From f005d9178e15280554c796aadd041609bdba69ee Mon Sep 17 00:00:00 2001 From: wangyang59 Date: Thu, 15 Dec 2016 15:58:45 -0800 Subject: [PATCH] gan tutorial v1 --- doc/tutorials/gan/index_en.md | 125 +++++++++++++++++++++++++-- doc/tutorials/gan/mnist_sample.png | Bin 0 -> 28721 bytes doc/tutorials/gan/uniform_sample.png | Bin 0 -> 20609 bytes 3 files changed, 118 insertions(+), 7 deletions(-) create mode 100644 doc/tutorials/gan/mnist_sample.png create mode 100644 doc/tutorials/gan/uniform_sample.png diff --git a/doc/tutorials/gan/index_en.md b/doc/tutorials/gan/index_en.md index ce91c80abb..e3841c4c9f 100644 --- a/doc/tutorials/gan/index_en.md +++ b/doc/tutorials/gan/index_en.md @@ -2,18 +2,129 @@ This demo implements GAN training described in the original GAN paper (https://arxiv.org/abs/1406.2661) and DCGAN (https://arxiv.org/abs/1511.06434). -The high-level structure of GAN is shown in Figure. 1 below. It is composed of two major parts: a generator and a discriminator, both of which are based on neural networks. The generator takes in some kind of noise with a known distribution and transforms it into an image. The discriminator takes in an image and determines whether it is artifically generated by the generator or a real image. So the generator and the discriminator are in a competitive game in which generator is trying to generate image to look as real as possible to fool the discriminator, while the discriminator is trying to distinghuish between real and fake images. +The high-level structure of GAN is shown in Figure. 1 below. It is composed of two major parts: a generator and a discriminator, both of which are based on neural networks. The generator takes in some kind of noise with a known distribution and transforms it into an image. The discriminator takes in an image and determines whether it is artificially generated by the generator or a real image. So the generator and the discriminator are in a competitive game in which generator is trying to generate image to look as real as possible to fool the discriminator, while the discriminator is trying to distinguish between real and fake images.
![](./gan.png)
Figure 1. GAN-Model-Structure
-If the GAN model is trained to converge to the equillibrium state, the generator will transform the given noise distribution to the distribution of real images, and the discriminator will not be able to distinguish between real and fake images at all. +The generator and discriminator take turn to be trained using SGD. The objective function of the generator is for its generated images being classified as real by the discriminator, and the objective function of the discriminator is to correctly classify real and fake images. When the GAN model is trained to converge to the equilibrium state, the generator will transform the given noise distribution to the distribution of real images, and the discriminator will not be able to distinguish between real and fake images at all. -The general training procedures are implemented in gan_trainer.py. The neural network configurations are specified in gan_conf.py (for synthetic data) and gan_conf_image.py (for image data). +## Implementation of GAN Model Structure +Since GAN model involves multiple neural networks, it requires to use paddle python API. So the code walk-through below can also partially serve as an introduction to the usage of Paddle Python API. -In order to run the model, first download the corresponding data by running the shell script in ./data. -Then you can run the command below. The flag -d specifies the training data (cifar, mnist or uniform) and flag --useGpu specifies whether to use gpu for training (0 is cpu, 1 is gpu). +There are three networks defined in gan_conf.py, namely generator_training, discriminator_training and generator. The relationship to the model structure we defined above is that discriminator_training is the discriminator, generator is the generator, and the generator_training combined the generator and discriminator since training generator would require the discriminator to provide loss function. This relationship is described in the following code +```python +if is_generator_training: + noise = data_layer(name="noise", size=noise_dim) + sample = generator(noise) -$python gan_trainer_image.py -d cifar --useGpu 1 +if is_discriminator_training: + sample = data_layer(name="sample", size=sample_dim) -The generated images will be stored in ./cifar_samples/ +if is_generator_training or is_discriminator_training: + label = data_layer(name="label", size=1) + prob = discriminator(sample) + cost = cross_entropy(input=prob, label=label) + classification_error_evaluator( + input=prob, label=label, name=mode + '_error') + outputs(cost) + +if is_generator: + noise = data_layer(name="noise", size=noise_dim) + outputs(generator(noise)) +``` + +In order to train the networks defined in gan_conf.py, one first needs to initialize a Paddle environment, parse the config, create GradientMachine from the config and create trainer from GradientMachine as done in the code chunk below. +```python +import py_paddle.swig_paddle as api +# init paddle environment +api.initPaddle('--use_gpu=' + use_gpu, '--dot_period=10', + '--log_period=100', '--gpu_id=' + args.gpu_id, + '--save_dir=' + "./%s_params/" % data_source) + +# Parse config +gen_conf = parse_config(conf, "mode=generator_training,data=" + data_source) +dis_conf = parse_config(conf, "mode=discriminator_training,data=" + data_source) +generator_conf = parse_config(conf, "mode=generator,data=" + data_source) + +# Create GradientMachine +dis_training_machine = api.GradientMachine.createFromConfigProto( +dis_conf.model_config) +gen_training_machine = api.GradientMachine.createFromConfigProto( +gen_conf.model_config) +generator_machine = api.GradientMachine.createFromConfigProto( +generator_conf.model_config) + +# Create trainer +dis_trainer = api.Trainer.create(dis_conf, dis_training_machine) +gen_trainer = api.Trainer.create(gen_conf, gen_training_machine) +``` + +In order to balance the strength between generator and discriminator, we schedule to train whichever one is performing worse by comparing their loss function value. The loss function value can be calculated by a forward pass through the GradientMachine +```python +def get_training_loss(training_machine, inputs): + outputs = api.Arguments.createArguments(0) + training_machine.forward(inputs, outputs, api.PASS_TEST) + loss = outputs.getSlotValue(0).copyToNumpyMat() + return numpy.mean(loss) +``` + +After training one network, one needs to sync the new parameters to the other networks. The code below demonstrates one example of such use case. +```python +# Train the gen_training +gen_trainer.trainOneDataBatch(batch_size, data_batch_gen) + +# Copy the parameters from gen_training to dis_training and generator +copy_shared_parameters(gen_training_machine, +dis_training_machine) +copy_shared_parameters(gen_training_machine, generator_machine) +``` + + +## A Toy Example +With the infrastructure explained above, we can now walk you through a toy example of generating two dimensional uniform distribution using 10 dimensional Gaussian noise. + +The Gaussian noises are generated using the code below +```python +def get_noise(batch_size, noise_dim): + return numpy.random.normal(size=(batch_size, noise_dim)).astype('float32') +``` + +The real samples (2-D uniform) are generated using the code below +```python +# synthesize 2-D uniform data in gan_trainer.py:114 +def load_uniform_data(): + data = numpy.random.rand(1000000, 2).astype('float32') + return data +``` + +The generator and discriminator network are built using fully-connected layer and batch_norm layer, and are defined in gan_conf.py. + +To train the GAN model, one can use the command below. The flag -d specifies the training data (cifar, mnist or uniform) and flag --useGpu specifies whether to use gpu for training (0 is cpu, 1 is gpu). +```bash +$python gan_trainer.py -d uniform --useGpu 1 +``` +The generated samples can be found in ./uniform_samples/ and one example is shown below as Figure 2. One can see that it roughly recovers the 2D uniform distribution. + +
![](./uniform_sample.png)
+
Figure 2. Uniform Sample
+ +## MNIST Example +### Data preparation +To download the MNIST data, one can use the following commands. +```bash +$cd data/ +$./get_mnist_data.sh +``` + +### Model description +Following the DC-Gan paper (https://arxiv.org/abs/1511.06434), we use convolution/convolution-transpose layer in the discriminator/generator network to better deal with images. The details of the network structures are defined in gan_conf_image.py. + +### Training the model +To train the GAN model on mnist data, one can use the following command +```bash +$python gan_trainer.py -d mnist --useGpu 1 +``` +The generated sample images can be found at ./mnist_samples/ and one example is shown below as Figure 3. +
![](./mnist_sample.png)
+
Figure 2. MNIST Sample
diff --git a/doc/tutorials/gan/mnist_sample.png b/doc/tutorials/gan/mnist_sample.png new file mode 100644 index 0000000000000000000000000000000000000000..f9c7bf7ddd7f148eac4fe347e9c38afaa8876760 GIT binary patch literal 28721 zcmXV22RP7s+b$s?gsf~rLXu>Zy_1z>m64q?Lo%{MD3T;8BZO=jWhU8VCp)C9kcjwh z=l#xgz3077=cNC5e$R8S=O3o6sdA8%iFD7NJqOiP6?O6dW4r$&A;90`wEmUu+2gpb zrYNW9`E35ZhY9WFKQdJp-Qx>C%jsx;7%|Q%@HqJKYU)QiV}XJfA8kcHzR{t%R#mAm!H>6P_ut`kLb&&yp~%4eS6Y2RG( z_V)ISt7&V?EtDDDA7Z{lNhs$ab7G~MVsE4KdT3>3Wv;B2meybAwIdf<$xW%0bT*^& ziH0?&WvK}3^Y`u}_Vn`7Ung>ObljwVawd9MOPJi%*;!8KQR8H9rp~@?9aixuDaj9= zeSvu7L{Hv@?R*iCpmtj1P0G~*3f<}FX|Y#%aXnBMBoi=y~eZ_{Iywgq9GIP*Ws3ahAeN}X6Q1&& zm0j(#nXtz5vx7QKI}|-{u1E ztj!J9>(GX$jQadZnjhHHYQ3^HKa_Nfl!B6ymtEX;UFY?s@srm+=06a-z5FdDg)@dbH!k&Z-qlv3ipt95b9YMde0a59$nhgp<#2<5bC50a98DR<>v z`Pk`n>lTZQ?>qhM-dQ)@tR7p@AHThZHta>25{4wiqhx&kKH^pjPvHv>@Ed5q_17;_ zIW;-iG>+g9;l^^yiOtp7*hi1j`hVh=)zf2Irj(SFL=xtrWwvcQU&tMPKH1C@E6aW_ zEKG-wR?xUKO~#=ovn$ze`xtc?LsdSD=bbyprwBOgN<~Y zB-FIDqJG=nUQ7vX)}hnq0@&~TSr8Nuz)g9*zxq=99scppsGp2CUSECa42spNnKoDd zyv`%_^neS!(n0bZsu#0+t&0rv`iI&*iv}Hp8E$?p>q=9j2_4mEr}F;QKu;HO^z2z? z9%nN%GdVfAwx_3A9$z~2#@PmcsJ=oj=6KO?hHZ>aRpCBU zyIK>^EJmEDVmNd2Yo<=Br2E{!n=6(1`T6~Mru8*hy+0m2pb9k|Xl$Hmjyp~lVTGfG zXN%t~BqT&}K8u7TAT1-K{thW2fe;O?+yfjtrbuNkovUR|av9xSG^?}S87xxn-v-7G zaVbA(vrf>^uXMlBC%^7pWt1y@XU!o&ZE|t3+MS}4^kuzTw=F-#Cap@+6kN1X{ z5509fAt+d4l%4L5azj4b8#Aqb42+Bk@$r|f11ieOa!ksXarZ<-M8a3m5G)2o$m^F^ zR@$sxm6bziu85)f-5dkd-gM7C85llX*U`~oCOq(Tg)U<1*JPt+7w5~DFV)jOTNH@e zcHT{8j%X0^-SYG*uGx6e|I}u$xoiB>d?wjk&eAe1G4Y6^aBxV7eaV-vUrqXscb6t{Dc|>8 zn`0bkwl0zo5^6v8P}W;|2ggfT%46YclLbf8+1rK1#l@MK_t;4SYh|evH8tB&l_BJe z1XTfe;rWiEs4z6&4`DPX$$t(O&7cjY>DpOXSWJG5G;CS@^=ofnlSP50jZL0WiF#_+ zqsNam6GYPM7KcW+otTqPUJJ>MU|suL>9JTy;(L6bj7;^TN9gJwqgYyWXrm@^2>AH; z2J+48zm!~$KVj-S`}t_DzLdoWlBgIvGX^o6c4qB#$s{2Gf%}!MI;qNx)tuDICa zLY5u{oAg;pNuw-1UD|YQzA^{b8()f#9XrOM_PKrdRqE2(+RdTT`7-B8CQ;i5v9Yc{ z`vg^ZGb17*So%Y?