Pith. sign in

REVIEW 3 major objections 5 minor 60 references

KroQuant claims that a learned 32×32 Kronecker block transform, run online before quantizing each linear layer, closes most of the quality gap between W4A4 and full-precision diffusion transformers while adding no inference slowdown.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 07:25 UTC pith:QWC7E2KV

load-bearing objection Learned Kronecker block transforms for DiT PTQ are a genuinely new and plausible idea with clean math and a fair controlled ablation, but the abstract oversells the results and the 32-block assumption is load-bearing and untested. the 3 major comments →

arxiv 2607.21446 v1 pith:QWC7E2KV submitted 2026-07-23 cs.LG cs.CV

KroQuant: Kronecker-Structured Block Transforms for Efficient Post-Training Quantization of Diffusion Transformers

Yann Bouquet , Alireza Khodamoradi , Kristof Denolf , Mathieu Salzmann This is my paper
classification cs.LG cs.CV
keywords post-training quantizationdiffusion transformersactivation outliersKronecker productblock-diagonal transform4-bit quantizationMXFP4efficient inference
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper is trying to solve a specific bottleneck: when diffusion transformers are quantized to 4-bit weights and activations, a few outlier channels dominate the 4-bit dynamic range and ruin output quality. The usual remedy is to rotate or rescale activations before quantizing and fold the inverse into the weights, but in diffusion transformers the normalization layers between blocks make this impossible to do offline, so the transform must run online at every denoising step. KroQuant's claim is that the right transform for this setting is a learned block-diagonal matrix with 32×32 blocks, parameterized as a Kronecker product of five 2×2 unit-determinant factors and initialized at the Hadamard transform. This gives more expressivity than a fixed Hadamard at the same block size and fewer parameters than per-channel scaling, while applying as a single tensor-core GEMM per block. Combined with LoRaQ's offline weight calibration, the paper reports outputs closer to the FP16 reference than SVDQuant and LoRaQ-with-SmoothQuant on PixArt-Σ, SANA, and most of FLUX.1-schnell at W4A4, with a quantizer kernel up to 14% faster than per-channel scaling.

Core claim

On the paper's own terms, the central discovery is that the expressivity-versus-cost trade-off for online activation transforms in diffusion-transformer quantization has a sweet spot: a learned, block-diagonal transform whose 32×32 blocks align exactly with the block width of the MXFP4e2 quantizer. Each block is built as a Kronecker product of five 2×2 factors, each factor parameterized by three scalars through a unit-determinant LU form, so a full 32×32 block costs 15 learnable parameters. Since the Hadamard matrix is itself a Kronecker product of five identical 2×2 factors, the transform can be initialized at a block-Hadamard rotation and then tuned per layer with a straight-through estima

What carries the argument

The load-bearing object is the Kronecker-structured 32×32 invertible block transform T, block-diagonal with blocks K^(j) = G1⊗G2⊗G3⊗G4⊗G5, where each G_k is a 2×2 unit-determinant LU matrix with three scalars. It mixes channels inside each 32-element block before the MX quantizer, spreading outlier magnitudes so the 4-bit format is used more evenly; it is learned per layer by minimizing quantized output MSE with auxiliary per-operand regularization, initialized at the Hadamard transform, and applied online as small tensor-core GEMMs, with its inverse folded into weights offline. The two kernel variants — precomputed dense blocks (K1) and on-the-fly Kronecker reconstruction from 15 scalars (K

Load-bearing premise

The load-bearing premise is that activation outliers in DiT linear layers can be neutralized by mixing only within independent 32-element blocks aligned with the MX quantizer; if outlier structure spans across blocks or involves long-range channel correlations, KroQuant cannot suppress it and its quality advantage shrinks toward per-channel scaling.

