pith. sign in

arxiv: 2605.26106 · v1 · pith:34TMSD6Dnew · submitted 2026-05-25 · 💻 cs.LG

Looped Diffusion Language Models

Sanghyun Lee , Chunsan Hong , Seungryong Kim , Jonghyun Lee , Jongho Park , Dongmin Park This is my paper

Pith reviewed 2026-06-29 23:10 UTC · model grok-4.3

classification 💻 cs.LG
keywords masked diffusion modelslooped transformerslanguage modelingtraining efficiencyinference-time scalingreasoning benchmarksattention analysis
0
0 comments X

The pith

Selectively looping early-middle transformer layers in masked diffusion models yields depth scaling without added parameters and matches performance with up to 3.3 times fewer training FLOPs.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces LoopMDM, which loops selected early-middle layers inside the transformer blocks of masked diffusion models during both training and inference. This produces an effective increase in model depth at training time without increasing parameter count, while the number of loops can be varied at inference to trade compute for quality. Across pre-training runs the method reaches the accuracy of standard MDMs while using substantially less total training compute and then exceeds them on downstream reasoning tasks. The central mechanism is shown through attention maps to increase interactions among masked token positions.

Core claim

LoopMDM selectively loops the early-middle transformer layers of masked diffusion models. At training time the repeated application of those layers creates a depth-scaling effect with no extra parameters; at inference time the loop count can be increased or adapted on the fly. The resulting models match the performance of same-size non-looped MDMs with up to 3.3 times fewer training FLOPs, surpass them on reasoning benchmarks including an 8.5-point gain on GSM8K, and outperform deeper non-looped MDMs trained with comparable per-step compute. Attention analysis indicates that the looping promotes interactions among masked positions.

What carries the argument

Selective looping of early-middle transformer layers inside the MDM architecture, applied repeatedly during the forward pass to produce depth scaling.

If this is right

  • LoopMDM reaches the same pre-training loss as a standard MDM while consuming up to 3.3 times fewer total training FLOPs.
  • Final models outperform same-size MDMs on multiple reasoning benchmarks, with gains reaching 8.5 points on GSM8K.
  • Increasing the number of loops at inference time scales compute flexibly without retraining.
  • Adaptive loop counts during sampling improve compute efficiency while preserving accuracy.
  • The approach outperforms naive increases in transformer depth when total per-step compute is held constant.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same selective-loop pattern could be tested on autoregressive transformers or other diffusion objectives to check whether the masked-position interaction benefit is specific to MDMs.
  • If the attention-map explanation holds, similar gains might appear in any masked modeling task where early layers primarily handle local context.
  • Because loop count is adjustable after training, the method supplies a practical knob for trading latency against quality on a single set of weights.

Load-bearing premise

The observed gains arise specifically from looping the early-middle layers rather than from any unstated differences in training procedure, data, or hyperparameter choices.

What would settle it

Train a non-looped MDM using identical data, optimizer schedule, and every other hyperparameter as the LoopMDM run; if its final performance on GSM8K and training-FLOP efficiency match or exceed the looped version, the benefit cannot be attributed to the looping itself.

Figures

Figures reproduced from arXiv: 2605.26106 by Chunsan Hong, Dongmin Park, Jongho Park, Jonghyun Lee, Sanghyun Lee, Seungryong Kim.

