ChebBooster: A Training-Free Approach for Efficient Diffusion Transformer Inference via Chebyshev-Inspired Extrapolation
2026-08-24 • Machine Learning
Machine LearningArtificial Intelligence
AI summaryⓘ
The authors studied Diffusion Transformers (DiTs), which create high-quality images but take a long time to generate them because they run a full model each step. To speed this up, they developed ChebBooster, a method that uses math called Chebyshev polynomials to predict future steps more accurately and efficiently without retraining the model. Their approach splits calculations into a heavier offline part and a lightweight online part, making it practical. Tests on various DiT models showed ChebBooster made image generation faster and used less computation while keeping or improving quality compared to other no-training methods.
Diffusion TransformersImage GenerationChebyshev PolynomialsExtrapolationInference AccelerationBarycentric FormulationRunge OscillationsComputational EfficiencyLatency SpeedupFLOPs Reduction
Authors
Chengjie Lu, Tianchi Deng, Zhengqi He, Chengwen Luo, Xueliang Li
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.