Correct way to load pretrained model

I have been following this tutorial,
and I just want to make sure that my understanding correct.

So in the train command, I set a directory where output or checkpoint during the training will be saved. Let’s suppose I made a directory called ‘output’, and I set the OUTPUT_DIR environmental variable to be the directory that I created which is ‘output’

Then when I run the inference,
I set the model_id to be the ‘output’ directory.

is this the right way to load a pretrained model?
cuz, no matter how long I train the model,
it seems like there is no significant improvements on the outputs to me.

Or should I get rid of unet directory in the ‘output’ directory, and then copy the unet directory from the each checkpoint?