Undergrad student, Fudan University
1 paper at NeurIPS 2025
We propose an information-preserving and data-free pipeline for ultra-efficient delta compression, achieving ultra-high compression ratios and strong performance across various models.
