tensorplay.nn.attention.bias API

Functions 2

causal_lower_right

tensorplay.nn.attention.bias.causal_lower_right(*size) → CausalBias[source]

Creates a lower-right triangular causal bias.

This function generates a lower-right triangular matrix to represent causal attention bias with a diagonal offset set so that the inclusive values are aligned to the lower right corner of the matrix.

The equivalent code for constructing this bias is:

diagonal_offset = size[1] - size[0]
tensorplay.tril(
    tensorplay.ones(size, dtype=tensorplay.bool),
    diagonal=diagonal_offset,
)

For instance, with shape=(3,4), the materialized bias tensor will be:

[[1, 1, 0, 0],
 [1, 1, 1, 0],
 [1, 1, 1, 1]]

Parameters:: size – The size of the bias matrix.
Returns:: The LOWER_RIGHT triangular causal bias variant.
Return type:: CausalBias

causal_upper_left

functionFull reference ↗

tensorplay.nn.attention.bias.causal_upper_left(*size) → CausalBias[source]

Creates an upper-left triangular causal bias.

This function generates an upper-left triangular matrix to represent causal attention bias with a diagonal offset set so that the inclusive values are aligned to the upper left corner of the matrix. This equivalent to the is_causal=True argument in scaled_dot_product_attention.

The equivalent code for constructing this bias is:

tensorplay.tril(tensorplay.ones(size, dtype=tensorplay.bool))

For instance, with shape=(3,4), the materialized bias tensor will be:

[[1, 0, 0, 0],
 [1, 1, 0, 0],
 [1, 1, 1, 0]]

Parameters:: size – The size of the bias matrix.
Returns:: The UPPER_LEFT triangular causal bias variant.
Return type:: CausalBias

Classes 2

CausalBias

classFull reference ↗

class tensorplay.nn.attention.bias.CausalBias(variant: CausalVariant, seq_len_q: int, seq_len_kv: int)[source]

A bias representing causal attention patterns. For an overview of the bias structure, see the CausalVariant enum.

This class is used for defining causal (triangular) attention biases. For constructing the bias, there exist two factory functions: causal_upper_left() and causal_lower_right().

Example:

from tensorplay.nn.attention.bias import causal_lower_right

bsz, num_heads, seqlen_q, seqlen_kv, head_dim = 32, 8, 4, 12, 8

# Create a lower-right causal bias
attn_bias = causal_lower_right(seqlen_q, seqlen_kv)

q = tensorplay.randn(
    bsz, num_heads, seqlen_q, head_dim, device="cuda", dtype=tensorplay.float16
)
k = tensorplay.randn(
    bsz, num_heads, seqlen_kv, head_dim, device="cuda", dtype=tensorplay.float16
)
v = tensorplay.randn(
    bsz, num_heads, seqlen_kv, head_dim, device="cuda", dtype=tensorplay.float16
)

out = F.scaled_dot_product_attention(q, k, v, attn_bias)

Warning

This class is a prototype and subject to change.

abs(): abs(Tensor self) -> Tensor

abs_(): abs_(Tensor(a!) self) -> Tensor(a!)

absolute(): absolute(Tensor self) -> Tensor

absolute_(): absolute_(Tensor(a!) self) -> Tensor(a!)

acos(): acos(Tensor self) -> Tensor

acos_(): acos_(Tensor(a!) self) -> Tensor(a!)

acosh(): acosh(Tensor self) -> Tensor

acosh_(): acosh_(Tensor(a!) self) -> Tensor(a!)

add(): add.Tensor(Tensor self, Tensor other, *, Scalar alpha=1) -> Tensor add.Scalar(Tensor self, Scalar other, Scalar alpha=1) -> Tensor

add_(other, *, alpha=1) → Tensor: In-place version of tensorplay.Tensor.add()

addbmm(): addbmm(Tensor self, Tensor batch1, Tensor batch2, *, Scalar beta=1, Scalar alpha=1) -> Tensor

addbmm_(): addbmm_(Tensor(a!) self, Tensor batch1, Tensor batch2, *, Scalar beta=1, Scalar alpha=1) -> Tensor(a!)

addcdiv(): addcdiv(Tensor self, Tensor tensor1, Tensor tensor2, *, Scalar value=1) -> Tensor

addcdiv_(): addcdiv_(Tensor(a!) self, Tensor tensor1, Tensor tensor2, *, Scalar value=1) -> Tensor(a!)

addcmul(): addcmul(Tensor self, Tensor tensor1, Tensor tensor2, *, Scalar value=1) -> Tensor

addcmul_(): addcmul_(Tensor(a!) self, Tensor tensor1, Tensor tensor2, *, Scalar value=1) -> Tensor(a!)

addmm(): addmm(Tensor self, Tensor mat1, Tensor mat2, *, Scalar beta=1, Scalar alpha=1) -> Tensor

addmm_(): addmm_(Tensor(a!) self, Tensor mat1, Tensor mat2, *, Scalar beta=1, Scalar alpha=1) -> Tensor(a!)

addmv(): addmv(Tensor self, Tensor mat, Tensor vec, *, Scalar beta=1, Scalar alpha=1) -> Tensor

addmv_(): addmv_(Tensor(a!) self, Tensor mat, Tensor vec, *, Scalar beta=1, Scalar alpha=1) -> Tensor(a!)

addr(): addr(Tensor self, Tensor vec1, Tensor vec2, *, Scalar beta=1, Scalar alpha=1) -> Tensor

addr_(): addr_(Tensor(a!) self, Tensor vec1, Tensor vec2, *, Scalar beta=1, Scalar alpha=1) -> Tensor(a!)

adjoint(): adjoint(Tensor(a) self) -> Tensor(a)

airy_ai(): airy_ai(Tensor self) -> Tensor

alias(): alias(Tensor(a) self) -> Tensor(a)

all(): all(Tensor self) -> Tensor all.dim(Tensor self, int dim, bool keepdim=False) -> Tensor all.dims(Tensor self, int[]? dim=None, bool keepdim=False) -> Tensor

allclose(): allclose(Tensor self, Tensor other, float rtol=1e-05, float atol=1e-08, bool equal_nan=False) -> bool

alpha_dropout_(): alpha_dropout_(Tensor(a!) self, float p=0.5, bool train=True) -> Tensor(a!)

aminmax(): aminmax(Tensor self, int[] dim=[], bool keepdim=False) -> (Tensor min, Tensor max)

angle(): angle(Tensor self) -> Tensor

any(): any(Tensor self) -> Tensor any.dim(Tensor self, int dim, bool keepdim=False) -> Tensor any.dims(Tensor self, int[]? dim=None, bool keepdim=False) -> Tensor

apply_(self: object, callable: object) → object

arccos(): arccos(Tensor self) -> Tensor

arccos_(): arccos_(Tensor(a!) self) -> Tensor(a!)

arccosh(): arccosh(Tensor self) -> Tensor

arccosh_(): arccosh_(Tensor(a!) self) -> Tensor(a!)

arcsin(): arcsin(Tensor self) -> Tensor

arcsin_(): arcsin_(Tensor(a!) self) -> Tensor(a!)

arcsinh(): arcsinh(Tensor self) -> Tensor

arcsinh_(): arcsinh_(Tensor(a!) self) -> Tensor(a!)

arctan(): arctan(Tensor self) -> Tensor

arctan2(): arctan2(Tensor self, Tensor other) -> Tensor

arctan2_(): arctan2_(Tensor(a!) self, Tensor other) -> Tensor(a!)

arctan_(): arctan_(Tensor(a!) self) -> Tensor(a!)

arctanh(): arctanh(Tensor self) -> Tensor

arctanh_(): arctanh_(Tensor(a!) self) -> Tensor(a!)

argmax(): argmax(Tensor self, int? dim=None, bool keepdim=False) -> Tensor

argmin(): argmin(Tensor self, int? dim=None, bool keepdim=False) -> Tensor

argsort(): argsort(Tensor self, int dim=-1, bool descending=False) -> Tensor argsort.stable(Tensor self, *, bool stable, int dim=-1, bool descending=False) -> Tensor

argwhere(): argwhere(Tensor self) -> Tensor

as_strided_(): as_strided_(Tensor(a!) self, SymInt[] size, SymInt[] stride, SymInt? storage_offset=None) -> Tensor(a!)

as_strided_scatter(): as_strided_scatter(Tensor self, Tensor src, SymInt[] size, SymInt[] stride, SymInt? storage_offset=None) -> Tensor

as_subclass(self: object, cls: object) → object

asin(): asin(Tensor self) -> Tensor

asin_(): asin_(Tensor(a!) self) -> Tensor(a!)

asinh(): asinh(Tensor self) -> Tensor

asinh_(): asinh_(Tensor(a!) self) -> Tensor(a!)

atan(): atan(Tensor self) -> Tensor

atan2(): atan2(Tensor self, Tensor other) -> Tensor

atan2_(): atan2_(Tensor(a!) self, Tensor other) -> Tensor(a!)

atan_(): atan_(Tensor(a!) self) -> Tensor(a!)

atanh(): atanh(Tensor self) -> Tensor

atanh_(): atanh_(Tensor(a!) self) -> Tensor(a!)

