{"id":"https://openalex.org/W7162692287","doi":"https://doi.org/10.48550/arxiv.2605.28020","title":"The Missing Piece in Pre-trained Model Evaluation: Reward-Guided Decoding Unlocks Task-Oriented Behavior Without Parameter Updates","display_name":"The Missing Piece in Pre-trained Model Evaluation: Reward-Guided Decoding Unlocks Task-Oriented Behavior Without Parameter Updates","publication_year":2026,"publication_date":"2026-05-27","ids":{"openalex":"https://openalex.org/W7162692287","doi":"https://doi.org/10.48550/arxiv.2605.28020"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.28020","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.28020","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.2605.28020","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5137215047","display_name":"Shaobo Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Shaobo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137192699","display_name":"Guo Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Guo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137209412","display_name":"Ziyue Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Ziyue","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137292925","display_name":"Zhengyang Tang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tang, Zhengyang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137239645","display_name":"Qingyang Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Qingyang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5054086732","display_name":"Xingzhang Ren","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ren, Xingzhang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137282104","display_name":"Dayiheng Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Dayiheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5137204982","display_name":"Linfeng Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Linfeng","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/T10028","display_name":"Topic Modeling","score":0.3993000090122223,"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":0.3993000090122223,"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":0.06239999830722809,"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"}},{"id":"https://openalex.org/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.05339999869465828,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/decoding-methods","display_name":"Decoding methods","score":0.833299994468689},{"id":"https://openalex.org/keywords/benchmark","display_name":"Benchmark (surveying)","score":0.6406000256538391},{"id":"https://openalex.org/keywords/conflation","display_name":"Conflation","score":0.597000002861023},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.41280001401901245},{"id":"https://openalex.org/keywords/similarity","display_name":"Similarity (geometry)","score":0.4041000008583069},{"id":"https://openalex.org/keywords/latency","display_name":"Latency (audio)","score":0.385699987411499},{"id":"https://openalex.org/keywords/base","display_name":"Base (topology)","score":0.34279999136924744}],"concepts":[{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.833299994468689},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7245000004768372},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.6406000256538391},{"id":"https://openalex.org/C130440534","wikidata":"https://www.wikidata.org/wiki/Q14946528","display_name":"Conflation","level":2,"score":0.597000002861023},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.44279998540878296},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.41280001401901245},{"id":"https://openalex.org/C103278499","wikidata":"https://www.wikidata.org/wiki/Q254465","display_name":"Similarity (geometry)","level":3,"score":0.4041000008583069},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.385699987411499},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.34709998965263367},{"id":"https://openalex.org/C42058472","wikidata":"https://www.wikidata.org/wiki/Q810214","display_name":"Base (topology)","level":2,"score":0.34279999136924744},{"id":"https://openalex.org/C108650721","wikidata":"https://www.wikidata.org/wiki/Q1783253","display_name":"Counterfactual thinking","level":2,"score":0.3075999915599823},{"id":"https://openalex.org/C28719098","wikidata":"https://www.wikidata.org/wiki/Q44946","display_name":"Point (geometry)","level":2,"score":0.3019999861717224},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.29030001163482666},{"id":"https://openalex.org/C204397858","wikidata":"https://www.wikidata.org/wiki/Q4437907","display_name":"List decoding","level":5,"score":0.2840999960899353},{"id":"https://openalex.org/C2989277270","wikidata":"https://www.wikidata.org/wiki/Q168338","display_name":"Behavioral analysis","level":2,"score":0.2621999979019165},{"id":"https://openalex.org/C18483071","wikidata":"https://www.wikidata.org/wiki/Q168432","display_name":"Anchoring","level":2,"score":0.25619998574256897},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.25130000710487366}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.28020","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.28020","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.2605.28020","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.28020","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":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"With":[0],"the":[1,11,137],"rapid":[2],"progress":[3],"of":[4,13,169],"large":[5],"language":[6],"models":[7,26,179],"(LLMs),":[8],"reliably":[9],"evaluating":[10],"capabilities":[12],"pre-trained":[14,25,111,138],"LLMs":[15,112],"has":[16],"become":[17],"increasingly":[18],"important.":[19],"The":[20],"challenge":[21],"is":[22],"that":[23,148],"base":[24],"are":[27],"optimized":[28],"for":[29,105],"next-token":[30],"prediction":[31],"and":[32,45,87,116,163,180,202],"often":[33,68],"fail":[34],"to":[35,60,77,136,160,189,198,205],"follow":[36],"instructions":[37],"or":[38],"produce":[39,61],"well-formed":[40],"answers":[41],"under":[42],"standard":[43],"prompting":[44],"direct":[46],"decoding.":[47],"As":[48],"a":[49,101,142,165],"result,":[50],"benchmark":[51],"performance":[52],"can":[53,84],"conflate":[54],"model":[55,139],"capability":[56],"with":[57,122],"decoding-induced":[58],"failures":[59],"task-oriented":[62,107],"outputs,":[63],"while":[64,133],"exposing":[65],"such":[66,82],"behavior":[67],"relies":[69],"on":[70,185],"costly":[71],"post-training.":[72],"Recent":[73],"decodingonly":[74],"approaches":[75],"attempt":[76],"reshape":[78],"output":[79],"distributions,":[80],"but":[81],"methods":[83],"be":[85],"inefficient":[86],"brittle":[88],"across":[89,113,177],"open-ended":[90,115],"tasks.":[91,118],"To":[92],"address":[93],"these":[94],"limitations,":[95],"we":[96],"propose":[97],"Energy-Based":[98],"Decoding":[99],"(EBD),":[100],"training-free,":[102],"reward-guided":[103],"framework":[104],"activating":[106],"behaviors":[108],"from":[109,187],"frozen":[110],"both":[114],"objective":[117],"EBD":[119,149,174],"augments":[120],"decoding":[121,200],"an":[123],"external":[124],"lightweight":[125],"reward":[126],"model,":[127],"steering":[128],"generations":[129],"toward":[130,153],"high-utility":[131],"responses":[132],"anchoring":[134],"them":[135],"prior":[140,199],"through":[141],"reward-tilted":[143],"target":[144],"distribution.":[145],"We":[146],"show":[147],"shifts":[150],"base-model":[151],"outputs":[152],"more":[154],"instructionfollowing":[155],"behavior,":[156],"increasing":[157],"behavioral":[158],"similarity":[159],"post-trained":[161],"counterparts":[162],"enabling":[164],"fairer":[166],"inference-time":[167],"evaluation":[168],"accessible":[170],"pre-trained-model":[171],"behavior.":[172],"Empirically,":[173],"outperforms":[175],"baselines":[176],"five":[178],"six":[181],"benchmarks,":[182],"improving":[183],"Qwen3-8B-Base":[184],"AlpacaEval2.0":[186],"8.8":[188],"44.5,":[190],"reducing":[191],"Mistral-7B":[192],"Math500":[193],"latency":[194],"by":[195],"18.9x":[196],"relative":[197],"work,":[201],"remaining":[203],"robust":[204],"reward-model":[206],"size.":[207]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-29T00:00:00"}