Exercise 06 - UNQ_C9: pretrained model issue with next symbol

I think there is an error with the model that I have in the system. I am passing all the unit tests before I arrive to cell UNQ_C9. I am loading the pretrained model you’ve prepared in the assignment.

The issue is the next token prediction is not working as expected. I know my code must be correct, you can see the lab ID: mpgnwzmklamn

The cell Exercise 06 is for sure loading the model that you’ve pretrained. It is not passing the unit test and it keeps selecting an incorrect token. It keeps outputting "'" instead of 'The'. Please see the workspace. Can you double check if the the pretrained model.pkl.gz is corrupted or something?

Thanks!

Hi @cmosguy

With problems like these it is good to refresh your workspace.

You can delete the model (as in step 5) you think is at fault and get the new one - the one that most students don’t have problems. If the problem persist it is highly likely that the problem is not in the model.

Cheers

Hello @arvyzukai,

Thanks for the help, it did turn out in fact that I had a swapaxis call in the the attention network using the wrong order and disguising a real issue. Therefore, the weights were not getting loaded in correctly.

Take care!
Adam