`3PN~rDWo9xt1#_P<=!95SE<(Qj)#AiDzOX^bW(xqF}(H!0F3q zV>FwiNCJN0(zNS|6L|vz29e4|mQB)EZQ$e zb@_7tTgTBl-iHq#&VAq{*)4nA`e?N&%lLVc01dQQ-uMVG2*Hfmv3~X|%Pq^KH90A7A+P%k%6=ZQ%ad@{wY@i?m8ow|*$H z)!|AyZc@&7XGG)V|NK4K7(Mg(tQ2s`gO(`@i6^)pEh<8_7_ax+Xc48phsop+O!m{;IQ6J3+&`#Asz@Wsv3S{N(afi_>GP>h(i}b_4k@HF#P&I!;uJ6B5Yz z`c^erTCh&-k!;fUQ0I;T&~qW5ej?=|bi=J3y%W#ezJy5{gZN5?n|T)z=fm%PxyH0g zXyx})Tk0tK-dxGn=Bsd@Z{FgSmX^kU>&7idg#umq^~(ulS07xzHxi1c(6vtXu!8H{U`*9{7 ze|FUfIziQ#)~2QdBqTuF;!)9zr>0SJbP)|^!VaTV!yHH4FkFZO$Gf#;rh0OAlyM{9 zmyR-F+@OEuM31aYw{fWz=I3|n{GNRvvA2sNN>VOZejM#e{rve_*|fJ@5e+H*KUbEQ zF`i}~)QB($o9z+*j2ipaIVSS;z=pT9=!m6=1V`~cxd(X84CEo-zJ0@GqbGiL5NnxQ z24rtblKP{d>Pr=-2$X46X9;wD?(A8EmbuUA5lxnoPGff!rbAQTAEKtV5KA=t+9CIJ z8IYi;uy8C>M_YTT-_)WJ%CTaUvY`NJmcIv zaoqL&><3kuA(clv55i8`%n)S#8csT-YvjQu)o zGbtQ5LdOEk2r}LlJ-uqeN9!wkTxNnxdV*_PYl7|aqWnvzt`sG_4X^joxQXM(ucHvz zmZgQL!#;ohygjSGMgLUPZh56cW@dVNdTJ`}(W6W~{$Tls&!0;>4CGUm^JdHD8kKNU ztG&HBJm+^X<(j;L!tqUl5OW|B^Zr8^nD?U_x^z502jvjiYgjk_n_bVzTQ$0OJzk4a)i|x8USUCIoZoeJ%ZBe8M zV5ej~OF~9gFHGL(%-l#s;2#~W1vJKb&I{zmy8T%bh$VUbyZ7%wi!ft81_cp@DN@tW zT)gmy;2rTnrmf%orrc`HQ!blo_Q}3~vlKaXvU}YZR58XisR$#OeQA}BQmdh<-3anS zmEhi!Y(3dYIB7{kDOkL|cfLGj(@xGP5)>f{!1L*rIV**}VK$E8esYr1!KpEq(6Lvb zy&>0M)csVaRr=YVr$rSc|8h(umWt4BwkTTjAV890Jammf}owY@5#QL_aX~V%(9S6C29|s zsl7da%j0qJDw_`0EIqGppAHRm1O^745)t7G508jAUY~n$xni(0J=23AK$xDO(@uI-n`lW=1L&P2?2q$$>u~+I}2e3)fjS0N_rAP zilf&a*-=JK#`0>lx3>dSywuJg*O~bB_3KxQYuDPWPX!TyxT6jBf~A9U=u5Rw$q~xE z(n`K53Z`+bH?xue?i z*90muf}JF467K`mV=dP+xv(&jXIhPU@@dKi(^DitH74lng2&=W!-o$!SKmu=aG0Z) zfGefzI_sPVJ>XUoDWI4)IEKQHZJE-~KF!N(^QG8sb*B2?+L4@euo%fN!e`FBT^Oze zp0WqG1=1_?+mV)|0=S=Oh{%cak(Irx!E=g%oIlPU<%tKx7k>4f=R@u$Q2sT!P4TG4 z#zy1ZzSUJX21howWXI}%`suoiksmP>FEKb6Up>lBQs_7;9i9>s6T{D}_fnf=BbHk& zido|GOJ#!WL`BX+5e@aBlsiAN^9H8O%^BQRzv~nj>F_qwCw z;@!J@{AD}A`f_q|Vq&PXdz%{^>V)T}TH8_=uN#J#>5a*O0=GrH1=naOUreVPDaW6JA7Qf_37!K_+f-{ z%n~j^*QNiwyS-9uG!OQxATPf#@;+OSUo8f#8&XE=8Evo$hd}}GbZMf7f__)o=E5I-~|+@-N8}uCE3i zUNzmb6fv~0NKxeMnsqbE)z8(RnVK3da}Fen)B}|USg+Z-7ZKqa5Us>cMN3C#ZEgLT zHuU!G+xzxYxUIT*|6R2UPiZz5e)HzdZbi4X1;-f~7<>v)a)68g$hru%3xWq9T||k| ze4$m#Ggiv+pFhbeie9{U@y>jb>TV8X8cHV7dr3hp@h24cV-z_PPG5hVoqau!_#jxy zorSLwW*Y&50A0^H<`0FKtEs7(9l}-MK}v~>5A^nyfR@Y;y`>~4cj$jJ3FKr7$LJzZlHY?4kT!>DdP8flK4kD=X2tZKC&b9;=xAB?rKKe&SJ#!_ zzd=gG+H4LSuL80Xpg1TV)!oGnaRG3txuc`nc`}41f{!G)4s~gm+t<_6)BO4C6XW+_ zLfL~3efA}#TQ#jGty>vh?D*gT)cEg<7c_RC_z*TRVIj`a{Vz*5i&^ADu)H~-V|{)7 zhPQ9d;z=B}OVw0-}~tzdOhzO7hi? zTkYeA5BXB`m1SknY})NhLPA4N803%P_ngiyY?Is8`mgnM&!HZPr*9M4@43#j zJ3HDlj)_a#_wW@Bp1t)$k7xAH@^Zkti^4}VcxaaVjzlVJr=u7srb@Qy@#kOG);9Y` zX}AC9g^TP|%x68Rm=sr%jRhV98Gu)uzVW5UQe@UmGKx8&6nN#KQSSEk=Hk55wD9HL z)7Sg-_-JXB4l)Sem6DN(Zme_~C+JKqEK@P4a69JXRFL#kIrZeRV|KuuvScs1X9p;~ zr(~%t$LoU87E`-~H@5FHo=Vf51rmn(@TK&IF{EkfmEm5O2zSa1!G+0P9SC>k23fEG)Qa$U-w{l4V5 zVUDIzidT5fOEg|=AHlwTY%;#&n^{E;#O%Flnwr)kaxLT|`0aMec4bG^R83d*3k?LEot-Jp zq<}(=iSeoxW@dgmR&nt4KutO=OG5Odr=H%UUVEQk6GT!>b5N2WcMv3sScMNdY(sOD zC}nK?1$FiAWWXK0{WZP4zlSS5xYeqB|2=JbrDq*L8e9kX4$N?8syVJZL(37bVP0|T zTdYyeE+?&Iz11$R2bHHlRKg#Bdt6uoIE@_T3Wc&Tt26%S#68D zeedKy9@@QUL?cVDd8u?_rxy2xavP%r(D2*duqh!XSl-z3*<5?beS5ldB+EcgOlX^iz$Ec8x6r(v7?QOp)A{q~ zO{=_Tb6$tW?XAy0e))|y?eERS>LT$vz?Jhnu{Q?tYlns^G1x)v0wR|4s{V$jcFj#r zuB^^pTxxPXwxGibm61QL*}la0ckktP{$~VxDsG~+i!#{&F?##Rr?nT^bnyQaWo`r# zslMJX6wwfqkkIt})WgPlXKzqg$)t7%F%=~xiI2Xh`8f5X&$1v=4Jr|fr@j3;Jd6nl zj{=jccES)77%qWRsHHGNt96kgr(vPxfyPcF^7^s6S$h0QS9#xo0p&|U%4jODyeKQX z$LT9$q_RpLHE59E=KR}7j~)S54`^P5!T|C=1#oGU+vgzWn$^=hvw~-HP=^_f;o>U| zXM>})SrA!3xeKoaySZ7HRaVOQZb8&Asq#)}n1n8X!3Ux6mgE(4^YMlV`lOf_XMO&< zW!_)wy=;1N>%fZDV6O3*0Co(~T;sQM*|&MrMoMkFUmfg8?XoWs_+h=Yv32a?t6c#b zEvUS_T$^XuRun8B80RafHmY&1OaPDl3se||P>Mr@FZ+LeBxb(YyK6%QfZi6vjnXjm=_%PElch?-6=-)8 zyqg6AwxUTe^5J%*815TrymOv0>6A?rnp}d$p+kpEzLk}giN6&fj&2OpOm zk0(P9=&sM=3;%1H_+rOV>%_`*>FX!vImhFzJKw6u`Mi`AWeH>+iA50Zz2_vFVk-en#h+fU+>A>yW!mX_XU6AuaW1MSRDUyFt==H zoOGWvNY@<%ME0xT(5%@noxsoCyh(Bg4$1{p)fasB=s7!X#?lkj;|mL#jdrozqigcSd`kx?!@8ZwngZN9H>9g}sl1cZcqK?ec7Al%gq zw=zk(WnRte|M?r`^bh?~T>W?;jsO&Qh@!#rIwmG2+S*RrTN{+u_sLON{^-dnJ$g^% z)T#1T-j~{CUcc?3y+3>fou{~XH9N0rg7s7|`zN?PYP$wY;nb6{d-}OX33yda#c&7u zPAtsMmU}E-{h|@z36YUh&LC%~)T)Ipu$Dc(1;9lI^kn8ae~|q6V)c&DLsj>Mud_2V zq2>diAfxZ^cx)_BX(mbTH4eKddL9fNMoLmr5)1yKrqlr zjZs}ghXUAvYdj&!y1TmrKrCSV*}95`+4f0jj`9t}f?T%>Zj$<0uOjG6I%hlF>?PvcYfIrRv1UEG~xxI7AgmTce zo;-O1E$iRLOfpm;YdB|OQV&EYrl*OGV^WQBx12ahKI4tWa9`qwEuPxEL#Wh+pCPa6 zFQ}H3`UJkL{7dcZUK(D_oaL6awqq5RR)7GhQ=%>G+}z(F@Re?{on^(q{0Vl-t!9sY zU~<*K0GOMR)XmwssiR|;%l;1L8xyCnlR--#|K&*%0M2~k+_@{c`j}|R<;>E_d9WAH zUJYOevNkp`d6t;Ci3{$BPp@|%JY`prO3+v{wB8|uPcwGerKijeHrxD83SWg?HxD1@ z4cUt9y?)~HLnJGGbBPF7CGZN$O)ym#*| zX5vm6Kyv_km72?%Xx=7K* z4Vp`F@Zc>E4;w*dDq32?L}k2sy!QkxFG4veNw~`=4bblJc%UT>l{sI#diA*mPoS;9 zfC9k5wQJV^LJf+w`Dh8)e!yvXq&|fb3kVEkv7hOX*&%Pb0oVqGWNCT%wSaSfW1~u5 zKMmoy?6wmSubr;Om>tw<8XgU&o_&%I1H!_kuSHMX70oX$eiTuCQ0@UP@3r3R%|DAp z`nT}+3_kUj+MVZ#l)Auyq<#MC{bvom%E*YdAy4S$uECjL513auYP zm`)E~ceVx}o0!9sE?e)7<@9Sqr=9d3)k4;15ov|tWmjOpO&yk`k&Q!A?XyX@762L* zw?nXU3=)OYwBul5@NKRekQd4FrFRuHM?eY8IS+9=CUW8g4PxH!a1e?fM4Jy_(z2iQ zst-9BWuy)H=Ye`U4h&@21y~6n>h>ilD#&;TNJ$HG--b-CLxgCvUcoIH6w@em;jPAS zXZ8Oyv@|d&cOml3%zie=`H)0JW1y4zaqfGTAEcrPx!W>_T8bnmKL&1<$2zzfdsl#m1w9rx%&ON#`F`4I4^IyYyr0*i%9{$ z<`@=1qtd0l1;R|%T!qdj>v1%;$-v~(-XWQMQhIv&jEoFec(3K-{3q7|(GL>ut2`dE z{crO~jm8e%SMUFtsng}++j#uw2fO^=!OB+_k@^pgEkTo@#p8T(Ff z(QaC|w@l5<#6NmO_&j)SZVuEMy1Uzldxy|Owm@kfsUKZ3+iTu}HhKJ+B&rsofMwy~ zwb}0dwh!k=Z+|&-8rENU%IEIx+VS2(EAYpxE?b@pPYDejI3L=AJ8-LwRC+|~UnZyQ zAqi-Ge(s1v2F_n>lU8$%Ki&=<9bMh&Mwpf0QnTR$-c!