9j2UddlmZmZddlmZddlmZmZmZm Z m Z ddl Z ddl m cmZddl mZmZddlmZddlmZmZmZmZdZeeed <gd Zd eezd eed ffdZdeeezdedzdedzd efdZ deded efdZ!d*dede"d efdZ#ded efdZ$d*dede"d efdZ%e ddZ&e dd Z'Gd!d"ee&e'fZ(Gd#d$e(e&e'fe)Z*d+d%ed&ed efd'Z+d+d(ed&ed efd)Z,y),)CallableSequence)update_wrapper)AnyFinalGenericoverloadTypeVarN)SymIntTensoris_tensor_like)_dtype_NumberDeviceNumbergox?euler_constant) broadcast_alllogits_to_probs clamp_probsprobs_to_logits lazy_propertytril_matrix_to_vecvec_to_tril_matrixvaluesreturn.ctd|Ds tdtd|Dsttj}|D]?}t |tj st|j|j}n|Dcgc]%}t|r|ntj|fi|'}}tj|Stj|Scc}w)a Given a list of values (possibly containing numbers), returns a list where each value is broadcasted based on the following rules: - `torch.*Tensor` instances are broadcasted as per :ref:`_broadcasting-semantics`. - Number instances (scalars) are upcast to tensors having the same size and type as the first tensor passed to `values`. If all the values are scalars, then they are upcasted to scalar Tensors. Args: values (list of `Number`, `torch.*Tensor` or objects implementing __torch_function__) Raises: ValueError: if any of the values is not a `Number` instance, a `torch.*Tensor` instance, or an instance implementing __torch_function__ c3VK|]!}t|xst|t#ywN)r isinstancer.0vs Y/media/conek/DATA/Code/OCR/venv/lib/python3.12/site-packages/torch/distributions/utils.py z broadcast_all..+s$Kq~a :Jq'$::Ks')ziInput arguments must all be instances of Number, torch.Tensor or objects implementing __torch_function__.c32K|]}t|ywrr r!s r$r%z broadcast_all..0s1Q~a 1s)dtyper'device) all ValueErrordicttorchget_default_dtyper r r'r)rtensorbroadcast_tensors)roptionsvaluer# new_valuess r$rrs KFK K G   1&1 1"&U-D-D-F"G E%.U[[F  MS GH"A Q(B'(B B  && 33  " "F ++  s*C&shaper'r)ctjjrBtjtj|||tj |||Stj |||jS)Nr()r-_C_get_tracing_statenormalzerosonesemptynormal_)r4r'r)s r$_standard_normalr==sc  xx""$|| KKU6 : JJuE& 9   ;;uE& 9 A A CCr2dimcv|dk(r|S|jd| dz}|j|jdS)z Sum out ``dim`` many rightmost dimensions of a given tensor. Args: value (Tensor): A tensor of ``.dim()`` at least ``dim``. dim (int): The number of rightmost dims to sum out. rN)rA)r4reshapesum)r2r?required_shapes r$_sum_rightmostrEKsA ax [[3$'%/N == ( , ,R 00r>logits is_binaryc^|rtj|Stj|dS)a Converts a tensor of logits into probabilities. Note that for the binary case, each value denotes log odds, whereas for the multi-dimensional case, the values along the last dimension denote the log probabilities (possibly unnormalized) of the events. rA)r?)r-sigmoidFsoftmax)rFrGs r$rrYs'}}V$$ 99V $$r>probsctj|jj}|j |d|z S)aClamps the probabilities to be in the open interval `(0, 1)`. The probabilities would be clamped between `eps` and `1 - eps`, and `eps` would be the smallest representable positive number for the input data type. Args: probs (Tensor): A tensor of probabilities. Returns: Tensor: The clamped probabilities. Examples: >>> probs = torch.tensor([0.0, 0.5, 1.0]) >>> clamp_probs(probs) tensor([1.1921e-07, 5.0000e-01, 