{"id":"https://openalex.org/W4396821298","doi":"https://doi.org/10.48550/arxiv.2404.18922","title":"DPO Meets PPO: Reinforced Token Optimization for RLHF","display_name":"DPO Meets PPO: Reinforced Token Optimization for RLHF","publication_year":2024,"publication_date":"2024-04-29","ids":{"openalex":"https://openalex.org/W4396821298","doi":"https://doi.org/10.48550/arxiv.2404.18922"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2404.18922","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2404.18922","pdf_url":"https://arxiv.org/pdf/2404.18922","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2404.18922","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5100748564","display_name":"Zhong Han","orcid":"https://orcid.org/0000-0001-7596-1699"},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Zhong, Han","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Shan, Zikang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shan, Zikang","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5059836708","display_name":"Guhao Feng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Feng, Guhao","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5076628427","display_name":"Wei Xiong","orcid":"https://orcid.org/0000-0002-2784-7696"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xiong, Wei","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Cheng, Xinle","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Cheng, Xinle","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102619061","display_name":"Li Zhao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhao, Li","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5085634514","display_name":"Di He","orcid":"https://orcid.org/0000-0002-0719-8289"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"He, Di","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101544240","display_name":"Jiang Bian","orcid":"https://orcid.org/0000-0002-5125-4882"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Bian, Jiang","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5100406711","display_name":"Liwei Wang","orcid":"https://orcid.org/0000-0001-9970-8604"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Liwei","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":9,"corresponding_author_ids":["https://openalex.org/A5100748564"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"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/T10933","display_name":"Real-Time Systems Scheduling","score":0.9172999858856201,"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"}},"topics":[{"id":"https://openalex.org/T10933","display_name":"Real-Time Systems Scheduling","score":0.9172999858856201,"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/T12810","display_name":"Real-time simulation and control systems","score":0.9000999927520752,"subfield":{"id":"https://openalex.org/subfields/2207","display_name":"Control and Systems Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.7846695780754089},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.48399969935417175},{"id":"https://openalex.org/keywords/computer-network","display_name":"Computer network","score":0.24729230999946594}],"concepts":[{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.7846695780754089},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.48399969935417175},{"id":"https://openalex.org/C31258907","wikidata":"https://www.wikidata.org/wiki/Q1301371","display_name":"Computer network","level":1,"score":0.24729230999946594}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2404.18922","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2404.18922","pdf_url":"https://arxiv.org/pdf/2404.18922","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2404.18922","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2404.18922","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":"article"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2404.18922","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2404.18922","pdf_url":"https://arxiv.org/pdf/2404.18922","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":false},"content_urls":{"pdf":"https://content.openalex.org/works/W4396821298.pdf"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2748952813","https://openalex.org/W4388335561","https://openalex.org/W2970530566","https://openalex.org/W4288261899","https://openalex.org/W4307309205","https://openalex.org/W2967478618","https://openalex.org/W4385009901","https://openalex.org/W4385572700","https://openalex.org/W2997152889"],"abstract_inverted_index":{"In":[0,176],"the":[1,32,38,70,89,113,117,185],"classical":[2],"Reinforcement":[3],"Learning":[4],"from":[5,18,93,137],"Human":[6],"Feedback":[7],"(RLHF)":[8],"framework,":[9,78],"Proximal":[10],"Policy":[11],"Optimization":[12,85,130],"(PPO)":[13],"is":[14,47,109,152],"employed":[15],"to":[16,111],"learn":[17],"sparse,":[19],"sentence-level":[20],"rewards":[21],"--":[22],"a":[23,57,64,145],"challenging":[24],"scenario":[25],"in":[26,37],"traditional":[27],"deep":[28],"reinforcement":[29],"learning.":[30],"Despite":[31],"great":[33],"successes":[34],"of":[35,40,72,115,148],"PPO":[36,158,169,180],"alignment":[39],"large":[41],"language":[42],"models,":[43],"its":[44,122],"open-source":[45],"implementation":[46],"still":[48],"largely":[49],"sub-optimal.":[50],"To":[51],"address":[52],"these":[53],"issues,":[54],"we":[55,79],"introduce":[56,80],"framework":[58],"that":[59,164],"models":[60,198],"RLHF":[61],"problems":[62],"as":[63],"Markov":[65],"decision":[66],"process":[67],"(MDP),":[68],"enabling":[69],"capture":[71],"fine-grained":[73],"token-wise":[74,90,104,146],"information.":[75],"Under":[76],"this":[77,102],"an":[81],"algorithm":[82],"Reinforced":[83],"Token":[84],"(\\texttt{RTO}),":[86],"which":[87,151],"learns":[88],"reward":[91,105],"function":[92],"preference":[94,173],"data":[95],"and":[96,132,170,189,197],"performs":[97,166],"policy":[98,119],"optimization":[99],"based":[100],"on":[101,184,193],"learned":[103],"signal.":[106],"Theoretically,":[107],"\\texttt{RTO}":[108,125,165],"proven":[110],"have":[112],"capability":[114],"finding":[116],"near-optimal":[118],"sample-efficiently.":[120],"For":[121],"practical":[123],"implementation,":[124],"innovatively":[126],"integrates":[127],"Direct":[128],"Preference":[129],"(DPO)":[131],"PPO.":[133],"DPO,":[134],"originally":[135],"derived":[136],"sparse":[138],"sentence":[139],"rewards,":[140],"surprisingly":[141],"provides":[142],"us":[143],"with":[144],"characterization":[147],"response":[149],"quality,":[150],"seamlessly":[153],"incorporated":[154],"into":[155],"our":[156],"subsequent":[157],"training":[159],"stage.":[160],"Extensive":[161],"experiments":[162],"demonstrate":[163],"better":[167],"than":[168],"other":[171],"direct":[172],"learning":[174],"algorithms.":[175],"particular,":[177],"RTO":[178],"outperforms":[179],"by":[181,190],"7.5":[182],"points":[183,192],"AlpacaEval":[186],"2":[187],"benchmark":[188],"4.1":[191],"Arena-Hard.":[194],"Our":[195],"code":[196],"are":[199],"available":[200],"at":[201],"\\href{https://github.com/zkshan2002/RTO}{https://github.com/zkshan2002/RTO}.":[202]},"counts_by_year":[],"updated_date":"2026-03-25T23:56:10.502304","created_date":"2025-10-10T00:00:00"}