{"id":"arxiv:2407.21783","title":"The Llama 3 Herd of Models","date":"2024-07-31","section":"LLMs: architectures, context, training","quality":0.075302,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/","source_url":"https://arxiv.org/abs/2407.21783","text_available":true,"authors":["Aaron Grattafiori","Abhimanyu Dubey","Abhinav Jauhri","Abhinav Pandey"],"outgoing":[{"source":"arxiv:2407.21783","target":"acl:2022.emnlp-main.340","evidence":"title","id":"acl:2022.emnlp-main.340","title":"Super-NaturalInstructions: Generalization via Declarative Instructions on 1600+ NLP Tasks","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/e43a06571b7e8d89abd5f20f8f520c8bc3d3eb32ffb156cb7fc9e0469fc14b0a/"},{"source":"arxiv:2407.21783","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:2407.21783","target":"arxiv:2303.17651","evidence":"title","id":"arxiv:2303.17651","title":"Self-Refine: Iterative Refinement with Self-Feedback","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/"},{"source":"arxiv:2407.21783","target":"arxiv:2311.06242","evidence":"title","id":"arxiv:2311.06242","title":"Florence-2: Advancing a Unified Representation for a Variety of Vision Tasks","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/edc2679fff98a10c70395e79c58b56eaf9ec0001691dbca2b33ba94a401a8398/"},{"source":"arxiv:2407.21783","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:2407.21783","target":"arxiv:2402.16822","evidence":"id","id":"arxiv:2402.16822","title":"Rainbow Teaming: Open-Ended Generation of Diverse Adversarial Prompts","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d4803e891a4b9c929eca007e3c3f003c330d2fc741109975c6ddd7d2c32828fc/"},{"source":"arxiv:2407.21783","target":"url:ai.meta.com/research/publications/code-llama-open-foundation-models-for-code","evidence":"title","id":"url:ai.meta.com/research/publications/code-llama-open-foundation-models-for-code","title":"Code Llama: Open Foundation Models for Code","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/0dc2d6030e514e45150f64f511c0f50fc010fdc1ddae0bdca7edcd189b084d41/"},{"source":"arxiv:2407.21783","target":"url:rlhfbook.com","evidence":"title","id":"url:rlhfbook.com","title":"Reinforcement Learning from Human Feedback","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/20e354559fad09ad4d6338d104c9c3f822a48e7fbf86cd8afa1ddf951b223658/"}],"incoming":[{"source":"arxiv:2405.18392","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2405.18392","title":"Scaling Laws and Compute-Optimal Training Beyond Fixed Training Durations","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/f70e803c548e8a0c524f42ec6203808e4b633b9cc2fb45f4a075af70150f41d1/"},{"source":"arxiv:2409.03137","target":"arxiv:2407.21783","evidence":"title","id":"arxiv:2409.03137","title":"The AdEMAMix Optimizer: Better, Faster, Older","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4c4e435f36a4ffe28fc019c45f447a6825e58858ce5616b9b4bcd9e9cd245233/"},{"source":"arxiv:2410.04444","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2410.04444","title":"Gödel Agent: A Self-Referential Agent Framework for Recursive Self-Improvement","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d04ae5d37f10547be79c728efed48a629e631ab5384155564e52558e12ff4362/"},{"source":"arxiv:2410.07041","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2410.07041","title":"Emergent properties with repeated examples","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/63a5d833b43fae5ed1f9f44f26b90813fa8a37dd44870e8e409d7a7c34fabfec/"},{"source":"arxiv:2410.21272","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2410.21272","title":"Arithmetic Without Algorithms: Language Models Solve Math With a Bag of Heuristics","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/397b826ba289bd4f14f457562942cb092c38550a4c49451e95ffcc7b376c8cb0/"},{"source":"arxiv:2412.06769","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2412.06769","title":"Training Large Language Models to Reason in a Continuous Latent Space","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68c71dc1a042dedd5e3c4d9c1ad32c9d046fd31b8dc79080bd6dff1c9800abc7/"},{"source":"arxiv:2412.13663","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2412.13663","title":"Smarter, Better, Faster, Longer: A Modern Bidirectional Encoder for Fast, Memory Efficient, and Long Context Finetuning and Inference","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/ae6c000a40b751c4486e08fd61f5f7788d8bb57f0838e234e27a5a67f894e923/"},{"source":"arxiv:2412.14135","target":"arxiv:2407.21783","evidence":"title","id":"arxiv:2412.14135","title":"Scaling of Search and Learning: A Roadmap to Reproduce o1 from Reinforcement Learning Perspective","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5a0b4b0b524698ee5f30eab659b7cd5104f5fe6497c8dec2d0c4b9d332350047/"},{"source":"arxiv:2501.00656","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2501.00656","title":"2 OLMo 2 