pax_global_header00006660000000000000000000000064141163343610014514gustar00rootroot0000000000000052 comment=2d2d8f4216ca5411e6e5c5f01d0006e0a7ce3515 chinese_wplc/000077500000000000000000000000001411633436100135235ustar00rootroot00000000000000chinese_wplc/LICENSE000066400000000000000000000406561411633436100145430ustar00rootroot00000000000000Creative Commons Attribution 4.0 International Creative Commons Corporation ("Creative Commons") is not a law firm and does not provide legal services or legal advice. Distribution of Creative Commons public licenses does not create a lawyer-client or other relationship. Creative Commons makes its licenses and related information available on an "as-is" basis. Creative Commons gives no warranties regarding its licenses, any material licensed under their terms and conditions, or any related information. Creative Commons disclaims all liability for damages resulting from their use to the fullest extent possible. Using Creative Commons Public Licenses Creative Commons public licenses provide a standard set of terms and conditions that creators and other rights holders may use to share original works of authorship and other material subject to copyright and certain other rights specified in the public license below. The following considerations are for informational purposes only, are not exhaustive, and do not form part of our licenses. Considerations for licensors: Our public licenses are intended for use by those authorized to give the public permission to use material in ways otherwise restricted by copyright and certain other rights. Our licenses are irrevocable. Licensors should read and understand the terms and conditions of the license they choose before applying it. Licensors should also secure all rights necessary before applying our licenses so that the public can reuse the material as expected. Licensors should clearly mark any material not subject to the license. This includes other CC-licensed material, or material used under an exception or limitation to copyright. More considerations for licensors : wiki.creativecommons.org/Considerations_for_licensors Considerations for the public: By using one of our public licenses, a licensor grants the public permission to use the licensed material under specified terms and conditions. If the licensor's permission is not necessary for any reason–for example, because of any applicable exception or limitation to copyright–then that use is not regulated by the license. Our licenses grant only permissions under copyright and certain other rights that a licensor has authority to grant. Use of the licensed material may still be restricted for other reasons, including because others have copyright or other rights in the material. A licensor may make special requests, such as asking that all changes be marked or described. Although not required by our licenses, you are encouraged to respect those requests where reasonable. More considerations for the public : wiki.creativecommons.org/Considerations_for_licensees Creative Commons Attribution 4.0 International Public License By exercising the Licensed Rights (defined below), You accept and agree to be bound by the terms and conditions of this Creative Commons Attribution 4.0 International Public License ("Public License"). To the extent this Public License may be interpreted as a contract, You are granted the Licensed Rights in consideration of Your acceptance of these terms and conditions, and the Licensor grants You such rights in consideration of benefits the Licensor receives from making the Licensed Material available under these terms and conditions. Section 1 – Definitions. a. Adapted Material means material subject to Copyright and Similar Rights that is derived from or based upon the Licensed Material and in which the Licensed Material is translated, altered, arranged, transformed, or otherwise modified in a manner requiring permission under the Copyright and Similar Rights held by the Licensor. For purposes of this Public License, where the Licensed Material is a musical work, performance, or sound recording, Adapted Material is always produced where the Licensed Material is synched in timed relation with a moving image. b. Adapter's License means the license You apply to Your Copyright and Similar Rights in Your contributions to Adapted Material in accordance with the terms and conditions of this Public License. c. Copyright and Similar Rights means copyright and/or similar rights closely related to copyright including, without limitation, performance, broadcast, sound recording, and Sui Generis Database Rights, without regard to how the rights are labeled or categorized. For purposes of this Public License, the rights specified in Section 2(b)(1)-(2) are not Copyright and Similar Rights. d. Effective Technological Measures means those measures that, in the absence of proper authority, may not be circumvented under laws fulfilling obligations under Article 11 of the WIPO Copyright Treaty adopted on December 20, 1996, and/or similar international agreements. e. Exceptions and Limitations means fair use, fair dealing, and/or any other exception or limitation to Copyright and Similar Rights that applies to Your use of the Licensed Material. f. Licensed Material means the artistic or literary work, database, or other material to which the Licensor applied this Public License. g. Licensed Rights means the rights granted to You subject to the terms and conditions of this Public License, which are limited to all Copyright and Similar Rights that apply to Your use of the Licensed Material and that the Licensor has authority to license. h. Licensor means the individual(s) or entity(ies) granting rights under this Public License. i. Share means to provide material to the public by any means or process that requires permission under the Licensed Rights, such as reproduction, public display, public performance, distribution, dissemination, communication, or importation, and to make material available to the public including in ways that members of the public may access the material from a place and at a time individually chosen by them. j. Sui Generis Database Rights means rights other than copyright resulting from Directive 96/9/EC of the European Parliament and of the Council of 11 March 1996 on the legal protection of databases, as amended and/or succeeded, as well as other essentially equivalent rights anywhere in the world. k. You means the individual or entity exercising the Licensed Rights under this Public License. Your has a corresponding meaning. Section 2 – Scope. a. License grant. 1. Subject to the terms and conditions of this Public License, the Licensor hereby grants You a worldwide, royalty-free, non-sublicensable, non-exclusive, irrevocable license to exercise the Licensed Rights in the Licensed Material to: A. reproduce and Share the Licensed Material, in whole or in part; and B. produce, reproduce, and Share Adapted Material. 2. Exceptions and Limitations. For the avoidance of doubt, where Exceptions and Limitations apply to Your use, this Public License does not apply, and You do not need to comply with its terms and conditions. 3. Term. The term of this Public License is specified in Section 6(a). 4. Media and formats; technical modifications allowed. The Licensor authorizes You to exercise the Licensed Rights in all media and formats whether now known or hereafter created, and to make technical modifications necessary to do so. The Licensor waives and/or agrees not to assert any right or authority to forbid You from making technical modifications necessary to exercise the Licensed Rights, including technical modifications necessary to circumvent Effective Technological Measures. For purposes of this Public License, simply making modifications authorized by this Section 2(a)(4) never produces Adapted Material. 5. Downstream recipients. A. Offer from the Licensor – Licensed Material. Every recipient of the Licensed Material automatically receives an offer from the Licensor to exercise the Licensed Rights under the terms and conditions of this Public License. B. No downstream restrictions. You may not offer or impose any additional or different terms or conditions on, or apply any Effective Technological Measures to, the Licensed Material if doing so restricts exercise of the Licensed Rights by any recipient of the Licensed Material. 6. No endorsement. Nothing in this Public License constitutes or may be construed as permission to assert or imply that You are, or that Your use of the Licensed Material is, connected with, or sponsored, endorsed, or granted official status by, the Licensor or others designated to receive attribution as provided in Section 3(a)(1)(A)(i). b. Other rights. 1. Moral rights, such as the right of integrity, are not licensed under this Public License, nor are publicity, privacy, and/or other similar personality rights; however, to the extent possible, the Licensor waives and/or agrees not to assert any such rights held by the Licensor to the limited extent necessary to allow You to exercise the Licensed Rights, but not otherwise. 2. Patent and trademark rights are not licensed under this Public License. 3. To the extent possible, the Licensor waives any right to collect royalties from You for the exercise of the Licensed Rights, whether directly or through a collecting society under any voluntary or waivable statutory or compulsory licensing scheme. In all other cases the Licensor expressly reserves any right to collect such royalties. Section 3 – License Conditions. Your exercise of the Licensed Rights is expressly made subject to the following conditions. a. Attribution. 1. If You Share the Licensed Material (including in modified form), You must: A. retain the following if it is supplied by the Licensor with the Licensed Material: i. identification of the creator(s) of the Licensed Material and any others designated to receive attribution, in any reasonable manner requested by the Licensor (including by pseudonym if designated); ii. a copyright notice; iii. a notice that refers to this Public License; iv. a notice that refers to the disclaimer of warranties; v. a URI or hyperlink to the Licensed Material to the extent reasonably practicable; B. indicate if You modified the Licensed Material and retain an indication of any previous modifications; and C. indicate the Licensed Material is licensed under this Public License, and include the text of, or the URI or hyperlink to, this Public License. 2. You may satisfy the conditions in Section 3(a)(1) in any reasonable manner based on the medium, means, and context in which You Share the Licensed Material. For example, it may be reasonable to satisfy the conditions by providing a URI or hyperlink to a resource that includes the required information. 3. If requested by the Licensor, You must remove any of the information required by Section 3(a)(1)(A) to the extent reasonably practicable. 4. If You Share Adapted Material You produce, the Adapter's License You apply must not prevent recipients of the Adapted Material from complying with this Public License. Section 4 – Sui Generis Database Rights. Where the Licensed Rights include Sui Generis Database Rights that apply to Your use of the Licensed Material: a. for the avoidance of doubt, Section 2(a)(1) grants You the right to extract, reuse, reproduce, and Share all or a substantial portion of the contents of the database; b. if You include all or a substantial portion of the database contents in a database in which You have Sui Generis Database Rights, then the database in which You have Sui Generis Database Rights (but not its individual contents) is Adapted Material; and c. You must comply with the conditions in Section 3(a) if You Share all or a substantial portion of the contents of the database. For the avoidance of doubt, this Section 4 supplements and does not replace Your obligations under this Public License where the Licensed Rights include other Copyright and Similar Rights. Section 5 – Disclaimer of Warranties and Limitation of Liability. a. Unless otherwise separately undertaken by the Licensor, to the extent possible, the Licensor offers the Licensed Material as-is and as-available, and makes no representations or warranties of any kind concerning the Licensed Material, whether express, implied, statutory, or other. This includes, without limitation, warranties of title, merchantability, fitness for a particular purpose, non-infringement, absence of latent or other defects, accuracy, or the presence or absence of errors, whether or not known or discoverable. Where disclaimers of warranties are not allowed in full or in part, this disclaimer may not apply to You. b. To the extent possible, in no event will the Licensor be liable to You on any legal theory (including, without limitation, negligence) or otherwise for any direct, special, indirect, incidental, consequential, punitive, exemplary, or other losses, costs, expenses, or damages arising out of this Public License or use of the Licensed Material, even if the Licensor has been advised of the possibility of such losses, costs, expenses, or damages. Where a limitation of liability is not allowed in full or in part, this limitation may not apply to You. c. The disclaimer of warranties and limitation of liability provided above shall be interpreted in a manner that, to the extent possible, most closely approximates an absolute disclaimer and waiver of all liability. Section 6 – Term and Termination. a. This Public License applies for the term of the Copyright and Similar Rights licensed here. However, if You fail to comply with this Public License, then Your rights under this Public License terminate automatically. b. Where Your right to use the Licensed Material has terminated under Section 6(a), it reinstates: 1. automatically as of the date the violation is cured, provided it is cured within 30 days of Your discovery of the violation; or 2. upon express reinstatement by the Licensor. c. For the avoidance of doubt, this Section 6(b) does not affect any right the Licensor may have to seek remedies for Your violations of this Public License. d. For the avoidance of doubt, the Licensor may also offer the Licensed Material under separate terms or conditions or stop distributing the Licensed Material at any time; however, doing so will not terminate this Public License. e. Sections 1, 5, 6, 7, and 8 survive termination of this Public License. Section 7 – Other Terms and Conditions. a. The Licensor shall not be bound by any additional or different terms or conditions communicated by You unless expressly agreed. b. Any arrangements, understandings, or agreements regarding the Licensed Material not stated herein are separate from and independent of the terms and conditions of this Public License. Section 8 – Interpretation. a. For the avoidance of doubt, this Public License does not, and shall not be interpreted to, reduce, limit, restrict, or impose conditions on any use of the Licensed Material that could lawfully be made without permission under this Public License. b. To the extent possible, if any provision of this Public License is deemed unenforceable, it shall be automatically reformed to the minimum extent necessary to make it enforceable. If the provision cannot be reformed, it shall be severed from this Public License without affecting the enforceability of the remaining terms and conditions. c. No term or condition of this Public License will be waived and no failure to comply consented to unless expressly agreed to by the Licensor. d. Nothing in this Public License constitutes or may be interpreted as a limitation upon, or waiver of, any privileges and immunities that apply to the Licensor or You, including from the legal processes of any jurisdiction or authority. Creative Commons is not a party to its public licenses. Notwithstanding, Creative Commons may elect to apply one of its public licenses to material it publishes and in those instances will be considered the "Licensor." The text of the Creative Commons public licenses is dedicated to the public domain under the CC0 Public Domain Dedication. Except for the limited purpose of indicating that material is shared under a Creative Commons public license or as otherwise permitted by the Creative Commons policies published at creativecommons.org/policies, Creative Commons does not authorize the use of the trademark "Creative Commons" or any other trademark or logo of Creative Commons without its prior written consent including, without limitation, in connection with any unauthorized modifications to any of its public licenses or any other arrangements, understandings, or agreements concerning use of licensed material. For the avoidance of doubt, this paragraph does not form part of the public licenses. Creative Commons may be contacted at creativecommons.org. chinese_wplc/README.md000066400000000000000000000217041411633436100150060ustar00rootroot00000000000000# 中文长下文词语预测 (Chinese WPLC)数据集 Chinese Word Prediction with Long Context (Chinese WPLC) 是天津大学联合鹏城实验室在小说上建立的依赖长上下文预测目标单词的中文数据集，创建目的是为了评测模型建模长文本的能力。 - [简介](#简介) - [数据采集与构建流程](#数据采集与构建流程) - [数据下载](#数据统计与分析) - [数据格式](#数据格式) - [评测脚本](#评测脚本) - [评测指标](#评测指标) ## 简介 该数据集在给定前文的条件下，测试机器预测目标句子最后一个单词的能力，选择的上下文、目标句子及待预测单词满足以下条件：当给定完整语境时待预测单词很容易被猜测出来，当只给最后一个句子时，难以被猜测出来。我们希望通过这个数据集检测模型在长上下文上提取信息的能力。例如： ```bash 上下文：随后他立即想到自己为什么如此气愤——他之所以气愤，是因为他害怕了。在他个人处于巨大危险的情况下，贝思抛弃了他。在海底深处只剩下他们三个人，他们互相需要——他们得互相依靠。 目标句：然而贝思不可信赖，这使他感到害怕，而且 目标词：气愤 ``` 在上述的例子中，目标词 “气愤“ 能够通过上下文和目标句推测出来，而单靠目标句，很难被猜测出来。 ## 数据采集与构建流程 - **Step1 数据收集** Chinese WPLC数据集来自网络爬取的小说，涵盖玄幻、言情、武侠、侦探、悬疑等类型，总量超过6万部。将重复的小说、公开读本（世界名著、文学名著、古典名著等）以及敏感词比例超5%的小说过滤后，剩余小说按照2：1：1的比例随机划分为训练集、测试集和验证集。 - **Step2 段落抽取** 使用pkuseg[[1](#参考文献)]对分句后的测试集和验证集小说段落进行分词，以段落最后一句为终点句子，在终点句子之前，往前累计总词数大于50的最少完整句子集合抽取出来作为上下文，并进一步将终点句子最后一个词作为待预测单词，终点句子剩余部分构成目标句子。上下文、目标句子、待预测单词共同组成一个上下文段落。抽取上下文段落需满足以下条件： >1. 目标词不是停用词。 >2. 目标词在训练集语料中词频大于5。 >3. pkuseg、jieba[[2](#参考文献)]、thulac[[3](#参考文献)]三种分词工具切分出来的目标词一致。 >4. 目标句子包含至少10个词。 >5. 每本小说最多抽取200个上下文段落。 仅当条件1不满足时，可将停用词前一个词作为目标词进行上述2-5条件检测，其余情况，将上下文段落抛弃。最终抽取出210万个段落。 - **Step3 段落过滤** 为减少数据集构建时间，需过滤掉相对简单的段落。使用以下四种组合生成答案候选： >1. 给定目标句子的预训练NEZHA[[4](#参考文献)]。 >2. 给定目标句子的微调NEZHA。 >3. 给定上下文和目标句子的预训练NEZHA。 >4. 给定上下文和目标句子的微调NEZHA。 当待预测单词出现在任一束搜索策略生成的Top-5个答案候选中时，将该段落抛弃。 为进一步减少人工标注量，在构建数据集过程将待预测单词困惑度在使用上下文和不使用上下文比值的对数作为指标。优先考虑指标大于1的段落，由此得到21万个段落。 - **Step4 人工筛选** 将Step3剩余段落经过随机抽样后通过100+标注人员进行三轮标注： >1. 在给定完整段落（上下文+目标句）猜测目标词，猜对后的上下文段落进入下一轮。 >2. 给不同标注人员重复第一轮。 >3. 给定目标句让三个不同的标注人员最多猜9个词，如果目标词都没有被猜到，则将该段落加入Chinese WPLC数据集。 第三轮标注中标注人员每人每个段落最多猜测3个词，以最大限度确保待预测单词需通过长上下文信息才能推断出来，而不能通过单个句子进行推断。虽然第三轮标注能够确保待预测单词不能通过局部信息推断，但是由于标注人员知识存在差异，该流程不能确保第一轮标注段落能被猜对。第二轮标注进一步确保上下文段落可以被猜对。为减少人工标注时间，在标注过程中提示标注人员待预测词的长度（字数）。 ## 数据统计与分析 经过第一轮后，只有14-17%的数据能进入下一轮，在第二轮中的数据中，有50%-60%的数据能进入第三轮。在第三轮标注中，只有60%的数据能够构成最后的Chinese WPLC数据集。数据统计如表1所示： | | 原数据 | 第一轮筛选后 | 第二轮筛选后 | 第三轮筛选后 | | :------: | :----: | :----------: | :----------: | :----------: | | **数量** | 21.3万 | 3.45万 | 1.85万 | 9301 | Chinese WPLC数据集每个上下文段落平均由3-4个句子构成，平均长度为120个字，如表2所示： | | 数量 | 长度 | 句子数量 | 下载 | | :------: | :--: | :--: | :------: | :----------------------------------------------------------: | | **Dev** | 4827 | 129.5 | 3.6 | [dev.json](https://git.openi.org.cn/PCL-Platform.Intelligence/Chinese_WPLC/datasets?type=0) | | **Test** | 4474 | 134.5 | 4.3 | [test_without_answer.json](https://git.openi.org.cn/PCL-Platform.Intelligence/Chinese_WPLC/datasets?type=0) | ## 数据格式 数据格式可参考下面两个样本数据，每个样本由2个字段组成，masked_text字段是上下文和目标句子，correct_word是位置上正确的词，数量代表待预测单词字数。 ```bash { "masked_text": "随后他立即想到自己为什么如此气愤——他之所以气愤，是因为他害怕了。在他个人处于巨大危险的情况下，贝思抛弃了他。在海底深处只剩下他们三个人，他们互相需要——他们得互相依靠。然而贝思不可信赖，这使他感到害怕，而且。"， "correct_word": "气愤" } ``` ```bash { "masked_text": "钟将也不躲不闪，只是简单的凝出一块雷光盾，只是他的雷光盾却不像楚毅峰的那样包裹住整个身体，只有脸盆大小，但是他的雷光盾完全是一块整体，没有雷电闪烁，没有电芒流转。甚至连一点雷电的痕迹都看不到，就是一声银色的坚实！", "correct_word": "盾牌" } ``` ## 评测脚本 本脚本基于 https://github.com/huawei-noah/Pretrained-Language-Model/tree/master/NEZHA-Gen-TensorFlow **版本注释** 版本1.0: 2021/04/22 **安装环境** 脚本使用Tensorflow 1.13和Python 3.6测试通过。python包``fire``是必需的。安装``fire``包： ``` pip3 install fire ``` **使用步骤** 1: 从[Google Drive](https://drive.google.com/drive/folders/1i4f_8LhaVDNjnGlLXNJ0rNgBP0E4L6V0?usp=sharing)或[Baidu Netdisk](https://pan.baidu.com/s/1Bgle8TpcxHyuUz_jAXOBWw)(code:rb5m)下载Chinese GPT的模型文件，将``models_gpt``文件夹移动到与脚本相同的目录中。 2: 将``dev.Json``文件移动到与脚本相同的目录。 3: 运行脚本 ``nezha_evaluate.py`` 。 ``` python3 nezha_evaluate.py ``` ## 评测指标 Chinese WPLC数据集的创建目的是为了评测模型建模长文本的能力。我们采用预测准确率（Accuracy）作为评测指标，该指标值越高代表模型处理长文本能力越强。 OpenAI公司发布的GPT-2在lambada数据集测得的指标是预测最后一个bpe分词，这会给模型带来额外的信息，降低任务的难度[[5](#参考文献)]。因此评测过程中我们不希望将目标词预测任务简化为最后一个子词预测任务。由于猜词过程中存在近义词，因此在评测过程中额外考虑TOP3准确率。 基准模型 为测试人类在Chinese WPLC数据集上的猜词效果，从验证集和测试集各随机抽取500条共1000条作为测试样本，并平均分发给4名人员，每人基于上下文猜测出3个单词计算准确率。 | Model | Top1 Accuracy（%） | Top3 Accuracy (%) | | :---: | :----------------: | :---------------: | | Human | 57.3 | 66.4 | | NEZHA | 0.0 | 0.0 | ## 参考文献 [1] Luo R, Xu J, Zhang Y, et al. Pkuseg: A toolkit for multi-domain chinese word segmentation[J]. arXiv preprint arXiv:1906.11455, 2019. [2] J. Sun, “Jieba Chinese word segmentation tool,” 2012 [3] 孙茂松, 陈新雄, 张开旭, 郭志芃, 刘知远. THULAC：一个高效的中文词法分析工具包. 