Pith. sign in

REVIEW 5 major objections 5 minor 78 references

DMQ: Dissecting Outliers of Diffusion Models for Post-Training Quantization

T0 review · 5 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read A post-training quantization method called DMQ keeps diffusion image generators accurate at 4-bit weights and 6-bit activations by learning channel-wise scaling and taming outlier channels with power-of-two bit shifts.

desk verdict Plausible PTQ method for diffusion models, but the evaluation is confounded by quantized models beating full precision, so the quality-preservation claim needs a clean re-test. read the letter →

arxiv 2507.12933 v1 pith:UQ3NWSEC submitted 2025-07-17 cs.CV cs.AIcs.LG

classification cs.CVcs.AIcs.LG
keywords diffusionmodelspost-trainingquantizationoutlierchannelsequivalentscalingpower-of-twotimestepweightinglow-bitinferenceW4A6
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

Diffusion image generators normally must be quantized gently: at 4-bit weights and 6-bit activations, earlier post-training methods collapse, because a few outlier channels stretch the quantization range so far that the rest of the tensor is quantized coarsely. This paper proposes a two-part recipe that keeps such aggressive quantization stable. A learned equivalent scaling factor optimizes a channel-wise rescaling that shifts quantization difficulty between weights and activations, using a timestep-weighted loss that treats early denoising steps as more consequential than their raw error suggests. A power-of-two channel scaling then directly shrinks outliers in skip-connection layers, with a voting rule over calibration samples selecting the scaling exponents reliably. Across unconditional, class-conditional, and text-to-image models, the paper reports that the recipe beats prior methods and preserves image quality at W4A6 and W4A8.

What carries the argument

The load-bearing object is the channel-wise scaling pair: $\tau \in \mathbb{R}^{C_{\text{in}}}$ for Learned Equivalent Scaling and $\delta \in \mathbb{R}^{C_{\text{in}}}$ for Power-of-Two Scaling. LES uses the identity $Y = (X/\tau)(\tau^\top \odot W)$ to move quantization difficulty between activation and weight channels, with $\tau$ tuned by minimizing layer-output MSE under the adaptive timestep weighting $\lambda_t = (1 - \Lambda_t / \sum_{t'}\Lambda_{t'})^\alpha$, where $\Lambda_t$ is a momentum-averaged accumulated loss. PTS applies $\tilde{X} = \mathrm{clamp}(\lfloor X / (2^{\delta} \odot s^{(X)}) \rceil, l, u)$ to activation channels and folds $2^{\delta_k}$ into the loaded weights as a left bit-shift $\tilde{W}_{kj} \ll \delta_k$, so extreme outliers are suppressed without per-step computation; the exponents come from a mode-voting algorithm with agreement threshold $\kappa$ over calibration samples. Both $\tau$ and $\delta$ are learned once and fused into static scales, which is what keeps inference overhead low.

What would settle it

Re-optimize $\tau$ and $\delta$ separately for each denoising timestep using the same calibration protocol, and compare W4A6 FID with the static single-factor version; if per-timestep re-optimization clearly improves FID on FFHQ or ImageNet, the timestep-invariant outlier assumption is false, and if the two are indistinguishable, it is confirmed. A supporting observation would be a channel-level correlation matrix of activation outliers across timesteps for skip-connection layers, showing whether the same channels stay extreme throughout the denoising trajectory.

Watch

Extended reading notes

Core claim

The central claim is that the failure of low-bit diffusion quantization is dominated by channel-wise outliers, and that those outliers can be handled by two complementary mechanisms without retraining. Learned Equivalent Scaling (LES) optimizes a channel-wise scale vector $\tau \in \mathbb{R}^{C_{\text{in}}}$ so that the reparametrization $Y = (X/\tau)(\tau^\top \odot W)$ minimizes layer-output quantization error over calibration samples; because early denoising steps have small but accumulating errors, the optimization uses an adaptive timestep weighting that up-weights those steps. The second mechanism, channel-wise Power-of-Two Scaling (PTS), applies $2^{\delta}$ factors to activation channels of layers with extreme inter-channel variance, such as skip connections, removing outliers rather than transferring them to the weights; the exponents are selected by a per-sample voting procedure with an agreement threshold so that a small calibration set does not overfit. PTS is implemented as a bit-shift on the loaded weights, so the paper argues the added cost is minimal. With both mechanisms, the paper reports stable W4A6 quantization, with image quality close to full precision on FFHQ, LSUN-Bedroom, LSUN-Church, ImageNet LDM, and Stable Diffusion text-to-image generation.

Load-bearing premise

The method assumes that the channel-wise outlier pattern in activations stays roughly the same across all denoising timesteps, so a single set of scaling factors learned from calibration data can serve every step.

Editorial extensions

If this is right

  • Diffusion image generators can be deployed at W4A6 with FID and sFID close to full precision, which removes a major obstacle to running them on memory-limited hardware.
  • Static, precomputed scaling factors are sufficient; because $\tau$ and $2^{\delta}$ are fused into scales and weight loads, low-bit inference does not pay a per-timestep rescaling cost.
  • Outlier channels in skip connections are the primary bottleneck at low bit widths; targeting only those layers with power-of-two scaling is enough to recover most of the lost quality.
  • The recipe transfers across model families: latent diffusion models for unconditional, class-conditional, and text-to-image generation all improve, suggesting the outlier structure is a general property of diffusion U-Nets.

Reading between the lines

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

  • Beyond the paper's U-Net experiments, the static $\tau$ and $\delta$ premise is most likely to break in transformer-based diffusion backbones such as DiT, where activations vary more strongly with timestep and conditioning; a timestep-conditional variant of the voting rule would be a direct test.
  • The voting mechanism suggests a general recipe for choosing outlier-scaling hyperparameters from small calibration sets: take the per-sample best candidate, keep only the modal choice, and default to no scaling when agreement is low; that recipe could transfer to other iterative generative models such as flow matching or consistency models.
  • The adaptive timestep weighting is heuristic in form (a focal-loss-style power law); a natural extension is replacing it with a measured sensitivity that tracks how much a unit of injected noise at each timestep propagates to the final image, which would put the step prioritization on firmer ground.
  • Because power-of-two scaling is applied only to a few skip-connection layers, the reported speedup is mostly attributable to low-bit GEMM itself; one could test whether the same outlier analysis identifies other high-variance layers, such as attention projections in transformer blocks, where bit-shift scaling would pay for itself.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. This paper proposes DMQ, a post-training quantization method for diffusion models that combines Learned Equivalent Scaling (LES) with an adaptive timestep weighting loss and channel-wise Power-of-Two Scaling (PTS) with a voting-based factor selection. LES learns channel-wise scaling factors that rebalance quantization error between weights and activations; the adaptive weighting up-weights timesteps whose small errors still have large influence on final output; PTS targets layers with extreme inter-channel variance such as residual-block skip connections. The method is evaluated on LDM-based unconditional and class-conditional generation and on Stable Diffusion text-conditioned generation at W4A8 and W4A6, where it generally outperforms prior PTQ baselines.

Significance. DMQ is a clearly specified PTQ framework with a credible mechanism: the analysis of timestep-dependent error influence in Fig. 4 is interesting, the bit-shift formulation in Eq. (13) is hardware-conscious, and the voting algorithm is a plausible safeguard against small-calibration overfitting. The paper also releases code and documents implementation details thoroughly. If the reported gains survive a more rigorous evaluation protocol, the work would be a valuable contribution to low-bit diffusion quantization. However, the current evaluation contains several confounds that make the central 'preserving image generation quality' claim hard to assess, so the significance is conditional on fixing those issues.

major comments (5)
  1. [Section 5.1, Tables 2 and 3] It is unexplained why the quantized model often produces substantially better FID than the full-precision model, e.g., FFHQ W8A8 FID 26.78 vs 31.34 and ImageNet W4A6 FID 7.81 vs 11.34. This undermines the 'preserving' framing and suggests that the evaluation protocol (DDIM with eta=1.0 and 20 steps for unconditional generation) is far from optimal for the full-precision model, while the calibration-time optimization in DMQ acts as an additional adaptation. Please report full-precision results under the same protocol and seed, report multiple seeds with error bars, and include results under a standard sampling protocol (e.g., DDIM eta=0 with 50 or 100 steps); if the quantized model still outperforms full precision, the reason should be discussed rather than treated as preservation.
  2. [Sections 4.1 and 4.2] The LES factors tau and PTS exponents delta are learned once from calibration data and then used for every denoising step. The paper justifies this through Fig. 4, which shows layer-wise average quantization error trends, not channel-wise outlier stability. Since per-channel scaling factors are only meaningful if the identity of outlier channels does not shift across timesteps, please provide channel-wise stability evidence (e.g., how often the max-magnitude channel or the selected delta* changes with t) or an ablation comparing static factors with per-timestep recalibrated factors.
  3. [Appendix B] The authors state that PTQD and TFMQ-DM were modified to quantize skip-connection, downsample, and upsample layers for fair comparison. Because these are exactly the layers where DMQ applies PTS, such a modification can change baseline behavior substantially. Please report per-method modification details, state explicitly which numbers in Tables 2-4 include the modified code, and clarify whether the time-embedding caching described in Section 5 applies to all methods or only to DMQ; otherwise the baseline comparison is not fully specified.
  4. [Section 5.1, Tables 2-4] All FID/sFID numbers are reported without error bars or multiple-seed statistics. For comparisons where the differences between methods are often smaller than one FID point, a single run cannot support the claim of 'stable' performance. Please provide at least three sampling seeds (and, where feasible, several calibration/optimization runs) with means and standard deviations.
  5. [Abstract and Section 5.1] The claim that W4A6 quantization 'preserves high image generation quality' is not consistently supported: in Table 2 (LSUN-Church) the W4A6 FID is 14.28 vs 11.28 for full precision, and in Table 4 (MS-COCO, Stable Diffusion) W4A6 degrades FID from 20.38 to 29.58 and sFID from 62.55 to 67.05. Please qualify the claim or report the average degradation across all tasks.
minor comments (5)
  1. [Figure 4 caption] The word 'qauntization' should be 'quantization'.
  2. [Figure 3 caption] The sentence mentioning 'the scaled activation \hat{X} and weights \hat{X}' should read '\hat{W}' for the weights.
  3. [Table 8] The header involving 'T /c' is confusing; it should clearly state that T is the number of sampling steps for calibration data and n is the number of samples per step.
  4. [Algorithm 1] The phrase 'for all channel index' should be 'for each channel index', and the tie-breaking rule for the mode of delta* is unspecified.
  5. [Table 8] The hyperparameters alpha, kappa, xi, and D are not accompanied by sensitivity analyses; at least a brief discussion of their ranges would help the reader assess robustness.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: learned scaling factors are fitted to calibration data and evaluated on held-out FID; the central W4A6 claim is empirical, not an identity.

