[{"data":1,"prerenderedAt":268},["ShallowReactive",2],{"project-47026186":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":33,"aspects":34,"strip":42,"post_text":45,"what":46,"said":70,"maker_in_thread":143,"shape":154,"grid":158,"attempts":229,"same_week":230,"maker_shows":32,"maker_hidden":260,"reader_version":261,"card":262},"47026186","Microgpt","Microgpt is a GPT you can visualize in the browser","title","Show HN: Microgpt is a GPT you can visualize in the browser","https:\u002F\u002Fmicrogpt.boratto.ca","microgpt.boratto.ca","b44",1771180835,2026,282,24,161,4366,4,"legend","learning_science","educational ml visualization","web_app",[24,25,26,27],"gpt visualization","micro gpt","neural network visualizer","learn llm","curious",0.207,{"positive":18,"neutral":31,"mixed":32,"critical":32},6,1,12,{"design":35,"performance":38,"docs":39,"execution":40,"idea":41},{"positive":36,"critical":37},2,0,{"positive":37,"critical":37},{"positive":37,"critical":37},{"positive":32,"critical":32},{"positive":37,"critical":37},[43,37,37,44,37,37,44,37,37,44,37,44,37,32,37,44,37,37,44,37,37,43,43,43,44],5,3,"very much inspired by karpathy's microgpt of the same name. it's (by default) a 4000 param GPT\u002FLLM\u002FNN that learns to generate names. this is sorta an educational tool in that you can visualize the activations as they pass through the network, and click on things to get an explanation of them.",{"problem":47,"audience":53,"pricing":62,"motivation":63,"time_to_build":62,"stack":67,"openness":68,"stage":69,"team_size":68},{"value":48,"source":49,"evidence":50},"Makes how a small GPT works visible by showing activations and explaining them","maker",[51],{"id":52},"post",{"value":54,"source":55,"evidence":56},"People learning how LLMs work","inferred",[57,58],{"id":52},{"id":59,"by":60,"text":61},"47030891","stevage","I'd love to understand how LLMs work, but this site assumed a bit too much knowledge for me to get much from it. Looks cool though.",null,{"value":64,"source":49,"evidence":65},"Educational tool inspired by Karpathy's microgpt",[66],{"id":52},[],"unknown","launched",{"praise":71,"objections":88,"requests":104,"questions":116,"alternatives":131},[72],{"value":73,"source":74,"evidence":75},"Well presented and easy to follow compared to other LLM visualizers","commenter",[76,80,84],{"id":77,"by":78,"text":79},"47032810","kengoa","Amazing work! Reminded me of LLM Visualization (https:\u002F\u002Fbbycroft.net\u002Fllm) except this is a lot easier to wrap my head around and that I can actually run the training loops, which…",{"id":81,"by":82,"text":83},"47040525","armcat","Really nicely presented, well done!",{"id":85,"by":86,"text":87},"47033632","youio","really well done",[89,93],{"value":90,"source":74,"evidence":91},"Assumes too much background knowledge for beginners",[92],{"id":59,"by":60,"text":61},{"value":94,"source":74,"evidence":95},"Output names are not very good with a small model",[96,100],{"id":97,"by":98,"text":99},"47032933","lucrbvi","It's just hallucinating training data, the model is very small so it cannot be useful at all",{"id":101,"by":102,"text":103},"47032899","RugnirViking","I was a little confused by \"see, its much better\" when the output is stuff like isovrak and kucey. What is it supposed to be generating?",[105,109],{"value":106,"source":74,"evidence":107},"Add explainer for what the loss value means",[108],{"id":77,"by":78,"text":79},{"value":110,"source":74,"evidence":111},"Fix sidebar showing 16x16 matrices instead of true dimensions",[112],{"id":113,"by":114,"text":115},"47031284","GaggiX","Wtok and Wpos should be 26-dim along one of the axis but it shows a 16x16 matrix be default, fc1 instead 16x64 with the default settings (not 16x16).",[117,124],{"value":118,"source":74,"evidence":119},"How many training steps are needed for good output",[120],{"id":121,"by":122,"text":123},"47028101","msla","About how many