mindspore_lite.GPUDeviceInfo

class mindspore_lite.GPUDeviceInfo(device_id=0, enable_fp16=False)

Helper class used to describe GPU device hardware information, and it inherits mindspore_lite.DeviceInfo base class.

Parameters

• device_id (int, optional) – The device id. Default: 0.

• enable_fp16 (bool, optional) – enables to perform the Float16 inference. Default: False.

Raises

• TypeError – device_id is not an int.

• TypeError – enable_fp16 is not a bool.

• ValueError – device_id is less than 0.

Examples

>>> # Use case: inference on GPU device.
>>> # precondition 1: Building MindSpore Lite GPU package by export MSLITE_GPU_BACKEND=tensorrt.
>>> # precondition 2: install wheel package of MindSpore Lite built by precondition 1.
>>> import mindspore_lite as mslite
>>> gpu_device_info = mslite.GPUDeviceInfo(device_id=1, enable_fp16=False)
>>> print(gpu_device_info)
device_type: DeviceType.kGPU,
device_id: 1,
enable_fp16: False.
>>> cpu_device_info = mslite.CPUDeviceInfo(enable_fp16=False)
>>> context = mslite.Context()
>>> context.append_device_info(gpu_device_info)
>>> context.append_device_info(cpu_device_info)
>>> print(context)
thread_num: 0,
inter_op_parallel_num: 0,
thread_affinity_mode: 0,
thread_affinity_core_list: [],
enable_parallel: False,
device_list: 1, 0, .

get_group_size()

Get the number of the clusters from context.

Returns

int, the number of the clusters.

Examples

>>> # Use case: inference on GPU device.
>>> # precondition 1: Building MindSpore Lite GPU package by export MSLITE_GPU_BACKEND=tensorrt.
>>> # precondition 2: install wheel package of MindSpore Lite built by precondition 1.
>>> import mindspore_lite as mslite
>>> device_info = mslite.GPUDeviceInfo(device_id=1, enable_fp16=True)
>>> group_size = device_info.get_group_size()
>>> print(group_size)
1

get_rank_id()

Get the ID of the current device in the cluster from context.

Returns

int, the ID of the current device in the cluster, which starts from 0.

Examples

>>> # Use case: inference on GPU device.
>>> # precondition 1: Building MindSpore Lite GPU package by export MSLITE_GPU_BACKEND=tensorrt.
>>> # precondition 2: install wheel package of MindSpore Lite built by precondition 1.
>>> import mindspore_lite as mslite
>>> device_info = mslite.GPUDeviceInfo(device_id=1, enable_fp16=True)
>>> rank_id = device_info.get_rank_id()
>>> print(rank_id)
0