Contents

• Taxonomy
• A. Loop Topology
  • A1. Whole Stack
  • A2. Sandwich
  • A3. Layer-Local / Immediate
  • A4. Cyclic Patterns
  • A5. Component-Level
  • A6. Hierarchical / Two-Timescale
  • A7. Parallel / Pipelined
• B. Loop-Count Policy
• C. Inter-Iteration State
• D. Sharing Strictness
• E. Training Recipe
• F. Memory / KV Strategy

---

Taxonomy

A. Loop Topology

A1. Whole Stack

The whole transformer stack/shared block is repeatedly applied.

• Universal Transformers — Mostafa Dehghani, Stephan Gouws, Oriol Vinyals, Jakob Uszkoreit, Łukasz Kaiser. ICLR 2019.
• ALBERT: A Lite BERT for Self-Supervised Learning of Language Representations — Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, Radu Soricut. ICLR 2020.
• End-to-End Algorithm Synthesis with Recurrent Networks: Logical Extrapolation without Overthinking — Arpit Bansal et al. arXiv, 2022.
• Sparse Universal Transformer — Shawn Tan, Yikang Shen, Zhenfang Chen, Aaron Courville, Chuang Gan. EMNLP 2023.
• Looped Transformers Are Better at Learning Learning Algorithms — Liu Yang, Kangwook Lee, Robert D. Nowak, Dimitris Papailiopoulos. ICLR 2024.
• Loop Neural Networks for Parameter Sharing — Kei-Sing Ng, Qingchen Wang. arXiv, 2024.
• PonderLM: Pretraining Language Models to Ponder in Continuous Space — Boyi Zeng et al. ICLR 2026.
• Scaling Latent Reasoning via Looped Language Models — Rui-Jie Zhu et al. arXiv, 2025.
• LoopFormer: Elastic-Depth Looped Transformers for Latent Reasoning via Shortcut Modulation — Ahmadreza Jeddi, Marco Ciccone, Babak Taati. arXiv, 2026.
• AdaPonderLM: Gated Pondering Language Models with Token-Wise Adaptive Depth — Shixiang Song et al. arXiv, 2026.
• Memory-Efficient Looped Transformer: Decoupling Compute from Memory in Looped Language Models — Victor Conchello Vendrell et al. arXiv, 2026.
• Fixed-Point Reasoners: Stable and Adaptive Deep Looped Transformers — Sajad Movahedi et al. arXiv, 2026.
• Nanbeige4.2-3B: Unlocking Agentic Capabilities in a Compact Model — Nanbeige Lab et al. arXiv, 2026.
• CHASE: Cache-Hole-Adapted Skip Exit for Looped State-Space Language Models — Zhenxuan Yu, Takeshi Kojima, Yutaka Matsuo, Yusuke Iwasawa. arXiv, 2026.

A2. Sandwich: Prelude → Shared Core → Coda

Only the middle recurrent core is shared/looped.

• Subformer: Exploring Weight Sharing for Parameter Efficiency in Generative Transformers — Machel Reid, Edison Marrese-Taylor, Yutaka Matsuo. Findings of EMNLP 2021.
• AlgoFormer: An Efficient Transformer Framework with Algorithmic Structures — Yihang Gao et al. arXiv, 2024.
• Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach — Jonas Geiping et al. NeurIPS 2025.
• Encode, Think, Decode: Scaling Test-Time Reasoning with Recursive Latent Thoughts — Yeskendir Koishekenov, Aldo Lipani, Nicola Cancedda. arXiv, 2025.
• Teaching Pretrained Language Models to Think Deeper with Retrofitted Recurrence — Sean McLeish et al. arXiv, 2025.
• Parcae: Scaling Laws for Stable Looped Language Models — Hayden Prairie et al. arXiv, 2026.
• How Much Is One Recurrence Worth? Iso-Depth Scaling Laws for Looped Language Models — Kristian Schwethelm, Daniel Rueckert, Georgios Kaissis. arXiv, 2026.
• Hyperloop Transformers — Abbas Zeitoun, Lucas Torroba-Hennigen, Yoon Kim. arXiv, 2026.
• Sparse Layers Are Critical to Scaling Looped Language Models — Ryan Lee, Jacob Biloki, Edward J. Hu, Jonathan May. arXiv, 2026.
• LoopUS: Recasting Pretrained LLMs into Looped Latent Refinement Models — Taekhyun Park, Yongjae Lee, Dohee Kim, Hyerim Bae. arXiv, 2026.
• Training-Free Looped Transformers — Lizhang Chen, Jonathan Li, Chen Liang, Ni Lao, Qiang Liu. arXiv, 2026.
• LoopMoE: Unifying Iterative Computation with Mixture-of-Experts for Language Modeling — Wenkai Chen et al. arXiv, 2026.
• SMELT: Scaling Laws for Compute-Matched MoE Looped Transformers — Shaowen Wang et al. arXiv, 2026.
• RecurTrace: Adaptive Latent Reasoning with Loop-Time Memory — Yuxiang Wang et al. arXiv, 2026.

