[{"data":1,"prerenderedAt":179},["ShallowReactive",2],{"project-47639567":3},{"id":4,"name":5,"line":6,"line_source":7,"title":8,"url":9,"domain":10,"maker":11,"posted_at":12,"year":13,"points":14,"comments":15,"rank_in_year":16,"shown_in_year":17,"top":18,"tier":19,"category":20,"subcategory":21,"format":22,"keywords":23,"reception":28,"sentiment":29,"stance_mix":30,"scored_comments":34,"aspects":35,"strip":39,"post_text":29,"what":42,"said":62,"maker_in_thread":110,"shape":116,"grid":120,"attempts":142,"same_week":143,"maker_shows":33,"maker_hidden":172,"reader_version":173,"card":174},"47639567","TurboQuant-WASM","Google's vector quantization in the browser","title","Show HN: TurboQuant-WASM – Google's vector quantization in the browser","https:\u002F\u002Fgithub.com\u002Fteamchong\u002Fturboquant-wasm","github.com","teamchong",1775314396,2026,165,7,376,4366,9,"notable","ai","vector search","library",[24,25,26,27],"vector quantization","webassembly","embeddings","browser search","quiet",null,{"positive":31,"neutral":31,"mixed":32,"critical":33},1,0,2,4,{"performance":36,"idea":37,"design":38},{"positive":32,"critical":31},{"positive":32,"critical":31},{"positive":31,"critical":32},[31,32,32,31,32,40,41,32,32],3,5,{"problem":43,"audience":29,"pricing":29,"motivation":29,"time_to_build":29,"stack":52,"openness":61,"stage":61,"team_size":61},{"value":44,"source":45,"evidence":46},"Compresses vector embeddings so search can run in the browser with less memory and download size","maker",[47,50],{"id":48,"by":11,"text":49},"47644881","you’re right that 32f is faster on raw query time, quantization adds extra step. main benefit on download size since gzip won’t help much, which matters most in browser contexts",{"id":51},"post",[53],{"value":54,"source":55,"evidence":56},"WebAssembly","commenter",[57],{"id":58,"by":59,"text":60},"47644125","refulgentis","Sloppiest slop I've seen in a couple weeks: - fork of a fork of a quantization technique - Only contribution is...compiling JS to WASM by default? - suspicious burst of ~nothing…","unknown",{"praise":63,"objections":75,"requests":92,"questions":97,"alternatives":105},[64],{"value":65,"source":55,"evidence":66},"Browser search demo and use cases look interesting",[67,71],{"id":68,"by":69,"text":70},"47640703","hhthrowaway1230","Awesome! Also love the gaussian splat demo, cool use case!",{"id":72,"by":73,"text":74},"47640808","glohbalrob","Very cool. I added the new multi embedding 2 model to my site the other week from google I guess need to dig into this and see if it’s faster and has more use cases! Thanks for…",[76,84,88],{"value":77,"source":55,"evidence":78},"Quantized search is slower than unquantized float32 search",[79,83],{"id":80,"by":81,"text":82},"47644184","netdur","I tried TQ for vector search and my findings is not good, it is not worth it if you cannot use GPU, however I got same quality of search as 32f using 8bit quant I wrote ann ext…",{"id":58,"by":59,"text":60},{"value":85,"source":55,"evidence":86},"Saves memory but makes text embedding search much slower in the demo",[87],{"id":58,"by":59,"text":60},{"value":89,"source":55,"evidence":90},"Only a fork of an existing quantization technique with little original contribution",[91],{"id":58,"by":59,"text":60},[93],{"value":94,"source":55,"evidence":95},"Benchmark speed improvements in the demo",[96],{"id":58,"by":59,"text":60},[98],{"value":99,"source":55,"evidence":100},"Whether OPQ-style trained quantization would perform better",[101],{"id":102,"by":103,"text":104},"47645167","ninja3925","So i assumed it would get crushed by OPQ (which requires training)",[106],{"value":107,"source":55,"evidence":108},"OPQ (optimized product quantization)",[109],{"id":102,"by":103,"text":104},{"comments":33,"answered":31,"key":111},[112,113],{"id":48,"text":49},{"id":114,"text":115},"47686425","I made some adjustment, can you try again? Is it faster now? https:\u002F\u002Fteamchong.github.io\u002Fturboquant-wasm\u002Fsearch.html",{"commenters":117,"max_depth":33,"maker_replies":33,"first_comment_mins":118,"lifespan_hours":119},8,111,88,[121,123,125,127,129,131,133,135,139],[80,32,31,122,81,32],"project\u002Fcritical\u002Fperformance",[48,32,32,124,11,31],"commenter\u002Fneutral\u002Fperformance",[102,32,32,126,103,32],"commenter\u002Fneutral\u002Fidea",[58,31,31,128,59,32],"project\u002Fcritical\u002Fidea",[114,31,32,130,11,31],"maker\u002Fneutral\u002Fperformance",[72,33,40,132,73,32],"project\u002Fneutral\u002Fnone",[68,40,41,134,69,32],"project\u002Fpositive\u002Fdesign",[136,34,32,137,138,32],"47642463","other\u002Fneutral\u002Fnone","himmelsee2018",[140,41,32,137,141,32],"47642215","aritzdf",[],[144,149,153,158,163,167],{"id":145,"name":146,"points":147,"category":148},"47640728","mvidia",964,"learning_science",{"id":150,"name":151,"points":152,"category":148},"47655408","GuppyLM",915,{"id":154,"name":155,"points":156,"category":157},"47673360","Brutalist Concrete Laptop Stand",786,"hardware",{"id":159,"name":160,"points":161,"category":162},"47625952","Blogosphere",783,"consumer_social",{"id":164,"name":165,"points":166,"category":20},"47624645","Apfel",743,{"id":168,"name":169,"points":170,"category":171},"47696562","Is Hormuz open yet?",484,"web_saas",false,"v2",{"id":175,"name":5,"line":6,"ai":172,"cat":20,"year":13,"points":14,"comments":15,"rec":28,"s":29,"top":18,"kw":176,"maker":11,"domain":10,"tier":19,"attempts":32,"strip":177,"url":9,"mix":178,"praise":65,"objection":77,"sub":21},47639567,[24,25,26,27],"100103500",[31,31,32,33],1794036697781]