APCD：面向可靠大语言模型生成的自适应路径对比解码

View PDF HTML (experimental)

Abstract:Reliable text generation is critical for deploying large language models (LLMs) in real-world applications, particularly in high-stakes domains such as medicine. To improve factual reliability, various inference-time methods have been proposed, including logit-level methods that modify token probability distributions and representation-level methods that manipulate intermediate model representations. However, most existing approaches operate on a single decoding trajectory, limiting their ability to explore alternative reasoning paths and making them susceptible to error accumulation. To address this limitation, we propose Adaptive Path-Contrastive Decoding (APCD), an adaptive multi-path contrastive decoding framework that improves factual reliability without model retraining or fine-tuning. APCD comprises two key components: Entropy-Driven Path Expansion, which adaptively expands the decoding process only at high-uncertainty decision points, and Divergence-Aware Path Contrast, which dynamically regulates contrastive interactions among parallel decoding paths based on their distributional divergence to balance diversity and coherence. We evaluate APCD on four LLM backbones across eight benchmarks spanning both general-domain and medical question answering tasks. Experimental results demonstrate that APCD consistently outperforms strong inference-time baselines in factual accuracy while maintaining competitive inference efficiency. These results demonstrate the robustness and generalizability of APCD across diverse models and tasks, highlighting its effectiveness as a practical multi-path decoding framework for reliable LLM deployment, particularly in high-stakes domains such as medicine. Code is available at this https URL.

Comments:	This is an extended journal version submitted to ESWA. It builds upon a previously withdrawn conference manuscript (ACL format). The core research work remains unchanged, with substantial extensions including additional ablation experiments and deeper analysis to meet journal requirements
Subjects:	Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
Cite as:	arXiv:2605.09492 [cs.CL]
	(or arXiv:2605.09492v3 [cs.CL] for this version)
	https://doi.org/10.48550/arXiv.2605.09492 arXiv-issued DOI via DataCite

Submission history

From: Tianyu Zheng [view email]
[v1] Sun, 10 May 2026 11:57:39 UTC (468 KB)
[v2] Wed, 20 May 2026 15:55:09 UTC (1 KB) (withdrawn)
[v3] Wed, 7 Oct 2026 13:32:01 UTC (906 KB)