[ThO VSSKJr SSKJrJr SSKJr SrSrSr Sr Sr S r S r g ) )Refine)unifyVar) TensorTypec[U5nUR5 [UR5n[ UR U5 g)z% Calls our symbolic inferencer once. N)rrefineunify_eq constraintssubstitute_all_typesgraphtracedrmgus _/var/www/auris/envauris/lib/python3.13/site-packages/torch/fx/experimental/unify_refinements.py infer_symbolic_types_single_passrs2 vAHHJ 1== !Cs+c<[U5nUR5 [UR5n[ UR U5 [U5nUR5 [UR5n[ UR U5 UR 5 g)z Calls our symbolic inferencer twice. This is useful when one pass is not enough to infer all the information such as the case for braodcasting. N)rrr r r r symbolic_relationsr s rinfer_symbolic_typesrsl vAHHJ 1== !Cs+vAHHJ 1== !Cs+rc/n/nUH9nURUR5 URUR5 M; [U5[U54$)zU Convert equality constraints in the right format to be used by unification library. )appendlhsrhstuple) list_of_eqrreqs r convert_eqr%sM C C 266 266 :uSz !!rc2[U5up[X5$)z4 Apply unification to a set of equality constraints )rr)rrrs rr r 2s *%HC ?rcR[U[5(aXR5;aX$U$[U[5(ab/nURH<nX0R5;aUR X5 M+UR U5 M> [[ U55$[U[5(a'/nUHnUR [X55 M U$[U[ 5(a0/nUHnUR [X55 M [ U5$U$)z* Apply the most general unifier to a type ) isinstancerkeysr__args__rrlistsubstitute_solution_one_type)mappingtnew_typetyps rr%r%;s!S  : H Az " "::Clln$ -$  %/** At  C OO8F G Au  C OO8F GXrcSnU(aDSnUH3nXnXUR5;a XnXX'XAU:wdM1SnM5 U(aMDURHn[XR5UlM g)z Apply the most general unifier to all types in a graph till reaching a fixed point. If the input and output graph are the same, we converge. TFN)r"nodesr%type)r r&flagkold_mapping_valnew_keyns rr r ^sy D A%jOzW\\^+!*$- !*,  $[[-gvv>rc[URUR5H!up#URUR:wdM! g g)zf A check equality to be used in fixed points. We do not use graph equality but instead type equality. FT)zipr+r,)g1g2r1ms rcheck_for_type_equalityr7ss7 BHHbhh' 66QVV ( rN)/torch.fx.experimental.graph_gradual_typecheckerr!torch.fx.experimental.unificationrrtorch.fx.tensor_typerrrrr r%r r7rrr<s2B8+,( " F?* r