MarkTune: Improving the Quality-Detectability Trade-off in Open-Weight LLM Watermarking

TL;DR

MarkTune is a fine-tuning framework that enhances watermark detectability in open-weight language models while maintaining high text quality, addressing limitations of existing watermarking methods.

Contribution

It introduces a theoretically grounded, on-policy fine-tuning method that improves the quality-detectability trade-off for open-weight watermarking techniques like GaussMark.

Findings

MarkTune significantly improves watermark detection power without degrading text quality.

It achieves robustness against paraphrasing and fine-tuning attacks.

The method generalizes well across different datasets and models.

Abstract

Watermarking aims to embed hidden signals in generated text that can be reliably detected when given access to a secret key. Open-weight language models pose acute challenges for such watermarking schemes because the inference-time interventions that dominate contemporary approaches cannot be enforced once model weights are public. Existing watermaking techniques for open-weight models, such as the recently proposed GaussMark, typically rely on small modifications to model weights, which can yield signals detectable to those equipped with a secret key, but achieving detection power comparable to inference-time watermarks generally requires weight perturbations that noticeably reduce generation quality. We introduce MarkTune, a theoretically principled, on-policy fine-tuning framework that treats the GaussMark signal as a reward while simultaneously regularizing against degradation in text quality. We derive MarkTune as an improvement on GaussMark and demonstrate that MarkTune consistently improves the quality-detectability trade-off over GaussMark by steering finer-grained, watermark-aware weight updates within the model's representation space while preserving generation quality. Empirically, we show that MarkTune pushes the quality-detectability frontier of GaussMark close to that of inference-time watermarking, remains robust to paraphrasing and fine-tuning attacks, and exhibits strong generalization: a model fine-tuned on one dataset retains substantial watermark detection power on unseen datasets. Together, these results establish MarkTune as a general strategy for embedding robust, high-quality watermarks into open-weight LMs.