CentriQ: Calibration-Free Quantization of Diffusion Transformers via Exact Mean Centering

Nataša Jovanović; Mathieu Salzmann; Saqib Javed

cs.CVOct 5, 2026

CentriQ: Calibration-Free Quantization of Diffusion Transformers via Exact Mean Centering

Authors: Nataša Jovanović, Mathieu Salzmann, Saqib Javed

Organizations: Tenstorrent · EPFL

Abstract

Diffusion transformers (DiTs) achieve state-of-the-art image generation, but their sampling cost limits deployment. Quantizing both weights and activations to 4 bits reduces this cost, yet existing methods fall short in one of two ways. Calibration-based methods are tied to a specific checkpoint and prompt distribution, whereas data-free Hadamard rotation, effective for LLMs, loses quality on DiTs. We show that this loss has a structural cause. Adaptive layer-norm conditioning adds a per-token mean to the activations, and at the widths of the evaluated DiTs, the Hadamard rotations used by data-free methods cannot spread this mean uniformly across coordinates. A single dominant direction therefore survives the rotation and sets the quantization range. We introduce CentriQ, a calibration-free quantizer that centers each token before rotation and restores the mean exactly through a rank-1 full-precision branch, so that per-token scales follow in closed form without data. Weights are fitted under a robust $\ell_p$ objective that tracks the dense mode of each group and discounts heavy tails. Across three DiTs, CentriQ matches the quality of calibrated SVDQuant at 4 bits, whereas calibration-free weight quantizers with plain per-token activation quantization collapse or degrade substantially. CentriQ outperforms the strongest calibration-free method reported to date at 2-bit weights. It is also the first calibration-free method to retain usable image quality at 2-bit activations.

Figures & tables

Figure 1: DC energy and per-token range across blocks. Top: DC energy fraction

f_{\mathrm{DC}}

of the modulated input and of the LayerNorm output. Bottom: per-token

\max/\mathrm{RMS}

under identity, rotation only, centering only, and centering + rotation. Median over tokens at the highest-

\sigma

step, band = min/max over layer types.

Figure 2: Effect of centering on one token. Sorted coordinate magnitudes after rotation, with and without centering; dashed levels mark the resulting INT4 step. Centering removes the DC coordinate, which otherwise sets the per-token range.

Figure 3: One quantized linear layer in CentriQ. Online per-token path (top) and offline weight path (bottom). The per-token mean

\bar{x}_{t}

and the row-sum vector

\mathbf{w}_{\mathbf{1}}

form a full-precision rank-1 branch that bypasses both quantizers.

PixArt- $\Sigma$
		MJHQ				sDCI
Method	Calib.	FID $\downarrow$	IR $\uparrow$	PSNR $\uparrow$	LPIPS $\downarrow$	FID $\downarrow$	IR $\uparrow$	PSNR $\uparrow$	LPIPS $\downarrow$
FP16	$\times$	16.7	0.952	–	–	24.9	0.973	–	–
RTN	$\times$	481.5	-2.273	6.14	0.995	470.1	-2.277	6.60	0.957
NF4	$\times$	479.6	-2.273	6.16	0.997	468.6	-2.277	6.46	0.966
HQQ	$\times$	462.6	-2.273	6.37	0.991	456.0	-2.277	6.58	0.963

Table 1: Quantitative comparison of W4A4 quantization. Best quantized result in bold , second best underlined . SVDQuant † rows are taken from Li et al. (2025) .

Figure 4: GenEval prompts on FLUX.1-schnell at W4A4, W4A2, and W2A4. Images in a row share the prompt and seed, and the marks give the verdict of the GenEval detector.

	W4A4		W4A2		W2A4
Configuration	FID $\downarrow$	IR $\uparrow$	FID $\downarrow$	IR $\uparrow$	FID $\downarrow$	IR $\uparrow$
No transform	40.62	-0.067	437.41	-2.270	363.14	-2.244
$+$ rotation (QuaRot-style)	25.65	0.768	455.30	-2.271	35.84	0.371
$+$ centering (exact rank-1 re-injection)	18.46	0.961	444.21	-2.274	21.81	0.744
$+$ percentile clip ( CentriQ )	18.80	0.939	35.59	0.394	21.56	0.728
BF16	19.45	0.966	19.45	0.966	19.45	0.966

Table 2: Component ablation across bit-width on FLUX.1-schnell (MJHQ).