full rationale

DMQ's derivation chain is a standard PTQ pipeline: tau in Eq. (6) is optimized to minimize layer-wise output reconstruction error on a calibration set, and the PTS factors delta are selected by Algorithm 1 to minimize per-channel quantization error on the same calibration set; all reported FID/sFID/IS/CLIP numbers in Tables 2-4 are computed on freshly generated images, not on the calibration loss. Thus the headline claim of stable W4A6 quantization while preserving quality is an empirical benchmark result, not a quantity that equals the training objective by construction. The adaptive timestep weighting (Eqs. 7-9) is self-referential in that sample weights are derived from the moving average of the same reconstruction losses, but this is a standard loss-reweighting heuristic and the weights are held fixed during each update, so it does not make the final metric an identity or a fitted value renamed as a prediction. The one overlapping-author citation, [23], is used only to list image editing as an application and is not load-bearing; no uniqueness theorem or prior-work ansatz is invoked to force the method. Appendix F's stated limitations and Appendix B's modification of baselines to quantize skip connections, as well as the observation that quantized models sometimes beat full precision, are evaluation-protocol and reproducibility concerns, not circularity, and do not change the verdict.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The central claim depends on several fitted parameters (tau, delta, alpha, kappa, xi, D) and domain assumptions about layer-wise reconstruction, timestep importance, and channel outlier stability.

free parameters (6)
  • LES scaling factors tau = learned per-channel, not reported
    Optimized via Eq. (6) to minimize reconstruction error on calibration data (Section 4.1).
  • PTS scaling exponents delta = per-channel power-of-two exponents, not reported
    Selected by voting algorithm from candidates {0,...,D} per channel (Section 4.2).
  • Adaptive weighting exponent alpha = 20 or 25 depending on dataset (Tab. 8)
    Controls emphasis on early timesteps; tuned per dataset.
  • Agreement threshold kappa = 0.85
    Rejects PTS factors with low consensus across calibration samples.
  • Momentum xi = 0.95
    Smooths accumulated loss in Eq. (9).
  • Candidate range D = 3
    Limits PTS factors to 2^0..2^3.
assumptions (5)
  • domain assumption Layer-wise MSE minimization on calibration data transfers to end-to-end sample quality (Eq. 6).
    PTQ methods commonly assume block/layer reconstruction suffices; not proven in this paper.
  • domain assumption Early denoising steps have outsized impact on final output due to error accumulation (Fig. 4).
    Supported by a single noise-injection probe; the weighting scheme in Eq. (8) is not derived from it quantitatively.
  • domain assumption Channel-wise outlier patterns are stable across timesteps.
    Static tau and delta are used for all timesteps; no experiment demonstrates channel stability over time.
  • ad hoc to paper Power-of-two scaling factors suffice to compress activation outliers without excessive rounding error.
    Limited to powers of two for efficient bit-shift; optimality not established.
  • standard math MinMax quantizer with static per-tensor activation scale is a valid base quantizer.
    From quantization literature (Eq. 3).

how reviews work

0 comments
Cite this review

Pith. "Pith review of DMQ: Dissecting Outliers of Diffusion Models for Post-Training Quantization." pith.science (2026). https://pith.science/paper/UQ3NWSEC

@misc{pith2026250712933,
  author       = {Pith},
  title        = {Pith review of: DMQ: Dissecting Outliers of Diffusion Models for Post-Training Quantization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UQ3NWSEC}},
  note         = {Machine review of arXiv:2507.12933}
}
read the original abstract

Diffusion models have achieved remarkable success in image generation but come with significant computational costs, posing challenges for deployment in resource-constrained environments. Recent post-training quantization (PTQ) methods have attempted to mitigate this issue by focusing on the iterative nature of diffusion models. However, these approaches often overlook outliers, leading to degraded performance at low bit-widths. In this paper, we propose a DMQ which combines Learned Equivalent Scaling (LES) and channel-wise Power-of-Two Scaling (PTS) to effectively address these challenges. Learned Equivalent Scaling optimizes channel-wise scaling factors to redistribute quantization difficulty between weights and activations, reducing overall quantization error. Recognizing that early denoising steps, despite having small quantization errors, crucially impact the final output due to error accumulation, we incorporate an adaptive timestep weighting scheme to prioritize these critical steps during learning. Furthermore, identifying that layers such as skip connections exhibit high inter-channel variance, we introduce channel-wise Power-of-Two Scaling for activations. To ensure robust selection of PTS factors even with small calibration set, we introduce a voting algorithm that enhances reliability. Extensive experiments demonstrate that our method significantly outperforms existing works, especially at low bit-widths such as W4A6 (4-bit weight, 6-bit activation) and W4A8, maintaining high image generation quality and model stability. The code is available at https://github.com/LeeDongYeun/dmq.

Figures

Figures reproduced from arXiv: 2507.12933 by the authors.

