Every Step Counts: Decoding Trajectories as Authorship Fingerprints of dLLMs
Qi Li, Runpeng Yu, Haiquan Lu, Xinchao Wang
TL;DR
This work addresses the challenge of attributing outputs to specific discrete diffusion LLMs (dLLMs). It introduces Directed Decoding Map (DDM) to extract structured, inter-step dependencies from decoding trajectories and Gaussian-Trajectory Attribution (GTA) to fingerprint models via cell-wise Gaussian distributions. Across cross-model, independent-run, and cross-checkpoint scenarios on math and code tasks, the combination of DDM and GTA consistently outperforms baselines, even when two models share checkpoints and identical training configurations. The proposed trajectory-based approach offers a lightweight, robust tool for attribution, with strong potential for practical deployment in responsibility and misuse detection in dLLMs.
Abstract
Discrete Diffusion Large Language Models (dLLMs) have recently emerged as a competitive paradigm for non-autoregressive language modeling. Their distinctive decoding mechanism enables faster inference speed and strong performance in code generation and mathematical tasks. In this work, we show that the decoding mechanism of dLLMs not only enhances model utility but also can be used as a powerful tool for model attribution. A key challenge in this problem lies in the diversity of attribution scenarios, including distinguishing between different models as well as between different checkpoints or backups of the same model. To ensure broad applicability, we identify two fundamental problems: what information to extract from the decoding trajectory, and how to utilize it effectively. We first observe that relying directly on per-step model confidence yields poor performance. This is mainly due to the bidirectional decoding nature of dLLMs: each newly decoded token influences the confidence of other decoded tokens, making model confidence highly redundant and washing out structural signal regarding decoding order or dependencies. To overcome this, we propose a novel information extraction scheme called the Directed Decoding Map (DDM), which captures structural relationships between decoding steps and better reveals model-specific behaviors. Furthermore, to make full use of the extracted structural information during attribution, we propose Gaussian-Trajectory Attribution (GTA), where we fit a cell-wise Gaussian distribution at each decoding position for each target model, and define the likelihood of a trajectory as the attribution score: if a trajectory exhibits higher log-likelihood under the distribution of a specific model, it is more likely to have been generated by that model. Extensive experiments under different settings validate the utility of our methods.