backward(self: tensorplay._C.TensorBase, gradient: tensorplay._C.TensorBase | None = None, retain_graph: bool | None = None, create_graph: bool = False, inputs: object = None) → None

baddbmm(): baddbmm(Tensor self, Tensor batch1, Tensor batch2, *, Scalar beta=1, Scalar alpha=1) -> Tensor

baddbmm_(): baddbmm_(Tensor(a!) self, Tensor batch1, Tensor batch2, *, Scalar beta=1, Scalar alpha=1) -> Tensor(a!)

bernoulli(): bernoulli(Tensor self, *, Generator? generator=None) -> Tensor bernoulli.p(Tensor self, float p, *, Generator? generator=None) -> Tensor

bernoulli_(p=0.5, *, generator=None) → Tensor: Fills each element of this tensor with an independent sample from the Bernoulli distribution with success probability p, which may be a scalar or a tensor of per-element probabilities. The tensor may hold an integral dtype; each sample is written as 0 or 1.

bessel_j0(): bessel_j0(Tensor self) -> Tensor

bessel_j1(): bessel_j1(Tensor self) -> Tensor

bessel_y0(): bessel_y0(Tensor self) -> Tensor

bessel_y1(): bessel_y1(Tensor self) -> Tensor

bincount(): bincount(Tensor self, Tensor? weights=None, SymInt minlength=0) -> Tensor

bitwise_and(): bitwise_and.Tensor(Tensor self, Tensor other) -> Tensor bitwise_and.Scalar(Tensor self, Scalar other) -> Tensor

bitwise_and_(): bitwise_and_.Tensor(Tensor(a!) self, Tensor other) -> Tensor(a!) bitwise_and_.Scalar(Tensor(a!) self, Scalar other) -> Tensor(a!)

bitwise_left_shift(): bitwise_left_shift.Tensor(Tensor self, Tensor other) -> Tensor bitwise_left_shift.Tensor_Scalar(Tensor self, Scalar other) -> Tensor

bitwise_left_shift_(): bitwise_left_shift_.Tensor(Tensor(a!) self, Tensor other) -> Tensor(a!) bitwise_left_shift_.Tensor_Scalar(Tensor(a!) self, Scalar other) -> Tensor(a!)

bitwise_not(): bitwise_not(Tensor self) -> Tensor

bitwise_not_(): bitwise_not_(Tensor(a!) self) -> Tensor(a!)

bitwise_or(): bitwise_or.Tensor(Tensor self, Tensor other) -> Tensor bitwise_or.Scalar(Tensor self, Scalar other) -> Tensor

bitwise_or_(): bitwise_or_.Tensor(Tensor(a!) self, Tensor other) -> Tensor(a!) bitwise_or_.Scalar(Tensor(a!) self, Scalar other) -> Tensor(a!)

bitwise_right_shift(): bitwise_right_shift.Tensor(Tensor self, Tensor other) -> Tensor bitwise_right_shift.Tensor_Scalar(Tensor self, Scalar other) -> Tensor

bitwise_right_shift_(): bitwise_right_shift_.Tensor(Tensor(a!) self, Tensor other) -> Tensor(a!) bitwise_right_shift_.Tensor_Scalar(Tensor(a!) self, Scalar other) -> Tensor(a!)

bitwise_xor(): bitwise_xor.Tensor(Tensor self, Tensor other) -> Tensor bitwise_xor.Scalar(Tensor self, Scalar other) -> Tensor

bitwise_xor_(): bitwise_xor_.Tensor(Tensor(a!) self, Tensor other) -> Tensor(a!) bitwise_xor_.Scalar(Tensor(a!) self, Scalar other) -> Tensor(a!)

bmm(): bmm(Tensor self, Tensor mat2) -> Tensor

broadcast_to(): broadcast_to(Tensor self, SymInt[] size) -> Tensor

bucketize(boundaries, *, out_int32=False, right=False, out=None): Bucket indices of self against a sorted 1-D boundaries tensor.

cauchy_(): cauchy_(Tensor(a!) self, float median=0.0, float sigma=1.0, *, Generator? generator=None) -> Tensor(a!)

ccol_indices(): ccol_indices(Tensor(a) self) -> Tensor(a)

ceil(): ceil(Tensor self) -> Tensor

ceil_(): ceil_(Tensor(a!) self) -> Tensor(a!)

celu(): celu(Tensor self, Scalar alpha=1.0) -> Tensor

celu_(): celu_(Tensor(a!) self, Scalar alpha=1.0) -> Tensor(a!)

chalf(): chalf(Tensor self, *, MemoryFormat? memory_format=None) -> Tensor

channel_shuffle(): channel_shuffle(Tensor self, SymInt groups) -> Tensor

cholesky(): cholesky(Tensor self, bool upper=False) -> Tensor

cholesky_inverse(): cholesky_inverse(Tensor self, bool upper=False) -> Tensor

cholesky_solve(): cholesky_solve(Tensor self, Tensor input2, bool upper=False) -> Tensor

chunk(): chunk(Tensor(a -> *) self, int chunks, int dim=0) -> Tensor(a)[]

clamp(): clamp.Tensor(Tensor self, Tensor? min=None, Tensor? max=None) -> Tensor clamp(Tensor self, Scalar? min=None, Scalar? max=None) -> Tensor

clamp_(): clamp_.Tensor(Tensor(a!) self, Tensor? min=None, Tensor? max=None) -> Tensor(a!) clamp_(Tensor(a!) self, Scalar? min=None, Scalar? max=None) -> Tensor(a!)

clamp_max(): clamp_max.Tensor(Tensor self, Tensor max) -> Tensor clamp_max(Tensor self, Scalar max) -> Tensor

clamp_max_(): clamp_max_.Tensor(Tensor(a!) self, Tensor max) -> Tensor(a!) clamp_max_(Tensor(a!) self, Scalar max) -> Tensor(a!)

clamp_min(): clamp_min.Tensor(Tensor self, Tensor min) -> Tensor clamp_min(Tensor self, Scalar min) -> Tensor

clamp_min_(): clamp_min_.Tensor(Tensor(a!) self, Tensor min) -> Tensor(a!) clamp_min_(Tensor(a!) self, Scalar min) -> Tensor(a!)

clip(): clip.Tensor(Tensor self, Tensor? min=None, Tensor? max=None) -> Tensor clip(Tensor self, Scalar? min=None, Scalar? max=None) -> Tensor

clip_(): clip_.Tensor(Tensor(a!) self, Tensor? min=None, Tensor? max=None) -> Tensor(a!) clip_(Tensor(a!) self, Scalar? min=None, Scalar? max=None) -> Tensor(a!)

clone(): clone(Tensor self, *, MemoryFormat? memory_format=None) -> Tensor

combinations(): combinations(Tensor self, int r=2, bool with_replacement=False) -> Tensor

conj(): conj(Tensor(a) self) -> Tensor(a)

conj_physical(): conj_physical(Tensor self) -> Tensor

conj_physical_(): conj_physical_(Tensor(a!) self) -> Tensor(a!)

contiguous(): contiguous(Tensor(a) self, *, MemoryFormat memory_format=Contiguous) -> Tensor(a)

copy_(): copy_(Tensor(a!) self, Tensor src, bool non_blocking=False) -> Tensor(a!)

copysign(): copysign.Tensor(Tensor self, Tensor other) -> Tensor copysign.Scalar(Tensor self, Scalar other) -> Tensor

copysign_(): copysign_.Tensor(Tensor(a!) self, Tensor other) -> Tensor(a!) copysign_.Scalar(Tensor(a!) self, Scalar other) -> Tensor(a!)

corrcoef(): corrcoef(Tensor self) -> Tensor

cos(): cos(Tensor self) -> Tensor

cos_(): cos_(Tensor(a!) self) -> Tensor(a!)

cosh(): cosh(Tensor self) -> Tensor

cosh_(): cosh_(Tensor(a!) self) -> Tensor(a!)

cot(): cot(Tensor self) -> Tensor

cov(): cov(Tensor self, *, int correction=1, Tensor? fweights=None, Tensor? aweights=None) -> Tensor

cpu(): Returns a copy of this object in CPU memory. If this object is already in CPU memory, then no copy is performed and the original object is returned.

cross(): cross(Tensor self, Tensor other, int? dim=None) -> Tensor

csc(): csc(Tensor self) -> Tensor

cuda(device=None, non_blocking=False): 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.

cummax(): cummax(Tensor self, int dim) -> (Tensor values, Tensor indices)

cummin(): cummin(Tensor self, int dim) -> (Tensor values, Tensor indices)

cumprod(): cumprod(Tensor self, int dim, ScalarType? dtype=None) -> Tensor

cumprod_(): cumprod_(Tensor(a!) self, int dim, ScalarType? dtype=None) -> Tensor(a!)

cumsum(): cumsum(Tensor self, int dim=0, ScalarType? dtype=None) -> Tensor

cumsum_(): cumsum_(Tensor(a!) self, int dim=0, ScalarType? dtype=None) -> Tensor(a!)

data_ptr(self: tensorplay._C.TensorBase) → int

defined(self: tensorplay._C.TensorBase) → bool

deg2rad(): deg2rad(Tensor self) -> Tensor

deg2rad_(): deg2rad_(Tensor(a!) self) -> Tensor(a!)