Figure 1
Figure 1. Figure 1: Overview of LoopMDM. (Left) LoopMDM selectively applies looping to a small early￾middle layers of the denoising network. (Middle) Under matched training compute, LoopMDM (red) reaches the same test NLL as a non-looped MDM baseline with the same architecture (dashed black) using substantially fewer training FLOPs. (Right) Increasing the inference-time loop count consistently improves GSM8K accuracy; the das… view at source ↗
Figure 2
Figure 2. Figure 2: Test NLL across language pre-training datasets. Test NLL as a function of training FLOPs on FineWeb-Edu, OpenWebText (OWT), and LM1B. All models are iso-parameter (170M) and trained under matched training FLOPs. Looping is applied to two mid-layers 1-2 in zero-based indexing. Solid curves show LoopMDM with varying inference-time loop counts (S = 1, 6, 12, 24), where S = 24 exceeds the maximum loop count us… view at source ↗
Figure 3
Figure 3. Figure 3: GSM8K accuracy as a function of training FLOPs. 14 layers LoopMDM (solid) is compared against MDM baselines with 14, 18, and 21 layers (dashed); the 21-layer baseline is sized so that its per-step training FLOPs approximately match those of LoopMDM. Inference-time loop counts S ∈ {1, 2, 4, 6, 8, 16} are shown, with S = 16 exceeding the training maximum. Results are reported under both Top-2 (left) and Top-… view at source ↗
Figure 4
Figure 4. Figure 4: Looping recovers global consistency under a restricted generation order. While MDMs can solve Sudoku using adaptive unmasking, we remove this advantage by enforcing a fixed left-to-right (autoregressive-style) order, where early predictions are made with incomplete context. This isolates the role of within-step computation, since improvements can no longer come from selecting easier cells first. We show pr… view at source ↗
Figure 5
Figure 5. Figure 5: Analysis of looping behavior. (Left) Average mask-to-mask attention at timestep t = 0.5 for a model with nm = 2, measured at the two mid-block layers (denoted mid[0] and mid[1]) as a function of loop counts S. Each curve shows attention at the corresponding layer after S loop applications. Attention increases with S and saturates near the training maximum Smax = 12 (dashed line). (Right) NLL improvement NL… view at source ↗
Figure 6
Figure 6. Figure 6: Adaptive allocation of loop counts across timesteps. Average loop count allocated by the adaptive strategy as a function of timestep t on OpenWebText with ϵ = 0.1, measured over 100 sampled sequences. (Left) Zero-shot perplexity evaluation on WikiText, Lambada, and PTB. (Right) Generative perplexity evaluation. Across both settings, the adaptive strategy allocates more iterations at intermediate timesteps … view at source ↗
Figure 7
Figure 7. Figure 7: NLL comparison with alter￾native looping strategies. Compare with log-normal Poisson loop sampling. Re￾cent looped transformers sample loop counts from a log￾normal Poisson distribution to stabilize recurrent computa￾tion and improve test-time scaling [19, 41]. We compare this strategy against the uniform loop-count sampling used in LoopMDM [PITH_FULL_IMAGE:figures/full_fig_p021_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Per-token loop trajectories across iterations. Each panel tracks the predicted token (pred) and its NLL at the focal position as the number of loop counts S increases. Lower NLL indicates higher model confidence. (A–D) show successful refinement where predictions evolve toward the ground-truth token (GT) via different mechanisms: (A) local copying, (B) semantic selection, (C) syntactic refinement, and (D) … view at source ↗
read the original abstract

Masked diffusion models (MDMs) have emerged as a promising alternative to autoregressive models for language modeling, yet the effective design of transformer architectures for MDMs remains underexplored. In this paper, we show that selectively looping the early-middle transformer layers significantly improves both training efficiency and model performance in MDMs. We call this approach LoopMDM(Looped Masked Diffusion Model), which brings two key benefits: looping layers at training-time yields a depth-scaling effect without adding parameters, while varying the number of loops at inference-time enables flexible compute scaling. Despite the simplicity, the results are striking: across multiple pre-training corpora, LoopMDM matches the performance of same-size MDMs with up to 3.3 fewer training FLOPs, while its final performance outperforms them on various reasoning benchmarks, including up to 8.5 points on GSM8K. It even surpasses deeper non-looped MDMs trained with comparable per-step compute, indicating that selective looping is more effective than naive depth scaling. Furthermore, LoopMDM can scale inference-time compute by increasing the number of loops. Adaptively adjusting the number of loops throughout the sampling process further yields additional gains in compute efficiency while maintaining performance. Lastly, with attention analysis, we provide evidence that looping is effective in MDMs by promoting interactions among masked positions. Our code and weights will be publicly released.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The paper proposes LoopMDM, which selectively loops early-middle transformer layers in masked diffusion models (MDMs). This is claimed to provide a depth-scaling effect without added parameters at training time and flexible compute scaling at inference by varying loop count. Across pre-training corpora, it matches same-size MDMs with up to 3.3x fewer training FLOPs, outperforms on reasoning benchmarks (up to +8.5 on GSM8K), surpasses deeper non-looped MDMs at equal per-step compute, enables further gains via adaptive looping, and is supported by attention analysis indicating promoted masked-position interactions.

Significance. If the gains are attributable to selective looping under controlled conditions, the method would offer a simple, parameter-efficient route to improved training efficiency and inference flexibility in MDMs, outperforming naive depth scaling and enabling adaptive compute.

major comments (2)
  1. [Abstract] Abstract: The abstract reports concrete gains versus same-size and deeper baselines, but provides no details on experimental controls, statistical significance, or exact training configurations, leaving the central claim only partially supported from available text.
  2. [Abstract] Abstract: The attention analysis is presented as supporting evidence for promoted masked-position interactions, but remains post-hoc correlation without an ablation that severs the loop while preserving the observed attention pattern.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address the two major comments below. The manuscript provides full experimental details in the body, but we agree the abstract can be clarified for better support of the claims.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The abstract reports concrete gains versus same-size and deeper baselines, but provides no details on experimental controls, statistical significance, or exact training configurations, leaving the central claim only partially supported from available text.

    Authors: The abstract is a concise summary constrained by length limits. Full details on experimental controls (model sizes, training corpora, hyperparameters, and per-step compute matching), statistical significance (multiple random seeds with standard deviations), and configurations appear in Section 3 and the results tables. The central claims are supported by those sections and tables. We will revise the abstract to include a short clause noting 'under matched training configurations and multiple runs' to address this. revision: partial

  2. Referee: [Abstract] Abstract: The attention analysis is presented as supporting evidence for promoted masked-position interactions, but remains post-hoc correlation without an ablation that severs the loop while preserving the observed attention pattern.

