Breaking the Compression Ceiling: Data-Free Pipeline for Ultra-Efficient Delta Compression

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Wang, Xiaohui, Ye, Peng, Huang, Chenyu, Zheng, Shenghe, Zhang, Bo, Bai, Lei, Ouyang, Wanli, Chen, Tao
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915550444126208
author Wang, Xiaohui
Ye, Peng
Huang, Chenyu
Zheng, Shenghe
Zhang, Bo
Bai, Lei
Ouyang, Wanli
Chen, Tao
author_facet Wang, Xiaohui
Ye, Peng
Huang, Chenyu
Zheng, Shenghe
Zhang, Bo
Bai, Lei
Ouyang, Wanli
Chen, Tao
contents With the rise of the fine-tuned-pretrained paradigm, storing numerous fine-tuned models for multi-tasking creates significant storage overhead. Delta compression alleviates this by storing only the pretrained model and the highly compressed delta weights (the differences between fine-tuned and pretrained model weights). However, existing methods fail to maintain both high compression and performance, and often rely on data. To address these challenges, we propose UltraDelta, the first data-free delta compression pipeline that achieves both ultra-high compression and strong performance. UltraDelta is designed to minimize redundancy, maximize information, and stabilize performance across inter-layer, intra-layer, and global dimensions, using three key components: (1) Variance-Based Mixed Sparsity Allocation assigns sparsity based on variance, giving lower sparsity to high-variance layers to preserve inter-layer information. (2) Distribution-Aware Compression applies uniform quantization and then groups parameters by value, followed by group-wise pruning, to better preserve intra-layer distribution. (3) Trace-Norm-Guided Rescaling uses the trace norm of delta weights to estimate a global rescaling factor, improving model stability under higher compression. Extensive experiments across (a) large language models (fine-tuned on LLaMA-2 7B and 13B) with up to 50x compression, (b) general NLP models (RoBERTa-base, T5-base) with up to 224x compression, (c) vision models (ViT-B/32, ViT-L/14) with up to 132x compression, and (d) multi-modal models (BEiT-3) with 18x compression, demonstrate that UltraDelta consistently outperforms existing methods, especially under ultra-high compression. Code is available at https://github.com/xiaohuiwang000/UltraDelta.
format Preprint
id arxiv_https___arxiv_org_abs_2505_13563
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Breaking the Compression Ceiling: Data-Free Pipeline for Ultra-Efficient Delta Compression
Wang, Xiaohui
Ye, Peng
Huang, Chenyu
Zheng, Shenghe
Zhang, Bo
Bai, Lei
Ouyang, Wanli
Chen, Tao
Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
With the rise of the fine-tuned-pretrained paradigm, storing numerous fine-tuned models for multi-tasking creates significant storage overhead. Delta compression alleviates this by storing only the pretrained model and the highly compressed delta weights (the differences between fine-tuned and pretrained model weights). However, existing methods fail to maintain both high compression and performance, and often rely on data. To address these challenges, we propose UltraDelta, the first data-free delta compression pipeline that achieves both ultra-high compression and strong performance. UltraDelta is designed to minimize redundancy, maximize information, and stabilize performance across inter-layer, intra-layer, and global dimensions, using three key components: (1) Variance-Based Mixed Sparsity Allocation assigns sparsity based on variance, giving lower sparsity to high-variance layers to preserve inter-layer information. (2) Distribution-Aware Compression applies uniform quantization and then groups parameters by value, followed by group-wise pruning, to better preserve intra-layer distribution. (3) Trace-Norm-Guided Rescaling uses the trace norm of delta weights to estimate a global rescaling factor, improving model stability under higher compression. Extensive experiments across (a) large language models (fine-tuned on LLaMA-2 7B and 13B) with up to 50x compression, (b) general NLP models (RoBERTa-base, T5-base) with up to 224x compression, (c) vision models (ViT-B/32, ViT-L/14) with up to 132x compression, and (d) multi-modal models (BEiT-3) with 18x compression, demonstrate that UltraDelta consistently outperforms existing methods, especially under ultra-high compression. Code is available at https://github.com/xiaohuiwang000/UltraDelta.
title Breaking the Compression Ceiling: Data-Free Pipeline for Ultra-Efficient Delta Compression
topic Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2505.13563