dequantize(): dequantize.self(Tensor self) -> Tensor

det(): det(Tensor self) -> Tensor

detach_copy(): detach_copy(Tensor self) -> Tensor

diag(): diag(Tensor self, int diagonal=0) -> Tensor

diag_embed(): diag_embed(Tensor self, int offset=0, int dim1=-2, int dim2=-1) -> Tensor

diagflat(): diagflat(Tensor self, int offset=0) -> Tensor

diagonal(): diagonal(Tensor(a) self, int offset=0, int dim1=0, int dim2=1) -> Tensor(a)

diagonal_scatter(): diagonal_scatter(Tensor self, Tensor src, int offset=0, int dim1=0, int dim2=1) -> Tensor

diff(): diff(Tensor self, int n=1, int dim=-1, Tensor? prepend=None, Tensor? append=None) -> Tensor

digamma(): digamma(Tensor self) -> Tensor

digamma_(): digamma_(Tensor(a!) self) -> Tensor(a!)

dim() → int

dim_order(self: tensorplay._C.TensorBase) → tuple

dist(): dist(Tensor self, Tensor other, Scalar p=2) -> Tensor

div(): div.Tensor(Tensor self, Tensor other) -> Tensor div.Tensor_mode(Tensor self, Tensor other, *, str? rounding_mode) -> Tensor div.Scalar_mode(Tensor self, Scalar other, *, str? rounding_mode) -> Tensor div.Scalar(Tensor self, Scalar other) -> Tensor

div_(value, *, rounding_mode=None) → Tensor: In-place version of tensorplay.Tensor.div()

divide(): divide.Tensor(Tensor self, Tensor other) -> Tensor divide.Tensor_mode(Tensor self, Tensor other, *, str? rounding_mode) -> Tensor divide.Scalar(Tensor self, Scalar other) -> Tensor divide.Scalar_mode(Tensor self, Scalar other, *, str? rounding_mode) -> Tensor

divide_(value, *, rounding_mode=None) → Tensor: In-place version of tensorplay.Tensor.divide()

dot(): dot(Tensor self, Tensor tensor) -> Tensor

dropout_(): dropout_(Tensor(a!) self, float p=0.5, bool train=True) -> Tensor(a!)

dsplit(split_size_or_sections) → List of Tensors

Splits a tensor of at least three dimensions into views along the third axis. split_size_or_sections is either the number of equal sections or the list of sizes of each section.

See tensorplay.dsplit()

element_size(self: tensorplay._C.TensorBase) → int

elu(): elu(Tensor self, Scalar alpha=1, Scalar scale=1, Scalar input_scale=1) -> Tensor

elu_(): elu_(Tensor(a!) self, Scalar alpha=1, Scalar scale=1, Scalar input_scale=1) -> Tensor(a!)

entr(): entr(Tensor self) -> Tensor

eq(): eq.Tensor(Tensor self, Tensor other) -> Tensor eq.Scalar(Tensor self, Scalar other) -> Tensor

eq_(): eq_.Tensor(Tensor(a!) self, Tensor other) -> Tensor(a!) eq_.Scalar(Tensor(a!) self, Scalar other) -> Tensor(a!)

equal(other) → bool: True if two tensors have the same size and elements, False otherwise.

erf(): erf(Tensor self) -> Tensor

erf_(): erf_(Tensor(a!) self) -> Tensor(a!)

erfc(): erfc(Tensor self) -> Tensor

erfc_(): erfc_(Tensor(a!) self) -> Tensor(a!)

erfcx(): erfcx(Tensor self) -> Tensor

erfinv(): erfinv(Tensor self) -> Tensor

erfinv_(): erfinv_(Tensor(a!) self) -> Tensor(a!)

exp(): exp(Tensor self) -> Tensor

exp2(): exp2(Tensor self) -> Tensor

exp2_(): exp2_(Tensor(a!) self) -> Tensor(a!)

exp_(): exp_(Tensor(a!) self) -> Tensor(a!)

expand(): expand(Tensor(a) self, SymInt[] size, *, bool implicit=False) -> Tensor(a)

expand_as(): expand_as(Tensor(a) self, Tensor other) -> Tensor(a)

expm1(): expm1(Tensor self) -> Tensor

expm1_(): expm1_(Tensor(a!) self) -> Tensor(a!)

exponential_(): exponential_(Tensor(a!) self, float lambd=1.0, *, Generator? generator=None) -> Tensor(a!)

feature_alpha_dropout_(): feature_alpha_dropout_(Tensor(a!) self, float p=0.5, bool train=True) -> Tensor(a!)

feature_dropout_(): feature_dropout_(Tensor(a!) self, float p=0.5, bool train=True) -> Tensor(a!)

fill_(): fill_.Tensor(Tensor(a!) self, Tensor value) -> Tensor(a!) fill_.Scalar(Tensor(a!) self, Scalar value) -> Tensor(a!)

fill_diagonal_(): fill_diagonal_(Tensor(a!) self, Scalar fill_value, bool wrap=False) -> Tensor(a!)

fix(): fix(Tensor self) -> Tensor

fix_(): fix_(Tensor(a!) self) -> Tensor(a!)

flatten(): flatten.using_ints(Tensor(a) self, int start_dim=0, int end_dim=-1) -> Tensor(a)

flip(): flip(Tensor self, int[] dims=[]) -> Tensor

fliplr(): fliplr(Tensor self) -> Tensor

flipud(): flipud(Tensor self) -> Tensor

float_power(): float_power(Tensor self, Tensor exponent) -> Tensor float_power.Tensor_Tensor(Tensor self, Tensor exponent) -> Tensor float_power.Tensor_Scalar(Tensor self, Scalar exponent) -> Tensor

float_power_(): float_power_.Tensor(Tensor(a!) self, Tensor exponent) -> Tensor(a!) float_power_.Scalar(Tensor(a!) self, Scalar exponent) -> Tensor(a!)

floor(): floor(Tensor self) -> Tensor

floor_(): floor_(Tensor(a!) self) -> Tensor(a!)

floor_divide(): floor_divide(Tensor self, Tensor other) -> Tensor floor_divide.Scalar(Tensor self, Scalar other) -> Tensor

floor_divide_(): floor_divide_.Tensor(Tensor(a!) self, Tensor other) -> Tensor(a!) floor_divide_.Scalar(Tensor(a!) self, Scalar other) -> Tensor(a!)

fmax(): fmax(Tensor self, Tensor other) -> Tensor

fmin(): fmin(Tensor self, Tensor other) -> Tensor

fmod(): fmod.Tensor(Tensor self, Tensor other) -> Tensor fmod.Scalar(Tensor self, Scalar other) -> Tensor

fmod_(): fmod_.Tensor(Tensor(a!) self, Tensor other) -> Tensor(a!) fmod_.Scalar(Tensor(a!) self, Scalar other) -> Tensor(a!)

frac(): frac(Tensor self) -> Tensor

frac_(): frac_(Tensor(a!) self) -> Tensor(a!)

frexp(): frexp.Tensor(Tensor self) -> (Tensor mantissa, Tensor exponent)

gather(): gather(Tensor self, int dim, Tensor index) -> Tensor

gcd(): gcd(Tensor self, Tensor other) -> Tensor

gcd_(): gcd_(Tensor(a!) self, Tensor other) -> Tensor(a!)

ge(): ge.Tensor(Tensor self, Tensor other) -> Tensor ge.Scalar(Tensor self, Scalar other) -> Tensor

ge_(): ge_.Tensor(Tensor(a!) self, Tensor other) -> Tensor(a!) ge_.Scalar(Tensor(a!) self, Scalar other) -> Tensor(a!)

gelu(): gelu(Tensor self, str approximate="none") -> Tensor

gelu_(): gelu_(Tensor(a!) self, str approximate="none") -> Tensor(a!)

geometric_(): geometric_(Tensor(a!) self, float p, *, Generator? generator=None) -> Tensor(a!)

geqrf(): geqrf(Tensor self) -> (Tensor a, Tensor tau)

ger(): ger(Tensor self, Tensor vec2) -> Tensor

get_device() → int

glu(): glu(Tensor self, int dim=-1) -> Tensor

greater(): greater(Tensor self, Tensor other) -> Tensor greater.Tensor(Tensor self, Tensor other) -> Tensor greater.Scalar(Tensor self, Scalar other) -> Tensor

greater_(): greater_.Tensor(Tensor(a!) self, Tensor other) -> Tensor(a!) greater_.Scalar(Tensor(a!) self, Scalar other) -> Tensor(a!)

greater_equal(): greater_equal(Tensor self, Tensor other) -> Tensor greater_equal.Tensor(Tensor self, Tensor other) -> Tensor greater_equal.Scalar(Tensor self, Scalar other) -> Tensor

greater_equal_(): greater_equal_.Tensor(Tensor(a!) self, Tensor other) -> Tensor(a!) greater_equal_.Scalar(Tensor(a!) self, Scalar other) -> Tensor(a!)

gt(): gt.Tensor(Tensor self, Tensor other) -> Tensor gt.Scalar(Tensor self, Scalar other) -> Tensor

gt_(): gt_.Tensor(Tensor(a!) self, Tensor other) -> Tensor(a!) gt_.Scalar(Tensor(a!) self, Scalar other) -> Tensor(a!)

