From e4c0e34459b536edbaa21b1232e60752cf080db0 Mon Sep 17 00:00:00 2001 From: mtorres Date: Thu, 22 Oct 2026 12:55:19 -0700 Subject: [PATCH 1/5] feat(akida2/portrait128): person segmentation example with trained models Refs HUB-43 --- akida2/README.md | 19 + akida2/model_zoo/portrait128/README.md | 306 ++++++++++ akida2/model_zoo/portrait128/colab_setup.py | 82 +++ akida2/model_zoo/portrait128/data/.gitignore | 4 + .../portrait128/docs/README.md.template | 306 ++++++++++ .../model_zoo/portrait128/docs/metrics.json | 45 ++ .../model_zoo/portrait128/docs/zoo_card.json | 26 + .../model_zoo/portrait128/models/.gitignore | 4 + .../portrait128/portrait128_benchmark.py | 191 ++++++ .../model_zoo/portrait128/portrait128_data.py | 200 +++++++ .../model_zoo/portrait128/portrait128_eval.py | 139 +++++ .../portrait128/portrait128_model.py | 191 ++++++ .../portrait128/portrait128_notebook.py | 391 ++++++++++++ .../portrait128_notebook_benchmark.ipynb | 236 ++++++++ .../portrait128_notebook_training.ipynb | 556 ++++++++++++++++++ .../portrait128/portrait128_train.py | 139 +++++ .../portrait128/portrait128_train.sh | 63 ++ .../akida_unet_portrait128.h5 | 3 + .../akida_unet_portrait128_i8_w4_a4_qat.fbz | 3 + .../akida_unet_portrait128_i8_w4_a4_qat.h5 | 3 + .../akida_unet_portrait128_i8_w8_a8.fbz | 3 + .../akida_unet_portrait128_i8_w8_a8.h5 | 3 + akida2/model_zoo/portrait128/update_readme.py | 15 + 23 files changed, 2928 insertions(+) create mode 100644 akida2/model_zoo/portrait128/README.md create mode 100644 akida2/model_zoo/portrait128/colab_setup.py create mode 100644 akida2/model_zoo/portrait128/data/.gitignore create mode 100644 akida2/model_zoo/portrait128/docs/README.md.template create mode 100644 akida2/model_zoo/portrait128/docs/metrics.json create mode 100644 akida2/model_zoo/portrait128/docs/zoo_card.json create mode 100644 akida2/model_zoo/portrait128/models/.gitignore create mode 100644 akida2/model_zoo/portrait128/portrait128_benchmark.py create mode 100644 akida2/model_zoo/portrait128/portrait128_data.py create mode 100644 akida2/model_zoo/portrait128/portrait128_eval.py create mode 100644 akida2/model_zoo/portrait128/portrait128_model.py create mode 100644 akida2/model_zoo/portrait128/portrait128_notebook.py create mode 100644 akida2/model_zoo/portrait128/portrait128_notebook_benchmark.ipynb create mode 100644 akida2/model_zoo/portrait128/portrait128_notebook_training.ipynb create mode 100644 akida2/model_zoo/portrait128/portrait128_train.py create mode 100644 akida2/model_zoo/portrait128/portrait128_train.sh create mode 100644 akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128.h5 create mode 100644 akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.fbz create mode 100644 akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.h5 create mode 100644 akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w8_a8.fbz create mode 100644 akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w8_a8.h5 create mode 100644 akida2/model_zoo/portrait128/update_readme.py diff --git a/akida2/README.md b/akida2/README.md index e5d0bd1..4533e4b 100644 --- a/akida2/README.md +++ b/akida2/README.md @@ -162,6 +162,25 @@ benchmarks for every mapping, and the steps to reproduce them. 3.991 + + Portrait segmentation + Segmentation + Portrait128 + 8-bit + 0.8861 binary IoU + — + — + — + + + + 4-bit QAT + 0.9091 binary IoU + — + — + — + + diff --git a/akida2/model_zoo/portrait128/README.md b/akida2/model_zoo/portrait128/README.md new file mode 100644 index 0000000..1f21567 --- /dev/null +++ b/akida2/model_zoo/portrait128/README.md @@ -0,0 +1,306 @@ +BrainChip Dev Hub + +# Portrait Segmentation (Portrait128) — Akida 2 + +## Model Card + +Float binary IoU: **0.8883**  |  Float pixel accuracy: **94.31%**  |  Parameters: **1,058,865** + +The model labels every pixel of a 128 × 128 portrait as person or background. Two +metrics are reported, both on the validation split and with each pixel's logit +thresholded at 0 (a probability of 0.5): + +- **Binary IoU**: the intersection over union of the predicted and true masks, + averaged over the two classes (person and background). This is the headline + metric: it penalizes a mask that is the wrong shape even when most pixels are right. +- **Pixel accuracy**: the fraction of pixels labeled correctly. Background covers + much of each image, so this is always high and says less about mask quality. + +The quantized variants below all share the same float model. On Akida 2 the +model is quantized with **`quantizeml`**: 8-bit weights and activations need no +quantization-aware training (QAT), while a lower-precision 4-bit variant (4-bit +weights and activations, 8-bit input layer) uses QAT to recover accuracy — 4-bit +PTQ accuracy is poor, so only the QAT result is reported. + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
VariantWeights / ActsQATQuantized IoUQuantized pixel acc.Akida IoUAkida pixel acc.Sparsity
8-bitw8 / a8-0.886194.19%0.886194.19%TBD
4-bitw4 / a4yes0.909195.48%0.909195.48%TBD
+ +**Akida 2 hardware benchmark (FPGA @ 25 MHz)** + +Latency is measured on the Akida 2 FPGA reference platform, which runs at +**25 MHz**. A projected latency at a higher target clock is also shown to +indicate expected performance on faster silicon. The cycle count is fixed for a +given model and mapping regardless of clock rate, so the projection is an exact +rescale of the measured cycles. + +> **Note:** the projected clock is BrainChip's current **target** for AKD2500 production +> silicon — that silicon does not exist yet, so treat this as a target rather than a +> measured value; it may still change before production. Power measurement on the FPGA +> platform is still under development, so only latency is reported at this time. + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
VariantMappingNPsPassesCyclesLatency @ 25 MHz (ms)Projected @ 1000 MHz (ms) (target)
8-bitMinimalTBDTBDTBDTBDTBD
AllNPsTBDTBDTBDTBDTBD
HwPrTBDTBDTBDTBDTBD
4-bit (QAT)MinimalTBDTBDTBDTBDTBD
AllNPsTBDTBDTBDTBDTBD
HwPrTBDTBDTBDTBDTBD
+ +### Benchmark plots + +📝 Planned: the full-model and per-layer benchmark plots for each variant are added +when the FPGA benchmark is run with `--save-metrics` (see +[Contributing and Maintenance](#contributing-and-maintenance)). + +The model is an **Akida U-Net** from `akida_models`: an **AkidaNet** encoder with +width multiplier **alpha = 0.5**, initialized from ImageNet-trained weights, takes the +**128 × 128** input down to a 4 × 4 feature map. A decoder of five separable +transposed-convolution blocks upsamples it back to 128 × 128, and a 1 × 1 +convolution produces one logit per pixel. There are no skip connections between +encoder and decoder. Transposed convolutions are an Akida 2 feature, so this model +has no Akida 1 version. + +## Requirements + +For environment requirements and setup, see the [Requirements](../../../README.md#requirements) +section of the top-level README. + +## Dataset + +Portrait128 is a binary person-segmentation dataset of **18,698** RGB portraits at +**128 × 128**, each with a person/background mask. It was assembled by Anil Sathyan +for the [Portrait-Segmentation](https://github.com/anilsathyan7/Portrait-Segmentation) +project from the PFCN portrait dataset, hand-picked portraits from the Supervisely +person dataset, and selfies collected from the web whose masks were generated with a +DeepLab model. It also includes augmented and synthetic-background variants. + +The dataset ships as a single pool. The scripts split it 80 % / 20 % into training and +validation, with a fixed shuffle seed, so every run sees the same validation images. + + + +Reference: Shen et al., *Automatic Portrait Segmentation for Image Stylization*, +Computer Graphics Forum (Eurographics), 2016 (the PFCN dataset). + +## Dataset setup + +The dataset is distributed by the Portrait-Segmentation project as +[Portseg_128](https://drive.google.com/file/d/1UBLzvcqvt_fin9Y-48I_-lWQYfYpt_6J/view) +(Google Drive). The scripts expect its two arrays, `img_uint8.npy` (images) and +`msk_uint8.npy` (masks, 0 or 255), in one folder. + +The scripts default to looking for the data at `./data/portrait128`. If you +want to store the dataset on a dedicated data drive, you can pass the path +explicitly to each script (see `--data` / `-d` in the individual scripts). +Alternatively, it may be more convenient to keep the dataset in its preferred +location and create a symbolic link from the default path (one-off step): + +```bash +ln -s /path/to/your/data/portrait128 ./data/portrait128 +``` + +This way the scripts work out of the box without any extra arguments. + +### Calibration samples + +Quantization with `quantizeml` is calibrated on real data, so a batch of 1024 +Portrait128 samples is needed in addition to the dataset itself. A prepared batch is +published on the BrainChip mirror, and `portrait128_train.sh` fetches it into `data/` +automatically: + +```bash +wget -N https://data.brainchip.com/dataset-mirror/samples/portrait128/portrait128_batch1024.npz -P data/ +``` + +The training notebook does not use this file. It builds its samples directly +from the training split with `get_samples()` from +[portrait128_data.py](portrait128_data.py), so that the process is visible and easy +to adapt to your own data. The same module can also write a pack from your local +copy of the dataset: + +```bash +python portrait128_data.py --extract data/portrait128_batch1024.npz +``` + +## Pipeline + +Training produces a float model, then quantizes it with `quantizeml` into +several variants, each converted to Akida format: + +| Stage | Description | +|---|---| +| Full-precision | Float32 training of the full model, starting from an ImageNet-trained AkidaNet encoder | +| 8-bit quantization | `quantizeml quantize` to 8-bit weights and activations (8-bit input), calibrated on real samples; no QAT required | +| 4-bit quantization | `quantizeml quantize` to 4-bit weights and activations (8-bit input), calibrated on real samples, with QAT fine-tuning — 4-bit PTQ accuracy is poor so only the QAT model is kept | +| Conversion to Akida | Automated conversion of each quantized model to Akida 2 format with `cnn2snn convert` | + +## Reference Models + +📝 Planned: pretrained models will be made available in a `pretrained_models/` +folder once the training pipeline has been run. They are handled using the `git-lfs` +package (git large file storage); for set-up instructions, see the +[Trained models](../../../README.md#trained-models) section of the top-level README. + +## Usage + +### Notebook + +Two notebooks are provided that walk through a) preparation of a trained Akida-compatible model and +b) evaluation and benchmarking of that model on Akida. + +[portrait128_notebook_training.ipynb](portrait128_notebook_training.ipynb) walks through the +complete training pipeline end-to-end. It is written to expose and explain the Akida-specific +aspects of the workflow: how a segmentation model is built for Akida 2, what the +quantization constraints mean in practice, and what the conversion step does. Start +here if you want to understand *why* the pipeline is structured the way it is. + +[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/Brainchip-Inc/brainchip_devhub/blob/main/akida2/model_zoo/portrait128/portrait128_notebook_training.ipynb) + +[portrait128_notebook_benchmark.ipynb](portrait128_notebook_benchmark.ipynb) walks through +evaluation of model accuracy on Akida and, if a hardware device is available, covers benchmarking +of model latency. + +> **Note:** the hardware benchmark section requires a physical Akida 2 FPGA +> platform with a connected board. + +### Script + +For straightforward reproduction of the training and evaluation results, run +the full pipeline in one shot: + +```bash +bash portrait128_train.sh [DATADIR] +``` + +The optional `DATADIR` argument overrides the default dataset location +(`./data/portrait128`). + +## Contributing and Maintenance + +This README is autogenerated from `docs/README.md.template` +so that the accuracy and hardware benchmark values are written directly +by the code (via the `metrics.json` file, also in the docs folder). + +When the associated model or training pipeline is modified to improve +performance, you should rerun the evaluations of the float and quantized +model versions, plus the hardware benchmark, including the +`--save-metrics` argument, and then regenerate the README from the template +using `update_readme.py`: +```bash +# Float model +python portrait128_eval.py -l pretrained_models/akida_unet_portrait128.h5 --save-metrics + +# 8-bit variant +python portrait128_eval.py -l pretrained_models/akida_unet_portrait128_i8_w8_a8.h5 --save-metrics +python portrait128_eval.py -l pretrained_models/akida_unet_portrait128_i8_w8_a8.fbz --save-metrics +python portrait128_benchmark.py -l pretrained_models/akida_unet_portrait128_i8_w8_a8.fbz --save-metrics + +# 4-bit variant (QAT) +python portrait128_eval.py -l pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.h5 --save-metrics +python portrait128_eval.py -l pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.fbz --save-metrics +python portrait128_benchmark.py -l pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.fbz --save-metrics + +python update_readme.py +``` +Then commit the changed files (template, metrics and updated README). + +Likewise, if you want to edit the contents of this README, you should +not edit it directly, but instead edit `docs/README.md.template` and +then regenerate the README using +``` bash +python update_readme.py +``` diff --git a/akida2/model_zoo/portrait128/colab_setup.py b/akida2/model_zoo/portrait128/colab_setup.py new file mode 100644 index 0000000..7936d03 --- /dev/null +++ b/akida2/model_zoo/portrait128/colab_setup.py @@ -0,0 +1,82 @@ +# Copyright 2026 Brainchip Holdings Ltd. Apache 2.0 License +"""Colab setup for the Portrait128 (Akida 2) training notebook. + +IF YOU ARE RUNNING THIS LOCALLY: you can ignore this file completely. +It exists solely to make the "Open in Colab" badge work, and does nothing +on a normal local run. It is not part of the Portrait128 model/training code +(see portrait128_data.py, portrait128_model.py, portrait128_train.py for that). + +If you ARE on Colab, this is the file that gets you running: + - Clones this repo (skipping Git LFS smudge, since pretrained weights + aren't needed for the default training path) + - Points Python at the right folders (repo root for brainchip_utils, + this example's folder for portrait128_data / portrait128_model / + portrait128_train) + - Installs akida_models, tf_keras, quantizeml + - Checks for the Portrait128 arrays. The dataset is distributed through + Google Drive rather than a direct download link, so it is not fetched + automatically: copy img_uint8.npy and msk_uint8.npy into DATA_DIR (for + example from your mounted Google Drive) before running the data cells. + +Called from the notebook's first cell like: + import colab_setup + colab_setup.setup() +""" +import os +import subprocess +import sys + +REPO_URL = 'https://github.com/Brainchip-Inc/brainchip_devhub.git' +REPO_DIR = 'brainchip_devhub' +EXAMPLE_SUBDIR = 'akida2/model_zoo/portrait128' + +DATA_DIR = './data/portrait128' +DATA_FILES = ('img_uint8.npy', 'msk_uint8.npy') +DATASET_PAGE = 'https://drive.google.com/file/d/1UBLzvcqvt_fin9Y-48I_-lWQYfYpt_6J/view' + + +def _run(cmd): + print(f'$ {cmd}') + subprocess.run(cmd, shell=True, check=True) + + +def setup(): + """Set up a fresh Colab session to run this notebook. No-op if not on Colab.""" + if 'google.colab' not in sys.modules: + print('Not running on Colab \u2014 nothing to do. (This step only matters ' + 'for Colab; local runs already have everything they need.)') + return + + if not os.path.exists(REPO_DIR): + # Skip Git LFS smudge: pretrained weights aren't needed for the default + # (RUN_FLOAT_TRAINING = True) path, so avoid downloading them here. + os.environ['GIT_LFS_SKIP_SMUDGE'] = '1' + _run(f'git clone --depth 1 {REPO_URL} {REPO_DIR}') + + os.chdir(os.path.join(REPO_DIR, EXAMPLE_SUBDIR)) + + # Repo root on sys.path for `brainchip_utils`; example folder for the + # local portrait128_data / portrait128_model / portrait128_train modules + # imported later in the notebook. + repo_root = os.path.abspath(os.path.join(os.getcwd(), '..', '..', '..')) + sys.path.insert(0, repo_root) + sys.path.insert(0, os.getcwd()) + + # Akida 2 quantization uses quantizeml (installed as a dependency of + # akida_models / cnn2snn, but pinned here for a clean Colab environment). + _run('pip install -q akida_models==1.14.0 tf_keras quantizeml') + + os.makedirs(DATA_DIR, exist_ok=True) + missing = [f for f in DATA_FILES if not os.path.exists(os.path.join(DATA_DIR, f))] + if missing: + print(f'Dataset not found: copy {", ".join(missing)} into ' + f'{os.path.abspath(DATA_DIR)}.') + print(f'The arrays come from the Portseg_128 archive: {DATASET_PAGE}') + print('To copy them from Google Drive, mount it first:\n' + ' from google.colab import drive; drive.mount("/content/drive")') + else: + print('Dataset already present at', DATA_DIR) + + print(f'Colab setup complete. Working directory: {os.getcwd()}') + print('If TensorFlow was just installed/upgraded, restart the runtime ' + '(Runtime > Restart session) and re-run this cell before continuing.') diff --git a/akida2/model_zoo/portrait128/data/.gitignore b/akida2/model_zoo/portrait128/data/.gitignore new file mode 100644 index 0000000..647613e --- /dev/null +++ b/akida2/model_zoo/portrait128/data/.gitignore @@ -0,0 +1,4 @@ +# Git to Ignore everything in this directory +* +# Except this .gitignore file +!.gitignore \ No newline at end of file diff --git a/akida2/model_zoo/portrait128/docs/README.md.template b/akida2/model_zoo/portrait128/docs/README.md.template new file mode 100644 index 0000000..08026d7 --- /dev/null +++ b/akida2/model_zoo/portrait128/docs/README.md.template @@ -0,0 +1,306 @@ +BrainChip Dev Hub + +# Portrait Segmentation (Portrait128) — Akida 2 + +## Model Card + +Float binary IoU: **{float_iou}**  |  Float pixel accuracy: **{float_acc}**  |  Parameters: **{params}** + +The model labels every pixel of a 128 × 128 portrait as person or background. Two +metrics are reported, both on the validation split and with each pixel's logit +thresholded at 0 (a probability of 0.5): + +- **Binary IoU**: the intersection over union of the predicted and true masks, + averaged over the two classes (person and background). This is the headline + metric: it penalizes a mask that is the wrong shape even when most pixels are right. +- **Pixel accuracy**: the fraction of pixels labeled correctly. Background covers + much of each image, so this is always high and says less about mask quality. + +The quantized variants below all share the same float model. On Akida 2 the +model is quantized with **`quantizeml`**: 8-bit weights and activations need no +quantization-aware training (QAT), while a lower-precision 4-bit variant (4-bit +weights and activations, 8-bit input layer) uses QAT to recover accuracy — 4-bit +PTQ accuracy is poor, so only the QAT result is reported. + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
VariantWeights / ActsQATQuantized IoUQuantized pixel acc.Akida IoUAkida pixel acc.Sparsity
8-bitw8 / a8-{w8a8_quant_iou}{w8a8_quant_acc}{w8a8_akida_iou}{w8a8_akida_acc}{w8a8_sparsity}
4-bitw4 / a4yes{w4a4_qat_quant_iou}{w4a4_qat_quant_acc}{w4a4_qat_akida_iou}{w4a4_qat_akida_acc}{w4a4_qat_sparsity}
+ +**Akida 2 hardware benchmark (FPGA @ 25 MHz)** + +Latency is measured on the Akida 2 FPGA reference platform, which runs at +**25 MHz**. A projected latency at a higher target clock is also shown to +indicate expected performance on faster silicon. The cycle count is fixed for a +given model and mapping regardless of clock rate, so the projection is an exact +rescale of the measured cycles. + +> **Note:** the projected clock is BrainChip's current **target** for AKD2500 production +> silicon — that silicon does not exist yet, so treat this as a target rather than a +> measured value; it may still change before production. Power measurement on the FPGA +> platform is still under development, so only latency is reported at this time. + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
VariantMappingNPsPassesCyclesLatency @ 25 MHz (ms)Projected @ 1000 MHz (ms) (target)
8-bitMinimal{w8a8_minimal_nps}{w8a8_minimal_passes}{w8a8_minimal_cycles}{w8a8_minimal_latency_ms}{w8a8_minimal_projected_ms}
AllNPs{w8a8_allnps_nps}{w8a8_allnps_passes}{w8a8_allnps_cycles}{w8a8_allnps_latency_ms}{w8a8_allnps_projected_ms}
HwPr{w8a8_hwpr_nps}{w8a8_hwpr_passes}{w8a8_hwpr_cycles}{w8a8_hwpr_latency_ms}{w8a8_hwpr_projected_ms}
4-bit (QAT)Minimal{w4a4_qat_minimal_nps}{w4a4_qat_minimal_passes}{w4a4_qat_minimal_cycles}{w4a4_qat_minimal_latency_ms}{w4a4_qat_minimal_projected_ms}
AllNPs{w4a4_qat_allnps_nps}{w4a4_qat_allnps_passes}{w4a4_qat_allnps_cycles}{w4a4_qat_allnps_latency_ms}{w4a4_qat_allnps_projected_ms}
HwPr{w4a4_qat_hwpr_nps}{w4a4_qat_hwpr_passes}{w4a4_qat_hwpr_cycles}{w4a4_qat_hwpr_latency_ms}{w4a4_qat_hwpr_projected_ms}
+ +### Benchmark plots + +📝 Planned: the full-model and per-layer benchmark plots for each variant are added +when the FPGA benchmark is run with `--save-metrics` (see +[Contributing and Maintenance](#contributing-and-maintenance)). + +The model is an **Akida U-Net** from `akida_models`: an **AkidaNet** encoder with +width multiplier **alpha = 0.5**, initialized from ImageNet-trained weights, takes the +**128 × 128** input down to a 4 × 4 feature map. A decoder of five separable +transposed-convolution blocks upsamples it back to 128 × 128, and a 1 × 1 +convolution produces one logit per pixel. There are no skip connections between +encoder and decoder. Transposed convolutions are an Akida 2 feature, so this model +has no Akida 1 version. + +## Requirements + +For environment requirements and setup, see the [Requirements](../../../README.md#requirements) +section of the top-level README. + +## Dataset + +Portrait128 is a binary person-segmentation dataset of **18,698** RGB portraits at +**128 × 128**, each with a person/background mask. It was assembled by Anil Sathyan +for the [Portrait-Segmentation](https://github.com/anilsathyan7/Portrait-Segmentation) +project from the PFCN portrait dataset, hand-picked portraits from the Supervisely +person dataset, and selfies collected from the web whose masks were generated with a +DeepLab model. It also includes augmented and synthetic-background variants. + +The dataset ships as a single pool. The scripts split it 80 % / 20 % into training and +validation, with a fixed shuffle seed, so every run sees the same validation images. + + + +Reference: Shen et al., *Automatic Portrait Segmentation for Image Stylization*, +Computer Graphics Forum (Eurographics), 2016 (the PFCN dataset). + +## Dataset setup + +The dataset is distributed by the Portrait-Segmentation project as +[Portseg_128](https://drive.google.com/file/d/1UBLzvcqvt_fin9Y-48I_-lWQYfYpt_6J/view) +(Google Drive). The scripts expect its two arrays, `img_uint8.npy` (images) and +`msk_uint8.npy` (masks, 0 or 255), in one folder. + +The scripts default to looking for the data at `./data/portrait128`. If you +want to store the dataset on a dedicated data drive, you can pass the path +explicitly to each script (see `--data` / `-d` in the individual scripts). +Alternatively, it may be more convenient to keep the dataset in its preferred +location and create a symbolic link from the default path (one-off step): + +```bash +ln -s /path/to/your/data/portrait128 ./data/portrait128 +``` + +This way the scripts work out of the box without any extra arguments. + +### Calibration samples + +Quantization with `quantizeml` is calibrated on real data, so a batch of 1024 +Portrait128 samples is needed in addition to the dataset itself. A prepared batch is +published on the BrainChip mirror, and `portrait128_train.sh` fetches it into `data/` +automatically: + +```bash +wget -N https://data.brainchip.com/dataset-mirror/samples/portrait128/portrait128_batch1024.npz -P data/ +``` + +The training notebook does not use this file. It builds its samples directly +from the training split with `get_samples()` from +[portrait128_data.py](portrait128_data.py), so that the process is visible and easy +to adapt to your own data. The same module can also write a pack from your local +copy of the dataset: + +```bash +python portrait128_data.py --extract data/portrait128_batch1024.npz +``` + +## Pipeline + +Training produces a float model, then quantizes it with `quantizeml` into +several variants, each converted to Akida format: + +| Stage | Description | +|---|---| +| Full-precision | Float32 training of the full model, starting from an ImageNet-trained AkidaNet encoder | +| 8-bit quantization | `quantizeml quantize` to 8-bit weights and activations (8-bit input), calibrated on real samples; no QAT required | +| 4-bit quantization | `quantizeml quantize` to 4-bit weights and activations (8-bit input), calibrated on real samples, with QAT fine-tuning — 4-bit PTQ accuracy is poor so only the QAT model is kept | +| Conversion to Akida | Automated conversion of each quantized model to Akida 2 format with `cnn2snn convert` | + +## Reference Models + +📝 Planned: pretrained models will be made available in a `pretrained_models/` +folder once the training pipeline has been run. They are handled using the `git-lfs` +package (git large file storage); for set-up instructions, see the +[Trained models](../../../README.md#trained-models) section of the top-level README. + +## Usage + +### Notebook + +Two notebooks are provided that walk through a) preparation of a trained Akida-compatible model and +b) evaluation and benchmarking of that model on Akida. + +[portrait128_notebook_training.ipynb](portrait128_notebook_training.ipynb) walks through the +complete training pipeline end-to-end. It is written to expose and explain the Akida-specific +aspects of the workflow: how a segmentation model is built for Akida 2, what the +quantization constraints mean in practice, and what the conversion step does. Start +here if you want to understand *why* the pipeline is structured the way it is. + +[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/Brainchip-Inc/brainchip_devhub/blob/main/akida2/model_zoo/portrait128/portrait128_notebook_training.ipynb) + +[portrait128_notebook_benchmark.ipynb](portrait128_notebook_benchmark.ipynb) walks through +evaluation of model accuracy on Akida and, if a hardware device is available, covers benchmarking +of model latency. + +> **Note:** the hardware benchmark section requires a physical Akida 2 FPGA +> platform with a connected board. + +### Script + +For straightforward reproduction of the training and evaluation results, run +the full pipeline in one shot: + +```bash +bash portrait128_train.sh [DATADIR] +``` + +The optional `DATADIR` argument overrides the default dataset location +(`./data/portrait128`). + +## Contributing and Maintenance + +This README is autogenerated from `docs/README.md.template` +so that the accuracy and hardware benchmark values are written directly +by the code (via the `metrics.json` file, also in the docs folder). + +When the associated model or training pipeline is modified to improve +performance, you should rerun the evaluations of the float and quantized +model versions, plus the hardware benchmark, including the +`--save-metrics` argument, and then regenerate the README from the template +using `update_readme.py`: +```bash +# Float model +python portrait128_eval.py -l pretrained_models/akida_unet_portrait128.h5 --save-metrics + +# 8-bit variant +python portrait128_eval.py -l pretrained_models/akida_unet_portrait128_i8_w8_a8.h5 --save-metrics +python portrait128_eval.py -l pretrained_models/akida_unet_portrait128_i8_w8_a8.fbz --save-metrics +python portrait128_benchmark.py -l pretrained_models/akida_unet_portrait128_i8_w8_a8.fbz --save-metrics + +# 4-bit variant (QAT) +python portrait128_eval.py -l pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.h5 --save-metrics +python portrait128_eval.py -l pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.fbz --save-metrics +python portrait128_benchmark.py -l pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.fbz --save-metrics + +python update_readme.py +``` +Then commit the changed files (template, metrics and updated README). + +Likewise, if you want to edit the contents of this README, you should +not edit it directly, but instead edit `docs/README.md.template` and +then regenerate the README using +``` bash +python update_readme.py +``` diff --git a/akida2/model_zoo/portrait128/docs/metrics.json b/akida2/model_zoo/portrait128/docs/metrics.json new file mode 100644 index 0000000..7ab4ce8 --- /dev/null +++ b/akida2/model_zoo/portrait128/docs/metrics.json @@ -0,0 +1,45 @@ +{ + "float_acc": "94.31%", + "float_iou": "0.8883", + "params": "1,058,865", + "w4a4_qat_akida_acc": "95.48%", + "w4a4_qat_akida_iou": "0.9091", + "w4a4_qat_allnps_cycles": "TBD", + "w4a4_qat_allnps_latency_ms": "TBD", + "w4a4_qat_allnps_nps": "TBD", + "w4a4_qat_allnps_passes": "TBD", + "w4a4_qat_allnps_projected_ms": "TBD", + "w4a4_qat_hwpr_cycles": "TBD", + "w4a4_qat_hwpr_latency_ms": "TBD", + "w4a4_qat_hwpr_nps": "TBD", + "w4a4_qat_hwpr_passes": "TBD", + "w4a4_qat_hwpr_projected_ms": "TBD", + "w4a4_qat_minimal_cycles": "TBD", + "w4a4_qat_minimal_latency_ms": "TBD", + "w4a4_qat_minimal_nps": "TBD", + "w4a4_qat_minimal_passes": "TBD", + "w4a4_qat_minimal_projected_ms": "TBD", + "w4a4_qat_quant_acc": "95.48%", + "w4a4_qat_quant_iou": "0.9091", + "w4a4_qat_sparsity": "TBD", + "w8a8_akida_acc": "94.19%", + "w8a8_akida_iou": "0.8861", + "w8a8_allnps_cycles": "TBD", + "w8a8_allnps_latency_ms": "TBD", + "w8a8_allnps_nps": "TBD", + "w8a8_allnps_passes": "TBD", + "w8a8_allnps_projected_ms": "TBD", + "w8a8_hwpr_cycles": "TBD", + "w8a8_hwpr_latency_ms": "TBD", + "w8a8_hwpr_nps": "TBD", + "w8a8_hwpr_passes": "TBD", + "w8a8_hwpr_projected_ms": "TBD", + "w8a8_minimal_cycles": "TBD", + "w8a8_minimal_latency_ms": "TBD", + "w8a8_minimal_nps": "TBD", + "w8a8_minimal_passes": "TBD", + "w8a8_minimal_projected_ms": "TBD", + "w8a8_quant_acc": "94.19%", + "w8a8_quant_iou": "0.8861", + "w8a8_sparsity": "TBD" +} diff --git a/akida2/model_zoo/portrait128/docs/zoo_card.json b/akida2/model_zoo/portrait128/docs/zoo_card.json new file mode 100644 index 0000000..c0d03ba --- /dev/null +++ b/akida2/model_zoo/portrait128/docs/zoo_card.json @@ -0,0 +1,26 @@ +{ + "domain": "Image", + "order": 30, + "rows": [ + { + "task": "Portrait segmentation", + "category": "Segmentation", + "dataset": "Portrait128", + "variant": "8-bit", + "metric_key": "w8a8_akida_iou", + "metric_label": "binary IoU", + "bench_prefix": "w8a8_", + "notes": "" + }, + { + "task": "Portrait segmentation", + "category": "Segmentation", + "dataset": "Portrait128", + "variant": "4-bit QAT", + "metric_key": "w4a4_qat_akida_iou", + "metric_label": "binary IoU", + "bench_prefix": "w4a4_qat_", + "notes": "" + } + ] +} diff --git a/akida2/model_zoo/portrait128/models/.gitignore b/akida2/model_zoo/portrait128/models/.gitignore new file mode 100644 index 0000000..647613e --- /dev/null +++ b/akida2/model_zoo/portrait128/models/.gitignore @@ -0,0 +1,4 @@ +# Git to Ignore everything in this directory +* +# Except this .gitignore file +!.gitignore \ No newline at end of file diff --git a/akida2/model_zoo/portrait128/portrait128_benchmark.py b/akida2/model_zoo/portrait128/portrait128_benchmark.py new file mode 100644 index 0000000..687dfb8 --- /dev/null +++ b/akida2/model_zoo/portrait128/portrait128_benchmark.py @@ -0,0 +1,191 @@ +#!/usr/bin/env python +# Copyright 2026 Brainchip Holdings Ltd. Apache 2.0 License +""" +Portrait128 hardware benchmark for Akida 2. + +Runs a latency benchmark on an Akida Portrait128 segmentation model, prints a +summary, checks per-layer activation sparsity, and generates summary plots. + +Benchmark details: + * The Akida 2 reference hardware is an FPGA running at 25 MHz, so latency is + measured at that clock. + * Because the FPGA runs at a low clock, a PROJECTED latency at the AKD2500 + target clock is also reported. Cycle count is fixed for a given model and + mapping regardless of clock speed, so projected_latency_ms = + cycles / target_clock. + * Power is not measured. The FPGA power-measurement path is still under + development, so this script reports latency only. + +Example +------- + python portrait128_benchmark.py -l models/akida_unet_portrait128_i8_w8_a8.fbz +""" +import argparse +import json +import pathlib +import sys +import time + +import numpy as np +import akida +from akida_models.sparsity import compute_sparsity + +from portrait128_data import get_samples +from brainchip_utils.hardware_utils import (get_mapping_stats, get_akida_device, + per_layer_benchmark, full_model_benchmark, + AKIDA_CLOCKS_HZ) +from brainchip_utils.plot_utils import (plot_full_model_results, plot_per_layer_results, + pretty_print_sparsity) + +# Measured clock: Akida 2 reference hardware is an FPGA clocked at 25 MHz. +MEASURED_CLOCK = AKIDA_CLOCKS_HZ['AKIDA2_FPGA'] + +# Projected clock: latency is also projected to the AKD2500 target silicon clock, to +# indicate expected performance on production hardware. Cycle count is clock-independent, +# so the projection is exact given the target frequency. AKD2500 production silicon does +# not exist yet, so this is BrainChip's current internal target (1 GHz), not a measured +# value -- it may still change before production. See brainchip_utils.hardware_utils.AKIDA_CLOCKS_HZ. +PROJECTED_CLOCK = AKIDA_CLOCKS_HZ['AKD2500'] + + +if __name__ == '__main__': + parser = argparse.ArgumentParser( + description='Hardware latency benchmark for an Akida 2 Portrait128 model') + parser.add_argument('-l', '--loadmodel', required=True, + help='Model to load (.fbz akida model)') + parser.add_argument('-d', '--data', default='./data/portrait128', + help='Portrait128 folder (contains img_uint8.npy and msk_uint8.npy)') + parser.add_argument('--save-metrics', action='store_true', + help='Write benchmark values to metrics.json') + args = parser.parse_args() + + NUM_SAMPLES = 100 + + # ------------------------------------------------------------------------- + # Model + # ------------------------------------------------------------------------- + ak_model = akida.Model(args.loadmodel) + imsize = tuple(ak_model.input_shape) + + # ------------------------------------------------------------------------- + # Device + # ------------------------------------------------------------------------- + device = get_akida_device(target_version=ak_model.ip_version) + if device is None: + print("No compatible Akida hardware device found. Skipping benchmarking") + sys.exit(0) + + # ------------------------------------------------------------------------- + # Sample + # ------------------------------------------------------------------------- + # Processing in Akida is activity dependent (because it exploits sparsity) + # and that activity is dependent on the input. That makes it imperative to + # use real inputs when benchmarking Akida, rather than synthetic random + # samples. + samples = get_samples(args.data, imsize, num_samples=NUM_SAMPLES) + + # ------------------------------------------------------------------------- + # Full-model benchmark (latency only) + # ------------------------------------------------------------------------- + # NOTE: power measurement is intentionally not requested here -- the FPGA + # power path is still under development, so we report latency only. When a + # power path exists, this is where full_model_benchmark's power fields would + # be surfaced. + map_modes = ['Minimal', 'AllNps', 'HwPr'] + full_results = dict() + for mm in map_modes: + map_mode = getattr(akida.MapMode, mm) + print(f'\nRunning full-model benchmark (MapMode={mm})...') + res = full_model_benchmark(ak_model, device, samples, + map_mode=map_mode, + clock_freq=MEASURED_CLOCK) + + # Projected latency at a higher target clock. Cycle count (mean_inf_clk) + # is clock-independent, so this is an exact rescale, not an estimate of + # host overhead. + res['projected_clk_ms'] = res['mean_inf_clk'] / PROJECTED_CLOCK * 1000 + full_results[mm] = res + + # Re-map without hw_only to populate ak_model.sequences for stats + ak_model.map(device, mode=map_mode) + num_nps, num_passes, num_sequences = get_mapping_stats(ak_model) + full_results[mm]['num_nps'] = num_nps + full_results[mm]['num_passes'] = num_passes + print(f' Mapping: {num_nps} NP(s), {num_passes} pass(es), {num_sequences} sequence(s)') + print(f' Measured latency @ {MEASURED_CLOCK/1e6:.0f} MHz: ' + f'{res["mean_clk_ms"]:.3f} ms') + print(f' Projected latency @ {PROJECTED_CLOCK/1e6:.0f} MHz: ' + f'{res["projected_clk_ms"]:.3f} ms') + if num_sequences > 1: + print('WARNING: note, model not completely mapped to hardware') + + # ------------------------------------------------------------------------- + # Per-layer benchmark. Minimal mapping mode, batch-size 1 + # ------------------------------------------------------------------------- + ak_model.map(device, mode=akida.MapMode.Minimal, hw_only=True) + ak_model.summary() + + # Check sparsity per-layer + sparsity_dict = compute_sparsity(ak_model, samples=samples) + pretty_print_sparsity(sparsity_dict) + + print(f'Running per-layer benchmark ({NUM_SAMPLES} samples)...') + per_layer_results = per_layer_benchmark(ak_model, device, samples, + repeats=NUM_SAMPLES, + clock_freq=MEASURED_CLOCK) + + # ------------------------------------------------------------------------- + # Plots + # ------------------------------------------------------------------------- + # Map without hw_only so ak_model.sequences is available for plot_mapping + ak_model.map(device, mode=akida.MapMode.Minimal) + # Plot filenames include the model stem so that each variant (8-bit, 4-bit) + # keeps its own reference figures. + model_stem = pathlib.Path(args.loadmodel).stem + perlayer_savepath = f'benchmark_results_layers_{model_stem}.png' + if args.save_metrics: + perlayer_savepath = pathlib.Path(__file__).parent / 'docs' / ('ref_' + perlayer_savepath) + plot_per_layer_results(per_layer_results, ak_model, sparsity_dict, + model_name=args.loadmodel, + example='Portrait128 segmentation (Akida 2)', + savepath=perlayer_savepath) + print('\nPer-layer results plot saved to ' + str(perlayer_savepath)) + + full_savepath = f'benchmark_results_full_{model_stem}.png' + if args.save_metrics: + full_savepath = pathlib.Path(__file__).parent / 'docs' / ('ref_' + full_savepath) + plot_full_model_results(full_results, ak_model, device, + model_name=args.loadmodel, + example='Portrait128 segmentation (Akida 2)', + savepath=full_savepath, + show_power=False) # power measurement not yet available + print('Full model results plot saved to ' + str(full_savepath)) + + if args.save_metrics: + # Updates the stored metrics used to generate the README performance + # tables. For code maintenance only, run against the pretrained models. + # + # Keys are variant-prefixed to match the multi-variant v2 README table. + # The variant is inferred from the loaded .fbz filename. Power keys are + # intentionally absent: benchmarking is latency-only until the FPGA power + # path exists. + stem = pathlib.Path(args.loadmodel).stem + if 'i8_w8_a8' in stem: + variant = 'w8a8' + elif 'i8_w4_a4' in stem: + variant = 'w4a4_qat' + else: + variant = 'w8a8' # fallback; benchmarking is only meaningful for a quantized .fbz + + metrics_path = pathlib.Path(__file__).parent / 'docs' / 'metrics.json' + metrics = json.loads(metrics_path.read_text()) if metrics_path.exists() else {} + metrics[f'{variant}_sparsity'] = f'{np.mean(list(sparsity_dict.values())) * 100:.2f}%' + for mm, res in full_results.items(): + mode = mm.lower() + metrics[f'{variant}_{mode}_nps'] = str(res['num_nps']) + metrics[f'{variant}_{mode}_passes'] = str(res['num_passes']) + metrics[f'{variant}_{mode}_cycles'] = f'{res["mean_inf_clk"]:.0f}' + metrics[f'{variant}_{mode}_latency_ms'] = f'{res["mean_clk_ms"]:.3f}' + metrics[f'{variant}_{mode}_projected_ms'] = f'{res["projected_clk_ms"]:.3f}' + metrics_path.write_text(json.dumps(metrics, indent=4) + '\n') + print(f'Metrics saved to {metrics_path}') diff --git a/akida2/model_zoo/portrait128/portrait128_data.py b/akida2/model_zoo/portrait128/portrait128_data.py new file mode 100644 index 0000000..3983585 --- /dev/null +++ b/akida2/model_zoo/portrait128/portrait128_data.py @@ -0,0 +1,200 @@ +#!/usr/bin/env python +# Copyright 2026 Brainchip Holdings Ltd. Apache 2.0 License +""" +Data loading for the Portrait128 person-segmentation dataset. + +The dataset is two NumPy arrays in one folder: + * img_uint8.npy: RGB portraits, shape (N, 128, 128, 3), uint8 + * msk_uint8.npy: person masks, shape (N, 128, 128, 1), uint8, 255 = person, + 0 = background + +For training, both arrays are loaded into RAM (about 1.2 GB); calibration +samples are read straight from disk. + +The dataset ships as a single pool, so it is split here: 80 % training and 20 % +validation, by a shuffle with a fixed seed (SPLIT_SEED). That seed is +independent of the training seed, so every script and notebook sees the same +validation images whatever --seed is used for training. + +Images are delivered as uint8 in [0, 255]: the model rescales them itself. Masks +are delivered as int32 0/1 labels with the same spatial shape as the image. + +Run as a script, it writes a calibration-sample pack for `quantizeml quantize`: + + python portrait128_data.py --extract data/portrait128_batch1024.npz +""" + +import argparse +import os + +import numpy as np +import tensorflow as tf +import tf_keras as keras +from tf_keras.utils import set_random_seed + +IMAGES_FILE = 'img_uint8.npy' +MASKS_FILE = 'msk_uint8.npy' + +VALIDATION_SPLIT = 0.2 +# Fixed seed for the train/validation split, independent of the training seed +SPLIT_SEED = 1 + + +def _load_arrays(data_path): + """Memory-map the image and mask arrays.""" + x = np.load(os.path.join(data_path, IMAGES_FILE), mmap_mode='r') + y = np.load(os.path.join(data_path, MASKS_FILE), mmap_mode='r') + return x, y + + +def _check_shape(x, input_shape): + if tuple(x.shape[1:]) != tuple(input_shape): + raise ValueError(f'Portrait128 images have shape {tuple(x.shape[1:])}, ' + f'but the model expects {tuple(input_shape)}.') + + +def _split_indices(num_samples, validation_split=VALIDATION_SPLIT, random_state=SPLIT_SEED): + """Shuffle sample indices and split them into training and validation sets. + + Args: + num_samples (int): number of samples in the dataset. + validation_split (float, optional): fraction used for validation. Defaults to 0.2. + random_state (int, optional): seed of the shuffle. Defaults to SPLIT_SEED. + + Returns: + np.ndarray, np.ndarray: training indices, validation indices + """ + rng = np.random.default_rng(random_state) + indices = np.arange(num_samples) + rng.shuffle(indices) + + validation_size = int(validation_split * num_samples) + return indices[validation_size:], indices[:validation_size] + + +def get_data(data_path, input_shape, batch_size, dtype=tf.uint8, seed=42): + """ Loads Portrait128 data. + + Training images are augmented with random horizontal flips, translations + (up to 10 % in each direction) and zooms (up to 20 %). Each image and its + mask are stacked into one 4-channel tensor before augmentation, so both + receive exactly the same geometric transform. + + Args: + data_path (str): folder containing img_uint8.npy and msk_uint8.npy + input_shape (tuple): input image shape (height, width, channels) + batch_size (int): the batch size + dtype (tf.dtypes.DType, optional): input data type. Defaults to tf.uint8. + seed (int, optional): seed for shuffling and augmentation. Defaults to 42. + + Returns: + tf.data.Dataset, tf.data.Dataset: training dataset, validation dataset + """ + set_random_seed(seed) + + # Load both arrays into RAM. Masks are stored as 0/255; integer division + # turns them into 0/1 labels once, here. + x, y = _load_arrays(data_path) + _check_shape(x, input_shape) + x = np.asarray(x) + y = np.asarray(y) // 255 + train_indices, val_indices = _split_indices(len(x)) + + data_augmentation = keras.Sequential([ + keras.layers.RandomFlip("horizontal", seed=seed), + keras.layers.RandomTranslation(height_factor=0.1, width_factor=0.1, seed=seed), + keras.layers.RandomZoom(0.2, seed=seed), + ]) + + def cast_data(image, label): + image = tf.cast(image, dtype) + label = tf.cast(label, tf.int32) + return image, label + + def data_generator(indices): + for i in indices: + yield x[i], y[i] + + def apply_data_augmentation(image, mask): + # Concatenate image and mask along the last axis + combined = tf.concat([image, mask], axis=-1) + # Apply the same augmentation to both image and mask + augmented_combined = data_augmentation(combined) + # Split the augmented combined tensor back into image and mask + augmented_image, augmented_mask = tf.split(augmented_combined, [3, 1], axis=-1) + + return augmented_image, augmented_mask + + output_signature = (tf.TensorSpec(shape=x.shape[1:], dtype=tf.uint8), + tf.TensorSpec(shape=y.shape[1:], dtype=tf.uint8)) + + train_dataset = tf.data.Dataset.from_generator( + generator=lambda: data_generator(train_indices), + output_signature=output_signature) + + validation_dataset = tf.data.Dataset.from_generator( + generator=lambda: data_generator(val_indices), + output_signature=output_signature) + + SHUFFLE_BUFFER_SIZE = 1000 + train_batches = (train_dataset + .shuffle(SHUFFLE_BUFFER_SIZE, seed=seed) + .map(apply_data_augmentation) + .map(cast_data, num_parallel_calls=tf.data.AUTOTUNE, deterministic=True) + .batch(batch_size) + .prefetch(buffer_size=tf.data.AUTOTUNE)) + + validation_batches = (validation_dataset + .map(cast_data, num_parallel_calls=tf.data.AUTOTUNE, deterministic=True) + .batch(batch_size) + .prefetch(buffer_size=tf.data.AUTOTUNE)) + + # The datasets come from Python generators, so their length is unknown to + # tf.data. Declaring it lets Keras show per-epoch progress and lets callers + # use len(). + train_batches = train_batches.apply( + tf.data.experimental.assert_cardinality(-(-len(train_indices) // batch_size))) + validation_batches = validation_batches.apply( + tf.data.experimental.assert_cardinality(-(-len(val_indices) // batch_size))) + + return train_batches, validation_batches + + +def get_samples(data_path, input_shape, num_samples=1024): + """ Loads image samples from the training split as a numpy array. + + No augmentation is applied. Suitable for model calibration and testing. + + Args: + data_path (str): folder containing img_uint8.npy + input_shape (tuple): input image shape (height, width, channels) + num_samples (int): number of samples to return. Defaults to 1024. + + Returns: + np.ndarray: array of shape (num_samples, height, width, channels), dtype uint8 + """ + x = np.load(os.path.join(data_path, IMAGES_FILE), mmap_mode='r') + _check_shape(x, input_shape) + train_indices, _ = _split_indices(len(x)) + + # The training indices are already in shuffled order; sort the selection so + # the memory-mapped reads are sequential. + selection = np.sort(train_indices[:num_samples]) + return np.asarray(x[selection], dtype=np.uint8) + + +if __name__ == '__main__': + parser = argparse.ArgumentParser( + description='Write Portrait128 calibration samples to an .npz file') + parser.add_argument('--extract', required=True, + help='Output .npz path') + parser.add_argument('-d', '--data', default='./data/portrait128', + help='Portrait128 folder (contains img_uint8.npy and msk_uint8.npy)') + parser.add_argument('-n', '--num_samples', type=int, default=1024, + help='Number of training-split images to save') + args = parser.parse_args() + + samples = get_samples(args.data, (128, 128, 3), num_samples=args.num_samples) + # Saved as a single 'data' array, the layout `quantizeml quantize --samples` reads + np.savez(args.extract, data=samples) + print(f'{len(samples)} samples saved to {args.extract}') diff --git a/akida2/model_zoo/portrait128/portrait128_eval.py b/akida2/model_zoo/portrait128/portrait128_eval.py new file mode 100644 index 0000000..f0e6e70 --- /dev/null +++ b/akida2/model_zoo/portrait128/portrait128_eval.py @@ -0,0 +1,139 @@ +#!