Furious","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1ca88dba5d5c59e8cb23b2a304e74370b8ec69077b714a30d426300f7ef2a0b9/"},{"source":"arxiv:2501.04519","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2501.04519","title":"rStar-Math: Small LLMs Can Master Math Reasoning with Self-Evolved Deep Thinking","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/473ad7b5a354658a1d241813741e15a8ab51e7c2ea34f0c3d22eb06884864a46/"},{"source":"arxiv:2501.14082","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2501.14082","title":"Communicating Activations Between Language Model Agents","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/e67d4addcda44093e242074332925d000726f87bd9538f6b318bdfc454a7dc13/"},{"source":"arxiv:2501.18837","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2501.18837","title":"Constitutional Classifiers: Defending against Universal Jailbreaks across Thousands of Hours of Red Teaming","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/c73837c3edb5465ee3f7cc7c6e16d23e16bdbdef5e1448aeaad2135dfd87de07/"},{"source":"arxiv:2501.19393","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2501.19393","title":"s1: Simple test-time scaling","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/2a88818f7605b71cd0160831faebd51b3274c24ffd29626b4d519390b4bffffe/"},{"source":"arxiv:2502.00873","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2502.00873","title":"Language Models Use Trigonometry to Do Addition","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/7766b2640c6b5ab53a295e8d60920cb2419bfbbd812810d9b35d4b7c323922ed/"},{"source":"arxiv:2502.03387","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2502.03387","title":"LIMO: Less is More for Reasoning","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/714392528bb842fb3a61cee5367942f697536f0131b0fcac85c6eb8d2212125c/"},{"source":"arxiv:2502.03407","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2502.03407","title":"Detecting Strategic Deception Using Linear Probes","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/8297cbfb305547d6946bba3998d8921b9578398ed371961d3f0eeadd871e4a49/"},{"source":"arxiv:2502.09992","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2502.09992","title":"Large Language Diffusion Models","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/a4863ed6f639560bd2feb6c66c13d1e3ac4cf42f8919a65c9818625cc6b6cde9/"},{"source":"arxiv:2502.19981","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2502.19981","title":"The Lookahead Limitation: Why Multi-Operand Addition is Hard for LLMs","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/fc024e63f0774183f41cf84f17bb5a4238ff43a8ec18a6f8afb9a1abcd113488/"},{"source":"arxiv:2503.02875","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2503.02875","title":"The First Few Tokens Are All You Need: An Efficient and Effective Unsupervised Prefix Fine-Tuning Method for Reasoning Models","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b9349378a2ea40a7998ba364a4587c40dff3cc9b22750e14b3b21e296a4ecb12/"},{"source":"arxiv:2503.14858","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2503.14858","title":"1000 Layer Networks for Self-Supervised RL: Scaling Depth Can Enable New Goal-Reaching Capabilities","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/dfa04045239215a4a49544cd31c178090abc3a587e98bd9af08e0ea96d658911/"},{"source":"arxiv:2503.20783","target":"arxiv:2407.21783","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:2504.06225","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2504.06225","title":"Encoder-Decoder Gemma: Improving the Quality-Efficiency Trade-Off via Adaptation","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/797af53c1331a8303ff731ef2763d562b29b9864407468ebbd7c1a89ab7c0eed/"},{"source":"arxiv:2504.13837","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2504.13837","title":"Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b90f70269aeb02311d74ffaadf47e139683424a47118dc18619000ba72174d17/"},{"source":"arxiv:2504.16084","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2504.16084","title":"TTRL: Test-Time Reinforcement Learning","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/ad54dd58d24afdfefc8aa5087d07c4aab2143e90ef06f6f7db8d3f3333452c48/"},{"source":"arxiv:2504.20571","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2504.20571","title":"Reinforcement Learning for Reasoning in Large Language Models with One Training Example","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b2b21c3df49619588c371411dba9467a41f81b3fd4da5441898b5145bb0de120/"},{"source":"arxiv:2505.13763","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2505.13763","title":"Language Models Are Capable of Metacognitive Monitoring and Control of Their Internal Activations","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/95c4e7c49af162594458624b663c82546ad42167854a2235442954821ea69c65/"},{"source":"arxiv:2505.15134","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2505.15134","title":"The Unreasonable Effectiveness of Entropy