Sparse Crosscoders for diffing MoEs and Dense models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Chaudhari, Marmik, Hundia, Nishkal, Gulati, Idhant
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915838597005312
author Chaudhari, Marmik
Hundia, Nishkal
Gulati, Idhant
author_facet Chaudhari, Marmik
Hundia, Nishkal
Gulati, Idhant
contents Mixture of Experts (MoE) achieve parameter-efficient scaling through sparse expert routing, yet their internal representations remain poorly understood compared to dense models. We present a systematic comparison of MoE and dense model internals using crosscoders, a variant of sparse autoencoders, that jointly models multiple activation spaces. We train 5-layer dense and MoEs (equal active parameters) on 1B tokens across code, scientific text, and english stories. Using BatchTopK crosscoders with explicitly designated shared features, we achieve $\sim 87\%$ fractional variance explained and uncover concrete differences in feature organization. The MoE learns significantly fewer unique features compared to the dense model. MoE-specific features also exhibit higher activation density than shared features, whereas dense-specific features show lower density. Our analysis reveals that MoEs develop more specialized, focused representations while dense models distribute information across broader, more general-purpose features.
format Preprint
id arxiv_https___arxiv_org_abs_2603_05805
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Sparse Crosscoders for diffing MoEs and Dense models
Chaudhari, Marmik
Hundia, Nishkal
Gulati, Idhant
Machine Learning
Mixture of Experts (MoE) achieve parameter-efficient scaling through sparse expert routing, yet their internal representations remain poorly understood compared to dense models. We present a systematic comparison of MoE and dense model internals using crosscoders, a variant of sparse autoencoders, that jointly models multiple activation spaces. We train 5-layer dense and MoEs (equal active parameters) on 1B tokens across code, scientific text, and english stories. Using BatchTopK crosscoders with explicitly designated shared features, we achieve $\sim 87\%$ fractional variance explained and uncover concrete differences in feature organization. The MoE learns significantly fewer unique features compared to the dense model. MoE-specific features also exhibit higher activation density than shared features, whereas dense-specific features show lower density. Our analysis reveals that MoEs develop more specialized, focused representations while dense models distribute information across broader, more general-purpose features.
title Sparse Crosscoders for diffing MoEs and Dense models
topic Machine Learning
url https://arxiv.org/abs/2603.05805