{"id":"https://openalex.org/W7151089626","doi":"https://doi.org/10.48550/arxiv.2604.02650","title":"Revealing the Learning Dynamics of Long-Context Continual Pre-training","display_name":"Revealing the Learning Dynamics of Long-Context Continual Pre-training","publication_year":2026,"publication_date":"2026-04-03","ids":{"openalex":"https://openalex.org/W7151089626","doi":"https://doi.org/10.48550/arxiv.2604.02650"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.02650","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.02650","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2604.02650","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5016559205","display_name":"Yupu Liang","orcid":"https://orcid.org/0000-0003-4074-463X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liang, Yupu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133024156","display_name":"Shuang Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Shuang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133040973","display_name":"Guanwei Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Guanwei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133061601","display_name":"Shaolei Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Shaolei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5110447612","display_name":"Suncong Zheng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zheng, Suncong","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":null,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.30489999055862427,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.30489999055862427,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"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/T12607","display_name":"Personal Information Management and User Behavior","score":0.07919999957084656,"subfield":{"id":"https://openalex.org/subfields/1802","display_name":"Information Systems and Management"},"field":{"id":"https://openalex.org/fields/18","display_name":"Decision Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T10918","display_name":"Memory Processes and Influences","score":0.0494999997317791,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/probabilistic-logic","display_name":"Probabilistic logic","score":0.6301000118255615},{"id":"https://openalex.org/keywords/dynamics","display_name":"Dynamics (music)","score":0.4616999924182892},{"id":"https://openalex.org/keywords/training","display_name":"Training (meteorology)","score":0.4269999861717224},{"id":"https://openalex.org/keywords/adaptation","display_name":"Adaptation (eye)","score":0.42010000348091125},{"id":"https://openalex.org/keywords/perspective","display_name":"Perspective (graphical)","score":0.41830000281333923},{"id":"https://openalex.org/keywords/training-set","display_name":"Training set","score":0.37709999084472656},{"id":"https://openalex.org/keywords/focus","display_name":"Focus (optics)","score":0.3718000054359436},{"id":"https://openalex.org/keywords/downstream","display_name":"Downstream (manufacturing)","score":0.3116999864578247}],"concepts":[{"id":"https://openalex.org/C49937458","wikidata":"https://www.wikidata.org/wiki/Q2599292","display_name":"Probabilistic logic","level":2,"score":0.6301000118255615},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6175000071525574},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.5212000012397766},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4878999888896942},{"id":"https://openalex.org/C145912823","wikidata":"https://www.wikidata.org/wiki/Q113558","display_name":"Dynamics (music)","level":2,"score":0.4616999924182892},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.4269999861717224},{"id":"https://openalex.org/C139807058","wikidata":"https://www.wikidata.org/wiki/Q352374","display_name":"Adaptation (eye)","level":2,"score":0.42010000348091125},{"id":"https://openalex.org/C12713177","wikidata":"https://www.wikidata.org/wiki/Q1900281","display_name":"Perspective (graphical)","level":2,"score":0.41830000281333923},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.37709999084472656},{"id":"https://openalex.org/C192209626","wikidata":"https://www.wikidata.org/wiki/Q190909","display_name":"Focus (optics)","level":2,"score":0.3718000054359436},{"id":"https://openalex.org/C2776207758","wikidata":"https://www.wikidata.org/wiki/Q5303302","display_name":"Downstream (manufacturing)","level":2,"score":0.3116999864578247},{"id":"https://openalex.org/C2777303404","wikidata":"https://www.wikidata.org/wiki/Q759757","display_name":"Convergence (economics)","level":2,"score":0.2994000017642975},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.29350000619888306},{"id":"https://openalex.org/C2775936607","wikidata":"https://www.wikidata.org/wiki/Q466845","display_name":"Tracking (education)","level":2,"score":0.2872999906539917},{"id":"https://openalex.org/C48103436","wikidata":"https://www.wikidata.org/wiki/Q599031","display_name":"State (computer science)","level":2,"score":0.2849000096321106},{"id":"https://openalex.org/C176777502","wikidata":"https://www.wikidata.org/wiki/Q4774623","display_name":"Anticipation (artificial intelligence)","level":2,"score":0.26809999346733093},{"id":"https://openalex.org/C112930515","wikidata":"https://www.wikidata.org/wiki/Q4389547","display_name":"Risk analysis (engineering)","level":1,"score":0.26249998807907104},{"id":"https://openalex.org/C77405623","wikidata":"https://www.wikidata.org/wiki/Q598451","display_name":"System dynamics","level":2,"score":0.2612000107765198},{"id":"https://openalex.org/C2776459999","wikidata":"https://www.wikidata.org/wiki/Q2119376","display_name":"Fidelity","level":2,"score":0.2574999928474426},{"id":"https://openalex.org/C114289077","wikidata":"https://www.wikidata.org/wiki/Q3284399","display_name":"Statistical model","level":2,"score":0.2547999918460846},{"id":"https://openalex.org/C2522767166","wikidata":"https://www.wikidata.org/wiki/Q2374463","display_name":"Data science","level":1,"score":0.25130000710487366}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.02650","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.02650","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2604.02650","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.02650","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[{"display_name":"Decent work and economic growth","score":0.5924726128578186,"id":"https://metadata.un.org/sdg/8"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Existing":[0],"studies":[1],"on":[2,9,45],"Long-Context":[3],"Continual":[4],"Pre-training":[5],"(LCCP)":[6],"mainly":[7],"focus":[8],"small-scale":[10,27],"models":[11,31],"and":[12,35,59,110,168,198,214],"limited":[13],"data":[14],"regimes":[15,125],"(tens":[16],"of":[17,19,74,120,126,128,130,220],"billions":[18,129],"tokens).":[20,146],"We":[21],"argue":[22],"that":[23],"directly":[24],"migrating":[25],"these":[26],"settings":[28],"to":[29,53,62,99],"industrial-grade":[30,80,135,221],"risks":[32],"insufficient":[33,133],"adaptation":[34],"premature":[36],"training":[37,91,143,187],"termination.":[38],"Furthermore,":[39],"current":[40],"evaluation":[41,212],"methods":[42],"rely":[43],"heavily":[44],"downstream":[46,173],"benchmarks":[47],"(e.g.,":[48,138],"Needle-in-a-Haystack),":[49],"which":[50],"often":[51],"fail":[52],"reflect":[54],"the":[55,70,79,218],"intrinsic":[56,166],"convergence":[57],"state":[58],"can":[60],"lead":[61],"\"deceptive":[63],"saturation\".":[64],"In":[65],"this":[66],"paper,":[67],"we":[68,94],"present":[69],"first":[71],"systematic":[72],"investigation":[73],"LCCP":[75,101,137,196,219],"learning":[76],"dynamics":[77,102],"using":[78],"Hunyuan-A13B":[81,139],"(80B":[82],"total":[83],"parameters),":[84],"tracking":[85],"its":[86],"evolution":[87],"across":[88,103],"a":[89,96,208],"200B-token":[90],"trajectory.":[92],"Specifically,":[93],"propose":[95],"hierarchical":[97],"framework":[98],"analyze":[100],"behavioral":[104],"(supervised":[105],"fine-tuning":[106],"probing),":[107],"probabilistic":[108],"(perplexity),":[109],"mechanistic":[111,215],"(attention":[112],"patterns)":[113],"levels.":[114],"Our":[115],"findings":[116],"reveal:":[117],"(1)":[118],"Necessity":[119],"Massive":[121],"Data":[122],"Scaling:":[123],"Training":[124,179],"dozens":[127],"tokens":[131],"are":[132],"for":[134,178,217],"LLMs'":[136],"reaches":[140],"saturation":[141],"after":[142],"over":[144],"150B":[145],"(2)":[147],"Deceptive":[148],"Saturation":[149],"vs.":[150],"Intrinsic":[151],"Saturation:":[152],"Traditional":[153],"NIAH":[154],"scores":[155,193],"report":[156],"\"fake":[157],"saturation\"":[158],"early,":[159],"while":[160],"our":[161],"PPL-based":[162],"analysis":[163],"reveals":[164],"continuous":[165],"improvements":[167],"correlates":[169],"more":[170],"strongly":[171],"with":[172,202],"performance.":[174],"(3)":[175],"Mechanistic":[176],"Monitoring":[177],"Stability:":[180],"Retrieval":[181],"heads":[182],"act":[183],"as":[184,189],"efficient,":[185],"low-resource":[186],"monitors,":[188],"their":[190],"evolving":[191],"attention":[192],"reliably":[194],"track":[195],"progress":[197],"exhibit":[199],"high":[200],"correlation":[201],"SFT":[203],"results.":[204],"This":[205],"work":[206],"provides":[207],"comprehensive":[209],"monitoring":[210],"framework,":[211],"system,":[213],"interpretation":[216],"LLM.":[222]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-04-07T00:00:00"}