Low accuracy when using Eval.py to test on dev_data?

[mfxss]

While training the evaluation accuracy was 89%. But only 14% using eval.py. I am using the same data.
I found out what happened. I was using an old version of data_helper which shuffled test data for all.
Thank you anyway.

[self-ms]

If your data is embedded and labels are available, you can use the following repository:
https://github.com/ms-unlimit/Transformer-Based-Machine-Learning-Framework