Inference-time alignment methods have gained significant attention for their efficiency and effectiveness in aligning large language models (LLMs) with human preferences. However, existing dominant approaches, reward-guided search (RGS), suffer from a critical granularity mismatch: reward models (RMs) are trained on complete responses but applied to incomplete sequences during generation, leading to inconsistent scoring and suboptimal alignment. To combat the challenge, we argue that an ideal RM should satisfy two objectives: Score Consistency, ensuring coherent evaluation across partial and complete responses, and Preference Consistency, aligning partial sequence assessments with human preferences. To achieve these, we propose SPRM, a novel dual-consistency framework integrating score consistency-based and preference consistency-based partial evaluation modules, which leverage the Bradley-Terry model and entropy-based reweighting to predict cumulative rewards and prioritize human-aligned sequences. Extensive experiments on dialogue, summarization, and reasoning tasks demonstrate the effectiveness of SPRM, significantly reducing granularity discrepancies by up to 11.7 on TL;DR Summarization and achieving a 3.6%–10.3% improvement in GPT-4 evaluation scores across all tasks. Code is publicly available at [this link](https://github.com/xiebin23/SPRM).

This paper introduces a novel generalized self-imitation learning GSIL framework, which effectively and efficiently aligns large language models with offline demonstration data. We develop GSIL by deriving a surrogate objective of imitation learning with density ratio estimates, facilitating the use of self-generated data and optimizing the imitation learning objective with simple classification losses. GSIL eliminates the need for complex adversarial training in standard imitation learning, achieving lightweight and efficient fine-tuning for large language models. In addition, GSIL encompasses a family of offline losses parameterized by a general class of convex functions for density ratio estimation and enables a unified view for alignment with demonstration data. Extensive experiments show that GSIL consistently and significantly outperforms baselines in many challenging benchmarks, such as coding (HuamnEval), mathematical reasoning (GSM8K) and instruction-following benchmark (MT-Bench). Code is public available at https://github.com/tengxiao1/GSIL.