hardshrink(): hardshrink(Tensor self, Scalar lambd=0.5) -> Tensor

hardshrink_backward(): hardshrink_backward(Tensor grad_out, Tensor self, Scalar lambd) -> Tensor

hardsigmoid(): hardsigmoid(Tensor self) -> Tensor

hardsigmoid_(): hardsigmoid_(Tensor(a!) self) -> Tensor(a!)

hardswish(): hardswish(Tensor self) -> Tensor

hardswish_(): hardswish_(Tensor(a!) self) -> Tensor(a!)

hardtanh(): hardtanh(Tensor self, Scalar min_val=-1, Scalar max_val=1) -> Tensor

hardtanh_(): hardtanh_(Tensor(a!) self, Scalar min_val=-1, Scalar max_val=1) -> Tensor(a!)

hash_tensor(): hash_tensor(Tensor self, int[1] dim=[], *, bool keepdim=False, int mode=0) -> Tensor

heaviside(): heaviside(Tensor self, Tensor values) -> Tensor

heaviside_(): heaviside_(Tensor(a!) self, Tensor values) -> Tensor(a!)

histc(): histc(Tensor self, int bins=100, Scalar min=0, Scalar max=0) -> Tensor

histogram(): histogram.bins_tensor(Tensor self, Tensor bins, *, Tensor? weight=None, bool density=False) -> (Tensor hist, Tensor bin_edges) histogram.bin_ct(Tensor self, int bins=100, *, float[]? range=None, Tensor? weight=None, bool density=False) -> (Tensor hist, Tensor bin_edges)

hsplit(split_size_or_sections) → List of Tensors

Splits a tensor of at least two dimensions into views along the second axis. split_size_or_sections is either the number of equal sections or the list of sizes of each section.

See tensorplay.hsplit()

hypot(): hypot(Tensor self, Tensor other) -> Tensor

hypot_(): hypot_(Tensor(a!) self, Tensor other) -> Tensor(a!)

i0(): i0(Tensor self) -> Tensor

i0_(): i0_(Tensor(a!) self) -> Tensor(a!)

i0e(): i0e(Tensor self) -> Tensor

i1(): i1(Tensor self) -> Tensor

i1e(): i1e(Tensor self) -> Tensor

igamma(): igamma(Tensor self, Tensor other) -> Tensor

igamma_(): igamma_(Tensor(a!) self, Tensor other) -> Tensor(a!)

igammac(): igammac(Tensor self, Tensor other) -> Tensor

igammac_(): igammac_(Tensor(a!) self, Tensor other) -> Tensor(a!)

index(): index.Tensor(Tensor self, Tensor?[] indices) -> Tensor

index_add(): index_add(Tensor self, int dim, Tensor index, Tensor source) -> Tensor

index_add_(): index_add_(Tensor(a!) self, int dim, Tensor index, Tensor source) -> Tensor(a!)

index_copy(): index_copy(Tensor self, int dim, Tensor index, Tensor source) -> Tensor

index_copy_(): index_copy_(Tensor(a!) self, int dim, Tensor index, Tensor source) -> Tensor(a!)

index_fill(): index_fill.Tensor(Tensor self, int dim, Tensor index, Tensor value) -> Tensor index_fill.int_Tensor(Tensor self, int dim, Tensor index, Tensor value) -> Tensor index_fill.Scalar(Tensor self, int dim, Tensor index, Scalar value) -> Tensor index_fill.int_Scalar(Tensor self, int dim, Tensor index, Scalar value) -> Tensor

index_fill_(): index_fill_.Tensor(Tensor(a!) self, int dim, Tensor index, Tensor value) -> Tensor(a!) index_fill_.int_Tensor(Tensor(a!) self, int dim, Tensor index, Tensor value) -> Tensor(a!) index_fill_.Scalar(Tensor(a!) self, int dim, Tensor index, Scalar value) -> Tensor(a!) index_fill_.int_Scalar(Tensor(a!) self, int dim, Tensor index, Scalar value) -> Tensor(a!)

index_put(): index_put(Tensor self, Tensor?[] indices, Tensor values, bool accumulate=False) -> Tensor

index_put_(): index_put_(Tensor(a!) self, Tensor?[] indices, Tensor values, bool accumulate=False) -> Tensor(a!)

index_reduce(): index_reduce(Tensor self, int dim, Tensor index, Tensor source, str reduce, *, bool include_self=True) -> Tensor

index_reduce_(): index_reduce_(Tensor(a!) self, int dim, Tensor index, Tensor source, str reduce, *, bool include_self=True) -> Tensor(a!)

index_select(): index_select(Tensor self, int dim, Tensor index) -> Tensor

indices(): indices(Tensor(a) self) -> Tensor(a)

inner(): inner(Tensor self, Tensor other) -> Tensor

int_repr(): int_repr(Tensor self) -> Tensor

inverse(): inverse(Tensor self) -> Tensor

is_channels_last(self: tensorplay._C.TensorBase) → bool

is_channels_last_2d(self: tensorplay._C.TensorBase) → bool

is_channels_last_3d(self: tensorplay._C.TensorBase) → bool

is_complex(self: tensorplay._C.TensorBase) → bool

is_conj(): is_conj(Tensor self) -> bool

is_distributed(): is_distributed(Tensor self) -> bool

is_inference(): is_inference(Tensor self) -> bool

is_neg(): is_neg(Tensor self) -> bool

is_nested(): is_nested(Tensor self) -> bool

is_nonzero(): is_nonzero(Tensor self) -> bool

is_quantized(): is_quantized(Tensor self) -> bool

is_same_size(): is_same_size(Tensor self, Tensor other) -> bool

is_set_to(): is_set_to(Tensor self, Tensor tensor) -> bool

is_shared(self: object) → bool

is_signed(): is_signed(Tensor self) -> bool

is_sparse_bsc(self: tensorplay._C.TensorBase) → bool

is_sparse_bsr(self: tensorplay._C.TensorBase) → bool

is_sparse_compressed(self: tensorplay._C.TensorBase) → bool

is_sparse_csc(self: tensorplay._C.TensorBase) → bool

is_sparse_csr(self: tensorplay._C.TensorBase) → bool

isclose(): isclose(Tensor self, Tensor other, float rtol=1e-05, float atol=1e-08, bool equal_nan=False) -> Tensor

isfinite(): isfinite(Tensor self) -> Tensor

isinf(): isinf(Tensor self) -> Tensor

isnan(): isnan(Tensor self) -> Tensor

isneginf(): isneginf(Tensor self) -> Tensor

isposinf(): isposinf(Tensor self) -> Tensor

isreal(): isreal(Tensor self) -> Tensor

item(): item(Tensor self) -> Scalar

itemsize(self: tensorplay._C.TensorBase) → int

kron(): kron(Tensor self, Tensor other) -> Tensor

kthvalue(): kthvalue(Tensor self, SymInt k, int dim=-1, bool keepdim=False) -> (Tensor values, Tensor indices)

lcm(): lcm(Tensor self, Tensor other) -> Tensor

lcm_(): lcm_(Tensor(a!) self, Tensor other) -> Tensor(a!)

ldexp(): ldexp(Tensor self, Tensor other) -> Tensor ldexp.Tensor(Tensor self, Tensor other) -> Tensor

ldexp_(): ldexp_(Tensor(a!) self, Tensor other) -> Tensor(a!)

le(): le.Tensor(Tensor self, Tensor other) -> Tensor le.Scalar(Tensor self, Scalar other) -> Tensor

le_(): le_.Tensor(Tensor(a!) self, Tensor other) -> Tensor(a!) le_.Scalar(Tensor(a!) self, Scalar other) -> Tensor(a!)

leaky_relu(): leaky_relu(Tensor self, Scalar negative_slope=0.01) -> Tensor

leaky_relu_(): leaky_relu_(Tensor(a!) self, Scalar negative_slope=0.01) -> Tensor(a!)

lerp(): lerp.Tensor(Tensor self, Tensor end, Tensor weight) -> Tensor lerp(Tensor self, Tensor end, Scalar weight) -> Tensor lerp.Scalar(Tensor self, Tensor end, Scalar weight) -> Tensor

lerp_(): lerp_.Tensor(Tensor(a!) self, Tensor end, Tensor weight) -> Tensor(a!) lerp_.Scalar(Tensor(a!) self, Tensor end, Scalar weight) -> Tensor(a!)

less(): less(Tensor self, Tensor other) -> Tensor less.Tensor(Tensor self, Tensor other) -> Tensor less.Scalar(Tensor self, Scalar other) -> Tensor

less_(): less_.Tensor(Tensor(a!) self, Tensor other) -> Tensor(a!) less_.Scalar(Tensor(a!) self, Scalar other) -> Tensor(a!)

less_equal(): less_equal(Tensor self, Tensor other) -> Tensor less_equal.Tensor(Tensor self, Tensor other) -> Tensor less_equal.Scalar(Tensor self, Scalar other) -> Tensor

less_equal_(): less_equal_.Tensor(Tensor(a!) self, Tensor other) -> Tensor(a!) less_equal_.Scalar(Tensor(a!) self, Scalar other) -> Tensor(a!)

