Learning to Align, Aligning to Learn: A Unified Approach for Self-Optimized Alignment
This work addresses the problem of efficient alignment for language models, offering a novel method that is incremental but with strong specific gains.
The paper tackles the dual challenges of low sample efficiency in reinforcement learning and constrained efficacy in supervised fine-tuning for language model alignment by proposing GRAO, a unified framework that synergizes their strengths, achieving relative improvements of 57.70%, 17.65%, 7.95%, and 5.18% over SFT, DPO, PPO, and GRPO baselines respectively.
Alignment methodologies have emerged as a critical pathway for enhancing language model alignment capabilities. While SFT (supervised fine-tuning) accelerates convergence through direct token-level loss intervention, its efficacy is constrained by offline policy trajectory. In contrast, RL(reinforcement learning) facilitates exploratory policy optimization, but suffers from low sample efficiency and stringent dependency on high-quality base models. To address these dual challenges, we propose GRAO (Group Relative Alignment Optimization), a unified framework that synergizes the respective strengths of SFT and RL through three key innovations: 1) A multi-sample generation strategy enabling comparative quality assessment via reward feedback; 2) A novel Group Direct Alignment Loss formulation leveraging intra-group relative advantage weighting; 3) Reference-aware parameter updates guided by pairwise preference dynamics. Our theoretical analysis establishes GRAO's convergence guarantees and sample efficiency advantages over conventional approaches. Comprehensive evaluations across complex human alignment tasks demonstrate GRAO's superior performance, achieving 57.70\%,17.65\% 7.95\% and 5.18\% relative improvements over SFT, DPO, PPO and GRPO baselines respectively. This work provides both a theoretically grounded alignment framework and empirical evidence for efficient capability evolution in language models.