This repository contains conda recipes for building MXNet. MXNet is broken up into a number of packages to allow components to be shared and different variants to be built and selected.

The receipes in the repository are as follows:

• libmxnet : C++ MXNet libraries. Comes in various variants depending on the BLAS implementation and GPU support: openblas, mkl, gpu_openblas, gpu_mkl.
  OpenBLAS variants are only supported on linux-64 platform. GPU variants on linux-64 and win-64.
  The mkl variants depend on mkl-dnn and mklml recipes.

• py-mxnet : Python bindings to the MXNet library. Depends on any variant of libmxnet.

• _mutex_mxnet : Mutex package on which libmxnet packages depend. Provides a means to select a particular variant.

• mxnet : metapackage which installs libmxnet and py-mxnet

• mxnet-variants : metapackages which installs a particular CPU variant (openblas or mkl).

• mxnet-gpu : metapackages which install a particular GPU variant (gpu_openblas or gpu_mkl).

The build order is:

• _mutex_mxnet (although existing packages can be re-used)
• libmxnet
• py-mxnet
• mxnet
• mxnet-variants
• mxnet-gpu

Only CPU variant packages were created for linux-64 and osx-64 for the 1.5.0 release. See the 1.2.1 recipes for GPU and Windows support.