In almost all tests there was missing a line that made that all weights and biases from the references were not taken into account when doing the comparison.
In some cases, some of the backward hidden states weren’t read due to the way I made the read_from_csv function.
In addition, I'm trying a new way to compare the values by just cheking if the difference is less than a certain threshold.
In this PR I try to solve this to errors:
In addition, I'm trying a new way to compare the values by just cheking if the difference is less than a certain threshold.