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 2401.17692cs.CL
收藏

利用先验感知解码缓解语言模型中干扰任务的影响

Mitigating the Influence of Distractor Tasks in LMs with Prior-Aware Decoding

IMCS UL;Charles University(查尔斯大学)

Raymond Douglas, Andis Draguns, Tomáš Gavenčiak

AI总结:

针对语言模型易受干扰任务影响产生非预期输出的问题,提出基于专家乘积理论框架的先验感知解码(PAD)方法,在多数据集和模型上显著提升了任务完成率。

摘要:

语言模型(LMs)的广泛能力可能因其对干扰任务的敏感性而受限:LMs除了能从提示中推断出预期任务外,还能推断出次要任务,从而导致不需要的输出。例如,提示注入攻击可使模型偏离显式指令。在某些“逆缩放”情况下,随着模型规模扩大到至少540B参数,这种不需要的行为实际上反而会恶化。我们提出了一个理论框架,将LMs解释为结合多个数据生成过程的专家乘积。基于该框架,我们提出了先验感知解码(PAD)——一种用于减少干扰任务影响的简单对比推理方法。我们将PAD应用于四个数据集上的十一个模型,发现在44个任务-模型组合中有41个得到了改善,任务完成比例的中位数增加了40%。结果表明,这为更可靠语言模型的进一步发展指明了一个有前景的方向。

英文摘要:

The broad capabilities of Language Models (LMs) can be limited by their sensitivity to distractor tasks: LMs can infer secondary tasks from the prompt in addition to the intended one, leading to unwanted outputs. For example, prompt injection attacks can cause models to deviate from explicit directives. In some 'inverse scaling' cases, this unwanted behaviour actually worsens as models scale up to at least 540B parameters. We present a theoretical framework that interprets LMs as a product of experts that combine multiple data generation processes. Based on this framework, we demonstrate prior-aware decoding (PAD) - a simple contrastive inference method to reduce the influence of distractor tasks. We apply PAD to eleven models, across four datasets, and find improvements in 41 out of 44 task-model combinations, with a median increase in task completion proportion of 40%. The results suggest a promising direction for further development towards more reliable language models.

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。

↑