I compared the Gemma 4 (E2B, E4B) models on my Pixel 10 Pro (LiteRT-LM, AICore) and on a iPhone 16.
Only when i run the AICore on my Pixel the LLM stops at some point mid sentence. This does not happen with the LiteRT-LM version. For example I ask "How to set up a meeting?" and the answer stops after 7,6s on AICore. The LiteRT Model on the other hand runs for 2 minutes without stopping.
I compared the Gemma 4 (E2B, E4B) models on my Pixel 10 Pro (LiteRT-LM, AICore) and on a iPhone 16.
Only when i run the AICore on my Pixel the LLM stops at some point mid sentence. This does not happen with the LiteRT-LM version. For example I ask "How to set up a meeting?" and the answer stops after 7,6s on AICore. The LiteRT Model on the other hand runs for 2 minutes without stopping.