keras-nlp

keras-nlp copied to clipboard

Currently, we use `tf.py_function` to implement BLEU score. This is because the graph mode implementation which we tried out wasn't very efficient. This notebook compares the two approaches: https://colab.research.google.com/drive/1TZ8XnrmMcU8ZE2J-3amb44p-U-hxER53?usp=sharing. We...

abheesht17

Make Decoding Functions Graph-compatible

1

Made an attempt to do the above here: https://colab.research.google.com/drive/1PBMzeBd-HyFE0o4VXwk19-kqHIhOZM49?usp=sharing. Ran into an issue: ``` --------------------------------------------------------------------------- TypeError Traceback (most recent call last) in () 1 inputs = keras.Input(shape=(), dtype="string") ----> 2...

abheesht17

Integrated gradients for nlp models. (Explainable AI)

4

**Describe feature** The [Integrated Gradients](https://www.tensorflow.org/tutorials/interpretability/integrated_gradients) (IG) can be a great tool to understand the neural network. **How API will change?** It would enhance as follows: ```python from keras_nlp.utils.visualization import IntegratedGradients...

innat

Currently, the [`rouge`](https://github.com/google-research/google-research/tree/master/rouge) package does not have functionality to pass a custom tokeniser. This commit takes care of that: https://github.com/google-research/google-research/commit/61ce9f0ca76025dac5b671c0631e443a9975a8a3. However, it is not part of a release yet. Waiting...

abheesht17

Closes [156](https://github.com/keras-team/keras-nlp/issues/156)

blackhat-coder

Add Notebooks for Examples

6

We can add notebooks (or share Colab notebooks) for existing examples in the library (with instructive text and explanation).

abheesht17

Issue discussing the nuances for the Random Swaps layer. Rough colab implementation can be found [here](https://colab.research.google.com/gist/aflah02/236ff822f8eeb29f95423f664afff73a/randomswaps.ipynb)

aflah02

**Describe the bug** When loading a model with TransformerDecoder, with the custom objects option, it rebuilds and reinitializes the layers within TransformerDecoder. **To Reproduce** In keras_nlp/layers/transformer_decoder_test, change the test case...

jessechancy

Stop Word Dropout Layer - Data Augmentation

2

This is an extension to the [Random Deletion Layer](https://github.com/keras-team/keras-nlp/issues/152). I currently plan to use this issue to link the next PR which will add Stop Word Deletion specifically to the...

aflah02

BERT example integration test

3

I think we are at the point where we need some automated testing for this. This runs preprocessing, a few pretraining train steps, and a few finetuning train steps by...

mattdangerw