ONNX export failure: Input, output and indices must be on the current device

[buimanhlinh96]

❔Question

Hi @glenn-jocher ,
I am trying export model from pytorch to onnx. I set --grid argument to ignore Detect layer.
But I met a isssue :ONNX export failure: Input, output and indices must be on the current device
How can I solve this problem? Thanks

[glenn-jocher]

@buimanhlinh96 yes I see this error as well when I reproduce your command. --device is a new argument, it seems it's not been properly tested.

But setting --grid does not achieve what you say you want, it includes the Detect() layer grid reconstructions and concatenation ops.

yolov5/models/export.py

Line 28 in 886f1c0

parser.add_argument('--grid', action='store_true', help='export Detect() layer grid')

[glenn-jocher]

@buimanhlinh96 this behavior is actually documented here:
#2297 (comment)

Workaround is to use either --grid or --device but not both.

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.