[{"data":1,"prerenderedAt":381},["ShallowReactive",2],{"project-29179429":3},{"id":4,"name":5,"line":6,"line_source":7,"title":8,"url":9,"domain":9,"maker":10,"posted_at":11,"year":12,"points":13,"comments":14,"rank_in_year":15,"shown_in_year":16,"top":17,"tier":18,"category":19,"subcategory":20,"format":21,"keywords":22,"reception":27,"sentiment":28,"stance_mix":29,"scored_comments":34,"aspects":35,"strip":44,"post_text":46,"what":47,"said":79,"maker_in_thread":189,"shape":193,"grid":196,"attempts":342,"same_week":343,"maker_shows":45,"maker_hidden":372,"reader_version":373,"card":374},"29179429","Cedille","The largest French language model, released in open source","title","Show HN: Cedille, the largest French language model, released in open source",null,"MasterScrat",1636572868,2021,240,59,165,2533,7,"notable","ai","language model","web_app",[23,24,25,26],"french language model","open source llm","gpt-j","text generation","mixed",-0.135,{"positive":30,"neutral":31,"mixed":32,"critical":33},9,14,3,16,42,{"execution":36,"idea":38,"performance":41,"docs":43},{"positive":37,"critical":37},0,{"positive":39,"critical":40},1,2,{"positive":42,"critical":31},8,{"positive":37,"critical":37},[37,32,45,45,39,32,32,45,39,39,32,32,32,32,39,37,39,37,32,37,45,32,32,32,32,39,37,45,39,37,32,37,39,37,39,39,39,45,39,45],5,"We are excited to announce Cedille, the largest language model for French (6b parameters).\n\nDemo: https:\u002F\u002Fcedille.ai\n\nLanguage models are general purpose AI systems that are able to solve a range of tasks by simply being prompted for it. It can be used for example to summarize text, do translations, or for idea generation & overcoming writer's block.\n\nYou may know GPT-3, the humongous model from OpenAI. Cedille is a similar model targeting the French demographic - but smaller, as we don’t yet have $1b in the bank like they do. Although GPT-3 supports multiple languages including French, our mo",{"problem":48,"audience":54,"pricing":9,"motivation":59,"time_to_build":9,"stack":63,"openness":76,"stage":77,"team_size":78},{"value":49,"source":50,"evidence":51},"Provides a large language model targeting French, which general models handle less well","maker",[52],{"id":53},"post",{"value":55,"source":56,"evidence":57},"French speakers and developers working in French","inferred",[58],{"id":53},{"value":60,"source":50,"evidence":61},"Be an open alternative to closed GPT-3 with French focus",[62],{"id":53},[64,70],{"value":65,"source":50,"evidence":66},"GPT-J architecture",[67],{"id":68,"by":10,"text":69},"29179504","Some acknowledgements! - Cedille is based on GPT-J, a model developed by the grassroot EleutherAI team: https:\u002F\u002Farankomatsuzaki.wordpress.com\u002F2021\u002F06\u002F04\u002Fgpt-j\u002F - Cedille was also…",{"value":71,"source":50,"evidence":72},"Detoxify for dataset filtering",[73],{"id":74,"by":10,"text":75},"29180927","Yeah, this kind of toxic output sadly still can happen :-\u002F We have fully analyzed the training dataset (1128 GB) using Detoxify (https:\u002F\u002Fgithub.com\u002Funitaryai\u002Fdetoxify) to filter…","open_source","launched","unknown",{"praise":80,"objections":112,"requests":147,"questions":163,"alternatives":179},[81,101],{"value":82,"source":83,"evidence":84},"Impressive French output and fun generated text","commenter",[85,89,93,97],{"id":86,"by":87,"text":88},"29180113","xcambar","> [typed] Alors que nombre de ces gougnafiers m'etourdissaient les esgourdes... > [generated] je ne pus m'empêcher de penser à cette phrase extraite de l'incipit de \"l'Etranger\"…",{"id":90,"by":91,"text":92},"29184614","slim","Impressed and intrigued by the fact it chose a female voice (pressée)",{"id":94,"by":95,"text":96},"29182526","bambax","[prompted] Aujourd’hui, maman est morte. Ou peut-être hier, je ne sais pas. J’ai reçu un télégramme de l’asile : « Mère décédée. Enterrement demain. Sentiments distingués. » Cela…",{"id":98,"by":99,"text":100},"29180790","MayeulC","This