titleSAE Interprets MoE: Sparse AutoEncoders Reveal That MoE Experts Exhibit Human-Interpretable Fine-Grained Specialization
authorsAnyuan Zhuo, Shaohua Zhang, Yuan Lin, Pinyan Lu
date2026-11-05
venuearXiv (preprint, not peer reviewed)
idarXiv:2611.08089
affiliationByteDance Seed
upvotes7 (Hugging Face Daily Papers, 2026-11-08)
linksabs pdf hf code
cite
dispatch Sparse autoencoders trace which narrow features each expert in an MoE model drives
relatedTraining Long-Context Vision-Language Models Effectively with Generalization Beyond 128K Context
Continuous Latent Diffusion Language Model
HarnessDev: Can LLMs Create and Evolve Their Own Agent Harness?
DanceOPD: On-Policy Generative Field Distillation
GenFirst: Generation Before Reconstruction for Stable End-to-End Latent Generative Modeling
abstractMixture-of-Experts (MoE) models route each token to a few of many experts, yet what an individual expert is responsible for remains unsettled. Existing analyses count how often an expert is selected on domain-labeled corpora, an instrument that fixes the taxonomy in advance, stays coarse, and records selection rather than contribution. We instead study specialization in feature space by training a sparse autoencoder (SAE) on the output of an MoE layer. Since this output is a router-weighted sum of expert outputs and the SAE encoder is affine, the pre-activation of every latent decomposes exactly over the experts. Aggregated over a corpus, this gives each expert a dominant set of latents whose activation it principally drives. On layers 4, 30 and 46 of Qwen3-30B-A3B, contribution is far more concentrated than gating, most latents have only one or two dominators, and the overlap between dominant sets concentrates on a few expert pairs. Auto-interpreting these sets reveals a division of labor below coarse domains, with distinct experts for Java and front-end web code, or for geometry and for arithmetic and polynomials, while the coarse domains themselves take very unequal numbers of experts. Our SAEs and code are available at https://github.com/Bytedance-Seed/SAE-Interprets-MoE.

bibtex

@misc{zhuo2026sae,
  title={SAE Interprets MoE: Sparse AutoEncoders Reveal That MoE Experts Exhibit Human-Interpretable Fine-Grained Specialization},
  author={Anyuan Zhuo and Shaohua Zhang and Yuan Lin and Pinyan Lu},
  year={2026},
  eprint={2611.08089},
  archivePrefix={arXiv},
  url={https://arxiv.org/abs/2611.08089}
}