Inference-Time Decomposition of Activations (ITDA): A Scalable Approach to Interpreting Large Language Models

Leask, Patrick; Al Moubayed, Noura

Abstract

Sparse Autoencoders (SAEs) are a popular method for decomposing Large Language Model (LLM) activations into interpretable latents, however they have a substantial training cost and SAEs learned on different models are not directly comparable. Motivated by relative representation similarity measures, we introduce Inference-Time Decomposition of Activation models (ITDAs). ITDAs are constructed by greedily sampling activations into a dictionary based on an error threshold on their matching pursuit reconstruction. ITDAs can be trained in 1% of the time of SAEs, allowing us to cheaply train them on Llama-3.1 70B and 405B. ITDA dictionaries also enable cross-model comparisons, and outperform existing methods like CKA, SVCCA, and a relative representation method on a benchmark of representation similarity. Code available at https://github.com/pleask/itda

Citation

Leask, P., & Al Moubayed, N. (2025, July). Inference-Time Decomposition of Activations (ITDA): A Scalable Approach to Interpreting Large Language Models. Presented at International Conference on Machine Learning (ICML 2025), Vancouver, Canada

• Presentation Conference Type: Conference Paper (published)
• Conference Name: International Conference on Machine Learning (ICML 2025)
• Start Date: Jul 13, 2025
• End Date: Jul 19, 2025
• Acceptance Date: May 26, 2025
• Deposit Date: Jun 1, 2025
• Peer Reviewed: Peer Reviewed
• Series Title: Proceedings of Machine Learning Research
• Series ISSN: 2640-3498
• Public URL: https://durham-repository.worktribe.com/output/4012842
• Publisher URL: https://proceedings.mlr.press/
• External URL: https://icml.cc/Conferences/2025