{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/from-bytes-to-ideas-language-modeling-with","title":"From Bytes to Ideas: Language Modeling with Autoregressive U-Nets","arxiv_id":"2506.14761","date":"2025-06-17","proceeding":null,"authors":["Mathurin Videau","Badr Youbi Idrissi","Alessandro Leite","Marc Schoenauer","Olivier Teytaud","David Lopez-Paz"],"abstract":"Tokenization imposes a fixed granularity on the input text, freezing how a language model operates on data and how far in the future it predicts. Byte Pair Encoding (BPE) and similar schemes split text once, build a static vocabulary, and leave the model stuck with that choice. We relax this rigidity by introducing an autoregressive U-Net that learns to embed its own tokens as it trains. The network reads raw bytes, pools them into words, then pairs of words, then up to 4 words, giving it a multi-scale view of the sequence. At deeper stages, the model must predict further into the future -- anticipating the next few words rather than the next byte -- so deeper stages focus on broader semantic patterns while earlier stages handle fine details. When carefully tuning and controlling pretraining compute, shallow hierarchies tie strong BPE baselines, and deeper hierarchies have a promising trend. Because tokenization now lives inside the model, the same system can handle character-level tasks and carry knowledge across low-resource languages.","url_abs":"https://arxiv.org/abs/2506.14761v1","url_pdf":"https://arxiv.org/pdf/2506.14761v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"from-bytes-to-ideas-language-modeling-with","repo_url":"https://github.com/facebookresearch/lingua","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"ok","spdx":"BSD-3-Clause"}}],"tasks":[{"task_slug":"language-modeling","task_name":"Language Modeling"},{"task_slug":"language-modelling","task_name":"Language Modelling"}],"methods":[{"method_slug":"bpe","method_name":"BPE"},{"method_slug":"focus","method_name":"Focus"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2506.14761","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2506.14761"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-24T18:15:14+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/facebookresearch/lingua","reach":{"status":"ok","spdx":"BSD-3-Clause"}}],"summary":{"ran":4,"unverified":12},"by_repo_kind":{"official":{"samples":16,"ran":4,"repositories":1}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted","community":"Not in the archive's code links for this paper; a community repository Syntology harvested"},"n_pointer_only_for_licence":0,"samples":[{"code_sha256_prefix":"e0ebc4027f1bd03b","entry":"dataclass_from_dict","repo":"facebookresearch/lingua","repo_kind":"official","path":"lingua/args.py","file_url":"https://github.com/facebookresearch/lingua/blob/HEAD/lingua/args.py","link_basis":"plan_row","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"BSD-3-Clause","inline_ok":true,"mcp_get_code":{"code_sha256":"e0ebc4027f1bd03b"}},{"code_sha256_prefix":"116b2f19f9a6148c","entry":"dataclass_to_dict","repo":"facebookresearch/lingua","repo_kind":"official","path":"lingua/args.py","file_url":"https://github.com/facebookresearch/lingua/blob/HEAD/lingua/args.py","link_basis":"plan_row","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"BSD-3-Clause","inline_ok":true,"mcp_get_code":{"code_sha256":"116b2f19f9a6148c"}},{"code_sha256_prefix":"15be2e56e073a228","entry":"flatten_dict","repo":"facebookresearch/lingua","repo_kind":"official","path":"lingua/args.py","file_url":"https://github.com/facebookresearch/lingua/blob/HEAD/lingua/args.py","link_basis":"plan_row","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"BSD-3-Clause","inline_ok":true,"mcp_get_code":{"code_sha256":"15be2e56e073a228"}},{"code_sha256_prefix":"bfdcca22e67a0a67","entry":"get_empty_buffer_state","repo":"facebookresearch/lingua","repo_kind":"official","path":"lingua/data.py","file_url":"https://github.com/facebookresearch/lingua/blob/HEAD/lingua/data.py","link_basis":"plan_row","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"BSD-3-Clause","inline_ok":true,"mcp_get_code":{"code_sha256":"bfdcca22e67a0a67"}},{"code_sha256_prefix":"a11baf6e8cca859b","entry":"consolidate_checkpoints","repo":"facebookresearch/lingua","repo_kind":"official","path":"lingua/checkpoint.py","file_url":"https://github.com/facebookresearch/lingua/blob/HEAD/lingua/checkpoint.py","link_basis":"plan_row","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"BSD-3-Clause","inline_ok":true,"mcp_get_code":{"code_sha256":"a11baf6e8cca859b"}},{"code_sha256_prefix":"9caf44ff8e00b1dc","entry":"dist_max","repo":"facebookresearch/lingua","repo_kind":"official","path":"lingua/distributed.py","file_url":"https://github.com/facebookresearch/lingua/blob/HEAD/lingua/distributed.