Figure 1
Figure 1. Images generated by Stable Diffusion [50] in full precision (top row) and in W4A8-quantization using our method (bottom row). Abstract Diffusion models have achieved remarkable success in image generation but come with significant computa￾tional costs, posing challenges for deployment in resource￾constrained environments. Recent post-training quantiza￾tion (PTQ) methods have attempted to mitigate this issue by focus… view at source ↗
Figure 2
Figure 2. (a) Smoothquant results large equivalent scaling factor. [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Visualization of activations and weights in a skip connection layer, before and after applying Equivalent Scaling and Power [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: (Left): Plot of MSE and FID score across timesteps. Quantization noise increases as t decreases and the same amount of noise has a non-uniform impact on the final output. The MSE indicates qauntization error of activations averaged for all layers. (Right): Normalized q…
Figure 5
Figure 5. Figure 5: Inter-channel variance of the activations. [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 7
Figure 7. Figure 7: Visualization of samples on ImageNet 256 [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: Comparison on latency between pytorch fp32 GEMM [PITH_FULL_IMAGE:figures/full_fig_p014_8.png]
Figure 9
Figure 9. Figure 9: Visualization of samples on LSUN-Bedrooms 256 [PITH_FULL_IMAGE:figures/full_fig_p015_9.png]
Figure 10
Figure 10. Figure 10: Visualization of samples on LSUN-Chruches 256 [PITH_FULL_IMAGE:figures/full_fig_p015_10.png]
Figure 11
Figure 11. Figure 11: Visualization of samples on FFHQ 256×256 generated by full precision LDM [50] and W4A6 quantized models using Q￾Diffusion [32], TFMQ-DM [22], and ours. Full-precision Q-Diffusion TFMQ-DM Ours [PITH_FULL_IMAGE:figures/full_fig_p016_11.png]
Figure 12
Figure 12. Figure 12: Visualization of samples on ImageNet 256 [PITH_FULL_IMAGE:figures/full_fig_p016_12.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

78 extracted references · 41 canonical work pages

  1. [1]

    Hardware approximate techniques for deep neural network accelerators: A survey.ACM Computing Sur- veys, 55(4):1–36, 2022

    Giorgos Armeniakos, Georgios Zervakis, Dimitrios Soudris, and J¨org Henkel. Hardware approximate techniques for deep neural network accelerators: A survey.ACM Computing Sur- veys, 55(4):1–36, 2022. 6

  2. [2]

    Improving image generation with better captions.Computer Science

    James Betker, Gabriel Goh, Li Jing, Tim Brooks, Jianfeng Wang, Linjie Li, Long Ouyang, Juntang Zhuang, Joyce Lee, Yufei Guo, et al. Improving image generation with better captions.Computer Science. https://cdn. openai. com/papers/dall-e-3. pdf, 2(3):8, 2023. 7

  3. [3]

    Align your latents: High-resolution video synthesis with la- tent diffusion models

    Andreas Blattmann, Robin Rombach, Huan Ling, Tim Dock- horn, Seung Wook Kim, Sanja Fidler, and Karsten Kreis. Align your latents: High-resolution video synthesis with la- tent diffusion models. InProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 22563–22575, 2023. 1

  4. [4]

    High per- formance convolutional neural networks for document pro- cessing

    Kumar Chellapilla, Sidd Puri, and Patrice Simard. High per- formance convolutional neural networks for document pro- cessing. InTenth international workshop on frontiers in handwriting recognition. Suvisoft, 2006. 12

  5. [5]

    Pixart-\sigma: Weak-to-strong training of diffusion transformer for 4k text-to-image generation.arXiv preprint arXiv:2403.04692, 2024

    Junsong Chen, Chongjian Ge, Enze Xie, Yue Wu, Lewei Yao, Xiaozhe Ren, Zhongdao Wang, Ping Luo, Huchuan Lu, and Zhenguo Li. Pixart-\sigma: Weak-to-strong training of diffusion transformer for 4k text-to-image generation.arXiv preprint arXiv:2403.04692, 2024. 7

  6. [6]

    cudnn: Efficient primitives for deep learning

    Sharan Chetlur, Cliff Woolley, Philippe Vandermersch, Jonathan Cohen, John Tran, Bryan Catanzaro, and Evan Shelhamer. cudnn: Efficient primitives for deep learning. arXiv preprint arXiv:1410.0759, 2014. 12

  7. [7]

    Pact: Parameterized clipping activa- tion for quantized neural networks.arXiv preprint arXiv:1805.06085, 2018

    Jungwook Choi, Zhuo Wang, Swagath Venkataramani, Pierce I-Jen Chuang, Vijayalakshmi Srinivasan, and Kailash Gopalakrishnan. Pact: Parameterized clipping activa- tion for quantized neural networks.arXiv preprint arXiv:1805.06085, 2018. 2

  8. [8]

    Consistent diffusion models: Mitigating sampling drift by learning to be consistent.Advances in Neu- ral Information Processing Systems, 36, 2024

    Giannis Daras, Yuval Dagan, Alex Dimakis, and Constanti- nos Daskalakis. Consistent diffusion models: Mitigating sampling drift by learning to be consistent.Advances in Neu- ral Information Processing Systems, 36, 2024. 3

Show all 78 references
  1. [9]

    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. In2009 IEEE conference on computer vision and pattern recognition, pages 248–255. Ieee, 2009. 7

  2. [10]

    Tim Dettmers, Mike Lewis, Younes Belkada, and Luke Zettlemoyer. Gpt3. int8 (): 8-bit matrix multiplication for transformers at scale.Advances in Neural Information Pro- cessing Systems, 35:30318–30332, 2022. 3

  3. [11]

    Deepshift: Towards multiplication- less neural networks

    Mostafa Elhoushi, Zihao Chen, Farhan Shafiq, Ye Henry Tian, and Joey Yiwei Li. Deepshift: Towards multiplication- less neural networks. InProceedings of the IEEE/CVF con- ference on computer vision and pattern recognition, pages 2359–2368, 2021. 6

  4. [12]