Appendix figures & tables15 assets

Supplementary material from the paper’s appendix.

Appendix

Setting	Value
Weight grid, group	$\{0,\dots,15\}$ , $g=64$
Scale and zero point	FP16 (PixArt- $\Sigma$ ), BF16 (FLUX, SANA)
Objective exponent $p$	0.7
Penalty $\beta$	10, fixed
Outer rounds / IRLS / proximal iterations	$\leq 3$ / $\leq 5$ / 20
Stopping tolerance	$10^{-3}$ , relative

Appendix

Table 3: Quantizer and solver settings.

	MJHQ				sDCI
Precision	FID $\downarrow$	IR $\uparrow$	PSNR $\uparrow$	LPIPS $\downarrow$	FID $\downarrow$	IR $\uparrow$	PSNR $\uparrow$	LPIPS $\downarrow$
BF16	19.45	0.966	–	–	20.92	0.984	–	–
W4A4	18.80	0.939	18.40	0.173	20.6	0.970	17.29	0.173
W4A2	35.59	0.394	13.62	0.457	28.61	0.683	12.85	0.429
W2A4	21.56	0.728	14.35	0.367	22.21	0.873	13.48	0.350

Appendix

Table 4: Sub-4-bit quantization of FLUX.1-schnell with CentriQ.

W4A4
Method	Calib.	Single	Two	Counting	Colors	Position	Color attr.	Overall $\uparrow$
BF16	$\times$	0.997	0.879	0.572	0.742	0.273	0.540	0.667
Q-DiT †	✓	0.741	0.424	0.378	0.418	0.073	0.208	0.373
SmoothQuant †	✓	0.619	0.293	0.272	0.317	0.043	0.143	0.281
QuaRot †	$\times$	0.819	0.543	0.472	0.519	0.118	0.275	0.458
ViDiT-Q †	✓	0.888	0.586	0.516	0.585	0.130	0.268	0.495

Appendix

Table 5: GenEval on FLUX.1-schnell. Scores on six compositional sub-tasks and Overall. Rows marked † are taken from the concurrent preprint of Lee et al. (2026) , which we include for completeness; its FP16 reference matches ours in Overall score. Bold and underlined entries mark the best and second-best quantized result within each bit-width.

	PixArt- $\Sigma$		FLUX.1-schnell		SANA-1.6B
Weight quantizer	FID $\downarrow$	IR $\uparrow$	FID $\downarrow$	IR $\uparrow$	FID $\downarrow$	IR $\uparrow$
FP16	16.66	0.952	19.45	0.966	15.96	1.097
RTN	32.56	0.721	18.97	0.936	18.06	0.993
NF4	24.40	0.834	19.19	0.926	17.97	0.942
HQQ	21.04	0.866	18.92	0.940	16.30	1.073
RoDS	19.13	0.891	18.80	0.939	16.19	1.072

Appendix

Table 6: Activation path against weight quantizer at W4A4 (MJHQ).

W4A16
	PixArt- $\Sigma$		FLUX.1-schnell		SANA-1.6B
Method	FID $\downarrow$	IR $\uparrow$	FID $\downarrow$	IR $\uparrow$	FID $\downarrow$	IR $\uparrow$
FP16	16.66	0.952	19.45	0.966	15.96	1.097
RTN	23.04	0.823	19.02	0.968	16.32	1.107
NF4	18.74	0.889	19.19	0.953	16.07	1.103
HQQ	16.40	0.925	18.87	0.958	16.15	1.096

Appendix

Table 7: Weight-only quantization at W4A16 and W3A16 (MJHQ).

Figure 5: CentriQ on FLUX.1-schnell with 4-bit and 2-bit weights and activations. The large image is the 16-bit output. In each grid, rows set the weight precision and columns set the activation precision, and every image shares the prompt and seed of the 16-bit output.

Figure 6: Further CentriQ samples on FLUX.1-schnell with 4-bit and 2-bit weights and activations. The layout follows Fig. 5 .

