The official implementation of the paper "Fixed-Point Reasoners: Stable and Adaptive Deep Looped Transformers".
A beta version of the training code is available on the dev/training branch.
Checkpoints for the experiments are available on Hugging Face.