PowerFlow: Unlocking the Dual Nature of LLMs via Principled Distribution Matching 文章

ArXiv CS.CL2026-05-26NEWSen作者: Ruishuo Chen, Yu Chen, Zhuoran Li, Longbo Huang

详细信息

来源站点
ArXiv CS.CL
作者
Ruishuo Chen, Yu Chen, Zhuoran Li, Longbo Huang
文章类型
NEWS
语言
en
发布日期
2026-05-26

摘要

arXiv:2603.18363v2 Announce Type: replace Abstract: Unsupervised Reinforcement Learning from Internal Feedback (RLIF) has emerged as a promising paradigm for eliciting the latent capabilities of Large Language Models (LLMs) without external supervision. However, current methods rely on heuristic intrinsic rewards, which often lack a well-defined theoretical optimization target and are prone to degenerative biases. In this work, we introduce PowerFlow, a principled framework that reformulates unsupervised fine-tuning as a distribution matching problem. By casting GFlowNet as an amortized variational sampler for unnormalized densities, we propose a length-aware Trajectory-Balance objective that explicitly neutralizes the structural length biases inherent in autoregressive generation.