Add AlbertClassifier

[kanpuriyanawab]

This PR adds AlbertClassifier class to keras_nlp, letting the user instantiate the classifier with AlbertaBackbone, out-of-the-box.

Closes #660 !

[kanpuriyanawab]

@jbischof This PR is ready for review.

Please consider following points while reviewing

1. Albert Presets are not implemented yet, so in classproperty from_preset() of AlbertClassifier, I've raised a NotImplementedError(), please let me know your views on it!
2. The Tests pass locally.

[kanpuriyanawab]

@jbischof @mattdangerw This PR is ready for review as well !

[kanpuriyanawab]

CI is green @jbischof

[mattdangerw]

This looks great! Left a few comments, mostly because we have made some changes to base classes here.

[kanpuriyanawab]

@mattdangerw @jbischof
Thanks for the comments ! Task has simplified the usage of kerasNLP even more !

Couple of notes about following commit.

1. Albert doesn't have classifier presets yet, thus I had to remove usage of classifier_presets from albert_classifier.py
2. when I am running tests for albert_presets_test.py , all are getting skipped on local

[kanpuriyanawab]

@jbischof This PR is ready for review.

[mattdangerw]

This looks great! Just some small issues. There are some docs changes we will need, but I think we can leave that to a follow up.

You still have some formatting issues, make sure to run our formatting script.

[mattdangerw]

We will need to add some more documentation for overall usages here, see BERT as an example -> https://github.com/keras-team/keras-nlp/blob/master/keras_nlp/models/bert/bert_classifier.py#L55-L161

But I'm also OK with this being a follow up, we have some work to do on our class level docstrings anyway.

[mattdangerw]

Let's just do this one as a follow up, I've opened #688. Other comments still apply!

[kanpuriyanawab]

I made current docstrings alone with Bert classifier, once the pr merges and we have clear cut idea on how the docstrings for all models are to be restructured, I can put a follow up PR.

[kanpuriyanawab]

Still WIP

[kanpuriyanawab]

merging with master messed up albert_presets.py changes. fixed it. Ready for review @mattdangerw @jbischof

I'm not sure why nightly is fialing in CI

[kanpuriyanawab]

CI is almost green.
Not sure why nightly is failing @mattdangerw @jbischof

[mattdangerw]

Thanks so much @shivance ! Not sure what is going on with nightly, but I suspect it is unrelated to your PR.

Nightly testing is important for us to have, but it will definitely surface errors in the upstream tf release as often as our own. I will take a closer look today. Thanks!

[kanpuriyanawab]

@mattdangerw How does nightly differ from dev and stable release?

[mattdangerw]

@shivance only difference is that nightly will test against the tf-nightly and tensorflow-text-nightly packages. Which are nightly "unstable" releases of our upstream dependencies. Often what will happen is the tf-nightly package will break our testing because of a bug in a recent tf change.

[kanpuriyanawab]

Got it ! Thanks ! @mattdangerw
By the way, I'm really interested in GSoC this year at KerasNLP ( I love working here and luckily kerasnlp participate in gsoc last year under tensorflow). Will it participate this year too?

What are potential projects ? (My guess is, ideally it should align with the current roadmap of KerasNLP)

[mattdangerw]

By the way, I'm really interested in GSoC this year at KerasNLP

I'm pretty sure yes! But we are just figuring out all the details right now actually, will follow up when we have more.

Also, confirmed that this nightly test is unrelated to your change. Will pull this in tomorrow, and thank you!