can’t convert cuda:0 device type tensor to numpy 해결

gpu에 올라가 있는 것을 이용한 연산을 시도해서 생긴 에러

• detach()를 통해 gradient 전파에 영향을 안끼치게 떼어낸 새 텐서를 가져오고

• cpu()를 통해 cpu로 복사한다.

input_label = inputs["labels"].detach().cpu()
preds = outputs.logits.detach().cpu().numpy()

• detach() ref: https://pytorch.org/docs/stable/generated/torch.Tensor.detach.html?highlight=detach#torch.Tensor.detach

• cpu() ref: https://pytorch.org/docs/stable/generated/torch.Tensor.cpu.html?highlight=cpu#torch.Tensor.cpu

• numpy() ref: https://pytorch.org/docs/stable/generated/torch.Tensor.numpy.html?highlight=numpy#torch.Tensor.numpy