代码拉取完成,页面将自动刷新
torch.distributed.new_group(
ranks=None,
timeout=datetime.timedelta(0, 1800),
backend=None
)
For more information, see torch.distributed.new_group.
mindspore.communication.create_group(group, rank_ids)
For more information, see mindspore.communication.create_group.
PyTorch: This interface passes in the rank list of the communication domain to be constructed, specifies the backend to create the specified communication domain, and returns the created communication domain.
MindSpore: This interface passes in the group name and the rank list of the communication domain to be constructed, creates a communication domain with the incoming group name as the key, and does not return any value.
Class | Sub-class | PyTorch | MindSpore | Difference |
---|---|---|---|---|
Param | Param 1 | - | group | PyTorch does not have this param. MindSpore: the group name |
Param 2 | ranks | rank_ids | The functionalities are the same, but have different names | |
Param 3 | timeout | - | PyTorch: the timeout value. MindSpore does not have this param, and should set the corresponding environment variables before calling this interface | |
Param 4 | backend | - | PyTorch: the communication backend. MindSpore does not have this param, and should set the corresponding environment variables before calling this interface |
此处可能存在不合适展示的内容,页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容,可点击提交进行申诉,我们将尽快为您处理。