Arm Software Developers @developer.arm.com · Apr 2 On-device AI is becoming a standard part of how modern mobile apps are built. Google's Gemma 4 running on Arm enables more capable AI directly in Android apps, with Armv9, SME2, and KleidiAI supporting optimized performance behind these experiences. newsroom.arm.com/blog/gemma-4...
1 likes 1 replies
?
Replies
the memory budget is still the real constraint. Gemma 4 2B at q4 is roughly a 1 to 1.5 GB weights problem, 9B is closer to 5 GB before KV cache, so app architecture matters more than the model card. are you seeing SME2 help prompt eval more than decode?