Add Google's LiteRT-LM

[Katehuuh]

Today, Google released https://huggingface.co/collections/google/gemma-4-qat-mobile. This enables faster inference -mobile-ct. Would it be worth adding LiteRT-LM? given that the format wNa8o8 for NPUs isn't supported by GGUF?

Mobile-optimized (wNa8o8): A custom schema engineered explicitly for mobile hardware efficiency. It features targeted 2-bit decoding layers, optimized KV caches, and static activations to maximize VRAM savings. Available for Gemma 4 E2B and E4B.

[Vali-98]

I have four main concerns on adding more backends to ChatterUI:

• Limited model support of niche backends - Executorch and LiteRT support only a handful of models.

• Model management in the app isnt built for multi-file models, and would require a bit of a rework.

• Possible SDK/device cutoffs - some APIs used by these backends are not available on all android versions, and would make the app unusable on older devices. This would require either app variants to be made targeting specific SDKs or just dropping old device support which isn't ideal.

• There are many pipeline differences between say, llama.rn and react-native-litert-lm, samplers aren't as granular, model splitting between cpu/gpu isnt available, etc. UI side this is a massive hassle to do cleanly.