{"id":"arxiv:2506.13585","title":"MiniMax-M1: Scaling Test-Time Compute Efficiently with Lightning Attention","date":"2025-06-16","section":"Post-training","quality":0.353911,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/","source_url":"https://arxiv.org/abs/2506.13585","text_available":true,"authors":[" MiniMax"," :","Aili Chen","Aonian Li"],"outgoing":[{"source":"arxiv:2506.13585","target":"arxiv:2006.16236","evidence":"title","id":"arxiv:2006.16236","title":"Transformers are RNNs: Fast Autoregressive Transformers with Linear Attention","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d119844a2dce8a3b156558743b90add84774d00894634b6478d03abb4f5216f9/"},{"source":"arxiv:2506.13585","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:2506.13585","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:2506.13585","target":"arxiv:2503.20783","evidence":"id","id":"arxiv:2503.20783","title":"Understanding R1-Zero-Like Training: A Critical Perspective","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b49a056f3ce8e269dd65e3b11605927f3c4d0ef074bc68e0b1d7f84e6c9c6c6a/"},{"source":"arxiv:2506.13585","target":"arxiv:2506.01939","evidence":"id","id":"arxiv:2506.01939","title":"Beyond the 80/20 Rule: High-Entropy Minority Tokens Drive Effective Reinforcement Learning for LLM Reasoning","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/e0f4a55bca612bce5895b6b98f89b0efc14cbd8cf45946a7743347cf6daee1a1/"}],"incoming":[{"source":"arxiv:2507.18071","target":"arxiv:2506.13585","evidence":"id","id":"arxiv:2507.18071","title":"Group Sequence Policy Optimization","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/69430db904efc7978803673806cc9338004ffd8cf92199f3349dfbfa8a44e440/"},{"source":"arxiv:2507.18074","target":"arxiv:2506.13585","evidence":"id","id":"arxiv:2507.18074","title":"AlphaGo Moment for Model Architecture Discovery","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1c22da54910a26f1f81b276f5cae3a81e22f7db024f7d1a5fa89ccd4f291ffa8/"},{"source":"arxiv:2508.07629","target":"arxiv:2506.13585","evidence":"title","id":"arxiv:2508.07629","title":"Klear-Reasoner: Advancing Reasoning Capability via Gradient-Preserving Clipping Policy Optimization","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/a232533e4e78ad28d2ffe9fafe799b66cf626b18ad523d0088d9cc12f1ba1b76/"},{"source":"arxiv:2508.11408","target":"arxiv:2506.13585","evidence":"id","id":"arxiv:2508.11408","title":"On-Policy RL Meets Off-Policy Experts: Harmonizing Supervised Fine-Tuning and Reinforcement Learning via Dynamic Weighting","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/48594da2e3f2cdfeb17195dd033d047235fc4b4ffb65030800f64c6faedc4fbe/"},{"source":"arxiv:2509.13232","target":"arxiv:2506.13585","evidence":"id","id":"arxiv:2509.13232","title":"Single-stream Policy Optimization","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/"},{"source":"arxiv:2510.13786","target":"arxiv:2506.13585","evidence":"id","id":"arxiv:2510.13786","title":"The Art of Scaling Reinforcement Learning Compute for LLMs","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/"},{"source":"arxiv:2512.00499","target":"arxiv:2506.13585","evidence":"id","id":"arxiv:2512.00499","title":"ESPO: Entropy Importance Sampling Policy Optimization","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/342b84df2bf40f64c57ea7839579c1f337374ad9c28b16ce7d477c42e24c67b4/"},{"source":"arxiv:2512.18552","target":"arxiv:2506.13585","evidence":"id","id":"arxiv:2512.18552","title":"Toward Training Superintelligent Software Agents through Self-Play SWE-RL","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/da99b397e01b725186243c38a2dc6072d3a5c81bd8f759e898f04ad09aa0f4b1/"},{"source":"arxiv:2601.20802","target":"arxiv:2506.13585","evidence":"id","id":"arxiv:2601.20802","title":"Reinforcement Learning via Self-Distillation","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/6073ec3d17b9a932a68e79ae078928f4635e40fc79cfcbc31709715d1f1e1ffa/"},{"source":"arxiv:2604.08690","target":"arxiv:2506.13585","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/"},{"source":"arxiv:2605.12969","target":"arxiv:2506.13585","evidence":"id","id":"arxiv:2605.12969","title":"Revisiting Reinforcement Learning with