Towards Universality: Studying Mechanistic Similarity Across Language Model Architectures
Keywords: Mechanistic Interpretability, Sparse Autoencoders, Universality, State Space Models
TL;DR: With sparse autoencoders, we find Transformers and Mambas are basically similar in their internal representation.
Abstract: The hypothesis of \textit{Universality} in interpretability suggests that different neural networks may converge to
implement similar algorithms on similar tasks. In this work, we investigate two mainstream architectures
for language modeling, namely Transformers and Mambas, to explore the extent of their mechanistic similarity.
We propose to use Sparse Autoencoders (SAEs) to isolate interpretable features from these models and show
that most features are similar in these two models. We also validate the correlation between feature similarity
and~\univ. We then delve into the circuit-level analysis of Mamba models
and find that the induction circuits in Mamba are structurally analogous to those in Transformers. We also identify a nuanced difference we call \emph{Off-by-One motif}: The information of one token is written into the
SSM state in its next position. Whilst interaction between tokens in Transformers does not exhibit such trend.
Primary Area: interpretability and explainable AI
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 4540
Loading