[{"data":1,"prerenderedAt":434},["ShallowReactive",2],{"project-44160073":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":42,"post_text":44,"what":45,"said":85,"maker_in_thread":189,"shape":200,"grid":204,"attempts":396,"same_week":397,"maker_shows":287,"maker_hidden":426,"reader_version":427,"card":428},"44160073","Penny-1.7B","Penny-1.7B Irish Penny Journal style transfer","title","Show HN: Penny-1.7B Irish Penny Journal style transfer","https:\u002F\u002Fhuggingface.co\u002Fdleemiller\u002FPenny-1.7B","huggingface.co","deepsquirrelnet",1748879348,2025,149,76,502,3837,14,"notable","ai","style transfer language model","other",[24,25,26,27,28],"style transfer","grpo","reinforcement learning","language model","irish penny journal","curious",0.016,{"positive":32,"neutral":33,"mixed":34,"critical":35},4,9,0,5,18,{"performance":38,"idea":39,"execution":41},{"positive":34,"critical":34},{"positive":40,"critical":40},1,{"positive":34,"critical":32},[34,34,34,34,34,34,34,34,34,34,34,34,34,43,40,35,43,43,35,43,43,35,35,34,34,40,40,34,34,34,34,43,34,34,34,40,34,34,40,43],3,"Yesterday, in the bygone hour of the weekend, I undertook a most singular and fascinating endeavor, wherein I delved deep into the recesses of my mind, and, with a fervent zeal, breathed life into a most remarkable creation. I embarked upon the quest, with the singular object of fashioning an artificial construct, one imbued with the verdant essence of the Irish Penny Journal, an ancient and venerable tome that holds within its pages the whispered tales of a bygone era.\n\nIn my haste, I set forth to construct a dataset, a repository of those fleeting moments, these ephemeral sentences, which sp",{"problem":46,"audience":52,"pricing":52,"motivation":53,"time_to_build":59,"stack":63,"openness":82,"stage":83,"team_size":84},{"value":47,"source":48,"evidence":49},"Makes a language model write in the prose style of the 19th-century Irish Penny Journal","maker",[50],{"id":51},"post",null,{"value":54,"source":48,"evidence":55},"Fun experiment in style transfer with GRPO, and potential for game dialogue voices",[56],{"id":57,"by":11,"text":58},"44161347","Good question - my best assessment is just the text classifier. IE was the LLM able to “trick” the classifier into believing the text came from the IPJ? And it came quite a long…",{"value":60,"source":48,"evidence":61},"Less than a day on a single A6000",[62],{"id":51},[64,70,74],{"value":65,"source":48,"evidence":66},"MiniLM2 classifier",[67],{"id":68,"by":11,"text":69},"44160932","My dataset is here: https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fdleemiller\u002Firish_penny_journal It’s fairly simple — I essentially just split the original text into chunks and then used some…",{"value":71,"source":48,"evidence":72},"GRPO reinforcement learning",[73],{"id":51},{"value":75,"source":76,"evidence":77},"TRL GRPO trainer","commenter",[78],{"id":79,"by":80,"text":81},"44161320","veggieroll","Thanks for the explanation! I'm learning and and think this would be a good next project for me to try, especially since I have a real world use case in mind with a similar…","open_source","launched","unknown",{"praise":86,"objections":106,"requests":139,"questions":161,"alternatives":173},[87,102],{"value":88,"source":76,"evidence":89},"Whimsical and fun project",[90,94,98],{"id":91,"by":92,"text":93},"44173402","sterlind","what a wonderful work of whimsy! well wrought. I'd love to have a library of these, so I could pipe text into `penny`, `brainrot`, `pony`, `newspeak`, `corporate`, `scp`, `trek`…",{"id":95,"by":96,"text":97},"44161189","fitsumbelay","this is awesome",{"id":99,"by":100,"text":101},"44162204","npunt","Love it. Immediately reminded of the text filters back in the day like the pirate one that would drop letters and replace with apostrophes and change certain passages into \"arr\"…",{"value":103,"source":76,"evidence":104},"Idea of a library of text style filters",[105],{"id":91,"by":92,"text":93},[107,118,125,132],{"value":108,"source":76,"evidence":109},"Output uses modern LLM-style words like delve",[110,114],{"id":111,"by":112,"text":113},"44160297","ekianjo","Nice work ! It still manage to use the word 