metadata
license: mit
language:
- en
base_model:
- Qwen/Qwen2.5-7B-Instruct
Language Model Watermark Evasion Attacks
Introduction
An adaptively trained Qwen2.5-7B paraphraser that was adaptively tuned against the EXP watermark (Aaronson et al.) using the method described in the work "OPTIMIZING ADAPTIVE ATTACKS AGAINST CONTENT WATERMARKS FOR LANGUAGE MODELS" (https://arxiv.org/pdf/2410.02440).