/usr/bin/env python +# Copyright 2026 Brainchip Holdings Ltd. Apache 2.0 License +""" +Portrait128 evaluation for tf_keras or akida models. + +Reports the binary IoU (mean of the person and background IoUs) and the pixel +accuracy on the validation split, both thresholded at a logit of 0. + +Example +------- + python portrait128_eval.py -d ./data/portrait128 -l models/akida_unet_portrait128.h5 +""" +import argparse +import json +import pathlib +import numpy as np +import tensorflow as tf + +from tqdm import tqdm + +import akida + +from cnn2snn import load_quantized_model + +from portrait128_data import get_data +from portrait128_train import get_metrics +from brainchip_utils.hardware_utils import get_akida_device + +tf.config.experimental.enable_op_determinism() + + +# --------------------------------------------------------------------------- +# Evaluation on Akida +# --------------------------------------------------------------------------- +def evaluate_akida_model(akida_model, val_dataset): + """Run inference with an Akida model and return (binary IoU, pixel accuracy).""" + device = get_akida_device(target_version=akida_model.ip_version) + if device is not None: + akida_model.map(device, mode=akida.MapMode.Minimal) + print('Running inference on Akida hardware device') + akida_model.summary() + + iou_metric, acc_metric = get_metrics() + + # Akida can't directly digest the tensorflow dataset, we need to + # manually iterate over the dataset to deliver inputs as numpy arrays. + # The Akida output already has the mask's (B, H, W, 1) shape, so it can be + # compared with the labels directly: no squeeze is needed. + for batch, label_batch in tqdm(val_dataset, desc="Evaluating on Akida"): + logits_batch = akida_model.predict(batch.numpy().astype(np.uint8)) + iou_metric.update_state(label_batch, logits_batch) + acc_metric.update_state(label_batch, logits_batch) + + binary_iou = float(iou_metric.result()) + accuracy = float(acc_metric.result()) + print(f'Akida binary IoU: {binary_iou:.4f}, pixel accuracy: {accuracy:.4f}') + return binary_iou, accuracy + + +if __name__ == '__main__': + parser = argparse.ArgumentParser() + parser.add_argument('-l', '--loadmodel', required=True, + help='Model to load (.h5 tf_keras or .fbz akida model)') + parser.add_argument('-d', '--data', default='./data/portrait128', + help='Portrait128 folder (contains img_uint8.npy and msk_uint8.npy)') + parser.add_argument('--save-metrics', action='store_true', + help='Write IoU and accuracy (and param count for .h5) to metrics.json') + args = parser.parse_args() + + # --------------------------------------------------------------------------- + # Model + # --------------------------------------------------------------------------- + if args.loadmodel.endswith('.h5'): + model = load_quantized_model(args.loadmodel) + model.compile(metrics=get_metrics()) + isakida = False + imsize = model.input_shape[1:] + elif args.loadmodel.endswith('.fbz'): + model = akida.Model(args.loadmodel) + isakida = True + imsize = tuple(model.input_shape) + + # --------------------------------------------------------------------------- + # Data loading + # --------------------------------------------------------------------------- + train_ds, val_ds = get_data(args.data, imsize, batch_size=32) + + # --------------------------------------------------------------------------- + # Evaluation + # --------------------------------------------------------------------------- + if isakida: + binary_iou, accuracy = evaluate_akida_model(model, val_ds) + else: + results = model.evaluate(val_ds, verbose=0, return_dict=True) + binary_iou, accuracy = results['binary_iou'], results['accuracy'] + print(f'Validation binary IoU: {binary_iou:.4f}, pixel accuracy: {accuracy:.4f}') + + # --------------------------------------------------------------------------- + # Persist metrics + # --------------------------------------------------------------------------- + if args.save_metrics: + # Updates the stored metrics used to generate the README performance + # tables. For code maintenance only, run against the pretrained models. + # + # This example has two quantized variants, disambiguated by filename. + # Each variant contributes "quantized" metrics (from its .h5) and "akida" + # metrics (from its .fbz): + # akida_unet_portrait128.h5 -> float_iou, float_acc, params + # akida_unet_portrait128_i8_w8_a8.{h5,fbz} -> w8a8_quant_* / w8a8_akida_* + # akida_unet_portrait128_i8_w4_a4_qat.{h5,fbz} -> w4a4_qat_quant_* / w4a4_qat_akida_* + # (The 4-bit PTQ model is a throwaway on the way to QAT -- it is never + # evaluated or stored, so it has no metrics.) + metrics_path = pathlib.Path(__file__).parent / 'docs' / 'metrics.json' + metrics = json.loads(metrics_path.read_text()) if metrics_path.exists() else {} + iou_str = f'{binary_iou:.4f}' + acc_str = f'{accuracy * 100:.2f}%' + stem = pathlib.Path(args.loadmodel).stem + + # Determine the variant prefix from the filename. + if 'i8_w8_a8' in stem: + variant = 'w8a8' + elif 'i8_w4_a4' in stem: + variant = 'w4a4_qat' + else: + variant = None # float model + + if variant is None: + # Float model: record float metrics and parameter count. + metrics['float_iou'] = iou_str + metrics['float_acc'] = acc_str + metrics['params'] = f'{model.count_params():,}' + elif isakida: + metrics[f'{variant}_akida_iou'] = iou_str + metrics[f'{variant}_akida_acc'] = acc_str + else: + metrics[f'{variant}_quant_iou'] = iou_str + metrics[f'{variant}_quant_acc'] = acc_str + metrics_path.write_text(json.dumps(metrics, indent=4) + '\n') + print(f'Metrics saved to {metrics_path}') diff --git a/akida2/model_zoo/portrait128/portrait128_model.py b/akida2/model_zoo/portrait128/portrait128_model.py new file mode 100644 index 0000000..aa80533 --- /dev/null +++ b/akida2/model_zoo/portrait128/portrait128_model.py @@ -0,0 +1,191 @@ +#!/usr/bin/env python +# Copyright 2026 Brainchip Holdings Ltd. Apache 2.0 License +""" +Create an Akida U-Net model for Portrait128 person segmentation, targeting the +Akida 2 platform. + +The model is a binary segmentation network on 128 x 128 RGB input. Its output is +a 128 x 128 x 1 map of logits: one value per pixel, positive for "person" and +negative for "background". + +Architecture: + * Encoder: AkidaNet (`akida_models.akidanet_imagenet`) with width multiplier + alpha=0.5 and no classification top. Five stride-2 stages take the 128 x 128 + input down to a 4 x 4 x 512 feature map. + * Decoder: five separable transposed-convolution blocks + (`sepconv_transpose_block`), each doubling the spatial size, back up to + 128 x 128, with Dropout(0.5) after each block. Pointwise weights in the + decoder carry an L2 penalty of 4e-5. + * Head: a 1 x 1 convolution to one channel, with no activation (logits). + +There are no skip connections between encoder and decoder. Upsampling with +transposed convolutions is an Akida 2 feature: this model cannot be built for +Akida 1. + +Inputs are uint8 in [0, 255]. The encoder's `input_scaling=(128, -1)` adds a +Rescaling layer that maps them to [-1, 1) inside the model, so the same uint8 +data pipeline feeds the float, quantized and Akida models. At conversion the +rescaling is folded into the Akida input layer. + +Encoder initialization: with `base_weights='imagenet'` the encoder starts from +BrainChip's ImageNet-trained AkidaNet 0.5 (Akida 2) float weights, loaded by +layer name; the decoder and head start from random weights. That is how the +training pipeline builds the model. Convolution weights don't depend on the +input size, so weights trained at 224 x 224 load directly into this +128 x 128 model. Without `base_weights` the whole model starts from random +weights. + +Usage: + python portrait128_model.py [-bw imagenet] [-s OUTPUT_PATH] +""" + +import argparse + +from tf_keras import Model, regularizers +from tf_keras.layers import Dropout +from tf_keras.utils import set_random_seed + +from cnn2snn import set_akida_version, AkidaVersion +from akida_models import akidanet_imagenet +from akida_models.layer_blocks import sepconv_transpose_block, conv_block +from akida_models.model_io import get_model_path +from akida_models.utils import fetch_file + +# ImageNet-trained AkidaNet 0.5 float weights (Akida 2) from the akida_models model store +IMAGENET_WEIGHTS = 'akidanet_imagenet_224_alpha_0.5.h5' +IMAGENET_WEIGHTS_HASH = 'd818dec7c924757de2e6970b996da9c807ccdf9a9044262091ce526678c0640c' + + +def _fetch_imagenet_weights(): + """Download (or reuse the cached copy of) the ImageNet AkidaNet 0.5 weights.""" + with set_akida_version(AkidaVersion.v2): + url, fname, file_hash = get_model_path('akidanet', model_name_v2=IMAGENET_WEIGHTS, + file_hash_v2=IMAGENET_WEIGHTS_HASH) + return fetch_file(url, fname=fname, file_hash=file_hash, cache_subdir='models') + + +def build_portrait128_model(input_shape=(128, 128, 3), + alpha=0.5, + input_scaling=(128, -1), + base_weights=None, + seed=42): + """Instantiates the Akida U-Net architecture for Portrait128. + + Args: + input_shape (tuple, optional): input shape tuple. Defaults to (128, 128, 3). + alpha (float, optional): controls the width (number of filters) of the model. + Defaults to 0.5. + input_scaling (tuple, optional): scale factor and offset to apply to inputs. + Defaults to (128, -1). Following Akida convention, the scale factor is + used as a divisor. + base_weights (str, optional): weights loaded by layer name after the model + is built: 'imagenet' for the ImageNet AkidaNet 0.5 encoder weights, or the + path to an .h5 file. Defaults to None (random initialization). + seed (int, optional): random seed for weight initialization. Defaults to 42. + + Returns: + tf_keras.Model: a Keras Model instance. + """ + set_random_seed(seed) + + # Transposed convolutions are only available on Akida 2 + with set_akida_version(AkidaVersion.v2): + # Define weight regularization, will apply to pointwise weights of sepconv transposed layers + weight_regularizer = regularizers.l2(4e-5) + + # Create an AkidaNet network without top layers + encoder = akidanet_imagenet(input_shape=input_shape, + alpha=alpha, + include_top=False, + input_scaling=input_scaling) + + # Add the decoder layers + x = encoder.layers[-1].output + x = sepconv_transpose_block(x, + filters=int(512 * alpha), + kernel_size=(3, 3), + strides=2, + padding='same', + kernel_initializer='he_normal', + add_batchnorm=True, + relu_activation='ReLU7.5', + name='sepconv_t_0', + pointwise_regularizer=weight_regularizer) + x = Dropout(0.5)(x) + x = sepconv_transpose_block(x, + filters=int(256 * alpha), + kernel_size=(3, 3), + strides=2, + padding='same', + kernel_initializer='he_normal', + add_batchnorm=True, + relu_activation='ReLU7.5', + name='sepconv_t_1', + pointwise_regularizer=weight_regularizer) + x = Dropout(0.5)(x) + x = sepconv_transpose_block(x, + filters=int(128 * alpha), + kernel_size=(3, 3), + strides=2, + padding='same', + kernel_initializer='he_normal', + add_batchnorm=True, + relu_activation='ReLU7.5', + name='sepconv_t_2', + pointwise_regularizer=weight_regularizer) + x = Dropout(0.5)(x) + x = sepconv_transpose_block(x, + filters=int(64 * alpha), + kernel_size=(3, 3), + strides=2, + padding='same', + kernel_initializer='he_normal', + add_batchnorm=True, + relu_activation='ReLU7.5', + name='sepconv_t_3', + pointwise_regularizer=weight_regularizer) + x = Dropout(0.5)(x) + x = sepconv_transpose_block(x, + filters=int(32 * alpha), + kernel_size=(3, 3), + strides=2, + padding='same', + kernel_initializer='he_normal', + add_batchnorm=True, + relu_activation='ReLU7.5', + name='sepconv_t_4', + pointwise_regularizer=weight_regularizer) + x = Dropout(0.5)(x) + x = conv_block(x, filters=1, kernel_size=(1, 1), relu_activation=False, name='head') + + # Build the whole model: encoder followed by decoder + model = Model(inputs=encoder.input, outputs=x, name='akida_unet') + + if base_weights is not None: + if base_weights == 'imagenet': + base_weights = _fetch_imagenet_weights() + model.load_weights(base_weights, by_name=True) + return model + + +if __name__ == "__main__": + parser = argparse.ArgumentParser( + description='Build the Akida U-Net Portrait128 model for Akida 2') + parser.add_argument("-s", + "--savepath", + type=str, + default='./models/akida_unet_portrait128.h5', + help="Save model with the specified path + name") + parser.add_argument("-bw", + "--base_weights", + type=str, + default=None, + help="Weights to load by layer name: 'imagenet' or an .h5 path") + parser.add_argument('--seed', type=int, default=42, + help='Random seed for reproducibility') + args = parser.parse_args() + + model = build_portrait128_model(base_weights=args.base_weights, seed=args.seed) + model.summary() + model.save(args.savepath, include_optimizer=False) + print(f'Model saved to {args.savepath}') diff --git a/akida2/model_zoo/portrait128/portrait128_notebook.py b/akida2/model_zoo/portrait128/portrait128_notebook.py new file mode 100644 index 0000000..44fafce --- /dev/null +++ b/akida2/model_zoo/portrait128/portrait128_notebook.py @@ -0,0 +1,391 @@ +# --- +# jupyter: +# jupytext: +# formats: ipynb,py:percent +# text_representation: +# extension: .py +# format_name: percent +# format_version: '1.3' +# jupytext_version: 1.19.6 +# kernelspec: +# display_name: Python 3 +# language: python +# name: python3 +# --- + +# %% [markdown] +# BrainChip Dev Hub +# +# # Portrait Segmentation (Portrait128) — Akida 2 Training +# +# Run Time: ~7 hours with training included (NVIDIA RTX 2080 Ti) / TBD with training skipped +# +# This notebook walks through the complete pipeline to train, quantize, convert, and evaluate an Akida U-Net model on the **Portrait128** dataset for Akida 2 hardware. +# +# Portrait128 is a binary segmentation task: for each 128 × 128 portrait, the model labels every pixel as **person** or **background**. +# +# The pipeline follows the standard Akida workflow: +# 1. Train a float model +# 2. 8-bit quantization (PTQ) +# 3. 4-bit quantization with QAT fine-tuning +# 4. Conversion to Akida `.fbz` format +# 5. Evaluation on Akida +# +# The focus is on the **Akida-specific** aspects of the pipeline. The data loading, +# model definition and training loop live in the accompanying Python files: they are +# standard tf_keras code and are imported here rather than repeated. +# +# Segmentation differs from the classification examples in this model zoo in two ways +# that matter on Akida. The model has to **upsample** back to full resolution, which +# needs transposed convolutions, an Akida 2 feature. And its output is a full-size map +# rather than a handful of class scores, so the last layers of the network run at the +# highest resolution. Both come up below. + +# %% +# Colab-only setup. Local users: ignore this cell — it does nothing for you. +import sys, os + +if 'google.colab' in sys.modules: + if not os.path.exists('colab_setup.py'): + !wget -q https://raw.githubusercontent.com/Brainchip-Inc/brainchip_devhub/main/akida2/model_zoo/portrait128/colab_setup.py + import colab_setup; colab_setup.setup() + +# %% [markdown] +# ## Setup +# +# The default dataset path is `./data/portrait128`, a folder holding the two arrays +# `img_uint8.npy` and `msk_uint8.npy`. See the [README](README.md) for where to get +# them and for symlink setup. Update `DATA_PATH` below if you keep them elsewhere. +# +# Setting `RUN_FLOAT_TRAINING` or `RUN_QAT_TRAINING` to `False` skips that training +# run and loads the model saved in `MODELS_DIR` by an earlier run of this notebook or +# of `portrait128_train.sh`. + +# %% +import os +import numpy as np +import tensorflow as tf +from tqdm import tqdm + +DATA_PATH = './data/portrait128' +MODELS_DIR = './models' +os.makedirs(MODELS_DIR, exist_ok=True) + +RUN_FLOAT_TRAINING = True +RUN_QAT_TRAINING = True + +SEED = 42 + +# Must be called before any TF ops to make GPU ops deterministic. +tf.config.experimental.enable_op_determinism() + +# %% [markdown] +# ## Dataset +# +# `get_data` returns a training and a validation `tf.data.Dataset`. The full loading +# and augmentation code is in [portrait128_data.py](portrait128_data.py). +# +# A few points that matter for what follows: +# +# * **Images stay uint8**, in their original `[0, 255]` range. The model rescales them +# itself (see Model below), so the same data feed the float, quantized and Akida +# models without any change, and a deployed Akida model can take camera pixels +# directly. +# * **Labels are masks**, of shape `(128, 128, 1)`: 1 for person, 0 for background. +# * **Augmentation is geometric** (flips, shifts and zooms), and each image is stacked +# with its mask before it is transformed. That way the mask moves with the image; +# transforming them separately would misalign the labels. +# * **The split is fixed.** The dataset ships as one pool, so it is split 80/20 with a +# fixed shuffle seed, independent of `SEED`. Every run is evaluated on the same +# validation images. + +# %% +from portrait128_data import get_data + +BATCH_SIZE = 32 +INPUT_SHAPE = (128, 128, 3) + +train_ds, val_ds = get_data(DATA_PATH, INPUT_SHAPE, BATCH_SIZE, seed=SEED) + +# %% [markdown] +# ## Model +# +# The model is the **Akida U-Net** from `akida_models`, an encoder–decoder: +# +# * The **encoder** is AkidaNet with width multiplier alpha = 0.5. Its five stride-2 +# stages reduce the 128 × 128 input to a 4 × 4 × 512 feature map: coarse in space +# but rich in features. +# * The **decoder** is five separable *transposed*-convolution blocks. Each doubles the +# spatial size, back up to 128 × 128. A transposed convolution with stride 2 is a +# learned upsampling, and it is the reason this model needs Akida 2: Akida 1 has no +# upsampling layer. Dropout after each block regularizes the decoder during +# training; it does nothing at inference and is dropped at conversion. +# * A final **1 × 1 convolution** gives one logit per pixel: positive for person, +# negative for background. There is no sigmoid in the model. The loss applies it, +# and a threshold at 0 gives the same mask as a threshold at probability 0.5. +# +# Unlike the classic U-Net there are **no skip connections** from encoder to decoder, +# so all the detail in the output has to pass through the 4 × 4 bottleneck. That keeps +# the graph a simple chain. The cost is softer mask edges than a U-Net with skips +# would give. +# +# The input rescaling is part of the model. The encoder's `input_scaling=(128, -1)` +# adds a layer that maps `[0, 255]` to `[-1, 1)`. Keeping it in the graph means one data +# pipeline serves every version of the model. It also means the conversion step can +# fold the rescaling into the Akida input layer's parameters, so the deployed model +# still takes plain uint8 pixels. +# +# **Encoder initialization.** `base_weights='imagenet'` loads BrainChip's +# ImageNet-trained AkidaNet 0.5 weights into the encoder, by layer name; the decoder +# starts from random weights. The ImageNet model was trained at 224 × 224, but +# convolution weights don't depend on the input size, so they load directly into this +# 128 × 128 encoder. Starting from features that already recognize edges, textures and +# shapes is why the learning rate below can be small. + +# %% +from portrait128_model import build_portrait128_model + +model = build_portrait128_model(base_weights='imagenet', seed=SEED) +model.summary() + +# %% [markdown] +# ## Float Training +# +# The full training code is in [portrait128_train.py](portrait128_train.py). The +# choices that matter here: +# +# * **Loss**: binary cross-entropy on the logits, averaged over every pixel. Each pixel +# is its own binary classification. +# * **Metrics**: *binary IoU*, the overlap between predicted and true masks averaged +# over person and background, and *pixel accuracy*. Background covers a large share of +# every image, so pixel accuracy is high even for poor masks; IoU penalizes a mask of +# the wrong shape, so it is the one to watch. +# * **Learning rate**: Adam with a one-cycle cosine schedule, stepped every batch. It +# warms up from a 25th of the peak to the peak `LEARNING_RATE` over the first 30 % of +# training, then decays. The peak, 3e-5, is low because the encoder starts +# pretrained: a high rate early on would destroy the ImageNet features before the +# decoder has learned to use them. + +# %% +from portrait128_train import train_portrait128, get_metrics + +if RUN_FLOAT_TRAINING: + LEARNING_RATE = 3e-5 + EPOCHS = 50 + # Freshly build the datasets for reproducibility + train_ds, val_ds = get_data(DATA_PATH, INPUT_SHAPE, BATCH_SIZE, seed=SEED) + + train_portrait128(model, train_ds, val_ds, EPOCHS, LEARNING_RATE, seed=SEED) + + float_model_path = os.path.join(MODELS_DIR, 'akida_unet_portrait128.h5') + model.save(float_model_path, include_optimizer=False) + print(f'Float model saved to {float_model_path}') +else: + from tf_keras.models import load_model + print('Training skipped. Loading an existing float model...') + model = load_model(os.path.join(MODELS_DIR, 'akida_unet_portrait128.h5')) + +# %% [markdown] +# ### Evaluate float model + +# %% +model.compile(metrics=get_metrics()) +float_results = model.evaluate(val_ds, verbose=0, return_dict=True) +float_iou, float_accuracy = float_results['binary_iou'], float_results['accuracy'] +print(f'Float binary IoU: {float_iou:.4f}, pixel accuracy: {float_accuracy:.4f}') + +# %% [markdown] +# ## Quantization (quantizeml) +# +# Akida runs on integers, so the weights and activations have to be quantized. Akida 2 +# quantizes with **quantizeml** (not `cnn2snn.quantize`). We produce two variants: +# +# * **8-bit** (i8/w8/a8): post-training quantization (PTQ) only. 256 levels per weight +# and activation track the float values closely enough that no retraining is needed. +# * **4-bit** (i8/w4/a4): quantization-aware training (QAT). With only 16 levels the +# rounding error is large enough to cost real accuracy after PTQ, so the model is +# fine-tuned with the quantization in place. +# +# In both variants the **input layer's weights stay 8-bit** (`input_weight_bits=8`). The +# input layer runs on Akida's dedicated input processor and sees the raw image, where +# precision matters most; it is also a small share of the total weights. +# +# Quantization with `quantizeml` requires **calibration samples**. Calibration runs real +# inputs through the model to measure the range of every layer's activations, and those +# ranges set the quantization scales. Random data would give the wrong ranges, so the +# samples should be representative images for the task. They are drawn from the +# **training** split, to avoid leaking validation data into the model. + +# %% +from quantizeml.models import quantize, QuantizationParams + +from portrait128_data import get_samples + +NUM_SAMPLES = 1024 +samples = get_samples(DATA_PATH, INPUT_SHAPE, num_samples=NUM_SAMPLES) + +# --- 8-bit variant (i8 / w8 / a8), PTQ only --- +qparams_8bit = QuantizationParams(input_weight_bits=8, weight_bits=8, activation_bits=8) +model_8bit = quantize(model, qparams=qparams_8bit, samples=samples) + +q8_path = os.path.join(MODELS_DIR, 'akida_unet_portrait128_i8_w8_a8.h5') +model_8bit.save(q8_path, include_optimizer=False) + +model_8bit.compile(metrics=get_metrics()) +q8_results = model_8bit.evaluate(val_ds, verbose=0, return_dict=True) +iou_8bit, acc_8bit = q8_results['binary_iou'], q8_results['accuracy'] +print(f'8-bit binary IoU: {iou_8bit:.4f}, pixel accuracy: {acc_8bit:.4f}') + +# %% [markdown] +# Note: the quantized model can be saved using the standard method. However, +# for later reloading, because of the custom quantized layers in the model +# we have to use the `load_model` function from `quantizeml.model_io` (a wrapper +# around the standard tf_keras loading function). + +# %% +# Just to illustrate how to load a quantized model +del model_8bit + +from quantizeml.model_io import load_model +model_8bit = load_model(q8_path) + +# %% [markdown] +# ### 4-bit variant with QAT +# +# At 4 bits, PTQ alone loses noticeable accuracy, so the quantized model is fine-tuned. +# During QAT the forward pass uses the quantized weights and activations, while the +# gradients update an underlying float copy of the weights. The network therefore +# learns weights that still work after rounding to 16 levels. +# +# QAT starts from a model that is already close to a good solution, so it uses a lower +# learning rate: a tenth of the float rate here. A higher rate would undo the float +# training rather than refine it. +# +# A model quantized with `quantizeml` is a standard Keras model, so it goes straight +# back into the same training function used for float training. The intermediate PTQ +# model is not kept. + +# %% +if RUN_QAT_TRAINING: + # --- 4-bit variant (i8 / w4 / a4), QAT --- + qparams_4bit = QuantizationParams(input_weight_bits=8, weight_bits=4, activation_bits=4) + model_4bit = quantize(model, qparams=qparams_4bit, samples=samples) + + # QAT fine-tune the quantized 4-bit model + QAT_EPOCHS = 10 + QAT_LR = 3e-6 + train_ds, val_ds = get_data(DATA_PATH, INPUT_SHAPE, BATCH_SIZE, seed=SEED) + train_portrait128(model_4bit, train_ds, val_ds, QAT_EPOCHS, QAT_LR, seed=SEED) + + q4_path = os.path.join(MODELS_DIR, 'akida_unet_portrait128_i8_w4_a4_qat.h5') + model_4bit.save(q4_path, include_optimizer=False) +else: + # Note the use of the quantizeml load_model wrapper to handle the + # custom quantized layer types + from quantizeml.model_io import load_model + print('Training skipped. Loading an existing 4-bit model...') + model_4bit = load_model(os.path.join(MODELS_DIR, 'akida_unet_portrait128_i8_w4_a4_qat.h5')) + +model_4bit.compile(metrics=get_metrics()) +q4_results = model_4bit.evaluate(val_ds, verbose=0, return_dict=True) +iou_4bit, acc_4bit = q4_results['binary_iou'], q4_results['accuracy'] +print(f'4-bit QAT binary IoU: {iou_4bit:.4f}, pixel accuracy: {acc_4bit:.4f}') + +# %% [markdown] +# ## Conversion to Akida Format +# +# `cnn2snn.convert` compiles the quantized Keras model into an Akida `.fbz` +# model that can be loaded and executed directly on an Akida 2 hardware device. +# The converter verifies hardware compatibility and maps each layer to its +# corresponding Akida primitive. We convert both quantized variants. +# +# In the summary below, note the following: +# +# * The input rescaling has no layer of its own. It was folded into the first +# convolution. +# * Each decoder block becomes a `DepthwiseConv2DTranspose` followed by a pointwise +# `Conv2D`. On Akida 2 the depthwise and pointwise parts of a separable convolution +# are two separate layers. +# * The last layer is a `Dequantizer`, so the model outputs float logits of shape +# `(128, 128, 1)`, which can be thresholded at 0 just like the Keras model's. + +# %% +from cnn2snn import convert + +akida_8bit = convert(model_8bit) +akida_8bit.save(os.path.join(MODELS_DIR, 'akida_unet_portrait128_i8_w8_a8.fbz')) + +akida_4bit = convert(model_4bit) +akida_4bit.save(os.path.join(MODELS_DIR, 'akida_unet_portrait128_i8_w4_a4_qat.fbz')) + +akida_8bit.summary() + + +# %% [markdown] +# ## Evaluation on Akida (software backend) +# +# We now run evaluation through the Akida model, to check that its results match +# those of the quantized tf_keras model. Here, we deliberately use the software +# backend (the default, since we do not check for and map to a connected hardware +# device): this delivers a bit-accurate simulation of the results that will be +# obtained when running the model on hardware. + +# %% [markdown] +# ### Run Evaluation on Akida +# +# The Akida runtime cannot consume `tf.data.Dataset` objects directly; it expects a +# 4D numpy array `(n, h, w, c)` of uint8. So we iterate over validation batches +# manually and convert each one. +# +# The output already has the mask's shape, `(B, 128, 128, 1)`, so it goes straight +# into the same IoU and accuracy metrics as the labels, with no reshaping. (Classifiers +# return `(B, 1, 1, C)` and need a squeeze first.) + +# %% +def evaluate_akida(akida_model, ds): + iou_metric, acc_metric = get_metrics() + for batch, label_batch in tqdm(ds, desc='Evaluating on Akida'): + logits = akida_model.predict(batch.numpy().astype(np.uint8)) + iou_metric.update_state(label_batch, logits) + acc_metric.update_state(label_batch, logits) + return float(iou_metric.result()), float(acc_metric.result()) + +akida_iou_8bit, akida_acc_8bit = evaluate_akida(akida_8bit, val_ds) +akida_iou_4bit, akida_acc_4bit = evaluate_akida(akida_4bit, val_ds) +print(f'Akida 8-bit binary IoU: {akida_iou_8bit:.4f}, pixel accuracy: {akida_acc_8bit:.4f}') +print(f'Akida 4-bit QAT binary IoU: {akida_iou_4bit:.4f}, pixel accuracy: {akida_acc_4bit:.4f}') + +# %% [markdown] +# ## Activation Sparsity +# +# Akida is event-based: its standard neural processors only do work for non-zero +# activations. The more zeros a layer receives (its input *sparsity*), the less time and +# energy it costs. Sparsity depends on the data, so it has to be measured by running real +# images through the model. Here, we reuse the calibration samples built earlier. +# +# The input layer is the exception. It runs on the dedicated input processor, a +# conventional convolution whose cost doesn't depend on how many of its inputs are zero. +# +# For this model, look at the decoder. Its last blocks run at 64 × 64 and 128 × 128, so +# they process far more activations than the encoder layers at 8 × 8 or 4 × 4. Their +# sparsity is therefore likely to have the most effect on latency; the per-layer +# benchmark in `portrait128_notebook_benchmark.ipynb` measures it. + +# %% +from akida_models.sparsity import compute_sparsity +from brainchip_utils.plot_utils import pretty_print_sparsity + +# Reuse the calibration samples built earlier (sparsity needs real activations too). +print('8-bit sparsity:') +pretty_print_sparsity(compute_sparsity(akida_8bit, samples=samples)) +print('\n4-bit QAT sparsity:') +pretty_print_sparsity(compute_sparsity(akida_4bit, samples=samples)) + +# %% [markdown] +# ## Summary + +# %% +print(f'{"Variant":<16}{"Keras IoU":<12}{"Keras acc":<12}{"Akida IoU":<12}{"Akida acc":<12}') +print(f'{"float":<16}{float_iou:<12.4f}{float_accuracy:<12.4f}{"-":<12}{"-":<12}') +print(f'{"8-bit (w8a8)":<16}{iou_8bit:<12.4f}{acc_8bit:<12.4f}{akida_iou_8bit:<12.4f}{akida_acc_8bit:<12.4f}') +print(f'{"4-bit QAT":<16}{iou_4bit:<12.4f}{acc_4bit:<12.4f}{akida_iou_4bit:<12.4f}{akida_acc_4bit:<12.4f}') diff --git a/akida2/model_zoo/portrait128/portrait128_notebook_benchmark.ipynb b/akida2/model_zoo/portrait128/portrait128_notebook_benchmark.ipynb new file mode 100644 index 0000000..3a33bc2 --- /dev/null +++ b/akida2/model_zoo/portrait128/portrait128_notebook_benchmark.ipynb @@ -0,0 +1,236 @@ +{ + "cells": [ + { + "cell_type": "markdown", + "id": "cell-00", + "metadata": {}, + "source": [ + "\"BrainChip\n", + "\n", + "# Portrait Segmentation (Portrait128) — Akida 2 Benchmark\n", + "\n", + "Run Time: TBD\n", + "\n", + "This notebook walks through evaluating and benchmarking a model on the **Portrait128** dataset, targeting Akida 2 hardware.\n", + "\n", + "For details on the dataset and preparation of the model, see the neighbouring README.md and `portrait128_notebook_training.ipynb`.\n", + "\n", + "> **Note:** the hardware benchmark needs a physical Akida 2 FPGA device. Every hardware step is guarded, so the notebook runs end to end without a board: the latency cells are skipped, and sparsity is still computed on the software backend. It will not produce hardware numbers on Colab.\n", + "\n", + "The Akida 2 reference platform is an FPGA running at 25 MHz. Alongside the measured latency, the notebook reports a projected latency at the AKD2500 target clock. Power is not measured yet, so the benchmark is latency-only." + ] + }, + { + "cell_type": "markdown", + "id": "cell-01", + "metadata": {}, + "source": [ + "## Setup" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "cell-02", + "metadata": {}, + "outputs": [], + "source": [ + "import os\n", + "import numpy as np\n", + "import akida\n", + "\n", + "from brainchip_utils.hardware_utils import AKIDA_CLOCKS_HZ\n", + "\n", + "DATA_PATH = './data/portrait128'\n", + "INPUT_SHAPE = (128, 128, 3)\n", + "\n", + "# Pretrained models committed to the repo. If you trained your own with the\n", + "# training notebook or portrait128_train.sh, set MODELS_DIR = './models/'.\n", + "MODELS_DIR = './pretrained_models/'\n", + "\n", + "# Which converted model to benchmark. Options produced by the training pipeline:\n", + "# akida_unet_portrait128_i8_w8_a8.fbz (8-bit)\n", + "# akida_unet_portrait128_i8_w4_a4_qat.fbz (4-bit QAT)\n", + "MODEL_FILENAME = 'akida_unet_portrait128_i8_w4_a4_qat.fbz'\n", + "\n", + "# Clocks (see portrait128_benchmark.py and brainchip_utils.hardware_utils.AKIDA_CLOCKS_HZ).\n", + "MEASURED_CLOCK = AKIDA_CLOCKS_HZ['AKIDA2_FPGA'] # 25 MHz FPGA\n", + "PROJECTED_CLOCK = AKIDA_CLOCKS_HZ['AKD2500'] # 1 GHz target (pre-production)" + ] + }, + { + "cell_type": "markdown", + "id": "cell-03", + "metadata": {}, + "source": [ + "## Model\n", + "\n", + "The notebook loads a converted Akida model (`.fbz`) from `pretrained_models/`, for example\n", + "`pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.fbz`. Pretrained models are committed\n", + "through `git-lfs`; see the [Trained models](../../../README.md#trained-models) section of the\n", + "top-level README for set-up.\n", + "\n", + "📝 Planned: the pretrained models for this example are added once the training pipeline has\n", + "been run. Until then, train your own with `portrait128_notebook_training.ipynb` or\n", + "`portrait128_train.sh` and point `MODELS_DIR` at `./models/`.\n", + "\n", + "The model takes a uint8 RGB image of 128 × 128 and returns one logit per pixel, as a\n", + "`(128, 128, 1)` map. Its first layer, `InputConv2D`, runs on Akida's dedicated input\n", + "processor, which takes the uint8 image directly." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "cell-04", + "metadata": {}, + "outputs": [], + "source": [ + "ak_model = akida.Model(os.path.join(MODELS_DIR, MODEL_FILENAME))\n", + "ak_model.summary()" + ] + }, + { + "cell_type": "markdown", + "id": "cell-05", + "metadata": {}, + "source": [ + "## Device\n", + "\n", + "`get_akida_device` returns `None` when no compatible hardware is present, in which case the latency benchmark below is skipped." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "cell-06", + "metadata": {}, + "outputs": [], + "source": [ + "from brainchip_utils.hardware_utils import get_akida_device\n", + "\n", + "device = get_akida_device(target_version=ak_model.ip_version)\n", + "if device is None:\n", + " print('No compatible Akida hardware device found — benchmark will be skipped.')\n", + "else:\n", + " print('Akida device found:', device)" + ] + }, + { + "cell_type": "markdown", + "id": "cell-07", + "metadata": {}, + "source": [ + "## Samples\n", + "\n", + "Akida's neural processors only do work for non-zero activations, so latency depends on the input. Random or synthetic images drive unrepresentative activity and give misleading timings: a random image has no smooth regions, and its activations are much less sparse than a real portrait's. So the benchmark runs on real images from the training split." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "cell-08", + "metadata": {}, + "outputs": [], + "source": [ + "from portrait128_data import get_samples\n", + "\n", + "NUM_SAMPLES = 100\n", + "samples = get_samples(DATA_PATH, INPUT_SHAPE, num_samples=NUM_SAMPLES)" + ] + }, + { + "cell_type": "markdown", + "id": "cell-09", + "metadata": {}, + "source": [ + "## Latency benchmark\n", + "\n", + "The model is benchmarked in each of the three mapping modes, which trade the number of neural processors (NPs) used against the number of passes:\n", + "\n", + "* **Minimal** maps the model onto the fewest NPs that can hold it.\n", + "* **AllNps** spreads it over as many NPs as possible, while keeping the minimum number of passes.\n", + "* **HwPr** (hardware partial reconfiguration) splits the model into several passes so each layer can use more NPs, reloading weights between passes on every inference.\n", + "\n", + "Which mode is fastest depends on the model, so all three are measured.\n", + "\n", + "The FPGA runs at 25 MHz, far below production silicon. The number of clock cycles per inference is fixed for a given model and mapping, whatever the clock, so the latency at the AKD2500 target clock is an exact rescale: `projected_ms = mean_inf_clk / PROJECTED_CLOCK * 1000`." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "cell-10", + "metadata": {}, + "outputs": [], + "source": [ + "from brainchip_utils.hardware_utils import full_model_benchmark, get_mapping_stats\n", + "\n", + "if device is not None:\n", + " for mm in ['Minimal', 'AllNps', 'HwPr']:\n", + " map_mode = getattr(akida.MapMode, mm)\n", + " res = full_model_benchmark(ak_model, device, samples,\n", + " map_mode=map_mode, clock_freq=MEASURED_CLOCK)\n", + " projected_ms = res['mean_inf_clk'] / PROJECTED_CLOCK * 1000\n", + " ak_model.map(device, mode=map_mode)\n", + " num_nps, num_passes, num_sequences = get_mapping_stats(ak_model)\n", + " print(f'[{mm}] NPs={num_nps} passes={num_passes} '\n", + " f'latency@25MHz={res[\"mean_clk_ms\"]:.3f} ms '\n", + " f'projected@1000MHz={projected_ms:.3f} ms')\n", + "else:\n", + " print('Hardware not available — skipping latency benchmark.')" + ] + }, + { + "cell_type": "markdown", + "id": "cell-11", + "metadata": {}, + "source": [ + "## Per-layer benchmark & sparsity\n", + "\n", + "The per-layer benchmark (Minimal mapping) shows where the time goes, and the sparsity printout shows why. A layer's cost grows with the number of non-zero activations it receives, so on Akida, per-layer latency and input sparsity are linked: a layer with dense input and a large feature map is slow.\n", + "\n", + "In this model, look at the last decoder blocks. They run at 64 × 64 and 128 × 128, so they receive far more activations than the encoder layers at 8 × 8 or 4 × 4. Their sparsity therefore matters most.\n", + "\n", + "Sparsity doesn't need hardware, so it is also computed on the software backend when no device is present." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "cell-12", + "metadata": {}, + "outputs": [], + "source": [ + "from brainchip_utils.hardware_utils import per_layer_benchmark\n", + "from akida_models.sparsity import compute_sparsity\n", + "from brainchip_utils.plot_utils import pretty_print_sparsity\n", + "\n", + "if device is not None:\n", + " ak_model.map(device, mode=akida.MapMode.Minimal, hw_only=True)\n", + " sparsity_dict = compute_sparsity(ak_model, samples=samples)\n", + " pretty_print_sparsity(sparsity_dict)\n", + " per_layer_results = per_layer_benchmark(ak_model, device, samples,\n", + " repeats=NUM_SAMPLES, clock_freq=MEASURED_CLOCK)\n", + " print('Per-layer benchmark complete.')\n", + "else:\n", + " # Sparsity can still be computed on the software backend without hardware.\n", + " sparsity_dict = compute_sparsity(ak_model, samples=samples)\n", + " pretty_print_sparsity(sparsity_dict)\n", + " print('Hardware not available — skipped latency; sparsity computed on software backend.')" + ] + } + ], + "metadata": { + "kernelspec": { + "display_name": "Python 3", + "language": "python", + "name": "python3" + }, + "language_info": { + "name": "python" + } + }, + "nbformat": 4, + "nbformat_minor": 5 +} diff --git a/akida2/model_zoo/portrait128/portrait128_notebook_training.ipynb b/akida2/model_zoo/portrait128/portrait128_notebook_training.ipynb new file mode 100644 index 0000000..0dc80cc --- /dev/null +++ b/akida2/model_zoo/portrait128/portrait128_notebook_training.ipynb @@ -0,0 +1,556 @@ +{ + "cells": [ + { + "cell_type": "markdown", + "id": "cell-00", + "metadata": {}, + "source": [ + "\"BrainChip\n", + "\n", + "# Portrait Segmentation (Portrait128) — Akida 2 Training\n", + "\n", + "Run Time: ~7 hours with training included (NVIDIA RTX 2080 Ti) / TBD with training skipped\n", + "\n", + "This notebook walks through the complete pipeline to train, quantize, convert, and evaluate an Akida U-Net model on the **Portrait128** dataset for Akida 2 hardware.\n", + "\n", + "Portrait128 is a binary segmentation task: for each 128 × 128 portrait, the model labels every pixel as **person** or **background**.\n", + "\n", + "The pipeline follows the standard Akida workflow:\n", + "1. Train a float model\n", + "2. 8-bit quantization (PTQ)\n", + "3. 4-bit quantization with QAT fine-tuning\n", + "4. Conversion to Akida `.fbz` format\n", + "5. Evaluation on Akida\n", + "\n", + "The focus is on the **Akida-specific** aspects of the pipeline. The data loading,\n", + "model definition and training loop live in the accompanying Python files: they are\n", + "standard tf_keras code and are imported here rather than repeated.\n", + "\n", + "Segmentation differs from the classification examples in this model zoo in two ways\n", + "that matter on Akida. The model has to **upsample** back to full resolution, which\n", + "needs transposed convolutions, an Akida 2 feature. And its output is a full-size map\n", + "rather than a handful of class scores, so the last layers of the network run at the\n", + "highest resolution. Both come up below." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "cell-01", + "metadata": {}, + "outputs": [], + "source": [ + "# Colab-only setup. Local users: ignore this cell — it does nothing for you.\n", + "import sys, os\n", + "\n", + "if 'google.colab' in sys.modules:\n", + " if not os.path.exists('colab_setup.py'):\n", + " !wget -q https://raw.githubusercontent.com/Brainchip-Inc/brainchip_devhub/main/akida2/model_zoo/portrait128/colab_setup.py\n", + " import colab_setup; colab_setup.setup()" + ] + }, + { + "cell_type": "markdown", + "id": "cell-02", + "metadata": {}, + "source": [ + "## Setup\n", + "\n", + "The default dataset path is `./data/portrait128`, a folder holding the two arrays\n", + "`img_uint8.npy` and `msk_uint8.npy`. See the [README](README.md) for where to get\n", + "them and for symlink setup. Update `DATA_PATH` below if you keep them elsewhere.\n", + "\n", + "Setting `RUN_FLOAT_TRAINING` or `RUN_QAT_TRAINING` to `False` skips that training\n", + "run and loads the model saved in `MODELS_DIR` by an earlier run of this notebook or\n", + "of `portrait128_train.sh`." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "cell-03", + "metadata": {}, + "outputs": [], + "source": [ + "import os\n", + "import numpy as np\n", + "import tensorflow as tf\n", + "from tqdm import tqdm\n", + "\n", + "DATA_PATH = './data/portrait128'\n", + "MODELS_DIR = './models'\n", + "os.makedirs(MODELS_DIR, exist_ok=True)\n", + "\n", + "RUN_FLOAT_TRAINING = True\n", + "RUN_QAT_TRAINING = True\n", + "\n", + "SEED = 42\n", + "\n", + "# Must be called before any TF ops to make GPU ops deterministic.\n", + "tf.config.experimental.enable_op_determinism()" + ] + }, + { + "cell_type": "markdown", + "id": "cell-04", + "metadata": {}, + "source": [ + "## Dataset\n", + "\n", + "`get_data` returns a training and a validation `tf.data.Dataset`. The full loading\n", + "and augmentation code is in [portrait128_data.py](portrait128_data.py).\n", + "\n", + "A few points that matter for what follows:\n", + "\n", + "* **Images stay uint8**, in their original `[0, 255]` range. The model rescales them\n", + " itself (see Model below), so the same data feed the float, quantized and Akida\n", + " models without any change, and a deployed Akida model can take camera pixels\n", + " directly.\n", + "* **Labels are masks**, of shape `(128, 128, 1)`: 1 for person, 0 for background.\n", + "* **Augmentation is geometric** (flips, shifts and zooms), and each image is stacked\n", + " with its mask before it is transformed. That way the mask moves with the image;\n", + " transforming them separately would misalign the labels.\n", + "* **The split is fixed.** The dataset ships as one pool, so it is split 80/20 with a\n", + " fixed shuffle seed, independent of `SEED`. Every run is evaluated on the same\n", + " validation images." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "cell-05", + "metadata": {}, + "outputs": [], + "source": [ + "from portrait128_data import get_data\n", + "\n", + "BATCH_SIZE = 32\n", + "INPUT_SHAPE = (128, 128, 3)\n", + "\n", + "train_ds, val_ds = get_data(DATA_PATH, INPUT_SHAPE, BATCH_SIZE, seed=SEED)" + ] + }, + { + "cell_type": "markdown", + "id": "cell-06", + "metadata": {}, + "source": [ + "## Model\n", + "\n", + "The model is the **Akida U-Net** from `akida_models`, an encoder–decoder:\n", + "\n", + "* The **encoder** is AkidaNet with width multiplier alpha = 0.5. Its five stride-2\n", + " stages reduce the 128 × 128 input to a 4 × 4 × 512 feature map: coarse in space\n", + " but rich in features.\n", + "* The **decoder** is five separable *transposed*-convolution blocks. Each doubles the\n", + " spatial size, back up to 128 × 128. A transposed convolution with stride 2 is a\n", + " learned upsampling, and it is the reason this model needs Akida 2: Akida 1 has no\n", + " upsampling layer. Dropout after each block regularizes the decoder during\n", + " training; it does nothing at inference and is dropped at conversion.\n", + "* A final **1 × 1 convolution** gives one logit per pixel: positive for person,\n", + " negative for background. There is no sigmoid in the model. The loss applies it,\n", + " and a threshold at 0 gives the same mask as a threshold at probability 0.5.\n", + "\n", + "Unlike the classic U-Net there are **no skip connections** from encoder to decoder,\n", + "so all the detail in the output has to pass through the 4 × 4 bottleneck. That keeps\n", + "the graph a simple chain. The cost is softer mask edges than a U-Net with skips\n", + "would give.\n", + "\n", + "The input rescaling is part of the model. The encoder's `input_scaling=(128, -1)`\n", + "adds a layer that maps `[0, 255]` to `[-1, 1)`. Keeping it in the graph means one data\n", + "pipeline serves every version of the model. It also means the conversion step can\n", + "fold the rescaling into the Akida input layer's parameters, so the deployed model\n", + "still takes plain uint8 pixels.\n", + "\n", + "**Encoder initialization.** `base_weights='imagenet'` loads BrainChip's\n", + "ImageNet-trained AkidaNet 0.5 weights into the encoder, by layer name; the decoder\n", + "starts from random weights. The ImageNet model was trained at 224 × 224, but\n", + "convolution weights don't depend on the input size, so they load directly into this\n", + "128 × 128 encoder. Starting from features that already recognize edges, textures and\n", + "shapes is why the learning rate below can be small." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "cell-07", + "metadata": {}, + "outputs": [], + "source": [ + "from portrait128_model import build_portrait128_model\n", + "\n", + "model = build_portrait128_model(base_weights='imagenet', seed=SEED)\n", + "model.summary()" + ] + }, + { + "cell_type": "markdown", + "id": "cell-08", + "metadata": {}, + "source": [ + "## Float Training\n", + "\n", + "The full training code is in [portrait128_train.py](portrait128_train.py). The\n", + "choices that matter here:\n", + "\n", + "* **Loss**: binary cross-entropy on the logits, averaged over every pixel. Each pixel\n", + " is its own binary classification.\n", + "* **Metrics**: *binary IoU*, the overlap between predicted and true masks averaged\n", + " over person and background, and *pixel accuracy*. Background covers a large share of\n", + " every image, so pixel accuracy is high even for poor masks; IoU penalizes a mask of\n", + " the wrong shape, so it is the one to watch.\n", + "* **Learning rate**: Adam with a one-cycle cosine schedule, stepped every batch. It\n", + " warms up from a 25th of the peak to the peak `LEARNING_RATE` over the first 30 % of\n", + " training, then decays. The peak, 3e-5, is low because the encoder starts\n", + " pretrained: a high rate early on would destroy the ImageNet features before the\n", + " decoder has learned to use them." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "cell-09", + "metadata": {}, + "outputs": [], + "source": [ + "from portrait128_train import train_portrait128, get_metrics\n", + "\n", + "if RUN_FLOAT_TRAINING:\n", + " LEARNING_RATE = 3e-5\n", + " EPOCHS = 50\n", + " # Freshly build the datasets for reproducibility\n", + " train_ds, val_ds = get_data(DATA_PATH, INPUT_SHAPE, BATCH_SIZE, seed=SEED)\n", + "\n", + " train_portrait128(model, train_ds, val_ds, EPOCHS, LEARNING_RATE, seed=SEED)\n", + "\n", + " float_model_path = os.path.join(MODELS_DIR, 'akida_unet_portrait128.h5')\n", + " model.save(float_model_path, include_optimizer=False)\n", + " print(f'Float model saved to {float_model_path}')\n", + "else:\n", + " from tf_keras.models import load_model\n", + " print('Training skipped. Loading an existing float model...')\n", + " model = load_model(os.path.join(MODELS_DIR, 'akida_unet_portrait128.h5'))" + ] + }, + { + "cell_type": "markdown", + "id": "cell-10", + "metadata": {}, + "source": [ + "### Evaluate float model" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "cell-11", + "metadata": {}, + "outputs": [], + "source": [ + "model.compile(metrics=get_metrics())\n", + "float_results = model.evaluate(val_ds, verbose=0, return_dict=True)\n", + "float_iou, float_accuracy = float_results['binary_iou'], float_results['accuracy']\n", + "print(f'Float binary IoU: {float_iou:.4f}, pixel accuracy: {float_accuracy:.4f}')" + ] + }, + { + "cell_type": "markdown", + "id": "cell-12", + "metadata": {}, + "source": [ + "## Quantization (quantizeml)\n", + "\n", + "Akida runs on integers, so the weights and activations have to be quantized. Akida 2\n", + "quantizes with **quantizeml** (not `cnn2snn.quantize`). We produce two variants:\n", + "\n", + "* **8-bit** (i8/w8/a8): post-training quantization (PTQ) only. 256 levels per weight\n", + " and activation track the float values closely enough that no retraining is needed.