What would settle it

Run KroQuant's calibration on a synthetic linear layer whose activation distribution has two large outlier channels separated by more than 32 indices, so no single 32×32 block can mix them, then quantize at W4A4; the method should fail to remove those outliers and the output error should be close to a per-channel-scaling baseline. Alternatively, take a calibrated DiT layer, randomly permute the channel dimension of the learned transform, and re-measure LPIPS or output MSE: if the observed gains over SmoothQuant largely disappear, the block-local alignment — not the learned rotation per se — is

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • At W4A4 with the MXFP4e2 format, KroQuant-with-LoRaQ beats SVDQuant and LoRaQ-with-SmoothQuant on LPIPS and PSNR for PixArt-Σ and SANA on both MJHQ-30K and SDCI, and closes most of the FID gap to the FP16 reference.
  • The learned transform strictly improves on its fixed Hadamard initialization: holding LoRaQ fixed, the 15 learnable parameters per block cut LPIPS by 0.035 and raise PSNR by 0.8 dB on PixArt-Σ.
  • The kernel cost is not a barrier: K1 matches or beats per-channel scaling on every benchmarked shape (1.00–1.14×) and K2 is 1.30–2.45× faster than a fast Walsh–Hadamard transform, while handling hidden dimensions divisible by 32 without padding.
  • The transform is a drop-in replacement for the SmoothQuant front-end in weight-splitting PTQ pipelines, so the LoRaQ weight correction can be reused unchanged.
  • On FLUX.1's double-stream architecture the gain is partial (best FID/PSNR but not IR/LPIPS on MJHQ-30K), which the paper attributes to joint attention mixing text and image tokens with different activation distributions.

Where Pith is reading between the lines

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

  • Editorial inference: because the block size is tied to the MX quantizer's 32-wide blocks, the same construction should transfer to other blockwise formats (e.g., 16-element blocks) by changing the number of Kronecker factors, suggesting a general recipe for learned rotation at quantizer granularity.
  • Editorial inference: the 15-parameter-per-block parametrization sits between diagonal scaling (32 scalars per block) and a dense rotation (1024 scalars), so one could tune expressivity per layer by choosing how many Kronecker factors to learn; the paper's fixed five-factor choice is a conservative default.
  • Editorial inference: a direct stress test of the paper's core assumption is to permute channel indices after calibration before applying the learned transform; if quality drops toward the SmoothQuant baseline, the gains depend on outlier correlations coinciding with 32-channel blocks rather than on the learned rotation per se.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes KroQuant, a post-training quantization method for diffusion transformers that inserts a learned Kronecker-structured block-diagonal transform into the activation path before W4A4 MXFP4 quantization. Each 32×32 block is parameterized as a Kronecker product of five 2×2 unit-determinant LU factors, giving 15 learnable parameters per block, with the Hadamard transform as initialization. The transform is applied online (because AdaLN prevents offline absorption), and the authors provide fused Triton kernels (K1 dense, K2 on-the-fly) claiming speed comparable to or better than per-channel scaling. After transform calibration, LoRaQ weight calibration is applied. Experiments on PixArt-Σ, SANA, and FLUX.1-schnell report FID, IR, LPIPS, and PSNR for MJHQ-30K and SDCI, plus kernel-latency benchmarks on an MI350. The paper also includes an ablation, kernel pseudocode, a normalization-commutativity study, and a limitations section.

Significance. If the results hold, KroQuant is a useful contribution to the DiT PTQ toolbox: it is, to my knowledge, the first learned block-diagonal activation transform specifically matched to MXFP4's 32-wide blocks, and it is shown to improve over both a fixed Hadamard at the same block size and over per-channel scaling within a LoRaQ-style pipeline. The parameter-count argument and invertibility construction are clean and clearly stated. The kernel measurements provide credible evidence that the online cost is competitive with per-channel scaling on the tested MI350 shapes. The authors are also candid about limitations: FLUX results are mixed, the Kronecker parameterization grows too slowly beyond n=32, and joint optimization is unstable. However, the abstract's blanket claim of superiority over SVDQuant and LoRaQ is contradicted by the paper's own Table 1 on FLUX.1-schnell/MJHQ-30K, and the central 32-block assumption is not directly tested. The contribution is therefore significant but its scope must be stated more carefully.

