Saved in:
Bibliographic Details
Main Authors: Ensign, Danielle, Garriga-Alonso, Adrià
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2407.14008
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910536517550080
author Ensign, Danielle
Garriga-Alonso, Adrià
author_facet Ensign, Danielle
Garriga-Alonso, Adrià
contents How well will current interpretability techniques generalize to future models? A relevant case study is Mamba, a recent recurrent architecture with scaling comparable to Transformers. We adapt pre-Mamba techniques to Mamba and partially reverse-engineer the circuit responsible for the Indirect Object Identification (IOI) task. Our techniques provide evidence that 1) Layer 39 is a key bottleneck, 2) Convolutions in layer 39 shift names one position forward, and 3) The name entities are stored linearly in Layer 39's SSM. Finally, we adapt an automatic circuit discovery tool, positional Edge Attribution Patching, to identify a Mamba IOI circuit. Our contributions provide initial evidence that circuit-based mechanistic interpretability tools work well for the Mamba architecture.
format Preprint
id arxiv_https___arxiv_org_abs_2407_14008
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Investigating the Indirect Object Identification circuit in Mamba
Ensign, Danielle
Garriga-Alonso, Adrià
Machine Learning
How well will current interpretability techniques generalize to future models? A relevant case study is Mamba, a recent recurrent architecture with scaling comparable to Transformers. We adapt pre-Mamba techniques to Mamba and partially reverse-engineer the circuit responsible for the Indirect Object Identification (IOI) task. Our techniques provide evidence that 1) Layer 39 is a key bottleneck, 2) Convolutions in layer 39 shift names one position forward, and 3) The name entities are stored linearly in Layer 39's SSM. Finally, we adapt an automatic circuit discovery tool, positional Edge Attribution Patching, to identify a Mamba IOI circuit. Our contributions provide initial evidence that circuit-based mechanistic interpretability tools work well for the Mamba architecture.
title Investigating the Indirect Object Identification circuit in Mamba
topic Machine Learning
url https://arxiv.org/abs/2407.14008