I just read paper(https://arxiv.org/pdf/2008.12298v2.pdf) in which facebook talks about running their model on devices. It is very recent from september this year but I see that they are still using internal tool “caffe2go” which was never open sourced.
The only talk about doing this and about this framework in this blogpost https://engineering.fb.com/2016/11/08/android/delivering-real-time-ai-in-the-palm-of-your-hand/ . As we have now pytorch mobile I assumed caffe2go is deprecated but facebook is still developing new products with it.
- Is pytorch mobile not stable enough to use on production and that’s why facebook still uses caffe2go ?
- Is pytorch mobile borrows concepts or shares some implementation with caffe2go ?