\n", + "* **4-bit** (i8/w4/a4): quantization-aware training (QAT). With only 16 levels the\n", + " rounding error is large enough to cost real accuracy after PTQ, so the model is\n", + " fine-tuned with the quantization in place.\n", + "\n", + "In both variants the **input layer's weights stay 8-bit** (`input_weight_bits=8`). The\n", + "input layer runs on Akida's dedicated input processor and sees the raw image, where\n", + "precision matters most; it is also a small share of the total weights.\n", + "\n", + "Quantization with `quantizeml` requires **calibration samples**. Calibration runs real\n", + "inputs through the model to measure the range of every layer's activations, and those\n", + "ranges set the quantization scales. Random data would give the wrong ranges, so the\n", + "samples should be representative images for the task. They are drawn from the\n", + "**training** split, to avoid leaking validation data into the model." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "cell-13", + "metadata": {}, + "outputs": [], + "source": [ + "from quantizeml.models import quantize, QuantizationParams\n", + "\n", + "from portrait128_data import get_samples\n", + "\n", + "NUM_SAMPLES = 1024\n", + "samples = get_samples(DATA_PATH, INPUT_SHAPE, num_samples=NUM_SAMPLES)\n", + "\n", + "# --- 8-bit variant (i8 / w8 / a8), PTQ only ---\n", + "qparams_8bit = QuantizationParams(input_weight_bits=8, weight_bits=8, activation_bits=8)\n", + "model_8bit = quantize(model, qparams=qparams_8bit, samples=samples)\n", + "\n", + "q8_path = os.path.join(MODELS_DIR, 'akida_unet_portrait128_i8_w8_a8.h5')\n", + "model_8bit.save(q8_path, include_optimizer=False)\n", + "\n", + "model_8bit.compile(metrics=get_metrics())\n", + "q8_results = model_8bit.evaluate(val_ds, verbose=0, return_dict=True)\n", + "iou_8bit, acc_8bit = q8_results['binary_iou'], q8_results['accuracy']\n", + "print(f'8-bit binary IoU: {iou_8bit:.4f}, pixel accuracy: {acc_8bit:.4f}')" + ] + }, + { + "cell_type": "markdown", + "id": "cell-14", + "metadata": {}, + "source": [ + "Note: the quantized model can be saved using the standard method. However,\n", + "for later reloading, because of the custom quantized layers in the model\n", + "we have to use the `load_model` function from `quantizeml.model_io` (a wrapper\n", + "around the standard tf_keras loading function)." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "cell-15", + "metadata": {}, + "outputs": [], + "source": [ + "# Just to illustrate how to load a quantized model\n", + "del model_8bit\n", + "\n", + "from quantizeml.model_io import load_model\n", + "model_8bit = load_model(q8_path)" + ] + }, + { + "cell_type": "markdown", + "id": "cell-16", + "metadata": {}, + "source": [ + "### 4-bit variant with QAT\n", + "\n", + "At 4 bits, PTQ alone loses noticeable accuracy, so the quantized model is fine-tuned.\n", + "During QAT the forward pass uses the quantized weights and activations, while the\n", + "gradients update an underlying float copy of the weights. The network therefore\n", + "learns weights that still work after rounding to 16 levels.\n", + "\n", + "QAT starts from a model that is already close to a good solution, so it uses a lower\n", + "learning rate: a tenth of the float rate here. A higher rate would undo the float\n", + "training rather than refine it.\n", + "\n", + "A model quantized with `quantizeml` is a standard Keras model, so it goes straight\n", + "back into the same training function used for float training. The intermediate PTQ\n", + "model is not kept." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "cell-17", + "metadata": {}, + "outputs": [], + "source": [ + "if RUN_QAT_TRAINING:\n", + " # --- 4-bit variant (i8 / w4 / a4), QAT ---\n", + " qparams_4bit = QuantizationParams(input_weight_bits=8, weight_bits=4, activation_bits=4)\n", + " model_4bit = quantize(model, qparams=qparams_4bit, samples=samples)\n", + "\n", + " # QAT fine-tune the quantized 4-bit model\n", + " QAT_EPOCHS = 10\n", + " QAT_LR = 3e-6\n", + " train_ds, val_ds = get_data(DATA_PATH, INPUT_SHAPE, BATCH_SIZE, seed=SEED)\n", + " train_portrait128(model_4bit, train_ds, val_ds, QAT_EPOCHS, QAT_LR, seed=SEED)\n", + "\n", + " q4_path = os.path.join(MODELS_DIR, 'akida_unet_portrait128_i8_w4_a4_qat.h5')\n", + " model_4bit.save(q4_path, include_optimizer=False)\n", + "else:\n", + " # Note the use of the quantizeml load_model wrapper to handle the\n", + " # custom quantized layer types\n", + " from quantizeml.model_io import load_model\n", + " print('Training skipped. Loading an existing 4-bit model...')\n", + " model_4bit = load_model(os.path.join(MODELS_DIR, 'akida_unet_portrait128_i8_w4_a4_qat.h5'))\n", + "\n", + "model_4bit.compile(metrics=get_metrics())\n", + "q4_results = model_4bit.evaluate(val_ds, verbose=0, return_dict=True)\n", + "iou_4bit, acc_4bit = q4_results['binary_iou'], q4_results['accuracy']\n", + "print(f'4-bit QAT binary IoU: {iou_4bit:.4f}, pixel accuracy: {acc_4bit:.4f}')" + ] + }, + { + "cell_type": "markdown", + "id": "cell-18", + "metadata": {}, + "source": [ + "## Conversion to Akida Format\n", + "\n", + "`cnn2snn.convert` compiles the quantized Keras model into an Akida `.fbz`\n", + "model that can be loaded and executed directly on an Akida 2 hardware device.\n", + "The converter verifies hardware compatibility and maps each layer to its\n", + "corresponding Akida primitive. We convert both quantized variants.\n", + "\n", + "In the summary below, note the following:\n", + "\n", + "* The input rescaling has no layer of its own. It was folded into the first\n", + " convolution.\n", + "* Each decoder block becomes a `DepthwiseConv2DTranspose` followed by a pointwise\n", + " `Conv2D`. On Akida 2 the depthwise and pointwise parts of a separable convolution\n", + " are two separate layers.\n", + "* The last layer is a `Dequantizer`, so the model outputs float logits of shape\n", + " `(128, 128, 1)`, which can be thresholded at 0 just like the Keras model's." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "cell-19", + "metadata": {}, + "outputs": [], + "source": [ + "from cnn2snn import convert\n", + "\n", + "akida_8bit = convert(model_8bit)\n", + "akida_8bit.save(os.path.join(MODELS_DIR, 'akida_unet_portrait128_i8_w8_a8.fbz'))\n", + "\n", + "akida_4bit = convert(model_4bit)\n", + "akida_4bit.save(os.path.join(MODELS_DIR, 'akida_unet_portrait128_i8_w4_a4_qat.fbz'))\n", + "\n", + "akida_8bit.summary()" + ] + }, + { + "cell_type": "markdown", + "id": "cell-20", + "metadata": {}, + "source": [ + "## Evaluation on Akida (software backend)\n", + "\n", + "We now run evaluation through the Akida model, to check that its results match\n", + "those of the quantized tf_keras model. Here, we deliberately use the software\n", + "backend (the default, since we do not check for and map to a connected hardware\n", + "device): this delivers a bit-accurate simulation of the results that will be\n", + "obtained when running the model on hardware." + ] + }, + { + "cell_type": "markdown", + "id": "cell-21", + "metadata": {}, + "source": [ + "### Run Evaluation on Akida\n", + "\n", + "The Akida runtime cannot consume `tf.data.Dataset` objects directly; it expects a\n", + "4D numpy array `(n, h, w, c)` of uint8. So we iterate over validation batches\n", + "manually and convert each one.\n", + "\n", + "The output already has the mask's shape, `(B, 128, 128, 1)`, so it goes straight\n", + "into the same IoU and accuracy metrics as the labels, with no reshaping. (Classifiers\n", + "return `(B, 1, 1, C)` and need a squeeze first.)" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "cell-22", + "metadata": {}, + "outputs": [], + "source": [ + "def evaluate_akida(akida_model, ds):\n", + " iou_metric, acc_metric = get_metrics()\n", + " for batch, label_batch in tqdm(ds, desc='Evaluating on Akida'):\n", + " logits = akida_model.predict(batch.numpy().astype(np.uint8))\n", + " iou_metric.update_state(label_batch, logits)\n", + " acc_metric.update_state(label_batch, logits)\n", + " return float(iou_metric.result()), float(acc_metric.result())\n", + "\n", + "akida_iou_8bit, akida_acc_8bit = evaluate_akida(akida_8bit, val_ds)\n", + "akida_iou_4bit, akida_acc_4bit = evaluate_akida(akida_4bit, val_ds)\n", + "print(f'Akida 8-bit binary IoU: {akida_iou_8bit:.4f}, pixel accuracy: {akida_acc_8bit:.4f}')\n", + "print(f'Akida 4-bit QAT binary IoU: {akida_iou_4bit:.4f}, pixel accuracy: {akida_acc_4bit:.4f}')" + ] + }, + { + "cell_type": "markdown", + "id": "cell-23", + "metadata": {}, + "source": [ + "## Activation Sparsity\n", + "\n", + "Akida is event-based: its standard neural processors only do work for non-zero\n", + "activations. The more zeros a layer receives (its input *sparsity*), the less time and\n", + "energy it costs. Sparsity depends on the data, so it has to be measured by running real\n", + "images through the model. Here, we reuse the calibration samples built earlier.\n", + "\n", + "The input layer is the exception. It runs on the dedicated input processor, a\n", + "conventional convolution whose cost doesn't depend on how many of its inputs are zero.\n", + "\n", + "For this model, look at the decoder. Its last blocks run at 64 × 64 and 128 × 128, so\n", + "they process far more activations than the encoder layers at 8 × 8 or 4 × 4. Their\n", + "sparsity is therefore likely to have the most effect on latency; the per-layer\n", + "benchmark in `portrait128_notebook_benchmark.ipynb` measures it." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "cell-24", + "metadata": {}, + "outputs": [], + "source": [ + "from akida_models.sparsity import compute_sparsity\n", + "from brainchip_utils.plot_utils import pretty_print_sparsity\n", + "\n", + "# Reuse the calibration samples built earlier (sparsity needs real activations too).\n", + "print('8-bit sparsity:')\n", + "pretty_print_sparsity(compute_sparsity(akida_8bit, samples=samples))\n", + "print('\\n4-bit QAT sparsity:')\n", + "pretty_print_sparsity(compute_sparsity(akida_4bit, samples=samples))" + ] + }, + { + "cell_type": "markdown", + "id": "cell-25", + "metadata": {}, + "source": [ + "## Summary" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "cell-26", + "metadata": {}, + "outputs": [], + "source": [ + "print(f'{\"Variant\":<16}{\"Keras IoU\":<12}{\"Keras acc\":<12}{\"Akida IoU\":<12}{\"Akida acc\":<12}')\n", + "print(f'{\"float\":<16}{float_iou:<12.4f}{float_accuracy:<12.4f}{\"-\":<12}{\"-\":<12}')\n", + "print(f'{\"8-bit (w8a8)\":<16}{iou_8bit:<12.4f}{acc_8bit:<12.4f}{akida_iou_8bit:<12.4f}{akida_acc_8bit:<12.4f}')\n", + "print(f'{\"4-bit QAT\":<16}{iou_4bit:<12.4f}{acc_4bit:<12.4f}{akida_iou_4bit:<12.4f}{akida_acc_4bit:<12.4f}')" + ] + } + ], + "metadata": { + "jupytext": { + "formats": "ipynb,py:percent" + }, + "kernelspec": { + "display_name": "Python 3", + "language": "python", + "name": "python3" + }, + "language_info": { + "name": "python" + } + }, + "nbformat": 4, + "nbformat_minor": 5 +} diff --git a/akida2/model_zoo/portrait128/portrait128_train.py b/akida2/model_zoo/portrait128/portrait128_train.py new file mode 100644 index 0000000..6e6f7ef --- /dev/null +++ b/akida2/model_zoo/portrait128/portrait128_train.py @@ -0,0 +1,139 @@ +#!/usr/bin/env python +# Copyright 2026 Brainchip Holdings Ltd. Apache 2.0 License +""" +Portrait128 training, for float training and quantization-aware training (QAT). + +The model outputs one logit per pixel, so the loss is binary cross-entropy on +logits. Training reports the binary IoU (mean of the person and background +IoUs) and pixel accuracy. For both, a pixel counts as "person" when its logit +is above 0, i.e. its sigmoid probability is above 0.5. + +The learning rate follows a one-cycle cosine schedule, stepped every batch: it +rises from learning_rate / 25 to learning_rate over the first 30 % of training, +then decays to learning_rate / 250000. + +Example +------- + python portrait128_train.py -d ./data/portrait128 -e 50 -lr 3e-5 \\ + -l models/akida_unet_portrait128.h5 -s models/akida_unet_portrait128.h5 +""" +import argparse +import os + +import tensorflow as tf +from tf_keras.losses import BinaryCrossentropy +from tf_keras.metrics import BinaryIoU, BinaryAccuracy +from tf_keras.optimizers.legacy import Adam +from tf_keras import regularizers +from tf_keras.layers import ReLU +from tf_keras.utils import set_random_seed + +from akida_models.param_scheduler import get_cosine_lr_scheduler +from cnn2snn import load_quantized_model + +from portrait128_data import get_data + +# Must be called before any TF ops to make GPU ops (conv backward passes, +# bilinear resize, etc.) deterministic. Has a small throughput cost. +tf.config.experimental.enable_op_determinism() + +# Disable quantizeml's run-time assertions (checks on quantized tensor values +# inside every quantized op) for training, QAT and evaluation. Set before any +# quantized model runs. +os.environ["ASSERT_ENABLED"] = "0" + + +class LogitBinaryIoU(BinaryIoU): + """Binary IoU over person and background, on logits. + + A pixel is "person" when its logit is strictly positive, the same rule as + BinaryAccuracy(threshold=0). BinaryIoU on its own counts a logit of exactly 0 + as "person", and quantized and Akida models can output exact zeros, so + without this the two metrics would disagree on the same mask. + """ + + def __init__(self, name='binary_iou', **kwargs): + super().__init__(target_class_ids=[0, 1], threshold=0.5, name=name, **kwargs) + + def update_state(self, y_true, y_pred, sample_weight=None): + y_pred = tf.cast(tf.cast(y_pred, self._dtype) > 0, self._dtype) + return super().update_state(y_true, y_pred, sample_weight) + + def get_config(self): + return {'name': self.name, 'dtype': self.dtype} + + +def get_metrics(): + """Binary IoU and pixel accuracy on logits (a pixel is "person" if its logit > 0).""" + return [LogitBinaryIoU(), BinaryAccuracy(threshold=0.0, name='accuracy')] + + +def train_portrait128(model, train_ds, val_ds, epochs, learning_rate, + regularization=None, seed=42): + set_random_seed(seed) + + if regularization is not None: + print('Adding Activity Regularization to ReLU layers') + regularizer = regularizers.L1L2(regularization, regularization) + for layer in model.layers: + if isinstance(layer, ReLU): + layer.activity_regularizer = regularizer + + # The model outputs logits (no sigmoid), so the loss applies the sigmoid + # itself (from_logits=True). + model.compile(optimizer=Adam(learning_rate=learning_rate), + loss=BinaryCrossentropy(from_logits=True), + metrics=get_metrics()) + + # One-cycle cosine schedule over every training step: warm-up for the + # first 30 % of steps, then decay. + total_steps = epochs * len(train_ds) + callbacks = [get_cosine_lr_scheduler(learning_rate, total_steps, pct_start=0.3)] + + model.fit( + train_ds, + epochs=epochs, + callbacks=callbacks, + validation_data=val_ds, + ) + + +if __name__ == '__main__': + parser = argparse.ArgumentParser() + parser.add_argument('-l', '--loadmodel', required=True, + help='Model to load (.h5 tf_keras or quantized model)') + parser.add_argument('-s', '--savemodel', required=True, + help='Model save path') + parser.add_argument('-d', '--data', default='./data/portrait128', + help='Portrait128 folder (contains img_uint8.npy and msk_uint8.npy)') + parser.add_argument('-b', '--batch_size', type=int, default=32) + parser.add_argument('-e', '--epochs', type=int, default=50) + parser.add_argument('-lr', '--learning_rate', type=float, default=3e-5, + help='Peak learning rate') + parser.add_argument('-reg', '--regularization', type=float, default=None, + help='Activity Regularization to increase sparsity') + parser.add_argument('--seed', type=int, default=42, + help='Random seed for reproducibility') + args = parser.parse_args() + + # --------------------------------------------------------------------------- + # Model + # --------------------------------------------------------------------------- + model = load_quantized_model(args.loadmodel) + + # --------------------------------------------------------------------------- + # Data loading + # --------------------------------------------------------------------------- + train_ds, val_ds = get_data(args.data, model.input_shape[1:], args.batch_size, + seed=args.seed) + + train_portrait128(model=model, + train_ds=train_ds, + val_ds=val_ds, + epochs=args.epochs, + learning_rate=args.learning_rate, + regularization=args.regularization, + seed=args.seed) + + model.save(args.savemodel, include_optimizer=False) + print(f'Model saved as {args.savemodel}.') diff --git a/akida2/model_zoo/portrait128/portrait128_train.sh b/akida2/model_zoo/portrait128/portrait128_train.sh new file mode 100644 index 0000000..0435054 --- /dev/null +++ b/akida2/model_zoo/portrait128/portrait128_train.sh @@ -0,0 +1,63 @@ +#!/usr/bin/env bash +# Copyright 2026 Brainchip Holdings Ltd. Apache 2.0 License +# +# End-to-end Portrait128 segmentation pipeline for Akida 2. +# +# The model's encoder starts from ImageNet-trained AkidaNet 0.5 weights; the +# decoder is trained from random initialization. +# +# Quantization uses `quantizeml`. Two quantized variants are produced: +# * 8-bit (i8/w8/a8) via PTQ -- no QAT needed (8-bit PTQ is accurate enough). +# * 4-bit (i8/w4/a4) via QAT only -- 4-bit PTQ accuracy is poor, so the PTQ +# model is a throwaway on the way to QAT (not stored/evaluated). +# The input layer weights are always 8-bit (-i 8). +# Conversion to .fbz is done with `cnn2snn convert` (accepts quantizeml models). +# +# Usage: +# bash portrait128_train.sh [DATADIR] +# where the optional DATADIR overrides the default dataset location +# (./data/portrait128). + +set -e + +DATADIR="${1:-}" +DATA_ARG=${DATADIR:+-d "$DATADIR"} + +# Download batch of samples for calibration +wget -N https://data.brainchip.com/dataset-mirror/samples/portrait128/portrait128_batch1024.npz \ + -P data/ + +# 1. Build the model (ImageNet-pretrained AkidaNet encoder + untrained decoder) +python portrait128_model.py -bw imagenet -s models/akida_unet_portrait128.h5 + +# 2. Train the model +python portrait128_train.py -l models/akida_unet_portrait128.h5 -s models/akida_unet_portrait128.h5 -e 50 -lr 3e-5 $DATA_ARG +python portrait128_eval.py -l models/akida_unet_portrait128.h5 $DATA_ARG + +# 8-BIT VARIANT (i8/w8/a8) -- PTQ only, no QAT +quantizeml quantize -m models/akida_unet_portrait128.h5 -i 8 -w 8 -a 8 \ + -s models/akida_unet_portrait128_i8_w8_a8.h5 \ + --samples data/portrait128_batch1024.npz + +python portrait128_eval.py -l models/akida_unet_portrait128_i8_w8_a8.h5 $DATA_ARG + +cnn2snn convert -m models/akida_unet_portrait128_i8_w8_a8.h5 +python portrait128_eval.py -l models/akida_unet_portrait128_i8_w8_a8.fbz $DATA_ARG + +python portrait128_benchmark.py -l models/akida_unet_portrait128_i8_w8_a8.fbz $DATA_ARG || true + +# 4-BIT VARIANT (i8/w4/a4) -- QAT only. 4-bit PTQ is a throwaway (_pretmp). +quantizeml quantize -m models/akida_unet_portrait128.h5 -i 8 -w 4 -a 4 \ + -s models/akida_unet_portrait128_i8_w4_a4_pretmp.h5 \ + --samples data/portrait128_batch1024.npz + +# QAT at a tenth of the float learning rate +python portrait128_train.py -l models/akida_unet_portrait128_i8_w4_a4_pretmp.h5 -s models/akida_unet_portrait128_i8_w4_a4_qat.h5 -e 10 -lr 3e-6 $DATA_ARG +python portrait128_eval.py -l models/akida_unet_portrait128_i8_w4_a4_qat.h5 $DATA_ARG + +cnn2snn convert -m models/akida_unet_portrait128_i8_w4_a4_qat.h5 +python portrait128_eval.py -l models/akida_unet_portrait128_i8_w4_a4_qat.fbz $DATA_ARG + +python portrait128_benchmark.py -l models/akida_unet_portrait128_i8_w4_a4_qat.fbz $DATA_ARG || true + +rm -f models/akida_unet_portrait128_i8_w4_a4_pretmp.h5 diff --git a/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128.h5 b/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128.h5 new file mode 100644 index 0000000..6db74ec --- /dev/null +++ b/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128.h5 @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4728add11ebd0bb752c8c6e391968330bcf53be9508e0c91207b0ba7ab50ce71 +size 4501976 diff --git a/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.fbz b/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.fbz new file mode 100644 index 0000000..f2ac49b --- /dev/null +++ b/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.fbz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4416cdcbb322100b43d624317f7938e44111ed215390ffff5d8b8290cc4af5b7 +size 1302167 diff --git a/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.h5 b/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.h5 new file mode 100644 index 0000000..c488331 --- /dev/null +++ b/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.h5 @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2dd085aae619fcd16576146f7e7f31a3ac830a01c6a963cbc20d10711ee1e1cc +size 4525112 diff --git a/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w8_a8.fbz b/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w8_a8.fbz new file mode 100644 index 0000000..8f1fc2b --- /dev/null +++ b/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w8_a8.fbz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:821fb2b8525857ae0d0e4b8f15de6bac8e7388c88efbdfa91ef05adf149a4dda +size 1302167 diff --git a/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w8_a8.h5 b/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w8_a8.h5 new file mode 100644 index 0000000..25aa18a --- /dev/null +++ b/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w8_a8.h5 @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:81e3a6ad391c9fb2e85742f491020656797c041dc952562942485571646fb04d +size 4525096 diff --git a/akida2/model_zoo/portrait128/update_readme.py b/akida2/model_zoo/portrait128/update_readme.py new file mode 100644 index 0000000..152a422 --- /dev/null +++ b/akida2/model_zoo/portrait128/update_readme.py @@ -0,0 +1,15 @@ +#!/usr/bin/env python +# Copyright 2025 Brainchip Holdings Ltd. Apache 2.0 License +"""Regenerate README.md from README.md.template + metrics.json.""" +import json +import pathlib +import runpy + +here = pathlib.Path(__file__).parent +metrics = json.loads((here / "docs" / "metrics.json").read_text()) +template = (here / "docs" / "README.md.template").read_text() +(here / "README.md").write_text(template.format_map(metrics)) +print("README.md updated.") + +# Refresh the Akida 2 landing README, whose model zoo table reads this metrics.json. +runpy.run_path(str(here.parents[1] / "update_readme.py"), run_name="__main__") From 4056847471ebd0b28fdafe58d52ded3aea97aef3 Mon Sep 17 00:00:00 2001 From: mtorres Date: Thu, 22 Oct 2026 13:08:59 -0700 Subject: [PATCH 2/5] docs(akida2/portrait128): sync review fixes and README wording Refs HUB-43 --- akida2/model_zoo/portrait128/README.md | 34 ++++++++++++++++--- .../portrait128/docs/README.md.template | 34 ++++++++++++++++--- .../portrait128_notebook_benchmark.ipynb | 10 +++--- akida2/model_zoo/portrait128/wget-log | 6 ++++ 4 files changed, 68 insertions(+), 16 deletions(-) create mode 100644 akida2/model_zoo/portrait128/wget-log diff --git a/akida2/model_zoo/portrait128/README.md b/akida2/model_zoo/portrait128/README.md index 1f21567..8a4d70a 100644 --- a/akida2/model_zoo/portrait128/README.md +++ b/akida2/model_zoo/portrait128/README.md @@ -180,8 +180,24 @@ Computer Graphics Forum (Eurographics), 2016 (the PFCN dataset). The dataset is distributed by the Portrait-Segmentation project as [Portseg_128](https://drive.google.com/file/d/1UBLzvcqvt_fin9Y-48I_-lWQYfYpt_6J/view) -(Google Drive). The scripts expect its two arrays, `img_uint8.npy` (images) and -`msk_uint8.npy` (masks, 0 or 255), in one folder. +(Google Drive). Download the archive and extract it: the scripts need its two arrays, +`img_uint8.npy` (images) and `msk_uint8.npy` (masks, 0 or 255), directly in one folder, +not in a subfolder: + +``` +data/portrait128/ +├── img_uint8.npy +└── msk_uint8.npy +``` + +To check the arrays before training: + +```bash +python -c "import numpy as np; x = np.load('data/portrait128/img_uint8.npy', mmap_mode='r'); y = np.load('data/portrait128/msk_uint8.npy', mmap_mode='r'); print(x.shape, x.dtype, y.shape, y.dtype, np.unique(y[:20]))" +``` + +You should see `(18698, 128, 128, 3) uint8 (18698, 128, 128, 1) uint8 [ 0 255]`. Masks +stored as 0/1 instead of 0/255 would be read as all background. The scripts default to looking for the data at `./data/portrait128`. If you want to store the dataset on a dedicated data drive, you can pass the path @@ -193,6 +209,8 @@ location and create a symbolic link from the default path (one-off step): ln -s /path/to/your/data/portrait128 ./data/portrait128 ``` +Here `/path/to/your/data/portrait128` is the folder that holds the two arrays. + This way the scripts work out of the box without any extra arguments. ### Calibration samples @@ -230,9 +248,10 @@ several variants, each converted to Akida format: ## Reference Models -📝 Planned: pretrained models will be made available in a `pretrained_models/` -folder once the training pipeline has been run. They are handled using the `git-lfs` -package (git large file storage); for set-up instructions, see the +Pretrained models are made available here, within the `pretrained_models/` +folder. However, those are handled using the `git-lfs` package (git large +file storage). For those to be downloaded with the repo, you will need to +set up `git-lfs`. For further instructions, see the [Trained models](../../../README.md#trained-models) section of the top-level README. ## Usage @@ -250,6 +269,11 @@ here if you want to understand *why* the pipeline is structured the way it is. [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/Brainchip-Inc/brainchip_devhub/blob/main/akida2/model_zoo/portrait128/portrait128_notebook_training.ipynb) +On Colab, the setup cell installs the toolchain but can't download the dataset, which is +only distributed through Google Drive. Copy `img_uint8.npy` and `msk_uint8.npy` into the +notebook's `data/portrait128` folder before running the data cells; the setup cell prints +where, and how to mount your Drive. + [portrait128_notebook_benchmark.ipynb](portrait128_notebook_benchmark.ipynb) walks through evaluation of model accuracy on Akida and, if a hardware device is available, covers benchmarking of model latency. diff --git a/akida2/model_zoo/portrait128/docs/README.md.template b/akida2/model_zoo/portrait128/docs/README.md.template index 08026d7..3df918c 100644 --- a/akida2/model_zoo/portrait128/docs/README.md.template +++ b/akida2/model_zoo/portrait128/docs/README.md.template @@ -180,8 +180,24 @@ Computer Graphics Forum (Eurographics), 2016 (the PFCN dataset). The dataset is distributed by the Portrait-Segmentation project as [Portseg_128](https://drive.google.com/file/d/1UBLzvcqvt_fin9Y-48I_-lWQYfYpt_6J/view) -(Google Drive). The scripts expect its two arrays, `img_uint8.npy` (images) and -`msk_uint8.npy` (masks, 0 or 255), in one folder. +(Google Drive). Download the archive and extract it: the scripts need its two arrays, +`img_uint8.npy` (images) and `msk_uint8.npy` (masks, 0 or 255), directly in one folder, +not in a subfolder: + +``` +data/portrait128/ +├── img_uint8.npy +└── msk_uint8.npy +``` + +To check the arrays before training: + +```bash +python -c "import numpy as np; x = np.load('data/portrait128/img_uint8.npy', mmap_mode='r'); y = np.load('data/portrait128/msk_uint8.npy', mmap_mode='r'); print(x.shape, x.dtype, y.shape, y.dtype, np.unique(y[:20]))" +``` + +You should see `(18698, 128, 128, 3) uint8 (18698, 128, 128, 1) uint8 [ 0 255]`. Masks +stored as 0/1 instead of 0/255 would be read as all background. The scripts default to looking for the data at `./data/portrait128`. If you want to store the dataset on a dedicated data drive, you can pass the path @@ -193,6 +209,8 @@ location and create a symbolic link from the default path (one-off step): ln -s /path/to/your/data/portrait128 ./data/portrait128 ``` +Here `/path/to/your/data/portrait128` is the folder that holds the two arrays. + This way the scripts work out of the box without any extra arguments. ### Calibration samples @@ -230,9 +248,10 @@ several variants, each converted to Akida format: ## Reference Models -📝 Planned: pretrained models will be made available in a `pretrained_models/` -folder once the training pipeline has been run. They are handled using the `git-lfs` -package (git large file storage); for set-up instructions, see the +Pretrained models are made available here, within the `pretrained_models/` +folder. However, those are handled using the `git-lfs` package (git large +file storage). For those to be downloaded with the repo, you will need to +set up `git-lfs`. For further instructions, see the [Trained models](../../../README.md#trained-models) section of the top-level README. ## Usage @@ -250,6 +269,11 @@ here if you want to understand *why* the pipeline is structured the way it is. [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/Brainchip-Inc/brainchip_devhub/blob/main/akida2/model_zoo/portrait128/portrait128_notebook_training.ipynb) +On Colab, the setup cell installs the toolchain but can't download the dataset, which is +only distributed through Google Drive. Copy `img_uint8.npy` and `msk_uint8.npy` into the +notebook's `data/portrait128` folder before running the data cells; the setup cell prints +where, and how to mount your Drive. + [portrait128_notebook_benchmark.ipynb](portrait128_notebook_benchmark.ipynb) walks through evaluation of model accuracy on Akida and, if a hardware device is available, covers benchmarking of model latency. diff --git a/akida2/model_zoo/portrait128/portrait128_notebook_benchmark.ipynb b/akida2/model_zoo/portrait128/portrait128_notebook_benchmark.ipynb index 3a33bc2..44bf7c6 100644 --- a/akida2/model_zoo/portrait128/portrait128_notebook_benchmark.ipynb +++ b/akida2/model_zoo/portrait128/portrait128_notebook_benchmark.ipynb @@ -70,10 +70,6 @@ "through `git-lfs`; see the [Trained models](../../../README.md#trained-models) section of the\n", "top-level README for set-up.\n", "\n", - "📝 Planned: the pretrained models for this example are added once the training pipeline has\n", - "been run. Until then, train your own with `portrait128_notebook_training.ipynb` or\n", - "`portrait128_train.sh` and point `MODELS_DIR` at `./models/`.\n", - "\n", "The model takes a uint8 RGB image of 128 × 128 and returns one logit per pixel, as a\n", "`(128, 128, 1)` map. Its first layer, `InputConv2D`, runs on Akida's dedicated input\n", "processor, which takes the uint8 image directly." @@ -123,7 +119,7 @@ "source": [ "## Samples\n", "\n", - "Akida's neural processors only do work for non-zero activations, so latency depends on the input. Random or synthetic images drive unrepresentative activity and give misleading timings: a random image has no smooth regions, and its activations are much less sparse than a real portrait's. So the benchmark runs on real images from the training split." + "Akida's standard neural processors only do work for non-zero activations, so latency depends on the input. Random or synthetic data drives unrepresentative activity and gives misleading timings, so the benchmark runs on real images from the training split." ] }, { @@ -190,7 +186,9 @@ "\n", "The per-layer benchmark (Minimal mapping) shows where the time goes, and the sparsity printout shows why. A layer's cost grows with the number of non-zero activations it receives, so on Akida, per-layer latency and input sparsity are linked: a layer with dense input and a large feature map is slow.\n", "\n", - "In this model, look at the last decoder blocks. They run at 64 × 64 and 128 × 128, so they receive far more activations than the encoder layers at 8 × 8 or 4 × 4. Their sparsity therefore matters most.\n", + "In this model, look at the last decoder blocks. They run at 64 × 64 and 128 × 128, so they receive far more activations than the encoder layers at 8 × 8 or 4 × 4, and their sparsity is likely to matter most. The per-layer results show whether it does.\n", + "\n", + "The input layer (`conv_0`) is the exception to the sparsity rule: it runs on the dedicated input processor, whose cost doesn't depend on how many of its inputs are zero.\n", "\n", "Sparsity doesn't need hardware, so it is also computed on the software backend when no device is present." ] diff --git a/akida2/model_zoo/portrait128/wget-log b/akida2/model_zoo/portrait128/wget-log new file mode 100644 index 0000000..3b74abf --- /dev/null +++ b/akida2/model_zoo/portrait128/wget-log @@ -0,0 +1,6 @@ +--2026-10-21 13:42:16-- https://data.brainchip.com/dataset-mirror/samples/portrait128/portrait128_batch1024.npz +Resolving data.brainchip.com (data.brainchip.com)... 146.59.209.152, 2001:41d0:301::31 +Connecting to data.brainchip.com (data.brainchip.com)|146.59.209.152|:443... connected. +HTTP request sent, awaiting response... 304 Not Modified +File ‘data/portrait128_batch1024.npz’ not modified on server. Omitting download. + From ae0eec0170e5eb51e45e76cfa6abe042f540d581 Mon Sep 17 00:00:00 2001 From: mtorres Date: Thu, 22 Oct 2026 13:15:53 -0700 Subject: [PATCH 3/5] docs: link Portrait128 from the root README Refs HUB-43 --- README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README.md b/README.md index 9fbd7d1..4450600 100644 --- a/README.md +++ b/README.md @@ -57,7 +57,7 @@ Akida 1 and Akida 2 examples are available today; more Akida 2 models can be fou | --- | --- | --- | --- | | **Chip** | AKD1500 | AKD2500 | — | | **Typical use cases** | Image classification, keyword spotting, object detection | Larger models, higher accuracy targets | Always-on sensing, edge inference | -| **Examples in this repo** | [Image Classification (PlantVillage)](akida1/model_zoo/plant_village) · [ImageNet / AkidaNet](akida1/model_zoo/imagenet_akidanet) · [Visual Wake Words](akida1/model_zoo/vww) · [Keyword Spotting](akida1/model_zoo/speech_commands) · [ECG Arrhythmia](akida1/model_zoo/arrhythmia_classification) | [Visual Wake Words](akida2/model_zoo/vww) · [Image Classification (PlantVillage)](akida2/model_zoo/plant_village) · [Keyword Spotting](akida2/model_zoo/speech_commands) | 🔜 Coming soon | +| **Examples in this repo** | [Image Classification (PlantVillage)](akida1/model_zoo/plant_village) · [ImageNet / AkidaNet](akida1/model_zoo/imagenet_akidanet) · [Visual Wake Words](akida1/model_zoo/vww) · [Keyword Spotting](akida1/model_zoo/speech_commands) · [ECG Arrhythmia](akida1/model_zoo/arrhythmia_classification) | [Visual Wake Words](akida2/model_zoo/vww) · [Image Classification (PlantVillage)](akida2/model_zoo/plant_village) · [Keyword Spotting](akida2/model_zoo/speech_commands) · [Portrait Segmentation (Portrait128)](akida2/model_zoo/portrait128) | 🔜 Coming soon | --- From df7cd8a25e3f5cf037813a09cce4e5050775cccb Mon Sep 17 00:00:00 2001 From: mtorres Date: Thu, 22 Oct 2026 16:52:14 -0700 Subject: [PATCH 4/5] feat(akida2/portrait128): 8-bit only, with one epoch of QAT as in the model zoo Refs HUB-43 --- .claude/knowledge/akida.md | 18 +++ akida2/README.md | 16 +- akida2/model_zoo/portrait128/README.md | 84 +++------- .../portrait128/docs/README.md.template | 76 +++------ .../model_zoo/portrait128/docs/metrics.json | 28 +--- .../model_zoo/portrait128/docs/zoo_card.json | 10 -- .../portrait128/portrait128_benchmark.py | 11 +- .../model_zoo/portrait128/portrait128_eval.py | 14 +- .../portrait128/portrait128_notebook.py | 132 +++++++--------- .../portrait128_notebook_benchmark.ipynb | 8 +- .../portrait128_notebook_training.ipynb | 148 ++++++++---------- .../portrait128/portrait128_train.py | 3 +- .../portrait128/portrait128_train.sh | 40 ++--- .../akida_unet_portrait128_i8_w4_a4_qat.fbz | 3 - .../akida_unet_portrait128_i8_w4_a4_qat.h5 | 3 - .../akida_unet_portrait128_i8_w8_a8.fbz | 3 - .../akida_unet_portrait128_i8_w8_a8.h5 | 3 - .../akida_unet_portrait128_i8_w8_a8_qat.fbz | 3 + .../akida_unet_portrait128_i8_w8_a8_qat.h5 | 3 + akida2/model_zoo/portrait128/wget-log | 6 - 20 files changed, 219 insertions(+), 393 deletions(-) delete mode 100644 akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.fbz delete mode 100644 akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.h5 delete mode 100644 akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w8_a8.fbz delete mode 100644 akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w8_a8.h5 create mode 100644 akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w8_a8_qat.fbz create mode 100644 akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w8_a8_qat.h5 delete mode 100644 akida2/model_zoo/portrait128/wget-log diff --git a/.claude/knowledge/akida.md b/.claude/knowledge/akida.md index 554cf2d..40dc866 100644 --- a/.claude/knowledge/akida.md +++ b/.claude/knowledge/akida.md @@ -161,6 +161,24 @@ pipeline instead, so the deployed model receives an already-shaped tensor. how much pooling can come before it. *Source: `akida1/model_zoo/uored_vafcls/uored_vafcls_model.py` docstring.* +**[Akida 2] `DepthwiseConv2DTranspose` maps to hardware only with 8-bit inputs and 8-bit +weights.** At 4 bits the layer falls back to software. So a model with transposed +depthwise convolutions, quantized to 4 bits throughout (`-w 4 -a 4`), no longer runs +fully on hardware: the mapper splits it into hardware and software sequences, and +mapping with `hw_only=True` fails with "Input should be 8-bit. Weights should be 8-bit." +Regular `DepthwiseConv2D` and pointwise `Conv2D` layers map at 4 bits. +- That's why `akida2/model_zoo/portrait128`, an AkidaUNet with five transposed depthwise + layers in its decoder, has no 4-bit variant. BrainChip's model zoo also lists that model + at 8 bits only. +- A mixed-precision 4-bit model should be possible: `quantizeml.models.quantize` takes a + per-layer `q_config`, which could keep these layers, and the activations feeding them, + at 8 bits. **(unconfirmed)**: not yet built or benchmarked; mapping such a model with + `hw_only=True` would confirm it. + +*Source: measured with akida 2.19.3, mapping onto `akida.SixNodesIPv2()` and on the +Akida 2 FPGA, 2026-10-22 (`akida2/model_zoo/portrait128`, HUB-43); +[model zoo performance](https://doc.brainchipinc.com/model_zoo_performance.html#segmentation).* + ## Backbones and blocks **[both] Separable convolutions: fused on Akida 1, two layers on Akida 2.** diff --git a/akida2/README.md b/akida2/README.md index 4533e4b..af8ebf4 100644 --- a/akida2/README.md +++ b/akida2/README.md @@ -163,19 +163,11 @@ benchmarks for every mapping, and the steps to reproduce them. - Portrait segmentation - Segmentation - Portrait128 + Portrait segmentation + Segmentation + Portrait128 8-bit - 0.8861 binary IoU - — - — - — - - - - 4-bit QAT - 0.9091 binary IoU + 0.9185 binary IoU — — — diff --git a/akida2/model_zoo/portrait128/README.md b/akida2/model_zoo/portrait128/README.md index 8a4d70a..326b4f7 100644 --- a/akida2/model_zoo/portrait128/README.md +++ b/akida2/model_zoo/portrait128/README.md @@ -16,11 +16,16 @@ thresholded at 0 (a probability of 0.5): - **Pixel accuracy**: the fraction of pixels labeled correctly. Background covers much of each image, so this is always high and says less about mask quality. -The quantized variants below all share the same float model. On Akida 2 the -model is quantized with **`quantizeml`**: 8-bit weights and activations need no -quantization-aware training (QAT), while a lower-precision 4-bit variant (4-bit -weights and activations, 8-bit input layer) uses QAT to recover accuracy — 4-bit -PTQ accuracy is poor, so only the QAT result is reported. +On Akida 2 the model is quantized with **`quantizeml`** to 8-bit weights and +activations. Post-training quantization (PTQ) is followed by one epoch of +quantization-aware training (QAT), which fine-tunes the model with the quantization in +place, as for this model in BrainChip's model zoo. + +There is no 4-bit variant. The decoder upsamples with transposed depthwise +convolutions, and on Akida 2 those layers need 8-bit inputs and 8-bit weights, so a +model quantized to 4 bits throughout can't run fully on the hardware. BrainChip's +[model zoo](https://doc.brainchipinc.com/model_zoo_performance.html#segmentation) also +publishes this model at 8 bits only. @@ -39,21 +44,11 @@ PTQ accuracy is poor, so only the QAT result is reported. - - - - - - - - - - - - - - - + + + + + @@ -110,37 +105,12 @@ rescale of the measured cycles. - - - - - - - - - - - - - - - - - - - - - - - - -
8-bit w8 / a8-0.886194.19%0.886194.19%TBD
4-bitw4 / a4yes0.909195.48%0.909195.48%1 epoch0.918595.96%0.918595.96% TBD
TBD TBD
4-bit (QAT)MinimalTBDTBDTBDTBDTBD
AllNPsTBDTBDTBDTBDTBD
HwPrTBDTBDTBDTBDTBD
### Benchmark plots -📝 Planned: the full-model and per-layer benchmark plots for each variant are added +📝 Planned: the full-model and per-layer benchmark plots are added when the FPGA benchmark is run with `--save-metrics` (see [Contributing and Maintenance](#contributing-and-maintenance)). @@ -236,15 +206,14 @@ python portrait128_data.py --extract data/portrait128_batch1024.npz ## Pipeline -Training produces a float model, then quantizes it with `quantizeml` into -several variants, each converted to Akida format: +Training produces a float model, quantizes it to 8 bits with `quantizeml`, fine-tunes +it for one epoch of QAT, and converts the result to Akida format: | Stage | Description | |---|---| | Full-precision | Float32 training of the full model, starting from an ImageNet-trained AkidaNet encoder | -| 8-bit quantization | `quantizeml quantize` to 8-bit weights and activations (8-bit input), calibrated on real samples; no QAT required | -| 4-bit quantization | `quantizeml quantize` to 4-bit weights and activations (8-bit input), calibrated on real samples, with QAT fine-tuning — 4-bit PTQ accuracy is poor so only the QAT model is kept | -| Conversion to Akida | Automated conversion of each quantized model to Akida 2 format with `cnn2snn convert` | +| 8-bit quantization | `quantizeml quantize` to 8-bit weights and activations (8-bit input), calibrated on real samples, then one epoch of QAT at a tenth of the float learning rate; the PTQ model itself is not kept | +| Conversion to Akida | Automated conversion of the quantized model to Akida 2 format with `cnn2snn convert` | ## Reference Models @@ -308,15 +277,10 @@ using `update_readme.py`: # Float model python portrait128_eval.py -l pretrained_models/akida_unet_portrait128.h5 --save-metrics -# 8-bit variant -python portrait128_eval.py -l pretrained_models/akida_unet_portrait128_i8_w8_a8.h5 --save-metrics -python portrait128_eval.py -l pretrained_models/akida_unet_portrait128_i8_w8_a8.fbz --save-metrics -python portrait128_benchmark.py -l pretrained_models/akida_unet_portrait128_i8_w8_a8.fbz --save-metrics - -# 4-bit variant (QAT) -python portrait128_eval.py -l pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.h5 --save-metrics -python portrait128_eval.py -l pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.fbz --save-metrics -python portrait128_benchmark.py -l pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.fbz --save-metrics +# 8-bit model (QAT) +python portrait128_eval.py -l pretrained_models/akida_unet_portrait128_i8_w8_a8_qat.h5 --save-metrics +python portrait128_eval.py -l pretrained_models/akida_unet_portrait128_i8_w8_a8_qat.fbz --save-metrics +python portrait128_benchmark.py -l pretrained_models/akida_unet_portrait128_i8_w8_a8_qat.fbz --save-metrics python update_readme.py ``` diff --git a/akida2/model_zoo/portrait128/docs/README.md.template b/akida2/model_zoo/portrait128/docs/README.md.template index 3df918c..e20629d 100644 --- a/akida2/model_zoo/portrait128/docs/README.md.template +++ b/akida2/model_zoo/portrait128/docs/README.md.template @@ -16,11 +16,16 @@ thresholded at 0 (a probability of 0.5): - **Pixel accuracy**: the fraction of pixels labeled correctly. Background covers much of each image, so this is always high and says less about mask quality. -The quantized variants below all share the same float model. On Akida 2 the -model is quantized with **`quantizeml`**: 8-bit weights and activations need no -quantization-aware training (QAT), while a lower-precision 4-bit variant (4-bit -weights and activations, 8-bit input layer) uses QAT to recover accuracy — 4-bit -PTQ accuracy is poor, so only the QAT result is reported. +On Akida 2 the model is quantized with **`quantizeml`** to 8-bit weights and +activations. Post-training quantization (PTQ) is followed by one epoch of +quantization-aware training (QAT), which fine-tunes the model with the quantization in +place, as for this model in BrainChip's model zoo. + +There is no 4-bit variant. The decoder upsamples with transposed depthwise +convolutions, and on Akida 2 those layers need 8-bit inputs and 8-bit weights, so a +model quantized to 4 bits throughout can't run fully on the hardware. BrainChip's +[model zoo](https://doc.brainchipinc.com/model_zoo_performance.html#segmentation) also +publishes this model at 8 bits only. @@ -39,23 +44,13 @@ PTQ accuracy is poor, so only the QAT result is reported. - + - - - - - - - - - -
8-bit w8 / a8-1 epoch {w8a8_quant_iou} {w8a8_quant_acc} {w8a8_akida_iou} {w8a8_akida_acc} {w8a8_sparsity}
4-bitw4 / a4yes{w4a4_qat_quant_iou}{w4a4_qat_quant_acc}{w4a4_qat_akida_iou}{w4a4_qat_akida_acc}{w4a4_qat_sparsity}
@@ -110,37 +105,12 @@ rescale of the measured cycles. {w8a8_hwpr_latency_ms} {w8a8_hwpr_projected_ms} - - 4-bit (QAT) - Minimal - {w4a4_qat_minimal_nps} - {w4a4_qat_minimal_passes} - {w4a4_qat_minimal_cycles} - {w4a4_qat_minimal_latency_ms} - {w4a4_qat_minimal_projected_ms} - - - AllNPs - {w4a4_qat_allnps_nps} - {w4a4_qat_allnps_passes} - {w4a4_qat_allnps_cycles} - {w4a4_qat_allnps_latency_ms} - {w4a4_qat_allnps_projected_ms} - - - HwPr - {w4a4_qat_hwpr_nps} - {w4a4_qat_hwpr_passes} - {w4a4_qat_hwpr_cycles} - {w4a4_qat_hwpr_latency_ms} - {w4a4_qat_hwpr_projected_ms} - ### Benchmark plots -📝 Planned: the full-model and per-layer benchmark plots for each variant are added +📝 Planned: the full-model and per-layer benchmark plots are added when the FPGA benchmark is run with `--save-metrics` (see [Contributing and Maintenance](#contributing-and-maintenance)). @@ -236,15 +206,14 @@ python portrait128_data.py --extract data/portrait128_batch1024.npz ## Pipeline -Training produces a float model, then quantizes it with `quantizeml` into -several variants, each converted to Akida format: +Training produces a float model, quantizes it to 8 bits with `quantizeml`, fine-tunes +it for one epoch of QAT, and converts the result to Akida format: | Stage | Description | |---|---| | Full-precision | Float32 training of the full model, starting from an ImageNet-trained AkidaNet encoder | -| 8-bit quantization | `quantizeml quantize` to 8-bit weights and activations (8-bit input), calibrated on real samples; no QAT required | -| 4-bit quantization | `quantizeml quantize` to 4-bit weights and activations (8-bit input), calibrated on real samples, with QAT fine-tuning — 4-bit PTQ accuracy is poor so only the QAT model is kept | -| Conversion to Akida | Automated conversion of each quantized model to Akida 2 format with `cnn2snn convert` | +| 8-bit quantization | `quantizeml quantize` to 8-bit weights and activations (8-bit input), calibrated on real samples, then one epoch of QAT at a tenth of the float learning rate; the PTQ model itself is not kept | +| Conversion to Akida | Automated conversion of the quantized model to Akida 2 format with `cnn2snn convert` | ## Reference Models @@ -308,15 +277,10 @@ using `update_readme.py`: # Float model python portrait128_eval.py -l pretrained_models/akida_unet_portrait128.h5 --save-metrics -# 8-bit variant -python portrait128_eval.py -l pretrained_models/akida_unet_portrait128_i8_w8_a8.h5 --save-metrics -python portrait128_eval.py -l pretrained_models/akida_unet_portrait128_i8_w8_a8.fbz --save-metrics -python portrait128_benchmark.py -l pretrained_models/akida_unet_portrait128_i8_w8_a8.fbz --save-metrics - -# 4-bit variant (QAT) -python portrait128_eval.py -l pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.h5 --save-metrics -python portrait128_eval.py -l pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.fbz --save-metrics -python portrait128_benchmark.py -l pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.fbz --save-metrics +# 8-bit model (QAT) +python portrait128_eval.py -l pretrained_models/akida_unet_portrait128_i8_w8_a8_qat.h5 --save-metrics +python portrait128_eval.py -l pretrained_models/akida_unet_portrait128_i8_w8_a8_qat.fbz --save-metrics +python portrait128_benchmark.py -l pretrained_models/akida_unet_portrait128_i8_w8_a8_qat.fbz --save-metrics python update_readme.py ``` diff --git a/akida2/model_zoo/portrait128/docs/metrics.json b/akida2/model_zoo/portrait128/docs/metrics.json index 7ab4ce8..064b91e 100644 --- a/akida2/model_zoo/portrait128/docs/metrics.json +++ b/akida2/model_zoo/portrait128/docs/metrics.json @@ -2,28 +2,10 @@ "float_acc": "94.31%", "float_iou": "0.8883", "params": "1,058,865", - "w4a4_qat_akida_acc": "95.48%", - "w4a4_qat_akida_iou": "0.9091", - "w4a4_qat_allnps_cycles": "TBD", - "w4a4_qat_allnps_latency_ms": "TBD", - "w4a4_qat_allnps_nps": "TBD", - "w4a4_qat_allnps_passes": "TBD", - "w4a4_qat_allnps_projected_ms": "TBD", - "w4a4_qat_hwpr_cycles": "TBD", - "w4a4_qat_hwpr_latency_ms": "TBD", - "w4a4_qat_hwpr_nps": "TBD", - "w4a4_qat_hwpr_passes": "TBD", - "w4a4_qat_hwpr_projected_ms": "TBD", - "w4a4_qat_minimal_cycles": "TBD", - "w4a4_qat_minimal_latency_ms": "TBD", - "w4a4_qat_minimal_nps": "TBD", - "w4a4_qat_minimal_passes": "TBD", - "w4a4_qat_minimal_projected_ms": "TBD", - "w4a4_qat_quant_acc": "95.48%", - "w4a4_qat_quant_iou": "0.9091", - "w4a4_qat_sparsity": "TBD", - "w8a8_akida_acc": "94.19%", - "w8a8_akida_iou": "0.8861", + "w8a8_quant_iou": "0.9185", + "w8a8_quant_acc": "95.96%", + "w8a8_akida_iou": "0.9185", + "w8a8_akida_acc": "95.96%", "w8a8_allnps_cycles": "TBD", "w8a8_allnps_latency_ms": "TBD", "w8a8_allnps_nps": "TBD", @@ -39,7 +21,5 @@ "w8a8_minimal_nps": "TBD", "w8a8_minimal_passes": "TBD", "w8a8_minimal_projected_ms": "TBD", - "w8a8_quant_acc": "94.19%", - "w8a8_quant_iou": "0.8861", "w8a8_sparsity": "TBD" } diff --git a/akida2/model_zoo/portrait128/docs/zoo_card.json b/akida2/model_zoo/portrait128/docs/zoo_card.json index c0d03ba..ecd0077 100644 --- a/akida2/model_zoo/portrait128/docs/zoo_card.json +++ b/akida2/model_zoo/portrait128/docs/zoo_card.json @@ -11,16 +11,6 @@ "metric_label": "binary IoU", "bench_prefix": "w8a8_", "notes": "" - }, - { - "task": "Portrait segmentation", - "category": "Segmentation", - "dataset": "Portrait128", - "variant": "4-bit QAT", - "metric_key": "w4a4_qat_akida_iou", - "metric_label": "binary IoU", - "bench_prefix": "w4a4_qat_", - "notes": "" } ] } diff --git a/akida2/model_zoo/portrait128/portrait128_benchmark.py b/akida2/model_zoo/portrait128/portrait128_benchmark.py index 687dfb8..ef0e8dc 100644 --- a/akida2/model_zoo/portrait128/portrait128_benchmark.py +++ b/akida2/model_zoo/portrait128/portrait128_benchmark.py @@ -139,8 +139,8 @@ # ------------------------------------------------------------------------- # Map without hw_only so ak_model.sequences is available for plot_mapping ak_model.map(device, mode=akida.MapMode.Minimal) - # Plot filenames include the model stem so that each variant (8-bit, 4-bit) - # keeps its own reference figures. + # Plot filenames include the model stem, so each benchmarked model keeps + # its own reference figures. model_stem = pathlib.Path(args.loadmodel).stem perlayer_savepath = f'benchmark_results_layers_{model_stem}.png' if args.save_metrics: @@ -170,12 +170,7 @@ # intentionally absent: benchmarking is latency-only until the FPGA power # path exists. stem = pathlib.Path(args.loadmodel).stem - if 'i8_w8_a8' in stem: - variant = 'w8a8' - elif 'i8_w4_a4' in stem: - variant = 'w4a4_qat' - else: - variant = 'w8a8' # fallback; benchmarking is only meaningful for a quantized .fbz + variant = 'w8a8' # the only quantized model; benchmarking needs its .fbz metrics_path = pathlib.Path(__file__).parent / 'docs' / 'metrics.json' metrics = json.loads(metrics_path.read_text()) if metrics_path.exists() else {} diff --git a/akida2/model_zoo/portrait128/portrait128_eval.py b/akida2/model_zoo/portrait128/portrait128_eval.py index f0e6e70..6f39423 100644 --- a/akida2/model_zoo/portrait128/portrait128_eval.py +++ b/akida2/model_zoo/portrait128/portrait128_eval.py @@ -102,14 +102,12 @@ def evaluate_akida_model(akida_model, val_dataset): # Updates the stored metrics used to generate the README performance # tables. For code maintenance only, run against the pretrained models. # - # This example has two quantized variants, disambiguated by filename. - # Each variant contributes "quantized" metrics (from its .h5) and "akida" - # metrics (from its .fbz): + # Models are told apart by filename. The 8-bit model contributes + # "quantized" metrics (from its .h5) and "akida" metrics (from its .fbz): # akida_unet_portrait128.h5 -> float_iou, float_acc, params - # akida_unet_portrait128_i8_w8_a8.{h5,fbz} -> w8a8_quant_* / w8a8_akida_* - # akida_unet_portrait128_i8_w4_a4_qat.{h5,fbz} -> w4a4_qat_quant_* / w4a4_qat_akida_* - # (The 4-bit PTQ model is a throwaway on the way to QAT -- it is never - # evaluated or stored, so it has no metrics.) + # akida_unet_portrait128_i8_w8_a8_qat.{h5,fbz} -> w8a8_quant_* / w8a8_akida_* + # (The PTQ model is a throwaway on the way to QAT: it is never evaluated + # or stored, so it has no metrics.) metrics_path = pathlib.Path(__file__).parent / 'docs' / 'metrics.json' metrics = json.loads(metrics_path.read_text()) if metrics_path.exists() else {} iou_str = f'{binary_iou:.4f}' @@ -119,8 +117,6 @@ def evaluate_akida_model(akida_model, val_dataset): # Determine the variant prefix from the filename. if 'i8_w8_a8' in stem: variant = 'w8a8' - elif 'i8_w4_a4' in stem: - variant = 'w4a4_qat' else: variant = None # float model diff --git a/akida2/model_zoo/portrait128/portrait128_notebook.py b/akida2/model_zoo/portrait128/portrait128_notebook.py index 44fafce..628c0c4 100644 --- a/akida2/model_zoo/portrait128/portrait128_notebook.py +++ b/akida2/model_zoo/portrait128/portrait128_notebook.py @@ -26,10 +26,9 @@ # # The pipeline follows the standard Akida workflow: # 1. Train a float model -# 2. 8-bit quantization (PTQ) -# 3. 4-bit quantization with QAT fine-tuning -# 4. Conversion to Akida `.fbz` format -# 5. Evaluation on Akida +# 2. 8-bit quantization (PTQ), then one epoch of QAT +# 3. Conversion to Akida `.fbz` format +# 4. Evaluation on Akida # # The focus is on the **Akida-specific** aspects of the pipeline. The data loading, # model definition and training loop live in the accompanying Python files: they are @@ -57,9 +56,9 @@ # `img_uint8.npy` and `msk_uint8.npy`. See the [README](README.md) for where to get # them and for symlink setup. Update `DATA_PATH` below if you keep them elsewhere. # -# Setting `RUN_FLOAT_TRAINING` or `RUN_QAT_TRAINING` to `False` skips that training -# run and loads the model saved in `MODELS_DIR` by an earlier run of this notebook or -# of `portrait128_train.sh`. +# Setting `RUN_FLOAT_TRAINING` to `False` skips the training run and loads the float +# model saved in `MODELS_DIR` by an earlier run of this notebook or of +# `portrait128_train.sh`. # %% import os @@ -72,7 +71,6 @@ os.makedirs(MODELS_DIR, exist_ok=True) RUN_FLOAT_TRAINING = True -RUN_QAT_TRAINING = True SEED = 42 @@ -198,17 +196,17 @@ # ## Quantization (quantizeml) # # Akida runs on integers, so the weights and activations have to be quantized. Akida 2 -# quantizes with **quantizeml** (not `cnn2snn.quantize`). We produce two variants: -# -# * **8-bit** (i8/w8/a8): post-training quantization (PTQ) only. 256 levels per weight -# and activation track the float values closely enough that no retraining is needed. -# * **4-bit** (i8/w4/a4): quantization-aware training (QAT). With only 16 levels the -# rounding error is large enough to cost real accuracy after PTQ, so the model is -# fine-tuned with the quantization in place. -# -# In both variants the **input layer's weights stay 8-bit** (`input_weight_bits=8`). The -# input layer runs on Akida's dedicated input processor and sees the raw image, where -# precision matters most; it is also a small share of the total weights. +# quantizes with **quantizeml** (not `cnn2snn.quantize`). We quantize to **8 bits** +# (i8/w8/a8) by post-training quantization (PTQ) only: 256 levels per weight and +# activation track the float values closely enough that no retraining is needed. +# `input_weight_bits` sets the input layer separately, because it runs on Akida's +# dedicated input processor; here it is 8-bit like everything else. +# +# **Why no 4-bit variant?** Other Akida 2 examples also offer a 4-bit model, fine-tuned +# with quantization-aware training (QAT). This one can't. The decoder upsamples with +# transposed depthwise convolutions, and on Akida 2 those layers need 8-bit inputs and +# 8-bit weights. In a model quantized to 4 bits throughout, all five of them would fall +# back to software, so the model would no longer run fully on the hardware. # # Quantization with `quantizeml` requires **calibration samples**. Calibration runs real # inputs through the model to measure the range of every layer's activations, and those @@ -224,17 +222,44 @@ NUM_SAMPLES = 1024 samples = get_samples(DATA_PATH, INPUT_SHAPE, num_samples=NUM_SAMPLES) -# --- 8-bit variant (i8 / w8 / a8), PTQ only --- +# --- 8-bit (i8 / w8 / a8), PTQ --- qparams_8bit = QuantizationParams(input_weight_bits=8, weight_bits=8, activation_bits=8) model_8bit = quantize(model, qparams=qparams_8bit, samples=samples) -q8_path = os.path.join(MODELS_DIR, 'akida_unet_portrait128_i8_w8_a8.h5') +model_8bit.compile(metrics=get_metrics()) +ptq_results = model_8bit.evaluate(val_ds, verbose=0, return_dict=True) +iou_ptq, acc_ptq = ptq_results['binary_iou'], ptq_results['accuracy'] +print(f'8-bit PTQ binary IoU: {iou_ptq:.4f}, pixel accuracy: {acc_ptq:.4f}') + +# %% [markdown] +# ### One epoch of QAT +# +# The 8-bit model then gets one epoch of quantization-aware training (QAT), as the +# model in BrainChip's model zoo does. During QAT the forward pass uses the quantized +# weights and activations, while the gradients update an underlying float copy of the +# weights, so the network adjusts to the rounding that PTQ introduced. +# +# QAT starts from a model that is already close to a good solution, so it uses a lower +# learning rate: a tenth of the float rate here. A higher rate would undo the float +# training rather than refine it. +# +# A model quantized with `quantizeml` is a standard Keras model, so it goes straight +# back into the same training function used for float training. The PTQ model itself +# is not kept; compare its result above with the one after QAT. + +# %% +QAT_EPOCHS = 1 +QAT_LR = 3e-6 +train_ds, val_ds = get_data(DATA_PATH, INPUT_SHAPE, BATCH_SIZE, seed=SEED) +train_portrait128(model_8bit, train_ds, val_ds, QAT_EPOCHS, QAT_LR, seed=SEED) + +q8_path = os.path.join(MODELS_DIR, 'akida_unet_portrait128_i8_w8_a8_qat.h5') model_8bit.save(q8_path, include_optimizer=False) model_8bit.compile(metrics=get_metrics()) q8_results = model_8bit.evaluate(val_ds, verbose=0, return_dict=True) iou_8bit, acc_8bit = q8_results['binary_iou'], q8_results['accuracy'] -print(f'8-bit binary IoU: {iou_8bit:.4f}, pixel accuracy: {acc_8bit:.4f}') +print(f'8-bit QAT binary IoU: {iou_8bit:.4f}, pixel accuracy: {acc_8bit:.4f}') # %% [markdown] # Note: the quantized model can be saved using the standard method. However, @@ -249,55 +274,13 @@ from quantizeml.model_io import load_model model_8bit = load_model(q8_path) -# %% [markdown] -# ### 4-bit variant with QAT -# -# At 4 bits, PTQ alone loses noticeable accuracy, so the quantized model is fine-tuned. -# During QAT the forward pass uses the quantized weights and activations, while the -# gradients update an underlying float copy of the weights. The network therefore -# learns weights that still work after rounding to 16 levels. -# -# QAT starts from a model that is already close to a good solution, so it uses a lower -# learning rate: a tenth of the float rate here. A higher rate would undo the float -# training rather than refine it. -# -# A model quantized with `quantizeml` is a standard Keras model, so it goes straight -# back into the same training function used for float training. The intermediate PTQ -# model is not kept. - -# %% -if RUN_QAT_TRAINING: - # --- 4-bit variant (i8 / w4 / a4), QAT --- - qparams_4bit = QuantizationParams(input_weight_bits=8, weight_bits=4, activation_bits=4) - model_4bit = quantize(model, qparams=qparams_4bit, samples=samples) - - # QAT fine-tune the quantized 4-bit model - QAT_EPOCHS = 10 - QAT_LR = 3e-6 - train_ds, val_ds = get_data(DATA_PATH, INPUT_SHAPE, BATCH_SIZE, seed=SEED) - train_portrait128(model_4bit, train_ds, val_ds, QAT_EPOCHS, QAT_LR, seed=SEED) - - q4_path = os.path.join(MODELS_DIR, 'akida_unet_portrait128_i8_w4_a4_qat.h5') - model_4bit.save(q4_path, include_optimizer=False) -else: - # Note the use of the quantizeml load_model wrapper to handle the - # custom quantized layer types - from quantizeml.model_io import load_model - print('Training skipped. Loading an existing 4-bit model...') - model_4bit = load_model(os.path.join(MODELS_DIR, 'akida_unet_portrait128_i8_w4_a4_qat.h5')) - -model_4bit.compile(metrics=get_metrics()) -q4_results = model_4bit.evaluate(val_ds, verbose=0, return_dict=True) -iou_4bit, acc_4bit = q4_results['binary_iou'], q4_results['accuracy'] -print(f'4-bit QAT binary IoU: {iou_4bit:.4f}, pixel accuracy: {acc_4bit:.4f}') - # %% [markdown] # ## Conversion to Akida Format # # `cnn2snn.convert` compiles the quantized Keras model into an Akida `.fbz` # model that can be loaded and executed directly on an Akida 2 hardware device. # The converter verifies hardware compatibility and maps each layer to its -# corresponding Akida primitive. We convert both quantized variants. +# corresponding Akida primitive. # # In the summary below, note the following: # @@ -305,7 +288,8 @@ # convolution. # * Each decoder block becomes a `DepthwiseConv2DTranspose` followed by a pointwise # `Conv2D`. On Akida 2 the depthwise and pointwise parts of a separable convolution -# are two separate layers. +# are two separate layers. The `DepthwiseConv2DTranspose` layers are the ones that +# need 8 bits to run on the hardware. # * The last layer is a `Dequantizer`, so the model outputs float logits of shape # `(128, 128, 1)`, which can be thresholded at 0 just like the Keras model's. @@ -313,10 +297,7 @@ from cnn2snn import convert akida_8bit = convert(model_8bit) -akida_8bit.save(os.path.join(MODELS_DIR, 'akida_unet_portrait128_i8_w8_a8.fbz')) - -akida_4bit = convert(model_4bit) -akida_4bit.save(os.path.join(MODELS_DIR, 'akida_unet_portrait128_i8_w4_a4_qat.fbz')) +akida_8bit.save(os.path.join(MODELS_DIR, 'akida_unet_portrait128_i8_w8_a8_qat.fbz')) akida_8bit.summary() @@ -351,9 +332,7 @@ def evaluate_akida(akida_model, ds): return float(iou_metric.result()), float(acc_metric.result()) akida_iou_8bit, akida_acc_8bit = evaluate_akida(akida_8bit, val_ds) -akida_iou_4bit, akida_acc_4bit = evaluate_akida(akida_4bit, val_ds) -print(f'Akida 8-bit binary IoU: {akida_iou_8bit:.4f}, pixel accuracy: {akida_acc_8bit:.4f}') -print(f'Akida 4-bit QAT binary IoU: {akida_iou_4bit:.4f}, pixel accuracy: {akida_acc_4bit:.4f}') +print(f'Akida 8-bit QAT binary IoU: {akida_iou_8bit:.4f}, pixel accuracy: {akida_acc_8bit:.4f}') # %% [markdown] # ## Activation Sparsity @@ -376,10 +355,7 @@ def evaluate_akida(akida_model, ds): from brainchip_utils.plot_utils import pretty_print_sparsity # Reuse the calibration samples built earlier (sparsity needs real activations too). -print('8-bit sparsity:') pretty_print_sparsity(compute_sparsity(akida_8bit, samples=samples)) -print('\n4-bit QAT sparsity:') -pretty_print_sparsity(compute_sparsity(akida_4bit, samples=samples)) # %% [markdown] # ## Summary @@ -387,5 +363,5 @@ def evaluate_akida(akida_model, ds): # %% print(f'{"Variant":<16}{"Keras IoU":<12}{"Keras acc":<12}{"Akida IoU":<12}{"Akida acc":<12}') print(f'{"float":<16}{float_iou:<12.4f}{float_accuracy:<12.4f}{"-":<12}{"-":<12}') -print(f'{"8-bit (w8a8)":<16}{iou_8bit:<12.4f}{acc_8bit:<12.4f}{akida_iou_8bit:<12.4f}{akida_acc_8bit:<12.4f}') -print(f'{"4-bit QAT":<16}{iou_4bit:<12.4f}{acc_4bit:<12.4f}{akida_iou_4bit:<12.4f}{akida_acc_4bit:<12.4f}') +print(f'{"8-bit PTQ":<16}{iou_ptq:<12.4f}{acc_ptq:<12.4f}{"-":<12}{"-":<12}') +print(f'{"8-bit QAT":<16}{iou_8bit:<12.4f}{acc_8bit:<12.4f}{akida_iou_8bit:<12.4f}{akida_acc_8bit:<12.4f}') diff --git a/akida2/model_zoo/portrait128/portrait128_notebook_benchmark.ipynb b/akida2/model_zoo/portrait128/portrait128_notebook_benchmark.ipynb index 44bf7c6..4e5b97c 100644 --- a/akida2/model_zoo/portrait128/portrait128_notebook_benchmark.ipynb +++ b/akida2/model_zoo/portrait128/portrait128_notebook_benchmark.ipynb @@ -48,10 +48,8 @@ "# training notebook or portrait128_train.sh, set MODELS_DIR = './models/'.\n", "MODELS_DIR = './pretrained_models/'\n", "\n", - "# Which converted model to benchmark. Options produced by the training pipeline:\n", - "# akida_unet_portrait128_i8_w8_a8.fbz (8-bit)\n", - "# akida_unet_portrait128_i8_w4_a4_qat.fbz (4-bit QAT)\n", - "MODEL_FILENAME = 'akida_unet_portrait128_i8_w4_a4_qat.fbz'\n", + "# The converted 8-bit (QAT) model produced by the training pipeline\n", + "MODEL_FILENAME = 'akida_unet_portrait128_i8_w8_a8_qat.fbz'\n", "\n", "# Clocks (see portrait128_benchmark.py and brainchip_utils.hardware_utils.AKIDA_CLOCKS_HZ).\n", "MEASURED_CLOCK = AKIDA_CLOCKS_HZ['AKIDA2_FPGA'] # 25 MHz FPGA\n", @@ -66,7 +64,7 @@ "## Model\n", "\n", "The notebook loads a converted Akida model (`.fbz`) from `pretrained_models/`, for example\n", - "`pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.fbz`. Pretrained models are committed\n", + "`pretrained_models/akida_unet_portrait128_i8_w8_a8_qat.fbz`. Pretrained models are committed\n", "through `git-lfs`; see the [Trained models](../../../README.md#trained-models) section of the\n", "top-level README for set-up.\n", "\n", diff --git a/akida2/model_zoo/portrait128/portrait128_notebook_training.ipynb b/akida2/model_zoo/portrait128/portrait128_notebook_training.ipynb index 0dc80cc..604b4a8 100644 --- a/akida2/model_zoo/portrait128/portrait128_notebook_training.ipynb +++ b/akida2/model_zoo/portrait128/portrait128_notebook_training.ipynb @@ -17,10 +17,9 @@ "\n", "The pipeline follows the standard Akida workflow:\n", "1. Train a float model\n", - "2. 8-bit quantization (PTQ)\n", - "3. 4-bit quantization with QAT fine-tuning\n", - "4. Conversion to Akida `.fbz` format\n", - "5. Evaluation on Akida\n", + "2. 8-bit quantization (PTQ), then one epoch of QAT\n", + "3. Conversion to Akida `.fbz` format\n", + "4. Evaluation on Akida\n", "\n", "The focus is on the **Akida-specific** aspects of the pipeline. The data loading,\n", "model definition and training loop live in the accompanying Python files: they are\n", @@ -60,9 +59,9 @@ "`img_uint8.npy` and `msk_uint8.npy`. See the [README](README.md) for where to get\n", "them and for symlink setup. Update `DATA_PATH` below if you keep them elsewhere.\n", "\n", - "Setting `RUN_FLOAT_TRAINING` or `RUN_QAT_TRAINING` to `False` skips that training\n", - "run and loads the model saved in `MODELS_DIR` by an earlier run of this notebook or\n", - "of `portrait128_train.sh`." + "Setting `RUN_FLOAT_TRAINING` to `False` skips the training run and loads the float\n", + "model saved in `MODELS_DIR` by an earlier run of this notebook or of\n", + "`portrait128_train.sh`." ] }, { @@ -82,7 +81,6 @@ "os.makedirs(MODELS_DIR, exist_ok=True)\n", "\n", "RUN_FLOAT_TRAINING = True\n", - "RUN_QAT_TRAINING = True\n", "\n", "SEED = 42\n", "\n", @@ -261,17 +259,17 @@ "## Quantization (quantizeml)\n", "\n", "Akida runs on integers, so the weights and activations have to be quantized. Akida 2\n", - "quantizes with **quantizeml** (not `cnn2snn.quantize`). We produce two variants:\n", - "\n", - "* **8-bit** (i8/w8/a8): post-training quantization (PTQ) only. 256 levels per weight\n", - " and activation track the float values closely enough that no retraining is needed.\n", - "* **4-bit** (i8/w4/a4): quantization-aware training (QAT). With only 16 levels the\n", - " rounding error is large enough to cost real accuracy after PTQ, so the model is\n", - " fine-tuned with the quantization in place.\n", - "\n", - "In both variants the **input layer's weights stay 8-bit** (`input_weight_bits=8`). The\n", - "input layer runs on Akida's dedicated input processor and sees the raw image, where\n", - "precision matters most; it is also a small share of the total weights.\n", + "quantizes with **quantizeml** (not `cnn2snn.quantize`). We quantize to **8 bits**\n", + "(i8/w8/a8) by post-training quantization (PTQ) only: 256 levels per weight and\n", + "activation track the float values closely enough that no retraining is needed.\n", + "`input_weight_bits` sets the input layer separately, because it runs on Akida's\n", + "dedicated input processor; here it is 8-bit like everything else.\n", + "\n", + "**Why no 4-bit variant?** Other Akida 2 examples also offer a 4-bit model, fine-tuned\n", + "with quantization-aware training (QAT). This one can't. The decoder upsamples with\n", + "transposed depthwise convolutions, and on Akida 2 those layers need 8-bit inputs and\n", + "8-bit weights. In a model quantized to 4 bits throughout, all five of them would fall\n", + "back to software, so the model would no longer run fully on the hardware.\n", "\n", "Quantization with `quantizeml` requires **calibration samples**. Calibration runs real\n", "inputs through the model to measure the range of every layer's activations, and those\n", @@ -294,17 +292,14 @@ "NUM_SAMPLES = 1024\n", "samples = get_samples(DATA_PATH, INPUT_SHAPE, num_samples=NUM_SAMPLES)\n", "\n", - "# --- 8-bit variant (i8 / w8 / a8), PTQ only ---\n", + "# --- 8-bit (i8 / w8 / a8), PTQ ---\n", "qparams_8bit = QuantizationParams(input_weight_bits=8, weight_bits=8, activation_bits=8)\n", "model_8bit = quantize(model, qparams=qparams_8bit, samples=samples)\n", "\n", - "q8_path = os.path.join(MODELS_DIR, 'akida_unet_portrait128_i8_w8_a8.h5')\n", - "model_8bit.save(q8_path, include_optimizer=False)\n", - "\n", "model_8bit.compile(metrics=get_metrics())\n", - "q8_results = model_8bit.evaluate(val_ds, verbose=0, return_dict=True)\n", - "iou_8bit, acc_8bit = q8_results['binary_iou'], q8_results['accuracy']\n", - "print(f'8-bit binary IoU: {iou_8bit:.4f}, pixel accuracy: {acc_8bit:.4f}')" + "ptq_results = model_8bit.evaluate(val_ds, verbose=0, return_dict=True)\n", + "iou_ptq, acc_ptq = ptq_results['binary_iou'], ptq_results['accuracy']\n", + "print(f'8-bit PTQ binary IoU: {iou_ptq:.4f}, pixel accuracy: {acc_ptq:.4f}')" ] }, { @@ -312,10 +307,20 @@ "id": "cell-14", "metadata": {}, "source": [ - "Note: the quantized model can be saved using the standard method. However,\n", - "for later reloading, because of the custom quantized layers in the model\n", - "we have to use the `load_model` function from `quantizeml.model_io` (a wrapper\n", - "around the standard tf_keras loading function)." + "### One epoch of QAT\n", + "\n", + "The 8-bit model then gets one epoch of quantization-aware training (QAT), as the\n", + "model in BrainChip's model zoo does. During QAT the forward pass uses the quantized\n", + "weights and activations, while the gradients update an underlying float copy of the\n", + "weights, so the network adjusts to the rounding that PTQ introduced.\n", + "\n", + "QAT starts from a model that is already close to a good solution, so it uses a lower\n", + "learning rate: a tenth of the float rate here. A higher rate would undo the float\n", + "training rather than refine it.\n", + "\n", + "A model quantized with `quantizeml` is a standard Keras model, so it goes straight\n", + "back into the same training function used for float training. The PTQ model itself\n", + "is not kept; compare its result above with the one after QAT." ] }, { @@ -325,11 +330,18 @@ "metadata": {}, "outputs": [], "source": [ - "# Just to illustrate how to load a quantized model\n", - "del model_8bit\n", + "QAT_EPOCHS = 1\n", + "QAT_LR = 3e-6\n", + "train_ds, val_ds = get_data(DATA_PATH, INPUT_SHAPE, BATCH_SIZE, seed=SEED)\n", + "train_portrait128(model_8bit, train_ds, val_ds, QAT_EPOCHS, QAT_LR, seed=SEED)\n", "\n", - "from quantizeml.model_io import load_model\n", - "model_8bit = load_model(q8_path)" + "q8_path = os.path.join(MODELS_DIR, 'akida_unet_portrait128_i8_w8_a8_qat.h5')\n", + "model_8bit.save(q8_path, include_optimizer=False)\n", + "\n", + "model_8bit.compile(metrics=get_metrics())\n", + "q8_results = model_8bit.evaluate(val_ds, verbose=0, return_dict=True)\n", + "iou_8bit, acc_8bit = q8_results['binary_iou'], q8_results['accuracy']\n", + "print(f'8-bit QAT binary IoU: {iou_8bit:.4f}, pixel accuracy: {acc_8bit:.4f}')" ] }, { @@ -337,20 +349,10 @@ "id": "cell-16", "metadata": {}, "source": [ - "### 4-bit variant with QAT\n", - "\n", - "At 4 bits, PTQ alone loses noticeable accuracy, so the quantized model is fine-tuned.\n", - "During QAT the forward pass uses the quantized weights and activations, while the\n", - "gradients update an underlying float copy of the weights. The network therefore\n", - "learns weights that still work after rounding to 16 levels.\n", - "\n", - "QAT starts from a model that is already close to a good solution, so it uses a lower\n", - "learning rate: a tenth of the float rate here. A higher rate would undo the float\n", - "training rather than refine it.\n", - "\n", - "A model quantized with `quantizeml` is a standard Keras model, so it goes straight\n", - "back into the same training function used for float training. The intermediate PTQ\n", - "model is not kept." + "Note: the quantized model can be saved using the standard method. However,\n", + "for later reloading, because of the custom quantized layers in the model\n", + "we have to use the `load_model` function from `quantizeml.model_io` (a wrapper\n", + "around the standard tf_keras loading function)." ] }, { @@ -360,30 +362,11 @@ "metadata": {}, "outputs": [], "source": [ - "if RUN_QAT_TRAINING:\n", - " # --- 4-bit variant (i8 / w4 / a4), QAT ---\n", - " qparams_4bit = QuantizationParams(input_weight_bits=8, weight_bits=4, activation_bits=4)\n", - " model_4bit = quantize(model, qparams=qparams_4bit, samples=samples)\n", - "\n", - " # QAT fine-tune the quantized 4-bit model\n", - " QAT_EPOCHS = 10\n", - " QAT_LR = 3e-6\n", - " train_ds, val_ds = get_data(DATA_PATH, INPUT_SHAPE, BATCH_SIZE, seed=SEED)\n", - " train_portrait128(model_4bit, train_ds, val_ds, QAT_EPOCHS, QAT_LR, seed=SEED)\n", + "# Just to illustrate how to load a quantized model\n", + "del model_8bit\n", "\n", - " q4_path = os.path.join(MODELS_DIR, 'akida_unet_portrait128_i8_w4_a4_qat.h5')\n", - " model_4bit.save(q4_path, include_optimizer=False)\n", - "else:\n", - " # Note the use of the quantizeml load_model wrapper to handle the\n", - " # custom quantized layer types\n", - " from quantizeml.model_io import load_model\n", - " print('Training skipped. Loading an existing 4-bit model...')\n", - " model_4bit = load_model(os.path.join(MODELS_DIR, 'akida_unet_portrait128_i8_w4_a4_qat.h5'))\n", - "\n", - "model_4bit.compile(metrics=get_metrics())\n", - "q4_results = model_4bit.evaluate(val_ds, verbose=0, return_dict=True)\n", - "iou_4bit, acc_4bit = q4_results['binary_iou'], q4_results['accuracy']\n", - "print(f'4-bit QAT binary IoU: {iou_4bit:.4f}, pixel accuracy: {acc_4bit:.4f}')" + "from quantizeml.model_io import load_model\n", + "model_8bit = load_model(q8_path)" ] }, { @@ -396,7 +379,7 @@ "`cnn2snn.convert` compiles the quantized Keras model into an Akida `.fbz`\n", "model that can be loaded and executed directly on an Akida 2 hardware device.\n", "The converter verifies hardware compatibility and maps each layer to its\n", - "corresponding Akida primitive. We convert both quantized variants.\n", + "corresponding Akida primitive.\n", "\n", "In the summary below, note the following:\n", "\n", @@ -404,7 +387,8 @@ " convolution.\n", "* Each decoder block becomes a `DepthwiseConv2DTranspose` followed by a pointwise\n", " `Conv2D`. On Akida 2 the depthwise and pointwise parts of a separable convolution\n", - " are two separate layers.\n", + " are two separate layers. The `DepthwiseConv2DTranspose` layers are the ones that\n", + " need 8 bits to run on the hardware.\n", "* The last layer is a `Dequantizer`, so the model outputs float logits of shape\n", " `(128, 128, 1)`, which can be thresholded at 0 just like the Keras model's." ] @@ -419,10 +403,7 @@ "from cnn2snn import convert\n", "\n", "akida_8bit = convert(model_8bit)\n", - "akida_8bit.save(os.path.join(MODELS_DIR, 'akida_unet_portrait128_i8_w8_a8.fbz'))\n", - "\n", - "akida_4bit = convert(model_4bit)\n", - "akida_4bit.save(os.path.join(MODELS_DIR, 'akida_unet_portrait128_i8_w4_a4_qat.fbz'))\n", + "akida_8bit.save(os.path.join(MODELS_DIR, 'akida_unet_portrait128_i8_w8_a8_qat.fbz'))\n", "\n", "akida_8bit.summary()" ] @@ -473,9 +454,7 @@ " return float(iou_metric.result()), float(acc_metric.result())\n", "\n", "akida_iou_8bit, akida_acc_8bit = evaluate_akida(akida_8bit, val_ds)\n", - "akida_iou_4bit, akida_acc_4bit = evaluate_akida(akida_4bit, val_ds)\n", - "print(f'Akida 8-bit binary IoU: {akida_iou_8bit:.4f}, pixel accuracy: {akida_acc_8bit:.4f}')\n", - "print(f'Akida 4-bit QAT binary IoU: {akida_iou_4bit:.4f}, pixel accuracy: {akida_acc_4bit:.4f}')" + "print(f'Akida 8-bit QAT binary IoU: {akida_iou_8bit:.4f}, pixel accuracy: {akida_acc_8bit:.4f}')" ] }, { @@ -510,10 +489,7 @@ "from brainchip_utils.plot_utils import pretty_print_sparsity\n", "\n", "# Reuse the calibration samples built earlier (sparsity needs real activations too).\n", - "print('8-bit sparsity:')\n", - "pretty_print_sparsity(compute_sparsity(akida_8bit, samples=samples))\n", - "print('\\n4-bit QAT sparsity:')\n", - "pretty_print_sparsity(compute_sparsity(akida_4bit, samples=samples))" + "pretty_print_sparsity(compute_sparsity(akida_8bit, samples=samples))" ] }, { @@ -533,8 +509,8 @@ "source": [ "print(f'{\"Variant\":<16}{\"Keras IoU\":<12}{\"Keras acc\":<12}{\"Akida IoU\":<12}{\"Akida acc\":<12}')\n", "print(f'{\"float\":<16}{float_iou:<12.4f}{float_accuracy:<12.4f}{\"-\":<12}{\"-\":<12}')\n", - "print(f'{\"8-bit (w8a8)\":<16}{iou_8bit:<12.4f}{acc_8bit:<12.4f}{akida_iou_8bit:<12.4f}{akida_acc_8bit:<12.4f}')\n", - "print(f'{\"4-bit QAT\":<16}{iou_4bit:<12.4f}{acc_4bit:<12.4f}{akida_iou_4bit:<12.4f}{akida_acc_4bit:<12.4f}')" + "print(f'{\"8-bit PTQ\":<16}{iou_ptq:<12.4f}{acc_ptq:<12.4f}{\"-\":<12}{\"-\":<12}')\n", + "print(f'{\"8-bit QAT\":<16}{iou_8bit:<12.4f}{acc_8bit:<12.4f}{akida_iou_8bit:<12.4f}{akida_acc_8bit:<12.4f}')" ] } ], diff --git a/akida2/model_zoo/portrait128/portrait128_train.py b/akida2/model_zoo/portrait128/portrait128_train.py index 6e6f7ef..c74137a 100644 --- a/akida2/model_zoo/portrait128/portrait128_train.py +++ b/akida2/model_zoo/portrait128/portrait128_train.py @@ -1,7 +1,8 @@ #!/usr/bin/env python # Copyright 2026 Brainchip Holdings Ltd. Apache 2.0 License """ -Portrait128 training, for float training and quantization-aware training (QAT). +Portrait128 training: float training, and the quantization-aware training (QAT) +epoch of the 8-bit model. The model outputs one logit per pixel, so the loss is binary cross-entropy on logits. Training reports the binary IoU (mean of the person and background diff --git a/akida2/model_zoo/portrait128/portrait128_train.sh b/akida2/model_zoo/portrait128/portrait128_train.sh index 0435054..d6fc6c9 100644 --- a/akida2/model_zoo/portrait128/portrait128_train.sh +++ b/akida2/model_zoo/portrait128/portrait128_train.sh @@ -6,11 +6,12 @@ # The model's encoder starts from ImageNet-trained AkidaNet 0.5 weights; the # decoder is trained from random initialization. # -# Quantization uses `quantizeml`. Two quantized variants are produced: -# * 8-bit (i8/w8/a8) via PTQ -- no QAT needed (8-bit PTQ is accurate enough). -# * 4-bit (i8/w4/a4) via QAT only -- 4-bit PTQ accuracy is poor, so the PTQ -# model is a throwaway on the way to QAT (not stored/evaluated). -# The input layer weights are always 8-bit (-i 8). +# Quantization uses `quantizeml`: 8-bit (i8/w8/a8) post-training quantization, +# then one epoch of quantization-aware training (QAT) at a tenth of the float +# learning rate, as for the model in BrainChip's model zoo. The PTQ model is a +# throwaway on the way to QAT (_pretmp). There is no 4-bit variant: on Akida 2 +# the decoder's transposed depthwise convolutions need 8-bit inputs and weights, +# so a 4-bit model can't run fully on hardware. # Conversion to .fbz is done with `cnn2snn convert` (accepts quantizeml models). # # Usage: @@ -34,30 +35,17 @@ python portrait128_model.py -bw imagenet -s models/akida_unet_portrait128.h5 python portrait128_train.py -l models/akida_unet_portrait128.h5 -s models/akida_unet_portrait128.h5 -e 50 -lr 3e-5 $DATA_ARG python portrait128_eval.py -l models/akida_unet_portrait128.h5 $DATA_ARG -# 8-BIT VARIANT (i8/w8/a8) -- PTQ only, no QAT +# 8-bit quantization (i8/w8/a8) -- PTQ, then one epoch of QAT. PTQ is a throwaway (_pretmp). quantizeml quantize -m models/akida_unet_portrait128.h5 -i 8 -w 8 -a 8 \ - -s models/akida_unet_portrait128_i8_w8_a8.h5 \ + -s models/akida_unet_portrait128_i8_w8_a8_pretmp.h5 \ --samples data/portrait128_batch1024.npz -python portrait128_eval.py -l models/akida_unet_portrait128_i8_w8_a8.h5 $DATA_ARG +python portrait128_train.py -l models/akida_unet_portrait128_i8_w8_a8_pretmp.h5 -s models/akida_unet_portrait128_i8_w8_a8_qat.h5 -e 1 -lr 3e-6 $DATA_ARG +python portrait128_eval.py -l models/akida_unet_portrait128_i8_w8_a8_qat.h5 $DATA_ARG -cnn2snn convert -m models/akida_unet_portrait128_i8_w8_a8.h5 -python portrait128_eval.py -l models/akida_unet_portrait128_i8_w8_a8.fbz $DATA_ARG +cnn2snn convert -m models/akida_unet_portrait128_i8_w8_a8_qat.h5 +python portrait128_eval.py -l models/akida_unet_portrait128_i8_w8_a8_qat.fbz $DATA_ARG -python portrait128_benchmark.py -l models/akida_unet_portrait128_i8_w8_a8.fbz $DATA_ARG || true +python portrait128_benchmark.py -l models/akida_unet_portrait128_i8_w8_a8_qat.fbz $DATA_ARG || true -# 4-BIT VARIANT (i8/w4/a4) -- QAT only. 4-bit PTQ is a throwaway (_pretmp). -quantizeml quantize -m models/akida_unet_portrait128.h5 -i 8 -w 4 -a 4 \ - -s models/akida_unet_portrait128_i8_w4_a4_pretmp.h5 \ - --samples data/portrait128_batch1024.npz - -# QAT at a tenth of the float learning rate -python portrait128_train.py -l models/akida_unet_portrait128_i8_w4_a4_pretmp.h5 -s models/akida_unet_portrait128_i8_w4_a4_qat.h5 -e 10 -lr 3e-6 $DATA_ARG -python portrait128_eval.py -l models/akida_unet_portrait128_i8_w4_a4_qat.h5 $DATA_ARG - -cnn2snn convert -m models/akida_unet_portrait128_i8_w4_a4_qat.h5 -python portrait128_eval.py -l models/akida_unet_portrait128_i8_w4_a4_qat.fbz $DATA_ARG - -python portrait128_benchmark.py -l models/akida_unet_portrait128_i8_w4_a4_qat.fbz $DATA_ARG || true - -rm -f models/akida_unet_portrait128_i8_w4_a4_pretmp.h5 +rm -f models/akida_unet_portrait128_i8_w8_a8_pretmp.h5 diff --git a/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.fbz b/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.fbz deleted file mode 100644 index f2ac49b..0000000 --- a/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.fbz +++ /dev/null @@ -1,3 +0,0 @@ -version https://git-lfs.github.com/spec/v1 -oid sha256:4416cdcbb322100b43d624317f7938e44111ed215390ffff5d8b8290cc4af5b7 -size 1302167 diff --git a/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.h5 b/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.h5 deleted file mode 100644 index c488331..0000000 --- a/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w4_a4_qat.h5 +++ /dev/null @@ -1,3 +0,0 @@ -version https://git-lfs.github.com/spec/v1 -oid sha256:2dd085aae619fcd16576146f7e7f31a3ac830a01c6a963cbc20d10711ee1e1cc -size 4525112 diff --git a/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w8_a8.fbz b/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w8_a8.fbz deleted file mode 100644 index 8f1fc2b..0000000 --- a/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w8_a8.fbz +++ /dev/null @@ -1,3 +0,0 @@ -version https://git-lfs.github.com/spec/v1 -oid sha256:821fb2b8525857ae0d0e4b8f15de6bac8e7388c88efbdfa91ef05adf149a4dda -size 1302167 diff --git a/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w8_a8.h5 b/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w8_a8.h5 deleted file mode 100644 index 25aa18a..0000000 --- a/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w8_a8.h5 +++ /dev/null @@ -1,3 +0,0 @@ -version https://git-lfs.github.com/spec/v1 -oid sha256:81e3a6ad391c9fb2e85742f491020656797c041dc952562942485571646fb04d -size 4525096 diff --git a/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w8_a8_qat.fbz b/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w8_a8_qat.fbz new file mode 100644 index 0000000..0b361df --- /dev/null +++ b/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w8_a8_qat.fbz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:868ade91f8e1d5c620ff116669fdf3a6a1a6b0626492e1e5eea505489054f5ef +size 1302167 diff --git a/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w8_a8_qat.h5 b/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w8_a8_qat.h5 new file mode 100644 index 0000000..514576b --- /dev/null +++ b/akida2/model_zoo/portrait128/pretrained_models/akida_unet_portrait128_i8_w8_a8_qat.h5 @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f0f5624b97fac7014ae1c76845edab4dac2897f3ef781653e791ea52f82ec015 +size 4525096 diff --git a/akida2/model_zoo/portrait128/wget-log b/akida2/model_zoo/portrait128/wget-log deleted file mode 100644 index 3b74abf..0000000 --- a/akida2/model_zoo/portrait128/wget-log +++ /dev/null @@ -1,6 +0,0 @@ ---2026-10-21 13:42:16-- https://data.brainchip.com/dataset-mirror/samples/portrait128/portrait128_batch1024.npz -Resolving data.brainchip.com (data.brainchip.com)... 146.59.209.152, 2001:41d0:301::31 -Connecting to data.brainchip.com (data.brainchip.com)|146.59.209.152|:443... connected. -HTTP request sent, awaiting response... 304 Not Modified -File ‘data/portrait128_batch1024.npz’ not modified on server. Omitting download. - From db25dbf075d21a83114bf63acd9d976a16d2ff6d Mon Sep 17 00:00:00 2001 From: mtorresBrainChip Date: Thu, 22 Oct 2026 18:19:53 -0700 Subject: [PATCH 5/5] feat(akida2/portrait128): FPGA benchmark results, plots and notebook outputs Refs HUB-43 --- .../skills/review-content/check_content.py | 6 +- .gitignore | 4 + akida2/README.md | 6 +- akida2/model_zoo/portrait128/README.md | 51 ++-- .../portrait128/docs/README.md.template | 19 +- .../model_zoo/portrait128/docs/metrics.json | 32 +- ...ll_akida_unet_portrait128_i8_w8_a8_qat.png | Bin 0 -> 281667 bytes ...rs_akida_unet_portrait128_i8_w8_a8_qat.png | Bin 0 -> 197533 bytes .../portrait128/portrait128_notebook.py | 10 +- .../portrait128_notebook_benchmark.ipynb | 279 +++++++++++++++++- .../portrait128_notebook_training.ipynb | 10 +- 11 files changed, 354 insertions(+), 63 deletions(-) create mode 100644 akida2/model_zoo/portrait128/docs/ref_benchmark_results_full_akida_unet_portrait128_i8_w8_a8_qat.png create mode 100644 akida2/model_zoo/portrait128/docs/ref_benchmark_results_layers_akida_unet_portrait128_i8_w8_a8_qat.png diff --git a/.claude/skills/review-content/check_content.py b/.claude/skills/review-content/check_content.py index d536f56..c0b76c9 100644 --- a/.claude/skills/review-content/check_content.py +++ b/.claude/skills/review-content/check_content.py @@ -118,7 +118,11 @@ def check_text(files, added): where = rel if path.suffix == ".ipynb" else f"{rel}:{lineno}" yield "machine-path", where, PATH_RE.search(line).group(0) is_new_zoo_script = rel in added and "model_zoo" in rel and path.suffix == ".py" - if is_new_zoo_script and not any(COPYRIGHT in l for l in lines[:3]): + # A Jupytext notebook mirror (`*_notebook.py`) must start with its YAML header, + # so it can't carry the copyright line; its .ipynb is the notebook itself. + is_jupytext_mirror = bool(lines) and lines[0].strip() == "# ---" and any( + "jupytext:" in l for l in lines[:6]) + if is_new_zoo_script and not is_jupytext_mirror and not any(COPYRIGHT in l for l in lines[:3]): yield "copyright", rel, "missing copyright header in first 3 lines" diff --git a/.gitignore b/.gitignore index ffa49f6..6c3e2db 100644 --- a/.gitignore +++ b/.gitignore @@ -220,3 +220,7 @@ __marimo__/ # Claude Code: personal, never shared CLAUDE.local.md .claude/settings.local.json + +# Logs written by wget / nohup when run without a terminal +wget-log* +nohup.out diff --git a/akida2/README.md b/akida2/README.md index af8ebf4..d8fe5f8 100644 --- a/akida2/README.md +++ b/akida2/README.md @@ -168,9 +168,9 @@ benchmarks for every mapping, and the steps to reproduce them. Portrait128 8-bit 0.9185 binary IoU - — - — - — + HwPr + 254.603 + 6.365 diff --git a/akida2/model_zoo/portrait128/README.md b/akida2/model_zoo/portrait128/README.md index 326b4f7..7c5a9e0 100644 --- a/akida2/model_zoo/portrait128/README.md +++ b/akida2/model_zoo/portrait128/README.md @@ -49,7 +49,7 @@ publishes this model at 8 bits only. 95.96% 0.9185 95.96% - TBD + 40.95% @@ -83,36 +83,49 @@ rescale of the measured cycles. 8-bit Minimal - TBD - TBD - TBD - TBD - TBD + 67 + 3 + 9900830 + 396.033 + 9.901 AllNPs - TBD - TBD - TBD - TBD - TBD + 69 + 3 + 9891115 + 395.645 + 9.891 HwPr - TBD - TBD - TBD - TBD - TBD + 187 + 8 + 6365085 + 254.603 + 6.365 ### Benchmark plots -📝 Planned: the full-model and per-layer benchmark plots are added -when the FPGA benchmark is run with `--save-metrics` (see -[Contributing and Maintenance](#contributing-and-maintenance)). +Two plots are shown below; click to open them. The **full-model** plot shows the +hardware mapping (NPs used per layer, with pass boundaries) under each mapping mode +(Minimal, AllNPs and HwPr): Minimal mapping schedules the model onto the fewest NPs +required, while AllNPs and HwPr spread it across more NPs. The **per-layer** plot shows +per-layer latency, input sparsity and NP mapping under Minimal mapping. Akida is +event-driven, so a layer's cost depends on how many non-zero activations it receives; +sparsity is measured on real dataset samples. + +
+8-bit (w8 / a8, QAT) + +8-bit full-model benchmark results + +8-bit per-layer benchmark results + +
The model is an **Akida U-Net** from `akida_models`: an **AkidaNet** encoder with width multiplier **alpha = 0.5**, initialized from ImageNet-trained weights, takes the diff --git a/akida2/model_zoo/portrait128/docs/README.md.template b/akida2/model_zoo/portrait128/docs/README.md.template index e20629d..b705adf 100644 --- a/akida2/model_zoo/portrait128/docs/README.md.template +++ b/akida2/model_zoo/portrait128/docs/README.md.template @@ -110,9 +110,22 @@ rescale of the measured cycles. ### Benchmark plots -📝 Planned: the full-model and per-layer benchmark plots are added -when the FPGA benchmark is run with `--save-metrics` (see -[Contributing and Maintenance](#contributing-and-maintenance)). +Two plots are shown below; click to open them. The **full-model** plot shows the +hardware mapping (NPs used per layer, with pass boundaries) under each mapping mode +(Minimal, AllNPs and HwPr): Minimal mapping schedules the model onto the fewest NPs +required, while AllNPs and HwPr spread it across more NPs. The **per-layer** plot shows +per-layer latency, input sparsity and NP mapping under Minimal mapping. Akida is +event-driven, so a layer's cost depends on how many non-zero activations it receives; +sparsity is measured on real dataset samples. + +
+8-bit (w8 / a8, QAT) + +8-bit full-model benchmark results + +8-bit per-layer benchmark results + +
The model is an **Akida U-Net** from `akida_models`: an **AkidaNet** encoder with width multiplier **alpha = 0.5**, initialized from ImageNet-trained weights, takes the diff --git a/akida2/model_zoo/portrait128/docs/metrics.json b/akida2/model_zoo/portrait128/docs/metrics.json index 064b91e..14005cf 100644 --- a/akida2/model_zoo/portrait128/docs/metrics.json +++ b/akida2/model_zoo/portrait128/docs/metrics.json @@ -6,20 +6,20 @@ "w8a8_quant_acc": "95.96%", "w8a8_akida_iou": "0.9185", "w8a8_akida_acc": "95.96%", - "w8a8_allnps_cycles": "TBD", - "w8a8_allnps_latency_ms": "TBD", - "w8a8_allnps_nps": "TBD", - "w8a8_allnps_passes": "TBD", - "w8a8_allnps_projected_ms": "TBD", - "w8a8_hwpr_cycles": "TBD", - "w8a8_hwpr_latency_ms": "TBD", - "w8a8_hwpr_nps": "TBD", - "w8a8_hwpr_passes": "TBD", - "w8a8_hwpr_projected_ms": "TBD", - "w8a8_minimal_cycles": "TBD", - "w8a8_minimal_latency_ms": "TBD", - "w8a8_minimal_nps": "TBD", - "w8a8_minimal_passes": "TBD", - "w8a8_minimal_projected_ms": "TBD", - "w8a8_sparsity": "TBD" + "w8a8_allnps_cycles": "9891115", + "w8a8_allnps_latency_ms": "395.645", + "w8a8_allnps_nps": "69", + "w8a8_allnps_passes": "3", + "w8a8_allnps_projected_ms": "9.891", + "w8a8_hwpr_cycles": "6365085", + "w8a8_hwpr_latency_ms": "254.603", + "w8a8_hwpr_nps": "187", + "w8a8_hwpr_passes": "8", + "w8a8_hwpr_projected_ms": "6.365", + "w8a8_minimal_cycles": "9900830", + "w8a8_minimal_latency_ms": "396.033", + "w8a8_minimal_nps": "67", + "w8a8_minimal_passes": "3", + "w8a8_minimal_projected_ms": "9.901", + "w8a8_sparsity": "40.95%" } diff --git a/akida2/model_zoo/portrait128/docs/ref_benchmark_results_full_akida_unet_portrait128_i8_w8_a8_qat.png b/akida2/model_zoo/portrait128/docs/ref_benchmark_results_full_akida_unet_portrait128_i8_w8_a8_qat.png new file mode 100644 index 0000000000000000000000000000000000000000..c27b2c1a4d8e07c11caea17bd13304872afd070f GIT binary patch literal 281667 zcmeFZc{JAV+cm5-7#a{NLn&j3l4O>UkSS!ykdPrm=9v--Au?5l3`OS5vr>_{2#Ly& z6v~uje)gyOe(&FNzxT7&`_KE=`?OXo71wor&ht2rV;_6(;|x+$k>5(eKtV!6vQ^=f z>=_c0Eh;1=q#|US@Gq&p(USN>)JaaqNyF}?q35Z}abW@Yy+-^Ir-5ju#;zzkEv9k%VL?4e@)Oc~Zw2l652` z3bN9gZt)ZET{UU7pKY4D&b^;S?!gTThIO7Yq16q!j~UPLKCY^dw76Nn=P1Kbj(Woj zlaK2^$UXIAq1SvKcw_zg{b$mSeB7J(KHT`Cv8!(IujRJEFa70?qaS(?c$(Y2-myWN zg!+H}rB0n@^lCTD{{Q(`6VHUOwCw-^D8(0{30V68X8io_3>8tl`0=j!E#LY$D@I2(i=fR()WaHZr!?de@kES zrJZA4S+S!_-6PZOA?+Q_&6FR#=M%VEUcA^SAHJtcw6%^`L}YjC-J>$r*4YBqoez~Q zI?@!@edb=9@?uM=dNRonuvv=opKWj2>JtAa2vXWr(p3_smKSENhj%eDrkXZJav9Z# zkM-n*f2j6+_WJeVB!^$_imIye=g*%{oa8)ulx}r(b!u)dATN*K_G3+GS(&hUe1Tc> zP+b6}$iwKw#IWnz_YQUEoGbHQ67sutje5_XuXGP2{Qrfia4zeslM*rB7V`?RZzuJztAd3?C@7cV~SvS;Duo*Dn3 z9=EbMo;Ni+tHD=US()|T@h{t0B>CO~v%>d=u?uCYIJlooBsE%fRd3Pb8~XopNRPQc$+E;I_`lN zAv=5P-B*`IGPU!w=4h2CTLbpBZ%pQjt>lVT z?vWf>;`?*n)eQ&LPX0{`R;N@|8$P$*8^oG#-@aYv+_{}jF{T06uPccLRCzCzTyePV zT6EC-%$YM`p%HJQBO_B>OxOm7xjLm*e(YGke*ILt)Y{8>5uat@n$^WiY3b=%(VzQD z?G9gT+4cOzi^23HpEF|ovbuCJp7O?x4e6h-DLb6+{u&D5pPK72$}0YbEzxqhOLE1g zuVmz{y`KXoeL33WbKb-r_kzx2OX{(Yvq19@zJzhs!2pKo{`w1ZNVQ7$qdU?V;;o3C8!L*;vy zzr5O~P{u;GOUkF}UFdOvdm@e;_$=um6V%=glChrG<6pi!W&N?nfL7@8!0}?n2D`qJ zgMD-B)~zdZpMBW9(D`O`S=UOjt5DW>YT#+gNJIGE;a}6k?g=^j14db?4|ws-UsFv^ zO!;oE9yb%86zotQ;?Vmyv2d;B|C@WS;7xfL4aD81NN}X;@{OO}PfzCzs=U<0=kM=N zdz>#s$lKeyth`)d)v5L6OS0liJ@u|1y{|iucN^nKm8%qbEzA(%7{(;BUP?;J&CTun zU|m3dK|y3_t;;t(g}2CEQ$sH(70N0r#kMhv`ir@KADHMZ{PM0ii%rGBK|oww+~M4o z6xB06@5LxmmG22DT)g?^V{NTo-=nq%2xCCw9(T+4l z?+C~-tkz3kI5;LbKO1plb?^9rfb{>+>h?-_Y#_n5aFkg0`fc962PuhSZmjECc=&db0VJ!)Z>!6*stV6~&gSSH zV0>XwQA>YN?*_vgai8T~J$Xjg@qK=Nekh_>pFfw4i;KIDWP91!`9-XePtYC-k0<^8 z`r{6@J}V41?}}bm#FG#;0nweT^Y}tjerLLBR*Qy;RZniHk?#tfL4{Mq&8ZJ5a+E*D zx+uSY|9&GRWJgY=^LPwf&X+G13UODju4{`IA-#0zlKLI)v)NogyNxx za1`C9hc=?{oVTzblb4qlw0c7(;ql{2R#ujtrG)#;22_<7p^V3<7#M=zzSYsu)2r+4 z-Tm!du}q#}^?56+%(HsIk~6Z-Sk+42BbXWOP475`T4!(-@B&R=w;#S}Su#T{dXTefYhClc4!uR&PXx9{GiAld)=UOB0T zHx8Z6$TGjn;k3Z3{;IQ6=3|Yw+(%&*y_4-2WSl|n+gQRzClUr2}^?(+z6{${K4q6&G?%~sXrxKTUz|1d9*W^^)*Li z_Hl7>X^&`*ym_P%`HfpMFd|(1uNhBN&{TJy;G=LTbVbCwtVEdjPTR$y@V3r&5-^8 zySOPCC$pe0id##Oc=31hWYVv`9Z@!@@QE>XnRt7+QE0p;uiZJ(`Jhy-64#WMW>R{3 zx~Z>`*K~k&KPtJYXYJXS%}eICNu};{-Wtrk%&~~<3|@_Lhmj3-c9Yhf7bV#3CJ$z( zFH0HpXcV0J9Y@Om3JUK+8IDFizbEcq!1jT_F!pCOG&UYNazvepk&*GWxq(ho&i&@` z{QPKTssi7im-LM2{Oi<9|AI>HvF9>3|He+ljZEt88fVVPHqU>5FN1@0dFE3iQzz|) zby&$0$Rg8tyW-h`^H10F8dPWsRaf0_TfJan(xt*LDY<93Av{)jR>Rc7BBOf?DULw2 zzgC8tw2&b-Sk{=FoIFm_TQo81`aF>dJm+({5^vw$!En?nP2p#>kllxyx`h`N z+AKv96B8|Za+$-J#W-~5zr1FBlc`0C%q@%fD*5Vv{rd1dyPBs%FG95Hd#gQ<1#V*w z_V?e=JNNqa>o2cUH`{)wx?@sPc*O35`0~PNQo%UZK4;>eO zST??VOi>{z2{r?IiIV0?IJ=vBJv-d=Ti$6<}i&7T^oMQbHP5TFWm41yN- zyGwTR=|4d*$r|IF$YD{~S(`~`3NGT}Q2G zX?E-|>yej}OFma>+Z55(rA{?kn483Yr!ir>L*WM6qGieN;(ICc`aej`4!fVDis9Au zyK}@mlUL)^DYAFPm#*Uln{v%1*v;ie7|w8U&qeif@w(^gJl+pHDr;O*QxiG*0>_iB z4_H#S$ZNd&;PZxtkUG6$EB3`~N~Ll-CwFH0l7f*R6Fm=W7DYam{;jq%%IjR-{l&-J zRCim1AG^<9?4@2DzS1>dWoe#LD^oM#xxKS9UF7ZC za!Z;=gCAz9Uj(q@->_i=89>kpltZ?0PUX94`gy-7$;n@P_!3~1xtT-xE*T{~FBS3; zYrK@NArS)uqobTm;%;hR-dC2IdZLfm$IE-#A={BbLXe%E9a%=MFG9bo3pG<=aR;YL zROr&cQ_^0?MpWa=4h}Ret*z|iyu7^r$eWf4wb`lmcjMy7Lg)|D zF`F!XZBvu*(){Gk2h@A_@(YWID9oJAI(u+%;OVvBkA}}kwr$(ClsOL%kGjU&ob2u% zD~rYFjg9vy1G$;eCo{3fh!~R%%B4D0&c4|ybO;HimGRVZjf{TO- zd7U+0*#1PB{czwR)0b4e?FTqGI8?)xU3xB6d(8Q>%7>95mDAJn>acwI`c+}5HA%|o zP7Gh}0fT;0hkM(d<@dC;dlo-LR@$_cflXodYiGKZjTjJTV@d7G&wDe%cjDqs?bjdC7CJ^>^zn%uSjZ45-_6Oboy*I3tn`S= zr_Za8vPNwTy+3NoyEKe6^ViZ6$MpAW7np^lF_HLt%#A5r+O5I5k17K@fJo&jcQ!6XDL$}K9FSbn3XwwpS; z5?3}ZF#YZ27HR)Q#ojJ&z6wV~MC@Ch|LE%ncxe4>i;R3lwb!vgN#XNm6f4+tRt=(|>Jof_Wa zTVG#a_v9*Rz>_CW^hXX~BK^A8=a;mZK45n;(nsA@(#^7iv!m^?pYrs}9Zu-$$8&wa z%du}|$eTo5VhSEW*j7hn)w}Fd)lblBOphUe&&JV2`{fIt`EMbk(CLD2ZDLpr-6ahGSZ%oZuzrVkqZS-2^>{q~v z6u0bj)9^Xk*2IGg2C5OMjpph)GgY4RwvDTw+Y++$eF4OJ#lL?2N{$F_buC&{jrH0J&f(0)Zx3 zMMd>!N^Rd)WSlF_b#Mpri4?xgq3l;)Ue1;6^tEG`m`gTWZvRh*?Ck8z!}VKF-xVaE z8}E@vX*-0~$KjF7`F8X{+v;g$WyjO$8tE#L`HaVGzX;|9yq+FznEqD!QLzrq#CA?j z&gi}e>FLj2zT|2!xpd*edaNP#ev~^2#VuuoQ#FwVKupLsLDZS(ah%4xm1%2i`h*mS z=ziAHa+>yRrHf7D0u>e26=b}ix}TFD`4#Ny{K@l+ilWkbsy$tMt1>U)?|`-%eAd*^d=CoJM4T;S{x2NG`G!dkc7x5qtEk?H9it|^NE3`*@%%^8Y zlwKD+c)+Kc%?y{#dcDbM zX$O7Lx^kp8v2-Ir3q;0m&umN29%UUr(96U7QGKQ_{fz4CfVx|wKz2%Xx%%b%mwXXU zW(KlU8%8S7fw;N52X%01=Q0tQsC>kR-p_;3 zxCs=gSU1ApKJHhsjVX< z2QDlvJ!@}QEqQh;9t#o@!_3TVzq;&RR#A~zT+NM-QjqjOf7nXySB>}58}~!@xNK9x&3iwk%)(b&Z3rUK5(H3(ZvtBrvlaGk$9w zZK4Cuml;vsEoMvlNh^Q3Ai0hI66KE`fi!tr?#Kx`HE#f#?m?%h1<% zXX{>5Pd)jlUr10;vxNFoPp-jXt2bQ2qN4l3Nc@H+Ns*C$02wJ4TkmFQ=Oxy=CxYL? zJNZRL(VD#t*~!lqnxz!O8w8@}JWjYpf7xRJ{jQ_0YP>yrrw~jEb$g_IvX|Gu0-Ba; zI4eJ*(2mW?$zf{&7eI5-+?)I_I`OnK=>(U*zzijt(_VK3I zH4khQ^}HF6Ze8Id4?d|ZP^_FWN*g+}okBM6*eE^%LcC+`S&1HN>gVd6r1n0)`;CR%vm%Wiii6dG z6m{u3s*Ry_F{60t8=@-ci`O9ypCGc|>}O>&H{_-BpvEVJ?H6XuUf1e&wq)i1ZR;6CbL@8a5fOEzgFhQJ)&` z&Y4EF4`SfrK55d{zl%*mf|=)R_7gOyf^Dv-7ni?wurDvqwwHO#Wj47@eynXkkmuWd zl+4yIrvZUZdVi6DY|9o+vF&yrYqVC4w5}Oyh1NNWMLtGp*?Izbf`h#I-DL@aGMNC= zu2eT;C3`tQ_3NHnN%LdNi@b_aT>C1_W##2Z%R>7lPT=?gGDQ2EC)%7aM$h(ZI9%$@ zSXxBa)ZjCFz?)u`Q+L&z!`0^QJn{_Z;vIFgO=5TWY8E|Ld6NTda`0r$k_zp^@7A0S z6k|AdEsUApi^n?tbDticml)}+44i37*ZX&$22c#>0}hDhnb*|cJ$h*qAUoPYT9V6? zA0*IjzjTSebgnn>24*etc3EyQ&}lO_5LYUQtaQy}pw?x6pcET;@7}#Hc_^lO52BL@8p`&; z!-q|QXsAS!9B0(keLk$eo1iKrB-FH4n7IFJTyFC6k4HTkMOtrX`iT}Pon2K`75GxF zJlpLd4wYuFB`+&0DLU%N-Om8DOw9$jct_eozz0MrYk3!ENG%2Z*^*$2wtr;A4Zk@~ zN5@V#F>`Zsu!xE$Ki7byOpDmVvSw0^o?S79_tA6>We4~>wos?gn>Q~{56LOW_yHq~ zns?81pA5iVR zL=T3vIODMjiY;2VK>V(2Y*cXA`el547^fK|uiB;Bhggj@F=62+_4Ow^o_uOc5X;an zFEH&xTm7gw-0%LliAJ24)%=wdw>%`y0t>-8FeM^kwi}H{)w)#=u*!%d^*kE0dYi$ zmD+2EAR2ZDiDtjtPowsu3mFpSyDy^dn+|`dbmh-0WX91e5DTwfMQT{S6F-_FG38p_kD zI=av)WZYK|ly<%*@EHU@83AthZYg9{yz&`Az0AbKJvEDCJdLY3mK@|EAtBEQNN~`q zrgQFbqtabN4R`15FS9OH=D*K!PByIeh|Hf5rLlQmp~bl*VL#-rm-<5+PzdiDBUo0A zO?p>q7mh-g^D{3!{rYdVuGG4BKl#$s;5EJZ8-6|yzs~4J_5z+wO%2MZ>j3RVr|Rr1 zS{6rzDvovEL>hobT48g;ty@$~A`Wcrv#gwXiKPJ}OAq@Cx%cgp{^`}^B}rdm4k2S9 zi|dVsL+YW=<;cG&BXM!N^edbc%Co-+uk(seJ~7yG=ZMI|2WmP^2|It6s`MWd{Jiu3 z$dYwwmzPhR_=`cw32kZ1))%)$6GfdVpfGBLvFzTvHyqr0U0>fh+C1=z1TWh*Gc+*3 zLN4Y$la}`Mw5^Dkm<&!jEj=@HDAEQX2`S0cSQifgvpXy_c0!iZ)Qn1d+7D)*ecG_r z=U~t=+kS|?ATS8*5E;Qk$I8@`t$U;|&_h@L=|p$VJ+{4;mX^CDy@Xp@S_UR2wm3RE zI*q+iD2aFI%NqXFcst|F0}3?v0pKl8F`NKyHY$1Q{5g;lMLJTSlu%BIp3(9=Gt>v~ zrJuTg2B*2GA8@7jGXuXu`RFWh^k&d`0sd0ei)>c=bZ6}&|Y+N|qIakP5h{Un) z;6Y{D?gBHVgwImPW|jy9sshfy*wXR<#f}Z@Qcs4SKu)?6&I+{u{+vV2NP3mqhtECp zWAR99yJ}Y#vu3=4gM%X_lCpDh4qLopQP(l{`1u%|zoHNKP4}4biTlEw>+BnXTGrYX z!BmXQR^|V5hzB~|f9{MK-SzkUChz|rc>GV?t!OrtIySJ-@*C=3c5$IcGr~3wiP;~^ ztkB^x|2@H4Z;VJF6D9cII|$_nfSV#WH@6OwphI;nj%uT=Ihd@dgXhkc_b1D2R*1(| zj%4<~i_)OlKu~TjV$Q1-u9JL<%?%Cm<3%I{kr(OE$jsxs3Wb{w8#emuPNQWWm33znof$9AGi=9Sfh@2EB)SEN`-|p`26ZjMxm6Viv=TNF6y+U^%IVog|WVr$5 zG{i(0R9G^E}C-gVcDeoRqWnGQ|4tfdPgRp>j>lPn!Oe3%->)u|ZujCT}6n-eEgcM8k=QQc) z=i{50%;6N_y*Nd(qZ-|{2+-z-%^r%Y{pTDsRHV;K$Qj_12e7wOC@3hL0GDJoe`l_S zo{o+kBw%BG>sLu5B5@E@n2_`@!Niq?mH%SFF4vR)W$u{?S{Z3>BaW%M+l34Jp;AV% z&Tyk93L9!Vr;ousjlPXBnX9u zV|Vgy5+-zcFfQ!cwd+P5GHe(K0-{0}C==f! z73y}-&`{CR`WF=mE>At%{_Eq~Drb-hJSFG4{Zfz*b{3UCoEh44tEb1sxLkYOFF84x z%?QsC8?Qu?PtYPdz+LPO(qa0&INe))eSM2(-=XGMqf7`yeh1YLyT$gjw6u(bm^y$} zq8=%f4eg0Q;y@;93M1o9XGyYj9|0?rf}rC`1xdUyiqUJ%H~2aS=|~l z=Zu>jA76O!3P7mWS=T%+|M6q#u1qbn;S$^TqJ1m0F%F|IoON`XHhD3$ucmlN{oUdJ z^84)Xh*(FuiG_=c47@MnZ*W^>V{Lu&WGI7iVsP|zYDuzL=?>chvrnMd@ueY_nQiL#%Oc2x! zy>R>5>JmR8KJ>ORgWc@-MD>Z%V9&j(l&?plwdb!?)s#>+SL^M>gR|?YC@4Hlmbor# zjGY1Xe#<(`_S#zzRPaBk(*UoYEuJMOK_ouMmaz?D&j?8-A|~dcx(D*pGvBq9>suLx z?jsMTI0t(^$C`Di^4Ye}d0<*s_C-V3l|0 zLg&Wm8^llTI%cb1Gr}oGghSxuJ$icjGKUcgAG9N{SKb1I!5DD$`t{8~nlUyE0VBN) z9!=?4EMo`CwRX++KU8MyoyunBB)Jcgm*GM3zo3C}>Z|xt$k57PILCGD2gFSZ(SIy^ z3kMJlP^?l8A#2lN1qg71k4_y7&G4f^E;BPE#>6AKZPX z1O&&Sqer`XS2su>cXJbljAIO^(S;Ym2YOz-cyU1flqL$PFoazeie_cWHwQRLz)=$W zW)*#a2l?iA6^W0E@&{VBh^8me1nHim&JR#Y4Lky6hL-2-7BE6qp)^U1dnLUHml!l) zV_VxWl#NEAQ;iDXQ6N@chr;yi?b}^wXo%8%);$=>2uhOr!_&J7>TtItyh7PVwT85= z5XU1oJt^1Qx2xNc%%;=HBh9TWE=38HW~jAlecP;+Z# zMh*MCseVZ7WWW8D(BAUu>m$(gjwQ*d;G|yV(azr$uq1|C- zOn0WwuGaW-%bg}O6bx8(GMs(6x&9Oq7vz*{I4)0~K9!D20$}wI3=AH+d*_aVsL{yi zibm{5IYDnBD{c;sO=r)ZjhxDs8kUonXN9c*l6d4K$Jhml5!P;vPjsI>6m8ps{9Za< zoSvR`n;RpXYLYWh4hU7BrW6a!(4MdTYNV~%Tvs|)S-Pum?Z?;7FK-{Ga@oRFad~ak zJKB=Y^nLK32+PPYY51qM?B_;(F24pg`o3zCB=_)!&6`tkOm2jS?+i$i@T7H1`yQc)+IdZY(iQh&_ibLA6Q9GOohV6B_%)NhY zr%|pyC`C#CGH|l8x-XtTzmGI-@xJ1yO1v=Pw;8E`{f$Y~iR_?W(JiODV7v~CDZes* z0N9Q1Dt6GN+Q6$g~Yuda{AY67ct#X(W?aty!GHaJU0X3(1ue#`YO)O1nkXLxlB zWNC$bSG~|3)1MCfdq{u(K|%h^ukBweuliZ%u{TeFT%(!yL5L=J_2}ty#a01$gr~i| zBDDplG%em_sFM;5frCSVKudl7L*jYZ!7f34O5)0P7->RD;1W5cm}WKAX42OPiXe zIaH$s17Cg{o3tHBlrR%Q3=(|p9`Ijr@}KXFuQOg=_^0eaZ&F{q zd&GPj{>nvBd3kWXwT(^4^VJy*bw0*Ft?Ee5VUzkN&CN=*pGQYG!^(vp`LIjClnl1H zsAkY~gvMTP|MxhYI`*!_CItM-h{M9nh&_BoNTk%%>aQD!VxVaOqbC{6gE0oB-b*fT za?WXwlt72G_*f$$ARxdtPMD}ch>YfSrSbUd>jrsoi9Caqg1B z0+j6Hm&G14p1uo@xMD+pzcuv;4z{M<{S3<-iKTy z+H7ukU~I&9dMF@G<$!i6zr{zNrGt=!*~vlVOI>V>zcc)^JxOY3bfIhzjou4&XGvb8 z8YbxM(eF$3N28i!`OQ7Q|LNW%yu2w$;1D4|g$&W~mmn2Bfw{u0hgrm717Z4Vo4KPM z?6ClIsy`ZBp4i)wkraoH?QLY9o%k@gVdKVRD2geEkTBgmJU;90DNG%UC(Mf1?%dg} z7|k63!$c}raSJGglAmwgz!(ZqT-B550817I^^DQV6i9V&XP+zG2IGIhI3mg^vbO4H zd2B7qRNRi=Td024fq9K?bzyYx=>$=&krF(CcDc>z(>M7FVn-$CXLfEvw*?(6-THbM zyiS8KyrOlKuK6U_wQ$Jho#3y<#U@+U9+3djJ%1j1kl>2eVG(yS*SFmMP? zrG@y^<+<`$soIP&!F42dAF7@LG0uv}pMkhk_w?FkIKKZR{-jXC)2QzX_PUrc+&2@C zefvAKK^g3fl<@EIQBuBY?y}ysN^CKo+b}cRoT;gVSa!Oo~HfTGt%(AQhZmKXsFeFE?g4Xzke}JM@NSRu$5#An1^r(!>O#6J3l>r z6~v15ay+52(nBnT1f6r*y{ebOk_XW#*C(5 z51I#g4-Zku8i&BT!F>W4V6=~oZ3CE#Cx30w5M-lxb!lLRBHS7T-3Y|?IZnmPOB{5S zKa$Q64AA+-#V@ptyaV8xB^1L{@pZ>x;o8O|drfm2JuQqxddT6JRyc%UixXnhMoki6x`38VCD{bGQH{j$X~m>-#sYph9fix;*$ea z2jt`&6m|W66EgpAuXq`(VD<3kM2(iAS0)$)r-`@fBlfReIRAmNQ6B|*RtTH&l`ABg zkIu-I`=Fgg*xU+Dx_#pY8AP9rNF@#Eu%|z_GP1F=CnHgD8B_=oR&AJEQqit*cmB5& z%|SANyhNA}62;vKGYa9gWIcKFrBL-UZ*3j4nJGw~gODTePRBumf^3@t;YW0<#1n#v zbT~bV6+|W?C8TN)2g&#-#Q!bMPj1I6hR$pt0gXi1=Lk`E#5^^l*5zx*M#8Br+&b}c z<<`eiJELWIOqxaj`I&?-lYorq{902yJ)T#4;mMVCdd%kxl1d3Z6(*{Ik&ys`$>QVN zM#KtMapK!I^}>tFf|eZ{(4qGT=lwa`8`&M1J)zkWQsI~Rc>v5Cn@HcJsjOn&fmo~$ zA3hMo$%h*65NJe@Tc16Bx(@WCOp>HG=`#6 z)qc2s7lZ<|$!c6pdBo>}@n;aC5`mWymtcd{bM%hFMM00+s+na25Y0m5TW||pzMXq9 z#RA$oP2GnU7OSh^&+FRTx2vnGpHfnixg%gU@cnxj_%*8^wwzMcUJGZ=e1@UI1j~_< zk7i_O@Y(gI5q71RQ?PyHGS}7BMK$M~E3F6DA=st?OE+P$2eTermuFD93BHk$g3MVF zWKhDVPf#BaTlXWOVZQ>9*6ZE(?N9OIKq99*1LqPg$&FP!Jtr8+1F@jU55Bwj>b`W! zuN@qxZ>1eJZ)0k#AygYkb4o)^E9I4ylsEt}t+O|%S=pv#PMq+^)?}UbMN@IY%F60$ zj1f3yNGJsDgVBXJn+3v74(*5n#%*pXF~TETR3g?$`@Y6lg@At}~@ z+TOs*nrb~e0^SL$m(g-tbY?a#OH}(!P@(WB|%sR`wG2Dk!!6 zKmca<_?;g-ctDP>KjjCUE&({)N@JZkHBCYX%NFoEy_)!;&MUOCo}S%i?;Rvu;U(|U(ofzl7i@^0D;yr zy~}I*#m+jU*us?wFuTb!$5wDqf79D&Ddh%XRf<&RIdn+vjCst`cxQSz)s7vvZE76$ zyS8XmKcn6tjW+O>x9Tme?s-PSP?V#3>eQ_w@;ir&{kYoTwPst=t8jYMyXPvtXxTHjva)!9kJ9yP+@@0s8r)@N{Vj?ljvYHj z*qL&iOiWC`_a~p#2QVX!CI|;Do>d43gash_EB?w1#peu9P8GNBi4*Gyk_dtUG@^rj zWguDz7dsqWTcoMj#u4bRz>9+4d1<@$YuW@%I{_Bk4;vAj7csSH)x`}{(nZE&;N3xw zUb}u>KJ2U2{=BgA5+acxCGN-2pUVok=zO>73kl!sAWP~LSwu{T;^W=iBf)T~JNs$2 zZlP)WW|Fs6?nl5FbCb6>kB9vZUoL*oG@xfuvtBcR1Vx z#=i=fWCBZj!FZlgEs=*Z#UH{hLAbOnXce_i$jA&LKjENdAZB5*pu?*Ywa4-FF;2LL z zCes-q|Hqa<(E$NdJzMAS$PD02^cpNzu^D`rd*V{>pU7BJ!NDQLqiI!!UckEP(p4Y&Gxb3u815r& zVYDl1(}3j4#ydN~?yx}`d7AX{%A#|xGy0ofzsIS)9f*{wXIN}SM{<5;aV{lz6b@$o z#n(cd!0XTObvO#?Fmm4%KoZm?Y<4i1?x3L>uLT;|5jA+;i#UG14=^t5KEnoCl!zHZ z6QniA|3aaO*ZVpmx6lyExEay&TjI6{z29HqOo@c_d(dhHwu=0Vtqi?I1*?Y)@(d*i z6Ey~dUqJf+$PR!bc{DB$3v$Zvw4n7{DuRwDsuPYtGK{cI&d{vBbf#}dvWhi%clTwM zqY5Lco*a(M-xYGhYrE_+DtCCmW!o)k<^siY;PdA|gwKVgpWlQr zT?@z;fIi3=I|Wnkju*9fJ+Sw;O)78#pL9G>nr=PzaR7go!9ery^qu{Wp0(XQO4*($ z(R{cX2OJ2BofLCguhu~x2*ju#oDCcc>yXEL3GXEr;hjM&#R{r@dGlxoVvEww&JODS zX2i-f)VQuD?aP<*xzK0X3cF0~B77?VMo-}bD(8c%Pr~!3re{8h7AAo9kt6YKCn1ED zl*{m9AZYc!sh$HlD2V`r`FxfcymBglRiDZ*5KaAco2gW6Gh*ZTdt}D>3E^`){M<$uK5zoOQ&& z1z7BLYJDUCRSH_Qky;^Hg#n(M#nseEqK1YZF>f1yobqeBVXxeR$FHe<_{W^HEZ7?I z6gOn?NH>1Ns#k8manSQeMzuA!Q1NaSI*zLJwtoi$yyF1@r1b27NAKhw<`Gwv)V9%VuqDlF^@<@0#!aMdNbQj z+E@cbE zxjl(rVH{)%^^4aS!vsp?3`*ZtZj$>o322V(JqAk@s<7ugJ*8G%ic!8V)&Q+@Zw`O{ z%<--UmDOrQYJBMqz92`^yZzkp^5t7a@)2T5K<&Jiq)~V8a$9NRI*@>!=D&G?u>Ood z)Up#8%M+M@Diwg4hIf{LuKWAuVC#*1M0)MoHM8{Z%~r5z9YVVvu$4(zO;BS!31Mjx zn7hwp5W9x4x=3m&`7ovfy)}?Ykm0X!@$TQh9+o0CJ_tf*J3@SZO%1}LM-|{a@s=9K z6Oqq65)zn?cb_rcwR?9^UH|*{>KFr1e~SW8pnOil-M-E85;5?q&U-cdh=(aQB>DCV z_D4~0=0dm0RuY(sBO}-f!X|DdCi2}0V&l&f`@=|7UBn0s!@g;_dwzYsXUBGAht>XZ zYYN@ow{M@ke{Vn=GBQH_^>_O$Ps92N@kY4>-o!%Ygub#NQii>Bbs){#KqmfIVdYD z`xK{b5VHofVKt&l?E^Tc9*0?LB-u>^!$sk`W;U;q)qrB)y3K6q!Fgmi2Fp_i&5>&d`{LV10nOo0~8=ppfswq=d+qr+8$!X&{LZ%nK`| zU&;P7Hfq)sC5>D!Yq@E|hT~KlihU-!%odJ%OHKdysxeZ)LRRLxRx`8N(fOll+}*n= zV04MgQSR`U&+i~wR|!eo%~xLXy^n_sDMQl@r*z{Cd|YgOk34qX?!g>X2!I8bSUW*L zWBXrf-F+8diuU3exV@-x%-H5Tj3}xzz=9!*LC8aHQ|BA+Z~aSn`r+8{Fx~@-eFpNr zJQw)V;&ngv|xyrYBE4y<;c-Gt_fOXJ~9>)mh*vDcQ99!FI^f8l~3gxu(b zNKej)@@F?rbYB&?+FyP9r4>5RNNFKDc$e6&@n&|w4e{EOO)i}b_|;5o3&y3k>^ek( zgRyaHErEx2?czh*h9HoJgCKgid+x=>4FHl5N+E-) zgq$3ku%&{>k0bma9r08&G<5y5>B#Sz{>Ok(*Yl zJ?HC@2jp1KeeW&Yj5eUT({*Dez`$VTWX&@i6cVrOT7JEvQ{z91BuB2=P`6b%iCU9!6FcBBZ zXh%x5awsP_lX^W85W8`v&z6|fteIF4I5haU4uf|}W1DL^M%oV;-F?+{lM;6=#Ek~_ z^D?;W9@ZX}yWSs?7@r&ZFKILrqMx94?+$r8{NrGZE&46jzbsFTij3SZw#ahp*d%As z0bg}5spM~!LEzUFB|8!OZ*mP75wzN?sE_JrIKyw=d=5IVxXsbFHC{x+X&F}BYk=N)WI?L{wS$DSIHISPNr5pagD7OBN4jNpH(ZQ7i`o-j_ z-$n-KyCe=hhM2!meq*`*)r7ZyZj+t0HSfY!$x2-oL2u3o1HInr+qaryIc8F-Wo;W< zsc=iiqj>lABt%w)*~?5m5dDjOt&f3V^8u33i{StQ2bNb>1}7)OA>{LzYq&4|H~?cr zu1h-kz%a{pS=k?!!#APh{bU}&$3QYc6$0|Tx(L?*`>9R5FgKGjQ=gr(3bT0!PWN-X_o(9 zH&}Xh@^(ER=b2(mg<`s4H%t01h}2tJasv-?nT5%XiD4osBI+BJ&e^Hi9b5`I&I; z=g*%oh%v?R8Bh$EkVaqyFGZiVZR>+XVEnGgGET^b!LZ8hD!j#mcu|Z8`9lza1S8wG zSIS3!I8g3S2H9wXwvJ$iGDz#tenr<7XDR^#fE{^_Yq@a43HRnc^ppuryJ<9mKt)jY z*e4l2y`y~D)mMmE*UnH|2h~A~50?-O0$!rOq<#Ss=vI|H*V*h{Xb7X45kd_xiA6O7 z(qpRdS~YHFseJkJCDZ2imv{NkojXU&?B@trc965zzU}Qzw(81Exg#ziE)J$dYs3MS zmxKgH9krMhqL9qb?|i`CLoO3Naj^voRr~fofK;zreJiXwS<~Si4sC5jm{S%gkG9AK7}Go-U7L5hAK2#7Zf|!}}97guJHs zVYlTN)=IBcKZOT$h(?x4kf@XW2T>6b;-UylQ40uJzrBK(QDQj#$?31oN_P_27y?VW zgD&o8Ls#0x6^au}G~Tpl0fwyK=C8vJYq$9zVJi@Buwik)^L1GD+y2d6okzVnlS5`m zXC~P|8*-8px0K;Z9yw!zpu{Aqg$Pf1LDPdIEtGgdI>ozpd2mKJZ>opBpg06D$KD|R zMX!6?cFYpUo<6;Wouw8Scic;7F7Qr08F( zOOr=ch(TtWpV+ViJcfr4A0}jhPx9{0-*B%FSX101vx$VbU#)=k?bf}L`ol%e>I6ZX ziW>U+ejH^QC`NBj- zR{qi8lE7B+X7Uhj^I^e)?%TL1fVdlIWCk?#D*&}nc^6IMlv*6$08!&~+#zWi)qC?h zD!A5e?UP|+6|u$MfQvg@qk&#s8VqCvx|`tnXWON2#Gu=~4p4QYj<^a1WtnR{fzZux z#2s6D?n_@&gDFz|Ets^&9&=WWr-MPo;gB@|yB;R)gt$+QV;ecYL-%9x%T}Rjv^POv zcHOnv%vuB@^j%wJYL3w+CKO;{O-Ost>h6htXnE3i?HMEh)Mh?Qef#83d-kX80?8{^C`McuA#iyE$uedZQ$|o66oy(7B^YD*49@U*y18W> zeKid8epI117crHG5R9#2RKT$Z5teV=D*!4>Fk&Wv!sULG4@=~fH zO`v6Y($=;O(g=NytAj%w=yX1XaZH&;O%QB)(@ZRn_BJSwB4PB&7*j%qQaZ_{`KkHC zhdtn*0;8kp0GkJghWxN3gFvm)$XguOkF);oDN6uB+eKG==Z;!#;u}1KFdW-3G7@({ z5)>paimI|VwB5p)O4T3%OB+?lbJv~Qo4rDk}AWX6NDe1)-M;3%I=#f8!QP*uZxQS?iRfo@zSE z+gsyZd2*?{7zT`FsNCg>GJUf`xaJK)E^$*Ae9DBTjOz2WwRJ1BF=Eyi?IEXPDymy~ z6VG3*>v2iR+3xKI`uatRtm7nPNHQ5(S@dXti7OScwP@Xlv1NwzZ$m@tK?jsoRz8ED zm`JRcgfczkhh}@7ii!$DdS1v3af<>m=EF4sUXHk#s43*_clY;Oc9m|+!Dc5 z3Gw~{6wj1=G~O^HFo4vTK@<221NTwbbx>w)LPltnbjmI#b4_4zF{8nc8za;oUyy@_ zybjtL?D8@o#HKn`r3P{Co1_IO?}3ue^s0V|bx51TWj)H==m|gv=h! zXHapfj%Y#AH4|2CqAS9{aj~K=k~&;M`;g#&_nurq6nI8qhfics?u!(Shzm-ugTPP5 z9Tq85xE53#x5yFv4@^vUN3FO0F%x!Q2MMuNlFqq)ed)8^TLRzNQs(`f`QfVNcd?kZkiK zZdg-r1Fb+bGVn2&fV{J66?}!ce3*t$39z0-P@gz)0?u7WGn0K;z4Zk4o1)C1 zH)*skMc&>6JSb-jAXkv&=yUkkF@n-|j1J6gBQ5|#`P>8)>Nfj1)N}=SaT=uNDED8B zK~e!R`+$Rxg7Hc!CZ?NkxJB_9Fd$#`SgA?FzD3YQ$cC~|3DS;;lOUPMbx5<+w#FSM z#U-jyEmOFvkNezSW(f~LTp-7FuN(UzcZ7qh3UVBD0Y6-82p0+oaaGzZ!Gka3KAK3^ zy|5yYQBc$Y60;|$Q5?Yt08!Rl47#Q1w6+$!BusTtjKA;cO+vRM@q6rp{$S+t%L#~f zSU=*bIlXivKph~Jfzc%2KuFC3XpWf(d<+ZUz}Q%(^I}B@1_Z^ueo^@@PjAL`2>kB* z)T-R!NTkn01Wrj=;uW5A9XPH|aqSaQ$=zY9B$(Sr4V{sUEg)qIp?pl)@iNJ-XC_odkxR$eP8!=U-$K7)%?lm zd0TpBQ^|ly^SoNlSh4aym9m!-0ZBSW@MnABamR>7EGj#3w_2xh(cd5Zfwb^nShV z#(n00lB_4QQPRr-2Or#VZ2XupQ`yYeuI^!Gc!TE-9&J`(;Ss*$;QAgiQ%Uu86o%O^ z>0sYmwHOHs4q$Pbm_86Cle8WS-;JXyHP7S%N(;-P_iDqJ{XMzd@IDg}BFmwfbL8Ct z;7KFa_D@{WM5vP5y$4l#9DFu{z|OF}Cv5i)o%2~;3nAC6W}QwHIQeYa&IQ}${V!?u zXRLYJ(je}#Pi?{kznUpV^NfsoqIlC@4Fh(r@tB;$QNt7S~v74~GGon!RA4wqfi-WL>>>?SQni{e%9U%XfW@cxm9IjGUa& z?b@`_wza)81oyDHwuh4rHW9tPTZf^j(}$>!$iGB{Q-0C9x4|dB?Zd@=lLUA0G0uvM zFO-FZ6eViT`?n~2cijJhI)yGmt4&wG{U|hialnk#X6o~rrSHmW8+K;)sUa@(z_P8EFIQq9t|`=N!xQ)rpw~V-c6h&jjYJTntGgTV z8SYh{7CROnL|p{xme2j3P1pQ|c?auWs)4#XIzq61{Pc-2o3@BtQ_qeZbLiNyGlv&! zq?R@RDGCKfGY%x1-pusxUor2=B1Mowr)sEt$Ksxx4?F#Q{bJCcRVvBND3EMC3)~D(4F7Sl-=| z5k0zh*HKrXxBJLTi_XRwQXFHWv!O#knM(B7)KrT`9w;^aQCGViL*Xv)#x#*mznXrn zE%B6;fF5juy1J{nAp|c-eQ}5pb-PGJqf{n2{oLi}Hw68_V>l4VFty$tL+r@>P+!dVgBh5++3TiFJG>V83_)MNtq!xn5uSrr19_c;lP$Bw=X?$z9wg2-@cPo znp|h-hY4E0#mnj3PiRQ8C)#Ew{+GkZfg!iD6BV9*qdq#!Zde87c`vUuIQHsKiFaBC ztNlNI`ON?2e>u$j`yl`KKlT4#Td_Ot?%SfNCu2K4KK)fxIw79$8HLh?E^rsUeKL75L|NYPZKl{r6DUZx&{xfp66~S3HfI_VJXoq%#cscVl0+~JY zme-er0w-NWzGH&?wuv*GEH_rmON z7a@JQLv$<3LmB=QXoNfz8IljLo_=*RT5+vZwp`K(srB&n@ev<9=1OPJV&)8f@LPEG=#U*#uwoWZXp&b4 z%w!c@vo=~R1ZGDXzwVJYZ}X%aVyy_Bh;@xizG<#1`|=taC^h#qqe!Rj#P=B0^!Ldt zA@R|Lwh_=oIS$L_g3}yX{FiCdzV9eP?Dq)xLp#0^k$w)Xan%(&PF?+5TxifH2xgIb zS0S!Z;Ayk+q9wh6O-#2=W|vA+21y1j8*da||9M8vz2oEKONi|_Ty;g*GL2lygzy2l zot*nOX*F%H7$KJbtbrE06)Y1$*xAL0AY&%bfM4!V@b>lqvAWu70N;MKrXGQ_N!lt! z{mS_#s?$fX zp#(BDvagKYS^oHH&|-Jr<-xQCMX1KY_-$YFrcK58^JY17*@kz^eEeo9pTg2c>cJyD z+P9Y{L^@x0e#R0JQlYWhF9I$~^-Vc*mx|vyRp0<*B!RpkXDQ#AGx^ojG{-6S-zCRj zN&^dkb@vb_UhW23STIl?7JFQv7(U`UHW?|*x^%$gZEjI?m`d)rj5f3R^6C-7r9loi zDbeT{wqg!7HE_fB;aXP<#9XGNynH&J=f2vt+6cQ0+jn|v0WN5m{u(kIMWSh5!;HvK zNy_w;*civwrL(n0?Wox%GTXo+bPG}vmN0Ls{}XM6a26b9;zNa_lr3tI^2mz_&Nn1o zcQ=0zoX#R@P>6c_4Va#+5(wJv$^?T8hQE*Kz4>B7CXbl@F4gPo9c3C!+`V`uX8*&rUJd`2MS|I>Gn6%7-S_AJ~Z=}@a3I@pKEX^)>eHLzvuW<|4ocpAJqKfWD{)y+=cE|}o}PvEh* zLKX2|HwC*0IlcT_r@$*3Se~SAb9HomU%XW<`!#^&K~3qQ-0eGel)_X(njEC`ei6z~ ziE%oeokjQdeACZTwJ$P~Mog(zBLx~V7hxD^%&{9Ut#h7NxJ}VlOKU}8I0GxIiP6#7 zw<4=_3oU|L;s^+%Wm4Ro=>@4~acOMzA`}!n+#v85F*hq=*?Y^`0OPkbfVR{kfSI^n zM3*i%G+fKZ*F^*(ZD>vf^8q+5icjfnR_y}9p0#&Yu=@Z3SZ@(IRL6v`0fCoOz)J;dvZ zMU}GYM@!UOGUpXiLzyCqP)JCr8liZ;QU7o99B(0S7YT7Xm*`aXdOY>zp2)}@_r{(j_IK#;;@6ffC7J_vceCkbt^$r< z-z*JJ^@o|6ec+u4Or1#|GTbdWjFIR^OzSbz*qZjq6cs&Vh_lf$nw96cefuYR2~cw{ zlkCknB7yH;RuK2CdFyVwP;N9*z%=h-*JiQNqNkM5u<4uE$t?v*gSbu;&z>Mnqz{b# zRug+LQJH18zQAlh^xu}qFGp+#(?Mbu8cI!=D4E+}@nsQU3DSGPSMuB+Ft*KA2X5qy znQP`b7dKC5G zva%X+*M#u~A5pR_J()?8r;#8=krM|opFLv+9ppAEcT^6NjxD*Hx7C-)Dilo9@N`)p zd*KL*lrI0Q`aAT}h9+^q%88jNY&nI3QO?P{Q*0ZJF0C)>+dsac?lDlq|4$KyvE}W3Xq>c673YcKI8O5TyT)(!gd3h%wHGz$Hb;Xp$r^66*fUY8G=~Fq2{s+E5x9@VFYtKDGgE;$f9>JD8>loHP->r!y#?#VtaPQ%(9SCL;tGjX;fF zUt`*N9bdn_G6FLuX|*74pJ4N)wM-OQDaiuQ2*5NSNlRO&1#DPlnN-qq$HP>T; zu-|TRJ)pJC1ivzY`4F{)m_o$OQ7YdU_rvBWgZm5tG*W=rqZsTtRUn=TTnJ&evwSEc zy{#a$Qft%F(q_lhzbAw}1fdH=g^&Bl_)J&3pPVG4% zv-;2x^H+m4T-ZP``g2v(i+R>!rIRqDo2z}>x#v`*OJ;Yv;dmnF9bn(4tVTrdJr$#v zL%ORzkJ9!7n?Vq-7yLqMeP4!J9q(8VM4U*|f*F z^}#QgMjf30mO4ovq^fiE0N3&ffVk|Qd3O$z*o!XHujlYL7(S=imGF}AnpT4Ped?iR zW9^RX*vt0FBM}gF&ZB@pZza1sVmLN79 zLbZ1}`Of)#d6=}!_Kx3@(U%FIM63QVeqaPi!4xK|@B<_@dIR)^==l-$0jQTlN@LEKeW`fV;}f><~@gNBBTy|Bwr${U^s25uP6N zFbsZtExnO9I^jz6-? zHtP{dS==9(;k1h1B>b&VrVGYSpOF$cO@FF>@J|CKIe;kj9v>{5I(dY5y36 z8~ZV0j8u4T^E$T7f8|CSbOy_H4HBZJ>b|;P3Z9{tlvya+3rOMgBApx@9wz($^(A<}9%g@V064{&j zY*A&=tOO@1`kjw1YV+~0lML!9c1ern&c*=-#YIJJ$yj(8f07#MYBkOc7CxPO-$PebZ2|6GwuC42E6=%1`0iDZ>FcDf8^xs{0-KbIV5lPyU&!oK&%Ai zctr6dmSD!@4?9|{dG2MVjMJ1QLVJS8B*_M&_dms4ie)M7Sr3PPvP{<58*gQlNQyye zI&yN2d#}OPFL{n)K1T5}u-5^V?Be2fNJD5%t)`RA;|lw^fe=H&WpCkCi&o=$=7^u5FJs6V#Ns)36~Ow0cSA!!(&+|0H%KLlPYR!9YA3Cc7fkO}-bvIZ zOcg4mdMA+|!*A{*ND4o6ZIn5`wLR3XVf72oc(~`RSTX<=bMwZmGot#Z8yB3>*Kqc6`yA;S7_Ch z6w&hp1RfZ6-L)nKfdcsBDUU)strx9mIYSK_3IEoISinRa>q@|#Pc}0d!|HFnwmXwP z`VTgDC_oz(lq>}D-M+ppd`pRn4X%Ps%9bBEVaX8+_zc zkA9r_4k-ZV^y&JC?)@0`USJPWNdLCotlZ{a?(7Ez%1Q)cx?EanzD8>9;gw2 z@k-R3^J|@;1mrY18K^3u(Rz_I0#|b3QU{yWKZggMOwaKMak*zSU^~tA2^ty$=)EvD z5q;(;N@+1vP-q?1yJoS!IhdOhfEo*mWk1T%A;#+$4sb2FOw3`NH|WynWG1C$j8g_8 zQP*Hzequu{uv>~Goe0A;oB2r^BCN`yjLXf42s5uUyk~xp&?&AIKHsHAnm?&_F9r9| zCj`ksj!{T_t^~U5S4}bA8ntM&-z>sB_RK@7BOwDUx=pa``)BL!{abRL=YF^^()DdD zcQ81ZZF@_dsZvgg+)b4e-S#a*+Iv`c##_Dk^$q9QAso}cuUgrlLH<5eDW~P`UA&1G zt5!k1MyobH*Shbu4NdSAJ>}o%9-^!_Hx0+j6Xt?X)XZ@2;X32N*|f7Yn`eY~#|Veao07Z3 zcY^6S=bdLN9lw>7=(W9c(elH?MraM1F^}<-Bx%JT4(4H|p?Yg%wSIUdS9M?AmR zMzs|&ZQ%S5pF)DLL3%OS|J;|82egbu9dc>6kGBPTb2VO~#CE|S97uxid1vtq0UvKj zR8%2$U?`-~@vswJIza(9V>E->r5S)N>BS-oHd}ODY$wg6Nq{#UjT~|7}Kz93* zItBX+=T9bHLoFqRoOOBp;}V1QgRjOKd|UH7))<%=2M%~QRH*|KKmB0AQ{U1RTb#Nq zfUHaWVQ%$EDFeesigLIA29KQQU3 z@`~^cHt8>VCEx_@P3;DT`>WJx*?r4?v>aW9NCk+CoR8ox%_I*kXC-34HXPF^%{WAxZE6@8_dzrk8o>*Kx<3l{#bF$Qji#*7L6gW9aNdU|M?KKp(- z|6O#r*c1=@B#JM34D$I*-L_@)DLcpi8kxQyb;BJ+94-&KrH4-Nw8v_j!3+Cs#F$B;+}XCUS|EV*CNPEOqZqcu-{8vULQ9FhxK zZ6`8sF8((0!Q@FFi`WRl`ui~Zb^u_dV{fY=#u;KmbK69!CY+O`BS3j54FDB4bBZn^ zb}O0k-nQ)`dj(E9Xy9CnfUp};i;!A5SC#80CwY2$o%kK_T8o#+U924>nu@I1=NxlbS=FmNzIdkOb7U2W3U%ljY z?Lpo^2`@k&9bTMj$m<5Ik^L$zt^-B*J_0)a(}y6~bqq4+6CMa2Czv#38Te~yOw$%DtrbqLY!8_D;5`5>Z*ml z2IB)0Xm5+Hf|MB(H8r&hJ8)GD{Etw)30+;!wM(Z0N2uRmGya4sv0gcNp2uXo)WyXr zP8AUc?XAg_3)1@JL{UK@GB9YZv3>jYhJpWoBVAAY8aVB(0|X5e;WV%2>X9{2SYbjZN*|KiHAh*KS zSMnXHD|a0$9w|BiS~s#B{^cU4(a6vqt1?8wIXPCbDA-F5-hxq+_sg=lt1(ad6NxsWUmg(3OU+>_Vmqr8PqN#@LcL3lK7wIE=h^=gw1XQc9@q`ByWfEW&S%`Qw>+ z$9G2d24$!4Kd1V8L~25I6z#GXxaLX|eG2OnZDPZaB4@$T5tMaypW`H$(>{iB^PhVU zdR-@o^WbAJNZ3C+z3HuL8ul9bGLbyMaBUZ7F^y?!`-y6$_p7VwIOHMCawA zKA*J@FstbE;BR(aadbvvOO7?&3yp+TuW6j9mP-@t}foOBi ztNyVx56i=kXv&wUuF;#5o&)PlP3lyceMXSZx-j~a>#KbORgp_H=tla?T!yk>284aj zUcE*$%Y0MWk(VffKW*PgyTW&iE4y!bf4>8Lvd+$Ng~>O2aG2__jh8Z+WRQfyl6Z#K z)bx=SDh&m`ZT5?fGs4t3O=hq={ zYB^>3jyh~r_Nd5GTGKpigZ<87{yV{D70*qkSFN4l@JAsAmYC0a&PCFoD2$CD+4CqG zlR(c{bUN<&`SUs&pwr9O1dE`V3GHv31Y&gGa!Ta_dNj*bT+QHMO=j990lJ3MUK54x zoFn-ddyxhnM@P5BH!IjLK^tb~&F!@_=fI*v!pqs(13kwu>l{bQrs$oiGG-GchM40C zOrcWp>Cm2>(U2Ak3f5AFY>;CV9vJi({eMn}cz%_j7X8G^Y5?}yY&3ak`um$wG@@qc z55vd0b@RMAb+k{95-UY>i@WSA606kFA}64j*yMS-|47m6?mE^cv2(FDJEf6Rjg}^s zIWM+yEz6Gu-ioFm6s!u8foc6j{z3gR4Qq!{^uJQ)Rpn!&^@uZa2A%sDNhIO?k(NT% zvP@PBbhiLNHGzeH(QY$Ib7`c%2uU_v(I*RD`NnI$O-ZeIh&bT@iSXdZ3e*x*KjMDI zT-+7&>H6)xOtz}jdUP)CiLXQlMiQ>Y(rQ$0zV4r*cco=z9gX)tjOx%#hE-*}(zbP{ zik2iU0HyqTx3&smBNTnd>BXUA(%_D%d-gG2zai;_?wRZtoOafY_q#YUxRCFwRGuC98E{(!^qR;$tJw`kMm8PCDJ&CENR74#crmep!NQ-zrPvxc;TmO;Zb z1(Hl7C9?@n9l5|Rf*SMddE+@v{?eE!X$2$3g0Z>X?2sD*O(r^5bzX}f}A)8FCZW1xCKj-fL_Sb41bQTX~)dt4dbvbxM?4^^}rJUf%W8L1OZ*8 z1kpB>#$-V`hD0}Hvumb;UYW=g3JlCHoQ_;EEDPrDN5GK+Z#V5sk#h^VaIiug<*!jr zEWD$XDUc<%@y0*WJm}RhVhuILlH{CD58`lVY=KhO($==YFW>GdiRu%=tan`pc6|2L z3WPBjD7_!#OX_IELc$;OSK8PzdxU0bwtqPurUM@|Fks>_$2W_2nw*ePuCW?!NDe}K zF}XX-Y8+2<6vs15aR|M%A8e3a4xc`*u#=StpE`RUPg;zfCpc`8DnUsJKQj${|6udK) z)cB)(j-T&h;#LXm_m5oZeao!ffuHu0+!&-gN+xQo;N_X5O$NA|oCY%RLVW43`;?%D zfDCHRb<3#PrD-c*b2%5Mww)A#Py;3!(e6wka3I>5+&X}f^5Jfy8&l)@R7_KF+Quc1 zKi1*a(O{mjafY_`!S(0gXlotJP4Mxu>BOMsNuRQZtUgPsZssxDP|iM=z0BU7O16}d zG*T?C905ztNryxseP*}bx8BJ)Q7j1Qydo&vc&Ct{J(|}nDw#>Oj;q@E@bG;ISETZs3tpS$-%*;Gjr)=9!z%U z=mitsgIdm2oU*IHc7s5r|EHPbE0rUdlnj8!>&ZLn#!n1{E!UEARpB#K&92sfpt{td z?Z#;dv8i;E)Ea*P@^|Rj^M=tGq#Y!PLo=LTP3*Pi%f=a`<}J|E;+O`87$XKoG^D9( zSBdEqJ{#{$^j=x6W-zOC;!r80_FcGsbm+gSYGc~Hn|=wulvpXHQQS5D`Br0+4MWmv zpETEQQuE5|*u9(!Yc4fioUm>ERf4IT?oF-|)0rWDSDCZ_4VeQe@{9}o!4=ftaVQ~%kN_mxeu zz%mpESAX1N(j+?cBSoFKcD3$4W_s$BOH5it_i?nju8geeqPQWSnKn<#B2CS#?Cho# zqs-cfx(K7N`2jQHwx9u%YrcK7Jl#IC(a*Pk4J@#_6CAuWRWUtoR28An$Fd=HM1+U8 zry@ZuIZJ7aa`dHcmXlA~w>`Jxw4AxlCH&3Ym ze`(q5dGk&?_t^jv7Pl%==o1tX5~SXyrIAFM$ZXF2;jY_py==tE+q(czVCr@YugrPG z_7wyT2U7z$aEod4FzQhYY3UGEeaM4Kg&HHHPZ}WA8`MUNX#n@{ZwTtMmD0`t(gA*o zi>VsT7dI5_jW$%oc<-#Q<3s}LvmUM$1J6a;y!jFO@&?O_*3iTHl@07Nu|qkmud*(otY!LJ9%cOl@BaV#!UXK7ue>_ngq4 zAltlwfK|&q@8cZ|1@d)EIEAEAul*jvwPRSf2gxOk4Joh?Ls7Sh3{}j4l6y_(lLwB~ zRrEQw7GZ=G0zf9$WCH_(YaYe~G$v?9i>7U(dwJDPAVHS(f&UR&VdC3WB+fb(Kj-VBH0Op?@JD#BoMo?C3R5TVzjm zgG+A!wHCcQ(R7vxQAAT~fbto4vGxq#ulP3Fe2gqk2MRRAR z%~d_?`MRKT+s|!#JyTO-U!B#wptfBrhw}~t-dy@MVvMp^j`55=o}od+u}wuspu zPnGz+ICJHEt;$OUcg=1^Ea6Qj(Rc2QGJ{^LKbQHe_ZHBv`nHmb!@C?x852I}8tt2W zj-BEZ^-Hl@)pIac@<}T;ez3#YCFSvxCw5QTjQ@Pz>H*TjmaSVSGB9g2NrA;-sFCTf zhYuc%e12_4wv8BXxZhT!e2wq zaJ+RU8`;Qr7>-l<5exF7E8}MNN$NO%qBmJDpBil`)a;_Wn-&9v<}DgeYe`enWq}GfO{blIT$Qb?f#M)o%;8s`WhT zRK5Mv@@=0Yo(;|4Z;|t={UD22jaKV77DrGS|Fh=ysp|Rz*-4K1t~ zam(klzFqHX(ywLugRa(Ftw9o7#?lZ@DM3x=4d`GneTmJzOIr&qfu6m(63Ee;&7!0B4?rm?SB{*pM8X z1f@sMa{z~wD~}sDE_1B@Mr2frIm0N9$6Jo%H}p4!?wa7VR?bTfySm2n7Xiq#LRMgP zcT#2)UjQQtrwZJ1X!0cRcC#%f;F01Z!*etOItv^CL-2L?FFIB7Gsg0_C+$XQ$Z9bqR)>)vdt9zp3 zpb00fOYc=j7@sCVCxGa@$)l(9YcHI(AxZb%@pCo(7acTEZ;b8JFM~_PiWW{qcm90) z?`j-E+iBZ{2;SE?Se?cvFEiV0gap-F&YOo()s`n{w%~x|?Q{{~n9zN^u4Y%-`w()_zL+7o%sjUOtd&yI*Wo~d^WQtlz77s4;v)k-a z3p4uX-u@cU;mQ*y8VMfw=O4>9>txbu!2-8|71ce%I}BDR6z|8{e3`?5>63A(%fQA) zv6_gx-taMYg?Cf(Z%6f3{F`&a(;xzR$q>&@aU*^YJJl=TaLL5 zksVH6>azT3Oyp4_Z0FPz`020en^#XP+-Y^x<-TR(Hy^hRn5XzM_2Vy!PW_7ee%e`A z=GAn^hJyIZhtzFzFT%*o7Ki51R*ys8J(V2f~Rd4=Y;q107rO;k&VZT|8POUaCOo$l<}FDt5mMB{*(DeZP>zc~ME zao)Hx+nqVq`-q;l%2r}P+OkcXahHZ4T6oTSGln%8PGxf=s*A#04;`}k#*NFiu+)Bgu}^AR858WpFEu7z0Ydv)YQTsQ=3Q6 zt~7P{W0;Hg@Cqjf=asIZT|L@G5Ma;GS#WN9G3yEY&!r>@S`~lXidtzDID55jb;L_$ zb@kAPJ5p0q1DaR)xOEsYv1p%bb%uIuN^0CtPp@7{2BskzPh7|I^c4>sD52PMdK{g! znFd?qv@IdoU1~YDkkQLHLQd(RdDKgD@qE0O;ap*is#6zw&-Qg)3g}$3^7F}n&haBW zj`*t-h2-?s{GPb*oQ8eC=8){u{Y`CS{gDzoR^?qCqqvHL}`Q zbpKOZ=jy`XD#F5dV2?xqO7MKV3g~Gp{XEC%GBXEm29Rg zS#mntq$qJu)cZWui%ls_meC= z(_GIhN+0j=Lgv_Hju$G{9iFnWDB}HlBf%Bylzjo4_1^t_ zsov~yLb34*ZOv+pYIWrAow|0-`DS(5r#iS*K_9oel)+gqhKD!gR>Y5UhQ(pT{PenS zn?5{r$wwvu$9I`TvJoNNJC+YH^ZJ&^${Bp`@<^AS*<#11G-llaZa@=tQ0Ss2kIDi% zS@kn*GJeaCZFGW|^1nrtNy1eDq`8Po>ZGKUzv%$`K24qc>e4LaYRapVwVv%P_%%rs zXSx%M!*!Zu41Wnv5*k<&@yyeDGi-p?&ks#{X*w=K&3gLT8hu31i!hx`Ic!Utr?R&a zt?|y3JQ$Po-}y%T?a{%m3rf|kS)q~b`K<5g(YN0`Y5`reuV|%P;QO+SpOM?nU#-Ys z=?%KJ$bass{ee!!BPPz!zm*O*Jt+BQ#;nMfhOyb7Ki~h}dw@i(lBHNo>jMxZyXEdm z2N&b$vW&8R-|ye>0t>mYm?`K}o}M>p_xpk5)LAD5mG|f%n;iY_HhYBC>%e+6TX}$< z5#KbI_$T?gzb=Zb-fYavK`P;nw`l-^>kZV0%OC8BkUTl}!o<{Z{)P@4k)Y!?aEnnH zA;d4|KlpYhHj6b%f5*+*+By+Ytue9P_25B$1Q|D$_y-0B9s2s|TlqPMzz0EvnRtL0 z+a)c2cA;$F*b;}J;ov)>+;aC#O`rMPUl=1biTlzO*C3&laR`K$WV+BDJo6p&GP|@U ztTWk!b(&B3udHgJ>lWP3+b^=U>F~Dur=eR2h&&D%x!CR0@#A`{?oG@&_7QK~0E_XN zPC+9=bA7Bey-GjlMdW<78S?t;(QNPGnVfQQ;!IinXy9G!#IiXDb^6CNT5oc~{_hzX zmqyxWEIeV=SfN-ngSz(2lPo>D1T6>7owWc8f_qrHkp}s4+MNdvc8qy3VD|arMjrIa z8d9o8ak#HI(bCknD{vO;E|zWjyNdYPBa=WX-ZU>a5>wuB785&2q9;#tHYJvT>=maI zjt{&^C)Ujt6xUh3W38us#x?vkI388!9ht}5WMR+w)dL3RF*uV94c57P_iOyiA5AbS zAp64cEN6)^54*LPf9vevY7MWAJjTn-dvCk=v@E9D^v#ejHC-~O7AcrXQ0lt7%RvEq8yDLILdXQ9%Kq8C38 zQnMZDskUJL#BX~AM^oc1%^lcl-?#FNY7na(KO`$gTPt(vrT^L*oJe*5sm^T3pV>9$ z_7iSh3tQWF;M$MBtQ!8wY^9<^Kju*w&->()tRdx79-cMcnbopQ*X6g@>>D$7Y-hAo zG%UAGTILyhV+C0NT=Y44&WFqN^}QQ4X?B0T%>>}7R!zyYS|}sO98ddl;_#`LIbX8p ztg+IX?vT&4qV{H+5!X)J&xinQ4Z)a^$(dxLSgY21V?e2b==vYa%Nu87ztePyBWj+R zd-QdCEY)Y6*?2_I*!zYtmV!or$0%46Yb?^F?!kWm%IDj%~o>YK3RSh+D9@jH1sH4s)Q?&dio1Q>A`?G1x1Y{|!GA`}$?b zE2oUJv6@!px%$MYYYz(?PL`GZP!si_nwlCr@Q_s3UAwJY5FcNBVZ~Vc3VW^Ub*83g zuBR8@3}k9}BHf}IH9zrP zJ45T}(^UOYas+Jsi4fvD#YX`vHwDGmHRW63fpc{F z!KzznPU>uz?!V}o4-#A6==G-sEB^ZUeDZ{)cV1SfqU+HENbmx`lfgQxTh>x?j41u< zKM$$vsf_^{?avhU;{c;^k&*!!FvIdo(&4QdhI%R(`Jb{M6ZF zZNK+2B3aZ~z>6_0aZTfz{I7+SAn{k6M6=u%q}Z2JiR^gM&=n;y~Kz!xa&(H#ws3q#%0XqtU?e^ zjEfSl;Je?G>%%>oe1%LX_~GU`<0*^f!!eK8-+6^YZ10A#FN8mE(=D?3N14zdWaz-W z(1L)nsh4sX{8=K5mAUG9t@;|9TfKrOK;ziG-WAO9X<7M7ey5wR=9GsFakg@HCc#9% zc%cs|A88QpsZ&IOu&>VY0fQFDv@#RSmL%52d_KkpY^2uh{fw0nZMgBJ0}4uYLKy3Q zV((6D{jXI<^CJ^ztq2Srs3W%WwtaF6*hiJk!Y0;~JW9OIaS{gd_;y^=vEF4aVQUS+DNH0w<2Tv zEw1_0g$qTx6NlV{H%dADlL^6Lwq6j7d4yLt-N>SSm23=Knq7@=A*eD#_T+YIq<)54 zGg|3a-dsA!h6t&;{2pYjys4hWqujs$)U9FZg{Hr5d7j-yTGE*&-)XgUG=DGzcXs0o{czx%C;Dn(RkR%C9=PTny#RXZvghA5ClwNK3pB`?|pRjJ| zurwPz+^H3gd4nR@GeYzldYxqal(5iQeZ$lBK^svrR8fr%(m~=iG%19Rz(z*LD|^Q< zkYXJ4e&yCwlS=jGGKlZF^XIz=Ff>A2Bn32s5-~Q&v*mVT$h&x~)R`nm5 zrB1^CZ7+@(oLAGtk#}7dVNb*k*j_bNFdZzlU{yvzLRZ*N=y`7R+CRB|UjT8+;r$u+ zyLA5M{O$Mo#SuI&<-voaOqZNLu3H~r6Lm8<6u^gM<<5u_avp;-|iY23?+G07L8s8WtA~*QWPmgOL!phBs2w_02#9upqwDv_8B-(6Dh(uN6O6 zastSuGG(&~D_?5lO*M$U>f_*8tY&IoepzdO{Ep)HqnS_CiAO+Ko@(mF5;Y;vRaV`J z-bB;IEoT0hF>9Y6Zd^NsQfzCRQ=`YsQz&*tKUdSMpdzXIu%}(=-4JUVo9}?t`))yp zdL)_!r5ZgbcCYAmfR4m<#H{G!*SV;!GpHeFYqz~&Oi4;;9^PfP%anB33*dUhqIK5Q z`>-GpS5BRHp&2h~?)>=l*NCcof47*y2A@4Y#~FhrwuFD15ovm#$PxiP5}0ZZUpg-` z>4!&c2U%T^I%U1gn1NE7w8P2!;5ZWd{EBw1Tif2Wl?`XcLvDJZtY_Q(B58T;sS%uL z<=)ja9{JOsn^~h$3~qi<0Y{^(a`i08S@j>nLk&Yk&0a(Ziqq zR+nzmi!ql4)boIH3 z4%U%Y-DjU39w&`hoqa!6Vzc4@X7mhdENRGU)%{$nvjEXR>y=Ft61LOfcUFhkfVxLV~Y2KYa}~y`1HPicy6lAR7Ds1b^nzvUwF8 z=r54sbny-E)(Wj$`Wy)qB6cX7Y717@FF^n9>S;l*!7*No~++!ksi7jt-?v-$|B5DwP((JxjER331m3F6Sg}Pn^y>A zp~ukgTFHLT$8XJ5o#_<|K@3_4J8v3F2BB}EJ8PDura4OG+lfsk3lRjpWQLYVN+1ZA zJIL}=%~k0Fu_G;z3 z-`wY2uFMGWWj85lI^L19znJCRNIyuhG$$Mgh8i?@aG1J(xO=^AnP9-h{ z-G>byK8Hrh0wS=PUNL)dw&q`;JZ7~@j?B)bj@sDdMi&)Tw@ai zOsL?Ak}`ISp5Ae}4i@GtH~5;~J8P^c9=mL-m!>US(huL_ruPKB2@y?_^oQbID)kSZ zRDK9B3o{?ABs5S>C1_T>UwFkV=XKe&4Hq)D13`o!!-%2vVf6ZvC8zCa9m;ezT6Mf_Yu&g?=9l365 zjW0dcnKrDJ^<3reVxAeW69jJPMyo2+>140>4mqBc>vwZTNDx5e3R`6(%8GzY-6oA2 z-(k2dC1^go({=8FW%O3~O}-4)FPN|OVFXE<-f2EVC*}Fw4nX?mLd5991H;n$Qx7+7 z)=bENfSe1sq;;O163xIo zioM0?BBFA{m(_xUmwC-g0lkndcZn#{IR%W+dR!LiKB9-phKt%JaCs`h!4!n~R|dw0OoLDoMg;43hIY^eW$Ahr!R%l`}Z1#TGKm zVcZwuN?6rf5C5w@Vl zUy_hYh?l^@45^X-S?p%Q-b9`y5v?FX-v)OYr+nmUkXU+|eNU!ElP`_TrMI*)OC^d8B0*tOn4b#j+gT8C-^B!9g{|J8XbMfNE1)eW~HBx+&vz0Ty{>r}+7S7ka47A8e zn?#3+CQTw;T!GJpZ*M=jdlLlW`1N*_^t8Lo@+c`k=l(^Bvj9bhAt7o9PCG*3bpmzvZ)NP|yNKKx2OV{P1|S?seu0b_2K9rCHo$^1B83LD;fI-vat^P; zbOYn-5TfHkc(%QQ?zd4H&?FD3#0P-{BM#AQ#!jB)47sbz(@MTl=3&z+Btmhej_|KDvh@% z1M6gzyBsUV@Wz;7pQpNi*`5ve_A`uOB5{mR56R# z2$5&)=yChOxBmP$axY5UyTxIr@2Qwz_mMq6RKB-~-AJR0py_4HOuf^CHXH+zix)>S zzC^CI&T%27)4)!7fB8xR3iE-^S2GwG#PRj%*8 z-&ahkcyHr~>_WCOB-jwnM~bght&kqN(D#!+?!B6FLnY$BPK4{RO zY{T>?o9M1RM z>wK9$f&l*SMPbD(N3E64aqLe`s2lCN3z@8z#hf0`Z7ilM1m{7u+|7d}+1cQC0BQo$ z%bcuJ)Y!i%V|on_ku+w|&u+fX#E1&!r<7k*{5Dv(EW&0f&rCXi8+qi**=vz?=6N;C zPLM?%#&ypMd^sY)a7)Bx=^8^FCy?ok(?n{*(-<~z;M0-`jk-mBw4Bg$1`b2(MQN4Uki93J`eMgqc`aetEhe^VMRK1}W5{?j~XJl2o>m9}-PHlq2 zqwYgiz3;uMuMX9;Iz!t8unhF6d9)x#|w4F68Ahr?Dod zo@QCf@}Mu(l2Ajv<=V^W)L(z)<{vJg3w@$mARH%x2^>Ixz=vih|8~M8q)SGf40g{I zaT?Frc=p{dqS2#|FBNNpgrJauR~kQ3V6Y3-GqDwn07d4pUzULKzP}n9+o?Mhh=w~y zz5yt|Xk@y5oa9h?iKuvmlEK*e#h}W2q(d_bc-O;+6Tvt}<=SqhnPF~gh7Ettn=7E? zkW6nj<47w{EcTiCZXhBDZ7mZDneso85npX>?aq~F9ZcKaq+}ryuKZ{poi6cMFI-4F z2Bt~2R6Co!3%|A!`5q|@iPr+MENCKV!E4>z&*rW6@7>$!+}~NB%~$F!APsX{g^i;) zLL9q`^odn9oGl{_S9y^99WV9JoK5vGNZ|G1lp}r*qfZR7n$S~s&YVPwT?oVg$U~_I z=QQ!#-=@bEKD_UVFAQ2f{8&mJL@!1A!BXopWrC+BnDbMZTlj(5N7!GNzM?L3f4K+Y2TW`ig6 zY%3v@h^|KgPnJuh?pX}lcnkIqp3goT_9J!Pux+;xf`>o6u2RHoa}tD*@Ivf{?X^y> z3ymh7D}SJVg_h{P(`SM*gG89HB-b)0<2*qnV`sf{x?PI#vw-D)1A6J*`Evo4&8E9| zl6gew0&MX5hkSRJNG#!mn;%jdG$=OGe$X>pZ?=h)T5K&T5*UT^9($`d7t9{Sp6D(| zn}tn1c2`c7HdnR79hAG?SMSqMB_cltt~tZEwCahJ+oeAU+cG-2n?Q!(vUfqtF6 zqlOfDv?56RvwDV(9vzqTL%KrVDXWS0jz6>Z%Ze8zoNwk8Pi9tc5*UYaDyg&^DX&`2 zDas8H-II{+BJ4oU(Q~LBItm{v^!CqX)t**oXl@-KA%PJeHXcI`YTkaR<{LD%WZ zM0zF4XRlBZY|adZ7XmuKH0aX;MetzoxoqtUhSJq{)ZCX~LTSe6djoZmsgYMR>;V!s zDPI$Dplh7w8Cxc?lOW(y`0U7NUj1xzNd!pL+hd)18rshd7G+M+`G)bX8EodhAK%xl zU%!c%Pr$KO-uNWF*jQ(@@sI%nrV;;>D7ZRx=#Z6PaNp8n0Zm3TyQ7b*reCfa;9^Gp##_$*NYmyGDx)joGq(9YeH7I>>oJ{=$N~US zLwP_;Y0b$+lY3ezPn?|n*(m!KQf93aDvcf*Z4S-*J>Bw2n}KuX!b;w~H*V z9WnPZuX32GYI3fg;$r*&xc3istXy7!wb@Wr#eF^E;}HxoAr_B8n8b~ZohUvH-FvlE zbj*iSi|Z$!Jn9I!pHu&DFo3BiPw}H(ld8GdQ_F?AYUH_i{uei zG^}?L#k-6d|aUl+OE*FTM3Z=0$A7qkSfAwjJW$+qrGS-|w92wWs)E zth{2yhxx7rHhrxgA3~glX6JBQ9_uOSb3M+M9e9m*8n|qDvzfnK9OE-$_pQHE&8e10 zskc6$#Z#_nROqBAx)?CoWXgX&udv-G{btX7TOZDDmy-E4XVs6NSv~F_`=I=vU*R!! z6dot(A7-&?H%extT5unTc6tWEZ1jZw3il3n9h|fw$m9BbsP~l%Ec5kj_pn!DXTRaO zHYL5v)f5+NS5J(7xb2k7?V7Br^R+!+^`7L^q=KKaa&DX4^6*1FF2jALMsTV)LYb1JuHvl2yolSOi@H-?QSwBsq&n^u=W$sej_kX_J zR}cS1jaegce;cI3*vj|`zcpM=^wAun%qa#9|G(zf*pSFQferuv=gE&3Y;Jb$_3z$k z_^bc?W3u+ID(dq8S0ZQ^hC)XuuyuFcOj}<3sNYrYy>9c0GKE7J*FU=i4n-vZCtC9U6F1R;5xipp5zxvDD!=tC~lSXByh7BG(Q?@%iOd>OWt}WzEAo2K{crGF$84(aQ`9|_6Hqy%9 zoTcB0)__jdCDD$deH2|G&3eNQ-VICU!_qk?CrIn1mL7fdt5>foaL`IWG}nXaxj+#x zf(Jw26v_|){K`!}Xxlgx0DOI@h6#MpU4V5CwxN*C>aJmBc`GX&?~??eQTP9fFqH! zm{B`tchLIxBLpDEv!W9jeTE~a%zsw+^x*%&+?&AFoWK45$3B+H7_yAWa*&d&qZnj& zFbTUNWSzgb;22=k+-!_ssV{{cm%B@9+Qj zJCFN2_e`D6`7H0x^}epx^?JQ7mY$Si`-blN=;WKeJ&eB2ENL>+2tC9ctZ?e3z%E(J zCuEyWI)5yC5#@E2V-92c!Q~hn8B2=pW0b?`o-cIBFnU0P9j@7lh=t+Ra-=TH0&+db)|c8`zl=XJdw&s#$T* zN-lFr6zIeUb(IgW_BTWov-4;@jd8w`?|s;*hyc@Rt6lJ5$Vmdh(b1Qdh_Z=$GZjK^ z-Z+oyDs!9kda=3N`fl9Ku09ZzMnQI5M`cUTR^9!my5DI;u@JSi*93D0qyukv;zMF9 zUd?Y-d#npP!{B`PyvuWq#`&8@&&&C0oV{!1M(lpOrNKjHDvwf+mdz1fSN-G#AWc*2 z<<_s1HLI|P-v|lY*+pFVmSHjlFqJML7;=!mqr%(FSByhx5 zB-vD&q7CLgxyEXX3D|nAE+1? zauu?kw_w*jaW!$-ubViawLL_=18mt#ConLkPt{E-X)i41pu)K5@6xXKkmha2f{%TL z0{eLMRaTs~V6%Q7R+Za7>P)!@vmUSDaSE2s7gU8rZsqAXaKX;c*Huse-RJM5J91Q@LD%&6v{e+?@ok;*U(yjwI-URl?AtS};} z3+WuZ|A~3T@+D>Ry>pj-s-Vhj6xeWTJ-39e)UvI8^SuFdYu1u1nuPW!ok*cw3i2Z1 zmbLQp*73^PuwD1$$6rP4zQHTKv2vftF%OuZiOd$nf^A5kF=2y2W{RN8JJDmX@|m%J zKy@9isVtDx_>Df@6O>&}Wrl~@nzXx>^{K#u@4E}y?K*y2a2i11GqnYFS8RVo*+dELt- z??y~KlxD;9yc`$}x=`ULV7BRTA|H@V;=pd-jy#ojk4)X7ZVYFTPOAN;IRC|quaa9{ ztgla=lTSJM=6OnwN5HA!k<=|0$eRd5OrepM7fAqQyS&%l%jC0)QmV&6)g5+$&^>hTPDhvxtdUaOIA7V0%bR;t{?W zWc>>~b5@$B|IndBH$7GdLwvFmmO8H~Eq+WnU2Eq3P4ukR9C3DVaESKph)AYee00Dd z+}B*Y5I5(_h%UD4X?b6(x`#9^Zoa0SPZ`-gGw;%3$@uP^eJ~qDP}}cGE;>#~E#Y3q zvsAM4{j_$Amn=E|;T1cy4#+OMpN-xK)y;js!s3Y{>e|cqFN`?O~FKv<#(f;-I z@4vO2AwB;V92IhG!n-!A>hq$$;)^mn9(Y^q$bu-o)uGj0Q>mN?pchSd$K1Jc^7-pi zZQM;yR!|DjbM@t(5!wy+Opfx4>Wsg?J0g+hx`JE-?W|JM{JSt}Z{a@hr3W@xu2!9j zzg^yjQ^*83bHGk)#v!`RP31D?R@9e={HyVb_kL6yQXIbV%W!)?z1f`H%&+hFXjbC? z$J_9?H+Prq7DrDBcS^F4ok#k%d4P-0t~L<_K9Ct6}M&IR?0+t>I9pR7E4 ze~jFJCyQ?CGjObc%i~&tqy$#+ti%_7`MLe|=v4oH+5A7Ku=ft#6$NARA6D2}foilr zz~KgKtnjw{Mr2q;e_w4-dbNKV;MgUKzVU3qy2UzC2#^_xhI=$vyu2u>JYT79ug`J`FSw zC;MR!R$=8=wqv`FkPSQA)KFBYROa9PJ$U6(dgqtJN9!6+{T)Uzc*UQ>8CPrO1m=)Dsoq*St4vSuyP*u$-r)0o2%GzlRLAx-XoWWr z^=>YK7Q+u`@c)6(VLZ<_z!)j#RU)az1)FKg2P9tvVb6j(H4&dn5vC9KF{0NtBO?z0 z-V5sqxHxSD{RncNdsGBa00BFL>nj;qY=P8%$LRS*Qv6q$g(>uj(;x7snaCWMPb7$M zPC$>E;lrRJ`J{LBt7ls}DqrhjBz?txTNeHF{jhnH(K$AVmc9`Qy;Z;%@#FM?FDt6w z)(|Em&*&4`OCw3#(jh3B8196}WoO&^{9C|K_pH)L@L8Jv3K zxMxaQueFj+UH`3187vNy#p_g8{puu8sT9mRd5C=yGPCmnx|2)H#}gjw02GCisASIDgi+`(orKy+ABCc z^5JbJSCmVrDj3B%2@#gQ_L+CDB<*^LtJ4m84_$O3A+^ImW>kB7cpQiDpsV!(2ew_Lb)G>EnX ztX!Db`o{qjx0}d@SDqO)bKmw^opps>}bSYT~M(Dq8lg_7pkgNi{JZMh5TK9)n|W7S63lr>OzGm zi2$?Ip~FB^(mSQrg129BgvYgR#gzzfRiS*kca- zT2aALGE9_}O~1R_BAzVo3WhYOjz&7H4jvm%0T;KP@q93Zr7p0KX-Dojm#NCGQB(q$ z+hOZ?>Q$X|va25TpMDSU!$RFq|LDza@XiJ|jR<%zx@Xh%Yw4v84LMkS^|wFk;CDRo zdaCkc?VQ}XXO@7FY5D_WF1sIh^zu;3d-?M%tF?5)=@e~hz z;pa{{f6vd}VPhBi#uT|b;C&kq5cgF{jqIp8d+9K>Sa!crt5&U&qA9=Mh-LsuL0yek z%92J7j43ZnWTLIyOp8l9BDR(x1=cq5F#;3@;x}tq^p%+U`<-80_TO20zY%-|9g6;x zTg|c1c6yf4?ebL){?gVXGfJo&ssJM$e)6J=(&RX@%z&u`4tqGJEQjn+rHC&`9J0bF zh|0@t>dL7BOJ7c>I>CiSkBRY%mv3@T zFM$%*QwOV7F7qy`edBNS?S^JCx=$tms2fi7c| z-^cDG2a?^Qr+L&BnfRp(1#OdE$=LWiDp-Ux+A!<)r{X+AEovn#_w5Vp*E8l`+#pAhL*=?cIL9;rRUs_e) zzFINk45H-?JgUndpL@~ra5!!mx=?l*XaMb#X7FAaeF?@OJR_*>Wnh{}r=J8y7aF#~ zs~75Gi-J|0T!+trZ!6oPNDC3gK`{(SN%ag#Z+U90&VBHe%pw(v=y|e%BN2t^}`F|&!2C+r!QNcG-Pqh zN)v(`^jgF+{qCdL@Ip&4WHP7UQztr98Z9+ymDcuJ8TD28VAR|T-`mg}IGS(Qx35*! z39)1%{%677nLxyXl@h3i@7YPD(fB*XzbVTSU29|rp}7~?a>~7!E(V4fui$=G`Sfto z?xWqY0J{mPFN`%X0?G_6OQBFo+eSc-`zfXEj}`?GyMU1Mq3P_dmQ!PRFaU;u+p?f+ zXJ89;OOS2a9}r-?YTe(Gpo_BDJICyTdxX&O{GB`s4&n#TD$g5r9A{1jA zk2i70mB=^)`B6HX_CPK_u2(c|JORCC8{TYQ>blwv`HOjr76}${l28p=sR!yWMbe3N zvSDj{eRr>N6xxqBw-1Q3h!eRp(8{2dN-!Wp`BkZVmIeS)QI_}Ab@#8m>rWsrfBfaB z!x7tYJ5T8Tse z<^R84tE>GjpizhbjiMkIR|lPtKe+KV49Gj5^W`_?+$l(QmX??QA&j)MyOQSr^^onI z{xdWO2nl>qY|!8)RZ$-m#y@`aMV)8`ew~f|Y8vVfp^F5Xd^v0}So!L|o~A0~xPsLD z%TIG1LzJ7C{Vk;x-2A~h517$3O3x)P#scF;H{P!;*ktH@s6sYhE27I}`x&z#!0nDH z7W>tg|E)`FrFs9?L({$yudIK4Wezjo*njy&jF9yI^;N`MjVCA|e)Q>4d(8+k|8NOm zIdW>G$2a-)7$2-VKry(WzVZx&Jf9W+uGVTL0>y6349;k0Mva#oF(vqb?T|zF`^R*& zHnZ*6zGduzxX8s<%+j(|F5SoGznsZ$D=#C zL~;AZ{{R`&uPq@fw)eX656}Pq#3Pi5VR(&N#Z{yaod4X#hE%p_A_qwa0it@XpJVW1 z4OEj`D1;mS*sR@#={{w~Dz_}N(GRJgj4nu$vZH1rOZzpJhi|;9XjJZM-zycGy?AYr z(y~|E4yv{Nrd~OEYpIbwtNZi43fcVMK2T5{dg(;eWcKeoP*@DO8~P-C{}*A(|1dIZ zm(o4AfvzqJ9e9A{D(v(K(x2<<1=KkBP?+nz%*ZPYT{=~6EVP`N{iux0B|xPoY%EIeNFgB8?xnf4{@Nq`#!*>vUDyVZ(#M9NAbuz$m|KLchuHrv zn(=FO^+|VSSXWn}V?fCmO%m4yD@dUvh`6&o9DTB;)@{2ghumjtf3+QM@pdA{025ga zI~BSm06$R6^j{gUAT$YT)0(yP1!S{zCcl3J zqLa1OxN`K2NGKo9POXK-q>x%!K_3rY3AP!4<}FAqQnyyE4&2z-?Z5b5XELDd;9js< zZVO{1E%*2^G+yZbBEV&GV+)_E@(6q1T(PRE*+m-cAIg*|^(<_B>9y~(jbGPjP|H+x zDf~$4@EGT*zfWxMyA)=LML?^%8lJEsSwp^EFjD46ZN%k$h4^#(pPPBbhgEZqo4L$qQ&KPCtUi*)w4@iWZnhour)jy3PF%G1nux zcNjHlv+&^{FgZ%D59?-vN4P3iRTN?zn6f*V8@BsMN;v`Pz!NVV-i7O++(j_U^!Km- zjJjA2*h#9yMH#_!Vn`meHe^nhnZh9}j|wsj6-w)*_p&C$zksFpphH7}PqM8H2)3<` zNVE;wyP|Z1eZuIMS9ix994cbxi7X-f$)@J46?mmwQOE7Lr9}8w2oVy%DR;v%?yZA* zx0U~Cbd(=k3vyFC_Oi44zJ)&2xO5iU1KXR)O8b~ z6n=-@^$w(U*aXTg67HD*_B_q$_*%@!kc-Sqn+j#ss#`-lnfr(^KrT|15AcFpH?3<;VcmAqLggNXUK(9npKLfG6Od-?f?$Al#4k?{AL=v}~bP(ErSt zSvJ{0^(r^QPILws)qM-oeYndAlSV%6`=4&t(L@2lmao;5J*&0siy~Fv=StShg?jQ_ zFmVusK`TYO!&W%^$>0ss-rH2D-Yn$m3j&t|I|cl*FrGd`+F>hD!9bw>6VrCG%^_XA zV-fe}hHNgZMsUwEso8$8#RVdQYO&+oH_E2ltikAQV|EJRUx#MK>5mS$RQ;lc66nxV zU$4jJ2IqZMh|3iMeCIA*64NM!^v3bNG;c9A1*XPz?n1d-ah@wxrVNZQaRwEe5D>Gi zbT82Od*Upyk0@JLBkZO3FCRpRIo9uX&pL}`L54TG*$^4XaPJ^$5$_yjv)8`a&p~(2 zF~Kstjt2Vbb;Su4U#q+O$NOUgZ~y|2E2JCIGD^P$N@az1{(Mqzov`Qb-=BYR zzd#hCXe>PLCzj^sJFez064?%%++Wm`W*G~sIbCH?pNNzJ@t8x@P2{sx7-ypq4jU_S z-0j)StQt(r)P+JpHIBMMTr|_LmF>>g6X)%Jkd*VtvFUKec|LFr^0>UQqqF$wc*JlM zgy1Uj1_5Mv_#RFwHb*W_)SC+yfF?JXq1gWIrIwNk(Ap3Qj#yobkO++!7%EMRUHyF> z72lNeO2wef)t46Y7hHT`Y*Hm;e9jZV{khIVN?#gs+H65Kh|mqnHiWZ^yf0LyhGS0j@%z@aY3vMWY^KPnw9Ay zB3Ic&kY5YsRJ6Us!Nd2bEhB66V{=zP@j-zz+a)FXJ6GCxa{o8>EuzOi9dLj8l#oVc zgqtlrK7v%Wb?oYNE5uc7Al|3qKkP+zbFr0mNE6w61a$R6i7Q@D zG*HwbDs93X7>29S9tpMNTxeMho5&gg1?4dn!tSOemGM3}%re*p5;lp>CXeU9c|Kw* zAz9=tD`nBzj%|YuS(DJi)W_lZS}n-nB<2=GiJ=}avbb{Z4*#b7S{9VF!5A!*A#13W zjSQst1YSDHR!=TlyQ;^M%0e(zV+D2Wv=3jn6|VIJ2y_UV#8ISIwN2FoWlX;w<}bov zM0k$40fW&8}{F`mJhKDbGsg^*I@WHvcD(GqOJ>*uOD~94!9Ia?iVr z=Fg)G;o*nkcfhxZyA!JnwDNYIkL&UUYNh|2TH!rv-|4i+SnZtc7#JbF4YFqwsM z{=~SQ{u3Cx@cL)T7D3Fzakvu$EU^v{zgjp(WPvwn04Rou@bjr860t8>Ze)-H9FYlN z_#$C579-B+?Oq<~!dBu)w4^B{oJpg+uNJy<#TZq%BjtL_wRlF^o!(@{}+x0{@9j)tmQ=VOLGLwzRSx z<}&PlauN+RG#jz!?~e^rbRode7gx)Sd1Y!}Em84@IuI9mh;dG`VA`K!MBH0_wW8|T zHsLGz+bf5x!oBN$z|5_^_5bCz zXXq}0_>~8qY_Xg}`suRA9_*GBo;!fD*8%Wd(0lqLy>xncmQ3+*+{6cLt3*?RhWHVE zYiZNIix(~=$lWH6p>Y-sx>!o1Rd3KZ$-K&?uh^+C6>`g;h)JWkU2!u!m=Q|)G6^Gn zo!alzFK))=RN+mEsEAvd4eWkOROp24VQ{iPI+@}>Z80@PbdCaL2tNfes>_}FfPvvE zF1IMV428V8=y_dnlftU8YjmPB*}$Nb)0F4o3}caL8uZM=DvjiR&021Nxt`Rg%!@q! zW~mrDq4+)K6iLXnnA%y#!x;w7kU`?R108myO~q{&78S^1)6;53XQz}OG3UiWanNU# zP+WwPH&(rkU3Dj2sC0BLby8M}s^9|^qhiBFn684)Ftu;hyGOelG!yPY52hqeKKx*t z_@c3fNW&a?e#$=|j~^$d3-I+PJUl+TbXN!pye`O&CQm)jmK94%=<6&$H2({r+Bmx; zAdU{#QenKU~Oi7MTLdrnY6u2&C=SD;mho6>>7TwzuZ!ym=uJZyX?kG7cVy3 z;|2p=tpN|&@vIq7V6PR1tM~8&jx-ZQ1c0@>lsDK-iOhDhO!=!7w({a+KysvSgG!fs{ zk*|wsi=_8vnoS)=B-DlCsCsc%9;LB`hAQ=a6FmTBzkOCh>Nng)h?HzQ^ixj+*Vi}I z#s!l2uxDGaJA%is8=bXQ?pboo550n|aW+slZrHG464ZjT$>L`@QXi{ATHpYF@@PUy zgPbzN4BD13iHNZ{Zw!xYc~*&o+6!@8PA?%2Y7(HY_r9M<-v(n99iA>P`v0uQRL#q? zp$jxcOvf6gGph3^we)#Hv+aO}Vk?As)oxAArM#oIX8Ba*2xCoWvoT`vtI^?A{+oAX zweM@5!r9f6r}<{2vskH^chT3>u}QNNQZs2~_7d!2X#PCq*i_evRuVVKYJe{TU%mTX zbRN941#irzrX0FAMTEWhqqc!YBxVo`Uj`BP1rpJDTxtJdb$fj`G&xb`d}#zu!QS<4 zy$?Bn2U!pndKYSnR$sc4nn{T}k-4PZUbaDX;q&HvBktF$=wqEn>2HOxTq(G66NwJG zP(c^*qEL$3wQJ2ZG5|!p>?v==l$cwVNJjZGFI<0_ozp0;r1~V&5w|p{wWPeJ^3Wt; zOwr0KYl4U~;J``Nh*RDu_Z#}lm?-HuUw2e`-00U`Wl#!Cb7`l$l39Gm5WXo#9+f7} zvm^Ql4fX4l`r9+>w=<92t$*;jE+b=bsn~wxFtVHeQs+pDyiIjk{g*b$R_PB5M?e&b z(~eHzDbF8T?SZxHynR0{j2!-KWpaorFYA)Zs{7A6hpg!@!Ae?9`=ck0M-jL4WK)xT zkdk3%z!C3avhvn$j5+kS*sN-v-O{c~Ns<>l2Mxc}>R=sBX2ggS4#zeC%A>iusk1Hd z>oe^oB0R(S;R|oK+A({kn`OMV%3c2ZtPGQEqu83cjQ_fF(tGKg_;(Y|lno;UPK0(K zA~jebIWY@{D4?fYUT(vn9ii+i$)?04b{b=Raoc=7OZBu1c_8*EU1jeFhKdwZQMR?2 zlgncYf~-;uv?s6{0?l0sx!Gk55Y}pckZLK~oV{N!WCBZs_~9PnA7YaP&w-@eXW({S zLzV63yhr8#j!wQHHX-`8n6?m`>!#%{58%NRLl3+xk5|=;wZgq68-eAbY^gFb0KEPB zN&VB<%TgZ-5;5^$j91@{Yt$q`m(T4{R*P+^#-1r%od9qcX8ro?CVp5!Ko?FE8j9FS zHoIzZxq&M4E6+-q9pi^+c*CWYhFEOu_j6ekb4BssBidA4Lp0+4%H|>F(`_n}soB>h z)0e7)lJVS>!!>IeX?rb=`pWDXxubX+GW#zC>vLqfZ@XhxCjHW_dRKMV~~u0pJ@Wbagw z*KuSW9#VfY!Iva_9ogU(F6$1SA}LZk>GLjCOB>LM{Et_B1siL+XR^RsxA$yf5J?E{ zBnj<46k%_%eugaVDSHpN#XCrQ6Dii%T;mVFkT0XIvJ}jLI4i)^xcK0-6B7*2zj!8z z#R7@X#ff>&GrO^oN<0uhl7x>P{8&4#I6VPPtm^kjzsax4=A2srR8vMgoztG2e&)m3 zVF}l-kBAh8(3)XWuCesZ+v-sv`*=14%5zh-hQ7-A4%uss-FSin|2Ta!faNaEZ*(8eyHfh@3^aqLmx(xbg_j)j+xwNOB{@ zb&mxI*nR@b85z{|T_ycyX4a#213Lzv57J?WvEMuL3Mh-tgZe04#AML&z>Gq8Uhxs0I4+Y1@GQ8&8Sb14^*%+#Ecz zcd88xzfAdN?3erKXgaO;jwB81uJeFaeawm8)ed?SO*P_;JY$73>YSZJ<~5ro~pCe=-gjOok$^d^x4+D zddHKf(M5Tr>0Wt`kLzM*XScnC{=4!c_jK@M&WU&lie8mQSF}%;&YcUjvZ|#|pSv+- z(;m`I#yf79D5J%yDq8co$x%=)YfLIq=U#0$noZ5vb-0&KG?Zw~cL2ujzEdqYcz?HB z5reLjmwUE4hCFCR?e=#K{o}xX6dU4FN0rr!Yagf<0V69_Hdq4Vh@1^J)4OxlWNY>$ zZ7a&bV~(wFAUObSH|GS)K(u7A13mvkzg98K3w4%pETjgmHB97 zTI7zCKBu~DZi&nZ_9Xome^hSq^j)1R85;-cze`)=TD&4ti_MWL7t?7$284Jn{T11# z$$JMNgO2v$IT)`=Z4_8idtrW2dxd|{d@SJ>iH0{lrXbI4Z;sq8kv3x%*}E3|oNc#@ z^p>e_)Eabt1Lv9ks0yiPUv1Mq3Dg#iH(N`+L)(=waH8B|3YFM)hAiB^gG>nk#y@`B zht!Z1sD8xFUK>kjdo3*Ta4I*tFA(jUwyl)w9)6lkVUmabCd6K$kp)*X_E|IKm$o*0 z|DGWBdcv6{5YO3LX6M$$0*D^%`)nWOZ$hS7vwS(+_H&)V^2Z!(VWofY&Dat%P*2y6 zv8q!?Yiy-#2bnE-D()jK5&wj^&%KD+F?txt!APKYAk-h`I^z++4H#%tN%=$7{@%0{ z*vf#+O2Q0=$41r>z(Xw)j$uL8n7uIG#hc|acFA%>0G3g@;ZwA}7 z!_#(psjg*I(esL09Fu0(nnk=IVY-ID6YaJO8KXIvI0TnqWTyN+WZPWK@M732x<^Tv>q0RBs)=uQ&INz4#_qZe!#*9xO6( zIyx{+$tw%hL{+TV#VC`|KpOTo)cRb)u7()d%<*A!BbEJ-qIjtGQbGL~%5vp5LVK75}VF2g^LgL3H?JM+*i@RA$$VQEM9~AP2cERBi zFmza7TS{|ZuoCHI;+sIMRcy6FkObzv-B{pZpnF|9g=szIylCedXgGOCGEcFhk-?p? zs;?5J6!xk-VGE;*Yp=st5~zKb(}|(FD$E49eS5G#0;q9BREO(%pOli?7Db_iA^q>K zW0c~|20w}`w3sK%pPz3!=R>X9-EO|0nj7@vyq|5Ko%s3TvMR-jZBHnP|H?O|Tr1uz z-l3wIS@Ck;hnn~AW?!;OrDA(SHfyUi?q9NT|G?Tq>jcbr*zw3dt4YP5H`R z#ch7B%C3CY^wyTIclM|`_@e3imx7z6Wm!X2oz~6s_tc*3(*ADj(EL3GFQUJmaRu*}$6w6| zs9Ua)dOPTr`S2rqb~~NeL3RC%r|v?K)kYa~&}7KI!NE{eGL6}|rT@`DL!mb6n~JgD zW_8oHS6{Y~Mk7qL1kOul0Mjpz==n=M>5;zia8$Xc+O1lsz>R|uW^Q(Gb?GXh^G;-il z{X;xXPMF8Bx6a%lJOr>-m@6{Sq3HmeCdA=}MRnNxN9El=#l0%u2nvB(!+b1!JFioh zZKLj&NkW*QUIVliEJ_?n;ap9#7>8(e!0`ffZ01AujoI;pT!EMOpoC!Q4jiC^kNIV9 zFpJg9Vc~OR^3OZeLgeT-C?{Sy_IavDW6s^yJ*}yBMXf58Pq}JMce2YrqmLmTB=27( ziR{4@Z4>b{aM;v!)0OWT<6#t0x#uL3BV8zy(8!~E(G3U^)}DmGKVVG^9EuiWpg1sLx2I{!BxaS2hqyuam@cUZ*KvWJ~B z8@4D@X5yoeoc_)0t8Be^G_K)*PLIB|^z|>SYZ-R4u?$0!ycwt;Y+yCRcv44m<*uVl zc?VJ>7gh2btYduct-T}+rO?QB>)is@a#4@9^UXglFq7c>xNg(AWiadG)<%>nVgrU! z>iO|V_eZ}AvV+};OlFWlApGmY$&Xv4CJ)QLR3f_RuDJ7Q&5inIBo68Bzp-{ko&5E? zPMO}Ve(UYK#A0#ZsQS(xZvC2GJ5}+3?^=#qjwSo!<1%|n?k*hM#u|F_-?jA!r|Grc z>yFnZmy`2#p|pz$4~_0TCP0Y$*o;>XcfnK%o$19XxL?0{q~wTWr0J4F(S)eo;7GP+W_*9{cvc^`ELbK=5p(HZ@g&5zz`TKd+QC<^j~RW zL+5e_KqfGJ($)#}&I?an*uO{IJBi(MgnRMmh0VY69x>E(OWN!2_DB$z>$P)1_N}vh zBRvUW?!m=lSC|}j{-Z!98K7C0TRmcFL-lr~(eay?XtgStjw00n(&%8hPCNj*&_3&0 zG_F&p&Y7{g=;re`?oleF*zag7u?<|8Ig~-0m&mLY^sM?%8F9jEAd5MX_M%*jf-5#Z zFIx1_qlV$p+;fYnB%CePyza?&^LNUmgDOAvOyIG4hc&v)CQDnt4bM7z8g~sIc*WOE zYi{^UcmEzS)~|sK(Yeh2qf$^#MV>KZ3r91CD@&UTe5n;eaJfeUgNc!pSB*L0Bv3*H ziC_;F%VaE?`y9GNBGI7n8PZbQv)@*!@<;CbbWG8FgUhS>m-4C9V)Jj8ZDjHnByoF~ zZql`GOtV&c2b2-Jq?;rTe8y?Lc#tz`MWw;}v!~*r&L3Xe-`{`c-18-;6TP&Cy=D8W zanp@xv^XNEVogLcb=aJv7HSZvWNXHU$XIsAP>Zxr!Kb)z!HF$$W!g}%bBmL~CA96% z6cym*if~98U9Z(@k6NTKo@5&hyjRYGOMC6Vda%05n6{v$U`sI(0Vb)msF(z}3{|r&rnd@dR6N&tTZxppFV88OrLNRhoV3nq;O}@I51%D^6n!6Jckb z00i&}TXnTYoE6r!7~AB$c%-Pb%b$bk2;jJH+~pl$bg6l?yahy7C))UNBpMymq}3Z- z8xDx_aHR(6DXV$Itdz3Cd!S;!i9z52LWS5QC%m(od}Qh|`CL`(<=GC_mA?4~UZne@ z`4{sj!~)f#=6jMKS~<(N;6fL#2wK`QEm3G#!dfGO4X^D69gU9gKa_jjS6U%MEex|)Q;-5QR_blD28z?tzEiQ?&x=4iyxY2DaX z9cg!aPF$u0{B)*C{eH2fzOa?qqLW^rv(SVy8Hr z;jXuA`W>QC8o#+fj}4f1SFbyw=k;QmRTj}}5+^{yWWVIw29TiI&6VoUy14*(pUTzo zT)1#yV^~o_Ql6XlZjdxxby*a#y>xG9eCr%__9V7_gd-V@YpP@=zz`s?4h&D060cOb zDtp>?Q_Y?K+K+TF<67xkZ~XGyK1|*-CqMIRlOB6Y_?K%WgQ>vl`;tpa6HorMJ(Y=W zceREkOXc|WyMUyw9qk3UjyfFB^C<;EloI{R0|mZL$_dzzl|SA$i#C2 zbq#OTzF;X|+xBBkA>E0%1@QXi@s6M?&}=-1yoW~Mi0>*$zk6{naqJxhuwu;t(`L)wC2F9*P~>*%h;xyh>Fq|xwuc+psAN) zOln&`cb`C}BQV9j@mF@U{vx2@enP=3xp&amRdes|0|MS2WT{-4%-#|b38i9QBe#(Z zH{N;EIkKqFAFG-fXgSPLY&$43oL~-`&3t)Z)E+V!oSuAClN`E{nyCD(I^{ECV`k+J z_{L`1A6J*xXo(5<@?lNeKphjB?uAv2{1eYB=dNA8rR$P!<2NrJwyura2FhQDJuW9b z>{b+YeYSyNFmx(`t-fE^7)qcEgSeFU{C4m|=@TiiN!*iE>EDZynonR5ODOsdRPG z#t4mo^P%F&v<$Me53y}KR^N_#{*p}~YS!z`Fpak|@#TTXQLR>cFJ7Y@zba?ik^NUm zm$!3WEFeYiu-`O)X*H|e!;emryhElB4^7oWy16${%Z7d1pqH6%e(sLJdp9qw(s(nU zC=ia2bZ@5VZ*)z8UPMh#xKYX`-ufo>N+f?NNaOytdiN`rdH3>DPE$FBh`!+b%NULk zWO5r0EJLXB?diiLs5=pkvt+-4HrBSFw94XYwYpoXGaU{G%HZeuA|{dsA4P-X&cSv4 zPVIUn&~c`#A&woH4vrygULi}2@fZJLn%dXJupKH=d8+87Q$_6DCoTOi@$qR4(g;+u z5Jb2n7l&#|`XthGBCWf|9G!P8xUe44e6K4Upm6dr9+El>wpexUzK7y*{I#wmhonu7 z0?qn$#$)}uaBN@+KErvB)Wj#0NILYlCZsy%F8a+kc6|;r?ACXy%;R{I9DXN7eFjCB z2nIazb7u=pn!VD2eAP0(s%mFkk>^mpk#vfI<;)X`*ON4$eMIfDDohCly7Hh6g#*0*|LC2_ou{fg7YCr0STkW z$2Vig!ft4PU5?AM@VHu1EeNIot416nQd)pD=&}l?S5^gFD+yEj&24J|cufIA`Pi6{e{$QW`@0HK4 zy-`jd36f6-OL@t$O}?bJNDkN=I@EBVlwTPBp^emhJlwcoE1?{7zX41A*LU#wbm(G} zp$B@B-MnkjO?h8(BwCnye)XTc^0emoCI{UiiJit(ANNro2Xbsk@QGN#DWiD`507X_ zc)80Qm0zEb!}3blIBa2Mes*-rSg!fgx!skY);VR(n9N$1s{#T7eoa$;r0N&|0}1r+ zUdC9^jbE#}>c5~oC%BnSh2u!56&Y&3sq;(|{TXIMB=s_D+xFAS$J934I_rtDv8Mco zOnaDV(uj#v#{2Xi6aB}Ug_W=T@V1ZF6Mm3fS*(7U25)T$H}$L24!_x$BMG%IU_STP z0U=GGjFsK+P*LDcu^tV%H(!H%nLF;BV8HeGnAn15`ecG^?qPLbS((6CmcF+OW+at9 zmWqYv)ROJo2d~$k`=9@5JM>XCdH$xz^-y$L)GAhf3)&7)rTIrol+NsESauw3o1#F! zDMPGrJ@S1H3iEfjmw9A{juGZuB zdm>D1rcEGr*Y5ZVAn`lIoub-kW!S>A=MSrDSH;Td{uUQ$hKsYW3|-Q#ZqU1`lt%&_ zeyE{;eq#pg9&U3uXobIjBA?V6bd8QfX9jyacc7-~L)FqJqKdLGwUu z2>m6gvAlVAa%fPm6QK%DC{+&17~%R1q7XY{IM8k1SrK)KcDKZxpQz zdk|+b7;&2<$_&nZSNOcKI)Ac8Xpk%L6F_n#6-v zB*zz)A@9VqT%_JO7S&1n6x0f`e>ZXp2QXwy-dbczFprqah9u1!X()qJ_tu57#srEL z`)s7P0oyeCMO7fPf#lMF0SzLXSIZ4Wk)zs;DxIjAW?$=q!4K#qv*Rh1!1|o1ySvLJmAIn6NNU&4N7KKDcWPtJ= zq$aWME7y=mLN1JV#m5z;>Y^L+q+(9e%X=;#av#;SUg=Wi@DJvupjN_s7M~>2Pu#i8 zWa2A&woMbbB-vZ3yhPw8-J-c6#Z_r5qQ9zfQh+{|n-r8l?g1jhL!Z#U6%TQ4)xzVt zMCs}P0594sp`tG=XDko_)B6@}E~`3Hw&+-$2mD$pHA!k0JP&2^-lfNZ=z6=*)yVom zuRci*ni4$y9Hp%ovvYyxn51H+)nAOdnzvN$f{V=likkLhsT#1*7L!>C3r@4282hKN zWp|;oJbGxTK9Dv^d4Hul@vbXY-mfP%V6`~4*K;VcW6-2{vG2m4l(%ZFj*fkFnnv>~r`h@+R2ai!?HH#2pxizvZ)2|MCO*kL1JiciQp)V}m5VQhCN>+)^$}vKnoH0sdkuJW*6~W-BRE?AOE< zPX&i4t*x-~%Qs&W$3}7xe|-fwW-IovHPs-zloYy9dYl&?g~V5}{j^0<+T~U>tR%w5 zilU;<%3hp8&C`rTCXW{#0sMzu`aD{0ShlGle_79? zPL7RS0%ba`;#f!LC2en9?fus5|HIDD7o1+MYV$NduP;3kJc< zEM>@nVs}*_cvCtDs8=w}TM6(bW37sUCj;u1ZzLllcyfkf5+9&uSNcuX4*@pz1ab{S zSy%k|B!Y=dC1%1Qw747Ll>~;g{Z{VKrbv~0OL}DqMEXa^AJ9XhVhSx+)Z@)+a9Kzq zWDcE;V((yEW98B(H)6-QiQ8zmu3gRQ)%!i&L;10GqxIC3pG#M)y_3{!*n0i`dnoWx zxkfU9o1V=CrhS>i;kqe_v*{{*);x@6=Ha)$Z7x>l#I6*&AVMB!bSmg$?U ziwb+m8Z;#yL>wyhIgVk)V=ET*7zAu8C#^#%>%jQ?5Zd>(t;J1Rz@ihIQbyNx0XKCF zD^X{G^mfN19+B=3jlNJW@m)D5ZN*XgidU6CnU1p{y+S7jTpPUJNY(Ll5p`{+fSYAQ z!ziPj1ned529jtg=fP>ut%D;8$?+g6wGP?IB|iT=1w4X4vr#}4_9*66o=jh@T-QCj zDh#lGO>!k7e6g{+i8T51-;E+26of$}SuMt1kRy{2eoG<7c(3~ z?E4m{t?Z6NOccpC6zJ^&t6M~#!A0Uzb1ZjsR#y^)UIR?!qh^sBO2wcGk+zT`^i>77Ncwhjo)<}^!Qtk7e-d2GG{)o9~uV5vMZ~1 z04WY`z_DZeHv-A&WaghtjWQpwZ8<6bnJhb^K=wBr?vYbVqvyX(yEmHi5!Nbs4H!Xx zS|`J6FT<))VMdu#+KcrMqeZs!r&`lT!_WLHNx*kp*xP=)r)L`FnE8D4eAE>~5_KB; z1F=2+$Dn&a{o*;m{S;W{T(X2m2K}-r2At$kkG67+e*`HZkvlZ^`V)b}rEN|mlgJ%# z(vVE508T)Nq{-wKbswBpt)ct3&AoVRt&v4R_>tusar`?;N|Lf(!=Ud0f4~7^-MibP z{YmWAwf5@^@foDI7R1pUY}SCVrR4`wyqi#M;T-LqwpPEE_C!#w-_IJg1p!Mq^|>xo zOqq$-0%^7#S4P&AR=EJIiW{86ZyoZbMBtu)L#9{;@ntpqns--(8*pBFzy>emlz2PC zSs1u?)S2HpbU-~!x?;yl9v{zh+Eap0`z&LXSQ&9hVcECoOs!v7S!}vx;&{u}s<}&x z+HC!9N@|gq3RhW}q@m;K**pc}ewe|~qYTd5&}STGl_LInXP z*E1K-_}MafPr)f`=HflGV2|_~UBrsg%0{cLDUTsNi^83jKmMY`M}IHmyZ-bFHJ&?_ z%a&aFY;|eE>lg^SNvl;YYgs&?5>E&3%=a_wdVtJli`Z+5#mFUaUmz6>uw~_jee5ZP zh}h>s{^8mE>Cog~lMkB(|I3f~&$-L~He>yF-)eC#T{1xLC3O5)+5f|v16>og)L}!p zD(B#9sFit`+={|_>=X2{;Lj(UTUCEqON?F)j8jjvaq!@^24q#<`9+JyS+3xbPTUY- z%ZUI1&|*J3RE;V=(#!p_jni}QF$gGU z>*+a9*N?hXyzk(f)usj)*IIn)rm~$lD!%U6^rG>@H`gc_uHoF%;G_!TnPKht~JT9TCqp`OS5Isl4 z^`4%`orn5=uWht5eVMggu~_50dP2KP)gI_vGrQU+Vo5@(PyJnqBg3-u+ve}7{MT-8 z!R6a&dACgmPc^w#I&*cw>P1RwB=`37E4TdTWq?13T3k&4kM@x4*QoU6H!yP<8?UO( zV(==1JU`W?HYLuNf5JZqbN{&|Ai)#kIcYMzbyMvr{tpk|T}C#c1Ui}CN;Ss)AC$8i zect%_rtmLC)ON;Wlp;G?1g`WtfB6@XSoWwd3fD?p{`2?o@7b1y*3zGUu1#WX{`m*W zzyJGhEw`_&EJkOBu{yxvFw~R$IVopyOnN*)1VEPf%MMC zH@|`xEuqHI*?iJwA+G=}-6ZSXA_htbs;Z6wU-SM^vWLbhl}{y?5{RdIO^x3KEd?&q zl9$mrQ*eGy13FKDEbG>35vQG4*ee1#`*CXgBblj7(?qCvuJj{5Arid-ovD0_lF4pG z^pc!Y_q3gNU`E8!tq3GD08>_ZS<9Yye%$@P@C#zr>N*O01&;=-rnHLK~VYIaRfbL)bWic5-2_7nzP zIg3Vf&ryui)R_`)^)9ZTD4RNicGSJfFtB7F)4*;6+2$f2+zq9x=6S)C@mUyP@Y_e(e1|sdT$1 zrc6?xciZq-lHEO3ov|~#Y9;NoDxQ2#f>n>&v%3&CW%O0gc=F+wwHo8G{`>k-yHR07I@Zy(Q!sH zbB7}EmbO?48BjBr! zq17YjPOZ0nLJHg>q2_}(PWZei?hkI&90(f2vU>&`NozVw`YLFEEwIED;y7V*qTvs2 zJ1#N{z8&pzOhPI3dpIzSxMwKGVM;eL0ybSwW%dPhlfl82w96Mqf3?S7Muzf~3&Tdd zFev<#qs#1PKsjx7bOuEj$bw!aZa0iq{+$ELZd_!f0_`Dozy?NiCcJ3Lx!PKiE}tS9 zA4QF5>s<17EVvpJ6bXYek0W9L8*zt1Ery(@Oyb~3pNZE5k!+_Ob9{;@@4VNG6a{(V%eFVu~5^CcI zd9_vt@9b?CaY^1bpu2Jeu`j0_KU=q9!>4Pe8_|LO(O-`cW|p7pY~6f(ya_zW`!kNl z#>RS_E3N!N_A=lE+)45eT^tK~Sj#{v$gd;N1vuY518?I)HV1>QWv87!-ehm0$v14h zKD)_mY2>%SIZ{eARh_S`|GpN_&a)D1epWqV1%t6X^1oI&hJ1N*KuNEre8Uq<({1Sb z|CO%A-<^0nt zq1@jB7yoWX?+6;rw`A)87esLNv5ji*yf1GEDcA5wC+I=2E+_r(&_p)t5 zpGW`jP^;M4`bWQAzx!;1S$VmaPR<)@{j9^ql(2(r+fnbn5B1EwF}Td@UCw?tUtb^U zb)eJGi_a#zc;@#{pWKC8i7)V-8@(_6IK!=kQ{4M2&inH77Ja`b+N!wr#vc`5{pfl* zF=?3FVI$uOqn-VIk9lTysTq23`G%M=1wsVYy?g9^>~OK2aXCXiMH}@`w|Ks~qbjtF z?<}|C-eosx{+=)dkQnt{P{^d2vl86K<{o%jF(*!%+u3P*jhx2c+OKHN&gb8=&6j`A zoI9>POe59y%cc*1OZfFc(y+WuE0(;fnbEWNA8zdxj$Bag=$bB;53lST{~@30uC@X+u2@hjU!UB7&RC@Vo8dNlL@^hqkBwfeU~>feUh z3_WBS(=dc6e$_Ch;lBkGv)dX1F|CnmPP6`LwVu33sa*g15P`@1ZG)DST6qA(dzI|A zw@wUlp#}X@2viB1ZJM1Gcbb3LEdNKx`|pN9P6jUqF|I(vP3uenSiR zCyP^6?1qC|mndDqp0sQ=T=^$3HG~oj$|d?kPcLed9<=*2C^@b*+UWgz^mBrrg z#nVohIVg*Dp+C%iqewq#XCdVi1AsLA2@Y8ICU}=O0L)r-lK0ycQmS*QD2KRuec8Tl zLQ9qz60Sn(6Lp_7(uxwE)ajC&a+HwN@QUwsQc9vM7Dj$z6lXeW~to`!oT#%Mit(%1zr>8%6v-QfL3BYZ%e zxQ!zD8N5awWtD1A5NOH-)MG+A)-~wuu&Hw?O@HpJIM&Kc5M3y=yJ71Z>fgPNuTH0s zlS)@O0uZBn>LBTEs?l;>I8r8&PRdy7^sNJFe?rUBZP>1V zlzsyUOGG@Re&8K1hWNTrgE45DfwuX+DuAUcRDB=a>XbGe8#W#hb%6Ij9sd26^Cl*1 zw_UmN?6q69`nw*7oo&0=EwAlj-|-nCD_)p(Y=5g$Zo6S+CjV3*`)5khowgH3XWUa| zAHH|lhVfLQH53>O;>HbDOje&O)3S!WXCtqEp!sB5@4L56Q<9pNp6pVPeRx@?n%7I8 z4*S}>ZU^@*Xz0WSBjIwY335Q>3Y!=;$duK-ki+b*RI!YkQ~9;e$Xcgk_IqAE`@T?0 zyWsmnzPVW>8?Szeo-54J<EJvn(a)p+wWIUk&11?<2HGuA-2%QO~IB`FMo zxiOyR9U6hu!2{NXZp4OTTas!BkGyydP*UAsCA-yB;Fa8JdU}#=_mjb=F3W@l(7gaP zgKoPYo%W>H)T5$=JwB@+fT1JElK!vS6rsJyUOMcujZ{0skIi7*wuy=N^|4@1f9%_N zm{e}c(c0}q>aE^-N39te*Vo_ko1{F2-oub+~0wr#?_T~AE}3PtI!%P94ALDkBY>uaX;z4ZL`u>M32pVB`OpFf_jP|iO3T}GYggRieosug`2 z&Yi1e+IWk;8-AF5eJoU%|57B-$Q4`DZw;Veop<-G{oME4_yNRNv~An=&Ko5@B|qf? zY=%6j3q{&}-(wxTonuWvXnDD}JN(=9#Vc3dVoKxiHTsS6NFV{$0x(rKr9*0f$BG4o zk_Kk4{=S4!%sq8;b6@vh2eG{Kn00m?Zp`PuT^~C8XJGMd+1WUHhC$_ItcQ$nN?Z4x z@`Zc$LXOcyeGEQ@GSyNeeVAzk)+w`D%LR#qh=fz%UIo&`A^X^KwSiUbb}U zG^ZTdEw7IWOU!Ba_jE7UaG16(bIr5dxid`32xVSP)oPyaQj_87{3P{v_dG(Q#H!FV;>nJH9dz(3?Ok#k~^%bamBI0f#r--yn z#5C~AKJQs2tB+oOQxkd+_{1~)hKAj<&-=kNZw0x>fAiY>OUg%U*GtzOoMKc)NBy#|+2WOfXT;X+rNpKvulUTs!Gc==K0Gvz~9zV9B2fH&~>U zj1vZQ8wT!jxUB&<8Xz?-Jqs~5moZVOopuZeTPbdn>?lJ-E6QuncH)oN*vZ$3^Q-JQ zpl9E)ccD;I5-Q$yyVoM~ziG|aSoWVq6|~h5xos1#5;pA5s8;lG|i-kwOy;kq&hAs0Wy`8=7wjC)VjEAhxaFDm+uZfnr|t~s5hLvOURjyxA4L^R z)--TipBiLCP0oxJ>wB5#`}j8NK-@hz8R76fxwLKoFzZb$$WSsgG&Zxh70H1pY9pt` z1PZSUADTq8oBi7#^6-gipPjsbX)Aqq#N{LEdxtk?V{|RR?zeEKU%R8$OR|S71fXd= zCPMp?92?UOFXIe)W1{e)0aQQj)0tX~;@v08JF%HSmn2iL#T(gz!sKx_r zciN{qPvC1_hx>jLO)g8QkMseVh%Qn^vSrS|u73CB_Jp(cD>ssh(;~D-)ga2gER^Br z@Gi$sA7u{!cvL8t{+TGbVOt%)yNDb`gzeJJlbPtkpdul_ra`9j$V7Rzj>{3ZHL@Uu z9Z$FIIw4H00i<-m#Nfu39u;Zu{jo9y-SiNIs<&VcW$Q1xC zVIadN0W?n%1K}ufo5nl{ccHZ@#yf8_YcHm4m^pmmJJDTu_z0I89(v=q2PqPyzHC00 zaC%YpNVY9mBoLNxaQoIwGI)zi$r@@!$<{mKrLF&O9u-J~Y=?MQjFt$XjzrAyM0y6V z$w#9xXmtOVm6>j2Ogu7|kU9*(^p>94cxN{-D>@Bi!Vg}{+{J?z#(u8NrNhKH1aa3P zX}K0VScGRYM?owO5-%{Oq$q=HDcsEXb6RiB^~A0?p%bLt;rO!+4*yMwLN&?p20_{B zP7;Hr;&xamHg;>vq!%G}R*StZWW`kkB1ZE?<60bN?){!~CoeIaefO&{j$x#C-Ahth zk#HE|6wi8Qgd59w&ZrJWORY*H0`fb~&2~7pDR*f*~5?yDJt%M2ox48mn2;W!w(!xPh zIG(W4c}`dh=wT{|C`ruAF+kFk{c#AWC%-A5n;}(Gs$KQ$kNk*vQo^> z$g=(B>Mk*VD)Izgm5apLV&qJ?Lje^3O)T` z3Z1Ic&o(g<#lL09G=Op0RGS9A&lNhGCJ%;+1WxfZji@4~-trMD_iZX{0w4_WM`~%ACI}uSZPl ziggiumIGmK)Sg4-jIM~m1dbaZ_8fu-?^P)5xh%eg z7=q`gN@?W|zlfK$F)24&Obh{fCvB3!U_s2f(>_alVZ+}$*QqX&3lmywDt;=*i^n|st1CjYaJ(UGeie6lZH$SS)DY(EV#!}_xvVHagmnOM zN*5cSK!Sb*w>_GBAOQAv?wDH3bZKgL;Q1TiR0E*?8>mIlC_IGUsqA}U{?(oH{E%y_ zygcStrtFt>OhgcNqGeP%eU=uQaJT`nSwJNT-Cq zq%h6l`8LiKC{`2KBVkS;-Zk`!a57Vu!4yh2CPns~FjPe1_(-`qQQJ@aN!W_7QT1En z1{3Xtt-pv*1a}5_11!=<14G!Smy1ZC#I}vHM;QAs7AI!MPUJ3;g5?vOiEt417L&T* zp)`yjh>|>QvHP+CgPk3yt1!)rq$K(DbJji#a79`WL?K@v>zhhK@JKD!M6pB6?AeKE zWbktyxRxvj;ps5*EUcVFauPBS@s(J=so}aC`%tQ`>TG+;qQoQ_ecZ10T+SW#{%luI z1q4*Hu$Efn#{}LfUWvCA^<4C(xEoW~UJLTIzzHjZ@+U{x2Xycjqnm<~a^|~(!GFVb zv)p(K@v8Ve+_Ze4T2)e9ybJ^8(Ad6`Q#P{+%;q)Md16|R9zBNZ$|x(pu`Z0yE@|Q! z$WzKDS0od;gCR22Y!2S@rsDdj*ntC+IWr<(6CaS2ed0b7ah1XP`#&yUvP2&6vBye6 zw+t3=7V)W8rGtc+Ndhzeeh~N3&{oX~Q}k|`e9%(cHd0nFx}F5}Uj%U_zhzXa`T$XV z{?kjkUYwKBMZ?X7cnxmyOQJBL9jP0|5M`P8^FT92gnLndy)sUZ)Uzafp6th-<_=Q@ zq=c&2KEA4M6U7@?I) zD+4q9lDO--n7L6lNg*W>uR#0@At}d_R zZOu_%%PN!WhPoln)k(Ym!56zQ1hpVc8*NC9{V23;D|11T{wl+{q5CbKe`F4b~ z*<$!7wU)1C#b;IOz#(Q^Qd^zgHn%VgtvcE@ymxsj0G(!o^Y^773cRIC$)83X(Q9Lr zq2Oss%7M|H#0Q3zW@nkl62GE|*HMh(MB%KFwJ-PDhS#B*F7Hd4h(B4sQn?PnI2Nds z6z$73mO*2&$S>L`;U-m*o4R+62XkFPf+evU>PDF##U+x`nmK#+J9So<%DET?${A;n z5RC)s8}|jAI<;B27UlR50G^aF2A&p>2~H$?)H;q~8p=K;i~re-yG;f}r?IzzuF0?5 zPR!Rs6=-bIp>Q_;=vd-qQA;{qZg- zujV)$ll z8U+s?MOnjp0TRjCz#IDwyTKIVFiWLs(YTn( ze29`OIG-|P+cNUDQA;Cd@GRM)prna<@PTDf0R`mi1p95})!