Minimization in LLM Reasoning","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/12b01b403e280a32d41fd991a27fa30f33315e09f0f42a4f43e9bece5ad9c53d/"},{"source":"arxiv:2505.17117","target":"arxiv:2407.21783","evidence":"title","id":"arxiv:2505.17117","title":"From Tokens to Thoughts: How LLMs and Humans Trade Compression for Meaning","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/e0837eb0f0d0e479142b8ae6d2ef144168e5e16c610ad22479076c9d606be084/"},{"source":"arxiv:2505.21444","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2505.21444","title":"Can Large Reasoning Models Self-Train?","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/206224ca50098e0bdc230140722feec107cfc90c1cef2ed8d3ed52355ef56bca/"},{"source":"arxiv:2505.21493","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2505.21493","title":"Reinforcing General Reasoning without Verifiers","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/06d2ae06e1a3bcda5db4695ecca7ec78df67002bddc755936e262799f3e15d15/"},{"source":"arxiv:2505.24832","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2505.24832","title":"How much do language models memorize?","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1d14faeb604373b83ad90ae653998a11cc09492020341f29419e0deed45a300c/"},{"source":"arxiv:2506.01939","target":"arxiv:2407.21783","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/"},{"source":"arxiv:2506.03106","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2506.03106","title":"Critique-GRPO: Advancing LLM Reasoning with Natural Language and Numerical Feedback","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b16b5772846ee30548a2d2c9cce1833ca3af6beb01c4a58f2dd5a69d809d55d4/"},{"source":"arxiv:2506.06632","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2506.06632","title":"Curriculum Reinforcement Learning from Easy to Hard Tasks Improves LLM Reasoning","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/79fda9609f9b88bb71f71f531b63ea6b5237d1c889b7ae59e9ca07c12dbe70d1/"},{"source":"arxiv:2506.10947","target":"arxiv:2407.21783","evidence":"title","id":"arxiv:2506.10947","title":"Spurious Rewards: Rethinking Training Signals in RLVR","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/f48e13a9656d8eff1882eb80988e07a820c4799401298947deac735c004e8592/"},{"source":"arxiv:2507.10524","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2507.10524","title":"Mixture-of-Recursions: Learning Dynamic Recursive Depths for Adaptive Token-Level Computation","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/7b25702c35f07446cea97d6a98afc489c8fdda54560c05fbeb1b3b139a9da622/"},{"source":"arxiv:2508.02513","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2508.02513","title":"Modular Arithmetic: Language Models Solve Math Digit by Digit","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/83aede757707b76dc76fc15431a436e5dec05563a9f2f28ebcc650775c878bcc/"},{"source":"arxiv:2508.05629","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2508.05629","title":"On the Generalization of SFT: A Reinforcement Learning Perspective with Reward Rectification","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/0a17fccfeffa8a04070bb16903a7217d8adee804cf42e0cfc8dd128d3390474f/"},{"source":"arxiv:2508.11408","target":"arxiv:2407.21783","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:2509.03646","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2509.03646","title":"Emergent Hierarchical Reasoning in LLMs through Reinforcement Learning","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/683657af0d699b7d854ce84d95bcd53c12e152aa738c602d0982cc0d11009f08/"},{"source":"arxiv:2509.14234","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2509.14234","title":"Compute as Teacher: Turning Inference Compute Into Reference-Free Supervision","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/a6f1443a7e55dd6421fdbe43e65e19c1962a19b135d737fd67f414d3324ce85e/"},{"source":"arxiv:2509.25123","target":"arxiv:2407.21783","evidence":"title","id":"arxiv:2509.25123","title":"From $f(x)$ and $g(x)$ to $f(g(x))$: LLMs Learn New Skills in RL by Composing Old Ones","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/7fdfe7abe26ee324fe7abad3ce802148163129a7c157d77bdec859c990868e03/"},{"source":"arxiv:2510.03215","target":"arxiv:2407.21783","evidence":"title","id":"arxiv:2510.03215","title":"Cache-to-Cache: Direct Semantic Communication Between Large Language Models","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b5a2a7fccb7e223ec3c3769c18189f6a6e4c18e2b074567db94cd7db2450723b/"},{"source":"arxiv:2510.05491","target":"arxiv:2407.21783","evidence":"title","id":"arxiv:2510.05491","title":"NorMuon: Making Muon more efficient and scalable","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/c9ea441f9c2808077269690a88c1b1407189ac1cb03fbc0df89979a4d1d41b98/"},{"source":"arxiv:2510.11686","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2510.11686","title":"Representation-Based Exploration for Language Models: From Test-Time to Post-Training","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/89b1f9c1bcd283b28d316cb6cddafdd9361be5cd510b2be2a1e761c20a0fd62e/"},{"source":"arxiv:2510.27688","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2510.27688","title":"Continuous Autoregressive Language Models","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/e954d29144273a3d9e4c1a5ac4fc43a163b8701659926e2bfa94011507f51414/"},{"source":"arxiv:2511.09149","target":"arxiv:2407.21783","evidence":"title","id":"arxiv:2511.09149","title":"Enabling Agents to Communicate Entirely in Latent Space","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02979751e22005b3284e4d7caa19baae335ca63c9ec46e27d833e95d088e9688/"},{"source":"arxiv:2512.02807","target":"arxiv:2407.21783","evidence":"title","id":"arxiv:2512.02807","title":"SR-GRPO: Stable Rank as an Intrinsic Geometric Reward for Large Language Model Alignment","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/ebb3b04baa11b5d0365ca4445da1c2f833108b680ea65272710ed66a57751c0e/"},{"source":"arxiv:2512.13961","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2512.13961","title":"Olmo 3","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/8daa11a9729da9879a7e95c15d4eef3bab8579d2885a14102ed09db111221802/"},{"source":"arxiv:2512.15745","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2512.15745","title":"LLaDA2.0: Scaling Up Diffusion Language Models to 100B","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/90a25474fdf076c254ea3ddbbc7ddbc1f2f49fb98065168d66b5ad9d8f7dafaa/"},{"source":"arxiv:2512.24695","target":"arxiv:2407.21783","evidence":"title","id":"arxiv:2512.24695","title":"Nested Learning: The Illusion of Deep Learning Architectures","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d19bf2367894704a701d0741e7db6d8b148719af1d634b55f5797aab5f617d06/"},{"source":"arxiv:2601.05053","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2601.05053","title":"Reinforced Efficient Reasoning via Semantically Diverse Exploration","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/74918d32c56962f9c5b2063026f5e2243a84e20c284eb3e37c882bd4681d1a11/"},{"source":"arxiv:2601.11061","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2601.11061","title":"Spurious Rewards Paradox: Mechanistically Understanding How RLVR Activates Memorization Shortcuts in LLMs","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/96002027cf4c274573f7320398481a0cd73dbf72f18d7f37c783605f5158ea7b/"},{"source":"arxiv:2601.20861","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2601.20861","title":"Evolutionary Strategies lead to Catastrophic Forgetting in LLMs","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4178758b0cb3ebc95f6a3cbc6423e8ea05c0ae649e11a2f4059a84b3faa6cde0/"},{"source":"arxiv:2601.21343","target":"arxiv:2407.21783","evidence":"title","id":"arxiv:2601.21343","title":"Self-Improving Pretraining: using post-trained models to pretrain better models","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/43ce1b189d6b21d445c0ce6d4d04b4352d268beb4346da43be19982402befde8/"},{"source":"arxiv:2601.21557","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2601.21557","title":"Meta Context Engineering via Agentic Skill Evolution","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/"},{"source":"arxiv:2604.08302","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2604.08302","title":"DMax: Aggressive Parallel Decoding for dLLMs","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/22790263bd218aac28626fe3dab2fcd0ed73ceef94aa8a948bef2cb8bc007f7a/"},{"source":"arxiv:2604.11962","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2604.11962","title":"The Linear Centroids Hypothesis: Features as Directions Learned by Local Experts","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/8428fd0cce9753855f4d0f801609b0205a0be52a459d53ae44e7860d00348485/"},{"source":"arxiv:2605.12466","target":"arxiv:2407.21783","evidence":"title","id":"arxiv:2605.12466","title":"Solve the Loop: Attractor Models for Language and Reasoning","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/eb46f81cd5718eb6b9256c583cf9dd2e97b793b4d14337e487c16285f96e2b9a/"},{"source":"arxiv:2605.12969","target":"arxiv:2407.21783","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":"arxiv:2606.31779","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2606.31779","title":"Bridging the Gap Between Latent and Explicit Reasoning with Looped Transformers","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bf07d5c19e507d9ec91a48f9067e6c29fbe8965d001923e2de0e96da143c72e1/"},{"source":"arxiv:2608.03893","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2608.03893","title":"Cross-Model KV Cache Transfer in LLM Families: A Closed-Form Linear Mapping for Prefill Reuse","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/c29465e80f45097327908d15926fd9b4a3c4c50431b2d525eb5aa9b92446fb2a/"},{"source":"arxiv:2608.11676","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2608.11676","title":"XBridge: Entity-Grounded Latent Bridge for Heterogeneous LLM Communication","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/a1cbafd1249485a1818a70256d0d3ac5806d55b6900f005e9098c57d37918e44/"},{"source":"arxiv:2608.23875","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2608.23875","title":"AI Finds A Way","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/afe34dd5ee58c2e9739f9651349c4c9180d95dabff51994a13339f48dfd0bb27/"},{"source":"url:ai-2027.com","target":"arxiv:2407.21783","evidence":"id","id":"url:ai-2027.com","title":"AI 2027 (Kokotajlo)","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3fd447e0ff58689ae00a4666e28bbd7faecd472dec7195ab1d524cb5402a1766/"},{"source":"url:thinkingmachines.ai/blog/lora","target":"arxiv:2407.21783","evidence":"id","id":"url:thinkingmachines.ai/blog/lora","title":"LoRA Without Regret (Thinking Machines)","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/09bfccf7014ee63d12eddd2c954194cd981ce104398b7b64c98873bf2f997b95/"}],"text_parts":[{"start":0,"end":15370,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-1.html"},{"start":15370,"end":30779,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-2.html"},{"start":30779,"end":46535,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-3.html"},{"start":46535,"end":62494,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-4.html"},{"start":62494,"end":78468,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-5.html"},{"start":78468,"end":94313,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-6.html"},{"start":94313,"end":110218,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-7.html"},{"start":110218,"end":124740,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-8.html"},{"start":124740,"end":139849,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-9.html"},{"start":139849,"end":155787,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-10.html"},{"start":155787,"end":171019,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-11.html"},{"start":171019,"end":186342,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-12.html"},{"start":186342,"end":202188,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-13.html"},{"start":202188,"end":217842,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-14.html"},{"start":217842,"end":232586,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-15.html"},{"start":232586,"end":240201,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-16.html"}],"contents":[{"title":"The Llama 3 Herd of Models","offset":0,"level":1,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-1.html#text"},{"title":"Abstract","offset":30,"level":6,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-1.html#text"},{"title":"1 Introduction","offset":1124,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-1.html#text"},{"title":"2 General Overview","offset":6538,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-1.html#text"},{"title":"3 Pre-Training","offset":10713,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-1.html#text"},{"title":"3.1 Pre-Training Data","offset":11120,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-1.html#text"},{"title":"3.1.1 Web Data Curation","offset":11526,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-1.html#text"},{"title":"3.1.2 Determining the Data Mix","offset":16913,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-2.html#text"},{"title":"3.1.3 Annealing Data","offset":18155,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-2.html#text"},{"title":"3.2 Model Architecture","offset":19785,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-2.html#text"},{"title":"3.2.1 Scaling Laws","offset":21812,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-2.html#text"},{"title":"3.3 Infrastructure, Scaling, and Efficiency","offset":26263,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-2.html#text"},{"title":"3.3.1 Training Infrastructure","offset":26490,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-2.html#text"},{"title":"3.3.2 Parallelism for Model Scaling","offset":30737,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-2.html#text"},{"title":"3.3.3 Collective Communication","offset":38687,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-3.html#text"},{"title":"3.3.4 Reliability and Operational Challenges","offset":40055,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-3.html#text"},{"title":"3.4 Training Recipe","offset":44751,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-3.html#text"},{"title":"3.4.1 Initial Pre-Training","offset":45034,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-3.html#text"},{"title":"3.4.2 Long Context Pre-Training","offset":46497,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-3.html#text"},{"title":"3.4.3 Annealing","offset":47401,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-4.html#text"},{"title":"4 Post-Training","offset":47822,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-4.html#text"},{"title":"4.1 