GigaToken: ~1000x faster Language model tokenization