[{"data":1,"prerenderedAt":34},["ShallowReactive",2],{"maker-matthewolfe":3},[4],{"id":5,"name":6,"line":7,"ai":8,"cat":9,"year":10,"points":11,"comments":12,"rec":13,"s":14,"top":15,"kw":16,"maker":22,"domain":23,"tier":24,"attempts":25,"strip":26,"url":27,"mix":28,"praise":31,"objection":32,"sub":33},44422480,"TokenDagger","A tokenizer faster than OpenAI's Tiktoken",false,"ai",2025,281,73,"curious",0.1,6,[17,18,19,20,21],"tokenizer","bpe","tiktoken","c++","llm inference","matthewolfe","github.com","legend",0,"5000000000000035303015053100033135351331","https:\u002F\u002Fgithub.com\u002FM4THYOU\u002FTokenDagger",[15,29,25,30],16,5,"Drop-in replacement with substantial speedup is appealing.","Tokenization is rarely the bottleneck in LLM inference or training.","tokenizer library",1794036814267]