Pith. sign in

REVIEW 4 major objections 4 minor 2 cited by

Fast Autoregressive Models for Continuous Latent Generation

T0 review · 4 major / 4 minor · reviewed 2026-08-16 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read FAR claims a shortcut head can replace MAR's 100-step diffusion head, cutting inference time 2.3x while holding FID nearly constant.

desk verdict FAR is a clean, incremental win – replacing MAR's diffusion head with a shortcut head gives a real 8-step speedup at comparable FID, but the IS drop and the single-benchmark timing make the headline a bit flatter than claimed. read the letter →

arxiv 2504.18391 v1 pith:ZNRKCRZL submitted 2025-04-24 cs.CV cs.LG

Tiankai Hang , Jianmin Bao , Fangyun Wei , Dong Chen This is my paper · ORCID
classification cs.CVcs.LG
keywords autoregressiveimagegenerationcontinuouslatentspaceshortcutmodelflowmatchingconsistencylossfew-stepsamplingmaskedcausaltransformer
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper claims that the main reason masked autoregressive image models are slow is not the Transformer backbone but the diffusion head, which is invoked 100 times per generated token. It replaces MAR's diffusion head with a shortcut-based head trained with flow matching and consistency losses, so each token needs only 8 denoising calls. The resulting model, FAR, is reported to generate 256x256 ImageNet images 2.3x faster than MAR-B (FID 2.37 vs 2.31) at the same parameter count and with half the training epochs. If correct, this makes continuous-latent autoregressive generation practical and removes the need for discrete image tokenizers.

What carries the argument

The load-bearing object is the FAR head, a shortcut-based network that takes a noisy token, a condition from the backbone, a timestep, and a desired step size $d$, and outputs a denoised velocity prediction. It is trained with a flow matching loss that fixes a linear interpolation path between noise and data and a consistency loss that enforces agreement between predictions at $t$ and $t+d/2$ through an EMA copy of the network. During sampling the step size $d=1/N$ is set by the number of steps, so $N=8$ head calls replace $O=100$ diffusion calls per token while keeping the trajectory consistent.

What would settle it

Measure the end-to-end latency of MAR and FAR on the same image size while varying hardware and batch size, and record the head network's fraction of total inference time. If the head accounts for much less than 63% of MAR's latency, the 2.3x speedup will not reproduce; the claim is falsified by a configuration where the head is not the bottleneck yet FAR still only matches MAR.

Watch

Extended reading notes

Core claim

The central claim is that a single architectural substitution, replacing MAR's 100-step diffusion head with an 8-step shortcut head that predicts the denoising velocity field, preserves autoregressive generation quality while cutting the dominant inference cost. The paper supports this with ImageNet-256 experiments: FAR-B reaches FID 2.37 and IS 265.54, close to MAR-B's FID 2.31 and IS 296.00, while running 2.3x faster; FAR-L reaches FID 1.99, nearly identical to MAR-L's 1.98, with a 1.4x speedup. The same head also slots into a standard causal Transformer, producing FAR-Causal, which generates continuous tokens one at a time and reaches FID 5.67 with 50 denoising steps, beating a same-architecture MAR-Causal baseline at 100 steps (FID 6.80).

Load-bearing premise

The claimed 2.3x speedup assumes the diffusion head dominates MAR's latency as measured on one A100/batch-128 setup; on hardware where the head is not the bottleneck, the speedup shrinks.

