{"id":"arxiv:2302.10866","title":"Hyena Hierarchy: Towards Larger Convolutional Language Models","date":"2023-02-21","section":"LLMs: architectures, context, training","quality":0.423283,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/","source_url":"https://arxiv.org/abs/2302.10866","text_available":true,"authors":["Michael Poli","Stefano Massaroli","Eric Nguyen","Daniel Y. Fu"],"outgoing":[],"incoming":[{"source":"arxiv:2312.04927","target":"arxiv:2302.10866","evidence":"id","id":"arxiv:2312.04927","title":"Zoology: Measuring and Improving Recall in Efficient Language Models","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/baa792b06bdd30f91619f1e852aa5ae4dbf688fe3b1f2ee0722b339b9993ac5c/"},{"source":"arxiv:2405.04517","target":"arxiv:2302.10866","evidence":"title","id":"arxiv:2405.04517","title":"xLSTM: Extended Long Short-Term Memory","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/40fd3e03894caeb940ad2021fb4052c0d2e809706b5022f7454072997b8ba2d5/"},{"source":"arxiv:2512.24695","target":"arxiv:2302.10866","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/"}],"text_parts":[{"start":0,"end":15472,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-1.html"},{"start":15472,"end":31126,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-2.html"},{"start":31126,"end":46922,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-3.html"},{"start":46922,"end":62896,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-4.html"},{"start":62896,"end":66901,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-5.html"}],"contents":[{"title":"Hyena Hierarchy: Towards Larger Convolutional Language Models","offset":19,"level":1,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-1.html#text"},{"title":"Abstract","offset":84,"level":6,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-1.html#text"},{"title":"1 Introduction","offset":1399,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-1.html#text"},{"title":"The 𝖧𝗒𝖾𝗇𝖺 {\\sf Hyena} hierarchy","offset":4924,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-1.html#text"},{"title":"Narrowing the capabilities gap","offset":6117,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-1.html#text"},{"title":"Scaling in language and vision","offset":7004,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-1.html#text"},{"title":"Toward much longer context","offset":7900,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-1.html#text"},{"title":"2 Preliminaries and Related Work","offset":8394,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-1.html#text"},{"title":"2.1 Explicit and Implicit Convolutions","offset":9890,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-1.html#text"},{"title":"Fast Methods for Convolutions","offset":13058,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-1.html#text"},{"title":"2.2 The Self-Attention Operator","offset":14775,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-1.html#text"},{"title":"Remark 2.1 .","offset":16504,"level":6,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-2.html#text"},{"title":"Subquadratic Operators","offset":16775,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-2.html#text"},{"title":"3 Hyena: Definition and Properties","offset":17721,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-2.html#text"},{"title":"3.1 𝖧𝗒𝖾𝗇𝖺 {\\sf Hyena} Recurrences","offset":18130,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-2.html#text"},{"title":"Remark 3.1 .","offset":19371,"level":6,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-2.html#text"},{"title":"3.2 𝖧𝗒𝖾𝗇𝖺 {\\sf Hyena} Matrices","offset":20645,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-2.html#text"},{"title":"Remark 3.2 ( 𝖧𝗒𝖾𝗇𝖺 \\sf Hyena generalizes 𝖧𝟥 \\sf H3 and 𝖦𝖲𝖲 \\sf GSS .) .","offset":22911,"level":6,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-2.html#text"},{"title":"Remark 3.3 .","offset":23990,"level":6,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-2.html#text"},{"title":"3.3 𝖧𝗒𝖾𝗇𝖺 {\\sf Hyena} Filters","offset":24247,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-2.html#text"},{"title":"Specializing filters in Hyena","offset":24993,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-2.html#text"},{"title":"Preserving causality","offset":26249,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-2.html#text"},{"title":"Proposition 3.1 (Causal Hyenas) .","offset":26590,"level":6,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-2.html#text"},{"title":"Efficiency","offset":27067,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-2.html#text"},{"title":"3.4 Hyena Algorithm","offset":27995,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-2.html#text"},{"title":"Proposition 3.2 (Computational Complexity) .","offset":28066,"level":6,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-2.html#text"},{"title":"4 Experiments","offset":28320,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-2.html#text"},{"title":"4.1 Shrinking the gap on in-context learning","offset":28338,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-2.html#text"},{"title":"How to parametrize long convolutions","offset":29253,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-2.html#text"},{"title":"Pushing sequence length to the limit","offset":30685,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-2.html#text"},{"title":"Comparing