{"id":"https://openalex.org/W7125147729","doi":"https://doi.org/10.48550/arxiv.2601.14243","title":"Jet-RL: Enabling On-Policy FP8 Reinforcement Learning with Unified Training and Rollout Precision Flow","display_name":"Jet-RL: Enabling On-Policy FP8 Reinforcement Learning with Unified Training and Rollout Precision Flow","publication_year":2026,"publication_date":"2026-01-20","ids":{"openalex":"https://openalex.org/W7125147729","doi":"https://doi.org/10.48550/arxiv.2601.14243"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2601.14243","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2601.14243","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":"article"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2601.14243","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5109693865","display_name":"Haocheng Xi","orcid":null},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Xi, Haocheng","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5123482337","display_name":"Charlie Ruan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ruan, Charlie","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Liao, Peiyuan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liao, Peiyuan","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5123467203","display_name":"Yujun Lin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lin, Yujun","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5123490424","display_name":"Han Cai","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Cai, Han","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5123532014","display_name":"Yilong Zhao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhao, Yilong","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5123505838","display_name":"Shuo Yang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yang, Shuo","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Keutzer, Kurt","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Keutzer, Kurt","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5123534921","display_name":"Song Han","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Han, Song","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5123524966","display_name":"Ligeng Zhu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhu, Ligeng","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":10,"corresponding_author_ids":["https://openalex.org/A5109693865"],"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/T10028","display_name":"Topic Modeling","score":0.2865000069141388,"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/T10028","display_name":"Topic Modeling","score":0.2865000069141388,"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/T10462","display_name":"Reinforcement Learning in Robotics","score":0.18330000340938568,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.13289999961853027,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/speedup","display_name":"Speedup","score":0.9384999871253967},{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"Reinforcement learning","score":0.7062000036239624},{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"Reinforcement learning","score":0.7062000036239624},{"id":"https://openalex.org/keywords/training","display_name":"Training (meteorology)","score":0.6093000173568726},{"id":"https://openalex.org/keywords/training","display_name":"Training (meteorology)","score":0.6093000173568726},{"id":"https://openalex.org/keywords/convergence","display_name":"Convergence (economics)","score":0.5945000052452087},{"id":"https://openalex.org/keywords/convergence","display_name":"Convergence (economics)","score":0.5945000052452087},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.5920000076293945},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.5920000076293945}],"concepts":[{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.9384999871253967},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7857999801635742},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7857999801635742},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.7062000036239624},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.7062000036239624},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.6093000173568726},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.6093000173568726},{"id":"https://openalex.org/C2777303404","wikidata":"https://www.wikidata.org/wiki/Q759757","display_name":"Convergence (economics)","level":2,"score":0.5945000052452087},{"id":"https://openalex.org/C2777303404","wikidata":"https://www.wikidata.org/wiki/Q759757","display_name":"Convergence (economics)","level":2,"score":0.5945000052452087},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.5920000076293945},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.5920000076293945},{"id":"https://openalex.org/C38349280","wikidata":"https://www.wikidata.org/wiki/Q1434290","display_name":"Flow (mathematics)","level":2,"score":0.4447000026702881},{"id":"https://openalex.org/C38349280","wikidata":"https://www.wikidata.org/wiki/Q1434290","display_name":"Flow (mathematics)","level":2,"score":0.4447000026702881},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.4309999942779541},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.4309999942779541},{"id":"https://openalex.org/C112972136","wikidata":"https://www.wikidata.org/wiki/Q7595718","display_name":"Stability (learning theory)","level":2,"score":0.4196999967098236},{"id":"https://openalex.org/C112972136","wikidata":"https://www.wikidata.org/wiki/Q7595718","display_name":"Stability (learning theory)","level":2,"score":0.4196999967098236},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.38679999113082886},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.38679999113082886},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.373199999332428},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.373199999332428},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.3005000054836273},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.3005000054836273},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.2937999963760376},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.2937999963760376},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.26010000705718994},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.26010000705718994}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2601.14243","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2601.14243","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":"article"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2601.14243","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2601.14243","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":"article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Reinforcement":[0],"learning":[1],"(RL)":[2],"is":[3,151],"essential":[4],"for":[5,31,66,159,172],"enhancing":[6],"the":[7,27,73,84,114,118,170,179,191,199],"complex":[8],"reasoning":[9],"capabilities":[10],"of":[11,34,77,117,181],"large":[12],"language":[13],"models":[14],"(LLMs).":[15],"However,":[16],"existing":[17],"RL":[18,39,79,138,146],"training":[19,36,80,94,126,139,161,200],"pipelines":[20],"are":[21],"computationally":[22],"inefficient":[23,173],"and":[24,81,96,103,127,144,162,168,202,217],"resource-intensive,":[25],"with":[26],"rollout":[28,61,192],"phase":[29],"accounting":[30],"over":[32,207],"70%":[33],"total":[35],"time.":[37],"Quantized":[38],"training,":[40,209],"particularly":[41],"using":[42],"FP8":[43,58,78,137,156],"precision,":[44],"offers":[45],"a":[46,154,203],"promising":[47],"approach":[48],"to":[49,152,187,195],"mitigating":[50],"this":[51,69],"bottleneck.":[52],"A":[53],"commonly":[54],"adopted":[55],"strategy":[56,90],"applies":[57],"precision":[59,65,157],"during":[60],"while":[62,210],"retaining":[63],"BF16":[64,208],"training.":[67],"In":[68],"work,":[70],"we":[71,133],"present":[72],"first":[74],"comprehensive":[75],"study":[76],"demonstrate":[82],"that":[83,109,141],"widely":[85],"used":[86],"BF16-training":[87],"+":[88],"FP8-rollout":[89],"suffers":[91],"from":[92,113],"severe":[93],"instability":[95],"catastrophic":[97],"accuracy":[98,220],"collapse":[99],"under":[100],"long-horizon":[101],"rollouts":[102],"challenging":[104],"tasks.":[105],"Our":[106],"analysis":[107],"shows":[108],"these":[110,131],"failures":[111],"stem":[112],"off-policy":[115],"nature":[116],"approach,":[119],"which":[120],"introduces":[121],"substantial":[122],"numerical":[123,166],"mismatch":[124],"between":[125],"inference.":[128],"Motivated":[129],"by":[130],"observations,":[132],"propose":[134],"Jet-RL,":[135],"an":[136],"framework":[140],"enables":[142],"robust":[143],"stable":[145,212],"optimization.":[147],"The":[148],"key":[149],"idea":[150],"adopt":[153],"unified":[155],"flow":[158],"both":[160],"rollout,":[163],"thereby":[164],"minimizing":[165],"discrepancies":[167],"eliminating":[169],"need":[171],"inter-step":[174],"calibration.":[175],"Extensive":[176],"experiments":[177],"validate":[178],"effectiveness":[180],"Jet-RL:":[182],"our":[183],"method":[184],"achieves":[185],"up":[186,194],"33%":[188],"speedup":[189,197,206],"in":[190,198],"phase,":[193,201],"41%":[196],"16%":[204],"end-to-end":[205],"maintaining":[211],"convergence":[213],"across":[214],"all":[215],"settings":[216],"incurring":[218],"negligible":[219],"degradation.":[221]},"counts_by_year":[],"updated_date":"2026-03-25T23:56:10.502304","created_date":"2026-01-22T00:00:00"}