{"id":"https://openalex.org/W7213530307","doi":"https://doi.org/10.48550/arxiv.2609.17564","title":"Beyond Static RAG: An Adaptive, Tri-Metric Routing Framework for Efficient Long-Context Inference on Commodity GPUs","display_name":"Beyond Static RAG: An Adaptive, Tri-Metric Routing Framework for Efficient Long-Context Inference on Commodity GPUs","publication_year":2026,"publication_date":"2026-07-23","ids":{"openalex":"https://openalex.org/W7213530307","doi":"https://doi.org/10.48550/arxiv.2609.17564"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.17564","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.17564","pdf_url":"https://arxiv.org/pdf/2609.17564","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2609.17564","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5135352532","display_name":"Saipraveen Vabbilisetty","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Vabbilisetty, Saipraveen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5151999089","display_name":"Ajay Kumar Boddepalli","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Boddepalli, Ajay Kumar","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5108920595","display_name":"Deep Narayan Mishra","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Mishra, Deep Narayan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5018306405","display_name":"Shashank Kapadia","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kapadia, Shashank","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135292949","display_name":"Haoan Wang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Haoan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5135361779","display_name":"Anupriya Sharma","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sharma, Anupriya","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":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/T10028","display_name":"Topic Modeling","score":0.8237000107765198,"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":0.8237000107765198,"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":0.11379999667406082,"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/T10286","display_name":"Information Retrieval and Search Behavior","score":0.012500000186264515,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"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/retrieval-augmented-generation","display_name":"retrieval-augmented generation","score":0.996999979019165},{"id":"https://openalex.org/keywords/long-context-inference","display_name":"long-context inference","score":0.7239999771118164},{"id":"https://openalex.org/keywords/prompt-compression","display_name":"prompt compression","score":0.7250000238418579},{"id":"https://openalex.org/keywords/gpu-memory-management","display_name":"GPU memory management","score":0.5830000042915344},{"id":"https://openalex.org/keywords/adaptive-routing","display_name":"adaptive routing","score":0.5460000038146973},{"id":"https://openalex.org/keywords/bm25","display_name":"BM25","score":0.6169999837875366},{"id":"https://openalex.org/keywords/type-token-ratio","display_name":"type-token ratio","score":0.6899999976158142},{"id":"https://openalex.org/keywords/vllm","display_name":"vLLM","score":0.5410000085830688}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7937999963760376},{"id":"https://openalex.org/C55166926","wikidata":"https://www.wikidata.org/wiki/Q2892946","display_name":"Oracle","level":2,"score":0.5690000057220459},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5415999889373779},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.5015000104904175},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.47769999504089355},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.4262999892234802},{"id":"https://openalex.org/C2775896111","wikidata":"https://www.wikidata.org/wiki/Q642560","display_name":"Router","level":2,"score":0.4187000095844269},{"id":"https://openalex.org/C34736171","wikidata":"https://www.wikidata.org/wiki/Q918333","display_name":"Preprocessor","level":2,"score":0.4174000024795532},{"id":"https://openalex.org/C78548338","wikidata":"https://www.wikidata.org/wiki/Q2493","display_name":"Data compression","level":2,"score":0.38089999556541443},{"id":"https://openalex.org/C187191949","wikidata":"https://www.wikidata.org/wiki/Q1138496","display_name":"Profiling (computer programming)","level":2,"score":0.3653999865055084},{"id":"https://openalex.org/C122507166","wikidata":"https://www.wikidata.org/wiki/Q628906","display_name":"Crossover","level":2,"score":0.36250001192092896},{"id":"https://openalex.org/C189783530","wikidata":"https://www.wikidata.org/wiki/Q352090","display_name":"CPU cache","level":3,"score":0.3582000136375427},{"id":"https://openalex.org/C186644900","wikidata":"https://www.wikidata.org/wiki/Q194152","display_name":"Parsing","level":2,"score":0.3458999991416931},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.32899999618530273},{"id":"https://openalex.org/C25797200","wikidata":"https://www.wikidata.org/wiki/Q828137","display_name":"Compression