2016. [4] Wei J, Ren X, Li X, et al. NEZHA: Neural contextualized representation for chinese language understanding[J]. arXiv preprint arXiv:1909.00204, 2019. [5] https://github.com/openai/gpt-2/issues/131 ## 项目信息 天津大学、鹏城实验室是 Chinese WPLC 数据集发布的主要单位。 ## 许可证: [Creative Commons Attribution 4.0 International](https://creativecommons.org/licenses/by/4.0/legalcode)chinese_wplc/nezha_evaluate/000077500000000000000000000000001411633436100165165ustar00rootroot00000000000000chinese_wplc/nezha_evaluate/README.md000066400000000000000000000020541411633436100177760ustar00rootroot00000000000000NEZHA-EVALUATE ============= The scripts and model are based on https://github.com/huawei-noah/Pretrained-Language-Model/tree/master/NEZHA-Gen-TensorFlow Release Notes ============= First version: 2021/04/22 Environment ============ The scripts are tested sucessfully with Tensorflow 1.13 and Python 3.6. The python package ``fire`` is required. You may need to install the ``fire`` package with the command: ``` pip3 install fire ``` Usage ==================== Step 1: Download the folder named ``models_gpt`` from the link below and move the folder to the same directory with the scripts. Step 2: Move the file named ``dev.json`` to the same directory with the scripts. Step 3: Run the script ``nezha_evaluate.py`` with the command: ``` python3 nezha_evaluate.py ``` Model download =========================== * Chinese GPT * [Google Drive](https://drive.google.com/drive/folders/1i4f_8LhaVDNjnGlLXNJ0rNgBP0E4L6V0?usp=sharing) * [Baidu Netdisk](https://pan.baidu.com/s/1Bgle8TpcxHyuUz_jAXOBWw)(code:rb5m) chinese_wplc/nezha_evaluate/nezha_evaluate.py000066400000000000000000001440271411633436100220730ustar00rootroot00000000000000#!/usr/bin/env python3 # coding=utf-8 # 2020/01/01 Changed for Nezha-Gen-Tensorflow. # Huawei Technologies Co., Ltd. # Copyright (C) 2019. Huawei Technologies Co., Ltd. All rights reserved. # Copyright 2018 The Google AI Language Team Authors. # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. from __future__ import absolute_import from __future__ import division from __future__ import print_function import collections import copy import json import math import re import six import fire import os import numpy as np import tensorflow as tf import tokenization import tqdm import random os.environ["CUDA_VISIBLE_DEVICES"] = "3" def set_tensor_by_indices_to_value(tensor, indices, value): # create value_tensor since tensor value assignment is not possible in TF value_tensor = tf.zeros_like(tensor) + value return tf.where(indices, value_tensor, tensor) class GPTConfig(object): """Configuration for `BertModel`.""" def __init__(self, vocab_size, temperature, hidden_size=768, num_hidden_layers=12, num_attention_heads=12, intermediate_size=3072, hidden_act="gelu", hidden_dropout_prob=0.1, attention_probs_dropout_prob=0.1, max_position_embeddings=512, type_vocab_size=16, initializer_range=0.02): """Constructs BertConfig. Args: vocab_size: Vocabulary size of `inputs_ids` in `BertModel`. hidden_size: Size of the encoder layers and the pooler layer. num_hidden_layers: Number of hidden layers in the Transformer encoder. num_attention_heads: Number of attention heads for each attention layer in the Transformer encoder. intermediate_size: The size of the "intermediate" (i.e., feed-forward) layer in the Transformer encoder. hidden_act: The non-linear activation function (function or string) in the encoder and pooler. hidden_dropout_prob: The dropout probability for all fully connected layers in the embeddings, encoder, and pooler. attention_probs_dropout_prob: The dropout ratio for the attention probabilities. max_position_embeddings: The maximum sequence length that this model might ever be used with. Typically set this to something large just in case (e.g., 512 or 1024 or 2048). type_vocab_size: The vocabulary size of the `token_type_ids` passed into `BertModel`. initializer_range: The stdev of the truncated_normal_initializer for initializing all weight matrices. """ self.vocab_size = vocab_size self.hidden_size = hidden_size self.num_hidden_layers = num_hidden_layers self.num_attention_heads = num_attention_heads self.hidden_act = hidden_act self.intermediate_size = intermediate_size self.hidden_dropout_prob = hidden_dropout_prob self.attention_probs_dropout_prob = attention_probs_dropout_prob self.max_position_embeddings = max_position_embeddings self.type_vocab_size = type_vocab_size self.initializer_range = initializer_range self.temperature = temperature class GenerativeModel(object): def __init__(self, config, init_input_ids, input_lens, target_lens, is_training = False, input_mask=None, token_type_ids=None, use_one_hot_embeddings=True, beam_size=5, scope=None): """Constructor for GenerativeModel. Args: config: `GPTConfig` instance. is_training: False init_input_ids: int32 Tensor of shape [batch_size, seq_length]. input_mask: (optional) int32 Tensor of shape [batch_size, seq_length]. token_type_ids: (optional) int32 Tensor of shape [batch_size, seq_length]. use_one_hot_embeddings: (optional) bool. Whether to use one-hot word embeddings or tf.embedding_lookup() for the word embeddings. On the TPU, it is much faster if this is True, on the CPU or GPU, it is faster if this is False. scope: (optional) variable scope. Defaults to "bert". Raises: ValueError: The config is invalid or one of the input tensor shapes is invalid. """ def get_predict_output(bert_config, embedding_table,sequence_output, beam_scores): """Get loss and log probs for the masked LM.""" input_tensor = sequence_output[:, -1,:] with tf.variable_scope("cls/predictions",reuse=tf.AUTO_REUSE): # We apply one more non-linear transformation before the output layer. # This matrix is not used after pre-training. with tf.variable_scope("transform",reuse=tf.AUTO_REUSE): input_tensor = tf.layers.dense( input_tensor, units=bert_config.hidden_size, activation=get_activation(bert_config.hidden_act), kernel_initializer=create_initializer( bert_config.initializer_range)) input_tensor = layer_norm(input_tensor) # The output weights are the same as the input embeddings, but there is # an output-only bias for each token. output_bias = tf.get_variable( "output_bias", shape=[bert_config.vocab_size], initializer=tf.zeros_initializer()) logits = tf.matmul(input_tensor, embedding_table, transpose_b=True) logits = tf.nn.bias_add(logits, output_bias) if bert_config.temperature != 1: logits = logits / tf.to_float(bert_config.temperature) scores = tf.nn.log_softmax(logits, axis=-1) #(num_beam, vocab_size) #beam score () #beam_scores = tf.zeros((beam_size,)) _scores = scores + tf.broadcast_to( beam_scores[:, None], (beam_size, bert_config.vocab_size) ) scores = tf.reshape(_scores, (-1,))# (num_beam * vocab_size) top_k_scores, topk_output_ids = tf.nn.top_k(scores, k=beam_size) output_id = tf.squeeze(topk_output_ids % bert_config.vocab_size) beam_id = topk_output_ids / bert_config.vocab_size return output_id, top_k_scores, beam_id def get_init_predict_output(bert_config, embedding_table, sequence_output): """Get loss and log probs for the masked LM.""" input_tensor = tf.expand_dims(sequence_output[0, -1, :], 0) with tf.variable_scope("cls/predictions", reuse=tf.AUTO_REUSE): # We apply one more non-linear transformation before the output layer. # This matrix is not used after pre-training. with tf.variable_scope("transform", reuse=tf.AUTO_REUSE): input_tensor = tf.layers.dense( input_tensor, units=bert_config.hidden_size, activation=get_activation(bert_config.hidden_act), kernel_initializer=create_initializer( bert_config.initializer_range)) input_tensor = layer_norm(input_tensor) # The output weights are the same as the input embeddings, but there is # an output-only bias for each token. output_bias = tf.get_variable( "output_bias", shape=[bert_config.vocab_size], initializer=tf.zeros_initializer()) logits = tf.matmul(input_tensor, embedding_table, transpose_b=True) logits = tf.nn.bias_add(logits, output_bias) if bert_config.temperature != 1: logits = logits / tf.to_float(bert_config.temperature) scores = tf.nn.log_softmax(logits, axis=-1) #(beam_size, vocab_size) beam_scores = tf.zeros((beam_size,)) #_scores = scores + tf.broadcast_to( # beam_scores[:, None], (beam_size, bert_config.vocab_size) #) #scores = tf.reshape(_scores, [-1]) # (1, num_beam * vocab_size) topk_values, topk_output_ids = tf.nn.top_k(scores, k=beam_size) #output_id = tf.multinomial(topk_output_ids, num_samples=beam_size, output_dtype=tf.int32) output_id = tf.squeeze(topk_output_ids % bert_config.vocab_size) #output_id = topk_output_ids[0] return output_id, tf.squeeze(topk_values) config = copy.deepcopy(config) if not is_training: config.hidden_dropout_prob = 0.0 config.attention_probs_dropout_prob = 0.0 init_input_ids = tf.tile(init_input_ids, [beam_size, 1]) input_shape = get_shape_list(init_input_ids, expected_rank=2) batch_size = input_shape[0] init_seq_length = input_shape[1] if input_mask is None: input_mask = tf.ones(shape=[batch_size, init_seq_length], dtype=tf.int32) if token_type_ids is None: use_token_type = False def step(input_ids, past = None, predict_token_idx = None, beam_score=None): """ In each step, one token is generated. past: Since GPT uses unidirectional attention, the precomputed tokens will remain the same. Therefore, we do not re-compute all the hidden_states, but store the hidden_sstate and reuse them afterwards. predict_token_idx: the current position of for predicting the next token. """ with tf.variable_scope("bert", reuse=tf.AUTO_REUSE): with tf.variable_scope("embeddings"): # Perform embedding lookup on the word ids. (self.embedding_output, self.embedding_table) = embedding_lookup( input_ids=input_ids, vocab_size=config.vocab_size, embedding_size=config.hidden_size, initializer_range=config.initializer_range, word_embedding_name="word_embeddings", use_one_hot_embeddings=use_one_hot_embeddings) # Add positional embeddings and token type embeddings, then layer # normalize and perform dropout. self.embedding_output = embedding_postprocessor( input_tensor=self.embedding_output, use_token_type=use_token_type, token_type_ids=token_type_ids, token_type_vocab_size=config.type_vocab_size, token_type_embedding_name="token_type_embeddings", use_position_embeddings=True, position_embedding_name="position_embeddings", initializer_range=config.initializer_range, max_position_embeddings=config.max_position_embeddings, dropout_prob=config.hidden_dropout_prob, predict_token_idx = predict_token_idx ) with tf.variable_scope("encoder"): # This converts a 2D mask of shape [batch_size, seq_length] to a 3D # mask of shape [batch_size, seq_length, seq_length] which is used # for the attention scores. #attention_mask = create_attention_mask_from_input_mask( # input_ids, input_mask) # create a mask for gpt : 1's in the lower triangle, counting from the lower right corner--by liaoyi attention_mask = create_attention_mask_from_input_mask(input_ids,input_mask,predict_token_idx = predict_token_idx) # Run the stacked transformer. # `sequence_output` shape = [batch_size, seq_length, hidden_size]. self.all_encoder_layers, past, predict_token_idx = transformer_model( input_tensor=self.embedding_output, attention_mask=attention_mask, hidden_size=config.hidden_size, num_hidden_layers=config.num_hidden_layers, num_attention_heads=config.num_attention_heads, intermediate_size=config.intermediate_size, intermediate_act_fn=get_activation(config.hidden_act), hidden_dropout_prob=config.hidden_dropout_prob, attention_probs_dropout_prob=config.attention_probs_dropout_prob, initializer_range=config.initializer_range, do_return_all_layers=True, past = past, predict_token_idx = predict_token_idx ) self.sequence_output = self.all_encoder_layers[-1] predicted_logit, new_beam_score, beam_id = get_predict_output(config, self.embedding_table, self.sequence_output, beam_score) predicted_logit = tf.cast(tf.expand_dims(predicted_logit, 1), tf.int32) beam_id = tf.cast(beam_id, tf.int32) past = tf.gather(past, beam_id,axis=2) return predicted_logit, past, predict_token_idx, new_beam_score, beam_id def init_step(input_ids, past=None, predict_token_idx=None): """ In each step, one token is generated. past: Since GPT uses unidirectional attention, the precomputed tokens will remain the same. Therefore, we do not re-compute all the hidden_states, but store the hidden_sstate and reuse them afterwards. predict_token_idx: the current position of for predicting the next token. """ with tf.variable_scope("bert", reuse=tf.AUTO_REUSE): with tf.variable_scope("embeddings"): # Perform embedding lookup on the word ids. (self.embedding_output, self.embedding_table) = embedding_lookup( input_ids=input_ids, vocab_size=config.vocab_size, embedding_size=config.hidden_size, initializer_range=config.initializer_range, word_embedding_name="word_embeddings", use_one_hot_embeddings=use_one_hot_embeddings) # Add positional embeddings and token type embeddings, then layer # normalize and perform dropout. self.embedding_output = embedding_postprocessor( input_tensor=self.embedding_output, use_token_type=use_token_type, token_type_ids=token_type_ids, token_type_vocab_size=config.type_vocab_size, token_type_embedding_name="token_type_embeddings", use_position_embeddings=True, position_embedding_name="position_embeddings", initializer_range=config.initializer_range, max_position_embeddings=config.max_position_embeddings, dropout_prob=config.hidden_dropout_prob, predict_token_idx=predict_token_idx ) with tf.variable_scope("encoder"): # This converts a 2D mask of shape [batch_size, seq_length] to a 3D # mask of shape [batch_size, seq_length, seq_length] which is used # for the attention scores. # attention_mask = create_attention_mask_from_input_mask( # input_ids, input_mask) # create a mask for gpt : 1's in the lower triangle, counting from the lower right corner--by liaoyi attention_mask = create_attention_mask_from_input_mask(input_ids, input_mask, predict_token_idx=predict_token_idx) # Run the stacked transformer. # `sequence_output` shape = [batch_size, seq_length, hidden_size]. self.all_encoder_layers, past, predict_token_idx = transformer_model( input_tensor=self.embedding_output, attention_mask=attention_mask, hidden_size=config.hidden_size, num_hidden_layers=config.num_hidden_layers, num_attention_heads=config.num_attention_heads, intermediate_size=config.intermediate_size, intermediate_act_fn=get_activation(config.hidden_act), hidden_dropout_prob=config.hidden_dropout_prob, attention_probs_dropout_prob=config.attention_probs_dropout_prob, initializer_range=config.initializer_range, do_return_all_layers=True, past=past, predict_token_idx=predict_token_idx ) self.sequence_output = self.all_encoder_layers[-1] predicted_logit, init_score = get_init_predict_output(config, self.embedding_table, self.sequence_output) predicted_logit = tf.expand_dims(predicted_logit, 1) return predicted_logit, past, predict_token_idx, init_score def cond(past, predicted_logit,predict_token_idx, sequence_predict,beam_score): # The condition to stop generation. In this example, we stop generation when the sequence length 511 is reached return (predict_token_idx - input_lens) < (target_lens -1) def body(past, predicted_logit, predict_token_idx, sequence_predict, beam_score): """ past: the stored hidden state predicted_logit: the predicted logit in the last step, which serves as the input to the GPT model in this step for predicting next token predict_token_idx: the position for predicting the next token sequence_predict: the sequence of all the predicted tokens so far, which will be output as the final generated sequence. """ new_predicted_logit, new_past, new_predict_token_idx, new_beam_score, beam_id = step(predicted_logit, past=past, predict_token_idx=predict_token_idx, beam_score=beam_score) sequence_predict = tf.gather(sequence_predict, beam_id , axis = 0) sequence_predict = tf.concat([sequence_predict,new_predicted_logit], axis =1) return [new_past, new_predicted_logit, new_predict_token_idx, sequence_predict, new_beam_score] #run the first step to initialize the variables in the loop init_predict, init_past, init_predict_token_idx, init_beam_score = init_step(init_input_ids,past=None,predict_token_idx=0) #init_predict = tf.tile(init_predict, [5, 1]) _,_,_,self.predicted_tokens, self.beam_score = tf.while_loop( cond=cond, body=body, loop_vars = [ init_past, init_predict, init_predict_token_idx, init_predict, init_beam_score, ], #BNTH shape_invariants=[ tf.TensorShape([12, 2,beam_size, 12,None, 64]), tf.TensorShape([beam_size,None]), tf.TensorShape([]), tf.TensorShape([beam_size,None]), tf.TensorShape([beam_size]) ], back_prop = False ) @classmethod def encodetext(cls, text, tokenizer): unicode_text = tokenization.convert_to_unicode(text) tokenized = tokenizer.tokenize(unicode_text) #bos = ["[CLS]"] #bos.extend(tokenized) return tokenizer.convert_tokens_to_ids(tokenized) @classmethod def decodetext(cls, text, tokenizer): return tokenizer.convert_ids_to_tokens(text) def get_predicted_tokens(self): return self.predicted_tokens def get_pooled_output(self): return self.pooled_output def get_past(self): return self.past def gelu(input_tensor): """Gaussian Error Linear Unit. This is a smoother version of the RELU. Original paper: https://arxiv.org/abs/1606.08415 Args: input_tensor: float Tensor to perform activation. Returns: `input_tensor` with the GELU activation applied. """ cdf = 0.5 * (1.0 + tf.erf(input_tensor / tf.sqrt(2.0))) return input_tensor * cdf def get_activation(activation_string): """Maps a string to a Python function, e.g., "relu" => `tf.nn.relu`. Args: activation_string: String name of the activation function. Returns: A Python function corresponding to the activation function. If `activation_string` is None, empty, or "linear", this will return None. If `activation_string` is not a string, it will return `activation_string`. Raises: ValueError: The `activation_string` does not correspond to a known activation. """ # We assume that anything that"s not a string is already an activation # function, so we just return it. if not isinstance(activation_string, six.string_types): return activation_string if not activation_string: return None act = activation_string.lower() if act == "linear": return None elif act == "relu": return tf.nn.relu elif act == "gelu": return gelu elif act == "tanh": return tf.tanh else: raise ValueError("Unsupported activation: %s" % act) def dropout(input_tensor, dropout_prob): """Perform dropout. Args: input_tensor: float Tensor. dropout_prob: Python float. The probability of dropping out a value (NOT of *keeping* a dimension as in `tf.nn.dropout`). Returns: A version of `input_tensor` with dropout applied. """ if dropout_prob is None or dropout_prob == 0.0: return input_tensor output = tf.nn.dropout(input_tensor, 1.0 - dropout_prob) return output def layer_norm(input_tensor, name=None): """Run layer normalization on the last dimension of the tensor.""" return tf.contrib.layers.layer_norm( inputs=input_tensor, begin_norm_axis=-1, begin_params_axis=-1, scope=name) def layer_norm_and_dropout(input_tensor, dropout_prob, name=None): """Runs layer normalization followed by dropout.""" output_tensor = layer_norm(input_tensor, name) output_tensor = dropout(output_tensor, dropout_prob) return output_tensor def create_initializer(initializer_range=0.02): """Creates a `truncated_normal_initializer` with the given range.""" return tf.truncated_normal_initializer(stddev=initializer_range) def embedding_lookup(input_ids, vocab_size, embedding_size=128, initializer_range=0.02, word_embedding_name="word_embeddings", use_one_hot_embeddings=False): """Looks up words embeddings for id tensor. Args: input_ids: int32 Tensor of shape [batch_size, seq_length] containing word ids. vocab_size: int. Size of the embedding vocabulary. embedding_size: int. Width of the word embeddings. initializer_range: float. Embedding initialization range. word_embedding_name: string. Name of the embedding table. use_one_hot_embeddings: bool. If True, use one-hot method for word embeddings. If False, use `tf.nn.embedding_lookup()`. One hot is better for TPUs. Returns: float Tensor of shape [batch_size, seq_length, embedding_size]. """ # This function assumes that the input is of shape [batch_size, seq_length, # num_inputs]. # # If the input is a 2D tensor of shape [batch_size, seq_length], we # reshape to [batch_size, seq_length, 1]. if input_ids.shape.ndims == 2: input_ids = tf.expand_dims(input_ids, axis=[-1]) embedding_table = tf.get_variable( name=word_embedding_name, shape=[vocab_size, embedding_size], initializer=create_initializer(initializer_range)) if use_one_hot_embeddings: flat_input_ids = tf.reshape(input_ids, [-1]) one_hot_input_ids = tf.one_hot(flat_input_ids, depth=vocab_size) output = tf.matmul(one_hot_input_ids, embedding_table) else: output = tf.nn.embedding_lookup(embedding_table, input_ids) input_shape = get_shape_list(input_ids) output = tf.reshape(output, input_shape[0:-1] + [input_shape[-1] * embedding_size]) return (output, embedding_table) def embedding_postprocessor(input_tensor, use_token_type=False, token_type_ids=None, token_type_vocab_size=16, token_type_embedding_name="token_type_embeddings", use_position_embeddings=True, position_embedding_name="position_embeddings", initializer_range=0.02, max_position_embeddings=512, dropout_prob=0.1, predict_token_idx = 0 ): """Performs various post-processing on a word embedding tensor. Args: input_tensor: float Tensor of shape [batch_size, seq_length, embedding_size]. In this case, seq_length is usually equals to 1. use_token_type: bool. Whether to add embeddings for `token_type_ids`. token_type_ids: (optional) int32 Tensor of shape [batch_size, seq_length]. Must be specified if `use_token_type` is True. token_type_vocab_size: int. The vocabulary size of `token_type_ids`. token_type_embedding_name: string. The name of the embedding table variable for token type ids. use_position_embeddings: bool. Whether to add position embeddings for the position of each token in the sequence. position_embedding_name: string. The name of the embedding table variable for positional embeddings. initializer_range: float. Range of the weight initialization. max_position_embeddings: int. Maximum sequence length that might ever be used with this model. This can be longer than the sequence length of input_tensor, but cannot be shorter. dropout_prob: float. Dropout probability applied to the final output tensor. Returns: float tensor with same shape as `input_tensor`. Raises: ValueError: One of the tensor shapes or input values is invalid. """ input_shape = get_shape_list(input_tensor, expected_rank=3) batch_size = input_shape[0] seq_length = input_shape[1] width = input_shape[2] output = input_tensor if use_token_type: if token_type_ids is None: raise ValueError("`token_type_ids` must be specified if" "`use_token_type` is True.") token_type_table = tf.get_variable( name=token_type_embedding_name, shape=[token_type_vocab_size, width], initializer=create_initializer(initializer_range)) # This vocab will be small so we always do one-hot here, since it is always # faster for a small vocabulary. flat_token_type_ids = tf.reshape(token_type_ids, [-1]) one_hot_ids = tf.one_hot(flat_token_type_ids, depth=token_type_vocab_size) token_type_embeddings = tf.matmul(one_hot_ids, token_type_table) token_type_embeddings = tf.reshape(token_type_embeddings, [batch_size, seq_length, width]) output += token_type_embeddings if use_position_embeddings: assert_op = tf.assert_less_equal(seq_length, max_position_embeddings) with tf.control_dependencies([assert_op]): full_position_embeddings = tf.get_variable( name=position_embedding_name, shape=[max_position_embeddings, width], initializer=create_initializer(initializer_range)) # Since the position embedding table is a learned variable, we create it # using a (long) sequence length `max_position_embeddings`. The actual # sequence length might be shorter than this, for faster training of # tasks that do not have long sequences. # # So `full_position_embeddings` is effectively an embedding table # for position [0, 1, 2, ..., max_position_embeddings-1], and the current # sequence has positions [0, 1, 2, ... seq_length-1], so we can just # perform a slice. position_embeddings = tf.slice(full_position_embeddings, [predict_token_idx, 0], [seq_length, -1]) num_dims = len(output.shape.as_list()) # Only the last two dimensions are relevant (`seq_length` and `width`), so # we broadcast among the first dimensions, which is typically just # the batch size. position_broadcast_shape = [] for _ in range(num_dims - 2): position_broadcast_shape.append(1) position_broadcast_shape.extend([seq_length, width]) position_embeddings = tf.reshape(position_embeddings, position_broadcast_shape) output += position_embeddings output = layer_norm_and_dropout(output, dropout_prob) return output def create_attention_mask_from_input_mask(from_tensor, to_mask, predict_token_idx= None): """Create 3D attention mask from a 2D tensor mask. Args: from_tensor: 2D or 3D Tensor of shape [batch_size, from_seq_length, ...]. to_mask: int32 Tensor of shape [batch_size, to_seq_length]. Returns: float Tensor of shape [batch_size, from_seq_length, to_seq_length]. """ from_shape = get_shape_list(from_tensor, expected_rank=[2, 3]) batch_size = from_shape[0] from_seq_length = from_shape[1] to_shape = get_shape_list(to_mask, expected_rank=2) to_seq_length = to_shape[1] to_mask = tf.cast( tf.reshape(to_mask, [batch_size, 1, to_seq_length]), tf.float32) if predict_token_idx != None: to_mask = tf.ones([batch_size,1,from_seq_length+predict_token_idx],tf.float32) # We don't assume that `from_tensor` is a mask (although it could be). We # don't actually care if we attend *from* padding tokens (only *to* padding) # tokens so we create a tensor of all ones. # # `broadcast_ones` = [batch_size, from_seq_length, 1] broadcast_ones = tf.ones( shape=[batch_size, from_seq_length+predict_token_idx, 1], dtype=tf.float32) # Here we broadcast along two dimensions to create the mask. mask = broadcast_ones * to_mask mask = tf.matrix_band_part( input=mask, num_lower=-1, num_upper=0) return mask def attention_layer(from_tensor, to_tensor, attention_mask=None, num_attention_heads=1, size_per_head=512, query_act=None, key_act=None, value_act=None, attention_probs_dropout_prob=0.0, initializer_range=0.02, do_return_2d_tensor=False, batch_size=None, from_seq_length=None, to_seq_length=None, past = None, predict_token_idx = None ): """Performs multi-headed attention from `from_tensor` to `to_tensor`. This is an implementation of multi-headed attention based on "Attention is all you Need". If `from_tensor` and `to_tensor` are the same, then this is self-attention. Each timestep in `from_tensor` attends to the corresponding sequence in `to_tensor`, and returns a fixed-with vector. This function first projects `from_tensor` into a "query" tensor and `to_tensor` into "key" and "value" tensors. These are (effectively) a list of tensors of length `num_attention_heads`, where each tensor is of shape [batch_size, seq_length, size_per_head]. Then, the query and key tensors are dot-producted and scaled. These are softmaxed to obtain attention probabilities. The value tensors are then interpolated by these probabilities, then concatenated back to a single tensor and returned. In practice, the multi-headed attention are done with transposes and reshapes rather than actual separate tensors. Args: from_tensor: float Tensor of shape [batch_size, from_seq_length, from_width]. to_tensor: float Tensor of shape [batch_size, to_seq_length, to_width]. attention_mask: (optional) int32 Tensor of shape [batch_size, from_seq_length, to_seq_length]. The values should be 1 or 0. The attention scores will effectively be set to -infinity for any positions in the mask that are 0, and will be unchanged for positions that are 1. num_attention_heads: int. Number of attention heads. size_per_head: int. Size of each attention head. query_act: (optional) Activation function for the query transform. key_act: (optional) Activation function for the key transform. value_act: (optional) Activation function for the value transform. attention_probs_dropout_prob: (optional) float. Dropout probability of the attention probabilities. initializer_range: float. Range of the weight initializer. do_return_2d_tensor: bool. If True, the output will be of shape [batch_size * from_seq_length, num_attention_heads * size_per_head]. If False, the output will be of shape [batch_size, from_seq_length, num_attention_heads * size_per_head]. batch_size: (Optional) int. If the input is 2D, this might be the batch size of the 3D version of the `from_tensor` and `to_tensor`. from_seq_length: (Optional) If the input is 2D, this might be the seq length of the 3D version of the `from_tensor`. to_seq_length: (Optional) If the input is 2D, this might be the seq length of the 3D version of the `to_tensor`. Returns: float Tensor of shape [batch_size, from_seq_length, num_attention_heads * size_per_head]. (If `do_return_2d_tensor` is true, this will be of shape [batch_size * from_seq_length, num_attention_heads * size_per_head]). Raises: ValueError: Any of the arguments or tensor shapes are invalid. """ def transpose_for_scores(input_tensor, batch_size, num_attention_heads, seq_length, width): output_tensor = tf.reshape( input_tensor, [batch_size, seq_length, num_attention_heads, width]) output_tensor = tf.transpose(output_tensor, [0, 2, 1, 3]) return output_tensor from_shape = get_shape_list(from_tensor, expected_rank=[2, 3]) to_shape = get_shape_list(to_tensor, expected_rank=[2, 3]) if len(from_shape) != len(to_shape): raise ValueError( "The rank of `from_tensor` must match the rank of `to_tensor`.") if len(from_shape) == 3: batch_size = from_shape[0] from_seq_length = from_shape[1] to_seq_length = to_shape[1] elif len(from_shape) == 2: if (batch_size is None or from_seq_length is None or to_seq_length is None): raise ValueError( "When passing in rank 2 tensors to attention_layer, the values " "for `batch_size`, `from_seq_length`, and `to_seq_length` " "must all be specified.") # Scalar dimensions referenced here: # B = batch size (number of sequences) # F = `from_tensor` sequence length # T = `to_tensor` sequence length # N = `num_attention_heads` # H = `size_per_head` from_tensor_2d = reshape_to_matrix(from_tensor) to_tensor_2d = reshape_to_matrix(to_tensor) # `query_layer` = [B*F, N*H] query_layer = tf.layers.dense( from_tensor_2d, num_attention_heads * size_per_head, activation=query_act, name="query", kernel_initializer=create_initializer(initializer_range)) # `key_layer` = [B*T, N*H] key_layer = tf.layers.dense( to_tensor_2d, num_attention_heads * size_per_head, activation=key_act, name="key", kernel_initializer=create_initializer(initializer_range)) # `value_layer` = [B*T, N*H] value_layer = tf.layers.dense( to_tensor_2d, num_attention_heads * size_per_head, activation=value_act, name="value", kernel_initializer=create_initializer(initializer_range)) # `query_layer` = [B, N, F, H] query_layer = transpose_for_scores(query_layer, batch_size, num_attention_heads, from_seq_length, size_per_head) # `key_layer` = [B, N, T, H] key_layer = transpose_for_scores(key_layer, batch_size, num_attention_heads, to_seq_length, size_per_head) if past!=None: past_key_layer = past[0] key_layer = tf.concat([past_key_layer, key_layer],axis = 2) # Take the dot product between "query" and "key" to get the raw # attention scores. # `attention_scores` = [B, N, F, T] attention_scores = tf.matmul(query_layer, key_layer, transpose_b=True) attention_scores = tf.multiply(attention_scores, 1.0 / math.sqrt(float(size_per_head))) if attention_mask is not None: # `attention_mask` = [B, 1, F, T] attention_mask = tf.expand_dims(attention_mask, axis=[1]) attention_mask = attention_mask[:,:,predict_token_idx:predict_token_idx+to_seq_length,:] # Since attention_mask is 1.0 for positions we want to attend and 0.0 for # masked positions, this operation will create a tensor which is 0.0 for # positions we want to attend and -10000.0 for masked positions. adder = (1.0 - tf.cast(attention_mask, tf.float32)) * -10000.0 # Since we are adding it to the raw scores before the softmax, this is # effectively the same as removing these entirely. attention_scores += adder # Normalize the attention scores to probabilities. # `attention_probs` = [B, N, F, T] attention_probs = tf.nn.softmax(attention_scores) # This is actually dropping out entire tokens to attend to, which might # seem a bit unusual, but is taken from the original Transformer paper. attention_probs = dropout(attention_probs, attention_probs_dropout_prob) # `value_layer` = [B, T, N, H] value_layer = tf.reshape( value_layer, [batch_size, to_seq_length, num_attention_heads, size_per_head]) # `value_layer` = [B, N, T, H] value_layer = tf.transpose(value_layer, [0, 2, 1, 3]) if past != None: past_value_layer = past[1] value_layer = tf.concat([past_value_layer, value_layer],axis = 2) # `context_layer` = [B, N, F, H] context_layer = tf.matmul(attention_probs, value_layer) # `context_layer` = [B, F, N, H] context_layer = tf.transpose(context_layer, [0, 2, 1, 3]) if do_return_2d_tensor: # `context_layer` = [B*F, N*H] context_layer = tf.reshape( context_layer, [batch_size * from_seq_length, num_attention_heads * size_per_head]) else: # `context_layer` = [B, F, N*H] context_layer = tf.reshape( context_layer, [batch_size, from_seq_length, num_attention_heads * size_per_head]) past_layer = tf.stack([key_layer,value_layer],axis=0) return context_layer, past_layer def transformer_model(input_tensor, attention_mask=None, hidden_size=768, num_hidden_layers=12, num_attention_heads=12, intermediate_size=3072, intermediate_act_fn=gelu, hidden_dropout_prob=0.1, attention_probs_dropout_prob=0.1, initializer_range=0.02, do_return_all_layers=False, past = None, predict_token_idx = None ): """Multi-headed, multi-layer Transformer from "Attention is All You Need". This is almost an exact implementation of the original Transformer encoder. See the original paper: https://arxiv.org/abs/1706.03762 Also see: https://github.com/tensorflow/tensor2tensor/blob/master/tensor2tensor/models/transformer.py Args: input_tensor: float Tensor of shape [batch_size, seq_length, hidden_size]. attention_mask: (optional) int32 Tensor of shape [batch_size, seq_length, seq_length], with 1 for positions that can be attended to and 0 in positions that should not be. hidden_size: int. Hidden size of the Transformer. num_hidden_layers: int. Number of layers (blocks) in the Transformer. num_attention_heads: int. Number of attention heads in the Transformer. intermediate_size: int. The size of the "intermediate" (a.k.a., feed forward) layer. intermediate_act_fn: function. The non-linear activation function to apply to the output of the intermediate/feed-forward layer. hidden_dropout_prob: float. Dropout probability for the hidden layers. attention_probs_dropout_prob: float. Dropout probability of the attention probabilities. initializer_range: float. Range of the initializer (stddev of truncated normal). do_return_all_layers: Whether to also return all layers or just the final layer. Returns: float Tensor of shape [batch_size, seq_length, hidden_size], the final hidden layer of the Transformer. Raises: ValueError: A Tensor shape or parameter is invalid. """ if hidden_size % num_attention_heads != 0: raise ValueError( "The hidden size (%d) is not a multiple of the number of attention " "heads (%d)" % (hidden_size, num_attention_heads)) attention_head_size = int(hidden_size / num_attention_heads) input_shape = get_shape_list(input_tensor, expected_rank=3) batch_size = input_shape[0] seq_length = input_shape[1] input_width = input_shape[2] # The Transformer performs sum residuals on all layers so the input needs # to be the same as the hidden size. if input_width != hidden_size: raise ValueError("The width of the input tensor (%d) != hidden size (%d)" % (input_width, hidden_size)) # We keep the representation as a 2D tensor to avoid re-shaping it back and # forth from a 3D tensor to a 2D tensor. Re-shapes are normally free on # the GPU/CPU but may not be free on the TPU, so we want to minimize them to # help the optimizer. prev_output = reshape_to_matrix(input_tensor) all_layer_outputs = [] past_layers = [] for layer_idx in range(num_hidden_layers): with tf.variable_scope("layer_%d" % layer_idx): layer_input = prev_output if past == None: current_past = None else: current_past = past[layer_idx] with tf.variable_scope("attention"): attention_heads = [] with tf.variable_scope("self"): attention_head, past_layer = attention_layer( from_tensor=layer_input, to_tensor=layer_input, attention_mask=attention_mask, num_attention_heads=num_attention_heads, size_per_head=attention_head_size, attention_probs_dropout_prob=attention_probs_dropout_prob, initializer_range=initializer_range, do_return_2d_tensor=True, batch_size=batch_size, from_seq_length=seq_length, to_seq_length=seq_length, past = current_past, predict_token_idx = predict_token_idx ) attention_heads.append(attention_head) past_layers.append(past_layer) attention_output = None if len(attention_heads) == 1: attention_output = attention_heads[0] else: # In the case where we have other sequences, we just concatenate # them to the self-attention head before the projection. attention_output = tf.concat(attention_heads, axis=-1) # Run a linear projection of `hidden_size` then add a residual # with `layer_input`. with tf.variable_scope("output"): attention_output = tf.layers.dense( attention_output, hidden_size, kernel_initializer=create_initializer(initializer_range)) attention_output = dropout(attention_output, hidden_dropout_prob) attention_output = layer_norm(attention_output + layer_input) # The activation is only applied to the "intermediate" hidden layer. with tf.variable_scope("intermediate"): intermediate_output = tf.layers.dense( attention_output, intermediate_size, activation=intermediate_act_fn, kernel_initializer=create_initializer(initializer_range)) # Down-project back to `hidden_size` then add the residual. with tf.variable_scope("output"): layer_output = tf.layers.dense( intermediate_output, hidden_size, kernel_initializer=create_initializer(initializer_range)) layer_output = dropout(layer_output, hidden_dropout_prob) layer_output = layer_norm(layer_output + attention_output) prev_output = layer_output all_layer_outputs.append(layer_output) past_layers = tf.stack(past_layers,axis=0) if do_return_all_layers: final_outputs = [] for layer_output in all_layer_outputs: final_output = reshape_from_matrix(layer_output, input_shape) final_outputs.append(final_output) return final_outputs, past_layers, predict_token_idx+seq_length else: final_output = reshape_from_matrix(prev_output, input_shape) return final_output, past_layers, predict_token_idx+seq_length def get_shape_list(tensor, expected_rank=None, name=None): """Returns a list of the shape of tensor, preferring static dimensions. Args: tensor: A tf.Tensor object to find the shape of. expected_rank: (optional) int. The expected rank of `tensor`. If this is specified and the `tensor` has a different rank, and exception will be thrown. name: Optional name of the tensor for the error message. Returns: A list of dimensions of the shape of tensor. All static dimensions will be returned as python integers, and dynamic dimensions will be returned as tf.Tensor scalars. """ if name is None: name = tensor.name if expected_rank is not None: assert_rank(tensor, expected_rank, name) shape = tensor.shape.as_list() non_static_indexes = [] for (index, dim) in enumerate(shape): if dim is None: non_static_indexes.append(index) if not non_static_indexes: return shape dyn_shape = tf.shape(tensor) for index in non_static_indexes: shape[index] = dyn_shape[index] return shape def reshape_to_matrix(input_tensor): """Reshapes a >= rank 2 tensor to a rank 2 tensor (i.e., a matrix).""" ndims = input_tensor.shape.ndims if ndims < 2: raise ValueError("Input tensor must have at least rank 2. Shape = %s" % (input_tensor.shape)) if ndims == 2: return input_tensor width = input_tensor.shape[-1] output_tensor = tf.reshape(input_tensor, [-1, width]) return output_tensor def reshape_from_matrix(output_tensor, orig_shape_list): """Reshapes a rank 2 tensor back to its original rank >= 2 tensor.""" if len(orig_shape_list) == 2: return output_tensor output_shape = get_shape_list(output_tensor) orig_dims = orig_shape_list[0:-1] width = output_shape[-1] return tf.reshape(output_tensor, orig_dims + [width]) def assert_rank(tensor, expected_rank, name=None): """Raises an exception if the tensor rank is not of the expected rank. Args: tensor: A tf.Tensor to check the rank of. expected_rank: Python integer or list of integers, expected rank. name: Optional name of the tensor for the error message. Raises: ValueError: If the expected shape doesn't match the actual shape. """ if name is None: name = tensor.name expected_rank_dict = {} if isinstance(expected_rank, six.integer_types): expected_rank_dict[expected_rank] = True else: for x in expected_rank: expected_rank_dict[x] = True actual_rank = tensor.shape.ndims if actual_rank not in expected_rank_dict: scope_name = tf.get_variable_scope().name raise ValueError( "For the tensor `%s` in scope `%s`, the actual rank " "`%d` (shape = %s) is not equal to the expected rank `%s`" % (name, scope_name, actual_rank, str(tensor.shape), str(expected_rank))) def get_data(file_name, tokenizer, ): datasets = [] with open(file_name, 'r') as f: for line in tqdm.tqdm(f.readlines()): example = {} passes = json.loads(line) text = passes['masked_text'].split('')[0] context_tokens = GenerativeModel.encodetext(text, tokenizer) input_context_tokens = [101] input_context_tokens.extend(context_tokens) example['input_ids'] = input_context_tokens example['input_len'] = len(input_context_tokens) example['target_len'] = passes['masked_text'].count('') example['correct_word'] = passes['correct_word'] datasets.append(example) f.close() return datasets def interact_model( seed=None, generation_len = 5, do_lower_case = False, temperature=1, ): """ Interactively run the model :seed=None : Integer seed for random number generators, fix seed to reproduce results :generation_len: Length of generation :temperature=1 : Float value controlling randomness in boltzmann distribution. Lower temperature results in less random completions. As the temperature approaches zero, the model will become deterministic and repetitive. Higher temperature results in more random completions. : do_lower_case: whether to perform lower case. """ dev_files = "./dev.json" model_name = './models_gpt/cn_gpt' vocab_file = "./models_gpt/vocab.txt" tokenizer = tokenization.FullTokenizer( vocab_file=vocab_file, do_lower_case=do_lower_case) dataset = get_data(dev_files, tokenizer) print("process data done") gpt_config = GPTConfig(vocab_size=21128,temperature = temperature) with tf.Session(graph=tf.Graph()) as sess: input_ids = tf.placeholder(tf.int32, [1, None]) input_lens = tf.placeholder(tf.int32, []) target_lens = tf.placeholder(tf.int32, []) model = GenerativeModel(config=gpt_config, is_training=False, init_input_ids=input_ids, input_lens=input_lens, target_lens=target_lens) np.random.seed(seed) tf.set_random_seed(seed) output = model.get_predicted_tokens() saver = tf.train.Saver() saver.restore(sess, model_name) top_1 = 0 top_3 = 0 total = 0 for example in tqdm.tqdm(dataset): input_id = example['input_ids'] input_len = example['input_len'] target_len = example['target_len'] out = sess.run([output], feed_dict={ input_ids: [input_id], input_lens: input_len, target_lens: target_len })[0] gpt_output = [] for o in out: c_tmp = o generated_text = GenerativeModel.decodetext(c_tmp, tokenizer) generated_text = [t.replace("##", "") for t in generated_text] gpt_output.append(''.join(generated_text)) if example['correct_word'] == gpt_output[0]: top_1 += 1 if example['correct_word'] in gpt_output[:3]: top_3 += 1 total += 1 print("Top 1 accuracy is {}%".format(top_1 / total * 100)) print("Top 3 accuracy is {}%".format(top_3 / total * 100)) if __name__ == '__main__': fire.Fire(interact_model) chinese_wplc/nezha_evaluate/tokenization.py000066400000000000000000000277411411633436100216210ustar00rootroot00000000000000# coding=utf-8 # Copyright 2018 The Google AI Language Team Authors. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. """Tokenization classes.""" from __future__ import absolute_import from __future__ import division from __future__ import print_function import collections import re import unicodedata import six import tensorflow as tf def validate_case_matches_checkpoint(do_lower_case, init_checkpoint): """Checks whether the casing config is consistent with the checkpoint name.""" # The casing has to be passed in by the user and there is no explicit check # as to whether it matches the checkpoint. The casing information probably # should have been stored in the bert_config.json file, but it's not, so # we have to heuristically detect it to validate. if not init_checkpoint: return m = re.match("^.*?([A-Za-z0-9_-]+)/bert_model.ckpt", init_checkpoint) if m is None: return model_name = m.group(1) lower_models = [ "uncased_L-24_H-1024_A-16", "uncased_L-12_H-768_A-12", "multilingual_L-12_H-768_A-12", "chinese_L-12_H-768_A-12" ] cased_models = [ "cased_L-12_H-768_A-12", "cased_L-24_H-1024_A-16", "multi_cased_L-12_H-768_A-12" ] is_bad_config = False if model_name in lower_models and not do_lower_case: is_bad_config = True actual_flag = "False" case_name = "lowercased" opposite_flag = "True" if model_name in cased_models and do_lower_case: is_bad_config = True actual_flag = "True" case_name = "cased" opposite_flag = "False" if is_bad_config: raise ValueError( "You passed in `--do_lower_case=%s` with `--init_checkpoint=%s`. " "However, `%s` seems to be a %s model, so you " "should pass in `--do_lower_case=%s` so that the fine-tuning matches " "how the model was pre-training. If this error is wrong, please " "just comment out this check." % (actual_flag, init_checkpoint, model_name, case_name, opposite_flag)) def convert_to_unicode(text): """Converts `text` to Unicode (if it's not already), assuming utf-8 input.""" if six.PY3: if isinstance(text, str): return text elif isinstance(text, bytes): return text.decode("utf-8", "ignore") else: raise ValueError("Unsupported string type: %s" % (type(text))) elif six.PY2: if isinstance(text, str): return text.decode("utf-8", "ignore") elif isinstance(text, unicode): return text else: raise ValueError("Unsupported string type: %s" % (type(text))) else: raise ValueError("Not running on Python2 or Python 3?") def printable_text(text): """Returns text encoded in a way suitable for print or `tf.logging`.""" # These functions want `str` for both Python2 and Python3, but in one case # it's a Unicode string and in the other it's a byte string. if six.PY3: if isinstance(text, str): return text elif isinstance(text, bytes): return text.decode("utf-8", "ignore") else: raise ValueError("Unsupported string type: %s" % (type(text))) elif six.PY2: if isinstance(text, str): return text elif isinstance(text, unicode): return text.encode("utf-8") else: raise ValueError("Unsupported string type: %s" % (type(text))) else: raise ValueError("Not running on Python2 or Python 3?") def load_vocab(vocab_file): """Loads a vocabulary file into a dictionary.""" vocab = collections.OrderedDict() index = 0 with tf.gfile.GFile(vocab_file, "r") as reader: while True: token = convert_to_unicode(reader.readline()) if not token: break token = token.strip() vocab[token] = index index += 1 return vocab def convert_by_vocab(vocab, items): """Converts a sequence of [tokens|ids] using the vocab.""" output = [] for item in items: output.append(vocab[item]) return output def convert_tokens_to_ids(vocab, tokens): return convert_by_vocab(vocab, tokens) def convert_ids_to_tokens(inv_vocab, ids): return convert_by_vocab(inv_vocab, ids) def whitespace_tokenize(text): """Runs basic whitespace cleaning and splitting on a piece of text.""" text = text.strip() if not text: return [] tokens = text.split() return tokens class FullTokenizer(object): """Runs end-to-end tokenziation.""" def __init__(self, vocab_file, do_lower_case=True): self.vocab = load_vocab(vocab_file) self.inv_vocab = {v: k for k, v in self.vocab.items()} self.basic_tokenizer = BasicTokenizer(do_lower_case=do_lower_case) self.wordpiece_tokenizer = WordpieceTokenizer(vocab=self.vocab) def tokenize(self, text): split_tokens = [] for token in self.basic_tokenizer.tokenize(text): for sub_token in self.wordpiece_tokenizer.tokenize(token): split_tokens.append(sub_token) return split_tokens def convert_tokens_to_ids(self, tokens): return convert_by_vocab(self.vocab, tokens) def convert_ids_to_tokens(self, ids): return convert_by_vocab(self.inv_vocab, ids) class BasicTokenizer(object): """Runs basic tokenization (punctuation splitting, lower casing, etc.).""" def __init__(self, do_lower_case=True): """Constructs a BasicTokenizer. Args: do_lower_case: Whether to lower case the input. """ self.do_lower_case = do_lower_case def tokenize(self, text): """Tokenizes a piece of text.""" text = convert_to_unicode(text) text = self._clean_text(text) # This was added on November 1st, 2018 for the multilingual and Chinese # models. This is also applied to the English models now, but it doesn't # matter since the English models were not trained on any Chinese data # and generally don't have any Chinese data in them (there are Chinese # characters in the vocabulary because Wikipedia does have some Chinese # words in the English Wikipedia.). text = self._tokenize_chinese_chars(text) orig_tokens = whitespace_tokenize(text) split_tokens = [] for token in orig_tokens: if self.do_lower_case: token = token.lower() token = self._run_strip_accents(token) split_tokens.extend(self._run_split_on_punc(token)) output_tokens = whitespace_tokenize(" ".join(split_tokens)) return output_tokens def _run_strip_accents(self, text): """Strips accents from a piece of text.""" text = unicodedata.normalize("NFD", text) output = [] for char in text: cat = unicodedata.category(char) if cat == "Mn": continue output.append(char) return "".join(output) def _run_split_on_punc(self, text): """Splits punctuation on a piece of text.""" chars = list(text) i = 0 start_new_word = True output = [] while i < len(chars): char = chars[i] if _is_punctuation(char): output.append([char]) start_new_word = True else: if start_new_word: output.append([]) start_new_word = False output[-1].append(char) i += 1 return ["".join(x) for x in output] def _tokenize_chinese_chars(self, text): """Adds whitespace around any CJK character.""" output = [] for char in text: cp = ord(char) if self._is_chinese_char(cp): output.append(" ") output.append(char) output.append(" ") else: output.append(char) return "".join(output) def _is_chinese_char(self, cp): """Checks whether CP is the codepoint of a CJK character.""" # This defines a "chinese character" as anything in the CJK Unicode block: # https://en.wikipedia.org/wiki/CJK_Unified_Ideographs_(Unicode_block) # # Note that the CJK Unicode block is NOT all Japanese and Korean characters, # despite its name. The modern Korean Hangul alphabet is a different block, # as is Japanese Hiragana and Katakana. Those alphabets are used to write # space-separated words, so they are not treated specially and handled # like the all of the other languages. if ((cp >= 0x4E00 and cp <= 0x9FFF) or # (cp >= 0x3400 and cp <= 0x4DBF) or # (cp >= 0x20000 and cp <= 0x2A6DF) or # (cp >= 0x2A700 and cp <= 0x2B73F) or # (cp >= 0x2B740 and cp <= 0x2B81F) or # (cp >= 0x2B820 and cp <= 0x2CEAF) or (cp >= 0xF900 and cp <= 0xFAFF) or # (cp >= 0x2F800 and cp <= 0x2FA1F)): # return True return False def _clean_text(self, text): """Performs invalid character removal and whitespace cleanup on text.""" output = [] for char in text: cp = ord(char) if cp == 0 or cp == 0xfffd or _is_control(char): continue if _is_whitespace(char): output.append(" ") else: output.append(char) return "".join(output) class WordpieceTokenizer(object): """Runs WordPiece tokenziation.""" def __init__(self, vocab, unk_token="[UNK]", max_input_chars_per_word=200): self.vocab = vocab self.unk_token = unk_token self.max_input_chars_per_word = max_input_chars_per_word def tokenize(self, text): """Tokenizes a piece of text into its word pieces. This uses a greedy longest-match-first algorithm to perform tokenization using the given vocabulary. For example: input = "unaffable" output = ["un", "##aff", "##able"] Args: text: A single token or whitespace separated tokens. This should have already been passed through `BasicTokenizer. Returns: A list of wordpiece tokens. """ text = convert_to_unicode(text) output_tokens = [] for token in whitespace_tokenize(text): chars = list(token) if len(chars) > self.max_input_chars_per_word: output_tokens.append(self.unk_token) continue is_bad = False start = 0 sub_tokens = [] while start < len(chars): end = len(chars) cur_substr = None while start < end: substr = "".join(chars[start:end]) if start > 0: substr = "##" + substr if substr in self.vocab: cur_substr = substr break end -= 1 if cur_substr is None: is_bad = True break sub_tokens.append(cur_substr) start = end if is_bad: output_tokens.append(self.unk_token) else: output_tokens.extend(sub_tokens) return output_tokens def _is_whitespace(char): """Checks whether `chars` is a whitespace character.""" # \t, \n, and \r are technically contorl characters but we treat them # as whitespace since they are generally considered as such. if char == " " or char == "\t" or char == "\n" or char == "\r": return True cat = unicodedata.category(char) if cat == "Zs": return True return False def _is_control(char): """Checks whether `chars` is a control character.""" # These are technically control characters but we count them as whitespace # characters. if char == "\t" or char == "\n" or char == "\r": return False cat = unicodedata.category(char) if cat.startswith("C"): return True return False def _is_punctuation(char): """Checks whether `chars` is a punctuation character.""" cp = ord(char) # We treat all non-letter/number ASCII as punctuation. # Characters such as "^", "$", and "`" are not in the Unicode # Punctuation class but we treat them as punctuation anyways, for # consistency. if ((cp >= 33 and cp <= 47) or (cp >= 58 and cp <= 64) or (cp >= 91 and cp <= 96) or (cp >= 123 and cp <= 126)): return True cat = unicodedata.category(char) if cat.startswith("P"): return True return False chinese_wplc/数据说明文档.pdf000066400000000000000000011645011411633436100226250ustar00rootroot00000000000000%PDF-1.6 % 307 0 obj <> endobj 329 0 obj <>/Filter/FlateDecode/ID[<1B527C657927194EBBD3392A45B0B7BF>]/Index[307 38]/Info 306 0 R/Length 108/Prev 321512/Root 308 0 R/Size 345/Type/XRef/W[1 3 1]>>stream hbbd```b``z"gIEɲDr2eSA$s Xv9} Lfٙ`rXDL*Hx>stream hb``` &  ea0Ph OkCI%S``h````2aꑀ30bb<@a c7CZR$>7019`3-``sZ9~Kiftr1<  $ endstream endobj 308 0 obj <>/Metadata 8 0 R/PageLayout/OneColumn/Pages 305 0 R/StructTreeRoot 12 0 R/Type/Catalog>> endobj 309 0 obj <>>>/Rotate 0/StructParents 0/Tabs/S/Type/Page>> endobj 310 0 obj <>stream hVmO0+_"!i0-Fj* ҪIkgRe ms'wgI#RrhLD)VDiRmQb[g t8 # N4=jIx49Utm$owTa. B '}>/\:8]XEC,'c:Z=QGH-Q8S78m'u*u~]l֨gXcˤn4VZ I-TYRn5_z/Oj{4kdenEYqìT)C`O!(|k9 j5 `mPacOiB2Y~+7Ɣ59UK#1hRC ='0O1`΁WsՆho7RͮܲʹN.vse?e&ֈD ÍeG'"X"xfPJj5>stream H|n]z3L6$ ؼ$-d85P˂ %OAQCrm{?}~ÇǗ뷯<\/~~1߇/7^_?}zswwZǛۿ<^_>?zÏ=?z9.]rO?3Ŋ yx|p}zsw.ws{ws}zus/7wvfne'}YkpӲs|B;voPaY.ڸBM(yΏ#q.rnev¹9Qh͹#))'tBu:R t[*oiXb݆=+llkسaϷ N<6&g߲֟M4:g3F;EkNCD7ŋ5;B\01WWiWE>͎x9Ee0ۛ~[ooZ\܉̂MދR)*;9R! gǢ7%gk}̺)wvyE#Zo)1WrCʺ~[`Cv+{uXJgYo`K[Y[bede #J:jkK\..١XD+hu;yhG?jqQQ6n뗍C?~N Q"E1|νe{mK,SomAuQM"/ Nh.;40+O+W`D΋bEM%z,橥 %#TEk'WyNUHhP&fҨ,aDD띝ssDہ˺5H"#(f AcJfOȟ~!X x,(3O`:\L̼,ZfpCx"Z/;yYEJE0Q { {FQL,3Fg*g>rF 숲))ȞQNG 1qeEQHY[vEMATb$bEbw9w Fw $J(Cv%QTM2,ȉbeͲЀlrWxѺ˧ stH5\M{ž {̹s*JMBRuIYl&25IE%$:c3f)JUrVn,ydY"e鞳V#vnl+{u-QrPhmTȎyeP~r<'+:}BN!!%!Ey)H#άi3B*i>qI:*ETL~ιBd{V<"/aWWITrab.FRz)Dg+J(kGQ\"KTE լ~¤VԊz (B)yg>[4tEKʔv7:Ht& (WSv͠izt;>:8H穊u*:4Dvq*S]3) )8eyvRyM'bfS 4ҨM5õmFs<AT]OCwviMAh!rC9`KbD띉wf4*h^_M9mDA# DKfCfG#jAugSvm@~&oe"s S;mW\WN?+r*]Զxt9o3B :Bg@WA)ȝGh%2{Ϯ^y }n8k'kss S:+1Q0q`>OE)CR!)CV8yh`0 NNz#qjP(Z[lN*M8lDыJwnP!>qh}u^&{Qv0h; P( Q S~"d|LC TEыJhɓBFq'*?r25h)H'>&'(5?_Ͽ^~{y>[%)% QԴ̼ԢҢT.0ҫ endstream endobj 312 0 obj <>stream H\Sۊ@}+qaPf /Bfv2DŘVUSu+wG?e8ŝ]y}ylCdxlp,y8_iX =K;.?1n|e/Ǻ~[wVÁYN?@s.a}퇸5fMρ5<,4Iv 5/J(TYˢx ^@ZG@UBڤP]&M%dɚPK\DЬ#R3:+D!mC/BHA:'9r^ KȋUN;:4*i^hiYt$ANh2++B!$̔IO'2t5MꂸZWN-6%׭XiJҵ(-VsIyQ3TptP]̥%.ؓsOwBx•J=%R($&]@% 6 'H0 X;<|kTʰqkOǶZnVLsuY 00 endstream endobj 313 0 obj <>stream H\n@ཟb"20'H@"Jb,,xRH1p?Sn}׎>\Cé!ݮC1ۮ/BY_(6˾;]*?8æcQ~49<Cyx?钺1ztViӾOa1ωMU;b5˯uX׺H]yw.;Px6_9/ʬK% yc!2"1g*rn8g#a9" al4G#a atF8GhAhssssssssss] d2+AWʮ])Q)lJ¦)lJ¦)lJ¦)l{J¦(<~ =c=c~mfmfmCCFiСl0;;tt4;N4;N4;N4;N4;Nt:K]/yݙezx?R y{i_vs}ȫ. 0z0" endstream endobj 314 0 obj <>stream H\n0 <BX+q؇41Q~6:i 8v+M8A9pn \ڹht_-7>) ʵC8VOv:߂й+z q}{tl(bK4%lSYyC1GHN17C[gEO]Z݄(Oy |`> i"N9NSlOK82'šX]]əTp,j$g9,Lɵ9-w|-ujբYKM\{?s Z>stream HT[k0~ϯУ`W:Xp$ 6Z̐vtkF[~Α]3ҹ|߹j3..»[71EOI6~YIV (XZa}Y턌"a004Z&kB+"-X+Ə lXF4|auXjh1WUxhXP'Fuv?xXu=??'\`-+v]m #:5yq\ftE[2zF"ENAC l@[:Ks> n\W/\4m{L]r@]( ޑ/ ~`즚Mgܙu9OI,O\^87EZ*hyWsf@kX4 )@E;)c6 ӿ2JyR7NS|P:t( hޖ{PjGbToŗ=ҟ[W]8͢me:e` -lpRkP]=6NH7+~Mt QR:a==Biu:,|>YmrS:F'tS۔ E6IE8#; Q+<ΩI zLJD!%LƿP@trg:%.J(G)o\? endstream endobj 316 0 obj <>stream HTK@ W-$;F'PoJ=SqdK9gO1~\~fח؄ȖS)TjPM" ή#h{?i mSB&S+|E^cʁ}J![Gآ-w3qz'J{8V^d*ޡbK:Cm,ΦE1G'-lWQ"a`\I>O_g/NnxaC}ū,A. KxXX)3)Pr1kA1ڊ oBo^,<(YHƌ֞Sk$(. otarq9`{kztysv[/h8iEV"Ét7R﷢u&|w/n5Ķy'+;co 46=z@45YShfh 0I% endstream endobj 317 0 obj <>stream HTˊ@ +hUK&Ky,$03k/sTJ#E3*GYIH;=|}eyw޷ATUӆBX$KK/S(d M,@2S}J4he :_DtNEg̈tyV1n*ՙ%Y0vnۛ^*eňAi?#RHkȘ1B^ɳW{赝ys`hLnR&sH-'eX{`λ.-By3: "vjamɩӋ1j0h& V^]0!` )RV\Ba[VۺV(v-7oQ1G~,V@amWKJB hVΓ5>̀ޙwxw٧P x8]tR_JHRd7/ _drwXڮV8tfT˶`?̎hJ)vI¸~Os?_3M"?]$ 0>M endstream endobj 318 0 obj <>stream HTM@ϯ0&N (<7$RWKï/?M?g0`'kƯ6ڈp >Ě 5!UJӚ^l8\?uqqI|&@f)%#Xa 9,yH]~,f_P:˞ 4 7@NgV5ِf K Z{n"6ho޴m;7JNJ+`)(J3afX^u^u1RIqXj"3HlVt澾k.bYĭr+;oDhv-(*Z':C7-Flq" U#Ar` Z=hG/Bbbiq?8˦ +SrleL6LP5#=ts&VDJԉcz^^|ڑHn őް;lU렶z`[PL endstream endobj 319 0 obj <>stream HtSKk0q}#f$̂^PJ-ɲdwƶi{| p7R(1?Eԏ!E:썘||h2ߦUiCq|n!ϟq /_%N[[%Fj|ln4WdUS7YJ@B2",ꀎ\DC@ 'd2*y0Ȣ̤ 5MSn.jmT OZ:;knpс…T6v8\S3:'a. h&ӭٷ0tkH>5y+jjvz͔*iƠJ*8͖je h)皓 Ơ\U!v2u+z52\ΪfOSyҴJ.\f%ϣNhoF|2EbZ ?- UH YIڱ^m5a_PL0QdTq"OAP1 endstream endobj 320 0 obj <>stream H[k0Ẉ`E3H2lC B&% !Z:]k>YX渹zeyu>l::F@=nӡ\I4Bwx}'l ˤsT4)U^Z`;c Ez\;@_Ln|XS lDYU)<bu"Ok_,SKh$1 NѺUZ&;ў룿Yɑ%Y6R/G}Õ"`Zx͔ |x~z!Ij߳Sk6^%:\EIGͽ6Ϝ_GY8$QS3$nҭdž Z"~%io9.dqwlƓyO}o|@DI1FߴT[hka-mY;t=>stream HTˊ0W9X= X/a!CH![~?X`[VuU$'/Q?SK||#}BbVk∉bx>?=/%0MKg7)j -%+hXW;Di;N0a3ecZkpH8 endstream endobj 322 0 obj <>stream HTj@+(4yh53$ubBC?V%YQUս H DC(cUS(UMERz> $\~˷ߢiAx+=}bY}"^]V*V4:N۱QA޾[ >stream HV0 9h@[\A((=Q< !q #\CرN$ܝ?o~Bȃ?*䬼L|XD,ׂ0"EFi'bz.Ɏ"c j C%wg%^Ddmo%4djL栞^ KUqǣN˙|eN-3)bݐousS9䖸{26r)3n71􃡮qkP|PYk,㓅_cc>NM$7<<XeZҙ36ne\c9txtܰ ,?w 4_Oj='jYöMBXF \4\JmC {_Sw. endstream endobj 324 0 obj <>stream HWMG~󻻳fc-Q'@JZ;c.gtI׻qlDZՉHpnqv8)CJ $B"{=*D8nMUWzի2h(C_?:pтօsᇾ;"c7Q{ +'>6mtѮdnN.T2!> Ķ̯~}']wMO/>eۥ׶OQg_ :o_7"sulg/2;(&娘,h-M,F2}\ĝ82V#DOlec+)qEwae"~kd~ *-Bx|Z -93]ui:Kwj^aԤEu僾HZvUV*[e|f^;oU2?|Wu;}×Z[ߌ޿qףn71:Sw^MN6_oz_zq W\~ =wٕ3K9xBjO?y|xTǾ=91~P_*1Q9ܻ! 1T~T4RurW*ߒ媬]VC 3rrj(D`f>-)stSuMY E(jǔwЃk>VKtdEzE*UfQJ!AX/S\7ќD7uf)/&@*ofË*6U"/&wO-68NӁ/zPlRZ9XegF2DP/sg"' f^:;.{FapT WKs~NlpFQ܈ž/R²ӓmATvvֻ ,{=I>Wpuz6e'}^}I{bQ-\|U .VuY C]xKM tYA:eY;h_y*\晃-6cEi ?&p/r03@-v != '\a^l #;rWܑ} 2S.vh'&s 8jK`ve牣1•`@:ʼnt s5iŞDG{]|n.& *Ug7Quw?_D=O]KሓD1Xt+.+\1D8SdB$@6-B43j]ol?in^oPzU1%P{8k-n92wEBUӯlHlp`<8|&?pސ Nv+W*fV[p@o)aj:J:Kȁ TiF%YG4>Otrq*/Q&_A TcGl&|R:+g➵HgO5av ّv.uqF ŚV.i)w%t9V1@'^VI1ڸ# 2N* /0Bm+4 pn`⏰G:CxDz@g_$3U)j0_K}K/(E]1Mi3vh AR-eYx0 !%]tnu ˞1ކk\Wsbc{?y%@_ޥI3i<( 'ӚVlE9YL9<ƖrJI6!]ql%0Dx+PgCc{ 5}ߝO :M/3HVe-ڌ`}#tjnR,vIiQmTT_Ա&5;eE%\.#_2CĽaw93< X~ڄD§ܔB NL`^ xTa,mfnOR_w+~ @QG7q K}3GpTYSˡKV 㷤b\߄5H =+TF_1)5 ,󼬸XZRp+->iw80z'q zzTL9SBSziJcZ[*넒5=Im{Q`8Ib's7oVӮE nkƪ4١gw~5 ҧY/M(\ ?<-'t՘s-Agqj9aaz)|< #5 z/֧Aq;I][_怆ꊲ2+;䍶M~X%9[1}}1'4Q|{ =S&Z'v?' W [e8!@Ss>?4^T^ӋwW Qf$/e=::$'亸֘3Ƴ_j_]M]gqnc;_8q88ey ,BXXvFB[ e!!TkUbB괪k;U*։lGK7q"}~\ax—6WsR-ϛ\,d/y62mT+0)+3,%N'D&3I38<$+[Hd;w6#c:ȔzGi Ҡ c KwdD]) 툭]e26G᝘[hOvw'EsV )]#"O+H&IHv!> bݯ;9ے댝̩YpU~?r6B/%>di@KA[\BH ԏt#oW^_Ɋ`! f;$E:[yşa= 3p Xpℯ#Q'3@)r+YpGOKiѮ<@6?ViIK쨠##-A%qoSS _B+yf"Rsh.0H@h5 <+ID<#6ҧEiĵz]^%|!ѹ;k*i[SU`*Ejs)և;#gHr?M>kn;ux$: %(IefqR[0$ ol',@]3‰ Vv#Qy{UH0׻Ỉ)%奚(P[ V䚴eGv t =F"D98]i鉂.OHrJkXu;Zd 66'|D*LSCRv^1[g2ЅN畮;zp>ݭ<]I= I6Y#yx&/U2C~HݑԒfO dkMڲWXM쓴UM~KHѥE&Z4)-DjB)4ĦEfP+Jngٮٶ!v!x "}qv &XHH/GDȇs^[xylOMxn&^vF̤9FlbYf*,|7&2-O-TČEz,Yʲ\|9#¯mAj8@xsc0]tli-W_9:NgEwսqXvGR00F\")+n.~V4|k6^~zӓΞ]W B*|t0;QKKesJ}:9Ts*"w3u ?X㎣UJFꚹqӷiȉozӿ4DrX)#ᗼO`'?Il'J=O U@&R H)ZPYeK!B_Ayg!]E0p)/ɮ&;_6ɲy  i ` 54b!T*!Qa_Δ`lkvu?:ŋj3h=F/~) {Aq4zIE2%gz-`f{d}h8LvH%T}G;nq/f{~1/ͽwqxLE14bJб RՠWi!\E+mDK]`֐ys\@!K;&Pk,$`nn԰<sU*ٕ;j^먼6`\[Y ڇXm5.gpO):e I(RfJ^R2FO'8\KWе2wv={t{ʔBzQh䌅#An>$iQIR=pzI NT /:~23(]Ww*bs+&Q;nR v' b2j$2&^7]nйDA6"CKt\1Ǻ)혘wFft^n6]}|2tReէc w A L*ȯ h[ieѳZfxg ]n)Pit戮 =e%Uhg3/!s[&,!\u֒tVQ3T!SW2QZ~o&D k2,8eOcP,Ow`nMd]:yYTb3)twx3Swgk+nNګ#.,!0uZLSrzDԵay{}3Z+؛˹SHc[CexRi$.ψUdNUyyxj{k!?XWW|2`L0o( ؗh6jG?2LEXƱSl36=<>E x='e,S5cF<73!?vr{/ǡ |SVAWE_= uhQ7mˈUJI4]j%&L!U7m1 t GBc[/~CCK Jau)b9Ǎp~j8b{;Z#ng|>DJft'Ejmԟ>a1bjgaZh<(27r(zXK^F,~Ju<*N[Qn.l渖hh@Fol_Jo 2&(/^]. -EEC瞷Mޱ,E nd2xer{S`5 MgP͕INw{Ռ|7BhqFDž pm;,oޯ")Y*=6#/;74? 0 ϼlX RUy}>81gpט6allsBY!|6!jiҬ͖IEіSE٤JUmIQeKjdʢ 7{}>=yVZ|ҨDH&Iii^b{dPR ZFW#D|iyF̋v:{N}Rn/xEΈʰS[,`)GCDy#LÏumK|X* &+b~ (J 7jtFoHQdWe o VNpAͼ p>1,Ј3gl~ו(r-GÎ 8'DCPK;oR'U]eWb4Hh E@L}b=at6|1ra^$PDװRT#R'\+,O0vfvUg\Ld 4րV*$]HOG[rZ~#]a@m6G/u1eceCtld2)$q7!rSI:$մEy0nWͭ_$G2E4k=]\K<֦ychzے˟/? MB,oOaU/ lO1M;ⱖ1>;Bǯk59"n@g 36 &o&%*D4W0冸nH0  b짂~y,Bg&/p]~hb sQPeg[]k2DOf,b),vc[Cl:7XG\@?dcͱ_6ftm۝[l49VwYV0WͨY]6HMlp+32F Z^F_5}3g)xzYkx;gߟlm!c6 +^;(@ c-8(8t[ss jW✜hiO a>r铑ٵg\d~feܙ>nܼZGn>zfDhh;r49Nx9\,;ЅaOHS7l )&=*<3$=9_8fQ5cv76cT0BwqDC,C+ %)=dY+\ZzPvؗ'vuTp#[U_ӨL6*68ʙ)Z2OhRɩTi ]MR `d b%}5$33x^`/BU ƚMLTAqz QO1qȻ烡dzxh\Ix{o(Q` MBs5p]R&VY7hE]16D`wt9u'xr=UױWtl. GaZS8oVUh31W lH?FWs4hww>3h1[$) Rg&_VJKՒcEh*(!W;dPR=fq?3\(V+=z^1Za &w?Ʀẉa,EԠ9\]xe1a}$W QA|וO)(\ TuI5T$xyWL-FNƍeQcU(7ɐ\Bt:gPﴜG3^ {|n?82iRiQng1fͧoN@g딗lkC14#M}.f O-qZ9`Ts`Ųy( \P> $;sDH95Q^qKSM骪b>cn(n@U~HO]fLyK^j^kcW@[Su zSE S`HxZRAt<Q#eV:O%k;̂NpOElW}lS}!ן xcc;1k<<c@! iEEFXEBEQU%BuZBB[ZhM}~=;s4#_suIs§GKq)6N&_usi}$ghr̺򸪴&@P_/m\Tڼl-)[̔55K٨5srw,HzkGU%/ 5аF jWkS;R½PJ&e4sg$|ponka܉ȁd@x# Tџ;>,~bf%HL 2u7ShOӣ1{yrrZQZ,E"ư{tyn>"j,>4SՔj asiːB.GXJc…Bp(7!6FT-sĂO}\9u=P-bLKdIC[&< %|g5AsNh }x%n%qQ(&fHp; Ӗ˗+OS)Y17wmzz_( $.~{6ªR.akY]eF#t,jN,B- Cb,Ћs몼 (Ĺ|Ce$Yg!_dQMxoPF訒VwdQvyr/a{-Tx@E/0NPtl=p( |!ږUBD&^לl x6R*(L48?Vf&遘DD3US?wy3N2QquRӈG֥*i#Tm"r K vmYg\Ig%X*olrfk Ps&*_&fn;10@k'BD :|މH)Qo 6M2‰M@ 3OqӠD,1L\SLCHk.YiZV6'qon18gB r$B 31%BTZ2DV$C2-v|268.[Xre/%S@+哈YLVP @Y-k^.cH}9ݻU?HiqKTuEjjTKN YT \{po~z#T>Ƨ|BexM!D[DGoA1!D͙=t?); q#;x8<pgI+کj y} ޺#-fB؝J&֬jry3 (XBL&_BP*J~AV TBdC @/Wӄe#b9v5D-X_iܫ?}yMt`gl8%M$4XnyrIeQ\|]}<>`⇵ @H8;uJ Ԟq<1_8-pw,Y><-\^;qB봖t-\F(!85<$-_w6K7PWoqR=!Z$4%p (IKُA~nPhcrQfa98R=8uxk^]%u;|˼XpW<:'NNt(]fglɪGK˟H P4s&&jT܍[U$pЖO| 82ąIَ2uXg_ܾzK%G1!PE\D!CbbD>rQ!ek4ի\iHFzZ$QX\{8` _c_2 wd9 iPǶ"P"Q|G[cc%4>Hx6s7o~cvQ?BnME#M\T=0ViUU[yGrPFEn|s\w*XI9=xn(3rNǑHC\$:B}@J5D0tL ՕCOӉB+!>39=קr!Ckţ84$5Hrv6q{;9$8e28KHl %LH='5@,B!JY@a@`e-ІPQMݠCEU*T!ѭ[;Mq:%ѽƅ#9ZmVpbHJJ:39Ik-g؈A6BC ݈ӼVkeV́,{f+VEʃ>omqMhDON؉DY6ҢSZu.W ltJ*ZjUP#4rcHF;Մ]j#7}_NEmpbu;;['R }ەgĽoG@yL%g mqa` @s~eMMඪ>HKNd?PmK8zO0Y?b%:Fn}k,˽sh,S*;k)SRa$EI.[C4p$Jk,8RGkv+c#NzNlM+~lsE*ִގzp ZoxeW.ObZ0ެPֺI4IIQ 9߅9>]*1}7ٍ";W~'MԚMު;="0׆znc5hgc ˍC?M[Еx'j Q֫/1H5@ bܲTW(ĜFCusb1Ƹ{友Хf?}{"&J 6M0WDR\uj p;B+1+@'QH:{ۉ|`aܰagme,`?oT* _?'7C_o0Y+-Oǣ6rut8?R,_Sn@iYҝ&]TiKl5Rz]4: KQV hYv5|p58WXg#Ԏdohv2^fM Dٛ= +[<dzF#( 3Ά#<#ͥl[!.F] 3ZmsL,n4U7N4@7A/!P{MCaltV,jO X;D įZӻSޘ={쮱mlI3$YvQϵ=Z}ly|>[lq|q13sv$vplϤ n4,iE,kQD F_ ["jL*(0B(H鞳i؏,ݝ|}ؽXp+m ,P(b ]yXK,d}"3YLJX$}ϻ+3OV|k9Z(J_Ya:]~puC qIx[;lAfg$[)JK$t-aod[wPqS ~TOjP+V\<ߩR=Á0Y>30!\Å&մ"2 ٛ,JxGGXN7|nay#ߧjVՅ> u92WG(„$\fIIeU A=I׃+HM g[_fhUt&]Aںî9G"u3Cd2o6/6Նg"Jkų+x$BeZ)n GS<*$KйsAAdѿ5`zO`#'(Bvu"{M(n}v"m-WRq}>VVd)V~`V̀NKm-:^ tҚ0A2"ͲhR>|t {굤}N#LH0~~erGѹضLg;ɾdD|ZˠYK'^ʹNZ{7LC18%KP5]: P׹j#7g;u_^Apw G7%0=1{pl?:S9mDs ɬ kz1\\4̀ʣ^lv:{j# Fl;?51o&]0P4#7 }kؙZlph2*-B 2Qt_r}>Jj /_кi-VZV7LUP `W(Xd7=K"^7%^g=Fbp΀&ёZ:!