API

MindSpore provides rich interfaces for model building, training, and inference. The functions of each module interface are described below.

Module Name	Descriptions
mindspore	Framework foundation interface.
mindspore.ops	Function interface.
mindspore.nn	A neural network layer for building predefined building blocks or computational units in a neural network.
mindspore.mint	Functional, nn, and optimizer interfaces consistent with mainstream industry usage.
mindspore.common.initializer	Parameter initialization.
mindspore.amp	Mixed-precision interface.
mindspore.train	Training interface.
mindspore.parallel	Auto Parallel interface.
mindspore.runtime	Runtime interface.
mindspore.device_context	Device and backend management interface.
mindspore.communication	Collection communication interface.
mindspore.dataset	Interfaces for loading and processing various datasets.
mindspore.numpy	NumPy Class interface.
mindspore.scipy	SciPy Class interface.
mindspore.multiprocessing	Multi-processing interface.
mindspore.tools	Tools interface.
mindspore.utils	Auxiliary interfaces such as dryrun.
mindspore.experimental	Experimental interface.
mindspore.ops.primitive	Primitive operator.
mindspore.boost	Automatic acceleration network interfaces.
mindspore.nn.probability	Parameterizable probability distributions and sampling functions.
mindspore.rewrite	Custom rule-based model source code modification interface.
mindspore.hal	Interface of device management, stream management, event management and memory management.
Environment Variables	Notes related to environment variables.
PyTorch and MindSpore API Mapping Table	API mapping relationship and difference analysis between PyTorch and MindSpore.