Editorial extensions

If this is right

  • The diffusion head is no longer the dominant latency component: at $O=8$ steps the head cost ratio stays well below the 63% measured for MAR's $O=100$ head across $K=32,64,256$ iterations.
  • Few-step sampling becomes tunable at inference time: the same weights run at $O=1,2,4,8$ steps, with FID 2.55 at $O=1$ and 2.37 at $O=8$ for $K=256$, letting users trade speed for quality without retraining.
  • The shortcut head beats a flow-matching-only head by a wide margin at extreme few-step settings (FID 3.86 vs 113.60 at one step), showing the consistency loss is what enables the speedup.
  • Causal Transformers can generate continuous latents without architectural change, inheriting KV-cache acceleration and reaching competitive quality for their size.
  • Continuous-latent autoregressive models no longer require learned discrete codebooks, avoiding quantization loss and codebook training instability.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The speedup ratio is hardware-dependent: the 2.3x figure comes from an A100/batch-128 measurement where the head is 63% of latency; on devices where the backbone dominates, the advantage of 8 vs 100 head calls will shrink, though it will not disappear.
  • If the head is no longer the bottleneck, further scaling of FAR should target the encoder-decoder rather than the head, and the method's gains should transfer to other diffusion-headed autoregressive models beyond MAR.
  • The uniformly sampled step size $d$ and the two-loss shortcut objective may be applicable to any conditional denoising head, including text-conditioned or video latent generation, where the same head-call count problem appears.
  • A direct test of the method's ceiling: train FAR-B for 800 epochs (matching MAR-B) and check whether the small FID gap inverts; the current comparison confounds architecture with training budget.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper proposes FAR, a modification of the masked autoregressive model MAR that replaces the diffusion-based head with a lightweight "shortcut head" trained with a flow-matching loss plus a consistency loss. The head supports few-step denoising (default 8 steps instead of MAR's 100) and can be attached either to MAR's encoder-decoder (FAR-B, FAR-L) or to a causal Transformer (FAR-Causal). On ImageNet 256×256, FAR-B achieves FID 2.37 with 400 training epochs and is reported to be 2.3× faster than MAR-B (FID 2.31, 800 epochs); FAR-L reaches FID 1.99 with a reported 1.4× speedup; FAR-B-Causal reaches FID 5.67. Ablations compare step counts, autoregressive iterations, head depth, CFG weight, and shortcut versus flow-matching heads.

Significance. The contribution is timely and, if the efficiency claims hold, practically useful: it is a simple way to cut the dominant head cost of continuous-space autoregressive image generation without quantizing the latent space. The use of a shortcut head is well motivated by prior work, the evaluation is against the external ImageNet benchmark, and the method is not derived from a fitted success metric. The central issue is that the headline speedup is an empirical inference-time claim resting on a single hardware and batching configuration, and the quality claim is weakened by the substantial Inception Score drop for FAR-B. Properly qualified and with reproducible timing support, the paper would make a solid contribution; currently the load-bearing claim is broader than the evidence.

major comments (4)
  1. [§4.3, Fig. 1, App. A] The 2.3× speedup claim is derived from measurements on a single NVIDIA A100 at batch size 128, with Flash Attention 2.5.9, while training used xFormers on V100. The head-cost ratio of 63% and the resulting speedup depend on batch size, hardware, attention backend, and dynamic token shapes; no timing harness or code is provided to reproduce them. The paper should present the speedup as configuration-specific and include at least one additional hardware/batch setting or a reproducible script before claiming a general 2.3× faster inference.
  2. [Table 1, abstract] The abstract states that FAR maintains competitive FID and IS scores, but FAR-B's IS drops from 296.00 (MAR-B) to 265.54, a decrease of about 10%. While the FID difference is small (2.37 vs 2.31), the IS gap is material and should be disclosed with the headline result rather than summarized as uniformly competitive.
  3. [Table 1, §4.2] The headline speedup of 2.3× compares FAR-B trained for 400 epochs against MAR-B trained for 800 epochs. The matched-epoch comparison, FAR-L (400 epochs) versus MAR-L (400 epochs), yields only 1.4× speedup. The paper should clarify that 2.3× applies to the specific FAR-B/MAR-B setup and should report speedups for matched training budgets, so readers can judge how robust the efficiency gain is across model scales.
  4. [§3.2, §4.3] The inference settings used for the MAR baseline in the cost analysis are not fully specified. Section 3.2 states K=64 by default and O=100 in the original implementation, but the text does not state whether the timing in Fig. 1 uses these exact settings, the same batch size, and the same MLP head configuration as FAR. Without this specification, the 63% head-cost ratio and the speedup decomposition cannot be independently verified.
minor comments (4)
  1. [Table 1] There is a typo: "Maksed Autoregressive" should be "Masked Autoregressive" in the Method column.
  2. [References] Reference [19] has a dangling page number "1" and inconsistent formatting; please clean up the bibliography entry.
  3. [§3.3, App. B] The role of the "desired step size" d is only fully specified in the supplementary (d=1/N for N≤16 and d=0 otherwise). A sentence in the main text explaining this sampling rule would make the method section self-contained.
  4. [Fig. 5] The CFG sweep range is not stated in the caption or text; please specify the weights used for the plotted curves.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: FAR's speedup and quality claims rest on external ImageNet benchmarks and measured inference cost, not on fitted inputs or self-citation.

