TRE is a parameter-free and single-run metric that estimates hallucination risk directly from the entropy signals of a single generation, without requiring any detector training or repeated sampling, and enjoys strong generalizability, efficiency, and robustness.
Abstract
Diffusion large language models (D-LLMs) have recently gained increasing attention, yet their reliability is significantly hindered by the hallucination problem. Existing hallucination detection approaches for D-LLMs mainly follow a training-based paradigm, relying on data-driven training to optimize the detector. Such reliance not only limits their generalizability across domains models but also incurs additional training cost and deployment overhead. To address these limitations, we propose TRE, a training-free hallucination detection metric for D-LLMs. TRE is a parameter-free and single-run metric that estimates hallucination risk directly from the entropy signals of a single generation, without requiring any detector training or repeated sampling. TRE extracts entropy signals within the D-LLM decoding process along both the spatial and temporal dimensions. From a token-level spatial perspective, we focus on revealing tokens as the most informative carriers of uncertainty, capturing where uncertainty is actively committed. From a diffusion step-level temporal perspective, we empirically identify the dominance of late-step entropy and hence aggregate these signals with a simple linear weighting scheme to obtain TRE. Extensive experiments on multiple D-LLMs and QA datasets demonstrate that TRE achieves competitive performance, while enjoying strong generalizability, efficiency, and robustness.
Diffusion large language models (D-LLMs) have emerged as a promising paradigm for text generation. However, similar to autoregressive LLMs, D-LLMs remain vulnerable to hallucinations, where fluent outputs may contain factually incorrect or unsupported content. Although existing hallucination detection methods for D-LLMs attempt to leverage uncertainty trajectories of the denoising process to better identify hallucination signals, they typically compress the trajectories along either the temporal or token dimension, overlooking the useful information encoded in the complete two-dimensional token-step structure. Consequently, they may fail to capture hallucination-relevant patterns, such as inconsistent convergence and cross-token fault propagation, leading to suboptimal detection performance. To bridge this gap, we propose a D-LLM hallucination detection framework that formulates the Denoising trajectories as Multivariate Time Series over learnable latent variables (DeMTS for short). DeMTS employs a trajectory-preserving token-to-variable assignment module to convert token signals into stable latent variables. Based on these variables, we propose dynamic multivariate temporal modeling to progressively integrate inter-variable dependency modeling with temporal encoding for hallucination prediction. Extensive experiments on two D-LLMs backbones and three benchmarks demonstrate that DeMTS outperforms existing hallucination detection methods while maintaining strong robustness, efficiency, and cross-task transferability.
This work proposes AURORA, a novel hallucination detection framework that shifts the focus from static representations to the weight-gradient dynamics of LLMs, and achieves strong hallucination detection performance across four model families and four benchmark datasets.
Z. Zhang, Hainan Zhang, Zhiming Zheng· arXiv.org· 0 citations
Object Hallucination in large vision-language models (LVLMs), where models generate non-factual content about input images, remains a critical barrier to their reliability in real-world applications. Existing mitigation strategies can be categorized into training-based and training-free methods. Training-based methods often achieve strong performance but are costly, requiring extensive computational resources, large-scale data, and time-consuming fine-tuning. Training-free approaches are particularly appealing due to their efficiency. However, existing training-free methods either require multiple decoding rounds, which adds computational overhead, or modify internal states in a model-specific way that risks degrading pretrained knowledge. We propose Test-Time Hallucination Mitigation (TTH) method, a novel training-free method that addresses both limitations. TTH introduces a token-validator module, implemented as a zero-shot Multi-Modal Classifier (MMC), to generate auxiliary logits grounded in the input image. These logits are fused with the original LVLM outputs at the token level for object tokens selected from a candidate pool. An entropy-based weighting scheme is then applied to enable robust and accurate predictions. Extensive experiments across multiple LVLM families and diverse benchmarks demonstrate that TTH consistently improves accuracy and robustness, underscoring its generalizability and practical effectiveness. Code is released at https://github.com/Mehran-TAM/TTH
Mehran Tamjidi, Hamidreza Dastmalchi, Ali Cheraghian et al.· 0 citations
Hallucination detection is essential for the reliable deployment of large language models (LLMs) in safety-critical applications, yet existing detectors often degrade under real-world domain shifts. To formalize this challenge, we introduce Multi-Domain Generalizable Hallucination Detection (MGHD), where a detector is trained on multiple labeled domains and evaluated on unseen domains without any test-domain supervision. The central challenge of MGHD stems from a previously underexplored phenomenon, termed layer-index shift, in which the most informative internal layer varies substantially across domains, causing detectors that rely on a fixed layer to generalize poorly. To mitigate this phenomenon, we propose a novel method Stability-Aware Adaptive Layer Aggregation (SALA). SALA defines a layer-wise instability index computed from the training data, which measures how separability varies across domains. For an unseen test domain, it provides an upper bound on the possible separability drop from training to test. Guided by this bound, SALA designs a per-layer projector that balances discriminability and stability. Then using these projectors, SALA learns invariant subspaces across layers and aggregates them into a robust detector. Experiments across four widely used QA benchmarks and four LLM backbones from the LLaMA and Qwen families (7B-14B) show that SALA achieves the best average performance compared with existing hallucination detection baselines.
Xinyi Li, Zhen Fang, Yadan Luo et al.· Proceedings of the 32nd ACM...· 0 citations
Latent Attention Denoising (LAD) is introduced, a principled and training-free framework that recasts attention calibration as a one-step score-based denoising process that achieves superior performance on both generative and discriminative tasks.
Zhiwen Luo, Siyu Jiang, Weilong Jiang et al.· Annual Meeting of the Associ...· 0 citations
This work proposes FADE (FFN Attenuation for DEcoding), a training-free method that attenuates FFN outputs to reduce language-prior dominance and shows that FADE effectively mitigates hallucinations while preserving inference efficiency.
Yichen Guo, Kai Tang, Fenglai Lin et al.· arXiv.org· 2 citations