How text becomes tokens: word splitting, byte-pair encoding, a vocabulary, and encode/decode that round-trips.