Hearing at the Fluttercon in Berlin about Gemma, I became intrigued to look at this model myself. It is fab to have an own encapsulated model and optimise it for your own content, so it is to keep in mind.

I first asked Gemma a few questions in Google AI Studio and got an informative interaction. The output can be viewed here:Gemma Android Studio AI Responses

Then I installed local models and asked the same questions with these outcomes:

Gemma-tflite-gemma-2b-it-cpu-int8: Running into problems installing the apps on devices as the model is quite big. If it installs it creates the recipe but later hangs on the next questions.

Gemma-tflite-gemma-2b-it-cpu-int4: It is installling fine but the output is not producing a full recipe and further questions are not answered fully.

Gemma-tflite-gemma-2b-it-gpu-int4: It is installing fine and providing a recipe but less detailed as the AI Studio response.

So generally I like that this is available now, but I need to fine tune the use - both on the integration and it also may be worth trying to customise a model specifically, which I try at a later stage.