<(W,L#CѨd/ =fޚבV}ekBRN&c^_CZFIc\ 2w]V-7kAF E,h2ٓZzwtqk-n~My=@ *W6fK*N5(*^guYܻzhp8"e_}O}:E_/;iiU+ $ЍHvS3+e8 8x0#fcqx7VUtZ\Ѩ^qv2'\ }W[)j;:&ۇ)Ml"|D;Iĉ@m,QYDc9 ֪ZÞ߯b ^C(P96fJW*`,Z(F|(7^F/w)C㵡:1v]² 4lڑ 7ټ.  \ݿ  mT*t=6:_Yhw$E' +)Kbr9l#Z=h:XW8V kp:Y{nDS/* /3=fU QW¸YB9sKJ҂ &Ɲ*`;дaGl(lzd`[&ᓀM'&CgeKz"˟e} }w>/]~ )RBQJ[T~|*'0s_e%|sw脘7&#(IAJr2h *䊣 p˺ˌ{k'esSSYkjpXa]oY77jEfX^NEJZ+&&E38_R(o=8pw{%_q )ؗD,+=KjOw`?L49Khl/{ 4 pX69 LXFCQҝYXeY ;V;!*^ܲC[[jJZ"a[fPZ5U?41.V,. ;]-MwsW8\[tW:̝8agׇLL_wPEODH.sDh'>}7߼2`{֭jj֯7^ay- AGeky[Ff\`G%g<Р97?, `-Qj,goUP^b4ɶ>EiPws\f9{PE^ҔFsL )Ea/9>82 3p/RJ P2[Z6D7ClɀnQ.[7c_@&݋/_qM)br N|;Dg6qg7 }c"mDy)ɉFvv`£:YZB٠iș9Q>'?R~Ɯ3͝%P D*KM_>(v pTZڭ賱Aۙh݄{BkP5rUQ-le3F$lQ‚kwP1J}QwNmk~퀴Y[ *CEAt'c$$)eu۪0PGjh:2*-R4:{F<~':F,(٬1R  IBhYp̥1 ec(!gT7=׏o`Ċb}`Q73#kM=LgdUxsZ9v "A7Swm߼VPfcr;gRȏFV\OM'hSη`kkCQ[kKeJX|$TU}[!v8w a¼:>$PWwc݊mw5d2 V,,`[20G<賡 4-$0||j+;|AǦۋǮF̪C{)6^VLoa"tt^|Qtp~}Pk ^Wid+LCBB$BGGj%'eFdЉ*&<ڕ:WJ\˰v)aNn!LS5=$Uy*Tb8tTMhLD/|o}tQaX0͋n^]'[,!dliP7~UU3Ehf/.]uj,:@y6tAS rUЊ؇N`ncqIIղe% (1\^}{XiuEX9/ja9]QB5xZA] h,;2z^0\v7܏J g߂` ?؛l+]akaAs'$;hQH0j0wWj8EFkUѪb.o qqpb$p5H X0)lkEqtxiXK1 (pn7[^Pҳ7qNWL Y Q_ `E*Th5k3muA:B џ*A=Q@( a!}l\YJJ$rɼ>?\5`Zt̙sDZo/c(X\<~:Adz'#d dD42+W@)u!OkBn\a<~x\t("Va| (* v PH°UQAVPx?D؈2:둆jvS'ŋşn X'J$`EwrOڤWH|NubSheaN}@)@PtOXQg)JRK(ʨWU2s`j@'ڦ[9=˛Dב:RHǚ}iq%DzDv677!Qm:{}}# w,7N8Nℋ1<|KCe(Bi|V? Uel0V !4V"d\'$ZAsyyrL y`5OoT&x"^*SCG~{SNTpXmZ#Bn bU68GuᜄjZyrbԚ쥥eNg 2lij1GT_B#p5:?WV O53Ck3}}J߳I6±pйVJqC~z85ºAqKpT:w& Dd @8Q#C\!Tl*?`\<%iv}E;:/M|Mid\qY%$<#JuHL#=[L#{gY~\i"HW^fĺnK\qRDJāg4f/ZpG0w4eee&pla[&&0yɇ/vc_v|۫ȍJw&| {F{@ ԓ^RF_w*+PaW:/:͝!ND1«Чl -m/U1Mn^l0TUL&;e_I%)Z2H 塗T~'~,O &0M]4{$8vm,-~XÍfx( J.1=p2Ϧ{#ÃѽlK)\1 {]!P;q56h:\2TgjwDARq&;(콯UܣDjRvr=*KI譳ۅ ^a%3֭%JWBp1 e5 cc ŞBm4zlGI(Rׂ}@ ݢoTA8H{R^`xNdzҗ̚m[aG4?Dps*mXS"x:X#CDC`5f [b՘[D_ki.,[A[mPau9u)c+606dR;ɲˤ(4T  ΠR$"p4Өur}wpP{} e'¤Q=ޖgg7_\<14@Z,,153ѝ6u%=|jf?j"S*S CQJJZ_(S?^vP5A=T%T#VxdҝKkMόoPj8>iD(\Լh֌s>rWlE CInO`hhkـCxt DAE~yH!q[C޺bcUZc?44-$cz:Y?򭍴ۄ>hVW{1p_Я@3> U+QM/bv?ݛZk,dk)ZYE+eY5\fr(3X+Ē#A_i”++?[[c9!И %M|ʔe]\_tެz͢X3:Kl݊CUV79Cʬ xE@3/YWSgYSTsǕg0C3 AP^%jvے."[EQm2)a1‹.u8:X&P0չ,Tp$C' Ar=s\-1f2/au5}$6/@ 1@"_!GqbFR8J=Jxz\Ѫֹ8XuϹ^oZ0uө^g$}M*.=cERgl%%{0+`ӸGZ@>"'k̹7<|yBg֞ڏE$'unqUP|cTmڶM9}CF:_r-.]քX%~Ѡ-]:_cFI9v\B溎& kLg.<5@\t#Cǭ -Kޠ73_(n^%[ tXɮ񖖱24P6=M;GVC4;8[3Lh^2(6Pj#š6+7̱۠i[m6eҺuJ,+ T-+d0rA(Lty,ceDfT,\QԜ`=id$L鵌&7˘]~Ja\2^Kw}R5懀j?Ag>? ;a5M =HڇĆ$%RWwE?B?RT|}`]1h#vX:y<<V gb?މFRaX x2-٥ +ƮNЖ\ESiK`3V: F[H)Z/OiHFD-5^⌏ɆT%_q0:}hF邷ս =ZeVy +_!d/ ʅ#>qI(Bs'g`6azOo*>׮3׿y^ֻ`.А=ApaGoOsfqbD.l^%HX81PB6E{ :7/өK6{ȅَ*>$)=A~L[Al"/4XDUb]!ՑjG1}86 @,Pqێ1H`/r`EQ`>_։h W%^$(a*I,c1rB/dkmV] @o]1l޳g3؋'-,2ᓏT ,+MZ^B0Ȱ | ?UBTz{_=b5xoA9zu,=$R@8#, ϪR }/ngف0Og2GSw;& oc? @:f)*B͍c!%s] Ԇъ;ˀA6rNߙe v!cI0x/?ey{;l_mDz(8kl/8vm˲BHa|,(˲,M#Z(tREQt:BPVZڭP=;_lR蒳?h p<465[K4ˆxe A3{vK+8JR$yA1.6`~DQ@z5rsLAAQtYԨϻl%m6>EE p+īZZ7=7$X7pᯚ2ב#W ّ|tHq`> [;,w>ͷdX}i >4,_My^ S,>og϶StjVXj5dʷ/Vb1B2K -Hd E$mc0] I ]CeX+&cO O}tyDm}k^wlḧ}u<#8 n!3C l'ȄH*q){ ׃~ΐ_6L""9 c4G*Begj,!XJOx}8t'98zf5_)HUϦ ,2 D ;mGǸƯ_"` 8=ȆGǻ|y庒~1rS&F,c\BJ*~o]@.!pxzޢ#,Dp[`J0rAkEdZ7 wZ _q*O@Дm㔝rh V}^nm8g)U@>;K(j-)D+6nd"s,[>^ʲΑ#9{-~w 84GeA&.+nl'}{P&kB"MV8> h'ĭ46hWe4+zE|_ե4="zSAMɾ5\x~+›7Z\JĞe^!W7lCV?ucj`H|_ii?`AZرikm-PK)^izM)VR~<⁻*?c1x3ܜ~]cܟd7{xw;/A:wL zv&O3><e.j08a1G3J"^ L2?}Mj]qdGCVwE‡*:طf9::9ĊGYAd}ވtSAF<ՠyx`}V0HkpBq57%QS=(5p+Li1N{1BfO#V:2n?O#$ 9)\{~}W4oHcQ"_1#cQI_ ^"]ݼCi|梖YXիQe32VeS8œ7zƛ-@vL{;8j #펏 mJe"/5'rHׂi wt"Di YDhVUmEx֊hŦs{%| W {Sx%m沮Pmftytv${1mgUN85f iJmx2fjJ`gy6R^9I<ֿ[D?{mjg& ' zڥ e+ ?U;![mC DI0r'q4x"?_al};ofAJAv\[hz!+$c|y Aߨi5f rYh 9LX&uM 1AµTC П.675$=v338}-ˇ-|Y;ٯ3iB_'@<}gB޵aZ;8fr 'E"IKKKf=y FZdmvŋ8}ȗ? _ ێ|Qۮ)T(oVe͟4P#7;TC[R`Dء$=̺v%plV*UxASu+/JZ +(?$@YvuHN'i"8B#r@( v3X Qw$AI % AD`'C?;eY'~ts1f1;Dh$6O&غd3 +۠ G1w=h@PkJ8.r #6 6фɱr̆u" `-v%g7,j;K\\oh /y 5kLT.5i5i PՙԜxrwpROaHЏ"Cψp}S A]Z  ȵ;ŹBLO..ѣ'VP`ʲghJ;WC.g34(,!iq ^#܇=Jgup޽sX8U(Z)8eO,j`ٹ+vl1!8z&aNܛ974(Pe,K0Cl*4mhbۄ4ѭB4 1hu!b?ΎCXٲ߻g?ݻ#A *[TzBPh; >V ]~϶+hyJOBEqw(6$XR3M'xEG TFt QKt[DPуT_, Ŏp0HEPP/LG `[̊tb='&+q$rf3 BJAQ;5/ %UŻ2, |C٥C>,A:ț8eBF`׏~uNy Ӛ"E'H}+GITgl S;w w&z&ܭޓ.@?+ݡ0 [,CK-Er!l7\v:"š;8d07jahl<½/' X; &ͻZW-P⌬*A!ˆT%p ^"d k W'W[4 aդL?UCQR49jR45t\,q ŒEբI7*57\:PihzE. ȥlL%"= KMrQR 1F\}}> CpUʑ~D b8F ]޹YQٔ;(ټoi ;܍^Cm7O&vnh m?=G&kv4bz?ўt+MK {S`YǙS,`j1f(| ^I'GHb605&݃V j*[r6BVp4@~;6334OufhuL/_!MY.^lϳZ[UM`f+6)$=bMzNWdK ͤ[OdQNjYC^k9ȐJi{U(VHWi 'K<uDkH3t}7ahkE:\F [yQe<0"2S6XvҎ$B)t tces}o+90\s98=޷/<7Mڎ؄vɨRgpBt*\SY%,%/D7l%x 4_pZdZFGǵ Lc>a4It43n2B`y@1݀՚ j04nB}tL2@-Zap ݽF=7ؓ:#\ 3_3ɢǾPN2#䪒MIT_0zB|CG҅Ѱy~{ρRʭ%ƇSF?wq5Y*.?\1?LN==zWp nǯl!dX܇nիX rH+BF2V..Vn7ʌ(G89SSBWm=?Xtϖ1 s/&B]\a0%%[$-p{FytU{q(yBYf$@F؛U^%0`D#{0ʅu$uub7y͓`1 EkъAtVx|FbQ2u2{;C!Hd,SV)XP)FSKÚڎYf˲kE3d(bJ-< 0ͅhN}ڛgc-3JGxkKݣbS(Іڌ㶀ϗ\ ua.Cu5ݐ@A~}GK[+OD!Ca8K(t뮰0i?6iNwL|N.C\!N|*W uY$*R-ۅD(JP"iac8l'L$qB:DZcHpLc`6d"F(-шE(BSGҍ4ES%C?kQįq -y~%,Uue^.KDp> k-gq8=LQÎfH6w7F#n^x7 oLlsUJEaj:ƣϤcX+hN)7YW[^^V/d0yVk&`  1XAtlW4+Q_/ʌ.̬67)4apP[:;HS lÌQYVM[B ]hߠ˙0f1.FoK·1&]2pmsy?e;5ܦkC\Y,Aa4xu9T\X:IE$u}է1qVc܏a_ŭ)b! -tǎ:oF\8]MK#JZgA\nZ .x>^mAЩéж쇠ev@-eh\[va1{Eo+YCV[@Cw}RtB2áoy -WCcvhRekiLT(;:cxpA`'DO߆ SLPi{8gJΙ3XKJ"0K*!7$:^d]p@#7z`km\܇P3/9^9`nV{SXOVB%kzgt~jhuI,,@o[葵7Bomxx۵P5Z87xWLqTe1)Dog@ZYֈ? g[hUa! Бabbl>9λc-m9o:L3ښϧ.$i5W{FX\B}( F3{$jaf3N03M&\qsS#TЏ}f47od^ FPP`NYgvSUZJפ+FL,CUG)ȼfFoS/KKGn1Q8\NFQ X#V~3d?Q@^ U_JB6SHe }|0xh#۪e-t3O6>f\/A ,'+O ;t (|nOsRq8cG۰SP>B3T I$ P{[$>h|TJPS(_eګA:uulLMc^qweABc~SgZ*VL4LcRWvE g!Lf]'(*.ӕ\y-~iY;5I\Dd\n@q=sn3^K? PK4V@wm]aGŋk;'vOd'3C6p3R밡bX:j~ܱ5_+Djӊ'gۇ]`1͇OXwmsOԹ7[dJ֕ׄ͡1lpn<>#oE&L2Oo( {ZLp;nY ݱh/T% .7fR QsXVrdoz54pb7Y-[Zn8::D? I!/=@-vfJ. ^x-:P]6Yp%yF{Auxsh$P %eg [::pg'$Z krbHZxYp0v2ϥ;@>AR#|8ڈ_r;T<\Fӆ8f(̏D$Ḍ876G>|<{kš1s@h0[5?R0idgL4Ga)2 juE @,Zrd᪳ˣ`+*(uc5E0~g0=GSTԼ-sھ.}0 J0N ipgN8%'qg%Yƻ"(%IRL1_j~3cSDbhX~q&s akl"`\KEڗ͖-Omu\P)*0`{OrV:֋ b Jعe.774Wؗ.olmtv*}7F*M ASbD2ʪ@=nKZtLAz jC+hXO+ lNB$xGyx:n9HTvvV쾞Dpr_gSJ˺z6!JJȟ{S]}ܹ"99"(&n{v+[Zp [+H=ۈ(z;LFk)":jwU (?@p:86$B8엣 c@wM}*L R?4D! X\q?B적#lq#ۓΨ)W\f+f &'h+ VҝN->0M|u`e.aP("ꏒWثGA^dҬ F*fs!5C( K !n n{VB[ c=&ig1Xtʚ“]T)D5Y|o:8?'otEno=1?en8 <BAZ]+}맨վr9x `p=MPT4q~7yJʠ<Zcs?^? y,_=,sz{[R)+XjL)-XE n&nf[nuͷKW 'ȷKt:Z*ݼ*Pn9z 5dY1*v0Oi \RL;vf ,,'R;g5 >+B-.]':& {C/YORtI.P<,C/)K> ?̞>8_}8 ({alUGX&G^b9Vi`4 :*փ{%yLvs@632(>=mR ڌ\;W<4B6=(Tgu/,, F\8D#ރHZCf$ZcD/;֜Q v{0qZձhK!4Zkxj7Lf "&>j[ZSc5 ڼ_yxz^7c6\EWlcN*.1-8UQ[58:RPfow7A-bÔbTc /r߹,nf S"8<\xg_]S}ߴ\ SLO}xxy3˽YN}*&jtlmZjub,*L M =1#cٮS@SGQ+U`fMd""L1;= $噒VӊY=zH鵝WqeojK( I׏xۿ<5;9!fmG"&]j)+:z3J) !vF/n6 ;:`JVƕE,anA;D,, eY0"@֛O2bmWjܖrqKQ):1-{p VWYFX5қԖ֠c@gpAʩ Pv 4g=;ۏBm+/=Nw+n5ä[;5CK %st 8\S@ ʤo)}X]QS 2mԀ_$pFSBK(VDz"jf@Γ (0&'h-@H&bNq7]Mbo=z3 ̡./+*+?8QX1#0BLNx:oɷ߇`‰CpĿR9n!EӉ@-V$<H(@pd /`vF uhŒUeP4' y  ]g-#?"m.e)pΪcE&mV+Ecv7|7@DQ y<;2bG&ئOj +ԁ[n14')6iZpklb| yyEG0I8[Iq!!t^* Iƞcc=Wd)4{(.>1`n F^72[I 0)w֜k(̕(BcCP0whw$ᅆw[;׋kև_z`d? u{>9qs3cs?N q<c ! k*BʟʀeV1Q`ZXŦmĶJ[P;ia|ݝ>O1aVO´)Oc vc{p}Ӱ`7.lz_HBzX2='09#SPhU"?JsٱT*iW LlQ$hR Ջ% 1ET>% GjLp߱s;S{O?M}*}^ >CkZ ><^DC5Beh,Ur؄AAT8:[ሼ]NRl+]D3Ypy2ᇁk[s (,hB)E̯5PeJ!!t{XE[\MC[D|̣0ފ~2 lnZwzM h8>m6''EwMc`"lDE:4 ŖuʇQ EuU\AM\ΏO; S:66,D6I(p(eGbX A4]K/aEeW -reeA|N3 J=/f#dx Erqn[$ؓa[n95 xV&)E~a^SPeۻ 3̀ ޻׉}D)9t[she{d zt ˱h/=fkѩj:XVXnW$fn] ^ֈZM- <͍nQ4EBT 2aKeeX,Y d.Kx0ܔ#M2Vv׵Ii-D3oRэgHTD4UGo 6gZ-]+[7gZ/ٽk^G8~⯪GiH'^7/TS.<8 )930&S2R#t%dܕ;j)YaK;VkON"/S8 IJfٿAUy/AhP.W{rŹc``[BQyTGD] ӣ۟?4b֭+*הt+bY{yZ^1[cX "q(r"Pg#S]R.$mg}!x6 D@_C"4c#p|0{79cb:aHcjh"T%Nlp[. Pk9fxby0ܼRs ΧZb/RvL5dŎ/ɔkkע?CzzH"͢3ȃ~f 3 =FnV)&LWŽtIjn-{\)II m^OZ\:#;2]* ":Fl!pgf]ʼnz"ьʍ\3wsӱuc>늽ulGoyAhJ,ɅG۬z:C>: zXV)ygU.=J'=b!0 6CW[M4td}_(\@ A0oSEPeq%jNڭ§V+̰x)rr];cQH (GgH6~M @p&0Of1_a*!%)47lË#eǶtC MIb;l=3-݄h$n>쨪_ <掐S<<`x< z2{8`[ 56ڴI/s;=Z?=/NY*z;9kIhd<䃯1mvD>d^UA(/3C@]E[ܷkήVڅJ:)vNf98xșAr)w;G=MY׭\0p.;KsJ(*lgJѻ)L?{S:s4G`sL+Pc*3M< =͹|97ؒK6*p '^wܔ@N8|C!zHʅ_%t !x[tXQzh].ręhi8.0r:=k?`04MlxӕU2s$]/KojOf{џvvM`K r=k{ /ʺ#%Jsx;rPsk(cV4y7jKEKCV}!*ӣ?/%,K۳+dtCG%s23~T/A}͐tcf0v6]6x@0;حUp=Y9t)+ ]A?; }ՏČ1J1mbFI#dMÑ;Z7yV=2+b1S+Mpvc1R x}7N5(@YJeP`5n^/ Vr4pAi(.V8Z+/2`?Dl[7AYyDH@TOR.^(ow'lצ=mT6"N*eŝ187h/H(T\F e*IS$ATE2܁IDc$iP<+4` F t Fm` ҪϪ9Qarԟ<At tG!:i7T򹀞ABprw.3j 潷rRU|F1t02ZSXܘsW}.MyC0 Q|bg*해G[$G}}tDtc[k< ޚ8E~ڗAuzRm]AՂ'Ȍ($xB+fї;X!B3~: ?"]t H;=^ˣ?b-5*ʚuRk((8F4͠q3,9]<.̻1y4xvׂǼk{_a %=(|dxw6_9*-y켊=VSƾkm-@ex`gr<}Qn"0CU7Y 􉹦ҴGPNW"SiuGc9w+(hڟ of0o_S>gqTN/zŇ,fMA鄷05p[4+uN{aIQS=Wu|Zڿq,4=zئYw ig8N:{0QMNJ'_ve:֯]ۼuk19z!JP'c|AlT#eٞXKQ-AF &l!W0EϝC' F>a2xadbt{/Tii͙yWdMy 0 \^w۷v>?WQD[ 6PZo AfPq< 7Î0)|>VWgUIdJĂ4q˚١,Ѡ;ٺj!X p9R:5UH*b d5U~C#OT@8~`Mʰ 0CbهOA-h,]!0iB dp81: v@tů¾4D.ҋJL2eE$bZa^$83&ړt=dп&^5冥8 &#*fmevBb00I[7ߜ2` ;U4Ӯ }`t_}Ս<7F\2; Jtw;|BOVb)U0VF]NSJJtuuKMR,NM-6iV.ꇁq -Y&II+DQT$ыKR8f%4.N(: VAjQb/ OMF}_|wxP4AȾGFhCE9/b)R`( #QcTFX{,qX4g]mm.xٹjwL{CzhAP%7$z]D-N+'nj6 9XƃoD-–X]^M23KU*BGIJ.0'yו<| GJeFR&RlR/TcqolT7:6PÂz6:n' Wн,O mf`TMM*4׹ۭ ޮTe*L𡨟`vpJQ",uvΖ~w G$˱LB}r竛G:3*%]Bx+F޻_}㝥֖ & /ꉔ4ZJijʮ[RFX[xm.W0˳e?gy2&İRcNf$]ΉXr1V&͂Dc^ 9劤F.{xaC@&mA5;%_ިj SL#+k6в J.zE+-]stkuK_>>ϣnjr-&Y]ގM&.jgMe򩌌_k`4MYSaguSO 2S,$ RTsa VO0g,Uw9v!<5_{#*'[irqIK *T8.BwxSta%<B[hdֱ_gkׯ`+D 64f3Wayfzob:r,z"J.D']e@g8}oy`S ( ܗFgZBP.˒O)6 P,?/rF *p {miPط:>c@r Ez.63lCGclc.0RY(,)*(CZdf,(,KNTAQo(,>??Þ@xMUހT^K/ys D1yM/5yKWȐ SZ'\zͪP@nKcͳ^)@XQ`JlZB&A0AcT[U^ 9ޮ:g zƺF RSM,kv~8jrI=/p>`ll*%.rA(PqfC9)?N+bh Sfo-,Z^_٤6,mVxmnǏWϱ"[HҢQz"!*z|/0fH{,Ur>UdȒdK[M5YfqHZD. n_E\D]8['À%O%)=`-u!fH)ƤIiD'e%t-7>(]ޞ=K {L/tU9WpvEۢD ~Nɤ5-zQa-|>Wª%S1SJu:gJܓ97iqVNdm4j]Վ_&%wVc&n/-[f'?2-=+J xDeW\޽7e-@r,d;2~mxvh6ui/kD}|NREGMr|=&:%Xsmи[3H9JK $r&' yes8!|cy^Nd9D a'BnC OДCe Zt[q[6ht}k`.тz4L1-] (3G3e3E Asu#ݲtQP墝dt)JĹRA]4ŴV"ޏ}[brb-Cxk7~/J;6"Ww^|V-wC"muw" mׯ= X}/ʏHQ0hl(< f}䵤 4UEҩ#b63((ӾNn3l=r[O2'EOJ$Mq3%r#aOvtxxmnx/"D%QCX9jy8N$98euf݀S:ԤMS;<&^l{/p#$ҋPSĪ$JwvTosv S{_ ?MO.M0q̲7W3L&=k{Ǐ'#ݴqEy "K5i7d;}xg\W>  dujb*zɷz\o\ky UL{+C0.0ti߹@w*,#Fn4H 7DZH P ٹ?0Z\~{?յF`uck[~Ou%P˻|v9.fA-A,GE ޺ŵh7: ޖt0?O=.2Kw?] k?:K]/CM 2ٲh;g6#9JlKR"i/I,T#|zaݠ'D#ePԴ`2'@A%XQw;6\Yvy"H0P +ZfJ1Ig}Т[X"\gUAVnz-3 6Bj$0V CAOIi" Qn'30s*cUz|ndl܄)ϙ\=j jʜÉco:FL 1vviks9 &ZXqB xPU٭24C lyOϮk,Ri<=,MP6+C -.gtVȈնIs6gV@A%6 #up 5$=SƆCϞ(@7nw4yqtm-Xjab m) 4H2*BSԝ<Œ!Y1Ż,gv9D#˓t`x@0hzW Ule>|+< &p% 1S'Dg٘*a1p8>V78i dTV~=›=?יr8]1,tzcЍ$]Hࡏ(Qi2(9vg l*z93ܤؙVjFeVS^꒦&WV4 |¬/X:eNJP H-a=X;Eʞ5q@pl ChstfTP };ck_Xa|g8huS-}?43*C!{@`oHK{6#̐ȷ{䠐q}>A5=jGWTXLAP%Rgؖ\3g|)MJ|j"R΄TQR:H$cU vq|ddZJba!@~_l8ضw<:(>p5Ue0 ޮ {avR a6mkmÎр<l EBVaB' lFj@gQo:oG?UePJWfVnΔ x4 z{ToCӋ$A'k| c!6 bf_bD(2d h^#+i ̺rܼpbibYGhg74<  ˔M b"0ph~oC75EwCC<^ m+$MyښgaD 1ʤI1,J b%T^EXx K F*$OedLe xg(@ԛ @z_u|%<8[U2;9QS9?8?J{[ %{$p)5.4f (f20!ds&멶f;ԧqg mddb^_iw:JVp,wXEYĚ)LS6iKmrU,-V ^꾿e}6_ =(:wq ܍ i_5XQT(zDp3G7űmOfD/-=5~מ\) CySaܴP=,ٵԈqZ Ȃ?wb%(j`,͠K壅Y=}wGk`C\)+UVM0+ s2"QI?IK(l-$&ZjTMMI]=S_4ٴ}鬴􏶏ni~x$rDwgYMa{ǺZkgI-m擇Z𕡡-N1e`i/bL7Hrm<s磇Y0'cqc,ICB(C)]&("T!@biQF)bEBC$TĶnC#~{g'u4;eG<_986u&F%z cS&`}Eft&sASo_yfcBNϚ5p|1@ * ʩ3VUx ˳?x"Cջ F@fgćWΫEnOsYYwq>&iԀzX3挽x@rl͜EBihzz{BXZtIzoJ^s:mu~2 9ϕf'*&03}{g`#owGZQbaX|\_o7 6HM"uhI N$aĀz|T\_O'<,{A9e\A7T@D,dJuw ?:bI&-\s25#+кb1ڟ?̉O}C} RRF8L-Ts4@L"}?'imfWXtp$e9!xIOJl{&B3.Ũ0&xh:"@xSl wHK\ūD=T肣 Zͩ@:ZP6^S!LA~tL#>Մ%Rmv)ISSS2nmMIel(QriqE7Mo,T #CoZo];7޷z@ُ0I=s#2s_ұ@Z ΃>-xR~ue-[P9[ZhL$#woOt;tBq)8^kT.үt-|EX 0`%L$)tIG8[o?aio,-9TU`?30}M `+ꚒB3s^8aq:680ſ1wMv+ui ԔB舡/v=OlVP)v] (4FbW5k wƿ=]=|f E"jijQg]BK& Z.=:R!mT):`ƫ<>t1F/~pN?a BGI&l mIURlin]32oh͚tXxl[AzzJޱ5fmvbu9*\> FVZ1 Gggg>t9=tk5? C>?+2NY.eW1m r-EeN_o+X7"Uo!U@Af"ް}<kT6 G~!4^ejtCKI^뜄#ɻ@|*y'A-p#jL'qLw/$ $-8y1Nɵ|hhÆ@8hqRB^2aex̎9ts VGxo:=1wnC:"hFsw /]Gglr JJF Yje+ Y9vZ63I^5AaȞ_!ר~?;]î;LN0&') rrX8AAHMBݭȤc#yG0 0oP{6xj2cr'Vz3ѹ?Y`JL9=Sj98X_pF hH(>XFX (~_-baOs\!n sikݝ%k'8E ]$}BC ^?!E,@> H9Q\,]r|=Z ;x&XfWj/iWTjdnDE"I9*8FaLd4A % llys9y'E9.{:/oߖ(凜bWֶBRB])""!08qD=,9ujy]ވi[q]c6u;eznˌQx}[$}'o{]]|Z^<2Wt^d/%Wń v)i N0áP0_09},[[=)BVo68-4:蜕5.P)@W"ةffJ=pC7arh G#$ɒJQ`3rO)1'0=b%Bsnhj-*X} 8:8*ԉu Ф7y*@$2}dtj,!y<5FJ)s>.xV6Ƕp߽T |-)u-[YkÚj֠5q{8 XG, V$yB鈌DzdW03> A*X{S)8E? |h^I'pG`,:z(W\x,'QnkzQ;مe²%e/7_Z4+6}M|Iq_ T pQ֕<sPzTv)HP#YTeǿ'}nZϿ1jjtr-z^~ rbCtmP¾ q+NXާ|S4a7ڸѩ9(ppjw,X$8xHL:wokA ߤb:@P\1ϷlYC L2˝{w%*~$^&1zvW )$X[,hӇ^CRe6e,HZ8kwS5_0>o6C 07(rds_Shzx#G v7'I08u6D.dLG,ID)gbKJOa~il{itS3AFIѥR9(CU6y,׿ާ $W}2nMr weJ}A[̅Ϧ9SD/V-Rv.Yl+޼&yڙhnzu O2Ts BKK)MaK^i7l{Jp[e[?ct8l! Ab fL0=@G"(fjknOV>vɅp^J&O>tP;JpR PhZm/@ 7<7ҁ**1)L,5ؕBmq- TtbVy7qlM:?΢ M_6Wn \90qXĚscA68vtz#KBK#]*!8I"s)cDiVel+ S}H(5АځϦz9xؕlPۼJ2zVun$t5fպ3ùF1= Jk3_|} s(pl3*AfJ`rQT E7lYhϛȓS8k,qLo۶a,WY[hkWݝ߂bi/P׶Hur]-M]gkǯƱq\c\7c'.&$x88N 4CƳ4Ye0EB S֊!& m6u$>1h6]{l_8XSUHtC ѓ7qJq9_V% Rtn\.)EYgǖp8nMwDoD@ow9jc$/s\Ab62e2aiBA/=`Ă R18&jg#Dy cl۱ih\_Md?0Wk+&Sxmѹi'T0[zMeW,Ugl.=O͚u\z<\zfif8ki3ȌA鰗N|3:sq7v缫fqyh^$=Ym;7j&Ւ^ uk۾9`HZ]5/.I$b[Wn3 #3*(^*,ԓP]$(@pjh,iQ֗;_Vb-O@{|ԅVKTc`QP;_iTS*BQN̍s* xdd_R2"~@e9nj>( nEW8;V[oKFi-)yetna4^Yȑ9seEgKt-DeL1Ex@ʮ2IY-'qKEÌo[p}=sZXNg!1^YDB/πYFO|]2]EBa@Y@Q{f`&, Ⲇv'aRm8]$ɣv/^-OؗnVόP&eaWHN5lGUk+ 0wTM\z4 fy@Qq?2H !xD_BF@,(ޤ>B5S (-,`H+F {QMhydu+1o"1Cm#}NZϊR3[,KX54%h˂3lV8yfh5LtzJ1e'~=frP튕?>'nB-%aPri8Jz`dzkN+P0JX}>+Gr/L,F#d!-dES]=qK<'tw3z^Ѡ(",ۗ߷ Dp;>5U:ɎN*hӋ浝8oW}u{2FqpqϯJwuhH[#-~_"*lɰӝǘ;KQ2 Ol%̗氞^!WRBI2Ya*zG&(MRʘ8| 2 yn_sN ,6q }>'$ՙ&`n%.$lS,>*NN=P݂.9*=q/= (pB3UퟀԪgdV?pz?=!W&k;]WOduzM2" Ր墛9n|l8D%!Q)חz;j*+C={P󳔑sRG+#qPa¾:״*&3u*!Si]O>cc00빑庄S> aQ/,CQְ@4""4JӄeeUeY4EUQui%Ve]-wgwz{*뺘/\%z+;mL`=0@d2dd7n×"BIzCJDt ID ӕj<+YY30S9Δon<.9B2Gpdʺ壆X~!HTR<3]+kٵ{^Nv?V߮@}prҴXSM}Gl[}BVH͡fyRF ub 4h&<3E2"aQj +(ZR.KhH;A+BZ{#PszqH8 }A7v_,h;VgumE B[RP/, iebTcplZv dE: I7XJX^#^o# f4N7`OQr#)vOXvoN!ʍ6s4{;T >\BFl::g,X*R iarW5o( w_Xd>w#^ - ЙD!+ 3?( 'Qim/i;7X G7.Nn!ɿ!nNT& MSJrU-o{S5>X=Wǻ#ܮ HUhpɥ1 !k䫙?Ժ`ҤBFʣBS,q`Kpкd!kJ-aY`%.TÞRRPb2TߋR%h{)9FGo".M %PmHEPCW@;CPG>UPX*)_Z3 wcCˤy&+hHiGO|rؽXw)5xz۟&~81]g./#5X)xѽ&CC'<&8M[=Q@;vWCl {vu:v ^W9Eiyb#2 rT`|Xq * QdKgXgg,AtuFSRlLڃ6e&պzi?˳5÷ BQg48gBt6LϠr 6{` DGЍ m 0?j %Y iU;e,aOk'^(Q2I£gs/},ԏec7N,qn++(C :Uu-?ik(z(9f"Ϗ)x޸E"kBȈ~~{C& $3 bCI6pY BY;hyL)-+N|>XiKnBE܇cjtG1UKDD4>|}!Vk=vӺ:0ynlh0dnܾPoO*Wn4D*n)qR}!G0̉SE Jǭ{;'9eA`ho;e Grh1+ ޾ Qv/ )콐ւ%gtlhp[//~;X ,|Z_6U0E2*Ih8ぴ>gՈ}Eh9E˔01MWo.*avX," 0r6N)0jk,-ɩVsiuj2BU 'Us31HO?& 5%L-rSxf13b|fSWꘟo(Nk*Y :_#fP[6td+ӟd[7q /XA:7*tzН .oSpWm@a}(dG4g2; nvgdQEè1+K(gweYWAG`^ق#|{;T"TN zDTGpaQ+ k{m`zbwE%N[N8O\"7͌?>K#mjggG&FG$*hCdgK߸`VǛ$=+bU+:c 㰡7syV:c)9e^[|q*j*xVdb՟Ɍ[ uE$g@LN#I6O=pc"k,5 FO3G&Iܾ&I[vW퓓*ڠ-Zjkd)tmq:rMC$)I%IeƂ[;gW {ÍBʼ}'+S<fTo_90bhJ w33^<< +ĉyx0Kgm^kzm C+3 [a=H,Q{-A-t,_=;t 12Jhq.ΤK 2LUǰ*`a,W@ďIV||F5&Ƃ?'7 D UWhqJ 1x~oFE) ~HkmѥPp_ e=xm #dcb J(3U"cs Gm}~>4-T̈NAPD-NR!M VNaqV_q >i:BOܢB 瘋k TdAYPyt+sQOek{@|xDk尤",ՒjcnPw#Ouon/W3}'5 >*2x~:\"M!'d6{{T֥b*1ݥftߺ )f{jZa*HdMm RO}z2wıg<%at2KoF?Ev7 X ﶝT&3>UIB@X0~*CXqz)g;j׮m lߞ[jiy(].q6qU*iOE bZ)o؉Brg?6x@/ӐS18Yů|,@?, An.Ǭ[iYAV X:۶V%MJ5vlc.CQK}cc}K{ eiX0G*LMEQPSձwVfKh_ FTvfnDe t։O*-B ^A#=# no_( wv-&w_sJG$)C<׶|z:M(ћʄ*Gg_ m<*(ݦO`;|N@4?53=Mnt%  _+пz;CnA2!(:^KK׮YrŠ0Ea%uVQD2g˹cJXl 5I4,I|Mw }qΝ^4N @ .d̕ʤD[fMV RǾdc!_d13FnHϹO_T$GBRɿ匽.FZGYN"s!7|+?PJ.{:D J2gT.>D|v 41+@E5{h\fBk7c7h'1UKr:#ִ8aMe]Q]jPf (o_tO-15.Mfְ%7+"rP^'#pM­anb,6'XGdL$^<$fh b_w h0C=m0yVe *p d 4 j3I3pFUq[Wcʐ+J/|owG+t3߳:OY%Rtj"]s,iD)y5krRczEwtO.t棏 "/AhpP;^Lpz5%E_bTa{/\q `*o;&svbt!̏|<PwʢN@zL^uz9N&hԫ>ɪhrt42I℔HxQ=Br8,,ǦOH-ǬPDSSS29 MijXs-dOaw֓')Pi 1Bvdd-*zй5P*F&x|DixJddcqMG4Y*j C!'lu_{wwAMܑ2j+k;/uRiVr>-_6L¬o~l4kóutE@c ;W Ո< Ia8_bVEC/1#G?Ā=ڢ61Dq$E=[ #'5 :ZuI^/+ks©:Gӹ3kBx Zn!z s챓dORPr_eMy[ey>t% lMEjfrζy*Nd[ ~*M`'hٞT&S/8p/&q$$ Q@YZQX - +6 !!RebҦHCB*k!6K엽p:z/w::?? w>rTP.RY·ia:vr^f ^hc^ ?kpfl`-=#ELIK ~KcڨU;6 ^+kP!] ;$-&J34]5;@IoܻUN9VO\vP77%15ڌUkzs_ 5˧Z]C ]2!3i",Ҳ"HDDNg=2]ʢCTb3So\pUKM"UrιR~l)+;CDݚUOuy`|-D 1 eoUv9χ?XMyBg1${oxuD\c-O#Sqt𼐘{ iJt>b\iK(ELҁ/3 x橍1l&"D .b}c-lPQj)+ \uᖜ_Ҍ\&Ӧ&4댧ؿ3^y` p !0~uA*Ke.rSX+,i{z&'>Xџ/}_65vXV Up+ܻzAg狠Seo ΀*3rX(W!(I*Ť$|F{'s <6ׁKKv?Mv8pI?y-ѩ̂7_7'΃aģ(Rݍfv5ϞG4 5ՅN.jz} [[jX^Zݚ'fj&e:̀~:A7De M'x%$oEH qN8z5֙'i1EQxg./ _n ?ـLU0.s  4[x=ׇ.NO%?Dҋb6=dZb/E`Tοp(¡GamjF[ nω^p@'x|(uqhļvvb-\!LRB$(_=z|hChy#YqcWl*n|0+Q|~VHw80Mh/OrZ-GwE}掕TmNRv*syTT_E'7VPgEIcLH'YzEҳKrPo+i3L)wvxQ?eb3Ti_o/.Cm0-/65a&H=WCBPLv+'ɛxZz{*=a|pYƿKGA)d Ҩڕ\D:4s`r$mj%;SdDZ*y &l )ۊ׶;6f ty^M5vX;Tv臨m<>'u`Bpsvb8&s$$Ɛ4&$+iVJicia@GC[B1(c VPubS7u~I0ֽW}"MG3hu^_*|VWFRM>5PU%5< va`+&[eĦeZxi5yw.,k\|ubF=1>QtuY;8:z5NSCE@&*%O=WoثzSe;T}zTwmBXJFcgT-ᯗOFM5'c׌ĺeeW^W^p!7t´v-64DkKkw:y>nXLqBDuhYX ;;''V+WE.،Nn&I[:Z"Hpg~[]*%vMfAwKR*V<:^O GpecғT/T*w@q0@y~2?%73Oe!Dh v2p8x3m%HP#yz_ nB- !H\ LmcgP>iĭ@|[ҢBB} ]:I΂ͷmGx;ͣGWoHWmb}M0=mhi|r l~Z-5 c|l@Uo&77yg -~1x<'8N3J0C1x)m!! MPmKIBZC^ʩZ{)xOhYKvw5;`QTR.fo{>dӓA f-@ڻQsx#EҸԨmRy A͈} 1˷0$U 1;:W*h͡/$w_.i2FCUN9acQjG4?OX x c88~r>18\lȅc 6>,l P^1LK&Yҥ͢heQZev*#*M포mChL;;vk[^۳V C0Y\".EѩhCts&mC*"VC6T9Q')5Z:+8? ڟڙ} j樕7dBj귲 P.@\k*g䀵Rv L,H9^H9[u- ;&Dm"H')kAHxI!z =EehOy$ /Q ZIJfTHEx<ޫn@EsGWP5$Ai!IBg 7t5^}pϸT`Yfqqߍ\E̹bXLeеzjbg+%p9yeUN:jfX%Ȣg[A-ShX:dfۣB|I&yuR:NAHw..dPUJ8T  vR٥$p4n4FBPSgWVsv9m\RaE4YfM4n~7ottL!̌YNlW?,Q+$-[mۂ`{KKNxz?HL+V&Y?Vմ&Z1a|1Vah#=ƟcX !]K"CoD6AdG\< ?4 R:ACo9f@?5(>@C\U&U4FW[=Cav5Ԗk u {hW_ц+W }H)ȊcwO7Z^FY2?*a:;w- Xte0KF6M0 +s'͐HmuYS LE0gcީy9U6%FÍ eFfYb2yl6Xk뵍flly8Xt,99.xM.yC/񦽐9\މ acc+/h|&0.#Lc\6~iQf,:q9Ic1їĭiS#D" ʪٳB=MVn1'9wV`B<РQP_:Ԋ~,+$ o<n5"iW/P#BD'teRrdeK,gr ^di9%zJZ:{ M$>Z}+V7T4| MSO/Y|觽}ߡC J%2OOK})ɽHZ:®ĕ93 'Z_TKHK:XnrOo:ؠpM6ĈP'p;S0|`7&KH\@X׍̯wdh+ XkLUQm!H.܉br8dg|H ུ)ќ(&8w @-=m|oKmу]Cp~A[2w>Ԁ-$nv*-7hK qM^qHJ T?uy\BPJbVRrhZJ/|YO P>K|Pi"MoiJm mA,XRRkR[|tKZkȗTDn"ߺQ 1󌺫1kŽ5fc!n7˚ooIywZ{M'Lm=BRaV;q=4/X<ƪ 1`2 -+>os1=iYQ,$^|6V\5:yR8H~LFb;G%԰qmѽl0,gkS %`L (>]}\x_i~аg9Y w yW~WO=kݶ #k"3J $͐@$07it~<.ufhH^s u6H`u6o}.4nP/Ť QB"U ޑ[$MI:$~#@h8ҵuT(As~7e!+Bb? QeWj^MMg3qٲ΢N+y-nc-̛9za1eSq/R},C,,<lmuߘW,:PYxtz1 U&cxX/1wH3v~TP[~H=^DI0hR _:?j6{Ed>4M%SsW*$(60蝥ڪզO-4S4D̑u|h۾?P2b!laW[uA Ueу]%@J@v,dEn(=DeI"'TBYl~@qDJD"ͯz~LDO{[uT{<bzMƍY7ź粹Q9%!V;M~,JHq `$yx$-n@մW|JN*i j;D5{R } (N_HXwm#_ߩ)[fA0x; 'IK1 )`Wڼ"gCn@SPHӴc2YI Pf~~/3%۞׃q` D=_D6zb#NXUܔ' vLWy( pmb(ec\^ 9QMhfpESЧ;YU b."_44 QǀZڇ? Qz~&-+#Sz[ze!k/I5z(B~Zt;M씭sim3=NkYS*E*jۈ G- v_-$00+V(sh>媯z3ŧ$NʓDz:GPPpL*Hӫ'gFD6VDг %7}Jv1NHH&aKJ|{ >0i@.Peeu~7eksDn] ep]aT@4SB $F~Ju2]сѾ̻N-L)(+Nߥ H'D%t a-vѕvB$rAUR]irwP"c9+X_ǣ=bTlpB̏"e;RlA6AI\aƣ4x0y̘ *aULb .yg@=jg١>"`dw-]/`t.-5+5is `DyBa`V,m[[(vv=,PDTV]s!G%9]d=c113.ݮg1ns.\VAxޛ)/!{~g99=u-v->p .}yJVCo}@ߛWg~_?Ԛ,9IŞ-A#vø4qLrS:X{GVƊ}=ľ}>55]؁2Rz+8=@@0B1 Miq}+i.o } kn/~Tvbwi%j'+q@AYޡ MՁv CV9C(W #$*Xؙ 1w8ٰG,!A^Vɲ4`CWyI4ftVtMm:. ڙeǐ?j_Td?66.se4(ągcAR=P%v ;΢6x4uG΄CtajYQ˞Nnkkε 3}-PU3XB`8ARK#<--aőK)>ODJ _ uc U'tWQ6wէȘA-6~u*|H B5d*j'+^ڦO(s~el }[=;װL_|?wW[/|Pen~_.䉃jT `.z\Zp,h8E@g ">yas6w-*i'X>+Fs N ~f:>%`MavdfNK3R8vY0xJEK%1wm"J/Prm. G8 Q,M9cMB1Z"8kNU/&zALze{qpc==ɟ:!Gn$Ѱ)^+yB_s]91yB&r⦆Oƞ}PKGi>x7K ϡa\f\g+8}Y!D#xgD3FlK`cxcR"N"mv < hhOȩ`g>pS'〗c9\S/0塕x< sD+gpkQQ`%=R|A, I29Hl8>JdB2sr{V q:U|STskss9՚J5 Oz,*KB4&wi!A2R/j%kKXJUeʗ|hdd:0z Ո_ zFCs?gFh_¥d޵1jHc4b E3 W翓 鈇H7Vl p4mY| ֢IS:@5YJd JEk>@k! C`.evi]@d 0 }}KrT ^$񖄃g&ucfE/.ryHZ躙wW1k*6W[ummDf5q> mF2SUd5Q@L)s=EdR8bENlj7͍-,Gr3fhݩM\VwmgtIB J$[%0@lLßb8MchfRdXx'tg${_YI`F4z\+ zk TՌ\ψNtja/q^M  iih,HL MJr) !a |ߝ+$[lٚvʹ$@ٯ\1-)}sI&ix ?S/ _7RCQ/<\Mw'"]G2we;.E^^O3!nniq{GVʖl#jBhj>-ƤSj["b)!vJ p/<-Y2Lh pdK{4QC<g [ dNCq#[;3}DžV?7zke&A Sa6*Kx6q{/糝[^:k\8clj'6\M$$!% $b4,2FZBuuQ؇ mEE(:ʘ4ۨX=g;$*so~Hf(>.2 t& a=Z 0QK"~-}]4.եT-Ųw`\/x5O\B4H4pjA,?в4}զھ8|q\#nz-stͷg =dXm+~0<;rNCA,:ӳ?wW!qx;54rw zC'-a*4uʹ TbtNJפ&KFeb|辁_㋁}|w;,> pbN*k G5uGI^GPݣf%HlƞUV_ůLbX,b<¶䖠kX7oIU l'SY~^nJOƒw{v-t/5D v r cZA]v/lw}t z4tHL?!L^WWk(/#R$Z:E 2Ie8UFr(udCb=D^nYK 5:Q;S@P;-۫%u]'8V^tmnǶY}gsܤy<FtBF& MwM`-8ek}̓?Bb:Ty;E^,so?zݠݸAa_mlܹcǎTV;shc/Hy4+W캲TLWDn />@PaT\HI5 - zԛRR]I@48qjO e_unikG^G%O4،}y9뗜c 8$Cw3-VTyڿ]R* WSM8М0G"`Nyl6fCYȐ<GЦ+p-үBc3h>f A#j Rۋ-1dÝf+ӏ=۰:ŒTQ-=7XCyZj5zk^3" H##MBn3I•d&5z)mͤՒeN6.=! Vwt&A?1#6t-ixV9"R*!j\ԓbqO:*+t0>aVsX7vTa֭]X-ttkvK]RLUIպu鱞|t)=5oEsΎU-Us/haF7u@IWPXHU Es}wF|7%f` J،4_K!^栯NG-tBYB=.P3)| y܃1l~u^\@5) 1պJf$0C2om~JH Gt\?FC22+$ lHIJֹ(½MTQxCn RFm (ShFmlHfk<>?6 "B bv̍ `_=Q뜵GtB}}*z-d9cA$\_U 2Z PRʜ5Pt܁lxu߮hY+R& B^W"#AW1UN;89cipԂ'7GBE^lRVK $_WR#V4ngAwҿ{7Ac|2 8ap 31GoR1cGNmN˷߇b-%x)ir8!P޾\N݅foL6+~Tx1q9\*0sxG|c~7?Th(B"&pP5w G P[%d+XhoL,zmekm*pҚ\ӬŶ(d?9J@#իUA0I"YdV|lcn>l'ėî/cO-I9S}RN,{o2 'u<)M[HlC1w =q'8o0Uc!M9_u9)OFrw/(^fXŲY,S`: 5cze6UpX}AWE\MU?c5!P[@ E աq7@Z="R8wtK/gJc1o z{$l 1?LٖdOu%XnI6-I24lK4Ӣ IЙܯT8 ^d `D0蔙zeNK XXp.Wh+;<^]*UrygEӬEPmj;;-9R`q.S|F_dL0@{QZZ%kT6'\?A*u%M@C;+xx}0;3)-[Vƾ 5I 0 &AHu\@dgtB}(O %BqUt}fWK*Xa8*8l19zͥeC* e\٢pQ)ݏ2~1ߠׂ^a=zDW}$0 $հ+lS9Ұ,_O,wљ׵{7T1&`W37G/ؑ 2xN5/_JP㽕]{<9Ji:미^mbj 1bqK=ϲDpXo|eؔ˷i($uondmmW?jI{kOXWp[!LTAQ^;]Jͱ}&w0F.=o;>;*E#N_ZT*W9<219(8z(5 HҿdfzW3^3f3@/gN!| l,H!u;EH.H Ȃ4L `̩L:efĝog, v6Cua9qC8v^@/ mIbHca`S^Pm凂!?P1(P= C0f4%ȉZy¨?@ '-Kʋl%Y&"htV=|WQeS|~n|<p 5[cKNf Ώ[ k{-w?l/S:H[D wdxD?RrE_e1FtFDvVד==ݟI&AI“5/H!?ÙDHD/o~t'q//庎]JvSa? Mo(o~l y.'?&N H2WXwȨ',|${kC@|m@TAs%-'7Q"BFG },zn! 'M(hACqfKI=kk4=!_"Ϛ[N˔~r 4H[*d?c`̝jLK>Rʟ&:=Jg$BINA!*1 "$rTVT0FqꍐPZwKy5նZx?l'wGx{l7ٿ:F>Avп_szìҽ]/(ɵcc5ݔCvG%Dz$]!LTݬ+iZ4f҅\DJ:RB12 ҮY#VCȏb=u*/&dBYΓLG߇= 3 :]9bGM(Sj eHN'짳Ugfw84qk)Ztj5j8jkڬ7_1Nن?Fa6*/JY UZ>Gm8#h KLrWg+x9 bY2q"RP(Jr M@ TZ׾w:ɕM,6+%̵5 0=I D&3s2Ӄ{]&҇R%b_FJ M )`x%;brG~ !;pz@:@#@|0L7O c &p>.峃ah`][%6tBmu)9W<䞁 ),:>nhY2&P2H}JaJ7k~gXoHKg@>G7W$ XEkZJjRJ_SBc6~m2Lm[c_xtU梅'>$,z-)m}0&Cy Y6PK[.tc,rmʓalid˧Ⱥrvyb3so!wLSvk7B`1\͏Kk jɍawІK?#m>+4;X-uu0}HIZck*>X_+O "ZqT(/}-/vZk-mmy)!2t !#ꐻvx8nj36n8.cLt9/&9s1, O>$~2pŌ2q HYɜY>/*hK0iGO>{ñ*f.ml#vTYnē XԲ=N=:=B8N#ϛI>ׁ{VNWak T?'tE>>> b7֮^ozJ|SmBXW^?3Y}h)s˸AX#U#ޑ3蟌@1w2(v$o9 S.LR[,A) }txIѴih0/kTkeՂ,3kj*-o[YR idb1c58L(q^ +D 0iƟQ̓޸XCWaGӠ7eBW)Ud"PLzdo<(^ RC~Dݳ!m>!i'XpE_ ۩F5,KY}Rv:gϪ+@3t.)0W̐R(W^':<6$:,XB؉6b6_b( -jC<Cz$o Ts^:yӲJ,!Җ,l^i p}\(ts9%Π _=]c,xy׵26,V/ jfX~?fJيB`|\b𝫉M=tS 6zIknF3~r"f@ uMjrYs F I݁/Ҟo.mX`x8UDk;9aYmҥ 2H$+*jjT*^.WzAftK@p!ÐpaX%K!:c ' Pմy8O*Ǿ쫇Ys4j|{;:b 197|Ո.kL#MaL`L%)m pT wd}?Ty|@\xTx=!=C(h4IZQS\cpG-xޏ4Whݴt?]ܲ>bE9rJno(a72/쩢 ^.*Kj`^[$c,?$.6(RKY'J6+4:X%~&.7 @glB*V>Ѝ38M-[7Ԫ㽣>8 \ 7zg5v ?2_T}&{ip½pиw3@qrYׁnn4 S؂,aXfJ_Z}]C<".q.$M_ m.oXɬ?)Y2XYݦXZ'/KTf>ob#?!;"mTA (tbe^ŠSϼy2ŦD(Hl[q}ߍBKκ;Onrle8 _ƻɣwWrEDP%3G\$Z%*ְ nY[.-9 kEcnʅ¶;-a$}&thz+"_CJ)䴝 pa7 f's EQϖ'є;5fnc4u2)?DIm2A|7EѺ)K7 ss%!V9ko˵x(:dEjٮ&d@D!G4 I1! ȩXYWz22ֱs;㨽^8ijqZ$}@Y6}زxϸ4a5ٳ[<51-T/}ytÜ 槆ώكǘkf1H$j]2L߽Qhc_UEk(Í>L8T* Bq{u?P'+tYPt&\HnhpA}w;pdJQ[Qz׎ұpCs6l zxH^G7@!s勽@r(HAG~Vu]N"(.3jxⶃ E[&jm[i a*oŹY/o}OΈ%IU!)zEœר2%!;\,lX4SxN_Eӵ,; WAU6.bkm{ &>B{& S hfEB %[k4^qw!?00p"Gu 35e[}5@yU=MASh)%ja"i2|)ȝU|O'&-EQj*XwrS΀F׳Hu̔s0'4>jE+/#lD;qc>/zC!)+K୯ȡ ވ f|7i#! a>ضCXZ| ɖI. PecmC/">\w8m 4L %Q* @ @^,cAAk '`y2AT; uzW_(įL\~lY්EP$gk*tWN/x@ C~o=~kZէ婕ڂ痮2w gf.u/쏖64kZշ9,nplsĭv\N Pba^Uc؍_WKCJbqš.l/q;0 z)MEȴ&$'Z,陙CLTcYʸO@<~ խV-ntr)GpW[G M35h4Or TM\|V$Wd[oGQq:x82Hi8pНU.