arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2410.13640cs.CLcs.AIcs.LG
收藏

潜在空间 Chain-of-Embedding 实现无输出的 LLM 自我评估

Latent Space Chain-of-Embedding Enables Output-free LLM Self-Evaluation

Shanghai Jiao Tong University(上海交通大学);Alibaba Group Inc.(阿里巴巴集团);University of Macau(澳门大学)

Yiming Wang, Pei Zhang, Baosong Yang, Derek F. Wong, Rui Wang

AI总结:

本文提出潜在空间 Chain-of-Embedding,利用推理时渐进隐藏状态的正误特征差异,在无输出、无标签、免训练条件下实现 LLM 实时自我评估。

摘要:

LLM 自我评估依赖 LLM 自身估计回答正确性的能力,这可显著提升其部署可靠性。在这一研究方向中,我们提出潜在空间中的 Chain-of-Embedding(CoE),使 LLM 能够执行无输出的自我评估。CoE 由推理期间产生的所有渐进式隐藏状态组成,可被视为 LLM 的潜在思维路径。我们发现,当 LLM 回答正确与错误时,其 CoE 特征存在差异,这些差异有助于我们估计 LLM 回答的正确性。在四个不同领域和七个 LLM 上的实验充分证明了本方法的有效性。同时,其无需标签、无需任何训练的设计意图以及毫秒级计算成本,确保了大规模场景中的实时反馈。更重要的是,我们从 LLM 内部隐藏状态变化的角度,提供了关于 LLM 回答正确性的有趣洞见。

英文摘要:

LLM self-evaluation relies on the LLM's own ability to estimate response correctness, which can greatly improve its deployment reliability. In this research track, we propose the Chain-of-Embedding (CoE) in the latent space to enable LLMs to perform output-free self-evaluation. CoE consists of all progressive hidden states produced during the inference time, which can be treated as the latent thinking path of LLMs. We find that when LLMs respond correctly and incorrectly, their CoE features differ, these discrepancies assist us in estimating LLM response correctness. Experiments in four diverse domains and seven LLMs fully demonstrate the effectiveness of our method. Meanwhile, its label-free design intent without any training and millisecond-level computational cost ensures real-time feedback in large-scale scenarios. More importantly, we provide interesting insights into LLM response correctness from the perspective of hidden state changes inside LLMs.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