oneflow.Tensor¶
OneFlow Tensor Class¶

class
oneflow.
Tensor
¶ 
property
T
¶ Is this Tensor with its dimensions reversed.
If n is the number of dimensions in x, x.T is equivalent to x.permute(n1, n2, …, 0).

abs
()¶ See
oneflow.abs()

acos
()¶ See
oneflow.acos()

acosh
()¶ See
oneflow.acosh()

add
(other, *, alpha=1)¶ See
oneflow.add()

add_
(other, *, alpha=1)¶ Inplace version of
oneflow.Tensor.add()
.

addcmul
(tensor1, tensor2, *, value=1)¶

addcmul_
(tensor1, tensor2, *, value=1)¶ Inplace version of
oneflow.Tensor.addcmul()
.

addmm
(mat1, mat2, alpha=1, beta=1)¶ See
oneflow.addmm()

amax
(dim=None, keepdim=False)¶ See
oneflow.amax()

amin
(dim=None, keepdim=False)¶ See
oneflow.amin()

arccos
()¶ See
oneflow.arccos()

arccosh
()¶

arcsin
()¶ See
oneflow.arcsin()

arcsinh
()¶

arctan
()¶ See
oneflow.arctan()

arctanh
()¶

argmax
(dim=None, keepdim=None)¶ See
oneflow.argmax()

argmin
(dim=None, keepdim=None)¶ See
oneflow.argmin()

argsort
(dim=None, descending=None)¶

argwhere
()¶

asin
()¶ See
oneflow.asin()

asinh
()¶ See
oneflow.asinh()

atan
()¶ See
oneflow.atan()

atan2
(other)¶ See
oneflow.atan2()

atanh
()¶ See
oneflow.atanh()

backward
(gradient=None, retain_graph=False, create_graph=False)¶ The interface is consistent with PyTorch. The documentation is referenced from: https://pytorch.org/docs/1.10/generated/torch.Tensor.backward.html.
Computes the gradient of current tensor w.r.t. graph leaves.
The graph is differentiated using the chain rule. If the tensor is nonscalar (i.e. its data has more than one element) and requires gradient, the function additionally requires specifying gradient. It should be a tensor of matching type and location, that contains the gradient of the differentiated function w.r.t. self.
This function accumulates gradients in the leaves  you might need to zero .grad attributes or set them to None before calling it. See Default gradient layouts for details on the memory layout of accumulated gradients.
Note
If you run any forward ops, create gradient, and/or call backward in a userspecified CUDA stream context, see Stream semantics of backward passes.
Note
When inputs are provided and a given input is not a leaf, the current implementation will call its grad_fn (though it is not strictly needed to get this gradients). It is an implementation detail on which the user should not rely. See https://github.com/pytorch/pytorch/pull/60521#issuecomment867061780 for more details.
 Parameters
gradient (Tensor or None) – Gradient w.r.t. the tensor. If it is a tensor, it will be automatically converted to a Tensor that does not require grad unless create_graph is True. None values can be specified for scalar Tensors or ones that don’t require grad. If a None value would be acceptable then this argument is optional.
retain_graph (bool, optional) – If False, the graph used to compute the grads will be freed. Note that in nearly all cases setting this option to True is not needed and often can be worked around in a much more efficient way. Defaults to the value of create_graph.
create_graph (bool, optional) – If True, graph of the derivative will be constructed, allowing to compute higher order derivative products. Defaults to False.

bmm
(other)¶ See
oneflow.bmm()

byte
()¶ self.byte() is equivalent to self.to(oneflow.uint8). See
oneflow.Tensor.to()

cast
(dtype)¶ See
oneflow.cast()

ceil
()¶ See
oneflow.ceil()

chunk
(chunks=None, dim=None)¶ See
oneflow.chunk()

clamp
(min=None, max=None)¶ See
oneflow.clamp()
.

