Python: torch.Tensor

Members

Use dir() and list comprehension to find the members of the Tensor class of torch that don't start with two underscores (i.e. exclude dunders) and don't end in an underscore:

for member in sorted([ _ for _ in dir(torch.Tensor) if  len(_) > 1 and _[1] != '_'  and _[-1] != '_'], key = lambda s: s.replace('_', '')):
    print(member)

`abs`
`absolute`
`acos`
`acosh`
`add`
`addbmm`
`addcdiv`
`addcmul`
`addmm`
`_addmm_activation`
`addmv`
`addr`
`adjoint`
`align_as`
`align_to`
`all`	Tests if all elements in the tensor are `True` (and returns the result as a tensor with one boolean element?). Compare with `any()`
`allclose`
`amax`
`amin`
`aminmax`
`angle`
`any`	Tests if at least one element in the tensor is `True`. Compare with `all()`.
`arccos`
`arccosh`
`arcsin`
`arcsinh`
`arctan`
`arctan2`
`arctanh`
`argmax`
`argmin`
`argsort`	Returns the second value of the object returned by `sort()` (i.e. the indices)
`argwhere`
`asin`
`asinh`
`as_strided`
`as_strided_scatter`
`as_subclass`
`atan`
`atan2`
`atanh`
`_autocast_to_full_precision`
`_autocast_to_reduced_precision`
`backward`	Computes the gradients of the current tensor (see also `grad` and `requires_grad`).
`_backward_hooks`
`baddbmm`
`_base`
`bernoulli`
`bfloat16`
`bincount`
`bitwise_and`
`bitwise_left_shift`
`bitwise_not`
`bitwise_or`
`bitwise_right_shift`
`bitwise_xor`
`bmm`
`bool`
`broadcast_to`
`byte`
`ccol_indices`
`_cdata`
`cdouble`
`ceil`
`cfloat`
`chalf`
`char`
`cholesky`	Deprecated in favor of `torch.linalg.cholesky()` (will be removed)
`cholesky_inverse`
`cholesky_solve`
`chunk`
`clamp`
`clamp_max`
`clamp_min`
`clip`
`clone`
`coalesce`
`col_indices`
`_conj`
`conj`
`_conj_physical`
`conj_physical`
`contiguous`
`copysign`
`corrcoef`
`cos`
`cosh`
`count_nonzero`
`cov`
`cpu`	Returns a copy of this object in CPU memory. Compare with `cuda`
`cross`
`crow_indices`
`cuda`	Returns a copy of this object in CUDA memory. Compare with `cpu`
`cummax`
`cummin`
`cumprod`
`cumsum`
`data`	deprecated
`data_ptr`	The address of the first element.
`deg2rad`
`dense_dim`
`dequantize`
`det`
`detach()`	Creates a tensor with the same data but is not connected to the current (autograd) graph. Compare with `numpy()`, `item()` and `tolist()`
`device`	In instance of `torch.device` which represents the device the tensor is allocated.
`diag`
`diag_embed`
`diagflat`
`diagonal`
`diagonal_scatter`
`diff`
`digamma`
`dim`	The number of dimensions of the tensor. An alias is `ndim`
`_dimI`
`_dimV`
`dist`
`div`
`divide`
`dot`
`double`
`dsplit`
`dtype`
`eig`
`element_size`
`eq`
`equal`
`erf`
`erfc`
`erfinv`
`exp`
`exp2`
`expand()`
`expand_as`
`expm1`
`fix`
`_fix_weakref`
`flatten`
`flip`
`fliplr`
`flipud`
`float`
`float_power`
`floor`
`floor_divide`
`fmax`
`fmin`
`fmod`
`frac`
`frexp`
`gather`
`gcd`
`ge`
`geqrf`
`ger`
`get_device`
`_grad`
`grad`	This attribute becomes a tensor when `backward()` is called to compute the gradients (for self). See also `requires_grad` and the function `torch.autograd.grad()`
`_grad_fn`
`grad_fn`
`greater`
`greater_equal`
`gt`
`half`
`hardshrink`
`has_names`
`_has_symbolic_sizes_strides`
`heaviside`
`histc`
`histogram`
`hsplit`
`hypot`
`i0`
`igamma`
`igammac`
`imag`
`index_add`
`index_copy`
`index_fill`
`index_put`
`index_reduce`
`index_select`
`_indices`
`indices`
`inner`
`int`
`int_repr`
`inverse`
`ipu`
`isclose`
`is_coalesced`
`is_complex`
`is_conj`
`is_contiguous`
`is_cpu`
`is_cuda`
`is_distributed`
`isfinite`
`is_floating_point`
`isinf`
`is_inference`
`is_ipu`
`is_leaf`	`False`, by default, for all tensors whose `requires_grad` value is `False`.
