Jd Kim

Will it be solved by using a seed?

I think it's weird, too. I think there is something wrong with my code.

It's late, but I'll check.

저도 처음에 헷갈렸던 부분인데 임의로 정한 것 같습니다.

Please check the CUDA version.

It seems to be correct. I will try again. Thank you.

Test accuracy is 94.25. Unfortunately, the performance has decreased. What is the reason?

https://github.com/google-research/google-research/issues/534#issuecomment-768527718 Author said that when using `hard pseudo labels`, `96.06%` performance came out.

> I had read that comment previously but wasn't sure if he meant hard labels, or the repo as is, I think you're right though. > > I looked through...

I agree. I thought that the sign changes in the process of calculating `moving_dot_product`. But I confirmed that it was not. I will test again. Thanks!