AI RESEARCH
LT2: Linear-Time Looped Transformers
arXiv CS.LG
•
ArXi:2605.20670v1 Announce Type: new Looped Transformers (LT) have emerged as a powerful architecture by iterating their layers multiple times before decoding the final token. However, pairing them with full attention retains quadratic complexity, making them computationally expensive and slow. We