{"id":"https://openalex.org/W7160491179","doi":"https://doi.org/10.48550/arxiv.2605.04431","title":"Towards Robust LLM Post-Training: Automatic Failure Management for Reinforcement Fine-Tuning","display_name":"Towards Robust LLM Post-Training: Automatic Failure Management for Reinforcement Fine-Tuning","publication_year":2026,"publication_date":"2026-05-06","ids":{"openalex":"https://openalex.org/W7160491179","doi":"https://doi.org/10.48550/arxiv.2605.04431"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.04431","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.04431","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":null,"license_id":null,"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.2605.04431","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5135593022","display_name":"Lingzhe Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Lingzhe","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135567398","display_name":"Tong Jia","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jia, Tong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135608778","display_name":"Yunpeng Zhai","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhai, Yunpeng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5113315265","display_name":"Liancheng Fang","orcid":"https://orcid.org/0009-0001-2722-2979"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Fang, Liancheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135589527","display_name":"Kening Zheng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zheng, Kening","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135566299","display_name":"Hongyi Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Hongyi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135584160","display_name":"Xiaosong Huang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Xiaosong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135631330","display_name":"Philip S. Yu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yu, Philip S.","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5135613707","display_name":"Ying Li","orcid":"https://orcid.org/0000-0002-7595-2474"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Ying","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/T10743","display_name":"Software Testing and Debugging Techniques","score":0.2840999960899353,"subfield":{"id":"https://openalex.org/subfields/1712","display_name":"Software"},"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/T10743","display_name":"Software Testing and Debugging Techniques","score":0.2840999960899353,"subfield":{"id":"https://openalex.org/subfields/1712","display_name":"Software"},"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/T10260","display_name":"Software Engineering Research","score":0.1860000044107437,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"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/T12127","display_name":"Software System Performance and Reliability","score":0.1639000028371811,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"Reinforcement learning","score":0.63919997215271},{"id":"https://openalex.org/keywords/reliability","display_name":"Reliability (semiconductor)","score":0.4864000082015991},{"id":"https://openalex.org/keywords/reinforcement","display_name":"Reinforcement","score":0.4742000102996826},{"id":"https://openalex.org/keywords/process","display_name":"Process (computing)","score":0.4733000099658966},{"id":"https://openalex.org/keywords/benchmark","display_name":"Benchmark (surveying)","score":0.4268999993801117},{"id":"https://openalex.org/keywords/fault","display_name":"Fault (geology)","score":0.3635999858379364},{"id":"https://openalex.org/keywords/construct","display_name":"Construct (python library)","score":0.35659998655319214},{"id":"https://openalex.org/keywords/fault-management","display_name":"Fault management","score":0.351500004529953}],"concepts":[{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.63919997215271},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5376999974250793},{"id":"https://openalex.org/C43214815","wikidata":"https://www.wikidata.org/wiki/Q7310987","display_name":"Reliability (semiconductor)","level":3,"score":0.4864000082015991},{"id":"https://openalex.org/C67203356","wikidata":"https://www.wikidata.org/wiki/Q1321905","display_name":"Reinforcement","level":2,"score":0.4742000102996826},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.4733000099658966},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.4268999993801117},{"id":"https://openalex.org/C200601418","wikidata":"https://www.wikidata.org/wiki/Q2193887","display_name":"Reliability engineering","level":1,"score":0.36880001425743103},{"id":"https://openalex.org/C175551986","wikidata":"https://www.wikidata.org/wiki/Q47089","display_name":"Fault (geology)","level":2,"score":0.3635999858379364},{"id":"https://openalex.org/C2780801425","wikidata":"https://www.wikidata.org/wiki/Q5164392","display_name":"Construct (python library)","level":2,"score":0.35659998655319214},{"id":"https://openalex.org/C108074857","wikidata":"https://www.wikidata.org/wiki/Q3067360","display_name":"Fault management","level":3,"score":0.351500004529953},{"id":"https://openalex.org/C739882","wikidata":"https://www.wikidata.org/wiki/Q3560506","display_name":"Anomaly detection","level":2,"score":0.3481000065803528},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.3434999883174896},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.33090001344680786},{"id":"https://openalex.org/C120936955","wikidata":"https://www.wikidata.org/wiki/Q2155640","display_name":"Empirical research","level":2,"score":0.32760000228881836},{"id":"https://openalex.org/C2164484","wikidata":"https://www.wikidata.org/wiki/Q5170150","display_name":"Core (optical fiber)","level":2,"score":0.29580000042915344},{"id":"https://openalex.org/C112930515","wikidata":"https://www.wikidata.org/wiki/Q4389547","display_name":"Risk analysis (engineering)","level":1,"score":0.2775999903678894},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.2721000015735626},{"id":"https://openalex.org/C84945661","wikidata":"https://www.wikidata.org/wiki/Q7366567","display_name":"Root cause","level":2,"score":0.2685000002384186},{"id":"https://openalex.org/C32896092","wikidata":"https://www.wikidata.org/wiki/Q189447","display_name":"Risk management","level":2,"score":0.2621000111103058},{"id":"https://openalex.org/C107094494","wikidata":"https://www.wikidata.org/wiki/Q428453","display_name":"Fault tree analysis","level":2,"score":0.2563000023365021},{"id":"https://openalex.org/C28944875","wikidata":"https://www.wikidata.org/wiki/Q1925224","display_name":"Failure causes","level":2,"score":0.25459998846054077},{"id":"https://openalex.org/C63540848","wikidata":"https://www.wikidata.org/wiki/Q3140932","display_name":"Fault tolerance","level":2,"score":0.25459998846054077},{"id":"https://openalex.org/C12997251","wikidata":"https://www.wikidata.org/wiki/Q567560","display_name":"Anomaly (physics)","level":2,"score":0.25369998812675476}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.04431","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.04431","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.04431","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.04431","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[{"display_name":"Peace, Justice and strong institutions","score":0.5552486777305603,"id":"https://metadata.un.org/sdg/16"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Reinforcement":[0],"fine-tuning":[1,172],"(RFT)":[2],"has":[3],"become":[4],"a":[5,83,137,183],"core":[6],"paradigm":[7],"for":[8,73,90,108,170],"post-training":[9],"large":[10],"language":[11],"models,":[12],"yet":[13],"its":[14],"training":[15,56,122,149],"process":[16],"remains":[17,75],"highly":[18],"fragile.":[19],"Existing":[20],"efforts":[21],"mainly":[22],"improve":[23],"reliability":[24],"at":[25,51],"the":[26,46,52,95,105],"system":[27],"level":[28],"or":[29],"address":[30],"specific":[31],"issues":[32],"in":[33,111,182,216],"individual":[34],"subproblems":[35],"by":[36],"modifying":[37],"RFT":[38,74,99,143,221],"algorithms.":[39],"Despite":[40],"their":[41,154],"effectiveness,":[42],"they":[43],"largely":[44,76],"overlook":[45],"problem":[47],"of":[48,98],"failure":[49,71,88,167,177],"management":[50,72,89,168],"training-process":[53],"level.":[54],"When":[55],"goes":[57],"wrong,":[58],"practitioners":[59],"still":[60,202],"rely":[61],"heavily":[62],"on":[63,132,159],"expert-driven":[64],"manual":[65],"inspection":[66],"and":[67,69,127,151,179,219],"correction,":[68],"automatic":[70,166],"unexplored.":[77],"In":[78],"this":[79,133],"paper,":[80],"we":[81,101,135,162],"take":[82],"first":[84,102,106],"step":[85],"toward":[86],"systematic":[87],"reinforcement":[91,112,171],"fine-tuning.":[92],"To":[93],"understand":[94],"empirical":[96,139,155],"structure":[97,200],"failures,":[100],"construct":[103],"RFT-FaultBench,":[104],"benchmark":[107],"fine-grained":[109],"failures":[110,144],"fine-tuning,":[113],"covering":[114],"5":[115],"fault":[116,119,156,209],"families,":[117],"16":[118],"types,":[120],"779":[121],"runs,":[123],"22,549":[124],"train-step":[125],"records,":[126],"1,457,288":[128],"trajectory-level":[129],"records.":[130],"Based":[131],"benchmark,":[134],"conduct":[136],"comprehensive":[138],"study":[140],"showing":[141],"that":[142,173,189],"are":[145],"both":[146],"observable":[147],"from":[148],"dynamics":[150],"distinguishable":[152],"through":[153],"fingerprints.":[157],"Building":[158],"these":[160],"findings,":[161],"propose":[163],"RFT-FM,":[164],"an":[165],"framework":[169],"unifies":[174],"anomaly":[175,199],"detection,":[176],"diagnosis,":[178],"auto":[180],"remediation":[181],"closed":[184],"loop.":[185],"Experimental":[186],"results":[187],"show":[188],"RFT-FaultBench":[190],"is":[191],"neither":[192],"trivial":[193],"nor":[194],"saturated:":[195],"it":[196],"exhibits":[197],"clear":[198],"while":[201],"posing":[203],"substantial":[204],"challenges,":[205],"especially":[206],"under":[207],"subtle":[208],"settings.":[210],"Moreover,":[211],"RFT-FM":[212],"shows":[213],"strong":[214],"capability":[215],"detecting,":[217],"diagnosing,":[218],"mitigating":[220],"failures.":[222]},"counts_by_year":[],"updated_date":"2026-07-01T08:55:40.977307","created_date":"2026-05-08T00:00:00"}