    Authors: We acknowledge the analysis in Section 4.3 is correlational. The primary evidence for the method's effectiveness comes from controlled performance comparisons (looped vs. non-looped models at equal parameters and compute). We will add explicit discussion of this limitation and note that a targeted ablation preserving the attention pattern while removing loops is non-trivial to design, as looping directly modifies the computation graph. This will be framed as a direction for future work rather than claiming full causality from attention alone. revision: partial

Circularity Check

0 steps flagged

No significant circularity; results are empirical comparisons

full rationale

The paper introduces LoopMDM as an architectural modification to masked diffusion models and supports its claims exclusively through direct empirical benchmarks against non-looped baselines on pre-training corpora and downstream tasks such as GSM8K. No equations, uniqueness theorems, fitted parameters relabeled as predictions, or self-citation chains appear in the derivation of the performance gains; the attention analysis is presented as post-hoc supporting evidence rather than a load-bearing logical step. The central assertions therefore rest on falsifiable experimental outcomes rather than any reduction to inputs by construction.

Axiom & Free-Parameter Ledger

1 free parameters · 0 axioms · 0 invented entities

The central claim rests on the empirical effectiveness of a chosen architectural modification whose benefits are measured against baselines; no new physical or mathematical axioms are introduced.

free parameters (1)
  • number of loops
    Hyperparameter controlling how many times early-middle layers are reused; its value is selected rather than derived.

pith-pipeline@v0.9.1-grok · 5785 in / 1118 out tokens · 28445 ms · 2026-06-29T23:10:53.646850+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

82 extracted references · 45 canonical work pages · 23 internal anchors

  1. [1]

    Path independent equilibrium models can better exploit test-time computation.Advances in Neural Information Processing Systems, 35:7796–7809, 2022

    Cem Anil, Ashwini Pokle, Kaiqu Liang, Johannes Treutlein, Yuhuai Wu, Shaojie Bai, J Zico Kolter, and Roger B Grosse. Path independent equilibrium models can better exploit test-time computation.Advances in Neural Information Processing Systems, 35:7796–7809, 2022

    2022

  2. [2]

    Structured denoising diffusion models in discrete state-spaces.Advances in neural information processing systems, 34:17981–17993, 2021

    Jacob Austin, Daniel D Johnson, Jonathan Ho, Daniel Tarlow, and Rianne Van Den Berg. Structured denoising diffusion models in discrete state-spaces.Advances in neural information processing systems, 34:17981–17993, 2021

    2021

  3. [3]

    Mixture-of-recursions: Learning dynamic recursive depths for adaptive token-level computation.arXiv preprint arXiv:2507.10524, 2025

    Sangmin Bae, Yujin Kim, Reza Bayat, Sungnyun Kim, Jiyoun Ha, Tal Schuster, Adam Fisch, Hrayr Harutyunyan, Ziwei Ji, Aaron Courville, et al. Mixture-of-recursions: Learning dynamic recursive depths for adaptive token-level computation.arXiv preprint arXiv:2507.10524, 2025

    work page arXiv 2025

  4. [4]

    Tiwei Bie, Maosong Cao, Kun Chen, Lun Du, Mingliang Gong, Zhuochen Gong, Yanmei Gu, Jiaqi Hu, Zenan Huang, Zhenzhong Lan, et al. Llada2. 0: Scaling up diffusion language models to 100b.arXiv preprint arXiv:2512.15745, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  5. [5]

    $S^3$: Stratified Scaling Search for Test-Time in Diffusion Language Models

    Ahsan Bilal, Muhammad Ahmed Mohsin, Muhammad Umer, Asad Aali, Muhammad Usman Khanzada, Muhammad Usman Rafique, Zihao He, Emily Fox, and Dean F Hougen. s3: Strati- fied scaling search for test-time in diffusion language models.arXiv preprint arXiv:2604.06260, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  6. [6]

    Piqa: Reasoning about phys- ical commonsense in natural language

    Yonatan Bisk, Rowan Zellers, Jianfeng Gao, Yejin Choi, et al. Piqa: Reasoning about phys- ical commonsense in natural language. InProceedings of the AAAI conference on artificial intelligence, volume 34, pages 7432–7439, 2020

    2020

  7. [7]

    Be- yond masked and unmasked: Discrete diffusion models via partial masking.arXiv preprint arXiv:2505.18495, 2025

    Chen-Hao Chao, Wei-Fang Sun, Hanwen Liang, Chun-Yi Lee, and Rahul G Krishnan. Be- yond masked and unmasked: Discrete diffusion models via partial masking.arXiv preprint arXiv:2505.18495, 2025

    work page arXiv 2025

  8. [8]

    One billion word benchmark for measuring progress in statistical lan- guage modeling

