TY  - RPRT
TI  - Transformers Don't Need LayerNorm at Inference Time: Scaling LayerNorm Removal to GPT-2 XL and the Implications for Mechanistic Interpretability
AU  - Luca Baroni
AU  - Galvin Khara
AU  - Joachim Schaeffer
AU  - Marat Subkhankulov
AU  - Stefan Heimersheim
PY  - 2025
UR  - https://arxiv.org/abs/2507.02559
ID  - 2507.02559
ER  -