ChebBooster: A Training-Free Approach for Efficient Diffusion Transformer Inference via Chebyshev-Inspired Extrapolation

Chengjie Lu; Tianchi Deng; Zhengqi He; Chengwen Luo; Xueliang Li

cs.LGAug 24, 2026

ChebBooster: A Training-Free Approach for Efficient Diffusion Transformer Inference via Chebyshev-Inspired Extrapolation

Authors: Chengjie Lu, Tianchi Deng, Zhengqi He, Chengwen Luo, Xueliang Li

Organizations: College of Electronics and Information Engineering, Shenzhen University · School of Artificial Intelligence, Shenzhen University

Abstract

Diffusion Transformers (DiTs) have shown strong performance in high-fidelity image generation, but their sampling process remains computationally intensive due to full model execution at every timestep. While cache-based acceleration has been explored to mitigate inference cost, naive reuse schemes suffer from low accuracy over long intervals, and Taylor-series-based extrapolation methods often face instability caused by Runge oscillations. In this paper, we propose ChebBooster, a training-free extrapolation framework based on Chebyshev polynomial theory that achieves stable and efficient acceleration for DiTs. Specifically, we adopt the Barycentric formulation to evaluate Chebyshev approximants with high numerical stability and minimal overhead, and further decouple the extrapolation into an offline weight precomputation phase and a lightweight online application stage. Extensive experiments across three representative DiT-based models, including DiT-XL/2, PixArt- $Σ$ , and FLUX.1-dev, demonstrate that ChebBooster achieves consistent improvements in visual quality and inference efficiency, reaching up to $3.68\times$ latency speedup and $5.12\times$ FLOPs reduction, outperforming existing training-free baselines under diverse generation tasks and resolutions.

Figures & tables

Figure 1: 3D Feature visualization of DiT forward features across timesteps. The naïve caching method (i.e., FORA [ 38 ] ) fails to align with the original feature trajectory, resulting in degraded generation quality.

Figure 2: 1D feature comparison between TaylorSeer [ 24 ] and ChebBooster extrapolation. (a) illustrates the feature trajectory of DiT, which evolves smoothly; both methods exhibit minimal differences in their fitted curves. (b) presents the extrapolation errors, where TaylorSeer displays larger oscillations in the mid-trajectory, yielding deviation in the predicted outcome.

Figure 3: Overview of the ChebBooster framework. The process consists of two stages: (1) Weight Precomputation (offline), where Chebyshev-inspired extrapolation weights are precomputed for different timesteps and stored in a structured weight table guided by the cache schedule; and (2) Forward Application (online), where full model forwarding is performed at sparse reference timesteps, while intermediate features are efficiently extrapolated using cached activations and precomputed weights. This decoupled design enables acceleration by reducing redundant computations while maintaining generative fidelity.

Figure 4: Approximations and Errors with diverse methods on the Runge function. Taylor expansion shows oscillation artifacts quickly in the near expolation. Lagrange interpolation starts solving the edge oscillation problem. Chebyshev methods give the best performance in both interpolation and extrapolation, while the one given equispaced nodes (Chebyshev-Uniform) is superior than that with the Chebyshev nodes.

Method	r	Latency(s) $\downarrow$	Speed $\uparrow$	FLOPs(T) $\downarrow$	Speed $\uparrow$	FID $\downarrow$	sFID $\downarrow$	Inception Score $\uparrow$
DDIM-50 steps	✘	0.506	1.000 $\times$	23.735	1.000 $\times$	2.18	4.29	251.56
DDIM-25 steps	✘	0.272	1.855 $\times$	11.868	2.000 $\times$	2.75	4.55	244.03
TaylorSeer [ 24 ]	3	0.343	1.473 $\times$	8.570	2.770 $\times$	2.34	4.76	247.00
ChebBooster ( $H=3$ )	3	0.302	1.674 $\times$	8.564	2.771 $\times$	2.25	4.59	246.49
ChebBooster ( $H=5$ )	3	0.342	1.476 $\times$	8.567	2.770 $\times$	2.28	4.64	247.27
DDIM-16 steps	✘	0.189	2.681 $\times$	7.595	3.125 $\times$	4.24	5.69	223.38

Table 1: Quantitative comparison on DiT-XL/2 generation with resolution of 256

\times

256.

Method	Refresh	Acceleration				Image Reward $\uparrow$	CLIP $\uparrow$
FLUX.1 [ 19 ]	Ratio ( r )	Latency(s) $\downarrow$	Speed $\uparrow$	FLOPs(T) $\downarrow$	Speed $\uparrow$	DrawBench	Score
[Dev]: 50 steps	✘	26.039	1.000 $\times$	3719.500	1.000 $\times$	0.9613	31.63
$\Delta$ -DiT [ 5 ]	2	18.085	1.440 $\times$	2480.000	1.500 $\times$	0.9360	31.58
[Dev]: 30 steps	✘	16.123	1.615 $\times$	2231.700	1.667 $\times$	0.9715	31.57
$\Delta$ -DiT [ 5 ]	3	13.381	1.946 $\times$	1686.763	2.205 $\times$	0.8969	31.53
[Dev]: 20 steps	✘	13.470	1.933 $\times$	1487.800	2.500 $\times$	0.9838	31.42

Table 2: Quantitative comparison on FLUX.1[Dev] generation with resolution of 1024

\times

1024.

