[{"data":1,"prerenderedAt":610},["ShallowReactive",2],{"project-49158333":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":29,"sentiment":30,"stance_mix":31,"scored_comments":36,"aspects":37,"strip":45,"post_text":48,"what":49,"said":75,"maker_in_thread":171,"shape":178,"grid":182,"attempts":571,"same_week":572,"maker_shows":47,"maker_hidden":602,"reader_version":603,"card":604},"49158333","Swiftlet","Run an 80B Qwen in 4.3 GB of RAM on a Mac, and a 35B on an iPhone","title","Show HN: Run an 80B Qwen in 4.3 GB of RAM on a Mac, and a 35B on an iPhone","https:\u002F\u002Fgithub.com\u002Fleonickson1\u002FSwiftlet","github.com","leonickson",1785776055,2026,312,141,140,4366,4,"legend","ai","local inference","cli",[24,25,26,27,28],"local llm","mixture of experts","on-device inference","ssd offload","apple silicon","curious",0.014,{"positive":32,"neutral":33,"mixed":34,"critical":35},8,18,0,7,33,{"idea":38,"performance":40,"execution":41,"pricing":42,"docs":44},{"positive":39,"critical":39},2,{"positive":18,"critical":18},{"positive":34,"critical":34},{"positive":34,"critical":43},1,{"positive":43,"critical":34},[46,34,34,34,34,34,34,34,47,34,34,34,34,34,34,34,43,34,34,47,46,34,34,34,34,34,46,47,47,34,39,43,47,46,43,47,47,47,43,46],5,3,null,{"problem":50,"audience":48,"pricing":48,"motivation":48,"time_to_build":48,"stack":64,"openness":72,"stage":73,"team_size":74},{"value":51,"source":52,"evidence":53},"Running very large models on consumer Macs and iPhones with limited RAM.","inferred",[54,56,60],{"id":55},"post",{"id":57,"by":58,"text":59},"49164441","CyLith","I know relatively little about the workings of LLMs, but I keep seeing projects like this that run massive MoE models using very modest amounts of RAM, perhaps excessively so. I…",{"id":61,"by":62,"text":63},"49164268","dghlsakjg","I know everyone wants to crap all over these setups that are impractical, but this is how progress happens. People will keep plugging away at this and figure out how to avoid…",[65],{"value":66,"source":67,"evidence":68},"Claude Code used during development","maker",[69],{"id":70,"by":11,"text":71},"49167509","no Anthropic involvement, I just used Claude Code heavily while building this and putting that in the README felt more honest than not mentioning it. Now that I think about it…","open_source","beta","unknown",{"praise":76,"objections":93,"requests":118,"questions":137,"alternatives":152},[77,86],{"value":78,"source":79,"evidence":80},"Progress on on-device inference is exciting","commenter",[81,82],{"id":61,"by":62,"text":63},{"id":83,"by":84,"text":85},"49169553","nc55g3g","Running a 35B on iPhone at 1 tok\u002Fs with 2.5 GB RAM… this is the future of on-device inference. Insane work.",{"value":87,"source":79,"evidence":88},"Useful project that makes big models run locally",[89],{"id":90,"by":91,"text":92},"49164685","adrianco","This looks useful, you can increase the RAM cache so if you have a Mac with 24-32GB it should speed up a lot and still run models that wouldn’t normally fit. I’m going to run…",[94,104,111],{"value":95,"source":67,"evidence":96},"Prefill is too slow for long prompts",[97,101],{"id":98,"by":99,"text":100},"49164178","brrrrrm","this is cool but like, are we just vibe coding NAND burners at this point? these decode times don't really tell the whole story, because prefill becomes the bottleneck. half an…",{"id":102,"by":11,"text":103},"49167383","agree, prefill is the weak spot right now. it goes through the same per-token path as decode, which is dumb for long prompts. The fix is on the list: during prefill we can