{"id":"arxiv:2606.06574","title":"Skip a Layer or Loop It? Learning Program-of-Layers in LLMs","date":"2026-06-04","section":"LLMs: architectures, context, training","quality":0.192674,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/","source_url":"https://arxiv.org/abs/2606.06574","text_available":true,"authors":["Ziyue Li","Yang Li","Tianyi Zhou"],"outgoing":[{"source":"arxiv:2606.06574","target":"arxiv:1807.03819","evidence":"id","id":"arxiv:1807.03819","title":"Universal Transformers","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1d843516d980c3404cb622a2a490edb535a3afe27a85a980335a54b5fcd2220a/"},{"source":"arxiv:2606.06574","target":"arxiv:2507.10524","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/"}],"incoming":[],"text_parts":[{"start":0,"end":15552,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-1.html"},{"start":15552,"end":31361,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-2.html"},{"start":31361,"end":46994,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-3.html"},{"start":46994,"end":53428,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-4.html"}],"contents":[{"title":"Skip a Layer or Loop It? Learning Program-of-Layers in LLMs","offset":0,"level":1,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-1.html#text"},{"title":"Abstract","offset":63,"level":6,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-1.html#text"},{"title":"Keywords:","offset":1330,"level":6,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-1.html#text"},{"title":"1 Introduction","offset":1394,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-1.html#text"},{"title":"2 Dynamic Inference as a Program-of-Layers ( PoLar ) in Large Language Models","offset":5821,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-1.html#text"},{"title":"3 Learning Program-of-Layers ( PoLar ) in Large Language Models","offset":14014,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-1.html#text"},{"title":"3.1 Program Representation","offset":14420,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-1.html#text"},{"title":"3.2 Program-of-Layers ( PoLar ) Prediction Network","offset":18115,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-2.html#text"},{"title":"4 Experiments","offset":23498,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-2.html#text"},{"title":"4.1 Experimental Setup","offset":23740,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-2.html#text"},{"title":"4.2 Main Results","offset":26823,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-2.html#text"},{"title":"4.3 Out-of-Distribution Performance","offset":30213,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-2.html#text"},{"title":"5 Related Works","offset":31341,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-2.html#text"},{"title":"6 Conclusion","offset":34552,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-3.html#text"},{"title":"Impact Statement","offset":35588,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-3.html#text"},{"title":"Appendix A Related Work","offset":37150,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-3.html#text"},{"title":"Layer Pruning and Early-Exit Neural Networks","offset":37178,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-3.html#text"},{"title":"Looped Transformer and Recurrent Depth","offset":39493,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-3.html#text"},{"title":"Dynamic Routing and Modular Inference","offset":40896,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-3.html#text"},{"title":"Appendix B Searching the Execution Program Space","offset":42439,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-3.html#text"},{"title":"B.1 Execution Program Space","offset":42777,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-3.html#text"},{"title":"B.2 Search Space Constraints","offset":43908,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-3.html#text"},{"title":"B.3 Monte Carlo Tree Search Formulation","offset":44647,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-3.html#text"},{"title":"State and Actions.","offset":44841,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-3.html#text"},{"title":"Reward.","offset":45050,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-3.html#text"},{"title":"Tree Policy.","offset":45330,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-3.html#text"},{"title":"B.4 Search Algorithm","offset":45831,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-3.html#text"},{"title":"Appendix C Experimental Results","offset":46247,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-3.html#text"},{"title":"C.1 In-Distribution Performance","offset":46644,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-3.html#text"},{"title":"C.2 Out-of-Distribution Generalization","offset":48211,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-4.html#text"},{"title":"Appendix D Empirical Details","offset":49420,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-4.html#text"},{"title":"D.1 Dataset Details","offset":49453,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-4.html#text"},{"title":"Revision note.","offset":50543,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-4.html#text"},{"title":"D.2 Training Configuration","offset":50856,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-4.html#text"},{"title":"Optimization.","offset":51049,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-4.html#text"},{"title":"D.3 Predictor Size","offset":51445,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-4.html#text"},{"title":"D.4 Direct Prompting","offset":51813,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-4.html#text"},{"title":"Instructions for reporting errors","offset":52464,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/part-4.html#text"}],"text_code_points":53428,"source_markdown":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5abe08b6c82ce9a4010a09d82dbee40f8b4b0f88539bf8911ae36048a97bb917/source.md"}