{"id":"https://openalex.org/W7160111478","doi":"https://doi.org/10.48550/arxiv.2605.00421","title":"RadLite: Multi-Task LoRA Fine-Tuning of Small Language Models for CPU-Deployable Radiology AI","display_name":"RadLite: Multi-Task LoRA Fine-Tuning of Small Language Models for CPU-Deployable Radiology AI","publication_year":2026,"publication_date":"2026-05-01","ids":{"openalex":"https://openalex.org/W7160111478","doi":"https://doi.org/10.48550/arxiv.2605.00421"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.00421","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.00421","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":"article"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2605.00421","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5135167814","display_name":"Pankaj Gupta","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gupta, Pankaj","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5121492347","display_name":"Kartik Bose","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Bose, Kartik","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":2,"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/T11636","display_name":"Artificial Intelligence in Healthcare and Education","score":0.9243000149726868,"subfield":{"id":"https://openalex.org/subfields/2718","display_name":"Health Informatics"},"field":{"id":"https://openalex.org/fields/27","display_name":"Medicine"},"domain":{"id":"https://openalex.org/domains/4","display_name":"Health Sciences"}},"topics":[{"id":"https://openalex.org/T11636","display_name":"Artificial Intelligence in Healthcare and Education","score":0.9243000149726868,"subfield":{"id":"https://openalex.org/subfields/2718","display_name":"Health Informatics"},"field":{"id":"https://openalex.org/fields/27","display_name":"Medicine"},"domain":{"id":"https://openalex.org/domains/4","display_name":"Health Sciences"}},{"id":"https://openalex.org/T11775","display_name":"COVID-19 diagnosis using AI","score":0.011800000444054604,"subfield":{"id":"https://openalex.org/subfields/2741","display_name":"Radiology, Nuclear Medicine and Imaging"},"field":{"id":"https://openalex.org/fields/27","display_name":"Medicine"},"domain":{"id":"https://openalex.org/domains/4","display_name":"Health Sciences"}},{"id":"https://openalex.org/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.009200000204145908,"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/software-deployment","display_name":"Software deployment","score":0.6934999823570251},{"id":"https://openalex.org/keywords/oracle","display_name":"Oracle","score":0.6434000134468079},{"id":"https://openalex.org/keywords/task","display_name":"Task (project management)","score":0.5745999813079834},{"id":"https://openalex.org/keywords/adaptation","display_name":"Adaptation (eye)","score":0.5275999903678894},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.4878999888896942},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.4528000056743622}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7785999774932861},{"id":"https://openalex.org/C105339364","wikidata":"https://www.wikidata.org/wiki/Q2297740","display_name":"Software deployment","level":2,"score":0.6934999823570251},{"id":"https://openalex.org/C55166926","wikidata":"https://www.wikidata.org/wiki/Q2892946","display_name":"Oracle","level":2,"score":0.6434000134468079},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.5745999813079834},{"id":"https://openalex.org/C139807058","wikidata":"https://www.wikidata.org/wiki/Q352374","display_name":"Adaptation (eye)","level":2,"score":0.5275999903678894},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.4878999888896942},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.4528000056743622},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4162999987602234},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.37369999289512634},{"id":"https://openalex.org/C2777904410","wikidata":"https://www.wikidata.org/wiki/Q7397","display_name":"Software","level":2,"score":0.35420000553131104},{"id":"https://openalex.org/C31601959","wikidata":"https://www.wikidata.org/wiki/Q931309","display_name":"Medical imaging","level":2,"score":0.2872999906539917},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.2856999933719635}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.00421","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.00421","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":"doi:10.48550/arxiv.2605.00421","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.00421","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":"article"},"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":{"Large":[0],"language":[1,27],"models":[2,28,87,125,148,161,179,203,229],"(LLMs)":[3],"show":[4],"promise":[5],"in":[6,19],"radiology":[7,38,58,70,78,216],"but":[8],"their":[9],"deployment":[10,44,189],"is":[11,168],"limited":[12],"by":[13],"computational":[14],"requirements":[15],"that":[16,165,199],"preclude":[17],"use":[18],"resource-constrained":[20],"clinical":[21],"environments.":[22],"We":[23,48],"investigate":[24],"whether":[25],"small":[26],"(SLMs)":[29],"of":[30],"3-4":[31],"billion":[32],"parameters":[33],"can":[34,180,211],"achieve":[35],"strong":[36],"multi-task":[37,215],"performance":[39,111,152],"through":[40],"LoRA":[41,107,166],"fine-tuning,":[42],"enabling":[43],"on":[45,53,90,193,221],"consumer-grade":[46],"CPUs.":[47],"train":[49],"Qwen2.5-3B-Instruct":[50],"and":[51,77,177,228],"Qwen3-4B":[52],"162K":[54],"samples":[55,96],"spanning":[56],"9":[57],"tasks":[59,135],"-":[60,80,129,204,210],"RADS":[61],"classification":[62],"across":[63,153],"10":[64],"systems,":[65],"impression":[66],"generation,":[67],"temporal":[68],"comparison,":[69],"NLI,":[71],"NER,":[72],"abnormality":[73],"detection,":[74],"N/M":[75],"staging,":[76],"Q&A":[79],"compiled":[81],"from":[82],"12":[83],"public":[84],"datasets.":[85],"Both":[86],"are":[88,230],"evaluated":[89],"up":[91],"to":[92,183],"500":[93],"held-out":[94],"test":[95],"per":[97],"task":[98],"with":[99,159],"standardized":[100],"metrics.":[101],"Our":[102,196],"key":[103],"findings":[104],"are:":[105],"(1)":[106],"fine-tuning":[108],"dramatically":[109],"improves":[110],"over":[112],"zero-shot":[113],"baselines":[114],"(RADS":[115],"accuracy":[116],"+53%,":[117],"NLI":[118],"+60%,":[119],"N-staging":[120],"+89%);":[121],"(2)":[122],"the":[123,150],"two":[124],"exhibit":[126],"complementary":[127],"strengths":[128],"Qwen2.5":[130],"excels":[131],"at":[132,190,232],"structured":[133],"generation":[134],"while":[136],"Qwen3":[137],"dominates":[138],"extractive":[139],"tasks;":[140,155],"(3)":[141],"a":[142],"task-outed":[143],"oracle":[144],"ensemble":[145],"combining":[146],"both":[147],"achieves":[149],"best":[151],"all":[154],"(4)":[156],"few-shot":[157],"prompting":[158],"fine-tuned":[160,202],"hurts":[162],"performance,":[163],"demonstrating":[164],"adaptation":[167],"more":[169],"effective":[170],"than":[171],"in-context":[172],"learning":[173],"for":[174,187],"specialized":[175],"domains;":[176],"(5)":[178],"be":[181],"quantized":[182],"GGUF":[184],"format":[185],"(~1.8-2.4GB)":[186],"CPU":[188],"4-8":[191],"tokens/second":[192],"consumer":[194,222],"hardware.":[195],"work":[197],"demonstrates":[198],"small,":[200],"efficiently":[201],"which":[205],"we":[206],"collectively":[207],"call":[208],"RadLite":[209],"serve":[212],"as":[213],"practical":[214],"AI":[217],"assistants":[218],"deployable":[219],"entirely":[220],"hardware":[223],"without":[224],"GPU":[225],"requirements.":[226],"Code":[227],"available":[231],"https://github.com/RadioX-Labs/RadLite":[233]},"counts_by_year":[],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2026-05-05T00:00:00"}