Does deploying exported ONNX model need caffe2?

[leo-paz]

Hi all,

I noticed that the deployment documentation mentions that exported onnx models will have caffe2 custom ops. Does this mean that you could also use this exported onnx model in a pytorch environment since the two have merged? Therefore also being truly ONNX and usable accross multiple environments?

If not, any help on best practices for deploying to sagemaker with the model I trained with detectron2 would be very helpful.