{"ID":2838973,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2511.16156","arxiv_id":"2511.16156","title":"Pluggable Pruning with Contiguous Layer Distillation for Diffusion Transformers","abstract":"Diffusion Transformers (DiTs) have shown exceptional performance in image generation, yet their large parameter counts incur high computational costs, impeding deployment in resource-constrained settings. To address this, we propose Pluggable Pruning with Contiguous Layer Distillation (PPCL), a flexible structured pruning framework specifically designed for DiT architectures. First, we identify redundant layer intervals through a linear probing mechanism combined with the first-order differential trend analysis of similarity metrics. Subsequently, we propose a plug-and-play teacher-student alternating distillation scheme tailored to integrate depth-wise and width-wise pruning within a single training phase. This distillation framework enables flexible knowledge transfer across diverse pruning ratios, eliminating the need for per-configuration retraining. Extensive experiments on multiple Multi-Modal Diffusion Transformer architecture models demonstrate that PPCL achieves a 50\\% reduction in parameter count compared to the full model, with less than 3\\% degradation in key objective metrics. Notably, our method maintains high-quality image generation capabilities while achieving higher compression ratios, rendering it well-suited for resource-constrained environments. The open-source code, checkpoints for PPCL can be found at the following link: https://github.com/OPPO-Mente-Lab/Qwen-Image-Pruning.","short_abstract":"Diffusion Transformers (DiTs) have shown exceptional performance in image generation, yet their large parameter counts incur high computational costs, impeding deployment in resource-constrained settings. To address this, we propose Pluggable Pruning with Contiguous Layer Distillation (PPCL), a flexible structured prun...","url_abs":"https://arxiv.org/abs/2511.16156","url_pdf":"https://arxiv.org/pdf/2511.16156v2","authors":"[\"Jian Ma\",\"Qirong Peng\",\"Xujie Zhu\",\"Peixing Xie\",\"Chen Chen\",\"Haonan Lu\"]","published":"2025-11-20T08:53:07Z","proceeding":"cs.CV","tasks":"[\"cs.CV\"]","methods":"[\"Diffusion Model\",\"Transformer\"]","has_code":false,"code_links":[{"ID":606827,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_id":2838973,"paper_url":"https://arxiv.org/abs/2511.16156","paper_title":"Pluggable Pruning with Contiguous Layer Distillation for Diffusion Transformers","repo_url":"https://github.com/OPPO-Mente-Lab/Qwen-Image-Pruning","is_official":false,"mentioned_in_paper":false,"mentioned_in_github":true,"github_stars":0}]}