An Adversarial Example for Direct Logit Attribution: Memory Management in GELU-4L

Jett Janiak, Can Rager, James Dao, Yeu-Tong Lau

Track: Full paper

Keywords: NLP, Transformers, Interpretability, Explainability, Memory Management

Abstract: Prior work suggests that language models manage the limited bandwidth of the residual stream through a "memory management" mechanism, where certain attention heads and MLP layers clear residual stream directions set by earlier layers. Our study provides concrete evidence for this erasure phenomenon in a 4-layer transformer, identifying heads that consistently remove the output of earlier heads. We further demonstrate that direct logit attribution (DLA), a common technique for interpreting the output of intermediate transformer layers, can show misleading results by not accounting for erasure.

Submission Number: 47