How to debug error `gradient computation … modified by an inplace operation" with `torch.autograd.set_detect_anomaly(True)`

Question

I am working on a Sequence-to-Sequence model with Attention in Pytorch. I implemented the attention piece, and now I am getting this RuntimeError: one of the variables