    Learned step size quantization.arXiv preprint arXiv:1902.08153, 2019

    Steven K Esser, Jeffrey L McKinstry, Deepika Bablani, Rathinakumar Appuswamy, and Dharmendra S Modha. Learned step size quantization.arXiv preprint arXiv:1902.08153, 2019. 2, 6

  5. [13]

    Erasing concepts from diffusion models

    Rohit Gandikota, Joanna Materzynska, Jaden Fiotto- Kaufman, and David Bau. Erasing concepts from diffusion models. InProceedings of the IEEE/CVF International Con- ference on Computer Vision, pages 2426–2436, 2023. 14

  6. [14]

    A survey of quan- tization methods for efficient neural network inference

    Amir Gholami, Sehoon Kim, Zhen Dong, Zhewei Yao, Michael W Mahoney, and Kurt Keutzer. A survey of quan- tization methods for efficient neural network inference. In Low-Power Computer Vision, pages 291–326. Chapman and Hall/CRC, 2022. 1

  7. [15]

    Efficientdm: Efficient quantization-aware fine- tuning of low-bit diffusion models

    Yefei He, Jing Liu, Weijia Wu, Hong Zhou, and Bohan Zhuang. Efficientdm: Efficient quantization-aware fine- tuning of low-bit diffusion models. InInternational Con- ference on Learning Representations, 2024. 2

  8. [16]

    Ptqd: Accurate post-training quantization for diffusion models.Advances in Neural Information Pro- cessing Systems, 36, 2024

    Yefei He, Luping Liu, Jing Liu, Weijia Wu, Hong Zhou, and Bohan Zhuang. Ptqd: Accurate post-training quantization for diffusion models.Advances in Neural Information Pro- cessing Systems, 36, 2024. 1, 2, 3, 6, 7, 12, 14

  9. [17]

    Clipscore: A reference-free evaluation met- ric for image captioning.arXiv preprint arXiv:2104.08718,

    Jack Hessel, Ari Holtzman, Maxwell Forbes, Ronan Le Bras, and Yejin Choi. Clipscore: A reference-free evaluation met- ric for image captioning.arXiv preprint arXiv:2104.08718,

  10. [18]

    Gans trained by a two time-scale update rule converge to a local nash equilib- rium.Advances in neural information processing systems, 30, 2017

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilib- rium.Advances in neural information processing systems, 30, 2017. 4, 7

  11. [19]

    Classifier-free diffusion guidance.arXiv preprint arXiv:2207.12598, 2022

    Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance.arXiv preprint arXiv:2207.12598, 2022. 12

  12. [20]

    Denoising dif- fusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising dif- fusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020. 1, 3, 5

  13. [21]

    Video dif- fusion models.Advances in Neural Information Processing Systems, 35:8633–8646, 2022

    Jonathan Ho, Tim Salimans, Alexey Gritsenko, William Chan, Mohammad Norouzi, and David J Fleet. Video dif- fusion models.Advances in Neural Information Processing Systems, 35:8633–8646, 2022. 1

  14. [22]

    Tfmq-dm: Temporal feature maintenance quantization for diffusion models

    Yushi Huang, Ruihao Gong, Jing Liu, Tianlong Chen, and Xianglong Liu. Tfmq-dm: Temporal feature maintenance quantization for diffusion models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7362–7371, 2024. 1, 2, 3, 4, 6, 7, 8, 12, 1...

  15. [23]

    Expanding expressiveness of diffusion models with limited data via self-distillation based fine-tuning

    Jiwan Hur, Jaehyun Choi, Gyojin Han, Dong-Jae Lee, and Junmo Kim. Expanding expressiveness of diffusion models with limited data via self-distillation based fine-tuning. In Proceedings of the IEEE/CVF Winter Conference on Appli- cations of Computer Vision, pages 5028–5037, 2024. 1

  16. [24]

    Scalable adap- tive computation for iterative generation.arXiv preprint arXiv:2212.11972, 2022

    Allan Jabri, David Fleet, and Ting Chen. Scalable adap- tive computation for iterative generation.arXiv preprint arXiv:2212.11972, 2022. 1

  17. [25]

    Quantization and training of neural networks for efficient integer-arithmetic-only inference

    Benoit Jacob, Skirmantas Kligys, Bo Chen, Menglong Zhu, Matthew Tang, Andrew Howard, Hartwig Adam, and Dmitry 9 Kalenichenko. Quantization and training of neural networks for efficient integer-arithmetic-only inference. InProceed- ings of the IEEE Conference on Computer Vision...

  18. [26]

    Learning to quantize deep networks by op- timizing quantization intervals with task loss

    Sangil Jung, Changyong Son, Seohyung Lee, Jinwoo Son, Jae-Joon Han, Youngjun Kwak, Sung Ju Hwang, and Changkyu Choi. Learning to quantize deep networks by op- timizing quantization intervals with task loss. InProceedings of the IEEE/CVF conference on computer vision and patter...

