提出分步规划验证填空法，让文本生成更高效准确
Plan, Verify and Fill: A Structured Parallel Decoding Approach for Diffusion Language Models
- 先规划关键语义骨架，再验证是否需继续优化
- 相比传统方法减少65%的计算量，保持准确率不变
- 适合追求高效生成的开发者和研究者使用
扩散语言模型（DLMs）为文本生成提供了一种非序列化的新范式，区别于传统的自回归方法。然而，当前解码策略多采用被动响应模式，未能充分利用全局双向上下文来引导整体生成路径。为此，我们提出无需训练的计划-验证-填空（PVF）框架，通过量化验证实现规划。PVF主动构建层次化骨架，优先选择高影响力语义锚点，并引入验证机制，在进一步思考收益递减时及时停止。在LLaDA-8B-Instruct和Dream-7B-Instruct上的大量实验表明，相较于基于置信度的并行解码，PVF在基准数据集上将函数评估次数（NFE）降低高达65%，显著提升效率且不牺牲准确性。
原文摘要 · Abstract (English)
Diffusion Language Models (DLMs) present a promising non-sequential paradigm for text generation, distinct from standard autoregressive (AR) approaches. However, current decoding strategies often adopt a reactive stance, underutilizing the global bidirectional context to dictate global trajectories. To address this, we propose Plan-Verify-Fill (PVF), a training-free paradigm that grounds planning via quantitative validation. PVF actively constructs a hierarchical skeleton by prioritizing high-leverage semantic anchors and employs a verification protocol to operationalize pragmatic structural stopping where further deliberation yields diminishing returns. Extensive evaluations on LLaDA-8B-Instruct and Dream-7B-Instruct demonstrate that PVF reduces the Number of Function Evaluations (NFE) by up to 65% compared to confidence-based parallel decoding across benchmark datasets, unlocking superior efficiency without compromising accuracy.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。