lgamma(): lgamma(Tensor self) -> Tensor

lgamma_(): lgamma_(Tensor(a!) self) -> Tensor(a!)

log(): log(Tensor self) -> Tensor

log10(): log10(Tensor self) -> Tensor

log10_(): log10_(Tensor(a!) self) -> Tensor(a!)

log1p(): log1p(Tensor self) -> Tensor

log1p_(): log1p_(Tensor(a!) self) -> Tensor(a!)

log2(): log2(Tensor self) -> Tensor

log2_(): log2_(Tensor(a!) self) -> Tensor(a!)

log_(): log_(Tensor(a!) self) -> Tensor(a!)

log_ndtr(): log_ndtr(Tensor self) -> Tensor

log_normal_(): log_normal_(Tensor(a!) self, float mean=1.0, float std=2.0, *, Generator? generator=None) -> Tensor(a!)

log_sigmoid(): log_sigmoid(Tensor self) -> Tensor

log_softmax(): log_softmax(Tensor self, int dim, ScalarType dtype=Undefined) -> Tensor log_softmax.int(Tensor self, int dim, ScalarType? dtype=None) -> Tensor

logaddexp(): logaddexp(Tensor self, Tensor other) -> Tensor

logaddexp2(): logaddexp2(Tensor self, Tensor other) -> Tensor

logcumsumexp(): logcumsumexp(Tensor self, int dim, ScalarType? dtype=None) -> Tensor

logdet(): logdet(Tensor self) -> Tensor

logical_and(): logical_and(Tensor self, Tensor other) -> Tensor

logical_and_(): logical_and_(Tensor(a!) self, Tensor other) -> Tensor(a!)

logical_not(): logical_not(Tensor self) -> Tensor

logical_not_(): logical_not_(Tensor(a!) self) -> Tensor(a!)

logical_or(): logical_or(Tensor self, Tensor other) -> Tensor

logical_or_(): logical_or_(Tensor(a!) self, Tensor other) -> Tensor(a!)

logical_xor(): logical_xor(Tensor self, Tensor other) -> Tensor

logical_xor_(): logical_xor_(Tensor(a!) self, Tensor other) -> Tensor(a!)

logit(): logit(Tensor self, Scalar? eps=None) -> Tensor

logit_(): logit_(Tensor(a!) self, Scalar? eps=None) -> Tensor(a!)

logit_backward(): logit_backward(Tensor grad_output, Tensor self, Scalar? eps=None) -> Tensor

logsumexp(): logsumexp(Tensor self, int[1] dim, bool keepdim=False) -> Tensor

lt(): lt.Tensor(Tensor self, Tensor other) -> Tensor lt.Scalar(Tensor self, Scalar other) -> Tensor

lt_(): lt_.Tensor(Tensor(a!) self, Tensor other) -> Tensor(a!) lt_.Scalar(Tensor(a!) self, Scalar other) -> Tensor(a!)

lu_solve(): lu_solve(Tensor self, Tensor LU_data, Tensor LU_pivots) -> Tensor

map2_(self: object, tensor1: tensorplay._C.TensorBase, tensor2: tensorplay._C.TensorBase, callable: object) → object

map_(self: object, tensor: tensorplay._C.TensorBase, callable: object) → object

masked_fill(): masked_fill.Tensor(Tensor self, Tensor mask, Tensor value) -> Tensor masked_fill(Tensor self, Tensor mask, Scalar value) -> Tensor masked_fill.Scalar(Tensor self, Tensor mask, Scalar value) -> Tensor

masked_fill_(): masked_fill_.Tensor(Tensor(a!) self, Tensor mask, Tensor value) -> Tensor(a!) masked_fill_(Tensor(a!) self, Tensor mask, Scalar value) -> Tensor(a!) masked_fill_.Scalar(Tensor(a!) self, Tensor mask, Scalar value) -> Tensor(a!)

masked_scatter(): masked_scatter(Tensor self, Tensor mask, Tensor source) -> Tensor

masked_scatter_(): masked_scatter_(Tensor(a!) self, Tensor mask, Tensor source) -> Tensor(a!)

masked_select(): masked_select(Tensor self, Tensor mask) -> Tensor

matmul(): matmul(Tensor self, Tensor other) -> Tensor

matrix_H(): matrix_H(Tensor(a) self) -> Tensor(a)

matrix_exp(): matrix_exp(Tensor self) -> Tensor

matrix_power(): matrix_power(Tensor self, int n) -> Tensor

maximum(): maximum(Tensor self, Tensor other) -> Tensor

mean(): mean(Tensor self, *, ScalarType dtype=Undefined) -> Tensor mean.dim(Tensor self, int[] dim, bool keepdim=false, *, ScalarType dtype=Undefined) -> Tensor

median(): median(Tensor self) -> Tensor median.dim(Tensor self, int dim, bool keepdim=False) -> (Tensor values, Tensor indices)

memory_format(self: tensorplay._C.TensorBase) → int

minimum(): minimum(Tensor self, Tensor other) -> Tensor

mish(): mish(Tensor self) -> Tensor

mish_(): mish_(Tensor(a!) self) -> Tensor(a!)

mm(): mm(Tensor self, Tensor mat2) -> Tensor

mode(): mode(Tensor self, int dim=-1, bool keepdim=False) -> (Tensor values, Tensor indices)

modified_bessel_i0(): modified_bessel_i0(Tensor self) -> Tensor

modified_bessel_i1(): modified_bessel_i1(Tensor self) -> Tensor

modified_bessel_k0(): modified_bessel_k0(Tensor self) -> Tensor

modified_bessel_k1(): modified_bessel_k1(Tensor self) -> Tensor

module_load(other, assign=False)

Defines how other is remapped before being swapped with self when a state dictionary is loaded into the owning module.

Returns a new object that is neither self nor other: the default is self.copy_(other).detach() unless assign selects the detached source directly.

moveaxis(): moveaxis.int(Tensor(a) self, int source, int destination) -> Tensor(a) moveaxis.intlist(Tensor(a) self, int[] source, int[] destination) -> Tensor(a)

movedim(): movedim.int(Tensor(a) self, int source, int destination) -> Tensor(a) movedim(Tensor(a) self, int[] source, int[] destination) -> Tensor(a) movedim.intlist(Tensor(a) self, int[] source, int[] destination) -> Tensor(a)

msort(): msort(Tensor self) -> Tensor

mul(): mul.Tensor(Tensor self, Tensor other) -> Tensor mul.Scalar(Tensor self, Scalar other) -> Tensor

mul_(): mul_.Tensor(Tensor(a!) self, Tensor other) -> Tensor(a!) mul_.Scalar(Tensor(a!) self, Scalar other) -> Tensor(a!)

multinomial(): multinomial(Tensor self, SymInt num_samples, bool replacement=false, *, Generator? generator=None) -> Tensor

multiply(): multiply.Tensor(Tensor self, Tensor other) -> Tensor multiply.Scalar(Tensor self, Scalar other) -> Tensor

multiply_(): multiply_.Tensor(Tensor(a!) self, Tensor other) -> Tensor(a!) multiply_.Scalar(Tensor(a!) self, Scalar other) -> Tensor(a!)

mv(): mv(Tensor self, Tensor vec) -> Tensor

mvlgamma(): mvlgamma(Tensor self, int p) -> Tensor

mvlgamma_(): mvlgamma_(Tensor(a!) self, int p) -> Tensor(a!)

nan_to_num(): nan_to_num(Tensor self, Scalar nan=0.0, Scalar? posinf=None, Scalar? neginf=None) -> Tensor

nan_to_num_(): nan_to_num_(Tensor(a!) self, Scalar nan=0.0, Scalar? posinf=None, Scalar? neginf=None) -> Tensor(a!)

nanmean(): nanmean(Tensor self, int? dim=None, bool keepdim=False, *, ScalarType? dtype=None) -> Tensor

nanmedian(): nanmedian(Tensor self) -> Tensor nanmedian.dim(Tensor self, int dim, bool keepdim=False) -> (Tensor values, Tensor indices)

nanquantile(): nanquantile.scalar(Tensor self, float q, int? dim=None, bool keepdim=False, *, str interpolation='linear') -> Tensor

nansum(): nansum(Tensor self, int[] dim=[], bool keepdim=False) -> Tensor

narrow(): narrow.Tensor(Tensor(a) self, int dim, Tensor start, SymInt length) -> Tensor(a) narrow(Tensor(a) self, int dim, SymInt start, SymInt length) -> Tensor(a)

narrow_copy(): narrow_copy(Tensor self, int dim, SymInt start, SymInt length) -> Tensor

native_channel_shuffle(): native_channel_shuffle(Tensor self, SymInt groups) -> Tensor

nbytes(self: tensorplay._C.TensorBase) → int

ndimension() → int: Alias for dim()

ndtr(): ndtr(Tensor self) -> Tensor

ndtri(): ndtri(Tensor self) -> Tensor

ne(): ne.Tensor(Tensor self, Tensor other) -> Tensor ne.Scalar(Tensor self, Scalar other) -> Tensor

ne_(): ne_.Tensor(Tensor(a!) self, Tensor other) -> Tensor(a!) ne_.Scalar(Tensor(a!) self, Scalar other) -> Tensor(a!)

