mindspore::lite

Allocator

Allocator defines a memory pool for dynamic memory malloc and memory free.

Context

Context is defined for holding environment variables during runtime.

Constructors & Destructors

Context

Context()

Constructor of MindSpore Lite Context using default value for parameters.

~Context

~Context()

Destructor of MindSpore Lite Context.

Public Attributes

vendor_name_

vendor_name_

A string value. Describes the vendor information. This attribute is used to distinguish from different vendors.

thread_num_

thread_num_

An int value. Defaults to 2. Thread number config for thread pool.

allocator

allocator

A pointer pointing to Allocator.

device_list_

device_list_

A DeviceContextVector contains DeviceContext variables.

CPU, GPU and NPU are supported now. If GPU device context is set and GPU is supported in the current device, use GPU device first, otherwise use CPU device first. If NPU device context is set and GPU is supported in the current device, use NPU device first, otherwise use CPU device first.

Model

#include <model.h>

Model defines model in MindSpore Lite for managing graph.

Destructors

~Model

virtual ~Model()

Destructor of MindSpore Lite Model.

Public Member Functions

Free

void Free()

Free MetaGraph in MindSpore Lite Model to reduce memory usage during inference.

Destroy

void Destroy()

Free all temporary memory in MindSpore Lite Model.

Static Public Member Functions

Import

static Model *Import(const char *model_buf, size_t size)

Static method to create a Model pointer.

Parameters
- model_buf: Defines the buffer read from a model file.
- size: variable. Defines the byte number of model buffer.
Returns

Pointer that points to the MindSpore Lite Model.

CpuBindMode

#include <context.h>

An enum type. CpuBindMode is defined for holding arguments of the bind CPU strategy.

Public Attributes

MID_CPU

MID_CPU = 2

Bind middle cpu first.

HIGHER_CPU

HIGHER_CPU = 1

Bind higher cpu first.

NO_BIND

NO_BIND = 0

No bind.

DeviceType

#include <context.h>

An enum type. DeviceType is defined for holding user’s preferred backend.

Public Attributes

DT_CPU

DT_CPU = 0

CPU device type.

DT_GPU

DT_GPU = 1

GPU device type.

DT_NPU

DT_NPU = 2

NPU device type.

Version

#include <version.h>

std::string Version()

Global method to get a version string.

Returns

The version string of MindSpore Lite.

StringsToMSTensor

int StringsToMSTensor(const std::vector<std::string> &inputs, tensor::MSTensor *tensor)

Global method to store strings into MSTensor.

Returns

STATUS, STATUS is defined in errorcode.h.

MSTensorToStrings

std::vector<std::string> MSTensorToStrings(const tensor::MSTensor *tensor)

Global method to get strings from MSTensor.

Returns

The vector of strings.

DeviceContextVector

#include <context.h>

A vector contains DeviceContext variable.

DeviceContext

#include <context.h>

DeviceContext defines different device contexts.

Public Attributes

device_type_

device_type_

An enum type. Defaults to DT_CPU. DeviceType is defined for holding user’s CPU backend.

device_info_

device_info_

An strucr value, contains CpuDeviceInfo, GpuDeviceInfo and NpuDeviceInfo.

DeviceInfo

#include <context.h>

An struct value. DeviceInfo is defined for backend’s configuration information.

Public Attributes

cpu_device_info_

cpu_device_info_

CpuDeviceInfo is defined for CPU’s configuration information.

gpu_device_info_

gpu_device_info_

GpuDeviceInfo is defined for GPU’s configuration information.

npu_device_info_

GpuDeviceInfo is defined for NPU’s configuration information.

CpuDeviceInfo

#include <context.h>

CpuDeviceInfo is defined for CPU’s configuration information.

Public Attributes

enable_float16_

enable_float16_

A bool value. Defaults to false. This attribute enables to perform the GPU float16 inference.

Enabling float16 inference may cause low precision inference，because some variables may exceed the range of float16 during forwarding.

cpu_bind_mode_

cpu_bind_mode_

A CpuBindMode enum variable. Defaults to MID_CPU.

GpuDeviceInfo

#include <context.h>

GpuDeviceInfo is defined for GPU’s configuration information.

Public Attributes

enable_float16_

enable_float16_

A bool value. Defaults to false. This attribute enables to perform the GPU float16 inference.

Enabling float16 inference may cause low inference precision, because some variables may exceed the range of float16 during forwarding.

NpuDeviceInfo

#include <context.h>

NpuDeviceInfo is defined for NPU’s configuration information.

frequency_

A int value. Defaults to 3. This attribute is used to set the NPU frequency, which can be set to 1 (low power consumption), 2 (balanced), 3 (high performance), 4 (extreme performance).

TrainModel

#include <model.h>

Inherited from Model, TrainModel defines a class that allows to import and export the MindSpore trainable model.

Constructors & Destructors

~TrainModel

virtual ~TrainModel();

Class destructor, free all memory.

Public Member Functions

Import

static TrainModel *Import(const char *model_buf, size_t size);

Static method to create a TrainModel object.

Parameters
- model_buf: A buffer that was read from a MS model file.
- size: Length of the buffer.
Returns

Pointer to MindSpore Lite TrainModel.

Free

void Free() override;

Free meta graph related data.

ExportBuf

char *ExportBuf(char *buf, size_t *len) const;

Export Model into a buffer.

Parameters
- buf: The buffer to be exported into. If it is equal to nullptr, buf will be allocated.
- len: Size of the pre-allocated buffer and the returned size of the exported buffer.
Returns

Pointer to buffer with exported model.

Public Attributes

buf_size_

size_t buf_size_;

The length of the buffer with exported model.