[{"data":1,"prerenderedAt":118},["ShallowReactive",2],{"maker-sillysaurusx":3},[4,30,50,73,98],{"id":5,"name":6,"line":7,"ai":8,"cat":9,"year":10,"points":11,"comments":12,"rec":13,"s":14,"top":15,"kw":16,"maker":21,"domain":6,"tier":22,"attempts":23,"strip":24,"url":25,"mix":26,"praise":28,"objection":-1,"sub":29},21877210,"pastebin.com","I trained a GPT-2 1.5B Subreddit Simulator in 3 days using 88 TPUs",false,"ai",2019,10,3,"quiet",null,82,[17,18,19,20],"gpt-2","tpu training","language model","subreddit simulator","sillysaurusx","shown",0,"030","https:\u002F\u002Fpastebin.com\u002F5qKQs2Hk",[23,27,23,23],1,"Swarm training across TPUs is cheaper than a TPU pod","language model training",{"id":31,"name":32,"line":32,"ai":8,"cat":33,"year":34,"points":35,"comments":36,"rec":13,"s":14,"top":37,"kw":38,"maker":21,"domain":43,"tier":22,"attempts":23,"strip":44,"url":45,"mix":46,"praise":47,"objection":48,"sub":49},27442644,"Show descendant count on Hacker News 'more' links","dev_tools",2021,6,8,90,[39,40,41,42],"hacker news","tampermonkey","userscript","comment threads","gist.github.com","000000000","https:\u002F\u002Fgist.github.com\u002Fshawwn\u002F3c1b71d69482afde15efb163925f2c7a",[23,23,23,23],"Descendant count makes it clear how much of a subtree is hidden","Collapsing or moving subthreads to other pages could hurt conversation flow","browser userscript",{"id":51,"name":52,"line":52,"ai":8,"cat":53,"year":34,"points":54,"comments":55,"rec":56,"s":57,"top":58,"kw":59,"maker":21,"domain":43,"tier":65,"attempts":23,"strip":66,"url":67,"mix":68,"praise":70,"objection":71,"sub":72},29128998,"How does JAX allocate memory on a TPU? An interactive C++ walkthrough","learning_science",78,28,"mixed",-0.1,26,[60,61,62,63,64],"jax","tpu","memory allocation","xla","c++","notable","3133100000000003053301330500","https:\u002F\u002Fgist.github.com\u002Fshawwn\u002F0e524d4a7a5d8fb152a86616559cc02a",[69,35,69,12],2,"Jax debugging and TPU VM access praised as enjoyable","Memory scheduling claims in XLA questioned as unverified","interactive tutorial",{"id":74,"name":75,"line":76,"ai":8,"cat":9,"year":77,"points":78,"comments":79,"rec":80,"s":81,"top":82,"kw":83,"maker":21,"domain":88,"tier":89,"attempts":23,"strip":90,"url":91,"mix":92,"praise":95,"objection":96,"sub":97},35026902,"Llama-dl","high-speed download of LLaMA, Facebook's 65B GPT model",2023,343,130,"curious",0.08,4,[84,85,86,87],"llama","llm weights download","65b model","llama leak","github.com","legend","5300003030033355353343333133451000333330","https:\u002F\u002Fgithub.com\u002Fshawwn\u002Fllama-dl",[93,94,23,82],9,46,"Running LLaMA on a single A100 gives output as good as davinci.","Cloud GPU cost makes running the 65B model expensive.","model weights downloader",{"id":99,"name":100,"line":101,"ai":8,"cat":33,"year":102,"points":103,"comments":104,"rec":56,"s":105,"top":106,"kw":107,"maker":21,"domain":111,"tier":22,"attempts":23,"strip":112,"url":113,"mix":114,"praise":115,"objection":116,"sub":117},49996495,"sharc","Pointless but mostly-exact clone of Hacker News",2026,16,7,0.22,53,[108,109,100,110],"hacker news clone","common lisp","web forum","news.ycombinator.lol","1355035","https:\u002F\u002Fnews.ycombinator.lol",[12,69,23,27],"Good job and awesome codebase, impressive hard work","Pages load about twice as slow as real HN","web app clone",1794036769215]