Thinking Like Transformers
This provides a formal abstraction for understanding and analyzing Transformer architectures, which is foundational for ML/AI research but incremental in offering a new perspective rather than a performance breakthrough.
The paper tackles the lack of a computational model for Transformers by proposing a programming language called RASP that maps transformer-encoder components to simple primitives, enabling programming of tasks like histograms, sorting, and Dyck-languages and relating task difficulty to required layers and heads.
What is the computational model behind a Transformer? Where recurrent neural networks have direct parallels in finite state machines, allowing clear discussion and thought around architecture variants or trained models, Transformers have no such familiar parallel. In this paper we aim to change that, proposing a computational model for the transformer-encoder in the form of a programming language. We map the basic components of a transformer-encoder -- attention and feed-forward computation -- into simple primitives, around which we form a programming language: the Restricted Access Sequence Processing Language (RASP). We show how RASP can be used to program solutions to tasks that could conceivably be learned by a Transformer, and how a Transformer can be trained to mimic a RASP solution. In particular, we provide RASP programs for histograms, sorting, and Dyck-languages. We further use our model to relate their difficulty in terms of the number of required layers and attention heads: analyzing a RASP program implies a maximum number of heads and layers necessary to encode a task in a transformer. Finally, we see how insights gained from our abstraction might be used to explain phenomena seen in recent works.