# ScaleRLT: Exact Log-Probabilities and Fast Deferred Rollouts for Reinforcement Learning on Looped Transformers

Authors: Dezhen Lu, Zixiang Ni, Runguo Li.

Batches later transformer loops across generated positions to accelerate reinforcement-learning rollouts. Replays the rollout schedule to recover its exact token log probabilities while using a dense pass for gradients, and evaluates the resulting training system on math problems.

First public: 2026-10-25.
ArXiv initial submission: 2026-10-25.

[Original paper](https://arxiv.org/abs/2610.34427) · [PDF](https://arxiv.org/pdf/2610.34427) · [HTML](https://arxiv.org/html/2610.34427v1) · [Reading-list entry](https://latent-implicit-thinking.github.io/papers/2610.34427/)
[Preprint BibTeX](https://latent-implicit-thinking.github.io/papers/2610.34427/citation.bib) · [Preprint CSL-JSON](https://latent-implicit-thinking.github.io/papers/2610.34427/citation.csl.json) · [JSON record](https://latent-implicit-thinking.github.io/papers/2610.34427/record.json)

## Approaches

- [Recurrent reasoning](https://latent-implicit-thinking.github.io/topics/recurrent-reasoning/): Repeat parts of the model or keep intermediate information in memory while working through a problem. Some methods also learn how many passes to run.

Reading-list version 0.12.0; literature checked through 2026-11-02.