{"id":"arxiv:2608.05806","title":"Hierarchical Latent Prediction for Language Models","date":"2026-08-06","section":"LLMs: architectures, context, training","quality":null,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/991d6d0a05c5eaa5f6c94943be460ba2a65770ccae4997df9544bb772604d430/","source_url":"https://arxiv.org/abs/2608.05806","text_available":true,"authors":["Chang Shi","Tim Pearce","Manan Tomar","Siddhartha Sen"],"outgoing":[{"source":"arxiv:2608.05806","target":"arxiv:2510.27688","evidence":"id","id":"arxiv:2510.27688","title":"Continuous Autoregressive Language Models","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/e954d29144273a3d9e4c1a5ac4fc43a163b8701659926e2bfa94011507f51414/"},{"source":"arxiv:2608.05806","target":"url:ai.meta.com/research/publications/large-concept-models-language-modeling-in-a-sentence-representation-space","evidence":"id","id":"url:ai.meta.com/research/publications/large-concept-models-language-modeling-in-a-sentence-representation-space","title":"Large Concept Models: Language Modeling in a Sentence Representation Space","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/a5cd322d52f1966361afb1d5225560e2d3ce4d22a117bd14cfa27ec669e09673/"}],"incoming":[],"text_parts":[{"start":0,"end":15073,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/991d6d0a05c5eaa5f6c94943be460ba2a65770ccae4997df9544bb772604d430/part-1.html"},{"start":15073,"end":18659,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/991d6d0a05c5eaa5f6c94943be460ba2a65770ccae4997df9544bb772604d430/part-2.html"}],"contents":[{"title":"Hierarchical Latent Prediction for Language Models","offset":0,"level":1,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/991d6d0a05c5eaa5f6c94943be460ba2a65770ccae4997df9544bb772604d430/part-1.html#text"},{"title":"Abstract","offset":54,"level":6,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/991d6d0a05c5eaa5f6c94943be460ba2a65770ccae4997df9544bb772604d430/part-1.html#text"},{"title":"1 Introduction","offset":1020,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/991d6d0a05c5eaa5f6c94943be460ba2a65770ccae4997df9544bb772604d430/part-1.html#text"},{"title":"2 Methodology","offset":3344,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/991d6d0a05c5eaa5f6c94943be460ba2a65770ccae4997df9544bb772604d430/part-1.html#text"},{"title":"Training objectives.","offset":4378,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/991d6d0a05c5eaa5f6c94943be460ba2a65770ccae4997df9544bb772604d430/part-1.html#text"},{"title":"Consistency conditions.","offset":6972,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/991d6d0a05c5eaa5f6c94943be460ba2a65770ccae4997df9544bb772604d430/part-1.html#text"},{"title":"3 Experiments","offset":7823,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/991d6d0a05c5eaa5f6c94943be460ba2a65770ccae4997df9544bb772604d430/part-1.html#text"},{"title":"3.1 Coding and Multi-step Reasoning Benchmarks","offset":7924,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/991d6d0a05c5eaa5f6c94943be460ba2a65770ccae4997df9544bb772604d430/part-1.html#text"},{"title":"3.2 Speculative Decoding","offset":8369,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/991d6d0a05c5eaa5f6c94943be460ba2a65770ccae4997df9544bb772604d430/part-1.html#text"},{"title":"4 Discussion","offset":8811,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/991d6d0a05c5eaa5f6c94943be460ba2a65770ccae4997df9544bb772604d430/part-1.html#text"},{"title":"Longer horizon prediction.","offset":8828,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/991d6d0a05c5eaa5f6c94943be460ba2a65770ccae4997df9544bb772604d430/part-1.html#text"},{"title":"5 Related Work","offset":9187,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/991d6d0a05c5eaa5f6c94943be460ba2a65770ccae4997df9544bb772604d430/part-1.html#text"},{"title":"5.1 Pretraining Objectives Beyond Next-Token Prediction","offset":9387,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/991d6d0a05c5eaa5f6c94943be460ba2a65770ccae4997df9544bb772604d430/part-1.html#text"},{"title":"5.2 Latent-Space Language Modeling","offset":10347,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/991d6d0a05c5eaa5f6c94943be460ba2a65770ccae4997df9544bb772604d430/part-1.html#text"},{"title":"5.3 Hierarchical Models","offset":11408,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/991d6d0a05c5eaa5f6c94943be460ba2a65770ccae4997df9544bb772604d430/part-1.html#text"},{"title":"6 Conclusion","offset":13210,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/991d6d0a05c5eaa5f6c94943be460ba2a65770ccae4997df9544bb772604d430/part-1.html#text"},{"title":"7 Future Work","offset":13571,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/991d6d0a05c5eaa5f6c94943be460ba2a65770ccae4997df9544bb772604d430/part-1.html#text"},{"title":"Appendix A HiLP Training Procedure","offset":13963,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/991d6d0a05c5eaa5f6c94943be460ba2a65770ccae4997df9544bb772604d430/part-1.html#text"},{"title":"Appendix B Experiment hyperparameters","offset":14984,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/991d6d0a05c5eaa5f6c94943be460ba2a65770ccae4997df9544bb772604d430/part-1.html#text"},{"title":"Appendix C Latent overhead and efficiency.","offset":15026,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/991d6d0a05c5eaa5f6c94943be460ba2a65770ccae4997df9544bb772604d430/part-1.html#text"},{"title":"Appendix D Input-Combination Modes","offset":16023,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/991d6d0a05c5eaa5f6c94943be460ba2a65770ccae4997df9544bb772604d430/part-2.html#text"},{"title":"Instructions for reporting errors","offset":17695,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/991d6d0a05c5eaa5f6c94943be460ba2a65770ccae4997df9544bb772604d430/part-2.html#text"}],"text_code_points":18659,"source_markdown":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/991d6d0a05c5eaa5f6c94943be460ba2a65770ccae4997df9544bb772604d430/source.md"}