Train tokenizer on en_zh_linear_1 (HF dataset format)

initial commit