Tokenizer Training

Build a subword vocabulary by iteratively merging the most frequent adjacent token pair.