major comments (3)
  1. [Abstract and §4.2, Table 1] The abstract states that KroQuant 'produces outputs closer to the FP reference than SVDQuant and LoRaQ on MJHQ-30K and SDCI.' Table 1 contradicts this on FLUX.1-schnell/MJHQ-30K: LPIPS is 0.336 versus 0.331 for LoRaQ and IR is 0.880 versus 0.952 for SVDQuant and 0.921 for LoRaQ. While KroQuant has the best FID and PSNR there, the claim of being 'closer to FP' is not metric-wise true. The main text acknowledges this as 'mixed,' but the abstract does not. This overreach should be corrected, and the conclusions should be scoped to the configurations and metrics actually supporting the claim.
  2. [§3.2, Eq. (4), and §4.2] The load-bearing design assumption is that consecutive 32-element blocks can capture all relevant activation outlier structure. The transform T is block-diagonal with 32×32 blocks and cannot mix across block boundaries. The FLUX MJHQ results are consistent with this assumption failing in double-stream layers, but the paper does not provide a direct test. No experiment varies the block size, permutes channel grouping, or measures how much quantization-error reduction would be obtained by allowing cross-block coupling. Since the central claim depends on the 32-block structure being sufficient, the authors should either add such diagnostics or explicitly reframe the method as targeting architectures where this assumption holds (e.g., the single-stream PixArt/SANA settings).
  3. [§5, Limitations and §3.2] The limitations note that the Kronecker construction has only 3 log2 n parameters per n×n block and 'grows too slowly with block size,' so the method is tied to n=32. This is more than a future-work remark: it means the paper does not actually deliver a general solution to the 'need' articulated in Section 1 for a transform 'more expressive than a fixed Hadamard at the same block size' — at n=32 it is more expressive, but the architecture cannot scale to the larger block sizes that would test the boundary assumption. The authors should acknowledge this directly in the main text and avoid implying that the method generalizes beyond the n=32 setting.
minor comments (5)
  1. [Table 3 and Table 4] The speedup ratio notation 'Scale/K1 Speedup' is confusing. The caption explains that 'Method B is X faster than Method A,' but the column header suggests the opposite. Please rename columns to e.g. 'K1 vs Scale' and state the convention near the table.
  2. [§3.2, Eq. (6)] The parametrization G_k = L_k U_k with U_k = [[a_k, b_k],[0, -1/a_k]] has det G_k = -1 for every k. Since the product of five factors yields det K = -1, all K blocks are in the determinant -1 component of GL(32). This is not a problem for the method, but it may be worth stating explicitly to avoid confusion with unit-determinant 'for every factor' language.
  3. [§4.1] The calibration and evaluation details are mostly present, but the evaluation protocol for MJHQ-30K and SDCI could say explicitly how many seeds are used and whether the same 5,000 images are used for all methods. The phrase 'following the evaluation split used in prior work' is a bit vague.
  4. [Appendix A] Table 4 reports a 0.99× Scale/K1 ratio at (m=4096, d=12288), while the text says 'matches or improves on per-channel scaling on every shape except the largest.' The exception is not highlighted in the table; please add a footnote or marker.
  5. [Throughout] There are a few typos and inconsistent spellings: 'KroQuant' vs 'Kroquant' in Appendix D.2, 'Pixart-Σ' vs 'PixArt-Σ,' and 'sDCI' in the assets list. Please proofread newly introduced names.

