Ryan Wheeler

You could reimplement the QKV / dense logic in terms of einsum for faster computation. An example layer [here](https://github.com/tensorflow/models/blob/master/official/nlp/modeling/layers/dense_einsum.py) and the use [here](https://github.com/tensorflow/models/blob/master/official/nlp/modeling/layers/attention.py#L84). This is how it is is now...

Bump tensorflow to 2.2.3?

8

I found a few tflite ops (mostly rnn related) that were supported in the master of tensorflow but not in the specific commit these bindings were built against. I was...

@boncheolgu Any thoughts on how to use this project to build with the hexagon delegate https://www.tensorflow.org/lite/performance/hexagon_delegate?

Using tensorflow_text with tflite.

8

tensorflow_text is great -- we were previously developing custom c++ ops and compiling them with tensorflow serving, now we don't need to. There are a number of text models I...