The issues of model training and test accuracy selection

[liyuntong9]

https://github.com/rusty1s/pytorch_geometric/blob/master/benchmark/citation/train_eval.py
I use the code in the above link to train my model and I found the best test accuracy is reported in the epoch 2. However, the training process is just beginning at that point with 92% training accuracy. The training accuracy will converge to 99% with the growth of epoch, whereas the test accuracy decreases less than five percent.
In this case, can I believe my model has achieved the best test accuracy?

[liyuntong9]

Sorry, there are some errors in my code and I already know how to solve it.