@Deepti_Prasad perhaps you can help me on this as I know you are one of the few NLP mentors.
I get the point they make that the weights are ‘shared’, but does this happen when we are still in the dual LSTM network stage, or are they saying we do this once we go to the cosine similarity juncture ?
I’m not quite at the assignment yet but I am finding this really confusing… I’d imagine you’d share weights at the LSTM stage, but have no idea how you’d do that.
What are they talking about ?
We had this question here before, have a look in this post:
As far as I remember its one base network shared with the inputs!
Thank you for letting me know/reminding me.
Perhaps I just wasn’t sure what question I should ask.
Just as I work through this… I (think ?) there is a mistake in the notebook:
Or in the args list it cites d_model as a default of 128… But I am presuming they mean ‘d_feature’… Or otherwise I’m not sure where this value is coming from…
I think you are right here! Let me try and leave a note on the repo!
