Can we use model.eval() and model.train() alternatively in Pytorch?

1

Hi! I’m trying to get two representation of the same input in model.eval() and model.train(), respectively. I write the following codes:

for id, batch in enumerate(loader):
    model.train()
    rep1 = model(batch.text)
    model.eval()
    rep2 = model(bacth.text)

However, it seems it didn’t work, so is there any ways to do that?

2

Could you describe what exactly is not working? Do you see any errors using it?

3

Thanks for your attention. I just found I put model.eval() in the forward process, and it works now, I’ll be more careful next time.