用于可控基于能量文本生成的分块 Metropolis-Hastings 采样器
A Block Metropolis-Hastings Sampler for Controllable Energy-based Text Generation
AI总结：
针对基于能量语言模型推理依赖单 token MH 采样的局限，本文提出通过迭代提示大模型重写整段序列的分块 MH 采样器，支持采样中确定长度，并在两项任务上提升性能与分布采样准确性。
摘要：
近期研究表明，基于能量的语言建模是可控文本生成的有效框架，因为它能够灵活整合任意判别器。然而，由于基于能量的语言模型采用全局归一化，推理时需要使用 Metropolis-Hastings（MH）等近似技术。以往工作主要探索每次只修改单个 token 的简单提议分布，例如 Gibbs sampling。本文提出一种新颖的 MH 采样器；与之不同，该采样器在每一步通过对大型语言模型进行迭代提示来提议重写整个序列。我们的新采样器（a）能够更高效、更准确地从目标分布中采样，并且（b）允许生成长度由采样过程决定，而不是像以往工作要求的那样预先固定。我们在两个受控生成任务上进行实验，结果表明，与单 token 提议技术相比，该方法不仅提升了下游性能，还能更准确地对目标分布进行采样。
英文摘要：
Recent work has shown that energy-based language modeling is an effective framework for controllable text generation because it enables flexible integration of arbitrary discriminators. However, because energy-based LMs are globally normalized, approximate techniques like Metropolis-Hastings (MH) are required for inference. Past work has largely explored simple proposal distributions that modify a single token at a time, like in Gibbs sampling. In this paper, we develop a novel MH sampler that, in contrast, proposes re-writes of the entire sequence in each step via iterative prompting of a large language model. Our new sampler (a) allows for more efficient and accurate sampling from a target distribution and (b) allows generation length to be determined through the sampling procedure rather than fixed in advance, as past work has required. We perform experiments on two controlled generation tasks, showing both downstream performance gains and more accurate target distribution sampling in comparison with single-token proposal techniques.
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。