1.0000e+00]) >>> probs = torch.tensor([0.0, 0.5, 1.0], dtype=torch.float64) >>> clamp_probs(probs) tensor([2.2204e-16, 5.0000e-01, 1.0000e+00], dtype=torch.float64) )minmax)r-finfor'epsclamp)rLrRs r$rres3, ++ekk " & &C ;;3AG; ,,r>ct|}|r,tj|tj| z Stj|S)a$ Converts a tensor of probabilities into logits. For the binary case, this denotes the probability of occurrence of the event indexed by `1`. For the multi-dimensional case, the values along the last dimension denote the probabilities of occurrence of each of the events. )rr-loglog1p)rLrG ps_clampeds r$rrs?U#Jyy$u{{J;'??? 99Z  r>TT) contravariantR) covariantceZdZdZdeegefddfdZe d ddde ddfd Z ed dede defd Z d dedzde dd fd Z y)rz Used as a decorator for lazy loading of class attributes. This uses a non-data descriptor that calls the wrapped method to compute the property on first call; thereafter replacing the wrapped method into an instance attribute. wrappedrNc*||_t||yr)r]rselfr]s r$__init__zlazy_property.__init__s)0 tW%r>instanceobj_typez!_lazy_property_and_property[T, R]cyrr`rbrcs r$__get__zlazy_property.__get__s/2r>cyrrerfs r$rgzlazy_property.__get__s?Br>z%R | _lazy_property_and_property[T, R]c|t|jStj5|j|}dddt ||jj |S#1swY,xYwr)_lazy_property_and_propertyr]r- enable_gradsetattr__name__)r`rbrcr2s r$rgzlazy_property.__get__sc  .t||< <     +LL*E +$,,//7  + +s A))A2r) rm __module__ __qualname____doc__rrXrZrar rrgrer>r$rrs&!a 0&T&.222(+2 ,22BBSBABB37D,/ 0r>rc,eZdZdZdeegefddfdZy)rjzWe want lazy properties to look like multiple things. * property when Sphinx autodoc looks * lazy_property when Distribution validate_args looks r]rNc0tj||yr)propertyrar_s r$raz$_lazy_property_and_property.__init__s$(r>)rmrnrorprrXrZrarer>r$rjrjs% )!a 0)T)r>rjmatdiagc .|jd}tjjs$|| ks||k\rt d|d| d|dz dtj ||j }||jdd|dzzk}|d|f}|S) z Convert a `D x D` matrix or a batch of matrices into a (batched) vector which comprises of lower triangular elements from the matrix in row order. rAzdiag (z) provided is outside [z, rNz].r).)r4r-r6r7r+aranger)view)rtrunrx tril_maskvecs r$rrs " A 88 & & (daRi4196$'>rd"QUG2NOO \\!CJJ /FR+tax88I c9n C Jr>r|cdd|zz dd|zzdzd|jdzzdt|z|dzzzdzzdz }tj|jj }tj js0t||z |kDrtd|jddd zt|tjrt|jn t|}|j|jd dtj||fz}tj||j }||j#dd|dzzk}||d |f<|S) z Convert a vector or a batch of vectors into a batched `D x D` lower triangular matrix containing elements from the vector in row order. rNrAg?zThe size of last dimension is z which cannot be expressed as z3the lower triangular part of a square D x D matrix.Nrw.)r4absr-rQr'rRr6r7roundr+r r item new_zerosSizerxr)ry)r|rurzrRrtrxr{s r$rrsV a$h, DLQ SYYr]!2 2QT]dQh5O OTW W X  A ++cii $ $C 88 & & (eAhlS.@,SYYr]O;Y ZC D  &a6affhE!HA -- #2QF);; )F)r)-collections.abcrr functoolsrtypingrrrr r r-torch.nn.functionalnn functionalrJr r torch.overridesr torch.typesrrrrrfloat__annotations____all__tuplerintr=rEboolrrrrXrZrrsrjrrrer>r$rsz..$99  *77 6e 5 ,6F?,uVS[/A,D D C&L ! D D= D TM D D 1& 1s 1v 1 %F %t % %-v-&-4 !6 !d !v ! Ct$ C4 GAqDM>)-1"5x) F # f F#fr>