Prism:通过分层搜索和自验证实现离散扩散语言模型的高效测试时扩展
Prism: Efficient Test-Time Scaling via Hierarchical Search and Self-Verification for Discrete Diffusion Language Models
AI总结:
本文提出Prism框架,通过分层轨迹搜索、局部分支和自验证反馈,提升离散扩散语言模型的测试时扩展效率与性能。
摘要:
推理时的计算量重新成为提升LLM推理能力的实用方法。大多数测试时扩展(TTS)算法依赖于自回归解码,这在离散扩散语言模型(dLLMs)中并不适用,因为它们在整个序列上并行解码。因此,开发有效且高效的TTS方法以解锁dLLMs的完整生成潜力仍是一个未被充分探索的挑战。为此,我们提出了Prism(剪枝、重掩码和集成自验证方法),一种针对dLLMs的高效TTS框架,其(i)执行分层轨迹搜索(HTS),在早期到中期去噪窗口中动态剪枝并重新分配计算;(ii)引入局部分支和部分重掩码以探索多样化的实现方式,同时保留高置信度的token;(iii)用通过中间完成的自评估提示获得的自验证反馈(SVF)替代外部验证器。在四个数学推理和代码生成基准测试中,包括LLaDA 8B Instruct、Dream 7B Instruct和LLaDA 2.0-mini三个dLLMs上,我们的Prism实现了有利的性能-效率权衡,与最佳N性能相比,显著减少了函数评估次数(NFE)。代码已发布在https://github.com/viiika/Prism。
英文摘要:
Inference-time compute has re-emerged as a practical way to improve LLM reasoning. Most test-time scaling (TTS) algorithms rely on autoregressive decoding, which is ill-suited to discrete diffusion language models (dLLMs) due to their parallel decoding over the entire sequence. As a result, developing effective and efficient TTS methods to unlock dLLMs' full generative potential remains an underexplored challenge. To address this, we propose Prism (Pruning, Remasking, and Integrated Self-verification Method), an efficient TTS framework for dLLMs that (i) performs Hierarchical Trajectory Search (HTS) which dynamically prunes and reallocates compute in an early-to-mid denoising window, (ii) introduces Local branching with partial remasking to explore diverse implementations while preserving high-confidence tokens, and (iii) replaces external verifiers with Self-Verified Feedback (SVF) obtained via self-evaluation prompts on intermediate completions. Across four mathematical reasoning and code generation benchmarks on three dLLMs, including LLaDA 8B Instruct, Dream 7B Instruct, and LLaDA 2.0-mini, our Prism achieves a favorable performance-efficiency trade-off, matching best-of-N performance with substantially fewer function evaluations (NFE). The code is released at https://github.com/viiika/Prism.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。