Modeling","offset":48662,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-4.html#text"},{"title":"4.1.1 Chat Dialog Format","offset":49253,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-4.html#text"},{"title":"4.1.2 Reward Modeling","offset":50004,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-4.html#text"},{"title":"4.1.3 Supervised Finetuning","offset":51093,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-4.html#text"},{"title":"4.1.4 Direct Preference Optimization","offset":51949,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-4.html#text"},{"title":"4.1.5 Model Averaging","offset":53886,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-4.html#text"},{"title":"4.1.6 Iterative Rounds","offset":54118,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-4.html#text"},{"title":"4.2 Post-training Data","offset":54324,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-4.html#text"},{"title":"4.2.1 Preference Data","offset":54688,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-4.html#text"},{"title":"4.2.2 SFT Data","offset":56850,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-4.html#text"},{"title":"4.2.3 Data Processing and Quality Control","offset":59431,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-4.html#text"},{"title":"4.3 Capabilities","offset":62152,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-4.html#text"},{"title":"4.3.1 Code","offset":62477,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-4.html#text"},{"title":"4.3.2 Multilinguality","offset":72326,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-5.html#text"},{"title":"4.3.3 Math and Reasoning","offset":76381,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-5.html#text"},{"title":"4.3.4 Long Context","offset":80921,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-6.html#text"},{"title":"4.3.5 Tool Use","offset":83997,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-6.html#text"},{"title":"4.3.6 Factuality","offset":91611,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-6.html#text"},{"title":"4.3.7 Steerability","offset":93449,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-6.html#text"},{"title":"5 Results","offset":94472,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-7.html#text"},{"title":"5.1 Pre-trained Language Model","offset":94776,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-7.html#text"},{"title":"5.1.1 Standard Benchmarks","offset":95864,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-7.html#text"},{"title":"5.1.2 Model Robustness","offset":99332,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-7.html#text"},{"title":"5.1.3 Adversarial Benchmarks","offset":102400,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-7.html#text"},{"title":"5.1.4 Contamination Analysis","offset":104349,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-7.html#text"},{"title":"5.2 Post-trained Language Model","offset":106989,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-7.html#text"},{"title":"5.2.1 General Knowledge and Instruction-Following Benchmarks","offset":108129,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-7.html#text"},{"title":"5.2.2 Proficiency Exams","offset":109624,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-7.html#text"},{"title":"5.2.3 Coding Benchmarks","offset":111001,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-8.html#text"},{"title":"5.2.4 Multilingual Benchmarks","offset":112579,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-8.html#text"},{"title":"5.2.5 Math and Reasoning Benchmarks","offset":114269,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-8.html#text"},{"title":"5.2.6 Long Context Benchmarks","offset":114794,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-8.html#text"},{"title":"5.2.7 Tool Use Performance","offset":116291,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-8.html#text"},{"title":"5.3 Human Evaluations","offset":117806,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-8.html#text"},{"title":"5.4 Safety","offset":121759,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-8.html#text"},{"title":"5.4.1 Benchmark Construction","offset":123101,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-8.html#text"},{"title":"5.4.2 Safety Pre-training","offset":124708,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-8.html#text"},{"title":"5.4.3 Safety Finetuning","offset":126439,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-9.html#text"},{"title":"5.4.4 Safety Results","offset":130852,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-9.html#text"},{"title":"5.4.5 Cybersecurity and Chemical/Biological Weapons Safety","offset":136141,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-9.html#text"},{"title":"5.4.6 Red Teaming","offset":144027,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-10.html#text"},{"title":"5.4.7 System Level