Adaptive Blockwise Search: Inference-Time Alignment for Large Language Models

Mohammad Atif Quamar, Mohammad Areeb, Nishant Sharma +5

LLM alignment remains a critical challenge. Inference-time methods provide a flexible alternative to fine-tuning, but their uniform computational effort often yields suboptimal ali…