A3. Layer-Local / Immediate

Each layer or small layer group is immediately repeated.

• MobileLLM: Optimizing Sub-Billion Parameter Language Models for On-Device Use Cases — Zechun Liu et al. ICML 2024.
• MoEUT: Mixture-of-Experts Universal Transformers — Róbert Csordás, Kazuki Irie, Jürgen Schmidhuber, Christopher Potts, Christopher D. Manning. NeurIPS 2024.
• Inner Thinking Transformer: Leveraging Dynamic Depth Scaling to Foster Adaptive Internal Thinking — Yilong Chen et al. ACL 2025.
• Loop the Loopies! — Zitian Gao et al. arXiv, 2026.

A4. Cyclic Patterns over M Unique Blocks

A set of unique blocks is reused following sequence/cycle patterns.

• Lessons on Parameter Sharing across Layers in Transformers — Sho Takase, Shun Kiyono. SustaiNLP @ ACL 2023.
• Relaxed Recursive Transformers: Effective Parameter Sharing with Layer-Wise LoRA — Sangmin Bae, Adam Fisch, Hrayr Harutyunyan, Ziwei Ji, Seungyeon Kim, Tal Schuster. ICLR 2025.
• Mixture-of-Recursions: Learning Dynamic Recursive Depths for Adaptive Token-Level Computation — Sangmin Bae et al. NeurIPS 2025.

A5. Component-Level

Only a selected transformer component is recurrent.

• Allocating Recurrent Compute in Looped Language Models — Ruhai Lin et al. arXiv, 2026.
  • Introduces the MixerLoop component-level design used in the survey taxonomy.

A6. Hierarchical / Two-Timescale

Fast and slow recurrent modules operate at different timescales.

• Hierarchical Reasoning Model — Guan Wang et al. arXiv, 2025.
• Less Is More: Recursive Reasoning with Tiny Networks — Alexia Jolicoeur-Martineau. arXiv, 2025.
• Universal Reasoning Model — Zitian Gao et al. arXiv, 2025.
• Hierarchical vs. Flat Iteration in Shared-Weight Transformers — Sang-Il Han. arXiv, 2026.
• HRM-Text: Efficient Pretraining Beyond Scaling — Guan Wang et al. arXiv, 2026.

A7. Parallel / Pipelined across Loops

Loop computation is parallelized or pipelined.

• Parallel Loop Transformer for Efficient Test-Time Computation Scaling — Bohong Wu et al. arXiv, 2025.
• LoopCoder-v2: Only Loop Once for Efficient Test-Time Computation Scaling — Jian Yang et al. arXiv, 2026.
• Bridging the Gap Between Latent and Explicit Reasoning with Looped Transformers — Ying Fan, Anej Svete, Kangwook Lee. arXiv, 2026.

---

B. Loop-Count Policy

B1. Fixed (T)

The model uses a fixed number of recurrent passes.

• ALBERT — Zhenzhong Lan et al. ICLR 2020.
• MobileLLM — Zechun Liu et al. ICML 2024.
• PonderLM — Boyi Zeng et al. ICLR 2026.
• Parallel Loop Transformer — Bohong Wu et al. arXiv, 2025.
• Loop the Loopies! — Zitian Gao et al. arXiv, 2026.
• Nanbeige4.2-3B — Nanbeige Lab et al. arXiv, 2026.
• LoopCoder-v2 — Jian Yang et al. arXiv, 2026.
• SMELT — Shaowen Wang et al. arXiv, 2026.

B2. Sampled at Train, Free at Test

Loop count varies during training and can be changed at inference.

