{"id":"arxiv:2510.13786","title":"The Art of Scaling Reinforcement Learning Compute for LLMs","date":"2025-10-15","section":"Post-training","quality":0.419492,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/","source_url":"https://arxiv.org/abs/2510.13786","text_available":true,"authors":["Devvrit Khatri","Lovish Madaan","Rishabh Tiwari"],"outgoing":[{"source":"arxiv:2510.13786","target":"arxiv:2001.08361","evidence":"id","id":"arxiv:2001.08361","title":"Scaling Laws for Neural Language Models","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/82302458a6d03bc051b2e4fdb8277893a81d4bd6a029969045df6a8380904eac/"},{"source":"arxiv:2510.13786","target":"arxiv:2206.04615","evidence":"id","id":"arxiv:2206.04615","title":"Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/70882486c1f8c46cb7b26cf100847f770eb7b39b69dee56b8a21141141ef814d/"},{"source":"arxiv:2510.13786","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:2510.13786","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:2510.13786","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:2510.13786","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:2510.13786","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:2511.07317","target":"arxiv:2510.13786","evidence":"id","id":"arxiv:2511.07317","title":"RLVE: Scaling Up Reinforcement Learning for Language Models with Adaptive Verifiable Environments","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/c0c18cb0ae7fe76a7239f38ce819c7ac588c1184dd27c5672f868ff4e2e26605/"},{"source":"arxiv:2512.18552","target":"arxiv:2510.13786","evidence":"title","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:2510.13786","evidence":"title","id":"arxiv:2601.20802","title":"Reinforcement Learning via Self-Distillation","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/6073ec3d17b9a932a68e79ae078928f4635e40fc79cfcbc31709715d1f1e1ffa/"},{"source":"arxiv:2604.02288","target":"arxiv:2510.13786","evidence":"id","id":"arxiv:2604.02288","title":"Unifying Group-Relative and Self-Distillation Policy Optimization via Sample Routing","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3d1160029d321f2069016f7702d4ebebee0de6cd31a8f296ee220abd6f69c99b/"},{"source":"arxiv:2604.08690","target":"arxiv:2510.13786","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":"url:thinkingmachines.ai/news/learning-to-replicate-expert-judgment-in-financial-tasks","target":"arxiv:2510.13786","evidence":"id","id":"url:thinkingmachines.ai/news/learning-to-replicate-expert-judgment-in-financial-tasks","title":"Learning to Replicate Expert Judgment in Financial Tasks (Thinking Machines)","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/c8b8e4f5970fc13702c823cc6753accda07d80b025c7699faf5e7f7a2f44a022/"}],"text_parts":[{"start":0,"end":15808,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-1.html"},{"start":15808,"end":31574,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-2.html"},{"start":31574,"end":47337,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-3.html"},{"start":47337,"end":62524,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-4.html"},{"start":62524,"end":73766,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-5.html"}],"contents":[{"title":"The Art of Scaling Reinforcement Learning Compute for LLMs","offset":0,"level":1,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-1.html#text"},{"title":"Abstract","offset":62,"level":6,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-1.html#text"},{"title":"1 Introduction","offset":1541,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-1.html#text"},{"title":"2 Preliminaries & Setup","offset":7552,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-1.html#text"},{"title":"Training Regimen","offset":8477,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-1.html#text"},{"title":"Base RL Algorithm","offset":9389,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-1.html#text"},{"title":"Controlling Generation Lengths","offset":11923,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-1.html#text"},{"title":"2.1 Predictive compute-scaling and fitting curves","offset":12480,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-1.html#text"},{"title":"Interpreting scaling curves","offset":13498,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-1.html#text"},{"title":"Scaling curve on held-out validation","offset":14101,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-1.html#text"},{"title":"3 An Empirical Study of RL Scaling","offset":14673,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-1.html#text"},{"title":"3.1 Asynchronous RL Setup","offset":15777,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-1.html#text"},{"title":"3.2 Algorithmic Choices","offset":17871,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-2.html#text"},{"title":"Loss type","offset":18348,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-2.html#text"},{"title":"FP32 Precision for LLM logits","offset":20075,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-2.html#text"},{"title":"Loss Aggregation","offset":20810,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-2.html#text"},{"title":"Advantage Normalization","offset":21350,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-2.html#text"},{"title":"Zero-Variance Filtering","offset":22281,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-2.html#text"},{"title":"Adaptive Prompt Filtering","offset":23142,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-2.html#text"},{"title":"4 ScaleRL : Scaling RL Compute Effectively & Predictably","offset":24066,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-2.html#text"},{"title":"Leave-One-Out (LOO) Ablations","offset":26170,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-2.html#text"},{"title":"Error margin in fitting scaling curves","offset":27250,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-2.html#text"},{"title":"Extrapolating Scaling Curves","offset":27755,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-2.html#text"},{"title":"Are the design choices worth it?","offset":28158,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-2.html#text"},{"title":"5 Predictable Scaling Returns Across RL Compute Axes","offset":30187,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-2.html#text"},{"title":"Model scale (MoE)","offset":30924,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-2.html#text"},{"title":"Generation length (context budget)","offset":31532,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-2.html#text"},{"title":"Global batch size (prompts)","offset":32038,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-3.html#text"},{"title":"Generations per prompt (fixed total batch)","offset":32645,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-3.html#text"},{"title":"6 Related Work","offset":33145,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-3.html#text"},{"title":"7 Discussion & Conclusion","offset":35042,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-3.html#text"},{"title":"Future work","offset":38433,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-3.html#text"},{"title":"8 Acknowledgments","offset":39471,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-3.html#text"},{"title":"Appendix A Appendix","offset":39912,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-3.html#text"},{"title":"A.1 Extended Related Work","offset":39936,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-3.html#text"},{"title":"A.2 RL for LLMs: GRPO and DAPO","offset":42892,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-3.html#text"},{"title":"Group Relative Policy Optimization (GRPO)","offset":42928,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-3.html#text"},{"title":"Decoupled Clip and Dynamic Sampling Policy Optimization (DAPO)","offset":44674,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-3.html#text"},{"title":"A.3 Training Setup","offset":46512,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-3.html#text"},{"title":"Datasets","offset":46536,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-3.html#text"},{"title":"Supervised Fine-tuning","offset":47023,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-3.html#text"},{"title":"Reinforcement Learning","offset":47307,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-3.html#text"},{"title":"A.4 What curve to fit?","offset":49177,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-4.html#text"},{"title":"A.5 Fitting scaling curves","offset":52486,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-4.html#text"},{"title":"A.6 Comparing algorithms","offset":54363,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-4.html#text"},{"title":"A.7 Robustness of fits","offset":56384,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-4.html#text"},{"title":"A.8 Interpreting Sigmoidal Curves","offset":58073,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-4.html#text"},{"title":"A.9 Forward and LOO Ablations","offset":59014,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-4.html#text"},{"title":"A.10 Controlling generation length","offset":59210,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-4.html#text"},{"title":"A.11 PipelineRL","offset":60897,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-4.html#text"},{"title":"A.12 Entropy Curves: Scaling Batch Size","offset":62243,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-4.html#text"},{"title":"A.13 Scaling on multiple axes","offset":63770,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-5.html#text"},{"title":"A.14 Downstream performance","offset":64017,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-5.html#text"},{"title":"A.15 Truncations and training instabilities","offset":64686,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-5.html#text"},{"title":"A.16 Comparing Prevalent Methods","offset":66701,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-5.html#text"},{"title":"DeepSeek (GRPO)","offset":66858,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-5.html#text"},{"title":"Qwen2.5 (DAPO)","offset":67223,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-5.html#text"},{"title":"Magistral","offset":68564,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-5.html#text"},{"title":"MiniMax","offset":68751,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-5.html#text"},{"title":"A.17 Loss Type - Stability and Robustness","offset":69041,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-5.html#text"},{"title":"A.17.1 DAPO clipping ratios","offset":69868,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-5.html#text"},{"title":"A.17.2 CISPO Clipping Ratios","offset":70632,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-5.html#text"},{"title":"A.17.3 GSPO ablations","offset":71227,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-5.html#text"},{"title":"A.17.4 GSPO vs CISPO","offset":72108,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-5.html#text"},{"title":"Instructions for reporting errors","offset":72802,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/part-5.html#text"}],"text_code_points":73766,"source_markdown":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72580bcba3d9b32b2f3e83d074f626234b26dc18126edd22a5eacef927cf87da/source.md"}