Eliahu/

LoWRA-Bench

The LoRA Weight Recovery Attack (LoWRA) Bench is a comprehensive benchmark designed to evaluate Pre-Fine-Tuning (Pre-FT) weight recovery methods as presented in the "Recovering the Pre-Fine-Tuning Weights of Generative Models" paper.

Task Details

Dataset Description

Dataset Structure

Data Subsets Data Fields Layer Merging Example

Dataset Creation

Risks and Out-of-Scope Use

Considerations for Using the Data

Licensing Information… See the full description on the dataset page: https://huggingface.co/datasets/Eliahu/LoWRA-Bench.