oneflow.Tensor.to_local

Tensor.to_local(**kwargs)Tensor

Returns the local component of this global tensor in the current rank.

Keyword Arguments

copy (bool, optional) – When copy is set, a new replicated tensor of the local component of this global tensor in the current rank is returned. Default: False

Note

This tensor should be a global tensor, and it returns a empty tensor if there is no local component in the current rank.

No copy occurred in this operation if copy is not set.

For example:

>>> # Run on 2 ranks respectively
>>> import oneflow as flow
>>> x = flow.tensor([0., 1.], dtype=flow.float32, placement=flow.placement("cpu", ranks=[0, 1]), sbp=[flow.sbp.split(0)]) 
>>> y = x.to_local() 
>>> print(y.size()) 
>>> print(y) 
>>> # results on rank 0
oneflow.Size([1])
tensor([0.], dtype=oneflow.float32)
>>> # results on rank 1
oneflow.Size([1])
tensor([1.], dtype=oneflow.float32)