In recent years, Large Language Models (LLMs) have shown remarkable performance in generating human-like text, proving to be a valuable asset across various applications. However, adapting these models to incorporate new, out-of-domain knowledge remains a challenge, particularly for facts and events that occur after the model’s knowledge cutoff date. This paper investigates the effectiveness of Supervised Fine-Tuning (SFT) as a method for knowledge injection in LLMs, specifically focusing on the domain of recent sporting events. We compare different dataset generation strategies — token-based and fact-based scaling — to create training data that helps the model learn new information. Our experiments on GPT-4 demonstrate that while token-based scaling can lead to improvements in Q&A accuracy, it may not provide uniform coverage of new knowledge. Fact-based scaling, on the other hand, offers a more systematic approach to ensure even coverage across all facts. We present a novel dataset generation process that leads to more effective knowledge ingestion through SFT, and our results show considerable performance improvements in Q&A tasks related to out-of-domain knowledge. This study contributes to the understanding of domain adaptation for LLMs and highlights the potential of SFT in enhancing the factuality of LLM responses in specific knowledge domains.<\/p>\n","protected":false},"excerpt":{"rendered":"
In recent years, Large Language Models (LLMs) have shown remarkable performance in generating human-like text, proving to be a valuable asset across various applications. However, adapting these models to incorporate new, out-of-domain knowledge remains a challenge, particularly for facts and events that occur after the model’s knowledge cutoff date. This paper investigates the effectiveness of […]<\/p>\n","protected":false},"featured_media":0,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"footnotes":""},"msr-content-type":[3],"msr-research-highlight":[],"research-area":[13556,13563],"msr-publication-type":[193715],"msr-product-type":[],"msr-focus-area":[],"msr-platform":[],"msr-download-source":[],"msr-locale":[268875],"msr-field-of-study":[268089],"msr-conference":[],"msr-journal":[268188],"msr-impact-theme":[],"msr-pillar":[],"class_list":["post-1021485","msr-research-item","type-msr-research-item","status-publish","hentry","msr-research-area-artificial-intelligence","msr-research-area-data-platform-analytics","msr-locale-en_us","msr-field-of-study-large-language-models"],"msr_publishername":"","msr_edition":"","msr_affiliation":"","msr_published_date":"2024-4","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":1,"msr_main_download":"","msr_publicationurl":"","msr_doi":"","msr_publication_uploader":[{"type":"url","viewUrl":"false","id":"false","title":"https:\/\/arxiv.org\/abs\/2404.00213","label_id":"243109","label":0}],"msr_related_uploader":"","msr_attachments":[],"msr-author-ordering":[{"type":"text","value":"Nick Mecklenburg","user_id":0,"rest_url":false},{"type":"text","value":"Yiyou Lin","user_id":0,"rest_url":false},{"type":"text","value":"Xiaoxiao Li","user_id":0,"rest_url":false},{"type":"text","value":"Daniel Holstein","user_id":0,"rest_url":false},{"type":"user_nicename","value":"Leonardo Nunes","user_id":40759,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Leonardo Nunes"},{"type":"user_nicename","value":"Sara Malvar","user_id":40753,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Sara Malvar"},{"type":"user_nicename","value":"Bruno Silva","user_id":42309,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Bruno Silva"},{"type":"user_nicename","value":"Ranveer Chandra","user_id":33344,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Ranveer Chandra"},{"type":"text","value":"Vijay Aski","user_id":0,"rest_url":false},{"type":"text","value":"Pavan Kumar Reddy Yannam","user_id":0,"rest_url":false},{"type":"user_nicename","value":"Tolga Aktas","user_id":42984,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Tolga Aktas"}],"msr_impact_theme":[],"msr_research_lab":[199565],"msr_event":[],"msr_group":[714067],"msr_project":[1018536,881235],"publication":[],"video":[],"download":[],"msr_publication_type":"article","_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1021485"}],"collection":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item"}],"about":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-research-item"}],"version-history":[{"count":1,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1021485\/revisions"}],"predecessor-version":[{"id":1021500,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1021485\/revisions\/1021500"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1021485"}],"wp:term":[{"taxonomy":"msr-content-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-content-type?post=1021485"},{"taxonomy":"msr-research-highlight","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-highlight?post=1021485"},{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1021485"},{"taxonomy":"msr-publication-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-type?post=1021485"},{"taxonomy":"msr-product-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-product-type?post=1021485"},{"taxonomy":"msr-focus-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-focus-area?post=1021485"},{"taxonomy":"msr-platform","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-platform?post=1021485"},{"taxonomy":"msr-download-source","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-download-source?post=1021485"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1021485"},{"taxonomy":"msr-field-of-study","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-field-of-study?post=1021485"},{"taxonomy":"msr-conference","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-conference?post=1021485"},{"taxonomy":"msr-journal","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-journal?post=1021485"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=1021485"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=1021485"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}