one of the variables needed for gradient computation has been modified by an inplace operation: [tor

ulatu = torch.stack(ulats, dim=1)
ulatuu = torch.mean(ulatu, dim=1)
ilati = torch.stack(ilats, dim=1)
ilatii = torch.mean(ilati, dim=1)
ulats = torch.stack(ulats, dim=1)

进行拼接的时候似乎会出现这种错误，不知道为什么。如果用循环相加来改变这个代码就不会出现问题了。比如：

all_userEmbeddings = torch.Tensor(torch.empty(uEmbed0.shape[0], uEmbed0.shape[1])).to('cuda:0')
for i in ulats:
    all_userEmbeddings += i

all_itemEmbeddings = torch.Tensor(torch.empty(iEmbed0.shape[0], iEmbed0.shape[1])).to('cuda:0')
for j in ilats:
    all_itemEmbeddings += j