Find Error and Improve a Reinforcement Learning Algorithm in Tensorflow 1.14

I need someone who is expert in neural networks implementations in Tensorflow.

I would like to apply this algorithm:

https://github.com/lena-voita/the-story-of-heads

in another context, but there is an error in my implementation.

I need someone to understand the above algorithm and fix the error!

Any code added must be documented, tested and professionally written.