InfiAlign: A Scalable and Sample-Efficient Framework for Aligning LLMs to Enhance Reasoning Capabilities Paper β’ 2508.05496 β’ Published Aug 7 β’ 9