An independent overview

RLT

Recurrent Looped Transformer

Reasoning that deepens with every token, not every layer.

Research by Yifan Zhang — paper & official project ↗