{"url":"/sota/language-modelling-on-penn-treebank-character","task":{"name":"Language Modelling","url":"/task/language-modelling","note":null},"dataset":{"name":"Penn Treebank (Character Level)","url":"/dataset/penn-treebank"},"category":"Natural Language Processing","categories":["Medical","Miscellaneous","Natural Language Processing"],"category_note":null,"description":"A language model is a model of natural language. Language models are useful for a variety of tasks, including speech recognition, machine translation, natural language generation (generating more human-like text), optical character recognition, route optimization, handwriting recognition, grammar induction, and information retrieval.\r\n\r\nLarge language models (LLMs), currently their most advanced form, are predominantly based on transformers trained on larger datasets (frequently using words scraped from the public internet). They have superseded recurrent neural network-based models, which had previously superseded the purely statistical models, such as word n-gram language model. \r\n\r\nSource: [Wikipedia](https://en.wikipedia.org/wiki/Language_model)","description_from":"task","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","rank":"the archive's row order at snapshot; not re-ranked","rows_end_at":"2025-07-28","rows_withheld_as_spam":0,"metric_values":"the archive's strings, untouched"},"metrics":["Bit per Character (BPC)","Number of params"],"metric_direction":{"note":"inferred from the metric name only (the archive records no direction); null = not inferred, chart draws points only","by_metric":{"Bit per Character (BPC)":null,"Number of params":"lower"}},"counts":{"rows":20,"rows_with_code":19,"rows_with_paper_page":20,"rows_dated":20,"rows_using_additional_data":0},"rows":[{"rank_in_archive_order":1,"model":"Mogrifier LSTM + dynamic eval","metrics":{"Bit per Character (BPC)":"1.083","Number of params":"24M"},"uses_additional_data":false,"paper_date":"2019-09-04","paper":"/paper/mogrifier-lstm","paper_url":"https://arxiv.org/abs/1909.01792v2","paper_title":"Mogrifier LSTM","code":"https://github.com/deepmind/lamb","n_code_links":3,"syntology":null},{"rank_in_archive_order":2,"model":"Mogrifier LSTM","metrics":{"Bit per Character (BPC)":"1.120","Number of params":"24M"},"uses_additional_data":false,"paper_date":"2019-09-04","paper":"/paper/mogrifier-lstm","paper_url":"https://arxiv.org/abs/1909.01792v2","paper_title":"Mogrifier LSTM","code":"https://github.com/deepmind/lamb","n_code_links":3,"syntology":null},{"rank_in_archive_order":3,"model":"GAM-RHN-5","metrics":{"Bit per Character (BPC)":"1.147","Number of params":"16.0M"},"uses_additional_data":false,"paper_date":"2019-12-13","paper":"/paper/recurrent-highway-networks-with-grouped","paper_url":"https://ieeexplore.ieee.org/document/8932404","paper_title":"Recurrent Highway Networks with Grouped Auxiliary Memory","code":"https://github.com/WilliamRo/gam_rhn","n_code_links":4,"syntology":null},{"rank_in_archive_order":4,"model":"Trellis Network","metrics":{"Bit per Character (BPC)":"1.158","Number of params":"13.4M"},"uses_additional_data":false,"paper_date":"2018-10-15","paper":"/paper/trellis-networks-for-sequence-modeling","paper_url":"http://arxiv.org/abs/1810.06682v2","paper_title":"Trellis Networks for Sequence Modeling","code":"https://github.com/locuslab/trellisnet","n_code_links":1,"syntology":{"n_ran":2,"n_unverified":6,"n_samples":8,"n_pointer_only_licence":0}},{"rank_in_archive_order":5,"model":"Feedback Transformer","metrics":{"Bit per Character (BPC)":"1.160","Number of params":"10.7M"},"uses_additional_data":false,"paper_date":"2020-02-21","paper":"/paper/accessing-higher-level-representations-in","paper_url":"https://arxiv.org/abs/2002.09402v3","paper_title":"Addressing Some Limitations of Transformers with Feedback Memory","code":"https://github.com/labmlai/annotated_deep_learning_paper_implementations","n_code_links":4,"syntology":{"n_ran":3,"n_unverified":0,"n_samples":3,"n_pointer_only_licence":2}},{"rank_in_archive_order":6,"model":"Past Decode Reg. + AWD-LSTM-MoS + dyn. eval.","metrics":{"Bit per Character (BPC)":"1.169","Number of params":"13.8M"},"uses_additional_data":false,"paper_date":"2018-08-14","paper":"/paper/improved-language-modeling-by-decoding-the","paper_url":"http://arxiv.org/abs/1808.05908v4","paper_title":"Improved Language Modeling by Decoding the Past","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":7,"model":"3-layer