Figure 7: W4A4 with calibration-free quantizers on PixArt- $\Sigma$ , FLUX.1-schnell, and SANA-1.6B. RTN and HQQ produce noise on PixArt- $\Sigma$ and lose the content of the prompt on FLUX.1-schnell and SANA-1.6B, whereas CentriQ stays close to the 16-bit output.

Figure 8: Weight-only W3A16 on FLUX.1-schnell and SANA-1.6B. RoDS, our $\ell_{p}$ weight quantizer, compared with HQQ and RTN at 3-bit weights and 16-bit activations.

Figure 9: Weight-only W3A16 on PixArt- $\Sigma$ . RoDS, our $\ell_{p}$ weight quantizer, compared with HQQ, SVDQuant, and RTN at 3-bit weights and 16-bit activations.

Figure 10: Weight-only W4A16 on FLUX.1-schnell and SANA-1.6B. RoDS, our $\ell_{p}$ weight quantizer, compared with HQQ and NF4 at 4-bit weights and 16-bit activations.

Figure 11: Weight-only W4A16 on PixArt- $\Sigma$ . RoDS, our $\ell_{p}$ weight quantizer, compared with SVDQuant and NF4 at 4-bit weights and 16-bit activations.

Figure 12: CentriQ and SVDQuant at W4A4 on FLUX.1-schnell. Each row shares the prompt and seed of the BF16 output.

Figure 13: CentriQ and SVDQuant at W4A4 on PixArt- $\Sigma$ . Each row shares the prompt and seed of the FP16 output.

Figure 14: CentriQ and SVDQuant at W4A4 on SANA-1.6B. Each row shares the prompt and seed of the BF16 output.

Explore similar work

OrbitQuant: Data-Agnostic Quantization for Image and Video Diffusion Transformers
Jul 2, 2026Donghyun Lee, Jitesh Chavan, Duy Nguyen +5Diffusion Transformers Quantizer
Cantina Labs · University of Southern California · *Work done during an internship at Cantina Labs. +1
KroQuant: Kronecker-Structured Block Transforms for Efficient Post-Training Quantization of Diffusion Transformers
Jul 23, 2026Yann Bouquet, Alireza Khodamoradi, Kristof Denolf +1Post-Training Quantization Diffusion Transformers
EPFL, Lausanne, Switzerland. · Advanced Micro Devices, Inc., Longmont, CO, USA.
DiRotQ: Rotation-Aware Quantization for 4-bit Diffusion Transformers
May 16, 2026Sayeh Sharify, Mahsa Salmani, Hesham MostafaTransformer Architectures Singular Value Decomposition
d-Matrix Santa Clara, CA 95054

Jul 2, 2026, cs.CVJ/K move · Enter open · S save

OrbitQuant: Data-Agnostic Quantization for Image and Video Diffusion Transformers

Donghyun Lee, Jitesh Chavan, Duy Nguyen, Sam Huang+4

Cantina Labs · University of Southern California · *Work done during an internship at Cantina Labs.+1

Diffusion Transformers Quantizer Video Generation

Open paper AskPDF arXiv

Diffusion transformers (DiTs) achieve state-of-the-art image and video generation, but their multi-step sampling and growing parameter count make inference expensive. Post-training quantization (PTQ) is the natural remedy, yet DiT activations shift across timesteps, prompts, and guidance branches, forcing prior methods to re-fit calibration data for every new checkpoint or modality. We present OrbitQuant, a data-agnostic weight-activation quantizer that bypasses range estimation by quantizing in a normalized, rotated basis. In this basis, a randomized permuted block-Hadamard (RPBH) rotation concentrates each coordinate around one fixed, known marginal regardless of the input, so a single Lloyd-Max codebook serves all timesteps, prompts, and layers of a given input dimension. We extend the same quantizer to weight rows offline, absorbing the rotation into the weights so that it cancels inside each linear layer and only a forward rotation on the activations remains at runtime. The same recipe transfers from image to video with no per-modality tuning. Across FLUX.1, Z-Image-Turbo, Wan 2.1, and CogVideoX, it sets the state of the art for PTQ at several low-bit settings. It also pushes PTQ of image diffusion transformers to W2A4 with usable generation quality.

Abstract

Explore similar work

OrbitQuant: Data-Agnostic Quantization for Image and Video Diffusion Transformers

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

DiRotQ: Rotation-Aware Quantization for 4-bit Diffusion Transformers