4 papers
Process Advantage Signal Shaping: A Paradigm-Agnostic Middleware for Process-Supervised RL in LLM Reasoners
Chao Wang, Hongtao Tian, Tao Yang +3
Group Relative Policy Optimization (GRPO) is a default recipe for process-supervised reinforcement learning of LLM reasoners, and dense process supervision -- via learned process r…
Calibrated Speculative Decoding: Frequency-Guided Candidate Selection for Efficient Inference
Xuwen Zhou, Fangxin Liu, Chao Wang +5
Speculative decoding accelerates autoregressive generation by letting draft tokens bypass full verification, but conventional frameworks suffer from frequent false rejections, part…
From Faithfulness to Correctness: Generative Reward Models that Think Critically
Qiyao Ma, Yunsheng Shi, Hongtao Tian +3
Through reinforcement learning with verifiable rewards (RLVR), large language models have achieved substantial progress in domains with easily verifiable outcomes, such as mathemat…
Learning More with Less: A Dynamic Dual-Level Down-Sampling Framework for Efficient Policy Optimization
Chao Wang, Tao Yang, Hongtao Tian +5
Critic-free methods like GRPO reduce memory demands by estimating advantages from multiple rollouts but tend to converge slowly, as critical learning signals are diluted by an abun…