LLM alignment remains a critical challenge. Inference-time methods provide a flexible alternative to fine-tuning, but their uniform computational effort often yields suboptimal alignment. We hypothesize that for many alignment tasks, the initial tokens of a response are disproportionately more critical. To leverage this principle, we introduce AdaSearch, a novel blockwise search strategy. It adaptively allocates a fixed computational budget using a sampling schedule, focusing search effort on these critical tokens. We apply AdaSearch to sequential decoding and introduce its tree-search counterpart, AdaBeam. Our comprehensive evaluation across eight LLMs demonstrates that AdaSearch outperforms strong Best-of-N and fine-tuning baselines. Specifically, win-rates improve by over 10% for harmlessness generation, controlled sentiment generation, and for mathematical reasoning tasks relative to Best-of-N.
@article{arxiv.2510.23334,
title = {Adaptive Blockwise Search: Inference-Time Alignment for Large Language Models},
author = {Mohammad Atif Quamar and Mohammad Areeb and Nishant Sharma and Ananth Shreekumar and Jonathan Rosenthal and Muslum Ozgur Ozmen and Mikhail Kuznetsov and Z. Berkay Celik},
journal= {arXiv preprint arXiv:2510.23334},
year = {2025}
}