Video2Video 跑不起来（OOM，多块显卡无用）

[youtianhong]

text2video 可以跑 （环境4块显卡，共64G）
1.使用默认显卡，video2video跑不了，
报torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 456.00 MiB (GPU 0; 14.57 GiB total capacity; 13.55 GiB already allocated; 192.75 MiB free; 14.12 GiB reserved in total by PyTorch)
2.设置使用不同显卡：在video2Video.py中设置如cuda:2,会报
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cuda:1! (when checking argument for argument weight in method wrapper_CUDA__cudnn_convolution)

怎解？