Mlm mask generator docstring adding example

[abuelnasr0]

for the contribution welcome issue #120 I added a simple example to show the best practice of using MLMMaskGenerator to train a language model

• Prepare data, we can generate random strings or load some small datasets from TFDS (tensorflow datasets).
• Instantiate a tokenizer, and call tokenize on the data.
• Apply MLMMaskGenerator on the tokenized data.
• Feed the masked data to a dummy MLM model by calling model.fit().

link to google colab with the example:
https://colab.research.google.com/drive/19GmGy4nCe2-AgRsHAqdVIMmjxR_WxSkB#scrollTo=YdPZIknsopS_

waiting for feedback.
thank you.

[chenmoneygithub]

Thanks for the PR!

[mattdangerw]

Some high level comments...

I think showing a more fleshed out usage is OK to do here, but I would scale it down significantly from what we are showing here.

We do not want to show keras_nlp.models usage here in our lower level layers. And we probably do not want to show fit() either. We should generally try to keep our usages fairly local to the symbols in question, so we avoid a ton of cross cutting dependencies when we update a symbol usage.

I might instead just show something that looks closer to a real world example, but is still just focused on the layer itself.

pad_id, cls_id, sep_id, mask_id = 0, 1, 2, 3
batch = [
    [cls_id,     4,     5,      6, sep_id,    7,    8, sep_id, pad_id, pad_id],
    [cls_id,     4,     5, sep_id,      6,    7,    8,      9, sep_id, pad_id],
]
# the rest of the MaskedLMMaskGenerator and invocation

[abuelnasr0]

like last push?

[mattdangerw]

Ah yes perhaps, sorry about that!

This may be me needing to sync up with @chenmoneygithub more. I do not think we want the examples here to depend on a totally separate part of our API, I will talk with @chenmoneygithub about it.

[mattdangerw]

Just talked with @chenmoneygithub, apologies for all the confusion here. We are scaling up our contributors, so our wires may get crossed a few times :)

But this latest version looks good to me! Thanks very much!

[abuelnasr0]

No problem. Writing docstrings and changes help me to get familiar with the library more. Thanks for the feedback.
One thing: should I run that /gcbrun or a mentor should run it ?

[mattdangerw]

@abuelnasr0 good question. We are playing with different options here, but /gcbrun would need to be added to project by a maintainer. Unclear if we are going to require it long term, but no action should be needed from you!

[mattdangerw]

This LGTM!

[chenmoneygithub]

Thank you!