aiqbench analysis
This research introduces CLVQ-VAE, a novel framework that uses discrete vector quantization to resolve the continuous mixing of features across layers in language models. By collapsing duplicated residual-stream features into compact concept vectors, the method significantly improves interpretability metrics compared to standard clustering and single-layer SAEs. The approach offers a more robust tool for analyzing how concepts propagate through deep neural networks without relying on continuous latent spaces.
Cross-Layer Discrete Concept Discovery for Interpreting Language Models
arXiv:2506.20040v4 Announce Type: replace-cross Abstract: Interpreting language models remains challenging due to the existence of residual stream, which linearly mixes and duplicates features across adjacent layers, causing single-layer analyses to miss this cross-layer structure. Cross-layer sparse autoencoders (SAEs) address layer mixing but operate in continuous space, where concepts split across many neurons without clear boundaries. We introduce Cross-Layer Vector Quantized-Variational Autoencoder (CLVQ-VAE), a novel framework which maps representations from a lower layer to a higher layer through a discrete vector-quantization bottleneck, collapsing duplicated residual-stream features into compact, interpretable concept vectors. Our approach combines top-k temperature-based sampling with exponential moving average (EMA) codebook updates, providing controlled exploration of the discrete latent space while maintaining codebook diversity. Across both encoder- and decoder-based models on ERASER-Movie, Jigsaw, and AGNews, CLVQ-VAE outperforms clustering, single-layer vector quantized-variational autoencoder (VQ-VAE), and sparse autoencoder (SAE) baselines across three evaluation axes: removing identified concepts drops downstream probe accuracy by up to 93%, LLM judges rank our concepts first in 66.7% of comparisons, and human annotators recover model predictions from our visualizations with 78% accuracy versus 54% for clustering.
Abstract:Interpreting language models remains challenging due to the existence of residual stream, which linearly mixes and duplicates features across adjacent layers, causing single-layer analyses to miss this cross-layer structure. Cross-layer sparse autoencoders (SAEs) address layer mixing but operate in continuous space, where concepts split across many neurons without clear boundaries. We introduce Cross-Layer Vector Quantized-Variational Autoencoder (CLVQ-VAE), a novel framework which maps representations from a lower layer to a higher layer through a discrete vector-quantization bottleneck, collapsing duplicated residual-stream features into compact, interpretable concept vectors. Our approach combines top-k temperature-based sampling with exponential moving average (EMA) codebook updates, providing controlled exploration of the discrete latent space while maintaining codebook diversity. Across both encoder- and decoder-based models on ERASER-Movie, Jigsaw, and AGNews, CLVQ-VAE outperforms clustering, single-layer vector quantized-variational autoencoder (VQ-VAE), and sparse autoencoder (SAE) baselines across three evaluation axes: removing identified concepts drops downstream probe accuracy by up to 93%, LLM judges rank our concepts first in 66.7% of comparisons, and human annotators recover model predictions from our visualizations with 78% accuracy versus 54% for clustering.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
arXiv:2506.20040 [cs.LG]
(or arXiv:2506.20040v4 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2506.20040 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 274 words
Related coverage
Cross-Layer Discrete Concept Discovery for Interpreting Language Models
arXiv · Jun 12, 2026
The J-Space Debate, Agent Swarms, and Pacing Frontier AI - Digital Minds Newsletter #4
LessWrong · Sep 18, 2026
DisParQ: Self-Supervised Part Concepts for Interpretable Vision Foundation Models
arXiv · Oct 8, 2026
Beyond Linear Concepts: Discovering and Aligning Non-Linear Concept Manifolds in Large Language Models
arXiv · Oct 2, 2026