    Ciprian Chelba, Tomas Mikolov, Mike Schuster, Qi Ge, Thorsten Brants, Phillipp Koehn, and Tony Robinson. One billion word benchmark for measuring progress in statistical lan- guage modeling. InProceedings of the 15th Annual Conference of the International Speech Communication Association (INTERSPEECH), pages 615–621, 2014

    2014

  9. [9]

    Boolq: Exploring the surprising difficulty of natural yes/no questions

    Christopher Clark, Kenton Lee, Ming-Wei Chang, Tom Kwiatkowski, Michael Collins, and Kristina Toutanova. Boolq: Exploring the surprising difficulty of natural yes/no questions. InProceedings of the 2019 conference of the north American chapter of the association for computational linguistics: Human language technologies, volume 1 (long and short papers), ...

    2019

  10. [10]

    Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge

    Peter Clark, Isaac Cowhey, Oren Etzioni, Tushar Khot, Ashish Sabharwal, Carissa Schoenick, and Oyvind Tafjord. Think you have solved question answering? try arc, the ai2 reasoning challenge.arXiv preprint arXiv:1803.05457, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  11. [11]

    Training Verifiers to Solve Math Word Problems

    Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, Christopher Hesse, and John Schulman. Training verifiers to solve math word problems.arXiv preprint arXiv:2110.14168, 2021

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  12. [12]

    A discourse-aware attention model for abstractive summarization of long documents

    Arman Cohan, Franck Dernoncourt, Doo Soon Kim, Trung Bui, Seokhwan Kim, Walter Chang, and Nazli Goharian. A discourse-aware attention model for abstractive summarization of long documents. InProceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 2 (Short Papers...

    2018

  13. [13]

    Recurrent stacking of layers for compact neural machine translation models

    Raj Dabre and Atsushi Fujita. Recurrent stacking of layers for compact neural machine translation models. InProceedings of the AAAI Conference on Artificial Intelligence, volume 33, pages 6292–6299, 2019. 10

    2019

  14. [14]

    Inference-Time Scaling of Diffusion Language Models via Trajectory Refinement

    Meihua Dang, Jiaqi Han, Minkai Xu, Kai Xu, Akash Srivastava, and Stefano Ermon. Inference- time scaling of diffusion language models with particle gibbs sampling.arXiv preprint arXiv:2507.08390, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  15. [15]

    Universal Transformers

    Mostafa Dehghani, Stephan Gouws, Oriol Vinyals, Jakob Uszkoreit, and Łukasz Kaiser. Uni- versal transformers.arXiv preprint arXiv:1807.03819, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  16. [16]

    Implicit chain of thought reasoning via knowledge distillation.arXiv preprint arXiv:2311.01460, 2023

    Yuntian Deng, Kiran Prasad, Roland Fernandez, Paul Smolensky, Vishrav Chaudhary, and Stuart Shieber. Implicit chain of thought reasoning via knowledge distillation.arXiv preprint arXiv:2311.01460, 2023

    work page arXiv 2023

  17. [17]

    Partition generative modeling: Masked modeling without masks.arXiv preprint arXiv:2505.18883, 2025

    Justin Deschenaux, Lan Tran, and Caglar Gulcehre. Partition generative modeling: Masked modeling without masks.arXiv preprint arXiv:2505.18883, 2025

    work page arXiv 2025

  18. [18]

    Looped transformers for length generalization.arXiv preprint arXiv:2409.15647, 2024

    Ying Fan, Yilun Du, Kannan Ramchandran, and Kangwook Lee. Looped transformers for length generalization.arXiv preprint arXiv:2409.15647, 2024

    work page arXiv 2024

  19. [19]

    Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach

    Jonas Geiping, Sean McLeish, Neel Jain, John Kirchenbauer, Siddharth Singh, Brian R Bartold- son, Bhavya Kailkhura, Abhinav Bhatele, and Tom Goldstein. Scaling up test-time compute with latent reasoning: A recurrent depth approach.arXiv preprint arXiv:2502.05171, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  20. [20]

    Looped transformers as programmable computers

    Angeliki Giannou, Shashank Rajput, Jy-yong Sohn, Kangwook Lee, Jason D Lee, and Dimitris Papailiopoulos. Looped transformers as programmable computers. InInternational Conference on Machine Learning, pages 11398–11442. PMLR, 2023

    2023

  21. [21]

    Openwebtext corpus

    Aaron Gokaslan and Vanya Cohen. Openwebtext corpus. http://Skylion007.github.io/ OpenWebTextCorpus, 2019

    2019

  22. [22]

    Diffucoder: Understanding and improving masked diffusion models for code generation

    Shansan Gong, Ruixiang ZHANG, Huangjie Zheng, Jiatao Gu, Navdeep Jaitly, Lingpeng Kong, and Yizhe Zhang. Diffucoder: Understanding and improving masked diffusion models for code generation. InThe Fourteenth International Conference on Learning Representations, 2026

    2026

  23. [23]

