{"id":"arxiv:2509.13232","title":"Single-stream Policy Optimization","date":"2025-09-16","section":"Post-training","quality":-0.068041,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/","source_url":"https://arxiv.org/abs/2509.13232","text_available":true,"authors":["Zhongwen Xu","Zihan Ding"],"outgoing":[{"source":"arxiv:2509.13232","target":"arxiv:1511.05952","evidence":"id","id":"arxiv:1511.05952","title":"Prioritized Experience Replay","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/c88bf0c11c1817dcc457c001dfef5cd120a0946c8a009a0fe40648dae03eb7d4/"},{"source":"arxiv:2509.13232","target":"arxiv:2203.11171","evidence":"title","id":"arxiv:2203.11171","title":"Self-Consistency Improves Chain of Thought Reasoning in Language Models","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/"},{"source":"arxiv:2509.13232","target":"arxiv:2402.03300","evidence":"id","id":"arxiv:2402.03300","title":"DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/636fb871904bc06f742cfdea069796f9e9665f95bd58dc63260eaff5154ef985/"},{"source":"arxiv:2509.13232","target":"arxiv:2503.14476","evidence":"id","id":"arxiv:2503.14476","title":"DAPO: An Open-Source LLM Reinforcement Learning System at Scale","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/548e934b912ef84691ace63bf1cfd670e7b0257d9d6986dde926f569e7b2f8b7/"},{"source":"arxiv:2509.13232","target":"arxiv:2506.13585","evidence":"id","id":"arxiv:2506.13585","title":"MiniMax-M1: Scaling Test-Time Compute Efficiently with Lightning Attention","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/"},{"source":"arxiv:2509.13232","target":"arxiv:2507.18071","evidence":"id","id":"arxiv:2507.18071","title":"Group Sequence Policy Optimization","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/69430db904efc7978803673806cc9338004ffd8cf92199f3349dfbfa8a44e440/"}],"incoming":[{"source":"arxiv:2604.08690","target":"arxiv:2509.13232","evidence":"id","id":"arxiv:2604.08690","title":"Skip-Connected Policy Optimization for Implicit Advantage","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b5ec1244fd1b770994920ee69c200ecc134262e6011043284200eb18eb0d6c87/"}],"text_parts":[{"start":0,"end":14865,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-1.html"},{"start":14865,"end":29992,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-2.html"},{"start":29992,"end":45740,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-3.html"},{"start":45740,"end":51334,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-4.html"}],"contents":[{"title":"Single-stream Policy Optimization","offset":0,"level":1,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-1.html#text"},{"title":"1 Introduction","offset":1996,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-1.html#text"},{"title":"2 Related Work","offset":5709,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-1.html#text"},{"title":"3 Background","offset":8706,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-1.html#text"},{"title":"3.1 Policy Gradient and the REINFORCE Algorithm","offset":8916,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-1.html#text"},{"title":"3.2 Variance Reduction Baselines for Large Language Models","offset":11024,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-1.html#text"},{"title":"4 Method","offset":12741,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-1.html#text"},{"title":"4.1 A KL-Adaptive Value Tracker","offset":13719,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-1.html#text"},{"title":"4.2 Advantage Estimation and Policy Optimization","offset":17421,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-2.html#text"},{"title":"4.3 Prioritized Prompt Sampling","offset":20911,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-2.html#text"},{"title":"4.4 Advantages over GRPO","offset":22094,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-2.html#text"},{"title":"5 Experiments","offset":24701,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-2.html#text"},{"title":"5.1 Experimental Setup","offset":24719,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-2.html#text"},{"title":"5.2 Empirical Comparison with GRPO","offset":25941,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-2.html#text"},{"title":"5.3 Analysis of Signal Efficiency and Stability","offset":28357,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-2.html#text"},{"title":"5.4 Agentic Training Demonstrations","offset":30882,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-3.html#text"},{"title":"6 Conclusions","offset":33063,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-3.html#text"},{"title":"Appendix A SPO Initialization","offset":34379,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-3.html#text"},{"title":"Appendix B Batch Extensions","offset":35198,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-3.html#text"},{"title":"Appendix C Comparisons against GRPO","offset":36396,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-3.html#text"},{"title":"C.1 Inefficiency of Dynamic Sampling","offset":36436,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-3.html#text"},{"title":"C.2 Variance Reduction for Policy Gradient","offset":38103,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-3.html#text"},{"title":"Appendix D Training and Evaluation Details","offset":42671,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-3.html#text"},{"title":"Appendix E Ablation Studies","offset":46315,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-4.html#text"},{"title":"Instructions for reporting errors","offset":50370,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/part-4.html#text"}],"text_code_points":51334,"source_markdown":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/source.md"}