This is the example code for running a powerful (as much as a 2019 phone can handle) language model on my Google Pixel 4 using the languagemodels python module. You can find the module on GitHub.

I run this using Termux on Ubuntu version 22.04 from Andronix.
I used the 1 and 2 GB RAM models for this project as the phone only has 6GB of available memory and the next highest model is 4GB which did work, but it was slow and eventually crashed termux after finishing.

I don't believe you need to have a rooted device for any of this, but I will say my phone is running rooted Pixel OS.

To run this you just need to 'pip install languagemodels' and then run either of the two python files in working scripts. Normal is a more simple one and just gets real time info about the date and time. Wikipedia version is a bit more complicated. Using multishot questions, it first looks for a topic in the user's question, then it'll search that topic and if it exists as a wiki page then it'll get a snippet as context to then respond to the original question. May Search for Paris sometimes.

Wouldn't be possible, or at least this easy, without languagemodels and its dependencies. (First one if counting my programming knowledge.)