• End-to-End Algorithm Synthesis with Recurrent Networks — Arpit Bansal et al. arXiv, 2022.
• Looped Transformers Are Better at Learning Learning Algorithms — Liu Yang et al. ICLR 2024.
• Scaling up Test-Time Compute with Latent Reasoning — Jonas Geiping et al. NeurIPS 2025.
• Teaching Pretrained Language Models to Think Deeper with Retrofitted Recurrence — Sean McLeish et al. arXiv, 2025.
• LoopFormer — Ahmadreza Jeddi et al. arXiv, 2026.
• Parcae — Hayden Prairie et al. arXiv, 2026.
• Stabilizing Recurrent Dynamics for Test-Time Scalable Latent Reasoning in Looped Language Models — Xiao-Wen Yang et al. arXiv, 2026.
• Looped Diffusion Language Models — Sanghyun Lee et al. arXiv, 2026.

B3. Supervised, Input-Dependent (T(n))

Loop count is supervised as a function of the input/problem size.

• Looped Transformers for Length Generalization — Ying Fan, Yilun Du, Kannan Ramchandran, Kangwook Lee. ICLR 2025.
• Universal Transformers for Circuit Computations: Perfect Length Generalization in Tiny Transformers — Takuya Ito, Ruchir Puri, Murray Campbell, Parikshit Ram. arXiv, 2026.

B4. Learned Online Halting

The stopping decision is made while recurrence is running.

• Adaptive Computation Time for Recurrent Neural Networks — Alex Graves. arXiv, 2016.
• Universal Transformers — Mostafa Dehghani et al. ICLR 2019.
• PonderNet: Learning to Ponder — Andrea Banino, Jan Balaguer, Charles Blundell. arXiv, 2021.
• Sparse Universal Transformer — Shawn Tan et al. EMNLP 2023.
• Scaling Latent Reasoning via Looped Language Models — Rui-Jie Zhu et al. arXiv, 2025.
• AdaPonderLM — Shixiang Song et al. arXiv, 2026.
• LoopUS — Taekhyun Park et al. arXiv, 2026.
• CHASE: Cache-Hole-Adapted Skip Exit for Looped State-Space Language Models — Zhenxuan Yu et al. arXiv, 2026.

B5. Learned Up-Front Routing

Depth is allocated before recurrent computation begins.

• CoTFormer: A Chain of Thought Driven Architecture with Budget-Adaptive Computation Cost at Inference — Amirkeivan Mohtashami, Matteo Pagliardini, Martin Jaggi. ICLR 2025.
• Inner Thinking Transformer — Yilong Chen et al. ACL 2025.
• Mixture-of-Recursions — Sangmin Bae et al. NeurIPS 2025.
• Skip a Layer or Loop It? Test-Time Depth Adaptation of Pretrained LLMs — Ziyue Li, Yang Li, Tianyi Zhou. arXiv, 2025.
• Think-at-Hard: Dynamic Looped Transformers for Improved Reasoning — Tianyu Fu et al. arXiv, 2025.
• Skip a Layer or Loop It? Learning Program-of-Layers in LLMs — Ziyue Li, Yang Li, Tianyi Zhou. arXiv, 2026.

B6. Convergence Test / Training-Free Halting

The model stops when hidden states or outputs satisfy a convergence criterion.

• Deep Equilibrium Models — Shaojie Bai, J. Zico Kolter, Vladlen Koltun. NeurIPS 2019.
• Scaling up Test-Time Compute with Latent Reasoning — Jonas Geiping et al. NeurIPS 2025.
• Two-Scale Latent Dynamics for Recurrent-Depth Transformers — Francesco Pappone, Donato Crisostomi, Emanuele Rodolà. arXiv, 2025.
• Fixed-Point Reasoners — Sajad Movahedi et al. arXiv, 2026.
• Per-Token Fixed-Point Convergence in Depth-Recurrent Transformers — Joe Logan. arXiv, 2026.
• Think Shallow, Solve Deep: Controlling Recurrent Dynamics for Reliable Test-Time Depth — Ivan Viakhirev et al. arXiv, 2026.

B7. RL / Oracle-Supervised Stopping

Stopping policies are trained with reinforcement learning or oracle supervision.