    Think before you speak: Training language models with pause tokens

    Sachin Goyal, Ziwei Ji, Ankit Singh Rawat, Aditya Krishna Menon, Sanjiv Kumar, and Vaishnavh Nagarajan. Think before you speak: Training language models with pause tokens. In The Twelfth International Conference on Learning Representations, 2024

    2024

  24. [24]

    Training Large Language Models to Reason in a Continuous Latent Space

    Shibo Hao, Sainbayar Sukhbaatar, DiJia Su, Xian Li, Zhiting Hu, Jason Weston, and Yuandong Tian. Training large language models to reason in a continuous latent space.arXiv preprint arXiv:2412.06769, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  25. [25]

    Reasoning with latent tokens in diffusion language models.arXiv preprint arXiv:2602.03769, 2026

    Andre He, Sean Welleck, and Daniel Fried. Reasoning with latent tokens in diffusion language models.arXiv preprint arXiv:2602.03769, 2026

    work page arXiv 2026

  26. [26]

    Diffusionbert: Improving generative masked language models with diffusion models

    Zhengfu He, Tianxiang Sun, Qiong Tang, Kuanning Wang, Xuan-Jing Huang, and Xipeng Qiu. Diffusionbert: Improving generative masked language models with diffusion models. In Proceedings of the 61st annual meeting of the association for computational linguistics (ACL), volume 1, pages 4521–4534, 2023

    2023

  27. [27]

    Unifying Masked Diffusion Models with Various Generation Orders and Beyond

    Chunsan Hong, Sanghyun Lee, and Jong Chul Ye. Unifying masked diffusion models with various generation orders and beyond.arXiv preprint arXiv:2602.02112, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  28. [28]

    Argmax flows and multinomial diffusion: Learning categorical distributions

    Emiel Hoogeboom, Didrik Nielsen, Priyank Jaini, Patrick Forré, and Max Welling. Argmax flows and multinomial diffusion: Learning categorical distributions. volume 34, pages 12454– 12465, 2021

    2021

  29. [29]

    Accelerating diffusion llms via adaptive parallel decoding.arXiv preprint arXiv:2506.00413,

    Daniel Israel, Guy Van den Broeck, and Aditya Grover. Accelerating diffusion llms via adaptive parallel decoding.arXiv preprint arXiv:2506.00413, 2025

    work page arXiv 2025

  30. [30]

    Scaling Laws for Neural Language Models

    Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B Brown, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei. Scaling laws for neural language models.arXiv preprint arXiv:2001.08361, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2001

  31. [31]

    Stop Training for the Worst: Progressive Unmasking Accelerates Masked Diffusion Training

    Jaeyeon Kim, Jonathan Geuter, David Alvarez-Melis, Sham Kakade, and Sitan Chen. Stop training for the worst: Progressive unmasking accelerates masked diffusion training.arXiv preprint arXiv:2602.10314, 2026. 11

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  32. [32]

    Jaeyeon Kim, Seunggeun Kim, Taekyun Lee, David Z

    Jaeyeon Kim, Kulin Shah, Vasilis Kontonis, Sham Kakade, and Sitan Chen. Train for the worst, plan for the best: Understanding token ordering in masked diffusions.arXiv preprint arXiv:2502.06768, 2025

    work page arXiv 2025

  33. [33]

    Race: Large-scale reading comprehension dataset from examinations

    Guokun Lai, Qizhe Xie, Hanxiao Liu, Yiming Yang, and Eduard Hovy. Race: Large-scale reading comprehension dataset from examinations. InProceedings of the 2017 conference on empirical methods in natural language processing, pages 785–794, 2017

    2017

  34. [34]

    ALBERT: A Lite BERT for Self-supervised Learning of Language Representations

    Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, and Radu Soricut. Albert: A lite bert for self-supervised learning of language representations.arXiv preprint arXiv:1909.11942, 2019

    work page internal anchor Pith review Pith/arXiv arXiv 1909

  35. [35]

    Lookahead unmasking elicits accurate decoding in diffusion language models.arXiv preprint arXiv:2511.05563,

    Sanghyun Lee, Seungryong Kim, Jongho Park, and Dongmin Park. Lookahead unmasking elicits accurate decoding in diffusion language models.arXiv preprint arXiv:2511.05563, 2025

    work page arXiv 2025

  36. [36]

    Effective test- time scaling of discrete diffusion through iterative refinement.arXiv preprint arXiv:2511.05562, 2025

    Sanghyun Lee, Sunwoo Kim, Seungryong Kim, Jongho Park, and Dongmin Park. Effective test- time scaling of discrete diffusion through iterative refinement.arXiv preprint arXiv:2511.05562, 2025

    work page arXiv 2025

  37. [37]

