列表排序 Transformer 中的结构发展
Structure Development in List-Sorting Transformers
AI总结：
研究一层仅含注意力机制的 Transformer 在学习数字列表排序时的内部结构发展，发现注意力头形成词汇拆分和复制抑制两种简单模式，并揭示训练数据特征如何塑造模型的内部组织。
摘要：
我们研究了一层仅含注意力机制的 Transformer 在学习对数字列表进行排序时如何发展出相关结构。在训练结束时，模型将其注意力头组织为两种主要模式，我们分别称之为词汇拆分和复制抑制。两者都代表了比让多个注意力头处理重叠数字范围更简单的模式。有趣的是，无论我们是否使用权重衰减（一种被认为会驱动简化的常见正则化技术），词汇拆分都依然存在，这支持了神经网络自然偏好更简单解决方案的论点。我们将复制抑制与 GPT-2 中的一种机制联系起来，并研究了其在我们的模型中的功能作用。在模型发展分析的指导下，我们识别出训练数据中驱动模型最终获得解决方案的特征。这提供了一个具体示例，说明训练数据如何塑造 Transformer 的内部组织，为未来帮助我们更好理解 LLM 如何发展其内部结构的研究铺平了道路。
英文摘要：
We study how a one-layer attention-only transformer develops relevant structures while learning to sort lists of numbers. At the end of training, the model organizes its attention heads in two main modes that we refer to as vocabulary-splitting and copy-suppression. Both represent simpler modes than having multiple heads handle overlapping ranges of numbers. Interestingly, vocabulary-splitting is present regardless of whether we use weight decay, a common regularization technique thought to drive simplification, supporting the thesis that neural networks naturally prefer simpler solutions. We relate copy-suppression to a mechanism in GPT-2 and investigate its functional role in our model. Guided by insights from a developmental analysis of the model, we identify features in the training data that drive the model's final acquired solution. This provides a concrete example of how the training data shape the internal organization of transformers, paving the way for future studies that could help us better understand how LLMs develop their internal structures.
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。
