Knowledge Localization in Mixture-of-Experts LLMs Using Cross-Lingual Inconsistency
This provides a scalable interpretability method for complex LLMs, addressing the challenge of understanding how knowledge is stored and accessed, though it is incremental as it builds on existing MoE and cross-lingual analysis techniques.
The paper tackles the problem of interpreting knowledge in mixture-of-experts LLMs by using cross-lingual inconsistency to localize experts responsible for factual recall, finding that deactivating about 20 out of 6000 experts reduces correct answers by over 40%.
Modern LLMs continue to exhibit significant variance in behavior across languages, such as being able to recall factual information in some languages but not others. While typically studied as a problem to be mitigated, in this work, we propose leveraging this cross-lingual inconsistency as a tool for interpretability in mixture-of-experts (MoE) LLMs. Our knowledge localization framework contrasts routing for sets of languages where the model correctly recalls information from languages where it fails. This allows us to isolate model components that play a functional role in answering about a piece of knowledge. Our method proceeds in two stages: (1) querying the model with difficult factual questions across a diverse set of languages to generate "success" and "failure" activation buckets and then (2) applying a statistical contrastive analysis to the MoE router logits to identify experts important for knowledge. To validate the necessity of this small number of experts for answering a knowledge question, we deactivate them and re-ask the question. We find that despite only deactivating about 20 out of 6000 experts, the model no longer answers correctly in over 40% of cases. Generally, this method provides a realistic and scalable knowledge localization approach to address increasingly complex LLMs.