  19. [27]

    A style-based generator architecture for generative adversarial networks

    Tero Karras, Samuli Laine, and Timo Aila. A style-based generator architecture for generative adversarial networks. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4401–4410, 2019. 7

  20. [28]

    Imagine flash: Accelerating emu diffusion models with backward distillation.arXiv preprint arXiv:2405.05224,

    Jonas Kohler, Albert Pumarola, Edgar Sch ¨onfeld, Artsiom Sanakoyeu, Roshan Sumbaly, Peter Vajda, and Ali Tha- bet. Imagine flash: Accelerating emu diffusion models with backward distillation.arXiv preprint arXiv:2405.05224,

  21. [29]

    Quantization for rapid deployment of deep neural networks.arXiv preprint arXiv:1810.05488,

    Jun Haeng Lee, Sangwon Ha, Saerom Choi, Won-Jo Lee, and Seungwon Lee. Quantization for rapid deployment of deep neural networks.arXiv preprint arXiv:1810.05488,

  22. [30]

    Srdiff: Single image super-resolution with diffusion probabilistic models

    Haoying Li, Yifan Yang, Meng Chang, Shiqi Chen, Huajun Feng, Zhihai Xu, Qi Li, and Yueting Chen. Srdiff: Single image super-resolution with diffusion probabilistic models. Neurocomputing, 479:47–59, 2022. 1

  23. [31]

    Alleviating exposure bias in diffusion mod- els through sampling with shifted time steps.arXiv preprint arXiv:2305.15583, 2023

    Mingxiao Li, Tingyu Qu, Ruicong Yao, Wei Sun, and Marie- Francine Moens. Alleviating exposure bias in diffusion mod- els through sampling with shifted time steps.arXiv preprint arXiv:2305.15583, 2023. 3

  24. [32]

    Q-diffusion: Quantizing diffusion models

    Xiuyu Li, Yijiang Liu, Long Lian, Huanrui Yang, Zhen Dong, Daniel Kang, Shanghang Zhang, and Kurt Keutzer. Q-diffusion: Quantizing diffusion models. InProceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 17535–17545, 2023. 1, 2, 3, 4, 5, 6, 7, 14, 15, 16

  25. [33]

    Brecq: Pushing the limit of post-training quantization by block reconstruc- tion.arXiv preprint arXiv:2102.05426, 2021

    Yuhang Li, Ruihao Gong, Xu Tan, Yang Yang, Peng Hu, Qi Zhang, Fengwei Yu, Wei Wang, and Shi Gu. Brecq: Pushing the limit of post-training quantization by block reconstruc- tion.arXiv preprint arXiv:2102.05426, 2021. 2, 6

  26. [34]

    Q-dm: An efficient low-bit quantized dif- fusion model

    Yanjing Li, Sheng Xu, Xianbin Cao, Xiao Sun, and Baochang Zhang. Q-dm: An efficient low-bit quantized dif- fusion model. InAdvances in Neural Information Processing Systems, pages 76680–76691. Curran Associates, Inc., 2023. 2

  27. [35]

    Magic3d: High-resolution text-to-3d content creation

    Chen-Hsuan Lin, Jun Gao, Luming Tang, Towaki Takikawa, Xiaohui Zeng, Xun Huang, Karsten Kreis, Sanja Fidler, Ming-Yu Liu, and Tsung-Yi Lin. Magic3d: High-resolution text-to-3d content creation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition...

  28. [36]

    Awq: Activation-aware weight quantization for on-device llm compression and acceleration

    Ji Lin, Jiaming Tang, Haotian Tang, Shang Yang, Wei-Ming Chen, Wei-Chen Wang, Guangxuan Xiao, Xingyu Dang, Chuang Gan, and Song Han. Awq: Activation-aware weight quantization for on-device llm compression and acceleration. Proceedings of Machine Learning and Systems, 6:87–100,

  29. [37]

    Microsoft coco: Common objects in context

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll´ar, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceeding...

  30. [38]

    Focal loss for dense object detection

    Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Doll´ar. Focal loss for dense object detection. InPro- ceedings of the IEEE international conference on computer vision, pages 2980–2988, 2017. 4

  31. [39]

    En- hanced distribution alignment for post-training quantization of diffusion models.arXiv preprint arXiv:2401.04585, 2024

    Xuewen Liu, Zhikai Li, Junrui Xiao, and Qingyi Gu. En- hanced distribution alignment for post-training quantization of diffusion models.arXiv preprint arXiv:2401.04585, 2024. 1, 2, 3, 6, 7, 8, 14

  32. [40]

    Repaint: Inpainting using denoising diffusion probabilistic models

    Andreas Lugmayr, Martin Danelljan, Andres Romero, Fisher Yu, Radu Timofte, and Luc Van Gool. Repaint: Inpainting using denoising diffusion probabilistic models. InProceed- ings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11461–11471, 2022. 1

  33. [41]

    Data-free quantization through weight equal- ization and bias correction

    Markus Nagel, Mart van Baalen, Tijmen Blankevoort, and Max Welling. Data-free quantization through weight equal- ization and bias correction. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2019. 2

  34. [42]

    Up or down? adap- tive rounding for post-training quantization

    Markus Nagel, Rana Ali Amjad, Mart Van Baalen, Chris- tos Louizos, and Tijmen Blankevoort. Up or down? adap- tive rounding for post-training quantization. InInternational Conference on Machine Learning, pages 7197–7206. PMLR,

  35. [43]

    A white paper on neural network quantization.arXiv preprint arXiv:2106.08295, 2021

    Markus Nagel, Marios Fournarakis, Rana Ali Amjad, Yely- sei Bondarenko, Mart Van Baalen, and Tijmen Blankevoort. A white paper on neural network quantization.arXiv preprint arXiv:2106.08295, 2021. 1, 4

  36. [44]

    Generating images with sparse representations

    Charlie Nash, Jacob Menick, Sander Dieleman, and Peter W Battaglia. Generating images with sparse representations. arXiv preprint arXiv:2103.03841, 2021. 7

  37. [45]

