EfficientDM: Efficient Quantization-Aware Fine-Tuning of Low-Bit Diffusion Models

arXiv:2310.03270v4 Announce Type: replace
Abstract: Diffusion models have demonstrated remarkable capabilities in image synthesis and related generative tasks. Nevertheless, their practicality for real-world applications is constrained by substantial computational costs and latency issues. Quantization is a dominant way to compress and accelerate diffusion models, where post-training quantization (PTQ) and quantization-aware training (QAT) are two main approaches, each bearing its own properties. While PTQ exhibits efficiency in terms of both time and data usage, it may lead to diminished performance in …

arxiv cs.cv diffusion diffusion models fine-tuning low quantization type