$kǖB"Ipj%-D2m4mLL[rrЁ-8_MgۚS aM'FGiu\JEwTtOW=^Z?߯JWj)XkT)) |*('0R8Nǡc=cqs2cn\cȍyB-.s3f]^+x-M^/M>'7s+t-ܦe{x QcNLH)*` bo_J/hF|ށҬm(*>_ rO#fG+#S{cr0WL{a?'EsI}wou3Ѓ*k֣WbŇexLӬ+&`G&~IKRUto7XV7oꪀfIWܓ<(LJř\s,{gΑ>''W`9.8l*ۦ~R%ْqSuCE~ gTI磁*yDVqM ٘`bU>}g~)Tˮ %(75եW~dFMfb2SueU1}CI5$Ί@2ْH)b@V|h9)R< AY0x'hAHw3[[qv PUBe9hl1}7LHK eEfBeڮ6F FH^Ϝ,be@fNJO:5rH+{MXoGOllѴc;S[ \?{̟z˔ wxE;3_k*Tn 6&K[Is&FG`vȱQ%p2?8zY`}VLЕ-Q#3Rr-.k" >:xWMS{8~,r 1~}jQ#EFӥiw'%W= тl0/#W{gZOEz=6ڮ wkTz=Xp'+Ng:SҌ^ͷ7ˤ˪ޣU($\Z6Tb`6/Լ9Ď w7 #R`0 lNxrXjr_ c <9@ ;z,1()D{8>0NئE~y9;})vx6&TK'$ZF:2 3ZU8JebZ1(=.O'B~%Ed=oJKɪ7U#17j>y <{k\j{O%bUwH`~ce^{xΏF,`fruEW~ ;H48}^" 2<[jjGC2Sh:2:^ɭZnc;OY~-G: +^_cwru@ѹ6['mYȜC)&}"K<ח%Nod>)s}һL伉>7v_nfS+:=d݅~t=| o2HkIbZucG f=}~p58l,|glz0;>bS ߋp`_*aND܉9._ 4Z5v̇(zk8 ̎z\jIIքiH%R> KEJRR*$)QQ0% jO01Ԓ/;@" @k ;wE],ϕ>@a>wv,P3S-zudhaԈc=_O,FXodOT؎n|X&/x Jaמh3oj~w҄U :[ $!yfE1gg.!XNαf*DbᖐQhضMk(:Q %%!EHQڇ_ClKI wj4y7RLB5+A)D&<#ÿ2,nx=vs4]IXW6zJѿz"إ]𼊧>h{P[sjo6[b;?>\*?:0X uįqmt=o3!`]9zA| O΄8-xoY_(UerefZ)OXhUF`ҵ{gtnWWݲPV$z۔m?jeZ k22Ui&L+>r(HTTš^z:bؘBynX Dq)k!48;5<$ c[8`eM.CkVAFc?BsztOp#?6rQ#-fɳa[E/n8sieˇGEr^S] 00N45kݲi 5F}g_;p3$]LL Yf4/v7=1<ُ0z&ŚDcm,W\"5>׫ZŒ8N@kO$]W (~v_n@(t=@hwl<+޾.7{$tr8trTmpuxZe;u>p^m 7SG #?pvfg^w"ε4IHQ!"'stnYڄ)䌱#a&Ց婫 ),+m6a  RA:L I 0 IT !d,i_H] E乪+v+UHV _喓vm h4h h۽O}#NRgrv]Q%DpX8_ ,xx|Ηn@5@5 ,&AߡJTnfk~w 漏XKl$j~Fd΢j=?ښoǜ c{WCr]!x*&qǰʖ柣Zq+E j? U0cXA9$@Xc,2ȵfv؇$Pεo4x| Q=x`%&݁0#P&<~f8qL@\ 6qӬ[]NGω{}Tnh4d Y|!W\6@Ìg@aƾ{0 a׋iQV4GspbCglw z.CX8qEV"D4.esS(d[o)ua^|zYI1Wc,il/9L6*2H!Fs̍;*aȉ egG3d"m%]Kf*m4(Dwc.@v9DCJ;ME(jF#1KZNV5juv)(=q]?TJ[V܇=*mR._bXQI롮ʚwx{iA\$f*ڼ&8,΁ v >]1Mw}RRxiQj)k*ZJ- a) (Sr!Řݶnΰ3.lb?f6c>y> ~?]̋ b7$N%knq.YWIq.|u~n.םJroS`6[6˘FaIa⥇8#=L;$&xW']e# HLo!7x_8: A&v2?)1c `sN(VX_dC†(d `߰{n}"mW!4A0mSc ؊'v*TR1zv{YZ//>'.'eQ;*q:20.-(twdiƠGm*K/;(]) ē^Ѵ@ 4e/Z^Ү+l Tԛv dy"Ag;!i*rKm)>8U"V z#X=p* {)J@{z:EpOQ\/AS'&~fũԱiXé |wˌkM`ViPD*Kg8/6ľBHIQ0mf(XT欷܇v1LX3V\#;OkkNH_=?==eEDuy ܷ=*JAXNm 5qpߓPTC5Bu݉$pT2֋+2TF*V<"! [* c* <ҥt:'\o` ie;L`n灥;wfw&PIȄ=|1AML,/ #ߗ m(F%Gf{yҽSNq1F2*^"&>)ܬhȼb\xE>QEz2ZY`u)/^ek̙ bkXӇ U QiP\ K',d gKK(}N)- f/Ykwtvo2,lYsŶLܛ2=q!^$Υ^,_#U{A Ri}tĺnh܁5VL?$3R a*k7 b|w~v3T8WxZO&j{WI `YEb]DmO8 T 4"1HDD !1ii@"_agUc-pᦞwgqKyf!⃐ t$`-\2,W0WR'e%8-IMT<+x~,C7{t겵njB롭ˌfkT8o" 4g$?R{Z; \zؚ/# Um-CӪ6CGh1|Llu;7%~LIR<>pޘJU=$[Xx;%$] 3Mn)V̄O=焯0~ݞ3ipAY0UJ;}k{L>S($m۰AK1 H]]j.x68uaypDze %Hj"j*  ^g(g8ٸyi{회]oې|Gcz`X 5<`i箣+Z_ݙat6wGIM6 @"h-coP/]KB +MBLiǀfXQP$);4RXcmTMjf=Cɾf7 y%9 }Ρi_ԑoiUQOKS6{Kx0XuDNb%ɚE"%/ϛRX(-J1dO`5'1aԲ0NbdlՕ {k浘4\?v7nX[2Whzp=ĕ~h3@#\+ flyZ&}Yg|h[&")F8؂)&t4rc&F=~@>]Xnv SCTʝG,ŬHXS˳KvRMtՋ/Ӭe "Q1M| W=CM\$y*e 286I*[=R6+RaNi5Eک-!W8;m^ ~Z:k XchtTkHW߂kq V&%c6%6 fM8dF!^-O"c`/,S2tn qOS=TKO? >\5[bX^Tf6GvnAomdu$4:I p?&ӏc 2Z91䎅 9vR>!ՓvsQXX8@ Πr| M|Ts" Pa˔Ti{Mɼ|$!^РrwHVeMߴtࡾcs¢9Z!%7̶,%g ,* b@̙x"*( <^W@0>늿{c//N|0``1<ˇϱ'c8141:iXPN1&41:mTMFBTԆS'{Ͻz\G׸p*BwW*lU*ex἟{EPV8@3# .Yߐ tR{,Eq}_n!U"-lk=~; Ü>iRtTg (27{9!!Gb5 LD 2_o],;xڣ' MB% 4UufrWf<&{nD5snn#Ug1ݸ~.w5<՜G xN 0ܞLڛ}A-MRot(^k'hG3M'D=2jEǸ"Q >JJHe">W,LJAAS.c@-;2 bK$X v#IKtv7֭lM{\JQmWAhAvD iAް#Xr!ozXXlKfP\ j d1]A,efD h 4߯}.sHU&;Ӷ*tLz}#3y{3۾BT =Du?T UQl ,N)(KP6!s46&蜙jE{pNzt;`qZMrR*.)5i^4:q KwXq.9'R1C,їɾL:ċ{Vn1ʱPQi,}}v2 d|R?Zb[ao#0FP`ĤըR}₱KeDQ MSE?C TX%d>9_<5Z|T"rVFb^pkzUq!_/Fp~]dr|Q RJS0e}6'۞q{;Vs|mư76`_1ˀWCD3T >+QpRB{T,RZJٙMEYtuNU0RB~4hIJhE;`P(@BM)X8hҴ {auk@mIxm]&y&)x+]K;AD'2BYiep >7}}}mN-\HgF~q=уiYЃ=G5XPhH!H8g;IR}H*B3=Q@MR)a֣P#0^:);6UkzL<][1ZV rQkss14 nT^!iZXH׃l𒧇y:Cn:n58eVqq .ω.9wІXf@0`3"Wf&\,%Jp9JdtFI*5YW RE6ه`劓xNbt8ogM>WH;# h! [6mNuS\nۦ3Nct{pot'!ɒ%Ya%8BeY/ɲp,Q,〠 1C`ꆐq\e\ڒ %id\3 餙e-ݓG2ZW\)0bCtMgNhCUa 47\b6SKP)ǧ/k#\Tv' J&@)Ⴍ˷PwOՂڝ>-u~ʖŦZJM]rrkU6gOq# mO1M(yD1Qs^ڱ#MЯK4J4=jj,(l̾ ϐkڙ)%k2^@KRB\ i c6o=pvog*5/)X`-LlPZ()\↲` j\4VGJ߇>Z~+ F{ƈùh`$,˶5z͆z8ЃRDoE_F}x{ޅZsΪ'UL?N-[o#Cͦ]C|csԅ]cِ)u 78q hFv{_nĵV>"%V!"F!FzP4V6Jmm n6'?0hvԈE\M*y߁m(]eᅙC,txV1ڐRaa| 8Ӵ,}ˇ^iI6_LJw_:C`E- g״QA2m(;Yo|RwBêWZt Z4.D$Z1 cs;|0:H~={ W]~d ̿l.(eH*W7?Y3|V)adjRQ7XKxʊ3׬쁷2-1;WRD{ce2 jQ`7 rG+@*?Fzli x`\,։ :㬶!(i)(X pRT[#-xlK*\vxbV,RI̽'-\89Vk+ATk(9W}'\d\j]UuY )u}PNܼvN+g@ AI^>w3-u\}lwCڄjDls >GGOǡ3wKz.}{qF?v/m:Z(V׎%psL*)7#(#9o^B9$0 KAEV,9K[$̂aHix&jEP  UEU"މ8)@JI [Ģ?mZ yJ%Gĺ|~q}m#O㶊ܪ?wp*K8}u׭ p- عxG-V*$Q7|1q's>ge97/57V*,0? KY7w@'zf- ľ$0^5tlÝ7_1>& fJ%"Eo*HSKЃsEQLMy8"eS 7)4 eE4[T9c&R5X5 0dHuAfB=uл1i'.*80*#i.N¯2Dͻ~_>! '}@gNU^F8SUR9gn,藤5Fj>du~Ql4s2q?]Z#@#͑XE<"K?G.ȢȳQ>8ys-lϖ/D*%r^W )eir[$όA9}7nƃ3eP7ʹ~r/zw|{rwMJE./ 9]{پx[A}qaP`qcgӸI^e?dH_:`1*p;Au0 2jQ 6gw9yh@ ߁ = :pozyøck3ֹ}t* X xWch.|iuq0>&_1I1C7Qˀ=Vkz\rp&Ew@Ng_aƹig}ѳYlϷqYcd9OW~ԍs\~ cqZomW booSO52gޕ;Bp9ix"2MpJ4@Ȅ6,mnf  A3w0e6NYeѼyw6WMq}lBxeV>@6O7dd[YugT["jϸxEhy;D4GInz|huic_Kv71εۜ ƂB{3fkjW6cdbB~+3vQ')$?αRAsJ9ٛ`̵Gzb+ B'-W崟rO=a|WQ^eP2tB˧6Ke!Ĺf)Ks46cؚ8Ki뾐iԓSIK)[I21 բF5)z>s<ҷ&~'ǜfk%̧֢cm'ʙU׮|{~|uG(\Evt_pY]~ֲKV0JɋqjaMu/k)6֩<i6'i{#U|\hwc}1~vJm͝YM B(YFk*K UMZZfOhgk(~\畇Wsʘ?fnT7Kϕ71Cs}qO=:3U\]7Eֽ!{(OGiW>IP#5]׎о{9~\Cԥw۔`.v~1E!ϓ]볝i) ;D~üt7BЇ>欗:v{󱎲Cy \3yy3cVZB=g0%y$k\%KQCgI)d>կ&T˱jV}BZCs KQ>e[GYխZƨ5kQ=Fgb꨷.#~wu!vVP~y&U!@y!ZR?C7XSCd-~]Pkⷞs([O}\=k4c6h#AsOG?z/xKfԿy&fPZq iY?学"~&Bu6AEԥW)-=gktN rMo>*eu̙ӳC =;N裌'W' %q뮹}߬gjebU"""F1"1F"F#"FDDTDTD\WZՋ?`=6ڋ}}_zxuNsa763V}|NԖ酏m.o# HnRg}s G3Kppjw l)^gSh>pEי'1{mWjcIb|Ht\E:/qn3n}t^0_=bOg_=0t:md<_R.6~g'fmuoe+lrVn5okM t[߷O'⦳+1yžk,ϡyN" *1Q0Ft:"s&>2[o9@e&|[VЭc-q9%x|>Kp*R=-'3_ǡT38v>7=}pum\N]mܵ{?cSVQAu0߉H{ۉwj[Y[$֞c<K_9R>yvf :Ӯ11+>cc, MA3<mdіMClY@ >8ؘ:f_g&q.[N"A4vVN)'?@G9Ξc ]:+#?:E=󞢿o}ߐdOr um?=(]ӻ܉w{;5__i.6Q<}M^#=/"+0vn¿gO3=oXǬ!u9sG7>Q#xmcF:c6>u" 0r=踷#H%_3!B>852l5sz\%tsE06OpLH"edlmkW\Up8/Gkgһh]53F"gf?;Е[6Vkbfhfg}Nt&7G`XekE?FIauO>؉2 zuci?"5Źsۙ-sA׷:,Ed猛:x!=lEX:_~?E1M<{*:'3 tωhCStF>-&٩wSoulY٭H-` As;RيTX}qWB-z#e\c"Zl,Huxt"SXc=~mGEU+^m8ymlsuu=n{ׁq)369ދT]),s64If2ktΚcyk5G?0ɸ^tnXcnh ?-1\5rXWa1FRI*Xof:}]mb܍Oyϑ< x7"ϳĮMUW_Cn'͹oHy~.֢Ǘ'#ՅjC.2[߆:\ v>ߎtog˶-[4(7c>NEdLG`n,W]FV#OU_*G~eWWY.GzVelĦfg "F-רo@SNپd}tC;R-ܕQDrd[YCU{=[+OOE~I y'}lߝ >e契L3pО` ?(6vv?d3giK{wճMkΡ7[^Hu`d|ZkFʝSȱq3t:_"Ŋk/c5YkF5}n_݋g}<nE|~讽tr=EpR758|Ssz_5>s.>I7?jc^gVUIk9'8пbO>b_3  z}׬]2vB嚵+kklzK/2l>2~E}CبOB^wN ݮ\]k~֛cAXw=|W`^F*.Dj5UlԞw=;Ы{z\/c#6 o&}}`>]BN!Ru;q- =!|#+?8þ8EU1FQQ#+\WqE\"+""ca5FUkU}Z/v뾨|3My~9<<;Q{Ŵ׹5WW}v?1(x (E^Q:8ٵ.:C֭(=N+ğطnԸgհsM,;Î(us6fqM7M%F'?NІ:v@' k3J;#zx«wݵ\G`9˺EjKDGQj+T M] 9:~uwGko=GCG|2FC2o7;Ch!;| ل_ozC{}wlMlz G['Ĩ˹ܯ12%n\ΞE#3?^ ~kع". Yten;&N;;oz|jpL_`)8}7S= f{=vԒ!{ )ʹI֝ZG}6e#^1h{> rJu-|sr.[G_׿/&عw*ed\9$'8o_ҫSyh&fฺ։ν{(oeú}{k5*Qjom{wkꮭtgyh=gg_qmsA=Ik-d,$o_A{սlտ{.kGegߎiʹ |*y7L: ͇J z-]SFF|}xvX;'p0ٯ;R6E)Xb̢<Ѕ<ܣxϐULJ2kgrķdrx&JvCxV`m| m?N6G_Ey[X\DOD S5ǽ(=;qs{p#>84dt=&tv zRGyA˖>64ʋmz~ -lǛ n&p'_i*лs U{ 7߷tGyLs?;2iuDO}YI(Ex 70Ÿ)6q:Z{ﱞns<4~7;~{F?'*^y} dBo&cqێk>\Lߢ<笁.t 6:̻F\֠e..'Y עXIOM|/q_?ޜDw[Uhs(T%ݧ^pV:MYw~ N+_nvEg, :2^_=rgpjI 8_u11LQf W᪰)| 3mXji_OFam&1D9Cf39v8 bt-[1nVMz_o9d֝EyރFĽ,kh:2W.oE dmEyߧg\$M*߯k7g5Ȩ  >N{?&P~Bq?^As?ٓ3*Kys@.87йT-hGQsӿ7v&T'o/4%O/|t9V+"G[E l=e^̄A@'ܽ(9;aU>״Qeo }w|XS}U |7fYՆ(쀽nE: YVQ~CX-$ME/&F=sh_2ˎ?@Щηj{\ez;>>j!޳BNj&Z]d |xG}q'cQq8"""*"""DEDTDDTUE&j\33535533טqwqxD< f2Lfd,^/fMzn[/"wN"-۷_ikWhY?7a=ݫmgc&f6އ{ry Y1]ŰmAOPKr(c4ؿVFa; U ~R8迎FOQI:>@*)8s1¼ЫC9̧8[tMO)bEg?['ɞ\%oLe]4w>zxt܁χ_CMtޅGʓN6Mw8mmxoCKUysv4DОAId¢x/xWg"&HkQh;*~rn"YKmBk-iHŵ}`=R׊z7u0B>0FF$?d/+]̛ٕg\ f6؛F׈;펔]T GR:ε[^g:k1;Y..AcylIhrWo2Os!pvf;1Xs~glݿG9y<^ ngAlP'Mvex;V`asge9R,\\ՅΓ-??dm"@x.rָ ?1Ff胯}o֍~Utn濰]여s~]y ]/f2ٺILAw*u-;V5u:s㜑>3z#g6Epcxv~0R>ٛfossceF=\c:3憡9ٴ7o1Fٙ9 DEf}H:ضFPh99^Dz6ѩffNwέe棑(mgLdmz:/lPTGٺvL~WLKvgoF=N] fuqn':>ѸDʿ8=SMt?\fׯtf 繻[U,-;_wYɺ)O=BgQʃ߉o3n<{h9F01^/Gyɭ9x=C.y*>F6/Ew#r(k,aWwWo {:rCc~%k]mdYp=w-қzwq,W[4%geީ`ަ@Y԰Eh?kW *oą׷y[F/s7hulq0m^ $w6>{QHoxss;-d0^dg]"d]_cmCs]B7:<]:Ysz'd|}LUtqʙmxoqOkwG|i=8SzCk<^B#tnEwS`8 pXmw_v'zcV9}V6M,8_̳fYqL+|NEϴK_~?dm{߱=:*+ X}ۜ6٣ټx?˙NǒVTzvqυ_ \.p_5WQoy3 ٛ3 N6YsOSwX 7oDvVw"_ϞA]G iǙn4<  tսp~)>Ƌ`$[x'x.]ߡ]މ|gw =̥oA8}o=x^1gcN u2{zZ?ϪYUbTDDĈ1FD1*"F1jĈ*QUOڿZzzjg?{s|Ϲyg L#p=1`ݍ/;+. 1(]Fv*X.Ns>c9}O#ww)^.?u|d\#ϙw}U* K: 0a.՘K|Wӽ!й:ؼW u/,):8rXOa+-^b\>C_W9i60i 6ow ^xtܗZ+ԐG. : cՂ N5xiWc0w7EdN,U,m˯rXK接<`}ؓ"s^/jrLzGUla=3'aoY#kK-03Ȕ-YfGK@|=\rKw+%;:r K5k, *`zج?V')_o rRsKuK^d=`n YGR/ lgrΏ0x\Zz~(?7{ Λ}ʮiekkhi_JOok|}ڏ%{ͷ˨̿gqkkݷ?u v>}l$O;b!$7ccR?R>iYk@د7 }}j̏c L9̹~|CNd{G0xO1 IZ_֋P>ڇ ;x A.~Ga]0)8XF?΢g[:Y 1[ |B`0>ێru9?J fϳ.?ROIYtwb+b}pE{cR2%zW6,.ZSy9~g~Vf5 ^kKX'-kziϕldRՉ'~f)>a'~}`G [ޤ;k凰N;!' 6z?!K:m3Z]p,G&"O=4]k]KwX=9-92w"],Mon)<)>oo٣]`Yy;Kr;wzt]cNfert"ہ_y{{޻,eh]ׅ^]p;yU7d94 7>w9z;5b=Xb=B|švXz3{=U_bqg.e=w⻈>@C<9T߷VA_'sw-'|X_ nWw-˾Πg^:OG?ΛYogv3w)=>w,Ͷ}_񭻠̚U9uk!~<{"yj?c0l&{ 摿l_UsbW|-&Uw,p Q  &'C .kQFN\d^=񖸼)K*}[ ؒ_`d~vt7سυFtu΋pUrQu1J;cĢx;∶oYigt@^ Gmskp^z;{6\zms kwkA{'Uć{ό֞a,ϰw/'Nci#yf3 >*@^9:K jfډsd.gSOOjgOWP`O eįjK.cq Xrq^{>^>Yn: =uY'LG೹wlnYSw|**{c|[6TC7e@h<oz7ٻ/[2Q } "mOp mb>d`-G{K:'73xcpp>|C~?*6गoHjJ7z;CNaҽ'A{ؑsTWj¼hQYg^._c*|2&Yis}w,,865x%`n?34SsboZZQDQJ(QJ(%DQ""BhmDDZkc/ڋXk^[s2^<~ys_ |Kt|/w5U,=@`:% !Xycȷ:8PE'">:.biMW=+;9f`€Nkf*nC]|cg`3 y*9?Nf] o\OdHwTZx|/U>Ξ{14yd®ʻ9HsTatx-rqǧ0>Mى}g/9u6V簹tu g&_̀ RCm;Uϥy~E߱S3ᨀ rohU %Φ,4ܩG3<<Й 3p#6a~ W*6R/e|)kO}`xTROCܧ=7S[N>v)bkhWvAp~?Z?={{l|JGV[$7lv5ďlmnڅRDoe/ż65횅߮&5p+ghYr <儿d+Ƹ 1W bZ eV{%|k o·6Y#8M.ukFHzVJrS.W|jXAJ[vn^bVrZDalP2d64UUKE+N~"^3Q﷐xܳZ}pstx}(7hI:- da)K1+,reKD[5hsmrǢ\m&w.YxUm/ FkUYkvן-⦝hZiph^ߏ"䱫)G-j:05zmރ*au[l۪E|u=}/d("׬OKG_gOسf9i, rz],rV}T?}N.Vv+,$>ӧww6q{y;o=3gWuq ? 9<9>|QfowN1^ cs߲h_o:Dqzp=>+n})W=-t/ |q Sx[Mw;pz O|q >3"=p ݳt=&ޝI`g= Z ֿ^W='yNxSl~Obu+_")xx?$yÏ$ ߂s ƇXxw=0;?͛>-sN{UMҝi|Gx._0.,k5qՈR5*""*""*DEDDD5Xc5ֺZڧx}1s>9s~Y 흴XW\c^יW8fMqipR'hb7_۴ KЬȸ>j50x:Gu]:CnY%zE=pCO|z8lj?#XKFG:gS 5nrٲOYw>gI37,b5j-ncZKNuwЛ&}ZY7=k;]?a/?]l'܁i1&`Ytv<ܕOYn'b>"I:ryȷ|ԻwP-.ꎾʫ,*X+M0(UY8gvW8ǼegA>,ruqLy=M|O,ޛv=ݱI%WSW;H4ԇlӉds$o".tHޢGd0K5-bʐXDapHW >*/GL]&<5c L{!+>ÿxϊW_(֖~d.6,b+18>9cZ9ƽ!y6z"?m>>\@'#CC6V~߷ Я0WLR2D{0_N:S)$5:ňt hmXRZWTBE;)"{tn]!ƥj:'Q~.2W)WPU +?*;o]urŹERܵC–F}|K1~Fw=Fe|m=znGG~#+Wmq*ɢ"R !}tDޮ9tnQG8Aɭwyא8_zYB7ϐĚb3gz4ՇYB~{wwrE>{EtΜB{]C}DrvźhߢG\#XUS8;h^[3]П|[yu}ίSddߞA{,r˚j;}DdQ rc(6(W&~!Uwp j w\^ ..ql/3@>z͝&+[Z S"DO-b@>8z0-w=`(?Ygע6VOXOn^3|xj>Z / 9쵑G1acVl'ꍶ.ՠWy>9䜤=fX¨7 ]ygw/N6:YxS0Y[h(t`h'cKd%;+VJWWt {su Us5 Q?H6>H=ʢ;mZdn[xw^9}_;xcJ5CCsw yDCi |\oU_ϻomiaHآ8I[ݷ `F/[Ecs _WQLX"P/xg׶U+Sd_r߁a{  |[sc)nR*^Jx#(U_8B&azSrzǜ;u"c;ՋpU+E ~l{L%wzL꽎wمo"ʾ(X ?{UcZ '1_O+ ?jCzQU,+"FĈcc1"FĈUQODQQQzYkڿ99yuιs9ߓ|$1<Ew .5;tn&Cue'/_1S<^bntvo)Cs]wm79 =h΅Y>0:[52]O|9OдѫșM)-r[8r%类jN/x/%i_{y7ЄF9ײ)3zgB&_HŒvYD-t(Z f+;ڴE",/ǔ ;~ޜctm^[wW}O6])tSwek%-F yCAx]y\v]O6R<f7U8ݨ`=-youY;Cc[OW^W.䮹Mq^Nru_J<#T$kj:5mxWF"RM3Jr#0~x |4bwYן4jb; [(<,+V -hQ?Nd{"OI`mܳ>}OD)sUx[)V)?n[) Q]wA8TԒyeF:vG {ZE>{K}ak9CNdex?T`<|"հ`;*7V𡰣CI76Go3ȭ,QW)w?ϱg5=(_L[ԯ56Wg-0ㅻ]0:PW.~P>;3Ш([2-Zbomݪ.TͫzrCZ@o&,pUg2SR`Eƹ}鿋f/.@_FS5@)+>W7iˌwT.pEnn7r6ɚf=p/m[1k9e>$1Z,¾¨Ow 9mY/?Sr=t7MϽm ,XxU֕sh }V8yTG-#[m,ce l_A1A{"f9qIgůY-{}tfb dj<.v"qA FۤxiYm{q!w>z`_rEV@>{ FzM^Gl 쾍\Ǣ_sܦO)6ư;;6g=vuBw+Γu]gW9O"k1f>jS_,b zYxbu+=-rⒿP]1Y`){X\G1Qk>^OkʽsqRE.Vd}'X_V}.ϠE}Mg Hk`;Ax-\()lafG.ZԮ v}9_جˉ/YO>Z/<{>dM|ݎS9.|~§/9!|uɹ ].ttp9}tqZ6ct{}ΰFrܫDmW_tv<@hCs{^?).eZwu: "~}9CuFoo3^IX[LcxdG8KWu+VHv;ibŜ6T}"vOM ȾrL'vla:v{O[C&aa贀MO8H'F-D9b<+y7/mFߌ߀vq_k tw|ךo;A>RN-)ՠ! W<û7kXgXI2w4osv ~v4 ;[g;E7F$v2ֹyc]΋!-}3i0v1X߃ȷҶso1jDbyoFM ~#gQv#1=7ڇv/n72y=]H -T yvPh/ DdTBHll:sg~=e F{C^mWWC>bޅ1{wkRg-2g>#Ndэ'-D~ajLƟDb6]\z;ss;[ڨ(O =7 #h]o=pu)zoqq Gۨf,D3Z_q^$n5VM{sˑ8v&k܇nE KlABt>wƪ57rP̝%oEbgMgV"MƉs|э0aa#mFb+|F.W~3XW+FZطbm2tG:G uo[w^DM|&M\Yݟ3Uhz͵f]|IGk52<0.Y|gt8or 监ov#@7 ꬭn{ιȚދĝˑ}v$ѺN~}N#s&v52.Ctub>ĦH|9̹xžj}oD|w>GVn$N.1<um]mGbn$oB_'>orVѫ2? bljW*:u"sј d_xq? {Xs_4ѧ+=Xo])lF'Wzw$r˘aŐJd:O˵9;dS/s_c:_{_& ldُQ=mV/'='1cȞvq;=z52_ݟXw]fzw-u?{70Gֽnd ǿuz:?*{F$-h#DuZwNaS˪}ievM졵|z"׷q-D/1{F :gMs\O?#{Gx3٩r]Dh}\bOOO04O*G6zŹ 0̊'k/W}mV9WX̗'6^EvwZ52o{'0r~D&s.G_d\g󯱙|\P> r8/O/ڊ{27VV>g5_Ndn/{*~ Yǖk;y{1/ƑU9", B$n4\α6mGbVd]]|\C?~ o?d#tcm%>#m+Z}x YW"'c[ȵ}8CbHDžs^uA_[,[rQYVߡk_;~,D]3>waLku\F0k7j/c변"Rg]r^ޭԸӉ:_tȷ2?OG=P\<}hi 781oX/^FO#S 32#C~wo%g[VߘfvYs s[1l 3|5cd:Yu^k v8f}Y䓳sXg}7g7Dȶ u[ieỻ<ۜ9to3?=Ykw8 *g&w[PН%4!K5?hs[uct55Oט; d2{ 6U>s޸&2qgspN9^ )򒳍2d8-N#uUt׽8+oE_^gحZw lpQdGwynU Ey]{;^[ /ln{q&le\O5kCO"W9e'>k <]t$56~YO>:Jv1-Ox]ߢަoXu_Y=~z~3Ǚc.u'x5ʘsm! vVN-}9r_B&5L w- qgM\}Y] >gzMmgA5% :==$?OimAs|a y<>>燂E2sֱ+{o0Y/#شw+ءjjȱ(3_YjSA}&1mM"ngݢwQnzm_z+qg};u9?AUS_/_p~iglR#\X;w U,ҵ e W=Pq:sY*&\Eܰ͹Ƣ%t~o;5,j˲Żq]YѵuNҥ-*a[,z#ߟw}YY-L ȠZ]o_FG.em`3?+fW|;9g!tMӝK]mr/qyԯ.`E|p̾a}η,[^".{[\I6[Ĩ?f^b>H~Ul-Xp',?t?%|Yv+s!2;P~)&;-r[29ߴvldE6}k>/9ȃVs{wg?*7:lړ-]ndgO8Gd7i̳vol_4;fu޴mdGv5sq OtF}|^=ΐq|Jp<n+'\kGhiq8DGdl2UyI2F>6%qEo7_$wO =cqҹx /Ղ]dv |!6UQ/uw!_#|{z資N},'\&sLUz6Ix>/#>ջ7($|4n}?&EwFN-Y#ggYK4ͽ=#zÙO\_ t?ICW=]=KyzȠ>sY7Bt^ʧtRq-T={UW߀MU;Y8WZ#pU#K<:-?ٳՠB13w-&>[U,p\79=l_jG.M`V= W,s3XԪWѲVsإl5x4w8#.zv CY _TUMfɱfs?[x W[[[@_/"ChrJ{ʼnbX+'ڪ$q9XZ5dP?ki&E_Li`C}8 ō{]?ق*3ȾxJ L]"=utew_~bQrm xl~9{q6}\ -`pz1`=Z7 .O{?jӼT<<( ɚt/֐|>u ugя|d\G#@8ެsVU9U*"""b1c#b!FDTUTUUTU-+:U*֪UVZkuVZuZZU7O<{_yyks?G-w_Dba ȶ:u^C3r^lb_ ?V6e.3dcJ<_֪mlz`OY7dyZ"ԃW3o` 4T[%W>z3uEamZ`Zت42ເGq8=O$&HZWzԡWBQwD_uGc5n ":?7.`m'NqC]YgE~R=ݰxc~W@$,] xKIwM eh-#9XPUxo%Ŏ$xgK?^n%줾=佖;iKMC[N>=nn%}ֻv2o x6ǼfHmʶ}uxv8*Б?;yK1VGvOZF%Q]r_x/-٪E^uud̈́Z|-:iNO2_,H8Ww] Kuo"ǝK~lvw\~-H}h1+6=4m_r%;E?F1y}vz⭵bw.T?*~ZbJ⯚/xסBF}.{)0_Cy=l-:,~52Nag_&E߸> DSo;Ȼcу'[t-z&a5 \)/q\.<׹6p뉾&zw>='-biO_s"/p7hoQ.|;5דwYh*wԵ絴q\ȥAW}k]66)w^\1TU3Щʽ8p*UĿw68cAoǺ5|8΄Ż.#ꓟ_)8;lEYd"]k40%]ydE?欅.boj0km,!%h( t=(ƒK-ms7|â/ (ed2[Đӿ dUqYl}RPəs>zN`7ړ&+N )eqƁ9ŹYF1h}kMtW/?Ne,bo2/&m36'i y%n8*? MGIM ' Qf`'pT(|5 \Oz<܇'a[INs<h1Dy(jO>;tx{7 ث %;Oa;"(PTmC*g=WȻcɞ~l 8DZ^ES=ہE_$t:?gc-bPu@yV1 u 2Xxk5V-jô(u?W"gO'vE_R9_5 ow>MV(H o޾9Q%6e[ЄF5Q2}=/Y\AV~"_ˆ9._l"yn;u\OMA>G~jbr}QE Ze? `բ5-j~:-5hJwc[%Now:hҥ+zH;;K{BMgۡY?g:ξe k^uý?~M!L#i|#_lyN~7rbp_oXk߁q }gtwt%_x~O6 gGBxuq? }?a>׏l7;@w2cfOpx}!?돯3V|߽?{u~?<.~/o {}x uSp>El|}<'WrlBcc߿q|6=FF>Z-3ljqj Ƽd1ԋ̗oZ&]ojkh(*m%tD6+F:XFźu .m=V}?fNw=^S XF,pLꏚ 0`JNp?~ 쵍m\}d\a$w-1W>y55}]K9Oo'x}CZfƏow,!\糎?H/MOq6kwR͘gZ7§= ][5o7 >܎;ܻ}mM. g,=l?^Jo6g-Ʈe|l6ы>yis^qn O7Y+'c@w*e1>ysp>l;aQTc2Y/`\bg}Ν|n! 8n,to?:,_Kx? mŒrrDM>', <*k=>WFEw-ř!:B&wKZε{=ڴ2"\xq~- > p _߲﫷Lx[%/0Q1{bxEUs;N>N]ˆ^d.Na-X__ qvSGSV;9_`8Cw;uu^'MSM.Z:ɹ8ہ;.zf:pFDBwY|b4vdoq} mz[g)9dNx8WNN!Yd>Jy o*n ]ŵh^X4e??Z/WPf_x\3ws:X:C6Xא[Vw0Z? -Msv磉΁7¨ܯ<6^J+Y<\f^w_ا9y{r\*O8X\njpg n{,wb߫:zG"t-eGy7-?"Vv敹68]/C ir{ql30 ^q^^cC|V]Ǘ[$bŷ"oy>?-]d?ħ%d 8/Z|jSwEl>CkS\quy<t 467d^ǹ67wgz#o1T 6rBv{\`W,4w۴Oj+ޤ[۷ |"G^Ao|A>6U-z xÙrqrf1??Clk׹yl>ཱུXoXZm71r"ovB#hvulv[%/"{>3wX;*۵X~xޯ[{  oY3{Şo [X+浿zbomYÿK>k_Zة;}Ec~l1TLʓ.8z!_bMnX%1`dro,q۪ŷ /kBGk 3[Ku[er.Z6+[/ -2}vTJ35?bz t71<ׂ^k;e8c9qtubmbſag5ٯq >?fb= cرsmTc[g-W:I=!;6. =Ǭ՗JؒLQNt.\v ㇓L{kkwB~r{aq)R/7. y1%oe#2"=bR7p A[ұ߀Ε/ f&ؽ!uwl dßuQ64[> l q 2d2 wM3g~V&ئ7/Ÿk8 rd.ǮS_gk3v~77]w`}|O>m~;? 0303@x2?հN[lc[dN8wA jADZ9$@(7BBЦ%(r+  "ttdEJt:PFvOnBBIgv9?KA_|O=y/֖~7+u_ Écs7zϳϛljc ~iymO#/㵳V~;|}R}?֙qԝuǼ?-֯zVʚvK+{vHR G_?`q f4&R{8nK<eO:NCA鋥4I_dC@ &P_% ]}Yv"KHco'#ݸ}f:BTA"ϝDSSU,u5B?uDZU!h֣M۱UY6aoZ3/GV#JCJ8s8;FMfZ06~xZky܃VW0>5h`%uYou4bYRhcE2,SԼУK9#l#Wk~f=eʟ W0 tc.FcWkxbYʽ@2syv H933З&ln;#nDkeE$n_@g:`;qo~cΜθy޸d4/B쏘Zm} Ϸ5;7 q>]=4'$s4tH7\X Q:鼳l֥7!:Fcu):iǜc>QxX\ڂ.<۫X#Gl1I>4Sz<m;q厭\1 Xƍ=e1{o 빱J'R<^zZ!l22lc4Kqc7;w ;WM\˘NB =_lTq1XYv*:vAXv(oy Dh%s_WkfsHS?G`Qdmg s&47.?ȹsj+Ҝ4>?yraczi:OşqauzpVu7G3~ķ,Z E} ^c>ǬxicJs-e{b u;VwnW86BN]s3#?'pcT~5n/@:vS9|v]O경~HGz}Yw^53l^w'~b*Ox%nn>J/zvFTxnsci%1M4W7]ǐeC?t}Z5{^}ۦQ-&X`Ns :Ig51<3bΙys7ʿYQ{n(od%&NW%Zpƒ? !f-.gnIDziib,{5/y?ߊ'n+nupēZeQ~mBT(]<fԪd٭+ >D4 \@@`U`DD*5A %-b[iJm2L$M6mjiu:mfuHЖ'u(Qc ͮvYv,w{e^pùGɑ.ҮS0GSxK|yyW]wwg\-Іk}cm}>87ܫʠqc2npRIdW6;/w3O6d5s`Yi7;g>ZmbҜYh3oфuI8CRdIQ`&Yx ߖ CӋ}ӖvOr|S9߲jwXj_ nVhZk`-v EӼ߆Ñ ;VX+9 ܪ-J< U26gK\g}vlt#W #lozl憎ݭ#9}9{[| G}c+IN v|9~|ti/fN? wR ~ `j:Ѿ᎗WA9cr^ɴwjNh Y6M^ЮךG:r3=c{`;:v>o|h߅SV$yVc]֮NI`*x0˝%A;Α!B'` IDuJɏtDIHf^-4|2i?Y<ǖ "xUg򬚖زglYG[[s϶,s~ǖ U[R%˟[*H) hUNxkZ5HZ;pzsSZWAn鸔^BKK\BE,Hko {Sϫ ue5꾻EZ~y2rr;I8_e|ߵLNL&WYQ<$y)\2Dϝ|6ŧnq;5xk&U͹ۤ*UQ` bQ1$6|cKlXC XY$$ abM|ȎWEV|0ů.R8UUEflXsVdĨbE*bya@*EiHPEJD?bHHpU, SELƂy gE|UDSH!0,4#d{|KkI 9\zol s|*nyFlʜ@_,=@@%Rfq:S& DImŖFҲtHST\Ү'qnMaqYH?,Hі>stream Hj` `1`d` $XYM>3R"JB`'@kg`![Ovbit;, 3AT@Fj.xt;:E!qU0| @ endstream endobj 326 0 obj <>stream HW{lGvm_µ*iVeN;@a.Ms즻.);i-IHR$,BJ-/UPZj;֩"h*Z mʫ<DID(P78V@?HoAD4C zǺ᷿Ğ ۅRߏˈ?4.&#ZkӖ]O-r7s/>/цk!y;_11}}SCd<[߲/'L;_$)AdQ7Lښgݔ8&AJ$4SIJBf/F2~2=g@\Œ=}Q$L~Dm bo`ZH_2R'h)\J>}xhk :E=SCFmA/AziM1?4=Cߤt}0N. #5 K=A;h>czYm^GaYz^B);L;t _Gi}D_cF{Y >=z. B. n[R7N{NKoID>Rh̫O%nqޅ{xa{ /x D~Pg| ^qW?ǾdUK;nm{خ;?#2m䖛o}ED"Iϝ'Q UW3Yh"X.b5,1eQ'7eS"JcٖhV6t!hOf1YAc|6-Dk>N957&8xM@rEӍt[&cjAȈ!t.81xxڕQds),ŵX|53&Q$h<ܹD9/z9㥚RXkgVF3, NDO@r^t~Jv!vFǻh8t^~ݔļhإL$l;#.8{H=72M<,9͎{yw،wV.|U j:e",.S䵵  XAEj)t2Mbji Rkh3iuXcMlElϽif:;gsΞs[6R𬮮tHlXI-UVNwvcWmU;TǢǤ $O3mv;/6HGZ4xGǜZ{tjA <]4K*&%j[[^ܶXt[JwzR. "khO:Rg~i-V;OH#UZJgA3 {R<G[ )=K]o?uuۈ\#ŀ2ҁ,_vP,`Hv{i{Nӵ{}}@Q <(PZaWXFu?z,Rc9maYm+|;hˎ][[0句[7=6ۗPާG6,!yDyy:X-jG $L¥3iGȩs,bQ8"8)u=2e潴PcuG[s>:%0釒8ޘ<#_H u(@"H Ąx+-Cc'4DC-Π$XlBbx.<PPbv _4+{ WcDŏZIW-lz!KpbP٠~xFkO_h2AdqCCv6n u/^~ ٢qVU^_#DcRp?WʙMn=E %P:S9#hi =᧤-~EKᯖX@1%D/8'LPcgke 2fOg& kc,5S%ִ-آ|ҷKRѥ|)T35_ߟl^G֬y08#s:>W}j"x7GD-qc(&6OֱIkD~DE9#@b=3q̗@<]ueE0tgRyp׭K%µF޻7jT\ēray1⎱X\$C/V,cnO0s{ƒx"3I1"(նGg]݆ =pxE Z S;^%ȱ,߉t&% BfOzo+,; >޷,j 2-;w9 %E%h&?Dt1ז$N4̜ģ%"kqTHae q&J(=rY6mE>JMmX̹uNTjx1~"aɚmG _/d" N A`ʺ\_,0ƣIG˻_-,u$ܟĮ"P FdF8\RZ^^bK$oAkC8Op95Ԕb'TH?DF4sL=4t$!rͤ:ZfIo;NrxM3O5evcw(:C4j1oVkЁͼ`wCNmH|hfܛq/Y(E`Kӗ>_J]|gشį)'~u g{\)&B*JqV?`CPP $w @hcaCbeD)1QB=sᚕ/ pG & fY!g!!#\gzݵNP %Ql$u(k<5QL$z8$·"~oyJ 6Wl1j'T4)bj[! Jc^PiVT?ș'V\]ҤECVg@kCz].w˖Qo4&0.g+$X0wH” )'4R#_9Hٹ0{OTT'5\S{_??:>SVrC@әHyGk9U'x"56oߺսzݞC)RZs/wlB֖֮k3ڇ<lLzEǦm;t ]#ZƫyչR6Ѻ ,}W3Wt2|IƺN$ sT t=P!C(Rz-M&7E=p ?T 3V"'"7۰PX,2>ms8Ka ʟܗaOI^(cv#%7k9j(dijڂ├݅5!"C׊dX=gO&Q(>T"᫾o%x:F.5e`FYޛw{EiLp g6 6.Uɻnd4C`nf_WG`;?g2j(w򜑱EDFY1Y{ܟBmء w9zg Vt"nJJYY]UBKӞp#7<鹹OG/LC_1>;é\6EZQ-"&,͔[?8!0>Ѽ@G6EI74S8K^PS86( VU2ZG2Sm5BB IHk5d~=Wo0d*M&Pu{oR jc$3噴K_3̓GjC髀Zmm ^$+GΫAA$iayvPI)TI[HXռ6>gZrfHͭP:[|:/wEߩ%Ź1+XwP  ؓ\S ]H^b*Rv3UUꆆ>eUjrr&~){fId^aTA1XՄ K:?ptX"sȬ5vלw9g`#F_Qh.a9IEFiHqLTFͲ 뒨Jp}p~pkMy߲D! aKUy|e9.eV$ ٕsRc׊"[%l+޾Onimm$3]D%`ssߓ=2cLKi6g姗\Y<Ӂ&-p̂Ori XLM}fk@ 㬆 D01Y%R}W(aOyYtxeNZ!1x*lO(B[uU[aNK^xr0[1W0Xk\t8k\o5(7?0̢ǶMPqMB}N#>MӃ܉O[a=,٤GK5TqGm>]RyΎ"X#\SW&$h=([kTl ֊%:3\2UfΜi|hLTgg  X=~C]"6߃& 4,S'cʤj9߹:W;v9Z;bi( ]ѽpTjWO +5BA;3B2H"}pg6]|b] TH|=;_@u-,{RUeHZt;5<7l\ƴ)1kH,PL34M$0Ǚf63Nƨ{3>_b#n{B,UjpHw#S/7%Y)8s\D@gcu?GAv\l.횙q| ?d-DJ{-uxȶmȌ1lJ^zL#LY=a_PӅ;u{[tD_zz@d4h"|&(෫QFS^R069'5!-N%ka)¼$_HG6h)/e01 O{zO #SNw(fU:lٍ.')+Wē{_,%t^?ۏ u:ijYv}z k_a3)1*ɳ{D5Ld)'Gϵ9)) KrϝKDDdDCL ."ֻI uX,QU7!"beI^HPV`=ENȆ1.@~:+&g0q3vm)V<3o2S POZ/O@9Qh)-*t߫ #7beXZjCر񀈨.)VT45)OQfۖpZ"ɓ(u&;C)~̢ٷGb]CL2p>A}~~#V}8 x%9)OhIk |  G eB ewzQC9..CwoؖSHj-r%d |g %ଯx}?G073nr:t UnW&XGE"B XQUsbLNTZkWp )\ u(ki`-]ëA<2V{#nf7S龦eGҮʡtM{jT:+0Uv0U*"|fLf#´'x"(=,j/$36*"!&`TVl))c>cMD{T=ݙLufYb V @"hb`qX"uFW pwD!BC|Rߒ7݃2iFC 7_CS)kD_B4ڠzUتϣR! d}_!C/Tz~Bn'w@" $!@!EEEaQ*UYzUmuUcwݮv۱YhYIŠNGN\ytӷzhml?]q0!7E7tVws-MFC{+0Tٖh,9!􀱣m+Z#I&p$ OG(fż̤SCLA_Ðd'*y&W II$%z$? "epPj *e9W[Z1g7]`95!uNv*W'!sm4Sۢi0{?z;FpV@Dm^fz!I#Y TRjZӹ?Jmo/`;|eXw4kѪۆ0D-39uĐ_pR&I:Az/VT*w4]H-Z+f׉i3~i[~EeigPxCt" S-ddlb)],'PB̦nW[ή(WD2̺5Fj"$;'#"c׶TomXkSf&Y[uu1ARXAUWg1Y} ["vo3I, EK,6-ogy&Qޠbx"CB' 9T{NFށɁFF!^q%DMPH NZ28]@< 7ٷH Зg@@KGIIaF%4SUAu&e!܄efoo_vÝ.,yKsc54ዹQaN(E< V-ܥ#7H@!pM iuڂv3nvݢ윚p٪{{ 62,F*BHrM-BG`̚f8~Uw|%EvxT]BQ45^a)5aN'OAVmR* IZ&ہo3B_ &o՜oP )<7V-,h4^˟L`ꮭ@@>Ƚj`7i4pߐ㏇Z>@6@Ps |[xd:DͦRe3Cw- .&6244E 'Roda'q*|ײS$>ڡQ ͞0Pz1%HOڟ!꺷KpMȗ7>3C te3ֵ l$pô_uj9 *(Oa ܦFMsꦖ8R)|poJiݒnh#gFrBYiFD*Gx"- BdR.a9Oi`*ʤJ[3҅ u3KJt Dg6uiiVuR x3֐="臰^['B ag$=#RD eIt8cn?,Pu@YGdLɥD#GґF)I٧$7.>֛VSƐsPn빳m:$),ʻ1OV%<>^F q<#1|ImȬg $鰈*jTnFիe6g zB4g.ZO#^Mߨ8k4/Ob$i'^a[=ndji޻=r珁6*~D~2oFЬZFؗE '?lJ}𴹵&19+BNw{ͻzڏޚե+kߑ^1B]/ņ_$p!vFp9K*S=7nNz"9 )UsaIJ;(qؼhRRq&' OM~{T [_L*S̪m3"] <^es>BvPQrT^/% {R J\3]+*\;JfI xWR\å=jʞ$`bk*gsE8|BffxC]]x#GZTn {Ր#Ȥ`!(-4#l赡8dKI,D01Shz#D%M0Dg(.vn[5+aPzΉWBi^F&n\3%U 3Ӝ*P[5r[.Ç U\Q"b㰯{gUkQ~f?bjGpRtV,JiXU3ԳxM)?0Q1K'o{>5pU|ކak37Ǒ<tMO}:423'͵kRJ)11&io!0iGlB¡F 'qfѭ-V1#HpXq%wW-")Xu(a'.r8dwXUy#}౔{ /hCo'GX戈ڹ6oyDUw` 7򀈪 YkQm&WU13Vst` 1$8nZ6邫`{^s!{MR1]oT-L}}nkԀ5E)\Gj-? K>'fW-Ҷ*lMSiYOO٧+t:N`~.geӪ F6`9)O;PvB /ȴ?Zav#je2s'JrQD&Tr*-Ufg'&"BlEyy@Qͪ+!?HΫvW~Z9Yat@b{% {a}%R\(v;Hdx5os=_v, @o?9Mwbɸ$~KZ~]s4;ҵ2WՏu1 ͙l"_$27uWoCq jU  ګקi-S~К8qZO4X 8BCY,K~E'5gRJiQ[1*9HnNt־16F%!:ԊMF܍QQi--R[8] /Bfl ۃ_Lf,UРw$D7]0}/b%{(>!cKR"M4nT"-D91?Aʏnj\-Igmv:#֮5r}#_Ywڳ7%~1v/US4˩<1anm-kK,.Xm=U^$c[]fg&K{IhK:3<-Ģ[q~E.6H֨5N\[;: Y^Eʪú=nd(hUUgn@#q[)'}BHHͩ~&-܅ZkJ0L?èYOV܇!vM#W%tr 5(D(&n5]?N@N AZ~kBAW̆|q-߉}DM;Pp󐚻'hGTH ( 7h@X$fTzH?5rn)Aބ;H= ?4IS_t*3^p1zUH:*#WP%^TpvsG[ ׼{y{X߈=4Z9%I#pIoQ(;p h`GϢV-r ٸ샼%w' pg xO+K$Eď=產\%K%  cE|h/V`['+'> iq&qπuz\#IjhKGᾥ/ RП*yonj !^b.PB,%-.Z˨]mVC!Zb5՘bW6plu"[2uV:J2" -=u?r7܉[^un>U:c/ᦼbsݖ7]g5S1vgxU Vo%TᅴZsPm:nkﺇK-RVzîW,wQW*|kufT[(4o;iq7oxpϘ.֏۾󡖒V#vyܰ%êžDQO -g;V;f5= aѓ34̔Mƫ oC7ԙ}!k6>oQ9]p1 _aa4 }NA]{* ~|‘|z-jQ7M?w<YHΥ\Zn 6fX+[+ 8vY}ۙw/*ܯ [ 's IȖ7m;Y-rlM]];8wb cqopCzX?|8kfK2azM5/5CgyK!&C!ɜ꩚#o?kn֭_&fkr=Iz{mt9ɏt{ŠO{`,'zik:'rP>b7 6$ʻ{[ I.B _'OI|LNc^;s(?ڕECΆf%lhkyѦ\>|d.Un|JUvw~hXo5z=y|+sQCdj^u>\p>ƱG;?x6OU:Cs5Hl{"fQۑ;ڗc&IO]B/&[:vl{ ;fٙOwSo֛Io);σjp<#H ~PAl-^c(|m_ 2zI+IQI0zC7Ղ|ZKG{XOo8zodXe=w13UdUIu.'#u<<~c꙱eJ o7%گQncT}!~td.}IH^]|cUԇCUIG[$vCWz|n~z >xήx)]7c":ی)#|k=IG(.ooϲўS*qB(~!=!&'У;V}^}/بzߟ%ƣ+:JuAb>[2c|VKc7.:>XWخ?*:ԋΒ 'YTY_k\ 1܉xV'$KwzsՕus.~7N/^CfC|[(6c|G\]ZCW O{#gx|o2ۉ27|833w%0sC?j*x{g55G %c!g.59~ quI{}V+Gj->ION0n`Y~ҙ9.3}v'w>q}'Y;bg=#'7k-|>{wޫ`/Qc|j,q[e=?GV=ɷV| V*Ljm쟫>{*jx;Q;k?gZ_->P\Op/ɗB2e=-Ȑ ->{-k6%ZsN(k1&іۑE=zlDF+2c|Od. >oث_YN\q\WD""+""BDTDD\u(UcԨUUUcc1kfn4:y9]#/T;8ybgچCǑj|YKSh9 \/BkEN]Dq{#ho1KzK #ߩ 2Ok0Ze|29qC!# &|s}#=F|F=Z|<\0=6!綝5B܊=ŗdzȿLՉ̅ӑqp9-[:j8o.2U:5} Nm\r"Ԝ;#ՖbS;R%މU/gQ?q lg_8f̱N z,{~$2=>uK%~ F3(ө/QS.yXK:98O$M~88fRaQl }gCƻ+ž3%48{ľ^MdZm9ndp^pkdLWo1ֈ/-ݺ^[\dz㫬#z#2ǚaƵuu9Vٲ:83c*2$r0"ߩq"{ L9yd<2ٟGix+tzj p W>Zet~)>uUhGBuqP2Ny{Vdŏ!'y =2f5.Fw9Ϙs)wGvŢt|WIzZ< 8t>p}ຑ=h#߯mN?kq_E?ץ/g(>[܎CE9|.V0t:2>װ&6`X! qst6;O/ڻk6udn7N{yu/819w-boݵJOWo-v[pSs\_C[_[P(BvǑ1ykoyNG=, %ٟu}ٛa =4ELꃿ#? /eoZ{inBSc΍LSc62`r׉|_[<ݯD?S\v;.vd߷_2]n?z;2,FrdMUgMc3++7Ğ6G'kyy!=1? W?w͚,0-t^f ΑS܍p>т̃e>~,xBkq*EVEgݛ鍣O3v+}sdd8cza}v#6fՊ} Cϰ#WZ?+8wۗ3j*o@׈[UzBU`,< r3טi|GeVO1 ;֊#=pu9UcT~k+Wߥwg|˙C`XνEGǩɽr0S\NDnjڋ9ۍgͺ]Os:|z~m蹯 g[ {wj\[|~+>#8#(E)QDQJDPDDDDDMh-hѢ56k5F[k1/0q>XL>>'~8sqтvMl?]"_.raKb_rΩZ&񪱯xfygʌu}qY꩚+U= sW5ߨEʸz=dr^k 4~et[{D6rZumV_5U }ɭ7w??wxKoo4Dzio| ~3˼ Cpzμ&%{9̢!y\st%𚗘,!lȶp {?Vk?W-za{ބE-ZľouyŢWSv,?Xx|L-u'2nآw(_q[9}+;þ˰9I9x4NEQ7ʨM+~W,AA1aڷ?