通过联合优化推理步数与每层精度，实现高效扩散模型部署。
DiffPro: Joint Timestep and Layer-Wise Precision Optimization for Efficient Diffusion Inference
- 后训练框架，不需重训，直接优化推理步数与精度
- 最多压缩6.25倍模型，减少50%推理步数，提速2.8倍
- 适合追求实时、低功耗扩散模型部署的开发者
扩散模型生成质量高，但推理成本高昂，源于大量去噪步骤和复杂的矩阵运算。本文提出 DiffPro，一种后训练、硬件精准的框架，可直接适配实际部署中的整数核，联合优化扩散变换器（DiTs）的推理步数与逐层精度，无需任何训练即可降低延迟和内存占用。DiffPro 包含三部分：基于流形感知的敏感度指标用于分配权重位数，动态激活量化以稳定各步激活值，以及由教师-学生漂移引导的预算式步数选择器。实验表明，DiffPro 在标准基准上实现最高6.25倍模型压缩，减少50%推理步数，推理速度提升2.8倍，且 Delta FID <= 10，验证了其实际效率优势。该方法将步数缩减与精度规划统一为一个可部署的预算方案，支持实时、节能的扩散模型推理。
原文摘要 · Abstract (English)
Diffusion models produce high quality images but inference is costly due to many denoising steps and heavy matrix operations. We present DiffPro, a post-training, hardware-faithful framework that works with the exact integer kernels used in deployment and jointly tunes timesteps and per-layer precision in Diffusion Transformers (DiTs) to reduce latency and memory without any training. DiffPro combines three parts: a manifold-aware sensitivity metric to allocate weight bits, dynamic activation quantization to stabilize activations across timesteps, and a budgeted timestep selector guided by teacher-student drift. In experiments DiffPro achieves up to 6.25x model compression, fifty percent fewer timesteps, and 2.8x faster inference with Delta FID <= 10 on standard benchmarks, demonstrating practical efficiency gains. DiffPro unifies step reduction and precision planning into a single budgeted deployable plan for real-time energy-aware diffusion inference.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。