{"id":"https://openalex.org/W7225078804","doi":"https://doi.org/10.21203/rs.3.rs-11540864/v1","title":"Post-Training on the Clock: Specializing a Small Language Model on a Consumer GPU","display_name":"Post-Training on the Clock: Specializing a Small Language Model on a Consumer GPU","publication_year":2026,"publication_date":"2026-11-02","ids":{"openalex":"https://openalex.org/W7225078804","doi":"https://doi.org/10.21203/rs.3.rs-11540864/v1"},"language":"en","primary_location":{"id":"doi:10.21203/rs.3.rs-11540864/v1","doi":"https://doi.org/10.21203/rs.3.rs-11540864/v1","is_oa":true,"landing_page_url":"https://doi.org/10.21203/rs.3.rs-11540864/v1","pdf_url":"https://www.researchsquare.com/article/rs-11540864/latest.pdf","source":{"id":"https://openalex.org/S4306525896","display_name":"Research Square","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"acceptedVersion","is_accepted":true,"is_published":false,"raw_source_name":"Research Square","raw_type":"posted-content"},"volume":null,"type":"preprint","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://www.researchsquare.com/article/rs-11540864/latest.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5113030548","display_name":"Uygar Kurt","orcid":"https://orcid.org/0009-0001-1861-2115","observed_orcids":["https://orcid.org/0009-0001-1861-2115"]},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Uygar Kurt","raw_affiliation_strings":["Independent Researcher"],"raw_orcid":"https://orcid.org/0009-0001-1861-2115","affiliations":[{"raw_affiliation_string":"Independent Researcher","institution_ids":[]}]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":["https://openalex.org/A5113030548"],"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.4140999913215637,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.4140999913215637,"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/T10028","display_name":"Topic Modeling","score":0.23749999701976776,"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":0.19779999554157257,"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/group-relative-policy-optimization","display_name":"Group Relative Policy Optimization","score":0.8820000290870667},{"id":"https://openalex.org/keywords/training-budget-allocation","display_name":"training budget allocation","score":0.6890000104904175},{"id":"https://openalex.org/keywords/tool-calling","display_name":"tool calling","score":0.7609999775886536},{"id":"https://openalex.org/keywords/function-calling","display_name":"function calling","score":0.9309999942779541},{"id":"https://openalex.org/keywords/language-model-agents","display_name":"language model agents","score":0.7250000238418579},{"id":"https://openalex.org/keywords/low-rank-adaptation","display_name":"low-rank adaptation","score":0.8240000009536743},{"id":"https://openalex.org/keywords/qwen2-5","display_name":"Qwen2.5","score":0.6420000195503235}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8004999756813049},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.6412000060081482},{"id":"https://openalex.org/C158154518","wikidata":"https://www.wikidata.org/wiki/Q7310970","display_name":"Relevance (law)","level":2,"score":0.51910001039505},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.5105000138282776},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.5038999915122986},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.41920000314712524},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4025999903678894},{"id":"https://openalex.org/C2777027219","wikidata":"https://www.wikidata.org/wiki/Q1284190","display_name":"Constant (computer programming)","level":2,"score":0.3950999975204468},{"id":"https://openalex.org/C14036430","wikidata":"https://www.wikidata.org/wiki/Q3736076","display_name":"Function (biology)","level":2,"score":0.3853999972343445},{"id":"https://openalex.org/C2778119891","wikidata":"https://www.wikidata.org/wiki/Q477690","display_name":"CUDA","level":2,"score":0.35589998960494995},{"id":"https://openalex.org/C51929080","wikidata":"https://www.wikidata.org/wiki/Q2425187","display_name":"Codebase","level":3,"score":0.3537999987602234},{"id":"https://openalex.org/C85847156","wikidata":"https://www.wikidata.org/wiki/Q59015987","display_name":"Verifiable secret sharing","level":3,"score":0.35030001401901245},{"id":"https://openalex.org/C139807058","wikidata":"https://www.wikidata.org/wiki/Q352374","display_name":"Adaptation (eye)","level":2,"score":0.337799996137619},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.31349998712539673},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.30079999566078186},{"id":"https://openalex.org/C2779804580","wikidata":"https://www.wikidata.org/wiki/Q102047","display_name":"Suffix","level":2,"score":0.28769999742507935},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.2797999978065491},{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.2797999978065491},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.2727000117301941},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.27230000495910645},{"id":"https://openalex.org/C147764199","wikidata":"https://www.wikidata.org/wiki/Q6865248","display_name":"Minification","level":2,"score":0.2721000015735626},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.26420000195503235},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.2597000002861023}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.21203/rs.3.rs-11540864/v1","doi":"https://doi.org/10.21203/rs.3.rs-11540864/v1","is_oa":true,"landing_page_url":"https://doi.org/10.21203/rs.3.rs-11540864/v1","pdf_url":"https://www.researchsquare.com/article/rs-11540864/latest.pdf","source":{"id":"https://openalex.org/S4306525896","display_name":"Research