batch…",{"value":105,"source":79,"evidence":106},"SSD offload is slow and may wear out drives",[107],{"id":108,"by":109,"text":110},"49164210","jbird99","At what, 10 tokens per hour? These disk swapping methods all have the same drawbacks - kill your drive early, and slow as hell.",{"value":112,"source":79,"evidence":113},"Not practical beyond chat-length prompts",[114],{"id":115,"by":116,"text":117},"49166284","pizza234","> but this is how progress happens. This is progress in the same way that a man climbing a tree is making progress toward reaching the moon. This project is essentially the…",[119,123,130],{"value":120,"source":79,"evidence":121},"Tunable RAM cache usage on larger-memory Macs",[122],{"id":57,"by":58,"text":59},{"value":124,"source":79,"evidence":125},"Support for Android, Linux and Windows",[126],{"id":127,"by":128,"text":129},"49172581","erelong","Is there something that already runs like this on android \u002F linux ( \u002F windows)? (ollama or something?) Or could this be ported to work on other such platforms? edit: AI mentions…",{"value":131,"source":79,"evidence":132},"Support for other models such as Gemma",[133],{"id":134,"by":135,"text":136},"49171227","lenerdenator","How can I do this with, say, Gemma?",[138,145],{"value":139,"source":79,"evidence":140},"How does web access work with local models?",[141],{"id":142,"by":143,"text":144},"49167198","gizmodo59","The web and connecting to other services is very important for almost all of my use cases. While I believe we are going to get better and faster models, the web index is…",{"value":146,"source":79,"evidence":147},"Why was Claude Code credited in the README?",[148],{"id":149,"by":150,"text":151},"49164713","crossroadsguy","I see this at the end of the README > Swiftlet was built in collaboration with Claude Code. Did this really happen (some sort of working with Anthropic or Claude Code team) or is…",[153,160,167],{"value":154,"source":79,"evidence":155},"BigMoeOnEdge",[156],{"id":157,"by":158,"text":159},"49171091","lern_too_spel","See also: https:\u002F\u002Fgithub.com\u002FHelldez\u002FBigMoeOnEdge",{"value":161,"source":79,"evidence":162},"TurboFieldfare",[163],{"id":164,"by":165,"text":166},"49165069","gitpusher42","Thank you for using TurboFieldfare as a starting point for this project and thank you for mentioning it at the README. I am glad it inspired more people to explore area of…",{"value":168,"source":79,"evidence":169},"Ollama",[170],{"id":127,"by":128,"text":129},{"comments":47,"answered":47,"key":172},[173,176,177],{"id":174,"text":175},"49167449","It's tunable, --cache-gb N on the CLI. In my sweep the speed barely moved between a 1GB and 6GB cache (43% vs 70% hit rate, same tok\u002Fs) because right now the bottleneck is GPU…",{"id":102,"text":103},{"id":70,"text":71},{"commenters":179,"max_depth":35,"maker_replies":47,"first_comment_mins":180,"lifespan_hours":181},106,651,116,[183,185,189,193,195,198,201,205,208,211,215,218,221,224,227,230,232,235,238,242,245,248,251,253,256,258,261,263,266,269,272,274,278,281,283,285,288,291,293,296,300,304,307,310,313,316,318,320,323,326,329,331,334,336,339,342,345,348,350,353,356,359,362,365,368,370,373,375,378,382,385,388,391,394,396,399,402,405,408,411,414,417,419,422,425,428,430,433,436,439,442,444,447,449,451,452,455,458,461,464,465,467,470,473,477,480,484,487,489,491,493,495,498,501,505,508,511,513,515,518,521,523,525,526,529,532,534,536,539,542,544,546,549,551,553,555,558,561,564,567,568],[61,34,46,184,62,34],"maker\u002Fpositive\u002Fidea",[186,34,34,187,188,34],"49164330","commenter\u002Fneutral\u002Fnone","arjie",[190,34,34,191,192,34