܅ (6ג=߲w.g%dȠ*j KKUhJ=t%֒?KQao-jg{`=6=n"c+^Os|oyF|hf/}Gvޢ>SW~6+mbg7}˾ERG',[|: ^E{O{MPeoc.k&r9TGFuZ{5552{>~RO<~&uBch^&g(ֽD*2|;zC~e ]t{\mzmXG8¢&*Ax TCAK9kb|A_nok*s\u9p?0p^ qXvޢn[p~:[S]9O⣟,l݋5io,zE^ {k,v^{ }T|yGЬw|뜹?9.wOoPAE=k5cIOC}zǿ?Qnѳfh` :^K7-dGZŴbz_3x-rLǢq`z3t~=8cz=`XHC hbCe_K@עu_u{Ən]3ADxyM;ѻ W\~7y(pe[93d{l%Y_^D#}f^I>cu쥺pU|+%?WmXL8KAd@~<,YZu_i;5/ZM=Uq֖ioYĽB:_A/ -:~ |? nGgc垽go7y]}џnk+ o|~ǯi=|^KL /~=9vXu\i~O񣿝&S_Nqx"&wJ=X2aW)_tv\3+zǯmy=KzLb+ ?ֿ̹w>7?1a/amx!Rbo"wاxsib[{.#b˰-꒮E ؽ\֫A={oWAi>/>+y>5lC9iE CND-xh!ܵ|u\sG06qAgn"󚳈iq__9G%hJ8%t*}\^cG4dfvgQ8ƁdςE=a4ƓO ~8͌E98lOE]"ҽ7tGyt m!TCު` ~ʢyKn6籠o3= :?g>}\zkep/c#wǵ6cˢ'Q򷳟έX]S!z[9B5ꍷAޏynCq?8p9Sx=dt=VzN=D++ӾChi UMm-xZ98tΏi^7[ae|:pD_N-b˨şLzx8ܤbI՛1[g}-']kߧ͸$I$Id$Ilm1RTTcayV^w{ ϊo%胏bH6v{9RjT:~:jgr ]@h>soM_F;ÁHg9~#]-q^s MnG{VxVi4gg](6mrOwZ ܱҝ578Ⱦm)Cصd:t*gk3<ų6mTc usz -kKeaqo>m^O7Dckk[?SE7۶![kANA:cpk 3˽^¾l=.:"ٕpǾk'~|{U]/va_0N 7C5yg;.vdxf٫#g \mmSȎelOf?22<7ߘ DeSYg&c ۱=] ܂V}q^3F#΃^Ʋuce<>k_͝ڲ;q;8cq);&Vwqo9/U#s[7i_k_cy|׺s v[Ocmy<`x<,s9g%auo4R4XlGGFñ~?Iyޱu@$[#jVϪ>biÞߊTh4TN"O $TxVzϔޣϑ}c;Qll2o8VUm0ܛIǨ?h3Dzyz4նdq>SG-[Bo4W_f?Jϙ -$3zf;; -0Cd#zԬ1 &+bk{e kG=tR_5<>aĝ!tBZawϴs֖#v:7~o;-t!)qmx=]H= ;?_ٳl;c chU5L8D&-ymEzsX@d=aݾ>k =[4:"uKm?upS>6~йǽQf&KF0r\d]ϊn#RQI)bVt;D? x qg>OJ xVeqF^W7 Dˑ޴m9\{񯑳]뉔gZ={4w兩ڳ4q1{%6D1bP%o321+w|^?D=y|kg`ԅ x:7_R~d+`>wk|p/zUWRÒ3wЌ,_8}9H?vmt2c ?kjegdZWvv3R>A\J.jʾj agOqN7x> F'/zqlKֱ0ܻ_Dwq:I7w_#rg.^o{/{Ishi`&k`a0y~I~ [f ~5]p=c~OH;@}X s<(u['dUL8~,GO]͵,F<dZ}̱k#cNgs.wϚ ۇ&# LރH~?|)v9&NEDĈ!F#FDDĊ"FDDUUEDDcTDDDEEETeUժSVꬳoZjsj{uuj|;o}x{~O'Ƕy#R-v|)V7^L6: 9寙m,S;NvoyN_|sݪbC1u+]{Zgv|%m3~-5_gz2~c;~_U_S_zk?f{=CW[@|tOο5!+ 9Xf@џJ8Ch#_K5 `qpu7ż`1f֪lpCøix0W_:N=YULХ dY6<ǝ쵅X3lsw 1z] 㔻#x-Bc~'ufv"}$#pܣ7kb6vD F_^[ 6m3ne\]2یG):L(3 u5_M1K3WoٲU1 WӑbDwׄ9tXyndaEoRKغ= ||~oe{2]rd?xSw6OGEUzoTzv\.q>:4>㼂;jF&Ç p6Q3'2:As${tW'# xjz*^G;j>_GUtv q_XP7͕-N"cزl5[n]+.Y!əmt/R?l=M{>py:R Fa;md :OKbi&g0̢ W_`[p3GAEcyy3f#(3}jlg:Igyݓ3~#GJ[j m&g3ѵEn{bu.kq~uG}G`8ޏ8tًgE΄?7}ܮY\uj ]6 KݬW[dxn`]bԛOD' ܂νv R(N}9g;.ƻNv2[wMp"ՐtA!)f.b[ށP*c{?dH9{9elאO#}CN :]w:W?y;m9{4)=<^x6k W3 V1BN+Wrr_.gp)GܹوrZwӣ 7˗yw|>G^t9@upE{/cԃ߅lUoT^<_]pW8 ~{Cd}H5}Ylx숻3;w ;~8{A3"{mhrY}Ey ^zs6;+>Cs5HȫP|W{H=ypl?þH^Sy|v8{1 L4\VVCi=kB)қ侧 /rf7e_xSWދhW9T.|A(ƾkzzXOn>PR(;m#p 0 |͡qr ӌd8kܟ$q/YȰ޹q#Wt=4ُ#OזaϰB5!ù 1.7r#o1֞l86J;o2_gk֪^NGlni1R!߽`.~^@'ŵ.?GH9qtgy&c u~,AG Fʳ G"y&6}-?AGנC,1~a{:dmo^JFBeoX\)r9>9%xU_f_k$l*qDk`q|5SgΎ;z~Y=ޜegUG%^ٱ _{w ~_OPYG;WnV,)~`o׽9i,r,_ÎEmX&~TvA"36D&[]SYc2=ߛd1<6+5m75%eIn uu %7㳍]kPl4,j˄P-aӱw |GI]sכmܷ\6H{vz6 ov{hN#`vpl {Uzl8:Gn']t(^/鿏no1EQ2ww0PVc]4n)4phkHn&7- |2}>UjX"gϢVx*G{8&XO箰w6S~v f|V:3Cݢ9FoͱwnQ 95OFt}o7}t-¸ΝX"#Э߱H8Wݑ'};1԰gWw, G{a=zG6^w=&8k/^+? qQ3 `Qg]ocYo#{[F*vѻg>cEёEϣDR?q Iw^Xc.Bw&k1Xo`j*#_Aն0'/9W8A+)NGDOamX5,r Tf(λM=-v"np^sG߳uI?Z=ĮE Q GyQ98WZWSB+ʍJ؂ 2_OvrEV={wuOoV_=l["-EMSYU<şV׀o,ń|6a}?cΨ*̶-ʾEi%3=g 7Y2/'}Z5gh lj7Yg/}V?mRm؁vbs|Bm|.܄Jw|E߭{ugmQgb+15>_Y.zNҚ܃*Ϫgܲsz( 7wߢ.(wjSQFowNWS#}eZ0])+`U;TVah#cș2eAMUwcD/;g'}soЭmu,bZzJ4Le 9G47ܦmLT+}_cX+ śaqw"7߁x"W+ct.YHʇ+'d8W{eκ^_G"4nשE.)@7,#8gÿx4n%xģ(oW'qz6G<|{?z)>ß?`~/{_'An7qkW|F-\wd~E׏`7]cOqezx[ס;oYxܳ2>߷6;+]>7\5nf}~\ϟ2%쯂e٢n}l9a]2ԋ+lf+g:FW8z׬_+Йk ߢŻT5޶:Kb_+UYGl{OWIgw*4Ƅ[|*]6L,Uo'.d(&-zQCvNdc{K9e^NC`Xqޡ 4+hOǢ-Ω=kzJ&ȫZE\ܠjqC>_n`řz33~_Akud-oEl-g<G9)eW ģj+MPK7;'Ņϊ.}w"twoW>֛E"ry~E6~GWX>]ߞ?XĶzuk ,%8N{j^_wUv3IŹji!}%{, G%>6a$ܗ,Q1 CaiciͅEwm>QH*Oքkݏ-vcP~j?G|KqXX3%[ ~/牥dN:_L4 ͪCp=Q>+ZĠ7l1/W-֕[YJZs訾_>;ꈪQUU*"""JFXZ{\c/뵗{Wz;'{}~/u?'9՛IFl^Mom|od=`!15>4Ew8Zi|Y 51lG qWϖcR1"̮-˖E92癚Eȱ[#[Dwd{%[xk>$Va$zUXd\m[kUC]Kc9,"%zGi_a}5[8i^e?"T۫kt6}n⳴wVYM­ۂEm 5]_(el5wT-|H*[[9gr[9P>/N- &/{ SBSv j\OG)t_c!WEƖ;}`bK*v} mΗ7_$,XϷ={md[ uj~řNUxlf`:x/A%ڌʱ~}yQ܍,ZfM[;ِ3t" Z"%XzI.=޼C殅.ؽĢVdG_O"õE~'~s^-]d]a>z ؗbHsWx >xnf]~yړ"Jto`;aoNmuz ?qv>~f_"4amlcWqf@S5ʭ=k}gпf{7`<1ٟςc0I"^|W=CW<=řE {~Y3O?B{RLU-rX,r=s?~>X=2P+rL3VW-U>9|n9S\S\,rY,ڹE޾@ΎUAka 5>Y@G:9 >OȲ:I|Nٿ1zo睎c8SݗK::MZdRP?pW2K1Į4U'\?A=h>Cs"qsQKo1ށ|ohыnCzȧ~ÄԢ0js|KU)H6իT#zw|ĢeK8g{y:sڵn߃%6P8- DSz+&ϓMw3hJhl?@E/]N|77ayf[OidL^$Līn+Ϥ^y\w?Jx. MK&%J>okr^&#Oc7z@[?k4<9¢WdkkhE=w DW!r'K ;Z?/U>Jft[>L9`"t{d sߛë "6C~v"uj )/~Zs±h:3lWA6?c?o>vN'`Z#$P=н]BUSC7s Vt|Zex̡{u.{cSsZ?}h S*iO0.b6ka^sYo }?pw\xkW~n|e*Xo"GySN˦|q-~֢GM yn`c ɭ|ѿgЖoe;).+P~8v,j:A-nrD[>=n G3.o" oU]?=e>_,mdorEE(#ȡ=7-盌X;߃O Ӷn;N91Nب9&4u`,Y Yȍ[Kؠ u]@Co#ާ#+}/b]dw;בUG}M6*Z ghp'zM\ԴDwW>'؂6vOI{ l'Y{2b*$jz&-It*kOW>xAN3L:Kfb譺mܴI\E}+BEٖE|0jE8U*Q5/6X+Ox;FEߨ[xf X\Em+Т*cS (\Ev}Ffq\"(Q"%J)(RDh""hME(hFkcc.Zke^^Zk?`s>_tv/>99yTü?eچECud1:~Eo 05]L:3-ŝUM[ܱ^gUmGif%Z*gx%בJ5*%|o[c/WuEi}ŚtI:꽠Y X_vuk' J:iWq_o-5RO,zik;i=*߾O~}&C| -ޫ8x7߶6lY;؇qpst.c~]/.XBފ-#0l,5{j{l,6ݰ~آobm[ԗc(As :#_~o\Y)ILsU+ED|?!t=qÉ{w/~n K|qo}.y{#|;Ww'|6c]~4~2eZ|EޙE-x>4oȉ=t9e8]upNώǶŻ~"U|M4Zraۺ'$O}ТoXVgMZѢOXeݺy`G4׻VI+.칼r+KoE}!c.-Ej[>ڮEY6MFoӺE^o˕-IZX/mng;_LetV,ܶEn}âRn(.P춮CfOPwG?8~k ^۸ލ/^'^yOuw㣍kZW?܍kh[o$6鍸?Z5)>v/?aa9ܻzOWEtXێݴ;b?c!еssg~zM=qyk_&\퍈k/qk#}2>$A>{7 1o}0C[N ׌xgb= wnp.q~Kr2?S sE]^y= Ѝ|ki\_"%8Q.+#<ƞ-Fik 6}=k9&𘩷`Y+_G=7fgukA=<w{3ssz=t's:Ip;_na4ՇL'Zɞo I=a,|e֫y,)2~,C+9vo_D>CF2lf%ϼW|,"兴^K^Hz/'+&-rb LG&goqքSStXdO)~*O*fѓ]uZa E#K909Cz*To|PK>gW;eZ֫^Br,'䯺}Vn4Mg_v*SF-%YdLvym{uz^Ѣ|-B/=|1a/W]L6Aˢ^HFy>::݀w6zZ"&?@07V΍!mVoehS \l< d**>kF< YM.moYeֺi>? Mp=O%fE~Ħ5tB4Y"c U0cLlTl+3;@ysӢ?!\/a3覞L9^f(W!Cq=z]Trֱg˽Z|;-5)~ p!:31ⴏMCmܴ1C>8 |,i~x-Dlj.=%x-zȧ.B˼omXC7`8߁,@u~^fgr[=[} 6ל'&x7*wԯZioѫ -E=~W6 g*`cײ ts~?g}֊O^/d854c+.#{a4+TyK&s6G z7\:[蓝{ρSZbR#s׳?B/}&~^2ًZ533NU3MAoE=Ӗ5x CS纈|XN>s5z#?C&)B@gׁ: D&1FC{ɵUdz*xf {|- qF{ypv;x Zy :mq|ۆy9^T[xd~by"!g;_s\:DXȳ:mYaNJ5|m*=coϓ{s5s0ɺ{ީ9LK߽ wOȷtd޼@gPόc"{{Ӭ!~86;^ulgv{(y:7g萾Ⱦ:ȹHxo 9gGf*F&4|G?>H_]XVy|MXgãΞS>_bW4 ͫ&z+x0ĿdE48T~:|do>M;5C6~SSMZ"nq/.pe{?}Q'yF)#O479xи$rvEJ)G &{!#oGo2r&q\wduPdnڎ-dzZY2Ϲ:؋n}Etn;s>a;C-_8\O\ۑ5Ͻ{~lF>>]plQz#{!s}{eÑ}S _½92y6Mk_"{o٩|_a{5Cz-ϡkC!bo'}JgkB }ѫᳯ8Fֻ'${։c G"g!>CdǮ^pq^ +߻ ~=OT"C?GuzϥwX6|\S E\/y^5St3אa>k0N#~-h]=O:5:Gϲd\1F"gX9 f8[;VgO7`?ǡY@<<Ŗt4y[odm!#YY w^YN;'ہL^\öYlG{-D-S L7g;c >umWGvL:_4:W1pkc|=8~"sWzsU񚽫^-x/xq8'يm{oمnع샽<`q^|Ŗsi|"G~,S% re<^ǿ{20qގ`h2/fֽpY}>r\^(ϰ R} 6wFd KqyYBz~/-b+c4|9_Mg38<.y'8Env}s66IU Ug}N'bWSdu/~X.Fd}#z. #׵K7"4˞w K5}`ݎ]%fNZzbdO{SƷ5:7i+OYu|ǻޭ:מgVُYx,vK8qwjd?#*rA{Al6Xs{`y®Kb k]YbMb]5?']ڑwkquApOGgOw 9vd_&[ ϟ#_۹d-@/zq?I28>E ψd ?[tSv=2_j~郾yy&s?qZDh֢EhZDk-"BkhE!""bĸ11ik]>}هϻe>_U;c8ԩS?U95[KMEzGJdO į_6O JZ^>L1k!̚O+u ;d;ϫWg;aR۲xBA} z5mG[Oûln3/[o%>mS_oػOߥwew\ 4׹k[O1Z.2tĹzd,;qOn[ S7ᯱjYXXͦşQoQQҹf@>#C转8o,j%z`=ŭf&w=`qeUr]Nz*𯀿εb|CV[wTiO3Yt 9s1EګpM;5hQ#{ 9hQe6r^ܮ#0#-dy#os^m/Vb *YBm"OM/!26Ӽl,'+5- Њ`51|/ހ@Cx\YԣE z1 /#=sڢ^-6f}٢&NQmwG1w=RM/0g=g~nG9 =?i9[,ޢʋ-G,$_,rS?_I1]x2Hd?$E|40Z7_縆%O} :wiuz$G8+T[-2{>"cYo߷~}`=}~lyYȻd{ 2>?sxa >Hdӆ;,f=Oq_Y)mS;839]_skDK0~nỆZw~Ǿ+"#ׂ61CE䎠+;{dLi;$\f%Keat&/ 3w>C +d kmk=ǿ vM3ծ|u=CxU٧Z5=:!SBL.C[g.Mq&::_2tnF +.!<16SΟYԑ !_E|?;+_S"W,꜎{YoM5l_t.Ew߫7`â~9NosOO=Z8``NO=~H?7$k<-;=Q~wuKj{X?zxxouEz-Gkkp?7 E=o< -UhCJ}užo6Չ&/Oo7h<~M염ƜK?M|(cK ;=k_Oculޟ%]{EtؼpjsU*7U>wG-R֡bbgqFR6.:V-r[ebԴzj+Y<ͺYߴw޾YQĮc.<}0Cv jjGҶml#xL֑'GɆh][Ԇ&rxYK]Ryw zO)CIzz3)^׼f9[5-Mh{ujO359EdaK.;vN~=|d= E/y2E-c@`ϧ9L2'383띘\^8`5xLYg ^aqwv`~;xϠZ >Tڷx+,㲏K<~O6ѫ?y!tW`1w;'mx5H#/9K0 #t_5`:F 6h#qg[<}>:/UsG6;nz}$oS>w,r\x"wX3 0iW"g]U_ F`Ѕ`l9!v '#zݟ#@8n1c1"""""DDDDDDEDDEU*֪"ZU9UYQ:9׹n}f8_~Y֐zSn[#U,`iXʽ=hfY%r_83K9~9EѻLmqt湥ު70s}/NjuA־5T.vydKw|])[ f͛u:!՝G(:^z}zI$ݱ4>-6{K}em_=տm/cN[as9`_yܼ~Y[E^zco=%T7e:1[Ks|UXfhZw^AoNY~fr8GVw1tZD4zľjkq- --,-t:z NZ>^ۜ YA%x/[0qks'Sy| 9ny ՂM?I-ƒ^;5؛@ _s8Lڝ0xqcyL _9 c gFq("h2ЖѣyH(C ?h*zNR_-[e }1l/X 2c1Br>2fde{,V"?ׂw5`e),[9$ݱcCV%KݟZ,B͠W uzrX΂NʍgјӪARd-`3;U'(3>h$KkՀ`RWTmB PNG\)΋ec"/Ո\u GNd)wu沼7i~M<J%-tԧdyt>5WR;(SCjF2cn_xчg]3^7g;EW=Hzi8 B9by^)u׫G?; X^jR61R/0߹cm~&kORޫMFJ9Nq7GןRr;Q a[?-39DbVDZ K;hl=F񕝪 qo/zĻ@9s[~RLUB+ҿ-,01?܁w_-'^8Qs/Y?zlKoځtව:1~x#q{oq^s<>/?u}_y^$G=nہCMpvٻ߁X/?_yOY!nkbkpA=[7{E;vpoݟ^s|P ap]@ޗ~:Q|@$hnp:ϻ-b=`:{/ﴛecW뀸Ьetu}\]^1 .vy]2㴁o~Wg]} -rܯg{{{..I_[~ w΂wix!Xʳmt}'k9q % ۶fB3zp'o1¾]GYŧsw+oX8܁ ]Cet[t3'|tޤ3yg6|һ/}b[̗z`v]#*a{۝ߚE?ݿz{yg]f Ynz([1;ۖz ճ(p]O]v s-`7c(4%6kU!~c}ω1 o*x[݂~%hGQ󘽴t ow?w xے>Xѣ ꜗ8+džVO5; aO𙶔ӕL8X4sv&< MBo8Xf?nR= 1W4n _B-C8cOҧM[}6rqd[/|u cG|l"+?Ǚb<9y|ٲk4B*إp_fC*5uaB5ĥhcκ|!,2%5|] :_cG׊55r]dþpΟrXoC &!\+!Uւ/͸ΔAΗ@u+*4-p/zzqOUKlx}^F߽>XgOq^|?וׄI1ԇ/Uϻ!#6]WvvM}e]ܚANf:D႓& 0&P 2F13I22YfpLtz}k'wϽ#IPy`kE>!M>]U`Uby!nr *0 {ho=LLJEi/[Qִoj-ZURUl!w,8>&yF{vy[tc\~#fmOKC8Wl j_)i}E(ƅq y9IY% z֕~8x,⟇pC>_rĈ 0 *muῼyUfzkZk ڭo%ko%@Aw :ͺ7v<+k~^٠|xmf0y=h\5ӞL_svioc:¨f04s҇|o|w鵯Zݧ5oETXh_eWChnk Ok [~VZ[h=lucQեֆN* =ۥOƍEaìas=|P04gkͺ=GSn~zUchm`l,jsEQc>ěƐ9aV+i`c~z?b76eh_(MM{M`} O@n_3sOd=h _d~6C>}?{`{^_3G}p_ޕ+Fx]*==>8_֥ZB:e=Ӛ>u*́$f}ȇ2՞X+~ɽ`7wT~oyKNBy W9gj~C{W}j 0{$8c4cFBW尭.з]_BV~9} z<}zO.Nķi=ך, rkG?^69cUo9x7NfvG}.IơG~w/xi~8YV@w _31Yrhl M7j^j\z؛v[k m|,6ف;ߊOǚZw}68C 1t;-gݓ>rBb4 Uǰ&٢s>C=][p'O}yr ч\ț5? mcW@r &C=o|Yzՠ4Яs/1ޅŗ$?mAw>ٛ!H,Jįi]\d L^롃vT%>o?>y0mc*v|S >!{?A .4U0 @9FwZѓclkeWRlct;uys;`ض,!^gR}aև!x%'z9'w,ztoc>+"<*uzGC4F,c9?rpzmJ}HuaT69#ڒ5S9[ҕ]S_L@Ena?frm\v랓̢ Ƕ8F9ܝاJoc_NEIi'cSQ|ٱqnGzvϹwPšS䘌XG1+>~,oL;=ez8y*Q1!*wsiF-N'雼*ŋK2q'c%c/WPh^G4\e*OЁ"]r&V$*>Ow.%{)JR񽸝;AW3}qFINx|e=͸_u<_Ln7SJZX<^Iɳe{kk$(3c̝?G32:c9?zMT|m|uZEӂ[N7&LU&1L#ux:i@aZmZNF+G^ڞ-S,7!v.'tu<2>6N~Ag`}V[fGhuh|x&d8!-\;I`^籑"]!!6`w,nx-4j"^xnvg- ? :w~y\uqT+ve= ;x}Ga?z{L'N t/M>dx_2vN!W3Ab͢{iy7=ͭWbߋs:crm!>_-iONϚ Ilh^ kQ%HG$_X/SY!_te2?.9I@јooI&m $!IMZBKBKMbh))Z(EBSuaԌ 2 ȅ3㨣֡ʅߞl}滙O}vg<R-NKDjdUɇ[3=Zi-)l(u-roGqQ~|*)j4sh9w\4 eU<׈毓xG].%dHDaD"5`RGd[]j.y(7u Y2px,ONG(>*A{)K$x ?RwM0gғg%uY8_}"nE'Δ0L+Džj Npn+Kp̿ ATx8M5¢c/[mG8[Y*G >XxL'0A$̲Ez[loIl5YK jہ\Tx(/󴲌&B8O8OڥrT8,<*Q^utUپuhagZfmaXֶ۠ju}Vae!m)NGyʳZﰝtDV[_ݩjnmW :a֦J[cz{tVe u]_ |̏:M^)[ߞ ^vSMӛDV(U5آe7[ƚ< 7H21ب߀M=ko6YZ?!IFAjSf׭6 Z]qRP\\;wMz" rv(F{P\_gb--D5'U]AXWƿ,뚫P?u5)= AΰxsuUlA6L)[:O,oOv,u5U~?c-m6Hӈ?ܴ2}xY7[1{!{j|o}w׵]u|O]FΥ[<4,m;jCߵ$þsQ з 5>X}WCuAo͕#2"ML轙C3?慾qnf~Y}|uو* ݞ.}eiזkJ[Ҍw񎙡36zf {F[Ji76yòujAW. ʏ{E3"//>u\6u-7N)55н:/- }fbJ*xjQ&)AJb#+Wx Ʃ W(eǕNFQ^5={jǪٝ^c\ڸJur AlpeS)&HPlF+騑붝kt5%pO{Is/yyw8ԑ{E$9Դ ѯB5*/T9x? י0 "[LOS-C[ endstream endobj 327 0 obj <>stream HyTGǫQ( chx"x8 hD4:*]⵺Q7xgg|Dx`2lhO꽪_}z՟cY!)qxǦubZf**;-eLXivN֦cjAO e?;)?0SR'*06)_SwϾX|+#22-y!WMz.\Rnmdb@_g!gy$&~YS:FO`_ȞeqaL*/aUp1)IG‡a>|+|cy/PI0Tx03o܀1,juk՞+i <1eJ^[Y.|Vm,4…v.tb8a( ҄BP Qpb31ILK=)RJE itGIIܿwHyr,rl;r]#'IQEU"(JRSjCTpUSۨSUTuVש;S"ElІhUjmvP;.ieڃڑu"c#V>6463463m!0cظ莎٦*}^ߤo?Տznow{þؼcч+B+*"+VTtW1bەm++WW>|ZՠaUPUhRQns'G3'T?V"4 Z{ xa,L agEx%eN(IK96ξOg˾0g(g#w/$`(^ÔXeCeM -[q~:Yͪ~bEcI[k%gqV]վ k?eȌ- gǸctެEJ:}#gS\?/W}y샽TD{٧pjr-«ZF w{;Z_]]j}W%k[,0>u73=YezLA~'3*McF'N13]75]h^v1ŘֺֻZ"\ 殦PW+jbΟ99utsw~xQ:qPrg=f'y9\N4[/[7[W[[-h]kt흯׿kʶmC0oۓ{].k˭Ky rssZ-6a!A^n\l&|Ͷz2'vcžb C$&1d1],qvGzqX.%^yM"x[q$vI3ܞ.{S+9_Ry4J$iWgHyu$]?#fi/=hÈh@GtD'tFtE7tGD/F,/!1A!aH@"LHB2##FbFc bc~H$a2?4L D& `,X\F1s>QXa,2a9V`nc/\ a؇ A<x8q ? < P m6؉[8oq)-tJ4 IE"2,Plʣ|Cs}GTHbZHVKQMI"MGuGɟP5FHAL!ԄB)50jI-H&SiI]6Ԗڑړ)t@u.ԕQN=K(zl*qQ 7"I:H@Ec!E8@00h(:N+:GtUF7:G#A=knVtXd47472>_?rWY2N 5m"َG;cn0{{춻˰!1X O"WQ5Q$G>  !a{U= P>Tu;ի^ģq.z?z+z'z#L8+;z3D<_Ņ؈L،ގލ^LEDEFo?#t؊ǟ~;V?4o?2U7o߅/_ _ 1eWǙE=;,o ou$ ;)ŏ|vdq=@=qw?83PX .v ]ɮ6g|6g'wvO' =ԗq{`+ٝ;7c <{'1qWScO|w6}}~s__Y}ѳ<Ѓ9}}ѩܽrrn]w~ۡŅo]WnF'dwtvplvC'eA!92;֧JųDEd֮S:I{@R1;z+"0AkHKGҜ\teޯ!x`yz@\ftz,c_z&OȖ#*킷Wdʢ?Qi02~hkeBo.֌tJ/{L[ƾ(c_62oVM*.He_PpQql4Q 2Z:tE58'-YI"iDf|u>ht4JswHU,| ǡN-goIL}AYV۷(޼\muH&'LAfU w`_¨vEͻ9`)AH] yYQ伜= UIsi.(iŀQ1m4ZI[mlM.r.0Zdzc*RB]&}hq8AZ?ĩ GD$T'Z1+OsE:ncԦ֣GU+(.gsU>few]K󜎖ʧrԝ0AT cX>gZۓګ3:PT taRK&H\0Q3"vLjOŨ<^BO~G2e L^mTݜTT'b> g(' O 9j>?T*)oݔ#FըZ2zfv͘ZVIBw Ru W´0rwj%|LiEQ ҭ@m@(c a B΀[;v:Sr*aԠA=ͤ4KK~^D1TFj$gpSZ9XekgNB5eiIADK6"Yz)ōqNwy|%drl/t6P܀=bZsubcRq=v+0( 'o*S E,$jESQtYǕWpϚ4pABG'bU PuH؊o|on2 [6ޫwآ%F{$.=v_==)1&&5nzw cbƤonħz{Yu{nӽ=lpr 8Yd|T=F#%ǡdk|^KpKV3dѶ;V>l{?kqТCR`5Dj2m )0=8ޯ6ϡ\):9By^FdXZڴVضIO&?Ql&ulf&Xc\$G$&48LY y8vdI+ \A!0dvg:1]p Ё(7 .DfX]V/Z 6A~W߄)9$5Cyi|MoS3B:.I`MmRC'+i( i?3(969 A[ ۥo/o\ͯz!9*# glU-x . '] N&v J8x䀀J@~I;}$CS;@q !jy]Pm^NlN{rͻ_z$<My`R[j5*k!r†K x7%ŢH9՘=HM/\]S})%'{<,e_s 7=;2gr~zdc&#XPG:FGHN#D Hh]Mkhh%:RKtlS}NDMY]b Ij4XθQv܌ʙJVPjqKz농VoX[("ЫGy Y:!B槨U hh!+EtTn R_3BbcP1RMb QB|;uWy:W}v68XTe8«qޚci;oe}[ȆOl9qR0wܡʂHM 4:=fɣyG_nc B dGqhIG LKm S\nR(l1"s6Dӈ9COEh> h{<kX! g3pol]FWgm_Qܺj5wl ˟*pkj|gkp6b.P`K<_ȓ0hDTBSCTmT5 x`<KBOV31mˏ^㥞śv,-=H6-̅NbJ{vJ}tE5Ri;7 1x̉3;=LB*NJ͡CO:Li $^a!aC_qbDBVf?([ڡ3 qY8x<_Mh!Y*,i0eZx)=Ô};1|= 3$e-=_<{hw6RRd9{$fY{d3q'V Trm=w@>KGrLh3룢mN3Fqq@?N`:{pbݮ02YԢQ~:H駸A#G|fئkѕS%]Щf" &5C^) nuGZ/}RU/Uu:ka*?j0 0Xk R`Vd0VB UE;-Ua -۸d°HcG4j_wEK!b{PsP0=T+cY̋ƷfPbsbs`'g,ԏqBXD`agj|4' IɂԉBCȩ\2dw{JLRwWVd,R_6> ?v>j݄Y^GJA좷V,73[lqOG Fl"I||ewxE%KbKEEpy" oQ|lJp|aKnZŦz< a]z7-]k/\rRiR\ބ)p%dp db֗z )vggbn~{溜6r&`2$6ҙ?S8&UpÕXCw\ml>I?(%u_y2 OȌq'z,mɣ>JNJ ?$H_/mŽtVFG :.d B.m2D^AЊZ p\X-B:d+ތ 2L)z^h3GÙc WƑD;c=jkd_w1u~ɥ >XuuzO퇸)HojTˊ+z=sYe/CEM^OlӭV5xy]@Nfp8sT|* _j8W[͢=+ ~l_4UK#t z5FeA̲L8& 0]Շd: ؜}<@t[ꐜҙw-TL/zg; {}g'[ZY`tl }q8{g<<&YFF%{F3G61u~tlSj:tY҇;cԜƋ(/j|oVe/Zyt0͂?vq z`/e~0xVи&+jD(NVBHn&[6?[fQcTMŗ١'GCT| Z+yi^&Yrj}AD;ysk4<h8*88l2k-mbb\II^[7c>o>/Bqj؀/arxh"'Z9>}.| D( 5UԬDQaQbAk55"(I%ĎN$F6!4qý?BL9s.w爉uVJGQm8Wuu ߹ ZHR7dl> m;/p//'[=Zd vV M/ 6> E7MG9R.\@WY 3ӏ xO_tɥ4y&U^͍ue^*WRీ>a)`Hйqf`KJAw{QN,lf@Ȉk4#&!/v!I깪(#*u)AcfƩ~U)Gt`MO0K̝z Ӏ=`80*$6BM_X.QJc=Nܣ,r'#Fkăps~5h1P&Yj&یhQlnjBZ"7ެЌUA9 a2!J6V~WIiU7P۔Jտ}~MEGt/h1FIf^OQK 9c~X>Ot+tlp\6',hXnZm;`3&"㙢iPEGQUt0֙bFK}&CHPED"8IVt!jZFXk2/Pڀ"SVzֲy[J}_~uaʟcfNa|g[r |-6!(. '4|X%B,E#zk].k! x %QH(+e4V[kwؠܟ<ԗ pgr͑4g=6r N2`ahX}$sarP$0ƌun 4e4%9"74UͅtBWnʭ!=)gOE:LP*0XBdޯ:#9~用lw~{B(34۰ dOTTsN$8QdCSo -ʇ/}CP;z8]f-^\or*"z?ܯϞ_tv1)?L V zQ[oR]N|ZwJVԆ@9E֑֑:/ugoRy,;~Jʬ&eeɣeެJ~nIah#QDNYf>W:uD8F SkW 7s\3d\Ϧw] O;h6r8_*0 waӚ :ae|Jjҥ+rK=xnFJj*A;+/%b/Vbr o)RRafg}KF䯑&Y:'*6:ygfdجJ"/&R7\<+7g_qeܦKKEʁIsIDȅE$-LNBɤ}Β1n=N7@bVpC=mN6"0-9vL-ZUN8m]^-Eȏ(ΈD01S< PƋj&dd)@x$mW:GI=c; ŮZSQ EQAF*Pi) 7zOrKHp NBbu[m tj=[F-q*JCQZ875lFzԉ;Q$G` /#/`_k#_8ͅ=O+c3=*ӄ4X#tV>r >FGUZpXCk?^\T: MB'iht9/Q<M{|N"zo }.{ZAymH:P|&x܀5_Ss45;F9> ; h/ (4";y0|&]b qvX`}WH%NH}u*o2g)fRJ.b B!Bx!1cK-5B%Pj%ѕRu2K:jS˰LRD9'?<9Ϲ׌[zu"49hh2e:ljiݿ^?~ ^JN<-s. Nw0{ާ{2S?2Xv9뱮;?bxNg77Oy8ǽkd{Rwp ʩ-(+OYߞuZwrH}`V{ v;{5l:iۓdN^nR˞LW!ElJVz$Udx쏠̓bLUVbM9jz٨?3$+y0:IQ֏JI'Cgm6—*;P j,·YSOQ+[;E݇=kl<@|T3.D};9izfvWGOlĞF6B+s.z3z@ῆZ[`]sՊ?=OsFչVq;\Σ9p5G+!c/> O .EJgm =r _N~q{sIbE_F%2^&}R/{| zf v,P; w~ƳtWe5컈 42>K<B/,g%sDZ(>#4sEl?">oBKtX%tCu4]شj5tNǐzk7Tk뇿?6+ٺQ9K~ =^)3(rB~zA2F-t./^z27ݥ tccz&]u`dc)MM{?N t:[r{"a,Ƀ'dn('o8s`4=ZҼ7醖E)= ߺ ҼF|\9elAG:Ni~d om ̞6c\mA io|C~֚ϾP[݁ͻTtx؝&FѩԷRՀ=LtwNy;a5]MQO'Ȼb:L9!Eٞt2 ߊ]eB'ԉU|TArޥ!Χ[A9=;y~KxΦ;8qdEs(ػf9?1ߓ y^5sǶ4֒Y=t,yahwqF") $ o<ћ&5I E:c;F{nK}`s?!>SlC <}xf_6?(/$V 6q,‘h|q#Nh }O6?r暘 P<]j rwط~Xbi׀vdJ[@N;#i9GTm8o#5ZCNβ‡gO#Xޯ{V]1x^^\KNWS1]!|DjMK]IL Qwsc" /?">w'Hn=odž >w 2|)#b,#E>}to6'* mMEC}$70~b=v ]_G{ѳg_x [S;$]ϞP.7q]{\TUTTĈc1ƈ1"bDĈbDETDDUTEE:ꨪJ/Zkk?n8w{z=}~d=, ]'E ~;St>Į)V Nwg!/7>w2sNx;ƹg  Xx#;zf=S*s;ٟՊ1{Bg,KO "6Fd>*qhպOk 2}rT7F[Djdls?ZЪ0ְC9SdY}o*:N82?|5Nhc_{7$:SuIϏg jdPN utvm ^~F9p>=t?E iaB:V[en36<ʫk˃C\.=Uk(N_l"Fр[-?Ed_k7w{=p4\Ǻϐ||yf~33z%4v adG"}*i??_F<"X{Z±J_smFkZClWV bi\/Dܱs;4g~m/w2CYKSzFրb4=7gmրs(xMpnߥڶ݁ c=drw܁l1#`wYEDžL6BsvU`\oA0FSNHY{&oٯ_#sXᴘ;7Y##s~Jdw >[>o涥ڃj3W=wdV}uV|\ə솫c-x"k{_WNy_G;>u6}u |gv;YwMԆ]~+6ҳ\yD[G*Jdni8K Ϲ <N~ҍ̧oqe6hG8w'2>+'fx-"nYd_):y?y6m=ϸkqya[C6.2u#si,A\z L 85?K~d-U8r@g:m-).*ݽBE췸Ox~zXؤ}UvyUtKcgAw{9;CCqGIj+7"b؟1~ֿF˿3f]̗ח.p<ڷ#k?-t֍̝5֎J۸Nл,hCd!F#ྉc{:g-lB%9 c㐹&؏Fֱ[}jЩF\בsҷ ޓy1_>׮͚Z\^$2lD5h"pGmpI4F#kBY칈~s-65T] =l[NbQ{7~[t $]e^2rd=7oj!O @dq2s o<й^Sp;_Z0q3tPR q~Nqp?ce{_γFd]x|∵|>8+ݟ v"o72z!^+yF6遯\BΝȿWYK֡ {+GhEgclar"!=%]esި˽sב}Dy6ٻ3ÿ+4^G[}4.x}-5tbi:.܄4t嗊U[̵:@ڱn52Nf<[iWXoZc$w8o5؍>|D.2oS<#b+__Rd?эa9ƠncȚ9d Yuxl 7Kq4}ӽ=kp 1R=6Ēun%؎S;uǓcuȽ2,%﷗x ꁜG٣7819b uޭ&ԯ˯ 6h"{~yMބjkԪTh5iO3Tef:։ϵkvfY!gX^:{k.XK5 m&>vkuJ:׽$3\t86ԠA=ϓIcdL%o/ӛvp] v%[w"=:ZQjsުwZ@3EO%9ss7s7]NߪGaxܴshi}ԆV8vletz;8g, #ukdb]x%~E)wc]b r?mps|:>{it}CͳMGox,Jvܹ#|#YQz}mZ2 &u`sLc z7]F\.? k/֔:5װa3'QRoEO9`+aʚ抟(|pZN:E-N(sUlu?q|^Ix\k}x3Jt=sĹQͮu_~uȦ诓&ָXg?&kߥ=f8=,WsB4 &t=t? b=A'G]RJQ~Q:ζײàYٻKoijZfҡCJw:[o;}|Ns@%@4d:o eY;Ͻ7|=9/&Y X IOxp^?wmz46\f_;[ POcfL;K4B׊gdNi-w*|}x!im"=uD}Z#>v,OXsq8@J<.>ݏ=yMm?W;kgU3/.=4I):o{,3| dڛ!|S,V;r]һ-mX:`G+JR?cOM:)N}絧i߲_i!Xb;^-EGeDo#[k_A{o]A'i~蘻;+~>?ǴA{l_^;U3,*Tϫe]֢Գ}^7XwV*+ߺMsOڣ Ct՟Q]ΗQƲ#t3W:jٮCqg|?Y(c> īhϐa7O<7ѯ6Kx2W {_Ksy=h=~(mEygu|tz+f9RV Ds~o%z">Ozs AnzuhWskבծ܇k=ut \>2*kC`lMO}U{;Wɾ@gGcg[d[ued]5_YK$>~>Uns6= +/}l@=rbI+A#J/e/w]v 7/EK7@7aLr&oGWM~mNb)cz?a .F=y`FA+d8VvXMt~<,U o)8qJzT:Vg(yg@ou~oᇫЅ&S+\b} }%})v f }֠C^#t߹&zC9]돫w;n/w%Fva]["I>/49W2Bsĩ m΢N?%4 Wbxz+;/]C9h,0d*ޏ/2[z?L+I+WI&$$$$$IffH|M$_d"kgfq5?|?=gs8>sysFZ?ܪZ/jֿo-Z$k:s;}UnSw65hLy u?ywEZHo` k}li\1)Yxy,l?T'/2&x+*Ys\O;XT%.Ib>xr7y8#]g(Mb,-ܓ6f2+tޚGs[ _IcdE_*-M^76iG: BUtntEETge,-rX"̚f0֣g~;,kYsߍوGqN︦nވ;]p?a=VsF-vߙSoN kk/e4óhC'}#08Y}›sݙ7Ak4r {v1h'ukӞF|Vv}Ί˘ȵp7u \Y|Ţ Is]37F 6p7!9Z|7uDm_}1V𩻙0qQ^?+5Ccҧ},\Gj9Co_hmlPZQAZ/Y֫FtE2:MS:}w>,+ޯѼr7Un [gLF1S{T.=C(ur8r gB-s-Ɵh}|aV[8SYS#TTZ,UWkF1`!w+JMF*)x;ho(gD=?{Iuwbs_"zynF>ш .ҥ܀ڧz5EkDܷ*9l+'Z ND1׻p4ӽ~k>4;՗h}-oKN벅{~_Xz{7-;ևq=~K)~TlODR|WL泬w~o _r[zߞcKƫH"}ҷYAl ݁qK2utL7"|jp|S5Kz[s?cNxw xG-bv者}mwX`nnIbD[y~SF$U&9;ڴU8N$7'* {M||y ՙhO8cdYtGznOel(Z=A{Dkf̵o?¸|p2j?'b2GMSqK-3f|z-8g 5αԲ}) {1mXY!-'wL#892O~e߁lZ7/cϒrv?,"3|U4}$WcyxU^ز <lco9Ya=N/bs(kcxcy:Db?rs;C?x] 2|-O?,< bv"~"k_nǗtD^[_FF1bOL"l~mVfL|u-ζ>MN{Ck/ %za璅s=K_<<֔.YK/swEdJp|[[_.%x=Dy]k {]X''>!%gml o=ױY#:W]S݊b/.WB9HlIuH/!'t^QP-St%x͡{=q{.xc=a+K ]{p1v nEmݚ|~k%榅}r|}s})_-1Tvq %L-GkK*rsj:w-dV,ԬQs}hnB>p>aOuNTw{j ]v{ȕ,^0Z^U~<:-Y <վ+u I|Y)ף]uY먵V VTququq*""BT*F5Zƨ?WZ1j5jsZ<9yc^ٶL"}zuץ_C@_b> V sst ҽZT*(>׃GjMuU|TYg-3Z3]h?w__og|coogQgoXqv:wAFΏaͯ:ߟz/q9Ćb/ KTDž@r3oOS`}t*Vec"z471(Agc?SDZo#_ʁFwupȘk}fǞ:2vQ9LsqՂ&y*5m/C [BNGcigd9x ZC A; q_}>@ ԙEzЭgg{A`)N]Gj8LX;-q{< d*zxw+-+a3x*,Q1c}+O{4ݠI戥xSn;'͡gvOq tz,"˛ߴN|Hib~uL{g2fj%+YG`f9sC] O ],Y^# _,mKq~4V73E |{78qd)çuKu[%y_Eۦ-ՓkXzo=N[#]_.).`ۜ7gRL;SUKOWTya&6)=nQLAt x{jza_; h,Ֆ^אs|}=.^gz>o%9(RT"k' jh2ͼ?s!WK #o/YWߍvۃkw-' R7v[Q/O#O 7v|֖`$tZX_᭷ɞmKU1}53"cekaNGun-<6'p䝳g:9SKI}30QT=ۖU 篱,#)~Yc}shމW奥ZRY{>W,2_sYk'XߤC94赁K*й)0^yOMb h2^TG<=Tcugzs^׽ubx6@/~q[ϱv3#΢Wџ-I,C h-8̶Saf%6ʇ@^1K m_fΨ=ßm>,傓@M&:ԠiߡLnueCח`X\E#;Ko?!KΟC_@:o?e~ucd[ )-{g}MuۄF66oKrl'Mv]hW8Nb-B~uۅ)61A)=xΖѵ .)qš*`[`1l=y #댧Nc`2s4\2 K˃êwOb#{)3> r6-6 4/>+Mu칺㇖g=v5x#ܶ-c/t}Xg>:{Wg5Vq;5~`KW+uK呪-\+]&5ݦd [mk8dT^ߡ+LoIjqR5d u= 1r%'W5\*aO߃ibC4{ctqZG߄О1A ?~F֦}x<kW\󽥜'̆B++zzsʛS?Ršj2:XU.ITwTX:Qx CV+~<{5%Xi'9ɞ?^Щ9Ee@urw92G^K2|([֙@<5ݧ8ojL5r2/क़}\eY׳_)[ -v Yz/@mwo̖ɧW5-$g`.o0[qǶ ׌mYҿ܆9qGߐ=vjS[F ob9|g{7:IUQc1ƈacĈ1FĈFDQQ"jV>}}Xk}9;='}>s>ϏSl;(o|/Zri 5FܲMxo"9fh\LJߥ-! B+7WZNvnsV^Iz2kdQ֡ԇWgE(m p-_ٸmcElb~$V,ޑ`[ ۉ^DZAqOر~hEIK'ۍ϶a:M [?SfXvꫲo`+_NYO wK,kwI:/A#{~g룷~Y z:xPg?v=wFŸKv=g%}ͮEx|T*>Y=@Oqc цu _m_I}߀39@]Jf3-ؒV沋j|'_ ՚jgVb<GisY#~Z\} MY]$U,? ҹ8}>K|uo[HkYw_W>xvn$ }j>-TLNeQ3\o޳6gDEm5"?-rW3j& nC`m,X(^T8EyliWѡu/[H^k'y鿁oT92مN**8aOWy̷s?~ͳ6A7~ye *oރW{=ud[׳7wGU\٫C z>ӄboS5N1$WڄsM9AghXZ6ֻyˢUu )cvlqN|%=KHrVq-r}Uw SMY#۬݇]qL 8?z\N$&7WUNn{ϢjYMCdsIu\"wYwc^9qv['A{c\1zƔ]yf?],^czh8Wj-ַS-|b+ݎ,D_A ;ȓ/;Dw_cݓA_FHwlzKr,u16,wrTREUv.}\R;CrǦE_m5o%^+XxMl`/ډFцYԫz3+xjXܛAgP_BHw~C^}S3l݅cUm[:Eݥ6֩_ƪY0_'}1bC)y졻T.u-z]'i ޯ}԰x|sXWǒbѣJg]:SϵiqG;+)gK[ޮMUYod.x-A/XcaY[W-%C̺g(Gs,z z(w|mYsx%wkqW,P9^=dc\؞k.޲^]c;w12UŶM}'gdMbgPDELF_"g ;u>Or8O/{ =UmLݙn^ $þ`eGe?>jbwyo2sno9yW_o0OLE{ O3g_sNw#vyM\\a+@8>O un?y+v|Wӌ﮾?<0?EczgQsA:ܔ/.q gи ^'~g\'?->x߿%\O|!g Iә",}NvZ 4gȪ&Z{9/?K~OY]&gs9\ivC0~"i!C|˚өP]sϣzZv*  A5ϳ^, XWaޱ@{qOqovϢ6,³б=jjy0xș#qv٢.Z+ل1΢'wYzKwe{D'`-\m:'SR<~=`עt߶/ز_1`XTw}\'5>|<X:=JҧYUm~R/V|l_:wUZ\}nXI#{MzS7h^>K=+c h"z:,pms˺v-v;>LEΪWSW;J4ԇtso"/KlKtQ=_MRͯ =W>lW ;ۄCPUxYQym{4!ㄧc|̃j㦊¯\G9W_(^2{@x LztW,.w+TݯK.E%mw}"Kv=?¼Mu}Ɲq\^<֪jUECcc11c1*"""DU/֪VG/Vޫ7'G_o^||9"??c;UT%[z5&n'^5OI|;Sd/ZԥM۱xnUztfull1x|]]/Bs^Ut)X|1Lu3t dQx-4yZ+q9~# UW)u ܊B)(e'9d]^[ӇZ~WT(%WQ]szwfnY!ɖF:ŖjV-y,EUy WZ"|Q |Ǿ~|];߆u ]aXQ~5N}Pz+y ˲yF=L_ޡYq=50^W.z[Ƕ>|}kzGS_c`Jo%? f yzu5X%8mYghB{.h6orNo J9o!8 {Q'wx>}=y,`pZksj7GƗ;?u߫د޲?;Q=f14ZұOqpr߷\`LT\J-"'Ig+_pwb.RAu?@n*-Q.շ=e}9X38n[Efsz-`>7Ch({{I/gfOYis| {[{Wzn5wjѿn{byCl8dMXMhOun{+O98C~~h"gע-r͒=-js<ٲ'5NѯeޘVo_0-douKd Y`64jg9k\߅3Dv`4 Ec캃ܵg%[WojQ+DWTܩۦ䗌ShNo$M5Hq|6JGM?K ,HY5 Oѻ悻EϽ5xl"'5 7Ky]I>n!ǑENp'Q>wگIvW-\өE]lQJbKи\n,V`clZ̆S*IՓdWĢ-Ͽ,r}/s)&h1{|ު-d[h-ɥ¢ϞqnnjIQSo5_j,]oG&e?ǣr}[#E@[:I&)WuKX[+wS.)}oޭgjs=-rkVKMȢ:N.򫞖oN:-e;TGآW88׌]2g?\k`o_=!yd6wZyuuwn^"0+ZCrIřu_]5ԡy,fVvk:o;E^S6ٟ;_Z9_#xrKwO? {/6͕_X1md9$1yk^%}Q-t~SX9,ߩwɹ]duasϱ} נ`]9>Njywg{UVM [, ϺjE.sq$Ԝ2 bMxj<.&vt{wyW6a|im!{p}N}]Y`-ܹf;8ڳ~w`ht0u6C[u,῜ʞ~=Ɔ׼x'cbvr?ӝdc齍"Cxm3>b'YWSMkmQMוz޵iW8HK̯Cu"wtSݢRgNZSYIӫLXMrtIy'Y_VqŢp%3OkuYu{"XOf] #Y}}>/l6t.%v9pN.~ W { q}}LI? ލl3Cl/3gО )#x_{载:w T}}9$q sd|&OY?g{`%hɝܟy.'?@^r]&svb/52N?A =#|o6.Z4G{U~5+^xqV1ic:aY Vyn |]e"=3q=ˬ ?C_AnadZv 57OG`d*X`٫ǚdTnY`)d*C M7Gݺ~期MYfS5"rg zoZ9᥽$$kSYH:ωd_udI@3KfYe l+y-jToˉfUq|FywiO~X[a:dWsw6ZZ߂gǒ>5w$El #0gm*`2AR{q$\$/Ы&/"޹|)M(.kresS-?*=m=nrlgez߀w5QUxLq&mi}?1Mx4_?s%]7UsH1bw"FOX+nյ[X4aFoeGs(u6e1xCyA!91d,)-WopƱkEh&Xyw>CJvAe񷹃 Ww{]vbj#z' G1g~pC߃~OXIW)ŽQFo#zci0Yȡ#ד 6Wg/)^~ ]/1zcpclŎY"c)A˪P}c?YsL6S;ocMd ;.C6{'앑ɱc~NE|,Y{loaxyM=c4of(4y ܇)G~-T|!ww8|K+Oy DnEMֺw,rff0wos~;M50dXOXksNvZ7፪v֩C/Lbz%}u?#O]\).B\|vS>M l#7}=@culX`z4k}emX]x`'z{  2/ Vm'h6o^T-k9 %t;-dm!T+Щ~/W_GDtϊ_X?Kî7-jUAo\N^gd \g7WRL;נS,X*v-bs!у0~k_e\E] =fke緗hV-^@knjE=S-ڵ"_8.DZ:uGFՈQUQ#""bDjGj8Xscck.jYb^lyg.^}|ok=ɹdƚyg]5 rb"oVAwVX9RF=ehr?Z,gŧnW8k|ẟ s>zwL7J|$g_,}z#y{{o_)W.y{\gvx>} SQDv;ͭ~3d|CWko ZSbK0!4z{,rO#pIS1t{;~W9_x۷׵x I:&:TXb]|5QGsK5m;}g0ԓdZU*%|%[تξbZŧ} |Ew.Ϲ2CxM*ao™Ulڲؾm{,pdnu-prY]/,p\HB^C,x n?y>W>@ ?op] (O&'KV$_ózk](O=-G; aOWj ^쟱`sM{q].~e l !"guOZoy/x煹Z`/AϜOSf[ Az wZ>y?M6;fߓ}se>i L =WW_O׼;w[<^2|u`q_Yw횰6-ۂN}AǦms-d)wIL#;|ʢ} sy@9\ae֯z7wsgMd)4W|H.-/GWX97zo%Y ]KxmX`źE]-X}/,0[FY+5Wu^2Kt-X-zO1.z=^1X,9kNl' =ӏ[ԇmxP_nZTϰSw?˿פDĵнWo/,@}G7-zs*䰻c=G|Km?(/&{ c+{gkM;-SI sm!NcױX}w}ElbS%w9r|7,j|-8 EeQ{{h]}tر緓EO(&1c0cD'+"et&%ի':߲&;fЭmG9z\=V,ZV~)#<~L8'l$>֠y`8_mEhWϣ+cD7΢Gp6ϱw y?bt _azƱ:~w'-j_y8c^~ y ^®t;I|k5=dƼlwZ]>w!LC:-zE~'y| {آYݲEݿ9m MӢOkb,+t"N=3xځEo0EU 5{컬 #qaJ ]F =th"Cd/@_a~>5oaa,#_ӡEYO#c]<&gg3;f_Ì3/u,pӝ+{Sxyc-Ee3Oi,ǹ=~N5Ku雜-ș .:+l?w:&M>-2kM|~x-SaC0[xOg6{=G~47=e~НRA@GDi!w|+Y$w='/l ~XN3ngb9WzG՞:rdSh/>oup3Uy;XX1o'7- vC4ϐ!Fڢָ !F=x^"e֞zu#$ltwIK8)s7.ksdQxU_Ku_ Aݖ<|jQ{] \~=$fzL-0j&X~{i tkˌqgyOn;uـ ;}i#d6qhr+drʚ޳+Fy]:9\ G5~|\5Wxvg>=|Ƿ=λĩbq7_ӵU>}Z}gesի]G )۶3~m:rĹ6t^G[g]TCpe [Esq]11wEtv?!g=?,] mhQõ-'ge. X дq-2&WoToP_A* ,Fz-zzWeFГ̇ئ"H̱oy]8C؁o>H ߿~[>c嗆E9g~Uiv:%j UW,uu W=Pבq<ஆ5s%.5%l2 11|Y{/U,B7bi"F mWV]ϡ{٦EGBaɿEDa>aEi[j8zY'p_EY#Y|>23f;gY1kXEF:|XMh3: t-=f(ˏcC`+&||`{;~)VO[ط5Qb[ykvQnQ[=ZĠl7\5--]G4?N1Ң?&Ev|zʱ}p躲d=hm@Np=Eg8cJgdbpu3lt2ָfvDGd6Uy%ޠ}Ht[{t77IT4]cxt.'^KzgBWI ߷ *nrݿu&xC.