InternVL-14B-224px，多gpu推理报错。使用device_map='auto'时，“RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cuda:1!”

[chengligen]

设备为8*V100，40G显存版本，使用device_map='auto'时，部署InternVL-14B-224px报错，错误如下。

RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cuda:1!