• Hierarchical Reasoning Model — Guan Wang et al. arXiv, 2025.
• Stabilizing Extrapolation in Looped Transformers via Learned Stochastic Stopping — Hsun-Yu Kuo et al. arXiv, 2026.
• RecurTrace: Adaptive Latent Reasoning with Loop-Time Memory — Yuxiang Wang et al. arXiv, 2026.

---

C. Inter-Iteration State

C1. Pure Recurrence

Only the recurrent hidden state is passed forward.

• Universal Transformers — Mostafa Dehghani et al. ICLR 2019.
• ALBERT — Zhenzhong Lan et al. ICLR 2020.
• MobileLLM — Zechun Liu et al. ICML 2024.
• MoEUT — Róbert Csordás et al. NeurIPS 2024.
• Relaxed Recursive Transformers — Sangmin Bae et al. ICLR 2025.
• Ouro / Scaling Latent Reasoning via Looped Language Models — Rui-Jie Zhu et al. arXiv, 2025.

C2. Input Injection / Recall

The original input is re-injected across recurrent steps.

• End-to-End Algorithm Synthesis with Recurrent Networks — Arpit Bansal et al. arXiv, 2022.
• Looped Transformers Are Better at Learning Learning Algorithms — Liu Yang et al. ICLR 2024.
• Looped Transformers for Length Generalization — Ying Fan et al. ICLR 2025.
• Scaling up Test-Time Compute with Latent Reasoning — Jonas Geiping et al. NeurIPS 2025.
• Hierarchical Reasoning Model — Guan Wang et al. arXiv, 2025.
• Less Is More: Recursive Reasoning with Tiny Networks — Alexia Jolicoeur-Martineau. arXiv, 2025.
• Simply Stabilizing the Loop via Fully Looped Transformer — Rao Fu et al. arXiv, 2026.
• Bridging the Gap Between Latent and Explicit Reasoning with Looped Transformers — Ying Fan, Anej Svete, Kangwook Lee. arXiv, 2026.

C3. Predicted-Embedding Feedback

A predicted embedding is fed into the next recurrent iteration.

• PonderLM: Pretraining Language Models to Ponder in Continuous Space — Boyi Zeng et al. ICLR 2026.
• DiscoLoop: Looping Discrete Embeddings and Continuous Hidden States for Multi-Hop Reasoning — Hengyu Fu et al. arXiv, 2026.

C4. Attention over Earlier Iterations

Later recurrent steps directly attend to earlier iterations.

• CoTFormer — Amirkeivan Mohtashami, Matteo Pagliardini, Martin Jaggi. ICLR 2025.
• Think-at-Hard — Tianyu Fu et al. arXiv, 2025.
• Depth-Recurrent Attention Mixtures: Giving Latent Reasoning the Attention It Deserves — Jonas Knupp et al. arXiv, 2026.
• RecurTrace — Yuxiang Wang et al. arXiv, 2026.

C5. Fixed Point / Implicit

The recurrent computation is defined through an equilibrium/fixed point.

• Deep Equilibrium Models — Shaojie Bai, J. Zico Kolter, Vladlen Koltun. NeurIPS 2019.
• Implicit Language Models Are RNNs: Balancing Parallelization and Expressivity — Mark Schöne et al. arXiv, 2025.
• Closed-Loop Transformers: Autoregressive Modeling as Iterative Latent Equilibrium — Akbar Anbar Jafari, Gholamreza Anbarjafari. arXiv, 2025.
• Fixed-Point Reasoners — Sajad Movahedi et al. arXiv, 2026.

---

D. Sharing Strictness

D1. Exact Tie, No Conditioning

The same parameters are reused exactly across recurrent iterations.

• ALBERT — Zhenzhong Lan et al. ICLR 2020.
• Scaling up Test-Time Compute with Latent Reasoning — Jonas Geiping et al. NeurIPS 2025.
• Scaling Latent Reasoning via Looped Language Models — Rui-Jie Zhu et al. arXiv, 2025.
• Nanbeige4.2-3B — Nanbeige Lab et al. arXiv, 2026.

D2. Timestep / Depth Encoding

The shared recurrent block is conditioned on the current loop/depth index.

• Universal Transformers — Mostafa Dehghani et al. ICLR 2019.
• On Expressive Power of Looped Transformers: Theoretical Analysis and Enhancement via Timestep Encoding — Kevin Xu, Issei Sato. arXiv, 2024.
• Inner Thinking Transformer — Yilong Chen et al. ACL 2025.
• LoopFormer — Ahmadreza Jeddi et al. arXiv, 2026.
• Universal Transformers for Circuit Computations — Takuya Ito et al. arXiv, 2026.