    Improved denoising diffusion probabilistic models

    Alexander Quinn Nichol and Prafulla Dhariwal. Improved denoising diffusion probabilistic models. InInternational conference on machine learning, pages 8162–8171. PMLR,

  38. [46]

    Elucidating the exposure bias in diffusion models.arXiv preprint arXiv:2308.15321, 2023

    Mang Ning, Mingxiao Li, Jianlin Su, Albert Ali Salah, and Itir Onal Ertugrul. Elucidating the exposure bias in diffusion models.arXiv preprint arXiv:2308.15321, 2023. 3

  39. [47]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. InProceedings of the IEEE/CVF Inter- national Conference on Computer Vision, pages 4195–4205,

  40. [48]

    Barron, and Ben Milden- hall

    Ben Poole, Ajay Jain, Jonathan T. Barron, and Ben Milden- hall. Dreamfusion: Text-to-3d using 2d diffusion. InThe Eleventh International Conference on Learning Representa- tions, 2023. 1

  41. [49]

    Swish: a self-gated activation function.arXiv preprint arXiv:1710.05941, 7(1):5, 2017

    Prajit Ramachandran, Barret Zoph, and Quoc V Le. Swish: a self-gated activation function.arXiv preprint arXiv:1710.05941, 7(1):5, 2017. 5

  42. [50]

    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. InProceedings of 10 the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022. 1, 5, 7, 15, 16

  43. [51]

    Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation

    Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael Pritch, Michael Rubinstein, and Kfir Aberman. Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 22500–...

  44. [52]

    Improved techniques for training gans.Advances in neural information processing systems, 29, 2016

    Tim Salimans, Ian Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, and Xi Chen. Improved techniques for training gans.Advances in neural information processing systems, 29, 2016. 7

  45. [53]

    Post-training quantization on diffusion models

    Yuzhang Shang, Zhihang Yuan, Bin Xie, Bingzhe Wu, and Yan Yan. Post-training quantization on diffusion models. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 1972–1981, 2023. 1, 2, 3, 14

  46. [54]

    Dragdiffusion: Harnessing diffusion models for interactive point-based image editing

    Yujun Shi, Chuhui Xue, Jun Hao Liew, Jiachun Pan, Han- shu Yan, Wenqing Zhang, Vincent YF Tan, and Song Bai. Dragdiffusion: Harnessing diffusion models for interactive point-based image editing. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognitio...

  47. [55]

    Make-a-video: Text-to-video generation without text-video data.arXiv preprint arXiv:2209.14792,

    Uriel Singer, Adam Polyak, Thomas Hayes, Xi Yin, Jie An, Songyang Zhang, Qiyuan Hu, Harry Yang, Oron Ashual, Oran Gafni, et al. Make-a-video: Text-to-video generation without text-video data.arXiv preprint arXiv:2209.14792,

  48. [56]

    Temporal dynamic quantization for dif- fusion models.Advances in Neural Information Processing Systems, 36, 2024

    Junhyuk So, Jungwon Lee, Daehyun Ahn, Hyungjun Kim, and Eunhyeok Park. Temporal dynamic quantization for dif- fusion models.Advances in Neural Information Processing Systems, 36, 2024. 1, 2, 3, 14

  49. [57]

    Deep unsupervised learning using nonequilibrium thermodynamics

    Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. InInternational confer- ence on machine learning, pages 2256–2265. PMLR, 2015. 3

  50. [58]

    Denoising diffusion implicit models.arXiv preprint arXiv:2010.02502, 2020

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models.arXiv preprint arXiv:2010.02502, 2020. 7

  51. [59]

    Bitsfusion: 1.99 bits weight quantization of diffusion model.arXiv preprint arXiv:2406.04333, 2024

    Yang Sui, Yanyu Li, Anil Kag, Yerlan Idelbayev, Junli Cao, Ju Hu, Dhritiman Sagar, Bo Yuan, Sergey Tulyakov, and Jian Ren. Bitsfusion: 1.99 bits weight quantization of diffusion model.arXiv preprint arXiv:2406.04333, 2024. 7, 14

  52. [60]

    Attention is all you need.Advances in Neural Information Processing Systems, 2017

    A Vaswani. Attention is all you need.Advances in Neural Information Processing Systems, 2017. 3

  53. [61]

    Towards accurate post-training quantization for diffusion models

    Changyuan Wang, Ziwei Wang, Xiuwei Xu, Yansong Tang, Jie Zhou, and Jiwen Lu. Towards accurate post-training quantization for diffusion models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 16026–16035, 2024. 1, 2, 3

  54. [62]

    Outlier suppression: Pushing the limit of low-bit transformer language models.Advances in Neural Informa- tion Processing Systems, 35:17402–17414, 2022

    Xiuying Wei, Yunchen Zhang, Xiangguo Zhang, Ruihao Gong, Shanghang Zhang, Qi Zhang, Fengwei Yu, and Xian- glong Liu. Outlier suppression: Pushing the limit of low-bit transformer language models.Advances in Neural Informa- tion Processing Systems, 35:17402–17414, 2022. 3

  55. [63]

    Outlier suppression+: Accurate quantization of large language mod- els by equivalent and optimal shifting and scaling.arXiv preprint arXiv:2304.09145, 2023

    Xiuying Wei, Yunchen Zhang, Yuhang Li, Xiangguo Zhang, Ruihao Gong, Jinyang Guo, and Xianglong Liu. Outlier suppression+: Accurate quantization of large language mod- els by equivalent and optimal shifting and scaling.arXiv preprint arXiv:2304.09145, 2023. 2, 3, 5