(rH2|Ayr7w9Oh!4!RAL(6FjYE}u>h zKuCac3PAewceDOcNWkpk;tc$I_4FkJH>4R@St&l8$YBZE4L;avFBv{Rcc;%*l;F2e zCn3h*utfn{vdC8vjXo_{w^y&Sn(&;=p=D!Yvj2HVf$GEX@Ovl)&J@Mgy@W;}d!$0^qWTGa~sCjuC+dW{-LhdCA zDD(WqnK-~jUay~bHFr0M(b~}wdb7kpU_T|RNwMw2`iZ$ZWMrOkF(0U)B7_q``OUJB zj$nw!V5mkvr+XHu9kcy=rgrB6oyRCc2&8zzcbqS_UUMZ8H~xY$!$cl}IauH{z(FEn z`H3oq+y5V<+yhk0)RMzR~@ArVJ zk52SwZH=SYg98QEB!o1jlPRB9LQBF~KtD<^F4p6G zI0y3wbpk_CH%nAV=q6OvIr&&&@*|?QRKfxIRnL`lR`GaXtgj8)VEPXYNl8foX7xrz zNhNp4xK6iX#yISn6Oe>!dHtY=KDK6vnzXWztzBgnb6_B=yTn698l0c(l5^tTWSJr- z2gwA?j!V89|7NDxNnGHyh_D#z>V7VFouLynk~O2hK8_>|TFHK%?}7U%UF>Xt5^Bw0 za4S=#fi3E0JsKX|Gtr|>d&|`|8}b<(kD4Tb*rw@@hz1UXK|v(??hHE#0a2Un_upPZFLl9*5gZvy<02tYwf1*4)Vjt%G98rFkH8LahisPhGS|kfkwILBAYDqP6`%DZv4Tm2`eL?H?|+PDzmY% z6&R44hoY1%>#o^mC{(RakVQ1;nwY#phXU0?Z$fBIcvf~(iX%LQMbZskQ9$fEPLRaf zB?E&e1_J0I=x4r;+$YTnz7u2vyw3*^q^LDBD6|Yy z2sEA-B@*DN-Fg3TQ}~zkCgbM@H%XQV`JcUsX{gP%?hjO&yy1oWq>~*B44VC-QZh@P zX1&;Ue)rZFq8vLrI}gt^X6n(SM>Xkg%;dR0W+yWuS+8P=Ja7C@v(-D7LU6*ZPFTpZ zW9ZO1(Wt2HO|iGHCvWf2aX#`pFO|KIt>QNEL%Uv%zF85L8+SuPRnUdYRl58>sE9KH zy8!GGjQEk<9WA|)SX4^F+*rM}eE#04G(Ppb{vA}VW{1-Zh0P`5O;eC8Xu%ixXkq@s zBL<8w|JM@Cf2$}W;yg^q{Rf$f!M*8ET;=BF{rTG~u+}V5u$_q^WPY#+0uVoKI7I%; zmwPoYEYJ|hJ;11WCIi)GWd{0_O;V4Hc^%=sfxIR4e}Q)+B3d9LySTcNk&^c17y?4R zd-rbp*NwMlA-1KarrwZ9YqfUO)TF5NJ}C>uW88^YwCI19VRrAUSFc?3TyTiIm%oLf zSw=<%7b$6LcR~R>5b&VM@=B%q6z5k+M`-TWMU2H&x>;wvR{0{jhj^2{Q}p;b8iH@x z+y1U=YrBCbReJQ(vGC9KB{<#y6S&kJDr^4ml-&oBr6N1Uv+RbwdM2q*l$Y>p#A;_2&RuG1Q-nE{_e#pqUZ)Fhtbk zhs@okzsP06d6D{?hHS`gjte6S>f-HR6T#Wp$A6Y`l87Mn2kx}N4P3ecjD*xBfF3)6 zQEHcXe>T;TBM_O&oxUSeYs&fXYo$kC>$wbMd*xqxE{*vsWW7JjN~xrS`TOb9Cqx>w zpApKXrKht3%ce8i8Hi5#FYm*%*Ee9*tZsHQ}Mdhlw`wG`Lqbwvfxo>a`z} z@DI3-bG!QI&)X9(MgG{x%E~%BpOmA5N(zkUA9=rH=IxE4$0Og9?p$+U8bb;}$K0B& zo{deH~YS(GL~* z!WV+#tsjpnE2Ri8)S4b51jcSeyIl1;mlJ|wMs`g9SSgN+*7LJnKYtq<8Upr1aTvwH z!V@sRF_>q6$xsVb*EHo4!;N2>ic;DIp zBC2DJ!fMx`-cwz|1Cu)i;|#InQNNw{044c$#SVkZ7}xoW$7o4;rakfUVMA+XBqILu zk0WiYJ>eAZ>C?A>4qHtT_5(pmO38#aj>q5Cf0BcPgdoqe}13Rlejo5Q#WmGD!m?)0t+Ri!Oqo?1c<^qKuiFofe2Z{(GNI_ zwSN&iVU%>U1&FaDQ_Rv6a~SC4@LwCBn!03S;tlhabg0RThMbg?pO24SO&96Su0M#) zfqDqn2&;h0o7_IQ{#v&CMP=zGr*-%S8vAIJ^r;yhc+@QDrD+) z_w1b3F+w@Ce3ckW4UPNrdT%gSy}i$Be}>4H83Z94jmdZ_ssIX+*FDYbw8g4&-(QeJ7f|YdvpJiQ{Mw|ABtXEhkZ{=n={}p8>(~2oATEMzXF7ab+3-%+ zO?dqHBc}{g<1a|K4SkJi`q@dW<_8BQAS<5z{DnZN>qPYyby>+{p)1YW`{9X9A%+Mi z%>GMpRuUI`h`Dys5%PH`p(-0?TY>3_MgM1Ox=Wkp7c2|A5aFN_Df7hkkbHDSU8kKV&5S zI2cmS!3ZG-9oA)cnRY880kSN#KIb+}R z9qwIe?2*P_Nb-VyVX7rR>cI5CNz%{$4dizCW4hqP*4`+`o73rErR7lPSG2P_uP#68TKDk8I70 z5MVjCY8=E)t^N!N4)#9-#XzrVt?#X)Df{OF0|A8Gybq39248GP&w6sUSRPLe{Xe@` zJtj>zi;yx1nVI$w=jW z4l-kV%%z9n9_@D08L#&vQO8i%(6)!QHP4o8mY&F<&6O)xb{#rb z`Rv|>p|^R>$;i~JC@MnDcvV(r@KPJ%Z6vcqq2JE76Rn`5`YFNycGz|MZahp(974Hb z*CZ`1?Md4-oY1(-dk%l8t+Bv_fBgJ8lEVjy6`%$p{)(u4Ke#>49XfWoi7$o#7$<44 zJo=u$Y>lw=VWlp355SoS^IO~4RNr19{&b#I+S?%gn38b@*2u{_F;nz!r%{s-V%QLgJOjWDCZiYm_E<#+Mr*htr>UuFzFd7> z-Q9OVzvGZdLHBx*sB)iSue+d=o0XMSoT-DEPFzA6ywUw_db+yQ^zR4 z$kKcLU$_DnYiKp|_wQd&*0ICK*x6S`A=D#PlA98xgV`gdZvaL&W zPGY-$_ZUOSg;^wc$xpmTi-fYC5VW@@6B%@x@5OxKSeIDySWoiolJHowP6ci&v2|X{Alea?1LCul;)zx72~Nta*5MX6L%U zy(J02CXPIG487OQ2I(ytok%C_OSZys#Fo=^d(s`e2#gO_Dfb)@u#9eyK)n2~s1+!_ zNWnlWGHjVe^905WKCyqRUqFVuogLuCucfHt?>zimT|g&t#JyZ2GyXw!)bpd z1nzaqyPIaKn+iNtsC;o}9c!EFcG#24e-OAyDq!d=cgRbXQbOQt+^~Vy6y+P&q!51EZBMoS2A65M+^^Bu>4d^LP;ZC&B|ti<|8JErG8Un{#2fi(p#P?$VA8wpaR0(@x-L5TugE7E9~=y4HB^HK%@6V#f91? zgB1{6-I%L+m-%QBfc$|J%b)clNa^KM9;FW3SN;j9c9`=o?r`WT=Jj6}W%{z8Vw^=J zL4cDpfGe81rlsZhv14J}YJDY#_af>RBv%2Xf~m~K#Z~($B0AdUs;3$st+dxF$;Ts{ zqwKyN{-0^YV242x1?8H7z3|4Q9C5U}G#!0B%2eN=uR}0L8Ef3+LJR>zwXVJ%TLEQg zXMnhFK)O&#E<6?8Sg2ft{)3^r_QURMfNpYg-KD>gl^-U+iOy z>2|i#AtEHoKn`5K{yUACnbeJ zdqqTIqNW1`wQ}`O7#4oukbpV>y$lzL@DCgmNDwv`)S}o`<>dqP_}@E?*WujG2`h#M z2LrplMGCO+_k2cBShDeY|2&0H zr6mC11L~?m2zl7Od#YMmew%AIz>@H^%E_wSDv zot|BMG)VF6vo&uN9rxuydJ@LgR;>@3XQjF`9qu#FlMXEoS*zvWWatict_@Yb|7&EA z^wU3Ee}x6Ne@X9LW~P>|Yg)GrEKGPQ9`)x(?{zFvtquR)NtC4uI&T>uh=cJ;GwH9~ zFA4qq&u&lNW&bn-U7_vF=o1Ydv(IoLCJZ*?q@X4BJ@ZUn$N1kpqIsq=Qlri(#5bUZ z-Ez$?vb}**#UZ=4>!+?dAcBbM8?E%96hF(pjEvxi4GmDf$Wfb+!iStGEKNjYhzJSu#jYXqg9&jM`%#Dqra!r} z`hEPnc8TNY^Edld8WZ;(CZ$PgwyKk5MfEqA< z3z`h77=c+0q4Hf3t2G)IAgY^y!*#Zyu?}ZpFIxj_AzD~)%2A92aJ~;VQ zf%Hd?82d)MWNIOFP1tM&cD3ef+h{HYz4zX2Oh;_LvaB4now*I60mB^&A?LgvwYOn8 zB(cH5xp?eUbfa;;*74RpqlgV^Hacp7hnrdfnYH&5I+mXg;n8>j*RRLtl(K%THbx_@#y{uZf?Sq7_}I; zo43vp%0YENQ?H#~`}2pAloS96;sP@CF3nuT6wD+W9?9Oz2O3{dAYAyW2P(a!#{#*I zJp2`_0wUPI3IcQrS3yS7z!8ZGx0It<+tOCXq^i7rH|xj~O5?n5-4;I*)-VCi2ei5k z!Ucnqgp_omu7Z!2^v~ib<;&lQ5dHe|2QaNdk#iQ@YlFh-avnBReDWi|Xz6BCBcnq^ z`VU|Zf)T6)`3g-)bcU)(l$>I?h1IS6d07DQPOch-PctiqMHPKn7x(L?U8t_PmN~vZN0C28?4|>cA-BVhI#N)&*be!DyM&xcuLG$Epot-4~o(T!x>JYZT@q z7q1TF+}j+2$Ao;GIW8y4`WSLZY<^pZKK;UuYnb@}76`EQ3T5lYq=gRc@1>rRHmR-IHVvP_ukLELmLx*aGl5zp4k=C=vVv^C{cfpgM zPEJcWhvr2h2^xr}Awr>xi;Ot%V_^YyA_aqhdzdJi)XeOxnwItuv;O*tPSK#y(4x#f zqT7LkgM+vk*h}IRA1~_Y9IC7rNV;>0!|ccT^v~c`!nEQ}V>KT>D5iE9n3(K_udozu z&fvuL#fTkiFkf(@Kww>6Tny&Yu+q5arvw%$x<)@!kr3i1=Xc2w`(MpNsuBCQb8g1C zI(i@6G@vu{Y#vYe1S{2#`%X1j; z&?;R@XtHAilaE*Bejuw%*#|UXa6Yi@3=9>?-M^SJDQ>*%vNag;#B$hP01Vps%M#I3M)+6e}w$ycGbZX8>J@nb+9u;j*cI@E7GImr+D_`0!C)UNHD7 zv=FFwNQk0C?{3e76d<6q>)9U;y_JZ8aHGTsn~r74(kQrJzXq}JrL3$B?q(?_WhAn9 zY-)TS^Qv|2j0^#GQn?)fx z(0!ABdhiWjxR;mL_OC`Z{p?FZ@kLda{(o&T8+V14W@n)?HgJHFGE_5zuh~)r*tZR{ z3@2e~>h5)>n-^wT4w6=OZ;_x8%ENvaC@r!ybR-+tNT71XP6PyulwS zv}w=xD^61tT~ZSx$1Fi@sTAh6a3i|x4VXl66zZqO35Vpgb zsrYUMyt5)i6VQwXp!#8j;tohXK4lQmK;VnGCwF`c@Z_iT5}c`g0sq>Nl4s4rh*lR3 zGQ3Xj@`5lfz72kA=(Y*G1}yRl4gs>G*$E z*$$No8Wgl{1RoEQWzYXDG0>;;f{&k<_dndjLR|?#w8ZG#^*#Y1!89F^g}OR<+X1wm zIogEMx?