Verifiable Rewards from a Contrastive Perspective","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/30e87f5eb9beda8e09b3945d6ba1707dec761abed83c49d129a21adaed02ce45/"},{"source":"url:cameronrwolfe.substack.com/p/grpo-tricks","target":"arxiv:2506.13585","evidence":"id","id":"url:cameronrwolfe.substack.com/p/grpo-tricks","title":"GRPO++: Tricks for Making RL Actually Work (Cameron Wolfe)","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d170bc3da5b8e6a055fd511719fcc3764635236bac638947bd0a018412245eb5/"}],"text_parts":[{"start":0,"end":15308,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/part-1.html"},{"start":15308,"end":31202,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/part-2.html"},{"start":31202,"end":47031,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/part-3.html"},{"start":47031,"end":53092,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/part-4.html"}],"contents":[{"title":"MiniMax-M1: Scaling Test-Time Compute Efficiently with Lightning Attention","offset":15,"level":1,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/part-1.html#text"},{"title":"Abstract","offset":93,"level":6,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/part-1.html#text"},{"title":"1 Introduction","offset":2290,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/part-1.html#text"},{"title":"2 Preparation for Scalable RL: Continual Pretraining and SFT","offset":10795,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/part-1.html#text"},{"title":"2.1 Continual Pre-Training: Foundation for RL Scaling","offset":11313,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/part-1.html#text"},{"title":"2.2 Supervised Fine-Tuning: Focused Alignment for Efficient RL","offset":13565,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/part-1.html#text"},{"title":"3 Efficient RL Scaling: Algorithms and Lightning Attention","offset":14130,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/part-1.html#text"},{"title":"3.1 Efficient RL Scaling with CISPO","offset":15267,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/part-1.html#text"},{"title":"3.2 Efficient RL Scaling with Lightning Attention – Challenges and Recipes","offset":23741,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/part-2.html#text"},{"title":"4 Scaling Reinforcement Learning with Diverse Data","offset":26920,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/part-2.html#text"},{"title":"4.1 Reasoning-Intensive Tasks with Rule-based Verification","offset":27335,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/part-2.html#text"},{"title":"4.2 General Domain Tasks with Model-based Feedbacks","offset":32774,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/part-3.html#text"},{"title":"4.2.1 Data and Reward Models","offset":33023,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/part-3.html#text"},{"title":"4.2.2 Addressing Bias of Generative Reward Models for Long CoT","offset":36265,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/part-3.html#text"},{"title":"4.3 Curriculum of Incorporating Diverse Data","offset":38276,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/part-3.html#text"},{"title":"5 Extending RL Scaling to Longer Thinking","offset":39437,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/part-3.html#text"},{"title":"6 Evaluations","offset":42612,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/part-3.html#text"},{"title":"6.1 Core Benchmarks","offset":42630,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/part-3.html#text"},{"title":"6.2 Effect of RL Scaling","offset":47940,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/part-4.html#text"},{"title":"7 Conclusion and Future work","offset":48644,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/part-4.html#text"},{"title":"Appendix A Contributors","offset":50528,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/part-4.html#text"},{"title":"Instructions for reporting errors","offset":52128,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/part-4.html#text"}],"text_code_points":53092,"source_markdown":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5beb2ec6c7e6362f9d957219d7a205dd1adcc53f739acc0c7fd5f1f2bded4d83/source.md"}