torch.autograd.Variable(target.cuda(async=True)) ---用async调用cuda()出现 SyntaxError: invalid syntax

 又比如：

./projects/personachat/persona_seq2seq.py:2220:43: E999 SyntaxError: invalid syntax
                    self.zs.copy_(zs, async=True)
                                          ^
./parlai/agents/vsepp_caption/vsepp_caption.py:160:38: E999 SyntaxError: invalid syntax
            images = images.cuda(async=True)
                                     ^
./parlai/agents/drqa/model.py:89:43: E999 SyntaxError: invalid syntax
            inputs = [Variable(e.cuda(async=True)) for e in ex[:5]]
                                          ^
3     E999 SyntaxError: invalid syntax

原因：

• python3.7已经移除了async关键字，而用non_blocking代替。(导致apache-airflow也出了问题)

• cuda() 本身也没有async. 构造函数如下：

  cuda(device=None, non_blocking=False) → Tensor

参考：

https://github.com/facebookresearch/ParlAI/issues/1141

https://stackoverflow.com/questions/56085657/pytorch-nccl-error-unhandled-system-error-during-backprop

**kwargs: For compatibility, may contain the key ``async`` in place of the ``non_blocking`` argument. The ``async`` arg is deprecated.

解决办法：

• 直接用cuda()
• 用 async_ 代替
• 用non_blocking代替。ps：两个关键字(device and non_blocking)可以用来调用cuda( ) ，device是选取显卡序列号的。

• • non_blocking (bool):
    If True and the source is in pinned memory, the copy will be asynchronous with respect to the host. Otherwise, the argument has no effect. Default: False.
  • 如果为True且源位于锁定内存中，则副本将与主机异步。 否则，参数无效。 默认值：False。Always good to take a look at the Documentation:

• 更新 Pipenv。 pip install pipenv  https://github.com/quark0/darts/pull/25 

https://pytorch.org/docs/stable/tensors.html#torch.Tensor.cuda