English
Related papers

Related papers: Training with Harnesses: On-Policy Harness Self-Di…

200 papers

Reasoning-oriented language models achieve strong performance by generating long chain-of-thought traces at inference time. However, this capability comes with substantial and often excessive computational cost, which can materialize in…

Machine Learning · Computer Science 2026-03-17 Gianluigi Silvestri , Edoardo Cetin

On-policy knowledge distillation (OPD) trains a student on its own rollouts under token-level supervision from a teacher. Not all token positions matter equally, but existing views of token importance are incomplete. We ask a direct…

Machine Learning · Computer Science 2026-05-22 Yuanda Xu , Hejian Sang , Zhengze Zhou , Ran He , Zhipeng Wang , Alborz Geramifard

On-policy distillation (OPD) is an increasingly important paradigm for post-training language models. However, we identify a pervasive Scaling Law of Miscalibration: while OPD effectively improves task accuracy, it systematically traps…

Machine Learning · Computer Science 2026-04-21 Jiaxin Zhang , Xiangyu Peng , Qinglin Chen , Qinyuan Ye , Caiming Xiong , Chien-Sheng Wu

Self-distillation bootstraps large language models (LLMs) by training on their own generations. However, existing methods either rely on external signals to curate self-generated outputs (e.g., correctness filtering, execution feedback, and…

Computation and Language · Computer Science 2026-05-22 Guangya Hao , Yitong Shang , Yunbo Long , Zhuokai Zhao , Hanxue Liang

Large language models are often post-trained with sparse verifier rewards, which indicate whether a sampled trajectory succeeds but provide limited guidance about where reasoning succeeds or fails. On-policy distillation (OPD) offers denser…

Machine Learning · Computer Science 2026-05-14 Weichen Yu , Xiaomin Li , Yizhou Zhao , Xiaoze Liu , Ruowang Zhang , Haixin Wang , Yinyi Luo , Chen Henry Wu , Gaurav Mittal , Matt Fredrikson , Yu Hu

On-policy distillation offers dense, per-token supervision for training reasoning models; however, it remains unclear under which conditions this signal is beneficial and under which it is detrimental. Which teacher model should be used,…

Self-distillation has emerged as an effective post-training paradigm for LLMs, often improving performance while shortening reasoning traces. However, in mathematical reasoning, we find that it can reduce response length while degrading…

Computation and Language · Computer Science 2026-05-21 Jeonghye Kim , Xufang Luo , Minbeom Kim , Sangmook Lee , Dohyung Kim , Jiwon Jeon , Dongsheng Li , Yuqing Yang

On-policy distillation (OPD) is a powerful paradigm for model alignment, yet its reliance on teacher logits restricts its application to white-box scenarios. We contend that structured semantic rubrics can serve as a scalable alternative to…

Machine Learning · Computer Science 2026-05-11 Junfeng Fang , Zhepei Hong , Mao Zheng , Mingyang Song , Gengsheng Li , Houcheng Jiang , Dan Zhang , Haiyun Guo , Xiang Wang , Tat-Seng Chua

Reinforcement learning for large language models faces a fundamental trade-off between sample efficiency and asymptotic performance: strictly on-policy methods discard trajectories after a single update, while off-policy reuse introduces…

Machine Learning · Computer Science 2026-05-26 Changyu Chen , Xiting Wang , Rui Yan

Harness engineering has emerged as an important inference-time technique for large language model (LLM) agents, aiming to improve long-term performance through task decomposition and guided execution. However, more elaborate harnesses are…

Machine Learning · Computer Science 2026-05-22 Boyuan Wang , Bochao Li , Minghan Wang , Yuxin Tao , Fang Kong

Reinforcement learning (RL) has been widely used to train LLM agents for multi-turn interactive tasks, but its sample efficiency is severely limited by sparse rewards and long horizons. On-policy self-distillation (OPSD) alleviates this by…

Machine Learning · Computer Science 2026-04-14 Hao Wang , Guozhi Wang , Han Xiao , Yufeng Zhou , Yue Pan , Jichao Wang , Ke Xu , Yafei Wen , Xiaohu Ruan , Xiaoxin Chen , Honggang Qi

On-policy reinforcement learning has become the dominant paradigm for reasoning alignment in large language models, yet its sparse, outcome-level rewards make token-level credit assignment notoriously difficult. On-Policy Distillation (OPD)…

Machine Learning · Computer Science 2026-04-14 Binbin Zheng , Xing Ma , Yiheng Liang , Jingqing Ruan , Xiaoliang Fu , Kepeng Lin , Benchang Zhu , Ke Zeng , Xunliang Cai

In natural language processing (NLP) tasks, slow inference speed and huge footprints in GPU usage remain the bottleneck of applying pre-trained deep models in production. As a popular method for model compression, knowledge distillation…

Computation and Language · Computer Science 2020-12-15 Fei Yuan , Linjun Shou , Jian Pei , Wutao Lin , Ming Gong , Yan Fu , Daxin Jiang

Chain-of-Thought (CoT) prompting often improves classification accuracy, but it introduces a significant throughput penalty with rationale generation (Wei et al., 2022; Cheng and Van Durme, 2024). To resolve this trade-off, we introduce…

Computation and Language · Computer Science 2025-09-30 Jillian Xu , Dylan Zhou , Vinay Shukla , Yang Yang , Junrui Ruan , Shuhuai Lin , Wenfei Zou , Yinxiao Liu , Karthik Lakshmanan

On-policy distillation (OPD) has become a promising paradigm for reasoning-oriented post-training of large language models (LLMs), especially when combined with reinforcement learning from verifiable rewards (RLVR). Existing OPD methods…

On-policy distillation (OPD) trains a student on its own trajectories under token-level teacher supervision, but existing methods are capped by a single-teacher capability ceiling: when the teacher errs, the student inherits the error. OPD…

Computation and Language · Computer Science 2026-05-05 Jianze Wang , Ying Liu , Jinlong Chen , Xuchun Hu , Qilong Zhang , Yu Cao , Jun Wang , Hua Yang , Yong Xie , Qianglong Chen

Knowledge Distillation (KD) can transfer the reasoning abilities of large models to smaller ones, which can reduce the costs to generate Chain-of-Thoughts for reasoning tasks. KD methods typically ask the student to mimic the teacher's…

Computation and Language · Computer Science 2026-03-17 Minsang Kim , Seung Jun Baek

On-policy self-distillation (SD) improves LLM reasoning by using teacher-side privileged information (PI) to turn sparse verifier outcomes into dense token-level supervision. Existing methods usually assume trusted PI, such as reference…

Computation and Language · Computer Science 2026-05-28 Jiazhen Huang , Xiao Chen , Xiao Luo , Yong Dai , Senkang Hu , Yuzhi Zhao

On-policy distillation (OPD) trains a student model on its own rollouts using dense feedback from a stronger teacher. Prior literature suggests that, provided teacher feedback is available, supervising the full sequence of response tokens…

Computation and Language · Computer Science 2026-05-26 Kaiyuan Liu , Ziyuan Zhuang , Yang Bai , Bing Wang , Rongxiang Weng , Jieping Ye

On-Policy Distillation (OPD) has emerged as a dominant post-training paradigm for large language models, especially for reasoning domains. However, OPD remains unstable in practice due to the high gradient variance of its single-sample…

Machine Learning · Computer Science 2026-05-11 Minjae Oh , Sangjun Song , Gyubin Choi , Yunho Choi , Yohan Jo