Safety","offset":149936,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-10.html#text"},{"title":"5.4.8 Limitations","offset":155763,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-10.html#text"},{"title":"6 Inference","offset":156622,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-11.html#text"},{"title":"6.1 Pipeline Parallelism","offset":156848,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-11.html#text"},{"title":"6.2 FP8 Quantization","offset":158315,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-11.html#text"},{"title":"7 Vision Experiments","offset":161296,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-11.html#text"},{"title":"7.1 Data","offset":163224,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-11.html#text"},{"title":"7.1.1 Image Data","offset":163294,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-11.html#text"},{"title":"7.1.2 Video Data","offset":169377,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-11.html#text"},{"title":"7.2 Model Architecture","offset":170857,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-11.html#text"},{"title":"7.3 Model Scaling","offset":174766,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-12.html#text"},{"title":"7.4 Pre-training","offset":177520,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-12.html#text"},{"title":"7.5 Post-Training","offset":179439,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-12.html#text"},{"title":"7.5.1 Supervised Finetuning Data","offset":180106,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-12.html#text"},{"title":"7.5.2 Supervised Finetuning Recipe","offset":182838,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-12.html#text"},{"title":"7.5.3 Preference Data","offset":184527,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-12.html#text"},{"title":"7.5.4 Reward Modeling","offset":186314,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-12.html#text"},{"title":"7.5.5 Direct Preference Optimization","offset":187527,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-13.html#text"},{"title":"7.5.6 Rejection Sampling","offset":188395,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-13.html#text"},{"title":"7.5.7 Quality Tuning","offset":189786,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-13.html#text"},{"title":"7.6 Image Recognition Results","offset":190400,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-13.html#text"},{"title":"7.7 Video Recognition Results","offset":192592,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-13.html#text"},{"title":"8 Speech Experiments","offset":197222,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-13.html#text"},{"title":"8.1 Data","offset":199162,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-13.html#text"},{"title":"8.1.1 Speech Understanding","offset":199176,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-13.html#text"},{"title":"8.1.2 Speech Generation","offset":201317,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-13.html#text"},{"title":"8.2 Model Architecture","offset":202869,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-14.html#text"},{"title":"8.2.1 Speech Understanding","offset":202897,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-14.html#text"},{"title":"8.2.2 Speech Generation","offset":204968,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-14.html#text"},{"title":"8.3 Training Recipe","offset":207564,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-14.html#text"},{"title":"8.3.1 Speech Understanding","offset":207589,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-14.html#text"},{"title":"8.3.2 Speech Generation","offset":211172,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-14.html#text"},{"title":"8.4 Speech Understanding Results","offset":213349,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-14.html#text"},{"title":"8.5 Speech Generation Results","offset":217514,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-14.html#text"},{"title":"9 Related Work","offset":219925,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-15.html#text"},{"title":"9.1 Language","offset":220353,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-15.html#text"},{"title":"9.2 Multimodality","offset":223764,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-15.html#text"},{"title":"10 Conclusion","offset":225976,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-15.html#text"},{"title":"Contributors and Acknowledgements","offset":228528,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-15.html#text"},{"title":"Core Contributors","offset":228961,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-15.html#text"},{"title":"Contributors","offset":232568,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-15.html#text"},{"title":"Acknowledgements","offset":237291,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/part-16.html#text"}],"text_code_points":240201,"source_markdown":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/source.md"}