AWD-LSTM","metrics":{"Bit per Character (BPC)":"1.175","Number of params":"13.8M"},"uses_additional_data":false,"paper_date":"2018-03-22","paper":"/paper/an-analysis-of-neural-language-modeling-at","paper_url":"http://arxiv.org/abs/1803.08240v1","paper_title":"An Analysis of Neural Language Modeling at Multiple Scales","code":"https://github.com/salesforce/awd-lstm-lm","n_code_links":12,"syntology":{"n_ran":3,"n_unverified":15,"n_samples":18,"n_pointer_only_licence":0}},{"rank_in_archive_order":8,"model":"Dense IndRNN","metrics":{"Bit per Character (BPC)":"1.18"},"uses_additional_data":false,"paper_date":"2019-10-11","paper":"/paper/deep-independently-recurrent-neural-network","paper_url":"https://arxiv.org/abs/1910.06251v3","paper_title":"Deep Independently Recurrent Neural Network (IndRNN)","code":"https://github.com/Sunnydreamrain/IndRNN_pytorch","n_code_links":1,"syntology":null},{"rank_in_archive_order":9,"model":"6-layer QRNN","metrics":{"Bit per Character (BPC)":"1.187","Number of params":"13.8M"},"uses_additional_data":false,"paper_date":"2018-03-22","paper":"/paper/an-analysis-of-neural-language-modeling-at","paper_url":"http://arxiv.org/abs/1803.08240v1","paper_title":"An Analysis of Neural Language Modeling at Multiple Scales","code":"https://github.com/salesforce/awd-lstm-lm","n_code_links":12,"syntology":{"n_ran":3,"n_unverified":15,"n_samples":18,"n_pointer_only_licence":0}},{"rank_in_archive_order":10,"model":"FS-LSTM-4","metrics":{"Bit per Character (BPC)":"1.190","Number of params":"27M"},"uses_additional_data":false,"paper_date":"2017-05-24","paper":"/paper/fast-slow-recurrent-neural-networks","paper_url":"http://arxiv.org/abs/1705.08639v2","paper_title":"Fast-Slow Recurrent Neural Networks","code":"https://github.com/amujika/Fast-Slow-LSTM","n_code_links":1,"syntology":null},{"rank_in_archive_order":11,"model":"IndRNN","metrics":{"Bit per Character (BPC)":"1.19"},"uses_additional_data":false,"paper_date":"2018-03-13","paper":"/paper/independently-recurrent-neural-network-indrnn","paper_url":"http://arxiv.org/abs/1803.04831v3","paper_title":"Independently Recurrent Neural Network (IndRNN): Building A Longer and Deeper RNN","code":"https://github.com/TobiasLee/Text-Classification","n_code_links":11,"syntology":{"n_ran":1,"n_unverified":4,"n_samples":5,"n_pointer_only_licence":1}},{"rank_in_archive_order":12,"model":"FS-LSTM-2","metrics":{"Bit per Character (BPC)":"1.193","Number of params":"27M"},"uses_additional_data":false,"paper_date":"2017-05-24","paper":"/paper/fast-slow-recurrent-neural-networks","paper_url":"http://arxiv.org/abs/1705.08639v2","paper_title":"Fast-Slow Recurrent Neural Networks","code":"https://github.com/amujika/Fast-Slow-LSTM","n_code_links":1,"syntology":null},{"rank_in_archive_order":13,"model":"NAS-RL","metrics":{"Bit per Character (BPC)":"1.214","Number of params":"16.3M"},"uses_additional_data":false,"paper_date":"2016-11-05","paper":"/paper/neural-architecture-search-with-reinforcement","paper_url":"http://arxiv.org/abs/1611.01578v2","paper_title":"Neural Architecture Search with Reinforcement Learning","code":"https://github.com/tensorflow/models","n_code_links":12,"syntology":{"n_ran":5,"n_unverified":0,"n_samples":5,"n_pointer_only_licence":3}},{"rank_in_archive_order":14,"model":"2-layer Norm HyperLSTM","metrics":{"Bit per Character (BPC)":"1.219","Number of params":"14.4M"},"uses_additional_data":false,"paper_date":"2016-09-27","paper":"/paper/hypernetworks","paper_url":"http://arxiv.org/abs/1609.09106v4","paper_title":"HyperNetworks","code":"https://github.com/labmlai/annotated_deep_learning_paper_implementations","n_code_links":10,"syntology":{"n_ran":3,"n_unverified":1,"n_samples":4,"n_pointer_only_licence":3}},{"rank_in_archive_order":15,"model":"R-Transformer","metrics":{"Bit per Character (BPC)":"1.24"},"uses_additional_data":false,"paper_date":"2019-07-12","paper":"/paper/r-transformer-recurrent-neural-network","paper_url":"https://arxiv.org/abs/1907.05572v1","paper_title":"R-Transformer: Recurrent Neural Network Enhanced Transformer","code":"https://github.com/DSE-MSU/R-transformer","n_code_links":2,"syntology":{"n_ran":2,"n_unverified":0,"n_samples":2,"n_pointer_only_licence":2}},{"rank_in_archive_order":16,"model":"Seq-U-Net","metrics":{"Bit per Character (BPC)":"1.3","Number of params":"5.9M"},"uses_additional_data":false,"paper_date":"2019-11-14","paper":"/paper/seq-u-net-a-one-dimensional-causal-u-net-for","paper_url":"https://arxiv.org/abs/1911.06393v1","paper_title":"Seq-U-Net: A One-Dimensional Causal U-Net for Efficient Sequence Modelling","code":"https://github.com/f90/Seq-U-Net","n_code_links":1,"syntology":{"n_ran":0,"n_unverified":6,"n_samples":6,"n_pointer_only_licence":0}},{"rank_in_archive_order":17,"model":"STAR","metrics":{"Bit per Character (BPC)":"1.30"},"uses_additional_data":false,"paper_date":"2019-11-25","paper":"/paper/gating-revisited-deep-multi-layer-rnns-that-1","paper_url":"https://arxiv.org/abs/1911.11033v4","paper_title":"Gating Revisited: Deep Multi-layer RNNs That Can Be Trained","code":"https://github.com/0zgur0/STAR_Network","n_code_links":3,"syntology":null},{"rank_in_archive_order":18,"model":"TCN","metrics":{"Bit per Character (BPC)":"1.31","Number of params":"5.9M"},"uses_additional_data":false,"paper_date":"2019-11-14","paper":"/paper/seq-u-net-a-one-dimensional-causal-u-net-for","paper_url":"https://arxiv.org/abs/1911.06393v1","paper_title":"Seq-U-Net: A One-Dimensional Causal U-Net for Efficient Sequence Modelling","code":"https://github.com/f90/Seq-U-Net","n_code_links":1,"syntology":{"n_ran":0,"n_unverified":6,"n_samples":6,"n_pointer_only_licence":0}},{"rank_in_archive_order":19,"model":"Temporal Convolutional Network","metrics":{"Bit per Character (BPC)":"1.31"},"uses_additional_data":false,"paper_date":"2018-03-04","paper":"/paper/an-empirical-evaluation-of-generic","paper_url":"http://arxiv.org/abs/1803.01271v2","paper_title":"An Empirical Evaluation of Generic Convolutional and Recurrent Networks for Sequence Modeling","code":"https://github.com/timeseriesAI/tsai/tree/main/tsai/models","n_code_links":35,"syntology":{"n_ran":2,"n_unverified":8,"n_samples":10,"n_pointer_only_licence":1}},{"rank_in_archive_order":20,"model":"Bipartite Flow","metrics":{"Bit per Character (BPC)":"1.38"},"uses_additional_data":false,"paper_date":"2019-05-24","paper":"/paper/discrete-flows-invertible-generative-models","paper_url":"https://arxiv.org/abs/1905.10347v1","paper_title":"Discrete Flows: Invertible Generative Models of Discrete Data","code":"https://github.com/google/edward2","n_code_links":2,"syntology":null}],"since_archive":{"claim":"Results that newer papers report for their own method, placed here by Syntology. A model pointed at the cell in the paper's own table; the number was read from that cell and checked against this leaderboard's metric, dataset, split and scale; an independent check that saw this leaderboard's other rows and every other leaderboard on the same dataset accepted it. Not reviewed by the paper's authors or by the archive's editors, and not ranked against the archive rows.","extraction_file_present":true,"measurement":{"test_papers":883,"papers_with_output":881,"judged_true":108,"judged":110,"wilson95_lower":0.9361,"measured_on":"2026-09-24","frozen_commit":"0e3de0df94"},"measurement_note":"blind adjudication of accepted entries on a held-out split of archive papers, rules frozen before the test","coverage":{"sentence":"Syntology has checked 6,264 of the 9,581 papers on this site that are newer than the archive; results from the others appear after they are checked.","complete":false,"papers_newer_than_archive":9581,"papers_checked":6264,"papers_extracted_not_yet_verified":0,"boards_without_verdict":2,"papers_not_yet_extracted":3316},"order":"newest first by month (arXiv date, else the arXiv-id month), then arXiv id descending","columns":[],"entries":[]},"syntology":{"read_at":"2026-09-24T18:15:14+00:00","claim":"Per row: N of M harvested code samples from that row's paper executed on a synthesized fixture; the other M-N are unverified. Not a reproduction of the row's number; not a correctness claim. n_pointer_only_licence counts samples the site points at rather than redistributes (a licence axis, independent of ran/unverified).","rows_with_graph_line":11,"rows_with_any_sample_ran":9,"distinct_papers_with_graph_line":9,"distinct_papers_with_any_sample_ran":8,"samples_over_distinct_papers":{"n_ran":21,"n_unverified":40,"n_samples":61,"n_pointer_only_licence":12,"note":"each paper (arXiv id) counted once, however many rows it is behind; this is the page-level figure"},"samples_row_weighted":{"n_ran":24,"n_unverified":61,"n_samples":85,"n_pointer_only_licence":12,"note":"row-weighted: a paper behind several rows is counted once per row; inflated relative to samples_over_distinct_papers by design, kept for readers summing the per-row syntology blocks"}}}