For evaluation

[JYKai]

Thanks for your code and thesis.

I really enjoy looking at the thesis and the code.

I have a question about learning and evaluation methods.

When training the model, the total iteration is set to 300000, and when evaluating the model, do you use the model weight file at the time of the finally obtained 300000 iteration?

Or, do you use the model weight file that performed best during training among the weight files in the middle, rather than the last obtained model weight file?

Thank you for read

[hkchengrex]

We use the final model.