{"id":"https://openalex.org/W7164942579","doi":"https://doi.org/10.48550/arxiv.2606.16231","title":"From Tokens to Regions: CUDA-Sensitive Instruction Tuning for GPU Kernel Generation","display_name":"From Tokens to Regions: CUDA-Sensitive Instruction Tuning for GPU Kernel Generation","publication_year":2026,"publication_date":"2026-06-15","ids":{"openalex":"https://openalex.org/W7164942579","doi":"https://doi.org/10.48550/arxiv.2606.16231"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.16231","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.16231","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.2606.16231","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5138714479","display_name":"Wentao Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Wentao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5107958486","display_name":"Jiace Zhu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhu, Jiace","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128056271","display_name":"Xing Zhe Chai","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chai, Xing Zhe","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138694472","display_name":"Zeng Qu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Qu, Zeng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138730659","display_name":"Qiaoling Xiao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xiao, Qiaoling","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5051934376","display_name":"\u6bb5\u7559\u6210","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Duan, Liucheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5121563724","display_name":"An Zou","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zou, An","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":0.13529999554157257,"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":0.13529999554157257,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.11949999630451202,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.11680000275373459,"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/cuda","display_name":"CUDA","score":0.9259999990463257},{"id":"https://openalex.org/keywords/kernel","display_name":"Kernel (algebra)","score":0.5738999843597412},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5544999837875366},{"id":"https://openalex.org/keywords/scalability","display_name":"Scalability","score":0.5202000141143799},{"id":"https://openalex.org/keywords/correctness","display_name":"Correctness","score":0.49459999799728394},{"id":"https://openalex.org/keywords/leverage","display_name":"Leverage (statistics)","score":0.47350001335144043},{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.40610000491142273}],"concepts":[{"id":"https://openalex.org/C2778119891","wikidata":"https://www.wikidata.org/wiki/Q477690","display_name":"CUDA","level":2,"score":0.9259999990463257},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8219000101089478},{"id":"https://openalex.org/C74193536","wikidata":"https://www.wikidata.org/wiki/Q574844","display_name":"Kernel (algebra)","level":2,"score":0.5738999843597412},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5544999837875366},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.5238000154495239},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.5202000141143799},{"id":"https://openalex.org/C55439883","wikidata":"https://www.wikidata.org/wiki/Q360812","display_name":"Correctness","level":2,"score":0.49459999799728394},{"id":"https://openalex.org/C153083717","wikidata":"https://www.wikidata.org/wiki/Q6535263","display_name":"Leverage (statistics)","level":2,"score":0.47350001335144043},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.40610000491142273},{"id":"https://openalex.org/C202491316","wikidata":"https://www.wikidata.org/wiki/Q272683","display_name":"Instruction set","level":2,"score":0.36309999227523804},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.3610999882221222},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.3465999960899353},{"id":"https://openalex.org/C2780586882","wikidata":"https://www.wikidata.org/wiki/Q7520643","display_name":"Simple (philosophy)","level":2,"score":0.32829999923706055},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.30799999833106995},{"id":"https://openalex.org/C12713177","wikidata":"https://www.wikidata.org/wiki/Q1900281","display_name":"Perspective (graphical)","level":2,"score":0.27639999985694885},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.2678000032901764},{"id":"https://openalex.org/C148027188","wikidata":"https://www.wikidata.org/wiki/Q907375","display_name":"Unit testing","level":3,"score":0.26489999890327454}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.16231","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.16231","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.2606.16231","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.16231","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":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"High-performance":[0],"CUDA":[1,50,67,78,112,190],"kernels":[2,19],"are":[3,91],"essential":[4],"for":[5],"scalable":[6],"AI":[7],"systems,":[8],"while":[9,96,184],"Large":[10],"Language":[11],"Models":[12],"(LLMs)":[13],"still":[14],"struggle":[15],"to":[16,21,47,104,152],"generate":[17],"correct":[18],"due":[20],"strict":[22],"and":[23,84,121,175,180],"implicit":[24],"execution":[25,60],"constraints.":[26,61],"Existing":[27],"LLM-based":[28],"approaches":[29],"either":[30],"rely":[31],"on":[32,127],"costly":[33],"agentic":[34],"or":[35,39,55],"reinforcement-learning":[36],"(RL)":[37],"pipelines,":[38],"adopt":[40],"supervised":[41],"fine-tuning":[42],"(SFT)":[43],"objectives":[44],"that":[45,77,110,165],"fail":[46],"explicitly":[48],"model":[49,173],"sensitivity,":[51],"namely":[52],"code":[53],"tokens":[54,90,120,151],"regions":[56,102],"tightly":[57],"coupled":[58],"with":[59,93,159,194],"In":[62],"this":[63],"work,":[64],"we":[65,130],"investigate":[66],"sensitivity":[68,79],"from":[69],"the":[70,147],"perspective":[71],"of":[72,149],"token":[73,83],"confidence":[74],"patterns,":[75],"showing":[76],"appears":[80],"at":[81],"both":[82,116],"region":[85],"levels,":[86],"where":[87],"most":[88],"CUDA-sensitive":[89,119,124],"predicted":[92],"high":[94],"confidence,":[95],"a":[97,136,141],"smaller":[98],"low-confidence":[99,123],"subset":[100],"forms":[101],"corresponding":[103],"execution-critical":[105],"structures.":[106],"These":[107],"findings":[108],"suggest":[109],"effective":[111],"kernel":[113,191],"generation":[114,192],"should":[115],"leverage":[117],"high-confidence":[118],"preserve":[122],"regions.":[125],"Building":[126],"these":[128],"insights,":[129],"propose":[131],"\\textbf{\\underline{CU}DA-\\underline{Se}nsitive":[132],"Instruction":[133],"\\underline{T}uning":[134],"(CuSeT)},":[135],"low-cost":[137],"post-training":[138],"method":[139],"within":[140],"simple":[142],"SFT":[143,179,182],"framework.":[144],"CuSeT":[145,166],"follows":[146],"principle":[148],"``from":[150],"regions''":[153],"by":[154],"combining":[155],"\\emph{adaptive":[156],"token-level":[157],"masking}":[158],"\\emph{region-aware":[160],"sample":[161],"reweighting}.":[162],"Experiments":[163],"show":[164],"consistently":[167],"improves":[168],"functional":[169],"correctness":[170],"across":[171],"multiple":[172],"families":[174],"scales,":[176],"outperforming":[177],"standard":[178],"advanced":[181],"variants,":[183],"achieving":[185],"competitive":[186],"performance":[187],"against":[188],"frontier":[189],"models":[193],"substantially":[195],"lower":[196],"inference":[197],"cost.":[198]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-17T00:00:00"}