bert_demo documentation

[tahouse]

The current documentation for bert_demo mentions to "Use the Neuron compatible BERT-Large implementation and public BERT-Large weights to generate a saved model using steps outlined in public BERT documentation here. " (link: google-research/bert#146)

The public documentation referenced is (currently) an open github issue discussion on the BERT repository -- not public documentation. No full working example is currently available at that link. If there is a known/vetted solution, can you please add it to the the bert_demo so that users can replicate the demonstration?

Thank you

[HahTK]

It appears you have been sent the bert_demo link before the right PRs were processed. Please look at the README again.

You will see that the right version of the README.md points to these links -

1. To fine tune BERT -
   https://github.com/google-research/bert#sentence-and-sentence-pair-classification-tasks which generates the weights.
2. To create a saved model -
   Use BERT fine-tuned model for Tensorflow serving google-research/bert#146 (comment)

The correct version of the README it refers to the answer to the issue used by one of the engineers to generate the saved model.

[HahTK]

Let me know if this is sufficient or if further guidance is needed.

[tahouse]

I recommend you copy the necessary code into Neuron README/repo. This way it is self-contained and can easily be rerun by new users.

[HahTK]

I updated the instructions to include all missing info.
Let me know if this works for you so that we can close the issue or fix other errors.
Thanks

[micwade-aws]

Since we haven’t heard back in a while we are closing this issue, please reopen it if more support from us is needed.