training steps are required to get good output?",{"value":125,"source":74,"evidence":126},"Is it character-based rather than token-based",[127],{"id":128,"by":129,"text":130},"47027649","kfsone","Minor nit: In familiarity, you gloss over the fact that it's character rather than token based which might be worth a shout out: \"Microgpt's larger cousins using building blocks…",[132,136],{"value":133,"source":74,"evidence":134},"LLM Visualization by bbycroft",[135],{"id":77,"by":78,"text":79},{"value":137,"source":74,"evidence":138},"Karpathy's microgpt blog post",[139],{"id":140,"by":141,"text":142},"47033044","interloxia","The linked inspiration has a good blog post of microgpt implemented in python. https:\u002F\u002Fkarpathy.github.io\u002F2026\u002F02\u002F12\u002Fmicrogpt\u002F It was submitted to hn a few days ago but only…",{"comments":18,"answered":36,"key":144},[145,148,151],{"id":146,"text":147},"47033544","the untrained model is literally just generating random characters, whereas your examples are at least pronouncable. you can add more layers to get progressively better results.",{"id":149,"text":150},"47028165","not many. diminishing returns start before 1000 and past that you should just add a second\u002Fthird layer",{"id":152,"text":153},"47031363","good catch - i intentionally cap node visualizations at 16 so it doesn't get super long, but the sidebar shouldn't have that",{"commenters":155,"max_depth":36,"maker_replies":18,"first_comment_mins":156,"lifespan_hours":157},21,158,27,[159,161,163,166,168,170,172,174,177,180,181,183,186,188,190,192,194,197,201,206,210,214,217,221,224],[77,37,43,160,78,37],"project\u002Fpositive\u002Fdesign",[140,32,37,162,141,37],"other\u002Fneutral\u002Fnone",[164,36,37,162,165,37],"47029022","krackers",[101,44,44,167,102,37],"project\u002Fneutral\u002Fperformance",[97,44,37,169,98,37],"commenter\u002Fcritical\u002Fperformance",[146,44,37,171,11,32],"commenter\u002Fneutral\u002Fperformance",[59,18,44,173,60,37],"project\u002Fneutral\u002Fdocs",[175,18,37,162,176,37],"47031059","BloondAndDoom",[178,18,37,162,179,37],"47033304","alansaber",[121,43,44,167,122,37],[149,43,37,182,11,32],"maker\u002Fneutral\u002Fperformance",[184,43,44,167,185,37],"47029358","WatchDog",[187,43,37,171,179,37],"47033310",[113,31,32,189,114,37],"project\u002Fcritical\u002Fexecution",[152,31,37,191,11,32],"maker\u002Fneutral\u002Fexecution",[128,193,44,173,129,37],7,[195,193,37,196,11,32],"47027779","maker\u002Fneutral\u002Fidea",[198,193,37,199,200,37],"47028571","commenter\u002Fneutral\u002Fidea","mips_avatar",[202,203,44,204,205,37],"47033370",8,"project\u002Fmixed\u002Fexecution","ramon156",[207,208,37,162,209,37],"47033701",9,"keepamovin",[211,212,37,162,213,37],"47038906",10,"prakashdep",[81,215,43,216,82,37],11,"maker\u002Fpositive\u002Fdesign",[218,33,43,219,220,37],"47029013","maker\u002Fpositive\u002Fnone","darepublic",[85,222,43,223,86,37],13,"maker\u002Fpositive\u002Fexecution",[225,226,44,227,228,37],"47031347",14,"project\u002Fneutral\u002Fidea","umairnadeem123",[],[231,236,241,246,250,255],{"id":232,"name":233,"points":234,"category":235},"47031334","Jemini",488,"ai",{"id":237,"name":238,"points":239,"category":240},"47051852","AsteroidOS 2.0",470,"hardware",{"id":242,"name":243,"points":244,"category":245},"47040375","FreeFlow",277,"productivity",{"id":247,"name":248,"points":249,"category":20},"47008163","Data Engineering Book",251,{"id":251,"name":252,"points":253,"category":254},"47011567","SQL-tap",233,"dev_tools",{"id":256,"name":257,"points":258,"category":259},"47014500","Sameshi",230,"games",false,"v2",{"id":263,"name":5,"line":6,"ai":260,"cat":20,"year":13,"points":14,"comments":15,"rec":28,"s":264,"top":18,"kw":265,"maker":11,"domain":10,"tier":19,"attempts":37,"strip":266,"url":9,"mix":267,"praise":73,"objection":90,"sub":21},47026186,0.21,[24,25,26,27],"5003003003030103003005553",[18,31,32,32],1794036693315]