~wY&6sb{Xr84f-dUSF(8yI_&}uZDrWT7&fSga3PWmj&8r@hg?6CU1OE;_d z>}{rVQdlXibepZy+=i@>tAAXMs&#hoIP{?-Vh;I76^XFgUt;8gZRx(^3GpZer3)dU z)%>^Wp%&B(U!eK`m2;pZym8W;L4W$$A*6+9F9S|88@{n!+VCEe;~}Zj48G65Yjwo0 z!w7IT@cD9CP@6#ri9A^HX2N%0hi=&a7nnehy~WG?LXza+s3!6Q0`oPEjT!m*53VX1 z<%;1qCeOYAQxD-%bwCcw95k#HeY_l7Bh z*PEK3PsI=fpW=feLF`RK%)E6Wh*lrJ{0U=Mg@)gnfEZl?jRj@t>FH+?y9&4Wgo{{O z7Z%-sCNQRuQ$oiqGCMCo>F2jgK#&)T7;gxhx6gZ^;ReN*P!1k#6q}4RJG&VU4^kAx zltIdIf|MjA+Wc|KsfXqm$C9>i{ZBeSoMdhT$2!lARh*}>vDc~}soWlDXT**UJoT{y zme*`0-anyoZB3G**KWeZX0JZN za@zj(BWS2wd+v<+yg*mRLQ_UJ9LWsHfq=O^U)L8`!q{k^@u14(!s zBa|1p31tg?iHx70_^ZI&IP3c+Cw89k!(!}{aI)9ePPltXC8^d}NBm$&^&^tLsp7EI z<=~UNKF}YvSMC9zykv{xq5xK|u)<#3Hi$?3I`uYA-bR@S@;-~9Y`zA-AfaOWY~2<| zSZ=dkFDo0`0K?4>M?C7Vdlo4IMZw{;nvXUX1ONM^2&^S!Xu6z~X_X)iAhkJL zb5~^Q-m_rzONDa93}Ib~>|g|Hj&6LdotT|$A*KC zkA&Z4*GP_IF<}t91Ih+ll3UFsD^De`-6^KYlIpR!SR!!3dqmlzQsDySLtYL~!I|9L z%VJ1KP>X2HReY-fjv|!HUBNWF_iCkk+sjohDSRq{kG2n=-;oO>#*$X5W(Gv@-JA-N zj@bTD(%Cioh1!rLL8d_@A%}c%pNEtjH}!>NX%+&tljO)`Q9W22IM&m?t6W1&2C9%u zx{38D<|8<%YL`dg>OrZUMt;lxS2TAMz)>IBC2neDamYWBgCItnJ-9U5)qYT$4FNPP zEQFezEHLm!j8L`hI6uEd1@Zq@_7Hm}+gL=ELaN6W$$><}xDFhid(qLm=-1a*n$`xP zTiCh}((!c#y1)8k@W}jUOOn@ zXq_-A+cTUPAMdfV?IRQ#4fIf_st~|;#rkW1{~t8}{K-7SqL5&PHu&1m&Isj@x&o=> zcim}mIEHax=-?pG6I1lXiSmm-;vB21ITcwX0f3J9c91dM8yIZ`23z%%GU9)qPo0F# z`)Lp#!vX9AVt+1qi(m9ONLxm?o#@l$*A<2Cl@%4caYN3B`S-jEM(Gb85hJ|Wh%LKy zO8_z|q86hwXRi0n0m!PSyMWL=2=bj9IN4{+8=eBOF-_&IM$H9ShmW#$QT6R>Z<|Bg zkuB4u27v*JnolIPvN(dhS#0&a{9+Tg5))#S%PQmR%~+KbZh8mwja%(Yi2u0w^Tb3c zmv5>I?KjBlVKC!g1O@~k^>ah(#@Dik?DxQ>mV+GQ40nSpeY0+OH`i3w>}K+I)!Xg# z7E#S-;)i(nDt*e)H-m$N3FTm;aSICzb8x(@3sRhx)e)7EVRO0A(ztbvp%mpC+mxow z*PQ%#?wwR;xm?Z_SPiqoIP))SS90bl5m|Crf->9 z*^SCjmg3kfafFyd9JjJ<#}_P2GP-y7J~Uw`0`EOXJ_G6;-tfnU2INHSA!Z_Miy;3e zWV9H_AAH61^IypOAq_^xoA#jFI8d99n`8p=YVWsP#D6NiH$*H2giW)6kT9r@9t~zW ztYcNX|G;=*|r1HBcHAMFRi}3v;<~SBgRWb zGA69xLGNUREJ?RmvIc)#d$LURk4J}LFkPug>$RWV^27MTm!~iVLKd5t#iMYVs6^-x zK4j+ni*H;UR*Qjlo4a(kDayAE<{t!gBFFe%dwg5wRj{J6avcTH%Q-BfbEx1uTf1Mu zfPt9UMjC8^KN0Ktcu`TM`6wc^zE!$o@h!)%6aVM52!}D+(~;LhN>$R)14L&uT?A%O zjVX4Ob))4A2Z4Fud#J={7~ed4Wn5fYdB@qgpuyrcBxWE>fLRzm^#Q?`r1yQ*fCryx z{B{B1<{A!oG&2v5kjXtjUF*{JA}Z$JN(?rF|uVOk?4>Lse}rZkac7f84)3s9U&uSv}5Z+~Yl7>k}Fj1i4tqk3)Rgp+WO5PA#ZFW<#VbX-`DZ+eHa^2pEWHvm7ua>7KS=Dq}&H8&!O>h2#3)E zai@t54wHwv|35WsV#>-^GI3iNzL<7S9y{f0@us4nC0oi6@K3cv57U~4&508`!mJBz zY3UlPK&n#)K{eyil3m#rmWIJE1I-A5LH}ohh zlL3b|ApQ+fSq?#!5^PeiwgQv~`}K7-gc8|r=Ux(Q@;E#q;_6+G5;lU!HoKdy$dC9H z^r9p;KqpNHI-oiQa)$LkLeZYQAbl3-UF(sWC=)Ad3)qrl7!SQ-)9}%u0TK ze(2x995in2e9K1^w=^SjI=v1+Cql~ZqnI`l_B!))KS^V%tFPa_d-rZ}@f)fAF>QpH zxLv}IAQ5m=0M~85*F5>4(dl{8yRVqKetejK&~n$|mNyJg<_GcsuiI3#qbo$@CW5;< zGY|SNePMzh`@=U#;fV+k)1*}ti;aQVKfelUMI($LMnX{LoVh4q;ed1@E^|*$ij*6K zZ?LpAH|OGV;)8?7{Dh*x*ZHXoO$HV+2YEw@7~BB-%s0To6RwBwDp)x2!jgI+v{>sq z_Xws$bq$SEIY*O>B}3w4ACSKQR)K>=;o2A*A6aDAt{(s+;3#mre3^djG9VU!QXBs} zFP{jdfC@TDq(wgh=KR;HsB&lbXzA$a*sZ;7YO+SH!W#ySOpWYCGT%`d?u6~8o1zKB zq&zmy5Tg=QUGlT$s7tP{g}@CHjV)F>zo+32gY!u{2V_G~-hr8yRtc60%8sA9Sa()H zg%VDZM;fL1P3Bv1bX*Xde(>T& z;%g+%xQ%&XRlE_Vj(iA7NpDY2*JF=BC}F3w&X3=48evy^5i&4Im$gRM{P`E>vM?UF zR1H9MC=FD(cwnPMwgMscx|C{4#TwP!eH2n;b)UvFZeq;Em&Z-KRgr>Z!>qsN55onP z#6G`HY@i|{RG21w8;Fg|<+(NL6Yc{vMhLP2p8yHo%EkySvjrsWU=OgF`b^~7>>za>2L0 z@2HhrcqV!H;VO^}K%`&~;H^Ww;uBAq4{~Gzzj-7lN$G-hYEYBnK>&I1eVvaKCvzLb ztd89~MR~vMh5?cPXl~Dw9w&Nw`UhS!=8`gwo|1p(STpC@R4+WlwQ+1PV0=^XabmgRQtgu!dg!KbQg{+)8G z1o(Fj@^c?(l%dj^NxaKavpwjEcJHHE%;HEhqqq4D?tXP?B#N!xGM({Dm3ZZ0AI(0c zo+u|HsS*z9WE0Zs06O~5^XVGHr*8eOmb6muFqviP^-Y;0g{83vpZ~bIk^Mzb_KTu} zFRt(VE7(7w3Wo+noc=P;I4xu!m?WOe6n-~8{~%7-`P}K?w(p_u?yW_qpUrQH9PV!G z*<)bwlsnYtquS$`>wiaF7XpU!=Mt8U*pHfuK<1|PM4swtriR|zV0>fi#+Kq%lgdV;EC|-3FV_V0A-ig9rBpMn*qS*T6QQAFyeCNr_xp zSpfocDz^tC{1cYiEj))Y<@EpvdSD?BJ>?20f|V7_@}RWLQ%zkyS0;uFf$T!W>kH*B zcJvE$DQ9JQwnSq@qRBj_tRh}T^!pA`(N}G4Q7_xjl;H7cCKQYi;I!BbDlmnmp`l?d zg-qP*(6>2eW@d{2ze;3F)+2nzI^Pclzsze)ktU`X$j}8kJC8t-UlFH>9t)q;#N917 zQ&2qY4t=7fIeK^~NxUqI{|V-bxj-&Cxdz#fT_Q$Ycy2!?g+yq>Wbj}4-NeEPw zy%!)K&`ET52D?{VD*|-lNzW9XcGcgJPvE5iyr`gX8ErN1S`6x_l%yn9FMu4VE?it( z?SzIMxu#}X_XW=cA+&NGSCp7x@v4!#!93Fbv?ZMr4O4oYnt5C8%eYa<`R!fd9ZxLH zVEYKkJH&qSi6Z|#SjHXUf{xs=fjQ?${V>H3vL0vUnwWJY1~*`@l3551!fO85c@mAP zF>!~C49muiMQe33!CfgkjHgE{j_xP~<_m~ij>muQt(0rX?EHMX&Q{U2aeOn}I{VLs zYk>w55OO@~cG|m;^2&DTNWne?fFS8La$V^I29$cZS->`E(G3vxh>QGD4fnClLX-fA z9R6!82B<}c&Upf#L}hpuKRq^f6;lz*IVp0iR@ojsGr;Sm1-Vln3jC2H$Lw7~mBV8~ z53K?hdY)KdZZ(#a<@~Rw%J+Kq=tfQ`ci^T-KD2`S z3T##6T8pEiq@*M)tjz=mGcunE5u4I+$cJevvVK zf^a&PNXFvfJfVnG6l~%udpcpD1QW@i)#y-+a$Mx)&yRr-z+!m)7GdE_{;y>fWMF&w z@A^J6DgF+!O9L=#0VdO8fAGrIHg<)?$Cm)ZmF3P+JMFFJp&R}(VPLb!IlAyvm(cTl z1kF4*H;2ioe#|x;Wj>GV*02_3AuKoAB{8ygH5rW&I61Mi?i9b#qX&baiSTgdu&O*=#*FWxQQ^T+8 z-rfYS*WlK85&qEAmscr%s}x6D*v2N>MLqH}L`lpj#;oNJt^}%O_=0bV%viUcHiXGp z!Vs|`;ipxbb)>n_O>6oTBqiUZ4g&ng-Rex<4*wV8lf!j3>2(ls6MK;KXIdq80mvVW zc6b@fjZIIT8bqE?pa?-3S~ub`#nuk6(<*g}E@^A;BmcjP$3l!O(%T3^(*Bd_ zx}&SXy70y6_Ir?j3f#jK47~?;=Vp}+P{RkF_$*pQf%>}!?jRz&g*}i>M&?8m+5B zLzxmhrjHz^12V3(EduV#I4^D(K~oLt0Nw~#9>#MM+($oIE6{TsNlz0QRRHQX-bDa6 z5l6(oil50TnD0roc`d06yzyXY6#S!tgEtlKmtfQN^mruPRdgbmgLJFQFB$vC*S zgG)q7*4crG?dfr|K7CKmvObsi<}nMNgQYhxU|Y$>qbwZl?Z?1~DW_&`ocx_Pas&-y z!;KsSl>V8K9e9CFxh4Qs`7;COG3KxLH&2{ikvS8OAWK41+2fR79H|J8(OIw)(qZdU z4T3e4(KH;ZDl63@fAE8cb*04=2tvTBeGwFB@6Ckj=BVZF+b3?LTG!NQ zE)y0{2s8uzAHF=SE3xH2ulYAl&ym>hsbi4vHZsq`6410_j#uQo)1F=BwTe~CQ?dy^ z0K4Z2dN2q2Ab6v24O;Ear*TlOYFTXKncAw?0che@y*n1L4I2pXUQGmUefuIZ@3-&Y zH@U2yQQs{s9pK-cXBE4+I0>=8AYLT}dL{2AAMhB=f$&f9SSA8GK=xKaK|#Ge$n!S* zv3E}|s0rWY0i(wT$R)OKrEBr9Tj+sB}Rk&~Rj zJzW5LUqE2Vg3ES)&LLu$wlbh}jJ>W9o-I&DiN#8>_FkkR>tW zEj~wx{NG^?IN!vQ-r2is(IYLt~B z5G}SpvZTXNPEKwBokqXnkv!