mindspore
Data Presentation
Tensor
Tensor is a data structure that stores an n-dimensional array. |
|
Create a new Tensor in Cell.construct() or function decorated by @jit. |
|
A sparse representation of a set of nonzero elements from a tensor at given indices. |
|
Constructs a sparse tensor in CSR (Compressed Sparse Row) format, with specified values indicated by values and row and column positions indicated by indptr and indices. |
|
A sparse representation of a set of tensor slices at given indices. |
|
A sparse representation of a set of nonzero elements from a tensor at given indices. |
Parameter
Parameter is a Tensor subclass, when they are assigned as Cell attributes they are automatically added to the list of its parameters, and will appear, e.g. |
|
Inherited from tuple, ParameterTuple is used to save multiple parameter. |
DataType
- class mindspore.dtype
Create a data type object of MindSpore.
The actual path of
dtype
is/mindspore/common/dtype.py
. Run the following command to import the package:from mindspore import dtype as mstype
Numeric Type
Currently, MindSpore supports
int
type,uint
type,float
type andcomplex
type. The following table lists the details.Definition
Description
mindspore.int8
,mindspore.byte
8-bit integer
mindspore.int16
,mindspore.short
16-bit integer
mindspore.int32
,mindspore.intc
32-bit integer
mindspore.int64
,mindspore.intp
64-bit integer
mindspore.uint8
,mindspore.ubyte
unsigned 8-bit integer
mindspore.uint16
,mindspore.ushort
unsigned 16-bit integer
mindspore.uint32
,mindspore.uintc
unsigned 32-bit integer
mindspore.uint64
,mindspore.uintp
unsigned 64-bit integer
mindspore.float16
,mindspore.half
16-bit floating-point number
mindspore.float32
,mindspore.single
32-bit floating-point number
mindspore.float64
,mindspore.double
64-bit floating-point number
mindspore.bfloat16
16-bit brain-floating-point number
mindspore.complex64
64-bit complex number
mindspore.complex128
128-bit complex number
Other Type
For other defined types, see the following table.
Type
Description
tensor
MindSpore's
tensor
type. Data format uses NCHW. For details, see tensor.bool_
Boolean
True
orFalse
.int_
Integer scalar.
uint
Unsigned integer scalar.
float_
Floating-point scalar.
complex
Complex scalar.
number
Number, including
int_
,uint
,float_
,complex
andbool_
.list_
List constructed by
tensor
, such asList[T0,T1,...,Tn]
, where the elementTi
can be of different types.tuple_
Tuple constructed by
tensor
, such asTuple[T0,T1,...,Tn]
, where the elementTi
can be of different types.function
Function. Return in two ways, when function is not None, returns Func directly, the other returns Func(args: List[T0,T1,…,Tn], retval: T) when function is None.
type_type
Type definition of type.
type_none
No matching return type, corresponding to the
type(None)
in Python.symbolic_key
The value of a variable is used as a key of the variable in
env_type
.env_type
Used to store the gradient of the free variable of a function, where the key is the
symbolic_key
of the free variable's node and the value is the gradient.
- class mindspore.common.np_dtype
np_dtype
expands Numpy's data types.The actual path of
np_dtype
is/mindspore/common/np_dtype.py
. Run the following command to import the package:from mindspore.common import np_dtype
Numeric Type
Type
Description
bfloat16
The
bfloat16
data type under NumPy. This type is only used to construct Tensor of typebfloat16
, and does not guarantee the full computing power under Numpy. Takes effect only if the version of Numpy at runtime is not less than the version of Numpy at compilation.
