RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.LongTensor [9604]] is at version 5; expected version 4 instead. Hint: the backtrace further above shows the operation that failed to compute its gradient.
I use torch.autograd.set_detect_anomaly(True) to detect the error, and the result show that there maybe some error (inplace operation) in WindowsAttention3D's forward function specifically self.relative_position_index or anywhere later. My torch version is 1.6
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.LongTensor [9604]] is at version 5; expected version 4 instead. Hint: the backtrace further above shows the operation that failed to compute its gradient.
I use torch.autograd.set_detect_anomaly(True) to detect the error, and the result show that there maybe some error (inplace operation) in WindowsAttention3D's forward function specifically self.relative_position_index or anywhere later. My torch version is 1.6