Vardan Papyan

A Two-Character Change in Transformer Architecture Promotes Ideal Token Geometry

Attention Sinks: A 'Catch, Tag, Release' Mechanism for Embeddings

On the Importance of Gaussianizing Representations

OATS: Outlier-Aware Pruning Through Sparse and Low Rank Decomposition

Transformer Block Coupling and its Correlation with Generalization in LLMs

Beyond Standardization – Putting the Normality in Normalization

Breach By A Thousand Leaks: Unsafe Information Leakage in 'Safe' AI Responses