Method	Refresh	Acceleration				Image $\uparrow$	CLIP $\uparrow$
PixArt- $\mathbf{\Sigma}$ [ 3 ]	Ratio ( r )	Latency(s) $\downarrow$	Speed $\uparrow$	FLOPs(T) $\downarrow$	Speed $\uparrow$	Reward	Score
DDIM-50 steps	✘	1.255	1.000 $\times$	95.362	1.000 $\times$	1.1318	33.0278
$\Delta$ -DiT [ 5 ]	2	0.469	2.675 $\times$	36.138	2.639 $\times$	1.1056	33.1066
DDIM-40 steps	✘	1.003	1.250 $\times$	76.290	1.250 $\times$	1.1364	33.0198
FORA [ 38 ]	3	0.973	1.289 $\times$	36.138	2.639 $\times$	1.0917	33.0869
ToCa [ 51 ]	3	1.265	0.992 $\times$	58.238	1.637 $\times$	1.0920	33.0869

Table 3: Quantitative comparison on PixArt-

\mathbf{\Sigma}

generation with resolution of 512

\times

512.

Figure 5: Qualitative Study of diverse methods on PixArt- $\Sigma$ .

Figure 6: Qualitative Study on FLUX.1-dev.

Appendix figures & tables6 assets

Supplementary material from the paper’s appendix.

Appendix

Method	IS $\uparrow$	FID $\downarrow$	sFID $\downarrow$	FLOPs $\downarrow$
Original ( $50$ steps)	251.56	2.18	4.29	23.735
Original ( $25$ steps)	244.03	2.75	4.55	11.868
Original ( $20$ steps)	235.35	3.27	4.93	9.494
Original ( $16$ steps)	223.38	4.24	5.69	7.595
Original ( $12$ steps)	195.83	7.07	8.05	5.696
Original ( $10$ steps)	168.99	11.19	11.12	4.747

Appendix

Table 4: Performance Comparison of ChebBooster on DiT-XL/2. This table summarizes the performance of ChebBooster across different configurations, comparing Inception Score (IS), FID, sFID, and FLOPs against the original model with varying sampling steps. The results demonstrate that ChebBooster achieves significant speedup while maintaining competitive quality metrics.

Figure 7: FID evaluation on DiT-XL/2 across varying sampling steps. This figure shows the trade-off between FID scores and inference speed when applying ChebBooster. Notably, ChebBooster achieves substantial acceleration while preserving competitive FID performance, indicating that the generated images remain close to the real data distribution even under aggressive speedup.

Figure 8: Structural FID (sFID) evaluation on DiT-XL/2. The figure highlights how ChebBooster maintains structural consistency in generated images while achieving high speedup ratios. Compared to FID, sFID is more sensitive to localized distortions, and the consistently low sFID values indicate that structural fidelity is preserved even at faster sampling rates.

Figure 9: Inception Score evaluation on DiT-XL/2. This figure shows how ChebBooster impacts sample diversity and semantic clarity across different acceleration levels. Despite fewer sampling steps, the method sustains high Inception Scores, indicating that the generated samples remain both varied and classifiable.

Figure 10: Extensive qualitative results on PixArt- $\Sigma$ .

Figure 11: Extensive qualitative results on FLUX.1-dev.

Explore similar work

Memory-Efficient Training-Free Acceleration of Diffusion Transformers with BaryCache
Aug 24, 2026Chengjie Lu, Tianchi Deng, Zhengqi He +3Diffusion Model Inference Acceleration Diffusion-Based Image Generation
College of Electronics and Information Engineering, Shenzhen University · School of Artificial Intelligence, Shenzhen University · College of Computer Science and Software Engineering, Shenzhen University
AccelAes: Accelerating Diffusion Transformers for Training-Free Aesthetic-Enhanced Image Generation
Mar 13, 2026Xuanhua Yin, Chuanzhi Xu, Haoxian Zhou +2Diffusion Transformer Image Generation
School of Computer Science, The University of Sydney, NSW 2006, Australia
BRACE: Taming Sharp Irregularities via Barycentric Rational Forecasting for Fast Diffusion Transformers Inference
Aug 4, 2026Jinlong Yang, Jinke Wu, Lizilin +1Diffusion Transformer Diffusion Model Caching
Sichuan University Chengdu, China · School of Artificial Intelligence, Sichuan University Chengdu, China

Aug 24, 2026J/K move · Enter open · S save

Memory-Efficient Training-Free Acceleration of Diffusion Transformers with BaryCache

Chengjie Lu, Tianchi Deng, Zhengqi He, Zhijian Gao+2

College of Electronics and Information Engineering, Shenzhen University · School of Artificial Intelligence, Shenzhen University · College of Computer Science and Software Engineering, Shenzhen University

Diffusion Model Inference Acceleration Diffusion-Based Image Generation Diffusion Transformer Video Diffusion Models

Open paper AskPDF arXiv

Diffusion Transformers achieve high-fidelity image and video generation, but their iterative sampling remains expensive, for each denoising step requires large matrix operations. Existing cache-based acceleration reduces redundant computation yet increases the VRAM footprint by storing intermediate states, which can directly constrain inference batch size. In this work, we propose a training-free acceleration method that performs stepwise forecasting for DiT sampling using a Barycentric Extrapolator. By leveraging barycentric extrapolation, our predictor is numerically stable and alleviates oscillatory artifacts analogous to the Runge phenomenon during forward forecasting. Across extensive experiments on both image and video generation, our approach provides a favorable trade-off between memory usage and perceptual quality, while delivering up to 3.30x end-to-end sampling speedup compared with baseline DiT inference.

Abstract

Explore similar work

Memory-Efficient Training-Free Acceleration of Diffusion Transformers with BaryCache

AccelAes: Accelerating Diffusion Transformers for Training-Free Aesthetic-Enhanced Image Generation

BRACE: Taming Sharp Irregularities via Barycentric Rational Forecasting for Fast Diffusion Transformers Inference