you're using a gpu model, for cpu, it's not necessary, as everything works without it. I tried with cpu only before.

Will add this to the manual today, thank you

  • 👍1
DenisovAV commented

Done, version 0.1.4

  • 👍2