full rationale

The paper's central claims are the 2.3x inference speedup over MAR and competitive FID/IS on ImageNet 256x256. Both are evaluated externally: FID/IS compare generated images against the real ImageNet validation distribution, and the speedup is a measured quantity from an A100/batch-128 timing experiment. The argument that MAR's diffusion head consumes 63% of latency and that FAR reduces head calls from O=100 to O=8 is an empirical cost analysis, not a definitional identity: the reported 2.3x is consistent with that analysis but is still a measured hardware-dependent outcome rather than a quantity forced by the loss equations. The FAR head is inspired by the external Shortcut model [19] and uses a self-distillation consistency loss whose target is computed from the model's own EMA; this is a training mechanism, not a derivation of the paper's success metric from its own output, and the generation quality is checked against an independent benchmark. The only apparent self-citations by the FAR authors, e.g. [62] for large codebooks, are background references in related work and are not load-bearing for any central claim. The concern that the speedup may depend on hardware and batch size is a legitimate empirical-scope caveat, but it is a correctness/robustness issue, not circularity. No step in the paper reduces by construction to its own input, so the appropriate finding is no significant circularity.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The method introduces no new physical or formal entities. It relies on standard flow matching, consistency distillation, and the separability of the backbone and head.

free parameters (4)
  • Default denoising steps O = 8
    Chosen by hand to balance speed and quality; Table 2 shows O=4 gives FID 2.56 vs O=8 FID 2.45 at K=64, so O=8 is not derived.
  • Autoregressive iterations K = 64
    Default from MAR; Table 2 shows larger K improves FID at cost of latency.
  • Desired step size schedule = d = 1/N for N<=16, d=0 for N>16
    Heuristic set empirically in supplementary B; no derivation.
  • CFG weight = not specified in main text, tuned in Figure 5
    Classifier-free guidance scale tuned per configuration.
assumptions (3)
  • domain assumption The VAE latent space for images is adequately modeled by interpolating between a Gaussian noise sample and a clean latent along a straight path (flow matching assumption).
    Equations 4-6 assume z0 ~ N(0,I) and a linear interpolation between noise and data; this is standard for flow matching but not proven for MAR's VAE latents.
  • domain assumption The EMA model f_EMA is a reliable teacher for consistency distillation.
    Eq. 7-8 use EMA self-distillation, an established but empirically validated technique; the method's few-step capability depends on the teacher being accurate.
  • domain assumption The encoder-decoder backbone and head are separable, so the head can be called independently for each token with conditions fixed from the backbone.
    The FAR generation process (Eq. 2, 3) assumes each token is generated independently given conditions; MAR makes the same assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Fast Autoregressive Models for Continuous Latent Generation." pith.science (2026). https://pith.science/paper/ZNRKCRZL

@misc{pith2026250418391,
  author       = {Pith},
  title        = {Pith review of: Fast Autoregressive Models for Continuous Latent Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZNRKCRZL}},
  note         = {Machine review of arXiv:2504.18391}
}
abstract

Autoregressive models have demonstrated remarkable success in sequential data generation, particularly in NLP, but their extension to continuous-domain image generation presents significant challenges. Recent work, the masked autoregressive model (MAR), bypasses quantization by modeling per-token distributions in continuous spaces using a diffusion head but suffers from slow inference due to the high computational cost of the iterative denoising process. To address this, we propose the Fast AutoRegressive model (FAR), a novel framework that replaces MAR's diffusion head with a lightweight shortcut head, enabling efficient few-step sampling while preserving autoregressive principles. Additionally, FAR seamlessly integrates with causal Transformers, extending them from discrete to continuous token generation without requiring architectural modifications. Experiments demonstrate that FAR achieves $2.3\times$ faster inference than MAR while maintaining competitive FID and IS scores. This work establishes the first efficient autoregressive paradigm for high-fidelity continuous-space image generation, bridging the critical gap between quality and scalability in visual autoregressive modeling.