neg(): neg(Tensor self) -> Tensor

neg_(): neg_(Tensor(a!) self) -> Tensor(a!)

negative(): negative(Tensor self) -> Tensor

negative_(): negative_(Tensor(a!) self) -> Tensor(a!)

nelement() → int: Alias of numel().

new(self: tensorplay._C.TensorBase, *args, **kwargs) → tensorplay._C.TensorBase

new_empty_strided(): new_empty_strided(Tensor self, SymInt[] size, SymInt[] stride, *, ScalarType? dtype=None, Layout? layout=None, Device? device=None, bool? pin_memory=None) -> Tensor

nextafter(): nextafter(Tensor self, Tensor other) -> Tensor

nextafter_(): nextafter_(Tensor(a!) self, Tensor other) -> Tensor(a!)

nonzero_numpy(): nonzero_numpy(Tensor self) -> Tensor[]

nonzero_static(): nonzero_static(Tensor self, *, SymInt size, int fill_value=-1) -> Tensor

norm(): norm(Tensor self, float p=2.0) -> Tensor norm.dim(Tensor self, int[] dim, float p=2.0, bool keepdim=false) -> Tensor norm.ScalarOpt_dtype(Tensor self, Scalar? p, *, ScalarType dtype) -> Tensor norm.Scalar(Tensor self, Scalar p=2) -> Tensor norm.ScalarOpt_dim_dtype(Tensor self, Scalar? p, int[1] dim, bool keepdim, *, ScalarType dtype) -> Tensor norm.ScalarOpt_dim(Tensor self, Scalar? p, int[1] dim, bool keepdim=False) -> Tensor

normal_(): normal_(Tensor(a!) self, float mean=0.0, float std=1.0, *, Generator? generator=None) -> Tensor(a!)

not_equal(): not_equal(Tensor self, Tensor other) -> Tensor not_equal.Tensor(Tensor self, Tensor other) -> Tensor not_equal.Scalar(Tensor self, Scalar other) -> Tensor

not_equal_(): not_equal_.Tensor(Tensor(a!) self, Tensor other) -> Tensor(a!) not_equal_.Scalar(Tensor(a!) self, Scalar other) -> Tensor(a!)

numel() → int

numpy(self: object) → numpy.ndarray

numpy_T(): numpy_T(Tensor(a) self) -> Tensor(a)

orgqr(): orgqr(Tensor self, Tensor input2) -> Tensor

ormqr(): ormqr(Tensor self, Tensor input2, Tensor input3, bool left=True, bool transpose=False) -> Tensor

outer(): outer(Tensor self, Tensor vec2) -> Tensor

output_nr(): output_nr(Tensor self) -> int

pdist(): pdist(Tensor self, float p=2.0) -> Tensor

permute(): permute(Tensor(a) self, int[] dims) -> Tensor(a)

pinverse(): pinverse(Tensor self, float rcond=1e-15) -> Tensor

poisson(): poisson(Tensor self, Generator? generator=None) -> Tensor

polygamma(): polygamma(int n, Tensor self) -> Tensor

polygamma_(): polygamma_(Tensor(a!) self, int n) -> Tensor(a!)

positive(): positive(Tensor(a) self) -> Tensor(a)

pow(): pow.Tensor_Tensor(Tensor self, Tensor exponent) -> Tensor pow.Tensor_Scalar(Tensor self, Scalar exponent) -> Tensor

pow_(): pow_.Tensor(Tensor(a!) self, Tensor exponent) -> Tensor(a!) pow_.Scalar(Tensor(a!) self, Scalar exponent) -> Tensor(a!)

prelu(): prelu(Tensor self, Tensor weight) -> Tensor

prod(): prod(Tensor self, *, ScalarType dtype=Undefined) -> Tensor prod.dim_IntList(Tensor self, int[] dim, bool keepdim=false, *, ScalarType dtype=Undefined) -> Tensor prod.dim_int(Tensor self, int dim, bool keepdim=False, *, ScalarType? dtype=None) -> Tensor

put(): put(Tensor self, Tensor index, Tensor source, bool accumulate=False) -> Tensor

put_(): put_(Tensor(a!) self, Tensor index, Tensor source, bool accumulate=False) -> Tensor(a!)

q_per_channel_axis(): q_per_channel_axis(Tensor self) -> int

q_per_channel_scales(): q_per_channel_scales(Tensor self) -> Tensor

q_per_channel_zero_points(): q_per_channel_zero_points(Tensor self) -> Tensor

q_scale(): q_scale(Tensor self) -> float

q_zero_point(): q_zero_point(Tensor self) -> int

qscheme(): qscheme(Tensor self) -> int

quantile(): quantile.scalar(Tensor self, float q, int? dim=None, bool keepdim=False, *, str interpolation='linear') -> Tensor

rad2deg(): rad2deg(Tensor self) -> Tensor

rad2deg_(): rad2deg_(Tensor(a!) self) -> Tensor(a!)

random_(): random_(Tensor(a!) self, *, Generator? generator=None) -> Tensor(a!) random_.from(Tensor(a!) self, int from, int? to, *, Generator? generator=None) -> Tensor(a!) random_.to(Tensor(a!) self, int to, *, Generator? generator=None) -> Tensor(a!)

ravel(): ravel(Tensor(a) self) -> Tensor(a)

reciprocal(): reciprocal(Tensor self) -> Tensor

reciprocal_(): reciprocal_(Tensor(a!) self) -> Tensor(a!)

register_hook(hook)

The hook is called every time a gradient with respect to this tensor is computed. It may modify the gradient by returning a replacement Tensor; returning None leaves the gradient unchanged. Hooks compose in registration order.

Returns a RemovableHandle whose remove() method (or context-manager form) unregisters the hook.

register_post_accumulate_grad_hook(hook)

The hook runs after the gradient has been accumulated into self.grad. It receives the tensor (the parameter) and its return value is ignored; unlike register_hook() it cannot replace the gradient, but it may modify self.grad in place. Only leaf tensors that require grad and are used in the autograd graph support this hook.

Returns a RemovableHandle.

relu(): relu(Tensor self) -> Tensor

relu6(): relu6(Tensor self) -> Tensor

relu6_(): relu6_(Tensor(a!) self) -> Tensor(a!)

relu_(): relu_(Tensor(a!) self) -> Tensor(a!)

remainder(): remainder.Tensor(Tensor self, Tensor other) -> Tensor remainder.Scalar(Tensor self, Scalar other) -> Tensor

remainder_(): remainder_.Tensor(Tensor(a!) self, Tensor other) -> Tensor(a!) remainder_.Scalar(Tensor(a!) self, Scalar other) -> Tensor(a!)

renorm(): renorm(Tensor self, Scalar p, int dim, Scalar maxnorm) -> Tensor

renorm_(): renorm_(Tensor(a!) self, Scalar p, int dim, Scalar maxnorm) -> Tensor(a!)

repeat(): repeat(Tensor self, SymInt[] repeats) -> Tensor

reshape(): reshape(Tensor(a) self, SymInt[] shape) -> Tensor(a)

resize_(): resize_(Tensor(a!) self, SymInt[] size) -> Tensor(a!)

resize_as_(): resize_as_(Tensor(a!) self, Tensor other, *, MemoryFormat? memory_format=None) -> Tensor(a!)

resize_as_sparse_(): resize_as_sparse_(Tensor(a!) self, Tensor the_template) -> Tensor(a!)

resolve_conj(): resolve_conj(Tensor(a) self) -> Tensor(a)

resolve_neg(): resolve_neg(Tensor(a) self) -> Tensor(a)

roll(): roll(Tensor self, SymInt[1] shifts, int[] dims=[]) -> Tensor

rot90(): rot90(Tensor self, int k=1, int[] dims=[0,1]) -> Tensor

round(): round(Tensor self) -> Tensor round.decimals(Tensor self, *, int decimals) -> Tensor

round_(): round_(Tensor(a!) self) -> Tensor(a!) round_.decimals(Tensor(a!) self, *, int decimals) -> Tensor(a!)

row_indices(): row_indices(Tensor(a) self) -> Tensor(a)

rrelu(): rrelu(Tensor self, Scalar lower=0.125, Scalar upper=0.3333333333333333, bool training=False, Generator? generator=None) -> Tensor

rrelu_(): rrelu_(Tensor(a!) self, Scalar lower=0.125, Scalar upper=0.3333333333333333, bool training=False, Generator? generator=None) -> Tensor(a!)

rsqrt(): rsqrt(Tensor self) -> Tensor

rsqrt_(): rsqrt_(Tensor(a!) self) -> Tensor(a!)

scaled_modified_bessel_k0(): scaled_modified_bessel_k0(Tensor self) -> Tensor

scaled_modified_bessel_k1(): scaled_modified_bessel_k1(Tensor self) -> Tensor

scatter(): scatter.src(Tensor self, int dim, Tensor index, Tensor src) -> Tensor scatter.reduce(Tensor self, int dim, Tensor index, Tensor src, *, str reduce) -> Tensor scatter.value(Tensor self, int dim, Tensor index, Scalar value) -> Tensor scatter.value_reduce(Tensor self, int dim, Tensor index, Scalar value, *, str reduce) -> Tensor

scatter_(dim, index, src=None, *, value=None, reduce=None) → Tensor: Writes values from src (or the scalar value) into this tensor at the positions picked by index along dim; the remaining coordinates of each written element come from its position inside index. With reduce set to "add" or "multiply" (None by default), a destination that receives several values accumulates them instead of keeping the last one. Returns self.

