Efficient Inference | Chunzheng Zhuhttps://zhcz328.github.io/tag/efficient-inference/Efficient InferenceWowchemy (https://wowchemy.com)en-usTue, 29 Sep 2026 00:00:00 +0000https://zhcz328.github.io/media/icon_huc9fe9d140a9bb016fbeeaa3ba1111928_724048_512x512_fill_lanczos_center_3.pngEfficient Inferencehttps://zhcz328.github.io/tag/efficient-inference/PACE: Pareto-Adaptive Compression for Efficient Native MLLMshttps://zhcz328.github.io/publication/pace/Tue, 29 Sep 2026 00:00:00 +0000https://zhcz328.github.io/publication/pace/<p>PACE performs direction-aware visual token compression for native MLLMs while preserving performance and reducing computation.</p> <p>Lei Tian, Kai WU, and Haihua Yang are affiliated with ByteDance. YiFan Zhang is affiliated with the Institute of Automation, Chinese Academy of Sciences. This work was conducted during Chunzheng Zhu&rsquo;s internship at ByteDance.</p>