Figures

Figures reproduced from arXiv: 2504.18391 by the authors.

Figure 1
Figure 1. Inference cost breakdown and efficiency comparison [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. (a) FAR introduces a shortcut head that could replace the [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Architecture of the FAR head, a shortcut-based [ [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Analysis on the proportion of inference cost attributed to the head network in FAR ( [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: The impact of different CFG weights on FID for FAR [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Visual examples of images generated by FAR-B. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. XYZFlow:Scaling Multi dimensional Shortcut Flows for Efficient Generative Modeling

    cs.CV 2026-08 conditional novelty 6.0 of 10

    Conditioning each patch's denoising on the full trajectories of earlier patches lets XYZFlow generate ImageNet images with FID 1.22 to 1.63 in only 2 to 5 steps per patch, at 7.2 to 8.5x teacher speedups.

  2. DiSA: Diffusion Step Annealing in Autoregressive Image Generation

    cs.CV 2025-05 conditional novelty 6.0 of 10

    Autoregressive image diffusion models can use far fewer denoising steps for later tokens without losing quality, yielding 1.4-2.5x speedup from step annealing and up to 10x when combined with fewer autoregressive steps.

Reference graph

Works this paper leans on

62 extracted references · 18 canonical work pages · cited by 2 Pith papers

  1. [1]

    Phi-3 technical report: A highly capable language model locally on your phone

    Marah Abdin, Sam Ade Jacobs, Ammar Ahmad Awan, Jyoti Aneja, Ahmed Awadallah, Hany Awadalla, Nguyen Bach, Amit Bahree, Arash Bakhtiari, Harkirat Behl, et al. Phi-3 technical report: A highly capable language model locally on your phone. arXiv preprint arXiv:2404.14219, 2024. 2

    arXiv 2024

  2. [2]

    Gpt-4 technical report

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ah- mad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. Gpt-4 technical report. arXiv preprint arXiv:2303.08774 , 2023

    arXiv 2023

  3. [3]

    Palm 2 technical report

    Rohan Anil, Andrew M Dai, Orhan Firat, Melvin John- son, Dmitry Lepikhin, Alexandre Passos, Siamak Shakeri, Emanuel Taropa, Paige Bailey, Zhifeng Chen, et al. Palm 2 technical report. arXiv preprint arXiv:2305.10403, 2023

    arXiv 2023

  4. [4]

    Qwen technical report

    Jinze Bai, Shuai Bai, Yunfei Chu, Zeyu Cui, Kai Dang, Xiaodong Deng, Yang Fan, Wenbin Ge, Yu Han, Fei Huang, et al. Qwen technical report. arXiv preprint arXiv:2309.16609, 2023. 2

    arXiv 2023

  5. [5]

    Estimating the optimal covariance with imperfect mean in diffusion probabilistic models

    Fan Bao, Chongxuan Li, Jiacheng Sun, Jun Zhu, and Bo Zhang. Estimating the optimal covariance with imperfect mean in diffusion probabilistic models. ICML, 2022. 2

    work page 2022

  6. [6]

    Analytic- dpm: an analytic estimate of the optimal reverse vari- ance in diffusion probabilistic models

    Fan Bao, Chongxuan Li, Jun Zhu, and Bo Zhang. Analytic- dpm: an analytic estimate of the optimal reverse vari- ance in diffusion probabilistic models. arXiv preprint arXiv:2201.06503, 2022. 2

    arXiv 2022

  7. [7]

