RuntimeError: one of the variables needed for gradient computation has been modified by an inplace o

问题

RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [1, 2]], which is output 0 of SoftmaxBackward, is at version 1; expected version 0 instead. Hint: enable anomaly detection to find the operation that failed to compute its gradient, with torch.autograd.set_detect_anomaly(True).

由于我对softmax输出的值就行了修改 ，出现如上报错

filter_value = float(0)
# 把大于top-k的值置于0，留下小的
indices_to_remove = semantic_p[j] > torch.topk(semantic_p[j], 1, largest=False)[0][..., -1, None
semantic_p[j][indices_to_remove] = filter_value
lu_s_similarity[j][indices_to_remove] = filter_value
s_pse_data2.append(lu_s_similarity[j])
pselabel2.append(semantic_p[j])

解决办法

对需要softmax的值就行如下操作，再进行softmax

lu_similarity = lu_similarity.detach()
semantic_p = torch.softmax(lu_similarity, dim=1)