Deleted the tokenizer.model and re-quantized with vocab-type bpe