[{"data":1,"prerenderedAt":223},["ShallowReactive",2],{"project-37398812":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":41,"post_text":43,"what":44,"said":75,"maker_in_thread":147,"shape":156,"grid":158,"attempts":187,"same_week":188,"maker_shows":32,"maker_hidden":215,"reader_version":216,"card":217},"37398812","Perplexity","Fully client-side GPT2 prediction visualizer","title","Show HN: Fully client-side GPT2 prediction visualizer","https:\u002F\u002Fperplexity.vercel.app\u002F","perplexity.vercel.app","thesephist",1693953759,2023,153,11,406,3109,14,"notable","ai","model visualization","web_app",[24,25,26,27],"gpt-2","token prediction","llm visualization","transformers.js","mixed",-0.043,{"positive":31,"neutral":32,"mixed":33,"critical":34},2,4,0,3,9,{"docs":37,"idea":39,"design":40},{"positive":38,"critical":33},1,{"positive":38,"critical":38},{"positive":33,"critical":31},[42,34,33,38,42,38,38,34,34,33,33,34,33],5,null,{"problem":45,"audience":52,"pricing":43,"motivation":56,"time_to_build":43,"stack":60,"openness":73,"stage":74,"team_size":73},{"value":46,"source":47,"evidence":48},"Shows how a small language model predicts each token of a piece of text","maker",[49],{"id":50,"by":11,"text":51},"37398858","Hi HN! I've found this visualization tool immensely helpful over the years for getting an intuition for how an LLM \"sees\" some piece of text, and with a bit of elbow grease…",{"value":53,"source":47,"evidence":54},"People wanting intuition for how GPT-style models see text",[55],{"id":50,"by":11,"text":51},{"value":57,"source":47,"evidence":58},"Found the visualization useful and wanted to make it publicly available by running compute client-side",[59],{"id":50,"by":11,"text":51},[61,69],{"value":62,"source":63,"evidence":64},"Transformers.js","commenter",[65],{"id":66,"by":67,"text":68},"37401157","simonw","There's a video of a previous version of this tool here which I found really helped me understand what it was demonstrating:…",{"value":70,"source":63,"evidence":71},"ONNX GPT-2 model",[72],{"id":66,"by":67,"text":68},"unknown","launched",{"praise":76,"objections":89,"requests":104,"questions":126,"alternatives":134},[77,85],{"value":78,"source":63,"evidence":79},"Helpful for building intuition about how LMs predict text",[80,81],{"id":66,"by":67,"text":68},{"id":82,"by":83,"text":84},"37403022","thomasfromcdnjs","This is beautiful. Needs to be a standard tool for all models. Great work!",{"value":86,"source":63,"evidence":87},"Runs entirely in the browser",[88],{"id":66,"by":67,"text":68},[90,97],{"value":91,"source":63,"evidence":92},"Red and green highlights are hard to distinguish for colour-blind users",[93],{"id":94,"by":95,"text":96},"37439687","netipulk","The highlights being very similar in red and green is a complete nightmare for me because I have deuteranopia. You should probably fix that.",{"value":98,"source":63,"evidence":99},"Runs on page load, which is bad for mobile users",[100],{"id":101,"by":102,"text":103},"37405847","skybrian","I’m sure it’s neat but it shouldn’t start running on load, because some people are browsing on mobile.",[105,112,119],{"value":106,"source":63,"evidence":107},"Llama 2 support",[108],{"id":109,"by":110,"text":111},"37401383","Scene_Cast2","Any chance of Llama2 support?",{"value":113,"source":63,"evidence":114},"Attention visualization",[115],{"id":116,"by":117,"text":118},"37401721","atgctg","It would be interesting to have attention visualized as well, similar to how it's done in BertViz: https:\u002F\u002Fgithub.com\u002Fjessevig\u002Fbertviz",{"value":120,"source":63,"evidence":121},"Analysis of user's own text",[122],{"id":123,"by":124,"text":125},"37402196","davotoula","Can you analyse your own text since the model has been downloaded to client?",[127],{"value":128,"source":63,"evidence":129},"Whether predictability matches human sense of information density",[130],{"id":131,"by":132,"text":133},"37400608","didgeoridoo","Really interesting! I wonder how well this syncs up with human intuition and general “information density”. If it’s a close match, maybe you could use this as a tool to help with…",[135,143],{"value":136,"source":63,"evidence":137},"BertViz",[138,139],{"id":116,"by":117,"text":118},{"id":140,"by":141,"text":142},"37403694","isaacfung","I really enjoyed playing with BertViz. Another similar visual exploration tool I found recently is in Edwin Chen's blog. I am pretty sure this is the best explanation of LSTM. I…",{"value":144,"source":63,"evidence":145},"TensorFlow Playground",[146],{"id":140,"by":141,"text":142},{"comments":34,"answered":38,"key":148},[149,152,155],{"id":150,"text":151},"37401390","This is definitely an interesting idea I've also pondered before. In my experience (just speaking from intuition) what's \"easy\" for LMs to predict often doesn't line up with our…",{"id":153,"text":154},"37401398","Definitely possible if supported by transformers.js. If I see enough folks wanting it I'll likely add it at some point.",{"id":50,"text":51},{"commenters":15,"max_depth":31,"maker_replies":34,"first_comment_mins":32,"lifespan_hours":157},71,[159,161,163,165,169,171,173,174,175,178,180,183,185],[66,33,42,160,67,33],"project\u002Fpositive\u002Fdocs",[131,38,34,162,132,33],"project\u002Fneutral\u002Fidea",[150,38,33,164,11,38],"commenter\u002Fneutral\u002Fidea",[166,38,38,167,168,33],"37402997","project\u002Fcritical\u002Fidea","tgv",[82,31,42,170,83,33],"project\u002Fpositive\u002Fidea",[94,34,38,172,95,33],"project\u002Fcritical\u002Fdesign",[101,32,38,172,102,33],[123,42,34,162,124,33],[109,176,34,177,110,33],6,"project\u002Fneutral\u002Fnone",[153,176,33,179,11,38],"maker\u002Fneutral\u002Fnone",[50,181,33,182,11,38],7,"maker\u002Fneutral\u002Fdocs",[116,184,34,162,117,33],8,[140,184,33,186,141,33],"other\u002Fneutral\u002Fnone",[],[189,194,198,202,207,211],{"id":190,"name":191,"points":192,"category":193},"37408150","smolsite.zip",602,"dev_tools",{"id":195,"name":196,"points":197,"category":193},"37391848","Puck",451,{"id":199,"name":200,"points":201,"category":193},"37393099","RecipeUI",275,{"id":203,"name":204,"points":205,"category":206},"37381268","Keep",181,"infrastructure",{"id":208,"name":209,"points":210,"category":193},"37364124","Shaq",155,{"id":212,"name":213,"points":214,"category":193},"37371237","RSSfeedASAP",151,false,"v2",{"id":218,"name":5,"line":6,"ai":215,"cat":20,"year":13,"points":14,"comments":15,"rec":28,"s":219,"top":18,"kw":220,"maker":11,"domain":10,"tier":19,"attempts":33,"strip":221,"url":9,"mix":222,"praise":78,"objection":91,"sub":21},37398812,-0.04,[24,25,26,27],"5301511330030",[31,32,33,34],1794036641362]