提出一种高效且低失真的文本生成方法，解决大模型拒绝不满意输出的难题。
Approximately Aligned Decoding
- 基于推测解码思想，近似对齐约束条件生成文本
- 相比现有方法，计算效率更高，低概率输出放大程度更低
- 适合需要严格约束但又怕算力消耗过大的场景
大语言模型常需拒绝不希望的输出，但现有方法要么需大量重采样计算，要么通过限制到极不可能的词元来实现，导致输出分布严重失真。本文提出近似对齐解码（AprAD），在输出分布失真与计算效率间取得平衡，受推测解码算法启发。该方法可在满足复杂约束条件下生成长序列文本，同时显著降低对低概率输出的放大程度。实验表明，AprAD在任务性能上接近不扭曲分布的方法，但计算效率远超后者。
原文摘要 · Abstract (English)
It is common to reject undesired outputs of Large Language Models (LLMs); however, current methods to do so require an excessive amount of computation to re-sample after a rejection, or distort the distribution of outputs by constraining the output to highly improbable tokens. We present a method, Approximately Aligned Decoding (AprAD), to balance the distortion of the output distribution with computational efficiency, inspired by algorithms from the speculative decoding literature. AprAD allows for the generation of long sequences of text with difficult-to-satisfy constraints, while amplifying low probability outputs much less compared to existing methods. We show through a series of experiments that the task-specific performance of AprAD is comparable to methods that do not distort the output distribution, while being much more computationally efficient.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。