Saved in:
Bibliographic Details
Main Authors: Bai, Jinbin, Li, Yixuan, Zhu, Yuchen, Xin, Yi, Shi, Qingyu, Feng, Aosong, Liu, Xiaohong, Tao, Molei, Xue, Jianru, Li, Xiangtai, Yang, Ming-Hsuan
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2602.01842
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914529923825664
author Bai, Jinbin
Li, Yixuan
Zhu, Yuchen
Xin, Yi
Shi, Qingyu
Feng, Aosong
Liu, Xiaohong
Tao, Molei
Xue, Jianru
Li, Xiangtai
Yang, Ming-Hsuan
author_facet Bai, Jinbin
Li, Yixuan
Zhu, Yuchen
Xin, Yi
Shi, Qingyu
Feng, Aosong
Liu, Xiaohong
Tao, Molei
Xue, Jianru
Li, Xiangtai
Yang, Ming-Hsuan
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2602_01842
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Prism: Efficient Test-Time Scaling via Hierarchical Search and Self-Verification for Discrete Diffusion Language Models
Bai, Jinbin
Li, Yixuan
Zhu, Yuchen
Xin, Yi
Shi, Qingyu
Feng, Aosong
Liu, Xiaohong
Tao, Molei
Xue, Jianru
Li, Xiangtai
Yang, Ming-Hsuan
Machine Learning
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.
title Prism: Efficient Test-Time Scaling via Hierarchical Search and Self-Verification for Discrete Diffusion Language Models
topic Machine Learning
url https://arxiv.org/abs/2602.01842