FutureBridge: Token Selection Beyond Local Preference in Collaborative Decoding

arXiv:2608.068192026-08-10

arXiv:2608.06819v1 Announce Type: new Abstract: Token-level collaboration allows a large language model (LLM) to assist a small language model (SLM) when their predictions diverge. Existing methods either use LLM-generated intervention tokens or rank candidates with the LLM's next-token probabilities. Both rely on the LLM's local preference, even though an LLM-selected token may be difficult for the SLM to build on. We present FutureBridge, which ranks joint LLM-SLM token candidates according to

저자 · Quanquan Li, Hongbo Zhang, Yihe Chi, Jingyu Li, Xidong Xi, Liuyang Song, Hongzhen Zhang, Yuxiang Huang, Jing Ke, Siyuan