BRACE introduces an anchored Bellman‑residual correction to address stale critic bias in asynchronous reinforcement learning for language models. By limiting the correction horizon to a prefix of policy tokens and adding a constant‑weight Monte‑Carlo tail, it separates policy correction from reward propagation. The method improves mean@1 on BrowseComp‑Plus by 2.4% and runs 2.46× faster per step than synchronous training while staying stable 50 updates off‑policy.
By Guanqun Zhao, Zijun Xie, Binbin Zheng, Jiafeng Lu, Enlei Gong, Zeyu Chen
arXiv:2503. 03660v4 Announce Type: replace Abstract: We introduce a sequence-conditioned critic for Soft Actor-Critic (SAC) that models trajectory context with a lightweight Transformer and trains on aggregated $N$-step targets.
By Dong Tian, Onur Celik, Gerhard Neumann
The paper introduces a method called score centering to address the training‑inference mismatch (TIM) that destabilizes reinforcement learning for large language models. By adding an additive correction term that cancels drift between training and inference engines, score centering stabilizes RL and can match or surpass importance‑sampling techniques, especially as model size and mismatch severity increase. The approach also composes with importance sampling, yielding further performance gains in staleness experiments.
By Martin Marek, Max Ryabinin
arXiv:2609.37119v1 Announce Type: cross
Abstract: Recent approaches to reinforcement learning (RL) post-training for large language models increasingly remove the critic to reduce training instabilit...
By Hongyang Li, Xiao Li, Caesar Wu, Said Mammar, Gr\'egoire Danoy, Pascal Bouvry
arXiv:2606. 03070v1 Announce Type: cross Abstract: Asynchronous reinforcement learning can improve language-model post-training throughput by decoupling response generation from policy optimization, but stale responses introduce distribution drift.
By Zehua Liu, Yuxuan Yao, Xiaojin Fu, Tao Zhong, Mingxuan Yuan
arXiv:2606. 29526v1 Announce Type: new Abstract: Reinforcement learning (RL) has gained growing attention in large language model (LLM) post-training, yet RL training remains fragile and can suffer from instability or collapse.
By Jing Liang, Hongyao Tang, Yi Ma, Yancheng He, Weixun Wang, Xiaoyang Li, Ju Huang, Wenbo Su, Jinyi Liu, Yan Zheng, Jianye Hao, Bo Zheng
Reinforcement learning (RL) post-training for large language models (LLMs) follows a efficient paradigm of "rollout then update", which inevitably results in off-policy training data. To resolve this, Importance sampling (IS) is proposed, while the token-level ratios compound over long sequences, causing severe variance exploded.
arXiv:2602. 04879v3 Announce Type: replace-cross Abstract: Reinforcement learning (RL) has become a cornerstone for fine-tuning Large Language Models (LLMs), with Proximal Policy Optimization (PPO) serving as the de facto standard algorithm.
By Penghui Qi, Xiangxin Zhou, Zichen Liu, Tianyu Pang, Chao Du, Min Lin, Wee Sun Lee
The paper identifies a failure mode in Proximal Policy Optimization (PPO) critics called Value Flattening, where state values vary sharply across intermediate states while critic predictions remain flat. The authors analyze this phenomenon theoretically and empirically, linking it to an implicit variance penalty and redundant updates from temporally correlated states. They propose Sparse Proximal Policy Optimization (SP³O), which applies the value loss to only a few well‑separated states, and demonstrate that this sparse supervision mitigates Value Flattening and improves policy performance on Qwen3‑Base across model sizes and evaluation suites.
By Yizhuo Li, Jianhao Yan, Yun Luo, Zhi Wang, Futing Wang, Rong-Xi Tan, Kanghui Tian, Ganqu Cui, Ning Ding, Peilin Zhao, Yafu Li, Yu Cheng
arXiv:2605. 05481v2 Announce Type: replace Abstract: We revisit a classic "chicken-and-egg" problem in reinforcement learning: to safely improve a policy, the value function must be accurate on the state-visitation distribution of the updated policy.
By Dillon Sandhu, Ronald Parr
The paper introduces Actor‑Critic with Action Chunking (AC2), a method that assigns credit to short action chunks instead of entire trajectories, enabling policy updates without waiting for terminal rewards. AC2 employs local readiness, reference solutions, and 10k‑token chunks to make critic‑based credit assignment reliable. Experiments on Qwen3‑4B with FineProofs‑RL show AC2 surpasses GRPO’s peak validation score while using 2.5× fewer decoding FLOPs and fewer training steps.
By Kaiyue Wen, Luke Bailey, Arvind Mahankali, Tengyu Ma
arXiv:2604. 00860v3 Announce Type: replace Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) has become a central post-training paradigm for improving the reasoning capabilities of large language models.
By Huaiyang Wang, Xiaojie Li, Deqing Wang, Haoyi Zhou, Zixuan Huang, Yaodong Yang, Jianxin Li, Yikun Ban