mindspore.mint.distributed.new_group

mindspore.mint.distributed.new_group(ranks=None, timeout=None, backend=None, pg_options=None, use_local_synchronization=False, group_desc=None)[source]

Create a new distributed group.

Note

This method should be used after mindspore.mint.distributed.init_process_group().
Only support PyNative mode, Graph mode is not currently supported.

Parameters:

ranks (list[int], optional) – List of ranks of group members. Default: None, which will be create the world group.
timeout (int, invalid) – Currently it is a reserved parameter.
backend (str, invalid) – Support backend Library. Currently support "hccl" and "mccl". When backend is "hccl", Huawei Collective Communication Library (HCCL) is used. When backend is "mccl", MindSpore Collective Communication Library (MCCL) is used. Default: None, which means "hccl" in Ascend.
pg_options (GroupOptions, optional) –
Additional communication group configuration parameters. The backend will automatically select supported parameters and apply them during group initialization. E.g., for the HCCL backend, hccl_config can be specified so that group initialization configurations can be applied. Default is None.

GroupOptions is defined as a class that can be instantiated as a Python object.
```
GroupOptions {
    hccl_config(dict)
}
```
hccl_config currently supports "hccl_buffer_size", "hccl_comm", or "hccl_op_expansion_mode".
- hccl_buffer_size (uint32): Specifies the size of the HCCL communication buffer.
- hccl_comm (int64): Specifies an existing HcclComm pointer. If "hccl_comm" is set, other HCCL config parameters will be ignored.
- hccl_op_expansion_mode (str | uint32): Specifies the HCCL operator expansion mode for the communication group. Supported string values are "DEFAULT", "HOST", "AICPU", and "AIV". Supported integer values are 0, 1, 2, and 3.
use_local_synchronization (bool, invalid) – Currently it is a reserved parameter.
group_desc (str, invalid) – Currently it is a reserved parameter.

Returns:

str. The name of the created group. Return "" in the abnormal scenarios.

Raises:

TypeError – If the list of ranks contains duplicate rank IDs.

Supported Platforms:: Ascend

Examples

Note

Before running the following examples, you need to configure the communication environment variables. For Ascend devices, it is recommended to use the msrun startup method without any third-party or configuration file dependencies. Please see the msrun startup for more details.

>>> import mindspore as ms
>>> from mindspore.mint.distributed import init_process_group, new_group
>>> ms.set_device(device_target="Ascend")
>>> init_process_group()
>>> group = new_group()
>>> print("group is: ", group)
group is: hccl_world_group