Outlier-Aware Post-Training Quantization for Discrete Graph Diffusion Models

Discrete Graph Diffusion Models (DGDMs) mark a pivotal advancement in graph generation, effec-tively preserving sparsity and structural integrity, thereby enhancing the learning of graph data dis-tributions for diverse generative applications. De-spite their potential, DGDMs are computationally intensive due to the numerous low-parameter yet high-computation operations, thereby increasing the need of inference acceleration. A promis-ing solution to mitigate this issue is model quan-tization. However, existing quantization tech-niques for Image Diffusion Models (IDMs) face limitations in DGDMs due to differing diffusion processes, while Large Language Model (LLM) quantization focuses on reducing memory ac-cess latency of loading large parameters, unlike DGDMs, where inference bottlenecks are compu-tations due to smaller model sizes. To fill this gap, we introduce Bit-DGDM, a post-training quantization framework for DGDMs which in-corporates two novel ideas: (i) sparse-dense ac-tivation quantization sparsely modeling the acti-vation outliers through adaptively selected, data-free thresholds in full-precision and quantizing the remaining to low-bit, and (ii) ill-conditioned low-rank decomposition decomposing the weights into low-rank component enable faster inference and an -sparsity matrix that models outliers. Ex-tensive experiments demonstrate that Bit-DGDM not only reducing the memory usage from the FP32 baseline by up to 2.8x and achieve up to 2.5x speedup, but also achieve comparable per-formance to ultra-low precision of up to 4-bit.