ratio","level":3,"score":0.32249999046325684},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.3111000061035156},{"id":"https://openalex.org/C180016635","wikidata":"https://www.wikidata.org/wiki/Q2712821","display_name":"Compression (physics)","level":2,"score":0.3077000081539154},{"id":"https://openalex.org/C2984842247","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep neural networks","level":3,"score":0.3034000098705292},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.301800012588501},{"id":"https://openalex.org/C174348530","wikidata":"https://www.wikidata.org/wiki/Q188635","display_name":"Bridging (networking)","level":2,"score":0.2888000011444092},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.274399995803833},{"id":"https://openalex.org/C162040801","wikidata":"https://www.wikidata.org/wiki/Q799897","display_name":"Bootstrap aggregating","level":2,"score":0.25099998712539673},{"id":"https://openalex.org/C183003079","wikidata":"https://www.wikidata.org/wiki/Q1000371","display_name":"Personalization","level":2,"score":0.250900000333786}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2609.17564","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.17564","pdf_url":"https://arxiv.org/pdf/2609.17564","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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.2609.17564","doi":"https://doi.org/10.48550/arxiv.2609.17564","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2609.17564","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2609.17564","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.17564","pdf_url":"https://arxiv.org/pdf/2609.17564","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7213530307.pdf","grobid_xml":"https://content.openalex.org/works/W7213530307.grobid-xml"},"referenced_works_count":3,"referenced_works":["https://openalex.org/W4376122773","https://openalex.org/W4400141368","https://openalex.org/W4404783788"],"related_works":["https://openalex.org/W7223566720","https://openalex.org/W7171702354","https://openalex.org/W7203944647","https://openalex.org/W7120051728","https://openalex.org/W7136170912","https://openalex.org/W7220801807","https://openalex.org/W7168016197","https://openalex.org/W7172201241","https://openalex.org/W4412887924","https://openalex.org/W6948196230"],"abstract_inverted_index":{"Deploying":[0],"retrieval-augmented":[1],"generation":[2],"(RAG)":[3],"on":[4,49,117,130,140],"commodity":[5],"GPUs":[6],"such":[7],"as":[8],"the":[9,22,74,156],"NVIDIA":[10],"T4":[11],"(16":[12],"GB":[13],"VRAM)":[14],"exposes":[15],"a":[16,59,63,77,121,150],"practical":[17],"failure":[18,56],"mode":[19],"we":[20],"call":[21],"Compression":[23],"Paradox:":[24],"neural":[25],"prompt":[26],"compression":[27,43,175],"can":[28,44],"add":[29],"key-value":[30],"(KV)":[31],"cache":[32],"contention":[33],"and":[34,62,72,87,103,120,167],"preprocessing":[35],"latency":[36,122],"that":[37,81],"outweigh":[38],"generation-time":[39],"savings,":[40],"while":[41],"skipping":[42],"cause":[45],"out-of-memory":[46],"(OOM)":[47],"failures":[48],"long":[50],"contexts.":[51],"We":[52],"identify":[53],"two":[54],"distinct":[55],"mechanisms":[57],"when":[58],"vLLM-served":[60],"LLM":[61],"PyTorch-based":[64],"compressor":[65],"are":[66,126],"co-deployed":[67],"under":[68],"tight":[69],"memory":[70],"budgets,":[71],"introduce":[73],"Tri-Metric":[75],"Router,":[76],"deterministic,":[78],"training-free":[79],"policy":[80],"selects":[82],"among":[83],"Raw,":[84],"Neural":[85],"(LLMLingua-2),":[86],"Lexical":[88],"(BM25)":[89],"pipelines.":[90],"The":[91],"router":[92],"uses":[93],"three":[94],"CPU-side":[95],"signals:":[96],"spatial":[97],"complexity":[98],"($L$),":[99],"syntactic":[100],"density":[101],"($\u03c1_{key}$),":[102],"type-token":[104],"ratio":[105],"(TTR).":[106],"Unlike":[107],"prior":[108],"semantic-only":[109],"adaptation,":[110],"our":[111,142],"dispatch":[112],"signal":[113],"is":[114,144],"hardware-physical,":[115],"based":[116],"VRAM":[118,181],"headroom":[119],"crossover":[123,136],"point.":[124],"Thresholds":[125],"calibrated":[127],"from":[128],"profiling":[129],"LongBench":[131],"qasper,":[132],"yielding":[133],"an":[134],"operating":[135],"near":[137],"4,332":[138],"words":[139],"T4;":[141],"contribution":[143],"this":[145],"calibration":[146],"methodology":[147],"rather":[148],"than":[149],"hardware-specific":[151],"constant.":[152],"On":[153],"out-of-distribution":[154],"holdouts,":[155],"method":[157],"achieves":[158],"0%":[159],"OOM":[160],"failures,":[161],"88.5":[162],"$\\pm$":[163],"4.4%":[164],"oracle":[165],"alignment,":[166],"49.3%":[168],"Combined":[169],"F1,":[170],"improving":[171],"over":[172],"always-on":[173],"lexical":[174],"by":[176],"5.2":[177],"points":[178],"without":[179],"additional":[180],"or":[182],"training":[183],"cost.":[184]},"counts_by_year":[],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-09-18T00:00:00"}