I have been running Google AI Edge Gallery with Gemma 3n E4B on my Pixel 7a for several days. Without Google Play Services, and with network permissions fully revoked for the app. It works really well. I am getting 4-5 tokens per second, and the camera feature is really neat and useful. The model is not anywhere near as good as Gemma 3 27B, but definitely enough as one to have with me on-the-go. I was surprised how well it works on my phone. But yeah, the app crashes sometimes, and other apps gets unloaded in the background, I guess because of shortage of RAM memory on my phone. It also consumes about 2% of battery per answer it generates, which is the price to pay for running such a large model all offline. But having a really good and privacy respecting LLM on my phone seems to be very close of becoming a reality now, and Google seems committed to make it a reality.