Understanding Verbatim Memorization in LLMs Through Circuit Discovery
This work provides mechanistic insights into memorization behavior in LLMs, which is an incremental but important step for understanding model safety and generalization.
The researchers investigated the underlying mechanisms of verbatim memorization in large language models by identifying specific transformer circuits responsible for initiating and maintaining memorized content, finding that initiation circuits can also maintain memorization while maintenance-only circuits cannot trigger initiation.
Underlying mechanisms of memorization in LLMs -- the verbatim reproduction of training data -- remain poorly understood. What exact part of the network decides to retrieve a token that we would consider as start of memorization sequence? How exactly is the models' behaviour different when producing memorized sentence vs non-memorized? In this work we approach these questions from mechanistic interpretability standpoint by utilizing transformer circuits -- the minimal computational subgraphs that perform specific functions within the model. Through carefully constructed contrastive datasets, we identify points where model generation diverges from memorized content and isolate the specific circuits responsible for two distinct aspects of memorization. We find that circuits that initiate memorization can also maintain it once started, while circuits that only maintain memorization cannot trigger its initiation. Intriguingly, memorization prevention mechanisms transfer robustly across different text domains, while memorization induction appears more context-dependent.