%4FS*y33(1^V9O8^JpSxpZ0F|u2_rKE8ec!+BL zqCz$PZniQwZ35%9bQnZ|6`CSM;HHiEJy?>c%XR8RX5~?tje)L4Cs?@j3T(>POW9@~ zv4U}+BB!j(b8htQz}P&(7J&aV?RQqXUoNp1YD<=jQ8ltJP`}_JR4q-G3={#`o+T}_ zdv{d5wKmBC@j#2HPr zW2JVx6JnR3<5Df>ic1_L0c^clC4#lC5b7P|(0U_hiY#x1hf|$dIp{l9gueP^`r7D)89K&E=cQoQ#zLUx?cmWqYCBwLV+B)<4L; zpR4l@QD6a7sgIs{{eAeNzl}ndL>k9m-SqE{&uY{Xj!(Gh^rrg6K7Zuz zDp16K?QbdOc2$+wx7}#H)6*{cH;PFNqRfb?Al#_H$6U+4%kUp2Q_u_*^0Cb1JtZf$ck$uVMAK*{0*DYri?w&Kj#Q^B2&lHUObHf;* zP!^ws<&T~qsPmvpop#bNN;X5AY!Cw>edI&=Z{0bUphLQd*(Teq|DC1e*uE~XJ#Q_?z6`xBIyyQbfiJgb%8?`$IRZ2m ziG&N$b!E_@L!{E<#0>*O>N-n-(9XdW({a2RsTsE3-W5ifV$8!33bLKHA+v!?-&*M2 zI5wA@oLpNAb4W7g15isyECiF6BucjpyU)MoF@!!K(_3zXoQTM}C*#mGPE1VTH!5EG zjOK=>sX6r%Xj;>6NgYnmkceFdtS2Lt?c3yvO8aEXIKhPg0*2$?m&G)%eb-2@I`-iSM#@IpZozSU6Ssl{c4uL9v^ zWMnM9d0K`()pQAO4ZMDVB6xw{LsomopycW|OKkJVXA9M}<3kJ%#d8b`@Lj{j17;CXQ(S@qIy1GyTUx4!v`ryL- z{?vTS2PXRvg6|QfX-0?{Tt6-ihwr8uAa3&-^b{FhHPeJo_JYt60!edsPkDza;FhuN zXy7B4{udGI8$Lcv>PrQ(fS3L3Gn}08U}5QizP__!5+0MHmx06cf3W@V7S@1yOKuqX zLV^D4VQeh@cm{ae;rxt%1&+)k<j9Pz)CddsRok!VqKV^ zW@HM0GLprn(+jEGZL>^dJdyf)u*1@Q zC|T5Cm}AglgS)!iP9CSjVBy3>=9<=vNf$c??JLD=PJ);I8VZ{09mqrL(}GJ_<;2C) zjTHil2o7Q40FuhBt+%lr->PTUAFdB26P_N-34v6=sy|%xS&?Li26cKTUBfJz#Ere{ z6}C*bVkkVGZ+T0%_D#(VR>h=azfeelBqJIG9A4`ob;c$Nq;tW^jXb~`FsIRcArp|Y z8GiFBE%ZXX1nf~4gm)@&eFH=&6Kybx9&sYtW4*OUo7(yhc z0h=JUn{H*=b_DtI!pP01tS9kKwOmca+}ttFRb0& z<)*N{trs|uyx!=Gr+2UiY57H;S(-qdL>1qwR zHeZdbNM9KKJdWD{OdN9hjrxEe0PJwnwE|^& zziP+5UJahwBypxlO%cG+^rX?+3!=P(GAX$6?d4dzZTG2R{R^Fte{Q`gO=PHXSy)CC z!mAJ8I0Ps|(Nr$Ik>O$HK!ZbvKH+dGl2F|2E|5S77Jken`nlq2E*y5_Y|C)U*zMy- zB2wfCv=_t$#wfe=7)NHrjgN$VfwbxJRZb!RbJAI5g?qr!^3qr0II{S9`!CuVO75eL zK+bh;Yz(AeIQiZ8c4QWVGsoskBB4!66oG8G-9tIFIPp6MIQ7gGGdmJ1eu1n+W0QLk zi8=8Y-+k67%xf5%U7OpU_MItoA8SHEghnu`{?gjg2nRI`b&7Z=1}nV)@Jt5OG#Mv7 zp^Z>3K$@49jmoa7;n1;z0K2qE9;T6+}crQW}(y?oL7JZX~6}mMV>9o#U`1ctzaV1LxA<#nqV0;lxGC&X-MB>g(S-Z%^ zVOJ~Jt(V8geLAl&-sjxB^ZKnQ%~xe3M#JXwscgRZm!w^ba7wVw2&=pkHmSr*NT3;H z%BYk!&p*)5j(IJ2t~1K;rct50j^JG}<{;jxf(2#Mw`VV2NwT@ZTF>see|S}=*2#D}rTWfcmtJ@xBebsS^;ruw4D>Hd?rQMMj#!HXensEy$3_r0 zz4JD3Y)t9DJrrW1wKLoCS-Hq8E+r+!>MJ(Mm)KabRE13Gxqct=8-~-(VPt|fX|2LL zsyx{$@UMUt3y5_O=$Rmy2h%9gdsWD^+Fo z8}|F_M4sQq+h~jAx6HF!8&_4R>Ef&FouE+h-Y^nw&Cchm>~+{^xgiHvtT!2$4$mU@ z{%$oBWec7C^hcgSAX~}oClS9n(vsC|KWC|lI={k_clHML{+An>-Y zwT;YOmb7~OpVmEm?6@;*oA0<|`h8;U&cG>Zr`VJfru>2eL+G}W^+sk(#oO=uO?5<9Ph>)e`uTO%54)#)r)cFohxW!D}A4Z17l)3KxQ-Q zvcDv@DcR@dIT)OJmm7%+g({E^bS;GpURH*|+=D zY@@=%!$}qKVRneg&+tHhU_q;LJkaYRZC4gCVZr5|hP?w>WvOO^Uqash-#C=Y(MNfoaf^way)RxOu&1{-WFZ9u zF}~A}JtonW*mI3+m_Av$Q$)i0-MvCaii}qC76h?*Mv8^BapKHr$hcYHI52$=mDuMm-`yT9xqvyN9_%xNk_S5yaj-3#TgZ$_D{# z%-{e+TAu*f&tkDMAH>ysjZVhC)*}6prS&>IBEnLQ$c~h-8?lp-knkcS>rE{#R&v~4 z)QVE8y7?^oW8%Y9ey$IBY8<38v@%Hn0Ra@fxaxPHxsRBcnV-tZ5#1y)I~Q`&JxYH7 zcFh7U1F?Re5ZFyM$>GCcbeOAnS@2GmqPwY^i6YrKo2%#yP|4<|S^Rc_2To*0zy13} zn>Sit5)wROahf>a!%Xg{C-yezjBh}9&s%Z+jZIBF2Aw*v5l)pDu3kf@P8J^dgDO*8 zZJ0y_nzzqPVzQ$5BR0PFQk4e1OVjsfoWk4vJFN%QaH0sVw%b)s$v#0lHVpA4ZRZazd@nZ3b(Gm_Te>#JSZgc=@BjoG`a@!eYwKDLqwO>quc zmf_(=JC?tW=S~XT-HO*?bC&CPUP!)BijvbkyCmy^7K!iANu`oETKcd{xaVEu-z`3& zS-HM$f&KdO?qR;+%F+OPvNYPp*YA5g@=uqVyn7zHd82z}$9ND8t|qK45;o-9=Is`!J2wsy1yNX@hL}xuV>qL*9 zY)rRwVUwuFCPh0uC>*wluvzTyNN&1l+GjpBIy96f?6^H?Wzs8K8?~-A%*7xaintx< zs`OKGK64*&-cf^Sr~2^W!;+tG$Yn&f^+fl>3QUJntQUIcGD?>NXdgxi$ibRGqw)N_ zKI3|LPR@0e*H3K8scHQLqekXN-U~kT9B$KP>Yv3#);&#XNKfLD#{R7F+*#;HFEZ}U zE`35xf{fSr<8sBmX*onH0qc3$f)p>IuE7)BwNVa!!%Q%P-gls&>c@9CalP) zKY0>IAo5G>jn-tW{OYR;`dH58Zk0%(pY3=zDUhLhBCW#we7S^#1hq!NMtmwEecRwM zjASV>?d>dEKIiJVz}I|8QJ9Ig4VU?1KPM3pku@VVpINL_S9i8{oyBA3A~gZR_DH_v zi*x4d?lrzNBUDy`wVsduv5%A;A+a&4=WR~mk;M}-NLw+Ax*QZaiNu`5BkNvRvz%`3 z@V|UJpt1UAR`pLa`GqsKTA0c%=Ia|vlQj6qz(ub+@P9xfKDCJPoOa2T#gLGY6j*AK z_V)+gp13O}RhJ0c6?teJBM+Pe`QKuR z@mlxw7irhwtJw})hX^M!*Dm_PED*F=oRz6C8RW&I5>kW_;{QTaR+UWH{xMAlFG*xH z?EnMmHCOF-{(I~UVti^f+kUzH*SNP=4mz!j`Z9W8$d&L&IcP}aJWlahV=D79FlPt# z+>hLIl5E8rcN_lIe5+tU;Jw!F`jo@*;g0oK6^2A7yUQBR(rBedfUJs2DumNVaGMzU zO!agKv-dyW^9kstvI%8T`s8}p=h_2NHBPJM&0KO)w2Tl{-)jO9{h1iV&!Y20p7BHZ zci*3d^QKQXuNacxzPaet&3|QRDj+a=1%Hvx)2U96IPIBHzw@{kL*9!~5s3f@uMm6V zL)9H50%V^){fypQB<2w|b^~53Ha0f830E90y;wLU`s<}HrC{Q_cW2Yp%fs2R?4%3m zndu7gN3R%of9g>7sOd4(SB&{ib~PfB^c#lYY8ZCvufO816|SABq2aSvWU4hBxIUDE z@(rw|0vpFX*Kt3$aRm8{Vjwm@xtJ@JqWUOqJSVBQ9r&0Q^k9Jl;lqwmLU$gRW(>@9 z?7BUdB$d7vfBYC!`K$R=uxfTIoBxaJQE8KtJi0G)IXTZbJ^%UuLLt7qynG_8cNgdO z+#sT%D-n`QU(4Yeau5Q9-X)}sog7d$Bv?kf;#NPlP)24#T7Lj(aZq}~wuRTuW1H*) zsvI9JT(Rh9lFzI{QX8~Pqd$IgJZ*NWn$16nl;I5H&P&$egzd<@XILwjt0+&Gic!sz zGFQY|vd?+NgYu>4sn69R~*-De2PLIfG>3 zB=_~!G$(>J{?j@nuq%(}^^#{c)`omDE0DzuRAF<7ByOZL8olj=kE!(BLGkI+K*k1R zL&HXEIsaF~rM3|PQpv4B&C;SqQWA+;q{SJZlD<9TC#iQY-E1{CohsIgs!js*VL|Fkn_CD4&KS|XU zUvxcGtQs!u9QbniLh@5S)%4oh@Na7`#p<7w%JW+@D>09t%^JAh@1ad=n=HLz<(*q1 zTmA-nsGK3HYBI!|iotW;xWe-n#fcohESx$s!v(xZPs&t>zL=`XO>e3~_jM`sv8f5W z$F3!(6?z?fXW8nuv&Ve)h?&>!4BBZO5E~mpd&xH+t+Q;Y(3(8qR!%i5Qik5OR29@S zwP-iQ#CJ zaZJ3x-Oqy;@fBu@o^8(i@i$QZg99!p9# zg}Dw}^s*K{2-sd6_zlhKi2arM zSOB0S^fC*FMnXyH(q$54?9y*M6ifvEde6ta&@t~OHrH(LUt|3NjqYk2EhKxHb0fnqlht1DZ``a z?t|S89*aqFG!6s@2fw@bnURv*=N<|!)a-tGpjjNGHeX5vLcDLTpw~kBiEZ?nKgQ8X zhI}b0QEm!heX6Au2Z&JJ^>9nwZ~r^HOPnUhm#0b6bEC(Hqg(yjeg>;!U&A@I--NKK z89rV5c~4qZe1rqnq(AE>^jd)^FM{mW0}EiV_1bu}LJeBsu-?GHqt{9y(dk9VVAaCY zH~Yz(tAR8$zI-r2AO}5_tjve$fTlMrImai%8qA*7EtkrW9bZwV@ zW^NH|!{ndbY%Qj(!h2HcoJ&}z|P4l=ql1!<$iV_@q z-%Sk#+-_kFD3}dac-}Ze9wdum4vGx<+ckTOhd0e=L-0=cKP<>wqyp*yx=XF`6h6hr zqH>0rNK0QdQ2@wxxS1r{^T=hLPz|O@xQKH}!HZE-rKp$2Yn>_Q2Gm>A_vKf}X`}oN zjvlM)pSi!M$a_wVcc3#z5QvE=KJ!nKyK575gwRn&8vEh!