Convert MindSpore dtype to numpy data type. |
|
Convert MindSpore dtype to python data type. |
|
Convert python type to MindSpore type. |
|
Get the MindSpore data type, which corresponds to python type or variable. |
|
An enum for quant datatype, contains INT1 ~ INT16, UINT1 ~ UINT16. |
Context
Set context for running environment. |
|
Get context attribute value according to the input key. |
|
Set auto parallel context, only data parallel supported on CPU. |
|
Get auto parallel context attribute value according to the key. |
|
Reset auto parallel context attributes to the default values. |
|
Parallel mode options. |
|
Set parameter server training mode context. |
|
Get parameter server training mode context attribute value according to the key. |
|
Reset parameter server training mode context attributes to the default values. |
|
Set parameters in the algorithm for parallel strategy searching. |
|
Get the algorithm parameter config attributes. |
|
Reset the algorithm parameter attributes. |
|
Configure heterogeneous training detailed parameters to adjust the offload strategy. |
|
Gets the offload configuration parameters. |
Seed
Set global seed. |
|
Get global seed. |
Random Number Generator
Get the state of the default generator. |
|
A generator that manages the state of random numbers and provides seed and offset for random functions. |
|
Return the initial seed of the default generator. |
|
Set the default generator seed. |
|
Seed the default generator with random number. |
|
Set the state of the default generator. |
Serialization
Get the status of asynchronous save checkpoint thread. |
|
Build strategy of every parameter in network. |
|
Check whether the checkpoint is valid. |
|
Convert mindir model to other format model. |
|
Export the MindSpore network into an offline model in the specified format. |
|
Load MindIR. |
|
Load checkpoint info from a specified file. |
|
Load checkpoint info from a specified file asyncly. |
|
Load checkpoint into net for distributed predication. |
|
load protobuf file. |
|
Load parameters into network, return parameter list that are not loaded in the network. |
|
Load checkpoint info from a specified file. |
|
Merge parallel strategy between all pipeline stages in pipeline parallel mode. |
|
Merge parameter slices into one parameter. |
|
Obfuscate a model of MindIR format. |
|
Parse data file generated by |
|
List of original distributed checkpoint rank index for obtaining the target checkpoint of a rank_id during the distributed checkpoint conversion. |
|
Build rank list, the checkpoint of ranks in the rank list has the same contents with the local rank who saves the group_info_file_name. |
|
Save checkpoint to a specified file. |
|
save protobuf file. |
|
Transform distributed checkpoint from source sharding strategy to destination sharding strategy by rank for a network. |
|
Transform distributed checkpoint from source sharding strategy to destination sharding strategy for a rank. |
Automatic Differentiation
A wrapper function to generate the gradient function for the input function. |
|
A wrapper function to generate the function to calculate forward output and gradient for the input function. |
|
When return_ids of |
|
Compute Jacobian via forward mode, corresponding to forward-mode differentiation. |
|
Compute Jacobian via reverse mode, corresponding to reverse-mode differentiation. |
|
Compute the jacobian-vector-product of the given network. |
|
Compute the vector-jacobian-product of the given network. |
Parallel Optimization
Automatic Vectorization
Vectorizing map (vmap) is a kind of higher-order function to map fn along the parameter axes. |
Parallel
Parallel layout describes the detailed sharding information. |
|
Broadcast parameter to other rank in data parallel dimension. |
|
This function is used to reduce memory, when run block, rather than storing the intermediate activation computed in forward pass, we will recompute it in backward pass. |
|
Specify the tensor by the given layout. |
|
Defining the input and output layouts of this cell and the parallel strategies of remaining ops will be generated by sharding propagation. |
|
synchronize pipeline parallel stage shared parameters. |
JIT
Jit config for compile. |
|
Create a callable MindSpore graph from a Python function. |
|
Class decorator for user-defined classes. |
|
Class decorator for user-defined classes. |
|
Create a callable MindSpore graph from a Python function. |
|
Recycle memory used by MindSpore. |
|
Make a constant value mutable. |
|
Used to calculate constant in graph copmpiling process and improve compile performance in GRAPH_MODE. |
|
Make the cell to be reusable. |
|
Make the function to be reusable. |
Tool
Dataset Helper
DatasetHelper is a class to process the MindData dataset and provides the information of dataset. |
|
Symbol is a data structure to indicate the symbolic info of shape. |
|
Connect the network with dataset in dataset_helper. |
|
A wrapper function to generate a function for the input function. |
Debugging and Tuning
This class to enable the profiling of MindSpore neural networks. |
|
SummaryCollector can help you to collect some common information, such as loss, learning late, computational graph and so on. |
|
SummaryLandscape can help you to collect loss landscape information. |
|
SummaryRecord is used to record the summary data and lineage data. |
|
Enable or disable dump for the target and its contents. |
Log
Get the logger level. |
|
Get logger configurations. |
Installation Verification
Provide a convenient API to check if the installation is successful or failed. |