Gemma 4 E2B on Mobile: Android vs iOS Reality (1)

#mobile #ios #android #ai

I'm still looking for a new job. In the meantime, I'm running experiments with the new Gemma 4 E2B model...

I just wanted to share a few notes on this...

The model uses the .litertlm extension, as it now relies on LiteRT-LM, the new framework to run these models in edge devices. It's an evolution of TensorFlow Lite.
On Android, there is a Kotlin-based API to run these models on Android devices optimized for coroutines.
On iOS, there is currently no native Swift API to run it, so you have to rely on the C++ API.

So, one of my experiments consists of evaluating whether it's possible to run this model in a device without a dedicated IA chipset.