The future of caffe2

Hello everyone,
I’m currently working on deploying models in C++ through the Caffe2 API. But regarding the torch c++ API, I’m wondering what’s the future of caffe2 in the development of Pytorch.
Is it still worthy to work with caffe2 or should I move to libtorch ?
Also a second question is what do you use or advice for the deployment of detection models in a C++ project.

Thank you everyone for your feedback !! :slight_smile:

I don’t have any numbers about the usage of Caffe2 vs. PyTorch in different deployment scenarios, it seems that at least some teams are migrating to PyTorch so you could check for similar articles from other teams.