    Tract: Denoising diffusion models with transitive closure time-distillation

    David Berthelot, Arnaud Autef, Jierui Lin, Dian Ang Yap, Shuangfei Zhai, Siyuan Hu, Daniel Zheng, Walter Tal- bot, and Eric Gu. Tract: Denoising diffusion models with transitive closure time-distillation. arXiv preprint arXiv:2303.04248, 2023. 2

    arXiv 2023

  8. [8]

    Large scale gan training for high fidelity natural image synthesis

    Andrew Brock, Jeff Donahue, and Karen Simonyan. Large scale gan training for high fidelity natural image synthesis. arXiv preprint arXiv:1809.11096, 2018. 6

    arXiv 2018

Show all 62 references
  1. [9]

    Lan- guage models are few-shot learners

    Tom Brown, Benjamin Mann, Nick Ryder, Melanie Sub- biah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakan- tan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Lan- guage models are few-shot learners. Advances in neural in- formation processing systems, 33:1877–1901, 2020. 1, 3

    1901

  2. [10]

    Lan- guage models are few-shot learners

    Tom Brown, Benjamin Mann, Nick Ryder, Melanie Sub- biah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakan- tan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Lan- guage models are few-shot learners. Advances in neural in- formation processing systems, 33:1877–1901, 2020. 2

    1901

  3. [11]

    Maskgit: Masked generative image transformer

    Huiwen Chang, Han Zhang, Lu Jiang, Ce Liu, and William T Freeman. Maskgit: Masked generative image transformer. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 11315–11325, 2022. 6

    2022

  4. [12]

    Generative pre- training from pixels

    Mark Chen, Alec Radford, Rewon Child, Jeffrey Wu, Hee- woo Jun, David Luan, and Ilya Sutskever. Generative pre- training from pixels. In International conference on machine learning, pages 1691–1703. PMLR, 2020. 2

    2020

  5. [13]

    Palm: Scaling language modeling with pathways

    Aakanksha Chowdhery, Sharan Narang, Jacob Devlin, Maarten Bosma, Gaurav Mishra, Adam Roberts, Paul Barham, Hyung Won Chung, Charles Sutton, Sebastian Gehrmann, et al. Palm: Scaling language modeling with pathways. Journal of Machine Learning Research, 24(240): 1–113, 2023. 2

    2023

  6. [14]

    Imagenet: A large-scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In CVPR, 2009. 1, 5

    2009

  7. [15]

    Diffusion models beat gans on image synthesis

    Prafulla Dhariwal and Alexander Nichol. Diffusion models beat gans on image synthesis. Advances in neural informa- tion processing systems, 34:8780–8794, 2021. 6

    2021

  8. [16]

    Dreamllm: Synergistic multimodal com- prehension and creation

    Runpei Dong, Chunrui Han, Yuang Peng, Zekun Qi, Zheng Ge, Jinrong Yang, Liang Zhao, Jianjian Sun, Hongyu Zhou, Haoran Wei, et al. Dreamllm: Synergistic multimodal com- prehension and creation. arXiv preprint arXiv:2309.11499,

    arXiv

  9. [17]

    The llama 3 herd of models

    Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Ab- hishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angela Fan, et al. The llama 3 herd of models. arXiv preprint arXiv:2407.21783 ,

    arXiv

  10. [18]

    Taming transformers for high-resolution image synthesis

    Patrick Esser, Robin Rombach, and Bjorn Ommer. Taming transformers for high-resolution image synthesis. In Pro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 12873–12883, 2021. 2, 3

    2021

  11. [19]

    One step diffusion via shortcut models

    Kevin Frans, Danijar Hafner, Sergey Levine, and Pieter Abbeel. One step diffusion via shortcut models. arXiv preprint arXiv:2410.12557, 2024. 2, 3, 4, 5, 1

    2024 arXiv

  12. [20]

    Seed-x: Mul- timodal models with unified multi-granularity comprehen- sion and generation

    Yuying Ge, Sijie Zhao, Jinguo Zhu, Yixiao Ge, Kun Yi, Lin Song, Chen Li, Xiaohan Ding, and Ying Shan. Seed-x: Mul- timodal models with unified multi-granularity comprehen- sion and generation. arXiv preprint arXiv:2404.14396, 2024. 2