Square","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"acceptedVersion","is_accepted":true,"is_published":false,"raw_source_name":"Research Square","raw_type":"posted-content"}],"best_oa_location":{"id":"doi:10.21203/rs.3.rs-11540864/v1","doi":"https://doi.org/10.21203/rs.3.rs-11540864/v1","is_oa":true,"landing_page_url":"https://doi.org/10.21203/rs.3.rs-11540864/v1","pdf_url":"https://www.researchsquare.com/article/rs-11540864/latest.pdf","source":{"id":"https://openalex.org/S4306525896","display_name":"Research Square","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"acceptedVersion","is_accepted":true,"is_published":false,"raw_source_name":"Research Square","raw_type":"posted-content"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7225078804.pdf","grobid_xml":"https://content.openalex.org/works/W7225078804.grobid-xml"},"referenced_works_count":31,"referenced_works":["https://openalex.org/W2908510526","https://openalex.org/W2973727699","https://openalex.org/W3083410900","https://openalex.org/W3134642945","https://openalex.org/W3168867926","https://openalex.org/W4379089709","https://openalex.org/W4385474529","https://openalex.org/W4388718089","https://openalex.org/W4391631327","https://openalex.org/W4402667093","https://openalex.org/W4402954204","https://openalex.org/W4403883066","https://openalex.org/W4403964119","https://openalex.org/W4405655184","https://openalex.org/W4411120021","https://openalex.org/W4412889889","https://openalex.org/W4415345646","https://openalex.org/W4415797489","https://openalex.org/W4416888804","https://openalex.org/W4417065980","https://openalex.org/W4417511763","https://openalex.org/W6959934489","https://openalex.org/W6967397561","https://openalex.org/W7127616265","https://openalex.org/W7133214447","https://openalex.org/W7160845055","https://openalex.org/W7163150380","https://openalex.org/W7196953391","https://openalex.org/W7206153012","https://openalex.org/W7213506460","https://openalex.org/W7214380358"],"related_works":["https://openalex.org/W7213506460","https://openalex.org/W4385574005","https://openalex.org/W7215040960","https://openalex.org/W7208220091","https://openalex.org/W7165666045","https://openalex.org/W4415348723","https://openalex.org/W4416765033","https://openalex.org/W7214088086","https://openalex.org/W7155218812","https://openalex.org/W7166871571"],"abstract_inverted_index":{"Abstract":[0],"How":[1],"should":[2],"a":[3,9,21,29,55,96],"language":[4],"model":[5],"be":[6],"specialized":[7],"for":[8,127,135,142,149,188],"target":[10],"task":[11],"when":[12,113],"training":[13,32,231],"is":[14,117],"limited":[15],"to":[16,36,63],"one":[17],"consumer":[18],"GPU":[19],"and":[20,44,66,89,103,111,124,144,147,211,226,232,237],"few":[22],"hours?":[23],"We":[24,50,217],"study":[25,57],"the":[26,168,178,219],"allocation":[27],"of":[28,59,83,99,159],"nominal":[30],"three-hour":[31],"budget,":[33],"with":[34,72,80,186,200,235],"up":[35],"0.3":[37],"additional":[38],"hours,":[39,175],"between":[40],"supervised":[41],"fine-tuning":[42],"(SFT)":[43],"Group":[45],"Relative":[46],"Policy":[47],"Optimization":[48],"(GRPO).":[49],"use":[51],"tool":[52,116],"calling":[53],"as":[54],"case":[56],"because":[58],"its":[60,67],"practical":[61],"relevance":[62],"language-model":[64],"agents":[65],"verifiable":[68],"outputs.":[69],"Using":[70],"Qwen2.5-0.5B-Instruct":[71],"low-rank":[73],"adaptation":[74],"on":[75],"an":[76,229],"RTX":[77],"5060":[78],"Ti":[79],"16":[81],"GB":[82],"memory,":[84],"we":[85],"compare":[86],"SFT,":[87,199],"GRPO,":[88],"six":[90],"sequential":[91],"SFT-to-GRPO":[92],"allocations.":[93],"Training":[94],"uses":[95],"filtered":[97],"mixture":[98],"xLAM":[100,121,180],"function-calling":[101,196],"examples":[102],"Hammer":[104],"irrelevance":[105],"examples,":[106],"covering":[107],"both":[108],"call":[109,129],"construction":[110],"abstention":[112,202],"no":[114,213],"offered":[115],"suitable.":[118],"Evaluation":[119],"reports":[120],"exact":[122,181],"tool-call":[123,182],"function-name":[125],"accuracies":[126,162],"local":[128],"construction,":[130],"BFCL-v3":[131,139],"Single-turn":[132],"FC":[133],"(agg)":[134,141,155],"external":[136,195],"function":[137],"calling,":[138],"Irrelevance":[140],"abstention,":[143],"MMLU,":[145],"MATH-500,":[146],"IFEval":[148],"selected":[150,160],"general":[151],"capabilities.":[152],"The":[153],"suffix":[154],"denotes":[156],"study-specific":[157],"averages":[158],"subset":[161,165],"weighted":[163],"by":[164],"size.":[166],"At":[167],"last":[169],"evaluated":[170],"checkpoints":[171],"within":[172],"3.3":[173],"reported":[174],"SFT":[176],"achieves":[177],"highest":[179],"accuracy":[183,197],"(82.82%),":[184],"compared":[185],"69.13%":[187],"GRPO":[189,206],"alone.":[190],"Selected":[191],"hybrid":[192],"allocations":[193],"improve":[194],"over":[198],"different":[201],"trade-offs.":[203],"Most":[204],"sampled":[205],"groups":[207],"have":[208],"constant":[209],"rewards":[210],"provide":[212],"relative":[214],"task-reward":[215],"signal.":[216],"release":[218],"prepared":[220],"mixed":[221],"dataset,":[222],"all":[223],"allocation-study":[224],"checkpoints,":[225],"Post-Training":[227],"Lab,":[228],"editable":[230],"evaluation":[233],"codebase":[234],"CUDA":[236],"MLX":[238],"implementations.":[239]},"counts_by_year":[],"updated_date":"2026-11-04T10:54:05.112534","created_date":"2026-11-03T00:00:00"}