&VfjddlZddlZddlZddlmZddlmZddlm Z ddlm Z dZ dZ dZ d Zd Zd Zd Zd Zd7dZdZdZd8dZdZd9dZd:dZd:dZdZdZ d;dZ d;dZdZ dd'Z$d>d(Z%d)Z&d?d*Z'd?d+Z(d@d,Z)dAd-Z*d.Z+d/Z, dBd1Z- dCd2Z. dDd5Z/d6Z0dS)EN)backend)#compute_conv_transpose_output_shape)cast)convert_to_tensorc@tj|SN)tfnnreluxs \/var/www/html/software/conda/lib/python3.11/site-packages/keras/src/backend/tensorflow/nn.pyr r  5::a==c@tj|Sr)r r relu6r s rrrs 5;;q>>rcV|}tj|}||_|Sr)r r sigmoid _keras_logits)r logitsoutputs rrrs' F U]]1  F!F Mrc@tj|Sr)r r tanhr s rrrrrc@tj|Sr)r mathsoftplusr s rrr!s 7  A  rc@tj|Sr)r r softsignr s rrr%s 5>>!  rc@tj|Sr)r r silur s rr r )rrc@tj|Sr)r r log_sigmoidr s rr"r"-s 7  q ! !!r皙?cDtj||S)N)alpha)r r leaky_relu)r negative_slopes rr&r&1s 5  A^  4 44rct|}t|tjd|jztjd|jz S)Ng@g@)rrr constantdtyper s r hard_sigmoidr+5sC!A R[ag... / /"+c172K2K KKrc&|t|zSr)r+r s r hard_silur-:s |A r?ctj|}|dkr|Stj|dk|||zS)Nr)r r eluwhere)r r%ress rr1r1>s= %))A,,C zz xAsECK000rc@tj|Sr)r r selur s rr5r5FrrTcbt|}tj||S)N) approximate)rr r gelu)r r7s rr8r8Js'!A 5::a[: 1 11rc|}|_tj|dg}tj|d}tj|tj|}n!tj||}||_|SNr9axis)r reshaper softmaxshaper)r r=rrs rr?r?Osw F |At$$vB//FBHQKK00qt,,!F Mrc|^tj|dg}tj|d}tj|tj|Stj||Sr;)r r>r log_softmaxr@)r r=rs rrBrB]sk |At$$""6"33z&"(1++... 5  QT  * **rc t|jdz }|dkrtj|d}nP|dkrtj|d}n4|dkrtj|d}nt d|jd|S) Nr0rrDr0)rrDr0rF)rrDrFr0zePooling inputs's shape must be 3, 4 or 5, corresponding to 1D, 2D and 3D inputs. But received shape: .)lenr@r transpose ValueError)inputsnum_spatial_dimss r_transpose_spatial_inputsrNgs6<((1,1fi00 Q  fl33 Q  fo66 B28, B B B    Mrct|jdz }|dkrtj|d}n7|dkrtj|d}n|dkrtj|d}|S)NrDr0rE)rrFr0rDrF)rrGr0rDrF)rIr@r rJ)outputsrMs r_transpose_spatial_outputsrQyst7=))A-1,w 22 Q  ,w 55 Q  ,w88 Nrvalidc@tj|}||n|}|}tdt |j}|dkrt |}tj |||||}|dkrt|}|SN channels_lastchannels_first) rstandardize_data_formatupper_convert_data_formatrIr@rNr r max_poolrQrL pool_sizestridespadding data_formattf_data_formatrPs rrZrZ1+>>K"?iiGmmooG)/3v|;L;LMMN&&&+622enn G&&&,W55 Nrc@tj|}||n|}|}tdt |j}|dkrt |}tj |||||}|dkrt|}|SrT) rrWrXrYrIr@rNr r avg_poolrQr[s r average_poolrdrarc|dkr+|dkrdS|dkrdS|dkrdStd|d |d kr+|dkrd S|dkrd S|dkrd Std|d td|d)NrUrFNWCrGNHWCNDHWCzInput rank not supported: z. Expected values are [3, 4, 5]rVNCWNCHWNCDHWzInvalid data_format: z9. Expected values are ["channels_first", "channels_last"])rK)r_ndims rrYrYso%% 1995 QYY6 QYY70T000  ( ( ( 1995 QYY6 QYY70T000   FK F F F   rr0cfdtjdfd}tjdkrjd}n jd}|jdkr |SS) Ncttj}tj|S)Nr_ dilations)rYrIr@r r convolutionrX)r`r_ dilation_raterLkernelr^r]s r_convzconv.._convsU-k3v|;L;LMMu     MMOO&# !   rT) jit_compilecSr)rusr _conv_xlazconv.._conv_xlasuwwrrUr9r0)r functionrrWr@) rLrtr]r^r_rsrychannelsrus `````` @rconvr}s           [T"""#"1+>>Ko%%<#<?6<###y{{ 577Nrctj|}t|jdz }|dkrt dt |d}|}t|tr|f|z}t|tr|f|z}|dkr|dkrd|dzzdz}d}n d|dzz}d}tj ||}tj |d }|dnd|z}tj |||||| } tj | |gS|dkr d|zdz}d}nd|z}d}tj |||||| S) NrDzJ`inputs` rank must be 3 (1D conv) or 4 (2D conv). Received: {inputs.ndim}.rGr0rUr0r0r0rr<rp)rrWrIr@rKrYrX isinstanceintr expand_dimsr depthwise_conv2dsqueeze) rLrtr]r^r_rsrMr`spatial_start_dimrPs rdepthwise_convrs1+>>K6<((1,!    *+q99NmmooG'30*//-%%<&(+;; 1 / ) )Wq[(4/G !  w{*G ! (9::Q/// - 54-;O %((    &# )  z'$5#6777o%%.4'7" 5 ! !" "  rc tj|}t|jdz }|dkrt d|dt |d}|}t|tr|f|z}t|tr|f|z}|dkr|dkrd|dzzdz}d} n d|dzz}d} tj || }tj |d }tj |d }|dnd|z}tj ||||||| } tj | | gS|dkr d|zdz}nd|z}tj ||||||| S) NrDz>`num_spatial_dims` must be 1 or 2. Received: num_spatial_dims=rHrGr0rUrrrr<rp)rrWrIr@rKrYrXrrr rr separable_conv2dr) rLdepthwise_kernelpointwise_kernelr]r^r_rsrMr`rrPs rseparable_convr@s1+>>K6<((1,! 4 0 4 4 4   *+q99NmmooG'30*//-%%<&(+;; 1 / ) )Wq[(4/G !  w{*G ! (9::>*:CCC>*:CCC - 54-;O %((     &#)  z'$5#6777o%%.4'7" 5 ! !" "  rc tj|}t|t|j}|jdd}|jd} t |j} t j|} t| D]\} } | | | | | <t| || |||||}t j ||||| ||S)Nrz)r^r_rq) rrWrYrIr@listr enumeraterr conv_transposerX)rLrtr]r^output_paddingr_rsr` kernel_sizefilters input_shapesymbolic_shapeie output_shapes rrrs1+>>K)+s6<7H7HIIN,ss#Kl2Gv|$$KXf%%N+&&//1 9+A.KN6  L 5   "   rfloat32Fct|d}|d}|r3|dkr|t|jzdz}tj|jt j|f}t jt j|d|}d|jD}t j |ddi}| |t j |dfd |D}d |D}t j |d }t|j}| ||t j|||St j|||| S) Nint64r*rrr0c6g|]}tj|Srx)r range).0dims r zone_hot..s 444S28C==444rindexingijc>g|]}tj|dfSr)r r>)ra values_counts rrzone_hot..s*EEE2:a,!233EEErcLg|]!}tj|tj"Srx)r rr)rrs rrzone_hot..s&999A271bh''999rr<)r=r*)rrIr@rprodr r>r greater_equalmeshgridinsertmaximumconcatr SparseTensorone_hot) r num_classesr=r*sparsevaluesindicesr@rs @rrrsg!7+++A } 7 !88#ag,,&*Dy)) A //)&!44EBBB44AG444+w666tRZ1--...EEEEWEEE99999)G!,,,QW  T;'''w666 :a4u = = ==rct|}t|jdkrdnd}tt |d||||}|r"t j||dSt j||S)Nr0rint32)r=r*rT)r=output_is_sparser<)rrIr@rrr r reduce_max)r rr=r*rreduction_axisone_hot_outputss r multi_hotrs!Aag,,**QQN Q+DfO y## .4$    =~ > > >>rc|}|}t|d}|r |j}d}t|do;t|tjjtjf o|jj|ko| }|r3t|jj dksJ|jj d}d}|r!|s|rtj d|d|dd ||fS) zCRetrieves logits tensor from maybe-softmax or maybe-sigmoid tensor.rTopr0rz"`zK` received `from_logits=True`, but the `output` argument was produced by a zB activation and thus does not represent logits. Was this intended?rD) stacklevel) hasattrrrr __internal__ EagerTensorVariablertyperIrLwarningswarn)r from_logitsop_typefn_nameoutput_ from_logits_has_keras_logitsfrom_expected_op_types r _get_logitsrs(GLv77&   &6BO$?#MNN N & INg %   69#$$)))))"1%  ( ,A   ! ! !7> ! ! !      L  rcltj|}tj|}t|jdkrt d|jd|jt|jt|jkrt d|jd|jt |j|jD].\}}|'|%||krt d|jd|j/t ||dd\}}|r"tj||| S|tj ||d z }tj |tj d tj z }tj |tj |z| S) a Categorical crossentropy between an output tensor and a target tensor. Args: target: A tensor of the same shape as `output`. output: A tensor resulting from a softmax (unless `from_logits` is `True`, in which case `output` is expected to be the logits). from_logits: Boolean, whether `output` is the result of a softmax, or is a tensor of logits. axis: Int specifying the channels axis. `axis=-1` corresponds to data format `channels_last`, and `axis=1` corresponds to data format `channels_first`. Returns: Output tensor. Example: >>> a = tf.constant([1., 0., 0., 0., 1., 0., 0., 0., 1.], shape=[3,3]) >>> print(a) tf.Tensor( [[1. 