TACG：面向扩散语言模型解码的轨迹感知提交门控

View PDF HTML (experimental)

Abstract:Diffusion language models predict many masked positions at each denoising step, but their efficiency and output quality depend on which predictions enter the evolving context. Confidence is informative but incomplete for this decision: at matched confidence, proposals with different identity and support histories exhibit different one-step revision rates. We propose Trajectory-Aware Commit Gating (TACG), a training-free decoder that separates the current token proposal from the decision to commit it. TACG scores temporal support for the current proposal by contrasting its logits with an exponential moving average of earlier logits, requires the proposal identity to persist through a History Gate, and promotes a capped number of additional positions per step. The gate decides which positions to commit and when; every written token remains the contemporaneous base proposal. On complete four-task LLaDA evaluations under the stated single-seed protocol, measured task-specific operating points reach 51.02\% mean accuracy with 67.82 mean model evaluations, compared with 48.00\% and 76.59 for Native Confidence. Under a matched per-event commit-count replay, history-guided selection exceeds confidence ranking by 8.3 accuracy points at identical per-sample NFE. History-reference interventions alter commitment timing while preserving the base-proposal write rule. Experiments on Dream and LLaDA2-Mini characterize quality--cost behavior across backbones and generation lengths. The code is publicly available at this https URL.

Subjects:	Computation and Language (cs.CL)
Cite as:	arXiv:2607.03236 [cs.CL]
	(or arXiv:2607.03236v2 [cs.CL] for this version)
	https://doi.org/10.48550/arXiv.2607.03236 arXiv-issued DOI via DataCite

Submission history

From: Chengcheng Wang [view email]
[v1] Fri, 3 Jul 2026 11:45:21 UTC (3,540 KB)
[v2] Thu, 5 Nov 2026 04:49:37 UTC (470 KB)