[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-mrvq-lets-one-search-index-swap-dimensions-and-bit-rates":10,"sections":35},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":24,"tags":25,"sources":30,"feedback":34,"feedback_at":22,"cost_usd":34,"total_tokens":34},9960,"mrvq-lets-one-search-index-swap-dimensions-and-bit-rates","MRVQ Lets One Search Index Swap Dimensions and Bit Rates","A new quantization method called MRVQ keeps one compressed index that can be resized for speed, memory, or accuracy tradeoffs, cutting RAM use sharply.","A new technique lets a single compressed search index shrink or grow on demand, instead of forcing engineers to store a separate index for every speed-versus-accuracy tradeoff.\n\nResearchers describe Matryoshka Residual Vector Quantization (MRVQ), a method for compressing embeddings - the numerical fingerprints that power semantic search - after they have already been generated. The trick is that one maximum-quality code can be trimmed two different ways: drop some of its residual stages to lower the bit rate, or drop some embedding coordinates to lower the dimension. Tested on the FiQA and NFCorpus benchmarks across four embedding families and 4-, 8-, and 16-byte codes, MRVQ used 17.8 to 22.0 times less memory than running three separately trained QINCo2 indices side by side, and 1.89 to 2.02 times less than a leaner shared-model baseline the authors built for comparison.\n\nProduction search and retrieval-augmented systems typically need different speed, memory, and accuracy tradeoffs for different workloads, which today means either maintaining multiple tuned indices or accepting one-size-fits-all mediocrity. MRVQ's single resident artifact means one index file can serve all of those tradeoffs at once, which could meaningfully cut the infrastructure cost of running dense retrieval at scale.\n\nThe catch: per-rate QINCo2 still beats MRVQ on raw search quality by up to 0.107 nDCG@10 on FiQA, and the authors' own negative results - QINCo2 breaking down when trained at high rates, a ranking-bound hypothesis that failed its own pre-set acceptance criteria - are a reminder that this is a memory-savings trick, not a free lunch.","[\"vector search\",\"quantization\",\"retrieval\",\"arxiv research\"]","2026-10-05T04:00:00.000Z","2026-10-05T15:26:26.062Z","2026-10-05T15:26:31.487Z","published",null,[],"ai",[26,27,28,29],"vector search","quantization","retrieval","arxiv research",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.03651",0,{"sections":36},[37,40,44,49,54,59,63,68,72,76,81,86,91,96],{"name":38,"slug":24,"count":39,"latest_published_at":18},"AI",6170,{"name":41,"slug":42,"count":43,"latest_published_at":18},"Security","security",860,{"name":45,"slug":46,"count":47,"latest_published_at":48},"Policy","policy",444,"2026-10-03T15:02:01.000Z",{"name":50,"slug":51,"count":52,"latest_published_at":53},"Deals","deals",323,"2026-10-04T13:00:00.000Z",{"name":55,"slug":56,"count":57,"latest_published_at":58},"Hardware","hardware",204,"2026-10-03T14:50:50.000Z",{"name":60,"slug":61,"count":62,"latest_published_at":18},"Science","science",177,{"name":64,"slug":65,"count":66,"latest_published_at":67},"Consumer Tech","consumer-tech",158,"2026-10-03T03:21:12.000Z",{"name":69,"slug":70,"count":71,"latest_published_at":18},"Dev Tools","dev-tools",97,{"name":73,"slug":74,"count":71,"latest_published_at":75},"Software","software","2026-10-04T10:00:00.000Z",{"name":77,"slug":78,"count":79,"latest_published_at":80},"Startups","startups",92,"2026-10-04T14:36:25.000Z",{"name":82,"slug":83,"count":84,"latest_published_at":85},"Gaming","gaming",53,"2026-10-02T02:50:39.000Z",{"name":87,"slug":88,"count":89,"latest_published_at":90},"General","general",51,"2026-10-05T02:35:01.000Z",{"name":92,"slug":93,"count":94,"latest_published_at":95},"Reviews","reviews",32,"2026-10-02T18:00:00.000Z",{"name":97,"slug":98,"count":99,"latest_published_at":100},"How-To","how-to",7,"2026-10-01T09:00:00.000Z"]