Need to alter LD_LIBRARY_PATH

[epowell101]

Also found it necessary to make sure this was set in the ~/.bashrc

There is some discussion here:
https://discuss.tensorflow.org/t/could-not-load-dynamic-library-libcudart-so-11-0/15711/10

[epowell101]

I actually do NOT think that is necessary if you get the right dependencies sorted, which is tough as on a cloud environment there are a couple of variables that are beyond your control. Therefore just using what worked for Sihao and other authors does not necessarily work.

The following env file saved as txt for sharing shows the Conda environment I'm using which is now working. Convert to yml and use w Conda
October_22_last.txt

[epowell101]

Here are some requirements of the data files that are needed for pretraining:

train_input_file - Training data TFRecord file:
File path: ./inter_data/train.tfrecord.{bizdate}
Passed via flag
Contains training examples
test_input_file - Evaluation data TFRecord file:
File path: ./inter_data/test.tfrecord.{bizdate}
Passed via flag
bert_config_file - BERT model configuration JSON file:
File path: ./bert_config.json
Passed via flag
Specifies model architecture
vocab_file - Vocabulary pickle file
File path: {data_dir}/{vocab_filename}.{bizdate}
{data_dir} is ./inter_data/ by default
Contains vocabulary for dataset