[{"data":1,"prerenderedAt":121},["ShallowReactive",2],{"project-41050468":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":27,"sentiment":28,"stance_mix":29,"scored_comments":31,"aspects":32,"strip":34,"post_text":36,"what":37,"said":50,"maker_in_thread":72,"shape":77,"grid":80,"attempts":85,"same_week":86,"maker_shows":113,"maker_hidden":114,"reader_version":115,"card":116},"41050468","Unsloth","Finetune Llama-3.1 2x faster in a Colab","title","Show HN: Finetune Llama-3.1 2x faster in a Colab","https:\u002F\u002Fcolab.research.google.com\u002Fdrive\u002F1Ys44kVvmeZtnICzWz0xgpRnrIOjZAuxp?usp=sharing","colab.research.google.com","danielhanchen",1721766620,2024,16,2,2092,3515,60,"shown","ai","llm finetuning","library",[21,24,25,26],"llama 3.1","vram reduction","triton kernels","quiet",null,{"positive":30,"neutral":31,"mixed":30,"critical":30},0,1,{"performance":33},{"positive":30,"critical":30},[35,30],3,"Just added Llama-3.1 support! Unsloth https:\u002F\u002Fgithub.com\u002Funslothai\u002Funsloth makes finetuning Llama, Mistral, Gemma & Phi 2x faster, and use 50 to 70% less VRAM with no accuracy degradation.\n\nThere's a custom backprop engine which reduces actual FLOPs, and all kernels are written in OpenAI's Triton language to reduce data movement.\n\nAlso have an 2x faster inference only notebook in a free Colab as well! https:\u002F\u002Fcolab.research.google.com\u002Fdrive\u002F1T-YBVfnphoVc8E2E854qF3jdia2Ll2W2?usp=sharing",{"problem":38,"audience":28,"pricing":28,"motivation":28,"time_to_build":28,"stack":44,"openness":49,"stage":49,"team_size":49},{"value":39,"source":40,"evidence":41},"Finetuning LLMs is slow and uses a lot of VRAM","maker",[42],{"id":43},"post",[45],{"value":46,"source":40,"evidence":47},"OpenAI Triton",[48],{"id":43},"unknown",{"praise":51,"objections":52,"requests":61,"questions":66,"alternatives":71},[],[53],{"value":54,"source":55,"evidence":56},"Lacks multi-GPU support, a common reason people don't adopt it","commenter",[57],{"id":58,"by":59,"text":60},"41077308","smcleod","Does it support multiple GPUs yet? That's been the reason most folks I know don't end up using Unsloth.",[62],{"value":63,"source":55,"evidence":64},"Multi-GPU support",[65],{"id":58,"by":59,"text":60},[67],{"value":68,"source":55,"evidence":69},"Does it support multiple GPUs yet?",[70],{"id":58,"by":59,"text":60},[],{"comments":31,"answered":31,"key":73},[74],{"id":75,"text":76},"41083440","We're currently doing a beta with some of our community members for multi GPU!",{"commenters":15,"max_depth":15,"maker_replies":31,"first_comment_mins":78,"lifespan_hours":79},3689,76,[81,83],[58,30,35,82,59,30],"project\u002Fneutral\u002Fperformance",[75,30,30,84,11,31],"maker\u002Fneutral\u002Fnone",[],[87,92,97,101,105,109],{"id":88,"name":89,"points":90,"category":91},"41007403","Sendune",429,"web_saas",{"id":93,"name":94,"points":95,"category":96},"41045834","Briefer",265,"data",{"id":98,"name":99,"points":100,"category":91},"41031975","billabear.com",260,{"id":102,"name":103,"points":104,"category":20},"41048194","Zerox",246,{"id":106,"name":107,"points":108,"category":91},"41048196","Overlap",229,{"id":110,"name":111,"points":112,"category":20},"40998497","Mutable.ai Auto Wiki",162,4,false,"v2",{"id":117,"name":5,"line":6,"ai":114,"cat":20,"year":13,"points":14,"comments":15,"rec":27,"s":28,"top":18,"kw":118,"maker":11,"domain":10,"tier":19,"attempts":30,"strip":119,"url":9,"mix":120,"praise":-1,"objection":54,"sub":21},41050468,[21,24,25,26],"30",[30,31,30,30],1794036658271]