Circularity Check

0 steps flagged

No significant circularity: KroQuant's transform is calibrated on held-out data and its improvements over baselines are empirically measured, not entailed by its own equations.

full rationale

The derivation chain is self-contained. Equation (1) defines the transform-invariance property AT(W T^{-T})^T = AW^T, Equation (2) defines the quantization error being minimized, and Equations (4)-(6) define the block-diagonal Kronecker-LU parametrization. The transform T is optimized with the objective in Equation (7) on 128 COCO calibration activations (Section 4.1), while all headline results in Table 1 are reported on held-out MJHQ-30K and SDCI test sets. No reported 'prediction' is a fitted parameter renamed as a result: the output-quality numbers are measured on disjoint evaluation data. The ablation in Table 2 compares the optimized Kronecker transform against its Hadamard initialization under the same LoRaQ backbone; the measured gains (LPIPS 0.347 vs. 0.382, PSNR 16.5 vs. 15.7 on PixArt-Σ/MJHQ-30K) are empirical, not guaranteed by the equations. The LoRaQ weight-calibration component is taken from the authors' prior work (Bouquet et al. 2026), but the paper reports the LoRaQ-with-SmoothQuant baseline in the same table, and Table 2's standalone rows without any LoRaQ stage show KroQuant alone reduces FID from 130.6 to 41.8 relative to per-channel smoothing. The comparison therefore does not reduce to a self-citation. No uniqueness theorem is invoked, and no ansatz is smuggled in via citation: the 32×32 block choice is explicitly stated as matching the MXFP4e2 quantizer's block structure (Sections 3.1-3.2). The mixed FLUX.1/MJHQ-30K IR result (0.880 vs. SVDQuant's 0.952) is reported as a limitation and is a robustness/correctness concern, not a circularity. Overall, the central claim is independently supported by controlled comparisons and held-out evaluation.

Axiom & Free-Parameter Ledger

4 free parameters · 6 axioms · 0 invented entities

No new physical or mathematical entities are postulated. The only new object is the learned transform family itself, which is a parametrization rather than a separate entity. The design deliberately fits T to the quantizer's own block structure, and no external theoretical guarantee is provided; the net contribution is a new parameterization plus empirical support.

free parameters (4)
  • Kronecker factor parameters (a_k, b_k, c_k) for each 32×32 block = Not reported; 15 per block × ceil(d/32) per layer
    Optimized via Adam on Eq. 9 with a straight-through estimator (Section 3.3). These are the learned transform entries; no final parameter values or checkpoints are shipped.
  • Regularization weights λ_A, λ_W = Set by loss ratios at initialization (Eq. 10)
    Chosen so auxiliary terms match the output-loss magnitude at the Hadamard initialization; they steer the optimization and are data-dependent.
  • LoRaQ low-rank correction components L^(ℓ), R^(ℓ) = Rank r=128; final values not reported
    Fitted data-free to residual quantization error as part of the inherited LoRaQ pipeline (Section 3.4); central to the final pipeline but borrowed without independent re-derivation in this paper.
  • Calibration and optimization schedule = 100 Adam steps, lr=1e-2, 128 COCO prompts
    Manual hyperparameters chosen following prior practice; no sensitivity analysis is provided (Appendix D.1).
axioms (6)
  • standard math Kronecker product inverse factorizes: (G1⊗...⊗G5)^{-1}=G1^{-1}⊗...⊗G5^{-1}
    Used in Section 3.2 to build an invertible T; standard linear algebra.
  • standard math The unit-determinant LU form (Eq. 6) with a_k≠0 gives |det G_k|=1 and includes reflection/rotation matrices
    Determinant computation in Section 3.2; true by construction.
  • domain assumption AdaLN and other axis-aligned normalizations cannot commute with a non-diagonal online transform
    Eq. 3 and Appendix C show LayerNorm/BatchNorm/GroupNorm/AdaLN incur O(1) deviation under a random orthogonal rotation; load-bearing because it forces the online transform.
  • ad hoc to paper Outlier suppression within 32-element blocks is sufficient for W4A4 quality
    The whole method chooses block size n=32 to match the MXFP4e2 quantizer (Section 3.2). The FLUX/MJHQ mixed results indicate this assumption is not fully satisfied for double-stream architectures.
  • domain assumption Straight-through estimator gradients through the quantizer are a valid optimization signal
    Section 3.3 uses STE to propagate through Q_b; no convergence guarantee is given.
  • domain assumption Calibration set of 128 COCO captions generalizes to MJHQ-30K and SDCI test prompts
    Standard PTQ assumption; not validated by ablation except via final performance.

