少即是多:利用稀疏自编码器在LLM特征空间中合成多样化数据
Less is Enough: Synthesizing Diverse Data in LLM Feature Space with Sparse Autoencoders
AI总结:
提出基于稀疏自编码器的特征激活覆盖率(FAC)指标及数据合成框架FAC Synthesis,通过识别缺失特征并生成对应样本来提升数据多样性和下游任务性能。
摘要:
后训练数据的多样性对于大型语言模型(LLM)的有效下游性能至关重要。许多现有的后训练数据构建方法使用基于文本的指标来衡量多样性,这些指标捕捉语言变化,但此类指标仅能为决定下游性能的任务相关特征提供微弱信号。在这项工作中,我们引入了特征激活覆盖率(FAC),该指标在可解释的特征空间中衡量数据多样性。基于此指标,我们进一步提出了一个多样性驱动的数据合成框架,名为FAC Synthesis,该框架首先使用稀疏自编码器从种子数据集中识别缺失特征,然后生成明确反映这些特征的合成样本。实验表明,我们的方法在包括指令遵循、毒性检测、奖励建模和行为引导在内的各种任务上,持续提高了数据多样性和下游性能。有趣的是,我们识别出跨模型家族(即LLaMA、Mistral和Qwen)共享的可解释特征空间,从而实现了跨模型知识迁移。我们的工作为探索以数据为中心的LLM优化提供了坚实且实用的方法论。
英文摘要:
The diversity of post-training data is critical for effective downstream performance in large language models (LLMs). Many existing approaches to constructing post-training data quantify diversity using text-based metrics that capture linguistic variation, but such metrics provide only weak signals for the task-relevant features that determine downstream performance. In this work, we introduce Feature Activation Coverage (FAC) which measures data diversity in an interpretable feature space. Building upon this metric, we further propose a diversity-driven data synthesis framework, named FAC Synthesis, that first uses a sparse autoencoder to identify missing features from a seed dataset, and then generates synthetic samples that explicitly reflect these features. Experiments show that our approach consistently improves both data diversity and downstream performance on various tasks, including instruction following, toxicity detection, reward modeling, and behavior steering. Interestingly, we identify a shared, interpretable feature space across model families (i.e., LLaMA, Mistral, and Qwen), enabling cross-model knowledge transfer. Our work provides a solid and practical methodology for exploring data-centric optimization of 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。