MXNet 2.0 API reference

MXNet lives

This is the Python API reference for the smolix/mxnet fork — a port of Apache MXNet 2.0 to CUDA 13 / Blackwell GPUs and native Apple Silicon CPU.

It is generated directly from the docstrings in the installed mxnet package (version 2.0.0+cpu.linux.20260617.1). There is no hand-written narrative or tutorial content; for install, build, fork changes, and known issues see the repository’s README.md, FIXED.md, and OPEN_ISSUES.md.

A warning to anyone wanting to use it: do so at your own risk. This is a port to ensure that some legacy packages, such as D2L can still execute mxnet code on hardware that isn’t ancient. You should not build new software using MXNet but it might greatly simplify your life if you have an old library that you need to maintain.

Note

mxnet.np / mxnet.npx are runtime aliases for the modules documented below as mxnet.numpy and mxnet.numpy_extension.

Packages and modules

mxnet.numpy

MXNet NumPy module.

mxnet.numpy_extension

Module for ops not belonging to the official numpy package for imperative programming.

mxnet.ndarray

NDArray API of MXNet.

mxnet.symbol

Symbol API of MXNet.

mxnet.gluon

Neural network module.

mxnet.optimizer

Optimizer API of MXNet.

mxnet.lr_scheduler

Scheduling learning rate.

mxnet.io

Data iterators for common data formats and utility functions.

mxnet.image

Image Iterators and image augmentation functions

mxnet.autograd

Autograd for NDArray.

mxnet.kvstore

Key-value store for distributed communication

mxnet.device

Device management API of mxnet.

mxnet.profiler

Profiler setting methods.

mxnet.runtime

Runtime querying of compile time features in the native library.

mxnet.contrib

Experimental contributions

Indices