ChebBooster: A Training-Free Approach for Efficient Diffusion Transformer Inference via Chebyshev-Inspired Extrapolation
Novel mathematical approach for optimizing DiT inference efficiency without retraining, relevant for ML researchers working on diffusion models.
AI Summary
ChebBooster introduces a -free Chebyshev polynomial extrapolation method that accelerates Diffusion by up to 3.68x with improved stability.
Excerpt
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