    2024 arXiv

  13. [21]

    Boot: Data-free distillation of denois- ing diffusion models with bootstrapping

    Jiatao Gu, Shuangfei Zhai, Yizhe Zhang, Lingjie Liu, and Joshua M Susskind. Boot: Data-free distillation of denois- ing diffusion models with bootstrapping. In ICML 2023 Workshop on Structured Probabilistic Inference & Genera- tive Modeling, 2023. 2

    2023

  14. [22]

    Classifier-free diffusion guidance

    Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance. In arXiv preprint arXiv:2207.12598, 2022. 5

    2022 arXiv

  15. [23]

    Elucidating the design space of diffusion-based generative models

    Tero Karras, Miika Aittala, Timo Aila, and Samuli Laine. Elucidating the design space of diffusion-based generative models. In NeurIPS, 2022. 2

    2022

  16. [24]

    Adam: A method for stochastic optimization

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 ,

    arXiv

  17. [25]

    Auto-encoding varia- tional bayes

    Diederik P Kingma and Max Welling. Auto-encoding varia- tional bayes. In ICLR, 2014. 1

    2014

  18. [26]

    Autoregressive image generation using residual quantization

    Doyup Lee, Chiheon Kim, Saehoon Kim, Minsu Cho, and Wook-Shin Han. Autoregressive image generation using residual quantization. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 11523–11532, 2022. 2, 3

    2022

  19. [27]

    Autoregressive image generation using residual quantization

    Doyup Lee, Chiheon Kim, Saehoon Kim, Minsu Cho, and Wook-Shin Han. Autoregressive image generation using residual quantization. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 11523–11532, 2022. 1

    2022

  20. [28]

    Autoregressive image generation without vec- 9 tor quantization

    Tianhong Li, Yonglong Tian, He Li, Mingyang Deng, and Kaiming He. Autoregressive image generation without vec- 9 tor quantization. Advances in Neural Information Processing Systems, 37:56424–56445, 2025. 1, 2, 3, 5, 6

    2025

  21. [29]

    Flow matching for generative mod- eling

    Yaron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximil- ian Nickel, and Matt Le. Flow matching for generative mod- eling. arXiv preprint arXiv:2210.02747, 2022. 7, 1

    2022 arXiv

  22. [30]

    Pseudo numerical methods for diffusion models on manifolds

    Luping Liu, Yi Ren, Zhijie Lin, and Zhou Zhao. Pseudo numerical methods for diffusion models on manifolds. In ICLR, 2022. 2

    2022

  23. [31]

    Flow straight and fast: Learning to generate and transfer data with rectified flow

    Xingchao Liu, Chengyue Gong, and Qiang Liu. Flow straight and fast: Learning to generate and transfer data with rectified flow. In ICLR, 2023. 2, 3, 7

    2023

  24. [32]

    Instaflow: One step is enough for high-quality diffusion-based text-to-image generation

    Xingchao Liu, Xiwen Zhang, Jianzhu Ma, Jian Peng, and Qiang Liu. Instaflow: One step is enough for high-quality diffusion-based text-to-image generation. arXiv preprint arXiv:2309.06380, 2023. 3

    2023 arXiv

  25. [33]

    Decoupled weight decay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101, 2017. 5

    2017 arXiv

  26. [34]

    Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps

    Cheng Lu, Yuhao Zhou, Fan Bao, Jianfei Chen, Chongxuan Li, and Jun Zhu. Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps. Advances in Neural Information Processing Systems , 35:5775–5787,

  27. [35]

    Dpm-solver++: Fast solver for guided sampling of diffusion probabilistic models

    Cheng Lu, Yuhao Zhou, Fan Bao, Jianfei Chen, Chongx- uan Li, and Jun Zhu. Dpm-solver++: Fast solver for guided sampling of diffusion probabilistic models. arXiv preprint arXiv:2211.01095, 2022. 2

    2022 arXiv

  28. [36]

    Knowledge distillation in iterative generative models for improved sampling speed

