Question about custom training outputs

[severin-div]

Hi,

First, thanks for your outstanding work and for sharing it!

I trained a model on my custom dataset. When looking on the outputs of the training, I see different .pth files: Some are relatively small (a few MBs) and some are quite large (more than a GB). What is the difference between those files? I noticed that I cannot do any inferences with the relatively small files in the "eval" folder but only with the larger files.

Thank you!

[lyuwenyu]

---

some are quite large (more than a GB).

including model.state_dict(), optim.state_dict(), ema.state_dict(), ..., you can load the checkpoint and check the keys.

---

I noticed that I cannot do any inferences with the relatively small files in the "eval" folder but only with the larger files.

these files are the output of the eval process, you can ignore them.