{"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/partially-shuffling-the-training-data-to-1","title":"Partially Shuffling the Training Data to Improve Language Models","arxiv_id":"1903.04167","date":"2019-03-11","proceeding":"arXiv 2019 3","authors":["Ofir Press"],"abstract":"Although SGD requires shuffling the training data between epochs, currently\nnone of the word-level language modeling systems do this. Naively shuffling all\nsentences in the training data would not permit the model to learn\ninter-sentence dependencies. Here we present a method that partially shuffles\nthe training data between epochs. This method makes each batch random, while\nkeeping most sentence ordering intact. It achieves new state of the art results\non word-level language modeling on both the Penn Treebank and WikiText-2\ndatasets.","url_abs":"http://arxiv.org/abs/1903.04167v2","url_pdf":"http://arxiv.org/pdf/1903.04167v2.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":"partially-shuffling-the-training-data-to-1","repo_url":"https://github.com/ofirpress/PartialShuffle","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"language-modeling","task_name":"Language Modeling"},{"task_slug":"language-modelling","task_name":"Language Modelling"},{"task_slug":"sentence","task_name":"Sentence"},{"task_slug":"sentence-ordering","task_name":"Sentence Ordering"}],"methods":[{"method_slug":"sgd","method_name":"SGD"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/language-modelling-on-penn-treebank-word","task":"Language Modelling","dataset":"Penn Treebank (Word Level)","model":"AWD-LSTM-DOC + Partial Shuffle","rank_in_archive_order":15,"of":43,"metrics":{"Params":"23M","Test perplexity":"52.0","Validation perplexity":"53.79"},"uses_additional_data":false},{"leaderboard":"/sota/language-modelling-on-penn-treebank-word","task":"Language Modelling","dataset":"Penn Treebank (Word Level)","model":"AWD-LSTM-MoS + Partial Shuffle","rank_in_archive_order":18,"of":43,"metrics":{"Params":"22M","Test perplexity":"53.92","Validation perplexity":"55.89"},"uses_additional_data":false},{"leaderboard":"/sota/language-modelling-on-wikitext-2","task":"Language Modelling","dataset":"WikiText-2","model":"AWD-LSTM-DOC + Partial Shuffle","rank_in_archive_order":23,"of":38,"metrics":{"Number of params":"37M","Test perplexity":"57.85","Validation perplexity":"60.16"},"uses_additional_data":false},{"leaderboard":"/sota/language-modelling-on-wikitext-2","task":"Language Modelling","dataset":"WikiText-2","model":"AWD-LSTM-MoS + Partial Shuffle","rank_in_archive_order":25,"of":38,"metrics":{"Number of params":"35M","Test perplexity":"59.98","Validation perplexity":"62.38"},"uses_additional_data":false}],"syntology":{"syntology_url":null,"atlas_url":"https://app.syntology.ai/?focus=1903.04167","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}