pith-pipeline@v1.3.0-alltime-deepseek · 14793 in / 19925 out tokens · 167603 ms · 2026-08-01T07:25:29.419101+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of KroQuant: Kronecker-Structured Block Transforms for Efficient Post-Training Quantization of Diffusion Transformers." pith.science (2026). https://pith.science/paper/QWC7E2KV

@misc{pith2026260721446,
  author       = {Pith},
  title        = {Pith review of: KroQuant: Kronecker-Structured Block Transforms for Efficient Post-Training Quantization of Diffusion Transformers},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QWC7E2KV}},
  note         = {Machine review of arXiv:2607.21446}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Post-training quantization (PTQ) of diffusion transformers (DiTs) to W4A4 severely degrades output quality, because activations entering each linear layer contain outliers that 4-bit formats cannot represent. The standard fix applies an invertible linear transform to the activations and its inverse to the weights before quantizing both. Normalization layers between blocks force this transform to run online at every denoising step, making its inference computation cost the binding design constraint. Existing options trade quantization quality for inference cost: per-channel scaling (SmoothQuant) is computationally cheap but impacts the magnitude of the channels, which can harm quantization accuracy; fixed Hadamard transforms yield better quantization accuracy but require large block sizes that incur a high online cost; learned full-$d$ invertible transforms calibrate best but entail a prohibitive dense $d \times d$ matrix multiplication (GEMM) per layer per step. We propose KroQuant, a PTQ method that applies a learned Kronecker-structured invertible transform to each 32-element block of the activation, storing less than half the parameters of per-channel scaling. The block-local structure runs as small tensor-core GEMMs, and on an MI350 GPU the KroQuant quantizer kernel is up to $14\%$ faster than the SmoothQuant kernel. Offline LoRaQ weight calibration then absorbs the residual per-weight quantization error. On PixArt-$\Sigma$, SANA, and FLUX.1-schnell at W4A4 (MXFP4e2), KroQuant produces outputs closer to the FP reference than SVDQuant and LoRaQ on MJHQ-30K and SDCI, while preserving or improving image quality.

Figures

Figures reproduced from arXiv: 2607.21446 by Alireza Khodamoradi, Kristof Denolf, Mathieu Salzmann, Yann Bouquet.

Figure 3
Figure 3. Figure 3: Median kernel latency and effective HBM bandwidth vs. token count [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

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

60 extracted references · 5 linked inside Pith

  1. [1]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Bao, Fan and Nie, Shen and Xue, Kaiwen and Cao, Yue and Li, Chongxuan and Su, Hang and Zhu, Jun , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2023 , pages =

    2023

  2. [2]

    ICLR , year =

    PixArt-alpha: Fast Training of Diffusion Transformer for Photorealistic Text-to-Image Synthesis , author =. ICLR , year =

  3. [3]

    2025 , url=

    Enze Xie and Junsong Chen and Junyu Chen and Han Cai and Haotian Tang and Yujun Lin and Zhekai Zhang and Muyang Li and Ligeng Zhu and Yao Lu and Song Han , booktitle=. 2025 , url=

    2025

  4. [4]

    Scaling rectified flow transformers for high-resolution image synthesis , year =

    Esser, Patrick and Kulal, Sumith and Blattmann, Andreas and Entezari, Rahim and M\". Scaling rectified flow transformers for high-resolution image synthesis , year =. Proceedings of the 41st International Conference on Machine Learning , articleno =

  5. [5]

    2024 , url =

    Black-Forest-Labs , title =. 2024 , url =

    2024

  6. [6]

    arXiv preprint arXiv:2104.08718 , year =

    Clipscore: A reference-free evaluation metric for image captioning , author =. arXiv preprint arXiv:2104.08718 , year =

    Pith/arXiv arXiv

  7. [7]

    arXiv preprint arXiv:1412.6980 , keywords =

    Kingma, Diederik P and Ba, Jimmy , biburl =. arXiv preprint arXiv:1412.6980 , keywords =