  56. [64]

    Integer quantization for deep learn- ing inference: Principles and empirical evaluation.arXiv preprint arXiv:2004.09602, 2020

    Hao Wu, Patrick Judd, Xiaojie Zhang, Mikhail Isaev, and Paulius Micikevicius. Integer quantization for deep learn- ing inference: Principles and empirical evaluation.arXiv preprint arXiv:2004.09602, 2020. 12

  57. [65]

    Ptq4dit: Post-training quantization for diffu- sion transformers

    Junyi Wu, Haoxuan Wang, Yuzhang Shang, Mubarak Shah, and Yan Yan. Ptq4dit: Post-training quantization for diffu- sion transformers. InNeurIPS, 2024. 3

  58. [66]

    Smoothquant: Accurate and effi- cient post-training quantization for large language models

    Guangxuan Xiao, Ji Lin, Mickael Seznec, Hao Wu, Julien Demouth, and Song Han. Smoothquant: Accurate and effi- cient post-training quantization for large language models. InInternational Conference on Machine Learning, pages 38087–38099. PMLR, 2023. 2, 3, 5

  59. [67]

    Timestep-aware cor- rection for quantized diffusion models.arXiv preprint arXiv:2407.03917, 2024

    Yuzhe Yao, Feng Tian, Jun Chen, Haonan Lin, Guang Dai, Yong Liu, and Jingdong Wang. Timestep-aware cor- rection for quantized diffusion models.arXiv preprint arXiv:2407.03917, 2024. 1, 2, 3, 7, 14

  60. [68]

    Zeroquant: Ef- ficient and affordable post-training quantization for large- scale transformers.Advances in Neural Information Process- ing Systems, 35:27168–27183, 2022

    Zhewei Yao, Reza Yazdani Aminabadi, Minjia Zhang, Xi- aoxia Wu, Conglong Li, and Yuxiong He. Zeroquant: Ef- ficient and affordable post-training quantization for large- scale transformers.Advances in Neural Information Process- ing Systems, 35:27168–27183, 2022. 3

  61. [69]

    One-step diffusion with distribution matching distillation

    Tianwei Yin, Micha ¨el Gharbi, Richard Zhang, Eli Shecht- man, Fredo Durand, William T Freeman, and Taesung Park. One-step diffusion with distribution matching distillation. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 6613–6623...

  62. [70]

    Shiftaddnet: A hardware-inspired deep network.Advances in Neural Information Processing Systems, 33:2771–2783,

    Haoran You, Xiaohan Chen, Yongan Zhang, Chaojian Li, Sicheng Li, Zihao Liu, Zhangyang Wang, and Yingyan Lin. Shiftaddnet: A hardware-inspired deep network.Advances in Neural Information Processing Systems, 33:2771–2783,

  63. [71]

    Shiftaddllm: Accelerating pretrained llms via post-training multiplication-less reparameterization

    Haoran You, Yipin Guo, Yichao Fu, Wei Zhou, Huihong Shi, Xiaofan Zhang, Souvik Kundu, Amir Yazdanbakhsh, and Yingyan Celine Lin. Shiftaddllm: Accelerating pretrained llms via post-training multiplication-less reparameterization. arXiv preprint arXiv:2406.05981, 2024. 6

  64. [72]

    Lsun: Construction of a large-scale image dataset using deep learning with humans in the loop.arXiv preprint arXiv:1506.03365, 2015

    Fisher Yu, Ari Seff, Yinda Zhang, Shuran Song, Thomas Funkhouser, and Jianxiong Xiao. Lsun: Construction of a large-scale image dataset using deep learning with humans in the loop.arXiv preprint arXiv:1506.03365, 2015. 7

  65. [73]

    Text-to-3d with classifier score distillation.arXiv preprint arXiv:2310.19415, 2023

    Xin Yu, Yuan-Chen Guo, Yangguang Li, Ding Liang, Song- Hai Zhang, and Xiaojuan Qi. Text-to-3d with classifier score distillation.arXiv preprint arXiv:2310.19415, 2023. 1

  66. [74]

    Adding conditional control to text-to-image diffusion models, 2023

    Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models, 2023. 1, 14

  67. [75]

    The unreasonable effectiveness of deep features as a perceptual metric

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. InCVPR, 2018. 7

  68. [76]

    Vidit-q: Efficient and accurate quantization of diffusion transformers for im- age and video generation.arXiv preprint arXiv:2406.02540,

    Tianchen Zhao, Tongcheng Fang, Enshu Liu, Wan Rui, Widyadewi Soedarmadji, Shiyao Li, Zinan Lin, Guohao Dai, Shengen Yan, Huazhong Yang, et al. Vidit-q: Efficient and accurate quantization of diffusion transformers for im- age and video generation.arXiv preprint arXiv:2406.02540,

  69. [78]

    Thus, we modify their code to quantize those layers for fair comparison

    and TFMQ-DM [22] do not quantizeskip-connection, downsample, andupsamplelayers. Thus, we modify their code to quantize those layers for fair comparison. Then we directly run original code provided by the baselines. To control for metric fluctuations caused by differences in ge...

  70. [2024]

    Implementation details This section provides a more detailed description of the experimental implementation presented in the main manuscript

    3 11 A. Implementation details This section provides a more detailed description of the experimental implementation presented in the main manuscript. Tab. 8 summarizes the hyperparameters used across all experiments for W4A8. The number of sampling steps for generating calibra...

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.