-*b$xg0foM$It`Yc~X z&4&FCDHofl!e$W=>=J5#H&Gyxw3Y^pmYpQkba9sV%N(!q^UbQXVj43aq*w!Eo`IFy zGUoX80W}IKK_IXYaO6n$CtHNRlO+%`o>47g3`hCnwk54bKAMZe^lJ@3`!4o2HYHA2Y>%H4Gs6WLNqQ7-M zJc-1xVdb@RKf3%fmLPw5nsU)^5&taG`PMJ7VwfqFx+2|vy!nlno2YbDOA@IU%4omu zX*AHSbcA$BJmJO+|0)WkiD?#@XWx=0Gg-y;N*QPV)z{W%i(9`$k#WP7H*Nn>1LlO< z`{S6WKxUf-vQ%aWG)CMn`{g%y19JAwRu?#BTOs4Dh87h#T?L^8fss`TRHUV!{xxQ< za{g$)xR_f6{j=QYxh1{+_#+s1VVV5G$JEEmggcIRI2R9>`GZU~G8V4`&-L@m{L?<~agqM`()S6On z=3xEd-dj(0ij*dyVpG$)l~`;4r%l!nb2BqPuQ4&`@%U(`t29c`0M3RLkcu%UHbv3u z$tvhIKuVcbuRH{bBE*COF?8R#0b0p(K?W&JqLrv2(0)o;tHm$HeQH%~p^~mWvVmgc5v)J@0z;NgGcEPWWE$JC`bR zCGaQxeUin2DimQDqku(z&A?#|Z#QgK$`cI?-57S9D-x8%;K61yYQvSzE%C!{NQ8FjlVcIF*x|#s3qhq4 zY@0>QN%{~6hPZ*evs4p_62XLkv*PH8BSd>jrF=FYh(dX7mrO0*{t2Xf`^F^I` z2q&jdd(E#R%&e=`_*{KQ@w?cEv@U#IJ4aov+zT7OsaXcB>C4>|M*Z_0^3z5}D9@1qD$E)u>8IpzN9#B5A4y zo{{9?Rhc{(%TV*!xY2Xy9Aw|#nVaR*N^juTlnyEyn-R%Gei2}^3dq#^L9@#|Q|tSE zH1Q9x0Em}C8T|UzvcwJFgA9_5K6Sf3{z|)B9=%81Zq7KmHDEHQA!3|O-^pU)npkDs z@~g%MxoS<@sTk27p1UKh{)sE@-ZC)rTTxkWB@_toA^uN>k63m!CP|T$r3BY$@A~|6 zG(YL!n_8y>7gD_@ANk%^zxC8ej)mRfW>hvq%F%Q6-*>g&cC_nJc!HO)Ft&C|2k2;-NCupGVNDU* zy-|3RTr*05#F=1d7;B~@a;iTLI_0rYgD&-aDyLb#jW(phIV46UQZR z_*W&n9Px9}=Q(GRq%(QbnXuOx7>6!{Pi^MVWkS~J^F8#*e-G|`qCT*d!EC^jQFW1f zsp3;95ncl-ixTUz)F3Nfc#ctfCmzDpz9WPqqPW~2oyx;3hE18=411*ygQQP+sc|l` z6+|X}P(?vpjef6ea0o0^y{7=6LY;E$`XtbBCUKOfP3>KB}hGQ!m6 zJWEmn##}Rb@Lr_ctNt5SH0}8Olw}pmAQ7F)g12R0IUeS3P1LckPDC zmLO705Zuv7T^U8ioY5+K9`ZfW5sIO^V0o^yyHq3=)1~i@Yq*^k4?f z(Pf+ieM*2#EFL$DQmpRf>j9Nr+-|{ei%*65%o)j(Z#)w;^QmnYQ^1^4esKI{!DA0- zpCC#Nfd+%wI+O17sD$puPAbzcMp^v+_=W!j3E0Z8NC>6W^)756Y;I<;6WjT2;nXBAOt|%An&C8i!zL2gKkmp)^UH zsi2U(?5whgE{<)FKJu>#=vbEcU{}Fx3Gxe#fm7BAm<;0!Uj*Ey5S&H)n#GjK`TE1f zus*M8pTs21lb$51O@MdSfRgUryOXnaWdKT6Sr&J&NmqW-($n%>J|%I_=(#9zu9T=i z_T6=XxEw*xSQN9Vl}2pT6cwe|?JB}eFxh`9tV9~wY6ExCK7oHJ7Ma>9D;_j!QFl1yHf8dN#<9F38Wl5$3QR~H;~Xy)Ef zujnHls(?8UW(lCx$HWBR8*UdKCuqqTHihN}G6;s+%H*~@s(d#6rMv>pAey0YrnFcW!k_`p81aIC)@&6ZzSXF8uun$tz~`lB~o`33{p>^t!FFL)^h8gUy_&t<*KpBaoSAkulcOrwyhH znWO&WVfQH3l*&w>ox3I? z1_GP^5yhIgkrx)apQW7?B&tl1YKWMFQg3**P~;|uhWrd_oJ@*EeK^Pr{!u0^D!)us zNBh`6KdsQjUP^?ik8YhVf%WyHMByV!kgGcfs(;?~S)0(Vk3~ZUes8-)_%09JNU8hQ zGVv470AhVR)V4MB`*xUgOFs2B@eIVaF*27qb<@_Qj+ULjYp}kvcVFiINd0_p&P2K6 z#k3el$)eXMQ}8L-`I;70V5{&h6Ic{tdK=bdJ8{L>DV!0w?Wy(;AP5wxe{2(KuwfGv zN}0G^th<*7Bh8xmpuw-tpMNT%qo|*fq>!6xJjZpsF)>b`>tz}q==J&6 zGAReYO<&!gD=YTAt$S8Dm9Tm;8sC#<`e+$YVwR}CY#>CpQLJcgyYPstrB4XT_#d`% z>Z+(9s-6)3X^1HE{~D`VczW8pEbhTQqsZ{R{c9@PzClqggF#-#8j1|K7=a5i>3La; zzgf^wBD08pPSs^e7p2;^BV{-d@|Pf~e;+cDqpUwB=Km$7o`)cssL~ij+bvjBIg&Fr zHbk{IU*B8#4tix(iZhXJYE}i>uFUBsiFuN;tDj$;IU`urJ}NC>MhN-uz=_s5M9pia z$sfhdEK!KM?0jcWF$9T{KuR-{nj)O&^yo7N8I~*gVf~&G4hex-uw@l2=lUHXXb=)1 z@?Ry-{I4-;&xGqQk0E8T60@)P#q1L4vz{@n5YAGFNF}sBg5WMgVA{@#^n&?lh--k| zG;Y&uymAyEnsjSxMZOeo6{s1LwE6~&ho~yIa$BaJE0@j;rXZT7U*6rb_QP*6QV;}V zh+w1u5pOpf&nJ)v+J?m)lYm)PByr6~sWO|%RXf+T<-9SYvO|e8LWhf3O%pH}HH%2C zT&EKPl?c|BNa01laF&3!2w;DG(-R|MHUgi(oodvRh(cg#LPRnI^Mz3B1&d?|ME-r$ zsIz@ae4sfm+FrbDZ%&J@Zc5VS`X`-9DWBoLq6qW?&nk#RD9l^)9~_Juf!C#=Hg_K| z4!vMtnzBM6YyIBRRz8LG!tuY zzi2_*tdR5bTbY}~JV9#;{l+pbU|5_RJF0z(S)%kx@S%ibNwoxloOqveZ4S0{y{C}Y zcQVH|u((eu-%cMtcXjX1n#UdxjF%jG45v#&VFJETejQ@!NWHPjQymrgLdiIxzOgR=VL*&x#P<7SznaTCy&FPcf%v zcWN>eWa7Q^_oFD1_0In{3I#y4iwB7xfj;%>H9d zW+b1b%}l&enMj^fK$WJsr&Yn->rPyEjT2*$t3#R^f6NtBvGk<-&ms0d%0s7%K$?Ox zVj9I#-Q9E~AFpufJ>dLkA2Nsx;^+Q?Qe4?~b9Jj~0MT=Oknkam_Ka}jAM)W7}X(B(YRmjZ#6CJC$vpjs8bXG`G8hlx`|#T27| z*mHUg`RY6vxnH6Ig7cot{lCBN$H#HE2&sXsA|2bsMk?gaRG$W;vVn%+;#%k)Gtjs` zAt@8$$j@ukWH2}?vYN!|ieAQ|a6w@TfLDArpluszOGuy-tF1aZ@9wQId<~f(IaUof zw`j`an>P1rkE%(?Te-|1{Hao?Sl-R~i-iSeIlTwl`W_d>VL^>PZDroo1nLT6Y=UGg zX0}1do`ab?6VX^@W=6x^JrTs@GTz1BkED-Vz80W{;+utntUfVx#y8fPnYz-JK#H;V zgnfTnRKN2|{oTuBm&UA*eeCo(pX=lIY^?J3-#sIKqo|>SI|CH-XSI+L#g9cgVx^+1 z3ZIhuid6-^#c|SEyp=DRA-N?@MEzM{t_$^zh^Ka$2uv4voj7!Jrfqx}U8sABSwyG( zuiND}KkVX@{Hi!*=3H?Do!Gr=l$6hUxE;{-Ar*TgRPWQ=Z+_s+l5sH8gQ<+34<`}e9(B5W823U zpscM#{lMa5pM)^IzW6)?1)_uUyCjzsx|SRiFTYar&USyq@JswZ{gFs8LW$N~@_&T*E+Z{V;bw zCPhL}2!(K7UjMgcVR_6#r%fu*PM_^;b*K`YxRRbvKFbsA_PYL43@n3v%Y-@&ixZ4D z=)KJFyj4c0o5!ik;r{Bdhh^ks`;W_8{7#RW8K7L&GN%RP`>e#~weg)DXMnM9uWr-C zAofk;sJ?nQ(TR zr)?3+8P~hw+Yl9uv+9M@>BOn|G5Rt?-W?EIQWd$TWz6j_!-;{@R|fF2`>4syTwKe@ ziGKMXho(NOp{HKK|8X63zIDols!%=RbG59tK8^%&P`&4}=sgj{q^aKKzcpy(_r*g@ zaQ#96S#CRhG{=PkNJGy0Yf*A?dZjhOia8$!U}D$U)OW`2#3ijHh~Gch)%>3HYb@OA zq3aH6__&qmycfL|QJZp!_NBxiN_HY@*=+o~#vA4{!8!IRx5`uc7MhH}Qh{;A5ArTT zmmt{Nz*hF^1XtaXGxwZIm`u&J76Fv*icM7aEVq-Z(*LZDR0s)S zk-&YWv}^CbByms9zX^K%)imrdFWkSlO=l@)Pt_r=`w#rCcKNXOm3p!WTy`)XRB7OB ze&g4^?^2h6@wH7GgK4SVAEO0+lVa5##I?4zcD(04In$)7wzq!d`JDWh2Ue-rHheVz zc9bce^YZXeptYswmod>ty*M**|g=t{?)Us`#^iCJ5fu~ZyAz_Rh~sqIHGzk zuU4D!oz+_Y;@qF5CLcj~9P=#3;9|D9c_Kg^>TzY39nz( zrkt@da*H9a$2efx#@koO6#s%jqJ@1%Y7N^@gnb7sKv@KvK(AFGsZgJAEjJT-vIoBS zE>3)*#Y7B$-z_n=gmldi-en!gHq@sEA9SF}x!l@SH*JbEh5d2b_Zq&b@cAV2{?V&f zkwF_h6q)z%1;}JA_FX|^eDHQd$Gz*ry){%a52|kp1&G>iKX|mA1QHTz!LGmun3Sg0 zdMVINN*0q0_X~@7ifrRB0<7%wu0&y7V(UD$)YhEyEks#RdMYIcm1NG`NbcXEvFMrr z8;-nYq5tnMrGA9~U1Zzl7MFna9T!{OHB7h-CAZa`v@cYlF^^}+)8;0^#4K9(F{z>t zBPKV*s%x;)25O?WFxxfGZ5%7HJrDE|8v=~jOOO$!SV%E}o zE3Z>b`{laW07q_be$?$}#p}f+5V4`iFH;Gz&6Kw-jSvy+$C&%f5tMBgoLnnr*~Ic~ znfsm8bZh9u>|~s^~3*3`M-2;5rxEc4|!i zgdGTp=W<^Wl#-+84!<7EMg+6AJb&-TpTn090g$RpE(uT(FBjhT)xAq_G zuK_oBu|!FCY54*neG+v{*{fA&3+(2&0x7sr5Od;2v8%~N1Q1|#1SijNh!((*GGr`r zu{Au3{T|8eDhq5ysg?%RxVX6X;>C-?26uwaiNgIGY|5U|x#W>df?1H$FV@SUFLMXr z7iSUz_f)6S!DG73-@tyV5Bu{W9FWJrrpdyo>X{B>43oiAS%NJL3yp>ALf%m{oHmckcNnHH!{GAg4^YjZ(dQ1xs?~ z`9%6kSm%|TKc0iUSi4F!`%^aQbeqa`5IEM+eq-)q8RdgwMkKqqmg^zUW^zZV|HG%d zLaLxQy+)FcYBF(9f-PT1f)z_gct7wZ2jyNCG0VnGK+VtiSeRw8F%7$4vAbN*>ZNm2 zQD9FI7jU+mF*Z`r-e-!5HsdL=pY^k0D4uV%Hs8^TqWEA3B(HK{#17H={k#*pAYbgl_qvRwBQZ7q%in6X zb8J3V)!o}jFHGD%6q$jRxiO|Swr$+YZ(KI!J{)hRGKOAkXP_^HTGwBXfj@so%i8}# zGLc*eu}n8neHb7JP&Zl3ktPctE|7Eh`fB{Hy#F4R{5rxZ7u&7l#yB$8w;e`bDplxn zDCjOxKD6GgnS*S*Vana#OJ^vKUm-Q#JxU}3sno#T9j;=cyM64zZDWQ!Yw4{0CHLXO z<|BtrjLDmEH+N`>tn7XV2h&Z~D`Ph)#CA1cDn(Me>m}_z6=n6UN1k+On>p;?-)0hD zKiA#nO8|r51v z*%bYUo?_AW3CUhwUfLbhyo!x|dvu#3D;qHPm6+t%9Cjl;n-dN8#JW=-|kXj*hSIu8(5&;RHn3pgDqs_N**qJ z@GiLxzJY=6aQKEkrRC)%RJYgW+If0;Wdrr?-1wG}jrBiP_k<}sMgisFWm8g8Qv2ic z!jxSPX{4WLuw<}Ch)C)wF{T?fjNsQ4e%WSx{3rz%@r04VY|4#j^Qxkn>ID}DE?v5G z()Vf%89aLq8a96NWOrmFpQwi&+goPA9j41VnT({I(wR4}i}I{l+1wI(8|U7(-%&rh zVEpPeYxJt;`O1u4`A&6dx+*_sp1%Hd?Zr1$2%ql!YoBSrb@|i2KM!f~^5x58?dTsl zGTaf{^{C$~3YDg&rsF+dB0L$%tI`26#oDE%rK!-J&#>!%eN$WtQE||yQGK^<+lIA8 z4>>tG-sin#g42EqS@EMZeOTeWeSMj;c?zVH-}7l+O6~U=Rrkn9Y3H`4SnaIw3JQZL zR18Z;yE-7>zWKE1c<-5}NfuSPxw)x~V(L{(p4Q2lzi?sqgoK1kWp);4UVoZ9loLCY z#Xal4wYL!b%szeoJpOI{Kk4hQ@N{2gXU|6(vc8jX?-8wsZkmhx`x$EL`Sa)VxYD)F z_1l75^!4>?zobd5->_i|J4y`)oWAlkHsJ2v*{p-yQ=T;@*79YTT56M;tmK~tcXq7t zurtYNI`sH}aP#x{_Nb=sgeBj`&bAKH?w|uB-#BgBw7}rN(ORzRM#jdUtE;_;S!Y^| z1D=M5g&FU-@4%o(ymPZ6jBYmRPI}TLg~skflBK#lgqayMe0cAP6DKO~dM}AV{q$@Z zvm;?)V}SOXEiI3s(=a1X81_5hd$i83KZ82>ky^Z!p)-3gFZ9T5e?><}*}i^v6}orp z_M^6Baf<(n5M`^w)EjyY)@N00CzRxxu5IlzB~<;+mM1+!H4YA?ZuZ+z_;*0S(7O8i zlkKJ(igITuEAN4!Ky~&9@9)3Z*h7~uPc~n%q4%)`)7Sskg_3o!{tl$a@fmHG!d3*B z4_vx%{`?DX=Kqt^(#k19EZ`wi!`Mzwh(&Ie_o-l zuS4mwTGc%#Wt`-Lt#?(sy12MNi3S`<_CIDie}31fsHk|f`FS^UQWp4W+L$D}dOd%> zB;}`P<@zghP_BUQv^zLJO5)v4R8qS2CRFIOZ5cyuYHhtp6T|S=oPeJh)_MPpKd0&P zr>d$d3Zo6m7rdtq(Rjy#u%1Bgbz7b5rscdj!fC~a;Vo|?(d;6#Jlsu)?w&2DyBacs2Q1InN zRFG7^eEFjN?%g|!-Mi(o@9HHh!L*CbPkD9}Mq(ySvC)AMgxix`tawoVhYodJ6K!}b zviW@tmR(QxA6XHuc?p$(ncK7}Q^w7hG2=yA+7@T;2^>cHIlCewBeiF2ye;xVHnqs} zM~~+3D6c6@bn+*_re$P!-Y`v;`?+>|a>S)eI?ylY>2%~E@X=kjE|5F1F`^KNbm`-M zF?u)wiNPJQIP6luEJa0G+g;M|{QV-YU0X&WJ;A}j;pVMd6BjOAh(KJbx-h|?w6M5m z&z{2YHl<+xpg|grKmH~fY{&dkEZ<~OYbw98QC10W+}jwBATO7y`1+MXQgWh$Re^JW z&a!2H@k3A6R>s}E8?Zg(-#2eoGb=J2?W9s&ubw^qR!)foU*&)}#QxhiXghMUv&EcR z4xujXH1X3a5UxlL95@d;?)0fs z^O$;=G+@AhjkmV;tEj9*gxfy6RY^%HnjmXoWu=QFgbe*Sc9LFTmWlruZ)$4#{=UOZ zR*}1N5f*VrJzt72Xy@QCY}0o_)tgIt!%{k6VP6Bho&}_5WDp@IrF1~LpKi@}ZEf@# zsl0vOjvYH9uU<`e+!;s8S|y5#!UzS@A+lKQF3%r7K3qFHTvu0@*|0Nazn+sqUXe-l zPK$!nW5zG(3)HF&9qk9RYpkrS_M)m7TuS6wLO3+5O5R&BxKEF6m|yf@V($!k)2~D= zOt4k}=J)>kZp_H|+^Mp%Ij}t=TA4Fh*zM5YC{veFR7c=?)6&tdhQ@xCqoYc~{rj)M za!WRB2(AryseNqIj8UUT>D_MsN?+=Qb3lL&u}2j15i>IBf9Sz?ntz!*L3lrTTwm)} z%0u8w>VZmq|2@a@>bSyRV{!a(9K6e_WtxeJ$)vl=Y(?A^z3J{q0JVxE6 z<+OMuEiEl#DJ_*%FWydChEVkz8X87`nWAg$;pY}JijZiZD@#OUc`X= zs#9Q&2tS)zX#IUBN7=cY5%Kv!Xcv$H`G<#(MX&4yqXZ&5k@(<&@VX{)EJqy~dH8U* zl=A;>-=6+cvq>!{(yt9K1K1iM^2wUeT9<4V%M(BJgqH{j}G2sb3Y;B%yBg? zY&7z8&-P@$;9y-86*7nB&Etn$f_)J;86MB;%&&Fi2I8IFV@8#}+U^H^ZQjd}SEMq4T7Knou`t#0X!JsOaX`7^+cz;5 zSV8T{N79EPia^Si4TcnpS|g*P7Qs55ev$0f>5A#MU;|$BOty1RBQIWDNW6*Qe~QJ6 zX>Z<~2DRyJ*s#H~a9Q0x+CzI`ErEiduiIxFo$$~&f{xmtDnYIfv23!Xr6p5}3XHsm zlNlEAR431xH4BTPE*Q-C1qKeIALz~3eBShvp)cRWM0Kjp0I10&bVr9V)ags)w1Psz zWowB|QPQMdy?Pn0T6M9eC|Rkl{L7b9(a}@k@Z8(l+NMmLcm!(7irPzB^3Sx)O#RjS znQh(+Hb|)vC`o+vUo}OSPoI{=9P2X2<5W_TYiwgZG{RS?v#nL-bBCVr^px(`uOIMt z2%+aSv!x&o7jFLd->EAPw%XX3e!F#B zonh}ZO7Mu9a~Cg8EJ<9$Rp;Ccy}zv^i+Z{1T+9}541HYu(x6URuN85`T7ph0LbMP(LY`W zSv!vC6dYO(?<7d69Utz^`jF%ja{SMFoJlklYpG)_oLyXWmoN8b2i|;|g_jV=i*QHc zcm=`!l?!p3K3mS4HX1A*5R8?7v z3&Hi}9*v?Az_{qu7h@(*@Q!G6Fwoo#yA>H7t(%{pKk3Epy#Ef>CUD$!V$HIBKupk7 z|7z!B;=S~(L$JU9LJ@~Jso8Xw7JmQ!J@WG9#R7g%y}UwOYK7Wens2lwdFY^Q)MZOn zt@20eyp)163ev=Yvn7a-+bEz!hOi<;zjdoBE?z(uK6B-okIw+8X%FA&vweL2^9Pytbs3muV0_R ze%sU3bXMe!29Be6^K_NG_zKS)r|a$At4x;i4FQVNNU=a zkZ;RK9-eLdts+_T@y3&^Up#+)p00uTNK>awk=&^@)A-(L$e66>E3^e&Y+zEw(`caR zG>d%&1QhFBr3dz2`enn_tmR&3;U z4jRPddffW42RFz^`zM=i*&+_qjCz(R$BBF?F2~!TSY%A6Lw3^Ryad5?3jD|&SMpvH z?|qCOkKWT>(2b9P2FdmLxi-Ex`@f~@9Z_Hk*hJ0=p2oSZlHRt0O21-`&5 zWO`h+!BAQi)}| z^q9pM=j7x>FqVMF)%~|t!4eK)F${_{rWiNx+_^+p)(ex6>#L@w#>z+okJz&!n8GD1 zbo4*7h?p@()$S z)nM0Y)2UDTw*g zOLb&K|JYbtk6baGn*FdKG<5X0yx575XY!4g0ns!~^_ns>GEVe+9;R$%7~MT>(r0Lo z)x0~q@LrH8XAYdQTEBic|NCDfH)MogA?gjAjg5`1V7$%JMfg*bzds!}_<3|zBsa+c zTY-|Pd7X{bo4U=T=(Y^-AfXad26ah!v>Dhz;@}>e2V2(F)otgx^_cf>WIFvQeK5hy zxMdJwAPC*J34~>Xvk36=>`3?PqSM@TnIauWBnoIAvu5?`NZyv-tpEJ4{P%$>ZdH`2 ziOD%^S7yGI)`)n#e3Nd^f_OqM4CC-tprP@1hugg=j39D*n0F${UZSIOsqlRA_IW7BV*I+j6jB z8d?z1UNE?C5Fj!6_@Dk`ffy3ama4WBw6sj$l)ZiXR;+SaMd_q#VBjlG85!x?%UEI$ zXXWFyZ<+>lM7g4n9^Ao1V^~*#2n$q3lr11OH4m(SH4%i`Ii`MLU zUY;;_N}pcx{rmEN|NYlcKUDB$LPKSp!2e%5YsoET=!hj`;qg#HoK0U$RR<2y)Pp$J z2XMdPOstid49T?gOZSF`hV68tC%x!DYPKhO?3AX_{YENJ`AkV*@A0Ua@ES>zRqv-J z{~}!`9m?6&JI0*y_rFLE`|rSR6>70_=S&iQE<3w^_4{MXgT_@n?PEpVcNwB)xcXs#&-%t_yE!8%S>2=h&r?m6bK=#uMu6$hu)4wz>Xle;?dlT`))ZajZVc znXbF^#(5sq%oJ7C;NySxQsc}l{1y~ex&Ap3nuEFKd%YM~2Ukm<39nrL71XPYAyRB9 z?S}2N);!zEiZbMdy2eJiTg&$Bk*C;{=#sLpj>Ak4)0Xhio2aU~;mv?Ydy9Jv`W5?% z%ZX}fVQGN){8V8}ls%$ucit!H8@S%QJ$Bcw>cL@emaWGvkq91Gv5rgtp`usK^-k7q z-br3P&{#FXypw_I7wz6dKUY;P8!%SWqAFa$yeB_nY+!S$`aZoLEeBaa^Ak3u}@({lw_$M3HgNn zcNc~`J<=@H)jQX~LYV9*GRpRuH*Or!Bij5r#QLU)L-sdr+%Wv9o0p$&fc*JISk?Yd z54&=LId&@hzDWuZh;@Q6_Z_k#Fa;uD3B*-KW@a=rv!ba{N_zT9K&>v7uJ}NlI!@;L z9a+M(?AePUl9Hzc$=%nXZRy618>2hms83I+{Q7kXP$}@63pKOvg9jRYjy*Q4Jj797 z-5#OyrMz6ABXNyaC&N8ZwdvlOa_;t#+mm#Lo;-eHBEg&5dR&WoRaMm#_y%SbR#w^b z)rS-~tRBh=_(4vb9Ub!6wAptfrr-eD!+JslMMz*vd-J}VUNJEQU6CI4t#Cf+ik*7FkncY ziYhrA{z5P8;biZIhQ0KeRwpJ#z+GI=^<-cZQp)e z=GVP@_m>t`$OYfAfw&K5=5js@t zzVzmZ9;1eR@w9ashkQL`7zekut=T@fRZq<{T+`+J#fwioI$^ftXU)<9XGiXjymaX^ z^iJ>Iz2^bnfvpD2OhPpDf#ulNKC-^PzP6!aRpNo4f5K}=T<^Z=&W>ZGK;6=e^z
0R^b2_3FY5DnZ6#DtKkDihn3zZrgIIDr?)BU=At5KZXHQa7dtt8gJT2`IeJPGyHpgWB#*H^M zQ&C0HhB!pS+QPzO-kLQ-K=(R?m{%z9MkJm*kxqT&3vbJREHp7mdZP%Lmy?xNwu>&R?_V(_7?fUgC+qU&$v@beQ zhbrgEhYw@WzdZsr8mwH|W!J7<2-yeh-@o58W~b$Z88iCvUEMuA`ViJvZ%rVd7Qgz9ZBWb$&-D_Ol!9qPd~}^%N}~h@^eYa5pWa_J-~NV z7fSeXl7FWEY;A3wICW|tR6%A{Gf#rbHbeNx&z}7cuA&)q*`%iPw*6u-E-UBnrNU;<{D=E~-v6T8<9>Vl&wz2UM z>!XL-9CN?C0D)7N6)RRa9y~ZvR`ySj)5>-07Bfx*^4az##(;+_@vcKstQ5 zS0Dq$bSTws;u&!oxF6%kcO|Wq20#lvqHB}BtCzxf?hN1IDfP>L^n#;J9GnQa=D@|g z2M6cC8FZ&B7TltzRzxiL{Pk<^3oDMpf{(x$Dq$ZDM#u=nu7|+4c1m4fw80)8Ud3Mq0&y6b6#)YL|iLcqxeZ#LjpWsqB5Yg4+DJ#}>KlW?KtXX&L?1#{OCMwQ~yjOlH zGIE}P%Fd1AqghxS<=)6Q1}=gD5{+K#qNFhhXP%*>vMio)`1pTlRdKZ%L>PNWIc-@6 zxcleUEhWlAFGwEI?4(~N4&D|sv&S!7V4(jRq1c0&o_S1i$>RzQk0t@w>(&Kptx>yo z?>3C9r4c9r=^;)GAGa24Nc3Kyr`MOmjE$GZO3IVYpL)~%c_fdXJn0Tsyd4*#;L_cN z@rSsoDeba^z$MF<%f*dV(9#+U9~m@z*V&BX3Q9vW)}f0I6BdDgBm7E#x=EWngADZ~ z$=Doraee%r3(MM*SG`Fxd)-rVv4QsnL|8i(b4jJt9sg!7hPKghv+H2yfP$1Wry2N& zzkk03AhBe{ieb+-LPYqsBn%rm^dIyew|9Ry%L;EVXhs-nNyy#Or>5G}u*`n<95Q4G zyh68bhq{mvRh;UiDe|}M*l~gbc5BDjrk0iovt|w9a1MQ@jk3a>PUf38Z{(pE*H$ZO zX=!0xF%ok%c}2y?1kp?1p2(hrI0lzbl$P#6G!+5>UY_)8brbB*rVQmsc3U~TFZFD1 zQM!3)`379OHoY*}ZER9f5<6iRy1HtTn^rQ!EZ#|3S=mj;QA7No0gaNJN09fwg6ug! z&d#5Rh#UZ=x?z%%pZM8?pj2^zrX&2CGkWgc-ncRR{qMiB7wz{zWRgP04HRc8Ll}yo zP%=B?-fzHwR757C&kJ;z$Kir9ryfy7P5StoYIjM^l9~_q-MQY7R^%8d0PS{2QnLBl zt(sqxA5xcnuBh-}SS5nO%`$-OG!kLK@LyvguBA953!vwPKdWLoJk%%>RB2h~3o86U z@GGTd73h#X;4bHznuf<}5yyd{r5{_ldp`V?k73Ybla&ee|Lys_AXLqvKTzf|cif_Q z#sqFJ2PKu<_nCA$a`L#oaIq}A?~1Is2NJ1Qxw!(Bi;Z{hnxAl2w!jX@yR^W2r{U*M{u-a9n=UtdNT~n+ei@23HqAl_{;KX>u+#a2hET-O=Zx;3kBB%- zo2_Gjsc2sPKfhM!m?QibFM}}GwF|LBTo!GK$pDIzAjH~}v#^Hwe}BnGY~S)v#OQ^@ zd0W0gcJV!^)_B+UxmTa1cV_On)i&~_TTiN+Ya8#JQ0gROryHMHJamatt5+Y6&{O__ zYJK?JeSadr`kc{O@p^LB)?dR6ptX6ad&*E13};{c(_lA1ao)Xs)3&(|7~o^Ku5ha! zHx4kL5hh@kpKoYpJV`CPaP8 zz|;9SKT@x~6HAVPg9aHCPo6k&;f`%wj@Y+yuBI*%)zn%|@9!KtV1>?R;FhR$wz=LY zso*mLLcNu=x*#X*2d5p{yzctOKX^DjI2KL#@39@@;X+fne8 zT^Y2{z>+bEK|dzj9pp4WfBCXJqX^f9MB36yZaOjBP`bI_xif2e}f^h&8p}Iy-5& z8QS-U9WqpF4(&c*K-``W?#Xxxay^eLbpl&Q)ObEUsx@fEx(3ZQvdk75JfY3g7LL`- zP0&ahV5)AEdke6>nYLbNbAr^gC;ZYJI=Gkk@dEfFuHm11Uxr^j3L~(mEE(!JLG8}- zZZ}e2h_J>uWN33|rnx28a^RRT#uJK!f@W<@Me;fOPm`Db_zyDU$m{+sg@s0K@Qg}Y zD^>nJp+~{x!9U4r1KH5^D$NKF44nM_vG+81P^8@tgS9{^gv-!oB}K*c{o*B%`t&EA ztJX^lf>Kef?|y94;@FCD0~c@HI1<`ewZ8u6HECuk9{~d`%UZ{9UzYo;&@vGLcUuRP ze)iJMMz&xtXWDy8ofI~oD=X83i;9X&Z!b_ACM_dlH+Np5a=hy~FZ*{^gT;Q@UY8_}bzvsD)Q4z z+4M10!s65RjjCSpr7`|SFQf8C*#jE>cfHn15&nuB4%#D^Uaq#v#o2ie4b`kR;Ju1- zll(T?QStDYh-^93%HFKc>yY@oeS(F}&m4m$$UC zx6&;sYj8^0r}bBEyEGjQ&vxyjJfMY}HjN@)Un(07nvHbTr>=T}hEqGa^Q`?6UEN2- zh2LkcZ~X3Z`*!Pp|Eg6kayQjHb74UE%W99yik@da-?TA!_j5V`3*z^5aEx-3W?LP+ zVJeze72csld3<8$x`tTA7VeZ*NEITp3cU5qQgp&W_ffoXE)6yPq8=r(6M8GFfK6jWE_J#zkk(-{~fL}Uh4OkXjwl( zO6uH+N93hFh&u2H7!Ic?R7YDgg1XXN{tIUB=YzfUUqtX_HrT|v*Z>$K>R*g1NXf^l z3z~8t!Rj$H)E{W0l|affcy5qtYgyGmTB>D3A8}YM?DVm4NZ|(w7dT6Yc>t;Tg;(^JX z%fEEy&ivc`_pkDzM-1ToAgX%2ps^ywA>aB%iz(cv(dN}-JGYJS8` z5&UV2O~lPwna3kzut6@aqxAFVXcg!7mLaqdx_0S(?6FzF72r!TkgIp!z6%9l*IZ1i z^m9#(58CMxt|nag#32slaU??}%A!v#aiB76TN`JAw4Iw6r~5 z^z6|?z7cnUfdG2HD^;{waW=0c=pf9DR3et;V@Z$`SPt*F<;Fsx&E@lnwy)Cdw9%;UKls9 z_leTIYu2n0ygd4}o`BJ<`=rR`ywdKZi}?s(1lnfa4Qnq6p1O5WO7ck8;vWC|s{j4C z%S^}_aD!R;(pz_U@zSsq7jE2;yK7A%Fn2?X1XMnM_3E^{*5s%X4+-s~Gzk?t+?M_L z^=mN$VA^3;z!lGSD&iEhM(07ycden9%sb&SY7s?kCrHegrf(=cZCYQTsXIA0?tM2&J$YlZfzSdt=Iqlv@0xA(Ef@Si41X5 zV{Y0FKN9L;D8DJwrk$J-5)hD*$L#b;D&i%#?c$YH{cqn^2AK?oC@@PeD=pQ%vy;*7 zm(q25deopFlidc;t&6q1l-38n{tqg>W87OLh(;yULzFHH&;eTK6DNN^u@%v(LRa3c z%jwgnPoX%bG71U~j-9&#mEbG767OOH|L-|PbTZC%Y|`zMW9wX+iaTGTcR3F?d)M04 zHEDH%2ghpx<-`=prF;8FttLYNo6npCG6-4DcU>YNulUe!zD%J%*ZOo?0v3Bdkay`% z*Z2zH@jt)1bYYU>FdW?$vX@!<(CZ&F-xRU6i1x=~>wHRCrH={^Q&M{J04P}os|_q6 z8s;TBDNRg9dI8bsj`IBB`}g+>Z<4bI@4MZ(oZKPr(z4fMZLq~XKW$&FnKK7q_&8`& zK!t1{oer8ky0`7V0QiMicTorU_wlP&19_)r60(qv3jlN@nw>bObGX=#fx^&rU62(v zH8+da^xG~yk)Np_Z2SvgFSoDw91jKAUvt;~|GRMLbQ0f)D`1Aqk3*?6^*r)V`8jh^ zDQTVtu)#TgwJSkE-L1pOI-giLUPUFyH~IgQ8Rs1uOe`}@tztHGKIPN1fHPb7H3+7P zrAk5gxaQr23qSKh_&6D{qs%&=4v|tLwft}ZZSMgC+~`)8l9+2 zxk;qaCc2KFIn#%?Ie~KPpMU-_R1*~*f&16atbFrfULJu~e89ESxLqoV!+ zu$NHMT%njsuza19BS^-k(avLCo9PdC=MDAl-+!puq6vJTTDkM{J|7HHbC82i8h`hi zw8a0S>OA0j-rv9fMPyTUIEY9Y$*NGHqGVL0$ZE(wGP5ejXvz$cN>-5(vME}GlD#7; z3Q3w8_w(wU-~G7n^SJN-@Aq?5-#(xB`x>w7^?F@;7DZD(|9ectz;q4+fR|U?x+`Ww zoQrNZAO3eA7@Gsg0=nB*e|@}V%FIuh0cG>u|MyfI5stU=u9xOO`_u;S-m!oGU0+^B z7Y<1Z??SJzUOdAi{i=W~B{f)Efk8o0eCuqEh-ODv*mfpCOSASdqM*24&H}-Q4wUNF!x;7sKz1a7ke>6{U!}#xE=A`#n6>|C)d$s$ zdQ`Z?v>;*Xq)Eoi=9e-}noy}IeGwk9J&0@^^%i!sGFST_2uv8|`@un(P2~_Bf^x0} zJ?S=><^=x}Pd_GRP6`I{)qF=`ImyaW0E#oCLWb#8-e^{T&NUhD!@rVr9x_s;U1NCi>@-)yYu|{-k3~867S!?cf8LWWNhrT`W@*o7?u0I8^5KQ z!{vMiA3a`sOrBPqSY*jz`<8lS(T9pP&})+CA2cQK7jE!v1Q*8bO%)OA%bUynz3$|* zxo{vzwkl1$9b|0Fo`Yp>!hs)+0%Y zAZT;L#w$t}4)3_RO1kp6ZuVR*v{!aCvG*e(=%77cpJzI%rU9S^>xVD9d*g zmY!#*cJ9(8GdSum&olLQx^ZfYao2dttRB;%50i%&9@p<~q`>w^4Ze=!eOl+klpg_A z!ZvXCI?bT(QyDDmw)Cnj!e}@B65VLPRBf9nQ<8A)(#UXqiV!5K=U9B#GTkeFN4;Fu zwryLM+jMt|yEwL0lK~$mCO+YTT?ssK)g!dZcZffNfD`Eh&-|Hl zG_q0n&s#-hg-%pAhq-5xAXoi(d#_?;F>qICXz01v*dWcn(-RW9>pNVQXESB1`+
    ?7>4z}6e4TZu1}Iowew<&SwNJ~)1AJt5x#wJnT?IjSBP(ifz7M{jCv0q9OUEU;NT#vqge3B zFa#s>$rMl7dBEhor+NfMjrHu=T^AauWB_x`$17DR2n=QcUMRQ%-@be2lGUP7qr(I{ z+3+`?Ue3fsnAz}1bPH2bQl1d{WqaRK_%<1tLNn&$L|5^Dgh!so@+8~}c5y0kt01@? zWn}f-LtTDTHT`8hVp$UQ`yJw4eb6e)p4C)0k5Hz`q9a*pqOQ(XmT&-;dx^6$@;kIh z_dG-!W4(6mwWldbQW#R)?W(w0HmX~(n;Wx`?R?!)!+wGAJ4IK=_uLs8Izjr~in6$kNJ3y$_|reGc7{Hd znwko4E=e_oINR85*|G>)4Qm8sy{chi2!Zb_6SR=B1Q-r{S!-3*JMwyYUX;j5!Eru3 zqYR4XdqfsqgKwlx3#l@K%MjDd!x}=K-?S@ByQ!$a-+u!5@0_)L+qPrquO@$`D>}^x=5f2?;*3<Te3V&0gn3hhMALuU?#OF_+y7O{O#*Cp?+szxf zy>j??6JZ6jvnrA0_nfO{MDMobR{RT2b5-)+LDK*VrzIn5of?k-7gweZ?9-XtR zam{d}7$0@zM*>! zW020S>Jb#GkEr;Zv(7f%P)0UC0d;T^sOeYEzln)Zjp0XLwXBqyr_y(A3BzW2lu?|b zh*fTU<9A4OdU^o80XgjZXwWvxAhf4CwuG~;E>?G|{XM@I6wHc_+yB+g?S^MfC>~{f zi`~pWtc`xa1TS?%PrqUP0{skab0$85uK`%qrnBBsFD4z4IG-flDm6{f)~*a}H9Sdh z74|PTA}va)p}0{}x%VK9nKIqy5g9YRy*g0l)XWxw%jW#XD_U z-MDe%5uXe{uliiRBuW(LJY@v~#*3s&nG8Wk1Xxlis4DkE7|ri6levwV2!9}@!z!%z<%YiuIyle3u(Z)cD`8+|C%(3Y^UuzBNTk*;8!K}4o5RAP4+P_%5_dNTuH zbs$8Av55@+wT%??-*p(2-XTJ%bm-sWvrk^)Fs-~@YkeS(q(CnI>{@{U$47lioA?|e3(P_yDRN+Od z=g4b~rXaufA1csB?ctXS26%Zo}99Zb;`Lho8x!JYWJDBEEX!a=;gD@ey35X=j7ha@$XL z{e%1yqyN3JT{Io4vbkMSy>4yZ*^gU)FY$hY&8vou8r|k$OT7s&JoZLj>`?&TR-`fS z;NT8Sq&5@My26&5E9ficl13EvLpZD3(Q=<#&}2_CgozB6{qQW5p*oF5NpD%D5Dz4& zNhu=ADvGs1nHq#IH5GlrxQh#>u3?R4BeWW~`9v!1+IiD1*Hb{fi9y?BwZ3m(CQnJr z3{6XyfgQ^vj?m$lJ#oT>t$;2(+3hgk`K@d(j(6km!MD7h+Cl*WcbkO-qR->o`Cpnh z)z&tdYRmo)b1r2XTxhlJz*r-`?}3anfY$GM=5!iv3|kdsvjmPBIy+^>7HxA@A2s#x z@QAKtCNS^ql>ro?+nCWNym&EnPZ%e!@<_ZlZK`K! zX{j^kj?xYH!coVoshI_^luC{y&fNC)HKU5D^dBESf0mJRIGqh#9Ul|z69W(H{{Bc2 z+fmwL#*T$me^#%rEK@O_)8{hdK&x*WeLDi>Yrae0<0o@MPF`E&83?>(a!6y?EPDFo=4ec!tA-32BBvcrC*27WmsfPkYusPWGcH-lnmosO!eKDKrPo~rq zlu}yLcK8NvYd@ea>0|C-W9aKmClE;Rd<-8WTY|&Koiee zCFIaQL4x5p*3s-Vn}x!o>~Bm~2*Zo(lfh@|<8Y-VD_(Yf^;%3h6A#beC+WUvdmZDa zjTSgLZQ^{Lv+!y=0F}%7vY&NiGz~H*zl}^LnJ*Yq%rA`Njq|IX$mVOFw33Ys4Cn z8vYxSsC4eU4GhFTkcaL}npN-xPUhWId+Z-1o$&KZLEQD=V0&!4`?|8M)U?VX2h1>J-I&;W!j63dz|FPXKt z73GYFGdAw~*`~#dhz+!SrXaT>V)*v5Mm_ft<~_lmkKcEs;R(r~S&eVW5X2q}j+y$M|7?t`f+j;Cvi_1$Y~syp3?dnBvS z1;EBe2Kr#jzHLiCrL>?G8r`;k8{;`Oc|2%HP+LG3?>~K*U}OK`S_@A~ou4Ho$qY%b zZmAE6M2!ZS@rw7n4|*MCnooQdJv(K$#JoOz`kZqwbjQyvH-JU&|88OZx^)1!O2P^fJW-r*Dgr5ay3ymqdsOYb z@KJc3Wy_Zf&>uw@z2$!8j4O93S!PZsD+SC)hSIY0Kph=pZj4xb^mM!3puDEI$%pb%RyB{toUkJyEkSTb^oFdCP@@GiBRt@Y-WYMc#rL6u z2PZMbi=kwYWDh54DO@)&VDXfAaA;UNW_r6`pFMndBh2I3q+1y_khU5SA3-)%NJ}!d zITiGY_9`RJ(^CfSq@V<<4Iz&4o@e8P)Sey_&`x>8h!IhhpmP!Fbu^pa@k~lh^_y~} zf-x0*u7A)q?g$5Cq6TC zG|fRF0^f_6&S}M6zdlU1+YX(RhYg{_)~6n0#*6_53Hn*DrsgkTWA&g+ z0w{+p;`XlfBb+AEm*qdxNTm)OoinU;m4!?i1~>25RXw=C>-R8VLERY}zN}^j?)>=b zq_zOyy0hMWnF@}oVc|n1X8wD|eVB4c1;fvrnHg>QAK{6SAagg}wVk>7qnXcp-Eqb< z>4jgjeCpny5eLfKlb6nqhvUn_*uG}m*(isn1E3Fh5h(ZW?<*o1$g6UQ^XLp6b2Gyl`b_8ZHf>u>bcrCS&daT54Oxbq5o^{bC0p;mZA{|^Xp>RgDX@t7l4_B$N--p5$ zWBe@BR4AmM_#S%8bDGkO2hbinF{e&4XdN7ZrH#z`h*AxkHQNGfU_yxAMtlfvKYiL- zG6}l??w6HMPNa<(F!w?GN>Fb_YT%Zm+fKkem<@#u^mE_WnhmoRtsqg!{1i*w(c{K3 zQ?Gc|-`{;T(|`!~k{6}YBv1-MvKYnHTTnxFZ5rlg|HAmSdj&K&LHTIJ9L~&|=GU{^ zk2jZFkkOA-yz2Rksi!|VPWC-!1jyQHlDv~skS{Twfk$m9o_iI_gA>) zWtNv0^&$r97o9k9V!1(YLkM6(C);~3HT4cBT+uowgYTXiumvSrs$ScQR)rTYU%otV z*(^qg=)}iu=u9Oi<~^vcJCyT)OzM$%M{@{m)O(FK@d*rs=s8qp{^K5I)f8qK&gU44 z@I>}jbbvee0n35_qwZ%u%~xFUthx2&Rws@iSdjlJJxwG>U7$M1IFYyCp*(E(8uecQ zJf8?t8bA8OL19dsz?p9!=Z!B(hE^KB9sIn0#RDzIkC*X2&$xwIbws+7LMHHVWv#0? z%zCT7*~M(!`uH3e7=rLD>Rb`Gm#mTc^YLTrr!9Ik;bDD%fCpaVNI{4JmN8k-B`fD+ zR8yEnEfjQ%W`zBib)`#$PCy8jR-R-U0Q5SCYBH(k+=0iy3#29^hcDD9DG zFS;AOC|tqI7mjDRj;HEhf6dHu{f+bXw|CduQDeG%-tOmj2Y^q}`bQ`eizz3KG&JfX z$uK6xbSSU$BPEppFk^+0`M7Zj6j0KF$MvJ$*QSQflFApE(5 z-Lj?vNqJ#&IGg*0O^zP!jsFmKVh-pPYE$VX0Yt{2_!V|LoWI*VfBW$vWo;7REReB~)wFZuuzVH8TgQ#}>}&#hRw(sbxh`#&3hWeVvuKXAesqa%+i z<9g0;{>VlP6@BNb3REpS4<6KB;86K>VITkH5DI;EHF78~uLVEUczFFTJuE8~ z<5tjQ-4cF{>n|?_^1*t?hA=X!HKy96&hYFUcC(J2*BQ)3xeQyGm*{??&Fa;qs;B%~ zYs~RTVfG;NoX_d^?%f7M^{@I~n(6=O*|Q+6H3oF1j+-J_9x-cyKIyBI;rjhCHC#y= zVb7R2);*tpdY*xpa0=R&L-!IwL+d}&?GFuxvoYuGM))2>ea~(L{F$e^`6!pk;@eOi zo!K|7?+`m<5SV2w7#wHh$(UD#lvkYCh?E<7@?_gT2jMYeAHN*|(A^bdn$QU_Gi`uT zvJW_++(^6F1#MuNcoWvo9niDd%o#)!>qS!Oiu>?IV-xOJ?aKkEpeS;rJ`yscqziz! z+BeFa2xzF4G?&y7LLQaN$UC=Ix9{7##@~I~-fDONd>+Wx{y3RclNzC7pRW$^Yw)e+ zhjMk*3v{UUp~6J0$_u;I;QW`((O-(a7J7nQ(_xwH-gMBU-n%RX<>(iHu77 zn#0jBgOE92!8~qs%sn?s#y+uh3Y}NcztT~sbi&NwM1z9F4q$#<3GJ&OFkx?_B zwjNpZ_JXX$v(l6kuoI{_1wdijP#0Qpmz94*0F%j)<6klZ*4Bq|`OfDjmo|eOnoOscjOU-`$da(6UE%9zS15JY3o%s0h z#0aDF`E-=bRN5CuA5YEdXrl6Q81#bri;QfIrb*AY2-4ZXRVa=_k`V z;j=D?L%;*QQ$i9QUhy8<8vl<2ygMxyfqSmSP2%;4Ur>#_o zpsO21IBrTmTQkkpcL07)HjjYV5y8JvG+w;~szv`xQ*5QU$o8B=)}rTg+a{JD;OkgtaPBAfC5$qz{;rNduRUbCj$+Nk?{gYkZm z*2fw%Yz^4If4^fr3y9sU2Oed3!(Jk6{>As7=`yN$V#d!x!tNk3(FUH0S z6NZ`FVeVjvgA=%;I=D!L9h1~LbkVA(TbF4=r_vcX(5do94MQw4eGCo0A|Cp`@TIJp zM>GW2m!~~?G!`$=B#BQ3irIuA0n@?br$skC=zfh_D}fe|4K+9|4`og-W!Uzg~O86;G4Ps zl@QT9oLUY>diZG^RE|$HZV{d1p_N#SdA^}6Zuh~1O_T1|Qr2;vO~b^lqfmnEFtvtCbzyf83kG@LmHvAI|5Viej=iVu@s_5NJgy`)>c-B zfske5?vj;`kerufgAN4YxOlcv2|vO8r1!942V>&^>t#?$|EUf=rmiu4$Z^U&D{E_2 z?bXskot_p%)-Yje@fl4!z|kH)F^!-G5gS9aIV%Hc0Cq71gO!*;OC3eTV^9ps38$Lw zyiD6_CZb|}0%gRCS`UT`Gd`k5QKDbyf$;30L26ZT9ni&&6`CT!aKT2rD zUQ28po)C)c%K>e~0ZHt0XqE=8%$qD8R#k`?#J`22YEM{L|5%W=eex?(+Z?6&GC*!1 zc2;zfvWS9u)m zUEZyDZ}k_6Bf6k-HmT5pZrAy@CvS8N7*eO3>Jd~mLK<+bCXZzBKWg!58xV8XYNq zKZ(ICz+ip@++#Ha`ju>l>Xb`2U&^XbzMvINDF11^S7Sb`k| zS3b$mlM2p1$cTrCWP9=3D`HY4E=w!0oE1wk29(yMTze6!Ac^&XduS?50O0#abhC2O zsaQ?OtPoz}^E%NFoS+@fptrx5lH$z*OLKmW&v|*UP%H?xw^QR#=IT-yJnIejmeh6( z#>)M8;dsMCf5&FyqnkI`c56|r^1ulVNZPUL1T#{J+p zxCfuH*bDZAkRm724H}?XX(02;YxLEwXLFf9QVQxWydEU<<}YB6A?1*9?tNVynD2hT zN8CJWB|Cr0$|m9N=voae_*Y@!#|O;ECX$3ha72O`GLNZDMCR$}Xp{8WFyZj$`aGy0 zkrB9Rzdt|ysRxdd*dzwj>Dr`8lWA9%X_%Ur+)eC)X$u@g>ND-S#>o2yF=+c$TpZ*J zZnq8ElF(F?YF#n0*;)Zw1~1ov z%0K}c7~q{KuI5hv&U>bvR%!Bfd(fKSUsUo}OLXI@=ntUtvt5c>{s-~%v?H5xKxFFpf} z6c~-}pg+fRvz6k2&YZdnJpqWBE~_Ybxnd8YJ9KC-#Z$Ik1P?FUi*=J)IFO_T8V>nZ ziHVz6=IU6@nH?P&ITi{e9MHS*@m@T8i6q?Tq}RxrEGb7K7KR@_Jd$^>-JwMMJpCKXp5X^oT za3n62tXsFP)`oN~t(qD3nap7CycV%ytLnwKyV~$nv!cGRl&$9k9T^1_cfH8k*ee4z zR8<5GtRbw0t~uU|s3<&2BNi-3du5vTD}vb!&{y2eyG# zu1gd9l<}5TpP@_+xuUl$dv4ukpnvEgO`P?9B%C zHY39Ds>aTkaY!NG%=HDPP^3*Bptp-pm>?EBKngU}b?!o@W1ZM}Aj03vU&#y6`gf;THc7K@Un);t^ zogKYm#fpZCzINxlVvP_yFbdc7dr`mc#xH6R0!H@qj{rmSR>+}@ii{K=NwN;-+|Tm$ z+cq~v_-9Q#^nW&xKDMs05LaJqm~PFB&c<@AWqRD=EG3()tSoUG=AxPm$F0B9WzgnxJ83SyRfFPIsOKX_!ymzogP3UKJ) zQZ0v;*NfN7iKcz;%G}1x5JCl>o+`05{vQ1c#ypRyi&%AQzTtY5pbOYuskxUm9xXA# z9W%GpQ7B81$TN^pk5SMXtXjd?F1YWlJ$v_>$>eU^Na+`3(@@Ut&suA{c);V;Ea{z> z&_8b?rc@&qdb)f`SiNkS)^qQ_7cH_xUV$TF@7?A*{~HF(^?<|u=Oc1uRVTHf=2u2cgu;$#jP4pH_szj;)Yu^6iY4wgclOh}S=dWi0 ztCuk{03g@lqKODE7O%3uxPF2VF(mM#vYoeZp%b|IegMSDZ+cgyp^hPQE91M~iGk@q z5fSPvhw+EdH7@qpvjS-;@$cBkJP45yoMtZyCPyZa47LpyiaE6Dis8d z|M{EgUQEfUQ>PO6MzFAq8#TL3tkp=$)e^E zK%#05Sm?Wn&>Q`}^O}lNr*U<2?uigzzmVsZ;0986xPZPMwraW4C`N z1~sst(<%{x)k>bjcJA!S)@M^uz?Sri za&vRV+G(ICy&ICH_EljbNa0ryf&e9WW1<9#mQFBKnGBbD)IS?$B&pZTSXTm z-VmJ{Dx>`v*b^fYU!O#FbX+ZMjhs8Mx=7lk&D*~*2ICR;|y{Fz=giQw9@O`H0V z@r9~GCAsOq0k!<)qB|9{4d!z#J9TQxZ_gJBbgO$W3532GcRh*H^C$8#klF|$l2~6@ zhMJ?T*#x%(r1U4uVET`6yd1?4SZc%9XOA9LB7DgCgGv_9Z45`qz!cL71jKrAz3O4C zFEVC~D2DEWv|!;PJOWffli#eUnLrmqpD*StGD{Q+fuW(8Hnbn^u7l#^;o6Q%8IaHY zA739J1~03u2k9$^AA+yqCzR zsBRD$Xe~{cn{vm+xqrYfTwoi5DR1S)(s^~OxH;S}l}2xArZ)}*efgDVb9vhjoZFOt zobx;}0pam%Aw&z~PGs0oVg(^$aiRUeO`=l1_jHyCQkS#%qYLeboO6dsIeCT?)IiW* z8m4n4Q@0vc!Xy*Hl^%*~(Z?O0xIv!DM<;QgsH`hcBJRM>q0nnk%V5X1l=G%YoC6u6vgHKICxm8Ch#Wih&0JLX0| zV6KSjW+Q0ws4-)dDW3q2MuUq97mE+2^XoK$3FkA4xu8dlKzD=sK!t_`ISDU#AY(QTMqF|A3NBz;5g3V z%S($py8ah42pd`k00W!o=Bd-gBSce6%Mq!UaKgAP?XNNu4j{~;1v>GvAlqv4Uaq}p z^G+20#xO~!W)pgLR2b8tvCu<%_r@X@+U1Hd%bE*?kT29PI(XoK?q8b8N(|eD4+1)q zo!Gx+L)M~(cAQ-OBFV_m)mS$eA)>nfA*-yQo_FUvYD8HGlDy%ta5HRe z+rlaGs|}pm-O1vij(69_reuCxwsfS5Lkpd338pE6A0) z_AbZz;69z3-U4T>G}S!gk6L;T6V^|*wVnGm_^uM?Chd(rElkZ!B4Ur-&-TT}#vVR> zdPYtGf0$j=>-A`dJ1)~B8ye=uQ@`1$B1oICye_{kLZz$+XCoqn@t9T#9vpJnytuen z|J{5Oy7VXS3Q_E$qD>8+Wp4g;d(%;L&E=(-Dd~{6U@-5Ay?^&^zEaVx(Or*y$4AQeG1_ILlc5$iC4f9Ry*{jzl z0$)HSZvggQcw%;Xh<<{5lN`}Nzg-H)-;i6}q5S&F{00myd{Zm_{DFWMwlj{DJ`ZnD z6Dn&bw}SU0%sdz=uV5u;#CfVxsZqMek+LyG(C7yfZKdpZF-T#s*ki)45oB7tr~)cw zyiXF=s8zKyHW~2i#mOm6L>?Ns$R*PC*hmwZ(eoU((l^OqiGP*E)N;5D_C9wpc!YxH=!KG#cM!GZcfgfq@?=fv}FGOQXJ1&mGhCP zNeSr2k77di;AIO1>L775Q2FbxKmgKOLL~z_f>*i$gp$rA*@Mqti*#Vhf}LFVXX`}B-n$71 z4fVm^S4grNurF?|J=#Nda>h%2AZ{CPFE4orrIyzIKLPj)Oyl~2T**;`x}ku);VpysTBLWW4dTAvZ1w86$&Mgkq$f?s(_Dt z&!nCcSSvqDtA+>rzlu^u#GG;QKO4Tj?3~?_XOx~`n%0LXbG=XWVe+p8r;5`ucBLF2 zpFV$n-p_*j1Ee#O{7ylmkN+`{4-{^Q#7G9dPSN zrc|Ke{+*v>WX!qvWE3^ec9=>T&F3tVL1Dy&^+ ze`As2kR$0P&?E|l7u2Jq^h?I3m>N#L>K}Zfl3vC!ukfwn0J@59*gsJe5Xm0Li?I+m zP^eEM)uUvXfq`xb(oti_I9w>5YPh@_RF-Q?h56`&R{$)?5);rX0uV+7uEtiG3%~x9m+!>GQy?Pl2VZQnBASl} zKAD!9-kRS`7Pq*HH-a_Yx@$FOOjMG3MR*15lL9i>hy_-5BR3~rk1`f~F|7UG6_W`=)X_)I;6eKq=xgYGBKgm3`iFyRE%7>fGTv#s((ZSi4gM}Y|dRT zXQ}bx1lbek45F4b)mxGQgTtfDM$Z3grLyg>5eI5vO@KSUyk{hbDAHzh6+(8wQ(5C8 z-O}1yjJ>QW^+nh#bYOn~sIk;lvayIC-KO)xC&!x|sa!rEa_l4uX<2l@qqL~I(RlJ% zdI%AZ&dH{6I}FCTdy-}kEWlw1t3EqL8hs2pZV#SM0X;zhNp1n6*KU3*nTuN>kY`|) zp2`%v4e(J92qUbdZeCoF8Rp|d4CChc@%K_o`i`|k@shuG>R+Dfp_eXS?geEutrFo( zz^^1r3a}|LYc&_v;^VwXV_sm+gG>r@tLpt?jv2{03n?ep%d(K!l{EyUFQ9%hM?4_h zo{)M69S&bvo3Qd`*2k!Mcdxam0rpxJNh_}IIbVMp$NR$C7W^i-C685%wX%xY9A=*B zGM{#RC&h0g1)yZm&*}E|1Aa}xQH{s1BQapKV=aIttKN)|GbMWbXRXFX9#GP*`0Kd3 z;QkgUEds!c1cru;Z5=&RUr+B5f!w~l+;Dv%q{m%gn_QZIQOHxYoR7$pcIE#fs0Wz) zhy(OdD6k7w*X|*JpUb!>IaO!mYrC=5wa*0qU*-u-`0v!y_ujt`t#yGjUDNjrB2Y_m zjcnKea9bH|$$(aNEKpKRQ_s{{SF*_VY!s>v88Ad`pSq%~GrFb1FB#q04%L8;I)QiV za;vi~)-@ahZ(t)?^mmShGUaXY3Q0E;<_UseLfy2SH&a;cS%-k;&X8i!;>GrVe=?*z zPu8ADG&gs+1Xcp)*yXe9wNW?2!e;qS8I^R~Ie!`aj^tBDy20Ov;Pf7#HrQ|HEZ%3CMJH}f!`>Y?OC<7s~5mx+qy zts_B)L#xubAVM)hl2#Yqo(WZsNs8>u1c>b3uiu2c(|4eL1jv?3dOZOSgepfoLv^=% z2r560RNqUZy0>Do_@9)?k?{^AkVGDYY08*@_8ZYO&gRdo%j?Xh1Q5Sp#Ig07HQVNR z>Vg!#?8J+58I5bSqK9Q%Pi2ch2pTnZQJhu-(ygc-`r&N8I1Sy}$6@MOfbPu(oY&ApCs%AW=V}5Cw{4F*>b?cQoaoo61E7r}#cWf&; zippHtSq?@M_;9+#!m*J90R>f>@Zi5MgyQ&&JMZoH@BgAClV-*_^oc-sxSjh`x*g$_ zL|=BBb>7)LW>shpo$|L-C+$;4Ff@w>F>@+5KiaWEpe;{Avggz#q z5_Cy*U+1Aq9AezZtM-^-uc}ZRy5vEb8U)~hT3u9CtsOP&+;py{Q9Qt`=-J(9_u`WT zPNXUyZt(8;jEs7=!0Za=%q7`Y#a_NVMl{B#1f!~?86f=$XB-vAt(U1jjJ!rMYeQr- z34vZzB^9dJ%z(UYlAe&FAIW%wNoB8CUt5QQN;I%XJp@;dr)r5=_@W;@*{zqWqcz_n zPrv%K8ptsoK0Pm1XTX5otHyu-{#_0cL&34%jSwE;23wf(up{MN>-+bcfdEsqVllh# zT3-PyIjmWNMC^xyt&qjov5p&i_cLIm9i#1CT5b1^|}d z_cVyh>osapSShSpf zzrGk{H+{Nl8L_SMU~(#6G}{W7)U7!lqkC)sV|{F_yYQ3#Ie&gx%2InfAX?{p~9oEhT zHrd~F?2ZP+U9;PrZE$wcyV-!bVbQ`xbLQNR%ET3rZy8d~@>1u}_{t!lbB|nds2x5W zlyi*~8Vxp+mzy9$u~yzh2|0#$V6v&9M=9wBY5AnQ**gE98t9qSU{P;PMW2lNO-DyZ zbLXBuYpWJjF%TaGQdFmPO#ov<@v5P#qdzhKzIw$9R+!FcWc0J9VC3tg{*>&1+cD_R zmQ_98wP%kKG&#|VgKj~a@V2a?cMlRRVpnv)+!ir-8rP%sQE{xNP*gO&>`!;wf93sR zQf@r$H$U2O!vw<1D8_x_^G2JAr_%eI!`k+-i}w|GS*Ud>z!!Ec`f2);%f#zS)~@*S zjatB^(sc{k3I={6W(QNo+?eNQ>s`7U54@05Z>EQ->`krP(Gu!XR;3!dFLv9+QD>d- zxD`4;o@H5Xn5W5$>KE=AJZ%stQS%?EGc@iGAKRZGE9!tQ{Vj&MWicL__vH9CP49mB z?n9Ah!F`ez=EjZnQPo@ni|@*fjgBBa@%eQoE;J!EwK1k<>UNm) z0Zufw6AELBV3CmQVVG)+PLT#q1ntjrKYxtsLx~bMsD}Ore{rB%oIftCJ;b<$@)69w zmF~gG)){|(QrlBQ4EPBbAqpQn+AHT<;XlqU9f5$rry(CuB)u3BkFzxR_~+-(-7LQk(Hge~LSaaXZZ=x}KwDzF!31cvz@$KZ7or-L`28Y2&%})*vwl$Q`)z zibo5FKc2Y+?YsByZ=%|LF!j8W7gQ})FhK7dlh+|VTl8COT}Dfm>Ky*M0H=EKw1*TF zQvVrPP*?RdgWF(Q6xxkwDS``j)NOrc$KH^Trr?ni@?;nX5#unHM@4O%Hs1L#gfFJf z46sddt~)E;qE@i5^j<;OI1wzQ=S?gdPlMCkUIfd0OZz-UJ^QkaMnxT}4oo^d?}Z+W z99eU^S-IT{uihNfkY*7|jfXY5;y$4@rdEC^($XsVYyq~nT>Ol?eppmLPMzniUo|u8 z)T!lPueUGS!9IIXU8o4GtYLd^A6hHQ)-c(6pXLAb#lh6|Ck6x#&a*&#t{X0>a+9PCqHL zP3HLR3!J+(GF_v=`I_+hB-ND5+_~x+8m7nBXt)(Rf{^69Uz`{3F~8b#-Hn60cGac7 zvN}}pM4wg1ootu5d4xMU)azGexwB_m$2-`W5&QJ6Zf}Pl%jxj&c>5)49}GhqH10Kh zS;owYXX!3AUG{y~2>+=zu$XFS4!KsmnqK8|F~t$UG!9CJB!pvosRu%y*2@9Q?e6Iq zU@YVqkAAb_=sD?T^u9Zl1!3kCwrkfvRB^aQHkBOo)5#^Vdwz^AQzmisX*Q2EGp=;` zJ8x!n+I}_FzXvn995Zv~`?!jM6_>L+9rfL^zv61ZiH&zwHQ)DL6DA80}e2LbePu z>@1{NekIS^uAH23HpNoV zO6G=w8C}V^ktD47v6ng$;61TM{5iW>G%Afh=IvNiNo&WHOvc5e@ydmaQUi!6W^cag z7P`)z3-;MA`s{5gR?|{Td(2;eL)PLwWY?y(ND~lcBKc8RVpipJC1R?l`sc`zBLir0tVXmN z^B=q%Bk~Ga=K(8BLrW&kq$A%d#n8@yLUX__2)s%9Iz2r-ACLYOz1KF74T8HT^ro)& z!>WU2O+LJi36-?UV$Sb+eF83ZazI2aDuvI?o6P;wc$mhe%umJG7$8K>2T(L1M)%#f z&&y~US0*uDLm9x(3Qy3E)UX<>8traNNE-y4YuATmE&YRM{GpS&otAcHKOIC4K=NM8 zN?sOj%XLXuCSenS8_uC5v8&6|$E+VTCQFreiCF^DK9-9KK-A99ftHtg|dj zIkg6hg0+cJM$`}B%|UBZwyW0=2*wp7D!l#V$?oF8B}GM}8H_eEimz7d(X*!`eFXLZ zJ0fpDojNkZxY}g+P27v=u+6d@%F*{q(e7IXuzZL9ntb3YnZmh=0T;oTXzb)$U_WSkt|fYS9Slj3zV_-Cuyiqo*I34|f4yu(3ORJhW$hA9 zPxBWW%77-40ZL4&CY(P`J@bhB%3xEkP{N*A>N7+UAFQ=q3{^hDv=`b431zmkvl7Hx z@pho((sfsE6W2n<@#t~3(8)%6-jq4(jf#%R8Wrf_T@4JTMf?Jm4=8%? zusz70**JSbx)$1XfW~TeefShemBWA4#Y?#MYb;bk0$ew@L>`hsjhqY_v?It|NH&AP^@77jD+dcuz4 zd6@TsAsKUm_5=stOZ9}=QrB3(VtUd?V6yXI$jDA?vW~25wXLQ~+YM(fc>jI?h~4vO zQiKgJE4kS^2auwsT|>Kaj}nD6a2L$UmaVbqGi4o(ib9}jZQ=E-{P8weHm_nfCzNRU zxSX&Zs1;~i(cafNwY<-)TOl%vKRu6&A}R=Uk*xvucfx-busf)RcLkamMK3R(VcN~L zg9XYa96?}x0`_5F&VuBzd~{)@k{^#y$Ywyih$IXsd0}Q5FFms0X=Jt6zNTj zp*CnJoGG$XCz|{&&Gc-gAXe{=3}s#1W`W*-Fet2Ur(z=biXb5?hVhj&#}$E2)(EWr z2<@+xnAO1?jIIPIl%>W*;&3Xo*C;M&qXUFJ4xtOcdJMl-blRMS3&#F-X-B zQkD&_F?KID5wHM_!wCr1)26A7KV%TZb1()*I!gB5je|^fZtl46=}^si+>a=*dK(xV zV9sSOXOF^G_Fq$6TG5)EDi!X;0{lWmX8}qQ$(uk5f(Oy#M~}h@0Ox-(2nmvHa2mo` zXAPxYxkn1)55*+%mXKnjjkaepM(SUuD+QGpIF49fwxhTxe=0PPh6>qQ&mi?H-AMGU zfpNPJ9U4zrt>gu;IS&SF^c%alBRoBYIM&IkUq65U-BChLSMs9vtF^);(gjz%I6WQ4 zFGaUZk$?UEy(dbhg>vu}cWI)a(Ii*Auhd3*aWp#GQV|LA8L@6hAKczfkj2k0G1@1V zbzhLChFoD$e0ORZ`Eeo%03-5)ML#a_3FBRBVV~Z^JPE$%M4mU}gIrY)MzK*<9Dxxp-;L~@Z(3n*f5_j4D|;w7nHTFJd#sSL(Cu`Kdt6z!+~rg%vKmcEUKvh-4#{Sfci6(}U#)!gH&=@OUOohULh@!4;UX*jBU0h@nmoc|C^f_4;xt zNeE*^Y@is#AjTGT2?FFDFo^}D2A`UbC*Po@ABBOOa4S%gY~Iv-lxMVev!blW0?2~Z%6 zJ7~x5j1j|N9teMT#`Q~)p)9K0NkwfA?HKQ7yYfSa4k4@eXV`5FtT-7cuQ4x{D772g z{QR{oSuGEze8`fwW6^96;z)}UgxQr1<+*_1(XyGu+0k(neC2TR0fl4)e{4m^BiEQX zCv^SjG;wlT0L6)4CIf)hNqr*U?NZI`u~eL^4ETVtPhO_Vsf`s zhAGq|yCI?odz`m1nmL4PHVx1)qJ)Gm(E*4O33w%f0hG|`gV~@!YCN@EcF{;WfZMs^ zQ*tx0OXWg~S?pNO)RayzFDFuiD0#udngFEXJmYWAY*ZVJRFu4sDbD8yqCM;wT#?IL zR>;n9_?@Uz#vHFk6eUwWhGSG#J;36hmw%Y`v6kM9n5j-zS`RTCt*?x&krB(<>KaoZ zifE^C)27@w;tQKtMU!^uP-k+GFVE3A$CDKkWDbtoQ|8SE_|fJ6xl7w~S4%0<8K zqjte$)ie*xiLB;Bkv{8=Fax8iOoEQI;D;wEc}ZrK3`Kdp1N1P8MY}Q+Gc*hhii2w9 z>g+rTV37IGPC!IU$W}+lo;ILJjT9p93JUWnNNAT57U>Gi)}2uigu%AYF2KkX@g1Ku z=0}sym~KMsjsld2p)`_DC4w^(6DC$Y5g|V>_TpQl6YB0ICvWA@egU|hOgpBoRryWw zD5Sg~S%M_F0`kHMq+N1nDOh$u@Z{4XE0~r@I!Z}p|CU?liLF#b1Eko29k*Org0piT zRn09{KGdmQ#?8TaN4q&CF0FuFW+DY4pXBi#XIx!odc1IuN~x0zVYZ=yoVktiezcU; zQE~91x~Vg3YkAuKg4yvjk!tQKe;MGXWaReGiT3FByDPG*nM6!nP(uPvs+fLh(Y?G8 zK1CM}^?%BEGlsP`r}@Ohw0#Hm<9OJOV8plE@wA#EpAr0Feq4}yR}x!o=Ewcf`j4*< zb%h1<%f1;uQS5XF5+WiwXY+~l?Waw9EMwhH!zeNNt{2qGuzwp5twB3>U=CFj&pBDk ztu@D<9{SkK6!P*;J*vyW{p|FVNp-xx8D?I znZ>lyI81{y7V!gxPMvUWc508G!ElZ1bVbl9!)mx+Tqm62PFojjwAAOpaKVVBy&xTe z;i*;~T)b>=trm-zYTEg$6iEM%tOdiPF`(*gXVhchuB)TdRe@Gouht)U*~*p2NS1ze z(>dLb`Ac>!hIH4u(JO4kI@>*7pCi}-pBcs?l9Av@N?sy`1mfLie_|ne#D~EH$NauJ zHOhL6%N!vA9J){|HZTQ);PQ_P>M? zC(=aGiYP?y2Ss`Ih#e<=;8(M8{7B@PXJ=*{q;A*Ry-vw$j%sz7XSK@s-&Xu}+G#AR z_xc(^Eyqu392~v2?+PF)kH<)0XPz2((!MW0rz3!{pFaHwrR7)H4jO?Rw)>xN0(knQ zuo-31@Q25@iBn{Z2j4HM3KgSUY$kUpOyVV8MOqUl_u;he3!zhCo{-t=dd5_yTFgH!YMK ze*i8l87gLhH!B$rA*~9jy49b(ECF@uW&!&nI9+g)T#UpknBV>BMX$w4I z5lc>-h;QpKKW&?2^*Lj zT^hXMUGUZ6rG{a-Hudh_zP+)HQI8&*DJFa9%v(U6ax^mXu9nx;OX&mEBN%*Ij~&}E z>_RWR#4mTbZRXH#U%MTrIE-9Y$+!yH(eCK>66&4b<>aYIcln2Q@5DboEI1Lt%e&L0Fr{?5KRRv|bQ|#HP zhv}|=b=rd+w=;5=PR+%nZ49Hn=yJdxQT%abyUEqzD)6evv&Si-pI1sJgp{v9P=g)U zt{K!31JjtC6UUBaAk}aWbmTfpPjTu`Lg}uHBOYciljY5WX2&aL?%E3CA}-im$CsXU zv{rv;PGfP|Z~>YzXqJI1;~lr4e3(ZiC5OB=4Of?aYGLbpDRbQ8k`?pk_dbO07;Nu* z;~p>NQZkg81jfD4!`{Awq*lNlcV9y+Ixe!J%`i@3BOc(AfMCBXd4WZLWjKG7805c2 zt~<*cMMzoKm{1_sYb{B+`%ESgnW1+eUq;fb3TeUs=r7ZK$&xd6xBMqtn@_pW<9k9F z>#?>~5(;#400yp@c`#9LU-aqIr-h#n?%%JI)9RT=7Z0=(ycugeC^PLG9n~qPGJrYS z7x^xTYvD=Ka3((Jn2YoY(Nap8{dclh&!2>_t@*WX4jbPe4Yx zE=!QS330czGD_BMU=+w)YrJr@6#Np#uoBx&ziniml1i7D!2t6f$WJ?MlxeOA(DvJv zX!tIe8g0p2%JjIy*hy^Ip`n>}9J1(x_Qh2QEn?PhubE=rs@KtzeD)q|aZrivKP31V zS9I{NSMTE&X3^5_ZzJZE{O>+@C}<`Edrk>VN18n4SuBm-?gIy=1>N?d9g#ik$?MhC z>($$&)owZhhQyCc?fN=XAI@C-%jfjz)A5Qrb7C<{T2OBZwcvx3d9V4trw02lyfH_O z8&J*@;cHH==xScWI=TKjd_gya&4(K)xhbbS^-fKC-z0jYF1Y_ zEhA1oJv>@jifsx{q;(N5%eHLUauIl#;Gj-i=076%y#7g(A|El2zf0~x>JmwzFbHro zMnBv)b?8;*T6a;RN{uH70KKyBHr5`E<4qk0U$&;4n=fi$tZ!lpSt>an)EkF~>8VJ7WgDsejq&Y>KyaWAybQPmBNvx|?E1-*_56}t_Ok%?F<7m;xf($3tK7pG^+Yt@4a|&*-ZJv_WqaN&G$+5JP1%Av3FaCfyW>F zZ)-JDE6QVNQexuMc@r;~_kqzdecSpW_cjplIA^w5SX<5CmOHy>5t`t+u_M1T?~OJ% zacIOkW^5rRJI#V`aO&hjfB~S2S;tO`^~KC~y2qudLmAYl)BW`HUW8Ix&l8l=Cx=Ad zBvwX;j;B0~8nVV2El%{wGa~@)c?dSyWwZ+RQ)k#%Sw%hWDT|KDa4}h{=^9_hui^pO z{_&385Q+$UcXafkpi6gYknW-59S}E)`~f~`iIJ(ul$WhueL@U2W7Yp~PX4#3LW&q@ z627rdW-%+0+(7>x#pfA>BrC=?39$GC#KN7Ja8eROL%*4r*HXJ*zvFa0JD{tAF!w~)5 zyu*pSr=wCu!DLNJKKqFCIFD*qib3NwgN=r4+-GD#q=&_VQZwkq)iDU2MYg4-@^hVw ztLqaGO92{OCn_r``4mr^`nJE;u0fm}GhSV*jyHz!Q4esC28PiT1izCgOr)M*rBK!xphBe!c{_Y4fDlHVk}`M@a7E-p#qk~<7r75y^j;L-12@i&x~ACD%qXP@0o z$3e61(N=Ql7H@AG_zLEDe|Y^$+n4ow;y44_p8fg-Vjw>e;|Zlbt8X=J)JWxxq1$4g zAV&5r*TjvaQ7m@t-@Er=AhQhYmjv#pePBwzw9OCbbqkstxBra8)lr6qhEoFjJ6zcz zz1#jUz%AN3wR*Gy!}$Ra5j0 zmwmpwAbr;VVe7r)dfxy4|Cg+cgCfd`lu=e@R!C{s^s3BK_Bi%ViPDfLB92ihQjuhj zh8Bs4L>#5aK1rxZ{jT@Czu({Gce(ujd4E2~IqLO#J|E+LzpeW|;v7c-Wk#;9qIQgq zi^e7s75x*`HhoD}A73UgM6OeZ`8Xfh?1~)u!&DF%o}Xcg<@aV@LK%SI`41LfhGq6Y z9Ts49j_3IfEC+e-7)8pHfx)t~Al zaw-MPXC$!@19lBhHJ-}bmv|P66*HG(ASR6hEq?#r^*L<@P>8|u$wbuIY`%AN*g*|@ zefh{o1eu(0Z1~j-a&&(^;?UJBBbvt)B0=f50)2X3S#Fq5#x~`UA+*i2K3957L9rAS z?W=rL>G6M(GE1c1Hd#0PIx6Vo~i!0=Zp$+h!~U zywYYX+05-IC(jp_)9d6VtU>*6;}*>M6nNl30LlD;elbu={s^p{;-zT12&OAbRa$x2 z3dgtO@{V=YFZtWr|H1vPi_YpxP7dXpszMyvpQ^C_30^Fq1U-Jr{eu@qOt3yyKpg8Z zl|+61d#e2)ggL4TNP>x%e9w!F5q3!0zQjq1FlD-4+i_bf z29C=vTgH3HNs9!K+KFj0HK?Y)<9zucNsdlU8g-$ZGBxE#yuT_ zFSZvRKI+!sTG25p-;`@SI$%Upc9vJ*)y?1aV;1eWa&3)sjQv2Hop;~(+{hR2s?G6% zY`TMct*7V~rRPGcO+Q*>dtksS6MZ)#y<5^-dP;_nL|Jeyd!K3BdrD>UM!JnwfP1{x z6O>|)xbafb`f!kw&+`Dg=k4@#&z()ROo-uZvGfO{iVrU4WD*z(z$%QI&O3a!Go?1g z%S4B}72m$CU4j8xP`R8T;PzemUf*~3%~Tfj^vnH5N*qvWy0%I>S&FxYZYu$LcpU3w z?$+);CEl4!=R^BOE=ulx+p;6I2X%RLT-z;-2mZ35ch8=J5j_;Tq7)wAii@MQ(I6sr z^AMaFxZu%1Iol}u9pge=p17UoJk>)PYc*-o+jhH1rs2^mr#CXuBvx(;M}Pg^VT8{_ z@ZqDQ=I8v_)HTq0YN-J&ubi30h2A)@uW;iJTN$_`pbpgijU^2Vqi0N+F~f>%8C@h_ zb1u5^q#i%niAtv*!B^>$vq4Jl z0P-pXX-mmyb6N6tSfbJ(yM1h3dn5_xK@7bPaJt}BBs+|X)<`snd?YqZWLAp_aCiLo zMbt_K=Kek0a$~U%N1i{d-IAS{O-ZqsZdcx}G-aQk#hLfa3PTwK#2%Qq#FFME7|~%| zl)!3W7TO5Yr}1yEG#PB$l3&b--{?>Q|7K+d{3QMU&JYR48LaPTJy z(%@Y7drptrRQ(##a8h>mIyc{$tC)<-xb=&O^xp;_n!Mo#zCw;k0gRwZ5<&Q5dLN{rLVNyqQU zc|2`!ueODuUaK9G1(m>2Os5>q!rHZ)O2?E;ofA|IWuM;DOZn`E*utyc7Beuta|^|j z(Q9f84g^r2pT_Q>7T@IkTdB9BpWydAEr|+3@eUV3DV3VmO~3inM$MXaMNmRI#?Gc{ z1~vJvHQ^(xD|P@XHBx;4@86lx&a0L}^9cDy_1t=CimzH z_82%-A}m$sE?w$Etm8l3cK1yvhrtzaUn~pLk9J|4454xg1po@QPB0k&MA|_14@E=a zv494kfcJQ2ZO{JwTY@-_rT-zisy}Y*LVhkO(VAcz4JeVB_iobN1qy_nI_o`pQmsMD zo+{{#&`x&*MqsX=2s_2=2}FsKc-p%X@%6Q8)5dnFG`d(y60@?Jqq7k^9jRn>6#NEH zvN>s;JPPv>AaZF1`q34%lIgra1Tq{1QcU6^>2oExf}qJp;O8L~2uqUST=YADZY9DB ziFpmJ{SGflMPRN?GXQl@DtZZnZYBOcK;yG(qtBgdA>WeDkc@v(M={w(9A_ezXM0Kj zNJs^X!)WWz0r{oqGgn@nC88z@HQbgm{el%KnW>?OB~41}bVB_Bx)JCHwq941mVD~* z6DA;05TBr=MT2>d#&3;YxpOPA0aehcg*yS7;0x9yJoYc_JmeA;ux7;#DdQXGSPvoo(DP zM7di>lmk7g(*7gP>EwL`ns~%FkZ+5#*20{a2e6GtbXI6AlI zsLJ1KVL_hK0`oPvnd&xeJ3uulDaph>|3fQ=8AH7|@*l`ZFrYaJM=mIEU>(F2)0`Iq z=ssn_9h7c^W}Z7Z3q?7`w`%j#*ftPWOxCB2ngC30aPY1f`;LTWcObLcl77QRoOBQ= zUvlkTC9IXBEg8nf5T*-7)1HAf{3QAD65))Jig-|m+j~1}4q6uUAS>eo32i$#cO{dV zP(`ndb@rO;U`4bu2y|`qg=R&cV`z z-)9%M;4iXH5`DB8=>pD0f^=&xT(VLN&b*QbM^n8c`BmRyhAaegjEU9PsC_f2hmBj> zqGwBIZ3V;Y59os-ZN*=jMAYN=4iVP~f&khFyvyD|-s6uLsS)#cE;d%prw0G-iv-t% z>{xcO7IBLlBw#Z9 zb`^6#1}x65#~a^?cYm_8Lewy59V-36s%|lNrfb)(!0=-zpa6OeH-B4HQBk3Kvl#<) zIv+pt9o(rs1I)oS+@OP~$-k5&)D%J&B+0P?P;=o7_^sGbZHFE>eAtJt8w#qiC_AhX z?W0UbIl@>E8RGz4F9Y;KDmaZfNH>Q&ftuM}U*C%pRTNlgmL8-RCGU|ah5|ZCsA&*mH7fNpOzDW}>E2JTk(Wxdk%lTb6DcmfaaU%f-TSb(g3=c1z zz2h*gOtALZ_3Xto$Z!Y09U`_N;PXJsIzwqRe-Me-b zUC+%4qjZm*a~Vx{5k3RiLSlL1He+0c+*gQk<}YtT1JNL({5`4QE1}8p(BDFckG(UQ zN`5keg(p>PTiIR}A7c{s9yrhdYmCSN5#X3{Ur#TC7F)&NlnpM?(GVqKk>Y&d#yXaB zt=ZliBmPo|gq?#}o(NbQ#Fr*JP{I}-{MwoG9|fDkGlFH|y-CPL z!82O4ZoQMs1&Q}xEH0Vo_8{y02g0{@@?dnBuY?cZoxn1TRFeOGp?Prvf}9pC7RJea ztND4tLQi?d-)L*a_6y6O50bbQ_N@$C;P7O5hv1zDSCH}G8saq;aW4R$T;woKJeKXr zuN)qRnu(Llzy(_$SkoS~4ifM|ZUppFL$wefIXP-50Njgn*4N_mtd?o-?GRqcoH}$o zLd44SFKT|4t5`v}_mqngo|F{Sjtyc9p$9`+){p@c>~4-oH7@#{B!S83vv=fo!cch> z{;yg5B#Q^XX1EiVx&)&$@`vfm%{hl@5)0USX?UL>GA}gbw=3XJk6yhro{j86Xfn-w zlix0 zB(>-jhko~-JxytzMe)l@ZuGln)I=bCoRsHQ2%E71 zOVB}cEbCDU!EQMl{6#^!vt&>@kCH#ZHEe6%BMPZ-eR8#MU^)@g#ff;fy>=J(?JtQG@zd;s7!!^Y*g+J!}8L?izzhuItgye!q=OjaNHluk@HQ`e%pY zLySPV?L1{hERfdZdGXx?XX%Z|;$Sr`HlRAV*G+%LQT-1?Oc#%|d;azrf*G@7!uT8C zYg3!$i2}V=KqAPxok1Ja=kD<3stN)lA8{yKQs1eyt6~31AwMVFoFQ}~_t0q5W|xP6 z-DEy>YUVkHi%s+u;#`^Oxf5LP|AOF=%maIB5Wvnq1ftZad(5RFsG)yWRmMH7pzW54 zv+4@5Q@gslikv&jXI%bAN_PQpRYq|$3bZ&Bwl(VP_+V&oXUaGUpYXvb?l(xK?`S&?Ez49K;=x!DXqAf>t~ zFUgl8pi+n={z$yHFH1SFMc{Mi_U&{3oqW%Cqh%L`-OJyDNbD8S5@-Ddbi+1Pf~~;# ze$G!gZu=_y%9S7{zE0p46iJi+Mp|d}O@HZG1s{Gi}Ohx zT;#I+_(FU*I%RM|9>p!y;HkZxn@x%QVYT@fNx4&M*8{j)&wg_JvzV6D4CikjX2IxR zCEstxy!e=ymB!_`#-!A&=raxnIxzSoDu6cQ zE-ujtMf19s5~9wVEz2O|*J3D-EuA)XDk*s`qNq?r(4Jqw9UHq&eLv8dDH%__&tJOK zRS`x`D1xpQOU=F}RR7!sMy=YtyqjbOaT9kNxJTRnh4PAWee-8tat z?9^SEBnQMjsLD_Bl%ZMu1RXc|KPbA*^Q{#losL;*7YaXr>1wuVz%OLS(|WLZkr4*vc-VEvK3kpYMgc7lo( z;oeb+(qb%vSd2B{{<^rp33FMHXWFfA@+NK~bRcylV%4kpe%sA?@mq%>0LmD=9wC=# z#R+cqt>k5&i;GJJUaw3;l_qIHbd;*ICy-^Je4TY7CZ;(Te#pz2c6PT9cAm2j@1t49 z<}IYvS&bWKHq;YN_<=?oj@H|@Jwm!78&f?x`!H5I_M3kN!bDQsTYtF8_tjbGwRF)U z)!Cc&f<_)W+LOIM^~VREAeDBf>>&sn zFJaTCkdlx+gr-ld;UO>Kvj+Q@gySJK$sf07^=i|zy-(9ou3dMBvL+=h%_oJlG6&_o z{M)xxfn6~r?c{W}2t45K&+0*#$}D6=?r(YcePRHzAn#O}omGW4KU}yfoo{?e*;% ztYr1UiwqZ%5wZ?#P7{(z%B3J8Qebqc2B`d!uI*YtL!uI8Nk`U1S`(@gt0Z@>7D@7QXc>pF0Gp7pzn&*>D*&RQ*8ytom_3;iz| zf3%|h6=3`(w(svV%?S+4tT5{I=@8W8CJ8l&U0bZ%HRI`r0L2$dc2}~>f&2HnpIxT9-`2L}x?^nUM!#ReU-;bEKsTcQ?d4pw-`tg^g&zS>{tfRz z4w;L5@g9L;pBF2j8|%_#OBn{EA5y;9LkXNGW(ZaT z4szR`fgKZac}3DvOa3XV!o;>=8ykL|4CCP3-@||D|HL!F2X#zelr%>Wo_$J4u%7zc zPyKT98go>M4QK;eKmC1!AR1%_sTf4GY?uAU3oMC!#28I+==OmZko5GaCfvVM3Phb7 z>K%q6^gL&f%OQ0Rpxr+*AeXurFcem;le#_D(s%du;4TTYwwX8cHr>*qnHC?CW2Ofj ztG)MrJ#`JY)1pdU@Vl9D72*G!H{NTTA*u-(R+aO^?vS?L;+YoNKL#l9K(%K($wQ?t zcX#=_kpXA2B<-JMY8e#{poB%(h(U8jel7FX!JJKbj22r;Z&_u>_5*L}J9w};^W?$2 zJ>nNOj>_w3P}ZNQ+?Fz4x!c4eEykHY&g;6k(A_%l0wXOX(}AVe#>g1U6n%&z45)7f z`yv&duoQUAtw^#9$lR%zanW0(*y6iIJiVePbMNrx=?qIruw?=CMasKz<3`o?E69bA z`ILN>Bc(%x^AUhPvStnfWyj`gYYqtu^$yei2%}IFx?*bUfFLz9Qv$VHjBA$}Y5hDV zHms7$y?d`-fkA8Hc~x+nj?K?i9@W)ARG2bg8ECCVFj3c$at5-E&Hr zt@m;51OCZ_nc>jMoW2UNj-2K{Cd{j)!T6sKZ%@-vQH(v{YE!jeGATEji*Th;dtj<@ z?tK<-X&b>HES?VwT7712@$1)^=0-c1tXuw&@`_}Zzvy_xiBDeE>tDu@{qvMYW6@Z2 zv&{1xiCYCUL!416>7L-Jh*`MU4y%c{NE%E+O6vKQa4fU?QmMVPq@;0JzI^ulmH&Yw z%9jQJ?O!-f4|>Ov|4G=X`=>)*l5q1MJkV(RW4pDp=EF7tqjv0*S$T8GuYnPtWBZL9 zL+O+b*M$PpjdLRu>x8pw91x93o65-L0hyonM?QgFipgv2)Mw5O2W2B0toV%HC6h<} zU8{|yg&q$yUPlc}#~5GFX0duV`v)h*H7(wb<0nj5J#dY$q8gl(=*-6B!1yc>W8KXR z1P2&*vV&UZ9z84_eyoE-O8NQ)MVA@$LjEc++}NJkRN3_wR4-Y)*jw**%#|ye97)c( zK+Ds6DoSoRuJcS^v{ysUz8ON^Bk z0d+JLiLX}by7=X}?AE*2X2W+sZM{2N?q_9F`AE{TBI118azyHyw%hknFiz0;aiPMsZ2m4U{-pakh0m3t zBRJkCIG9}}(9;3DVNGtz(T)ni@>8-_$y^}Y5v0n;?XteQ5Cq<_L1FQ?S;I!EWb5xbN zNA^iHlEqwxOM=DJ^HWos6YuOzVSm1m*1s)foXXbr%gWosVEG?EE|~>QqA&FJ8?n1Y z@B1s;nEYp+k`H=dUBY8e`#9YZS-k!F>*LS#Dr;KdsvXb4Y^uJ#mSJk7U0trbLFNInDoncc{VQn#;Q7`}H{oM=xg-3>FRTRpfe@l}zwY(%xxAp)h zCe4pr3@LkMa=?VIJh$+&63oaVIUVT+)8Vom{)IxDT zn<3F5Pd zwG$km(UFC`>oh+((jf4F0}x&2)QvZAxN- zG9v7nKBotBq_(`#nxxstFW_6=c9(%#!fP>W}cm0yV-cxQ}{}frU}zz-L~x8H&bxa>812*K0;IBjdLy^ zhF_a*G_Fw0@q~-qc>TOY4?K>W`FtVys`6)$S1mbwE6*kYYujac0R9zl0jun|kAkIM z*D9Y-(5M1dfl7y7@U%*s=`C%*|Ab9rlyl{=AtHDhTH>Lk9F;6h1eNgxQy9o>KlHtj zClvY9E|#d77er9u^`Fh>{ zosN6?NRcsf-5$y@pph0v_-L#Mh$`auG9VU`)9ff<)sm?t8}>!>LK%)422bBvlX84cYW;MsO7o8 z{~elEMK$iLX{Z8HRXEcgGljx6u9-Do_3a>*!8Dx;Sqvt4_?Whz_mx?_F&J*+hwEU^D>+m~ zTbhXd*-*QC$%pmtv-OxmU8)y7Xvt%*z~%}C)XJ*rs!j5Tl~#iN^J8qBEtinp1w(xL z$B6CQkB0|*a8)9&#MUvFH-U*gmHn{E#SJBr%wz!c7LHP@3>B?6`(Gy3%Q&&hbMS!V z@HUfZ9=?}TrY-Uhnsrj>J&{O>V&z6G+?=oW$T>aBQ}X^x-m`1h7$9uUv!u&E6x)D} zi?E#_-rmt%rlSc;LxD$4`r>L!j%wdZRQC4*z75|3nFc6O+eQ4cLN#`aHbS#ZuY*C| z{m<3l%TiemiEmaDbkk5GZDw8?-?L7z{Nczea%$$RLBJq%?~ECEiFoy=#AnqLDOLK1e{~ zM?3-AJ^xaB3!Mcgu{(ceyyY5qfF^cUgKAV~LqHaV&hBfofV{rwmOVkQe0A5M3fKSL zMQ0SaAJ|g9P5~h zX00oKB;H{$%(E-TPz;1}r|rY~a_{)a<1mDGKD1};!%dorNQ#1*#7zzG$L;7TMTxu? z*OE;YfJp$Gfy!VH9jaJ$zp*Jr(iH*8dtu#28|fD^nYcCx$oT>KPBF_%*LB#?se#z* zaY%}no7p#-Wqc@0rcu&}isb9{=g$)`2SD2N!7CF?kMW3JNDJbCk75S7F~BiJRqWz) zLy`zCgU|*rLrP4a2GtyOQrer08Dj~AfxdAZs*EmMw{HFL{Bj?Ze7iJcI+O4v2(P=~ zcb5+&{&nPBqnMBD_L{*gL9GW*W3pb(Y%b}?SRu)I=Vf=O?%vXrn)NZRw z?}OtIBM$M>2>{tKjhf4#nkr9K z!N3I$4yuZli?_;9|EF?q*n#iCwE@jUnvPA*`i zq;-68=|U4@Tut?=RYrGq8tQU*Jo`o4m5R)_($o~=iZ1_2El1;AgWo;*Nxk&p-^4m*}%_nDp`=}e>F37Rd&9Pne6ILj{fI#0N3}A!yKOFQ+tI^HL zAV4$^IXHfZ#^02sj}pgb=}MhY4&d{I(JDmRCW>fLGL8b76BV>g6&FJQcSPl{U(-0A z#o~j$%?cz~eD$K&5ZK@F_}w>OKgnc@eXw=dQQa|znahqq>nGz2qe>+Ih3e`AZKLdS zMtU3;6PUPU*^>|9n(ay{;;q=R1K_e=8fPM1kmL>O0SgW*iEBu!wW%Q=p)sGrp|J=M zPJV+ZS!t}rsYV^?W5^_eJ@8^;SHe|hiF7246L&JlkA=CpCkOR-DnzHuZaR1>x?zm+7!!(u`mzpaUQF%vK zn3^Jd=n5&vTaqJ5Cf~8XcF{MZvhe0UonA_mg_$A(uc|oY4wC~azn-(=4wL9Eo+0UL z9XG=%N|9_#i;FlwzEJ|~CE`xpE7k3O;$z|V@aOauC%Wui31grGRvATSG(WXZuU_GU zmgk8)y7K4GY%a4!ob-=LWfeB2s3_TIlBqmM#cfWt>Z-Cy@;$!OxN(r*0bygQOfMDH zI!%f)Pwb`970}$U0~>$2E1z9o_GCSe5{YE`lFDV}S1vDqK9;UrnRgu$Ykn%a@}rD0 z}TZU~qzrDg8;N(JOPBO9eT^tn)zjwf`WWgC#o1I=+!82Qs3V}J0Z>YRO#(5J1 zk+^?fYME{9tP@IS^ySB`<;#}c=F@Q^S=!hf0M@^E*cCI8P^~OLS>HU5y$B-e1dDof zU=4Nf@Z+A;sB#2h@Y@uBf%?xz=&t-wsPBiN^@$jm#;RLPoheBZ!m3j_X44wkl`g&B zTz-T^rcn2Xq3NB*O(0i*Fj}1W7bvK0;)e)UYq|Lt0{Oil6_phgdoN!eyzX;epjbPw zW{5st0OGA91J_1pB8r>t*{M5Qbz zIQwfXYDh_cp>c3X#rvX$6L%kcx7^JVr6D2pEKbhC5F|3^3u~Ii_6xY=p}Tw%Yhxc_ zH(4DuUMCAEI5eZiL8o>b?SP7Yb(%^PEqkThoh5`Tl(~fg-3L?722!x`w zt2BV3Lb=N&k%Ifb*7}4;ZvWAvy;@7tFWO*{@lr+01Lg#nKYyF`JD!2qX=zZfpx(D? zoxsw*LkADXBI}y#^UJ<0l zg3epvlXXZHubZqBFE6>CIcMMOCPlz!aGEvZA(XdW*gc_?*$;pPHaSs-idPXj&?!?S znM`KCjZDDd+-(>W*%2Unnq_@vUlTF0%HXchz@`K%OXyJl;!`J22H>wt-1eIOkAgFV zR3TMQge-E%7#{b!ot!MX@b@VE zqj$`ycbfL-bwR->#6<^HYhZz8;2;P0+>2^+0{L?0?3WAVp4g0CwT{twCnzhjIlS$< z#j(|rT$?qVa9g9Y8 z`Qa;bd4x98#M3C; zEu$^L5A3@2eV&*1wfzzP`i-kf#+@VNk|CrvRTL#<>Zhg6kQE)e`qUeq(DY#gpe2h| zUzH{W2q}#Fnf>#6H!DuG3|!H>439?)e;*7m?Cdrj!x#c66EM(d7YW|-ryWmxy`9rD zfEzBVUO0u@UdU9LFTLUuyP0AtS^p~M_8toDMI15CPH;!$`u2a~wuY1HF<;(z<^_xs z$WuqrPDEEwN7LDKar?EDU93kpW-}-g{KbNvgtd`2SQ0wq(&yhA7UNPxxzh9Z_gs5t zCqm(XoYNw2w>o%N!m~c$C_&adN9WRBI(%S}eXO^m44_`OH@wqw;uhFW@1eleV^1NR zAn5CmY<`po(du98yE<^r=E0!Tx?wixe&;QH79`gAUzs4ne=FZ38zgkZxQcnE+gWpS|#|1=DsYA zZe9wTbt-GnlyLg&r*0W*`leAifZgle9Y(qwS*H^BKbTt_wbVw58gn=HErXkW8 z!xmbwE}=pzwx0yX0kD-l8x4s17+v!Rs9uA3{{Br$Cn=+8>nfiV0OU*i!@WF@)As`f zgB+wK8$`E8O&8bO^vkf642x{BUraC^z>olG7g<^3dM@1$gSv$+or@uzv_><+t$MB( zl@2}^sf41o7JSfsZ>L2!Ba=-KsO(ucx*x(c$N4TNxq0gUH^ibauS!3l>qHB^oIW!2 z-k&5IwZMkY`S=mp6{rTL4|y+27v5Xv(666vYPazPaC7_|mYbXVc6nO{c-lsOphCQV zZS_hD?~N0`a;AOl^^P5SJF@XxHpya8QBzZ~pNmgAycE06xyWN1(R0IX>FCHpBxUup z{%$#!1NX&8M0iPu$eO1AmkATu6})1xZ%J#z8o$vv5JdeXxA@mI_?X+A7deHi35u2Y zBOM)S6f3HWudjV?BEY&+HCtECWm9Y8w>&G`#m>~B7Z}M`TS;?fM$OpqRBkmL9nCV^ zXuBzAMz^QCQW8N{4sd1)-W`GE}#*)_-< zZ0cCjf5r^j6viDdBDAzBj&SAmQtO8yJ5B=;Ffin5h9Z@POuw{HZd_zFn(m!@Be?mp z{eo@?&8k=U2Mc8$pe0qWm9M>F*3KvlcFLP@JDndK5p71~1W;r<@eBtR`NjsEdo zWJZP5b9C&a)e-o>S-&N`lrt)Jb+!OC$N|&>ZNCrh+c#o>3&8-q8%I7n5<=sck9JbDZ-$@&LPdVaS2{3swWQ?c!l&i_k&@+K)WOmvO zd4C)m8_|)brr*B28Z+mCl}GI?Y0DUL!kV}fhaS@^x{@|hAS449>^B>gCP=vN+V)JN zvr~2&i_ng697%XHeY0jF;}Ekez}Uob#_9W_b)GH_{`XQT83=sv=UpByyFG%Pf9>inEb2;C%g-N1z=_C2+X z(vLuobB+7rruqBgoi2!#L*N4X6*C_?Ht)4R|Nm47gj#;6^1l^@kI{<2ZKqvZW3djG znR9rjUaLM*N3z~^V;wfv>d@Fr(ZiN2Pock2>Mzp$o9rnM6rlQs)gx=FP%PW17uEEX zQ`pvH+JkWcu61k4?$F3fzrHYiX)LIZsddL?xwJG^MLeVv6tMk@YY<>aCz1PRGsh+l z!UI-C6ljh)yGQ*`umQhu3N&18O68GXr7Mh$jDnDTqy_bPZ+8e3oNi0mAeMXU{D1M? zg!h97tj*MDRD~}f`m@5xsYrjHIoGBI(IZL~7bv_xh~(i2=VP&-ji()@cGn)Z-j@+H zx@ig65-wtva&*?Ylb#{@Y@n>24driDP8ZKIfR6Q5Jh=5r?I-lxYZa+2RRg*EUL_S2 zH?d^kKB;o8`7LqmZ-yO%mxTS<;bm6}Mz@p5GS&7onMeo(|ynfaWoChe@ zf+HgA-dq8rA z{>(NUmC>40jn@D`CZ?o>*|>V<-RA_3S$`}x5PuyO6y3BMt&IEk#{w+PU9zMxGS-co zH#?G-Eoykm-E(W-PemeLmQWo;J34MLs3Zy1uZFpB5R=DxY+wx}gE(69OQD>WJv;9) zu!f~6lLzxlSpqUj1%e{Ff^R3(TVgEtP7%u{x*}oUneek`{peQsLIiCT4RNVzM(IvO zo7v{g*Gwx$eZJO!NA2voFCe}F5i}CRol-HXvMob~XE;{nt=*lIlOu*Oj*r+`EV2!)EF@#NndCRS?LJ`aX@uB_2-)))5Dpa}SrM-`j zot-WL@K}C4ll@f*eHbwLSBzU&C6+$*H#JRZ3>wp;j}_=22%Wn@rUsK2#4TN4=Gt&B z{La;$M!}(lETy@EtLYYiNu}RNEOz`Y$9oANQCt=ke%&C}Kw+b0oKgzgrNVE`)mieK z_*s8?U+2@!YjO*7$7;u~J*MnK*vC6!RJK;sbhbx8!W&m-FO_!y2M7rdua7>#`-_`> zOf;29t^m!v;qgPni;1j&>7_NO znE01th>2Q}`Q%A!VUJU5=ExX6PM{s!q>yTvKXH^epp0xC|0gJ2t##`L!Oqj>apX7_ z)qRscm-ThSJqXZj)R-|R^2~xGs;DkxT(RN{Gn0n9RM{=dmpKn&Q8IR|ZmO`DI8oxK zQ`0WT#c?gDfc?VOLFiIYq@=xFX(T=)z@U_2m+&3O+86P&iJsI{;OZITOf`Qowrm7r z{EGNN4HTq6YjHF=-b^E?xL|n@Y9yIRSyNWFjEe^?L~!DSY_6kQlWDhD{K_n;2kSIQAdO6i%wLt?B5qJke+xUTqxVk;DAD|b68Ht zOxr0_k`QQ|qk4!p)VT$g-FI=euKdaoTh@pFI(20rj3K{O=~zWmDKnZJRz@13@^>Q| z!Oc%+l|VaB8NPky4c}fViC5Y#Mhmx9DP zKu@D&0*4e`rB7b-N^en%nN z&gEs#ROo+VzhbP~*vJUdEd}96>gK@pbMH*fGDv(jsoO=yr>J5Ev3r%NRelUHu zsQ7#<6@39*C5^n`DTu70WXe_nF*w0os{mR@$cP8g4v4y_9in+L9*ZovZ!yz<9SD;ZrrK%jgn$b-he1K%J@vobr_bL+ zXSWiPalwL3Ti=l1l(vLMiovi3Rx>)}D0+%gX z8BA50C1#V^r*6Jh-4xxm-R!&Mrp~||LzcuT~fW-l$wpgJi;!?7^{C5X~>T*fy z00P2Uc%Zy}4-8Irc0rFlYwFe@aJ=z)PeerbSp<20NUoWfFeW+^jm$=GP4N~oN3)N% zK^df<&Eh5;&jIDHqN1X@FiePp|82W2I?ITqyVyM8n$d3LvvYYv(c6Dwtn&Ti9Z>W| zyC}1<*Oh*b#AlWF?3otDc=Ycn%G2xx4rVzxyQmp=srj##4H?oX^M)8_X=@3exwWxN zZ|1h}hfm}!9V_tyI_1<2@(8%I6 z+qUgzZ0>Fw_eYV1!r+&hhgY61RG=mi61ardt+x!zVj#|QztJAc8>dG zo}9(U!!G8=_zjHp4h|06cq3w=rGQzvW=+e(%@uO(tZ9H6>g(Cj+%i%lai?_d-Mbm! ziYYR8$QF-60KBI38IfxEXeR7obS(23Tej7{m_4GY^pVA5AT4s5IqiGLi(&N+kmx|;cu2q%4bd4)Z5vrCJg?OA$* z)=YJ6is}jKPvQfZEYX~1NN#+B~KuD&X>6sS#Q^Z@b&hF?Wwk3A>| z+LCJM8^pG5><%LRnsJK9@4g^bhS?kDikOC16UrX`QeN)KiNvYT&rM3ILom2+1;`bD zZcc8*LsF|D0k*~TTMpq~ziY21C#X!t`AT_{%x`{(?)0U#4*e#+(3i zNW6P@T<{~*v~6hhWIZCe(kj&+zERFtCKi3&4Ynh33lujHYD2oXun72qk}%*_1V{7q z#-{#J4!&&4A~w77t6&9bpR{v`TLi3;NnOs+RDKd>xaeZi9`b=w(9%iv zgETBG$L&xXGcGLZ+}efh`GELf5cH-z=?Yjso_xxWS&)P2$SLHv6ZE2kbeuN+%9ir8 z?9i(t{slA>V>TEMok|iwO=?3D(s~8b@};cIY4-92lZ3m&+u^z3Vivg| z{g;NhUwbug=9u3d`&blV3uP1X#ROaLip6>hiee>VykgH6RV0AF2L!HrMiW>x(e@} zn-eK}RQ47+@Ex=feG5^<{sjnfi>h0a6R_|5^w~B6+4t)aFOHw)^&HcFoj$@7p|oVl zc=`I8av0Q{nADvT_yC$|=wHuMr%oj{&Vy$s2vZBe<6WaoWo&8&oK6fJ{h7N*+8Y?q zyIyE(wu_LI^`Vl!)!A7pJS1I5)oRSKo5w?ev#uMU6T)GHBB}%L)9&JuK)I{RUorfu zCB)Fqt5=6GAoojU*+KzLO7q|C31Bq%fv~sJIiVS8{}*@D%i)i*v$r2Ve%!92ko-LZ zgQDl-E}lOxSt_WCeCei4Q1Xi86pa^RtR@`J{>G;Sm&M$J+$XM=!FXcxjVN_l&a5F2Zjae~QEW~Cofj*bCaX(HMJ z-{_8)vcPW15`BXD&KF*0Hu(MHBkkW@LCFTeN#G{Hb>k*YHr3pReuESDhj@_b7HrI{ zpb4faB#8Ct6O$cFf`dPds_=J!33eoajVvn{ssuF#{MZTPf9ePCB7tCpCZ_F|3HdiR zjiKb>a2-73oI@Q(7{nE}!K8yIs3afmL^K*e)F)+V1=`f$rmb%(`_nJ7m=!1(=7Vnqphv(#UDB}eYY6O(yA!ppWHrjC=CO=XMNO0P>w>=!$uu0i{Iqo(i# zOQg?fo}0`Eifmtk?NeeH4P8ML5b*5LPe0geX4;2m5{JdDLfj;LyYTgrG|WMUkU-7o z04{F+rS+^?P|Bct?`xPk5ehXY)Rbs%AY`=SyemHc3l~1)+S(qQwrrV(rR}D3Wq^kL z-ZRQo|Ner$`0TR#9Etx?^CnRb?h-zi$|ICE*!sN3p?Bt6%F*$XT;Ba_X=woj~w zf_JV8AF-PXNydcmb02Z*+MVg+?F9y<|M@y<8B$sPkt}G*aJ#nVZ8KDH2{*9@Z)&Wn z+6n$hlJ}NY@7dB815_{&egJE?*X(N-vP;rU6N-la0uGWP;WFw620i-r__@op2VJVQ zJO6Rc$M?O|yxq)qbg?>RW?Ns%ICISKKg#JByAYi4%j3_#=jPQZqrmqlu8F|B+_n<0 zsBa}+ceq+t(stv{wp^e+ZvWOOZN1DJPH|`V2`h%k!lr15D0PGSWV*z5|CKya6cGcz z-n@BJtBjz_DP4}A896mu6U`gZAud=frJw z2;=3SLs1GUhr};_)Yacvf22MsfxNuPZ`y!XN503s-%$37Jf=qkN^Ue3O!}b4@pt8A zlK=XUwv#C#o}|%DMCJ4c5kw&TYn44w9=mzqW4RziIVCzMgZL{>^ErS`FTEj^3^%qM z$S6S!`Z3egW|6V~O5l=t^e6- zAjrDij2TaE1BrEUs)g3IL?4E6po>qdy3*|>o zq-V4E9jhPs&;)EW)n;+o7LK__xaZ`ewjX=aXg`$|=UgIZ>$1-l3AW3REaWJ44pd(f zKuG=rtrz^a6-1j_IMYP9_1aOtN4IX~ueV%hJwC|w^b&l~zk>a&qwMd3%#$OVcZLwy zeqCiGfo|zb4;n;sSG(U2okZf&+O-!foT=_YN_Nl}P)oG(jv!9W-pHFTNTp}(x<~ZS^qi)eA4&tP2zHQ>t&j75c;6#L%^-fVN3vwo}w5(L!^l5P~c)Gxp>vdsk}&-+D5;gF%&zILn=#cSSrE!r3ZeWJWPj8!@!-(9i94W+3x+~oNa?3%{oQho$;4wC9-HYmr)W~)%tfl8 z@;rnawj*n%)u!3MZ^FV8JKBb3D`?fgleHm1{DoTxPTob%3ylEqMOPn}s_P zNpILZoSj!g))*Ft!qQ_5W==ElccJYu)H z?C*YQGU{OHyjygAE3BdT}G};>G#1 zv?d4rcAWCV(-q-#YqpQw=PfdT)^>dB1qpUolUE(Qe>{X@HJB1i@E@`Bu(ON!#*z;P zZ}yPNoc1E|*|W-o>Ih+hEV7NO1azcuzCv1KzRN10HXf{Gi>3YhhNbC7(oq3dic0Qj zK`#J-wQQp>?Jy!W4e?O@^<@km)!>C!4h``R#Ws~+=KD(@<3$91w*ml?lOiw|5 zT&CNJ@%(saYDjT{P>BYC;>-7>)xi*}hV5Ij2nns!|JHHynMbZc7sx4!z)F&xn|q4Z zg>JO41E$Sj0M!6yvxgUre@t_?dgaPs@oo5_%aRdX7RcCOFvyZac?+A&#HCcLbS_T| z{y~K9HvdQM3l5;nILX+h`iM&Rf(02=^UCfpC1l+tSLIJ)KWMUI3FY}+k1-q!Zn3)+`F8Dogake|KmTsE%h|L3lb*&w%fb6Hy zsoE)*oSDc7f_;BH>iVabJSp$cjdvjGeSV%y@uakYqQw$2U>(bS@eutVdOC~U~uFo{0 zP?6Kw&bMv0DuTXNR0`gultk%2huWNCyH3U*XtyCP!iobzht2gwsT;Kq;Sn&@B*!YY z!$VerURuoOvYuSx0F5G9P=ED*4db`o|MF`I@hC%9U@zw_TqxrdWbTponaZ$$$U#=@ z7U&{TPf-in=9P>aKVH&-k$lZlzCU{RjZE=ED06yh`9&6t%%%aXk?_?r4WI-sCi%YO z)dt?*Lgt#NkJx%~g@h_dX2)&N{1~TgSH7`d%zZUS1B)KiwErJFKkh)!&V*P#IcgXg z@V`XXT37z7Z(Gy7xy3{)465PJ(IgfXih-J4;l}59#q2XNRboK5rw?yslZGEGz0z(iZViM_f zV9Be7(bQbFDO;K`blwRn+A$C~I!9Y=`)Z}+Y5H5_BsKo+yL2z=+9|>|%96h&y3uk8 zlnGkC9EIzF#;3WA@Xo&(b`~m?WBN%Iq_u9>u3wcl9GDA!kFT3yFkSw6{>(K-?PHZ; zd@!}3DJvlS4lL0jn@J_AJEup#`=K>bkceaddIMCaO!IZRhw4WtNvW1QkgT}yNas07 zzjo2(zoKUlz!5V7J>r7|SJ&?BR>!(;l%42fc7f=10-rd1_%IZ+giDZBck6KT{ToW3 z@kY2g>gsIgrGb3vUedn@Fe|fyc|jgvSd!_0;mR>8h@Qch^+4y0m|f^^AikFki{xBH)01C`!kDY zTks25E@OV9wAH`C3Vht^g{=jVB_4JXf-k>N9?C0Nh#RT{yE=Q%hi zNDfHLx#a#Z>)ZytDX}G6;;H{qj)lNu5_i#R*BP5Y7tgKXdBd~NQb@?>!%aVK!YI_U zY11D<^KSJbln#j_y3)GDT3g(MaHDFe_%vH$12Vt!>1FAy%K!