    Pith/arXiv arXiv

  8. [8]

    2024 , eprint =

    Playground v2.5: Three Insights towards Enhancing Aesthetic Quality in Text-to-Image Generation , author =. 2024 , eprint =

    2024

  9. [9]

    2025 , url =

    Muyang Li and Yujun Lin and Zhekai Zhang and Tianle Cai and Junxian Guo and Xiuyu Li and Enze Xie and Chenlin Meng and Jun-Yan Zhu and Song Han , booktitle =. 2025 , url =

    2025

  10. [10]

    Open Compute Project • OCP Microscaling Formats (MX) Specification , year =

    Microsoft and AMD and Arm and Intel and Meta and NVIDIA and Qualcomm , url =. Open Compute Project • OCP Microscaling Formats (MX) Specification , year =

  11. [11]

    Proceedings of the IEEE/CVF international conference on computer vision , pages =

    Scalable diffusion models with transformers , author =. Proceedings of the IEEE/CVF international conference on computer vision , pages =

  12. [12]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Rombach, Robin and Blattmann, Andreas and Lorenz, Dominik and Esser, Patrick and Ommer, Bj\"orn , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2022 , pages =

    2022

  13. [13]

    U-Net: Convolutional Networks for Biomedical Image Segmentation , booktitle =

    Ronneberger, Olaf and Fischer, Philipp and Brox, Thomas , editor =. U-Net: Convolutional Networks for Biomedical Image Segmentation , booktitle =. 2015 , publisher =

    2015

  14. [14]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages =

    A picture is worth more than 77 text tokens: Evaluating clip-style models on dense captions , author =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages =

  15. [15]

    and Montoya, Evan and Munechika, David and Yang, Haoyang and Hoover, Benjamin and Chau, Duen Horng , year =

    Wang, Zijie J. and Montoya, Evan and Munechika, David and Yang, Haoyang and Hoover, Benjamin and Chau, Duen Horng , year =. arXiv:2210.14896 [cs] , url =

    Pith/arXiv arXiv

  16. [16]

    Xiao, Guangxuan and Lin, Ji and Seznec, Mickael and Wu, Hao and Demouth, Julien and Han, Song , booktitle =

  17. [17]

    Imagereward: Learning and evaluating human preferences for text-to-image generation , author =

  18. [18]

    Advances in Neural Information Processing Systems , editor =

    ZeroQuant: Efficient and Affordable Post-Training Quantization for Large-Scale Transformers , author =. Advances in Neural Information Processing Systems , editor =. 2022 , url =

    2022

  19. [19]

    CVPR , year =

    The Unreasonable Effectiveness of Deep Features as a Perceptual Metric , author =. CVPR , year =

  20. [20]

    SageAttention2: Efficient Attention with Thorough Outlier Smoothing and Per-thread

    Jintao Zhang and Haofeng Huang and Pengle Zhang and Jia wei and Jun Zhu and Jianfei Chen , booktitle =. SageAttention2: Efficient Attention with Thorough Outlier Smoothing and Per-thread

  21. [21]

    The Thirteenth International Conference on Learning Representations , year =

    ViDiT-Q: Efficient and Accurate Quantization of Diffusion Transformers for Image and Video Generation , author =. The Thirteenth International Conference on Learning Representations , year =

  22. [22]

    Attention is All you Need , url =

    Vaswani, Ashish and Shazeer, Noam and Parmar, Niki and Uszkoreit, Jakob and Jones, Llion and Gomez, Aidan N and Kaiser, ukasz and Polosukhin, Illia , booktitle =. Attention is All you Need , url =

  23. [23]

    Proceedings of the 32nd International Conference on Machine Learning , pages =

    Deep Unsupervised Learning using Nonequilibrium Thermodynamics , author =. Proceedings of the 32nd International Conference on Machine Learning , pages =. 2015 , editor =

    2015

  24. [24]

    Proceedings of the 34th International Conference on Neural Information Processing Systems , articleno =