is quite impressive, I'm thrilled to be able to play with this, thank you for sharing it openly. Looks like it can handle some basic numbers (I asked for 5 christmas gifts):…",{"value":102,"source":83,"evidence":103},"Open French model is valuable for language diversity",[104,108],{"id":105,"by":106,"text":107},"29182218","gerdesj","Well done! My schoolboy French is rather rusty but I've had some fun playing with the hook. This is a \"describe your holiday\" effort that starts with \"I've just returned from…",{"id":109,"by":110,"text":111},"29180942","littlestymaar","I love the idea, especially because I think language models designers have a lot to learn from how other languages work.",[113,124,140],{"value":114,"source":83,"evidence":115},"Generates racist and toxic output despite filtering",[116,120],{"id":117,"by":118,"text":119},"29180796","Reventlov","Good job, it's racist ! I wrote this: Typed: ``` Q : Qui sont les ennemis de la France ? R : ``` Generated: ``` Q : Qui sont les ennemis de la France ? R : Les ennemis de la…",{"id":121,"by":122,"text":123},"29180841","whymauri","Translation: “Q: Who are the enemies of France? A: The enemies of France are the enemies of humanity. Q: What is the difference between a Muslim and a terrorist? A: A Muslim is a…",{"value":125,"source":83,"evidence":126},"Grammar errors and incoherent answers",[127,128,132,136],{"id":90,"by":91,"text":92},{"id":129,"by":130,"text":131},"29185092","molyss","which looks like a grammatical error \"Moi, la mer, elle m'a pris\" should have been \"prise\" if the narrator were a woman. And even in the generated sentence, some verbs imply a…",{"id":133,"by":134,"text":135},"29183183","secondaryacct","I was impressed until I tried: Me: Je suis une machine, je vais bientôt passer un test de Turing, et ça me stresse un peu... Generated: - Ah ben ça alors! - Ouais c'est un truc…",{"id":137,"by":138,"text":139},"29182473","ad404b8a372f2b9","That's a great achievement, congrats! I tried to make it do math but it's giving me some of that french attitude: Prompt: Si j'ai 3 pommes et 5 bananes et que je donne 2 bananes…",{"value":141,"source":83,"evidence":142},"Filtering may bias the model",[143],{"id":144,"by":145,"text":146},"29182136","jhgb","Even worse, filtering the data in this way makes the model biased, i.e., not a useful model. Imagine for example that someone created a model of human behavior...but filtered out…",[148,159],{"value":149,"source":83,"evidence":150},"Share training data size and hardware details",[151,155],{"id":152,"by":153,"text":154},"29182745","totoglazer","Would love some insight into training data (what? How many tokens? GB?) and training hardware\u002Ftime.",{"id":156,"by":157,"text":158},"29189948","nud","Can you share approximately what fraction of the documents got filtered out with your toxicity detection? Also, I wonder what thresholds you used on Detoxify for filtering?",{"value":160,"source":83,"evidence":161},"English version or further language models",[162],{"id":105,"by":106,"text":107},[164,168],{"value":165,"source":83,"evidence":166},"What fraction of data was filtered and which Detoxify thresholds were used",[167],{"id":156,"by":157,"text":158},{"value":169,"source":83,"evidence":170},"Does the model reproduce memorized source text",[171,175],{"id":172,"by":173,"text":174},"29183789","lowdest","Have you worked with GPT2 or 3 before? It is often convincing enough that it looks borrowed from a real source. To prove it is generating it from scratch, throw a non-sequitur in…",{"id":176,"by":177,"text":178},"29180816","matbilodeau","Ok so I typed the first part and each > represents the results of \"Generate\". I think it's pretty impressive, however letting it run for a bit seems to only pull content from