用特征空间提升数据多样性，让大模型训练更高效。
Less is Enough: Synthesizing Diverse Data in LLM Feature Space with Sparse Autoencoders
- 通过稀疏自编码器识别缺失的模型特征，生成针对性合成数据。
- 在多个任务上显著提升数据多样性和下游性能，最高增益达12.3%。
- 发现跨模型共有的可解释特征空间，支持模型间知识迁移。
后训练数据的多样性对大语言模型（LLM）的下游表现至关重要。现有方法多依赖文本层面的多样性度量，但此类指标仅反映语言变化，无法捕捉决定性能的关键特征。本文提出特征激活覆盖率（FAC），在可解释的特征空间中衡量数据多样性。基于此，我们构建了名为FAC Synthesis的多样性驱动数据生成框架：先用稀疏自编码器识别种子数据集中缺失的特征，再生成显式包含这些特征的合成样本。实验表明，该方法在指令遵循、毒性检测、奖励建模和行为引导等多个任务上均显著提升数据多样性和下游性能。有趣的是，我们发现不同模型家族（如LLaMA、Mistral、Qwen）共享一个可解释的特征空间，支持跨模型知识迁移。本工作为数据驱动优化大语言模型提供了可靠且实用的方法。
原文摘要 · Abstract (English)
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.
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。