0. 0.] [0. 1. 0.] [0. 0. 1.]], shape=(3, 3), dtype=float32) >>> b = tf.constant([.9, .05, .05, .05, .89, .06, .05, .01, .94], ... shape=[3, 3]) >>> print(b) tf.Tensor( [[0.9 0.05 0.05] [0.05 0.89 0.06] [0.05 0.01 0.94]], shape=(3, 3), dtype=float32) >>> loss = categorical_crossentropy(a, b) >>> print(np.around(loss, 5)) [0.10536 0.11653 0.06188] >>> loss = categorical_crossentropy(a, a) >>> print(np.around(loss, 5)) [0. 0. 0.] r0zPArguments `target` and `output` must be at least rank 1. Received: target.shape=, output.shape=WArguments `target` and `output` must have the same rank (ndim). Received: target.shape=NQArguments `target` and `output` must have the same shape. Received: target.shape=Softmaxcategorical_crossentropy)labelsrr=Tkeepdimsr.)r rrIr@rKziprr !softmax_cross_entropy_with_logits reduce_sum clip_by_valuerepsilonrlog)targetrrr=e1e2s rrrsN !& ) )F  !& ) )F 6<1 H"L H H9? H H    6<C ---- H"L H H9? H H   flFL11B >bnrL & LL=C\LL  & Y(BFK u66&t7   bmFD4@@@ @F !!3):):#:F M&27;;v#6#66 = = ==rcP|dkr-|t|jdz krtd|t||dd\}}t j|}t j|d}t j|}t|jt|jkr'|jddkrt j|d}t|jdkrtd |jt|jt|jd dkrtd |jd |jt|j|jd dD].\}}|'|%||krtd |jd |j/|sZt j |tj dtj z }tj |}tj||}|S)aNCategorical crossentropy with integer targets. Args: target: An integer tensor. output: A tensor resulting from a softmax (unless `from_logits` is True, in which case `output` is expected to be the logits). from_logits: Boolean, whether `output` is the result of a softmax, or is a tensor of logits. axis: Int specifying the channels axis. `axis=-1` corresponds to data format `channels_last`, and `axis=1` corresponds to data format `channels_first`. Returns: Output tensor. r9r0z4Only axis=-1 is currently supported. Received: axis=rsparse_categorical_crossentropyrrr<zBArgument `output` must be at least rank 1. Received: output.shape=NzRArgument `output` must have rank (ndim) `target.ndim - 1`. Received: target.shape=rzcArguments `target` and `output` must have the same shape up until the last dimension: target.shape=rr)rIr@rKrr rrrrrrrrrr (sparse_softmax_cross_entropy_with_logits)rrrr=rrresults rrrEsi" rzzdc&,//!333 I4 I I   & Y(IFK !& ) )F WV7 + + +F  !& ) )F 6<C ----&,r2Ba2G2GF,,, 6<1 +"L + +    6<C SbS 12222 H"L H H9? H H   flFL"$566B >bnrL & LL=C\LL  %! GO%%q7?+<+<'<  V$$ U ; ;f<F Mrctj|}tj|}t|jt|jkrt d|jd|jt |j|jD].\}}|'|%||krt d|jd|j/t ||dd\}}|r!tj||Stj |tj dtj z }|tj |z}|d |z tj d |z zz }| S) apBinary crossentropy between an output tensor and a target tensor. Args: target: A tensor with the same shape as `output`. output: A tensor. from_logits: Whether `output` is expected to be a logits tensor. By default, we consider that `output` encodes a probability distribution. Returns: A tensor. rrNrSigmoidbinary_crossentropyrr.r0)r rrIr@rKrrr !sigmoid_cross_entropy_with_logitsrrrrr)rrrrrbces rrrs !