Poster Session
in
Workshop: Scientific Methods for Understanding Neural Networks
Structure Development in List Sorting Transformers
Einar Urdshals · Jasmina Urdshals
Abstract:
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.
Chat is not available.