<Цi4%tSSKrSSKJr SSKrSSKrSSKJrJr /SQr"SS5r Sr Sr "S S 5r g) N) OrderedDict)AnyTuple)RemovableHandleunserializable_hookwarn_if_has_hooks BackwardHookc\rSrSr%Sr\\S'Sr\\S'SS.S\S \S S4S jjr SS jr S r SSjr SSjr S\S\S\S S4SjrSrg)r aA A handle which provides the capability to remove a hook. Args: hooks_dict (dict): A dictionary of hooks, indexed by hook ``id``. extra_dict (Union[dict, List[dict]]): An additional dictionary or list of dictionaries whose keys will be deleted when the same keys are removed from ``hooks_dict``. idrnext_idN) extra_dict hooks_dictrreturncb[R"U5Ul[RUl[=RS- slSUl[U[5(a[R"U54Ulg[U[5(a[SU55Ulgg)Nc3N# UHn[R"U5v M g7fNweakrefref.0ds P/var/www/html/ai-image-ml/venv/lib/python3.13/site-packages/torch/utils/hooks.py +RemovableHandle.__init__..!s'K 1 A #%) rrhooks_dict_refrr r extra_dict_ref isinstancedictlisttuple)selfrrs r__init__RemovableHandle.__init__s%kk*5!))1$%' j$ ' '#*;;z#:".1s9_K^C#%%K^s)r!r r r%r&s r __getstate__RemovableHandle.__getstate__-sQ    &'')4773 3'')477E9_4K^K^9_4_` `r)cvUSc$[R"[55UlO[R"US5UlUSUl[ [ RURS-5[ l[U5S:dUScSUl g[SUS55Ul g)Nrrrc3N# UHn[R"U5v M g7frrrs rr/RemovableHandle.__setstate__..?s'I1 Ar) rrrr r maxrr lenr!r%)r&states r __setstate__RemovableHandle.__setstate__3s 8 ")++km"U1X-"$D "''Ia'I"ID r)cU$rrr0s r __enter__RemovableHandle.__enter__As r)typevaluetbc$UR5 gr)r+)r&r@rArBs r__exit__RemovableHandle.__exit__Ds  r))r!r r )rN)rr)__name__ __module__ __qualname____firstlineno____doc__int__annotations__r rr'r+r1r;r>rD__static_attributes__rr)rrr sp GGS=A L3 Ls Ld L(a JS#$r)rcSUlU$)z Mark a function as an unserializable hook with this decorator. This suppresses warnings that would otherwise arise if you attempt to serialize a tensor that has a hook. T)__torch_unserializable__)fs rrrHs"&A Hr)cUR(aYURHHnURUn[US5(aM%[R"S[ U5S35 MJ gg)NrOzbackward hook z on tensor will not be serialized. If this is expected, you can decorate the function with @torch.utils.hooks.unserializable_hook to suppress this warning)_backward_hookshasattrwarningswarnrepr)tensorkhooks rrrSs^ ''A))!,D4!;<< tDzl;99:(r)cB\rSrSrSrSrSrSrSrSr Sr S r S r g ) r ]a A wrapper class to implement nn.Module backward hooks. It handles: - Ignoring non-Tensor inputs and replacing them by None before calling the user hook - Generating the proper Node to capture a set of Tensor's gradients - Linking the gradients captures for the outputs with the gradients captured for the input - Calling the user hook once both output and input gradients are available cnX lX0lXlSUlSUlSUlSUlSUlg)N) user_hooksuser_pre_hooksmodule grad_outputs n_outputsoutput_tensors_indexn_inputsinput_tensors_index)r&r`r^r_s rr'BackwardHook.