flow.S/B/P

[clackhan]

>>> import oneflow as flow
>>> b = flow.sbp.broadcast
>>> b
oneflow.sbp.broadcast
>>> type(b)
<class 'oneflow.sbp.sbp'>
>>> p = flow.sbp.partial_sum
>>> p
oneflow.sbp.partial_sum
>>> type(p)
<class 'oneflow.sbp.sbp'>
>>> s0 = flow.sbp.split(0)
>>> s0
oneflow.sbp.split(axis=0)
>>> type(s0)
<class 'oneflow.sbp.sbp'>
>>> d = flow.device("cuda:0")
>>> d
device(type='cuda', index=0)
>>> type(d)
<class 'oneflow.device'>
>>> x = flow.Tensor([1, 2, 3])
>>> x
tensor([1., 2., 3.], dtype=oneflow.float32)
>>> type(x)
<class 'oneflow.Tensor'>

[clackhan]

格式参考：https://github.com/Oneflow-Inc/OneTeam/pull/456/files#diff-ad2703813429d9c97f01909677d9b77526105ad73d3d89c10ff842d358bd95d0R33-R37

[guo-ran]

规定成这样，是确定未来不会有其他Partial了吗，比如partial max