'delve' in the first sentence, which is a giveaway that it's written by a LLM.",{"id":115,"by":116,"text":117},"44160353","dragonwriter","Its considered an LLM tell because its a term that is rarely used by the median modern casual writer; its not at all uncommon even in current literature [0], and its even more…",{"value":119,"source":76,"evidence":120},"Example contains a historical error about Napoleon",[121],{"id":122,"by":123,"text":124},"44160557","_1","Kinda of strange to pick an example that is just wrong. It's supposed to be written from 1840 and says Paris is the seat of Napoleon almost 20 years after he died.",{"value":126,"source":76,"evidence":127},"Output is more complex than the source journals",[128],{"id":129,"by":130,"text":131},"44161394","KaiserPro","I'm not sure if you've tried this already, but removing the translate step might give you a more authentic output. In the journals that I saw, the language was much more simple…",{"value":133,"source":76,"evidence":134},"Repeated adjectives in adjacent sentences",[135],{"id":136,"by":137,"text":138},"44160445","rndmio","The repetition of adjectives in adjacent sentences is much more of a tell than using the word delve, imo.",[140,151,155],{"value":141,"source":76,"evidence":142},"Publish training code or notebook",[143,144,148],{"id":91,"by":92,"text":93},{"id":145,"by":146,"text":147},"44160717","kamranjon","This is really cool! Do you have any of the pipeline code available that you used for training? I am curious about how you created the reward model. I love little projects like…",{"id":149,"by":80,"text":150},"44160428","Have you written anywhere in detail on how you gathered your dataset and trained the finetune? I have a few use cases that are like this, but I'm not sure where to start.",{"value":152,"source":76,"evidence":153},"Remove the translate step for more authentic output",[154],{"id":129,"by":130,"text":131},{"value":156,"source":48,"evidence":157},"Add a penalty for LLM giveaway words",[158],{"id":159,"by":11,"text":160},"44160307","Haha! You’re right. Maybe I’ll add a penalty for that and some other giveaway words and make a revision.",[162,169],{"value":163,"source":76,"evidence":164},"Why no supervised fine-tuning was used",[165],{"id":166,"by":167,"text":168},"44161635","throwaway314155","You mention no supervised finetuning. May I ask why? I'm curious if you could get similar\u002Fbetter\u002Fworse results by just finetuning the LLM on your dataset rather than generating…",{"value":170,"source":76,"evidence":171},"How the reward model was created",[172],{"id":145,"by":146,"text":147},[174,181],{"value":175,"source":76,"evidence":176},"Antislop sampler for suppressing overused words",[177],{"id":178,"by":179,"text":180},"44160833","Der_Einzige","This is easily fixed with the antislop sampler. https:\u002F\u002Fgithub.com\u002Fsam-paech\u002Fantislop-sampler",{"value":182,"source":76,"evidence":183},"Pirate and jive text filters",[184,185],{"id":99,"by":100,"text":101},{"id":186,"by":187,"text":188},"44160825","latchkey","Reminds me of this: https:\u002F\u002Fwww.unix.com\u002Fman_page\u002Fdebian\u002F6\u002Fjive\u002F",{"comments":190,"answered":191,"key":192},8,6,[193,194,197],{"id":57,"text":58},{"id":195,"text":196},"44176873","I put my scripts up in github. It’s a bit scrapped together at the moment — but fine as a reference. https:\u002F\u002Fgithub.com\u002Fdleemiller\u002FPennyLM",{"id":198,"text":199},"44161888","In some other experiments, I’ve noticed that SFT can be very rigid and hard to generalize from, and later found this paper: https:\u002F\u002Farxiv.org\u002Fabs\u002F2501.17161 Also I wanted