Figure 10. Loss changes over learning iterations.

As can be seen, although the training dataset shows more stable progress, the test dataset is less stable, although it progresses together as in Fig. 9.