Balanced Actor Initialization: Stable RLHF Training of Distillation-Based Reasoning Models
Abstract
The development of alignment and reasoning capabilities in large language models has seen remarkable progress through two paradigms: instruction tuning and reinforcement learning from human feedback (RLHF) alignment paradigm, and distillation-based reasoning fine-tuning paradigm. While both approaches prove effective independently, the third paradigm of applying RLHF to distillation-trained models presents significant challenges. Our investigation reveals two critical phenomena that emerge in this paradigm: Sequence Length Collapse, where language generation dramatically reduces during early RLHF training, and the Reward Hockey Stick Curve, featuring severe reward score drops followed by gradual recovery. These instabilities fundamentally compromise the model's alignment and reasoning capabilities. To address these challenges, we propose Balanced Actor Initialization (BAI), a two-stage weighted model merging approach. BAI first merges instruction-following and distillation-based reasoning fine-tuned models, then further combines this intermediate model with the pretrained model to preserve foundational knowledge. Through comprehensive experiments across diverse benchmarks and detailed analysis of training experiments, we demonstrate that BAI resolves Sequence Length Collapse, mitigates the Reward Hockey Stick Curve, and enables continuous sequence length improvement during training. Additionally, our analysis reveals that balanced merging ratios achieve optimal trade-offs between training stability and reasoning capability preservation. Our work provides the effective solution for stable training in this third paradigm, enabling more capable reasoning models that combine distillation efficiency with RLHF alignment.
Keywords
Cite
@article{arxiv.2509.00309,
title = {Balanced Actor Initialization: Stable RLHF Training of Distillation-Based Reasoning Models},
author = {Chen Zheng and Yiyuan Ma and Yuan Yang and Deyi Liu and Jing Liu and Zuquan Song and Yuxin Song and Cheng Ren and Hang Zhu and Xin Liu and Yiyuan Ma and Siyuan Qiao and Xun Zhou and Liang Xiang and Yonghui Wu},
journal= {arXiv preprint arXiv:2509.00309},
year = {2025}
}