to…",{"commenters":201,"max_depth":191,"maker_replies":190,"first_comment_mins":202,"lifespan_hours":203},49,20,36,[205,209,213,215,219,221,224,227,230,233,236,238,241,243,245,248,251,253,257,259,262,265,267,270,273,276,280,282,286,289,291,293,295,296,297,299,302,303,304,305,307,310,312,313,317,320,323,326,329,332,334,336,337,340,342,344,347,349,352,355,358,361,364,367,369,371,373,374,377,379,381,385,387,390,392,393],[206,34,34,207,208,34],"44160867","other\u002Fneutral\u002Fidea","joshstrange",[210,34,34,211,212,34],"44161055","commenter\u002Fneutral\u002Fnone","jsheard",[214,34,34,211,208,34],"44161086",[216,34,34,217,218,34],"44161285","commenter\u002Fneutral\u002Fidea","acdha",[220,34,34,217,208,34],"44161331",[222,34,34,211,223,34],"44161105","shinryuu",[225,34,34,211,226,34],"44161923","Mtinie",[228,34,34,211,229,34],"44162316","inkcapmushroom",[231,34,34,211,232,34],"44162600","visarga",[234,34,34,217,235,34],"44162846","JohnBooty",[237,34,34,217,80,34],"44161464",[239,34,34,217,240,34],"44162167","BizarroLand",[242,34,34,217,235,34],"44162754",[244,34,34,217,240,34],"44175405",[246,34,34,217,247,34],"44161719","wongarsu",[249,34,34,211,250,34],"44162673","bee_rider",[252,34,34,217,235,34],"44162738",[254,34,34,255,256,34],"44163488","commenter\u002Fneutral\u002Fdesign","inhumantsar",[258,34,34,255,235,34],"44163566",[260,34,34,255,261,34],"44165115","MangoToupe",[263,34,34,207,264,34],"44161441","speps",[266,34,34,207,235,34],"44162695",[268,34,34,207,269,34],"44162862","killerstorm",[271,34,34,207,272,34],"44163270","qingcharles",[274,34,34,207,275,34],"44165456","vunderba",[277,40,43,278,279,34],"44160610","maker\u002Fneutral\u002Fperformance","sjkoelle",[57,40,34,281,11,40],"maker\u002Fneutral\u002Fidea",[283,40,40,284,285,34],"44162367","project\u002Fcritical\u002Fidea","dwringer",[99,287,35,288,100,34],2,"project\u002Fpositive\u002Fnone",[186,43,43,290,187,34],"project\u002Fneutral\u002Fnone",[145,32,43,292,146,34],"maker\u002Fneutral\u002Fnone",[195,32,34,294,11,40],"maker\u002Fneutral\u002Fdocs",[166,35,43,281,167,34],[198,35,34,281,11,40],[91,191,35,298,92,34],"project\u002Fpositive\u002Fidea",[129,300,43,301,130,34],7,"project\u002Fneutral\u002Fexecution",[149,190,43,292,80,34],[68,190,34,292,11,40],[79,190,43,292,80,34],[306,190,34,292,11,40],"44161444",[308,190,35,309,80,34],"44161820","maker\u002Fpositive\u002Fnone",[311,190,34,294,11,40],"44176886",[95,33,35,288,96,34],[314,315,34,316,250,34],"44160544",10,"other\u002Fneutral\u002Fnone",[318,315,34,316,319,34],"44160749","blululu",[321,315,34,316,322,34],"44162396","projektfu",[324,315,34,316,325,34],"44161962","w10-1",[122,327,40,328,123,34],11,"project\u002Fcritical\u002Fexecution",[330,327,34,211,331,34],"44160960","Philpax",[111,333,40,328,112,34],12,[159,333,34,335,11,40],"maker\u002Fneutral\u002Fexecution",[115,333,34,211,116,34],[338,333,34,211,339,34],"44160541","observationist",[341,333,34,316,250,34],"44162863",[343,333,34,207,339,34],"44163788",[345,333,34,316,346,34],"44160706","freedomben",[348,333,34,211,250,34],"44162919",[350,333,34,316,351,34],"44160708","tiahura",[353,333,43,301,354,34],"44160715","arscan",[356,333,34,211,357,34],"44160441","secondcoming",[359,333,34,211,360,34],"44160528","esafak",[362,333,34,316,363,34],"44160764","numpad0",[365,333,34,316,366,34],"44161056","Suppafly",[368,333,34,211,366,34],"44160983",[370,333,34,316,212,34],"44161195",[372,333,34,316,366,34],"44162748",[136,333,40,328,137,34],[375,333,34,211,376,34],"44160623","rsynnott",[378,333,34,335,11,40],"44161211",[380,333,34,316,325,34],"44161955",[382,333,34,383,384,34],"44160700","other\u002Fcritical\u002Fnone","mbStavola",[386,333,34,316,179,34],"44160892",[388,333,34,316,389,34],"44162810","jerjerjer",[391,333,40,328,112,34],"44177047",[178,333,43,301,179,34],[394,333,34,316,395,34],"44161312","officeplant",[],[398,403,408,413,417,421],{"id":399,"name":400,"points":401,"category":402},"44157177","Kan.bn",509,"productivity",{"id":404,"name":405,"points":406,"category":407},"44126612","Command Line Handbook",454,"learning_science",{"id":409,"name":410,"points":411,"category":412},"44127653","Onlook",408,"dev_tools",{"id":414,"name":415,"points":416,"category":412},"44123098","Typed-FFmpeg",342,{"id":418,"name":419,"points":420,"category":412},"44159758","Vanta",265,{"id":422,"name":423,"points":424,"category":425},"44147803","Patio",255,"web_saas",false,"v2",{"id":429,"name":5,"line":6,"ai":426,"cat":20,"year":13,"points":14,"comments":15,"rec":29,"s":430,"top":18,"kw":431,"maker":11,"domain":10,"tier":19,"attempts":34,"strip":432,"url":9,"mix":433,"praise":88,"objection":108,"sub":21},44160073,0.02,[24,25,26,27,28],"0000000000000315335335500110000300010013",[32,33,34,35],1793991122323]