On-device streaming omni-modal inference safeguards user privacy and eliminates prohibitive per-token API costs, but faces a critical bottleneck: the continu…
机构:北大
来源:arXiv 2609.34653 | AI4Papers 论文推荐平台