operators","offset":31100,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-2.html#text"},{"title":"Generality of 𝖧𝗒𝖾𝗇𝖺 {\\sf Hyena} operators and filters","offset":31856,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-3.html#text"},{"title":"4.2 Language Modeling","offset":32329,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-3.html#text"},{"title":"4.3 Downstream Evaluation","offset":33353,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-3.html#text"},{"title":"4.4 Benchmarking","offset":34432,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-3.html#text"},{"title":"4.5 Large-Scale Image Classification","offset":35395,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-3.html#text"},{"title":"5 Discussion and Conclusion","offset":36410,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-3.html#text"},{"title":"Acknowledgments","offset":37460,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-3.html#text"},{"title":"Appendix A Experimental Details","offset":39167,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-3.html#text"},{"title":"A.1 Mechanistic Design Synthetic Benchmarks","offset":39271,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-3.html#text"},{"title":"Long convolution comparisons:","offset":41411,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-3.html#text"},{"title":"Operator comparisons:","offset":43325,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-3.html#text"},{"title":"A note on Transformer performance","offset":43751,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-3.html#text"},{"title":"A.2 Language Modeling","offset":44863,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-3.html#text"},{"title":"WikiText103:","offset":44890,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-3.html#text"},{"title":"The Pile:","offset":45811,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-3.html#text"},{"title":"PG-19","offset":47473,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-4.html#text"},{"title":"Architectures","offset":47796,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-4.html#text"},{"title":"FLOP computation","offset":47950,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-4.html#text"},{"title":"A.3 Downstream Evaluation","offset":48795,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-4.html#text"},{"title":"SuperGLUE:","offset":48826,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-4.html#text"},{"title":"Models","offset":49132,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-4.html#text"},{"title":"LAMBADA:","offset":49409,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-4.html#text"},{"title":"A.4 Image Classification","offset":49730,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-4.html#text"},{"title":"a","offset":49760,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-4.html#text"},{"title":"CIFAR-10:","offset":50449,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-4.html#text"},{"title":"Appendix B Theoretical Results and Details","offset":51381,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-4.html#text"},{"title":"B.1 Proofs","offset":51428,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-4.html#text"},{"title":"Proof of Proposition 3.1","offset":51444,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-4.html#text"},{"title":"Proof.","offset":51475,"level":6,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-4.html#text"},{"title":"B.2 Analysis of Data-Controlled Mechanisms","offset":51985,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-4.html#text"},{"title":"Isolating the surrogate attention matrix","offset":52788,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-4.html#text"},{"title":"Operator decomposition of the surrogate attention matrix","offset":54173,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-4.html#text"},{"title":"Appendix C Discussion and Additional Results","offset":58002,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-4.html#text"},{"title":"Vocabulary size scaling","offset":58051,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-4.html#text"},{"title":"Single layer recall","offset":58826,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-4.html#text"},{"title":"C.1 Learning Arithmetic","offset":59347,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-4.html#text"},{"title":"Appendix D Samples and Visualizations","offset":60533,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-4.html#text"},{"title":"D.1 Hyena Matrices","offset":60575,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-4.html#text"},{"title":"D.2 Hyena Filters","offset":61797,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-4.html#text"},{"title":"D.3 Positional Encoding and Filters Initialization","offset":62840,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-4.html#text"},{"title":"D.4 Downstream Examples","offset":65187,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-5.html#text"},{"title":"MultiRC","offset":65216,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-5.html#text"},{"title":"Instructions for reporting errors","offset":65937,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/part-5.html#text"}],"text_code_points":66901,"source_markdown":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/617c7279bfc8f8250cb7fb2d0bb4a2f7be1d1d692516fd45f80c7229632b16c7/source.md"}