Supports both imperative and symbolic programming
Runs on CPUs or GPUs, on clusters, servers, desktops, or mobile phones
Calculates the gradient automatically for training a model
Supports distributed training on multiple CPU/GPU machines, including AWS, GCE, Azure, and Yarn clusters
Optimized C++ backend engine parallelizes both I/O and computation
MXNet is developed by collaborators from multiple universities and companies. We sincerely thank the following organizations for supporting MXNet and sponsoring its major developers (alphabetical order).