    Ho, Jonathan and Jain, Ajay and Abbeel, Pieter , title =. Proceedings of the 34th International Conference on Neural Information Processing Systems , articleno =. 2020 , isbn =

    2020

  25. [25]

    2024 , isbn =

    Chen, Junsong and Ge, Chongjian and Xie, Enze and Wu, Yue and Yao, Lewei and Ren, Xiaozhe and Wang, Zhongdao and Luo, Ping and Lu, Huchuan and Li, Zhenguo , title =. 2024 , isbn =. doi:10.1007/978-3-031-73411-3_5 , booktitle =

    doi:10.1007/978-3-031-73411-3_5 2024

  26. [26]

    International Conference on Learning Representations , year=

    Fast Sampling of Diffusion Models with Exponential Integrator , author=. International Conference on Learning Representations , year=

  27. [27]

    Proceedings of the 36th International Conference on Neural Information Processing Systems , articleno =

    Lu, Cheng and Zhou, Yuhao and Bao, Fan and Chen, Jianfei and Li, Chongxuan and Zhu, Jun , title =. Proceedings of the 36th International Conference on Neural Information Processing Systems , articleno =. 2022 , isbn =

    2022

  28. [28]

    arXiv preprint arXiv:2202.00512 , year=

    Progressive distillation for fast sampling of diffusion models , author=. arXiv preprint arXiv:2202.00512 , year=

    Pith/arXiv arXiv

  29. [29]

    Proceedings of the 40th International Conference on Machine Learning , articleno =

    Song, Yang and Dhariwal, Prafulla and Chen, Mark and Sutskever, Ilya , title =. Proceedings of the 40th International Conference on Machine Learning , articleno =. 2023 , publisher =

    2023

  30. [30]

    2023 , month =

    Li, Yuanzhi and Bubeck, Sébastien and Eldan, Ronen and Giorno, Allie Del and Gunasekar, Suriya and Lee, Yin Tat , title =. 2023 , month =

    2023

  31. [31]

    Lee and Deming Chen and Tri Dao , booktitle=

    Tianle Cai and Yuhong Li and Zhengyang Geng and Hongwu Peng and Jason D. Lee and Deming Chen and Tri Dao , booktitle=. Medusa: Simple. 2024 , url=

    2024

  32. [32]

    Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month=

    Li, Xiuyu and Liu, Yijiang and Lian, Long and Yang, Huanrui and Dong, Zhen and Kang, Daniel and Zhang, Shanghang and Keutzer, Kurt , title=. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month=. 2023 , pages=

    2023

  33. [33]

    CVPR , year=

    Post-training Quantization on Diffusion Models , author=. CVPR , year=

  34. [34]

    2023 , eprint=

    Efficient Quantization Strategies for Latent Diffusion Models , author=. 2023 , eprint=

    2023

  35. [35]

    2023 , url=

    Yefei He and Luping Liu and Jing Liu and Weijia Wu and Hong Zhou and Bohan Zhuang , booktitle=. 2023 , url=

    2023

  36. [36]

    2024 , eprint=

    TCAQ-DM: Timestep-Channel Adaptive Quantization for Diffusion Models , author=. 2024 , eprint=

    2024

  37. [37]

    2024 , isbn =

    Tang, Siao and Wang, Xin and Chen, Hong and Guan, Chaoyu and Wu, Zewen and Tang, Yansong and Zhu, Wenwu , title =. 2024 , isbn =. doi:10.1007/978-3-031-72992-8_23 , booktitle =

    doi:10.1007/978-3-031-72992-8_23 2024

  38. [38]

    2024 , eprint=

    MixDQ: Memory-Efficient Few-Step Text-to-Image Diffusion Models with Metric-Decoupled Mixed Precision Quantization , author=. 2024 , eprint=

    2024

  39. [39]

    2022 , eprint=

    Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow , author=. 2022 , eprint=

    2022

  40. [40]