],"49164361","other\u002Fneutral\u002Fperformance","hedora",[194,34,34,191,188,34],"49174301",[196,34,34,191,197,34],"49164537","apimade",[199,34,34,191,200,34],"49164951","cududa",[202,34,34,203,204,34],"49165015","other\u002Fneutral\u002Fpricing","xbmcuser",[206,34,34,203,207,34],"49165034","formerly_proven",[209,34,34,191,210,34],"49165071","zmmmmm",[212,34,34,213,214,34],"49165106","other\u002Fneutral\u002Fnone","jack_pp",[216,34,34,213,217,34],"49165442","kaashif",[219,34,34,187,220,34],"49165452","fragmede",[222,34,34,187,223,34],"49165740","darkwater",[225,34,34,187,226,34],"49165583","cvak",[228,34,34,191,229,34],"49165174","root_axis",[231,34,34,191,197,34],"49165653",[233,34,34,187,234,34],"49165363","flaburgan",[236,34,34,187,237,34],"49166196","dtj1123",[239,34,34,240,241,34],"49165766","other\u002Fneutral\u002Fdesign","foxrider",[243,34,34,240,244,34],"49165925","SJC_Hacker",[246,34,34,191,247,34],"49166085","HPsquared",[249,34,34,240,250,34],"49166195","naasking",[252,34,34,240,241,34],"49167112",[254,34,34,213,255,34],"49166112","kaelwd",[257,34,34,213,197,34],"49166218",[259,34,34,191,260,34],"49166684","Azantys",[262,34,34,187,62,34],"49168739",[264,34,34,187,265,34],"49171980","ksec",[267,34,34,213,268,34],"49165647","sh3rl0ck",[270,34,47,271,188,34],"49174288","maker\u002Fneutral\u002Fperformance",[273,34,34,187,268,34],"49221333",[275,34,34,276,277,34],"49164927","other\u002Fneutral\u002Fidea","gpt5",[279,34,34,213,280,34],"49165073","anon373839",[282,34,34,213,277,34],"49165196",[284,34,34,213,280,34],"49165298",[286,34,34,213,287,34],"49165569","xyzsparetimexyz",[289,34,34,276,290,34],"49165699","zozbot234",[292,34,34,213,62,34],"49191333",[294,34,34,187,295,34],"49165550","submeta",[297,34,34,298,299,34],"49166090","commenter\u002Fneutral\u002Fprivacy","mihaaly",[301,34,34,302,303,34],"49166197","other\u002Fneutral\u002Fprivacy","hobofan",[305,34,34,276,306,34],"49167234","dudefeliciano",[308,34,34,213,309,34],"49165389","marci",[311,34,34,187,312,34],"49165716","josu",[314,34,34,187,315,34],"49165765","l33tman",[317,34,34,187,309,34],"49170836",[319,34,34,187,312,34],"49179383",[321,34,34,302,322,34],"49166168","croes",[324,34,34,187,325,34],"49166200","cromka",[327,34,34,187,328,34],"49167012","bestham",[330,34,34,187,325,34],"49167758",[332,34,34,187,333,34],"49168556","JKCalhoun",[335,34,34,187,322,34],"49172589",[337,34,34,276,338,34],"49166271","myrmidon",[340,34,34,213,341,34],"49166586","Bayart",[343,34,34,213,344,34],"49168369","mannanj",[346,34,34,213,347,34],"49175476","RachelF",[115,34,43,349,116,34],"project\u002Fcritical\u002Fidea",[351,34,34,187,352,34],"49166773","AlecSchueler",[354,34,34,191,355,34],"49166780","rhdunn",[357,34,34,187,358,34],"49166781","sipjca",[360,34,34,187,361,34],"49168975","Hugsbox",[363,34,34,187,364,34],"49169478","trymas",[366,34,34,187,367,34],"49171948","mv4",[369,34,34,187,364,34],"49179941",[371,34,34,187,372,34],"49169972","bluerooibos",[374,34,34,187,367,34],"49172004",[376,34,34,213,377,34],"49166624","hagen8",[379,34,47,380,381,34],"49166764","project\u002Fneutral\u002Fperformance","Perz1val",[383,34,47,380,384,34],"49167105","efficax",[386,34,34,213,387,34],"49167484","jarek83",[389,34,46,390,333,34],"49168516","project\u002Fpositive\u002Fidea",[392,34,47,380,393,34],"49169151","ac29",[395,34,34,213,265,34],"49171918",[397,43,34,213,398,34],"49164282","AHASIC",[400,43,34,213,401,34],"49164324","Mistletoe",[403,43,34,213,404,34],"49164394","sudo_cowsay",[406,43,34,213,407,34],"49164425","bobbylarrybobby",[409,43,34,213,410,34],"49164913","swiftcoder",[412,43,34,213,413,34],"49164481","