{"id":1191467,"date":"2026-10-27T15:52:48","date_gmt":"2026-10-27T22:52:48","guid":{"rendered":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/publication\/bridging-compute-and-data-optimal-pretraining\/"},"modified":"2026-10-30T09:04:48","modified_gmt":"2026-10-30T16:04:48","slug":"bridging-compute-and-data-optimal-pretraining","status":"publish","type":"msr-research-item","link":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/publication\/bridging-compute-and-data-optimal-pretraining\/","title":{"rendered":"Bridging Compute- and Data-Optimal Pretraining"},"content":{"rendered":"\n\n\n<p class=\"wp-block-paragraph\">Classical compute-optimal scaling laws assume an unbounded supply of fresh pretraining data, yet pretraining is increasingly entering a regime in which compute grows faster than the availability of high-quality data. We propose Compute-Data (CD) scaling laws, a unified framework that bridges compute-optimal scaling, where data scales freely with compute, and data-optimal scaling, where the corpus is fixed while compute can grow without bound. CD scaling extends classical scaling laws by introducing a token-effectiveness function, <math><mi>\u03b7<\/mi><\/math>, which quantifies the value of a derived token-produced, for example, through multi-epoch repetition or paraphrasing-relative to a fresh token, ranging from a perfect substitute to having no value. We fit <math><mi>\u03b7<\/mi><\/math> for two data-expansion strategies, multi-epoch repetition and paraphrasing, across model sizes from 14M to 600M parameters using the Dolma-3 corpus. We find that token effectiveness is far from constant: it depends jointly on model size, the tokens-per-parameter ratio, and the amount of derived data, and it saturates as the corpus is expanded. The functional form of <math><mi>\u03b7<\/mi><\/math> implies diminishing returns when substituting compute for data as either model size or data availability increases. It also partitions training into three operational regimes&#8212;compute-bound, data-bound, and model-bound&#8212;and shows that classical compute-optimal allocation is suboptimal across most practically relevant settings.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Classical compute-optimal scaling laws assume an unbounded supply of fresh pretraining data, yet pretraining is increasingly entering a regime in which compute grows faster than the availability of high-quality data. We propose Compute-Data (CD) scaling laws, a unified framework that bridges compute-optimal scaling, where data scales freely with compute, and data-optimal scaling, where the corpus [&hellip;]<\/p>\n","protected":false},"featured_media":0,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","msr-author-ordering":[{"type":"text","value":"Tian Qin","user_id":0},{"type":"text","value":"K. Hamidieh","user_id":0},{"type":"user_nicename","value":"David Alvarez-Melis","user_id":"38814"}],"msr_publishername":"","msr_publisher_other":"","msr_booktitle":"","msr_chapter":"","msr_edition":"","msr_editors":"","msr_how_published":"","msr_isbn":"","msr_issue":"","msr_journal":"","msr_number":"","msr_organization":"","msr_pages_string":"","msr_page_range_start":"","msr_page_range_end":"","msr_series":"","msr_volume":"","msr_copyright":"","msr_conference_name":"NeurIPS 2026","msr_doi":"10.48550\/arXiv.2607.25271","msr_arxiv_id":"2607.25271","msr_mag_id":"","msr_other_authors":"","msr_other_contributors":"","msr_speaker":"","msr_award":"","msr_affiliation":"","msr_institution":"","msr_host":"","msr_version":"","msr_duration":"","msr_release_tracker_id":"","msr_highlight_type":"","msr_date_display_format":"","msr_main_download_label":"","msr_external_link_label":"","msr_doi_label":"","msr_published_date":"2026-07-28","msr_startdate":"","msr_presentation_date":"","msr_highlight_text":"","msr_notes":"","msr_longbiography":"","msr_publicationurl":"","msr_external_url":"","msr_secondary_video_url":"","msr_conference_url":"","msr_journal_url":"","msr_year":2026,"msr_month":7,"msr_day":28,"msr_microsoftintellectualproperty":false,"msr_pub_id":"026c135d2cc5ef3c98ad2d122522ff93380b6239","msr_publication_uploader":[{"type":"url","title":"https:\/\/dblp.org\/rec\/journals\/corr\/abs-2607-25271.html","label_id":243109,"id":false,"viewUrl":false},{"type":"url","title":"https:\/\/arxiv.org\/abs\/2607.25271","label_id":243109,"id":false,"viewUrl":false}],"msr_related_uploader":[],"msr_original_fields_of_study":[],"msr_s2_paper_id":"026c135d2cc5ef3c98ad2d122522ff93380b6239","msr_s2_pdf_url":"","msr_citation_count_updated":"","msr_citation_count":2,"msr_influential