MURANO: Design, Run, and Reproduce Mechanistic Interpretability Experiments as Composable Pipelines

By Alireza Bayat Makou · Paper · cs.CL

This paper presents Murano, an open source framework for designing, running, and reproducing mechanistic interpretability studies of large language models, intended for researchers across disciplines. These studies often combine loading, recording, attribution, intervention, and

Cs.cl

View original

HomeResourceLoading…