如何把loss的tensor换成标量

在pytorch中计算平均loss如下代码:

t = tqdm(trainLoader, leave=False, total=len(trainLoader))
for i, (idx, targetVar, inputVar, _, _) in enumerate(t):
    inputs = inputVar.to(device) 
    label = targetVar.to(device) 
    optimizer.zero_grad()
    net.train()
    pred = net(inputs)  # B,S,C,H,W
    loss = lossfunction(pred, label)
    loss_aver = loss.item() / args.batch_size #pytorch中直接用.item()转换为标量
    train_losses.append(loss_aver)
    loss.backward()
    optimizer.step()

而mindspore中item不行

data_iterator = validDataset.create_tuple_iterator()
for data, label in data_iterator:
    output = net(data)
    loss = lossfunction(output, label)
    loss_aver = loss.item() / args.batch_size
    valid_losses.append(loss_aver)

会报错

ValueError: Can only convert an array of size 1 to a Python scalar

float也不行

data_iterator = validDataset.create_tuple_iterator()
for data, label in data_iterator:
    output = net(data)
    loss = lossfunction(output, label)
    loss_aver = float(loss) / args.batch_size
    valid_losses.append(loss_aver)

会报错

TypeError: float() argument must be a string or a number, not 'Tensor'

那请问如何操作把tensor转为标量呢? 已经翻了一下午mindspore文档了都找不到呜呜呜

解答：

loss.asnumpy().item()解决的 先转为numpy再换标量