self.parameters() access in pytorch lightning

[arjunsinghrathore]

I have noticed that if i call self.parameters() like this in the init of the pl.LightningModule class, then I'm able to access it :

for param in self.parameters(): print('param : ',param.weight)

but if I call it similarly either in the training step or the optimizer step then self.parameters() does not return any param. Is there something special I need to do to rectify this?

[rohitgr7]

I don't think paramhas an attribute weight. Can you share a small reproducible example?

[arjunsinghrathore]

Yep, I realized that. I needed to do param.data

But just one doubt I still have is at what places can we access self.parameters(), as in I noticed that in training_step it does not return any params whereas in training_step_end for example I am able to get all the parameters of my model through self.parameters()

[rohitgr7]

it should be accessible everywhere.