sh17SrSSKJr SSKrSSKrSSKrSSKJrJr SSKJ r SSK r SSK J r SSK J r SSKJr \R "S 5rS S 1r\R&"S S 55rSSS.SSjjrSS.SSjjrS SjrS!S"SjjrSrS#SjrS$SjrS%SjrS&SjrS'SjrS(SjrS(Sjr S(Sjr!g))zFUtilities for manipulating the torch.Graph object and the torchscript.) annotationsN)IterableSequence)Any)_C)GLOBALS) registrationz^(.+)_(([ifstgz])|(ty))$inplaceatenc\rSrSr%SrS\S'S\S'S\S'S \S 'S \S 'S \S'S\S'\R"\S9r S\S'SSjr SS.SSjjr SS.S Sjjr \ r SS.S!SjjrSrg)" GraphContextaExtra context for symbolic functions with all methods from torch.Graph. NOTE: This class is not meant for external consumption. Please do not depend on it outside of torch.onnx as the interface may evolve. Attributes: graph: The _C.Graph being constructed. block: The current _C.Block being constructed. opset: The opset version. original_node: Current node that is being converted from. params_dict: Mapping from graph initializer name to IValue. env: Mapping from Torch domain graph Value to ONNX domain graph Value. values_in_env: Set of all values in env, for constant-time lookups. new_nodes: List that tracks all new nodes that are added (used to make sure metadata is propagated to all new nodes). z_C.Graphgraphz_C.Blockblockintopset_C.Node original_nodezdict[str, _C.IValue] params_dictzdict[_C.Value, _C.Value]envz set[_C.Value] values_in_env)default_factoryz list[_C.Node] new_nodesc.[URU5$N)getattrr)selfnames r/Users/tiagomarins/Projetos/claudeai/copy_bank/venv/lib/python3.13/site-packages/torch/onnx/_internal/jit_utils.py __getattr__GraphContext.__getattr__7stzz4((outputsc$[X/UQ7SU0UD6$)a]Creates an ONNX operator "opname", taking "raw_args" as inputs and "kwargs" as attributes. The set of operators and the inputs/attributes they take is documented at https://github.com/onnx/onnx/blob/master/docs/Operators.md Args: opname: The ONNX operator name, e.g., `Abs` or `Add`, or an operator qualified with a namespace, e.g., `aten::add`. raw_args: The inputs to the operator; usually provided as arguments to the `symbolic` definition. outputs: The number of outputs this operator returns. By default an operator is assumed to return a single output. If `outputs` is greater than one, this functions returns a tuple of output `Value`, representing each output of the ONNX operator in order. kwargs: The attributes of the ONNX operator, whose keys are named according to the following convention: `alpha_f` indicates the `alpha` attribute with type `f`. The valid type specifiers are `f` (float), `i` (int), `s` (string) or `t` (Tensor). An attribute specified with type float accepts either a single float, or a list of floats (e.g., you would say `dims_i` for a `dims` attribute that takes a list of integers). Returns: The value representing the single output of this operator (see the `outputs` keyword argument for multi-return nodes). r%)_add_op)ropnamer%raw_argskwargss ropGraphContext.op:sFtJhJJ6JJr") overload_namec4UR"S/UQ7UUS.UD6$)znGenerates an ONNX ATen op node. This function is for backward compatibility with the old symbolic functions. z aten::ATen) operator_soverload_name_s)r+)roperatorr.argsr*s raten_opGraphContext.aten_op_s7 ww    )     r"cURRSUR3nURRn[R "XV5"U5 [ X/UQ7SU0UD6$)a*Creates an ONNX operator from onnx-script function, taking "raw_args" as inputs and "kwargs" as