search

rinongal / textual_inversion

MIT License
2.87k stars 278 forks source link

Many questions about the code inside the main.py #133

Open JosephKKim opened 1 year ago

JosephKKim commented 1 year ago

Hi thanks for releasing the nice work, I've read the paper and now I'm trying to understand the implementation.

Q1. Does anyone know why there are two lines of code that are exactly the same in line #745 and #747 in main.py? Is it related with the comments below? taking care of proper multiprocessing?

Q2. What is the difference between the data.prepare_data() (at the line #751)and data.setup() (at the line #752). I think data.setup() is enough.. Am I wrong?

I wish that someone who knows the answer (even if you are not author) replies this issue. I'm trying to append this issue as new questions come up. Thanks!

Joseph

rinongal commented 1 year ago

Q1: That's just a bug. You can safely remove one of the two.

Q2: That's actually part of the original LDM code so I'm not entirely sure what their motivations were for calling both. Reading through the Pytorch Lightning documentation seems to suggest that setup() should be enough. You're welcome to try.

JosephKKim commented 1 year ago

Thanks! I'll take a look at the Pytorch Lightning documentation. I have one more question, is there any reason for using pytorch lightning? I'm asking this because the only module which you referred using pytorch lightning is taming transformer.