Transformer/Transformer(Greedy_decoder)-Torch.py on gpu

Hello, I want to put the Transformer (Greedy_decoder)-Torch.py code on the gpu, using model=model.to(device), input_data also to (device), but the error still appears "Expected object of backend CUDA but backend CPU for argument #2 'mat2”

从 Transformer/Transformer(Greedy_decoder)-Torch.py 开始，重现报告的 CUDA-versus-CPU mat2 错误。跟踪 model、input_data 以及涉及失败操作的张量所在的设备。当示例能够在 GPU 上运行且不存在后端不匹配时，即视为完成。

由索引模型根据 Issue 内容生成。