LoopFormer: Elastic-Depth Looped Transformers for Latent Reasoning via Shortcut Modulation

TL;DR

LoopFormer introduces a flexible, budget-conditioned looped Transformer architecture that adapts its reasoning depth dynamically, improving efficiency and performance in language modeling and reasoning tasks under variable compute constraints.

Contribution

It proposes a shortcut-consistency training scheme for looped Transformers, enabling adaptive reasoning depth across variable-length trajectories.

Findings

Robust performance on language modeling benchmarks under compute constraints

Graceful scaling with additional computational budget

Effective latent reasoning with adaptive loop iterations

Abstract

Looped Transformers have emerged as an efficient and powerful class of models for reasoning in the language domain. Recent studies show that these models achieve strong performance on algorithmic and reasoning tasks, suggesting that looped architectures possess an inductive bias toward latent reasoning. However, prior approaches fix the number of loop iterations during training and inference, leaving open the question of whether these models can flexibly adapt their computational depth under variable compute budgets. We introduce LoopFormer, a looped Transformer trained on variable-length trajectories to enable budget-conditioned reasoning. Our core contribution is a shortcut-consistency training scheme that aligns trajectories of different lengths, ensuring that shorter loops yield informative representations while longer loops continue to refine them. LoopFormer conditions each loop on the current time and step size, enabling representations to evolve consistently across trajectories of varying length rather than drifting or stagnating. Empirically, LoopFormer demonstrates robust performance on language modeling and reasoning benchmarks even under aggressive compute constraints, while scaling gracefully with additional budget. These results show that looped Transformers are inherently suited for adaptive language modeling, opening a path toward controllable and budget-aware large language models.