    Eric Luhman and Troy Luhman. Knowledge distillation in iterative generative models for improved sampling speed. arXiv preprint arXiv:2101.02388, 2021. 2

    2021 arXiv

  29. [37]

    On distillation of guided diffusion models

    Chenlin Meng, Robin Rombach, Ruiqi Gao, Diederik Kingma, Stefano Ermon, Jonathan Ho, and Tim Salimans. On distillation of guided diffusion models. In CVPR, 2023. 2

    2023

  30. [38]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. In Proceedings of the IEEE/CVF inter- national conference on computer vision , pages 4195–4205,

  31. [39]

    Improving language under- standing by generative pre-training

    Alec Radford, Karthik Narasimhan, Tim Salimans, and Ilya Sutskever. Improving language under- standing by generative pre-training. OpenAI, 2018. https : / / cdn . openai . com / research - covers / language - unsupervised / language _ understanding_paper.pdf. 2

    2018

  32. [40]

    Language models are unsu- pervised multitask learners

    Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, Ilya Sutskever, et al. Language models are unsu- pervised multitask learners. OpenAI blog, 1(8):9, 2019. 1, 3

    2019

  33. [41]

    Gener- ating diverse high-fidelity images with vq-vae-2

    Ali Razavi, Aaron Van den Oord, and Oriol Vinyals. Gener- ating diverse high-fidelity images with vq-vae-2. Advances in neural information processing systems, 32, 2019. 1, 2, 3

    2019

  34. [42]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022. 6

    2022

  35. [43]

    Progressive distillation for fast sampling of diffusion models

    Tim Salimans and Jonathan Ho. Progressive distillation for fast sampling of diffusion models. In ICLR, 2022. 2, 5

    2022

  36. [44]

    Multistep distillation of diffusion models via moment matching

    Tim Salimans, Thomas Mensink, Jonathan Heek, and Emiel Hoogeboom. Multistep distillation of diffusion models via moment matching. Advances in Neural Information Process- ing Systems, 37:36046–36070, 2025. 3

    2025

  37. [45]

    Learning structured output representation using deep conditional gen- erative models

    Kihyuk Sohn, Honglak Lee, and Xinchen Yan. Learning structured output representation using deep conditional gen- erative models. Advances in neural information processing systems, 28, 2015. 1

    2015

  38. [46]

    Consistency models

    Yang Song, Prafulla Dhariwal, Mark Chen, and Ilya Sutskever. Consistency models. In ICML, 2023. 2, 3

    2023

  39. [47]

    Hart: Efficient visual generation with hybrid au- toregressive transformer

    Haotian Tang, Yecheng Wu, Shang Yang, Enze Xie, Junsong Chen, Junyu Chen, Zhuoyang Zhang, Han Cai, Yao Lu, and Song Han. Hart: Efficient visual generation with hybrid au- toregressive transformer. arXiv preprint arXiv:2410.10812,

    arXiv

  40. [48]

    Gemini: a family of highly capable multimodal models

    Gemini Team, Rohan Anil, Sebastian Borgeaud, Yonghui Wu, Jean-Baptiste Alayrac, Jiahui Yu, Radu Soricut, Johan Schalkwyk, Andrew M Dai, Anja Hauth, et al. Gemini: a family of highly capable multimodal models. arXiv preprint arXiv:2312.11805, 2023. 2

    2023 arXiv

  41. [49]

    Llama: Open and efficient foundation language models

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timoth´ee Lacroix, Baptiste Rozi`ere, Naman Goyal, Eric Hambro, Faisal Azhar, et al. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971, 2023

    2023 arXiv

  42. [50]

    Llama 2: Open foundation and fine-tuned chat models.arXiv preprint arXiv:2307.09288, 2023

    Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, et al. Llama 2: Open foundation and fine-tuned chat models.arXiv preprint arXiv:2307.09288, 2023. 2

    2023 arXiv

  43. [51]

    Givt: Generative infinite-vocabulary transformers

    Michael Tschannen, Cian Eastwood, and Fabian Mentzer. Givt: Generative infinite-vocabulary transformers. In Eu- ropean Conference on Computer Vision , pages 292–309. Springer, 2024. 2

