Poster
in
Workshop: Symmetry and Geometry in Neural Representations
Structure Development in List Sorting Transformers
Einar Urdshals · Jasmina Urdshals
Keywords: [ Developmental Interpretability ] [ Copy Suppression ] [ Head Specialization ]
We present an analysis of the evolution of the QK and OV circuits for a list sorting attention only transformer. Using various measures, we identify the developmental stages in the training process. In particular, we find two forms of head specialization later in the training: vocabulary-splitting and copy-suppression. We study their robustness by varying the training hyperparameters and the model architecture.