Update docs for upcoming 0.2.0 release #158
Conversation
|
Also, note that this is currently cross linking the keras.io docs, so we should land this after we land the new keras.io section (or I can just ditch those links and submit them later). |
mattdangerw
force-pushed
the
update-docs
branch
2 times, most recently
from
1864b80 to
c87453b
Compare
| '`pip install rouge-score`.') | ||
| ``` | ||
|
|
||
| ## Keep computation inside TensorFlow graph |
There was a problem hiding this comment.
Maybe worth pointing out the difference between "TF graph" (tf.function) and "XLA compilable" (tf.function(jit_compile=True))
There was a problem hiding this comment.
Good point.
Though I fuzzy on what we should suggest here. I'm going to say we want both, but for preprocessing and postprocessing, XLA is not a requirement. I don't think you can use the tf text custom ops with XLA.
| self.units = units | ||
|
|
||
| def build(self, input_shape): | ||
| super().build(input_shape) |
There was a problem hiding this comment.
Is the super call necessary?
There was a problem hiding this comment.
I have been unsure about this for a while! Core keras has two styles...
super().build(input_shape) or self.built = True
You should have one of those lines. Is either more canonical?
| self.w = self.add_weight(shape=(input_shape[-1], self.units)) | ||
| self.b = self.add_weight(shape=(self.units,)) | ||
|
|
||
| def call(self, inputs): |
There was a problem hiding this comment.
Might be worth documenting mask and possibly training
There was a problem hiding this comment.
This is the style guide, so will keep discussion of how to actually handle masking out of here. But added a not that masking behavior of the layer should be documented in the class level docstring.
| reduce the friction of training a tokenizer, we should offer a standardized | ||
| experience for training new vocabularies. | ||
|
|
||
| ### Scalable examples of popular model architectures using KerasNLP |
There was a problem hiding this comment.
"scalable" -> should we mention DTensor support somewhere? This will be important for model parallelism.
There was a problem hiding this comment.
Sounds good. Added a line.
ROADMAP.md
Outdated
| to have KerasNLP components available in a pre-trained model offering of some | ||
| form. | ||
|
|
||
| We are working with the rest of the Tensorflow ecosystem (e.g. TF Hub, |
There was a problem hiding this comment.
TensorFlow
(also we aren't exactly working with TF Hub, are we?)
|
Submitting this without the keras.io links before I cut 0.2.0. I will push the keras.io links back as soon as the section is live. |
You can browse what this will actually render like here:
https://github.com/mattdangerw/keras-nlp/tree/update-docs