{"id":"https://openalex.org/W4416034072","doi":"https://doi.org/10.18653/v1/2025.findings-emnlp.1246","title":"Instability in Downstream Task Performance During LLM Pretraining","display_name":"Instability in Downstream Task Performance During LLM Pretraining","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4416034072","doi":"https://doi.org/10.18653/v1/2025.findings-emnlp.1246"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2025.findings-emnlp.1246","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.1246","pdf_url":"https://aclanthology.org/2025.findings-emnlp.1246.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EMNLP 2025","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.findings-emnlp.1246.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5102570223","display_name":"Yuto Nishida","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yuto Nishida","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5073805047","display_name":"Masaru Isonuma","orcid":"https://orcid.org/0009-0004-6152-178X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Masaru Isonuma","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5112645578","display_name":"Yusuke Oda","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yusuke Oda","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.31112385,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"22883","last_page":"22895"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10982","display_name":"Motor Control and Adaptation","score":0.08489999920129776,"subfield":{"id":"https://openalex.org/subfields/2805","display_name":"Cognitive Neuroscience"},"field":{"id":"https://openalex.org/fields/28","display_name":"Neuroscience"},"domain":{"id":"https://openalex.org/domains/1","display_name":"Life Sciences"}},"topics":[{"id":"https://openalex.org/T10982","display_name":"Motor Control and Adaptation","score":0.08489999920129776,"subfield":{"id":"https://openalex.org/subfields/2805","display_name":"Cognitive Neuroscience"},"field":{"id":"https://openalex.org/fields/28","display_name":"Neuroscience"},"domain":{"id":"https://openalex.org/domains/1","display_name":"Life Sciences"}},{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.05469999834895134,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T12177","display_name":"Resource-Constrained Project Scheduling","score":0.04089999943971634,"subfield":{"id":"https://openalex.org/subfields/1803","display_name":"Management Science and Operations Research"},"field":{"id":"https://openalex.org/fields/18","display_name":"Decision Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/task","display_name":"Task (project management)","score":0.550000011920929},{"id":"https://openalex.org/keywords/downstream","display_name":"Downstream (manufacturing)","score":0.43650001287460327},{"id":"https://openalex.org/keywords/control-theory","display_name":"Control theory (sociology)","score":0.34880000352859497},{"id":"https://openalex.org/keywords/instability","display_name":"Instability","score":0.31380000710487366}],"concepts":[{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.550000011920929},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4560000002384186},{"id":"https://openalex.org/C2776207758","wikidata":"https://www.wikidata.org/wiki/Q5303302","display_name":"Downstream (manufacturing)","level":2,"score":0.43650001287460327},{"id":"https://openalex.org/C47446073","wikidata":"https://www.wikidata.org/wiki/Q5165890","display_name":"Control theory (sociology)","level":3,"score":0.34880000352859497},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.34220001101493835},{"id":"https://openalex.org/C207821765","wikidata":"https://www.wikidata.org/wiki/Q405372","display_name":"Instability","level":2,"score":0.31380000710487366},{"id":"https://openalex.org/C112972136","wikidata":"https://www.wikidata.org/wiki/Q7595718","display_name":"Stability (learning theory)","level":2,"score":0.30880001187324524},{"id":"https://openalex.org/C169760540","wikidata":"https://www.wikidata.org/wiki/Q207011","display_name":"Neuroscience","level":1,"score":0.2793000042438507},{"id":"https://openalex.org/C99508421","wikidata":"https://www.wikidata.org/wiki/Q2678675","display_name":"Physical medicine and rehabilitation","level":1,"score":0.25209999084472656},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.24230000376701355}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.findings-emnlp.1246","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.1246","pdf_url":"https://aclanthology.org/2025.findings-emnlp.1246.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EMNLP 2025","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.findings-emnlp.1246","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.1246","pdf_url":"https://aclanthology.org/2025.findings-emnlp.1246.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EMNLP 2025","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320320912","display_name":"Ministry of Education, Culture, Sports, Science and Technology","ror":"https://ror.org/048rj2z13"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4416034072.pdf","grobid_xml":"https://content.openalex.org/works/W4416034072.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"When":[0],"training":[1,17,125],"large":[2],"language":[3],"models":[4],"(LLMs),":[5],"it":[6,35],"is":[7],"common":[8],"practice":[9],"to":[10,37,123],"track":[11],"downstream":[12,28,55,116],"task":[13,56,68],"performance":[14,57,105,117],"throughout":[15,72],"the":[16,21,24,39,44,52,76,97,124],"process":[18],"and":[19,78,93,110],"select":[20],"checkpoint":[22,40,88,91],"with":[23],"highest":[25],"validation":[26],"score.However,":[27],"metrics":[29],"often":[30],"exhibit":[31],"substantial":[32],"fluctuations,":[33],"making":[34],"difficult":[36],"identify":[38],"that":[41,67,99,112],"truly":[42],"represents":[43],"best-performing":[45],"model.In":[46],"this":[47,82],"study,":[48],"we":[49,84],"empirically":[50,109],"analyze":[51],"stability":[53,118],"of":[54],"in":[58],"an":[59],"LLM":[60],"trained":[61],"on":[62],"diverse":[63],"web-scale":[64],"corpora.We":[65],"find":[66],"scores":[69],"frequently":[70],"fluctuate":[71],"training,":[73],"both":[74,108],"at":[75],"aggregate":[77],"example":[79],"levels.To":[80],"address":[81],"instability,":[83],"investigate":[85],"two":[86],"post-hoc":[87],"integration":[89],"methods:":[90],"averaging":[92],"ensemble,":[94],"motivated":[95],"by":[96],"hypothesis":[98],"aggregating":[100],"neighboring":[101],"checkpoints":[102],"can":[103],"reduce":[104],"volatility.We":[106],"demonstrate":[107],"theoretically":[111],"these":[113],"methods":[114],"improve":[115],"without":[119],"requiring":[120],"any":[121],"changes":[122],"procedure.":[126]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2025-11-08T00:00:00"}