The paper introduces CHASE, a cache‑hole‑adapted skip‑exit mechanism for looped state‑space language models, specifically Looped Mamba and Looped Hybrid Mamba‑Transformer. It shows that looping these architectures improves performance on controlled reasoning tasks and remains competitive in pre‑training benchmarks while using fewer distinct parameters. The cache‑hole adaptation allows selective skipping of recurrent steps during inference, maintaining perplexity close to full computation and achieving significant speedups.
By Zhenxuan Yu, Takeshi Kojima, Yutaka Matsuo, Yusuke Iwasawa
The paper critiques the common practice of evaluating depth usage in depth‑recurrent language models by truncating depth during inference and measuring performance decline. It argues that this method conflates three distinct effects—fewer block applications, reduced computation, and an out‑of‑distribution readout—yet is usually interpreted as measuring only the second. To address this, the authors introduce the Depth Control Protocol (DCP), a suite of positive and negative controls that isolate each factor, along with a training intervention to confirm causality, specifically tailored for depth‑wise weight‑sharing architectures.
By Ha Van Dau, Thanh Tung Khuat, Nguyen Thanh Dung
RecurTrace introduces adaptive latent reasoning for language models by allowing each looped layer to attend to its own past states and by using a halting head to decide when to stop iterating. This approach overcomes two limitations of prior latent recurrence methods: limited access to earlier computations and a fixed loop count that mismatches input difficulty. In experiments on MathQA, RecurTrace achieves 56.9% accuracy with an average of 2.0 loops, outperforming fixed‑depth baselines and other adaptive methods, and it also improves generation accuracy across a range of model sizes.
By Yuxiang Wang, Kunyu Feng, Yingda Shen, Haoning Xu, Junyu Wang, Zhizheng Wu
The paper investigates how temporal recurrence affects the required depth of neural networks in streaming tasks. By treating depth, expert width, and parallel experts as a compute‑allocation problem, the authors compare recurrent and non‑recurrent models across various compute budgets. Experiments on Sokoban and FineWeb language modeling show that recurrence shifts the optimal compute allocation toward fewer layers while maintaining or improving performance.
By Ivan Anokhin, Johan Obando-Ceron, Irina Rish, Sebastian Risi
arXiv:2606. 04438v1 Announce Type: cross Abstract: Mixture-of-Experts (MoE) and looped architectures scale models along two orthogonal axes, namely parameter capacity and effective depth.
By Wenkai Chen, Tianshu Li, Wenyong Huang, Yichun Yin, Lifeng Shang, Chengwei Qin
arXiv:2605. 08696v4 Announce Type: replace-cross Abstract: Over the last two decades, language modeling has experienced a shift from the use of predominantly recurrent architectures that process tokens sequentially during training and inference to non-recurrent models that process sequence elements in parallel during training, which results in greater training efficiency and stability at the expense of lower inference throughput.
By Benjamin L. Badger
arXiv:2607. 10110v1 Announce Type: new Abstract: Recent work on looped language models suggests that many reasoning problems benefit from greater computational depth rather than from additional independent parameters.
By Zhenxuan Yu, Takeshi Kojima, Yutaka Matsuo, Yusuke Iwasawa
arXiv:2608. 15533v1 Announce Type: cross Abstract: Linear attention models eliminate the quadratic prefix computation and context-growing KV cache of softmax attention by replacing pairwise token interactions with recurrent state updates.
By Junqing Lin, Jingwei Sun, Guangzhong Sun
arXiv:2609.27373v1 Announce Type: cross
Abstract: Recurrent language models repeatedly apply shared network blocks to refine latent representations, but standard inference recomputes global attention...
By Ke Wan, Chen Chen
arXiv:2608. 12435v1 Announce Type: new Abstract: Transformers owe much of their strong long-context retrieval capability to a token-level memory that grows with context length.
By Ming Zhang, Kaisen Yang, Shu Yu, Ermo Hua, Ning Ding, Xia Hu, Bowen Zhou, Chaochao Lu, Youbang Sun
WaveFront Decoding (WFD) is a training‑free, self‑speculative decoding framework for looped language models that reduces decoding latency by batching draft and verification steps within the same recurrent‑block calls. By exploiting intermediate recurrence outputs as draft predictions and weight sharing to process token states at different depths together, WFD arranges mixed‑depth states into a diagonal wavefront, allowing shallow‑depth drafting while deeper‑depth verification proceeds concurrently. Experiments on six Spec‑Bench task categories show WFD achieving up to 4.81× speedup on Huginn‑3.5B compared to autoregressive decoding, outperforming traditional draft‑then‑verify approaches.
By Hyeongju Ha, Jae-Joon Kim
The paper introduces Recursive Quadrature Filters (RQFs), complex‑valued temporal filters that act as band‑pass filters within diagonal state‑space models. By making each layer’s bottom‑up input prospective through a parameter‑free two‑tap update, the authors mitigate depth‑dependent gradient attenuation in deep continuous‑time recurrent networks. Experiments on RQFs, S5, and ORGaNICs show that prospective variants match or surpass non‑prospective controls, achieving high accuracy on raw‑audio Speech Commands and the Path‑X task with few parameters.
By Shivang Rawat, Mirko Morello, Flaviano Morone, David J. Heeger