`is_meta`
`is_mkldnn`
`is_mps`
`isnan`
`is_neg`
`isneginf`
`is_nested`
`is_nonzero`
`is_ort`
`is_pinned`
`isposinf`
`is_quantized`
`isreal`
`is_same_size`
`is_set_to`
`is_shared`
`is_signed`
`is_sparse`
`is_sparse_csr`
`istft`
`_is_view`
`is_vulkan`
`is_xpu`
`_is_zerotensor`
`item()`	Compare with `tolist()`, `detach()` and `numpy()`
`kron`
`kthvalue`
`layout`
`lcm`
`ldexp`
`le`
`lerp`
`less`
`less_equal`
`lgamma`
`log`
`log10`
`log1p`
`log2`
`logaddexp`
`logaddexp2`
`logcumsumexp`
`logdet`
`logical_and`
`logical_not`
`logical_or`
`logical_xor`
`logit`
`log_softmax`
`logsumexp`
`long`
`lstsq`
`lt`
`lu`
`lu_solve`
`mH`
`mT`
`_make_subclass`
`_make_wrapper_subclass`
`masked_fill`
`masked_scatter`
`masked_select`
`matmul`
`matrix_exp`
`matrix_power`
`max`
`maximum`
`mean`
`median`
`min`
`minimum`
`mm`
`mode`
`moveaxis`
`movedim`
`msort`	See also `sort()`
`mul`
`multinomial`
`multiply`
`mv`
`mvlgamma`
`name`
`names`
`nanmean`
`nanmedian`
`nanquantile`
`nansum`
`nan_to_num`
`narrow`
`narrow_copy`
`ndim`
`ndimension`
`ne`
`neg`
`negative`
`_neg_view`
`nelement`
`_nested_tensor_layer_norm`
`_nested_tensor_size`
`new`
`new_empty`
`new_empty_strided`
`new_full`
`new_ones`
`new_tensor`
`new_zeros`
`nextafter`
`_nnz`
`nonzero`
`norm`
`not_equal`
`numel`
`numpy()`	Compare with `torch.from_numpy()`
`orgqr`
`ormqr`
`outer`
`output_nr`
`permute`
`pin_memory`
`pinverse`
`polygamma`
`positive`
`pow`
`prelu`
`prod`
`put`
`_python_dispatch`
`qr`
`qscheme`
`quantile`
`rad2deg`
`ravel`
`real`
`reciprocal`
`record_stream`
`_reduce_ex_internal`
`refine_names`
`register_hook`
`reinforce`
`relu`
`remainder`
`rename`
`renorm`
`repeat`
`repeat_interleave`
`requires_grad`	This tensor's gradient need to be computed if set to `True`. (See also `backward` and `grad`)
`reshape`
`reshape_as`
`resize`
`resize_as`
`resolve_conj`
`resolve_neg`
`retain_grad`
`retains_grad`
`roll`
`rot90`
`round`
`row_indices`
`rsqrt`
`scatter`
`scatter_add`
`scatter_reduce`
`select`
`select_scatter`
`sgn`
`shape`
`short`
`sigmoid`
`sign`
`signbit`
`sin`
`sinc`
`sinh`
`size()`	Returns the size of the tensor as `torch.Size` or `int`
`slice_scatter`
`slogdet`
`smm`
`softmax`
`solve`
`sort`	See also `argsort()` and `msort()`
`sparse_dim`
`sparse_mask`
`split`
`split_with_sizes`
`sqrt`
`square`
`squeeze`	Remove the indicated dimensions whose size is equal to 1. Compare with `unsqueeze()`
`sspaddmm`
`std`
`stft`
`_storage`
`storage`
`storage_offset`
`storage_type`
`stride`
`sub`
`subtract`
`sum`
`sum_to_size`
`svd`
`swapaxes`
`swapdims`
`symeig`
`take`
`take_along_dim`
`tan`
`tanh`
`tensor_split`
`tile`
`to`	Perform dtype and/or device conversion.
`_to_dense`
`to_dense`
`tolist`	Returns the tensor's items as a (potentially nested) list. Compare with `item()`
`to_mkldnn`
`to_padded_tensor`
`topk`
`to_sparse`
`to_sparse_bsc`
`to_sparse_bsr`
`to_sparse_coo`
`to_sparse_csc`
`to_sparse_csr`
`trace`
`transpose`
`triangular_solve`
`tril`
`triu`
`true_divide`
`trunc`
`type`	Returns the type if dtype is not provided, or cast the object to the given type.
`type_as`
`unbind()`	?The «opposite» of `pack()`? Creates a list of tensors along the first dimension.
`unflatten`
`unfold`
`unique`
`unique_consecutive`
`unsafe_chunk`
`unsafe_split`
`unsafe_split_with_sizes`
`unsqueeze`	Compare with `squeeze()`
`_update_names`
`_values`
`values`
`var`
`vdot`
`_version`
`view()`	Reshape a tensor without copying memory (similar to numpy's `reshape()`)
`view_as`
`volatile`
`vsplit`
`where`
`xlogy`
`xpu`
`zero_` (?)	Fills tensor with zeros.

Python: torch.Tensor

Members

See also