If, after you reed a csv file and there is Arabic character in your column (originally not), then you might have been using "ISO-8859-1" encoding, just change one.
The training of GRU, LSTM and RNN is quite sensitive to comoputational resources. If you're training GRU, LSTM or RNN and it seems to take a lot of time, try restarting the software (like jupyter notebook). This might save you a lot of time. And by the way, it's normal for a RNN to train slower than GRU and LSTM, the main reason could be that the RNN layer is importing an old tensorflow whose code is not well optimized.
If you tried using plt.savefig to save the plots generated my matplotlib, and you get blank png files, this might be because your saving process is after plt.show(). It should be before, for instance: