tensorflow2_2_1_2-gnu-hpc-doc-2.1.2-bp153.1.15 >  A `CY!M@eee%{s痿Z `Ik: Fn}Ĭ[j#}$Kc=Azᶻ&i'' #*%,&]}ZG`dÆG:RMO\FDPĩŅˀ~oc>;rd"D@{2I'Z7} r}N3tH#Nk=,:!'KT۩9LJ2. {K<򑣷Z0! adb2f34f36d016e9b5885b27a25d3769ad930892ebce8ef0ce6622811d323ffa8c548ae628c52a8b01186a8d11ad349940898fc88`CY!M@eeeJ)k3HKi:6YWHuS䝌dn"׀cM 3]UڵE3}>NGTI3'YDgd.,!4x)>@WstwMD/kפH?'Bh]/kz?$`w[}-x`bMaz<)NFIAMd>R (A1C&Y!)7{ K$6Y5A+klb݀e_Y^xmGK(>p;=l?=\d & 1 V  =F    (  `t(8?9?:K?F:G:H:I:X:Y:\;]; ^;Lb;c<:d<e<f<l<u<v<z<= ===XCtensorflow2_2_1_2-gnu-hpc-doc2.1.2bp153.1.15Examples from the tensorflow websiteThis open source software library for numerical computation is used for data flow graphs. The graph nodes represent mathematical operations, while the graph edges represent the multidimensional data arrays (tensors) that flow between them. This flexible architecture enables you to deploy computation to one or more CPUs in a desktop, server, or mobile device without rewriting code. This package provides examples from the website.`CW)obs-power8-01%SUSE Linux Enterprise 15 SP3openSUSEApache-2.0 AND BSD-2-Clause AND BSD-3-Clause AND FSFUL AND MIT AND MPL-2.0 AND OpenSSL AND Python-2.0https://bugs.opensuse.orgDocumentation/Otherhttps://www.tensorflow.org/linuxppc64le,Amm`CW%8mC8mC37dbed59a6aef0803f24a7dac146679491c460faa8bc3834974d786acf56a40fc2fc832a4c6c9c5f43fc26b74a9e3ffad77aea5e82f38a5132f780d8cbdb3f46rootrootrootrootrootroottensorflow2_2_1_2-gnu-hpc-2.1.2-bp153.1.15.src.rpmtensorflow2_2_1_2-gnu-hpc-doctensorflow2_2_1_2-gnu-hpc-doc(ppc-64)    rpmlib(CompressedFileNames)rpmlib(FileDigests)rpmlib(PayloadFilesHavePrefix)rpmlib(PayloadIsXz)tensorflow2_2_1_2-gnu-hpc3.0.4-14.6.0-14.0-15.2-12.1.24.14.1__q@__[@_Z@_X_Wr@_P_>e_>e^^^@^˳@^k@^}^|@^y@^t@^ku^f/^`^Y^Nt^F^E:@^B@^:@^6^4^2@^2@^0"@^0"@^0"@^*@^)]]V]]]>]/ ]#0@\]@\t@\@\@\\\\\@\b@\ac\R@\A\8@[@[=@[%@[u[ZN@Christian Goll Christian Goll Christian Goll Christian Goll Christian Goll Christian Goll Guillaume GARDET Guillaume GARDET Martin Liška Christian Goll Christian Goll Christian Goll Christian Goll Christian Goll Guillaume GARDET Christian Goll Bernhard Wiedemann Guillaume GARDET Christian Goll Guillaume GARDET Christian Goll Guillaume GARDET Guillaume GARDET Christian Goll Guillaume GARDET Guillaume GARDET Christian Goll Guillaume GARDET Guillaume GARDET Christian Goll Guillaume GARDET Christian Goll Guillaume GARDET Guillaume GARDET Guillaume GARDET Christian Goll Christian Goll Christian Goll Christian Goll guillaume.gardet@opensuse.orgChristian Goll Christian Goll Christian Goll Guillaume GARDET Guillaume GARDET Guillaume GARDET Guillaume GARDET Christian Goll Guillaume GARDET Guillaume GARDET Christian Goll Adrian Schröter cgoll@suse.comChristian Goll Bernhard Wiedemann Guillaume GARDET Guillaume GARDET Todd R Todd R cgoll@suse.comJan Engelhardt cgoll@suse.comcgoll@suse.com- fixed hpc flavor and Leap15.2 builds- updated to 2.1.2 with following fixes (bsc#1177022): * Fixes an undefined behavior causing a segfault in tf.raw_ops.Switch (CVE-2020-15190) * Fixes three vulnerabilities in conversion to DLPack format (CVE-2020-15191, CVE-2020-15192, CVE-2020-15193) * Fixes two vulnerabilities in SparseFillEmptyRowsGrad (CVE-2020-15194, CVE-2020-15195) * Fixes an integer truncation vulnerability in code using the work sharder API (CVE-2020-15202) * Fixes a format string vulnerability in tf.strings.as_string (CVE-2020-15203) * Fixes segfault raised by calling session-only ops in eager mode (CVE-2020-15204) * Fixes data leak and potential ASLR violation from tf.raw_ops.StringNGrams (CVE-2020-15205) * Fixes segfaults caused by incomplete SavedModel validation (CVE-2020-15206) * Fixes a data corruption due to a bug in negative indexing support in TFLite (CVE-2020-15207) * Fixes a data corruption due to dimension mismatch in TFLite (CVE-2020-15208) * Fixes several vulnerabilities in TFLite saved model format (CVE-2020-15209, CVE-2020-15210, CVE-2020-15211) - using fft2d.tgz instead of fft.tar.gz - removed fft.tar.gz- fixed json-feature-name.patch for leap15.2 builds- updated disk constraints, as sometimes the build fails with too low disk space- Package C-headers for standard tensorflow (boo#1175789) - fixed build gcc10.1 errors for Tumbleweed with following upstream patch: * added file Provide-overload-to-cope-with-const-ness-change-of-N.patch- Package header files for Tensoflow2 Lite - boo#1175099- Revert memoryperjob constraint support and use again %limit_build macro to avoid OOM errors- Lower memoryperjob to 1300 MB (as done for tensorflow)- Use memoryperjob constraint instead of %limit_build macro.- fixed build with json_cpp 1.9.3 (bsc#1173314)- fixed local CUDA builds- updated to 2.1.1 which is a bug fix release mostly for external sources which are not part of this package (sqlite,libjpeg-turbo, Apache Spark) * Fixes a versioning bug which causes Keras layers from TF 1.x to be used instead of those from TF 2.x- fixed broken builds which were caused due to missing dependency on @com_google_absl//absl/strings in various BUILD files - added patch: fix-google-absl-memory.patch- added mkl-ddn as source and do not use system mkl-dnn (bsc#1168839) - removed patches: * fixed-mkl-sgemm-call.patch * added-mkl_dnn-as-syslib.patch - added source: mkl-v0.21.2.tar.gz- tensorflow2-lite-devel does not requires libtensorflow*- removed hpc-mvapich2 build (bsc#1167735)- Use pip install --no-compile (boo#1094323)- Lite flavor should not provide python3-tensorflow nor tensorflow- removed sources of bazel sources and replaced them by internal packages * rules-cc.zip removed * bazel-toolchains.tar.gz removed * bazel-skylib.0.8.0.tar.gz removed- Lite flavor should not provide "tensorflow", otherwise tensorlfow2-devel and tensorlfow2-lite-devel conflict and break armnn- added Provides: tensorflow, so that Kerase works with this package and fixed Leap 15.2 build- Fix name for libtensorflow* sub-packages- openSUSE has no CUDA package, so disable cuda build for openSUSE- addding changes for CUDA builds- Add 'Provides' only for hpc flavors, otherwise it matches the package name- Add provides/conflicts to avoid to install tensorflow and tensorflow2 as some files are provided by both packages- removed mkl-dnn as sourc and force usage of system mkl-dnn for x86_64 builds * removed file mkl-dnn-v021.2.tar.gz * added patch: added-mkl_dnn-as-syslib.patch * added patch: fixed-mkl-sgemm-call.patch- Add 1.25.0 as minimal version for grpc-devel- Add 3.1.5 as a minimal version for double-conversion-devel (2.0.1 from SLE15SP2/Leap15.2 is too old) - Lower required version for protobuf (3.9.1 from SLE15SP2/Leap15.2 is fine)- removed AVX2 flavor, this should be fixed via mkl-dnn- Fix build on hpc targets- added shared library packages libtensorflow2, libtensorflow_cc2 and libtensorflow_framework2 - removed the AWS sdk support as this forces a SEGFAULT * remobed file aws-sdk-cpp-1.5.8.tar.gz - dropped following source files as they are not needed any more * removed file backports.weakref-1.0rc1.tar.gz * removed file gettid.patch * removed file grpc-v1.24.2.gz * removed file libjpeg-turbo-2.0.0.tar.gz * removed file nsync_1.20.0.tar.gz- Do not try to install *.pb.* files in Lite flavor- Define package name at 'tensorflow2' instead of 'tensorflow'- Generate *.pb.* files and package them, to be used by ArmNN - Do not build on %ix86 - Do not build %arm, except for Lite flavor- updated to tensorflow 2.1.0 which is a stable release and has following breaking changes: * Deletes Operation.traceback_with_start_lines for which we know of no usages. * Removed id from tf.Tensor.__repr__() as id is not useful other than internal debugging. * Some tf.assert_* methods now raise assertions at operation creation time if the input tensors' values are known at that time, not during the session.run(). This only changes behavior when the graph execution would have resulted in an error. When this happens, a noop is returned and the input tensors are marked non-feedable. In other words, if they are used as keys in feed_dict argument to session.run(), an error will be raised. Also, because some assert ops don't make it into the graph, the graph structure changes. A different graph can result in different per-op random seeds when they are not given explicitly (most often). * The following APIs are not longer experimental: tf.config.list_logical_devices, tf.config.list_physical_devices, tf.config.get_visible_devices, tf.config.set_visible_devices, tf.config.get_logical_device_configuration, tf.config.set_logical_device_configuration. * tf.config.experimentalVirtualDeviceConfiguration has been renamed to tf.config.LogicalDeviceConfiguration. * tf.config.experimental_list_devices has been removed, please use tf.config.list_logical_devices. - renamed the project to tensorflow2 so that the original tensorflow v1 API compatible release can stay in factory. Following changes were made to achive this: * added tensorflow-v2.1.0.tar.gz * added tensforflow2.spec * added tensforflow2.changes * removed tensorflow-v1.13.2.tar.gz * removed tensorflow.spec * removed tensorflow.chnages - following source files had to be updated * updated abseil-cpp.tar.gz * updated bazel-toolchains.tar.gz * updated eigen.tar.gz * updated gemmlowp.zip * updated license.rst.txt * updated rules_closure.tar.gz - following new souces had to be updated * added aws-sdk-cpp-1.5.8.tar.gz * added bazel-skylib.0.8.0.tar.gz * added fft2d.tgz * added rules_cc.zip - for the following souces the system libraries are now ues * removed aws-sdk-cpp-1.3.15.tar.gz * removed double_conversion.zip * removed file unicode-org-icu.tar.gz * removed file 816a4ae622e964763ca0862d9dbd19324a1eaf45.tar.gz - these patches were removed * removed file support-new-bazel.patch * removed file tensorflow-make_aws_sdk_work_on_aarch64.patch * removed file tensorflow-fix_lite.patch * removed file remove-keras.patch * removed file grpc-namespace-corrections.patch - these new patches were added * added fix-lite.patch * added removed-docker-tools.patch * added right-json-location.patch- updated to tensorflow 0.13.2 - dropped grpc.tar.gz and grpc-v1.13.0.gz as system grpc is used, this fixes the broken builds which were introduced with gcc9 (bsc#1152671) * added grpc-namespace-corrections.patch in order to use system grpc - dropped re2-2018-10-01.tar.gz as system re2 is used now- added remove-keras.patch which removes keras sources and uses distribution keras libaries * removed keras-applications-1.0.6.tar.gz * removed keras-preprocessing-1.0.9.tar.gz- using now system protobuf instead of building it (bsc#1151150)- Ajust %limit_build to avoid OOM errors - Do not use %limit_build for lite flavor- added additonal dependencies- fixed installation location of shared library- removed bazel mirror from as much source links as possible - added support-new-bazel.patch support newer upcoming bazel versions- Fix build for lite flavor: * tensorflow-fix_lite.patch- Call ldconfig for devel package in post/postun- Fix aarch64 build with upstream patch: * tensorflow-make_aws_sdk_work_on_aarch64.patch- Add Lite flavor- updated to 1.13.1 fixes boo#1133490- Update _constraints to avoid OOM errors- Build and package libtensorflow_cc and libtensorflow_framework- added fix_mvapich_mpi_bzl.patch which fixes detection of mvapich2 mpi library - fixed python3 build- update to version 1.13.1 * Major Features and Improvements * TensorFlow Lite has moved from contrib to core. This means that Python modules are under tf.lite and source code is now under tensorflow/lite rather than tensorflow/contrib/lite. * TensorFlow GPU binaries are now built against CUDA 10 and TensorRT 5.0. * Support for Python3.7 on all operating systems. * Moved NCCL to core. - drop merged patch mpilibpath_configure_py.patch - drop obsolete pyton3.7 patches - disabled jemalloc for now- enabled aws and googlecloud support * removed no_aws_and_googlecloud.patch- Fixed build issues with python 3.7 what introduced the patches * python3_7_compatibility.patch backported from upstream * python3.7_unicode.patch fixes a minor function call * python3.7_async_keyword.patch avoids the new keyword async- Fix build with python 3.7- Build and package libtensorflow.so as some packages may link to it- Add constraints on HDD size to avoid no space-left error- Fix python3 provides - Minor spec file cleanups- Provide python3-tensorflow- updated build command to fit bazel-0.19- Trim pad wording from descriptions.- Updated to Tensorflow 1.10 as with this release it supports the partial use of systemlibs. Still a lot additional sources are included which are * closure * weakref * double-conversion * gast * farmhash * nsync * gemmlowp * abseil-cpp * boring-ssl * google-apis * cub * highwayhash * abseil-pypi * eigen * arm_neon_x86_sse * fft * grpc * re2 Although some of these libraries are available in factory they could not be used as explicit versions are needed or bazel or the build system links them in the wrong way. - mpilibpath_configure_py.patch changes the search path for the mpi to also include lib64/ - no_aws_and_googlecloud.patch removes the dependence of aws, googlecloud and kafaka apis, as this version is not compiled with the support of this apis.- Initial commit of Tensorflow 1.4 not all requirement could be met by the distribution packages and the sources have to be included. This is true for - Eigen - protobuf - grpc - lmdb - json-cpp The build itself is now based on bazel and creates the pip package which is then extracted from the build environmentobs-power8-01 16150259612.1.2-bp153.1.152.1.2-bp153.1.15tensorflow2_2_1_2-gnu-hpc-docLICENSETHIRD_PARTY_TF_C_LICENSES/usr/share/licenses//usr/share/licenses/tensorflow2_2_1_2-gnu-hpc-doc/-fmessage-length=0 -grecord-gcc-switches -O2 -Wall -D_FORTIFY_SOURCE=2 -fstack-protector-strong -funwind-tables -fasynchronous-unwind-tables -fstack-clash-protectionobs://build.opensuse.org/openSUSE:Backports:SLE-15-SP3/standard/90823700cd9addc99bc079091eb2ac7f-tensorflow2:hpccpioxz5ppc64le-suse-linuxdirectoryASCII textUTF-8 Unicode textWZrHB>),utf-8671ee5fc37c19fc165a23f02f93afe4bec3a70127c1ecfe39abfd835be744d39?P7zXZ !t/(] crt:bLL.ZjIa OYHյ<gv:uDrGdo>QÑn:EAڗGb=6G)W^&p' pQu u}**]o{%}|,,Fyk}Y$MJ*7FR&ğoȓ>_}@qZ=Iڇ)7v/\'l]_ RŧvGaT.A &FeO 4Aز, X|>@n<;@~[WQ@uX+~F/1uC^ФIt۾l MmnggU=>ݴުa8]l,O 0QW˖#cQvXD?wk*tUz#+C_93 ҃'m qV`F?e;1yR}@7I2ѡuw})aŬ0,M*qK[9~1 Z+{Xv N~#k3ӫw!3pTD>5yę~i;vv9t: #*(ȧԝC61r #=`; XԍWD96$Dt @ xIXO^^eqz-x _R19yRg'0H"=AS-M`:h!gWcUj(RIkwI 3_2^|ƍ) g]bMgrfwjBz4ȿjSᶭ "YI$@.*7tiEvVK) #d74 f !O,Sᄩ2l"(2Z1v_{3ҝ$C.tnNdmJSnGI(3ωD|ec'<˨x w"*AAxH v UiXTM| 4bF?UN,Vi{x!$l )L v _^U&u,)7Кs 9N{j1_v@zVMFu=p`!I =9DkܩRW$3d視T O/b:GG5-M/B7L8$O8l_(pFgE>BVY4Ե>#*'c{"# ׋xx~IvOU`)WxlU |ǁ><"iE4+}]'ppG ǣCb[Bb8a*5ã*YX*ZIki Ϙ̅L]2t56߂io;}Wͽ>a1+=yԍCd $28U3٬݅+d Fѵ3[w`IU(wC$OE-TV}S6*M I|9*Vυݔi,Lxf P$S#D ZĿQpY'eE߯fU![{rʣV3ڎ\TTPrz9^ \0afXt}H =DdU!J mm 4 V}V(u8+#*D&BgV*9M"BT u$v zc_{ v|>VoUumN^=핏]oPoYO Y'G @ ZڑfH3s=zhFX*!۸HBvj>E6u,t0sy,~)kYs<fv>~6[^SF \IFsbO s%Jv>$/, =*/X:7ઇ(so5umgj9ұGsq ay?|r:acR3' RL|FM}9OnJ6f̹<~j7@+ĿZ 赝-kR5~{Ńs-EzҸir-9ݶhæP0^: ۈK@q9p< xM}(+ܸ`xAq5i"`(ԑ8zG ynMJPȔ*ӕ7+j75.[*Tlb4a{tGQcvnc0a )P/hVu>%fòW8@/#mE['.,rZmc|m]t=Y/5c‘RH?uQ1Mz)e`Jh@)R:NT4'*ܝW@lk^ <#Iʩ)h`Q/ R2bV##c (@EIs+0$% }9s@c 1|v7}m(q[~Vo,~>0!߿b:&Rš4]3Zp-BUR0%ΡW5/>I"R^w WB8y_Upfxlx CYW.7aCd2uAx?.u :XpׇP{K7dXx)u}$cI~3Fgd'>#8q.$4y39/O1g.e5D9JdrWmO(w&r-%5*<3"n+/ge_eyV˒]yNSW䡡S\t@ԅg zF0yV1ȹ{  +nj~H Ƹ1[<$W&LBccaw05úٯ;8=F1jOAwa,&F&K!P/Zwt&\t+8 pAq{uZ7W_+ց܍,S"O(lmK\geڑq5Ls:|C.o '=fU¥Z6&@/ '\j mu֯&FxR%>cSFĀdⲖ nvpE&ɛtlEt.Ή_w:1 |I lXR|4ZZo^5t6$Ev"-HpfJ(4p{ j}os{Ajܳ*V<&AV,G:GP}F #fiѪʏNKT嫴 /ՔbG%>Ϸ+K_E='C.Iy`H9Qln+/Fa[q{k)0b#O#?Hl$qQ`^?9i? t~X|ʰ%Lfʆ$bXkUvLS̖-g H( oVXߙ44Q>%dL"[$uC쉬7? KߝPׂo-S:y%Qۑ~iGZDa!}V׎OC+72!#*%aֈYSFkCDS(P} I]_jPص=Ai=+)}qxct:mlI &>,٭ )VKˢĹ﷧*~Pol_4Z%>78r%(`qAWѮ͈]"Ы âOܙЃ`0#`r'KLr;ShMÀʬ6-qhG=X&c( 0ʿ|}rjB *h{urjJ[8 0sB#r30Pi9M`W y\+#h1&cT#}w[;j$'(nDn] `p2,NV'M#a#<:!]!q >Z4Gt53O/auL߽[ᅴbK= }˽? ̫>-['^m.G#x2y]c(4#$wuRŗxuR9GFXSy:isv |>_@wi2Dc~v#5^fdQ3Y_AG[Xx%=/Φ a%CC>tmN+U!ae;+w 2ɍ/ yz f ÍhXᡡZ$NԒz 88 31L\پ2T_'Fn*EMIRqS*sD;=B!Uw9^k GG{;l[Lg]>J%#HA nɇ|Ȃ,=Zi*q_; ݦ:u7ESM\X-o€W(N1G8;OŎN3f t(*I==*Sň 2ݑRdL*^ԣ6x*[x'(KC('g,&a +AD!gg:?8xp5eL9Qj89#K6 m7O^e>y J=PMz>{>ˬYvkAIC& t \v ˙z0 MBecӸS$ϰ_]`akqԕQ%a?R?;bNQ9r'aa&knT u6Kj/ 6ƴ&VCzXa39 NA\ƚqT>㝖`I@crlor/~okaLq=AoKDw%qc>2:4*HӰOV\ y5E_Քw!%ܴ/%eL!)|("_ǘÔ;R @ldSƾ_u&)GkJL~AћOcB=yAzTcMӀs5фz}4.NbLcH5f~@mex-9vs[0bxӴr%„ROJw #P\W!f,+ %PVC\Md5V ώH#\پougaq&  p`#R >O^k*0,1{3 V'P\`eBx0zlJ\g_|BR_-G."!~v&of%@<Ðٿwڰ?>~ $c4b5P/'œ|1H|>R->yQxJT^NĖ }9Wk9,`烝d R'x _h/9:񇥙# 5B1TRiv, cTYD](taߢ,Q "(ޭ3s˙$I2FUeE6&<<힂'o;I%dE U_XCMu1Zs{ < */Cg `.Z$B畠Et/O;w~Z<$G^U֤X",5L%O('P <$wkRtj=ܢkp}Խ5p%-mLF5 4Td?زN|p4[Ꭼ)(!uhJNڻ3+|VD+SƬJ56/YD[AFl%;emϔ}9ZcE:p],vcgk= <׉C^]ys&bcO}Jn)˶O[{t{oУP&cn^ mݾ[2*x3h AOiid w# bYYj)]-!gm O 30IJa"j^߃@A)̭,MTCkE^ׯeFdweEctnPAAWNk]K 881uB\A满rCcٵ$MB [DP Р= am&:n`T!G*3QF=R{d;MU@O1ϿC /BA*S+;p.2s0d,ux1$'-k)2}ZwkܓM? R"ptnZ _vj NYaӥn#?[k=iA4YsA<Dse{6Lnnj^+S@Wk-:ClعimS(rQwi⮪9 + k/>(aāDL練Ba^OCRR468K 4nԼPcV)/*&NN\_Sߊþ~9eGev> R}$4nO3e4?]8̺Rb騍1 pCPܒfWzR XRw;\=;C}6xiIhß$;?]*&i N +P^MM+|PbaQZ WS9M':iFp0#ʆH.fpB ^a V]hkc?_ecRAZX:'U4[\m9yqҐ3Kh[ܭ 2Z}v!l嗶`e=qu7kñC-qU̽l ,IZ"4-{:t^Ox٣n_,9c~\":~?hOP$ 5lozv:W !Ff7g%;D62eֆQ`?$E7YEiYJiK+v{~pB}VJQw@ʝ0dPA5Pkݚzݦl/O9Q QouF%ExEm%)% ;@/9pΘ0X+j `h&sY%X5;f}juiXw5bL>؜C HQ1zPcP3I(YaOHB?zϔFT՛kJp궑>$-Rr!fՌ-,,R58 d~߅[vY vrګlIpիupI#!$AZg^PYQ~d r_K1<#K}e}j"sJxIK_`')skKO0u\ :ܽ".sˌe_%HVXwL~r8=T=}4P&VSLf IȅTty\2>EexE݃1H_M c[R'2C:n씡۷+ 2Ϳ^X@ .S]r\ķS4iybm^UӓH(FWhJ2@WYUAaCn\9!)jnw:CK_["TӹmwwV %DŽA`ɇgFpC K %ձBd`NY^IZZ=ʧ͗d=!--J>ʗ_KwRձ4k2 ҜddN! d| kf%t9{Nz˳w:xc{F&ҝYMDZ_]ZFbmYsa[tF[E\[4'UeǺ{( @((9Nda5I*K(T+t[)?yHR-@m':|_)/BSKݴ̬߯كUϰw<7GZ岎G;F+&ޗFBC;!d(GUjd8eVlxڱ{.pF ӱ)Q[{<XY bbHR4u]a[nHlNR%yL ]\ʸ7u]1}*dUy':;9VkTGbk?Mk(WB5н؆-&12+~U@LBS53ѯx%4m)Pî8PGi(hӸ49 j? MJ` ]n@|쒸zg*us *yP&QS>3sIZVewLhb6\]*zHiecu/L/7sD)4vo.vj1!jcY¨%^W0#" y2+t+ v mat4*fۡ(t1+kF  s$};n Օ?UkČYNU3T^ص+@3lCᦚJjqƁYߒBbՅ^:M⊃*:BXl5[$<8Jy%uOQ&F5Ы#T+Wz V_4vʴvSU(ϛFw[ɲUƘH>hn+1l#~/{0Mqr<ϵҠg戽W=t;s,v'نܣO~$06| c$zV@"0V"qaj+Pԅlj\ 8/zDh#l@7l6rwzH}ih7>PKP9Qi"[~|UnQ:ď`Ca`:hSJ>\liOW&32X6iSO?zCS 9uOSτ8?F+2~v8A&+*B~TAS5K ;)+t v{C;jO3nX??)dCϙ+ *7&*Kǘ_DUΩnٲ./$#%?ԝL%\6K( #?p38q|k7B$>zKl?_p*Tg3$0dZiY8QȀzٮNy AlvAJh9X/7^Yb>7N%M:Ff;0401Ahm:ث ~TI)):q&O2ZC?u߂fӅBr)Kv" %% T_V+"v~Tc]t$^3RVA ^KMV1AD5i9:oP3H:.&+:RgM_ )+!;`*8.L_+[w"L`vGdYA 弜ʉf&K5ðiO KBi#m4)X1di՝LP ͝h}؟~sT6B] 2_"G(ڨ y^“) OL[YWn s(/L<34"K :jebصgdڡ1L}6f. 9cT9lӼD Qf"kB[ug+s iE^<(:JRr3r%(-TFENd> ,RHaM |́]eP]hɼTl4FOTZ $fM' ``͔ܺ.Iύ m(U8?dI}R\Z'|-|W; :yk[·0j i{pBԮd=7yjrjZauNOȰygߏTvZ6 &טhsaVzfJm$N]rl-}'Q{f4e:!Uj6QFFؿ-ZH|ߥrU!*uUz%dWɶ \4OY6bOu, Mlkx,vU;$w{`k< LhAMѓ#<@y1`'⵮wÚ~<+,뿀ȣ]s\Hdx[5wN͓[]4Cg06L;UZ/LciB|:׹A; ps,SE܄h7E7(:hJ7X!ߝ=Lû a(Z}X%V|8BH>C[ja+ Ռi:gheQ:S,'p%d,6eĄU{__/;M:uO>PO̔5>llǹ\Iչc+1@- zIXcdk vL‚CTLѠs+Wd dԓ3N͕jE`"lR>aVF\jjZ7U=EȟAM[\Wj|&&T uo'14#W_Upy;`-wƥ42?bu4)l/"_ziqeˉ iwZ]N F!r17_$ l⿫%vs;"]g5bY 2Ğj!nE[f6 ` ӍSU75#5Ͽy+dr q} icuK. 43}U+ pS\G8Jy - 529q_oL(˶n|@-"R0&~jSaG@@#7yd\0ƴ+'[prR %:-fGEsqcT6p F%X^[jWL7)Ǡ"L?pK4]C=ح/V_V̓s V`dt+ֺr# iu OcDd u_o/Mץ*54ooLQ'K"}v>)sfV4B`Cɝ TEVjj<ϖ"(dc)9y '* ʹ]-5BE_z30w`l7bH! s]c3suᕗo^[*UtY]u٦ɩ~w?fBҼE'Y=A֞3.܌I S kzoz*.3}h;"1%JNPQ?ًگ#i:Tc;~& gw/ffʌ8I&5a eˡ})Cmy*91Bw@Q9s@%x%8ibHsjx>7bցVwL>:bXUӗ;;qԅKsTO^{d$^PE8h.|n6s՘v# V 3 S8R>nH vH*"XP.pr:W-{uKY t#> Nr 5 b~ƴOxH Mu$mZEL\x6feU n2^$\A]:-5'$^lu921mqG4g4Is5ģEwbU  PoSuNJŽQ #oR7pQ͛=8%8lK]GFSNJ(*Z,o͌[tsQoP>Z>pZȦŴj_EAN2w{M 1) "}vq2yoRCN 2Cѫ^^"Agim#Q Z)}D&#VAKmPԮHb豵јkKi(v#Ӗ.i342i|Y^"Lb@oJ:->n|`%˒wJjegL"e4Y&\4"[yB$ Guq=(ue=#w?sg0:VUJFSx@hho:S 豔+I.=Ȭr52NOj ,%<ʂ&GҺ_0^CƼQpҖቮY'rob &/tJƄ1{?v7zD>lOyL ֟pn3O%k5m&+nIh@e&v1- M;n"}й(JoWdh$.olwKFL_yaWpn+zgrE ;6Zp=uo?h <(-l3,Aƨcx%3"VrcLZe=jPaAVJLiD(B>- 7n ؕWU0WӸ0][K !]f>dDh!GоDxkBTrL@ lE#(b,ܕ>W3)5,'rD^$fv链C<0bKvxpZ㝧PI;Op,tu5v>ɶK4sy-Ԝt 'pim{l<#rWmf)=oAWol؜/R-gӔm Kܲګ/F>f&F¾qCmJu`u n ^orYoB2B!'52dƟ=Xt"KIR6 퇔aגv%B acπFVq4n@1YeO"t8~m *PBu%c+&& q2t"uxH- p1]/\) OdǕ_.]NC,K4d}Ւ\-OSxX>b Mn X>{~MQRv<7q(HjvS =ٳmŁJ EHP- t<&\X55" jB9M7vx˺MQT$DVR=$*G{Ҭ{ȫ_rhIWۘj˅ o8ԍ2j|6_}ӜNt r7sЉ؄cOQcQoC%NBJHL1x<4&ͬM~;UHη_{|*,RMH@awr,ϟ *ʘ޽ڋdU*_:LBNwhO7X+mDŽ?32u(ߚa`Q!oO| `n ^Vie9ְ+;9IPu,lS\IN V =u[$$RnԲ4gfX PfkJqCU~ʛw2cB]>)ɢT_ 6)<#Y_s( D7i Cp!XR,}Q98dAÊH\~'+M^aeEDY7.=Ke2VJ :-bFcg$u _ao{P;Lֳ˦G]y_̨<㬣iq<IsXȖ _, 2O0AŢVӀ1T7?LJ0YTJɠ~~O_pOOBٝ6\}[QGmӿQKπD @5tMHՎ ]."!8k1u04t5 ]Zϼ(r"bc~Iw'm1|ayR )FvtnWY:erbOtm)1n=/a9ڰGP= o{O(xũI8N-۝"q$,zIAQ JiAMO֭{zC ? 6%F}a߈8#qs$} , |`-Jo$5N+y-:'d?3?=4DǏ!=\j1Ӊީ,]U;@Az}1~(3Qgڜ(o2y $B*ly(Ivd9=uvC;.T2 cQnqDec DD7Hn;g0Y2д Eqa#'M3/'C@ s] ]'csi[8҂~ ߀d-lb~kY9Bi `ambd%vUL}(Pf |18~8ȹem<WdUNQ93,'p[/{LKz#nmV:*!M~㾮PrBH Hq)Q DN`KZi7+NP>PMf&< vVU`;Ba&ӿDS3&޺sO$:$B(M6bYj.kMߪk (SIM0*qW z5OtΜ] C-3bmr-9i9:K Aы86 {V B> =5Kٌ''0cnmh)q)r:Gq[>VRW%a]oo9jt@U:sGxLu;+MR(ovsRk^!^z5DeW7IBB̹Xcbɤ-#Sx]# П7:< mp_J$nb|!:E0̀?/ϼ!L)*C8sB8l5ACߘY Ab3o/U[tZciq- 1N hV_j! ٺ?fLQ[xa'P~!FzRfE8S2y;KQP[QBs2̘iw@F7G #\4ư vAh@+B=Xlђ,gTy|QNωL~5!B;&ظ=}(,C?dCq*Y˚ hmNN AqR?3= Жa#bAzw/rapʦZ<tRUY)]yMB@y{PlV9yW|ǿmWI)/!X\%P=}ʢ339w֦_#ݭ:JR:nn-~r*wT|Ep t|9ڃH:ۍS]Xe~%VaxDx!_[9i6irGLPiSx 5# gI[DԬ^\(7< :="n>̫'8H"Vʋ-p 1KgsnSYPoMX(W;2v>XET$ҏlT͈?E!RحHLmp*M.iJKx~26k<)?BW!Sn hq!@EЃ7J|Qn՘u1L'>yFmxPL@hwVa9]( T ot{1 c h=iJ}̃!7Z}ɂ70WR/l;h:6TYNzX}s T.4qDSYqf83Τ 5 qw;Z?ӴC?π&a;z5~"t+Z`{Y(K9 KjBЗl.u8N@7ѾWmq#jm"MܤTVs۴$ ׇeaAf-7`$9r& JTpO74Vys# `jKR˨}a Ӆ/ffksq#eEv,mS`Yw0咡0 [xOSErGdV)ުyڑGVBVπhkʔ fr9[zCP2tx9xǪԷ15ܻ9gmmNΰ@t]KCA.Bj(+b*.'mE Ǟz { BNkYy[“q~ &1ELT3"HtM=˨䃐@N Vj4v)͏ >϶wM.,Rl@TRѢkdvC({S'<LDJ!\ȴ nuLJ;pؚ6ߙfJvkQgk<6wY1"@#34n:Ygl^giì6Jgs}u@pdZ5a+mPĚaf$ff??|(/N(JQ~p=v|w>~ts ldECϿK 2Av\MPFPʲiQ}lKb60y]1S]BLQk}2ls\n%@Ok Uμ \%2DX}, cIRZt^[/h;" < TU0ȚZ 2q)zrة3ψl#۟GtB-!ĽOm|%W(ɜb)M6hx@8a'wd=$*3Hq!ӅōN1&3zRm>\|N/U2y;"ٛg7FXuU[CPkf4o4E:Q ȖGl*wQ q9/r^]q6 Ǩ:yǪ$Dz q} XFhPZ X5 Xh'YG3jdmz҉`3㯴$PuI&HD#t8ې}/ʷ?|;vQ+窪:-ADiIU| ^uHII;RIh V'?j>]>ϘN#SfU]iiѓy>d.3/ۜ/mj]y#GbK%=Ӛ3eѹ  @:<uJt9I 5_B5w.aq d ǂ$@!I\w5_G Y%{5 x/ɑQ Դ?xcXBЂ~ FNq Xq/R6"΃]x?{݀/ZY~t'Խ"9o)ss\kv\ -1u'dy2>Tb9EI0 z[^~T*zRyħ E6PL).r[ ң_qɎKts]=|Dg`WfL6Mҭ./ *3\Xcg32dzeZ R RsO'1PTdH'?쩅Z/f?V+}mmF!rOCFiʹ|I9Og]|F~*HM2喍L,$H`?O9wy]d%heAcf}oSpqSs*`:cK".п?nחTW1fE%#TaowVxP3t -]2p:^qM>}%9ύG?l[.s3|  {pM*7Ҟ۷bഽD. L׷+ ~?g1;šJ5#8^ֶ#]VZjZg69!{g;VRlupEH +r6 dn! YULm4 $W >DRSܼ:A9@ť#M"Za0_W/zF5i_M4pL AHZ[R;, !8:=$؂hen1u R,Y">-O uTf`fQ"c~jȞq^`o=iZMS~Z6jt49غɯ4][Ί@e)(ZβPВ-ʷ1Z 1P]ҚBe>n3\%E Q v FT2T~T֓/StǭǴ/ S^O[UEn{Cӏ5 (; TJ][9Dxv|I/%W .u(c=;.W"8_푈%6W[a~E^xwBVK׍Dna7tUAXt`8]uHs[P,8!{6Y8HHR e3٬ |!}јa4}wX&DC d5ڔҭ5CWѼwELӮC*U?}Kf㭰AqW&' \)%SBRq 9DK_=p4k{/f`Ɨ L@E_ήC/ߙ;w9\JFrҚO~qS{U{"ދ"U:#F]peE_)!+s ٮ BSּ*'Dex+(/4͉ (0αu@`)l%h^ƒjևm$;U Vǚ} S}ׁa^OHScg|1o8SgOa1iFko{Ǖ;?Ћ#uTQlP+/e=G%&zOCAJ "yG:@ L >nhLR49xjwoUfx "T6Sy%/[(ĩYPN~nO$P(gF}ee6_yKN`˼YeQ6'$u}ǎ x8go=b<`R IV_AnQ&CyJqVCz,{2;5 #h|RaB:E`zi5^)Tޭ:YDxZkK- "vy)gLZtZL_5{dsI16u<|mA%nGl&lL?gp tz8GJ=YFqXWjXmA*t Pknsx v2C[$m^,;tڧ"Fy &0n:m&߲ lE͊z/|n,\ZSC/ZUPUpxWP"K[x9 *Ni lG;(&H;Wy;%McL#G7\|w?ӺQ[o{uzgDxӜTuTE5 T5^JroYGmv-z=a&O9A&4n)$U3‹\7-ζ绔supuŽ^9hٺbYgD:pR TA;̾E`\Og>>).f Q,2X[@B5Jx '4FƊ8<#mMPr/\X9R@~^ZŨk/u ᲁ(^ҋh8xwa. k#MzN77^W4;m{9H^t 痵6+B(G g+ߦ j0I$N1)Yy7^B pvu|rP#It/ٞbʤVy!pOȈs渻rsG[y1:Tk;B6GWʚ-V9]Y'/شKiq3_dh vZ-e6:oF2P:x5u1G5gw! SfF|PچB[X)6]@ ȷG8#P̂ר7?q> ^' 0 S{Oؓ MY>8KM(NZR $˥1Mįg~'~&(l<ְ8{uM 0jfaGePM1NwN{CV$Y439(♃յ$#kWzM#bюe^5n0O5DaL&̊FBdfC Z!00̻$[vFq"UVn<aoNV(t8lc@+hw4\tn2eMh C8`T;Lرq(+[ mȐE?ϵ_\ax}RLƜW:x o@Ew]NӔN^ cPg'>$P|赃xie6kT;#!5B,F2e1@i )9v \baͪJc^Kά\ :/^i'._&9}=>g&lAĂ nD%&(ݻR(D+ezP]axWzM3)qDwӠ}1FHHPǧo/) "(/E\ A/J\,oDKo ڼq,G<uƤIO*L{$=MLm"AtyA1O 3d(ڴv:?b{u52rH) خGP<'Yz5,9,9HwQ2b=J~ǔ3!<J̯.zR|5K݁2k 0xm MpWGOdiȥ>>^=" ƚ c :=^hА!=LItkU* dMHHe)}vPp'`Bq9d~ﲈ!}E#֐Z*qghlZOݟ6'2LO*>x/K. Ez\m`6yTh7uT+&ߺF!>~\0j[YOQ ko9BhMNbjl r$DEԥٍ>~>,Do5m-6ktٴg4)OhL1 yHuJnH^M"BU=\WȅVaũ(f*!v́Z#,{ZC#t yCLeMHH; 0?(v=ED͏.v=a &J[$8Lo:fi _i>Q zGXP4 E,J$fzQ5{쀺iO)b >+ՠ}/u&;U}氰FchWoNM1`;R]Vΰa&%Ca D%{Xj2a.V̢Ey۱sapQLgX\Mqrĉ^B˜DZlR/-Ho=_&$d$BWAʡVvXB^. eު^qpcɅipb,q8C*-T*wU4H ݜJ߲t/s DKvj{Brj @WGAbiPt9f…\‘@S/4fgYTdHߏvMQôX^ l\ !'H K?I-{&Hg3lVgAB敏d:Y4, @T-_S_!uS/VN?3\-`|[_gjVx$V*<ꦣTddVf0zt뻱{CGuarvJvME+H܇jnە lyrw Q6Ac(OXw>0e9V"ɧ QF@ g]n"~06_viQDRf}<֜KwZ xV-ԯ1B7}㏔U Pl%u%˖p}NJү_ZR*4)2] \O/`BJEmՎT nTom`rkա,f23BY4Z~ꆸ1LU1Ѭ}z]^R"᳉2g!0wf[f;2-m{sRYn_^՟1*J0:NjC^JO\<^vϗdǸ:#,A[3rn ,\tx_c(O`xp1.]@Χ eQՄ8ճ8AG} R b5sVcO4䝓btVT\Exͅ AP(K/L e%k_*%$\Rʭt'[}M_z~x$citKovgDWУ¹ -)`m"@Ur%NcƵb< kbit/N0x L`T^cT_$c`v%]yP*/JΤ6B:G% X`&$eiTVoGڍ}QvL0Duoi,X\A=8gN>?o~ NT&ue>##ϴ$;Bz-`/"̈́;7VRu#>N-ghN6j<X('vnLl}~xVOChAW, 4:1ns4pH1x I ́5h{/:R93G*Nv-O:2a\]p) zG]7܂Sbc sG`D䱷xd ]|4Nv=uX Y5ń|8"q;;&aAq N\4ijCK_1[٣9m/6Ǐ! s=&=F3oI I_n㧽L Xc rFO,$ 鬭iTjsE# _w 58w&HjoOݞ7>RP!-5 ]lQ,g&޳a\R u̝^ sj6X? 9ĦpDhz5_aET,u4{قax ?4a}:*KE@;Kwg, EMkPmTC-oIǼR|WL2 &@f*,Cd物/+ gQKݫr?ĚQ| |ߖ;rYH_ dY 橊C[<սvvG-=܅ϴ 3I4GLl_,+2Ca)k2 LИ/w<^#{+9G7 < Boyf3NTpW/=PGox1$nBΉ= E7\ᄭ~dSc;1BCk?$GIƛ7]ӜZ<:*V9 |h`zu6th>+ Kv YZ