CircuitSeuss",[415,43,34,187,416,34],"49167386","jsjohnst",[418,43,34,187,393,34],"49169311",[420,43,34,191,421,34],"49164692","ethersteeds",[423,43,34,213,424,34],"49164981","harrouet",[426,43,34,187,427,34],"49165843","KeplerBoy",[429,43,34,187,424,34],"49167677",[431,43,34,213,432,34],"49169494","xprnio",[434,43,34,213,435,34],"49170812","dgently7",[437,43,34,213,438,34],"49164675","greggsy",[440,43,34,213,441,34],"49165189","dd8601fn",[443,43,34,213,62,34],"49168909",[445,43,34,213,446,34],"49170250","red_green_yell",[164,39,46,448,165,34],"maker\u002Fpositive\u002Fnone",[90,47,46,450,91,34],"project\u002Fpositive\u002Fperformance",[57,18,47,380,58,34],[453,18,47,380,454,34],"49164455","zamadatix",[456,18,47,380,457,34],"49164460","ianmurrays",[459,18,47,380,460,34],"49165247","spockz",[462,18,34,213,463,34],"49165348","fodkodrasz",[174,18,34,380,11,43],[142,46,47,466,143,34],"project\u002Fneutral\u002Fidea",[468,46,34,213,469,34],"49168644","pbronez",[471,46,34,213,472,34],"49168804","morgoo",[474,475,34,213,476,34],"49165914",6,"throwawayffffas",[478,475,34,213,479,34],"49166515","sbsbdbdbfndj",[481,475,47,482,483,34],"49171163","project\u002Fneutral\u002Fexecution","tredre3",[485,35,43,349,486,34],"49165648","sallymander",[127,32,47,488,128,34],"project\u002Fneutral\u002Fnone",[134,490,47,488,135,34],9,[492,490,47,488,165,34],"49171900",[157,494,34,213,158,34],10,[98,496,43,497,99,34],11,"project\u002Fcritical\u002Fperformance",[499,496,46,450,500,34],"49164279","kennywinker",[502,496,43,503,504,34],"49164921","project\u002Fcritical\u002Fpricing","selcuka",[506,496,34,203,507,34],"49165188","hdgvhicv",[509,496,34,203,510,34],"49166338","wccrawford",[512,496,34,203,507,34],"49200363",[514,496,47,488,500,34],"49165466",[516,496,46,450,517,34],"49165075","fsuts",[519,496,43,497,520,34],"49165488","IsTom",[522,496,34,191,355,34],"49166925",[524,496,47,380,520,34],"49167231",[102,496,34,271,11,43],[527,496,34,213,528,34],"49170472","piyh",[83,530,46,531,84,34],12,"maker\u002Fpositive\u002Fperformance",[108,533,43,497,109,34],13,[535,533,47,380,500,34],"49164266",[537,533,34,213,538,34],"49164267","wat10000",[540,533,34,213,541,34],"49164383","Alpha3031",[543,533,34,213,538,34],"49172365",[545,533,34,187,404,34],"49164398",[547,533,47,380,548,34],"49164725","petu",[550,533,34,187,290,34],"49165627",[552,533,43,497,548,34],"49166695",[554,533,34,213,476,34],"49165946",[556,533,34,213,557,34],"49189132","a96",[149,559,47,560,150,34],14,"maker\u002Fneutral\u002Fdocs",[562,559,34,213,563,34],"49164872","iamflimflam1",[565,559,34,213,566,34],"49165331","vermarish",[70,559,34,560,11,43],[569,559,46,570,150,34],"49179698","maker\u002Fpositive\u002Fdocs",[],[573,578,583,587,592,597],{"id":574,"name":575,"points":576,"category":577},"49124218","Elevators",1680,"learning_science",{"id":579,"name":580,"points":581,"category":582},"49170165","Inclusive Color Space",627,"creative",{"id":584,"name":585,"points":586,"category":582},"49149966","Isopolis",350,{"id":588,"name":589,"points":590,"category":591},"49140396","Cycloidal Gearbox",338,"hardware",{"id":593,"name":594,"points":595,"category":596},"49145937","Kakehashi",255,"infrastructure",{"id":598,"name":599,"points":600,"category":601},"49119425","Gander",211,"dev_tools",false,"v2",{"id":605,"name":5,"line":6,"ai":602,"cat":20,"year":13,"points":14,"comments":15,"rec":29,"s":606,"top":18,"kw":607,"maker":11,"domain":10,"tier":19,"attempts":34,"strip":608,"url":9,"mix":609,"praise":78,"objection":95,"sub":21},49158333,0.01,[24,25,26,27,28],"5000000030000000100350000053302135133315",[32,33,34,35],1794036707019]