    Tinygsm: achieving> 80% on gsm8k with small language models

    Bingbin Liu, Sebastien Bubeck, Ronen Eldan, Janardhan Kulkarni, Yuanzhi Li, Anh Nguyen, Rachel Ward, and Yi Zhang. Tinygsm: achieving> 80% on gsm8k with small language models. arXiv preprint arXiv:2312.09241, 2023

    work page arXiv 2023

  38. [38]

    Discrete diffusion modeling by estimating the ratios of the data distribution

    Aaron Lou, Chenlin Meng, and Stefano Ermon. Discrete diffusion modeling by estimating the ratios of the data distribution. 2024

    2024

  39. [39]

    Plan for Speed: Dilated Scheduling for Masked Diffusion Language Models

    Omer Luxembourg, Haim Permuter, and Eliya Nachmani. Plan for speed: Dilated scheduling for masked diffusion language models.arXiv preprint arXiv:2506.19037, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  40. [40]

    Building a large annotated corpus of english: The penn treebank.Computational linguistics, 19(2):313–330, 1993

    Mitch Marcus, Beatrice Santorini, and Mary Ann Marcinkiewicz. Building a large annotated corpus of english: The penn treebank.Computational linguistics, 19(2):313–330, 1993

    1993

  41. [41]

    Teach- ing pretrained language models to think deeper with retrofitted recurrence.arXiv preprint arXiv:2511.07384, 2025

    Sean McLeish, Ang Li, John Kirchenbauer, Dayal Singh Kalra, Brian R Bartoldson, Bhavya Kailkhura, Avi Schwarzschild, Jonas Geiping, Tom Goldstein, and Micah Goldblum. Teach- ing pretrained language models to think deeper with retrofitted recurrence.arXiv preprint arXiv:2511.07384, 2025

    work page arXiv 2025

  42. [42]

    Pointer Sentinel Mixture Models

    Stephen Merity, Caiming Xiong, James Bradbury, and Richard Socher. Pointer sentinel mixture models.arXiv preprint arXiv:1609.07843, 2016

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  43. [43]

    Exact expressive power of transformers with padding

    William Merrill and Ashish Sabharwal. Exact expressive power of transformers with padding. arXiv preprint arXiv:2505.18948, 2025

    work page arXiv 2025

  44. [44]

    Can a suit of armor conduct electricity? a new dataset for open book question answering

    Todor Mihaylov, Peter Clark, Tushar Khot, and Ashish Sabharwal. Can a suit of armor conduct electricity? a new dataset for open book question answering. InProceedings of the 2018 conference on empirical methods in natural language processing, pages 2381–2391, 2018

    2018

  45. [45]

    Cotformer: A chain-of- thought driven architecture with budget-adaptive computation cost at inference.arXiv preprint arXiv:2310.10845, 2023

    Amirkeivan Mohtashami, Matteo Pagliardini, and Martin Jaggi. Cotformer: A chain-of- thought driven architecture with budget-adaptive computation cost at inference.arXiv preprint arXiv:2310.10845, 2023

    work page arXiv 2023

  46. [46]

    Scaling up masked diffusion models on text.arXiv preprint arXiv:2410.18514, 2024

    Shen Nie, Fengqi Zhu, Chao Du, Tianyu Pang, Qian Liu, Guangtao Zeng, Min Lin, and Chongxuan Li. Scaling up masked diffusion models on text.arXiv preprint arXiv:2410.18514, 2024

    work page arXiv 2024

  47. [47]

    Large language diffusion models

    Shen Nie, Fengqi Zhu, Zebin You, Xiaolu Zhang, Jingyang Ou, Jun Hu, Jun Zhou, Yankai Lin, Ji-Rong Wen, and Chongxuan Li. Large language diffusion models. 2025

    2025

  48. [48]

    Your absorbing discrete diffusion secretly models the conditional distributions of clean data

    Jingyang Ou, Shen Nie, Kaiwen Xue, Fengqi Zhu, Jiacheng Sun, Zhenguo Li, and Chongxuan Li. Your absorbing discrete diffusion secretly models the conditional distributions of clean data. 2025. 12

    2025

  49. [49]

    The lambada dataset: Word prediction requiring a broad discourse context

    Denis Paperno, Germán Kruszewski, Angeliki Lazaridou, Ngoc-Quan Pham, Raffaella Bernardi, Sandro Pezzelle, Marco Baroni, Gemma Boleda, and Raquel Fernández. The lambada dataset: Word prediction requiring a broad discourse context. InProceedings of the 54th annual meeting of the association for computational linguistics (volume 1: Long papers), pages 1525–...