{-HLkvmjtP7nGQi5T zl$;*rT+M_R7Y&=k!;Y=$K(*cElTj$k;=sj?TuJR{H6^{Mlzl`2~8= z{=Py3fl|YRXN_jP4#!OF7b=jtiCikHB$PNkeGI4o9cBvXL8^_~7tS!Tz*4D!tGbg{ zFZN=`XOHS;l&Eo_Byq??0)cw=4{p@$e{wW zd?B(aWWZy$q2&PHmRv}<1mrt&vF1h()u@w*B>N!&##gRgGeLntE$@ZP6RA7jJn7of zzY%`gKK&T*E=*BnZ0tPx(hehBoz<>cdwhHxUwGf_YB|)Vcnek?EY)8=KNN4?y^*zg zDE1&P*eV5uDmR4N?1^L$bUScy^>V-e>7d#wsF79?rYp!a$BI4_B5V|^yyfY;OV=@SFG>Cl2GCZmTT5-jA|C}rk$S5Jl__c;Nk$R7H!F=J zx&dMF>d!YvUrNugn2J>S)#rONu=K;ZO2Q=DR*iJ3y?gi8{HSX+qh><p7%E*(qZFrV4Ky@mwNp!IUF-tqF-cCZ?0Z>iZqp#n+vjE5nr|)AjjsMg8 zvS2=-8)h3%;qQsvLu6QoXWH0|<+eM;VG&!iSJ_=x*EBS+0R*TCH^yE~QgCy{jmRL- zy{aPz4?YNQrZ*$0AI?J_n_ikdNT{j7UoUQK+Q*yf9J^#Jr?B!GRzOOFfs={3td$`g8epaBIG-)<)uh z3_x*NJC;(08!fFy;3k9H6ty?KWGV8~=Kn4qWBu}bG$>=eZDJA|R@;vi>TkEQuLwO&x4IfL+LF09th8Vnj3%|r?elN0F(|%o-X@GPvDoioAa-422 zx^T|8IudS|os;wCo*@v>lcjD@oQPK6{{Gk5bvFWAohRR~i@E8elr@4JiNyZuwQDj& z>Epx}`8-7*0+mMr&&!xASmkqeQMq8|k1$Z*`S;(W7&%}E=0p#h;2gtsEwOe|R&ut> zFl^Ri`p>K8*457tkoM@@d34j`*5{xO^_4NLF3XHDz=4CBH6T5Yxp%!w;Sr5Vw-{YOtG>N?%}`4){nJndBGqZPG=nNUOmd8k~AD1AAi{J;N5Uy7)w7j934{< z^{L-OkU46y3s%)AJIH|;={gY$NyO?&PPOhh>|HVoivm%ll0SDy?SsRGzw(v8-{(>X ze@_32m`0Rev40#3_=WHN z?1-xSu-76whT+^^T`B)dtNG8LM_01%0sd@1C9&=wQZ(9?fr@2apTzYc;{1}I4eHmA zE#@L%ddiDgNspgADgNv)UQMz09+r%yjzpsEx}1&(rrUv=-?=zSUD|+15HC}BbcN&x zRDw3>VwF}I1*#tSV>&XAtiyfCRQi25V1M%j6N>9l?hhH>BK?B+VPTtFwj%}X08Ud= zipX$u1lj19#PH-;B<+CYA+i(T?-{vukgwA1z}sw6b4lDhfe_kT9BLywcXngKb@ve? zX6vAvt97CQa__^QWO{wF6*God&=Qjp-kfGT@cWQ3xS&q;y)>pxMLbi+7;I&2f$7 zzL|A%vj%38?byb+fE#Sx`rE4HG&*UiKfI13-N66gNF_#PiAbq~!@POcr*G|t0E%k* z0PL}l`WYCd3nOnGmMqCQIN|fhk9V(3O{9&1%=Clz?}X+Dq`>*mPO&M@vaII#6+%;{ z4XjwPLPUS__wf%cD6;0INCq=YIN{K?sNeip_CB7)f=MXb*-80QFApwR<3 z+8JXq3-zCt?XKOV^P_>5;6!fRP&S%#StbMCRvjoQ>6-|n9@)|*#z#EeCJLh z_Ffuy>lgCOCG%alCro$%r5YY>Dn%`R7enTuAqyPCNvkH)e97C~fn$Q?77^WIS$x5a25`iRvaFE| zVz^T=$Cf}}Pz}1FBq%FF!@iW3CQ+8Js{A@%g5aUMoRkmcl``{XCy{<|D7z$%hmzfh z`30oLJV0$AX+(yu$n_=3Lv*UNPIMZ=wZXaGV$Fyqfq*ssD{P?P`whqo*aaRm%FL`j zeYY$S(W-LR?;sLn&(MxhjptdA5rn9Jdtarh$;`Av*WCPyQ9`*r^urhVG>P> zLczNC0rzpD@Q@awajLXhAisV_uxfy$NlP{i-hWRMEJG7pYb!{V$XLt2u?lt_P9 zJKBKb5>j637i^)d`$@}i=NJm(BN?W+82jS4oAZ4?prRyw2)9kascHMfd(R$E$Tmqn zBrR#q!GF+oWgZ^bgzFiZz+C0~OJI?bmMSqfsB%rX_Xw?%D0XoOOOQ})Ad)sCZzNU^ z#hG|ekTA-a5x%(5I?MbV4Kx+6|NHyS&dH~OlhR+kP?)JOe&65J?-wJZ=R z1#_^Z&uKgq>cY0w7Q#f3d5PLm^l0#r&Kl<_jI!DH zjdd;Q;jE6gAAdZdEK9p&gT;bt!U%IDX9l-S{g=akx!6;Mc?XWT38X`AN=JzjV9u}) z`jF0~XJXLVNmGI9^dZ-CT27I`Y|<@|VziOaSpaU4Sz_D~gi96+cRMn$J6`rv)Nb-a zXuR-!>WglWy73jaNf7zowAlq%StSQ|yx*QZ<~+%j5(`NxJb{Pl+`$qOfZQR3%EGsj z7naQ5Gz3{#tQ9KwoorZ% zvy&AsyOoR$e)}QA1Yi_K2-K0)N|j!=49cXL96fNSsfY!%%4o(ow(9*4t0bYKDc5rg zHC(ucoMw<%&7i6sZw#*d@p&f9pDLw1tGEBg*(;%sYU6{bU}S6LoQb0VqRl419+z-9?6gnK4h9RJ`|p}8TbV`M2@~~2Q~(zD_DJXBa4x8P66^EM3-R=ds#AP zCF6|mDylwHcg)HTaB4sismL8P^SgPeqB%j|n2wSak6IHRDAk7(hm;*OHc4K7)C2vi zn{kPGg2w+(gFhN&C+NmgguNtocu=wZlT$YSPpD>w2tAR?fn__Oo&2B9uF|;z#L_pn zHlT;?B$pP=u|%9c#aR_g@u9ULX5SyZ1n5gi?e3X*xfy-TIf~jc!)MQ)C1C_ZauVB5 zfow#KnT`zB2q7jZ{idpYN}7qbcNzI1?oWjVUDm4_EIU7v;Hi z{f{m70tza2KvA$^i46r28y0Ld8wwIs>?LAUz=|M6RP2fkML=P&mxv%uEMP}Xf`Etx zr3r!zGw*kkocHtoe&>(pImZ~7x$o=RW$m@r-ri&S*>@ZWah#L0FTop4qN$g##DN3} z8r#rSiPsrk$Ks7O&)t0&C&Zb1;cxQ)Xx+uXlOH6ecEbC|9`kfmR1*SX5+PIQf&`>~ zQtgq`chK^R3m8B7B<}*Q8K_h|-L#UtZ^rR~KIC6P{zO=gk8?7Wi4dLSWrT&zr`vS6 zyikj9_-OI+X?%Sml`Q>GB9nus+yXi)W^VWB(oC2AH5h7~0-pAK#M{nnhl){O+SgRz z$P>SR`^(Wa**J0cG0x&!y~-t*Hkia3MsBDPxRc{=j2UsHNf}jxUL~pn|4JY$HmS}? zNWXvod~o`&yY8Iow`OfLxKXkSbp2(j+umb0IEnnA*pC?1{fZ%!9u7rXT8yym(+;w& zrzIth|BHNcNJ;{D17c&b5i>1W1${y!gEMe;JlN^~qj?Yc0G!sI#zuK@?joRGNL^>y zWnlfca2;j;9ej>-zry_dl*pmNPnhfKN6bEq`$f%WMK4aA7*OaNhpx^_CzwtyPCodV zn@a-*x-uHc-*M~JAkFsD9Ft`IEBRBgs{!CXMUJ=o+&h>xTI79>S4U-KBN2=Dtgr`zDtnE{c7oi~bws!H~ zS^i6<*AF$k>aCxtVsc2^Zaf*kPxl$lVK$c~?o(MPK4-8@zxx z$r!!cim1z%_hCrwVEgnD{2m??6Ns&Mi}_)fW?n+rFb&tiRTsa#y%w~#>zwglIQy4@ z?iM@QRyVCd(yL|X<+Om`>mJGfK_Yd3UsYL&&^RUKRq7yc;D<)WPWi9a1i~mvRxuuU zp6XVrntSW_r2#64D}6uE&%}-KqJHy>uA&UzRf5_lrX*%8s=z`ll%3Y(MwrjE{u;S$paH`P2~OkwIY-F^SxG z@%bEm`eVY9qwO86t-DgBF~>cBS9WT>{4tFvELu$p(SKJ31DV_Xc*^gY4s*F)AVbVx z${>?YEMn-#4m$hS9TuYYiux9Y!HMtAUz)Wa$Jx{w%tai;tYQ8YO(pdI42J&o0@0f6 zVJm`3r@OQwyvmnL*|3wy*zlL7SLn9~Q#6LG=&IRiE$7%H~3}NCk*vudg)(u4C zew+{CxOPG{@yzf-OiYa4kM@SIZL8v5YR>hRQysSr*lDUu*QidHK93HDg@-lv?5`UR z+D(eeAaAiLa;E$*|2p1Ollmqmk#BzVz3`^K0&pwb^wQN+2|6vWfQ=-(-@Tkx2U`B6 zZt5SlEf0R=Z7>)LFCV@nwi^-E(Irgn7x8AIj6?Li4B!jYXp~X+%4}VsAYTIy7A2G3 zw~|KKPqn{zE#=HsT1|*;ow8t25)cz%J}@0fKbVC)OgO{E;0C*hsM2*h;7wy7dwgN2 zdSqo+xeH1v`3U@+wOYR=MpwvUxcXBn^}sjTaaxZ$Fa=RiWyH>%3DM!1<+;mcaf3MI zBRuiJw!-BO{|DHsqJl*n(%y(hih65|STxF*2Q3Hq$;HH142?l+lu!=vc-*5Em($+X zHJg>vIIGts)c6;^eo_3jpa4%%Kh%WD=|}ki5XT$b&Sz#L=q58j|4bbf+1Soe$EQ~u z`4A{%;j^;=cURI&+HE}X(_ILp?72W3=wH10($G5q;ZR!d|J0jNRt4=Sfz{_a=3w^B z++i>14*vG}vmcGL=?9RC6O7c;{CWUb#cjOi3otK+)GzM1G31IeE*-~FnC3ejhk}82 ze~8${<<6~J8v?GjI5K%0Sn|LUuc#cO2wCR_<517M{&YO7&LlU=nHc%5PM*#~_XxwJ z52b_PDak=NVz<5Hh>Z;*PL416p)+JOZWC=d!4aEmMsM%lE#hnGL=30LjChuIP$wd- zBdbisVRb$t3Bu%z&kOXMjZic}+a-i_J|^0Vl$xBu7c42J@Csz!nT^b;4f)c*lX0yq z3S`rg5iM2XzK_jQyN=7`_aR>Wmxdg)|8f4?d=6kcAk1 z#4pZrS+?wPh2Fm}aTX>J#ECC@OHW1|pOx@&T~Rhz&$;A4$BC~xdO!E+&zyEHXVYxMdkVIY3XT?1yV6S9Rbqxp(Qr`XYcuUuRi=QkeNK5`r^1Zy%KcXT^#x3Dr*>T5WUxjq_>h| z-JuslAQ`F~hZ=D<1&nau_bz#mKiCAVcLKG0;@?oE_vdt zK8+$dF{X1>4GefC$rrS?KLNJk%dc-dn`r}sA$vWp^cZGZE-po5)gqJ$`10|iRKNb2 zTPn;wJ+F$7BO%U$IhMuxjL8g93;&7V-@JjknNEBXBP@ivcd5WAflYoz!Bx?0;-ER( zvg26ztTn5J`H&%P`K)I5tlGb_;0Vx>jG|Pn6zk&p&3x_|caPcvt<&X>l5&FMTXf}+ zZiv&IoK(`9!_p?7KYhw3Uv@ra-D^!;{-qoCc;5*djXGXztm>7#Hr!iVa*kWZRWN&_%V{znL582C_JEY0JtS|7Y#UPk1hD8CF{*>V`WwuA@hk(2L z6LPL6?|I1!LQ#-uodO%Im~KpBBs%Ypf88kc`u?eADL{oQnBMiWn7%B&*&p5u!;?NM zXGOx;*Un=t3zHv2@mS}zVMnunG^RMk?ipMy-;zUbrX)mMP47>X^mTF>S?-+V37JkC zmP|1TK6(6zgg3V)wlp&QGr!0{1H9y%*Ldhyu6>{DV^9Gi#%eW8vw=|AW$4hu%B{Dc zQIWIE)F=i`0_;^|_*b{x5UK3EL^jI#P*4@aW!*e-(vqboLl@aJo`$QHV91}_zbM9b zp59ZGJ9aAIPU_wUe{Gmw)owiVN3~{<_Qgr~6)E4X$1n%2hweHIN^e#Pn`uBkMVX zz*L+kZCBnqY)JRjbA0+G5Q1?7ETnHCRZc-jNliz09-p@g{Y1iB6PiA+JI&2aO_kGv zhmN1K)h})qr1y>R{{gG~_u7lO>AzY&oqP}AOd;$K+JF3bAXs>e-vw3M^6x@vnmN-r zMzMSF#dt0o%V!dIIcgh`!psfq#!Q(SOfaIaz0|wej3dH1r`uu2V61ft$#4gVR5ppv z0akD1woIov53PTd6$R#_0a94Z2^V8x7(n4+hYIRbQM~1PVIa_RW6hknlXI7{GKf_IQ46#LbYnuIPUs&hj#;U8q;lzePWGIwF>QmvPgzEj_;O#-()wS z9R**b_pfGjnC4(RP^>!{9(MRZFn=VAJ^op)hRkTTv0}=JXF)wQ5E_90YAZ;g!Vq`P zef#<~8Lt+kP{ovQnQyPSAILrxgGpinPK_GE95)qr5zsS_n)u&pL$~)Q_`s zmvw$C>pgEIqUx1AP3sf{)K9?Ruiv_LDPl=3dPPu()|51(zzPaISen56LbRv-o9Wl` zrkuoetdUe;YtkPwkKfaEf_y5wIriQdT0`$;=cs2`E=Z3ht3>WTYBIG_9M_qTh2_GE zNQA7BlwdBFygePfq4WGxOi3bOi1>;7b)P6r;Osjap}sr}|2`Y4rxjB#+4BPFq3;I- zq^Bt=AfXy@Sxf?s0sq0haxQBJK}RThjB(#ZMyzE2pRBnk8@T&2HRwqO7oS{VBVY_H z%Q8qY7S7H&h(Kkx_4&HS!-mZ|(SiL0_{`0pf6`*s7mlKH_JNR)hL^8+HtGA@X|l)W zxFgs7)#=a{egS(xL>cs~(f{6u`W0$u2*? z#01S8*rd%|eq9Vp7AP<6h`k`-{QH@W9*BGkkUZzL6`kyVs%lcy-u(BEaisvZN=(M) za|9$A!50;IRTKGe2Rc=cm z0o3I(CcX|3>0RxmiQy^D66*Ai zsOOrDx5|c96+xp&N;vSJ%!f|$!f`k3LSWkj^JniSJQZd{UDzvgQzin;apruh^~Ano zP&6fAJ$^^>_?BBLuJp1S^Hhwy7M=L3<1thKAmetQj#si}30IJoG%;?rW-Q~m%v_JK z)py0ayVe`>__rP6JcsjxY+YJk_zY5^U^^wWE$gbwB<4)!wc>z4l*cEAB$kb4=;HHD zRC2FimA7Veoo;NJTO)QxV^CS%UaC+LRB>+D_Ve4?0I&7IGC_VI{$fJ4!w{^CVk0$S zXFJ)LBYV$@k(*^jV~r=YI2@D48osY{0=2MRzL*TO)zh=PFJ_!caj=$NzkO?eANNPs zBpkdAptFd@Vk$PZZb|769-PJ}eK%jw;tt7GtbeJ2G2^7wrS66{G z*7YVe2-~r)_toWDk8vV;4kU#gZCfp^KQ~qF=c$U77~9LRaS(NnV==PF`$xvg&cyli zM{*g3$H-joIdu~2vgSX`qInBjFTS%vNT7Vg`J*%G&@U*+?V*ST9B0lfQUAI4g>j94 zuTb`CTTmy6RTV$(6mYDTg6icgb!_qR;ec#mQ~`D>s@qk+|IQLYeE#SIZ>KLKSp#9Z zB^l|{KDC4j3GdU0(n59(aGv z>Uv|uD`I*w>{vxAAT6b=V=I+<4 zMlmlHb2;ApWy1bT6?I|Dn%$Pn?D#`j@x;-aUmazr|3~%DfrWzxLG4FP$6#DvA%@sc zQ|2Y%k<%YT=N1Yv^XB^pGjy+Yf|*9lAG)eW;?|4$`g5GRO$}BOvWA^USX7>W8Ys&b zOgt%+>8>tX&%%%)NXMLwnb`K-o%Kvr(GO@W36|k$9ENxA*fAy~PV30po?8EZ5(1Jm zaNHUai;*m<#;#}_u(APV%w`qxH>D`To{%ud+uB}9%i0-m!P4sTrmEln(Hoob1&lJP z;>xDuoWN%s=g6aO^sG^`n@_mCR#sq>-S|Q9J-YHxYFjXAUXWDy>|FEZ9RYBsAn+T< z++D22g%-;WP?XsIl?>3&psy-Ye`OITXZZPvk#z*bMD_%CY5qemhP8M-ZTV3MDj^=N zuwr<6oxw70ikoiD?#!OcI|A4Sf?@E-#Rbs>gG)Kr7CqZBD2Na|9(KE!3R5vtY^G*q zh3a^5qAtUw&^OL9a7Sm47*EKzaCgKV7TQ4n50nvyAy=%|bY#;DNq0B49gE!E_p(LD z{m5`*(7;-wqVI9Qow@0Cf-fDFw z@&TH5<%EY8`hqot}^gs zbl{ijyANgu_SY{Me!9M<>Q3(weVebgSyKLRZu+I;S1PhJ5XhXr#=-HFfJF!z&$DW5FuxYdg2Qh%MQKVwK%8zKE*zV(V z=@0c^dY*Z=HXC{)kT6yl6oBX7KP5P;4N8lq2`0o4&t{^`1_?0YJyx3Y%HSqJ{ z<0hj3`NlmhUPI@3g>BY~-9&08N*Hg|;yDHbA!yn!@H>;c=v2u-Z@;5hn7PEp#Juy; z1hpF!ctu0GF0Y;YKZ8A*n?3$uz8zJ)=Wz%o^5@9+EfmDf7O5fU=iK^LBzk4;;h%2w zi#r0O7(3#z#`Vn3JFjE!#m5)FWw^;vYMK!9GR09kyl1V+J5n z)=skgvjUS_@ik!J>jJ<`ciB{}u39e43cx)yO!EuGY*%Vh^6h+_2W6(jV^|45$zHIN z)YGzjg<`U8rGdUaV?IAr3$t;*0XSgva+h<&VmEL8P|6M&nWF7Fc1Wv3>O)APZ^vm3MiDW++XU0YmKq0fB(_Vz8;2~dtL^<)uuK~7SZZYs7 z9u?23gF@GZ0XB`x2x>|hKwcl+L(xKuA6-}7d>w)xN|2H8JveTb%*ESO;kVTNE))9| z;Atf0ljI0l%}wDN`QzLcek&d=x!8UQg`N8W9}hEz)ClW@YqF3=_!{(6Bdu7Lp~Z6N zw9hkhW&wTsQ)L1q=<(LX-R8u|tqyC4R_q5?F<_ZH<>^SU%>e0#Gj$r?h;uRH@eV}DWH3zvZ#|xh z6!T<3=i+l`&dg&nbRD@WiAOvkgm2D0h0EdcRfI?chLuq7gyO?{+CfwzX(wsgN1tC) z_)iClvqb11oEVr=>j2P{&_oRhN&! z)=*Sw?db-k)RpBm!a{&Qz+H~I*m6gcwp+-JCpf9HbMZgnTg~k#F5V1aFR4E>4xyDu zEAf%SugicN!DrtJ{w@|ea_s3s+dBEk!m)*z+sY3mNQgR*1d_!04c@8v?F2Ba>i@R*!2h>Pm9Pb~c3y4~u`*U!{`_)W^GmtKRx`?OVpPudk>@QfM=1S$kG& zHwD&0dntR1c7wps*JNPk@$t)-{w11*`L97uT~(wM@!aJR0;58yb66_L!n z@zRmKqTGl!RqM?F$Rn+*yo-(|Q==$xacQ9i6@?zi!cKk~ zOdDzD1W|hOL`%AOGq-yC?7M~R*^`rEK)4iw0O(j#S~ae0Q+S{L)miB#wlnJb@7(Eq z=gtU-R}U)MtM`-9p*e|oQdAG%Ul~ZpIL~{`Llb@(pYlvm#O}3sycrMVjI` z-dKWaL0QZbo#*D}1OEjFg>6?tcdlRuj(9tWUlZ{FlHz7aAa*YX#pTfyikl3B(k-BN zC@4=eh2pRua8L=;hRapV#G1hm4m=G+9Z0p_#%Y48H68GF`d;Ccb?OwFRlP=v1cj=t zdxg%ArI^-Wl(CDH*acim^tybNrf8^fwFyziosG^;NlwNCtP!hwWR0G*j3D~&fX;=i z&q^l97t9{wQ`0YGYub+wbrZD%XCBn9v3~8Kl&#Q*nXK%DN5DR^wlXHa^0@b>*@p^q z|K@n-@OVWF1HOydx=r#j?^*XWZBM-{x(7iYgaJ#d`HTx>Gp~zz=Z*fjs_tO$C{}}< zcky`MR0n)Yn9 zt7mJm2jf-qJyR=gelt%m%)9`4RqO(vX0o(o|FWrXS*-Jg9!Ejxa;46nGHiYSUgX}E z!#0sfsk(fL<_|w~okAzJhzDY2#X6CAuRFk~kUToSxv|>F#EWE6uWsEIS~!~Dx*J%@ zPk+LNcBuUE_3c=w&z<#8!Nc2}){4QFtOPag-TSu!_7VxHYbtKAQ*2eB^L&I{`@dcM z&@SO_Zo6>oCJ^Gayz#XmwcxCB?6?$^XT>Jd$#qXaTo%bRTLl7k0&Z4|;CF$4=$|4m zmZlukW~5FPh-k?0^63!(+VqS)`uE?f)YuG8ha~usfWbdZOe-}YSoP4?-}0hrc3V!# zS-8e8^PdL2EB0u`&)|(jTdAz9cx_>1pYc&8uSr1i_lX~dIX06fh4C*aXB4y^jfi=) zl~o2PuwhO<$s)WF?S+V*8>%QrjsT`4M?YLo#H1od#yPlxmpc_m$9HG(f5Hvv)9o{VA3dfGv)dsMEs9G**TVI-xrk@;ROZTiL1kU9uG9JlzM7gh9rf z9H?NRICmrb)}x5x08B-Q#Pr?f#4RfT4L*9z#a~`9HfCQ=8?Al|;k5=ui1=gHjBLzBegNx*8cZUUNVKoYdX2hAY!%hB}Ur zw#jZzC*xes-x^Hl?K;#q#%0YH?yz{O=&Jz$#Qy{^(mK(`O8&|pCGUL`Rb<#T`Hwri z)cM8B3g!;)A?D)qp9XLi|9LQ zw1CE0-~_XdHYbcnOsyKr%s>OCNT1zwP1^(}jTXnYoLJoh+l-rTFcTeptVXT)}z*e*2z zofzMy!Uf<@+|4-RDVd>&s;SDEfsRX!4h!23qAwjml+#v*t%?R_G6xaFQWOq}%@y39 z6#6hMZj9{Ie#f?FRn3__TiOz!mK`BkZ{FzP&1>li%ue+s1U>(10)Z$FDW9nxp*f^= z!T>WFW~q_|^vPdYjVmu;F!UpMny;^9Z+L*i$g!}6az`!~B=Q5TE_If*nr z5+Y=_0`~2}N7F9GKU9n9n)a|?IL8vM;b(Wvsh>7X>miOj?^pf<=mgg7Pzkc#SN42^ zk_UWcBnL|UTV8!{er_%o$nsiBVq(nCqme?37W1);ucE`m9@E_m7LeSVRZBpU;Z6RQ zmkxOn^wug+wyN`Mqg09*%G%qTs_Z_N%VJMRb5YgQt79fk^w0Pz)xO-ED#Eod5xvph zb7#)2TTyhr=Sy}@`9R@eOJ!QxayK{eH0TnCC=S&?41iCvnzhWl7^$HxJ*lyKI>abw z*M2LA_Go(Q9|Iz7)XXc5m_xYg)SvNt6S1fzO4DiY35lb+mvw3-?7}$3Z>^yCbgZP; zX~1BzOC044WvZt~wcW9l7G)h-D>n}cto9LmM$`%L1Xj`!QY!iW-Tb+KsjN#ULBg+W zUBXC2&WR8WUCb5qc=dtgwIjc?`fFo*luud8k$*u8(I&!dX4r3~;x$l=x0;Z5wUe}< z6dL0mn&a4peOYWotYce&%z+KYENfm0e?M~q{lk~a)hBMpzNa?dz6^e2 zJg~51Ro6*u&&@cxmHR~4_JMBJe(0KUOFFnq&Pa?0FoDxJoz9|BCE5wojTI||6;EDs zx(9q_B-36hE<4X zYtB=qVB7OJEsn)lt#oz$me?*rE<+lzQ8whC!=!D8s1(HG0{-@v?*HWy4vGb$eFK?-v60L_hc_4Pt5lRIgLi!j>-B9EcZ1 z+_nrDf-UhEQztclF`wE{zjdWax4a}% zrNMY0&z1P@ZVn~)OL&7bBP(iYa>@JeTD-IIyMmo2&J8_wSD1Wqm;e(D<$Tnh&zdtX zJon!F`kH)lIdT5HFNf2&vX%w)EL*yEAB^kP!G#1*EA2S1f8^l6;?Nd_l;rjcdKxwJ z4+y~JwD*&G-_5hzjC&JJ0j_*BsK&&Yp0DxWy=mO_6S%51*AT#8jQidnZAt7^qL49w&!aeTT$KN^XvdI{+S(Iu|;5mx7s#JZ3oKwe6UW8@ z|MDXA_)x9|8iC{*J{lrkr&Jyyav^Gz-wM^QU%$SO;x9dQdNu#Y>w#0tS4B5UqgEl< zM9tA5c%|R|Kbg$PkQm@(yRy^njTayCiGz;u!%#{b3U>iy z*RGNA05RU9r|9X4W*wGZXEU?*io3bR;{Bqm++6Z1hGJ9QERk$5l5g`qsp)hLEcM+- zjx<(?V1d@7`_)$&Om%rwff`qAhWz**zK?{F145h4u~no+1{G#1M=kNB8Gwa9k7>Q? zE$FFXxyx;g(ml6FaZC+>L@|2MI-+6X1wdX;A*dlEr`BIg1O5Hq7xsHZ041nv^3(ex zL=^?RcI1dmobV)J+4i$`g_Cm^Ro;u7H|cVTD0)mxHCR}Qy7iAi>L4-15$}}P#YYD( zq+~-oUJU8Mw?d~?tG+}m2Fj!KJx|>J<2bbFR3+ov*o(&<4#mMbMMuyg{PMIpbXLu< zje!T>IDlvs4{ic+>9+#e3(Ne?%B~4mjO-O~2o;n3>g)~g5rApb3#?sJSkpUwLCB8J zVpV_r)~#KC`9_~SSqqpe*tS7L8nYc0vL*|Hz1i+;HBZeSZyPD zi%dzoSknlu0b(|TqFgE}X6u%tN9V0G9}XlIXe}?9t>O1=Hrik}4m&+jHVAqKIIvhv0A=eYN8A^R!vvP?)_x*MV>W0Re%vKXjRXeRD|_!qAhO|Ty6sV1+oE!T?A1y7FlgXFX^!Hyj^@AuGx?>=`v++dVel5YPjOa2){38M-cR&_~;XOuKOMerChL6{m5kBt(3RWU*|aQ zav~nDV0QbW!v{CnF1~i%!89>~gk}|QeD;DrCo+c*61C@!vAEX)y-@$}w^9*rYsiU+ zlvmAeugr+vc@8@5=s=tZ;#p4l;wZ>VKIFdE!HvJ;&H?9lo^7puy?QkTO^Ug4MQ6z4 z^5J{dm^Ga-;+zm4;eTvYRZP`a6P`wm9J$%{+H+J}77P;Je`w>)3>~(Vw%k^W<7nX4bt*Hx(-iUpJ6lD!($C_TInD!ib!25ikz24 zaoIMhAvacJ*Y+|PGjpaMGWn9w^pJ+ln(0m$_@h(&g>3sklDAY4B4|!*zOR$Z_z#O} z(r=0-v2k-+?o^KIZO`h%wx;y}A*e&Q1P4E9zRUX$5099%QG#f)0L4nzU18|e#PFEO zGtM7~Z_tY|_sqN4k9m3>((R)mYhZaYIMsfeS>|NfZ_^&CltfGVf+tFsv8YV%k>M`D{kQkfH-wYAH7fgVi-)8(0Xm`*Msei z|CwIZBh-jLgOMU}B%mhPz{%+7=0q?IZP>S~rQc`7K}4gWE zTG7coVIm>+%pi6dyz7c1gzH^@9h;@YAn9FR-gf9hVp6dXc02H2-QM2>1VrmG#^bpQCj zP3x(qPDA|Fkyb`rXS7?kWOKt#SU`SBefaj|Nk(wU}l)M>1(miS-!0*_79W&+CcVNWxiu+(`kLN@yoDliL12;R_`*g_(^$IaN@CvtYKGl`sH+|0|hp}VHc zLofX2QE&P_tLlmQ%iVw8)}iRCKQ{0A_kptXqe;78oA6yn1LlL^ZU^0nntfv$m-9n( zHbrv>h=&5DayN>Z4ai_{?uP1)%bIHP$Ne;OC+(9t=efEjnWr<|srN71O3P$t_Y1>c z!4*J9d~jfc!fMiD1EiGLpHc5#7Mbvf>I3RB7pVh5y&k=dl(kW^CE_#_H-8S`<|G;u z#-=Ap{hZiVeuYvs-YZK9oho?k=%}vuJ#OZh8gFRI1Oia^w~fcETjwnL_3zD|qidYGGs zhQx8;Wjqq+M#y0~oa{Cg_YROw=RDcO%)hKIJn|AY zF&qHK9tEs9|0{#2b0jHsA!W=t%!Kz~4=3xosVy8Tp-$9mPXkDZA1ptKLS-5~vx(+M zz9YEsqfqluuj7=rejx(snwgn3un03321bEX%)EG;YWS%B)qGR+d8)5+|7u!w^D<_g zQq_Z_eLojejA2f)rh4vmPEnlrwZFBo zvLEb_OC`t@5AE+g$9Po%H;$EZ4?mZF2hC7UqLpRZo7S8kU=nGXGa*8-?RVF$sz+NczY`$Xtw797cVoS2RTB%wpG46x$qxRh3Kac!!2kef6F^C zfc}PM-!m_CS~HUc=Ohcl~K!%)-C^HfPi(OZwtRCQwT`x*aZ#yF1ee=J*DG2)+13~2)3 zu!55c@$EZz+)mB>Ve^mxI*}=`V%K|kOks*38(%|^>`Z+3lhuCyD4xe51ry&GaNl`( zM%sbv08&P=p6jeY&OCp|s<_Z+LWx)ezK1u;BoTXgXbu(`9e0CBd=8LtIzuO7iPrNSulvh!Xqp+XwueRC~bPhW7I0l~j zDfj?S1UHTPaD%rpfoUu66d@@^CS62wCQH;>gAIhyP6y|Te>V>>S?#1{Lrr#Aw%V-f zIby_d<}$G2eU)A%@9&FGNw5o{Ultm|K%8HIQ++SkII5xzN-vFifb4=&W5olV?i`?9 za4yX)7;txfY&?U#h~fD~1bp?kbB~w^nLiA@xrP+$-4jI)I!h zw-3wn8k=bIYdbWn4L2R%WFC$2<9J5v2Gl}&2=m3Cs04(YLLmQ~@~GGrq)1@Qr-Cyt z8wB-E8M+f8Vb!@Hw5^Xborrmc9i-{)rq2OC&2QM%egi@ZiK&Oj9#~^m%@Qyn5i;J` z)&2(`f?BbEc^p?Bbi%X3>mQPgSd6=v2fttKDn~0jVY_|`P3JgHY1D^Lnr(nhml#GZ zP9GBU=RfK`T}FS`43C#0LjI!;WsZ;o${(T*02a;d@E4V81wkRqYdVby3cCZr=JIVW zKz)2HiDu8vVNJ~i`F-+mg0IA9yVhJm>P&pIuND=QC=!7#d#;;OkqGTkPm z#jwRJSb^^6pDjT6UA5r0R99`y6UaXv7y_E z-}>08@%a`MIKi$G(>~IxP>57-xw`(;^)7e4YY8L8MML;SB3MU81H1~=x0XE622Gk+ zQLjK(xqH3WY9ctPSA-!wo$efUnm=&(@D3nWcj<3sEmd+zNAHsTcZFNe7lU%4IHBsl zf?`W`-GH7h^mX90XI(s{;}*-!xRvMd`uu;~c>(%yq~EurR;y=C_3lQ?LKb z1*GtrS!wgH&mg1Sk zMg@Zzfv;qD2c8uTXXspLMskGDF`B4}Re5@$9$O`hgb84C+$^3{(#C7D!W$1K>Qp6L z;s(G!w^+aZ4fTr)PzH?l>vw05TpPT`Y!ehzELWQ`xfIq-L^!s{M+kwOFf7!jUr{O1$ zpX6xa;Sj{P0VI^>>mEl&!T}Z{4WNv30YT?cZn3;zz@mA~4pfhIU`fk3f-~d?hnWor zeLelTp=jjm@O@R4=eg2CF z;&d?c77CbOEyK+cg6Hrud%cdlIamhyD`gb8~o==r|Eu;k@` z@s}JTTFY*7l?w>HHRY9&P-bC(Oc?l%2O#@^Yn`CS*~|C?H#=>`XV}VL3LmotaWNE` zTAeWklV3&aP}_pY<~pa=gs1!V?|(}C6yLcukdUk=Pd%$?MMum?9HKW~6VLo7nhn zW!XX+q+^HMM=%y5nQK>hnlMY}-`Tu)Th1Bhx#B0UuQty%gA9J)<8AHwDH>zOay;tM zig9Dd`T$^qp#&({f86@Z4QEgsD=Tuip+b1<^!Z#Cu_Bh$^UV8QzBX^%m~ncKlO8ki z3ADZD+lbjG#pzDPZdy<{{d=g!YqgkQRJ-OUftSE?-nTEDHd!eC>M!%({&{rgLfq5p zdeh6^XR|VNdU|O;s?5ZvuRAl<28J2WAMod?{A2!BL&YOjH(n9@#xv2;Hi~evln4#@ zT>6Trn~o}D)S`$vKO1^@#Q#(L($^yS)1dOBf zVdm(m_>4hN);aRF7+(MBSv_;%+U@nQ!!DTlCCp%O-&~VF~H#dY<-3mTD_2C<-P~(WkcR zQrUW7BKG~PwDep#dtu8EqhS<`jVg07+j+8L+WEte8@^@D01 z(LIzkZRSjCDp5<%t6!Z>{vzh`%0@H%rJy$gF$l@rlMha^5M1PaZtf_eg}8*m1V8U! zGz7cH9vFnm*QGfhKtvY42-Uw++*zuKY1o0M;-U_3)tV&{&ZR7!$)G6@N4~Sb-CY`K z7GdiuCoO<`HXZdnUCM5}IRA5=GAj?V%M=N;od#>M630H zn**2sGkd|)sRV00BK+1?t&p_1axY;6a*g)FMPdtlc-Xp&4j20{>STJcRx}oOoCpp8 zVcqrzoV3bo`410GFMu zG??qAU!7y!k)<{*+ZOt{Uz*5>coSfID)bgZU320u1TEigg zRgk9{`%pam0s?f7n1j9uzykUzlBqj)?yP}QuN2}y%Mhp5W$?J!xF!apK!siRYtkFx zm1e>H>WWMV<x)Vp0FmQSpo0;ouLWbJ%EhAN@`#Pdih$mp0OOC&3_x3vvJIJfhw-fb9?ZmzCVxs-6h zcWFA36WQ$WAaPZP2mcWwNnium5bU|9p8_>c7@TF_Ugj20Z}hXuTWv&TX{r1RTp_8v zji@U~0^^o6joTsv_rl=ozqWeyp3bUfj^Mjfvb_6?vG+5+U(-<(3IG41VZD z!S8p4hJVOU8@C4p^clF5RUlFIFm;4yasQuJRuiJz0t=xS%egLzd@ACu2xi zP9*E`<%AcNS6y$6hb$Q|o`#Ny#HNiDAxzr8SlxlA7tHw&%etX!aA8TUx zJ44r(7@Bx7X5l#o)7h34W`&BmEE!T9oD_H!Zwm=c$(_n9ZX732dQsl*fK5Nw&^#OR z6U2P@%k}`+S6YfyYu21W1@V;O=OS9CGpA0$O&Al^`i7=5zY{UO8;4b||8GsiON#mK zdBz`jL#5u#dBqS;PVrNXg1R%*I)pmI#!vbp|+1H!3x>&(wKRr89&UqjIMd+I8YEg zG(MhJ^GoO`*mx3bURhP04#MVGc_}tF6@^=9MtOA*V7iGpcP}MfINT6PRVy|E5(2u% zI#u{ft8NUBe;c5-6q=w;3vss@fKx3SocfDUhb;RgMMi$LPH@&B%I^ig~p@ThD}@$|&GDa~TPPjM`o{ z`~;T@duPz4!wkG0;TsF5Y9fXjQW$aPqEx^`A{N74L)J$;gWM~LgN|9gxZrx{x*u@jp1{Dz&8vmKsFLaLxXYDlpRYGqIX+AZdATm0P}B}WfVXGA!T(uPgv@#g7J^yNOH>v1qTBjA+y0|)4p z-5aJkBKq)Grr@Yy^H4o_b(y^1$Lqz#e(Uuk5C+LvXD{eDA zL9ug-qv_e}ujGcbr)}Ux3tujH?=dQiZX7GwHtym@mRhuU@iZZHCC1a8-RBD6x|J$q z*Z9xusF*QOywtel_pz%8@#4^$FL`ho)&pn8^-7)Zpd1n&Fx_aOx69w(P(JkDCp=a| z;HHwnXV0A3jnmYlgQi;n*QDTP^4Vj?i<4lslrdw-Q&B2ZSY5bWCyy17D4+Kppo>7k!wcgxT&CHQq4-)jMr=`EDBJ` zNq|Cz67 zC>XSa1#$bJ?ptvSWIN%@_$A=catZ0N$I_UIVH>x$z5Ed|;oi-%H;fTvam?9?hb*CL zfolv96j0h1k_;hGdH96j%~TT8HR@_AXGZfX=~@_0^3}%Fc);m=nP6?5%Gu)no`)4I zaU9Oi8DOWq_$*88t-)qdQv)-KIC~OqXADOdVeTi>6$-Iad<1f3e8vSe=6<{>z`Ef4!f_~j%y8g7 z*>Hi`+KQt2=$lUhiuNlfS{{Lg^pV_b=>o?|I&!1Ty{ z$7{rvrV#3lz;8%btfcMLsAP;aG6gfRZD71M zavP}5Zo=;{(OerhR2QDf9J~U6CY`^O6c5)azI^@WGe3R66tQY~`SrsFx?5t`w8e`f z9EKhAEM%ixOZOKkh1jJuhF$A>p*-<(GWy+hn%zRKztkpuUmcw+m|lmJS%^hVI9KDmU8oF7FM@h z<$7oI(;D_FZ8nW%h{LtvHF$=oJD8pNCf%awzcKp|WPPew$O6_vZW^*5wFTwEcMgV= zD$29Nf1#b57{?1IynG3Fr1y&2EQRS zS`P{m*8PGl7PgMJJxtlJ_1!0zdDCyU$^uZ;kY5tz;4?i*mPr`@-p}mTYNiol)<855 zhn8)A)Wg`s!~kr{uGcflai>KVcv?=T%5*L;N4^!-(uO)k{3bIsMn~NM(q-kCwpK6` z>bu>xGPQgfGyh(jTfS+L_HQyXThJ#l8f;K-FbEL^pW)KXht>aAh4agtI~yI*c+hUX ze*IdUPv_0EfWdTzneOIq+GJhGYU9P!HSb4Rne45_;eu&GWn(p+zZ_TDW@Vy*{Toi9 zV`atX>NDj~4ty<@=~;|;1!N+D!DG8s*LXU_%)J!(k#!5HrvU6uprMr}>tk_-0WRD>`4MC~!$nU-&3@~abYWs%EX@+sG}HG7N>hIpaGQ`r^D8(nTl{$mBW zK3$Eri3aHKJY*XyE2$AS-*`i~XH`Vw$Zg)tWcDJN+eARb$-fJtC&nUz_z01qAo?A( zDIIu9&~~sUSwdnP`Ga%Jfhy0@@tr>FOn53fZ?$7!OQs`sgAMV6NmLbPMi_+8sXO3G^O*NnjXOpbxFW-!GO{WwmWR3N?ll zR)}ifSHZMe#86FtZ>>XcpC{9hY>We@EyL`8txr_VG*=!OzREp{4(;dafcEJN6xYpD zX8SV^P&s=+$)rpxZk__s!C`HOs+KQ{QvBOa|MVf4XY`Es)TZh`_0kd&M#!>`ZOOE9 zb(Bs539uULk2Nri#ZruSfT(N8~iY2`>yq8pJRDz}eO?Bz&X# zvh&^}=uo^aUPaL&sT)Yr1wq)2emehVwhryuO&Bp^gfzKcP3UG{wp(tzn}i4OCulbD z{f-msXeg7DrZotrg8|c$U3i7|G;du-?TC3}@uL^E3-B_#5Kjpv@AA_)29`_e)mF<4 zIF=*9_TBv%Kb(nfGVc%rpVPl?nbF#jXQVoZ_qflNEeXw2h)=@QPWD7I#gqz*@iB0I zOI~&E)2BHRgb9pwV)GcUrerca*bam{?MZin3Wb>?82o{qox)WEk<^R=sKaR@N(Jz5 z-d+Qv4Ge!5Zcx-Qwf1#-Mq*MsFlvMOQN+|!7LS(Ww=VP!$4sf;UU>W|5tLwMC^8^02Lml5Wg`-iYtD6n?u(uFrI%k!gO)zpMp7un6go{i~+PK?$*KnOIEND zGv*hFkBR&sdSqE&%uyK2Zw0vXX3+DF5(jU_{x!WEZKSN<@>st>a4%m|4N7PRcE=P)s5NuZe8N?_=$uU}gh3YF;@`tp}$EGwK*pg!n0BS<9( z6YJCL=o=c=1!MldlCV!oPR zW}h(NNmhf!2(pzb&aZ=q4a)`iX)|P%p@A3U9g%E9cYx0}>TdM8b3T;(;#+|gpxb-& zXIXL5{E73YxX^?2|Lc*Fqt3cBmb`xut98*wXlvKzRw{>U8VP-vRScd5k89{Ve9(rK z%q(h*W*NfbLg@0ozBgR`lbju7Gh@L<=}NabpX3Fa%FO*S%6M$R#8(5z;^2`Zry|OR#x!Iy7%mv2t7)? zGl~8qVe+gqm)@n>2!V)IJ^6d6TjYy3(JXalv&CX`gth{4nEKtR9 z_keSU0JS77{u(svUbNUZfBg7SxS^u*@@Dr`W4?3(X!Qyj(lIsq(cvJP_qd<)mo9ze zG5KtQ)85stiu*PAGiLD3d4M3tkZ(`I39G<=3zWGboDJC_0oB33ntH$1JWQKds!<>7 zaMAP=+_U+FnK+A~$1P(1g=(XJxu}Y$F6AJAeVwFJhBPWB$tsVTq@bE~2ouDD*MX4T z2e`dxwX4WJcrgVS4Fa~HNLuW9(f>@tMvW3gwm}TU9IPD|LG;eNFWDM`_QulM`ZG`P ze(|k!%Sb{EWX{DyVr1HNfaGx@>&Sfi^4eAQ0@AAcC} zw2syj$Xw#~-}1)2ayt4DJG-bxI!oJ({u)~i37F|rY=queD{*uSE?mvHa_z^A&HS|F zw73QLwc!p2rz+vUyqeURkb|;Og%#4S{UHFHOp{F?2YYkd(CDsbTCE zprV+?yULet_oJ3A=BH)Am3zZWzHB^4G)J zNOn<#c#(Y1&;`ZrdCyT952|GwmQJ^eW#G-!!b=Q&1Z|%4=O4IUP+SwRWG&*yVbIci zKvE69SI(J(``z$y=Ls9H()1Vq}YO|z7mk#VV=2x?DO zoY_L36Z`q4#SHZb=7o+6u9PreQ1S=g-Hy6}?pT&8bP!f-vu!`pFw<35C0M?S0-1?2 z4)A*Gnvj*{v@kG=yBcMDVAESyXMz>~oH@f;Qc3@E{)#@{_Ym3ZCS%Vr26Y3Z%q?@e zK~ShSqCfOY!4N~_d@$v;RG!3)gqXLp=EN}L&gS&aV>EvmSC?e<@>>&U$9c}_{wo5Q zK6F~9G(U8SVxza+`6Te2%;>3|lrSKVv#O#OGgA8nV_(Uu6rw@lsfbvXpFPDJ=F@+0 zYiz`{!xYjCO)qb(Ty;`O--V;<>*o1$BNI>}aZo36p!6f{V5e^8koABQr1Zk%RwlbJ zQgrp+!Y>w>1_io{$k2r&d~_HfX-k>eI^93e5J*~?8sz!8F;@ zC?;hY>;bd7*hNI+s%7eJ12>ZVf=h35mc-KV4^%)VUic!|j2kDjY~S0-QuK(C`pj1X zxd)y}zXT^oW?EPZpHyOTEdL#9V9Xy)#(zlL_||nxN^UR`7>m3zVm*&wet}U9SkLt* zt~0+G-?{fD;G%) zk}rH`8__Thzyy*RsB1{;3RIG+MpWSokwpuan8I6zw4fZPu|9+@+im6bz?SXXYfKS2 zi1`m?Njp$>QBwey7BTv-q4RP3&{3MK2n9bq?fPK9ZHX)nAT)_ps+uvMY?md|G4vU{ z=3@)$ukFm^bD74ds$_=`H^J7kDJY<1GuR7F8RwxVkUUDQSzm!4(6qbF6_7P-|8e~x zMqz^op8%05DlHZFbHty!xM^k^%A4TfuUm4B-<)>KwI#v~X3fI&gH#<88(UjT{B>Y$ zCUYiMvR!#|6^e+2r%z+8)ga4afQ8b2Bs1K1A3rt_X>UbQEgU|!F>r$dsMW6BgSa4c z(6U3cs}~GFG0mk417Vl)MJU?=ltycsfX!`o8TvHs!wo=%FL4d^OWF0!Cl zzB}}3xVJ4)h;?7ngs@b1?blobSHWd;RrI7Wc`IMHA^zZIrp{4icXMlg>!db;Ib?RJ^jT>fvRx(;lq03 zTE;;^@Ot9S#ZHy^I7Neni)WlHVWJ@K>-M<4vI1DhpM!jK_y%1f2OVJo@Zda3eqHk^ zYCYM}MJTVM;A$CO{c$b<^}+8RfBazy7BBq>?j#}|`Bei`vqk&%O+Y&TQ?jP;A(K7? z`THjlHNzM@z#{CwczsfoPE6ofm!;R+{M>`^LYg$>BY7ui)@Gvste4mm0i7QRpiw{Sg>>< z_^|G~JH}RRZq)Kl$(57Blh=t*iwOnbBt}=k%1+(AnnKC3)HzNoAxk^pEeXqTW!PqZ z7-F1>3+mnHGM?l_qZ|H(e!dfErnvd)yNsW z=U5UD*w6xoTq_}O}cSq+&Mto28|nU1(buTD9fuzbR!`)2HWiO;(KD2 z?F~IcjPpuAOtL#x`_7Lb8hJKkvXx~F;i8^*gt0JPg;C8$gv zBJ6sOetpFf)F?|zo9d{EPasu|iSl?J{34{Rq0y4%#j%05>>lq}94n({U7Orf4F>U! zdRYD@jf0qCQpG4g`yYT|PY~!i^YXavx-_H-Tk17_m!F@Ksfo<_L?sP>xIqAep9C^K z_sv$lq-9>GA*;@$R;~N=v>sri*wy6bvL}65czF0cY&kN)Xvn6Q8&HS)-2)-WQ?u(c1}>|;_3elelUl?HFBG19l08Gk+LU~ z=d2(F-DWu6SgbLqsBB9#w0)9Gu3a0;KbW^{Ss?YQ^uuBZLI=E3D#t29uHWt7 zTGjh+N(9V&iXiY2PK>}LuCAiT70I|by~zSna!13z&OtjRM@8%r2r^3_e+ZyKd7a$n z`gbp6?<$$FO7bEDJ3U~76sB1JvnJ@YmDcXw%a^(wYw7t&&284HTI5A9y3ub>;C4o3 zb@GZo0`iXbG=Id>QEAsn6K&S5ROu9Bd3s^Ar_UfZ*l}^^hm6(g(4i^68jY;e!iT_C zpz5uD-ihiBC-gT8earj^I3Ga&J^)PmLx$W)`=WnJHE@Xi2m3DAH_P!{+WXd+;tfBR zQe$mQ2XSp^am8~-^G*E8m zcw-y+^-d^Ru9g-3LT7A?&B%y;UA1TKPfI%#Bri(UAz2~ast|&_nfpUbLKiMfp<`J< z*f-ZN1nKUjBQpxiU0QDCgd_|75XN01vE?umZJ*37PE)T)tgjPs+q6%(DFNS+zAq!$Scpa)$F^VTZVm!$SLJxcGI{5hAcgM=tK_VeJ8^4#dJI zdC{B;W4vVkvvgjeBQ4|8)5~|i*Z_KpHhfY@Gm~{gSA9@4YTZ9)ceck2?caMis4-2=8dd6|=80`XES5wOZ7` zpeSrUnNk*C+Z^7pV@I|%rNEqmq4uZ~ElcH`c6ZbHk_X1><(?pqRO z!U;{W9aiuPNylZ1?Cdzq3es@$UPXMO0Bf5TwI!ouMWQD%4tdrR?hL#L-6RUunenTu zUxz5!P>USoPvCc<$Xh@!_oLr(-tHl)D-j-H$*3Ufm3ilIlM)r)w0&506n)CbD)AJ$1v0N}J}@%ry6; z6?Kn%d?@~ERm2c?h$VjV%O(z09OMTPCx|^P`?HX0ZAz%;PRRtzv}m|v1he0oBr-(z zGm5Bvm24DZoNsG;@qBMQHSqTD^mzkW-<+qu2`ZTAUzvX6EL)+=gR#!uon5Vyx;4|1 zswQ;cM{2F%``Zsc5~&9H@U8OKW{&;SjaSrK+PM(S>$7k1oEpj&c0|wi~uQuJ@UQd26g@lX|U}C(6TXnk7|8BygES(*! z*5m337T>-roH-C@=nwS*R;BO9rmB7Pq!z=R$gr?l(%9J2XYIww@Pw{Ion(qZwY{+v zw&j?uG3bcsEiE28tz0=TcVc~-M}YCrBZM&B zuzWx06y*swkM+Zdix=nRewp*1ZRm)^5pX3$6lbvM0{CZ^H&|k$F=OKOD_2lja!-Wj z4j~!(;>DG|RfCx)PM}f95!-R&&LzF*%Sjr3=~7cT1y!_?JZoHrfzePT^GM1%ZH{t{orR4S5SZ zNF8AD>ZC}DqN4QydH?@Z=$#<20%H8%VxdE=YPT|PH)2Cm(mG>U6z&P)`r}@wLfRXt zM%gHFZ^Cf~Mgrv!dZNjSbIixiJr4D)*qd;$+RpDu!-u9rCXoq_ci(pU^qLUtlA{xH zT;9GAKlGd`$YYKMFC2_!V5Ivv=-{Ep-HE1G*0X0mEHVqp!a;F*$Yv98{@ag^S z;i6Tyh$OzdWvh{!28vv(Yz_$OSjyp5xTpggcuNK?B_ZiZFZ7k_JE?LS)F>*WzBw#dV38ZT77+yi{M;xN3)u2o2b<~Bw=<0B+ALRaB0N?W z<%Oq%f(TQe!|J9c^X*aV4v|JcHZO`#96xR{W5$(<$`eF4lf1CB{NEoxLf;;sC1}Fx z4?kjSX&E)^MYMFRU(9A&N8RW$twY8~3WJAEc6R2uUuF)f9KmX9{>jR`IrT#5+bE50 z-j>|VWtM4(-S%^MeRnuYAbG;v3`ORKpB!2?k@mH*3Lx;@-hUEdS#;fzE z&}f7F3*7<%a8d5>A0?M3@8FAAqSJXaSER9|j35`JFgc*PlB?Vlxn1o}DF0je{^hD$n*J=I%Yc{B(;$z z%z)L`qUhc5GwsZm$z0r_TYnA2buDAl*FqgcYBx1Zpwx^7`jB}5e0-w(wmbK1@Lcsj z;eh}p^SS;ZYLBM2lUV`M#8tCswoi)&Y6b!`*j!&7Jnh^gM6(iHRrWO^I@IrL_dRT7 z;V$WAHUAlj_m-Y3p1!`-`Wm&n`1hERUO^6$i z#|1`VZD#|+(ITkdO`Tz1Z59wsx%^`)O|Kw!lHo?@Biug+>IVt}38;>aj!xamTCmE^ z@uus@rk4=(6PTe9Vj#|dEL12mES*1a*R3f?K(6OeE}(3>HJfWI+tc$y2Q-rQW=pvi z_owOn`2o|Gjpm}WnLXS3ftvmg2W)NTQyEh!*f}`VLjEmn7XM3TJ?-3{%8e03pKMty zDcwe{?YMJI5H|74_RuPlsM{RcZX6HVUF!?pIcWW#`sFnwsZ^?8Dl1}aPgl;x*tZ3(JANPoNWK z7-$_3VtK}YkfS~4 z-uRQk`lHh(#GYfQhh}yiAzJeP{!z_@N6VC-jiZ2bFO@hdXvU6QNkorqcnIa^Koc0LUWYAeso9Oyi=fr z->@X$X7#uZL^5RcH{`j&{rV2+#UPz*7L%L>OvP~%MjYcXsI=Vl_$VY=XIlkND3CR7 zDeOTVH{hl$p5V~Ur9Aub@#zVh(#=}m!(osmaVZS1YJgRAWAFx zaJLprug!wX!~r*>(hkd+%7}LkluO)VC}L`2Hv$`s2j*nnsWCXnq{Ta?L*VQ@XyS^> zXIO@lqoX_Zjq!iPFrQLdbSI}itE5aL)5qe0y2n50QX%hUl!Rp5)vEKyDDF_<+7&V{ zM>dt!gTCTX`tKiQi~rj07~)W>1ME4>0jqZqu|zn=KEAS!pAOr z)SpO`LyjK)oi1eA)u?0noMY&`WqZiz64}aFxdAUT*-VJbBnWfo>i#(7bS^Ck?Z0%> z#YYkIAy(xLVw9^3=P7j`W+Ecs#;^}r#N7g7?q6;5H2sVT7{^IvrMH5=O542m5OhSuI{16zte$qwsu(7;p`pfD`6Yg_h(Q^6GNMkls5OppX@YiT|K9<_<; zS#gBhP9ju7z?xL*O2E}8EIfVBOr;qmzbp$VI}REAbNFa_1|?=BFKuWq&>1M zJ$LS$^_z$b7v?&uBj$6-!v;N8Z^gq8=#)W}|0n?;NAhV`< z`3Tc7CpP{Kxp0cXB$FYjR7YL`lLmvkdsfR@Wm=MG;jqz_MZRSP&5R0-V7m2Tw1gAj zLOa5`RF9G4`$$e8zx<>BB3|D>co2aB@HsUzFW>T(+cmz;Y;+zo5RT{O=6aHo2dB;& z7|RiT#KYf%bv!Akh|kblULhl@OvS5)Cw}?`n{&ftvG6yQ?p=c)-QKGh$;;CspyRt=|-8Wlfj2OFM;I@Sb`r^ij>c z1HAdPmw!~JEyv?aQUkh*T?;b;!4TOVAzONHW;@y!sp~H*6TniGm=JhlxGK8n1c52Y z4IGz6>)s>J+HVoIs5593r-?fl@r|j4VH086eh+alj&E~kXmT%eg5ZtDX)Py}*8+E}Z7wO3N zoj<=Ms*-dx%Ti*TkJ*v~e>0}t4kQKy%loJ|bFFpP)&01BM3imKlK0a%p0yk}x+Pi! zQlJs?$GG_T8VarsO8tGp%dBj!p`eN!%k>SsDv6*pn+i7|LlM$(9R73tdcRWgz|eZM z<^%2KZiRJ8b{f};tUE+3_L%jmm2pyWOF>MAMn}A58D2gfeNm0S>=o1!1sf$!QIGkS zS1a-7y8{L?Eah?N?f2|0sbBbD48|&&eTzk5M>i+oh+(DcX?RDWCS}@bBTA}mF1vX% zv77}m+ay7f%qUiC@R;O7A;L{@BHgcKk+jOZ4tt`5!IrJ zi(>%QBv1}D#fAvaG?aVJ<^et-N8ut80$sgUE;tfFQ{f0W%$Wv>do^y!94-olGiT4%>}HfmHMa}94fR4K;X%FLa@WQi)kT&md6OvWEUr5) zc0_B?9_T~n0iX})2#QSMV9()Wa`VSAX5Gw|hLmdn*Jme13=EOv2ho|@R1m(g9b8kU zxzK+q6x?kZb?OLA5pJ00R-Ez`z}A5;jCi%~G}p#IzrJd+w?!|l&_C|)WF5_iSd#+R zqaJ;VG{7|eovu!bz9C{sj_4ORrmnl=4{5a@T5xH3Sc!7@g8M#8uMN9ECE%<6n?G5# zc5V2P=wTP?K^>v0isiq0T|a0$oIElB%ddtn(#=KCeqe47{nNvG86@s+Yh7^UUWT<< z$RC|MpI)lHBlPbdb&^v4`tmE<>91cm%YXeFHtOcv!sxbp_w2EW`kIS<%twQ-8;8<* znvgz+56jFaMF28Nx}%P|x@TW!oX=Q|FZ%~y4o-zZQgQrx#bMjJ*y-w4k6wzWTa%PX zKn;=8L93727@SWPA`@>c*9TGm#z2lCJ-tM@Ht2&iIRPWaB3Z8mZx@He40nvSiY( zAkNzbs3n9j>oam~tA`Xr+;q+i|G_xaTp>?;^3lG=3*s+-f$`dotig^lNJx9P;*GZv z3kfgFK5?jQQ}yd>AF0q`>J%a%mAF$3PfbH@5_a+~>~Y4*NG@nP2{hzkNb0`?r!G^| z%m{8K^Cr{=lB>(-vjc<;uLL)tE{tbqB4er%pjxx?d5F0vLBB3tY`POCjvaNHY7N~= zBVZ>6F3pU}dQ-?pIIy_qj~-c{0a51z0I`TIZyGCd{ZAvp4PmVoT@4O)jLJA2YBfls zi#8(w_&ntqwwA%emNlj;ji&?>o5VCS*uzd4^&N3EH1Pq!z$Ug#Y2Ie#L2uaWd> zgUA)od?19ppOl1Lt~QvQJLi6nA#i2Z)kC{@#uAiQzO0NXMc_T|Hp9-2WNlHEun&g- zQr6^nrF516R9URclv9%byMo#%-~eO%Cxz<2D1VDM{~g`+(&%w7%2%#gx9-B<(Kqs< z&a82iTxqo`TpzKcWz`#|n3VJgnRqHzo4C2QLxleF=*SQrFv%roeFZj#OBM%SN#ZyZ zp%H#DhxxYC>E=FjL;onwog|EaayiiPKu&Zk$>@_YIAvuUbaZf5S+7wfKQeZ-5a6jQR?-(X;V!_48%h_eCm`L4LY9Od)gG#Z8-5Gi~C^O zW6^!4J2o|?lNiH!pchoEzd?+)p-sl(eBzv`M*ov~?AtoKUpVTCA0&(=RKc`NC$)jC zWX@+``R>IL{zzLJmVv6&GqSqkLTI*N`SQsUuYuT6tkLixLdk^Per5#}=S{=(pXJ$D zlu9g?Lj(drq*PA@(sxBqtD_KuD@Xhp0&1SQ)u@Gw+Z|;ZD93(ZHQlDDq$ropfeGTk z>`F*8BZNBgi*Arz-b2l%MFg`K9Cr+eX`lhNbKvUon_ttdMs`Q~2PlTdtDRUK33C*y zEE0!l1?>xr>`5HZa6>x3_`g&JI%plXvZOD4O_KS69Tf{!te67F9W1E~7*AvbD;fs% zfIY&HOqanQ5>uc~o`rN+J^Ua+3sh2{k&{Q4bQj@Tz3-&{eln+x%D&^wU>md!oiP&w^K#(_)B`M?(c}7iH2W(hp8#AB{t%! znjHG205Lt3*kOgD&2S9jypsV{Fe`)6iG}OeuBx9mlSbu1V&ZP1ROl9F20X1Kvw;|@ zy}rUHQ-O@7l94V!OtzALLnTmOAyJ;bYBn5z)>vd5lAFMi31XB5nI$b|uZ!Tw6&hHg zX~cES^?QaR$_KQt+9`O381;!YM`1|_V3x@o-=q$HbFBP9=-IUt(*J;5Njif}yMq5RrJvI(ut(@~*VuF*x8oNuSM^Us zV?|-cuI0t-$S5iWQS3w z7Y$$3n4BD$yRCgKGBOZ=|F6huGwVUal=pYG%EWF};~t;@O+UHu;)eXn^H;X2P>#9y z2rtKI&U2YYi_9PpvD3lmXd`*KAXhWjkIe+CP3APLrx2wuc=9+_uw-QMmvnG8Y~tYS zp{D<5E~h8;`Vua~fs8O}Kv)wsD8%f-cfJLL#Q%;j07VF9*M3)CyLK&ay`Kz?#gsih z{6vbf#_v?BeVh)`rIba>IEMMzaTEnJdyP6594s2-=zo84;b?5X%N_HwVJgqG=H0Ql zNy_fdhx>vfj$G>giaaWDd#itflr+@XA;!HH0YeC$WhiZ?8kz5$Z4>_DHC>+N<_d(y z!B#Z;9eGq^I<^|I=|{J)+KW4QOJs*C&Wki8{##lNQD)}r=})yVAVFKbo`DG4>TjZ~ zweswMNVukFd(f5^X+tv+--NdR!r50_Yq(3{`(AxNUj~kCP_N#60OI5;Jm2ZU9kVSK zFQ?;?jWyNY4s5RFc$Q#JT%0^g!tP$2nFKWj!@cM2;v={ZLEppm=8}Z61W41Q**uxfi4xdPD&yHLY*9oPcP z8}i3+-65h@K=EW^VzP7AZYUgi1+sGYRW`>J@)Usi0fD^43YB|m5*cYXsK^GGti}pd z!K7QroO9(onPv9EptKh`90NIaS#()bU&hD9)!VZEvS`#L&;+@T2#t00jS`k8XF3Ta zBcECN!$D73UNcYDYEGT!%Q#yzh7Wfl*`aDt9;k5qOP7DSAS7%|SZwiJdL(6B+oLh^ z?>Tq293#(SJYC*d3e#rh8>eTo<>RtMV27~gkR-)S9ZNPsi`j3!pay_)?~S4N0GGok z)7)1BICS##CRACoe#HAP+6YI|IEI)(GGpUwD(OYYQGVU@*9DBI>)&9UdkwAOl6=+FHia|@%btJo&H$cN zL%a3wKfU9fJAj~HO`CwMZ!@UF8?G8;48B3?#Fb(I$c9~97~Wfd&R>KeT-oEk^jCF0 z0rDG_x`d;2+x}*7he04G_-K8dhOJt+E|`{j+Z)D?%I(R9p4qEu1i5su{9{)37Pu3# zKb3l^bp>MD2}Ji`tk4}|fRROH zRG!O?D#qo=GD(R-WgH0DP-HZm`KU|iD(s7Z!h3^$vaDwCW<^NAm18P+S5~)f_p*yV z5jGh_dvgGZhq|H|W{-0-{`FMTACg;6Ah7^8puC9>(Kfs(>hy6AdFR&t;V2z`P2YUR) z@XRLtEX9xW_2vG5Z#BQR%hmv4scO4+Mnonsu5Y{c?Poeolbj=vx-p-&$|FI^H;tpr zrhM6^i<4{r2G;5Ep3$~EAY+1itSm2%+x-*F7vsn!;EYc+7QawVe*UA3Evku-d?{H`|W(V1`Qi7p)`;TLO72++sj|{r}_h6P^+X? zzHMy8_g92%6*j#V&7dxN#8h%(JhaOOc-2^c%F+yZByoz2<>)hTFh`nTM#vm}GcYDC z_}ZVtiPcmWW#zs~v13tK^&{Onuphe-Y{Xr{xR6EG5hKU7CKpD59RqXN=hjWiI86aO znwD(wa+**W{oHHk9KTdcGp&wt?io`;&Ahz}UI@n`KDtSZ!x4aszMzMAEZAf+Cor&N z0&_~JVjQ>r!JY<`aO377ysV`bGp*DLx<#~CJJtk@87@2 z08)zg7*>!!+iigV>17p<<}`IUV}L|4OhH{kfwNh@y!zk1x%?@z+XU_kvJ&qaMNrQX zQBzwk0be!hn#EM~q7

    69p)6$ZJ-hOd7Ltop2gKH@`BAqCNG=le>sEG@0ZDZYuCw z8l70w>o-!$k3W9A!?IjPlpsx{f6^+dHdI1-rKU5$0`7mDb~D zifYv3p>+=aK;Od1of;g3{Ds`2Ec)#Ls?$|BIci(aJW{28a&5BTP)(kd)Kxu1;4V6fz)0|ddu+)UoB#YjHKYr+{=xAj1H|V8XbSZN zm3qTidC3Z%kEG zjv})Wq~_EqFI)pP6rB3*VgU3mS2hQ#WhjRhCUTLgi>#do9l-f7b6Kg~D*wMpYJ*0$ z_(kX%$~1{Byjx{t*C_g|_6}wkmY@(hu&j&&{cRh=F44P)Mf1Q|_vqBf*`I7e&!6u! zaG*uc#QB-`+>tdl{f8)lHk<{nI)|%zCJ%MgLA4huKt6Iewj$t)jyrI z*3MCLr+xq?BrJ?{6cTjb(jyOI5Xy+W7p8JmpJ9f$RtARDU12 z@M~RtWh3xH&?OVjJf^Q!RV0p@J-aPb<5?c09 zqXpvV_duVJ#p$DZP~5=iO?(PQGhSFNpaVIkh=nM3pe~c)MF@HxrulQ~ap2laugjOeL|}Wcr7nbn@ol%B(SB#g<5qKGMj3x_MIL_CF33QM(0kQY6vIV zs#)UZfrN-M?WQx=0Qxkon`-M1nVH*go7f>{Lg#J*H6x>#M&J%73?sVYvyhARw}+}?lt%AgJbSi-&O1&uDmQ}r#CXAKw|`U$>@9I$`Yrv{uZ|n# zM-Vj)-jf(QFO7t_>hXqBGK(P{R5gCN$}I{v3`kQM4hBXqLXvuA$ z3)eLIo?%ocT1U>V+s@fz8Db`L`(#!d2xPefQV&#@&h zwfT1j&F)q{PTWa`{YPsVQX%OT*g&KL;4Es|rj3*xSE|CRB4Stgk)G)NgelJe6#yhv z{C}AlGUFIL{sV{acQp0+)({Gielh?f!rt;O7{Mj`_y&^b8mX-is}^7s1z9SP60HLl-nYMNY>+^mu$kx<2N z38WPiK2n*eH?70sZ%lk^f#^#mWrH>_!Il39BVqu|yJ`V9%ZS-zfyrXz za_nC{=NdoX_gA*cBE7SBa^7RXUNe^YDn zdqzf2{Cw@udXk?gPqVBC&;MeJO2;Tw{W2mfY%j4GR+d&Yu-j=8>;*bu3Y@diuwi)e z>3SDGJKKox6I7pFU9e4dKrh}NY*MR)w5ns;9?FVV3l}cjz%-aJHx(F-%*Q~7L~2dN zD;Ww7wdbwg*nOo8Fq0t;baL2c_r617EQd=c5>})ZG)RcvgP=`rf`5RqL0WgWAMR3b z@?|+%Ro352OA|QFWekPu(?Qeby-Uo<(B>1Za(1@McVaEe;60quj94#mVvv>%#%>r6 ziIzxk<;;>yl4#?C*)o0`EM-sEm&akFC=`0H>eYc86aCDdaKm}+v1ISy#IbvDuxrgC zD+)M{Sq+Y8875JRgD#UJ0@^2J)QeCa+hl*k_B3WSADnwJ2FoYcmMk&OrSfadG|3n7 z=3aVPXfErNJXNainIC55v-II?wI22p^Ep+g?xmO`K0ds;IstSks*AdlLq6vn1A&px z0#%pAU#RIox?3EjvW@l*w%78sOU5}2nR(Nf)6mMZ3%RkLa4yMdyLdP`GDtwjqH$?=5zME+3@g zUuKD7a=;?5m|tlW)%{3*Wb||p#l$`t8k}$a-bRFjb-!O)U#XQnQ z3~)NoW5AN9XS9me4jU@!DB2#yI3?L}XeYITy*zz;zb+WzO{f8M0I-DfR<04-2lOu*iLb@#Bif)(maH2u$f|9y56V;PsErz?Gi=-;eTNd zGW+<6uK-$vB~bTk`?;(Q8Fa3!5rWQ>{V%QT_-_tL6tAlQ>PT8*u5f$6Jh@o_ASAX% zVk3&FSc%Af6vKtvkJwl#5h!)Q^V~7B;suFMvPF+h>Rel{MCpSNFO%Vw;`Hx2s2$eB zl}-Doita)3bI>3Wkbx?gh>D2&d&ILtvoZaqnVZ~rRoJA%&y=r2 z8~pUk{^L$w$%v9$g_$4n66`WIznr%@qBiOnrtOTfvifyf2|^)t3n*O;j+nv8_`4(y z0LA~RiSKLrD*ez1$Hd0g;%$v%&kDm%cPxq%c3T{PncAxRbQL>L2QqQKF{G@hBWBc`%lXUoFQhLZV-2*^DCs0t0#;7+s(`!(w|7yT zoaR{LQrkvV2{p+0(4BO$QeQvY?yRVudrjIIJ-uu;emW6mi4@M}ASFb{`5%hr%_FC1 zE%h!O5+jZ4!!(F580u{3`n{Mlq}Na7J&Syro0pQ+;e-a$#2$9v*ib=zTdk*NM@rNa zTOH~{BH~ysN$ssGn<1};CD$P8s}AFMz;Od9b>u120{5kFGLu>lgF#KW=eyKP98){E z2V@#DO_^Wl)*8_XLuWHlrOlwF8d!`5C*xI-<@EPP&xRb@l39fynropZdjj;0vC8ex zm6Ar5lacj~X2*B9>x3y&#_%LFKYjAxXCu!i=b#By5#(mlki3B~*hwQ>iY7taSRy*5Ri9Ejai5Lm&B(fCONi1Q z4~FNf0IC)cpu-QF9A1%(x<%3(MKfP&02IGOqCj9jl$*T?#IX2QQ9^|&SyG&=q8sx6-C;vQ>W%#S5ZjGY-66_hr&W{9_@CIRm_bO&x+W|IYqlvO<`f# zEm~yVTgk~rvUgw5FC^ZgC1{`5dk0>vRPv!jYcygmkC^<1HKzujtw zogEfY9>`tHtKCj%H;Pk-_bpx~k9s2XC*<>7V$;-i_-c#I-2Uq}3|MO2PR%`j2L5n$MukT%JQ5x0XB+C1^Xebd#a z6PvzDQu!_GsMEKLj!vLAvDsMlqW%pxce0d7%gE68G@DJ6xUuOR{Hps!v-)V9jXF;w=FJB(rAG^hD9 z{$N81A&;rBeCMeTU zF3+eJy0J5d$gC%sY$18;GB%X^Qgp;lRU{_ho7QQ&2n|RO%eLra#G~wRZdQ!|6E!`@ zjki_ebWNik+qb#`lD9`bj_Y36jIe;*bM4wScr_DEVMMJn6M+h?u2=Ge&tFoqYjaBr z-(J6RuZp4Dhu`@7_)N1S#nSS6rKu^ZovUzyF3Xu+gj}?vP7w8T|3QN;_!KUaTzIxX ze%jNg2Z80xOifYFo;iJbG?wP+6l+&Vhwy;=(~=iz~9?L3?EDrL?2N(#=7 zP}2H6d!qei?>z;gNZW`{j{yK-&zT|0l_kOEIzgQLp3W4}ro;#OspRL<5=1%>?^@b= z6Cff7b9N3giC)6owdZ#q=hE>E$>02%s}ErEOev)BNluk?=GH!20;p?hifq*M<;$0A zt5h4$SHfM%G*Burz9YoBr!xuiCbWTv6K`9YOroPI@^ax6DAY1 zm<ut+f__MpM`Ns!F!U?{4Z1>$JkcLfe%}jvi!nPuLWmA7yxZtij`8VSw3l@m_$8$sdE+vSOZxnbnUn0%PT>{4G2CQ(L=FKQAum z@tARZUaJRWeMaE&gz`AunfN9cYqAIC-PwWze(0>T(Hxrl$y`amR`#=x7xPs$Je%|>? zlT|E~1}O$~IC&Yauc@7Nl&3G}eJ7-Gs)&Z@#Sc*K*VfROMu3x3m2Icp+K}qhZi@_5 zLTfS@$jSA;lbwYuVao?$DUjJc;n>}HyRhlf3zNax+#Q%s64hYjuA{MPLz1F-`a$J3 z7#$lJCFHeJZVbs^=D5?#eW!4re22+2*ozX8qbzSHGI+{xgjcppqf?810F~V5iPGwM zczjARQj4(ZQVHU_Kn<>AoUlJbdv4%m9}b0=QC z-m3|0P430R0(;VrpEaq{w+tO`(W#;Xx&n|NXLl9%9EPJ_uL`v^aq`dX#Ej(S?)nR; zI0EinDEO`2_g!82jZTBi?CG7N$!8#iDCysu5cSPhdWS?rc*DW?eqB#NL?LjyLAfuZ z2pN01TJ4HQg%>N>NC3TU-+uk>@=A(65@uHV#z};rVGFu6LN+WVx3ukSV^20^O3ur5t4kXo?Vr)49RC$XLc+x`bK*b}N zge3jSf6XEI0qKAE3AcfQA%I@Rfi!@a*TCM&J=l5Jur?6d)Pi+nQl3aAfucbde$oOb#~S;IBa}`C9Ue;2F@LO0&dE&B7%hK~S(+a|0)i zupK~Z-OgRR?5PF#8S-0&eC6K+ao9?^0>;@R%HIhLb~{hX@>7P;*v}Sm-&Fr|KO%u^ zS~5BUl@3l$w%?N6_yBNkVLs|LW73O3sOthxTi$nK=m1SE8VQubl9*I5b>hVBgmF>%?xbyZ0Js#PynyylU|B2x@#dYo7>5!?=#{1f z5+SFQna5lOg2T#tp*{)ZO-`?H_c{k|A-Il|YvM+P?C-nmWlxIu(=>Xt_o~@>zI#?x z{#r*?hWOcAkt4uVEtZfdcWVtKwzA&eMYL|~-`Y}GDKxpS{#^f1@P(ZK2daKh{r`}vBA zw08WHV;yw$>Yvy^oBtZZ(IZpTuPbdIWh(u#0*HFQd2v`sh{dNnYCr`f1EHx7@Clb1 z*Eb5f9N}AIr?(t!2rIoftv+}gRgGc+VoKS=WjmU9Zg$3{D_3GEZ-Yx4G8g3V;Q%%$ za-0?tPe)LcAK`WAm+wK;_TxFxogdzO@%wDX_cZ~}70)bI!@c}8u{?AC*Gll#_U1_j z@?&`3^Ljv4caVmR)f?{=UAc=pN?gCR1d^tJVSJlWRoUpJ>WE8LyPUX|^=~B?P$D)z z-t3}<>So1YokObP8Wq$fomEM34(@4#TxcN33Y2FT*>U(wXH6043RO-Er7Z!XU zBxxk6X5aT5WCRuzT2|EYxWCYMVqN_6jdRivblEbj^{z|XBS@X7#cX_gcamuG_=JQ# z*^^C8cX}l0mlB0|MBxMNjO6S9vuSI{c291;n~q~^IHENC$yejxt3XPJ; zs=yr>UVZ{y)&%|+AL|o$t6wGGly)*f1u7udBJYwJJBG<3?B=fStdL?7`IJTQ#OA*R zvXty*VoZ_|H44NRWBAkjHk&tYbPlyBI>r+how|HBk}iOlMMX>=QpxmAOAr-Ej4j(% zHOJ zARvoVa{LFe!e*e^zP+{L2-xdX?xjDQHXX&pv!?m)sfbN?_MzexDfL>&#k(D{q2ku& zCN)pac&C;LCg_0nSgy)C@#xXyJ)WL+mN&}IqMGHf9t%mS88_9G6v>e{oU?b$m-3(g zBr~cr?=3i(3Fv>Gn`oOMae^YbNAJA7OzG|nYA_rPXjW9o9*_I#{*+EAi&E?*1c+y) zqY%|^a->fx5du%>1w^_P{|{dev}ha(j9ZURJCW3l4qr5KTs2^W9+k@8?{p0h9X=d4 zjGVMjKFmTR3QbEPlcg3>2JoZ6BI+rK(LaPx!t(rd_s1*X3b$X8#MSC)x4;AY!HZ?q z0=oH6G~9lb>^g@+Pf!&!A_sY4H3EQGS5;5gsC-9cqYv#TV)kYV;RO~MWuR)TT9Mp| zNM{-HL!@vZd-y3;s4ONKDkaPMLiTYv@~26Y3_bjWIwZbbl@Si4$tX3>aU3?t($iW~ zW|XyP-xXfKj4fMk&36ral|<eHvcH=&6 z7)C7a?L3|ov0#s1T3Vc$^vFnX?57N>@HjVU;J|u{3pSD4uc#M=`+i0P|5b%c{{G9P z=V&qi8ZpYzwnyv`-i7kSnxi8&_MdXJ?=}*Zrtq2$vo9f>q&?jP%CB5`p^m>tTRpI> z9pvPBh&RLwmrCqwCrO{aBCZ3iUKgP3C-J&9+!%Ulpe&b%4ttm6<&R{IMXuo{wNzaT zkrwcSndj3r=yDSmV$^HOR=B$P=Snlzpv!}w+v?gfd@APRUGCSR6(kBzP^yaR4o8qN z95FA#@HK0)Z>4_V5YkFPJ@Pfbj!c#-nKBO$t-PY*O5QP=BaRw9g=og<>ptY@=`Sm+ z$_+XI_<3pBcQS9*l=hkZ_dXdDNJ zd*5O+0iD9sJV4vV(L3JMvHpNNv)&`#I!eynBvPex+l^5Mn$TP{m{4P}EFQ(vrCj;$MGm3{5MkS|?b-!r7 zd0xs-^aGyE$<#FF)YWh!Wh`Oi8qrb^5@(jb5uNh!uE*dRhGyu=Brfj3&-8qz7B120 zxY28l;`EZv)C{Wvjp&5s0e-Z+H|tP)WaS&d+FzJ=`*?dAD}VaRdI}84tKP>H(|F5c z98NZn;iOv}TC01904I+_O5^bJEq9(wW~(`hCw?nV-*>AYl;JaK&va)_*Xnuap|hg& zh}e?*otGz#tE3k234G93P)BBBcP5!i=L06Mjpidu^fniax4a zFLOk6GXB5h32}eHv3dNp@j*&8GI|-#5|h`{At_8-KFTjt77zG!{rC0ZARz{_-B!M& zGMj)j9XN3USDzkiL9%TSm0BRN09UfE z@|5mAtq7cSJi{dDF-+*Q^|;H-Bom*>P!5S~Kn=;}n*vbQmELV|>yAv~-qI1ln$;i} z=N-s`=#^FqrinEs*KhuX{OoWjEJ^ZLgz)y!3yz}$)eX9QW-X}g&~i@?x=Is;%!n4v zId*(g3#*yIJS;!GUWlxvpasDSfd&CwdzdU9q!N>thy%zlZp|Sg#u;%Looq`JDBJlE zfUGy!&5|Dpz&KhAX!gk*pG)wr`2p<#+;>q>f{oGD)m^@*fvr(E7cEwp7X-9+|H&+VX(Pf_|S zvj%9fv&s$mft``>x@k(Bed!|nuXcfqXk>W?gi_6BJOO1CHEga>cSGEuMxb3NpJmSf(N}&+M0ix zX(9*Q+iOcd5E+(YGxKzv9er9(8UK(nNzz_JEzX+|t}2=PI{G&afkyp?#M<6i>q%WL z?GW305*_pWJaB0Py$^iX2SDlIYv*aK#bWhwhCDyH+jC?>gxEyUWxrsUE7W0{&JJLf zH60R7O+g12ocKs)Y&EcLapdsdQBk?K%0JM4N}EGZV{v+HqiW;E6=i|2L~@t`|h3h=IWBsoJ5tLDdXi&DlAIdxuOdfgK)dG;_LhT*_=JGnCH4+s)R@DSfWE?(}~{9 z#WpB3G-*j^Rf_ZN*tvxRmoP~&V1zze1bk{?YcIKzBh|9L-T<46B4x>5L}j6!h62eqEC55rTlx)^_*~$1K<$N%A2TR%z$ne#w*u7?LYexS_a`1HG2;4L9| zpLn#}qY5g%XYL4C!LNAQ4c*+cI=2fY5ElgO2YSK!S%|EIq1Go_4E=ZlGI@`e*T)OT z<6g;S2~xMA;FivxbPwupHD9BfN<~HYKgT9p;9y9{D~!0sP4Ae^ zA@Zig@99j$vX0q>8{9k-xUE(+ppouK!Ra!!t)^zD$oX^IaPN2d<_q>b1vrTN#%3xv z$us-6^v08xKO#rbBW}GX&+A;0GxGR@|1^=imVWq3njPYStC)t*S3-!Np^Lwgozl?K z(0bOJ+ncY&B{W?ULe=`l2GR0FBoWI8jD;9DcJ%0~V$##yEx&Zu-@Bf+XNsxg&!t=@ zexbn-I#x8uI7XHSj=}yWHoV8JV3_>ErpS0~pzrt=g9y34w7(f5cxIm8$58TP2^%2* zuMaPGxayYk_K@Y@d?r23!*nK88c&bwW;z|xg0^}dNjE}V4c2#W#F-8MQ^y>+#ye7; zc(bw*tjqI>WkRVG`K<$uNB(gduo;~H2IWJT=<@a06!|3|8%O_DR-wrck|DuDn$V*SL%05BsIkT1_4mh(dmFJ4$jkC8(5AS2w~Yq}=0&0ulq>Q-H}s+{^Cq*i;1A0yWf zgSMj78J6+V zG~D8MbF{N|dN=YNx!Y)S>2lAC50vdb6k)?xl4^18VvGMyPnZ0(PDSfGv}!e;7q+~2 zIP&VNDD%!&8aZ6phhz^n%un*SxwW<0>ha4$fOh><^0x;rPIaSIGv*Zkfd|iXe=~Zn zk8d^CV*-$~v$QkFqcvYNE zj*Zss)M=ZncbgHFS6(K4PtPr0re86uwIMd@Cq^yqCM759;#5nn?WulvAJjI-Xbk(E z<}OVdD}vP|*h3+!ln)*AK%M??dPGoAO=^*ODu3h1<)+EW>v3Kufav^zJQl4Ip+E3gUN1;V9kgU3Mn{2Fzpc zUI$lRdn;ViRUPVY-A}>+EdSjT(U-lipx;DZ*@zxX5mQxVF`*!u5w28ckw~mTY>U3tO?rPCsQn%tZ(lWn3-c=rrzj2R|ig~SR_ zP#1m7_eYTj=g_fe12z9PKz7!}-E<|t8Un|!J$o*eaTVAn;L36b)=`KkFn>)w1+F<6 zEQ)>JrzG#~B7h>%>_ldBCER`RpuUh5+)M@)FLpBe~8wrOZgD{UdY4=wk^p@NlMxd0t^J8IQ(u+%~7;+8;fckjL77I z+`y2e&EAw=0T(ZN!OzZ9#iU%XLyImRr$o8)2o8&J1r08!)7HZC8 zQAYC-2f!QlM53?%&aaMtc26RMOf@)b6plX^arp}w*T*%WX9r{_UBkn}72^|5Q+jpl z*^^}B26}34)$~QRGK%aMR1di9M_I_F@ckR?TWx^0y@2i}2pTRK?1o0X8yBbHmb00u zx(^LnG-%v7f!+u6-RQ=EC_@OQZxs1%Ds9-6v zB!mlVPYrv;Mw%2Mp7g3g+d#D5;%6GZlKD$)O0-Negkg1Ytx=eseehp-RaRDJ=I8I@ zcDIvuu#(prnLxBfM)&>F=G)nj2rG}Ox z<&z-M(%Z8wEHpsX|M@+i1%xVx;gX1oI(F;k(cbCx)%q0lq%)b~7BQbQ=g|W;+&o5C z9B40I6`C7zinr73pDQo%--e7#SQ)^>%jKg3^5@M5;gT_O&nEG1oPC&*3xZo7x389T9lDIbPh@X-2 zlsuv~3SNOb4c?_}(G$~yFjQtVKwf&AKbNFmUD8aLE-%0blB^n{G6GAp(Ajmk4gfV} zaS(S)&u@g18slh5n{^CJC2z}Dd0}5F^SJC#r!!XONLe{Jrl_46qZ0G-WvT5ZmwTOm z7rv0o+xSBrQh@=XWpE`5x3qGo1(y=NMzotEOPn;BL<9wIu4n6{7s#&JWC{h+c^Tn* zq|2lU6Se`?(@1Y98pi-7(1BZb#?jjl=BiWU#VC2~wy??Z=-aG_C# zF7!_kK`zBgKT<@Nvq&+$#&WElDFqg7gOGF^FsIw){n{jTjpAPk)9(jS$UXGoM;kDS z{Q`t%Qz^tU1%>)+4Dav~L#NEDXlDX(f97w1hM$HSl*$HX?CwT~(cG3|c%;)+bUU$c z1n+BMvY#hdtOJy~pROAs7*G16p~HsNRDdDxr*XC8d=#sN1PZi>9RCp2#S?a&%LuHt zKy@wo*gGKb==fqDh;pBKW?t>#>FV+d)5zbXvDWYRxztwVCS_zhr)Mmr?nXM^gjFfo z=D=#ZuUsjKt{f=GZ!3B8`}S#Qb}}$%O~rI8ahBY?Yy`A2t4KeaR|&f(-aD>H#o_o_ zXqsIAMx}2rhPfrxzc?!z_b&Z=T*aV&)_&IF-bY@vgF^267r23O zh?|HGtrd_z^4G1QC!3H*+dzl$;pd z(@mNiUM%v?iobX7o_g(wUjtMi8p}I;UX)_M3R}HeeZPr?*Rqahv=t&21{rr@4%~8P z#n?eagiH@^Ag0p?KzLh0LHHY&Q*tQZv$*zcNOt&`;_r>PW!0>1;?D-~yhpsCLOeGt zv^xBU3Di@ObOy=z-=-Xo8=RqVEQ;8@yDl)sUQ~g}ZEHbI!?M=le8JdIQvv4Uj=oxh z7(&TwLx~8V7+gy7c}rZI%N^d+R%+rs1ZcNI2yhf~f`@jdRAv^S_gb8BXC#ZVH7{)( z#bzXg4`Fw*k)~I09`^;R7nO=4sl}5iyjzmrME)-Jcq6q6$wx{63eQi#kWg+-wr<@@ z1cf&0lAWHO^{Gj>fmAc{wlKr>xE!?ml1{~5yL2grBjErlsgK_uW-|D26bu{SYr-;- z6;YSag>C2hMKx{_8r7eY42Ir5dm^p=7A4DuhLxa$o@5y?oEt*i*|`Ysx)!1=`%Zz` zpTQ7hPHlc69;hk8+zxhnzYpEwe2_Fh6}UD%vn+oZ`xmf$&Z84Un!x9$(K?B344m+M zCDow3s}t7f4&w+le1e8cybfIX2|#3LN^9MV@STaO1i~bJ{dcMY(FyaVvae;hiqB(c zxzZgu>HxSJJgG9FP%Og8;1H&8D|!Mx$|NB_qX_jgRofN4*(O!TXwNT`aYUVZ2y}}M zYtq%*&fsMfOlICs5Yd1~24ih%s_?XO<)H`jO`VIKxb`?|??c~kqjsIOumd3l%$`bq zpiN>hjb=_(mWi0PdJUizet?Q<#>3M*KKWtb00$We1iN~9@C=^wf+b5PaA0KqXaiJi z!kH;>L`_mK?egCWvJHrlox{y9jk#o>k(e_6>Z3V4h$*K=96KbVD+GT;5-Map7o}q! zG2W)sb2&Lv&CI0Y7t|MQ$J|%n(x#%AbGrYKA?Nrw=PNnd1%ndU4q*2nq6X846WP)? zFkqj1H7z|c9sXf40BHU{;CSJcA1J^{iSXrJca2hixVs|<`_!ydLA|52i5loi$v13E z*`?>gi*x5V^lIH%N9R5)>`84*B@nB|OGzx|`?GzTYMMg<(%(g{f&a+)@+Ikk7_p8U zDGi6atX;ccg&`OHc%TYNj*U_>3Ld()hsy|@j4wY9V74pN^(Wrk^a>j8$skpS;gC!n zR}6_>UXl)1HtY7LsOOSAJl)D_;fiVUCP+X!Y0I2misK3LL2lo^9k*x))z{gdW!_ST6l9MP&FO2gn_Rs97WbBNf_1Zk?_~y+g4wC8C*5~LApsC(0r*t=Sr3QJK;Ojxb>9N%?nn6>@+lH3o z)amkM@FqZQQ3-f~IwhQc@r4z?n7g<7K4Q^4@WNL-o_piaYYE%t*vTIX3WTOZ1gst& zKre~sgPldFxp?uVBNsq9KppRY+fj{(`2V#1$e||zNTTTwViTR_W`8z?Y1Jf(nuFX% zFUT?xv{dA^HwOccvrZ)H+ONZDzybjqIdrNr`)5{!jqC%l~A`bZQ;`Yjqk6MGWwAWLEa6`$Rg18V&n%KO@-%>=hA;9eG;FiC6 zi4(KsB^JcsE_^^+*%#CGuUxRid_BRw9*`Fv&oqJ4E5JpkB9B;} ze~M9I6vOG|0_1W1ohgOUl4QcoDOh*pMbJM|LI5*C2lkBeA78d%6IYlZTP`DLjyw(Q z_#5122VWaWd>q~D@KgOA8^}EFY;Mm~G;_hF2#^l3%ZD?mv7@5r&tf9trh^Hc5$RSl z->rYQAF;uweFYMQlslkNNY{gbxF-}aVydriL@H{+&y;ZL z(r48)OWSCZ#?f}MyF|tGdk7n4`{gy_K(bh4buPXsA9pS?Op&bkYDPZL6 zq?tWm`5y=bRv__;gnAlt+gArp(aUN2#`go>?rG)Qf+m5I|5C*_f( zO(5_+^hYsXTeoakOqT%XtiBK&9KXcX5>X820B(S^A^-y!B=*?-@u`s~a8&9>iCr0i zMY5N~Zb8#-ohKp?YRj(j!lU|gdz_)OzLG8B2^9hI$<56(f6V*^ky_pwt-%w%gv;CjiUAl$T5l9ElD3LxD#f?&<+na67ELkB&=tD{rdg9N5QW5WN&$%}apKM6OO^sww-j=ZwB#*?h^qXmTd3TDw*Tv=K(E2nc^+#}Q!8%QlP1WGP4xG?0O`1Ao`* zhpd(YMe9Yv89NGDkpST^M7#Y52{;E%VjzfN2O5;~6ZVQBhp<`^LGlRFu5&v~x_a!d zI@~Thckh<`o%iqG4_))|k7ctrVPW{A!-!TeiPI}^C^p4T&jyifYe`n2(NIVV`7Q{_ zyf3Nf9M<6|g#k_!Qtb7Q+oZDaYj%PN@k!5=USre+xll2XP=-%?|3d z6+{|3&7Bj#IduF}+z2A?{BNAF zgi%W5wV>+Q)wI1z;Q50u`6aI}UlYk4A`B5FgS}spY(uQGQ!ltagjHJa3ES-Z-3pSz z>xliHY>Y!W@UbvURz<;}XuSHq;TbIju01Ts(`@cZu%x#^ar zXGnUobBv?AjZ)kQM;`7}T$2B&EeGvl zuQ0mOv5xG7YHa(B8^23W1eC&wj7rUyBY?>f)7u z>Vllo?A@a!1Bgj>Ifx3xI@V_h)H3Bo_vy>ckrLfytm`vaQN0dx((lB@N8}%N5XxQ|`eV8cXf4 z%9Sf2#6S@+gT3eFP!h`r5LiSwDeyIJZT zcCyg#Jv%eAhr(C)WPNSxh)Nx?H^HWEa{(Dnu+5^JqMXvn5b;oR0>yCSwp-U_^l;cw z?LK}N4{slxeOiw=FoG=`b^pe9S(fh*cLU*`|8J5~Qb>vlm~2XK7p?sPRU7=?-Ddj^ zAEvV##?IswE40u73V)kw>yjTJQ;w_?+So$7*?*GMY%?=kR58pj&a8H}Bb-anse!vq?VoEk8aG1rLKMnSCU>HL`6+{g*f< zygahLLAWp8oiRN3qt+89)T0b6D=X*$YFl{0x7-r5-^D2J1Q$bSzKuM+2-gX}k?YJF z*Ca0J$VHHUlnaCts1TD!MjE1jA+pFWNjo=FjbMsQCKu{70J7Z^D_j}OLi1gJ@s)w0 zD8a_H3AaBIwW+m!_@eoNKHeQq2-=X@~0gAjbz7m^6Jn_Vo=^wt*w3xCh!%z5<-$Js7B zqW4~@RVSKV+iWPpQiRsgCvNt z=D}a~4q;r7g8RSwne!c=9%*w@`Z^F#zp7uaSl;qzE*|yKR)z48ar@m5ELkTIqR_f8 zhChzslHF1veFjNN-YB*+zc2}yL;+~-o0WDm8GXqCR7GB!fcWKK zDJF05KXjO$QB2ys(#YE2=6e+Tkq3Tl948_hH3glB5ERY(b*t;dQKL~mi*cx$TD=AhS}9;WF3g!eZJJmHuTnCyvGh#^7g1A~ zPdVcL=;6a%X8nA3fvXDfJ0vu87Tu%tMkKELNg~_h)l#A9z`}kC5^$EiI-oIUnLp%J zLHN_#bXjWa+JDOPF^HOgsIlMSLpViI9;u84Ys~g>tllGqFTGkXTHuY%Al6tqiTO}c z4e3}ci0gFY`7mr+Y}Y|x$&4ho_z3ffxfYXar=2M|IZ6G=B;B|>4j@fxzKMO$g@xJO zP6B3vN3dkkA_HFgYzvFY!t%jLW1|cCz;8sw7|6Z$Q+1dplQfMsJU|cJC#SzT!3T9x zGs-bf=HYdCg_n}f!taCk5}{o?;R|&9^C|qDdPSN&dounyv5=OJ zyz( zS+7WWB+il}iYf6SstjQ`izzt|DeK#o(VKs4MG!WH0!Y-ykpDit_*iX&|J({$w}hLR zFuX_~68JY4fib4Q`V^yDQmP@}E0_||?)En2@noI!|CjcNA!{wOR9a#XPew1Ev;K(= zM^hQh2=E|O5|PEy3(3qmqMQ=z|A(WjcRi>&fhyi&J=f3>7KjFSv+&0cp>=rFt0nIPav0Z09O5#{ zYH%2n?K^h7I>nFrk#;U;Lkrn~z>rs(YB(#g3L=))CkhitN~Z-D2L5-m;f42Z-JpxF>Di_+y_wbT7nQBJHm7k9imswX zpeZ=X2*>P$>l}C_+`&+$A_r*^gb0f*1K7RS&78KKJ5LuTqELMc3xhdsIA4ZuE)sGC zvl0Z3R={mjN$XkHW>ncQ^j-1czT3YlWabi9`YpsS7aiz})6+W=Y=XJoLKofgdPo84 zp`;lgsgGl0-56BKg6v|1f9X{S6oMEj9poIbMoXyrTl{xS;|tS;&7y96#9!dMPo$HL z_?w_H9f-7F+qvMvBZpx?Yuspq#)nKW|C6E+0#7&?89Bx*+{!<3w{zK#Z2~@_2RE%` z&zR6VLH;l-m?&vEJ~0sm#jS1-<&y&1(W3QaDss*`YD}F#5S=S_a%FtRYO0Sp$queT zQUvhAfW`LHVUGlA=y8tfUQi~wQZNi&mLnn$&z17x-K$0lPme2|GwNAsS)27`fXBXCKVya|pXgN; zY7`Hj$2c0)aq@tuYc@-v$*yD!qOMoP%l6c((rr0nH^Ku=S!c#~7J>0~)_lvlk`q`9 z?-b4_pH`E7cl^~;a`OH@&_cUtZu_9cg2y8xl+_iyg+s`Kgv0LdwRdmU{LLoJgNfQg zaOG^{^fHN;tDg3feF|;b`m_bv0q8hI9ejXcP0ZU<m4a%9DmZs-k-GBt;%Kp6-y+u7{TccX^)F8&laweu3%=R+lKiD8nQ(OocdEyr zLF#n3_eqTd#Ha6c1@m_CZ=z=Eecc*v9izdc0M}flmcRYhkd&Zc!ZM16z@o|0mB-kz z0BB-Mql6RMg5bRyI2bateyor3wxt@BKJh*CNcJLpEW}*dPv&2oqI(>27O3tZL0vw) zm$)e(4)IXPPYfF5L-8FaAw&UIPz7~VEM0TAu8N99*!q>0lxGxd>#!pi+b^zqJp~`( z1NT^3G)*PJLMRuo4gR3g6+h&yS1J#oqXY|pHy}Xp&W^;;?GGb}=~O|j6NGPp!&r2c zVm~{aDWIdP=%LhgK9u*yR)*KMy_}6!RF&P2c2zK~YN(-}Koc%&wrhiz@8QE5NZju$ zjcJQ$u_q8j>TNb(ijT~LfBu;_0DFrz4B~qoWk<^;4sCgk%dl7$Cv?}Mz?gts}%J2DxRFw0zn6HPdVpmK=D(PZ)@=B4SLSDTOaKsYd zz5|!9$8$>fMzjllUF7T?<)F}~48T%-vx9?*Ld;xvdTcq}en?55DF0r(0l|_3nNKU? z3)?V1$va+~Y^p}OWAake_tOuMT(Xz$QKhTiZfh%-hr?SHb}$;5#K>v@2rh_(!Pezw zwLSwRJ%PK}rr6Kj8q^M5%0&1a^{}XR{7)yhfh=$tiL~rrV|a1IVv^WKdFVA?*>c3m z!(c3c&;tqI5sszJdAh9SLL;2@}u(@J-EML{m+eNc$r}5ccNPj=|h$+^A-+t za|vAiEQ7Z2m4SXF0*QKpWNbND9dLByri(Gk!s#?UqCscOIIfCQ3M$feLy#e=jdilt z{p_qsEtTXdc8Y8zdo}*pqH*K<2&D%U6@d$eum9xUtp+~Y12)y- zXx28Sg#ybr_q}`FxRBjy!1jZAINNpliz$aF)EU$?{OFD6HN`6cS6}z8T1KXL4m45E ztg{slO`eYm4coZd8iNb{$`Zw!qF?vs|f%=<8h4Vib4pAsH`KF$;9jN z2r5)PHd$l;&Fl}yrLlD4n}kseM7W!@Z-9_slc?rCW@i4eY}qcJOW}7-jkmY$n^2_m zYo9nw3~yn7zyJ7Ag}^he%lQ91f={k)ARW-zA95{mI_~GLob%{I3pSx+&WKrEV7Gt$ zH8_8U{08jNwcG>3uMZNY5H}I6=_WENX!Rbd3lYaNRg9>l9Fom#!QkG0Uh&Re)_+`J z*S0$f-3J$1%m0Ia9_gatZG>p(k@hj-XY`XNO&GOG_+Y@Mr~eR%t+@JatlCm{542$2 z&YiGp5c}AT5vW-Z>*KfI1W4E~%!Y4j^l{6>|K;<=l6~~bxvuzGC7;{C{_;j_GZrQl zC9iDhdE7!lAvX?&#mJw4K-2+Le{RrHY=6=mcf$B(d7OM{>faC9+4W&YB+1li^0dv2 zIqT57O(NJq6Vcsl{~`SH)vK+F2Le45u>Ui={SuL%2!hBp=#4CW|p zH+N!Eq|Fkb{zqM5Eck_ax_D#l2r*>B09TyHS+J^FR2z2WDOKT@Shkf=c;!zW!pC^| zIN!cCBMrvH#H=f9r6#@vz`hEtc?;iv{n~`Wl!*Z2mu7y}4~PZgxxaA~eqel#1~iRe zUJaHjktki3v6B`a;^D2Mi`g zCo{AVx+7haaNYP>dgXN%vR{QexI9kG5je17oNuK)_fKXKaE1Pe1CNDe=|%yIxJ-M+ z&7g4J{y4bgbZC1vJ1~>jzUIA3{O>{hS4IPr!~(%3qlMux_H{(Y5xdJyDL7QYqG*`i zqEVv{Sw$rQe^YYm7Ww{P2%|q;>iIi*3L#Qu{rHxbhi`SvV_o(bCD6e2T%SZRT&!0G zpbTzah}=e1RrONyYkU$65XQ^<#O?!rkuhq53)3dFtQEG^-%4ShiLbkQg3w@iVl4Tb zHb8b{)%%vD(lE;WA=P(Lhx7R}nIE~&?!-!Hd2^ib3?Y{=@L(}T3x!xMb3reITuWvW zhKe9+5^re{ny_U-p}<;UAxQzJxt@$7x01;}Fo-j)B>9U&1u$)+v9C2`$zgPIatj`8 zG)GK!bDynN41l+T&-|{=@V2_5g_hqfM_NSYh2dCTKu=xVrX8%To$NSWjB14i!!2@7 ze4(|*8}G2$8*pq50p01cgl{FJksvH;n2`DH_T3^0KGQodGd;R)kO7%uP^D2 z9NCeMb$dDVYt=O-R;H#JxjQHSV%F9B--|#SEVMQA%dT94|ChF=B>XDwog2m7Y7EqB zfxr~cmK{iEZEx!`+#O0`Q5QO#VrfTxSv$dh^#|8f)}{k}Mt>eEYgjpT%i`!K<(U4< zP?}P~(Ydm@vx0(M74QokdIK@#3Ugyl(jImu{`(DPm?vnkv>tQmd$>6d;JO)KP6ZoK zgPsaBtPqNjm}E2cQll=6CCBNt=_21Dc!Jlf``bu_kndV?_C-EZzzp7Rg#FZ2t`pE! zRSd7PNhHrI4T*FS6KFV_6FEB()bJ8mz>SOvdwaW6j_1v8d>xxKuO?S%yImJS#8 zOpaAM?Rx9iy_g)5O_5J5UKB^+Ep;b1tU=&vwfkfg38}uUQ4XV;Ssa}QV7CwlN+u8n z6j9WD4=5lYi}v*u`8dl#{)k>pI{*T53HQ|5@U4M0Am|wato#MD7Xm-9LIVXpOJeRR z+*#;$_tMj?77ZHP0mjC5CViJOVZ`Il+!21`M7WF#AxBe=L(rT0O2tKsIe~M1w%Qo~ET?X*TdvR)~?cQ{VJ8Z~q<$ zf~69;x<@Rh%5wYdV8UHpbYywuU~A*pT|n`^ogPA}11L?yM~*bgas^xT^zxbsky^3T zt`mwjS$sSbJp}tdKQgvur?by&7a38qUt^rGV!0d;DIInra%idhI`j z!&rV2gcNVzsZ*!$_h_|oJ`c+Ycx0F>fTMVYa01ZzSbX&31w>k3Y z*N!o%Ehb0qqC>|Y*T0;V#;jwwU+LJhcqse>B65A*8YTz72i5hd-YfJIj{##YZu|FJ z%YQahI6Hn~O-^VIT7E4JCU4Kcw z(7tlM*R`&I;wUv!_7Mnt$BtQsoJf)BYAx;fzZHf2q8}!lDCQe@WimJY!78$gK*!qx zk~_IjR?Zt=@c-&lSX6_={9$?z06b32`g5IxEFnM2z3R9BNq9_4$KuVZcu&VDwDf=J z)@Q?1Tz(P>PqDoTLI687KBQs*Z^mG{1z$l`?E`&6LPZ9t7pKLMdKq(bb2Kde3pjmpjEfY^&Q~;p){1+U)_s4%*p2)U7725vJc)cM!&ep(NZ) z!a+o(5SXC}zU;ok3)4FB zY&L)W1Nw0{4T!>1s-36}JSaJXn5UD|CJ+~Zji838f1pk-zzW6a8;@=W(4bi0;w=ZM zO9U2qPLLXyV)wt^RpC(jeJ^FM;MjCqvPQXA6*HHtytUk03U+TMp@Ed=1BMJyRfqvK z<-DIt4dA>`VYw3bXgEj}s zix*4G^4!>XQV;BWyWD$d7CHbmGL_oCzB@bwF^gO%<}YA(L7|~@f?QKzz{*lGg5xCe zm$X42|6zn%R#vv|cwdmBky5+Sg=CUNSfsVgx7jRHQ$dHA$W8uMiW@Z)PyS3j4OoOD z=mPF1Ay4>&S+Qv)P5{q9myxPOCnPk5ogzj0>n)MlotW4+BF)D zIY%7GUlRe&$xtKMRFjG`BJ-0PGxq;cnaBauM#E z`r~I^aje=&6ME!Mi66iOP{GVU^%Xna+|14loj+|_Q^ue(p>07226meJJ;PdT{a8%= zJp&H;J!;39VE7j^@^KM>De>f+vdu$FdaHHN@Uc~vbvd%1Xym3}K-aO9p8LR~v3$Uh zx*>0v`q9(OK{N$;yGP%;{eg5Zx`?C4j!E6S+MxR{Y8C{DoKJky-%lI+HzqE~J_+_K z{*l9W`GHKYjcX{o?waWSDd$w>O!x$ooR%?NYthPex2xAdy6^K*GXcZ1A{7&xn@8ll{Ks_3kJ*WcV&)7LhjvIlUxtdc9}UMXMA;^X;Il1CY0>6Au1hXU49{gFQO#Cq^bpfhQw zMb3g7h)Fse>uAytwMUeq)?FP)BB7Yd`T9 z1TOk8LfAT}05t3>T6byylKR#79>zvcJS9Q3MFv0|X&9JihsRE+pVv5PHyBvBF>}V% zzT3>Etfo_CT$1%^Ex8MWm4-}nq^V>~zBmWVEf4en>kI#bfOEv$fJLB3CH}q=;;QV& zNqN{*A*droa6M)ef@s@M;H@;{%3x+0_doA!rZ7FNqaKn1kU*GT%#a-^{7#EMmErpz zL~*~VD9kzdn0OMFc&P2LDVvyUyOCqoIJS&D9!EbLUtP%@xlVkxB~WTppm=n=;}`9t zYEuETJC84OOs5)ek&3_ z+*2_-)VQ;l{(%eZqgt+Yg8qr?uF6`Yhx-MA7g^=_@>R7T0>+4lWPmc+J=75m zIslDYg1i;L`9lk97@y1#EVoFg+G^!xWiys6`D({kI9FVh*7lw*Im3iS#ZC5VGY0GP*hYDM2rt< z@f3P94Gg4+i8P=1*GKox;?%cTIzeIF^1tgiolOCb8GeeJDBKHM$nA~I-c@!Wo)8ij zvc6UXOia{vg20L37pCV}#K`p44?YhiH3~;qmge!|pFx+T&XQ7txVFYkdz_OO9KLP? z2DAhm7w#>5!=SLRCP;GWRHF=hyG6$GOGWVoW+8!2npcL%l$_!#i_3nGVZ-LEiUO0T zzHp;?dk?lo-`kp0yba8Jkx=2F%gFEAOUlL;L&im5`HMltu;_3jD9FB$)c`;E>6%t< zyArPoT(<*85)&$K2Dak>@7Ylz{AvnWv+rLHXDkAFVzSavIIcu*d1M^UbsIN6B#Xiw zwERA`WkxWllWc9KxZ5dHHpX&NhQaX`*>op;mrfjxgKw05_gwXS{x1$H8?@T$TJ<~1 zqO1-KUDOlUwy1zeLJ|KI@`_l)ZZC(oruA&%p|vvNzGW0wx|z`mlM&UTy7zCUq_WcH z+Pe#s4xukE>l;j%;5MtCPuGIMZ_K}b{W_kcG9U^zSXhT1s?bQb-_SgQIuZ11p$olm zE-h)cFnG(Yg|H}^voLcG^Dw&m=x5K=sEsBYeNrVw<@)CwBIk6Ay0k7aNcq>XI^L2_1=iI|OYhyP2=vHF(!4a>CNC%Q+VG zrn037VS%PpVu14WJBDoLAQ71dUF_q8A*nk-vQajIj!T9mV=FiTY?N({>kThP^Qe;QEce}=n>c_j4hm2+v01S~JTwNNkH1Dl z{klp+Fh)fnLWi^e7;mYfIi#Us!wN%_qt+0lXGkgkNU4H>2j)@ImIw1`#5WfnRU^7q zP@9De(gf^gPOWua=1Iwj66f7my$!`67Q~)>&4h#FVxv-1@=hTp21F_ab3rY4>Ifl7L6W@qjk7~*cTG}0SKJj1`%%!(Bn? z^&)6|12_j7iXjW{Dbi&ZYOzRdDA<4T7LjwW731C!Ajo^M|B{@nXsB45+8VH4C|yIA zCv_6G6xg=PSi*Ns>@Rm)N~bo$8-Y38NRfDK@a|zsC`2o`9pWs=$LpyOW3?0dqeuJm zxTaD#_d?sDGL{<{Q4YciM}CiR*yzZG9IF6jbBrhhcavm!4TZJ11@n)}J)3%$c{0z* zrDv%6aB!WjI0F(EQQo(5c;6NVb6U^iU|@JNet;_%H9YgXjmyNWF+rQ2HCb?LEwhq{ zeRskSoL&zglB3#r5;btd#+35o+_Hk7mQ^}k&=8U}K3-Q(K7tQJ3S)yU7iubp@+;vu zgG5VnGnyT(16=8FvC^M0z!gk}>S(5mOJuhiC@*k(PQ#`d_lt&_5ek8Zndo%jW^pDI zR#}}u?tRhE-xz9yLOh?rr>bR>&S2cXxrqI62gD~38d6X0mdaQPg@|%$jXta%q~sk7 zNYL*k*`|%ItUvMGyElZDAv2ith-^pkl=>VX&-dL`$U*yPB#ht#t8OI(JgBuM6fzKJ zu_r@fm9d1qT9)2tCTYU!JvZ}I{X6xY>;vK5pfu7tj%)9&+qaQ=cB&IdOtGqpx<6cu z9C?AyC_k}(S*1GmX?$shLGE`)N*8&$YPW-5C7*eVs4LDU=SX+~T8cGqotp&R$v3;U z-Wpas+vB%Ep|7`PjDC-kk?XTO12YS0v$vb8V}Tq*;Z^v??2D1CE{H$39b@*116*lX z0j^93xWcqwaM!^X+_zod@DvO>$}+- zDl?0^V*cqrTI1jk49zJ_52t-G2?&qQQ7T-*Y1A(V)&P-vq>VLXS61eQ?XPlds>;{0 zpI-L(MSP!JQ`o`vVqTwpgMkrHP~nl$m65m*e2{eOnJKG}cOSp&(w`$xWiw1y4gGV3 zH5;nii{n7=-o5{0=D+vTrzPoCgfT4J$AH3$b%|^-qb%wlV{cifw=|k1Up^PT56uFK{VOJ}c3c;UB)XtsLJ+#RMl zq5!8jU+Z0P!6NKR>>caK`U?2fpDr99q&6kS3&I67+xOQiQjQcj043A?sh&t_$un!8 z4+vKHvR>X-p~JPNfsQacq8>-9@#rRj(~$N_J`*8Hwj}&btH1X0Q_AIGj5{eqAG~?f zf!p~2m_fG3op2>ALJ1fL;y#f5BMifjrfnWUD1A(m=j9yR-p>; z3%F~$nJJ8d;XJGs#(Gp*p1Quwci>BPh=f~=GoN&dB#?!hJ)z<*E$41LDV?0VrI&5e zIK!P?|DBK-Yh&3U7$u|z5v%bx(q*(w?vxfAI@wcv%B3`*%y^$D7Da+tbLjXU_q)1Q zZ6l}_W;fVr-FGk|?caX;jet^Lgl51rvGlL{ZQ6gdJB$JQ&cU#^hr8B{GB7X@I0S0I zJv>M#qumfZ@%2Sdcea`_V#6j)MBl=nlZEi?z<3ZBC$%fgAz2_i{O&HOvGd_Nbno8% zZz8$k90(HTIvpSomPHl`?w-*kH6!w5m zEyVuH(Us|ce4lO%fyXoDx>KJ)l*~9FBW@?)