%Zg,n;/}u sLMv|b=Gt}>H9rwQoPFo{,x}[oZtiLE05l۶5,Cp\79cl߆j{أHt֘UE|vEM\}e}m]'y}xfk2>裷h;~ -ΆٝߊE4:̒c;~D+}\tsom| 㾟h9VOU}I>WִkɊIZzE_ᣴt&|B4~e|~qTy X1xmK ֢:XC5l<i(~, 9d tڇj Ϝ -Ekf̅t&<,rL)\[[[zƮG{*oN Z Z3ؿn% E]ԁWӢq~m`[10lUGuxҝ۬ 18k ,juſw>CVGc`F6>>OՍؼo{ w?Lq3G>3pUmTVWC$S/H5ϗܫP|,LhQOp7wLneM6|oJށjuZX:NQ񟠏{>T|; r geQ#,EoZ^5Y:ER{lt>4*!rnFya' OٳO;%p "Wo4N9Ww}k#ߞE;I~>x;Y;Sբ?H6R-ڪʟM͖.:./3T+ W\W\WDD\WD3*55553Sy̼^wx~<lp^/O#d:1w(W|7q"&/Тs]/ZE`;]=_2:Ͻwo'+^ Qt߀*޳WPr_b夋h_k%=o]dp}ꕜyp-g#y8סEd3%SCmdp&پZ[o:[[]Kw4ff,ޞrb0Y; Wt 8.w||}uup;]wnw~̗-z-=}7ҬU^иe7UEohYN{"/|b!7?)hwΟXl՘k`U/b,A40mVGer:tvm[n<&-rE?_GO-j)X--Sdr< cE &%߸> D*N#1wwǢSOl.5XLjZ~C2p-n-9\ysM}? T|γW%=~?ϙj鳿E7;m\KrN+rUQcϫibKAW}k ]66)7tܴSU|p3-)ql4`.X]jl^OC9\˒kv򾅼Mt20ZLŴ׻nܿ_Js&mo<3' %\ݶEu9Uw0.O w^8_p%b6[9f{<'k8mՌt` d8L*_ٻ~0W7cp_zxedP}+ad+!k!-.Φe8n7 5|o!-˱dX*:LYnTih(gs:uA!sZ#Ifҥl~kN6ajo2OZ9dpG-bAm6dU\8DߒşT9TrK*MD S,C{x{Q`98:6(%-D | 1t*?bc?~)h!gq&>L,PH) ٨WA)hz۝p~El6nkZ$wO5Y=:ˠN˕Q}xUo={;|GfYP1f%E|v<.r=oWAVIvv!!lީ f g9ۂp%;wk6y+?V,/[ylYK?y 9y"_Y g ءhQgW^v+o Y0̈́7-s-մمs 3mTWh[G❏x;;ҡYOAKo_ԌE? vU}ȭ th@c6Q2}-/Y\DV~"_ˆ9._l"yn6-:ڧڦuQ#?5q9ʾ(Yďbt٢?Q7 ja_ǤEOjQ5Cq_Hr_=| `(&]E𧹓ܝdn!` Zy҂z S=ŝKӁv >4OO_Xgڿfu{ͽ;~N}F5x?r3>8 ^`OHwn7pw߀o Xwtk߀q{>ww?zoJ߰%x~Ol+zu.76Nw<^yw||Ľ/{?2x /Y}OZ{?nw^e<9:cq߆|,vp }<YA]*\a9L!;E_Oq߭qbET p+v^ Ts'ը*XGe2<\śz \glѧVwoaJ,7)|o\E&ɰDI`;Cl{Gd22nr~uܺEo\Ak/sأh\seJ7ig lc~d6[H42HBeH(V]sE$cVD$6ґ[HdTY|ڮj׹.8_Qk܇<9N<ĮY$G `gM=U g3EVm@-wj# C|Gm_ чkn/<K]mwEjlҮVd;Z @ܫ#^?m%)V#jc}·<(gmgekjg/>F?.Ʒҙ>#G?:Opc`yXd< 9Ɋl *^^+zVØ5,.Ek"|W4oXT2:🝯GշVGE"3PoGt'4Ok!C՜~clmѺ~|Eh*>gݫސ˟pz六=W yhG"8na^c,g P c7{>l71q˭ wpC>3G8gW~5(Qj< ~Yc9|NE9o@pqY۟٩DZvRWu2i?c|xFv7Z6WF[#ٱl{LS# QmXO,׏p<:[M3љ#Z1O;ѳguFt-&f1M⴦| Wz}/q=\<x K+[(!{W=fЛh:O17 ܛo͜kbouH{4G} W܀?=`r~8&waVpVKͮO8_N&@WmoÿY%~7t}ȁ>ado?qzX1k5XmukWiΘju 4V8ǻKy.glګa>}:Kkelw4O`nBk UNjF} {|^[9zE|:ǹ_w|؛y{bcnlqV]/!yvx9 fҘ^ݫg?| :z^pg9x{E^m9R]sw܏bW'j>hDC&V;XNQсN79||ljo)]zspV׭6/F|ϧ86/s+4vQ}po)Ls'ح[ h57:fl7M|Wf}xQ|}1qss}|[9ׇlbGXo!t=Zcd>/siZḃgC<{ :5&w]u`ۀF]ѽҝbm!w+a5Ơck/cݺ_4^gu3ٽ.`>mm6+WʴAWw{5u 5GbԽjtx|[As}U -3ZYqto9UF}53pQGC8M/c;wz;ی-f|9s҇{>`1[*.P iL.pT$8"z O7ClLǠcqcv*k44f&_ddw]{dy/k-w>W ݼԌU*MĜ)2t["u1Q(#dSY t]žj'͇CXdQN&w3INdѽFt`xR雉=c*u,<ɤr?} k=&Z|Mn1U*q /y&Oh%bi 5-CO|>+a"o]S]*6JԦk+qO)J/\g̅w"Z7.{zt@##yFr/g*^+:Sy$ҝs/˸Cew HNw~uy{'3CMC/&mp.r/+1zsݛ<҂'άǵ ;._ɸU*9 )P'0 د0[,{ovo^q/c;Н͏OtgsYws}]zQp?vgֹQYyWȽݔ}9wܣ܇\d7w7W_L׻RgR7[tӻ/%*;'ݒ/rOg܉lےuGsnM.Nۅ_ rv%݃dރ* 89E`9KZ@]hDw%EnR`x53RSK'l&(,6 b9YXgs<9=tOwkrwuZMZƨ-B*} ^9&;^_{.|qV37^/X>Uˡ>Dz-6߮6o g鳚봦U_uNMlfz4k:hQCwIvv$;u_L|_,=>}SZ@ﾁG3~R~Xy}/a=ݑoyl{yztHL#Zد5>{VkIpn_>Цy qvIm;6*iT,Hp?\uop M,7[㶱iuhku@[nA~֯|'Xּ]bYq_<; rSs?zM~}Qwelm^s6]߱;ZC`}_{L8mQ>[k.߷]a?U{ϋZf0zjUv5ZSfNxd(%e^`rywV0[u.m~GodtGW{y9.#.KG7|ܠ9k53-}UZt3iu_t=$;.Xb3I=$<^NrGrKg~P:;Qv^> ~{$Ot)rQ(ߙ?ݶzڿyfuT^r7>moqqѕc5߿=? 0U>w.OƲ?-Gq|yz1ݿ9qlT-3qٕ8XVyzt|~㎖"qnΫo>䪴CiJzu/1qɢd.V zЦfg3Q8mocrzRO&gRyE HoRn*J~%5\(2}@gB% 3g4 4AOH#F8󱲳y."tH}h_ۡ>~4fy\6wv">Fi5Ja+'e9<%B@I)itUj&٭zB\ڐ0׃Êzٵ^OE=[zzgdQ]%b}PWYpάƙKz>Z/bjjVSԑ\y+/UZzzTu-zu+x&it֤chMj-9Fqz]zǺft_&5{G+;^҆-mgu'In҅-lIBŲy3G۪;ZƄ_ y=={KS7u5QpJ8ѓG tTϺlryB꽿rp3;,]w(^w ϋ>aN>EqH{4> s:ù!6]n_ 0a( endstream endobj 328 0 obj <>stream H PWǿ %8Юz j%*)VQQD((>≂x3#q/TF7b+U[[uUz " wgV2>6)5G`7'M7K>3'^ v̘п]Dԑ8>t@Ƿ~S#$3q&%՜Q|1ISR"lS3pzVhRZbj!:@(ATg #}qLOi K~ DX|d,@: Zr`o}F*hG*m}8g跡\,., ^mSւELc]Y/!lKbY͖2; -!Q(NuQ'ZbX(! !ލ,@ I'.R7/II{(!rc9BVd1dyLUJC\hEU)zC T*DŤ,WruFm>mRNUBu_=j.H{_?G?O_ߤ/o՟ " QCO2*S'6&.&.6n6]ekZJ+Զh{ f.iWv^zϭHWKjsusv}JM5kn?H0o>dUCJ}cBc:ad^&6.5.c3ccΨ8yNS<lllt;:Cp;98[v=۞eiϰGð=^QYVmy_ml鶩dk/kwk7kgk'dFXZXor޹Z?̡Vg(ge5JRB222b|bI6Zifyyȼ\{qbUbCO?L6cfX,bV}% 5~Ur+ra0>xDp E@PW R팊Ŗ/kPwVܮ"lv .{O7a6xa\q 4c<~1p1y-s³h7뙼Yⷘgy/`^Yk|܃ pr󊰙ܘs`d\ϰZ҅tzRoqdX#:o3uQwA>B$KԗQ 42R a4>!Q4>?4>DO(&|&V-tx=?|1>j|э5_=HWHAT)R(գ0O !5ԄR8EP$5ԂZQK֤P4ԆTh;mÍxn@;~N ZMLOh (R鹩h]Q^a]6Q1k*-v`zL;i#mK oh2c#U.P)ݡ:tct5MeTA'ݧ.I7}Gnye:(Cts*'+ՐlKzuBD`D:~o5?m^w> v>Kj($(P ;y"TL X _LcnB.@o>6}|ۦA[9MI45';w풜?:l')mЀ)- AP)t jJJQHc _ݝy_͛!u~ξ˞d`KH~go!u8k5Ⱦ̾ɞ*?aϱٯ!Uw(βc󠧕abg؏Ϡ ?No{i^ͧנӓZr]iWڕv]i=xo`װ{I^FN_blռXK)ȹ}Ysȱ"l@*yBnϮJ*_#W?=}WC~}#'~葻w>tpvf59}mݻ}\Y˓)[˒b/rTy T=j~ed̯ow*US['[D/DˑI_P#df𛺸t _5<̖wywszBfO',^s$jʓπ6)Re"ĨqE̦nm&}%`'*^3h rjqEP:|̇xYA^:Q dYl hR'2cԄ?oqk0c5 J@ Y16y аBYmX,T ՘m0 X4L7BV,X@pvkL|.9b5..}ۤXhںɋ8B.6e#[K!|9 hd媡 w-p;Ԓ(CPM܈tvQՉqnv +PrƶL0y9jc5㖜ԚAg6 [dl9$6WD=a zisUt`5~C"HLΌ3LcuAOh|EU5pԝ%ܔEYn; UVD!$G0ŸB`#;6]I]"O`i˥Tوc짉OXQyhFVS c}цvlW̉c+!r!d4CD4yPW[9uY @bږ0Z1Qm˺Xlsl?¢]m EY4>LFYA%B^Uwy{Th6f$;xʺvy>`"e!r[(kOG90*}{fќ%{\Upqʆ\thu0R"荄vgL-#I^zJz/腈C`2W3Eԇ΄3[^jz KO1cJgh'[޻G iÕK#@DEn`/nǰ}*= ƪ&:Wd P:`Qɝ^ WKTݑ/M鑛)Jq+6VfTA=Fi:v]{1?Nw2߱dpqߨc^\q?[8F¿_*chװe$8hڒoZmS-aJ.hIsln/eО7 dڜw ;uw5~Snǽ=!H77Q?oδR D/i*UbŪ(\[=PxKՔ K%{;DK@+gwyLa%OS~ ^rS98Ϧ1G AtCאpv6tʻ:,,eV+8d:/-VO>= qu GDyR BU(qHXZj*\,Ve:T LLBW~Pi8ajJE;xY?~N%j7ha TBujAV{Îj|2PBNʑu^_Cd\եi-փ]s6ϊvf!'o7 W߹1g*_+־Rrӿb⫋K_!T^f"ɳSyoV^Uٺa~G/}ѳ~_|q˯_g$~fIon`Ey=> Lʶ3v'glw]AQ]g=g~c]E%fն"TG@0hP#$uJPSO>W@aHڤv$#FPĚJ2{~yymJKsSz;wUb}IuCNBS*2F^X4|?K,EХpFrDTJSl^yS]qOXy è/ϕpA8=ko^uM%g;`|8[<:Z9;10(@3tě]Xh21ЙPob\"JźV9yGȨxܽ>~}}u醚y+w~toq2$Œ•W34l<CZ@r_ (QT4)t`B^`WH!)ȅDOv{=nmON8;9I}6C1k?l~x''5 ϓm8|43CG46Zc^rmψ >OVNpı6 DR)b؆}8 iG[p6'g+ 78_&:%@%=0ͬ r̆e H(PKcR f\RVHւ+5IuJc D,eH*{nV*5$KKVIdpu=sj/׸[!˫޶yKG=eEpQkg7D"BDgtر( "Km 2xs$~Od.d.-Vm QbL+zrJg=ՎމsW,~d/&u5&v[ܯTk?tlsΝ(Y^!X: *FFYY {g cF ' x >f|f#"Kc^$8̲I%Ѱk_vBnQfIܽ7|+e+pxɜU9p>\<#jA ̈́dנXI굶Mq1c" 4#D~$V˘ +#ӥ d6X3;6j+\^N,z\8&VQ|pU7|߷ןw?󗍛;7nY` Z>-ydGsT|l$x0:.X㬠W`{m"7)|~[0B XFtOųK;PϺ2%oxS`" v|xƪJIl1f6Dϡu&XIHk(U W4f3ehc Ǡ9MRj<7P`#ҧ99cYN`{>9>U(ƭ|檪gE-Yh[ 6ǽ/^˱cgOBj9_6^oԫ[f&3`BbJ.f7e+WU!K9~6N\Kuu͡N*MWnRrzN>hA#$qZ}EQnNim{amKxk[K7cj;*^TVp^'$HT.m(8[h hJLJ ^v,F c5> VI(k ,{u[;t2ގ^]9f?8%b|fY^^RBӹR.6Q` A{EhD%@:S$ +h܆ m TXL 0c1B6܈ce4Bo>W^:5(q;^vwM^خSF~kw :lݑ5Sc]uwZlWtۿ'>O5Χ)خ/6yySS?Y8* Vf{7du:`-Jvhj%kֿKU {#˽asג[קo.f(!/I@;:joA <';ܷ;_? ޛ2 [ݲujdZXWOւqƤ΃ Z,q7KЗE2$WYoz:R=!3$hw&oL*MXiroKJS,i L' tvd)нtwJS†?9H}ipC`hniy;/5ㆴNN g2@m/~;umw?; G@ d4r4rgudc<.`cShWlΓ![B;#[ J&1pqdA؁<)s\F^\!'̓J=&9QGdy`| d0ϙ'y.LX Q^WAn!9Y)o\%+ t`>Pa^ MX{7{5ltNFJ;lWX+[ Z_!1.Wz[zwu} ^o@zW޵0~Y[1KޗC ?ֿQ@W@ (! Xe3p|A4 nۢ-#aw_՟zNXm DyZU  s9 aw^E0@C8ûraԒ`_*1?l Xzfڏc!b?kYX[ kP臂YA&|P:^H}}1Z}K Z5jI8X֘Xmba7ރS_ZЦRJگ9ҷij/smL ba1ވܬS{{k_Gב]Sp˕FcWKkw}v]4ZB |SKL[7ui-1u`6{֋]ȸ=>LLž2vi/;5_X E)e9C9G_@JHuv˦FS͠|5Ԇ9.-嬉|ڹu=S{E3~ z0&b\Z׊(fx8Br[gFrQ}h/E1=6y S;?+nsG#ȌِñOrE^܇ȦX>#Z3b/&}?k E\Ӟu՛B~ ާaXµrB>5u>cw.0c,Qi?|~ֺA^2%p2.;!]^ҎRq??}gg4Rf0hK;(p2DI*Hu0QUSw(%"DRJD)!"""JQZDhx"ZkOӞgbx^ ,?Ř7|sn,q=\?` lz?p^XW1(?7QG:/).vP}X+';`hMuVzXmzȿN߉G{5uǜ]VXK`>wu.1$yHwbقn Ȝ\I/m-ti/Y=iyESCvj͹*zXjp Mdd4+SV-s=0zȼtK~~; sY*-cIdeOtOuNƾ-umK1;§nomnmdx}_oh)2 tnא˙vn[^ɾd]6wqZ駷n{ҏƬI>q79# |Xy=dX1SG6;lb+Klb/߇k9jܿ9"}Izff^3?;Y2Ţ;]뒻XǮ;yl5Ω}>ǝӝcsSSA"dSdx)y=7jʕՕb~Μ3wjA; Cwy3g{8)0mO=^ϙEkj*0|mtk(opW~s1H_yyk{܍~}&k9Qg?Bqs12W8&T"Bd MA߅n>:B&zڷY>s?{U{?aW컖?#l=^C5j\ +b[WIlՀ=jGZd}`O5oWb>[9>]$].e0g~&-kI&rKŚKk6{eաs 7*q#\5'^YOǟ9 WD#o=|2=oA*pYm&]Fe} D g/[Vkvr'2L׋w(78[X󑱰~ʘ1鿥)?c,k5rI~y_C/G19k/5ioW}=-K'AￋgW'|3 ÷` $g{^"㵛^>=_5ؿC["2Ts3qϚ;>;/3Fk^f};{6YaM?Sl5x[3Xt/2k<_عYU 3Q>< {d+%3]GƦ28+Eqdv ^qeޯSq_`:/˚gf;~'GqkYHxࣵNd .)ʥ^Sb_WGƢ_?cp6lC `c>| źk6Fc1׫v lFn}'G56} }/vЄxi7MAwndML3kLBc"Wۮ3;Bү#yztҽ+')Fm {؅`#έOm̞֗6e+`,Gy|:|+`:x na2~kɠg]"w~U8aF֕ڇfp1V1?w=+#0K8#ֶ"k>go!?(xwLC-z.~}lksf/ڶZc؎|žXCsS Z}+SZd,/l/H>ܓ2mcc%w?/"ߠlXódz@ b8+δ#}ѵ2T~)?9Gy9~ HD?cCX/mqɪ=W5{pn$h?`o3Bjb/(=Z *kکBz|WI\\v ]:t 50|qn20&q8pl=~-s^z @j`;2cx32WmѴ.?W:6ZY1H]cp֙;N/;s&]dl7(Yw#z[/q aqf&O+Z{U?B_oco}N%ol7oA>G/#}98f3mp"kq{#kl!_$_IMuW-c`lq"ƈc5"""ˆ1FDDDUDTT*DEUתj]f/ vl8_A_y hu0{|q֌b}s0nq^d:n`9lWnrk6]8@#s#|d웜/;9U/p O}UFǹ ~ƑCk@F#Vp\TWC1'h9x'Ȧ/u%cdW1ZNd <2n t#ROoE&xXd}إ/FᎩSub zUCW"O ڏ\gD9V6vkH-Fa+͗X4{mH #[#oWg=z=R8vF(p7iWKJ~wT[)?)ƼwG3[j1r*(7))2GΡZuGt=Sp̕!cqZS5]%rr[@Y/7̱g;[gFs>W6.U>\W9>YBWl[ϵ]q_TDoGILKV8g5.M~,Kw/go +Xc>gپ 2tr܅w]!8uN#26Ws 6},Sc5`eYZD^:ѝ_D|ӑ=&!g}*p|R!k|5{c7t)E.r/EuzuwMM+ on tx~Og?Yw~-E.`8߂_y ?/?fo Zj 9xYcu>ҟʩO oaJ2V;.߯yXd9Nwalfdz㚹gXk;7"{ ?Y{{1+]ȺYF oWY55_3_̰7;2}C2t} ;{o{~ oEcBKxA86/o V.F;1 %8]Kl*2UmGGg[~N\_FNч>g#:=fެWQd>{^#7"kSTO'V-{G^Ոw#t ;77'ڕ̵^]\nށsȾe7j%W =^Sw>F!/6arO+~υ>E.Am,1Y+?A߇5#kI~CB<7!Y_c g8#+ow?B֧[G+}hx۷Ս#<쯚맑5Q͟#sdr u:XsְqY~sx[۟Fஃ]/qY޵K:+5tp}E#zJ@6~>[gEքa-uQžB?W~]o>jb=GPDB'2v;/tRlg,sxY`uOXv5} 9cyd-Ql*>,Gmqp8pݍK7^oGu{{ ζۊ{l@F8fȷ S68Ż'ы~A:P=dEs6 ^zuk;W|\+)u#knd ^Թ䣭Q>328>>q,swEaȚFՏ;_N]OQk-M?Zx^eQa)V`Y 1Nd0V{CP{ #kF^"~g@= Wl'G.|d?{wXb]-t3ثU7;^aȇ7yxIhkF9K(73Y+k`8+?/eYgI6rG88"QGEEժQQUU5FըcUcjUķ5Zkժ1Xk?}{~p~_m$X(}徺J{/8woL{ob/rDϹ϶'Ff:;]}+8=!ah?>~?Jޫ2;w ^TEGWx'4q#bYnb]b.lކrsv]W{?ßN<\DḰa .2_r|׳$.x֠od+l7J=s$5ԤE~,b~6 mr;qpt{3 x(94^{]s<jśVzF Z?W1vL(slީ'gψ&}֢s&뫛l;u܀V=f>9:e(}yLtoFE.q!(_ʟC}x|=:E|3'w{Wd6}OgpU ~~w42HёgEŔcu2:T^[e]28Ww{Q>s{vp~}t85h-/y߭|W jq1Ⱦq2?tӹ%>|>5@glv޺Q@7p=A&ooAw]v;:;4:}{۬iXs]1zaz؝R'=`=JOz>By K0%D fOݟIsQb~j}nrofSz+3Qutn!ml'{(?N Ṅ|yĞgNPqe<|N/`KrŹr d}[vy 9>Oմkeɞq!{I Ybӿt?\C/+'Q_K31чwa=&2“eΆ~"dy;]N//ex{ dy¿t@W$|>q5F駖C/9RF>ociR> J߳˿l uu >_~[LCG9O߿E](=k{b9HoUC̵~ |7 #ʜo䇽( QN~w\E7E1zt`[. dzRSkgkvwuͶk trֆ}`)ej&˽Y¾l9Kv0tg"mƾkOַvyL'7akI΋ddvgN6{dYӲnTͳo㉦;5[g(l{9>csm6xl`?ǎ^,Y{֡:5~w_ޝO<{9wGɷv9#AuRڷOLJϝwKxzcгߺqNGy$LWy9V#l9{siMz1:9;8~i]}]8M׃ynwk_ֱ-Fɛֻs\mh'-8'9Gc_D^$[‹zVV\>}ݟLF{tg|Y.%5ԫ83Co0ȾXkq;iQ} o547ܻx6OT] k_w6uDfy*wүf4a*sLCnb|;u}pΠ_ɹvKȻƿ,G-ޟKw2 @ zyԪ1NW Qr{7ч5S3$|Q?na['4Wy3 =xXy3sȱhuG~LQf;[Ec-t:7?o=zCg7y: Sp5OA[:i<]8&]@c>>ok ]K/Q2yf뀣ۇowwєm! x.n|;c9?f]V([|Z.Cȧp@sٍⓒe2{sOkzQr(1m[F;)71txLM~Yjkع[ =FG\9D9 ^V ~A=g\dXǥuDD#"≈x*BDDDTUETDDDx"''"""JUUUDըQ5j?Fƪϥƚ?ڿ}}m{ιU;0~"yl| s{)R-ױ&e}Kd|p_Fdžl-$;ƔT7wM0z7w Z~e#lQl ?i4Z=Rq9DKPW:+OqܽH. rؾ^t쏣FDKo>4+7f<|q]:ýT*:U ˬmgC0 M.ڶK?Tֱ/]찙m#94S~H:=B# \%=Ne9=o5_ "Rnc CC5I=5l\{9HeZY6 :=?Gl ~%;#~=w Ǔr*zY[Lu%-W5w =[t#@k/dˑ(m V)vv ئZf#C zedv_TѭHu}K̻9yuy}G2&ȼ&z޸d7<[Oi<ǘsV6;yyk!gmf~W(g#YXo$ۙl?4ߋ]+q4ή!XlcʙL;me5wnVпn;_\fOۦ [ 5enlimhUU_>FZ}ك|Ng_<_8|BB/~`G? z]ŎTzs1Ek)?58ɺ\ݍ0O#qF k;n~C0dWFљA=h@KY[F 2Ad6<Ν쵂}ИOf0YnAwAoq] };>n2?[qFq&3z8mFX:QhU|z3)G1kg."{>){9Fu.6 pܣpb#K?Dzx\:lAt녾ϵn:M_2CΔkk*9alC<_`g2PLf̷#{wNG)xD&MB{[OfcP|$2|eLfg1 k঱gR~lv7,_7WnCVv:Cxy?oDʙucels _p6QfOn^m%_fwz ;_fns@9" z*^;wC9w#i|ڶO݃/mg,B#R\seH:,;ok֥=;d~ `F"3ؤ{wçp 7ͣb@6^f[{[;i!R,u[`^CLhu} fʙ j濾H=hL11WyfuQzfYt'/g|]#Zjm龦Ιgt-s'cXg[W;H+pa>^<B+wr&zHޱR`] UΕŵY:zu#'1hCvݕњ 3zHx۹.Bf}oҹ)o y^rwL}d.D!ct*B! fNa[އoGx5R˺1[O*ި{6wྸ#d[D>ڋ{o&R=)@oW[?<_T"ſ:ۺs.|f?*oA*2@W6S>/g38 zJwAEDE$.<֝tO? ?mMȫipv \z)(N_u~bW1ާr9?D3u g'߻E֗Yӫ>.cww v|"8 |>ϛt #hrY}5y^zx1VB +o9HȫP\G9d݅FIϷ~[#{Ok;]E Npwػ8Bg^ė^m.H4OGDRJRJD)DQ"""ZD%BZk"hsimXmcOkm99q-mk{hc|y}?W}zQv]dXT>{;*ܳ7 c IڶPa ߛJa~7)2Ι_6Pay<;DWp2}žr +6nEL3_34""7zܽ)lLw.X+ҹv{}=OӚ _ MqnisM:!2pnzŮ<5tn#=֝qlAaߊOG|>G_[̀em!s|wwl%qG>|w:ΏSKp yG<_.AKƌ!XOβ|{θ | }.{;mZiw'^k9=DCqW]d/c}j#}X}{{ԠLXB+tga {!:Q.8Gsd~wy‡.>=8Ug܏/j[S)==!<ƉqQ?7xc1q3il v8O}߂R';D%{5@{|*Яc96L|+oU8+#x ǜ}%̾}:\G||1~5?WW{|X,`+}rli.\}ocߋ<'"Os3I"x}RWDzŌUg2UnE.k#wzĹ.X}͎Y \^}˹l']Oܷ/YIƉb@*{I}dFyD}K O%a~muqFRf^¡>Ru3i*~,pn1\ bkٳU]l.ecڸ{kkI5h'܌{w lT VCuP}gXA%^TI!mcdۄjeJ>8G_濗0-re-,SEӽ3\z}Ĺ&u̥x+~nO%>Qq`Q{EO{بiW8mz'l&OpN s縨VmX,Yet_sLzdsȲ-^w/}"OȢQrx^w1Fȥ}#XG]=Lת]PVG|w,mg`(=ϠE{~ U;,rP=pTNT| ;hCP]E of}`5F9wwڋi ǶEl,r{n%ɧ9kն!6E^Tt9-U*U=IFuj={ :R.߷Ѫ^ElVM'nC^ʺf!Z4剎E_MT|f'dtn.wsGT>wS_ٷ#{",:Xt@~O徭,r߲皉գ7&ZocQ5_[jg{=gUxYs&Wruε-nUYEߣ35Բ?=\Ѳg5&|܃*ϪgܲseQtF)(~FaSH^L78,̅g0϶X} '~_B?h$^w$_QɣZ^G9 zyT/-|I>UW$ye/svgw곜,{]F:|I8зsC's +yCsR; \yR{|"d ְ1nCR7Ql*wsƭ<x_Bwl5ih@ǝDk͢ :{޵'},jJ}"UvXȑT_:?g',za>WIWŤ)iD.;Bg'|mݖWE."dNk >G4nb~yk>qٴWw oCV7`#XjOyEIsENE-Uw]oG,gv \E.Йጿ߲fE1%rU-јMߚW&O8n݅j}lvo'0?{y(%By""JDDD(JR(QJD%"<5hOk6,,f5YbV,Ff9uuN{==^kh?Wh|-yl˿Fh?0Bw=:?x|de{4M|)-7_M򞠹!5\xC ~{}a0 _/9˹!z< xd4c|.zGe]žYN{DZװ܅o0?Oc^3DktWT@}Μep-$?_=gzXZܿC/1ԋ>;󿾪s{:3Ϳzo];fZm>6V[:K^ o{Bx7Yt­|>E.v-M1N|u/ʷEٽp*%6;A?߅4 p~"VswihͲlOfcΩY!詞R(IJ}by".пd!c=dm3f.gg⿈>k#k ~+`dTl+$Nc]z]}wۦE)woE͢xluƼcQc/ͤl/qQU|)pWS^^~]Kx8I{+K5rU򜶞땤\wt>^Q|0SQ! 塰[dXj@3QU<ĚQz(xlWl<ʄ>Z2m6;̌DžJ?TkJ:_H4G }uu/ ,{|VA/x<_:/YWZWmQf}iujy9To&6~׫qD|kmsjâW 7X@{ELNw]f[C͋W,x욶OU-Y_2~Y/ݫmՌ[ku<$ZK\c/YdzE U/mu+ [ _|=)|W6}\3w D72r, d+gg,&PU\fc"eUsJoT;ߪyśo7kgt}N2.][:gZ{ߵ7ӷ\ׅ3zK>@jZ5s3/CŤ˸UCUNqv|xř :Kmsk;f2Y_u\ŕh]:E|Nj36x\c~EM4vs)zU0os/fs`,/c٤MYLKsnS{Lm<7(~߸_G6|5_O4k:@qvMoq>ی7'ؔu}&YkG?rY,?tC|%z ɖS||k]|)[]xpY',O=Gt[d`׾E:]=c:&]=A_d_~5|~+h9rwmBEO3:vy2bO:]fQZ]{9oݓ7k K0vՒ{W8s{*n ;,2痓mpfQdGq=_U{8:uj.'鄗rH@\G lXԼc]^ZSG?&>7u=bM;-|o3{G'M9[,rL5"7,~9r_^<Æ׽{ocz)-jX-js쵟q>T OupV u֔mZE:kְCy<6~-jAfu;l6}ȦMq=ns.x>&G>`~gWux?vÞ};gԛހxFEܡ׬V:WW. ,Jo:BnnE~f5,zx._5}duxťbW^beыnb=R\]眯k`*TWcxX[6m>ZIwyzX/çgqw- +Ȯ?⛇zʍm7ZM>'[9H>A7zOe=ȢNK,zIF EEoJLm\0{0eZ-,\97 s}39`M<$k^:7\>VmwBϓijmQ<мd+t}NsL ]-z59Lgd5$fq;es?~|Rm]߶xFo&v}F%q\"ZiZhZ"!"FĈcĈ1"q-5Fk>Z\a>>WUp:ߪ:u&\sK25d 4>Ä2o)V_7[oǛ#S,ޓC{=g[On+3צuzXA_Ǯ݇m8 EG sI9q #x+q }l]M =0}1} pöW𧼤cE {{g|-p d.𡀷>mKw=xi"晓klʭMS_{ ./k"7׳I_AyM} V&>]dܐmW/1wʕMx.hř-Ζk1n`Fպn>d!v-ew;Jԙnqv 6,bšz6sm 9_jG+WII/nSO#ڇmw_0W[Iyڳ8``۳Cp$x#&k-bȱ=NaG3no"ߊE*>t[?!«u?[ĥK8xΉl+[u;4ci?T*Y/-+v7F>]CWƶ,Ivu 5J:iWjQǭa`p}Pm?ZzwCL2K|&kfY|_kk6q~Cvjg tn$,jM:vY3Jz{J\sD~^6\̡SoŞ{}㻬]j[k|E /]'`U|zEOJZ}{WȜ}1߂~o\X)dOc>g .Qq;Y|'~Op8qsN2nO=ߗ3xGr?י37ŷcs~g+pvk˯:\m[u+XW"s 7~OL}A昱g3Gm#wJs/~v<>Z; ozWSPwn͢SSֽ8Q&},9dutU,%+ȭ[~ c|K|l1k< >_Kツȸgѳx_dvX:f{kQ6UG2mQ}aثަQE-W'5dct}{?B_Lǭk> ݗ?O'%/"Fu|I]>yiu2p=`|ҿCK_Xڋk1=WhK {UI6c\6GG|~\`ƁE[ށy==ȍQiqy\2ߢ-8X8s[: W:1?Әi|htF8?3M{ڢ?+1Vf}}tߘEQCֱ k9Ck:% s3uM|=l$s 9ͫIpŝ0CJIVg୙mݳ[<g*j_׾+%*~#+{_R$OxO>?)'ɮ&ymHuy9d8~R11FcDĈ1b#"FDDDUDTU8*:G/X\ǪZ^ՋZ{{ثͻz{>|y>: h6yWZ|d_so.=L-Dwx*}y.{ٕD1?}? ,XNlVJyj9vO1}Zv"?M֜%O&׊+3vKwا;l;æ>~\@5 ?Mt"1OU0B:~]s:o:& syc-q:kVqgz1^]İkV$VG1ƆPk ]7E-|mc]Qxsc~\o5o,"WY3q?i/gbߋ<{-c1C "a:}Ʌ{H|H =4 E<.>Q 7X1%'75W17] {|- }<~=8F߿B8Do{{~=WȶwvyNȚZć;g݊<{k/׊~XN{woK><߉mO"#b5Bf+{])`#r3vsfy :C[HL$_!U~ǿd|>qhtQ$?¦Ȟ! ԓvɛhkӘ~^m}n ٮ1ZɳJٽB6|^sm|/E.Ρ?"Ur;!z͡O9ZF~d +7=?mַ߱1;94:nDNHdLTt q+șy|w'"ڴw)^'#{qdEH.{ѝ/pњC߾3'{"CP s9'[=XZ]OO{sأf#a$.5V6lD3%2".Y-lZ/GbT_v^WifZCV[mAsDz,ƓHWZ3R>f$n9zU";|'~wd!9MI^9[Ԑ;~6\YR|ﶋXWE摧s=5qَS6f-dC^]%Fuy`.rαO>ȳ3d}O?gѹx  W| 3g]<^4$WS>=lx:#l8;xCzANq-Mh_Cs^x#>7sn\ιۮ>ٱyC7I^fd/l9=½M7rκ~futc;ijC#q{\>I%cj!9}@Ξ#9Oo^$Of]w{lѣ lR a}+dὈ-g<[9-[yNc'ΧYgz;!{ElD ?.twٿ; xt3&2ϑm/W@>D>D.|ʚzj-{֫ r=,b%9'u (n `|X 1bȞ彎; |v$Y"l7̎N3^FwI'"JDD -Ѣh#B-ZDhb1""F11֊1k/^{{k?~ݒߣjzg^?NSOթsvuO_l--.b#q9B+1.}ud `<@]!z4=s C#-%sYw'[ />m-j_9yw59 OhS{纮x W ;pUJjzw}> <>#s#;-X;W w̽mn/Ա|^fâg1rUbsߋ$g>?,})DV; roǿ[st,QޖY8CV>>08 -{ssb.t& W`Q+^Y9 n,r.ml/bg8_`yuO9_$u,2؅ĉε;% WD<|y`\z`xNw+c:xdq\ }ku  8Q~MóObյ+ŽIo}뾬Ir,r]59O]2^*_CU7{Jw+ƽwۢ`V-j=_;{aKwX=ľ͚+p]pGd/>hꝩM;αvZ#;lҲxWk 6z2o9҂&k+p:q]]Xgo-elѷ-bsVq:7pc_ğ][kW;ĺbQwu=h7ܖOkgjxU\ŖX8=gGY[7_2z!3EG:XEg]M|$1;pEwq]SY!o/b_f<Ga65OQȱ2} _.0+$<)36\]!Ϡ }^Cy.c1]gU7f 6>H:׾oQ{]o-ŬswzWp;ϫvkdTN)VގX_٭[f>e8ເFyl⽂l#a^`q?j K8+YNB)0O})k٫M}A3Ĺ 65eeeaQG-_H^@]@{ac7koF;oЏQi/ϋdlqǞ=C辬8L1mQ[|#Wn_s)n%;f]+Ur]i$;E_Z1~M߃+孟;sTxlk/߹Iګm0k0#sڢF^1k9:fQI|9/ǵz#|I.{S_be"#}uvG<+T[,kı?ߎ8MC6[?>v|~=lv|^kw"7k|m_no؃ω=j'0vGo[j,wCt?2Wp"#3{`QU,jΔMu@yUt,'Y8=܎<ȝ s ״>~x9_w3zu{QwY;H2t=›jBn;:#%V-[a^z.׉rq?$m}y.lV>BnCv6aWn癹u<>߿>,Gm"<<<2Sx{}h@;MRx)`+:4^5K, s@/=:lb6_C{:@2w߆w_)"꽤>^FzLUb/h] qXn{]p G<:w_s6Ǿ󘅟īm*n}kS6'KY;殮^^ %⭘iN,Տrt] gˌc<~~g/fK7< ~j6:Tol[Î&Zab.2et\9ExK]?f~{{a﹛a= ]l[ܵ!(8=b^V[C5U֪-?2{ЭW:>$nKp[=/:';MK@26BxnϊQx?N 7'_-Iv8ceF_,-K?ǞzK]Koe3կ,- syo/95l=>yn:6ZEM+[#ҭG<6d-~_!|/y}tйe| ns%c|-tj&qZk[2_"UޯyL@[~ lԿMܱㆥYkad-tW5nY}|eO.sE6=Ⱦʸt`9;`Կ[EAs_қEW-Օbެ;f`_3/};_fp K9~EѻL9tY楥ު޷4s}/}j`鏠w$kYz;k{K6 w~HW{n~9Kw&c[c_/ĩ !q_7.#,SMuk^iԽǸk)/}@Mգcқ8l`Do.X~g}pGFw .:!o=Yg_gk7*d˶\nࡷ༥ Ϲ@L{o ^s?uЂcaqKAW 69H'5Kzp6^m?M@ql豧sq<rK<ůx(g#cơ梋qA[F:#FE/c`GMP)YeK}̾o? / < .3N ( <}7p%+wؓmuh-e:`_9SKRO;tΐcޜ< dAK{㌋{.ϯjZw{äjJ!.7 P /:=o-֞oSVZ{p [&SJδ t'@)l}ҝ":x)VzIo|1 mg!6Tm[+| ʙYKoR LW =OYy|c} ?hw z[Z?5K=#t[z;\R,ťVs\mZ~l mKl{Rowz] Vݾʸ!O }o=~|~;Ώ{3g3!>q~d<pk/.??۰/ σ3%g~?e&~ nq vc~KvOٿW.rv'GOXa_e3<`-|k: a}Ow70~<Aw_by_W~oN5-{þ b?a%<2ʗE:?c -W}v.]=K[@CWzy=Tf&8{ew[σ7\G\늸""""qE1+""FFDHck֚NPbZk5kQVZUZUUU}U}}Nd/{~AtʻXY6ЃNu(M{_vPG7=fop^B=PuqyKyG1t E>'&s>Gz_6H,x{-פoGz~w^uxi-h_gSt}}W9qx[8W5\p;o~}=@kç]c`}˳1|>A.O]]tg=;au~U˳~gyOl϶r_[%b7Xo v.)daV(Ÿ7պwU?1˽"}dbyv]Da^u9kkg pqŧ@fMlP߆zoC#]bo _LM#hca"]7&:<9[.i|}d7k+~}xxy}A>ɖ 3(zL"g&c։O=;kan9'+?p4}˳ 8S0 qo8{5zX^;3lح[Qⴁ=ux>5h&o`סE RM7Fn1iX΍ͣCSb= V_܇,Z~}k@?tU|؂;F: ӱ/6-.sņh|ز܇ՙM+gk|}Z P9AB&=LϮo{r_qu`Y7-CS\)?_9*pw9칼?rn_&& % gv^WȒ.uFFÜK zۓ{={}szྂM{o%2@x xf;}nû1ߒga=k}Z|äpekL9o=<ma=$_-y/87-߇~:-oE+}h-'~`.-'ؿx 2}=lf=7u_zomrpUNZ{oo&p6#W7,䏃}=#x8xL>p 1Vl< b.*GS5kǠYg+g=<7=6~GVҡFLM4&TY|Nx6,W>y͌B4UBɕB]}˵oAm{߄Y> la!=W^G;kGͦq:W^B},gU@߳ ^lM輇v#FɁp޷|Y zݽwp]S˯,q9W/ۖ*~~;Z%}A9 1 >8›f^zbWQK:x!|^|CMz_~w{{O6r=}o9?W+ȞO$ʼy}'kKt?]C[^Y>,==V񷟷6}<rc{Esć<'W~+s{,'kpυ]t{{q)gصoky=F8`Gd^ Gѷm0zĽ@:ny  `rϲ 犝yD;|YNxior^̓A{x}U#;,[|OH?BGr%O=p>ʜQ-K;F^l ݎa8Eq7-{4@l6[״{U:S_{Wzprq-sujyt; 5;;wׅjbtʫX,Ow͢ϻuS7?rq9VSMUS\?K+I0a(9ޞY6}]&g~\?ݟ$ C߱K+WW% xTX*_:Jҷ]Wy Ub[$w/^JPmIUOR"c__- |H MKrҏ7CEi"lUR~"ÐHY|PP9$;.GrEꙊKvп|\J9Ukɭr4~XQ`̹&- i ",E)B6 AJ]ATQ "T0FԶQaAZѨFKR5hZ&Tۦ̝ ! ms9gT9E-x-_!q*ߦj djm\;}9q]%꾚5?ǖdr:ɖ{2uO|>(q9u}}?Ϙy|tnʻalzϨwNoiV&'jcUW[/r/?tRO^Ιs:ӝNK\?nљt?Tk:)C-(v7[ڑLt=vkGخo3K_ԧ6OWV9v\_}酾ޭ:ߞT&tZh֦҄/\_uFkz]jSr6''W\.{imxEڸM򁷿^Cwe *ʡIL.OO^LNOKtS6Ԙ%U4"yXo5>^u_eV|I~[ԤdƔxxy;?c;SB. U'}J7rO=pFvHiYS<鉚Lt\~U~vO4ۏ,/=ΓX)vww&\k2Y7xE'w67&<Gҏ4!kwym <6{6?*^<=#kڸ76i7OQ^iw s_<&yNɽRxIKs/լ[f7ɽO;:=u;Nft}$O׎rdMJjսS{nc}6[m*Oj.{04?dڜc5{nmt5=ǫꙍWrzVMOYMJDz_~u)7xU]J#}uA>:ל!㶱Ӽ Ky,~X]:9!}u+Ϗsm6d{:qzD+ç>7 {zZ>k]P \_sޱ]x~ *Ѳ8M+_(M,EG|yL]Xpǂ>\򅞎wF.)Y|']wyȄSsTR:9(nT˹yO.u=;Om9@!ސ ח4Fe%ꩳt߬qB\ZM=S yRB-y๏D=J^N=Xy\5\3:io[0cܶ1K<]ծZkg0OW=9m^W_r4[{Iכ=?qhԬ]͚+5U4M5X{۸nKM>U-^GZǽz1^vI\a?Ijcڒ!!Z Q4".mR}#I_/؎3g Fڛ|;Z;z`^v;=93gwf5pf:N|0n_,K;NO_|ԭ_Gli2.Y͔֠sH~8Fzx;C&YzG b#gx\E;&vZ+l31C2$=&S'o“>>>/Rotate 0/StructParents 1/Tabs/S/Type/Page>> endobj 2 0 obj <>stream HWm$fO*`XW0d 8>/?UR[#ڇ9OoR=US7c#~whds$E$l0z>iw 4?Dks|:8:輗Bg@4*+ )c]78K-hb͡{O%s]8Vy\ʝ·fs Nk([+8nԽao ث".aﮱg}sxB@7BD!`(.'tN^5`Qwb N>TFOJvZ#D"~8#7vHC#ÙR9SyOƨ%~GJ_Vp'Rʢtz.xE CgmfZw2Sw0 K."rT>܊j"Iu&0A/{]*FE~x='/ViB3+0у`Ћ=椹=0NȹsY< #ݨ"Ԣt|21P@ޑeJ -XFn\Bֲޣ4FQ ї7 a]]_e%ܻp KHf(4r2ZtG-AwPd'z=g>&~|#0Kz?Fр9 BѯMS!b ȡ{]8[H7j+0]),䂭ER%&Ή=C`њTV beqTnabݤB|ԎEJz>}UD%uP eVZB UY).7+mE&H'3u3>[2VRjf[iU# wyZ5ҠsB.ff1idM\ҍՀ_YvIJAܯ{Y}=wN$ZNu/| Q@^NˁA9+r[` p98pɹ~op[x\蜥QJTBJ!Cy3'|f0qKއC@yGbRpAɼk) _$DTy%wBu!aI&}LEtIOGeZ}/9k!0OԂҋ?Y4Qo)5[~AN3op Ti'XW5-r {+T,MtGZ.,UT{B_tC%ؿ׫V|aW71^r$:xҗ&B0=w%kKl lL ݆9N.nhy&X\EK,0~HXj7xvʹZN4{fZDH*얤V W„KR>=ePy+B`cȉaPB.3m{*]WլgJtI qB%mr\^_oSѯ!S} $ʹ<9d̹wJ_3<-m@ -ǫRˍ#hoO)]$_3k+j+T#(rZZH`iÕp k3$#ҼirDJt6&]۪UyaQIovH 6n;)Gd'cI.䮿!w=/@qT-H扖*7$8NDQ5 b~:?~YIËFy dBu7WŲX/V%)erLivV(6|`m;h~e\a *|IzSیюVU-uXez>>XmQf j¢xp򵸘n[>*oJ E}6󅤠NJ R?\hɖ]_w*swSL;Pܜlx+b 5i(^ Z5rQV"ÆŪR]޺\ԱU[>|]-_)lE""k.Jb}\c]-qg9&Sc8P1AKvk][c+czΒ-hg|mUlyF`Ћu38NDFSʵ!͑u 7[`qR%ψ_ji ЍTb~r!,% $$6 Jlu[Lrm﫪uߜ(UsC_7ocs3TCC+WP5g =3ϥ38ޘkbDr[{]?|xA.J# `Q mJɜG~ 'Л dѺy6a"/졸XK4,|4Re~ >lpa]'Y,z.4&Hߞ@z88(~ydRbBF1CT9vVZ4h8?[ Aڥg՚Vn^bsX),&e B zD)&I,檷Hs{A\&KqW f=Է{XX=!7Nmjo!EiI Gq@wQz**PAsb~ىuyǏH2gRy$#gÔhfdLd dכJXT88L_'Žhx_-v5u፸HK!aI],:t/{Q:o`o ^!%zƱهB.ڀ*͟ ^*=YI;>>>/Rotate 0/StructParents 2/Tabs/S/Type/Page>> endobj 4 0 obj <>stream HWiaixS glJdkAWXa"ݜrv%Vdczn?~+VvmUW{VAsa5cT X *\tpufV7ww^s|Ǐ-AfVw덢WۢWS/^~]Ŏ|d=r` =ڊ`\)gYZ;'B2F=-V1 VMj=[* @m6PKI9 r& sUVPҙTA0Lb8֢Z5[z Yu>koƌ+)1gnVZQs&Tc?" }H.ǹdnѡ6;z+r$opaD !IwoňYLSr'=ڢU-gޑ6쐬B?a̧)i1vkV"zՖt)p :ɦjЭW~HktYj5Ԭ%>qc^[b(&+ :/\j\+bj‹Œdq?A>zGd7VPwѼXF£?0&3Bf\6Sጊ0 Nf _zTh>!j =6=sNauQ SL-$‚\'5ҺJb{K>ү +3J&ڱ+}g$Ļbuy]+GVH=b` CCL  lU7qƵmGEKn5|K$vz`%zuJ#EV|9X|{IBRW0O2P8΁Éuw%$'T3HY5Vi$hUgv@r!g-j+r>_2\j??: ֞qfh9ajsL@̝63GcJgP3@<3I rWLR|_K9WE 9@]} c!ʇ'0j @Q,(<4F˷,63xL!t9Ot}oc+RHW>܂0 ^ƜG+!;G yߨ\]SġBA#>SQ$PiU$2lc!Oc'JGxCKSy 59z{P<" dOBEL+XDޅŌ\E1;O(u8+f$ß$2p-qbQPYp<:99u&(uI)o}y4x0Ai9I“-<ٜz*FhRAYZd 8wh1vT!25- 𘙽)b2VLy]6ε83&nቷpEw%{v'J4%A`F33.F8hxIY:q %սdYBp QqtUT7ψT ѕƮ5^YH#'qќih9x<e)>t/9E46 L 9т~2\Oo_uޯO<2ȘFVOEHۥ#,.y W^e3P=e,2H%6/٥řp*$"MhIƇQummo':3;ׇ훗7!fHq 1a [k7gWôH?N9zi o"m9WKBbitf{o >^02OK;+Iٔd~beufMZ g/pX,p^8"]x,T<1 'IVAO:@ZuŜ)@l_5ϋ9S46}NIr\Y1N-t<'X-HyrDfo/3LΈY]/ eqW F`J|[y1Q7/KՏl[ĴHLOcXrX$IDcOvhI×f :]TMDnwo wU`_胘ڸ$U&$Œ&ju|k Cݐ0L7WKo0 WibQ-EԱQa^ kKl)Wl.mDS B\ېٔ*F?DFtp7\mfYPټeul%}\%:}!E s"}]vω܈5Կr ^\!y4@Z U/:31eWUW͙" NgZˬ)-`f:ԪɌT`mQv4@Lh:։TM}%ʺx7%2SvWqf5w^;H 3# n'ΆZ"U9;_25p_: B|rp}EW27TtϽ?qW̬ɵYU-~7Fү)(+kJU8Sc ypA[G9(יUF9=Tnr}=YoRHB|{yab6vWKЩL .꠫5̯sZ:4 "MeWB_ UP7B2@jF4PCHFH8pquB3arxt{b3Ejp2d-r/br~56,=ܦ oH2ȬTЅQȜnPq)`In\" L}&OQ$~?t241sk @vmpxzáGtF>X  <쾲cr^N^;-`Cp0{Z[ 8^BƠ~ za{D @6x"Hke s&4kxI~}HN@bּَ|@2%@\Uݧb?e <LGb?7p}: y`rϐau(. ^*z/;nݛ R/ endstream endobj 5 0 obj <>stream hޔY[o+|yh;"@ܤH5b~P\D# |+.W+kãs#$% -d'+񂜈?&AQ(P:H)P X4hBa0J݁Z%FK)T\A1 gd8 yz!C ֤EEP < }bC;<C;%Dc!/%z) )Pw48gA=9 M* O }\$`>.xȳPC.ԞDA><}؅l96##ssٷZ`ȍ.eEX?|yV8mwzv߅L`sTӠӻf_MwtvZ~\oO__^.wݪ{^[wۧO~ :ipDnכ%R|ާgjg߇O]p\~Owm^zIoVYy~|NA+Q>W򳞡OV_7LӘV 6]' 5X5X746؝/8D V WW+;Q|Du jhl'op`R&Js:-M'}ĦRQP\=o_ M ʜwי7H` 6r.l_%L!|%nϬ$c'9_]&ۦF:IKAsU6KqQ*,8vJIFVr&of3׫gW\,ClCS[?8ߵKԓMR>E|R`|>e'Y)uWuӤ6Ddm=NƐIMB^aܚpQSSgQW)P9KW270M$yQW3L덗e<,]Qŕ_DP*wϏèEo):78Yid/Aq컫?ǂ~#;()Ī%G&#>0z.*]z$PMw. .5zo5S] Y1L'+z_.d_%/WmVџQ-b_<~V>#Ɨs o>stream hޔX[k%7 +~l6ǒ/aYHh!evۗt)MRRh}?ifvxxO7[1;j#9_qf< 9;bbw&H;K .Urt9߳U+Qq+8#r"@4YEyWU Q)+"^b7G nQhQ_ba!x-p"JV)`UI9(c 9DQ D̄rQ e* D8كP"8+3Ģe L@.d3sU52l"j<H]`O E AwK`?D f9iv ' g֕GW, tIJh ٌv UZ}{87鯻wL МKu3Rn?yA 7ʸ2lsG=8ujXOzERLVKugQO["߶ }&c* a{ddoOFZK322Z;2riHFkIn4Yi=̱9vcsU)n2^v%T&^NѮ,dFGKi)|%HK饣EZ_(+DRzZjK&k _6cf1kLi)۸h|8mnMVPWqѕ[]וvtP0読.*;FF{_I?2):g?[qwt/2#Mjtҡ7Ǖ^F6/t>hІ%ti0-%|0q7Kz`$4;(A^7T-o+~@ކv j=M ɆIs$'n5{ŷfd}PiDQk{?jٛoXxпVqLO8ѧ[іqQjJIܼ7X$bz/RZEu"vծfMs7 OI<_ˤo%5$YgDK-;k%ʜQ]SW nl%\hҗvKkEߪV)nbw>>R/9 ˯t%]x=7{pxwj٬g\kWWfK۝K/=Xw?/"] endstream endobj 7 0 obj <>stream hޜXMo7 +:x%)  $Ht60rk#gJh%a$,\0ufkW Ib~LcJzy= P3\`M/J A+,,R0Yn7@W+34 Uf1hn3^4@ -CJ'Cs΂.΂drsm=J}n47\@cn)oU}-9e3;CІ"I*7.N.*:^C&OW"k= b KY,,hn8:,hV ,sCuC;#2b^p!c)KaL K̇f'XG@۷a|]#P-W}L0Y̺eĤ2ؖ1bc90kshyqiow!qybyG^d𣃺\~}׷wv_~?#l>ra<,lx\>`r1iĔP0%oOA:b2oxw^z^us0LspO/tH\ גf)f1 R'p}P߱TK f6^>P͛_qQ٢:_f}&UftVO$ϴݴ94DƇ͠⨺<0fv8H ך.H.H.H.>Ѷ8Eϋyy%\"kYq8u ^M}(ɠsK4k)gM㍓uwLj+iS˞:wRYŻ6[ o] )Z eڇ΋7ԵIۚ;^1>+XN5QANNƹc<+shVy͝_qz0qBXeVfd4o3!|6uH)ه{::.srd_ +:a%ˬ jY!-vKҶ*>C'z̿ { endstream endobj 8 0 obj <>stream 2021-04-23T21:41:36+08:00 2021-04-23T21:41:35+08:00 2021-04-23T21:41:36+08:00 Acrobat PDFMaker 20 Word 版 uuid:a88a85fd-aabc-4287-b22f-5e03417c5fa4 uuid:e67fe1d5-be4e-4d0b-82d5-edf8578e8b05 39 application/pdf hbge Adobe PDF Library 20.9.95 D:20210423133851 endstream endobj 9 0 obj <>stream h260U0Pw/+Q0L)66)IcRYZlg`p endstream endobj 10 0 obj <>stream hlJ1ES,BwtK[Jaq)lF2"j*">e=cp.Pe=Cdyp/Tx>O<~$&BH6+ZLfT7 N~<5tZh` FwƏE5sGGm!8< tYc<[ʘE20WR=Nl9K endstream endobj 11 0 obj <>/Filter/FlateDecode/ID[<1B527C657927194EBBD3392A45B0B7BF>]/Info 306 0 R/Length 87/Root 308 0 R/Size 307/Type/XRef/W[1 3 1]>>stream hbb&Fr L O@"9H 3 "UM#) a`%GIjsGaMW@LKi 6 s endstream endobj startxref 116 %%EOF