    2016

  50. [50]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. pages 4195– 4205, 2023

    2023

  51. [51]

    The fineweb datasets: Decanting the web for the finest text data at scale.Advances in Neural Information Processing Systems, 37:30811–30849, 2024

    Guilherme Penedo, Hynek Kydlíˇcek, Anton Lozhkov, Margaret Mitchell, Colin Raffel, Leandro V on Werra, Thomas Wolf, et al. The fineweb datasets: Decanting the web for the finest text data at scale.Advances in Neural Information Processing Systems, 37:30811–30849, 2024

    2024

  52. [52]

    Path planning for masked diffusion model sampling.arXiv preprint arXiv:2502.03540, 2025

    Fred Zhangzhi Peng, Zachary Bezemek, Sawan Patel, Jarrid Rector-Brooks, Sherwood Yao, Avishek Joey Bose, Alexander Tong, and Pranam Chatterjee. Path planning for masked diffusion model sampling.arXiv preprint arXiv:2502.03540, 2025

    work page arXiv 2025

  53. [53]

    Planner aware path learning in diffusion language models training.arXiv preprint arXiv:2509.23405, 2025

    Fred Zhangzhi Peng, Zachary Bezemek, Jarrid Rector-Brooks, Shuibai Zhang, Anru R Zhang, Michael Bronstein, Alexander Tong, and Avishek Joey Bose. Planner aware path learning in diffusion language models training.arXiv preprint arXiv:2509.23405, 2025

    work page arXiv 2025

  54. [54]

    Let’s think dot by dot: Hidden computation in transformer language models

    Jacob Pfau, William Merrill, and Samuel R Bowman. Let’s think dot by dot: Hidden computation in transformer language models. InFirst Conference on Language Modeling, 2024

    2024

  55. [55]

    Parcae: Scaling Laws For Stable Looped Language Models

    Hayden Prairie, Zachary Novack, Taylor Berg-Kirkpatrick, and Daniel Y Fu. Parcae: Scaling laws for stable looped language models.arXiv preprint arXiv:2604.12946, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  56. [56]

    Language models are unsupervised multitask learners

    Alec Radford, Jeff Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. Language models are unsupervised multitask learners. 2019

    2019

  57. [57]

    Exploring the limits of transfer learning with a unified text-to-text transformer.Journal of machine learning research, 21(140):1–67, 2020

    Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. Exploring the limits of transfer learning with a unified text-to-text transformer.Journal of machine learning research, 21(140):1–67, 2020

    2020

  58. [58]

    On the constant-depth complexity of k-clique

    Benjamin Rossman. On the constant-depth complexity of k-clique. InProceedings of the fortieth annual ACM symposium on Theory of computing, pages 721–730, 2008

    2008

  59. [59]

    Simple and effective masked diffusion language models.Advances in Neural Information Processing Systems, 37:130136–130184, 2024

    Subham S Sahoo, Marianne Arriola, Yair Schiff, Aaron Gokaslan, Edgar Marroquin, Justin T Chiu, Alexander Rush, and V olodymyr Kuleshov. Simple and effective masked diffusion language models.Advances in Neural Information Processing Systems, 37:130136–130184, 2024

    2024

  60. [60]

    The diffusion duality

    Subham Sekhar Sahoo, Justin Deschenaux, Aaron Gokaslan, Guanghan Wang, Justin Chiu, and V olodymyr Kuleshov. The diffusion duality. 2025

    2025

  61. [61]

    Winogrande: An adversarial winograd schema challenge at scale.Communications of the ACM, 64(9):99–106, 2021

    Keisuke Sakaguchi, Ronan Le Bras, Chandra Bhagavatula, and Yejin Choi. Winogrande: An adversarial winograd schema challenge at scale.Communications of the ACM, 64(9):99–106, 2021

    2021

  62. [62]

    Social iqa: Commonsense reasoning about social interactions

    Maarten Sap, Hannah Rashkin, Derek Chen, Ronan Le Bras, and Yejin Choi. Social iqa: Commonsense reasoning about social interactions. InProceedings of the 2019 conference on empirical methods in natural language processing and the 9th international joint conference on natural language processing (EMNLP-IJCNLP), pages 4463–4473, 2019

    2019

  63. [63]

    Reasoning with latent thoughts: On the power of looped transformers.arXiv preprint arXiv:2502.17416, 2025

    Nikunj Saunshi, Nishanth Dikkala, Zhiyuan Li, Sanjiv Kumar, and Sashank J Reddi. Reasoning with latent thoughts: On the power of looped transformers.arXiv preprint arXiv:2502.17416, 2025

    work page arXiv 2025

  64. [64]

    How Much Is One Recurrence Worth? Iso-Depth Scaling Laws for Looped Language Models

    Kristian Schwethelm, Daniel Rueckert, and Georgios Kaissis. How much is one recurrence worth? iso-depth scaling laws for looped language models.arXiv preprint arXiv:2604.21106, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  65. [65]

    Simplified and generalized masked diffusion for discrete data

    Jiaxin Shi, Kehang Han, Zhe Wang, Arnaud Doucet, and Michalis Titsias. Simplified and generalized masked diffusion for discrete data. volume 37, pages 103131–103167, 2024. 13

    2024

  66. [66]