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"BSD-3-Clause","inline_ok":true,"mcp_get_code":{"code_sha256":"9caf44ff8e00b1dc"}},{"code_sha256_prefix":"d09c7072d82024f4","entry":"dist_mean","repo":"facebookresearch/lingua","repo_kind":"official","path":"lingua/distributed.py","file_url":"https://github.com/facebookresearch/lingua/blob/HEAD/lingua/distributed.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"BSD-3-Clause","inline_ok":true,"mcp_get_code":{"code_sha256":"d09c7072d82024f4"}},{"code_sha256_prefix":"cc85b8b7fd6d3080","entry":"distribute_data_to_rank","repo":"facebookresearch/lingua","repo_kind":"official","path":"lingua/data.py","file_url":"https://github.com/facebookresearch/lingua/blob/HEAD/lingua/data.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"BSD-3-Clause","inline_ok":true,"mcp_get_code":{"code_sha256":"cc85b8b7fd6d3080"}},{"code_sha256_prefix":"9e7610ce761e17d3","entry":"find_and_sanitize_chunks","repo":"facebookresearch/lingua","repo_kind":"official","path":"lingua/data.py","file_url":"https://github.com/facebookresearch/lingua/blob/HEAD/lingua/data.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"BSD-3-Clause","inline_ok":true,"mcp_get_code":{"code_sha256":"9e7610ce761e17d3"}},{"code_sha256_prefix":"d9d11e6287ad4384","entry":"get_device_mesh","repo":"facebookresearch/lingua","repo_kind":"official","path":"lingua/distributed.py","file_url":"https://github.com/facebookresearch/lingua/blob/HEAD/lingua/distributed.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"BSD-3-Clause","inline_ok":true,"mcp_get_code":{"code_sha256":"d9d11e6287ad4384"}},{"code_sha256_prefix":"81d8582d1fab7bda","entry":"get_num_params","repo":"facebookresearch/lingua","repo_kind":"official","path":"lingua/metrics.py","file_url":"https://github.com/facebookresearch/lingua/blob/HEAD/lingua/metrics.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"BSD-3-Clause","inline_ok":true,"mcp_get_code":{"code_sha256":"81d8582d1fab7bda"}},{"code_sha256_prefix":"22fbaa2caf605450","entry":"log_stats","repo":"facebookresearch/lingua","repo_kind":"official","path":"lingua/probe.py","file_url":"https://github.com/facebookresearch/lingua/blob/HEAD/lingua/probe.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"BSD-3-Clause","inline_ok":true,"mcp_get_code":{"code_sha256":"22fbaa2caf605450"}},{"code_sha256_prefix":"784760add5d1bc21","entry":"lr_cosine","repo":"facebookresearch/lingua","repo_kind":"official","path":"lingua/optim.py","file_url":"https://github.com/facebookresearch/lingua/blob/HEAD/lingua/optim.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"BSD-3-Clause","inline_ok":true,"mcp_get_code":{"code_sha256":"784760add5d1bc21"}},{"code_sha256_prefix":"90403d20388a5636","entry":"lr_inv_sqrt","repo":"facebookresearch/lingua","repo_kind":"official","path":"lingua/optim.py","file_url":"https://github.com/facebookresearch/lingua/blob/HEAD/lingua/optim.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"BSD-3-Clause","inline_ok":true,"mcp_get_code":{"code_sha256":"90403d20388a5636"}},{"code_sha256_prefix":"4e2454a79a8338d0","entry":"lr_linear","repo":"facebookresearch/lingua","repo_kind":"official","path":"lingua/optim.py","file_url":"https://github.com/facebookresearch/lingua/blob/HEAD/lingua/optim.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"BSD-3-Clause","inline_ok":true,"mcp_get_code":{"code_sha256":"4e2454a79a8338d0"}},{"code_sha256_prefix":"7255d9a8985034c9","entry":"mul_tiled","repo":"facebookresearch/lingua","repo_kind":"official","path":"lingua/float8.py","file_url":"https://github.com/facebookresearch/lingua/blob/HEAD/lingua/float8.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"BSD-3-Clause","inline_ok":true,"mcp_get_code":{"code_sha256":"7255d9a8985034c9"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}