Hi author,
Thank you for the wonderful work. It's really amazing !
When reproduce and reading the code, I observe that you use EMA for tokenizer training but never checkpoint or use it ? Moreover, in the paper, you do not claim that you use EMA. Is there any miss-information here ?
Thank you
Hi author,
Thank you for the wonderful work. It's really amazing !
When reproduce and reading the code, I observe that you use EMA for tokenizer training but never checkpoint or use it ? Moreover, in the paper, you do not claim that you use EMA. Is there any miss-information here ?
Thank you