Rapidly evolving LLM architectures and heterogeneous mobile hardware make supporting new models in on-device inference engines a recurring and costly system …
机构:上交
来源:arXiv 2611.14356 | AI4Papers 论文推荐平台