Diffusion Language Models (DLMs) enable parallel text generation by denoising multiple masked positions in a single pass, but confidence-threshold decoding often commits only the safest predictions, leaving informative uncertainty underused. We propose Plan-Verify-Fill (PVF), a training-free draft-and-verification decoding strategy that preserves the high-confidence decoding backbone while selectively testing lower-confidence candidates with high expected verified utility. PVF uses a calibrated planning-token vocabulary to allocate its limited verification budget, accepts proposed drafts only when they remain consistent with the backbone trajectory, and otherwise defaults to conservative confidence-gated unmasking. Across mathematical reasoning, knowledge, and code-generation benchmarks on LLaDA-8B-Instruct and Dream-7B-Instruct, PVF maintains comparable accuracy while delivering up to 2.3× higher throughput than baseline methods.

In recent years, Large Language Models (LLMs) have demonstrated remarkable capabilities across various domains. Within the training pipeline of LLMs, the Reinforcement Learning with Human Feedback (RLHF) phase is crucial for aligning LLMs with human preferences and values. Label smoothing, a technique that replaces hard labels with soft labels, emerges as promising techniques to enhance RLHF training. Despite the benefits, the choice of label smoothing parameters often relies on heuristic approaches and lack theoretical understanding. This paper addresses the challenge of selecting the label smoothing parameter in a principled manner. We introduce Confidence Aware Label Smoothing (CALS), a method that iteratively updates the label smoothing parameter based on preference labels and model forecasts. Our theoretical analysis characterizes the optimal label smoothing parameter, demonstrates its dependence on the confidence level, and reveals its influence on training dynamics and equilibrium. Empirical evaluations on state-of-the-art alignment tasks show that CALS achieves competitive performance, highlighting its potential for improving alignment.