ONNX: export failure: Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0! (when checking argument for argument other in method wrapper__equal)

[taoxunqiang]

感谢这么棒的工作。
YOLOv5 + mobilenetv3 small转fp16的onnx：
python scripts/export.py --weights ./runs/train/YOLOv5-mobilenetv3s/weights/best.pt --include onnx --device 0 --half
出现如下的错误：
ONNX: export failure: Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0! (when checking argument for argument other in method wrapper__equal)

[Bobo~]

@taoxunqiang 导onnx 的时候建议不要指定gpu，不用--device 参数

[taoxunqiang]

@Bobo-y 程序里面要指定gpu，不然会出现：
AssertionError: --half only compatible with GPU export, i.e. use --device 0

[Bobo~]

@taoxunqiang 🤔，我刚试了一下，我没加的确可以的啊，而且转 onnx 一版都不指定gpu 的

[taoxunqiang]

@Bobo-y 也是加--half吗，我需要转成半精度的

[Bobo~]

@taoxunqiang add nothing, only add --weights