clamp_
(min=None, max=None)¶ Inplace version of
oneflow.Tensor.clamp()
.

clip
(min=None, max=None)¶ Alias for
oneflow.Tensor.clamp()
.

clip_
(min=None, max=None)¶ Alias for
oneflow.Tensor.clamp_()
.

clone
()¶

copy_
(other: Union[oneflow.Tensor, numpy.ndarray])¶ The interface is consistent with PyTorch.
Tensor.copy_(src, non_blocking=False) → Tensor
Copies the elements from src into self tensor and returns self.
The src tensor must be broadcastable with the self tensor. It may be of a different data type or reside on a different device.
 Parameters
src (Tensor) – the source tensor to copy from
non_blocking (bool) – if True and this copy is between CPU and GPU, the copy may occur asynchronously with respect to the host. For other cases, this argument has no effect.

cos
()¶ See
oneflow.cos()

cosh
()¶ See
oneflow.cosh()

cpu
()¶ Returns a copy of this object in CPU memory. If this object is already in CPU memory and on the correct device, then no copy is performed and the original object is returned.
For example:
>>> import oneflow as flow >>> input = flow.tensor([1, 2, 3, 4, 5], device=flow.device("cuda")) >>> output = input.cpu() >>> output.device device(type='cpu', index=0)

cuda
(device: Optional[Union[int, str, oneflow._oneflow_internal.device]] = None)¶ Returns a copy of this object in CUDA memory. If this object is already in CUDA memory and on the correct device, then no copy is performed and the original object is returned.
 Parameters
device (flow.device) – The destination GPU device. Defaults to the current CUDA device.
For example:
>>> import oneflow as flow >>> input = flow.Tensor([1, 2, 3, 4, 5]) >>> output = input.cuda() >>> output.device device(type='cuda', index=0)

data
¶

detach
()¶

device
¶ The documentation is referenced from: https://pytorch.org/docs/1.10/generated/torch.Tensor.device.html.
Is the
oneflow.device
where this Tensor is, which is invalid for global tensor.

diag
(diagonal=0)¶ See
oneflow.diag()

diagonal
(offset=0, dim1=0, dim2=1)¶

dim
()¶ Tensor.dim() → int
Returns the number of dimensions of self tensor.

div
(other)¶ See
oneflow.div()

div_
(value) → Tensor¶ Inplace version of
oneflow.Tensor.div()
.

double
()¶ Tensor.double() is equivalent to Tensor.to(flow.float64). See to().
 Parameters
input (Tensor) – the input tensor.
For example:
>>> import oneflow as flow >>> import numpy as np >>> input = flow.tensor(np.random.randn(1, 2, 3), dtype=flow.int) >>> input = input.double() >>> input.dtype oneflow.float64

dtype
¶

element_size
()¶ Tensor.element_size() → int
Returns the size in bytes of an individual element.

eq
(other)¶ See
oneflow.eq()

erf
() → Tensor¶ See
oneflow.erf()

erfc
() → Tensor¶ See
oneflow.erfc()

erfinv
()¶ See
oneflow.erfinv()

erfinv_
()¶ Inplace version of
oneflow.erfinv()

exp
()¶ See
oneflow.exp()

expand
() → Tensor¶ See
oneflow.expand()

expand_as
(other) → Tensor¶ Expand this tensor to the same size as
other
.self.expand_as(other)
is equivalent toself.expand(other.size())
.Please see
expand()
for more information aboutexpand
. Parameters
other (
oneflow.Tensor
) – The result tensor has the same size asother
.

expm1
()¶ See
oneflow.expm1()

fill_
(value)¶ Tensor.fill_(value) → Tensor
Fills self tensor with the specified value.

flip
(dims)¶ See
oneflow.flip()

float
()¶ Tensor.float() is equivalent to Tensor.to(flow.float32). See to().
 Parameters
input (Tensor) – the input tensor.
For example:
>>> import oneflow as flow >>> import numpy as np >>> input = flow.tensor(np.random.randn(1, 2, 3), dtype=flow.int) >>> input = input.float() >>> input.dtype oneflow.float32

floor
()¶ See
oneflow.floor()

floor_
()¶ Inplace version of
oneflow.floor()

fmod
(other) → Tensor¶ See
oneflow.fmod()

gather
(dim, index) → Tensor¶ See
oneflow.gather()

ge
(other)¶ See
oneflow.ge()

gelu
()¶ See
oneflow.gelu()

get_device
() > Device ordinal (Integer)¶ For CUDA tensors, this function returns the device ordinal of the GPU on which the tensor resides. For CPU tensors, an error is thrown.

global_to_global
(placement=None, sbp=None, *, grad_sbp=None, check_meta=False) → Tensor¶ Performs Tensor placement and/or sbp conversion.
Note
This tensor must be global tensor.
At least one of placement and sbp is required.
If placement and sbp are all the same as this tensor’s own placement and sbp, then returns this tensor own.
 Parameters
placement (flow.placement, optional) – the desired placement of returned global tensor. Default: None
sbp (flow.sbp.sbp or tuple of flow.sbp.sbp, optional) – the desired sbp of returned global tensor. Default: None
 Keyword Arguments
grad_sbp (flow.sbp.sbp or tuple of flow.sbp.sbp, optional) – manually specify the sbp of this tensor’s grad tensor in the backward pass. If None, the grad tensor sbp will be infered automatically. Default: None
check_meta (bool, optional) – indicates whether to check meta information. If set to True, check the consistency of the input meta information (placement and sbp) on each rank. Default: False
>>> # Run on 2 ranks respectively >>> import oneflow as flow >>> input = flow.tensor([0., 1.], dtype=flow.float32, placement=flow.placement("cpu", ranks=[0, 1]), sbp=[flow.sbp.broadcast]) >>> output = input.global_to_global(placement=flow.placement("cpu", ranks=[0, 1]), sbp=[flow.sbp.split(0)]) >>> print(output.size()) >>> print(output)
>>> # results on rank 0 oneflow.Size([2]) tensor([0., 1.], placement=oneflow.placement(type="cpu", ranks=[0, 1]), sbp=(oneflow.sbp.split(axis=0),), dtype=oneflow.float32)
>>> # results on rank 1 oneflow.Size([2]) tensor([0., 1.], placement=oneflow.placement(type="cpu", ranks=[0, 1]), sbp=(oneflow.sbp.split(axis=0),), dtype=oneflow.float32)

grad
¶ Return the gradient calculated by autograd functions. This property is None by default.

grad_fn
¶ Return the function that created this tensor if it’s
requires_grad
is True.

gt
(other)¶ See
oneflow.gt()

half
()¶ self.half() is equivalent to self.to(dtype=oneflow.float16).

int
()¶ Tensor.int() is equivalent to Tensor.to(flow.int32). See to().
 Parameters
input (Tensor) – the input tensor.
For example:
>>> import oneflow as flow >>> import numpy as np >>> input = flow.tensor(np.random.randn(1, 2, 3), dtype=flow.float32) >>> input = input.int() >>> input.dtype oneflow.int32

is_contiguous
() → bool¶ Returns True if self tensor is contiguous in memory.

is_cuda
¶ Is True if the Tensor is stored on the GPU, False otherwise.

is_floating_point
()¶

is_global
¶ Return whether this Tensor is a global tensor.

is_lazy
¶ Return whether this Tensor is a lazy tensor.

is_leaf
¶ Compatible with PyTorch.
All Tensors that have
requires_grad
which isFalse
will be leaf Tensors by convention.For Tensor that have
requires_grad
which isTrue
, they will be leaf Tensors if they were created by source operations.Only leaf Tensors will have their
grad
populated during a call tobackward()
. To getgrad
populated for nonleaf Tensors, you can useretain_grad()
.For example:
>>> import oneflow as flow >>> a = flow.rand(10, requires_grad=False) >>> a.is_leaf True >>> a = flow.rand(10, requires_grad=True) >>> a.is_leaf True >>> b = a.cuda() >>> b.is_leaf False >>> c = a + 2 >>> c.is_leaf False

item
()¶ Returns the value of this tensor as a standard Python number. This only works for tensors with one element. For other cases, see tolist().
This operation is not differentiable.
 Parameters
input (Tensor) – the input tensor.
For example:
>>> import oneflow as flow >>> x = flow.tensor([1.0]) >>> x.item() 1.0

le
(other)¶ See
oneflow.le()

local_to_global
(placement=None, sbp=None, *, check_meta=Ture) → Tensor¶ Creates a global tensor from a local tensor.
Note
This tensor must be local tensor.
Both placement and sbp are required.
The returned global tensor takes this tensor as its local component in the current rank.
There is no data communication usually, but when sbp is
oneflow.sbp.broadcast
, the data on rank 0 will be broadcast to other ranks. Parameters
placement (flow.placement, optional) – the desired placement of returned global tensor. Default: None
sbp (flow.sbp.sbp or tuple of flow.sbp.sbp, optional) – the desired sbp of returned global tensor. Default: None
 Keyword Arguments
check_meta (bool, optional) – indicates whether to check meta information when createing global tensor from local tensor. Only can be set to False when the shape and dtype of the input local tensor on each rank are the same. If set to False, the execution of local_to_global can be accelerated. Default: True
>>> # Run on 2 ranks respectively >>> import oneflow as flow >>> input = flow.tensor([0., 1.], dtype=flow.float32) >>> output = input.local_to_global(placement=flow.placement("cpu", ranks=[0, 1]), sbp=[flow.sbp.split(0)], check_meta=False) >>> print(output.size()) >>> print(output)
>>> # results on rank 0 oneflow.Size([4]) tensor([0., 1., 0., 1.], placement=oneflow.placement(type="cpu", ranks=[0, 1]), sbp=(oneflow.sbp.split(axis=0),), dtype=oneflow.float32)
>>> # results on rank 1 oneflow.Size([4]) tensor([0., 1., 0., 1.], placement=oneflow.placement(type="cpu", ranks=[0, 1]), sbp=(oneflow.sbp.split(axis=0),), dtype=oneflow.float32)

log
()¶ See
oneflow.log()

log1p
()¶ See
oneflow.log1p()

long
()¶ Tensor.long() is equivalent to Tensor.to(flow.int64). See to().
 Parameters
input (Tensor) – the input tensor.
For example:
>>> import oneflow as flow >>> import numpy as np >>> input = flow.tensor(np.random.randn(1, 2, 3), dtype=flow.float32) >>> input = input.long() >>> input.dtype oneflow.int64

lt
(other)¶ See
oneflow.lt()

masked_fill
(mask, fill_value)¶

masked_select
(mask)¶

matmul
(other)¶ See
oneflow.matmul()

max
(dim, index) → Tensor¶ See
oneflow.max()

mean
(dim=None, keepdim=False) → Tensor¶ See
oneflow.mean()

min
(dim, index) → Tensor¶ See
oneflow.min()

mish
()¶ See
oneflow.mish()

mul
(value) → Tensor¶ See
oneflow.mul()

mul_
(value) → Tensor¶ Inplace version of
oneflow.Tensor.mul()
.

narrow
(dimension, start, length)¶ See
oneflow.narrow()

property
ndim
¶

ndimension
()¶ Tensor.dim() → int
Returns the number of dimensions of self tensor.

ne
(other)¶ See
oneflow.ne()

negative
()¶

nelement
()¶ Tensor.nelement() → int
Alias for numel()

new_empty
(*size, dtype=None, device=None, placement=None, sbp=None, requires_grad=False) → Tensor¶ Returns a Tensor of size
size
filled with uninitialized data. By default, the returned Tensor has the sameflow.dtype
andflow.device
as this tensor. Parameters
size (int...) – a list, tuple, or flow.Size of integers defining the shape of the output tensor.
dtype (flow.dtype, optional) – the desired type of returned tensor. Default: if None, same flow.dtype as this tensor.
device (flow.device, optional) – the desired device of returned tensor. Default: if None, same flow.device as this tensor.
placement (flow.placement, optional) – the desired placement of returned global tensor. Default: if None, the returned tensor is local one using the argument device.
sbp (flow.sbp.sbp or tuple of flow.sbp.sbp, optional) – the desired sbp descriptor of returned global tensor. Default: if None, the returned tensor is local one using the argument device.
requires_grad (bool, optional) – If autograd should record operations on the returned tensor. Default: False.
For example:
>>> import oneflow as flow >>> x = flow.ones(()) >>> y = x.new_empty((2, 2)) >>> y.shape oneflow.Size([2, 2])

new_zeros
(size=None, dtype=None, device=None, placement=None, sbp=None, requires_grad=False) → Tensor¶ Returns a Tensor of size size filled with 0. By default, the returned Tensor has the same torch.dtype and torch.device as this tensor.
 Parameters
size (int...) – a list, tuple, or flow.Size of integers defining the shape of the output tensor.
dtype (flow.dtype, optional) – the desired type of returned tensor. Default: if None, same flow.dtype as this tensor.
device (flow.device, optional) – the desired device of returned tensor. Default: if None, same flow.device as this tensor.
placement (flow.placement, optional) – the desired placement of returned global tensor. Default: if None, the returned tensor is local one using the argument device.
sbp (flow.sbp.sbp or tuple of flow.sbp.sbp, optional) – the desired sbp descriptor of returned global tensor. Default: if None, the returned tensor is local one using the argument device.
requires_grad (bool, optional) – If autograd should record operations on the returned tensor. Default: False.
For example:
>>> import numpy as np >>> import oneflow as flow >>> x = flow.Tensor(np.ones((1, 2, 3))) >>> y = x.new_zeros((2, 2)) >>> y tensor([[0., 0.], [0., 0.]], dtype=oneflow.float32)

nms
(scores, iou_threshold: float)¶ See
oneflow.nms()

norm
(p=None, dim=None, keepdim=False, dtype=None)¶

normal_
(mean=0, std=1, *, generator=None) → Tensor¶ Fills
self
tensor with elements samples from the normal distribution parameterized bymean
andstd
.

numel
()¶ See
oneflow.numel()

numpy
()¶ Tensor.numpy() → numpy.ndarray
 Returns self tensor as a NumPy ndarray. This tensor and the returned ndarray share the same underlying storage. Changes to
self tensor will be reflected in the ndarray and vice versa.

permute
(*dims)¶

placement
¶ Is the
oneflow.placement
where this Tensor is, which is invalid for local tensor.

pow
(b)¶ See
oneflow.pow()

prod
(dim=None, keepdim=False) → Tensor¶ See
oneflow.prod()

reciprocal
()¶

register_hook
(hook)¶ Registers a backward hook.
The hook will be called every time a gradient with respect to the Tensor is computed. The hook should have the following signature:
hook(grad) > Tensor or None
The hook should not modify its argument, but it can optionally return a new gradient which will be used in place of
grad
.For example:
>>> import oneflow as flow >>> x = flow.ones(5, requires_grad=True) >>> def hook(grad): ... return grad * 2 >>> x.register_hook(hook) >>> y = x * 2 >>> y.sum().backward() >>> x.grad tensor([4., 4., 4., 4., 4.], dtype=oneflow.float32)

relu
()¶ See
oneflow.relu()

repeat
(*size) → Tensor¶ See
oneflow.repeat()

requires_grad
¶ Compatible with PyTorch.
Is
True
if gradient need to be computed for this Tensor,False
otherwise.

requires_grad_
(requires_grad=True) → Tensor¶ Compatible with PyTorch.
 Parameters
requires_grad (bool) – Change the requires_grad flag for this Tensor. Default is
True
.
For example:
>>> import oneflow as flow >>> a = flow.rand(10, requires_grad=False) >>> a.requires_grad False >>> a = a.requires_grad_(requires_grad=True) >>> a.requires_grad True

reshape
(*shape)¶

retain_grad
()¶ Compatible with PyTorch.
Enables this Tensor to have their
grad
populated duringbackward()
. This is a noop for leaf tensors.

roll
(shifts, dims=None)¶ See
oneflow.roll()

round
()¶ See
oneflow.round()

rsqrt
()¶ See
oneflow.rsqrt()

sbp
¶ Is the
oneflow.sbp
representing that how the data of the global tensor is distributed, which is invalid for local tensor.

selu
()¶ See
oneflow.selu()

shape
¶

sigmoid
()¶

sign
()¶ See
oneflow.sign()

silu
()¶ See
oneflow.silu()

sin
() → Tensor¶ See
oneflow.sin()

sin_
()¶ Inplace version of
oneflow.sin()

sinh
()¶ See
oneflow.sinh()

size
(idx=None)¶ The interface is consistent with PyTorch.
Returns the size of the self tensor. If dim is not specified, the returned value is a oneflow.Size, a subclass of tuple. If dim is specified, returns an int holding the size of that dimension.
 Parameters
idx (int, optional) – The dimension for which to retrieve the size.

softmax
(dim=None)¶

softplus
()¶

softsign
()¶

sort
(dim: int =  1, descending: bool = False)¶ See
oneflow.sort()

split
(split_size_or_sections=None, dim=0)¶ See
oneflow.split()

sqrt
()¶ See
oneflow.sqrt()

square
()¶ See
oneflow.square()

squeeze
(dim=None)¶

std
(dim=None, unbiased=True, keepdim=False)¶ See
oneflow.std()

storage_offset
() → Tensor¶ Returns self tensor’s offset in the underlying storage in terms of number of storage elements (not bytes).
Example:
>>> import oneflow as flow >>> x = flow.tensor([1, 2, 3, 4, 5]) >>> x.storage_offset() 0

stride
()¶

sub
(other)¶ See
oneflow.sub()

sub_
(value) → Tensor¶ Inplace version of
oneflow.Tensor.sub()
.

sum
(dim=None, keepdim=False) → Tensor¶ See
oneflow.sum()

swapaxes
(dim0, dim1)¶

swapdims
(dim0, dim1)¶

t
()¶ Tensor.t() → Tensor
See
oneflow.t()

tan
()¶ See
oneflow.tan()

tanh
()¶ See
oneflow.tanh()

tile
(*dims) → Tensor¶ See
oneflow.tile()

to
(*args, **kwargs)¶  Performs Tensor dtype and/or device conversion.
A flow.dtype and flow.device are inferred from the arguments of input.to(*args, **kwargs).
Note
If the
input
Tensor already has the correctflow.dtype
andflow.device
, theninput
is returned. Otherwise, the returned tensor is a copy ofinput
with the desired. Parameters
input (oneflow.Tensor) – An input tensor.
*args (oneflow.Tensor or oneflow.device or oneflow.dtype) – Positional arguments
**kwargs (oneflow.device or oneflow.dtype) – Keyvalue arguments
 Returns
A Tensor.
 Return type
For example:
>>> import numpy as np >>> import oneflow as flow >>> arr = np.random.randint(1, 9, size=(1, 2, 3, 4)) >>> input = flow.Tensor(arr) >>> output = input.to(dtype=flow.float32) >>> np.array_equal(arr.astype(np.float32), output.numpy()) True

to_consistent
(*args, **kwargs)¶ This interface is no longer available, please use
oneflow.Tensor.to_global()
instead.

to_global
(placement=None, sbp=None, **kwargs) → Tensor¶ Creates a global tensor if this tensor is a local tensor, otherwise performs Tensor placement and/or sbp conversion.
Note
This tensor can be local tensor or global tensor.
For local tensor
Both placement and sbp are required.
The returned global tensor takes this tensor as its local component in the current rank.
There is no data communication usually, but when sbp is
oneflow.sbp.broadcast
, the data on rank 0 will be broadcast to other ranks.For global tensor
At least one of placement and sbp is required.
If placement and sbp are all the same as this tensor’s own placement and sbp, then returns this tensor own.
 Parameters
placement (flow.placement, optional) – the desired placement of returned global tensor. Default: None
sbp (flow.sbp.sbp or tuple of flow.sbp.sbp, optional) – the desired sbp of returned global tensor. Default: None
 Keyword Arguments
grad_sbp (flow.sbp.sbp or tuple of flow.sbp.sbp, optional) – manually specify the sbp of this tensor’s grad tensor in the backward pass. If None, the grad tensor sbp will be infered automatically. It is only used if this tensor is a global tensor. Default: None
check_meta (bool, optional) – indicates whether to check meta information. If set to True, check the input meta information on each rank. Default: True if this tensor is a local tensor, False if this tensor is a global tensor
For local tensor:
>>> # Run on 2 ranks respectively >>> import oneflow as flow >>> input = flow.tensor([0., 1.], dtype=flow.float32) >>> output = input.to_global(placement=flow.placement("cpu", ranks=[0, 1]), sbp=[flow.sbp.split(0)], check_meta=False) >>> print(output.size()) >>> print(output)
>>> # results on rank 0 oneflow.Size([4]) tensor([0., 1., 0., 1.], placement=oneflow.placement(type="cpu", ranks=[0, 1]), sbp=(oneflow.sbp.split(axis=0),), dtype=oneflow.float32)
>>> # results on rank 1 oneflow.Size([4]) tensor([0., 1., 0., 1.], placement=oneflow.placement(type="cpu", ranks=[0, 1]), sbp=(oneflow.sbp.split(axis=0),), dtype=oneflow.float32)
For global tensor:
>>> # Run on 2 ranks respectively >>> import oneflow as flow >>> input = flow.tensor([0., 1.], dtype=flow.float32, placement=flow.placement("cpu", ranks=[0, 1]), sbp=[flow.sbp.broadcast]) >>> output = input.to_global(placement=flow.placement("cpu", ranks=[0, 1]), sbp=[flow.sbp.split(0)]) >>> print(output.size()) >>> print(output)
>>> # results on rank 0 oneflow.Size([2]) tensor([0., 1.], placement=oneflow.placement(type="cpu", ranks=[0, 1]), sbp=(oneflow.sbp.split(axis=0),), dtype=oneflow.float32)
>>> # results on rank 1 oneflow.Size([2]) tensor([0., 1.], placement=oneflow.placement(type="cpu", ranks=[0, 1]), sbp=(oneflow.sbp.split(axis=0),), dtype=oneflow.float32)

to_local
() → Tensor¶ Returns the local component of this global tensor in the current rank.
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.
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)

tolist
()¶ Returns the tensor as a (nested) list. For scalars, a standard Python number is returned, just like with item(). Tensors are automatically moved to the CPU first if necessary.
This operation is not differentiable.
 Parameters
input (Tensor) – the input tensor.
For example:
>>> import oneflow as flow >>> input = flow.tensor([[1,2,3], [4,5,6]]) >>> input.tolist() [[1, 2, 3], [4, 5, 6]]

topk
(k, dim: Optional[int] = None, largest: bool = True, sorted: bool = True)¶ See
oneflow.topk()

transpose
(dim0, dim1)¶

tril
(diagonal=0)¶ See
oneflow.tril()

triu
(diagonal=0)¶ See
oneflow.triu()

type_as
(target)¶  Returns this tensor cast to the type of the given tensor.
This is a noop if the tensor is already of the correct type.
 Parameters
For example:
>>> import oneflow as flow >>> import numpy as np >>> input = flow.tensor(np.random.randn(1, 2, 3), dtype=flow.float32) >>> target = flow.tensor(np.random.randn(4, 5, 6), dtype = flow.int32) >>> input = input.type_as(target) >>> input.dtype oneflow.int32

unbind
(dim=0)¶ See
oneflow.unbind()

unfold
(dimension, size, step)¶ The interface is consistent with PyTorch. The documentation is referenced from: https://pytorch.org/docs/1.10/generated/torch.Tensor.unfold.html.
Returns a view of the original tensor which contains all slices of size size from self tensor in the dimension dimension.
Step between two slices is given by step.
If sizedim is the size of dimension dimension for self, the size of dimension dimension in the returned tensor will be (sizedim  size) / step + 1.
An additional dimension of size size is appended in the returned tensor.
 Parameters
dimension (int) – dimension in which unfolding happens
size (int) – the size of each slice that is unfolded
step (int) – the step between each slice
For example:
>>> import numpy as np >>> import oneflow as flow >>> x = flow.arange(1., 8) >>> x tensor([1, 2, 3, 4, 5, 6, 7], dtype=oneflow.int64) >>> x.unfold(0, 2, 1) tensor([[1, 2], [2, 3], [3, 4], [4, 5], [5, 6], [6, 7]], dtype=oneflow.int64) >>> x.unfold(0, 2, 2) tensor([[1, 2], [3, 4], [5, 6]], dtype=oneflow.int64)

uniform_
(a=0, b=1)¶ Tensor.uniform_(from=0, to=1) → Tensor
Fills self tensor with numbers sampled from the continuous uniform distribution:
\[P(x)=1/(tofrom)\]

unsqueeze
(dim)¶

var
(dim=None, unbiased=True, keepdim=False)¶ See
oneflow.var()

view
(*shape)¶ The interface is consistent with PyTorch. The documentation is referenced from: https://pytorch.org/docs/1.10/generated/torch.Tensor.view.html.
Returns a new tensor with the same data as the
self
tensor but of a differentshape
.The returned tensor shares the same data and must have the same number of elements, but may have a different size. For a tensor to be viewed, the new view size must be compatible with its original size and stride, i.e., each new view dimension must either be a subspace of an original dimension, or only span across original dimensions \(d, d+1, \dots, d+k\) that satisfy the following contiguitylike condition that \(\forall i = d, \dots, d+k1\),
\[\text{stride}[i] = \text{stride}[i+1] \times \text{size}[i+1]\]Otherwise, it will not be possible to view
self
tensor asshape
without copying it (e.g., viacontiguous()
). When it is unclear whether aview()
can be performed, it is advisable to usereshape()
, which returns a view if the shapes are compatible, and copies (equivalent to callingcontiguous()
) otherwise. Parameters
input – A Tensor.
*shape – flow.Size or int…
 Returns
A Tensor has the same type as input.
For example:
>>> import numpy as np >>> import oneflow as flow >>> x = np.array( ... [[1, 2, 3, 4], [5, 6, 7, 8], [9, 10, 11, 12], [13, 14, 15, 16]] ... ).astype(np.float32) >>> input = flow.Tensor(x) >>> y = input.view(2, 2, 2, 1).numpy().shape >>> y (2, 2, 2, 2)

view_as
(other) → Tensor¶ Expand this tensor to the same size as
other
.self.view_as(other)
is equivalent toself.view(other.size())
.Please see
view()
for more information aboutview
. Parameters
other (
oneflow.Tensor
) – The result tensor has the same size asother
.

where
(x=None, y=None)¶ See
oneflow.where()

property