SparKV is an adaptive framework that splits KV cache loading between cloud streaming and on-device computation to cut Time-to-First-Token by 1.3x–5.1x for on-device LLM inference. It models per-chunk costs and…

#OnDeviceLLM #EdgeAI #LLMInference #KVCache
https://arxiv.org/abs/2604.21231

A new edge AI architecture (AO-DA) splits persona and logic into separate inference paths on a single INT4 base model with hot-swappable LoRA adapters, aiming to keep on-device LLM reasoning robust against persona-heavy…

#EdgeAI #OnDeviceLLM #AIHardware #LLMAgents
https://arxiv.org/abs/2610.09772