    2024

  44. [52]

    Jetformer: An autoregressive generative model of raw images and text

    Michael Tschannen, Andr ´e Susano Pinto, and Alexander Kolesnikov. Jetformer: An autoregressive generative model of raw images and text. arXiv preprint arXiv:2411.19722 ,

    arXiv

  45. [53]

    Neural discrete representation learning

    Aaron Van Den Oord, Oriol Vinyals, et al. Neural discrete representation learning. Advances in neural information pro- cessing systems, 30, 2017. 1, 2

    2017

  46. [54]

    Tack- ling the generative learning trilemma with denoising diffu- sion gans

    Zhisheng Xiao, Karsten Kreis, and Arash Vahdat. Tack- ling the generative learning trilemma with denoising diffu- sion gans. In ICLR, 2022. 2

    2022

  47. [55]

    Qwen2 technical report.arXiv preprint arXiv:2407.10671, 2024

    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.arXiv preprint arXiv:2407.10671, 2024. 2

    2024 arXiv

  48. [57]

    Vector-quantized image modeling with improved vqgan

    Jiahui Yu, Xin Li, Jing Yu Koh, Han Zhang, Ruoming Pang, James Qin, Alexander Ku, Yuanzhong Xu, Jason Baldridge, and Yonghui Wu. Vector-quantized image modeling with improved vqgan. arXiv preprint arXiv:2110.04627, 2021. 2 10

    2021 arXiv

  49. [58]

    Scaling autoregres- sive models for content-rich text-to-image generation

    Jiahui Yu, Yuanzhong Xu, Jing Yu Koh, Thang Luong, Gun- jan Baid, Zirui Wang, Vijay Vasudevan, Alexander Ku, Yin- fei Yang, Burcu Karagol Ayan, et al. Scaling autoregres- sive models for content-rich text-to-image generation. arXiv preprint arXiv:2206.10789, 2(3):5, 2022. 2

    2022 arXiv

  50. [59]

    Language model beats diffusion–tokenizer is key to visual generation

    Lijun Yu, Jos ´e Lezama, Nitesh B Gundavarapu, Luca Ver- sari, Kihyuk Sohn, David Minnen, Yong Cheng, Vighnesh Birodkar, Agrim Gupta, Xiuye Gu, et al. Language model beats diffusion–tokenizer is key to visual generation. arXiv preprint arXiv:2310.05737, 2023. 3

    arXiv

  51. [60]

    Unipc: A unified predictor-corrector frame- work for fast sampling of diffusion models

    Wenliang Zhao, Lujia Bai, Yongming Rao, Jie Zhou, and Jiwen Lu. Unipc: A unified predictor-corrector frame- work for fast sampling of diffusion models. arXiv preprint arXiv:2302.04867, 2023. 2

    2023 arXiv

  52. [61]

    Fast sampling of dif- fusion models via operator learning

    Hongkai Zheng, Weili Nie, Arash Vahdat, Kamyar Aziz- zadenesheli, and Anima Anandkumar. Fast sampling of dif- fusion models via operator learning. In ICML, 2023. 2

    2023

  53. [62]

    Scaling the codebook size of vqgan to 100,000 with a utilization rate of 99%

    Lei Zhu, Fangyun Wei, Yanye Lu, and Dong Chen. Scaling the codebook size of vqgan to 100,000 with a utilization rate of 99%. arXiv preprint arXiv:2406.11837, 2024. 2, 3 11 Fast Autoregressive Models for Continuous Latent Generation Supplementary Material A. Environment The sof...

    2024 arXiv

  54. [63]

    Embed. Dim

    The sampling process is formulated as: ˆh0∼N (0, I), (10) ˆht+ 1 N = ˆht + 1 Nfθ(ˆht, c,d ), (11) where N represents the number of sampling steps, with d = 1 /N when N ≤ 16, and d = 0 when N > 16. The variable t denotes the current sampling step, with t∈ {0, 1/N, 2/N,··· , (N−...

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.