attributes. onnx-script repository: https://github.com/microsoft/onnx-script Args: onnx_fn: ONNXFunction from onnx-script; An example can be found at https://github.com/microsoft/onnx-script#example raw_args: The inputs to the operator; usually provided as arguments to the `symbolic` definition. outputs: The number of outputs this operator returns. By default an operator is assumed to return a single output. If `outputs` is greater than one, this functions returns a tuple of output `Value`, representing each output of the ONNX operator in order. kwargs: The attributes of the ONNX operator, whose keys are named according to the following convention: `alpha_f` indicates the `alpha` attribute with type `f`. The valid type specifiers are `f` (float), `i` (int), `s` (string) or `t` (Tensor). An attribute specified with type float accepts either a single float, or a list of floats (e.g., you would say `dims_i` for a `dims` attribute that takes a list of integers). Returns: The value representing the single output of this operator (see the `outputs` keyword argument for multi-return nodes). ::r%)rdomainrversionr custom_onnx_symbolicr')ronnx_fnr%r)r* symbolic_name opset_versions r onnxscript_opGraphContext.onnxscript_opps`F#==//07<<.A  -- ))-GPtQXQwQ&QQr"N)rstrreturnr)r(rAr)torch.Tensor | _C.Valuer%r)r2rAr.rA)r)rCr%r)__name__ __module__ __qualname____firstlineno____doc____annotations__ dataclassesfieldlistrr r+r4atr>__static_attributes__r@r"rr r s" O O J%% !!  *00FI}F) #K#K+#K #KJBD   B  (R+(R (R(Rr"r r#)r%n_blockscRUR"U/UQ7SU0UD6n[U[5(aUSR5nOUR5n/n[ U5H8n UR 5n [ R"X S9n URU 5 M: U[U5U4$)aCreates an ONNX operator "opname", taking inputs and attributes. Args: graph_context: The context for the current graph. opname: The ONNX operator name, e.g., `Abs` or `Add`, or an operator qualified with a namespace, e.g., `aten::add`. inputs: The inputs to the operator. outputs: The number of outputs this operator returns. By default an operator is assumed to return a single output. If `outputs` is greater than one, this functions returns a tuple of output `Value`, representing each output of the ONNX operator in order. n_blocks: The number of sub-blocks to create in the node. attributes: The attributes of the ONNX operator. Returns: A tuple of (output_values, new_contexts, node) where: output_values: One or more output value of this operator (see the `outputs` keyword argument for multi-return nodes). new_contexts: A tuple of new graph contexts for each sub-block. node: The node representing the operator. r%r)r) r+ isinstancernoderangeaddBlockrJreplaceappendtuple) graph_contextr(r%rOinputs attributes output_valuesrR new_contexts_ new_block new_contexts radd_op_with_blocksr`s>"$$VTfTgTTM-**Q$$&!!#L 8_MMO !))-I K(  % -t 33r"r$c UVs/sHn[X5PM nnUR5VVs0sH upxUcM Xx_M n nnSU;aSU-n[URUUU URUR U[ RS9n URRU 5 US:XaU R5$[U R55$s snfs snnf)aCreates an ONNX operator "opname", taking "args" as inputs and attributes "kwargs". The set of operators and the inputs/attributes they take is documented at https://github.com/onnx/onnx/blob/master/docs/Operators.md This function is monkey-patched onto Graph. Args: graph_context: The Torch Graph or Block. opname: The ONNX operator name, e.g., `Abs` or `Add`, or an operator qualified with a