& ) )F  !& ) )F 6<C ---- H"L H H9? H H   flFL11B >bnrL & LL=C\LL  & Y(=FK u66&7    !!3):):#:F 27;;v&& &CAJ"'++a&j11 11C 4Krcd}tj|j}|dvrd}t|d}|rt |||\}}nt |||\}}|rt j|t jj t jj }t j|t jj t jj }t||}t||}||fS)NF)float16bfloat16Tr) rstandardize_dtyper*r_compute_moments_sync_compute_momentsr rrminmax)r axesr synchronized need_cast ori_dtypemeanvariances rmomentsrsI)!'22I+++ I  =.q$AAhh)!T8<<h-bjnbjnEE#HbjnbjnMMD)$$),, >rcntj}|st|||Stj|d}tj||d}tjtj||d}tj||d}|tjjj |}|tjjj |}|tjjj |} tj || } tj || } tj | tj| z d} |s*tj | |} tj | |} | | fS)Ncount)nameT)r=rg)r distributeget_replica_contextr ones_likersquare all_reduceReduceOpSUMr divide_no_nanrr) r rr replica_ctx local_count local_sumlocal_squared_sumy_sum y_squared_sum count_sumry_squared_meanrs rrrso-3355K 34222,qw///K adT:::I billMMM- $FFFK  " "2=#9#=y I IE** "$5M&&r}'='A;OOI 7  2 2DW**=)DDNz.29T??:C@@H .z$%%:h-- >rcFtj|||S)Nr)r r r)r rrs rrrs 5==D8= 4 44rMbP?cX|dkrdgt|jz}|jd||<tj||}tj||}|tj||}|tj||}tj||||||S)Nr9r0r)r rroffsetscalevariance_epsilon)rIr@r r>r batch_normalization)r rrr=rrrr@s rrrs rzzc!'ll"jmd z$&&:h..  Z..F  Jue,,E 5 $ $  %  rcPt|}t|}tj|d}tj|jd}|dkrdn|}tj||}tj|||||d}tj||S)Nrrrfloat64F)rr label_length logit_length blank_indexlogits_time_major)rr rr result_typer*r ctc_loss)rr target_length output_length mask_index result_dtype compute_dtypelosss rrrsv & &F v & &F WV7 + + +F &v|Y??L!-!:!:II M WV] + +F 5>>""   D 74 & &&rgreedydct|}tj|}|d|d} }tj|d}t j|jd} tj|| }t|d}|dkr'tj ||||\} } n@|d kr'tj |||| \} } ntd |d g} | D]X}tj |j |j|| f}| tj|d Ytj| d} tj| d} | | fS)Nrr0)r0rrDrrrr )rLsequence_lengthmerge_repeatedr beam_search)rLr# beam_width top_pathszInvalid strategy z2. Supported values are 'greedy' and 'beam_search'.r9)sp_input default_valuer<)rr r@rJrrr*rr ctc_greedy_decoderctc_beam_search_decoderrKrrrappendrto_densestack)rLr#strategyr&r'r$rr num_samples num_stepsr*decodedscores decoded_densests r ctc_decoder6#sv & &F(6""K(^[^K \&) , ,F   i 8 8E WVU # #F'wGGGO8E44+)" 5  && ] " "E99+! :  && * * * *   MPP _RZ[)4L M MRY//2/NNOOOOH]333MGM733M &  rc*ddlm}|j|jkr td|jd|jdt ||j}t jt j||z }d||zd||zz }|S) Nr)log10z Input shapes z and z" must match for PSNR calculation. r ) "keras.src.backend.tensorflow.numpyr8r@rKrr*r reduce_meanr)x1x2max_valr8msepsnrs rrArASs888888 x28 +BH + +28 + + +    rx888G .27++ , ,C g eeCjj 0D Kr)r#)r.)T)r9)NrRN)r0rRNr0)r0rRNNr0)r9rF)Fr9)F)FF)NNr )r)r r!r0TN)1rr tensorflowr keras.srcr&keras.src.backend.common.backend_utilsr!keras.src.backend.tensorflow.corerrr rrrrrr r"r&r+r-r1r5r8r?rBrNrQrZrdrYr}rrrrrrrrrrrrrrr6rArxrrrFs 322222??????"""5555LLL 11112222    ++++$      >   8   F   """"P   ;;;;D  <<<<D  %%%%P>>>>4 ? ? ? ? ! ! !FP>P>P>P>f<<<<~----`.@555 ?C8 ''''@-!-!-!-!`     r