AI RESEARCH
A-IO: Adaptive Inference Orchestration for Memory-Bound NPUs
arXiv CS.AI
•
ArXi:2604.09752v1 Announce Type: cross During the deployment of Large Language Models (LLMs), the autoregressive decoding phase on heterogeneous NPU platforms (e.g., Ascend 910B) faces severe memory-bound challenges. This study reveals the ``Model Scaling Paradox'' caused by the static deployment of single-sized models.