I have a strange problem coding the call method of the Encoder class;
indeed the output of the call method returns a tensor with Nan values.
Adding a “print” statement after every step I have identified that the error is coming from the previous question UNQ_C4, specifically in the MultiHeadAttention layer which outputs a tensor with Nan values. This was very surprising since I correctly pass the test for the UNQ_C4 with the normal output.
I know I cannot post my code, but if anyone is willing to check I would really appreciate it.
Thanks in advance