-3&@W8`{Q1>8KPx*Mo_0 zVaEa(pzy=9SW4a5Om|=(v>TG$C)s|EW?-i$pUb^HdVs&&B{xRp86`dnc|Hp1nKopP^+K{>F4LWdpIFZ6YxsP9m_s>KO2Bg>%4sG6 z=T6e<_&W}>ByrkL;Hh|`hsJrvN7FHgbSXQ}VOphKP?ZmI&r*`gUxps}?_+UoV z)=|~SD0OaQw;3CU*s6FbRpT&4>G5MtDp8lRx;hg>!|QZmfWE^HF~mYh-)Qyr^+f>U zEnzw&_Ws_i$)WS$!NG~wm$`_&w-$CFbLm>4Hhf?tWiT3cOdKXMDoPu?`#aJFO;w(h z4qob0s@R`(mbaZ2maB6->#GGz!8&S~A=>WN=)v?2y8M+dHt15U#$oS%;j%$4(Rp;Z zzxDMLwLT=_0r6c)kBs`Ww1A)SM)L~J?1TOfSMOVG^(ZVHY_#IPyY7*D6#?=}QzQTIO z)Qxb~8d4@ECT^HdtcqD;p7@mLR-R|r!?ZM9s=%y}AvKhln0Qa6z_??sD~V15AYHhi zZN_pXP*QdiKIk4Kam(V@U1G4v)D&k3#6*aIdLikC#{rxBMntD&*Xhm_A;SyXyR#b>$pH)pYxZ748{cfciWS(K zDCFQ9`*}N6JCtmD-I>35TJ}N82kr+lKA0~b^!;u-Bvo$4>pw`j-{@?<=NuThyHn!q zKu%6zN6A`hM#sMNfIdpt)co(kwwUnW$7vmjI)@S$*YqD6`}=e~4UlnY`fpjKKmH9u znOa%b&uWomOeU~PWa2l}UnD_xUoKlC^4g?Ws!hzL0rn>1_js*T5!y$p;Tf<(wY7bK z^w?OAv?Cf`K+^8@dl=bINd}mEtC?f?yDC1%cawwMTDZ_v!UnhD?L8wRzlaCw#Pw>o zujhCGfhAVJ@tnjAsb!_#==^Amh?zo^<#SVbqoWh3P^*K5SOAyB_jb@zM~4u&WH)1= zAQetLim+H+ z5-zUrHPM z=3SC|_lh+@4!Rz3GP>k@z$B=={rt#XEtasPu|4e1*I#w9&6jj#dlYGrretP@0_4(H z%mijgRRIsb&Jb6Ofl^JkLqC1o{OJ5x{-XMsWtPAp!}+M1x>`r95Gvc zyQrzkoEz)64{XXdFLT@F-ry$}Yz_`!H}~JwAuq>X6{u!2BZm84?$I=`CCLh8tB>T1 zYw)tRxXzA0I?m%po9M~I!xI+JoS2xXs(|e#33Lx2?XZ9ge*x`8gYfC);nKFCzOaCS zIXxf725#J%oRUkwH!hVuQsLbNjpH8+_6^t!y$!6dzWUGY5D0IS4C}$3Rq6WW67Cxd zwfwwQGBbwM4Km=C+Py23_bdmnRsSSFk+^DRn6cO$!1{LiRk&Ef+>qrnz__^zSu zjJ9-S`{Xx)Vu`xDYL=F{01hV_O*-xIVi^E8ik-Y3#xdP%?(x@2PuLuBu75W|gfGLf zH=RqEHZ_<><*=6F@@KZ;VWZF4V9?y2%NYB)4PR0{YWA{zRc;3tgSk^5AY0-VRm7~~ z%4dEnkZ|FI;~JTmEdXMK_kpbfZ<58hd!lMjgKyM^KN&*>+XqMZPS%G$`4?v6z3DbR zjKH(~$P4FUM)Gh@-hOvyzTcg`F7j8&on;Q-;f%i(-C8l*?>C~+fF@Kuf6r_sn)$x` z9FA|Xng3Z2Z7i3Ah2m1U=w&fhd8^sn^6c$>o@7doKf6Wb<^)-Te}Qbtx8cSn z)?TSEnErmGxi~Z?>T25E3xP!+A7j%{11TAw{#jCpEL{%x*z&u!e`f2@AYC;3pHpI> z2gEAvvLz!WRFN@Z7`Ryyb{j9p{O@GQ`XwzAv3q5`bW~p5K=sONYok&4baBaF#}I+F zXWl%>Bo{olRgDpAFXEpGJZI-K(l^S=%KB?3kRr@fPDI>G!Il_sHLP2)2W4fs(5a=Q zwMgtE^--l}Q(SBeWWT>R$8mj9q9sOD0?^|`~>*V;sn%Yc`p2QZ5A zdxZ2-?Vuf{rE;(U!^dmfA3Hrg4Qqo@Tvv~C(|DsK9GS@m5-S$2koe^DJpoziKC+3LGw}uF^oNs2|tE~3%u@Z&Hc9FaIUstvglIo#gX#CAgk&87dCrJ z+7GVd)zn)PVbCz7wHg8P%TOzo_weut=P{E1l=kI|IG0{)cW=6iqe+}vj*iy@Nl7^s z6=n$mhs`fjTOK?Dwa&E@P2D2P8#4KR`>H9mwGMB%`BZ_@22NbYC;mx*Aq29VT*Y^! zXEBjtEy@1je4|i6l5(o5k+MMUeE@Jte#5jI@5WGUf@J7Cib+6ANjiV^o<1;5A-$Id z>O#v)truCI|BYXTA))P)v&4S$3K{r5L=fK?(nG4Oq9UiP{3D?(FHbhnU>*i@m}=)} ze`TS3XQT@B3dr75YV%UeQ17 z|GVE|%CNG2kfqYK0Q2R!nTz62Sctpw40!b9-x~a(sm#}WOaa|G{*7JK9N%81069-) zA7?*IFS-LHGH%_*i-+r>mEAoG8tM0`?7x*fxVj)vv3Pi-od2l(KPhOZftr2jC>>v~ zXz@^1)>24E(8Q@Eh+5P|$lUQ!okXV_2G(ZV@j;uH1;|Z~%f}Yy(r%UJyeR^+?1Cja*7K zXfHgWy&%CP==3Wq!OuXlA5xbkn=iKjwbB}VdwYJE>W0}AJ$8>ab>$^@8g(zVh z(B0eHtJkkqW|v*-M?D&oa<%DK&MA0;O$8=PQM(uXDz-j8z)-eZ_o?X(X#cqWUTdhx z{5tMcPL{Zu0+YeXja_cjp(0big`y>P*IjSGbT#pV0`g`>tJOH^$=ob*f6$0X5Tb(} z7`%=}DBM8}7@zTJ;%p6pX!koJ<_o%L?wcZ)Dho?jlCAhoKpyCpOC#libNy)dtl6oy zfkgai5i-ai49`{V27~l+RENE7ExQ(5Jw7;1l3uhQ^A6DfR_G$JrIz`I!bg%fqOUh!UTN`#TGv zl?MrwJ*WI{{}Em;v(L(}>m>Xut)uv|m>Nd;e*{Qx@Df1RXO^wMn{J6nLT4{X<778X zBSCT?I(=d>MFfe+E<3kJhGk6rA6|fJorHsLusE6uvK(Qa>vw#*tDL9*#BaxbDp;X} zl^Uq{XyKnPIT8 zJEZ#F@uoy*$YH;x@F5mN#OUE~np`#>`=@6x-Fg&#ATD0`_n39ioDJy*#j_7YKXP7U zO>K-K>Gce=ecB#Ivf)ItB7r)dTxz(G8emO4oJoyXR7`D z#~DFtQtM4V;_=ci#S-GlBtFyjTs!xrx#`a-)dHnv9=;*5`g!s|Xl)Y@x1d)z7fW|d zE?ZFwRLn2g*&94pxLW)n#TNMVyO%@zDd>e&V^N;bLw7dAD%w}V1?F`{g*yC<(-Lm-{Ynhi z@x$rf?5KK`%Xt$oM2k4vlCy!J*nfysw7iSes%ghvPG6=@V0`urYiZVr;b5dS#I#U| z)rwk6c)Ws>iUQDC^#v;h%vous;dR`Xw@UOiZ@xGUr{=!Juwn={pU7*!tj3)cd*l6p z2c19(^XCp_05v~E&HgAQZc5+8Di#ce_{bS!MM(*KM}VVj;_xOFE76a;?V2S;?r^(_ zjirX=88Y|D1OE_szrW#%jw;M*wzRtI3W&`2bJsRW@L13k02#Xm_ax-b8Z!n^=k~e% z)+w5Z)1~#2f0i`+ld7-!aajI%X|--;)z7vNWRj$sZT+lwkIHKk;8vyY_TL@R6l_@h zoIrt$S-GoC;t{8|i6}Q}F*uw5nwc(_aPRF4XMJX|9N&EtHBn}pLS`C z_4kw8=byT4AMEw7yH}MUsx2$5$`IVDVU+G7)wbnN^SV>*;Zfhbz3;GIh_)QGS`<&K`Ap~wlcD{6lPVj)}p zYR9v?RRvjs+O@c>N;!9WdU})#jKj|2;Gi<=-a=;5fuOq6+Fg^uf`3nOn0o2A zeE+?RVTx$vY5ge(T1$h43mzou?K>4a!pqg|mDn{2O{f;YPiOFyUWhFTTzxGG`h*)2-eNn8PiF!wAfv9V%+$qg4gIU`cidZ8 zkpMKF5rj*5x$|z4d?qF`@$AJ5P7X;bvjT0{BTrWaRH~OjV=XcH0Br+E6kP^viUYW2 z@|^c=n&wYcIa`$T8rUiZiSkgY&HQTs>9fz^TNB3Cc@0DnNUi@Vz1jh`iWG5H&edz{ z1+>lWa$qO9a2h%M-iSH(d4a-B&Ddbze}gWCz$=0Mgos%lHG~BODSG`}aXl{GULH8EDLFmQ%x8LAOFyVOCob{|RIA%r63MPZGO+Vz_g{D>iH00qOSfB59Sz(vC{cF@LkLxEHCnTjNiLE|ZblF80SQ zWR$k%)aB`t)_{Jy`O}ly@cTO=ube=JM}`1XVA6Nfg{odj79Ls#@M9hW_zzV_MUQve z0$o7wo@@x*+K4$;y)k0^AEyei1=7zlbv9aD2eC z`Af!?BDhnk$q^5c^3P_XNibM|qn`WRr((_L#g~i>X&69x5P`v1#-mm3{>>RA^|NR9 zGC-n*x}JxP>O<4{VEz>q+sC$)U;rP2St5nxZaKqTPy0aRrStXXHMJ4~Kaf^LXWm z(FEDd&bC?T)oY+9-0#QM#DTn>-O4KO!u`+r@9G}r7tWf7jEbtfMAW)wsY-h;Yk+0P zjrHT>tq&vgISKDF3=nZVU`$v-mvjro`nW;519)bUwq-zBl86y&3z1!uzjyBhP5#ch zaevm!sWnYz*k+rudl~#k!FELdFgy(wCDX;T$z^4#f%FgMRa8_mQr`#xC4jaCW7qpk zNN zJmciX_yu26QAEA!gxn;V7D-6sOjU1(H3{g;j(x)khASm1kRI--s}ytb6C;uCa37v( zHkQDZm#l#qf+dHD-5EmJ50dU`iu9vcD)ef$R{I#M0lsHa{suP6L8AkSaS}sb+f6qB zYYU zNEQP#1w)MOEFoA5_|b3H*~~0OnsE*f_s(?^1=x7C38E;yh;-sC#)1_|4(7|ovVmn) zi`ws2b<&aG3)OoN?VtCwli}M8rOa1NlB(}AmlQs-vpyCpnA*3Qp(N>Ax@o%?);R|g z5JcH8#l-`?8|VG%+71?$b4`D4;Fo{aB{AcKJNN!Er^$_AGbTOkoY=rW1U5+Zy}8H0 z*9XL63eq^CP9t%Zr%T!<>C4agx- z+TXm{;?JP;cLa|Wy2ai1#%+b3?nC&8=)Qui*yLSvM(z_5&Y6-M?))K1q%wYgsUMf;?{VmSAz{+4CD? z($f6{c*|cus?m5V;?WdF3hbT~Yx}7B&g#6)ONyUZgZv+zu!>FvN`7F2ZGwM1FP#sg z3h&QxEoS;>&h7-ph&VNRyqeAU>z%ZWUsWgFj^<1(7=4)dfM_bvoHpK3#6ytIKF;E9 z4&0u;z9tUb!x`6%Cw!_8Kc6kpd_~4mcPA}6n+$YtjhTi7Sc1>3$*q9iZsN2oc#(Dz zp^5qJlnimvSnc}qY6OU~5bZ0(NUY#Mc-z-`-dQ|YTL1aM&Ch2YZo>@y_!b+PP1Bye zDO4e~ben)pZomC?0!;5mMm*SLa_!``i`B$Z-(b@ZS=y!er@lVt!