    2023 , eprint=

    Flow Matching for Generative Modeling , author=. 2023 , eprint=

    2023

  41. [41]

    2024 , eprint=

    InstaFlow: One Step is Enough for High-Quality Diffusion-Based Text-to-Image Generation , author=. 2024 , eprint=

    2024

  42. [42]

    2023 , eprint=

    All are Worth Words: A ViT Backbone for Diffusion Models , author=. 2023 , eprint=

    2023

  43. [43]

    2021 , eprint=

    A White Paper on Neural Network Quantization , author=. 2021 , eprint=

    2021

  44. [44]

    SpinQuant:

    Zechun Liu and Changsheng Zhao and Igor Fedorov and Bilge Soran and Dhruv Choudhary and Raghuraman Krishnamoorthi and Vikas Chandra and Yuandong Tian and Tijmen Blankevoort , booktitle=. SpinQuant:. 2025 , url=

    2025

  45. [45]

    NeurIPS , year=

    PTQ4DiT: Post-training Quantization for Diffusion Transformers , author=. NeurIPS , year=

  46. [46]

    Bulletin des Sciences Mathématiques , year = 1893, volume = 17, pages =

    Jacques Hadamard , title =. Bulletin des Sciences Mathématiques , year = 1893, volume = 17, pages =

  47. [47]

    2024 , eprint=

    Q-DiT: Accurate Post-Training Quantization for Diffusion Transformers , author=. 2024 , eprint=

    2024

  48. [48]

    Proceedings of the 41st International Conference on Machine Learning , articleno =

    Tseng, Albert and Chee, Jerry and Sun, Qingyao and Kuleshov, Volodymyr and De Sa, Christopher , title =. Proceedings of the 41st International Conference on Machine Learning , articleno =. 2024 , publisher =

    2024

  49. [49]

    and Li, Bo and Cameron, Pashmina and Jaggi, Martin and Alistarh, Dan and Hoefler, Torsten and Hensman, James , title =

    Ashkboos, Saleh and Mohtashami, Amirkeivan and Croci, Maximilian L. and Li, Bo and Cameron, Pashmina and Jaggi, Martin and Alistarh, Dan and Hoefler, Torsten and Hensman, James , title =. Proceedings of the 38th International Conference on Neural Information Processing Systems , articleno =. 2025 , isbn =

    2025

  50. [50]

    2025 , eprint=

    SpinQuant: LLM quantization with learned rotations , author=. 2025 , eprint=

    2025

  51. [51]

    2025 , howpublished =

    Dellinger, Eric and Khodamoradi, Alireza , title =. 2025 , howpublished =

    2025

  52. [52]

    2025 , month =

    AMD Instinct CDNA 4 Instruction Set Architecture , institution =. 2025 , month =

    2025

  53. [53]

    2018 , eprint =

    GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium , author =. 2018 , eprint =

    2018

  54. [54]

    2021 , eprint =

    Demystifying MMD GANs , author =. 2021 , eprint =

    2021

  55. [55]

    Proceedings of the 37th International Conference on Neural Information Processing Systems , pages =

    ImageReward: learning and evaluating human preferences for text-to-image generation , author =. Proceedings of the 37th International Conference on Neural Information Processing Systems , pages =

  56. [56]

    2022 , eprint=

    LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale , author=. 2022 , eprint=

    2022

  57. [57]

    2023 , eprint=

    Outlier Suppression: Pushing the Limit of Low-bit Transformer Language Models , author=. 2023 , eprint=

    2023

  58. [58]

    arXiv preprint arXiv:2410.09426 , year=

    FlatQuant: Flatness Matters for LLM Quantization , author=. arXiv preprint arXiv:2410.09426 , year=

    Pith/arXiv arXiv

  59. [59]

    2026 , eprint=

    LoRaQ: Optimized Low Rank Approximation for 4-bit Quantization , author=. 2026 , eprint=

    2026

  60. [60]

    2015 , eprint=

    Microsoft COCO Captions: Data Collection and Evaluation Server , author=. 2015 , eprint=

    2015