Caffe2 improves Caffe. It has first-class support for large-scale distributed training mobile deployment . It gives new hardware support in addition to CPU and CUDA. It is flexible for future operations such as quantized computation. It is stress tested by the vast scale of Facebook applications.
Caffe2 helps researchers in Facebook train large machine learning models and deliver AI to mobile devices. Developers have access to different tools, enabling them to run large-scale distributed training scenarios and build machine learning applications for mobile using its cross-platform libraries.