$h-Kt&ScP*CP-H zyclwXCmj0(TR7WzsF+ zGQe!Hj>BTP~9!#G74LF##@EvxVgaJBm@oJ-a1eqsynP6mc^ z>G{$ogG|(?YUzXK#eM3I>2=zD=_*MorV>lV3ke1`~BLfw5jNsNHRBj^!7i?hT{TIH<;IQJZ zH!<2dPZOmYV$XBh@CJeB_G?)_f!6?a{@DCZ(pzo4d zc;(Sn=5r3x>vVK3pdeiG{tg>|+nImEZyF?LW{Peo59mF3o)qJpNSj+xA<`$8_x*m` zOgNY8KAB|9qf%-DGr>Q~W33N4Lk))&5N7e~0OY~UZ2+`XIn`)q=O^$Yqh>{U(Ed*n zP#No5|GcF>Yq~61VS9h@yQDT-pvon|rcCJIJ`3F?#VRF$Ob_LUF& zor-lVa95LndJ7*^jfMOuIX(R$y7GE6H8Rf-($g+aZ0yQD_8;m$KwYT;o4+RfYBzyo07z_E3nJv36j5u>^2r;;pjv}4Yx?o9<4CC2_LM! z1{2_IKkKF>;#K1~o{LCXO~@HTH=@OlP80EG%=8{P48UFzzsKDAV`DT^pJQm==rE!I20f2px?d0MLUI9-Tf zuDivEFatrvPqCKPm#7-MXx_E8OdLLzkM^`$sJ*g-fN;{pmeVL17sX}DYTveoEmTBn?WD%R}hwk1;n0VaQ?F)FTBD{CKo6>V;g5By1V6h>Yq z!7$i+iW03u^HqOhtIwI-Y%AhJ8JD)!)T3oltK*n>c8LH;pr^vwx_$Fi|8DdNp?sZc zmr#gZJ~1v-dEVy3&SJ&<%3j<)&#u$!$zHbZKhB*F0ZfHY-h!fL=Fy`7b{5O$R}SJz zpM5_ok)#_gdQt6*+u`hZE4yms41>*Mte}V`xu(WO0qYOjLmrKu*EQ22EBa}c51QNs z-Iro-*-Wdhf*F&+=0#@skcR*eX3Mj;bQd1l-DG34!9b~#$;bCe4F5P`9KDzSIAQn- zad>{~i(~+r7#dA0cIBN97c_d3M79|3_e6;K$__YZ<5W=r?y_k1$#pMP3Ej-^vzd!_ ztoBuU_{s;->OfwYV7KfE!s;^8 zxAkoKV|?7ZI+Dqj?-c&%@VrxaEtVL^872QI8m{(S@~i~YQ;k#!#Y3SYa}9 z8E0y>W;JGmocL@**A-_KnU^-pLt&{Y7_B=0ea~>2D8?nO1Oe0%bMkeZ zszSQP5pXKua5O0B7sQL_EX0z#wxD_KH)mTX>imc1HESWoLd*r8m8)~+37Aj!@)!>* zMcR@n+iLFnWluM2r)4h|x7Kd>;~m9J7d>O17<3>K@aAL8(Pnya z#ua~JP_bcp)ui8yEo7fIcCX#r+vA$i9aWmqjbUs^%gkKaE^v#U#}rotl;#qvMTjT(N@g!?@>cI>H8DiD_sw51jQm z+!bD`yZBM2K`H$KU7~FJg7)B{5Lkr_n4<&PK1Irsl*SWkMp--wz(b|WRFO8f5yhxJ z(p>Md52@ynhdq7$O~wza=M8W(&~>A^L#|>xQl_-j^!`w+p~;Um&}y%FX*hS^KFcB4 z+gONTfNWtfX6E{!b(Ttsc2If2t-2!fiPnQA${0N{X{8Hsl&?b6B3oL`Bg4cKJlG=A z?dEKDGZnKx^~onB#kiF8)mb)e_`j2iKE+(p)8))IJd_{PT5PM*Ce2i~DRRq2oAma{ z)1`b7Q4@SzALdCc3n~|7me$+2^-HJh#pPUMPBEAEJpJ=EX}7KMa|D^?e*TbHT%uY0 z#31iC(JY&y(Oi6x6T!EJ-!-%E8Sgf;arm0we4LW;7(YprAbGxRW6XVZ%3VO`%J80J zK3zBS%GF8sYE!tYS*}&ZN6R(c`>8Stb3z2tv+}mLD6a>%?z{J~j7iDQ)%sNkxWarr&XVuR%bpf1WLg2B_eF>7iY7i7p2GvNz1rHpG@dw6&cg_q7 zV-4JgYNE-vgvu*Xk5A<-DB=_e&)VQJhyBB?V{hu_0!Pn{52J+VVua zAL<6?a~rg=KQRRP6yk=ZlQYca7L6vkW~VcyNKZt%b3!uAcY^Tb%SE*AXp93s%<=p` zxQPF~N7Zq4?B+0(C!hvZD5cokZF5ye-HhLPpaRxyaNz%FQWXRp+O%|uw;V;a%eIxN zo>UIvg0{=uCWXU^1qHkjhTxAyZsh3BW&a2L8u&;gncs3+JOFH^s3TQ>q|Dx7DZkG_ z$OZ&w7^;pHmmNVLYEkt7usGAIS+?+?F4nF3tA71!3~;q758xO9y`}>x%wyD>c1H3E##Pf{0kl5UE0$S8yUw=ULi{9l*9*HT+-sMs4f3A0GZB zaBbh}(-@s9fO3WWyu4Wu`35pM;GRNXL$U_{ckkzR5%A7Xm9lA<7Ek2|g<=R?yUv&L z>60f1%aw-NVfLshxrD|c_u9JxKIAKgXkgxk}g@@aP zl|8`<>I1r+Dz?Pz>XCWjXhq{W$|WE?5k$2yB#T)BCP2sG&S+*yi2&5~%>6LR z?D=O-Gk`q8+uJ+ee&cxykj#D}dtanuMA9K1xOVSU9y_a(9T@9Vz&!p0ts1;` zG3WE%>|NRm4c0_LP!jxCRrT)@>jhI#9byg_^?sCCr$#$2^E>a9j`k>2tP_=k%Cf$H z2!(pvX8Jn_at;T5FMw8NU9@E@zr*2H-&Vxv!95&emUhrQwkD{L72#gL>*CEP?g<{9 z*mNU67XGCGdl;w(bWc-rTKirCGQiM;P46Ku1a&mGcccDCu zu8VtXI^u=0j2?7EQ4H(};O8e3b#V$^hmFm7_3J6HfGmv;{P@9?EX<8A`r&N4#e#x@ zl)St=v^4JP>r-7vQGTLFMQ?B~eF8@Q{lY0Y>jjh44gwm~`eZ}UvWl!Mt%Lw@{Ntlh zs4oRPKmFO-20cHs>!`XQ!{7Q$eJDN?-_w@3_-ksc&YAAN@_+Y*i>XnUes}TfJyxu>0qlsBUscw zbgu08WV$4ot-;(cKdr`5QNdG80jXASS z=6Flv_-;D{8nFiu1cgBkg2dx_`C2VG%Kri7m55fK?r$%3!5;4jdowYHTdc zc{@J}{!giy70RF}IxJ?_mEhWw3OiU8F0B5JK<;YE&h!(ghyD=UOnsnI$U^NWYFdxk z>rM(-w*5||0B?Yx$@WTB7pe~5(T@GkiD9Qy^>^4~1Q_|D(V?&mM=XRZsc4w?XCPeoVAjQHtv|jO= z{LoSJP`egrN%3D)@stR78!fWk_oM2m_P~K+k^~&KA~Com$cxsEbvrjKC%vUwS6~_N zSWZi!Wm;%c?)g!rE9E|)gO$6aq06FFe&ZCxB9jpiD5t@6r-$0Frk_3%)>Q}4Grc<# z^I~}*H>pn>2JYFOlgI`zve&5{7dbaj;}f6(sjiis2!J1XbL?imV796gf;y3AsI!Q8 zBzl;~uT_2GcR2*q7(}T130h3PRg#x~4(j>$X~IWBT5L)=-zdN}#c$Ys4JsQ3WF2j1 zSNN&Gmb42#txi=aHQsIJYWq0)tM*3or%zWl=2J6y{isDvpDqo@Rn8RULS1nf?90z@ z-h8>cmFYTM{iEa&{fzC;Y7g`yd$4_dWR_3k2xZqWb(vz;$Tq%dUpT+Q?*bFD<<%?9 z4cMZ1(Ssse-A|_hFg}N8v}_2VVN-ttw0f=>`{_#?f}^{>_ArnUP@FRzciXcw!&|RYaQ zo_@jFI#sIGd}d$6{5#R~RX-tgIIB$=!XiiE_TeyZraKcP(=!rNbuD~>Hk7smZgd)? z>kGQnF2CY_TTcPjh4c)V0KXY_TDtQ1a0QwdYC+mW`)8`b!v2w93>tA_+YHxAO%*6ri5k^@m*9dX^%J{}W)1 z#*$CQP_0q|%Wexiql(HeZZu3%j9N6Ysx=CzZ|rUgEJ@PO6x9HOs1u$z-5O=2E2qX7 zxKRrB+~AN9`ZpdF!VcZ*dnHfkXMVkK-H!leQC9I+2Ys?jr3OSd?He7^|EOS>x_-Z+ ze)3=h-t?l@C<5`^2(anL;SaOXS>z!nqQ7k!aE7>`%=vH z*+%JAn}WR)EWM>cwn0(la^_6v4e>5rzXKb}rM@;+2SL9?gKax1?d*?WO!=@iQ{@f& zn_~{ROo)hy4I6_P2U0#Q=sNgTZl%94fRrL!z&b@v&MN;XdUR*hRRdyCw7TOGkML?0 z*52`8aixFah$#Npie(EUfic?%hg|>OOK~RJsvtbD8*+n!EjZ z_wygM=#1K9h;lA0)N&}PeSFXIslD3l=;mc6tv5fWqend|eOAI*ZMTyScgr+ML}rAN z+BVkA_7?4h9yvP;Q;9geU0gHCvxbo8lzCcl@uhp8&jVj?QSxGT^wF9~Evd`lUVfS-Yc5-EZmwi& zmAj|sYtAEH^bHMfJ$gg{_T%$JL{qXFuG822LMB}2Pl-Jq)x64GCs}N}LiJ;C z(ERAYK0ZFai1nW9%mZJIGCK^Pd#{gQSzthi6xK7W3)YCKdsU)kjFmAtCz`{rgT zZeHFWEQ6=G6qbL2v_K;)ObMak@wJPJii%)A_YeJH_$xR8#1el$2XHZ(@*^qnP&I-jn6(1e)dvXAs4PIu`aEnAuofUS6iY{m1a_4*ffhj)JXGf_hz4WMuDP*9O#D zrO$nZg{ArM=Be|xJrv)vv$Hwtr!2&2F9d{WZ!7TMw9Aq;MPL&Hnubo@rlTl9f(g@q(sTwJ{D_cph-UQY1U>*Ubw zn^Sng+?cfzDAmr?sC*;Kl<(MaMe$5a_V?9Qj_svEPpD;^7|7KFelZ$n*s6SSG%&HYA;={a~&@DQHnld7?Y(G=Ns7gdF@0m zbJxK}^twT(y1M$+Fz4rv)42nIG1(~Wv^6hRyCDP>6CwE~HZ~y= z2{IvJVRO4?|xCUU#amU%zg=J-|bXYYh9oG1TaeTyl#s&m&lzoTP=RsY4&w z@MBu4YCA0s30z&j3x*Ka>~&=|14dNKv11tFIM zQoXk5yM~5LK@aGx1SlOG9I~JY8u3;c)X!Irxb!M4?2?Y2UX3P4Y(W8WTwI*Ksp&~= zTr4as2%AflWC3mG!?nYt-oU_o&d9)r^2zMt=P>EO>9){g{_A=izxvs>4L;{RZYcAZ z_Gv$S)qc?0(xL@X`q#lOS4wK?hoGPn!4EtNUyOcAO~tV&KG48xm^s~I{`4s!85tQ( zV8df|b^O2v_rby!+WPvZAwweC4i4c^ zEBxwvZEa3{y?B9M8`bu73wh~AKww}P%=jar>({I8qK_w%a&kH&YJSa3&&(*PsZmOb zi)&u_@D@lXA8T9N!N|avn6_KDZY}OfOG#yQcFGhyZy|#le+TAn>3N6ODurOlBY1?R zmdn3{6qYAO*M!=j&B8<}8Wwbw8kH`(kioFf)4%b$b0s7q!l*B_C}aZord&v>a%X4t ziO-+^YGG*^cJK3={H)7We*Rs*Hrb|mrMq{ran78{R(Ysm+H$_xeY)>SB@~|sP~NX< zUc7MY^kw&R>!o9kSl;K*zn{QkvN;1E-4d_*^eptxM=bxZe6&Zs+)d)Z|EskdQX@p- MuJoOZTbi%_7b3Xb=>Px# literal 0 HcmV?d00001 -- GitLab