scatter_add(): scatter_add(Tensor self, int dim, Tensor index, Tensor src) -> Tensor

scatter_add_(): scatter_add_(Tensor(a!) self, int dim, Tensor index, Tensor src) -> Tensor(a!)

scatter_reduce(): scatter_reduce(Tensor self, int dim, Tensor index, Tensor src, str reduce, *, bool include_self=True) -> Tensor scatter_reduce.two(Tensor self, int dim, Tensor index, Tensor src, str reduce, *, bool include_self=True) -> Tensor

scatter_reduce_(dim, index, src, reduce, *, include_self=True) → Tensor: Reduces the values of src into this tensor at the positions picked by index along dim, applying the reduction named by reduce: "sum", "prod", "mean", "amax" or "amin". When include_self is False, the original value of each destination element takes no part in the reduction. Returns self.

searchsorted(sorted_sequence, *, out_int32=False, right=False, side=None, sorter=None, out=None): Insertion positions of self values in sorted_sequence.

sec(): sec(Tensor self) -> Tensor

select(): select.int(Tensor(a) self, int dim, SymInt index) -> Tensor(a)

select_scatter(): select_scatter(Tensor self, Tensor src, int dim, SymInt index) -> Tensor

selu(): selu(Tensor self) -> Tensor

selu_(): selu_(Tensor(a!) self) -> Tensor(a!)

set_(): set_.source_Storage(Tensor(a!) self, Storage source) -> Tensor(a!) set_.source_Storage_storage_offset(Tensor(a!) self, Storage source, SymInt storage_offset, SymInt[] size, SymInt[] stride=[]) -> Tensor(a!) set_.source_Tensor_storage_offset(Tensor(a!) self, Tensor source, SymInt storage_offset, SymInt[] size, SymInt[] stride=[]) -> Tensor(a!) set_.source_Tensor(Tensor(a!) self, Tensor source) -> Tensor(a!) set_(Tensor(a!) self) -> Tensor(a!)

set_data(): set_data(Tensor(a!) self, Tensor new_data) -> ()

sgn(): sgn(Tensor self) -> Tensor

sgn_(): sgn_(Tensor(a!) self) -> Tensor(a!)

shallow_copy_data(): shallow_copy_data(Tensor(a!) self, Tensor new_data) -> ()

share_memory_(self: object) → object

sigmoid(): sigmoid(Tensor self) -> Tensor

sigmoid_(): sigmoid_(Tensor(a!) self) -> Tensor(a!)

sigmoid_backward(): sigmoid_backward(Tensor grad_output, Tensor output) -> Tensor

sign(): sign(Tensor self) -> Tensor

sign_(): sign_(Tensor(a!) self) -> Tensor(a!)

signbit(): signbit(Tensor self) -> Tensor

silu(): silu(Tensor self) -> Tensor

silu_(): silu_(Tensor(a!) self) -> Tensor(a!)

sin(): sin(Tensor self) -> Tensor

sin_(): sin_(Tensor(a!) self) -> Tensor(a!)

sinc(): sinc(Tensor self) -> Tensor

sinc_(): sinc_(Tensor(a!) self) -> Tensor(a!)

sinh(): sinh(Tensor self) -> Tensor

sinh_(): sinh_(Tensor(a!) self) -> Tensor(a!)

size() → int[]
size(dim) → int

slice(): slice.Tensor(Tensor(a) self, int dim=0, SymInt? start=None, SymInt? end=None, SymInt step=1) -> Tensor(a)

slice_inverse(): slice_inverse(Tensor(a) self, Tensor src, int dim=0, SymInt? start=None, SymInt? end=None, SymInt step=1) -> Tensor(a)

slice_scatter(): slice_scatter(Tensor self, Tensor src, int dim=0, SymInt? start=None, SymInt? end=None, SymInt step=1) -> Tensor

slogdet(): slogdet(Tensor self) -> (Tensor sign, Tensor logabsdet)

smm(): smm(Tensor self, Tensor mat2) -> Tensor

softmax(): softmax(Tensor self, int dim, ScalarType dtype=Undefined) -> Tensor softmax.int(Tensor self, int dim, ScalarType? dtype=None) -> Tensor

softplus(): softplus(Tensor self, Scalar beta=1, Scalar threshold=20) -> Tensor

softshrink(): softshrink(Tensor self, Scalar lambd=0.5) -> Tensor

softshrink_backward(): softshrink_backward(Tensor grad_output, Tensor self, Scalar lambd) -> Tensor

sort(): sort(Tensor self, int dim=-1, bool descending=False) -> (Tensor values, Tensor indices) sort.stable(Tensor self, *, bool? stable, int dim=-1, bool descending=False) -> (Tensor values, Tensor indices)

sparse_blocksize(self: tensorplay._C.TensorBase) → tuple[int, int]

sparse_mask(self: tensorplay._C.TensorBase, mask: tensorplay._C.TensorBase) → tensorplay._C.TensorBase

sparse_resize_(): sparse_resize_(Tensor(a!) self, int[] size, int sparse_dim, int dense_dim) -> Tensor(a!)

sparse_resize_and_clear_(): sparse_resize_and_clear_(Tensor(a!) self, int[] size, int sparse_dim, int dense_dim) -> Tensor(a!)

sparse_sum(): sparse_sum(Tensor self, int[]? dim=None, ScalarType? dtype=None) -> Tensor

spherical_bessel_j0(): spherical_bessel_j0(Tensor self) -> Tensor

split(): split(Tensor(a) self, int split_size, int dim=0) -> Tensor(a)[] split.Tensor(Tensor(a -> *) self, SymInt split_size, int dim=0) -> Tensor(a)[] split.sizes(Tensor(a) self, int[] split_sizes, int dim=0) -> Tensor(a)[]

split_with_sizes(): split_with_sizes(Tensor(a) self, SymInt[] split_sizes, int dim=0) -> Tensor(a)[]

sqrt(): sqrt(Tensor self) -> Tensor

sqrt_(): sqrt_(Tensor(a!) self) -> Tensor(a!)

square(): square(Tensor self) -> Tensor

square_(): square_(Tensor(a!) self) -> Tensor(a!)

squared_difference(): squared_difference(Tensor self, Tensor other) -> Tensor

squeeze(): squeeze(Tensor(a) self) -> Tensor(a) squeeze.dim(Tensor(a) self, int dim) -> Tensor(a) squeeze.dims(Tensor(a) self, int[] dim) -> Tensor(a)

squeeze_(): squeeze_(Tensor(a!) self) -> Tensor(a!) squeeze_.dim(Tensor(a!) self, int dim) -> Tensor(a!) squeeze_.dims(Tensor(a!) self, int[] dim) -> Tensor(a!)

sspaddmm(): sspaddmm(Tensor self, Tensor mat1, Tensor mat2, *, Scalar beta=1, Scalar alpha=1) -> Tensor

std(): std(Tensor self, int correction=1) -> Tensor std.dim(Tensor self, int[] dim, int correction=1, bool keepdim=false) -> Tensor std.correction(Tensor self, int[1]? dim=None, *, Scalar? correction=None, bool keepdim=False) -> Tensor

stft(): stft.center(Tensor self, int n_fft, int? hop_length=None, int? win_length=None, Tensor? window=None, bool center=True, str pad_mode="reflect", bool normalized=False, bool? onesided=None, bool? return_complex=None, bool? align_to_window=None) -> Tensor

storage_offset() → int

stride() → int[]
stride(dim) → int

sub(): sub.Tensor(Tensor self, Tensor other, *, Scalar alpha=1) -> Tensor sub.Scalar(Tensor self, Scalar other, Scalar alpha=1) -> Tensor

sub_(other, *, alpha=1) → Tensor: In-place version of tensorplay.Tensor.sub()

subtract(): subtract.Tensor(Tensor self, Tensor other, *, Scalar alpha=1) -> Tensor subtract.Scalar(Tensor self, Scalar other, Scalar alpha=1) -> Tensor

subtract_(other, *, alpha=1) → Tensor: In-place version of tensorplay.Tensor.subtract()

sum(): sum(Tensor self, *, ScalarType dtype=Undefined) -> Tensor sum.dim_IntList(Tensor self, int[] dim, bool keepdim=false, *, ScalarType dtype=Undefined) -> Tensor

sum_to_size(): sum_to_size(Tensor self, SymInt[] size) -> Tensor

svd(): svd(Tensor self, bool some=True, bool compute_uv=True) -> (Tensor U, Tensor S, Tensor V)

swapaxes(): swapaxes(Tensor(a) self, int axis0, int axis1) -> Tensor(a)

swapaxes_(): swapaxes_(Tensor(a!) self, int axis0, int axis1) -> Tensor(a!)

swapdims(): swapdims(Tensor(a) self, int dim0, int dim1) -> Tensor(a)

swapdims_(): swapdims_(Tensor(a!) self, int dim0, int dim1) -> Tensor(a!)

swish(): swish(Tensor self) -> Tensor

t(): t(Tensor(a) self) -> Tensor(a)

t_(): t_(Tensor(a!) self) -> Tensor(a!)

take(): take(Tensor self, Tensor index) -> Tensor

take_along_dim(): take_along_dim(Tensor self, Tensor indices, int? dim=None) -> Tensor

tan(): tan(Tensor self) -> Tensor

tan_(): tan_(Tensor(a!) self) -> Tensor(a!)

tanh(): tanh(Tensor self) -> Tensor

tanh_(): tanh_(Tensor(a!) self) -> Tensor(a!)

tanh_backward(): tanh_backward(Tensor grad_output, Tensor output) -> Tensor

tanhshrink(): tanhshrink(Tensor self) -> Tensor

tensor_split(indices_or_sections, dim=0) → List of Tensors

Splits this tensor into several views along dim. indices_or_sections is either the number of equal sections, the list of boundary indices at which the split happens, or a 0-dimensional integer tensor holding those boundaries.

