Build your own Tokenizer

Why do language models miscount the r's in "strawberry"? GPT-2 never sees the letters. Its tokenizer hands it three pieces: st, raw, and berry. A tokenizer is the program that splits text into pieces like these before a model reads it. In this challenge, you'll build one that splits text exactly like GPT-2's.

Along the way, you'll learn about byte-pair encoding, UTF-8, and more.