Compositional Capabilities of Autoregressive Transformers: A Study on Synthetic, Interpretable Tasks

Rahul RameshEkdeep Singh LubanaMikail KhonaRobert P. DickHidenori Tanaka

Rahul RameshEkdeep Singh LubanaMikail KhonaRobert P. DickHidenori Tanaka

Feb 2024

0被引用

0笔记

摘要原文

Transformers trained on huge text corpora exhibit a remarkable set of capabilities, e.g., performing basic arithmetic. Given the inherent compositional nature of language, one can expect the model to learn to compose these capabilities, potentially yielding a combinatorial explosion of what operations it can perform on an input. Motivated by the above, we train autoregressive Transformer models on a synthetic data-generating process that involves compositions of a set of well-defined monolithic capabilities. Through a series of extensive and systematic experiments on this data-generating process, we show that: (1) autoregressive Transformers can learn compositional structures from small amounts of training data and generalize to exponentially or even combinatorially many functions; (2) generating intermediate outputs when composing functions is more effective for generalizing to new, unseen compositions than not generating any intermediate outputs (3) biases in the order of the compositions in the training data result in Transformers that fail to compose some combinations of functions; and (4) the attention layers select which capability to apply while the feed-forward layers execute the selected capability.