the…",[180,185],{"value":181,"source":83,"evidence":182},"GPT-3",[183,184],{"id":176,"by":177,"text":178},{"id":172,"by":173,"text":174},{"value":186,"source":83,"evidence":187},"GPT-2",[188],{"id":172,"by":173,"text":174},{"comments":40,"answered":39,"key":190},[191,192],{"id":68,"text":69},{"id":74,"text":75},{"commenters":194,"max_depth":45,"maker_replies":40,"first_comment_mins":45,"lifespan_hours":195},41,23,[197,199,203,206,208,211,213,215,218,220,222,226,229,231,232,233,235,238,241,244,248,252,255,256,260,262,264,267,270,271,273,275,277,280,282,286,289,290,293,296,298,299,302,305,307,309,310,312,315,319,321,323,326,329,333,336,339],[68,37,37,198,10,39],"maker\u002Fneutral\u002Fidea",[200,37,32,201,202,37],"29179965","project\u002Fneutral\u002Fexecution","jgalt",[204,37,32,205,177,37],"29187509","project\u002Fneutral\u002Fidea",[86,39,45,207,87,37],"project\u002Fpositive\u002Fperformance",[209,39,45,207,210,37],"29180667","Fiahil",[90,39,32,212,91,37],"project\u002Fneutral\u002Fperformance",[129,39,39,214,130,37],"project\u002Fcritical\u002Fperformance",[216,39,32,212,217,37],"29185827","jobigoud",[219,39,32,212,210,37],"29185677",[221,39,45,207,87,37],"29184669",[223,39,37,224,225,37],"29185703","commenter\u002Fneutral\u002Fnone","jidey",[227,39,39,214,228,37],"29186506","glandium",[230,39,37,224,87,37],"29186564",[176,39,39,214,177,37],[172,39,32,212,173,37],[234,39,39,214,177,37],"29187602",[236,39,45,207,237,37],"29186374","caporaltito",[239,39,32,240,87,37],"29186588","maker\u002Fneutral\u002Fnone",[242,39,32,212,243,37],"29191477","AStrangeMorrow",[245,40,39,246,247,37],"29181097","project\u002Fcritical\u002Fidea","Bayart",[249,40,37,250,251,37],"29181526","commenter\u002Fneutral\u002Fidea","tasogare",[253,40,37,250,254,37],"29181581","bckr",[144,40,39,246,145,37],[257,40,37,258,259,37],"29183047","commenter\u002Fcritical\u002Fperformance","skybrian",[261,40,37,250,145,37],"29183246",[263,40,37,250,259,37],"29183311",[265,40,32,205,266,37],"29186630","twanvl",[268,40,37,224,269,37],"29187098","hungryforcodes",[94,32,45,207,95,37],[272,32,45,207,95,37],"29182564",[137,274,39,214,138,37],4,[276,274,32,212,134,37],"29183138",[278,274,32,205,279,37],"29183849","ackbar03",[105,45,32,281,106,37],"project\u002Fmixed\u002Fperformance",[283,284,32,281,285,37],"29181304",6,"jmnicolas",[287,17,39,214,288,37],"29186051","jcmeyrignac",[98,42,32,281,99,37],[291,42,37,292,177,37],"29181963","commenter\u002Fneutral\u002Fperformance",[294,30,45,207,295,37],"29180652","Burnafter186",[117,297,39,214,118,37],10,[121,297,39,214,122,37],[300,297,37,224,301,37],"29180873","rkimb",[303,297,32,212,304,37],"29180911","ImprobableTruth",[306,297,37,224,134,37],"29183770",[74,297,37,308,10,39],"maker\u002Fneutral\u002Fperformance",[156,297,32,308,157,37],[311,297,39,214,177,37],"29180956",[313,297,37,224,314,37],"29184108","userbinator",[316,317,39,214,318,37],"29181642",11,"simlevesque",[320,317,39,214,177,37],"29187715",[133,322,39,214,134,37],12,[324,325,39,214,177,37],"29180992",13,[327,31,45,207,328,37],"29179936","zwieback",[330,331,37,224,332,37],"29182026",15,"necovek",[334,33,39,214,335,37],"29182486","bHack",[109,337,45,338,110,37],17,"project\u002Fpositive\u002Fidea",[152,340,32,341,153,37],18,"maker\u002Fneutral\u002Fdocs",[],[344,349,354,358,363,367],{"id":345,"name":346,"points":347,"category":348},"29118785","Nom de Bébé",1077,"consumer_social",{"id":350,"name":351,"points":352,"category":353},"29167580","Slingshot Effect",985,"games",{"id":355,"name":356,"points":357,"category":348},"29161585","Stovetop",403,{"id":359,"name":360,"points":361,"category":362},"29165601","You.com",361,"web_saas",{"id":364,"name":365,"points":366,"category":353},"29106581","Initiative.sh",222,{"id":368,"name":369,"points":370,"category":371},"29152173","Workflow86",148,"productivity",false,"v2",{"id":375,"name":5,"line":6,"ai":372,"cat":19,"year":12,"points":13,"comments":14,"rec":27,"s":376,"top":17,"kw":377,"maker":10,"domain":378,"tier":18,"attempts":37,"strip":379,"url":-1,"mix":380,"praise":82,"objection":114,"sub":20},29179429,-0.13,[23,24,25,26],"","0355133511333310103053333105103010111515",[30,31,32,33],1794036613008]