_citations":2,"msr_reference_count":30,"msr_s2_open_access":false,"msr_s2_author_ids":[],"msr_pub_ids":[{"provider":"s2","id":"026c135d2cc5ef3c98ad2d122522ff93380b6239"},{"provider":"doi","id":"10.48550\/arXiv.2607.25271"},{"provider":"arxiv","id":"2607.25271"},{"provider":"dblp","id":"journals\/corr\/abs-2607-25271"},{"provider":"corpusid","id":"290635341"}],"msr_hide_image_in_river":0,"footnotes":""},"msr-research-highlight":[],"research-area":[13561,13556],"msr-publication-type":[193716],"msr-publisher":[],"msr-publication-cta":[],"msr-focus-area":[],"msr-locale":[268875],"msr-post-option":[],"msr-field-of-study":[246691,265497],"msr-conference":[259048],"msr-journal":[268188],"msr-impact-theme":[],"msr-pillar":[],"class_list":["post-1191467","msr-research-item","type-msr-research-item","status-publish","hentry","msr-research-area-algorithms","msr-research-area-artificial-intelligence","msr-locale-en_us","msr-field-of-study-computer-science","msr-field-of-study-machine-learning-296"],"msr_publishername":"","msr_edition":"","msr_affiliation":"","msr_published_date":"2026-07-28","msr_host":"","msr_duration":"","msr_version":"","msr_speaker":"","msr_other_contributors":"","msr_booktitle":"","msr_pages_string":"","msr_chapter":"","msr_isbn":"","msr_journal":"","msr_volume":"","msr_number":"","msr_editors":"","msr_series":"","msr_issue":"","msr_organization":"","msr_how_published":"","msr_notes":"","msr_highlight_text":"","msr_release_tracker_id":"","msr_original_fields_of_study":"","msr_download_urls":"","msr_external_url":"","msr_secondary_video_url":"","msr_longbiography":"","msr_microsoftintellectualproperty":0,"msr_main_download":"","msr_publicationurl":"","msr_doi":"10.48550\/arXiv.2607.25271","msr_publication_uploader":[{"type":"url","title":"https:\/\/dblp.org\/rec\/journals\/corr\/abs-2607-25271.html","label_id":243109,"id":false,"viewUrl":false},{"type":"url","title":"https:\/\/arxiv.org\/abs\/2607.25271","label_id":243109,"id":false,"viewUrl":false}],"msr_related_uploader":[],"msr_citation_count":2,"msr_citation_count_updated":"","msr_s2_paper_id":"026c135d2cc5ef3c98ad2d122522ff93380b6239","msr_influential_citations":2,"msr_reference_count":30,"msr_arxiv_id":"2607.25271","msr_s2_author_ids":[],"msr_s2_open_access":false,"msr_s2_pdf_url":null,"msr_attachments":[],"msr-author-ordering":[{"type":"text","value":"Tian Qin","user_id":0,"rest_url":false},{"type":"text","value":"K. Hamidieh","user_id":0,"rest_url":false},{"type":"user_nicename","value":"David Alvarez-Melis","user_id":38814,"rest_url":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=David Alvarez-Melis"}],"msr_impact_theme":[],"msr_research_lab":[199563],"msr_event":[],"msr_group":[330695],"msr_project":[],"publication":[],"video":[],"msr-tool":[],"msr_publication_type":"inproceedings","related_content":[],"_links":{"self":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1191467","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item"}],"about":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-research-item"}],"version-history":[{"count":3,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1191467\/revisions"}],"predecessor-version":[{"id":1192181,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1191467\/revisions\/1192181"}],"wp:attachment":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1191467"}],"wp:term":[{"taxonomy":"msr-research-highlight","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-highlight?post=1191467"},{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1191467"},{"taxonomy":"msr-publication-type","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-type?post=1191467"},{"taxonomy":"msr-publisher","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-publisher?post=1191467"},{"taxonomy":"msr-publication-cta","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-cta?post=1191467"},{"taxonomy":"msr-focus-area","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-focus-area?post=1191467"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1191467"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=1191467"},{"taxonomy":"msr-field-of-study","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-field-of-study?post=1191467"},{"taxonomy":"msr-conference","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-conference?post=1191467"},{"taxonomy":"msr-journal","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-journal?post=1191467"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=1191467"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=1191467"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}