Everything You Always Wanted to Know About Storage Compressibility of Pre-Trained ML Models but Were Afraid to Ask
Summary: Exhaustive analysis of pre-trained model file compressibility across granularity levels, showing general-purpose compressors fail to exploit PTM-specific patterns. Propose Elf, an error-bounded float transform that removes shared exponents, and Elves framework; achieves 1.52× compression (~1.3× vs zstd/SZ3/quant) with negligible accuracy loss. (summarized by gpt-5-mini on Feb 09 2026)
Incoming Non-self Citations Over Time
Authors
- 1. Zhaoyuan Su (University of Virginia)
- 2. Ammar Ahmed (University of Minnesota)
- 3. Zirui Wang (University of Virginia)
- 4. Ali Anwar (University of Minnesota)
- 5. Yue Cheng (University of Virginia)
BibTeX Citation
@article{su_vldb24,
title = {{Everything You Always Wanted to Know About Storage Compressibility of Pre-Trained ML Models but Were Afraid to Ask}},
author = {Su, Zhaoyuan and Ahmed, Ammar and Wang, Zirui and Anwar, Ali and Cheng, Yue},
journal = {PVLDB},
series = {{VLDB} '24},
volume = {17},
number = {8},
pages = {2036--2049},
doi = {10.14778/3659437.3659456},
url = {https://doi.org/10.14778/3659437.3659456},
year = {2024}
}Incoming Citations (Sorted by Pagerank)
Showing 3 of 3 citing papers.
| Rank | Citing Paper | Year | Venue | Pagerank |
|---|---|---|---|---|
| 8,221 | NeurDB: On the Design and Implementation of an AI-powered Autonomous Database | 2025 | CIDR | 5.4640314e-05 |
| 10,386 | NeurStore: Efficient In-database Deep Learning Model Management System | 2026 | SIGMOD | 5.093636e-05 |
| 11,078 | LiquidCache: Efficient Pushdown Caching for Cloud-Native Data Analytics | 2025 | VLDB | 5.093636e-05 |
Previous
Page 1 / 1
Next
Outgoing Citations (Sorted by Pagerank)
Showing 5 of 5 cited papers.
Citations counted here include only citations to other VLDB/SIGMOD/CIDR/PODS papers in this database.
| Rank | Cited Paper | Year | Venue | Pagerank |
|---|---|---|---|---|
| 148 | Gorilla: A Fast, Scalable, In-Memory Time Series Database | 2015 | VLDB | 0.00029250767 |
| 1,364 | Chimp: Efficient Lossless Floating Point Compression for Time Series Databases | 2022 | VLDB | 0.00011020069 |
| 2,023 | Decomposed Bounded Floats for Fast Compression and Queries | 2021 | VLDB | 9.2950046e-05 |
| 2,788 | BtrBlocks: Efficient Columnar Compression for Data Lakes | 2023 | SIGMOD | 8.1205155e-05 |
| 5,021 | Online Deduplication for Databases | 2017 | SIGMOD | 6.3982122e-05 |
Previous
Page 1 / 1
Next
Semantically Similar Papers
| # | Overall Rank | Paper | Year | Venue |
|---|---|---|---|---|
| 1 | 10,589 | Morphing-based Compression for Data-centric ML Pipelines | 2026 | VLDB |
| 2 | 6,485 | Tuple-oriented Compression for Large-scale Mini-batch Stochastic Gradient Descent | 2019 | SIGMOD |
| 3 | 3,269 | ALP: Adaptive Lossless floating-Point Compression | 2023 | SIGMOD |
| 4 | 8,794 | AWARE: Workload-aware, Redundancy-exploiting Linear Algebra | 2023 | SIGMOD |
| 5 | 6,190 | Progressive Compressed Records: Taking a Byte out of Deep Learning Data | 2021 | VLDB |
| 6 | 10,584 | QStore: Quantization-Aware Compressed Model Storage | 2026 | VLDB |
| 7 | 11,770 | An Evaluation of Methods of Compressing Doubles | 2020 | SIGMOD |
| 8 | 1,644 | Compressed Linear Algebra for Large-Scale Machine Learning | 2016 | VLDB |
| 9 | 9,520 | Experimental Analysis of Large-scale Learnable Vector Storage Compression | 2024 | VLDB |
| 10 | 3,540 | Elf: Erasing-based Lossless Floating-Point Compression | 2023 | VLDB |