[{"data":1,"prerenderedAt":604},["ShallowReactive",2],{"project-47593422":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":35,"aspects":36,"strip":47,"post_text":49,"what":50,"said":89,"maker_in_thread":199,"shape":201,"grid":205,"attempts":566,"same_week":567,"maker_shows":44,"maker_hidden":596,"reader_version":597,"card":598},"47593422","1-Bit Bonsai","The First Commercially Viable 1-Bit LLMs","title","Show HN: 1-Bit Bonsai, the First Commercially Viable 1-Bit LLMs","https:\u002F\u002Fprismml.com\u002F","prismml.com","PrismML",1774990878,2026,430,153,71,4366,2,"legend","ai","local llm models","library",[24,25,26,27],"1-bit llm","quantized model","local llm","llama.cpp","mixed",0.056,{"positive":31,"neutral":32,"mixed":33,"critical":34},18,35,5,14,72,{"performance":37,"execution":40,"idea":43,"pricing":45,"docs":46},{"positive":38,"critical":39},8,10,{"positive":41,"critical":42},0,3,{"positive":33,"critical":44},1,{"positive":41,"critical":41},{"positive":41,"critical":41},[33,33,42,48,33,48,42,41,44,48,41,42,42,42,44,42,18,42,33,44,44,33,44,41,41,41,42,41,44,44,33,44,42,33,48,42,42,42,41,44],4,null,{"problem":51,"audience":49,"pricing":49,"motivation":49,"time_to_build":49,"stack":63,"openness":87,"stage":88,"team_size":87},{"value":52,"source":53,"evidence":54},"Running capable language models with much less memory and compute","commenter",[55,59],{"id":56,"by":57,"text":58},"47595244","volume_tech","the speed is not just about storage -- at 1-bit you are reading roughly 16x less data from DRAM per forward pass compared to FP16. on memory-bandwidth-constrained hardware that…",{"id":60,"by":61,"text":62},"47596521","andai","The site says 14x less memory usage. I'm a bit confused about that situation. The model file is indeed very small, but on my machine it used roughly the same RAM as 4 bit quants…",[64,74,80],{"value":27,"source":53,"evidence":65},[66,70],{"id":67,"by":68,"text":69},"47595610","wild_egg","Don't have a GPU so tried the CPU option and got 0.6t\u002Fs on my old 2018 laptop using their llama.cpp fork. Then found out they didn't implement AVX2 for their Q1_0_g128 CPU…",{"id":71,"by":72,"text":73},"47597268","freakynit","Open access for next 5 hours (8GiB model, running on RTX 3090) or until server crashes or the this spot instance gets taken away :) => https:\u002F\u002Fofo1j9j6qh20a8-80.proxy.runpod.net…",{"value":75,"source":53,"evidence":76},"CUDA",[77],{"id":78,"by":61,"text":79},"47595849","Does anyone know how to run this on CPU? Do I need to build their llama.cpp fork from source? (I think they only offer CUDA options in the release page, which I think might…",{"value":81,"source":53,"evidence":82},"Qwen3 base model",[83],{"id":84,"by":85,"text":86},"47598182","fxwin","I'm very skeptical of the advantage they're claiming here. The whitepaper [0] only compares these to full precision models, when the more interesting (and probably more…","unknown","launched",{"praise":90,"objections":114,"requests":147,"questions":167,"alternatives":187},[91,106],{"value":92,"source":53,"evidence":93},"Impressive performance for its small size",[94,98,102],{"id":95,"by":96,"text":97},"47598483","nl","I ran my custom agentic SQL debugging benchmark against it and I'm impressed. Results: 8 passed, 0 failed, 17 errored out of 25 That puts it right between Qwen3.5-4B (7\u002F25) and…",{"id":99,"by":100,"text":101},"47596857","simonw","You can run this model on an iPhone via the latest update to this Locally AI app: https:\u002F\u002Fapps.apple.com\u002Fus\u002Fapp\u002Flocally-ai-local-ai-chat\u002Fid6741426692 For its