use proper tokenizer.json and change to batch_size=256

remove old tokenizer.json. it's not trained properly

add pre-generated config.json and trained Thai tokenizer.json