D3. Per-Loop Scalars, Gates, or Norms

Small loop-specific parameters modulate the shared computation.

• Loop Neural Networks for Parameter Sharing — Kei-Sing Ng, Qingchen Wang. arXiv, 2024.
• Sliced Recursive Transformer — Zhiqiang Shen, Zechun Liu, Eric Xing. arXiv, 2021.
• LoopMoE: Unifying Iterative Computation with Mixture-of-Experts for Language Modeling — Wenkai Chen et al. arXiv, 2026.

D4. Per-Loop Low-Rank Adapters

Iteration-specific low-rank adapters relax exact weight sharing.

• Relaxed Recursive Transformers — Sangmin Bae et al. ICLR 2025.
• Think-at-Hard — Tianyu Fu et al. arXiv, 2025.
• Retrofitting Recurrent Depth into a Pretrained Language Model: Installation, Extrapolation, Transfer, and Retention at Two Parameter Budgets — Mark Shapiro. arXiv, 2026.

D5. Experts Specializing per Pass

Sparse experts can specialize across recurrent iterations.

• Sparse Universal Transformer — Shawn Tan et al. EMNLP 2023.
• MoEUT — Róbert Csordás et al. NeurIPS 2024.
• Mixture of Universal Experts: Scaling Virtual Width via Depth-Width Transformation — Yilong Chen et al. arXiv, 2026.
• Sparse Layers Are Critical to Scaling Looped Language Models — Ryan Lee et al. arXiv, 2026.
• LoopMoE — Wenkai Chen et al. arXiv, 2026.
• Loop the Loopies! — Zitian Gao et al. arXiv, 2026.
• SMELT — Shaowen Wang et al. arXiv, 2026.

---

E. Training Recipe

E1. Provenance

From Scratch

• Scaling up Test-Time Compute with Latent Reasoning — Jonas Geiping et al. NeurIPS 2025.
• Scaling Latent Reasoning via Looped Language Models — Rui-Jie Zhu et al. arXiv, 2025.
• Nanbeige4.2-3B — Nanbeige Lab et al. arXiv, 2026.
• Loop the Loopies! — Zitian Gao et al. arXiv, 2026.
• LoopCoder-v2 — Jian Yang et al. arXiv, 2026.

Uptrained / Retrofitted

• Relaxed Recursive Transformers — Sangmin Bae et al. ICLR 2025.
• Encode, Think, Decode — Yeskendir Koishekenov et al. arXiv, 2025.
• Teaching Pretrained Language Models to Think Deeper with Retrofitted Recurrence — Sean McLeish et al. arXiv, 2025.
• Think-at-Hard — Tianyu Fu et al. arXiv, 2025.
• LoopUS — Taekhyun Park et al. arXiv, 2026.

Training-Free

• Skip a Layer or Loop It? Test-Time Depth Adaptation of Pretrained LLMs — Ziyue Li, Yang Li, Tianyi Zhou. arXiv, 2025.
• Training-Free Looped Transformers — Lizhang Chen et al. arXiv, 2026.
• Recirculation — Michael C. Mozer et al. arXiv, 2026.

E2. Gradient Path

Full BPTT

• Less Is More: Recursive Reasoning with Tiny Networks — Alexia Jolicoeur-Martineau. arXiv, 2025.
• Scaling Latent Reasoning via Looped Language Models — Rui-Jie Zhu et al. arXiv, 2025.

Truncated BPTT

• Scaling up Test-Time Compute with Latent Reasoning — Jonas Geiping et al. NeurIPS 2025.
• Universal Reasoning Model — Zitian Gao et al. arXiv, 2025.

One-Step Approximation

• Hierarchical Reasoning Model — Guan Wang et al. arXiv, 2025.

Implicit Differentiation

• Deep Equilibrium Models — Shaojie Bai, J. Zico Kolter, Vladlen Koltun. NeurIPS 2019.

Activation-Compressed BPTT

• LASER: Low-Rank Activation SVD for Efficient Recursion — Ege Çakar, Ketan Raghu, Lia Zheng. arXiv, 2026.