size (1.2GB…",{"id":103,"by":104,"text":105},"47597484","robonot","really impressive for the size. Curious to see what happens when someone trains a 100B+ model natively at 1-bit.",{"value":107,"source":53,"evidence":108},"Runs on old CPUs and phones",[109,110],{"id":67,"by":68,"text":69},{"id":111,"by":112,"text":113},"47605891","w10-1","anecdotal experience report: They link the (free) locally.ai iPhone app, but the bonsai model doesn't present in the list. You have to get it via settings. On my ancient SE-2,…",[115,127,142],{"value":116,"source":53,"evidence":117},"Comparisons made against full-precision rather than quantized models",[118,119,123],{"id":84,"by":85,"text":86},{"id":120,"by":121,"text":122},"47595021","hatthew","I feel like it's a little disingenuous to compare against full-precision models. Anyone concerned about model size and memory usage is surely already using at least an 8 bit…",{"id":124,"by":125,"text":126},"47602738","WhitneyLand","A bit misleading to say they take 14x less memory, no one is doing inference with 16-bit models.",{"value":128,"source":53,"evidence":129},"Output quality is poor in some tests",[130,134,138],{"id":131,"by":132,"text":133},"47595856","fgfarben","> User: what color is the sky? > Assistant: .es | ) for the and > 1 \u002F up to to > ( or > > in : > the2 > it it?1",{"id":135,"by":136,"text":137},"47598712","druskacik","The 8B model response to my \"Harry Potter knowledge-bench\" question is too funny not to share. > *Fathers of Harry and James Potter*: - Sirius Black is the *father* of *James…",{"id":139,"by":140,"text":141},"47596285","UncleOxidant","Are you getting anything besides gibberish out of it? I tried their recommended commandline and it's dog slow even though I built their llama.cpp fork with AVX2 enabled. This is…",{"value":143,"source":53,"evidence":144},"Memory claim of 14x less is misleading",[145,146],{"id":124,"by":125,"text":126},{"id":60,"by":61,"text":62},[148,159],{"value":149,"source":53,"evidence":150},"Ollama and MLX support",[151,155],{"id":152,"by":153,"text":154},"47595248","keyle","Extremely cool! Can't wait to give it a spin with ollama, if ollama could list it as a model that would be helpful.",{"id":156,"by":157,"text":158},"47598651","vx_r","Eagerly waiting for mlx to merge 1bit quantization pr to try this out.",{"value":160,"source":53,"evidence":161},"Larger 1-bit models",[162,166],{"id":163,"by":164,"text":165},"47596886","ycui1986","i hope someone do a 100b 1-bit parameter model. that should fit into most 16GB graphics cards. local AI democratized.",{"id":103,"by":104,"text":105},[168,180],{"value":169,"source":53,"evidence":170},"How to run on CPU or Android",[171,172,176],{"id":78,"by":61,"text":79},{"id":173,"by":174,"text":175},"47594783","OutOfHere","How do I run this on Android?",{"id":177,"by":178,"text":179},"47598693","afaik69","any tutorial on how to run this on linux cpu only?",{"value":181,"source":53,"evidence":182},"Trade-off versus larger models",[183],{"id":184,"by":185,"text":186},"47594812","_fw","What’s the trade-off? If it’s smaller, faster and more efficient - is it worse performance? A layman here, curious to know.",[188,195],{"value":189,"source":53,"evidence":190},"Microsoft BitNet",[191],{"id":192,"by":193,"text":194},"47597836","est","is this somewhat similar to Microsofot Bitnet?",{"value":196,"source":53,"evidence":197},"Qwen3 quantized