    Seed Diffusion: A Large-Scale Diffusion Language Model with High-Speed Inference

    Yuxuan Song, Zheng Zhang, Cheng Luo, Pengyang Gao, Fan Xia, Hao Luo, Zheng Li, Yuehang Yang, Hongli Yu, Xingwei Qu, et al. Seed diffusion: A large-scale diffusion language model with high-speed inference.arXiv preprint arXiv:2508.02193, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  67. [67]

    On the Reasoning Abilities of Masked Diffusion Language Models

    Anej Svete and Ashish Sabharwal. On the reasoning abilities of masked diffusion language models.arXiv preprint arXiv:2510.13117, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  68. [68]

    Lessons on parameter sharing across layers in transformers

    Sho Takase and Shun Kiyono. Lessons on parameter sharing across layers in transformers. In Proceedings of The Fourth Workshop on Simple and Efficient Natural Language Processing (SustaiNLP), pages 78–90, 2023

    2023

  69. [69]

    Remasking discrete diffusion models with inference-time scaling.arXiv preprint arXiv:2503.00307,

    Guanghan Wang, Yair Schiff, Subham Sekhar Sahoo, and V olodymyr Kuleshov. Remasking discrete diffusion models with inference-time scaling.arXiv preprint arXiv:2503.00307, 2025

    work page arXiv 2025

  70. [70]

    Chain-of-thought prompting elicits reasoning in large language models

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems, 35:24824–24837, 2022

    2022

  71. [71]

    Dream-coder 7b: An open diffusion language model for code, 2025

    Zhihui Xie, Jiacheng Ye, Lin Zheng, Jiahui Gao, Jingwei Dong, Zirui Wu, Xueliang Zhao, Shansan Gong, Xin Jiang, Zhenguo Li, and Lingpeng Kong. Dream-coder 7b: An open diffusion language model for code, 2025

    2025

  72. [72]

    Qwen2 Technical Report

    An Yang, Baosong Yang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Zhou, Chengpeng Li, Chengyuan Li, Dayiheng Liu, Fei Huang, et al. Qwen2 technical report, 2024a.URL https://arxiv. org/abs/2407.10671, 6, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  73. [73]

    Improving Sampling for Masked Diffusion Models via Information Gain

    Kaisen Yang, Jayden Teoh, Kaicheng Yang, Yitong Zhang, and Alex Lamb. Improving sampling for masked diffusion models via information gain.arXiv preprint arXiv:2602.18176, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  74. [74]

    Looped Transformers are Better at Learning Learning Algorithms.arXiv, 2023

    Liu Yang, Kangwook Lee, Robert Nowak, and Dimitris Papailiopoulos. Looped transformers are better at learning learning algorithms.arXiv preprint arXiv:2311.12424, 2023

    work page arXiv 2023

  75. [75]

    Beyond autoregression: Discrete diffusion for complex reasoning and planning.arXiv preprint arXiv:2410.14157, 2024

    Jiacheng Ye, Jiahui Gao, Shansan Gong, Lin Zheng, Xin Jiang, Zhenguo Li, and Lingpeng Kong. Beyond autoregression: Discrete diffusion for complex reasoning and planning.arXiv preprint arXiv:2410.14157, 2024

    work page arXiv 2024

  76. [76]

    Dream 7B: Diffusion Large Language Models

    Jiacheng Ye, Zhihui Xie, Lin Zheng, Jiahui Gao, Zirui Wu, Xin Jiang, Zhenguo Li, and Lingpeng Kong. Dream 7b: Diffusion large language models.arXiv preprint arXiv:2508.15487, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  77. [77]

    Hellaswag: Can a machine really finish your sentence? InProceedings of the 57th annual meeting of the association for computational linguistics, pages 4791–4800, 2019

    Rowan Zellers, Ari Holtzman, Yonatan Bisk, Ali Farhadi, and Yejin Choi. Hellaswag: Can a machine really finish your sentence? InProceedings of the 57th annual meeting of the association for computational linguistics, pages 4791–4800, 2019

    2019

  78. [78]

    Core: Context-robust remasking for diffusion language models.arXiv preprint arXiv:2602.04096, 2026

    Kevin Zhai, Sabbir Mollah, Zhenyi Wang, and Mubarak Shah. Core: Context-robust remasking for diffusion language models.arXiv preprint arXiv:2602.04096, 2026

    work page arXiv 2026

  79. [79]

    Character-level convolutional networks for text classification.Advances in neural information processing systems, 28, 2015

    Xiang Zhang, Junbo Zhao, and Yann LeCun. Character-level convolutional networks for text classification.Advances in neural information processing systems, 28, 2015

    2015

  80. [80]

    LLaDA 1.5: Variance-Reduced Preference Optimization for Large Language Diffusion Models

    Fengqi Zhu, Rongzhen Wang, Shen Nie, Xiaolu Zhang, Chunwei Wu, Jun Hu, Jun Zhou, Jianfei Chen, Yankai Lin, Ji-Rong Wen, et al. Llada 1.5: Variance-reduced preference optimization for large language diffusion models.arXiv preprint arXiv:2505.19223, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

Showing first 80 references.