[vednig]

They've shared some interesting optimization techniques for bigger LLMs that's all, not exactly low powered devices as in power consumption. Still a good read.