models",[198],{"id":84,"by":85,"text":86},{"comments":41,"answered":41,"key":200},[],{"commenters":202,"max_depth":203,"maker_replies":41,"first_comment_mins":202,"lifespan_hours":204},97,6,269,[206,208,212,215,217,218,222,225,228,230,234,236,239,243,247,249,251,255,258,259,261,263,266,269,272,274,275,278,281,283,285,288,290,293,296,299,301,304,306,308,311,314,316,319,322,324,326,330,334,337,340,343,346,350,354,356,358,361,365,368,370,374,377,382,383,387,391,395,397,401,406,410,414,416,419,422,425,428,430,433,435,438,441,445,448,451,454,457,459,461,465,467,469,471,473,477,480,483,485,489,493,496,499,502,504,508,512,514,516,519,523,526,528,530,534,538,540,545,548,551,555,558,561,564],[95,41,33,207,96,41],"project\u002Fpositive\u002Fperformance",[209,41,42,210,211,41],"47598689","project\u002Fmixed\u002Fexecution","Imustaskforhelp",[213,41,41,214,96,41],"47599054","commenter\u002Fneutral\u002Fnone",[216,41,41,214,211,41],"47599219",[99,44,33,207,100,41],[219,44,42,220,221,41],"47597107","project\u002Fneutral\u002Fnone","voxelghost",[223,44,41,214,224,41],"47598549","IshKebab",[226,44,41,227,221,41],"47609424","other\u002Fneutral\u002Fnone",[229,44,42,220,100,41],"47599562",[231,44,44,232,233,41],"47598112","project\u002Fcritical\u002Fperformance","newman314",[71,18,33,235,72,41],"project\u002Fpositive\u002Fnone",[237,18,33,207,238,41],"47597336","logicallee",[240,18,33,241,242,41],"47597694","maker\u002Fpositive\u002Fnone","TRCat",[244,18,42,245,246,41],"47597729","maker\u002Fneutral\u002Fexecution","ggerganov",[248,18,41,214,72,41],"47597790",[250,18,33,241,211,41],"47598399",[252,18,42,253,254,41],"47598733","project\u002Fmixed\u002Fperformance","ramon156",[256,18,42,257,72,41],"47599158","maker\u002Fneutral\u002Fnone",[67,42,42,253,68,41],[260,42,42,245,140,41],"47595726",[262,42,41,214,132,41],"47595808",[264,42,41,214,265,41],"47596222","cubefox",[267,42,41,214,268,41],"47596500","ddtaylor",[270,42,41,214,271,41],"47596663","boxedemp",[273,42,41,214,268,41],"47620556",[139,42,44,232,140,41],[276,42,42,277,140,41],"47607737","project\u002Fneutral\u002Fexecution",[279,42,42,277,280,41],"47611799","gorgonical",[282,42,42,277,140,41],"47618108",[284,42,33,207,280,41],"47718841",[286,48,41,227,287,41],"47594754","alyxya",[289,48,41,227,265,41],"47596481",[291,48,41,227,292,41],"47598611","hrmtst93837",[294,48,41,214,295,41],"47603979","guerrilla",[297,33,42,220,298,41],"47596004","drob518",[300,33,42,220,265,41],"47596399",[302,33,42,220,303,41],"47597858","londons_explore",[305,33,41,214,265,41],"47597896",[184,203,42,307,185,41],"project\u002Fneutral\u002Fperformance",[309,203,42,253,310,41],"47594903","kvdveer",[312,203,33,207,313,41],"47595145","adityashankar",[135,315,44,232,136,41],7,[317,38,41,227,318,41],"47598837","WaterRun",[320,38,41,214,321,41],"47599175","TheLNL",[323,38,41,214,318,41],"47601162",[84,325,44,232,85,41],9,[327,325,33,328,329,41],"47598260","project\u002Fpositive\u002Fidea","riedel",[331,39,44,332,333,41],"47595913","project\u002Fcritical\u002Fexecution","kent8192",[335,39,42,277,336,41],"47596068","liuliu",[338,39,42,220,339,41],"47602116","riidom",[341,39,42,277,342,41],"47607912","dragonwriter",[344,39,44,332,345,41],"47596957","dodos",[347,348,41,227,349,41],"47594785",11,"Archit3ch",[351,348,42,352,353,41],"47595432","project\u002Fneutral\u002Fidea","wmf",[78,355,42,277,61,41],12,[163,357,33,328,164,41],13,[359,34,33,328,360,41],"47600157","Udo",[362,363,42,352,364,41],"47595854",15,"plombe",[366,363,41,214,367,41],"47595922","Geee",[369,363,41,214,265,41],"47596343",[371,372,44,332,373,41],"47604043",16,"iJohnDoe",[375,372,42,277,376,41],"47605696","m0do1",[378,379,41,380,381,41],"47601080",17,"other\u002Fneutral\u002Fidea","ide0666",[60,31,44,232,61,41],[384,385,42,307