namespace, e.g., `aten::add`. args: The inputs to the operator; usually provided as arguments to the `symbolic` definition. outputs: The number of outputs this operator returns. By default an operator is assumed to return a single output. If `outputs` is greater than one, this functions returns a tuple of output `Value`, representing each output of the ONNX operator in order. kwargs: The attributes of the ONNX operator, whose keys are named according to the following convention: `alpha_f` indicates the `alpha` attribute with type `f`. The valid type specifiers are `f` (float), `i` (int), `s` (string) or `t` (Tensor). An attribute specified with type float accepts either a single float, or a list of floats (e.g., you would say `dims_i` for a `dims` attribute that takes a list of integers). Returns: (Union[_C.Value, Tuple[_C.Value, ...]]) The value representing the single output of this operator (see the `outputs` keyword argument for multi-return nodes). r7zonnx::)rr= n_outputsshape_inferencer#) _const_if_tensoritems _create_noderrrronnx_shape_inferencerrVoutputrWr%) rXr(r%r3r*argrYkvrZrRs rr'r'sL?C Cds}2dF C$*<<>C>41Q$!$>JC 6F" !--#))44 D""4(!|{{}   -DDsC CCcdUcU$[U[R5(aU$[USUS9$)Nzonnx::Constant)value_z)rQrValuer')rXris rrdrd s2 { #rxx   ="2C @@r"ch[U[R5(a&UnURXU5n UR U 5n O[[U[R 5(a<Un U R X5n US:a#[SU5Hn U R5 M [W R55n [U 5U:XdeURS5n [UR55HupU[;aM[!XXS9 M U(a[R""XU5 U $)z:Creates an node 'domain_op', taking inputs and attributes.r#zaten::)r )rQrGraphcreate insertNodeBlockaddNoderS addOutputrWr%len startswithsortedre_SKIP_NODE_ATTRIBUTES_add_attribute(_jit_pass_onnx_node_shape_type_inference)graph_or_block domain_oprYrZrr=rbrcrrRrr] node_outputsr keyvalues rrfrfs."((++||Iy9% NBHH - -}}Y/ q=1i( )(L |  )) )    )DZ--/0  ' ' t%31 33D}U Kr"c[U[5=(a* [U[[[R 45(+$r)rQrrAbytestorchTensor)rs r _is_onnx_listr9s2 eX & z UELL)0,r"c8UR5S:XdeUS$)z,Convert a scalar tensor into a Python value.r#r)numelxs r_scalarr?s 779>> Q4Kr"c[RU5nUc[SUS35eURS5URS5pe[ U5(aUS- n[ XS35"XR5$)z7Initializes the right attribute based on type of value.zInvalid attribute specifier 'z<' names must be suffixed with type, e.g. 'dim_i' or 'dims_i'r#sr]) _ATTR_PATTERNmatch ValueErrorgrouprr)rRrrr mrkinds rrzrzEszC Ay+C51C C  QWWQZ$U   46 $T 11r"cxUR5R[RR 55$r)type isSubtypeOfr TensorTypegetrs r _is_tensorrUs& 668   1 1 3 44r"c[U5(dg[R"[RUR 55nUR 5$r)rtypingcastrrrdevice)r tensor_types rget_device_from_valuerYs9 e  ++bmmUZZ\:K    r"cSU;a[SUS35eURSS5upSU;a[SUS35eX4$)z(Parse node kind into domain and Op name.r7z Node kind: z& is invalid. '::' is not in node kind.r#z) is invalid. '::' should only apear once.)rsplit)rr8r(s rparse_node_kindr`sU 4;tf,RSTTZZa(NF v~;tf,UVWW >r"c US:H$) Check if the domain is official.r r@r8s ris_atenrj V r"c US:H$)rprimr@rs ris_primrorr"c US:H$)ronnxr@rs ris_onnxrtrr") rXr r(rArY_C.Valuer%rrOrrBz-tuple[Any, tuple[GraphContext, ...], _C.Node])rXr r(rAr3rCr%r)rXr )T)r|z_C.Graph | _C.Blockr}rArYrrZdictrrr=rrbrrcboolrBr)rz torch.Tensor)rRrrrArrr r)rrrBr)rrrBztorch.device | None)rrArBztuple[str, str])r8rArBr)"rH __future__rrJrercollections.abcrrrrrtorch.onnx._globalsrtorch.onnx._internalr compilerry dataclassr r`r'rdrfrrrzrrrrrrr@r"rrssL # . '- 56 "F+ ~R~R~RJ ,4,4 ,4,4 ,4  ,43,4f <!<! <! #<! <!~A"!$'$$ $ $  $  $$$ $N  2 5   r"