tfhub-bert

1. Setup local cache directory for tensorflow hub

   mkdir "${HOME}/tf-hub-cache"
   
   cp ~/.bash_profile ~/.bash_profile.bkup
   echo 'export TFHUB_CACHE_DIR="${HOME}/tf-hub-cache"' >> ~/.bash_profile
   source "${HOME}/.bash_profile"
   
   echo $TFHUB_CACHE_DIR

2. Create a new conda environment (let's call it bert) and install requirements

   conda create -n bert python=3.7
   source activate bert
   
   (bert) pip install -r requirements.txt

   NOTE: if source activate bert does not work, try conda activate bert

3. Download a Bert Model, and copy vocab file. Feel free to replace BERT URL form list of available modules. You would also need to change the local vocab path accordingly.

   python download_bert.py
   cp $TFHUB_CACHE_DIR/ecd2596ce849110246602e3d4d81e2d9719cb027/assets/vocab.txt .

4. Get tokenization.py from the official BERT repository

5. Start jupyter

   jupyter notebook