,386,41],"47601151",19,"ant28",[388,389,33,328,390,41],"47595491",20,"marak830",[392,393,42,220,394,41],"47594600",21,"syntaxing",[111,396,33,207,112,41],22,[398,399,44,232,400,41],"47600006",23,"Nihilartikel",[402,403,44,404,405,41],"47595541",24,"project\u002Fcritical\u002Fidea","bilsbie",[407,403,41,408,409,41],"47596028","commenter\u002Fneutral\u002Fidea","MarsIronPI",[411,403,41,412,413,41],"47596949","commenter\u002Fneutral\u002Fexecution","syntaxpr",[415,403,41,214,409,41],"47601144",[417,403,41,380,418,41],"47596316","txrx0000",[420,403,41,227,421,41],"47597136","kennywinker",[423,403,41,227,424,41],"47597761","humanjir",[426,403,41,227,427,41],"47600037","adrian_b",[429,403,41,227,72,41],"47597979",[431,403,41,227,432,41],"47598455","kgeist",[434,403,41,227,427,41],"47599946",[436,403,41,227,437,41],"47601561","orbital-decay",[439,403,41,380,440,41],"47608679","in-silico",[442,443,42,352,444,41],"47594485",25,"stogot",[446,443,41,227,447,41],"47594516","trebligdivad",[449,443,41,227,450,41],"47594519","jacquesm",[452,443,41,227,453,41],"47594530","SwellJoe",[455,443,41,412,456,41],"47594939","jjcm",[458,443,41,227,132,41],"47595934",[131,460,44,232,132,41],26,[462,463,42,220,464,41],"47597573",27,"wshell",[466,463,41,227,72,41],"47597925",[468,463,41,227,464,41],"47599011",[124,470,44,232,125,41],28,[152,472,33,235,153,41],29,[474,475,33,207,476,41],"47595261",30,"ariwilson",[478,475,41,380,479,41],"47595382","onlyrealcuzzo",[481,475,41,380,482,41],"47597918","tim-projects",[484,475,44,232,140,41],"47596307",[486,487,42,307,488,41],"47596952",31,"ggamezar",[490,491,42,352,492,41],"47594397",32,"yodon",[494,491,42,277,495,41],"47594414","woadwarrior01",[497,491,41,380,498,41],"47596219","mchusma",[500,491,41,227,501,41],"47597114","NooneAtAll3",[156,503,33,235,157,41],33,[505,506,33,328,507,41],"47601314",34,"naasking",[509,32,42,510,511,41],"47599615","project\u002Fneutral\u002Fpricing","p0u4a",[103,513,33,207,104,41],36,[173,515,42,220,174,41],37,[517,515,42,220,518,41],"47595266","najarvg",[520,521,42,220,522,41],"47615521",38,"rpdaiml",[177,524,42,525,178,41],39,"project\u002Fneutral\u002Fdocs",[192,527,42,352,193,41],40,[529,527,42,352,507,41],"47601278",[531,532,42,352,533,41],"47604381",41,"aaroninsf",[535,536,41,227,537,41],"47735658",42,"aimemobe",[56,539,42,307,57,41],43,[541,542,42,543,544,41],"47602192",44,"project\u002Fmixed\u002Fidea","steffs",[546,542,41,214,547,41],"47602672","Havoc",[549,542,41,227,550,41],"47603459","Jowsey",[552,553,41,227,554,41],"47595559",45,"zephyrwhimsy",[556,553,41,214,557,41],"47595717","tacotime",[559,553,41,214,560,41],"47596505","noman-land",[562,553,41,214,563,41],"47597257","Dwedit",[120,565,44,232,121,41],46,[],[568,572,577,582,587,592],{"id":569,"name":570,"points":571,"category":20},"47536761","nullclaw",340,{"id":573,"name":574,"points":575,"category":576},"47557921","Git bayesect",337,"dev_tools",{"id":578,"name":579,"points":580,"category":581},"47580350","30u30.fyi",256,"other",{"id":583,"name":584,"points":585,"category":586},"47571183","korb",205,"consumer_social",{"id":588,"name":589,"points":590,"category":591},"47589856","pg_textsearch",203,"data",{"id":593,"name":594,"points":595,"category":591},"47534283","Turbolite",185,false,"v2",{"id":599,"name":5,"line":6,"ai":596,"cat":20,"year":13,"points":14,"comments":15,"rec":28,"s":600,"top":18,"kw":601,"maker":11,"domain":10,"tier":19,"attempts":41,"strip":602,"url":9,"mix":603,"praise":92,"objection":116,"sub":21},47593422,0.06,[24,25,26,27],"5534543014033313235115100030115135433301",[31,32,33,34],1794036697426]