E3. Loop Schedule

Random Unrolling / Curriculum

• End-to-End Algorithm Synthesis with Recurrent Networks — Arpit Bansal et al. arXiv, 2022.
• Looped Transformers Are Better at Learning Learning Algorithms — Liu Yang et al. ICLR 2024.
• Teaching Pretrained Language Models to Think Deeper with Retrofitted Recurrence — Sean McLeish et al. arXiv, 2025.

Shortcut Consistency

• LoopFormer — Ahmadreza Jeddi et al. arXiv, 2026.

Deep Supervision

• HRM-Text: Efficient Pretraining Beyond Scaling — Guan Wang et al. arXiv, 2026.

RL over the Latent Trajectory

• Prioritize the Process, Not Just the Outcome: Rewarding Latent Thought Trajectories Improves Reasoning in Looped Language Models — Jonathan Williams, Esin Tureci. arXiv, 2026.
• LoopRPT: Reinforcement Pretraining for Looped Language Models — Guo Tang et al. arXiv, 2026.

E4. Stability Machinery

• On the Residual Scaling of Looped Transformers: Stability and Transferability — Shaowen Wang et al. arXiv, 2026.
• DeepLoop: Depth Scaling for Looped Transformers — Shuzhen Li, Yifan Zhang, Jiacheng Guo, Quanquan Gu, Mengdi Wang. arXiv, 2026.
• Stabilizing Recurrent Dynamics for Test-Time Scalable Latent Reasoning in Looped Language Models — Xiao-Wen Yang et al. arXiv, 2026.
• Simply Stabilizing the Loop via Fully Looped Transformer — Rao Fu et al. arXiv, 2026.
• Stability and Generalization in Looped Transformers — Asher Labovich. arXiv, 2026.

---

F. Memory / KV Strategy

F1. Separate Cache per (Layer, Loop)

A separate KV cache is retained for every loop.

• Scaling up Test-Time Compute with Latent Reasoning — Jonas Geiping et al. NeurIPS 2025.
• Scaling Latent Reasoning via Looped Language Models — Rui-Jie Zhu et al. arXiv, 2025.
• Nanbeige4.2-3B — Nanbeige Lab et al. arXiv, 2026.

F2. Share First-Loop KV

KV states from the first recurrence are reused.

• Mixture-of-Recursions — Sangmin Bae et al. NeurIPS 2025.
• Parallel Loop Transformer — Bohong Wu et al. arXiv, 2025.
• Scaling Latent Reasoning via Looped Language Models — Rui-Jie Zhu et al. arXiv, 2025.

F3. Active-Token / Gated Single Cache

Only active token states are stored or one cache is updated with a gate.

• Mixture-of-Recursions — Sangmin Bae et al. NeurIPS 2025.
• AdaPonderLM — Shixiang Song et al. arXiv, 2026.
• Memory-Efficient Looped Transformer — Victor Conchello Vendrell et al. arXiv, 2026.

F4. Compressed or Sub-Quadratic

The recurrent KV/attention representation is compressed or replaced by a more memory-efficient mechanism.

• Looped Latent Attention: Cross-Loop KV Compression for Looped Transformers — James O'Neill, Fergal Reid. arXiv, 2026.
• LoopQ: Quantization for Recursive Transformers — Rui Fang, Hsi-Wen Chen, Ming-Syan Chen. arXiv, 2026.
• LT²: Linear-Time Looped Transformers — Chunyuan Deng et al. arXiv, 2026.
• Allocating Recurrent Compute in Looped Language Models — Ruhai Lin et al. arXiv, 2026.

F5. Serving-Level Scheduling

The optimization is implemented at the inference/serving scheduler level.

• Relaxed Recursive Transformers: Effective Parameter Sharing with Layer-Wise LoRA — Sangmin Bae et al. ICLR 2025.
• Efficient Parallel Samplers for Recurrent-Depth Models and Their Connection to Diffusion Language Models — Jonas Geiping, Xinyu Yang, Guinan Su. arXiv, 2025.
• Depth-Adaptive Inference of Looped Language Models via Continuous Depth Batching — Kristian Schwethelm, Daniel Rueckert, Georgios Kaissis. arXiv, 2026.

---

Contributing

Contributions are welcome. Please keep new entries consistent with the format:

- **[Paper Title](paper-link)** — Author 1, Author 2, et al. *Venue, Year.*