__init__hs8$,  $(! #' r)cTS/U-n[X5H upVXdU'M [U5$r)zipr%)r&indicesvaluessizeresidxvals r_pack_with_noneBackwardHook._pack_with_noness1ftmG,HCH-Szr)cJUVs/sHo2UPM nn[U5$s snfr)r%)r&rirjrmrls r _unpack_noneBackwardHook._unpack_nonezs'&-.gsc{g.Sz/s c4^U4SjnURU5 g)Nc >TRcgTRTRUTR5nTRHcnU"TR UTR5nUcM&[ U5[ U5:wa#[S[ U5S[ U535eUnMe STlTRTRU5$)Nz.hooks  (&&t'?'?T]][C4;;T->->?;s8s3x'&(..1#hZs3xj(RSS(!%D $$T%=%=sC Cr) register_hook)r&grad_fnrYs` r_set_user_hookBackwardHook._set_user_hooks D0 d#r)c\/n/nSn[U5HVupg[U[R5(dM&UR U5 UR U5 XWR -nMX U(a[R "5(dUS4$[RRRRR"U6n[U5S:Xa [S5eUV s/sH=oRcMU RR5S:XdM1U RPM? n n [U 5S:Xa [S5eU"U S5 [!U5n [#X85H upXU 'M [%U5[&La['U 5nX4$[%U5"U 6nX4$s sn f)NFrzCCannot set Module backward hook for a Module with no input Tensors.BackwardHookFunctionBackwardzaError while setting up backward hooks. Please open an issue with a code sample to reproduce this.) enumerater"torchTensorappend requires_gradis_grad_enablednnmodules _functionsBackwardHookFunctionapplyr9rwr~namer$rhr@r%)r&fnargs tensors_idxtensorsriarg new_tensorstgrad_fnsarg_listrmrnrzs r_apply_on_tensorsBackwardHook._apply_on_tensorss  oFA#u||,,""1%s#!2!22 & %"7"7"9"9: hh&&11FFLLgV { q de e'2D{!iiIPQPYPYP^P^P`eCQCIAII{D x=A  PQ Q 8A;:K5HCSM6 : /Ct*h'CDs*F)>F)F)ch^U4SjnTRX!5up4[U5TlUTlU$)Nc(>TRU5 gr)r)r~r&s rr)BackwardHook.setup_input_hook..fns    (r))rr9rdre)r&rrrl input_idxs` rsetup_input_hookBackwardHook.setup_input_hooks4 )//9D  #,  r)c^U4SjnSn[U[5(dU4nSnTRX!5upE[U5TlUTlU(dUSnU$)Nc4>U4SjnURU5 g)Nc>^ T RT RUT R5T lT R(ar[ T R5nT RHMnU"T R T R5nUcM%[ U5nXR:wa[SUSU35eUT lMO T Rm T RcT R//T R5nT RH]nU"T R UT R5nUcM&[U[5(a[SU55(aMT[S5e ST lT b.T Rce[U 4SjT R55$g)NzABackward pre hook returned an invalid number of grad_output, got rvc3(# UHoSLv M g7frr)rels rrKBackwardHook.setup_output_hook..fn..hook..sRlhkbdY]S]hkszoBackward hook for Modules where no input requires gradient should always return None or None for all gradients.c3.># UH nTUv M g7frr)rrlocal_grad_outputss rrrs Z@Y1!3A!6@Ys)rorcrbrar_r9r`rwrerdr^r"r%all) ry grad_output expected_len user_pre_hookhook_grad_outputs actual_len grad_inputs user_hookrlrr&s @rrY8BackwardHook.setup_output_hook..fn..hooks$($8$89R9R9D9=%I!&&#&t'8'8#9L)-)<)< ,9$++tGXGX,Y),4$%():%; %5".066@\Q]P^0`#aa,=)*=&*%6%6"++3"&"6"6r2t}}"MK%)__ ' [$BSBST?JsE4J4JsRlhkRlOlOl".0o#pp&5 )-D%%144@@@ Z@Y@Y ZZZ2r)r|)r~rYr&s rr*BackwardHook.setup_output_hook..fns" [H  ! !$ 'r)TFr)r"r%rr9rbrc)r&rris_tuplerl output_idxs` rsetup_output_hookBackwardHook.setup_output_hooks_% (N$&&7DH00:T$.!a&C r))rarer`rdrbrcr^r_N) rFrGrHrIrJr'rorrrrrrrMrr)rr r ]s+ ( $6# J3r)r ) r collectionsrrrTtypingrr__all__rrrr rr)rrs: # Y;;| :^^r)