See tensorplay.tensor_split()

threshold_(): threshold_(Tensor(a!) self, Scalar threshold, Scalar value) -> Tensor(a!)

tile(): tile(Tensor self, SymInt[] dims) -> Tensor

to_dense(): to_dense(Tensor self) -> Tensor

to_mkldnn(): to_mkldnn(Tensor self, ScalarType? dtype=None) -> Tensor

to_padded_tensor(): to_padded_tensor(Tensor self, float padding, SymInt[]? output_size=None) -> Tensor

to_sparse(): to_sparse(Tensor self) -> Tensor to_sparse.sparse_dim(Tensor self, int sparse_dim) -> Tensor

to_sparse_bsc(): to_sparse_bsc(Tensor self, int[2] blocksize, int? dense_dim=None) -> Tensor

to_sparse_bsr(): to_sparse_bsr(Tensor self, int[2] blocksize, int? dense_dim=None) -> Tensor

to_sparse_coo(): Convert a tensor to coordinate format.

to_sparse_csc(): to_sparse_csc(Tensor self, int? dense_dim=None) -> Tensor

to_sparse_csr(): to_sparse_csr(Tensor self) -> Tensor

tolist(self: tensorplay._C.TensorBase) → object

trace(): trace(Tensor self) -> Tensor

transpose(): transpose(Tensor(a) self, int dim0, int dim1) -> Tensor(a) transpose.int(Tensor(a) self, int dim0, int dim1) -> Tensor(a)

transpose_(): transpose_(Tensor(a!) self, int dim0, int dim1) -> Tensor(a!)

triangular_solve(): triangular_solve(Tensor self, Tensor A, bool upper=True, bool transpose=False, bool unitriangular=False) -> (Tensor solution, Tensor cloned_coefficient)

tril(): tril(Tensor self, SymInt diagonal=0) -> Tensor

tril_(): tril_(Tensor(a!) self, SymInt diagonal=0) -> Tensor(a!)

triu(): triu(Tensor self, SymInt diagonal=0) -> Tensor

triu_(): triu_(Tensor(a!) self, SymInt diagonal=0) -> Tensor(a!)

true_divide(): true_divide.Tensor(Tensor self, Tensor other) -> Tensor true_divide.Scalar(Tensor self, Scalar other) -> Tensor

true_divide_(): true_divide_.Tensor(Tensor(a!) self, Tensor other) -> Tensor(a!) true_divide_.Scalar(Tensor(a!) self, Scalar other) -> Tensor(a!)

trunc(): trunc(Tensor self) -> Tensor

trunc_(): trunc_(Tensor(a!) self) -> Tensor(a!)

type(dtype=None, non_blocking=False, **kwargs): Returns the type if dtype is not provided, else casts this object to the specified type.

unbind(): unbind(Tensor(a) self, int dim=0) -> Tensor(a)[] unbind.int(Tensor(a -> *) self, int dim=0) -> Tensor(a)[]

unflatten(): unflatten.int(Tensor self, int dim, SymInt[] sizes) -> Tensor

unfold(): unfold(Tensor(a) self, int dimension, int size, int step) -> Tensor(a)

uniform_(): uniform_(Tensor(a!) self, float from=0.0, float to=1.0, *, Generator? generator=None) -> Tensor(a!)

unsafe_chunk(): unsafe_chunk(Tensor self, int chunks, int dim=0) -> Tensor[]

unsafe_split(): unsafe_split.Tensor(Tensor self, SymInt split_size, int dim=0) -> Tensor[]

unsafe_split_with_sizes(): unsafe_split_with_sizes(Tensor self, SymInt[] split_sizes, int dim=0) -> Tensor[]

unsqueeze(): unsqueeze(Tensor(a) self, int dim) -> Tensor(a)

unsqueeze_(): unsqueeze_(Tensor(a!) self, int dim) -> Tensor(a!)

untyped_storage(self: tensorplay._C.TensorBase) → tensorplay._C.UntypedStorage

var(): var(Tensor self, int correction=1) -> Tensor var.dim(Tensor self, int[] dim, int correction=1, bool keepdim=false) -> Tensor var.correction(Tensor self, int[1]? dim=None, *, Scalar? correction=None, bool keepdim=False) -> Tensor

vdot(): vdot(Tensor self, Tensor other) -> Tensor

view(*shape) → Tensor
view(dtype) → Tensor

view_as(): view_as(Tensor(a) self, Tensor other) -> Tensor(a)

view_as_complex(): view_as_complex(Tensor(a) self) -> Tensor(a)

view_as_real(): view_as_real(Tensor(a) self) -> Tensor(a)

vsplit(split_size_or_sections) → List of Tensors

Splits a tensor of at least two dimensions into views along the first axis. split_size_or_sections is either the number of equal sections or the list of sizes of each section.

See tensorplay.vsplit()

where(): where.self(Tensor condition, Tensor self, Tensor other) -> Tensor where.ScalarOther(Tensor condition, Tensor self, Scalar other) -> Tensor

xlog1py(): xlog1py(Tensor self, Tensor other) -> Tensor

xlog1py_(): xlog1py_(Tensor(a!) self, Tensor other) -> Tensor(a!)

xlogy(): xlogy(Tensor self, Tensor other) -> Tensor xlogy.Tensor(Tensor self, Tensor other) -> Tensor xlogy.Scalar_Other(Tensor self, Scalar other) -> Tensor

xlogy_(): xlogy_(Tensor(a!) self, Tensor other) -> Tensor(a!) xlogy_.Tensor(Tensor(a!) self, Tensor other) -> Tensor(a!) xlogy_.Scalar_Other(Tensor(a!) self, Scalar other) -> Tensor(a!)

zero_(): zero_(Tensor(a!) self) -> Tensor(a!)

zeta(): zeta(Tensor self, Tensor other) -> Tensor

CausalVariant

classFull reference ↗

class tensorplay.nn.attention.bias.CausalVariant(*values)[source]

Enum for causal variants used in attention mechanisms.

Defines two types of causal biases:

UPPER_LEFT: Represents upper-left triangular bias for standard causal attention. The equivalent code for constructing this bias is:

tensorplay.tril(tensorplay.ones(size, dtype=tensorplay.bool))

For instance, with shape=(3,4), the materialized bias tensor will be:

[[1, 0, 0, 0],
 [1, 1, 0, 0],
 [1, 1, 1, 0]]

LOWER_RIGHT: Represents lower-right triangular bias, the include values are aligned to the lower right corner of the matrix.

The equivalent code for constructing this bias is:

diagonal_offset = size[1] - size[0]
tensorplay.tril(
    tensorplay.ones(size, dtype=tensorplay.bool),
    diagonal=diagonal_offset,
)

For instance, with shape=(3,4), the materialized bias tensor will be:

[[1, 1, 0, 0],
 [1, 1, 1, 0],
 [1, 1, 1, 1]]

Note that these variants are equivalent to each other when the sequence lengths of the query and key/value tensors are equal since the triangular matrix is square.

Warning

This enum is a prototype and subject to change.

as_integer_ratio()

Return a pair of integers, whose ratio is equal to the original int.

The ratio is in lowest terms and has a positive denominator.

>>> (10).as_integer_ratio()
(10, 1)
>>> (-10).as_integer_ratio()
(-10, 1)
>>> (0).as_integer_ratio()
(0, 1)

bit_count()

Number of ones in the binary representation of the absolute value of self.

Also known as the population count.

>>> bin(13)
'0b1101'
>>> (13).bit_count()
3

bit_length()

Number of bits necessary to represent self in binary.

>>> bin(37)
'0b100101'
>>> (37).bit_length()
6

conjugate(): Returns self, the complex conjugate of any int.

denominator: the denominator of a rational number in lowest terms

classmethod from_bytes(bytes, byteorder='big', *, signed=False)

imag: the imaginary part of a complex number

is_integer(): Returns True. Exists for duck type compatibility with float.is_integer.

numerator: the numerator of a rational number in lowest terms

real: the real part of a complex number

to_bytes(length=1, byteorder='big', *, signed=False)

Functions 2

causal_lower_right

causal_upper_left

Classes 2

CausalBias

CausalVariant

Help improve this page

On this page