TL;DR: We observe copy-suppression and head specialization in list sorting transformers.
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.
Style Files: I have used the style files.
Submission Number: 13
Loading