get_set removal

[Riccardo231]

Starting to work on the code; still a draft, maybe have a look so to tell me if somebody is visibly wrong.

(I thought i was working on my fork but instead i created a branch on the main project, i hope this is not a problem. Thank you)

[milancurcic]

Looks good, thank you! It's exactly what I had in mind. And the same approach with set_params.

[Riccardo231]

Hello, I've also removed set_params()

What I wanted to discuss with you is that in the layer embedding_layer there are only weights and not biases, so I cannot properly use the get_params_ptr, I think you already noticed because you didn't implement it.

What should I do? I think that adding biases to the layer wouldn't break anything.

Plus, in self_attention_layer I cannot find the submodule and the get_set procedures. Where are they?

When I'll be done with these two layers I can finally remove every instance of the getset procedures. Thanks, I'll wait for an answer @milancurcic

[jvdp1]

LGTM. Thank you @Riccardo231

[Riccardo231]

Thanks. Could you provide a feedback on the question above?

[milancurcic]

Thank you, @Riccardo231 for the work on this, and sorry for the delay. Thanks @jvdp1 for the review.

I think we can already merge this as is because the functionality remains the same and a decent chunk of the code is removed, so it's already a great net positive.

For the embedding and attention layer, we can just skip those for now and address later, with @OneAdder's guidance.

So, I'd say, as soon as you're ready to mark this as ready for review, you can also merge it as is and we take it from there.