LoRA-WiSE
收藏资源简介:
LoRA-WiSE数据集由耶路撒冷希伯来大学计算机科学与工程学院创建,包含超过25,000个权重快照,来自2,000多个独立的LoRA模型,涵盖不同的数据集大小、骨干网络、秩和个人化设置。该数据集旨在评估从LoRA权重中恢复数据集大小的方法,特别是在生成模型中的应用。数据集的创建过程涉及对多个LoRA模型进行微调,并记录其权重变化。LoRA-WiSE数据集主要用于研究模型微调和数据集大小恢复的技术,特别是在机器学习模型的隐私和安全领域。
The LoRA-WiSE dataset was created by the School of Computer Science and Engineering, The Hebrew University of Jerusalem. It comprises over 25,000 weight snapshots derived from more than 2,000 independent LoRA models, covering a range of dataset sizes, backbone networks, ranks, and personalization settings. This dataset is designed to evaluate methods for recovering dataset sizes from LoRA weights, especially their applications in generative models. The creation of the dataset involves fine-tuning multiple LoRA models and recording their weight changes. The LoRA-WiSE dataset is primarily used for researching technologies related to model fine-tuning and dataset size recovery, particularly in the privacy and security fields of machine learning models.
数据集详情总结
数据集标题
Data Size Recovery from Lora Weights
作者信息
- Mohammad Salama
- Jonathan Kahana
- Eliahu Horwitz
- Yedid Hoshen
所属机构
The Hebrew University of Jerusalem
数据集链接
描述
该数据集与从Lora权重中恢复数据大小相关。详细信息和相关研究可以在提供的论文和GitHub项目中找到。

- 1Dataset Size Recovery from LoRA Weights耶路撒冷希伯来大学计算机科学与工程学院 · 2024年