%gUWIR(Efs(iNW2>x>yB4&l~iWyRL) zg7P<9XVBk{rTwWOcJqVhM>@6Vi=aka;)49yxXBcNy2OjN+()`&cngUa1$`Pm{WkQE zDU1AQ1Upb;sb`*_>Z@&9iF>^HOsSFRl%>zuUWSjiJL8*v^Z#ZG`2TX7V2Xm`PL#Ce zqToGB*TTn9Eoea@3lJmwSi98##R%L_nY6u(a@~_#Hd||@T zo-pJ(T%q^wmfddWuJdniXapA!Gn8?h{$o+}4yLYFp&=o%G%~L;&(84!2XSzex3~8T zulsztS4DVr#ZosY6gbQ+tbu4_yQ*({+ySe^js#*&Wl~kuZ`9bPmEnYy$R>*xF9wlR zMF1eeJZDYdM>a0dvA!NEv^*Fd-C#Hasa&nb; z-*ifxXL9AK+a6t%1>@rkDN%P~2IJV2*5j<$fV=P8x2z_)iQAa9f0gC6yFL9%ZPBPs z&AjxoIfGE@su_13?0+7k$#}AI9Msa^EB|Ei)i^SifaW<4%tsgEW_BG)t||o31XnC} z&zw1P5Bh$r0l<{+n-&VtR_uQ!BCwP1Hg~ST{n?u^ZufcbLXUI8PR5H2jqkX^3Tw{* zFQ!klKL9$2*kwmQ0KXuP0VQ)EZpALH{(V~aRjvqlxDrPt1jwRKD#6z_M4$t2TKK%H z>+&-$Hda*r~4ZnkbNrN9m_X_%w>ANEPa(CiB}KoQ=gGOVkG@0Yf&yDZ;~c z8H4TX)*U2(leQagv}P0C99(3(fcdnEp;mGr$TURO)h-6Fi`^VbjsE?OTk0#A4E10E z(It(}RIx&A0OC|FFu}frlqb}SZZ!xG#o09TTT?FSHKXD1e z4jTYp8rit%SVQG+>e`=paivsq$dJ~E%j+oU8zws~pL)XUadPr^JJf8b*PMf8ZihRv1C>3Caz)W9gP)2`ng>d&61 znR$dC^~`>Nx}-_ITW~?0R>ULcc0EPrGlB$=tiHM zs2l&$$aGLJPrUc>nY`=lYy%`>&dV#x%Yzuyo4PJ<+}UI~E_wM^xafG!s3-uYjzz(c zHf`55DKXLahiiWWX<0mM5|*AiLC?X-=zH0mMSKBS4r406efy>iWrWu-!6e^S8j%I; z^PH+U!_B_tRKET;8>z$BNKrMp7R%$jP6Wpjzeho_m=%cW0qBwl@=TX4>m6Q0-fyK4 z*$=c5FAUe7gt>xasBVP@3FpFIw0Tz5wa6nX=oRpKBW|8bijUV&{B`l7F+25wAVsAA z6G_W{g0Yos4%k4kk^ax}X6*;oD8HqwI=D7}hT6-;6lrweoXiSSGT@cQV4tkAgzP%JLpz`ZRM zb19H?*%|RNOf(_X8REWB=lzh2k@#8c6}%17$Z1NKtNErZ+iF4G6WTVn3;G`tcziB=J5RgJC!mAiBdVXBMKi&rtPzsDZ-t zMU*##h`55IiMF|~RW*UXw?gb}q?yeBb_TM`RK+7URKNz-nf%WthLn7z(@dhCF|yFY zuWn8hJPr}n(Klg|rh%{Xk*HPe#&k#2Dsmq{s;OoO*_Hj8m_m}HPS1j zS24JB1tV153UD(GKqE+i^e_LJTHN4Is*I&47k3lHXx+nQ%gSsYM2qNAVTaN?m2Tm` zC)x;*4%;MCCse6Z^Y!25bJE(68s*0zZYsEs;N6PN9p_Lb8X8h2ox;?72k@XFK~Gg2 zPNPpK3Ay5VBHI8VRj4V9i)LH*D4u(aD@QgBBK90WWeVjxC&B^7R*(Do_k9}B{~hQ+ zSOyk(&sg|X#FoK$p}zj^H0DhQt!pvpFcWFe|Ep_qez z`fcjq;(@mlr`ML}g|*E`##oe(GU(TB^ONSVF@!m^^WAZ{DaS}75aHkJa6nrGSa3Lw zFUSNJIR%>$M*I6bwm9$R$AQ1d1~2wvT}ita;`!rEm1?UFpzWL0dX9@F7ycQHuI)}Wxyn1!1+w!!|&m%gA*lIT1Y;euN zWSv7%tLN#^sy``h{v-Lq<6E-_*BsZ)=+XV+wpRx&g)y+1!(Jas5c72nY?NX>2}GY} z|Dhbr`4S!%BRGV@51PRrIo#E!>UFzU!X-9HL!%1QgPMgGE9LkL0F&i*9p;~Ncb+11 zL+YnIJk;nsmxV_sCGBe$MKej=WXn8RNKWPcetw&8m4@YIf-5va-#M9mzQjV@xMI4D z&r|KKq0!VjNv~m>{KoA~s4%9M!8Kwz^F5<6nOChJZSGK}+&uft<~g~WD~cS8%Lhsa z&e6BeiYY9kL*IPCuPke**h6%C>$jt&iv-F8IH{+lqzHOW*;I0~)0nOq?YUza&{5zZ zy5q07Y}skU<^kSLhv5wjH1HlwevoB-W5#r$B3IAcKs%n&(h85G$!s8 z9z}k)zdC(uH|S*jZuqP1tGbd-MbGIKh?5$AS>T3TFkn5vi<>qdp zXlAQPD0X%~Om*0|7boji@Q`0V@*VZ#6iJ9=_M{r!O+Myqd{WH7^h?mR+7(`R9#Z@M zFESLiGYv<@H-FpqpdPj@#B51qtBE))G6#@N2(&e_J{JLshK#;6_0^`#b~86K>%|Ke zft`4e;iM7&T6v*!i0Ew@BS~souDq!ogv@>xQB=SizrIUwN}EZEB!A;eTDS@Aff$xS zAqYjj_MAavtrux7wS6&M9MKED8qSkEMemjZ3s0}yLt8mKtf8`GL9xC;-tdCO+K$&7 zuN+~_DV7x!7QIG;VFG2!))Ltlr00E^A9^L0-p&HN`=%#NT- z37+ZKw;A$Z5As^#xk*}DT7une3F}<~dHYSomaB#(;uG9q#0D+hHs?ueMofhcm0b*H zb|Se4z^^$>*cLqH&wR!f#IL%|I1Yd_xSM`<)Jxl76p?1n zE8p6iG8y6>PefWDNV&Jp#hAQAh*AC%5egE%mz@d}ak00qe5W+YMQ*#l>>?htwG{}_ zZFpP?7j}cFo2sBu^``gT2dm{Xx2)F0&~T}*B?Owsw~gZ;J=)LYhP5>Z#D3}4j%6q` zE?8cj`0-98a4fy+a2Dot!ee1wA^lzo#=I*Y?eBj6D@AMnYFrN@W?!P(4xuBGFJ&3p z2Fkz}+!0vP>H*Ux{ue``aE-QOb14G<2X4NcZvjK`IF4I%WP1+Gvpb7Kv-$gAB)Fql?4cQsAtA2|v36Wn*Shzy5EXh8} zF4+*jMh|cABV`xidCIop zTRC@t#Y|-X`lqj$+@5^Z)&On-;cz8$qYc4*dSPz%pNM%AVuKa-SA@qc^c%0!i+{Bg& zS9thSVx^aIcm%OF#K<~>C6{sjomNnPr6qegdK6|@imUBEI!Qf z(5lT{o`;SfpNeo*1J0DnSb+4LRHU~?qIGFi259#&)tU_hK))iUS3G+0!e@@Y+A~Ce zXP|VSpHR#-*<2!^JWHTuW2E9of~|EMd6%zXVfVuaeg8W41AxiA$##HBb+dp|d0)6nU6(Xe(2Ig`J1vEFW9ewuihg92ag`l${32ND~BtR{S{kCZ(UE z+*Z-D8TWY7ljr<3v}Kpl)&Y(ywnA|mo}}$gajC0gp=?BPYxTyfrq%z@hI?%Ox(QiB z`rHrBP583uZ#ugO;w1*++3*Fz)`i{W&VGau?rIY{4@!;I$yvbnKL|OUFvNn9f?pXi zUsvZ)VnRY7Bz{my;gza?+Wgnh|T4DZeRrl*@M~3 zE~tAyYsB^cNC6cjL#HMLvg!+Ku9K{nM-l6>PXCejda#Ue=uh*8A~*5B`U{u}5T_X0 zM~m|;oP~mS=Ltpj)u>CCTB4I)RKqvg;U0(#j4Pi6x=m=d=qHO&TeoeC34R;t6L@uK zC@<%9Ra4sr?1eR4G{UkYzrTYZ?{hSTo#4&3UU8hk$4aWkJ9{x_0|3Dk_9w0YFvTm@yG! zCG3~n+WmLHs0B2o`@E>DUC&hMRb9I_ z#G_M<0P{ZY0$^&c|0yb~9>v2s4~KtSUW;`eJ#pd`wTz!^GnuxfWi9Kmc8a7MV3ffu z%KONz7}t;g13MxKl?JZU$c;_Kj<0|?TC(NVVM;IWzD@GY8BipEqu`qnXN+Qz4lIu zELWlK{==-@n43fWDDHJH02zZtPM}d<9p}UxXe9t*#I@Ys8gt>Z;?yxXidsWpnw#d6 zI4<2FX33r^-zZE`#eyT^T5Cr{*i!mI!C-qX9!?-YvcFUYpIh3k5Qv%uW8lh-6hKw` z!7c(O`RG1Ygrl>@l6H$rIt+Fuf(@wR4Pk;-;f`2+P*V16be(5k7wR&Yv8Ekf{B_@E$j)Bb@(Z2z z03b z{^J&;GM4N_;j398y26{ey|uq4JY`dzDebO8sx(@^5ZQ`H&#LDd&gVCUSR&@H3Nig6 z%Cp)mW%J5s(+$OR_{z0wOEB%&+l*a@eu6pjO8lw?g)UVM$gdVWkN?l7x4|dewJ^37 z_iqA^>ed&jsb`rTMVOfhBSh+!cdZCommKvY&mEBkY-R^lWs$e+iGN1NriB^cRaZvm z=SS98h&wH1XA1wT$2T}vmD!IZBI1>SN#d{!fiDClt<|DS0-N`Zb$(2|+;Xc$jizAs zwLBTqx3Yo!DX86+aARdJivR4p{CBjOKdk&~j3`Gw6VA7nL%=O?{q}D8Y@f8!F;!l@ z8=q#Cq;NXR^_g}Sf%)dKv`lZ+oMla4>eweQAxG2A|_REEF=`;YxyWMS?m*du= z=*a1}gU45W$*(w{HTSWfUgekviB0Xkv|YP4zn{K;4F~-S)zjyvK|62{)r`&XXFl?^ zlYUrb3cD)SL!>P)i0c|+8Z!0xtb>P?-*X)nEuJ=Q4`ZE2aMMBn2#?onAS5nD!c)kG zV1MfeWZEU1m*?1sPd}aM*`{wiJ)jozsu6L?Uuue(E-t~(dk|d4#Lag>L;y{!2<;b& zHXWh6*4vP~zv{<>7N!^H>rH?8=u`PYYAXMan!S2G=0s$1zETproGUivAR8SK+fMD4 zbL9=zv`wb0=*?m2lWoK_@{FC6hDL>JWQkrCVc&?vFf^mlm8DO50|e{`)yeHO_<8Ao z-BDm|AZ(|XHP3kXa3Yjw4T|a`ls{iO1=y8YA5P65Ov)sPJc~$PpB3MWTJ`{M@(GW~ zexK6%yssWOWX=!$9$`5@P_dN!?&5QRJNRX^-ILJBlIw#WU%9R8lUn_m0P;j%S=EE- zKwE4B6W?6^6gBepzE;caT}KQYI1^|`n|3~oD>FkrB!T?nr%ydOMQw2G{Po>gkh6rb z%*QWZ&V4q~Og=LI&c^AXWg5!FxyN86$8W7u_b-U-tn`wIZzrlw$%0E;SP|+oQ`Qx*% z4rvxbFbzy=RKg5Dq|CIM)lr}A4cz4pB0@?Vl|X@ahd)01 ztmy=d?y_)bQSaRO!p*YryP7`xD<2!;{Z%aEa9Hs772x51ugiS$IH@XQuPU7JOyH&* zTH0jJK`7MX(>}EB!3yKA>^WQT`OWpPW!?S5!^5pIce}YghK4t^iD(}J)5;hlU!kQ< zyr*rvc6p8+=rLpL6BTg9SPtInmBX>?8Ty0I#Cs-1VqK+A%&bTG&C>`pJ60UVy? zu(zwax(0cu6=NLe>^VuPHcr>)>M(NM&uE!+HZypqRVE?P6C4{F(43<7Yyzt&DoWP> zYCmYub$DdC&d-&zU|96&U2VVM{E;pV8Z@|TToL*b7qyk}5Dq)LE{JrxHGJgACp3k{ zzkgQtA%{-apJ{!l<}5z?vx!AyKN=r?inIw3Y9>b7)nTBrP zhs!w92Nw+nad0dd5p>eeb2MelbC`uJ95|KKF<9TePKa*x!3u8z~g#O&VM>p_~2@}td4{Eo^37+X`UZS zbq+fJ-Nei3ZE*^(EHV2o)$^C3F6O17r-Fl*5DX;OGq;xv$K2SWvZ@l@SlMv3+kjB zlj*8MW&|00KH2l!ld)5H5VTA)sKY`&SADMR`HXY;n8X_U`n9Fq!Q9YRUv4*_oPKAM ziTJdTR8#$W9(VGL7*oGzM#Ppw*)O7BX#V^%s9XCkT~21JZZ$D6eqld$5d5CRvpXJ- zd|GmMT^2CM6~><{-`&2x?6PBFC6j1_oHFK5hrmojhTj@i?$gnE!|#Vo8gxQ2&N=NJ zTRE<)*KY?+j#8eU_tCYsY8Ax?zQPpDKhw);l-gn%f9T<@rknStPhn(t%<5xCl&dw0 zFAZ$kTH4taweEPTc%W7%=Ed?(c>$rrSDrESP?uEXpUDEaaR5k~AipX7`cLjM``yj8 z0Z9dnu)u!$4#M{0taVgm_Fvs)di>AL@T7u6CP!I76$w~qVP~*$Unjl4IJqO@WKm$S zQU_RL5O-ooMs@Mfpc=YRiExsTt_GE*95PJy9~qjm(}DK&c*3>-<1=4xpT#`D7%Hwd z$0>E%aMuw*7rhS|>W^-wv&En+W!5MAce7=fZh2*ihpazO^tfAL@wa-*mfbiXhrz)o z6U2w|;B7ukMXZa(f(d&+`*TL8K8t&vb2|7_N}Zg<6QO#}1Im}YW9IU;lXcIpl_`&p z4XG^odPHO140hLNK?E8RXENrG^`K}mvBcf^lOwHxb{f4YONEf?c@WbV6 z*62i46?*-8dL!&tdA7Pj(_IodnUxpr?vWRnq3SvQJdYXt1Q-O*6YE#d3j1G-)~iZ3 zV%BTN8A6zF~FgF@IcI;ul zb!FC#TC_N_b?erCCeeE(-mdurg1d?I6q5D0FuI>vYvagr<-vb?b-WrNXY%>bAs;_ z`fC1C4gvt?K^nC$y-qwLG|zJCuemmT?ydD9R>ljFbRIx-?SB2bz2`Vjx(avlT~bC@ zToXbd%p6Z%3NCxr{KD~Rtfav>;_8sSx@v0G!{{ix@qH{0SBwseB@b}Zi(TSBxqgP; zflEVG;?GS0W?6g&hvaiVtg%Tw`2PL-evjY$u=d!u?=Z*r^XR_cLkmp&hIa=V=+nDp ziLbNC1#_>PH*Y2!AKEuII137AZm-BkakGpl=lWB)J)3ssYmwnVyqgktI;>Yt4$v&c z+9lMnY-gl#+r8f2CYwHZj%l0f<>VB=aWuNRu9x0F+s0hYzq`4>K=cw+SucydH|X8c zWu_MUR|L?UmC4I$-HH@;o6PYXsO{W!i2#RQ7-#m5v_C%_4@jUS*Y-NsckePVpFI7?fO~FIHWc;EpeVkSN)OmuCEGtJ%1@a)Br&~)~?lN z)Q9JaO5-@+p%%G5zQ=ZWxaWT5*+2W*Cr^1D0#?6oHcPK_CpoguZ8k%J5Y!pX50KQWb9v?q(;;YUW<{143448qo{p6yn&c_BVEpE&Hlc{!PF`?Bxoc(ty zbuRio{t`KZOg6!gQJepwJySCpST3=5t>75&bLYBq;#P6IhA718{SKq{+`W4s;(^hQ z8z<^kcmxq>EVQWBsZ-YQj>Xf6h^(q$ zbJ}ipK4C|>)xfRs(hN6Vj3n>vVLfDPKOp-qf+GY~VxhuHHhg!`cpq%roRd0Q+xZxYl3`g18Om zX%Ej^TsC-CvS#(_0W9F)_rUG$AhI~>01ejfK^9oIe*Jy^3jTNJX62axi*;w_4WEy- zuH(U~rcIxIVDNI|yD{z5u#8&qF)VczoZx%$@l$mR>a6&vsjNFDy{_}ju+%;mgYwI0 z==-=#a#GR!`SaK%1ouUv+(MD{Jf0_w?>p5^V}fVysk0rR1j5p-wa@dNS`WDG?!zed6j7BIQ0+c|CX7&X!0#_J(u7}% z(ojYCg=c4r7YQ65e*Lc#C%p_mlf{@F`IY(f*ImooPggGy2XZ&y(?l;AZ}RHs-BC! zv1^UPZ8kfU`r(Yl<>25E0iNF8aUAJO%Mv@_z18+!9Ce;!=|f&QBclVKEkroYFrm$Y zsL2O3#E36=#98pTMs$>eNk$E?&uzw+=R3KK@cGiaabE>NQ!!u=MDVmFg7Ehd*1;q{i`eF636w;!Z+a(6Z=$nX>2FgXjim%HvTejH-C_r)Qu9v!K1}ro$O|W zr7DX5$*8_oZ8`LG!RMn21p?A&_njd>^Ru$t zU>=@=qs*T^Gb3obRVFK{o5HNSk3=!h^z01GlJ4%cSGfnc`vao}(F}6(q{8_pr@5M% zvrK14Iakb{>5s5Ic*)~~bFByW?p>djMs|LyVP30vbVRpn{pOwAk=bj{b(Nn2RmbdE ziP7PH9{eTkbBEsWuVhOh&aFL=If!h));5}>(mL50;{wwnyGQ3q$zO}#0b94G=t3|# z647mah1T3_T8QvG0$eJlJnQ5Jv*9Td?Ye#uDM1gZ^FdX)>54*ea zKsHo--+z*Z2ww~Z*yiz{X^3ACU^#eUzBx-gdZ^gtEpa@H-vrdZ;&NdRBt(&60I-b< zF^$HBE4&~I(8Q;tsi9%y#vbl9y7Qu3?tV;q0=UuIWUlKffM&?`TEGQ6nVu8j8q+|( z0@*ZFCCN`bYZ@}Bb25Y>SlBaO?Be!ch)QTsuU-(nvP{un{dtju#n4w-Q>lwh#Xuyz z2`5L7Oet8FaCDosOZkME9p#d6UA0!1D+lA<#=q`J1Vf@b5yI?;uV1}rAENoyn8zd+ zFFFWMJl5JEF)?wd|B5xEwiK#D*Hdg|cp3~GBfnLI4@ADj%r^M&D^6!NLndJJ^8Cyf z^+VDoGir7b)*VZ(-9_^9;zes&#mJ0u#_4J88C~e5Soh=mxC1}+x$Q!_{=p*iHOHGP ze*FA7k!N~9)1=F_Ki_pA8yhlSG4>l?2h$s&cOg-l8^@;lK1BKL9FC8`%)+B`YurY5 zi=UaTjeV$as=t=;#xOf=IBL|W_n$wD19WShxvkN-iT5%yN&7DY;$mWCQ7y)V_rP8X zsP>s>1aoBMj{vF$7qwf9QowK27IiWLMSFdMJN(kl{2CGF(lDm3ykAB`Bh)>9kqkdc zSAhZ`mVvWATiN|F;M>qrb_b`39EIg0Lz_J6ICRxcgikh|#dKl##=N81%Zs7=e0p-E z1)uMDTHup=N9JDG6&f0v-G%k??Fb@$+D3nMGyqC;&z+a~{rf(Ag8H1h3L_J7kBxI8 zpIO2>)7&l)7V1FN;-0n^>m3kGTjg+m{BDKM&u^nYFfvOf$6C+@45M7Z=>xCvk4}Hi(I^ z$G1|Ji8h+LF1fhpHkJVJFhX)GI6=JJm#04~wh}J_DSfnN1S-T=Nu^$68WaVkbu8`V zy5iiKXJ((978i!ay?~x)n(?D1QQ7$Z-uYf!htcP!=S408X5uggRbR&e^r)qZ>||lu6r(-sS0t8gR08 zwBz(=??KgMPdgR=vIHz&EkiwebN&*KK7ec7lHnz}WmQEmfpNyQZCcV+bsFt3J0+u} zv~&{S-K;$;TWif2`&Ih7{-E06$gMrzQ0<6mH-wa1#e{9kG z&w%c@C>amD_}7zcZNUNZ$HAyx2i;_XJrQ&l<;6<-(Q4b63L>wN^*tnkI#hh3Hxkt{ zAzXZuMg@2C!!5Wc!Y&K?tG8izwb zORZ&WiuL>{Ko0D*ZafWac=C-?^wApr@QnTz|gzDj3qh-+wpv z8PQiTs?o3RI^JaSyCq#^Y(|DP6_e)4lP3j5cztJ6xN&K>;lnK$9xj#PWW4-a1h0o7 zLHF@IcDowEtu*JB5t}c|FJ&F_Cl1pfyXfR+)62d1+dp#z-S{hmiP1rF z!er|AHX&`M=Xb%Z$ciu1i?ewIXqb|4Z>KR%7R=iib zR~K0(G^n#K^6}}F*evyC<!X7O&-6pK)A5Keq}iB4BP@0l>B#<_XJhWp`7 zhQDilfh`(V9E@Uz0HqN~MU@>-VHULoI4C0;KDP4irI`p+m(V!Vr{m|vXV zhfj8ZICIOf^dhcYmbhoqJf9^a3*wvN6!m}Fd-Jdy*Y@rI&ag}wqL5jpGAxx$AyZLN zsc4ZQNl2zti-ii2+?i!45>c5FDy%GuBvVQ$%aEZ=l_^4{;rBVR*7LA^f7|=t`~LN9 z+a8vly6@}0uJb(3G+9;GKq5y$~Mk1S&Db%0-#dtZCCjq@p%}=O#sT+-Cgc$3nlxPro3ve55B@fN&sR ze}>CO<;HX*uy~ST~puB@bHM8HF!h|rozlX$Il+!%6cvAQc)HuB$$HE)UKy7 zsVe&PDbux_a-f0Y2w5-?oOaIgr~s1Q`ikoqtuPXO0%F?e%!xCX5uH1uc+-dW9ms5A z<58o4!zsP4kWSG4(E%~8t%2Ph&1PN4Hc<$@g?w}imtwh-mJL&$M;BM&E z&u2aw4pz*0gFOEy%737ZZ?ltqWI_!6H9Ql^;MzND3r)B$K? zt31zWPcTsz5@k8aad^}rTl zw(Ui6w}ZprCej|+*`)Jbg;|7_zm5RWv*10Djy+TzuG=9Gjoj@2-y=7}9RQ2}WvA$S^5V(JhIIi3@*!qjYtC?V9dTsiu{o+`83T z8FGSZRUC9!<$<}iM&V^IB`NV^B5CHhThN?#JTXYUiLvX53Irg+E%?rWh#eS%MO^TjsuoKKZ^gzeZ? zty{wr6v1Zi9_nL6GPFC_<2?C_G-0Bp{S_HgtWi`Ygne+J!;u$hq3sCIw1afs166KV4#WKvrYT_?*-tKK3qsQVN9e&+|F{dl3%Bo==n~M*?%VYSSj51d4D~g5@pIXIQ zG%I+GYl_r~JY zGLHgCTS@byH5sRgbqF-*e$G&NVLXf3s44u$dIG7uFx>ssw4qhUUsse9pD~sjd*^c$ z8o;E+L$xW@`2-;6s#7j3(-3e?Pfwb+UY%&i}i{kfKMks#Y@0kuetTf>BG_G zvWn4%OPWt~CLsI|6?AGNrEr6;@`XchK(LhgwdS+KttHl7mM*t%uE_y40*$?wH{iS%M9Xp@#Q>mZI`yaTVpY5 zr!kdUyYN32R8;>icy_HMzN9Hsm!Qz=th-;@M<|A!WOd!zW$?+9i2Z zYu-F!y#r;%g+pU|k3ZaG@vGY#`NS?|=dTC#)Zb(M%AMa4G$;?O8=u&=sG<6L6@Z=< zv}v!nv8%Uu(N-%DE6U8Ju_391C|U&!DFKmOrkN=-4*;gfvBh9DbqJuE zcSN6S-j+C>ZSB7hSJ0vwSTto_S%|GS9jh8Cc&uLGtu?X;<>@*#)j#O@rV*SN`mhnf zH;Y@sn+$5o^%mnFjg7l%3?sDIZrIrUGp#pbO3D))%?(<)*{ApD4J}pZL^W{=B)LlF z{(~qAoxJH;5AmaiIB5VKT8+1l2)A*hJky?h=6Kjd&daNuLF!pTsOBc@38fJ#Wlg)PSmTUm>E&GA}La=kB-+7=!hF_skhD{%(z}up10DyBy_f;<0>d4aZYx5`2pDa zxcdqP+-@5VIDZmBS(8-B^ZSL|y%-m#&s&Y#yFTUi?M7f!6IMi7k)AeW7bX!`RaUg= zKMy@8B**@xLvkcyl8>8sN_uR2^*1u!YWk#cySvBLbS9ta3Dprjpmhdsa%#!@X(pcC ze><@O;=RYwPdaLs#Ds*SOvXzqj4FE-(o#cLQfPic&FdGh+6Ph3IbqNxHl;8N$WX_T z=!%OQkd$(U7x~5O9wj$g_xIP{!Ih1nV$$zPdVOF+>;9I7qRrWcnpgE^+TMYkNC+|w zFZ%lvNB-iJ&Cy3@GjQ?q^!)LmRBSh4wfpppwh31b#!;%U1n6`tsr{^pMvM=`=H*Cs z5i;hUM8aDqSD-@C$I&lJbUEf=V)~EbIG0^@$cnRzOpJ+r9XVb!cXC2FP!Yy%8VaYH zwr>O*<+uAN%Cn1TK&iO@o7hM_xkYECz=S!n*;rPXKurD7%I9 zR)gkmMHg65-U|JC7oL6~)Qb}H1n3=8l0j$g@kcVFf;Lr>-Ey+B#sZjXP3m{{>@K-K zB%U6dNb8oD;#W2kwd{|_v?I0a!pkM_Z-_WYBIdJ3w z?VTi?igCrgwY5zP&)^?6dhyF!I+V;iq&5Laa3<>x{`wH6GhF*IzpioChgLTGvku0G zI3Q+VJSub}F!6QH-=3_<7Ldb782J^}%y?U$KebC|qu8?jhAyQCm$n^Zit2UbOwKPU zMz|I7^Ne-g9k#U7s14)FTkY)aaKiH&%qCQzr2KRVv5utZCXY+xrRK=|aHEA$Wi8aW%N z*dWLD^EMQ^V~f82ZS@xU02YYBj-`ZONQEu{yo@?!S^b*yQL1G&d)y`p`oEUC{Xu_C zyfep-%^_2@kleF#vkGD&x3K7}sj4D9%PC;IzI^=)Eb%lpnNfMD>6PZ;H%_i#Q|L2y zIVuk#@l!fwr^4?llaBb)9$dycKv2CZwni$e+B z`1Wp_=~EkyGr+-5>$iw@1Y53B_Ch$SsTmxLFccY>V+J7*lB_gR=+VX0e)5fT@$u$l zJIR7swrwkAyk=#T>1JeYxfkug9X#h&AtSrOMhE9}#&s5~%R^Itra#E=L@CZWbCkKC zU*8#7AJvB_dqo&~nB{0ZC_f#UXB-GPld@#nvCA0hU<&&+}fn zef;)2d!zbwb*Han%$3O1HA1@=<>g93pMehq2H^dL?<{ zF}-jAW}O8-1irWh3I4KR(LEoZWQ(U=CyKBgoSf@1bCK(P)hu+qzjk=<)F*~W4q2vg zo1k$=QSi?3dw~wK^i8YPy-PF7-;$+jmf7@f0I9hSoZFM9Br#e$W z7*0J%zK~zJmoS-3IS51*yD7@<82F`jQoRSP6th(|YR3{+xV;?#Elv9;-z8di;TqC) zE}lPszWDuWAHD>kc6f=~JYouHFD2mEYGJ z%jJ20pB%1K*8%flwJh5hZFfxc$H9%FGHT}>$;|g#x9%ny5`%zg0umEYP9y50w|{%d zw{L}aACrliBPhvG-G?mWI<(iJVXJB=Gjt+$S$>5C5IB0^A0H3#!-7r;*DClL6~kSZ zbb7fb3-kW2LkV>Ml^Zv@=e%znx;?&vtIR+pGKpt|3zX@Ikt1ISI!(O_53t#CPqDF) zrIuJ+r-q$ev!<4OGk5JS%ye9)C6OTrvM3ovJyq2}@+iR5(bZrQEf^ucTw`zaVzZck zjd{vZyAAcsKw?Hj9@ii4l7LvcI$)ld+JoD%Lts?*SDUbox0W^Ju5`lHRde9tnZZv! zQAhACC6wk2MFu(je^j&u1 zqh}eC9&N0n>18$b%)Fr66@hYM0($ITFlok&us{x8JavwH%A>{&{S0?Q+hP(ol0>P{ z;D8$oN%R`h-pw%X2eQXO2CB+UVdIOeclZB069nXxoXwM4J9B-kwys?t9TA-^|Qbn5bhLrTSUa_y{Ao4NYu+M3$0vE#aRt{ z6GN%0t3hv0PXT4=6z|E6H;g+=`M_aaW#DxiZ*rF>7cVd#)yn#>l9CdGXh--TIm!P7 za=;Qh01#aSskqk*nWMaFU?T)P3FV=0U*7;ubA89>9Q0aNkwXuMIh^Km60$(@<(p*u z==esXKj=Xcb!}12NwWZ80R6YJWg<04Zua*Vy`=WErJ5~T?0VT(#xk*@x_}7oqP^pf zqTIF2cCwIn=72HG@s;(&X3uO}B~lhlq@7DbdJchma(UnbrCE)&laGBwvCG2W3RrfH z{pL)*YM9zdhzIEB=d2`C+ve!bWvS(F!P(#Z7#Srz&n=IeLQpweTtVZYclW`Ar|b<0 z$qY&ID*YJF;u=HqLv~SbIs$6fSHJe*eOm8?axD~vcV)erc=7vYXXi2GP}%sdc`At! zB`!-AeJ3S(VeF;9)6%R6hFw5nTy79e4*_ZzdQ}N!HJ<7=r*B;?EzqIu?991s9oEpY zEV&#B#&*@4oUOfx%bEGDCjDvGRX*`zn($rH_pef*j&?^s{;W;qE^wV|AWq?=3N?mR z_j1s%Fw_^D>F8axbMi|aI76p76p{^b-VTzLBk7Ic;3z$$^Crm1E8emq2dhTOU2z%h zWpQ5~B8uIlxUrgMIZrlj+SHs()osa=7Br#!QB`?iuutU_KB=~3`2g25OU`Qb;n)sG zuZ3uHZ^S&7(>6#8kE(3v&KH(1lIw;Od4q2h@xL>K9My7XAsaZlUl*T!N?ALD=_>Ey z2U32Xii=DA@NLwn%ECB(D)$}26>3%q^9C7V#)H*}I#|*!`$oFYn6Kr-jSu|>b2b{- zR+o_qxAqRS-M(6vQeUR-k#gfAu11b+Y`mXCS`66fM_KuUW(0e39?)85)e7fr+l~E* zkije>8VQ=N$Ln$IxJcvc|nLT6RDi|F!pDM5=3OiABV=cBQ{-Fb)7)O zKYae2%%Q@piOGTpu&wXhHM8{Etw~SN9Q+lqhgr`t*#M<(df-$nBsqOXAf1b2o{WFp zkFV0A|Ey!8R3DCECHVdaK+$T(XNbr&0J3%*DM6-061@(b1o6?~QZE$HDKykcgLm=~ z^Pw+Zy;?`r7)COQ*l0LqrgYcxxTo`xWqby*8QZ(eIQ2bv&?zF9=FXPRHv)$f`%jVo zspqHTMYtB(Q9Wtii3hU(>ddteWH*xX?&oyGuKzL>f?~zk@i8+VKHi+fp{aE75>J>VUC?BoY{#ds170Y&w!!8}_VzO>3=3ss4RE9P?>q zwLA7VS_30wduXw0BfA1(JNI(X2(*iWXoFe^Oi0m7`#XG!A93{(8&U+HeflIpqo0K0 zlNny%m>-IY?%dS@<|eZD(HP3$6Q+d!f?v!A?#%%LZt!W2T_@ZCUX#vc1E70jKxtr% zd^w%8jiRU}tRJAc=!xmk*cTC_MemRCIObs2BFYdlgib^c25qcfyS6s4P4LThq-~;52~tKA!6-xndroQ! z`b-rLf1z<>eCyGJd) zQOsStL}~k~1ewABe12qeqc#wKXXf5NNqe)rz#XZm8{kT6+H0ycY|4yD@y0QhAtbM6 z`xMvhdB`l9i!B^rKLLM7zIXa7^pn4EGbLOI{?C|7niP1`4rBv{DfD}$kWsVItHI4p zdC>4&0y*5drA1*~D~X*QwNMYSrTeNUo9EnX#egfC3RiNly0(j9hy^kK6vq3qwFFo` znnQthZpr(0cp~kXu~(tsS^M!YW|l2nsNbyFcJ(|ZacvdSC(fuXxqnFXX0OXQPUL@@ zL>2oBvcsGh1CsCE5A{zQVDYQKV7A2*hIYG9xqxP}Q>=SdlliBKuqz*^ijs12+6rz@ zQWRJ1?5l?EnWXAa={MY8SMgudDO-_Fv^x(r(EwOSMrfou$smrnIC0FDjg1G&oGkCB zc|%I_BSSOV?!r;5J!$J24u=ENFRf<7&zZA3w;kiu4pJ_F_l|XI2+PQxs!5CE6~esD zhI%?hC>GUqY&s{On|7L;FFsX&|LGoScZ$Nni{prbb~q)8P4OKFMmf%LObrtdB;_mi z(-g4jhpjt_#*ssBrTEL_G|p zkHW|kCd9(w9s3_%*Mzv~X+Qg;4h@V9N^T0U60!b0)DSfx?5|kH7^c)&$)P&u>P(UW z)Oad{d{*$fPL>ih%VJkkBEVIU%8ZbM z?U`aN=?~j|*0ML9KM%o=xq93SWB zWeIY1u;kUJzS%bzoI5!6+*b7Ua({5}sfEN&8>}-T*Q+rE!_Y4~`Y&i=Z3y<)i5?0m zbaj3^v6p==Dz4gzvooz`U4_$CkL+p_|L`k3gbjjQl@4q_H8KIRQ4PX^4#L0qNMjw5 z@g+w;BR(om=n-907kom7Wkx(r#(m{j;FKeOEM2O-AR%LWHQ~UK+LYgx7vZ&1-+kcJ ze!X|K;lRE-xA|}qS$k~6(z*Tzy_~72)(^3G1$0M(b+@$EU5+OmUTNxQa++;}mMjp+ z1Y#-xu%sIDe7z!fz5g7eW`X6lr#mepK@<5_s@Y}%xdf6D9R1U6$E};+*TfDCu<_2L zS3pYF*e0Il?StPC-IxN>`g%~GNW~To`t{suRL+xmkjFD!x@omv(u0UHSx_#8hdzV1ilTkYUf^svT`O-=_5TZg?r1ajh~YVQCW9Oz21V(ybR8mNetGs zj7;rJt%yiC!&rKU6j#t3%iwomG$?oSJ!-7Hma@I`d5`?Cfu4*?HAQqclcFZ~#f$J= zu}dJD0)aFD#etPj5*SX6B;(0_`!-}jkbP8h7Ho*+@`>!U*;ekm>7*m@Iuw@FY22>- zO}@E6MFU4}XHK7AWYI;s`1n|dZOet5Ecu2UWO+y)wqGw|A_sD!+aeN+@GM7-;oYy< zBx;M9q_y@bwBJ_Fdr$x{Z${9M-4sdzPj<*BR94Mu*7Q}idX{#a`9h=Jh6*67Jn<|quM_u2@MlJnVI%6KZ!p3^wOdk zqSw1!L~C**p;0W!G!zF?=U=Vn@Xh+GaMi9b|_RH@It$uztjq z-80Hh&bbpNV8;DZPU}P!)~1S>CVlJ{U+UAqeb=cSD^$7`FjG z&m<%)23813F@TmZT7Dy+R4p7)h~d;u*#By7s*N>J!QKNdfVcVko`)W%1Y{{wVpd@r z38c6|^=1y(1q%6U$^8+|;_gX{@u9UoGTuIAvr^gA)wNL7=Qk0{x%)-1&Tas8>Gxc+ zXweq7fJ<9woM$j{WvVyHp-yDZehgo2w5N0?$_f$HM&%9VI${!D@nq5mj%=o;7LEZ} z2V!Q%t$|#O1&wjNh{6))LadnHnDE=jg%DuM6Q|He$#Z0*n?VnT!>a`UA)V+U`li&! z-(wbSSyzBDG8-Rd;DVfTBIYAbY2G$-&X(?r1V6k|bmFBANl@XOIDMw<+&xk;rN{Px zqP8fnoB*+eg5%J2kNw3%cG%l}cO@hqxtt!}s%1-mUteFLcLkn)wj!|UUPc)soE3bZ zqO0Bqw?Y1wyFHy9=z8@kN|jHL1@jD1(Id}5E0lL9om`B&Cq z#9m7(4>`zLJ~iX)Ws=H%HY5|s!YxlR26qvXK@(5?<#T78=E%vwsd$-B7dLWS5f7o)C)!Q-sK&f#9X(A~chm_bjkv<*(%C>(yt16JXHh1}! z`(qnk=M-8i0u++<0Suo4q#k!_h6%f)xzyifhoWZ^dPkrprq66&tzb<)+z9@6((Bvr z()K`hDO0sX^ONb)S&@!oU-H*Pyio3Z!UI@*rr0x0aV?V)4CmYR6slodg<(i%6$Ld6 ztzF83=eL4vt;<0S|7n<7cD?e$5w`kBG99u7X`wfNo2_l?X<>FGyaRj$}!9$qq}xXb_S?0R8veI`p_G&28D zo7**{a3o|Lx*vW*W{v#Z`)ANC?R9{C)K#dFae!e4&IHh5fHG7Nao$6#iZiVyUR?1< z{Cz77R|6<4!v{_szW(`_+0T(s8hY*P=bEi5QZoWW=HBO;jFP8BFZB4h*}ns|KRr96 zAZRiIhcdPSK+>MXi~8J=9ajQOnHh+axpCpx`q%Z{8ygMzY<7fXEe!sJiMv+rm-BdG z8A~jW%# zSQ2p}S)VK`rB$e96jz}9W^Zma{TK>~h-go8E%23tgaD?KY6|@G$z=u$)?~$szrqZI zhQu{4Dfzk(0TEbq&C1Ge3q_H{iA$HHUbRjnqUO`Q{QKrs^KCP{6#_@G@O=qC&&Vxs zWy{_>FNg4D5V@y^Cstbi{$YmZW!vFcrPeyLAV*0%VjgG$&72*8dUcszgH>IR`HQPd z0PrP^EFqP{UM+e1K!e2SGAb1}@@jH=>CMC6f(!Evn|K75T5a$4J>E7W>)KDHz_G72 zzWa`8r-!a6V5eQ%)4#XOW~5FRHAbeUnSLIS3o6Bx=g$us-(5p6|4#$bB5@;gV~tBO zFWq|#=rf4eu1Ush3`Eq|sa4PrJ);;|njCS{d2g1oedyzzTZt$#pPd!3&OBUtUv7sE z30+336*FfxIwtwi&vt;&r%bsvbaa$mC&YvR0{IuKKHqubNmDWvAZ>lI2F{amo@l6n z^BG6pOQ*jBPDhPLJyMyexoUH_9;B@WcSG%tWi6#039)?U%o?>l66)|kUf+-DMoPDa zN;p!xnHp-r8bd3Jw`UivN_*ly+YNET5hiSIaAXiZLa88EX2tpg_2HP{4JTVI&9 zCN$aY%|C&NOu6qp^&&@^BgqGvK11xp8FW@tUI2Yi+K)_+NEj&l(GUKvaWi+e9XZ*FIv?(Hj{Ku20ezOPF z7|ALTKnK4NBNGk@Es70a&}x^q?Mr+ra;#!Jx&|=Ak>cHRSwqr{%cuj>5xW~WEt6eG zTIJa)dOXh;M>*rpIqw_AKlY_xRgfUSh8e+cvVblqmAes=j(M(lnlPzH41A79%tA&s ztH9Z2xqLT|nUppx@V2dpGMZqSs$=%oZ`i_i<<_R}|1$BZP=b~@~X-I35 zFGC;+3r8n>Q<)Cq_BYcjK>!5%7M(IQs5B+w7qE$Y7I7q#KgTv6hi`K$o}+;Juw2LQ z*CiRM#xr;zOR)w~8XPfyL?<8+4@N^8#F=r#0ROrYs)+qFfIVFLc>=qc-3bdPV@YJn z7>AXWp00>Pn5(P-`jWZqz;H|e=-%;1dy~j`T2|{?T;_-vMUYW8;d{y-zwlq}G(&?^TJC*~>n6UPYtwBvYM`IAUv}`3#x+eHVa-RisVwim*poaC~ zaUGF44m4?txQb4WUo-EC=$fUu8eLEr#BH#N7Y|8-odi>e36~4HZ-6qSa|>I(l@JGE zu;HerrYq_AC1DUCUMJO0$#M>t2TC6W*#>4_j% zuFYfNq(%8ZinVcyDDROnOTUv$HnQ4x1!#t-Q(HurU};x?n@{59CY_F=m9oa7 zE!!dLK@}*jY-;2hBvW1g?OY|7q9PbZf;J8WP2E$rGsnAgSu5N>UQ&+8Tv~DsKc3z} z+ykT`SDDmxDpwLNWeY z{RvpEo1s13p(gisgKPNZd+;QW+kT^$>%_SVk>+`91QM3UtYmNpTr;|oU)%8DR63Ac zZ39gzC~k$aN+aQyaF-P<(<5dv_JAKc4j2CtSq*~m7jVAmM=Btp$33bhm+SX#7_s8IJNXuAxacKCF9tAj1q+b+8r?f94uJ-+ z{=z)pz~Lu*^}_cm`Cfk3v0aKLl%;MIfTROT=5)gy6CEeVid0Lq)CqoMRyy&1IF|tD z`NLg->D%BY@hgd=h9ZO0HMye7N`|VseH(+K=yfX&Cs@p8Dj1aXvjGg%K~fJ&;g4n% zH;rl)qxiHGxdNbW33KHf7|ok^71bn>eGC;9G8A)#_~>~qOOqxuw{KIYOo_eNuKH+s zvYncf$K6y@vovl`*gqyo+wifJc0XpjT&KwDY-m^}_h!;5yO(T_YG4b&8|99fygs#s z!W4qr)85vN@+jFWio?A}{3y(@M2)F1y||+E5S!fw6|^8<6yetqgNc9*zmtI6dy9zD zCie)iG2E~K?wepywD(-$87oW)73lIfDb&wCLDhMu@9AYxW7jwB!rf%CX~$m`S=H(0 zeJnvk)$qXCgmksX(M({@Uw7fG1e64DC-c?nOSA(XNzeqRRNN^ zN;az2s#SdjfMwlL|H$%nIx-bUJu=vx7VBj#yiRkzJCabqW3BQ}Gq~p ztw{@{k`zLOA{xm=YqIJTk||>LLEAU1L^_9gTRJ;jPdA^L%?&(-n(tA9PkXJ^AUgX& z$i)OBY`_u5c07Y(q>8(_%gXKdwXELb_Hp&8?Ic#vcPfr7Tt1cM2auN>xre^MK64!OIBJUmH954kbRBR-DB-)X(S!>I zvZ+mex4-0Tku7BpIf=4>^}3&o)ZSq`k;uPtIht98pe(C!3(dZ#+NjZ0@>My9 zJK$ojkQ|mGEpqvc`>$9owaKj}cvJ{H541$<4`@q29!X=OSZxJyq9#sri8OFz?49_f z3lDJoqpv~q(1eBB&c#UP3Q^My_Kk-3n$`Us$a}QgZDtzp6)+j~ELw6c+%F$!s}~edOLiE8RQt7U9cp<5SYp zn~((Bu85LnO`-t%pH+O1QcKAl(Ao6)2xojp;;MDq$&(+B4Xxb$hDSB=&&6G&hXn10 zkdzb>(!FLd$sksZQGq;r+L2&i)bID}BRn`cc`}Rm*{fG|@H!A`hU`7fG5on58hf{fs$LX-vTm2%AWDc$H*EJxe;5E!{VOW=egG zh_}8F!V0+%;crD@vpwg%JL*=A*<(N~?-)KDej12oI#~)GR47?$5YQo=8^W3NG4We{ zOAMuEJ_p>EtWS%e${42TwrSHk^ zUNbd=`42>ZFoGy)4VZ)4znb$R7X~oFpd%>1(Voc2*}=tM97McKx7^&!6^cZ*K(#ML z4?9(9x+IQ%pg*CF(TaaHnPuJp1!2?)+|ySP_mfaMpe8bbOHyM-0TpI0qQsVZGuDKeRgk4lXSpgL z>@RM0ge_Ya5<>a`3$mjcaB|9e;hgftXk7f!=7gova>ja6K45I`{@26ST3vwXB@&5U zJR!=Fo-?L*$h3uXm`Tt`89>%M%kVLwMEdkh=8s#p@h%Fs0E!kN&m_-6R{cRt{`p97 zI+uyE8oir{btGRE>o;r||JRoLj~+=Lxpshl{Mfw*z0MpLNjA%%%v4*o?F9s}>UrEwc#UV4>uv$mVJdM1o^|)q{fCbkp zmfjsMNQOisMP3EvA{8AsDGB-wr@j52v)^)Cm<>3DDx&u6*{{@{2}LI4z9$K${7_fh zB0eWYOla;Y;;kdZ96P*$)XlJ(KTrjq#c7O=^KZ)Kmx=)8{n=&{`md8PbV{qi%M$==s#WP!3l(&@%V>i3f@B zg(Q%-i(Pzq_>?}TMSt_rPT_!ne?$M|{Wp0}Ar@&O99Fn|3)v7ZD$~XJs;{e^GIS)s zCO_fF8g^tk*(d+|bP0h;k(o_+(U!2x!g>m3M-B}#Cu}|2;}k_Y#I|u^gs!FA z=?+7>3>H3~Qn*0{*oqLBf{J6kaIo4)PW!U|0?;EDt}xe;tWZrYzw77qFQgx?&n(q{ zFk&srT7aDtmSIg2e(2fGnO=8_<^1E2$1`Vq@Bm``CBQm0b@jk}+dG+s?-$UmCUp~< z1=}{>=kOn(GHVL3WB+vt4WjEP)&(Yzd8NwJ6bfPNE1F+hi8$xTQmxhO2KCJsh~q$^ zqY0jOPCAZ+T)0G#;Se;T6^&RQX}dF-zmh00+|L=GCYKS2j4bfp%uTm8NNi^CdCEUt zyyTJ6NUkvqM7KVg(pB0UPLzl-9XqzCuRF z7wuu1wH%=ucub3fmq-q{-Mb>n*~bK4yS=w7Z7E9EB_#O7=bN<076|O>658;>5|)qO zpMU=0mp^6?D^!*rDU^LO5|z;~KZ~e84Z7(n>r(Ko9-Ezwhg z%&?+-$;#(y(3&J)r>_!!rX%S?uN0~C>C2a^?4U&`>Z0<#qT51ZVcOqFF<{UjCl-x( zh1nJPu3nwaIa5wij#HwQH3H^ZjCrF!tS%KwR`*KogZu!jTf}5I)&;`GBMeOMZlu7l zq!v$5x>MteSnEo(D|6e@T5^@GFf6`EWj{MgEWA)m#3`$ADOGJVpb4wMqEA;m9s^j8 zc^r+}Q_h{VE82`V57GnF6Rq=i_U27EW2%)TMb$Hd{Iy*{KWa-mKf{%rZFxxN(M}m zM=bL2!V(TTl9qw`J<|&G>P$@;yR!Ea-!=mmOGa;>g2zM$@|27rrX8^Z5${TH5J>Z^ zgpuFh>5vcHwkUYUcb^dd0tmdXuI?~iE@Q}2y%;1hjeq;?o<|I2UQja;{2&O;E{efc zqe+vMAlH%SU20q_TaN8PjWJ_xdZf(LFBJQ1V`GP_-#4w_caBHD=W<5DPL&y$K0U`z!^cc@^?bnbWYmdt52c(S5mba zY`v`fzM^f!=6#qrLZ9YAXdxB_jkHgpyG@E9ee(qog+s(!lT-K<5VC=@)@m}Dr;(xF z8HPsH@lwrsW7_8>HEYzsst-Ed4GG2HOtocnPDbEV&S3}>Zrk^(Bb4>&c`q(uPj3e! zNf1B}Eo>1&2rM}rWFDf)$fq%}xmKKglx;c2J}Lrc_P z$nt5DW0_n;p_}>dRpp2JcffT4A(zG3DeYZrNh(A>!oi?pabGmakb{I%m$p@7Dsl*z zg~hFS&TP^K+&%+m$#rF$_TMC_cIT_OpnG^C>o3< z96ps$huvZFi>SFIGe~Jn-U8{$>EtI0VbpMvQBaL>kaQ79a$run_>wg!W#Yn07*I!! zIyqk{eCjFuF)1ZuKZUQQv@sQaS?A{`uDrMpv=gdOuvPH3tK{&4#A3JJk{x4F)QXmV z^m_-#ItYaabu1`9lGUSsl+!B)`MgBwR(HNUYVJG1W*B9#qpK^tNnJ7g;K;aaHMAn| zWjj{CZsymiEEnn+@d6eBV5-?hrj&pt{wg+$oJMF-$zsA#3X4;uWFw+1y>&~K$xTG^ zLFr5fdUVmNH9wSnS7*vOU4h0w_x*l;nD+l!W#1$D!D>27rn5ok&=75T54u|r8!lfd zr!nva#eEVLy+GZN?)@x$ehathNJ?IxjwYu1M}Wg?R?N_ zC7=Cu)p2^fg{wn^x^l-34K_y;P0jTHT|z&z2UgQ_lwHKRyix2*kM{Z3I5TZ1WvMKA zejf*?j!0RTGBBDlBc1Ut92bqRUVZSkBl39@LcDqwQWO(FrHIh>WW4lTR}a6#BYjr- z3vznWixfnoviQR-?#UD6a6EYUHfxs3DShlv9|BfO{h4*??g8~_43>(8O)h{&%K?iz za8Dv*@3$g=nSg0fv|EQ6`*qtONS~V89iL7NHAByF83nJnulytMmaCQI9wp}ift^y` zqrt|^C;aejSuSi3-j(Pe$t@3urjN!X@4TxP&=W_-RRmY+XUeMnMB$cOuY@gy2*U>kK$vgERgEo{e#mi)zg_3(TCC&QKi`1|kVJ9S8O)X$Vxf&C0OHAMnI3;M}Zr>@_1 zPgo~bg2jK((@G|g-&9RjIXS{5(qCyZ;odnR(ABn$*kF080-_E4(8N;;TS9}*(NCg~ z0GcE~j6}BqJ>W-$EA}*o;j`7%IMkwbB1pVS>^~CUjOIXb0EP^ zA^YBoNS-^yQ(DC#0{1X|!GaUEZ}@*ZRl76s2(*m?i|8bk;LYh(A008C1@)x)2)gd=28UCO?`kG>vu01#~f%=>Wbwsq>%F}Cj|2`Y$e?Q(&H}!v?-hY?K5C8SQi}R;d^3Om1@9aJO$_EX<=ik+D6V{vIH;HIA zZ3fRmibwsPM5TT8>XVc$KI6x$(`2&+nLpZ=D%bw6jx8_Bt*zM4QLSt6{Evx0Z0-k7 z>kj&Fh`7^VY?eUbtcucbb^`WW$@xGh&Spp>ueY-b-hht?4?*0&_`7Xox6V%rK|Orl zMto0vs$a<~=BH@(y0tC77);BRiiR&q?kVc;T=ZH_uJFROKmFSO`FibB|EI9!<;18o znJW5}jE$cX7)&xYMPhV9rIN_*^`g1JDF1#{ynp&*1P)s`g|bOxQ@ewLh8v3=okwU3 zl1&ao&V-Jcr<`WY5W6;Nf|=Jp_35Oirv^+i5`s8v8Ds^pV%3Zbg5>k|aFkwtX=aZP zYia4ULuP$hs(EZH|L6(YMoL0@ZhQAJgh1MvXYTsT7iPLf`-?O1rxN+WcI_(f4+!$A zCbz#|?VYL`KZWlyd0OqM-)kvIi)tvq6HcPcLGaZMiv1cBA?PRIO&U-}3ig|{6cBwR zsO@x`tl<8Vs63>!1mYJb?vT(>P1;mm^FWoqlfH|0VrbM7M0S{Dv<8yV#(|FfX&g$_ za>g#=tZn{D{8y76KW<0LHKzYQ%+48)OlFulPe7(7;xP?N^OsLmI~2bVq+RN(Yp*2J z*iTmpz?zTWd<`{ut&_CuB4C1~k?gnpfSVUW&d zAvftxpySuU@7^ZBdLA0zPkYn#zbm!aLigtlpjBZy?e0RL8HJ?PI29e$!IO<*pa%N` z=Bge#JSoIE|1H&KlVX){VJZrQj*_mDWCd500V75bZ8%bS?tuf`nZ>*YA8BDq>3b)J`!Je8Rmlv$BnfowS67Lk7(XDklR}91-X{l1PWdBc=Gg-%L&ZL z{gR%{1K8Mc`(j{Ko-+Fa9KIQ1l|&FLDv}1Ex5ycBZE26{Nv8m7B}dYGK*E}Jv`AuE zqIDZKOa`16MKLA16Kh!c@KYD;M2zVK72O2vbFJ^v0h#1tTubJT*`_2Vjf`=bfiD>( z$tCh`x+^yT|A`_{bDGtj<==G3vgK;I5NZ9Cx!)o(Mjs`TWZ7PTvL1!V4Ac@+LiPTQ zAtXp=6Nec;ckTZ}PNv-nrc!gKW~)}0|NeV`Z5lXY`pg;Q${ullW*Y3>cgH=37js-O zA9J`SNXcs|7zEc2iJ8g=z06BsL%MVw16s?%0;C-duw>L4ckX&%YgeH9+!zKgT&x0| z&W{~MgK7awFcB$g8A#576s)#Br`cW*&fA z_|oxoX#Ei8vWa3BLUA>Y8(j;pa8L*Wjr_F+rpTtYP&BUZ+>DKD5ehQqzX!^rm_DZN zDdYf{FWiNKbp6_n9=jrrqXiP#nlvOrl|}ry0JKOXGKyhLgJ7W>ecJxbM?e1Tk%#Bu z7=bUlSyTwM2<<9Hf)pH~VQ=`bVou4{VE6k%M9n&C@!xH3?6AFPS11E4|NScdj7v}d zq|<164p4ptSK*-K=-2Jd!v_z9q2S-l%n`w|*8=XTE@on>RaNYC*5}ChH1>CID4P?d zRY?Cjq9xm2+Mx&VQSY1-M!v%@o>A>@f%=MRgEe5^6bfntL2NynX1gfr^4PGLn7AqEqoF01><=@-L--pH`b_4Ee zKp~Adm_7`HME+jHu_lw9lih{)0^Ox0;};PXAe0c?gBh~)@T(}AG;O*q>^2>i3WZ3U z!P-IW-OwC@G(|pa7oL`;$|*zZX5?f$)RKxsD#^Y=k=ezlMZgy1XZHXC1EEtW*rA#6 z^Ye8WTRuQQeEbLHbSzO(C7rjq9$bcgKii$!_ zZ~~PM$yfhLx8$#|D$k!HT~#J{PoU$Zg0)Xz@->a_B_?)At|BBGW~Q~(n^DWSvu#>i zgE`~;(z153Q4(+8eix872wa3-!K0J*hxVG>Pp}4*P>Cr$YZJ|e31z$tI#_t-llU@; z_y;GCJ9if{d_kspfxOJED8EI!_U)6mxP=OvnW|>q8$)-_9{&~im}w|BclY(CMY1&4 z-KgS%>4IosRk(-mJ`LS(*3chtgsSMpI;R>v`@vOyA2-h7<><`tjN8pML9huK(NEXv@nDg3^eh{ zllHWwB$AwfAo#0_flCVpK>SPe;>RMxBm|E$4>v#MGJMY~A8}#R?y=t1@YmMT_IVUuBKDbng5Vm`kw|Ixmbc;L`j&#*VK7+=P-s z2u!(Od+CR#DhyFkl9*n)d2>wfv^4+FPzS%(9V#onIw*#bPxRfa!SIjmE~1jk&pz*# z=hrp!N~9;lbW{|id}Gc6V0~4O8`4LbHm@HQ3ssNM?RWCZ;?FwhqjFcY??0A6;Rq?r z03>4wZO=_V#I0h7pGyC*_vw-Mn*pU|1QM@QV6eS zsA!5(^EhGYUzR9zuWr7j6K!7TGD+xe#xRP}oHHpc|wZ*sl!M;hza#td6fnAaD zZ2*?J(78W{Vz>|1-0oG-v`&3zQ-v z`UHI(sXQ(E5IUfN?9Ganvf*W_<3R)*G>VK0-k?slD)fU`=;gX`UqpIA8kx-cXRgwA zbx-U?H<9#opX)PAsyqFq6@g?)VKl48wG9wvvI}L7A@J>L(jEb|Mcs&m?G)qWBt9YN zwT%Qol_nygw>Z#psVGSB(1&P1yWgQmkg>0YuYA^1SrQMd(QB3?<`dqVW3(pjOkyY? zq<#MdZAdCIqMl%df*o63IZXNBl1>wSWttJUaU{K48BzC`FDe2&`4NJxBh4YA;DE=5 z&A5X8QnBKZ>DFn>8>nk+w1;f_~_;K4MLT(ywd(f@oETR8n#Z~y!5)qj^Q(DZ)$%a2#^!!1*D*7<{qZ;d x@8Bmt#pfU2?th`#AD7mD|NH+rxc9kAka?amf6rU3Pxfc;@3D{HK7QXnw&PewtGvVWJoj~9*Lj`idEGaa6r@)y-L#ZK zp{$@DKcY;b{3b}DEUNu&G5)3Uf!tU8A!>D0&5CAr!OB+GQja39Yh`}X%<7`y*{wEu zmX{38Oa=J%?BUzXyVbzT%KVZjKflSpUcqN(sn37Ky%!%s`Q7~Z=}Qy}^BVHsFZw|x z%9LLy6zY+KDwiL1*V$fXIypHzP{g=e*I+;6Wh9Vn{VBEf9^Hk z()_E7{y2K@k9CVyA3S<@=hloHTQ@9F4mr&6>+x0GLzCG$qXm1@^0rrbw@jzTWSl8_ zB+?dGRps~LLxw)DtL}ycw7C$+9duc6$Ush33HB=$s!QH!OcKM53Vu)I`H|B(iyY;NT zBG#Ym+SizXFT27{9?vbW`V5Dz{-*4d3ht8;5Aq*yKMf1xN;E7zc6IRzW%o_{#F=#S z9(Z{@o1Pp?%Lux2XL(Omf^^pOsK4`Mk8n?I3hn02n?lwRsSU#Y37Ng?-J0e_Z3dBn+E zH*bEbO3?SB)1Ndpp7G*QRamiENTsbXpghgi?E3ZV>kK3)IoCr+Su!tv@ZB>pJ=PH2 z7n?Dv9(sRc;N>sBN2@31tC+^Ay)Y`@c`{a8qHE;XboX`-`%g(h0ebH&7>hy%h@@{rT=J>G^O{I-7S+l_2P=$@AaE{d^4Wy=eM=c z7NSs;$j-_+p6$9~1$Eqhd~$O0D)Ig+7Bf?JW^O@|mD&y;4L&~){%AZg+)=->fV-_I zC^@aShr8X~(Zx1~$-hlXNKCtL_LVG*@@&!v_6qO?XyP0e?pH?40|RaI3=QkqqV z@>83hYK6qpnT~eN_Z21c1Ws+`&y4TmI=*GR%9cr5+LLvkjMq-RgzjbQdCiULhuVrZ zzkdCC<=CF)TuiQ+UHpi|?Gn%W<_7@7*Z_a!a3fBOZa=1!A z$Z&GBM?K9ZdK)`?1`AzFzwptX>9ei1DVMZwy>nWc`pD4GP)1SFmb%=BPfOXzD57-a zKAk@NL7<3h`}Si37d|}K&>#4clR8qN;pOGks^X6&Eq&@K8S(b*5s9G>0t&ecH1kxk ztZSrC>@}_3ua#<*H<78GVNc!>@NmrEEc1lA`oruF9mj#|fi5%SZ(MI^zHC-gPO;Qm zv2pK_Am@qj#O{$woyx}7S7qOFt+*0tE@v4{kCzlt9QWaW6v~w2s}=lxDQ&s9C$%UP zOY@wClq<7lMS&9bo#CPBEXU%QR<92B_1#b!`O*&?W<8&#!A{}jhZQBqLrR(T`Njqd z_`9aYhhlzP#bqgzpP3z)pkLI9zrRHDuFw$NiyLvcN{-SmnxXD zwBpU%bN>Aw{$noFBR=c-wYz(t-&%)I6OqZEdCc`FbpPb(Z^|D6RUY}I?dRW2p_s2& zvk9SA|M=k5DCWaYo<2=PWUOEjRat*KM`f%lHmhQwIXB}Ey@9W7ZK229H~DGDtlzL9 z%zKwcXRsS<$&Mh$Cu8*vvBf6Gjvh5E3p+8<LZ|R!DPf6VItCdgF6;G~|1(ZDni*Sa^`wPj)$*CwQ-A%k$6|cAQT)u39_pW^xK5VG7EZD`+ zIbz#Ua&+O6m1X27q3}_-byo#9AD;p;>?v%;k^byek(RCb)z5XW zvf8)*>2J45(BSdW+v}?y z35f=uwtWl`Q*~R(LJtZG>a2{7{bk`|gI8C6v03|AGsARnB5;eS)$!6$#Y)@=6B}FX zt#v#R_~1=K7kwj6MH>lt@?_GKo5p^*?#>=-{pCj5So!!k@6yivPaD-=7&Hy9Snu#u zp!QOc#6vOLn2-Dond4tK%Rk!x=|a|NC45Xd;zp15wXZLBmDH3{U8*%tvA*4K`w6Gf zDt+}V=Zr7S%_os7tb3|Yu96tOyYp0(Z{G$jYq^3|CMa#@62vZt`gMV%&I z6eLNh4!*P3cd#l%n8c*@rWP=wE_{5bp6Qqp-m54z9ikuPRKfUAZ;`!rvgmJCGcD_j zyf@M-3IZh@hC4!oIz6Rklnx#Go76-kuhU}b~+3wdH2XepvU_AztK z5PlQOtbO@4<6?VpFhkltS~EqNq#XGw1g_j@gS)Ah2hNs9sw&d^c*j~bMbEctyaL^zLRpP?d!%$W*tY}Yp$-AbD| z<;ySQ&gLxjPtqI+u*{Ks{M@1?vF{}vVYdijceWwZiIF7jv8AId{q*V6NV5CnhWct# zO53rS!l!D?vaCOQCv>?0gxF-A8~XaOz*a+qp%`TPfKRKsSA5-0mEEVJ)S8AR-|W_O zkh|m^>B;FciLj6xlJggeF5Z*&aL26o``4rG=_6g0{u7a^v4M4|C?|#1E& z&FV(R7y_0W{0MIAmvX9F)?RVgw3;J(vef<7UKV~oymzQb$}HnV zpqG~(ZvW$RHC`cWqw)x^u8+gDA0CMaG;~>idUBLubN1}nJNaHbZIJ;eaAJ45Q0jzs zwRyz?3Z@A^8@ylKKhYK2R_%raGl06g!KC^{-X*(ZcQ)wSp4Ln;-QuK(_@wLm>*YV# zQ}{^aoJqmD25!~vh2nn2jW?aF4@%zt`*|h0i%X9l+UDF^$FPgdoG{AXz5HB7)M-2C z-ohxif-14VUzaS|e4ClS5GyJ<*_|L>dCA=Td0i%p))AZ5Js-;lFY)G)UGq>M$&K}< zu$1_C@-^M(5?_6GtlhLUZPqGhpn`errWve5>&C-Jj$B*K!eWxFRY+Sp%ko7$N34x| zv_-$6n`cjD#K{82^s83~*f}|K7#l zaAboUY5Yt!9VIzsGc-jBy<}4>s~2@9-23-`4m%m?y_%KP^x^u|t97GJCooftjA&tbR1$^BR z{JN;9=;icijXO5oX5-43eE=#35o`bWkpVV{*+kV{yDQw_SX%9knicbT36s!Deg2CMmO23qMNhW&rnIrr z{6|eCt1gYz+4_HF5jK5fEwO0H693GF9;RErCeGw~uoypFzkdCNF;&Y#XK(yfVEWJr zx<#FJ<;L6VckRVS)^i-lVLmp}SsqX?y}{Mk(=f(6=v-cFOzo@+5}K05RG-VN0BWbf z5J7ZH%NfWgXB<9AIsCo?HXvBzLdQ=_G_l6$`bP*<-q1+~XHYqsd$|xlVu0ruc*ecf{bQrdU7HA{y7B zJeLb*CSB$`?w*~2Uu=tzs|%a{uKYzP<)DGn7}x=C$sdg@snE6}2BWpg90{?Jf+GGBYY_XlMvucEn@Nrc|qrvzG>> zL`ADjGk{vz`S=Re`%o5-U0=FRZUnpeYlU@lyMuC?jS0;k$l*n}wkNBET-UqZOJ@Qs zTO}Vqw%DK6*I$;I`aI}*LJYRB|LD}zRQJjhhiBKW{mRb4@%DoD?VPg-I`Ot~!*#X| zi8b1$T`sjREf|BqY#Rv%x7F}T5dLr(56wL!gu~w+Ou@bd3l|1dBY3)!^;Nmw$8or2 zNVjVGUG`!db$x5IJ#L$D;DpBCJ#lBcIRynTel9kxd$V1T>B4MBXLGdf%h5i0{r&3v zsC#=&R(T|aK9lvSE^8L&w0#uX()amG#}Jdl;Nt8AHST%X zk@`h}i4|R?p`PJ|Ixf?#B$4%sc#pq{C%qnH#&O_tUq|x>3Z?$*4zLd?eVLOA_$KA2 z$K1;vA6$q&Kip+@=9Z_YaCyephsIVP_f@JjxB$X1MDY@3ly3Rc$lv=MznRmt9ENIW zXlP8T;*XMF`U{zqh0}IqeEV{n0Dsz0n~%Ws^=sE!sucR_GvxK~v@8sxG7U&Kitm+c zEoI2lz#D8S^Z~)~>eVapN)zOJJUle2C&TtHT)5Dro^8MR0rcoLT}ejeGN=uwvYe+c z=4XBTvW!L|`y?r{db>1|1Wi`$*|SGI;exw-6_S^HkNbu_hroBBAJ1}eNsALTetTne zyc%!C7-H(wfdf~<>+k+~SOx8V*CYyo|MUS5q_OIB+mTd@^DXc1F@yk(9_Tw>Klxzv zm%wUOht7okHr>7-CjssAi}?F!$!1y8r?|Pfh4gxAl9Q|W>zu~w7~|b01u1ruJ;@^_ zviwyKPDZFujVq!)p|1&wO^SJO=Jf(v)s5BcOq(_t(JEeAoLjMB_p1&YCkqF@_Bf8* z*2(BCs-|M~UvG{Sj<&1&unB90@19gtWHHNV^3xC8GZzj#$1%QhaDr~dPX~z73@lBb7)^y zkYwI;h<4gSYIGq*iODy%eE~)311i}GW`6At3EBn*VUpu5YrC%3qH7@R6vJLC%_dPS zD`T+1c~TyX!_t*23lsa=0v!~rJInTX)ufGgE-f6j1tl{*HB^Ky@cv3x@vv{-E`xbe z7 zp^CJZO?kY{&8?Z{?Wr67z#SxP7FgOdSu(0wjGEJJ!dy6&(QMeO-FMM5rd4iGQ&lyA=#)*TrP{bNCs_s=Ppkr!)iT#yYR&g5e_=?Q9&Zb(h{x98VrM@L zfG+g=iVXXa5@f{}=;Nug+r^e|*nPldYQU}In+gAkLxTaWg-!W=!Th`$REfE`xLXX4 zAOw7*6B2+?DZ@Rn8wuN{t0L?~wy7N|q`V%srl`YcR7`4Tgyv*lO|qHflU?Ha6Amc9 zr1$3JfDxZHGf^-x3)&%VNMu4mWT)*ZJRo_!XEI)Uy-tQJ=cruh ztCV!rRhwJk%?^{R911M#9AlJJ{`tAxd3Cv0@#Xu1n<{l&6355#g<@Nb3wp44vHxN5xZ>Wtc|&rzL4V`xR4`NW z*%jK(Yrk2}%*@ynuuRKvOGro*sdY~;%?lXp>;tD;nNTFD@i+?#*Flo_ zKpKQ1--<@r4tzO^MN~w;R{DUOHvD$2;&ICfvg)y+TK z{n4!Nm^`SQ@DHD!ideq1uWR|5AN-VaZ2E5@=@ZpUE?esKgcofS6fCLD+{?(v#}?>wavvmNqr(@zh4~XYEdIwa>x= zLz!G-b80gjY~=&Rs~EgH@V$L%;QX1uRe__EroV6)Gu3)&{@V1=)r{4f91T;nrX_~G z^qN1fqFm{-?DJBhP#j+&<1wvWd(MeP{IU`XcX`fj<`D$A(YO948#9VZn9^%g2M@km z+9G$HJOI(YZajy$ABt+0Ng46EcP|)l{b9bw3xiS(qasV?3Z7`^Dt*^Mjzb%YAg5!wH=Tl1v55 z)T4dH4wOW6%}aqAQ*FMi#@_aKzEAxVJu+#TugbgL`7k9qUD+{K9;lpXbkcup)Y-yO zBzw1@!{|kt9rpTYUxS%_)#S)&P0dG!zBbZE6;XymGq3(y*a=A@I4fb)HOOa+i+5pd z$go#W)w*0dPgRYr=)|I60T4uC$d6U&2yvCiN#ZH>RjgV(?Cja5<9YXYNNmZ=x-~R# zD!*ZLu6WOiD8{wQHvFl$v$KtRQLFJD@O(KBQ09lbsR+;Lxp3|Uj7TrO<;~FIz48FhBA+tKs*gsH!k4l(AEpeaj zpd)A;Wzpj1**4U_Fl&%3{g?pk5Urh2ZT0soc8|keK~D|PREt-!UbJTzf{`1>zlu5C ztE+Eb=CMWOY;rH!xSq19@pbFg>Dk-#f?}~k6H%>sL{?T7aLB(5SrB4O!7xkz&`{~u zkN)dcuWnMw`t#2}D`r(APss!432S|;N;LKX&u8|~77dKTNYjcKZQ;z2u%3mtnln?K zrzeE6Br~l#=t!em>R(L&eX#V>2AffOildY3AL?g-dn5$l1eQBW@6)GGgr`Jx?CC#d zu1#n&U|YYWER_84gI5$sf@koR6 zphp-XpT1aiFQGgRJB!wwvZa4)Of^W-Nv;)|$8N8UuE9l;lOI&-f&P4Q)^6Eyap$Gn zA*TEH@5}2!_hQ`dL0eAh8ed-nt-#NRo5l!J43cErzDD>$JdA+Y%}S$uye`-S1kHZT z@c|I;#PhQ>a@J3WHX?et|8aH)7DN?op+mE%`O*%f<`_psyaHdUp02KZGN9^zSbng)+yP^M5kxzXy7l8 zH2yBa`t;JBeRxlQDT}S{0t}imVuGx2oz;EA_h)c2+%9*Q1Lo7Jjuolxu&Hpl0$dUD>r-SEpVzc7S?TxRSD{jz&`_Hj8m zK~|^X6SM$S_02mJ9z8egBc)7ENy!xeYOeWC`dOZUBcWSaj|8 zwRb>nf<|ioHqd+wO-4@=AYIp?oD+9j1(#><2QvBeT=%fQDq0*l6BKUWPKlXs%|a#+ zPD@SGot>Q-c0Cy@*(5KP)-@%=~H))5Hh&vcUOK{ zcn583{f*nV6Z|sui&&pRCu7VwPmib?2cWqH%|KrEIqsZ0Xp-s|wu3FyW^mAZOfyDK zvG3kpd>6dn`Ojrl5IO4f99z9q@4R>+i0fw;5~_IF!SL^{FfIwXX()n7>=_|DYbaC-M!LI*M(;Y;gN+${}fsa!+G)Y4QE0aj7=~M zxUFIn0%b3nHb7eYGKSb_P>f{#rem*X>v_4PyN=Qaz$0oThvM;UO0KoM9jaBC(MbEr1V^h`VN&Y%9@!^v;BD>7)XoWT9GuTZ~2yK%?|B6 z1s68B#zBV1zf{LHl;-ZCP}1G!NYyW$*A|13HY{aSNyPC&_dNk3fE}Ee;74@IsBr>` zCw-S)M_(EzW(K%%B2u3a|q;;uM(mP7QX$c9xf+W^j_Ped~7q0h~n_Qdys|(X7 z#`?@kD^O&Y)p~QVZ{Pmzh3065mq08M_@<7I4m;-yqgL(3ugkP?b0?BbwVocjzSL%t zP|f%J{0vJAvG-WsF>l;xh<-V@cg5C2|YT`NxL*W?epTDYnu)$@G{)zC~VG1=2RceH!C z?G{=Y;kRYW5~>1Y96v(yHcfjJdig?wvM=^xA?=Y@P|+LJvxWD(y)OudE9~v7 zM0x72Z?t*gKJf)BsrHquZgU(r#$&kN+edN$=mg82wv@I&fv-hDePr31-}hnJ`gFMDz-B@_Oma*tjnNWo zMts^iYkLs!!vL*K>2x`|8p5=8CyjvV*@(n$n;H!Wuo-MxTwM~MR_b|S-NwPn_xIS{ z0o45S5jTcLMgCYKP2t}3n}H4gEK%0vO0(|Bx_iZEHn{G&_<6U#kSweNkEC4cRS^58 zj2SaXn4uHq!^>XOrP*$FZJM(G))^Vwt?k=a9_cf(;F5ENAtd)Y9WI?2B9Pxf<$0Lj z0~Wk;^bxb|c0tV)vsj%h=f9e*n9Zhly&mFEjq4>j^85>A?B4qb;Z4hId(6AsfKAT0q71 zca{Kl<9-SR@eWB^Q}RV;tFp549g-QCN^063Hd4ExeRdH!-eo#kgPUrRYYgh*Dq5Zq+wwOuC`R&bVn%T$B&fXBDcj}b) z9;XSbr$|UE$5t0=9}!wUcCpCLqB5=p-U&u(U;2;&)^x`|Ys#m&oA4wm5h{7rKFE8Y z67&_%K~cbCk|e=}g$>!7`|j_2lZ=qi(9P%G-HEGWn}j_EQ6*dvqt&G{`F#hHYKR?p z@~+97nmszb;Qf;}&xQs$8%2nL$V;>K0aL1?lli>1SCmkN8;^b3Kk6{Je=+4s^6BN- zyMLt|*fRiF#3Lb5$B078k5GPgOLj8v{d*-x$8HOa6nP62J$(jFQUUZADy3p`nqewM&;SjaQJBlPd$fH|c~GKplzZ z1vih}mO|c84m~}+4zvvlyGp&@WM`9-*foiw$s%GIa=?VD93X0q=Myg81(If`q@;Fu zcsM>!85V$btgPCN&H?w(Jfqs|s8ht6~X&D0JPz{$c z(Lj=#fu|ASlNv>ciZlC;~7O7fBJlKw_T$)*7#PJqvI43~pLmyV?~8wYiVVxS6474oN{`9FE`WH4i<$0;ksMXI#1v2lTm zH}6vK7mg(SVjN=8LDyTD3N88-amc`>W3m3h8Oxt@H1{Ugt12;?1_t$~{s6fJM$7UKmd!#=JJ$Ma=zyp3A?IQ7^ zJIw6gt&`^b#85-;?ITSSSZQHg9nl~O`k+3ggm}22c@63O6>E$9A z$XfdP%$%gm>cv?k9Y5b*{aORE&OY5Yzxu}$^(&NZAAF$rQBt#$IG{sH`LKEUh=FcbXhc+5hm1yY-5eXf9{PTW3GSDE(uT;$-=K1-`UD40qFxhhGNj zO%Q_6E}cwGB0wl4tMNwYWM#tm_}HN@Llt#kMEm)bRa^a*(3#+uLRPaATmJ0^c-EeV ztcDuh2fsxbZ#6qQ;GoQO&gS|*{v@uB;|BPeE2^3fJ&6w`9Fmr2D}hen(@qb2{(LvF zC>}g`uzn@({}m@5Ca9F=p4}7*A5kGnGXsXyD3sdlvm0ESdbZ+Y@YDDR-w0dHIs@%v z6Ytq7_T5c>@=b#K)}FLO7TUdE(l1gd53K&pwc6Z9tf4=DVvhW;b-DiU_2>T2uKdcO z4tO9Q3YkP8R_w(dY|HcS@+_M3FgF$~NN4%JsCUt+GxP)P-l=EhE#U)tl6-i88#>D=d;3&|LO=uxg;-SE$B|rpl0qo`dOu>PaUlfk$L;L~ zZlxdxqG6XY1OYn%gA;*8tB$$-eZ#|pe8`$LYtmVWk(@#~x@K-qQ7GShQH&xX570v( zDB{Zv-dy{=8kRb+L80hUyWs2_w0vOh1-B8GfB@~pi4&X=W^vU2ehaDc_!~QAjHA!X-8GQ zaDhQQ0ue*l=LR4qiR^-M1=7pD18j&gHe{U;Ser;RKiKom^bjHuDN2kyK+H-CDU^4-@+2~UqSK^}7NjgR>Y35Z~bfHN`GQb#58*vVWUf4+_(2+~12CDUbsb zvv8reOU;gZB3dsz)E^&gwQmcsB`V2C`={&+)%aj07MAD_ALKFy^LaJ_h%e86uFwbc z9DFVl%mVRc)5hV#OoRyO&L*Wzv{baf(WPXWP#U@S>^afkG!{TS6lo@Gl8$Q7j}wt` zflA~{iDi}en@)nHlOG4FAaO+Gnuns+5mVFC(WQ7LJSU0uq{tBum?-=qEh!&x=P#1} zq=KhR@Y6=w@gIXsD{9+IvbX`w*^bv=*z6Q6UnY{aRP_|gXPHoT5vmHrydGd%9}Wr8 zUla(b9I6`;1Zh!)!x>m%I8NWc*n+hwy|YnZm;eBPE=CefS(y;359yq^)lfuCCq5)@ z)WuF2R1|(hHU!UA&MVu#4Ek3=Jf?@BT;kyqwvT0Nc5rus`TFFhk=6jaO%hfNor-gty0C4CgDRO22YQu zFvmQ$aHGA0gr)`iR7Cb@klojfz`dRDw3dS1P6Q)`@qisgdd|3r7x){2w)Ejo(rl;O z>Js!lvp@Z5DPUB_rFY?i5lJebp`6ECnF%LGpOKe&3594M5DCeF#4d^(9)U7qjDV%j zu(+Rl{r95rUR63=U%{X+-C4v=Veb{YuGlQ-JazfHt1H3Zhuu~tiZ#GALfpe3C&_RG zkvfT|7s?UW&Yd#o3rYaJmocp+EAsgK8A9LU=hhAUtHx}Qz?0WNKt!a3)!!JAjtzS| zQ6chFbQ#E8>)zVa=)6BI<4=D~xVSP?v{mFmt>ZbRX7(m6A3_P!nl)#4_JP!9wQE@o zf9xT2XrI#UhOFRmQ4d{(qyaF0lhAfVdBN%n5;LS>Yn0*H@3_9K8E!+608to0|?&{5_-ryYb#dXEiZ=D6=iXmo8YvIbs&zn1q&8D@be1?ClPU}tx%|xecLvxp2W)3 zb0_AyurfFoNk9hdw>XrSmxsMdw;Js}O)ViN+aXeg!0VL5DjN)VPW$DTS=){1#y8Su zGQyo(<)_fk9-pw~aP99mPuqh{0cXDgNgA+_Tu4?xf*neJY--iYzRwc8LH5Kfmbq0u z^s%RHJ0#6I%iB3m&^g0qQdBmIw8l9tcIaAA&~!s-OVY==He>7ec8PMOXdEjer-uz( zfu4B;fP*yH^bmAT>Y<)nL`p>898!pXecSsNk7Lgcx7=40 zP2`1uQVc}@;NioE#u*PDFwv?&Pe3>03XpVCCze0tnjKW))%NDoGIDQ~5$V!ve&=q4 zc;J}q;?URntg)jctE*OZzs!f1jwndE^Hf5p!Ec9%2VkZ-258VLm@E$93pY;3wwbb;3oF3&2e#Y z3L}ZIuaTmJ^&xNv2{sg^iU-DJ`}S6AM^7Yff_ z0RdvBPNOo5bhRQ4SolVni#T*0iO##(M~UIezBM)*m>{ZLT>%?~yBLP<#OW zTYiEzd}mR!I!M*9{RXv(6FVDi&ygcX3^3|q6Ivx3^IRWcfy*Pp7w}xq_TMxm-&~OZ zsWT^wBinXi6Gg!G3j?eGwFC}{v$lat*X?WvySWKO7pp^eoOtu{6)UXI8s25Y^DBj6 z_maFb0MrRk1xudJ1q9Q+A4pW`2x3nTZ4yR=Dv@4EBkPmcgXt9jq4g2&kEHd9ExpNL z2~zp{MEC9T55oWk$TR|}#9Ek}K@)e^dU&w|p5IgKk4`{}>)Nhwpid5|H`G_8+g<9Z z&(PQ`Y-**Z4?tlf{grG*%2w$-iG;bt{`dwDk^-C(TvU71B}dctna7W}QcI9VdlYnF zU&UC0x4syj!F>)gG%zt-s>n%2ojv8ZSFgJw3oauLD$l8?VkD12mgLBmJ5#U;H70II ze`eN@i5;$1m1-SXCkpz^W5?m|_kZx8`@fbk{?&h{YRn^>-r8_*xCWp{AOGJkv9zDtpM2up563>u0`;Hzv)|olkO9MNHTJu%B=J%ybORzz9 zbTs_xj!29IAWtbA!n|3Bw}$BnWo3Gx^MoPZ6#=%wsQB?;L>5Q>s&gFrh>q%E%`v5y zT`_4KR1C-{A?(Th0K1nxY(Md6(c;BMfXO>GlA`cXpTTn#i3=x$09d0C_%Kz7o6IAA z^&hWjC*~6Og@2aoEpd*J)JU8oD5l&ZBBz(E+)|7nhbi!xsC8#MOed8XwEFVtB_`q$ zA^g_H{ z6K)YjSqW}&)0EcrsO0iU9en+JzkWqkDn)gVLeIs+%}t#g>r(~7g>$>MBq-JpCi|TZfcRqpP3~(d=r?1qn;|J&1dtbHh$NJyv0u5}lQPHU)+P!s}9rSG3#13RNQv zWhEb0E;xLW{7(61a&{m$keUy;auQhItsn^P5NJg(sVd0e_rPHj5bfU9l6RkE_w4bx z)#qIMBc;A-BWp=)MI)f?jgTtqS1#e}f6wmOj_3!19am%^QPWkqN zcrnqUJcT+%Rn{ zk~F*UJRI3OcZXY5i6e#lI4U5xJy<0o0MM|v!>u|>>R0}L{Vq&ESXP*04g>wFAcm8F zMJ#ZLHkSMZO!dRrtE5^)XW{!P zhr8uujOJ<5-~-O9#GO4#uoL{Afv10)_n|U>+}7RTntatf83T5Pn0up#fbb*f4(h6? zdE*r%atTLrjtYtmrE?10q+ry*yTzHb9f(ksHwfFT`w>)(;HEDq=ss3S;EBLK4Ac(W z4>ya~y?ZCZFeCl1A1)aTgFSfk=%TIq6}NK(FEQvrRYrma?`rT+8(uA1-4-kgonxPA ztr`ZRcYkz1nKDNHCcOztY#DUw1-X6#dkuMbA6#)UF>Gutr7`I;+MfNS?| zC1M=&U=fom&6?~Lqbj46mBZHTZ%BF*jw<;JcSFztVIotR{^C-b0`sx&#Kzob9>_b@ zEo~5{m6k+$DkK`P6sB_~PRzVvs)b56B-$u6b5GDYhN*-r3k|Vil98?go!+gdGMpxj zB%e^RXx=4WCyUBq2m?1+_+exb(i|uFRU{g#lD-)e8p0oa?%v%>-AQBI$xgb&Xf6AA z`)Si&SUf^dRz=Id_oClFGDj=UrR5rjN$txM=*@hPyFw9%va{>{*+y*Y@j`2X@QcrfW9VA$QP3j-y1sDncZS}c%$ z?$2!&bd!v60>U5;9j6l~g3sG!^S)Q+z@fufaZc)`uOB{PC(4%!o7RS*v{aXU{De_A zK#`rO!bs_fe%=8A(P*|#(91|~3=aiM)4Y=>l(rGkGM(j`SWjJ=D!`Zu{D{O=xSIV~ znBCA}&g!v3QS&VAMuI^*pQ!Q&<1e}+E_HyD7Qy5;Wf z1zZ5eHkcq;cyrNhTg42S)g;LivW@_dD_1D{%<45kUuAVxM0e4@p-Z_N7#J8kfhNH8 z*3SgkzkJQ6TNJ2M%h@Ewbw=P_?PMJPxgJIDd$d!6La)vh=pX<5UKVA7LYXrD^7bYp z9*|6ye$11#r^Ec`+hVtYUvoS%=iSA~SJTe}LLTpLzBfns{CM$yszEE~-~E3sQUCty zUoRuWa(VP#HhWgGH-2fILhDNos&dvq!P2@@=8v;7r_G8Ek*odjb9n(izHzbP#?vyk z2Z^VI-a=3M=a(46ye6&iufO4c{?l0|1Zy~6;}k2^{5H!bqHFfBDVJ}{zuu9!q-7ZCxDd3@Y|zr~aAix_mIhX4yWi2i(h6g~(SSGMiq;(D)wlNH*`iDvr! z=gO16{+pHgH>C4FlZt+P+kaa2fBW#cb7HpUvh+o(s2@8OTob;oITjPiOfV z1Qg+|!M_~1ci~&y=ARqw7W@Cxd-}H+{U1CLMKah$q$a}sVrYT%r9>y%w9lwTw4tLk zG!e5=EEqdOTW3_Cp7P_o2?s9Ae_|4U%6P27Tn6AbDd89#I|Z5S;|sK<$^g+7Fo5;Di37lfBndWthrZw7xfHMffi|DIKpA)BlVw*ZeU7j zm)i3K&{IO|>|9V$ez056_sd{Lecwq_Fy zpi%I#oj|iVSBD~1`TaXEHC~KUNaTj}&Yw2`WCAaRf|Y!sC~(*9oNZKMJu3rWgxOCs zd&oi<1K#Gz`!l#H6wbebk{6mM!;aJSlmp$`IhhN0IiOJNGc0+lk_p8L-Uby6+LYiV z1v3H$(Sm>HmkC=87o@5Bf>#n?MTSP^MU~>tQFg&8@EsUKRE6_}P+pE7({N+}RT(%M z1{==OVMwo$a8=&AeY=z>LpbEc=k4;#Z_p+N#7+RXSkpNE^^7TJP5~7PnbvYqamY#l zY($9wqth_^k$xHEP}FAAmk04oUx=BZ!{-Qxlw=dC0iG5!hh#DhY+#4?x1W25!gW*J zdv5vB1{k4NB7B(#n^X)j8tvP65{D2d;kL`bjloMP4-7=ZM`@Py{k3oj2bGvzaGgXJ zu99;B%S;?6?XU@-08}$!fa{T=tDwhp>uHk{1Ia4;Ws$hv|GnhCa;;gr2^gFKYYee_ z+;VqM)Yt*ugkW=k$bImrfpf0A-0W`O9hYkbhdP?1tXI=IFU=3qhv9haZ2$C>nDmKD z*$cx|gv&-CghHW+1j9zgN91+T*U$-=Cs!paBV)NU5M1HsT~c5N3D8j4e1+%NMT`9M zF+aOS;b%IaC|YV9I(+eBl(t>#I>PSd=jYQoAnbGjO;(V@ScubV?_Om@CRuOznFx%* zb0un0fe$}9ndKxrv^X>B;KIerxm3h)Ungg0ruRYn1k+T%<9uviQB$4WY^ z@@Isg(KH0XPR2BdXa&!VR6|@DG$Fc5(#6x|F@-m5nS!p4S)0UDU8>!y+TWOC*Z5lbNyv^vbdMnt7!N9t{G1(_1A02PwM z@0ghZ5+VBJ5Zqw3$s`2y>bP@mi8!X}Uq9w6lJ6AnhI;4&?Is2!8wN?VGuzTyy*19q zsXlZCs&&A1k6iAg!31yy}7eNwBgw+awq~ zeix)qLXQi=6GqPB=}oWHyyT9>dj?Ug18$WGByDms5ai(qoVTE!VV}5DC9E-CZC*fJ zSt4z)AJ34SR|EI+VGw)qexqeH#SV;wJkRv>^sFQj3ehz&&&t=41w$Db{U9tWET%kI zDwG=~JZ>W5XttRkxq@_wBnFTrt5{VqW&KdtECvE`D#v_^;!MH0-&f#8(ZtjS3VnQh zy5ZFyP;iJthI4w-;0AH7kyGXf6NrIs!eH`7R|?~${03^(j&YZOvU8P zJ&-BKVFw*l@jY>dZ!%YLiqyc=tjMX6Cj5Kv119-{OxVNrpi&BVkMf?Zov=82||1Eg9N?az5N zl)~18E5;jS=#sE*5Dgl~;#CS1iL+CYuKmF{m_!?Ia0NflDJ3OU%D!{nR~{H>!<*qC z3Wv!D6_y51HPgn8;fOhHdeqGoB1Ftb(I+jM5__ z6>5DswuOJ{9Q8zWS6b7&NTjf>Y0N9cg+(IeaR~{X!iYJxAwB2pp7PCt26t0!dcYT} z+wGlCr!=yQKPZ*V9ZWyu^0$fV03L`l{izhcCC2f7^5J>U&kf7npSL3j zMoldwffAsba^MI>>lLq7W*0RpKALxvq2qQDPjPG>wFJ8M@$aPZkB@rqmAhus9Iphr z;9h9zJVZ!2`T)N}M&q5+aQ}uw0_TtDynm=3ba8oUI1O_G`sHYGXB z!3yF9u5jMVZuwFaZARZL;?>(<}Tow#& z0_j=_+K+>}apOkN?y7-9nn>Xp2j|^4C8!wz3NFtWNrE0iWGNi-hJaBvH#d(NA_J5Q zKh1x=&KJBM#G5+33`jymqWeS! zVT+e@&bwx=Cs;2W{JD%;0z>$J@w?rCrdaYWkl-pPrlG0x>B^&bu@d0ipAs$#D8AHP zxIPSZGH>^CMeBK4gL3OGMbTRD%qt4HcFe62z)8FpptaRe$x{lG(X=F*HPjzIF+U6F zG~&Ktp~`vpuKEE|%h&O5H<1xj1bFA&LgQ5uayYP}17{fKk2!*KvpM{exIwe;W5*jD z3}bI#=XbB-_ct?H{C5Y!HhW)lg^*+`*p*~6Z#k}*S;!vBCJxKwdV5&_OFlz0D94@k z8EHZnKR41x=Uvrop!Smi#Fu)zHWs31m}QK3lSA^z!6;<7 zESwn>Hh6eIk%O>7GHggDAaWE~zu2j%TDrr$heL6^tC8?)XL)2O=2ztiV+&{+icG!} zgKMG^sJT!Dh4P4Jt(4fTAS93k20$W}_men+k$Jc2#^5Fy#5wBW17d5#z$9> z{GK9#pA1tvh&xoq1dIxXYZR}|&k~Nl(rY&HF{92vpSMrdn0LoY*ZO&XR!5*2LUYXf zk^GOj@c;3z5xvjP<*o$#ZXYpUby2#8ux-kkosPw>9R{@M!tVYrk{M`g;$Nm- zk=813;?%3K`RVM{TR;|^)mKRtgLuKPj#x!M0@>gip%FhGda$qIO7kAW70)5HLx-Wg zz+}q>d}<5~kHbvDa3~+(A~$g+L_q$4zV#FaRS5P4G~M};z(!C<0d}=^kkhCi&fAMd zi$_knB-89*6ryIPCnLcu+mjx0>YIhxwfT#Z{s9G01kLUU^iueAH+f+8POppXV&4As)>f#nYNz0 z^FPGjZdb4_x;%T7Rhx!hQBN9tVOcDU*j|l^T_~N&k$BI&3eK^N1hzUhyD5S*0 z(wi}+O%e<;bfM~I;rh!@aq!AcfWH)_xG1%_L)gdhnML{emTJO459CY~6i}1=%Ba&% zv5f6EcLWyW@QtPG=2w>It#PY25pNFyN)svoQ5OdT(0}r04AWsRj9l>1pB-!hnl>5= zYdMmEcoz_e<3_C=<~Pz0 zH~>ic7ZFb=Vt|O$h8PmGtC+=ky^rHTYFExj_iNPqU4m}2jC*%BRMoO4ZnFO zYHI}%#EHqhV;o~x{8VE2&lUwUb(aG+6S49O5U{$v+0e6qqOum9-=EzQ@N_RW$KIrp zW21(mG%&P4k{l7=allCM1elcytQi-TShZT)+NQRJ;)tYX3l5Yh#(YwMOPqWXaZ2V7 zS*#TyfeuZLFn>sVHh)b9VDloN7(~niZBbB}E|Rwt6M&X#J9h1or>mq|8HQHh`LEQ@ zZTMg%kRQ|%%!WUT2%=&8K6&NDDrO@`!jRNYRc$-Yh9MuAndwZ-%x%vLA*FFlHw@0- z&TmMDMyJ9_1pp~X=tU7A`8b z(w?~@TM2wd2iTX7!!R=Wk7EPxpc3;pP6LJgfcAlEIbNXEth*wpSjfeIa4p6-c9|_D^qVLxZW@|nZ zr^fV0fHz~1EKTOAU01R3Q8(>3e+>W2FuENBq#b$)Im*3yezP$bKU=Bb z28pAJq#BevKZf*?BkfRAVQfmX!foQ%6}Nh)F)i3XPeEqhwDGX-+k7SutRrkV=+=EW(s0$$ zc|FE~yszcg7MsCTSs>syFK{TA{||fb0T$)Cbq$Xx8Vh=2ED>AOQB(w!ricP+PzMAQ z1f*LiBSjPx1rbOLn4p3VO*$$~ibz$Y8I975bVdaWML|RXDbl{R2gZ=(oUdH(_m}Jc zKd$SX^S(07Jo7yFz4zK{t-bG%nwf&I5*&vw2@t1QR}5q+-lcQ*-akK+ac?e4rzNob zf%>k(_WFn@fRz^yUGzXk{{~!V!lE4twdeCxk_Z*rLFvVa8nfx#s>N3 zHtoY+zj40{`?VAM4}2W+25j6RtvKl7^Uth3RA{2nXE_U|-+%Wzz-OASg6rFfBI4g5 zyJ*9#yF~pAVG0iz#d9$1Eq$le5@h$)1m!IKkc2W6*$rw;678*GiC zu2f!}yEYyiXWzq@1`sE*eU&o2w`fYe*+sMSgZ~NPsQDD;_ubho78p`D|JwM+n_YxA zf~CKEFR%G$?83r+?dq}{caaQ+c4zH30@^gOCk5RvsOMIr61e@$E)ay&yZCVmJi}$i zhrVwFaLZbsW!y$H%VKnt00cAc=k`({Ode?N7? zh);X)H=gR-=QlivWQTNu0HO5>RuwdKuVV4IAA!k}_h7A4vp7jf5K+Lcjk_=RCY=Ht z0~}J~(6mtt@oh(d7FM(9K0Os2$M-PaZ^mlc0H&3B10h7Zsb@R;QaIU=)q#6YfBBaM zor_0;7{&?5HEq-!;c)`!D!D7n9QV=BuT$X z@wnH&Ur5UtJikfeKFO2s^l3uIfh^d1NF(G(f(AY389wgyjE8>U25w?ZGU1>xX%3ZZ zIMEo0avbWiW@FkM&BP}sDmGplPC@A2durUf+&lNxsib7gA6ki#^Vrgr5J+eYWn0RU zZEpKUj9{%7(Z91BN>8uxU;b_fZtO%BETZ0X*6a&Id835j96vZ{nK5I=j2i5x1bD%z zS0ZHE_&TZYU!ERimtG+#R>S-fPfcsMePBAZ&<8@J>0KBJf2}fq)lQy7GZ)d<<{&YC zi-DHt$@>j3xs7lul(VPmPvL{6 z>GmJbfYeFFK*NR?jvy_0>eh+3wR;fyk81xq?yv6cNEf-f1mYk5;WgxtMSjfGY?GzN#=L?C&SL_$p|V8t#o z0U(tz;-CgSIo-8~JXsu|b=+S;i?K#u-ZkvD0X=;HF${G7(5E*bZWHyQ(9A`BjMVUo zE*mwm?7nMJ|IBd+shn_7?*WmR8uBEQrs9&X8W0J}pz@;M#xK9V=~#Z~-`&YUP%{;w zQISkkp!tzUkXKN$UPB)i{0C)!3x-76gYZ@6pfMQ4d7+;l6N4`P*ej29!#R!tShsV> zVFrfdd4vckx2sqrg{8*9+{OuqB2inp*`YfcrZzE_`-KbLyMN)|e*Et}ZQOvXkg7Q0 zC?)q5lIV-HN9qWX7_hfftLQVkCsqZR?*SM@WVo8^IDi8m6h0P;&oxl*ZxCs#F%D-r z3sJZlgcOs7tP)9ai&RFmae6aJy2Ze*fAd{kTo~lhBmV+)*wjIgq?KS-Vt}gm6%vNW zdu=D{<=-9$^PX2QZ+0ggJF)K2qt_6UHL#=6iAu(EHMj9~lj8*xcQ&uk^8%?^rvbpd zfdUftHGJ}|5A(*|9G58jx0k=`mpcA>H{m<9=k?mH9QXTs4;-bhj!)m-`hRl>{V&>& zCe%;Ar$Okh@BIN75O!hLkROoY^sZ#4b?4@s%8X>=}*YdNtFg!IjY6cPljboQ%$IG;K@sVnI5*`uU;?r zDTo!8^_e5o8Q4*eOYD2H!=U7hH6pqOs%^=M?{@uJl0P$s5P2Q0&LRrTAAx z*C#UtscU=)9LqcN`omb-yD@R^c!kRPRhu?F)y3Hc5g?(o$#{cnqH5wb{v%mDg3&Fl z$j#K3MzSeVxDp4)(VT$Yg3d-lMf4cPl#WD`ao{)Xj&P4>&>iqN?Y8-0w~ASZES2{BLfRK_1dvGQse1_jUD=H3^Vs}Dw9l<5$`;PJ@8eP|YF_qqT#>M0HgC zE)y>WbL9hhJREMLvwzX7^uufAZI-Tw5R39Hg!zWr%@2QSa2^rKcS_pkfp#45k1Ex7 zKYy^<$;{j8uvGSkkj%C^_P8`Gxg+rpBM-Y74Ier1)uc~tPeyu*PJ%7JchaYpCkBuw zmPRR%q=%#)(8^pfJaBF==<^45Hsr=u{tGb?^BG-aE*Dv*{O0SO?C8^<%+6KBgsxSS z-J7bGJ%G!3cpgTi_r7ppQ@?NwV1M@TCkuo%jo`RA=aC-@?cMw~qb;-f7h}@^31$9b z!(S@=?w$9oAUdmW!1+aOO4O`OG5a=6!Dg<)gu(-QNP@f$gs4v*f9E3H&pz<6?zBUA z4EuG}v1O7M1f`V74+hqk*Od-m`}8(E1cYz!3g=GPL{i*J{fWQic3=43X+pmPuE;z3 z7dNahU2LSi086ChqerhmHpVd5-KDzqv;38Gw#Ir@BYH)*I(Q~d6{=M<`WS~y97RxM>qP9WPaEVG)|+w z1vfjS6sJHu*o=N%PB>24i2*ew>N%gk46vLzF}3o)@Y>J;#FmrEq6!nZsaZ}V3m}`a zNn{eu#l^$Fjh4%+*bQWP;!7mTGh?(-OT_=7w~ z|M9=-bN-v*k!Q;N{Bcv3VSuyI$Fo2G{B+jjQByAecM6w#K+d(q4g&)|A_*xrT>ejT%^6v@65OqE?cj6Bjv&xzZ6XwQ5L7t zdN670?b!=nCs>a>NiJ(=FgnNUNqxWIr4bn`F?(S5u@96oX5>HUm!TrLI^1HTpaKLvTSyCPA|^;*d6^b`HOuO#b(MHMp4q_4I`@j~D|-%2=vYE;eoAqRlhh9^ zU%{*{%}^uW^HIoua8c-!?88Sm)VkJ+$%Lhb1VZlg;`|IbJ2Qo1^C*;I?hWY0mlo4a3Lc4b_e%V99?kYj)B zCsgy&jya2NI%;IM&i~qhxfK8WlxBR6{wO3K>^t^mK5U+s_BbyfPxg)d*9Y}~+vCj0 z?TZ~-X(OIlvNI}dGY4`Z;WGOuh9StG2kL&H7@WbG5v?VBpgcVG*6_ zDZ&v0Mq}Tby@OkRQ+_K$!S-M5(Hr41bAL@}xnxAcqIiQZQuMyy-M`9Mks-Gq)+9u3 zT9#?~-0ysV(OcUGft?qQhBfsKeCw;GH)Vo%=c~10@m*E)Fcq=}-F|4_qY?6Li1+OE zXO1PdOj7zSXYpayT!Z6}#dCo~GMV|9ZIYc-sI!DV{Jqf_YqxU3$M)BI3=Qj_f3Kh? zREC^CC9#UWTRThdlYj$bY9Q}A@?Nwv$uoD!df zHWedp4ZAlo*dk767csOJ{e5V3hev zE#Ivh>eNxBR-W?ywa{d*FJcA9|5e0}yzuT*DoScspL%*Q*iQm%9;935Uo-KDmpN_7 zF%AtKIpd!gIe4Zu()LN2x5D8w9SQe2%UPUOhvBTZ4HhZd+0K^ogWA&>g3lNKn6e|{ z&g|61FBu|=ymw`|_{6VTA9SwhPVDcA6DoWc(qmGrQ5;Aej&q(R)jXtlx@jP{XIt0= z7tIU1#oXi}yr{l=ZLxbAiy4YP^HM!JxScUsz!FLQ@uf?-`7bQ$? zSJ5~*^x+9sP|8XAs+r{s$!L+Gqw$AxRG!#5c~wimx)Ihbd7@@<&P6>{o9a3uAVMO<-uPn`Cs# zD8}7!SMI%~35mV-y|bipY)#HR?Uv7&Yfp*$c}1U<}S_RmuuFDm#xK` zRxB2qWlwFaVJ7s7fI`K&1}Q~}fa+o_8;sjYcx$si9rsc}cK6C&Z8^MS*#j-5Vx=wtggl2*^M&Tq`w@mxcB?(K8HXSDi&ib4t{xG z{k}Y|#Lw9o(-xJ!67+q5&i~-h;qEA%Bx6>4A|`IF1N<+w1 z-&_MhUz&2^^x?cwBM8RQZ8J7|gZ&hN&(qvz;{=y#P0i#DQ_pAH8)`l6e)zB<$htJ3 zCxFQ^IaU#P{9;e+tOxS#n~lr8YmE0_u6ZB8DEpVJe@a}Wo8yqd42Ht0Z`tfWpq^e~ z)*X%tE=RYm$s_rJx>8fPI7)D9d@v||?G7|H7l?v&)S8F=AGInW-`nXlcFGTZ8s6&D z{Fz7O_Bi~~$es9)H-=r@n^|X>Tr3!rTxS~Cgf7qmj1JM((!p1akKFz4NuDp$SUo^# z9lN~VI%I18mckDwp@&5Lk)t|)9i~(=^SKrY&z=WiGDW3-Va#>x&1%sD^Hgr!m^HAo zTirb?#Yb*tmiIw1s|e2@#Ug5z_1-KyESn{n8kttEysGDNO0Z#OPC)sN!ruD&1|yYq zS$WSGT5jr&Dx997;nff2IFVH!iyJmyiD`Yg^@rHG_7~5JWhUg!X0>=M?9AAi^dLN5>yd=qlM2Wo5sj?pncuFYWBamZD(U^{jI6 z9Zefs=-FnjRVpYqSkJw0@kGnvKfK}CWtH3g4_@%M>uz-&(WvKs?VW_$NQi|&PTc1~ z5r@#+Kdf|vucvaq#0IiVZ}MevaExMH{RaqgD`9ntmEa37`;L7faIqHFDKE!Nk$fDl zonF)j#x4+ZWg{VH*P+*5+io;|G8kp1lO-h&BBa%UXy!0?EI9Tf&N%zXO+C4u!%*M<(gHKPawpGCy(mB!~q z((*i=2VYf-@;gzv-bq`p?q;_yReA>AF`&b-Mb;iOev`W^G;#A2&a_+fJCAgXL|~p@ zy#A8W4`ZZ=d?BEC3S;KPX}KW&q8H%}N^}dquf|?s#Zll_7`tr>^fJc&nc<=~a}26c z3)#5Wd?h&c5*N<0u`~fjU;zOc-cMK~65H6%@*T*dqdzNUMll}#KZX2n|Ib@9{&GWZ zu-gfLxOftCfi}E+ocLCb;R@BFQO13lyQg(veY&JwkK4U+mTty3cqLmDP6tEu%4HUP zATIzq@pR0BPv~BaX-b9@ss-FtF>K61Fk0M7EnGFvs=uof!_iGKEx2y~SpkaWcyAT< z+o6|Th$pwq6SvVauqu!r{@}U5CU1#=2Ri>{F@Su3$)>I%_(AqB*P2+!g#|A8=!^SL zaS(EUt9A~T+mvE2*MPr_ox@uJ+_+-tCu@iw=HO^ic{r%^v6=iRuR0I48A;f^twk4i zrcZ2Rg=ONyIrgBqF%~4WTGUn3r58C);R#R*{wMGRQk)swQK`;`ca9z)EFW~DTE`xz zb=mg4IM-q{@=3dGcAiZ-tPXk^Uui9=CzF)bIo?8T-wp!A22C`kFf7m8>~K89zS(~h zj+4Yu1AmCy&K1%120dtk(5Z9iqEu3~u7%X8dVq}<1ZOp9QRXKd_hp_s22RUY#e zjb%Du_5TO+gI40-wXZmz3MoqyVTR)zMN4!@=$$J>9@kb_q^4FAQj$_pJC zY}v}$%DE2)r!CkPkFKVqzan&aubCyt8rKKYW#h`gcX4G{Q|Okwa2LKA-Rqb$@T_AB z{>3U~&jE}Zas5thQ(_qw12?syzvQHEY!9gaIhue`o$4!bS2AJ45HQ3cUJqU2!-af)KAUu6@&;eOI*VVbA5dF z6z9GNi(5B$2-Uz@=-|KToWOjoJNTqFgAw?4Yf1vw9ol~O<4H_kxDbQ7tVD+q4=bSYiEg9+ArnuScMRpWC!=BO7HJ9fNeV5OMzr^<5gh|H3gZ(THxN7*1 ztPsP3nya{3|FB@*(S-w(#jBHfydk8juunag8L$$ z$-<=5iX7(=P?tg&R#Qans#vG%(M5bfRi??CD}^m-3l+n6@^>p5p{^7FmK$`WcVu*A zXJYyi`bY9E?^vW`_7(s#&+v4+4}c<*g@NR~(-v*oKpr;eufBxBG3=`lN<`bH`d9M7 zXdq6zA3Y}8(aLOXb)A_6uT1c@_vc2$Zo5Q_S z&5sTxIDL#IfXw3HkHeFa#32-#Q;LP0BnB@@M^|u%V^m%1LP_@SG_HB<`eaVp>Kztuf-3^}wJ*0Vgc`0nZW6xmbn-Pm7t>1aZzvZSY}LZ#Z4cl8HIQwwb{^2XipY((-3y*<#}d{H%KYGG z4WomN(Av{SiF-Dt*X`b^PBx;pAnP~N1z1zirnOT4)C*TvTD%6Q56ManzTP3fS|1q0 z6``m~1`U|q_YsPAj%3%-nKpHW5ME{L#ic=(~ zpn-1nUTy6fkD-Bv5z%S$rQ_2fv#Y#fxGlo{&2(bw#p8`Bk% zS;l*^oO@t#W!=IZoy(a8JzBF5zO?xUIS=RP%`(K0R_+G`*V);955@4oaCWt23QcjH z{p9Ka@)=8QNXzut)R~7^ffP4|hrMvZ=l%P*-&%biJe~Dh`rLaAS4s|TPS5PP7}>44 zQgZl2%BIkc*+T)q5@G$`w|DKLT&Yw6gNO|p8J4p!FeV9?Bg87fD8iz^^t7Qw*l zkFyn1>{Kv@YIq=LM3%s|7ZaiN$k~<=h^q}KtyjIl%1f>uLXcQd+=j| zvkFHaEgLeJ?+9S5Ba7e>*dpfKuZ#=rgw$sWVeVsfHbx{h(b~WnZ2FF`3J)I^!F9)N zPQwZ?!;mAET>asfFqE*5o4Bu^2-@q&gvZMaF)kTzS--je7huc z?SZY>;HeHRYXml@gu`WE3V8%6*UWrC4{Ozn30J;la60~=WXq(5-pthXy(7dJYZw>= z1Fg55HyS(`oQPd7%1Xk?@hw=0vA0PgtGTK7srF~Jjku8&=qz2jaZUZY+yxp!mWgPN zdSJeI9j5wwPpnz&@cQbThMW=c+zbrN`A&7i5|#}@Z}{$&k{S<-%?>+b@_>z3_4*wh z4>SLH4>TWhNdZ+*t^-n;zd#mafNiliSFq5l`;h8;Ryn5G4Pu|Da=7#FjP#F?c@B%O z69J!+@rjajdSp1dEY8*viD%yWBl=lVaalcF9o!YaSijcRR(-tJJSp?Nv_U3Xi(&>< zM&Sa?iG5{*jN6PQM0wf5dX~`0$p<}HBC1$7!6z_)@Hof#`bBB;ZrdbDMtRxA77VSn zNq5lymIo~k!T(Bs*acUp^oEwM&=JS8hd^Cy<9MF5KZW@3Xa7v^2{4->e*5X-WccH~D0p0#VI+aOAu0AJoQvyuJOd*QscyhP<;!1A z-|z#a-u6stZ|-^JX1BC290Gx}{~>63`NQEs)4dm;32+=x)$RU)-h*wMZAX_@x=fN(kJvR_YM`6rvxK*!#%6 zCv|DSCn>;Ug)Uw#6jpo}QO2*W*%r;B=JNGuny#H(%4OwoeomR+y|v=+&<#U2f-N8ooNo-Cg2vw{KjZHov@9baKSa#K3n|RaL2rPqtPbHu2W*hSi!@mdu-Q zuTI*TNbW}|eHVpnmc>BDGibc~6m;t4Z~f0YoV1a)y6YEo9N`a=b9$T_idrviZOta z%eG(0qk5z9)%@EZsS+{2{VC`9&r`L+&X%voEVhZcs8#d_j-gM+LVmdPhaz;Zw;AYZ zXn2ulci!s3<4|IYLDq)=Xl!=fx?}`X?QQ9~eoMP}yh+;^Gp)?8qS;3;L;Lsu+(c64V zO%!xLB*9 zDzh9&Y_PIx%}{(NhH%Ibpz&hj_8$jTBpB2HMsQciCK>eWnftEMDSw_G8yl<5RsBlj z1(sp-Q2W3LjZJPwzlbVl!LdKSJ@H7(=*-7y3pT7}nYRo^p;^BLQO@alje+3ID{^NX zGzwJ9`v#6-n7Y5bvlBMU)EjCV4%Rn&jSp$1Bz&4R&@b9Ex{R85LwGKEekp(gM|%c*&|M?gZ;2R)21g-(ovIOB%0pH zS%86NJT9aOVX=g=Hr1q{rwbt9Q?WLPKv{$eIXGvi-C88+<4D2oJ$o*rhLA*a>#6Kc zpM-FP8uCGl-g*8rB9k#TMSsW0aED3tLcF;QVuyr7_vQ%AMj#iJvu)1kN*C(4kAPnz7B;jgwi@}BEQ%kTCHaVr1OCK{ZX390}$JxI#{6^ zbcHx&lqRY)0S>cBqUw0JUAady^q_phL#JY__OeVK74ySlYei=?em2WC^s5f{<;t7F z0zgG?maTkmdTD$mclyTHVUBHj4F!Os!MNTZC<+IEfSUrLHZ7kpv)bmG#N0)C19~k; zOxB2-HLnM)ow@?)Tp|y+Z!XMVFG^VLq#;TLO2tiYqq_;3_CFMJx_n;r?n?UZ!HL9h z!cnV=eg{9D-s_8pR>HcJk09rZ?SKyU(<>>Xh3 z0`M48dQPGGuvpyo{xI!TiZ6%~?6a@v4XTlQ9C8E8Pupe->BS^`l5FZScsm*gcx;o4 zG*MtwNOPEY=Bv#y>%qX5Vll1lVT4H@(n0Ouh4LrDd1RQ-f6r6fQf`-6YIA^vX>5F2 z0nR(titZ6ft6fS(Kzmx>$Q!D0fG@o;uYDDI7gWfva_ar_x8MR$`kb+W{$NIb3S=Uf z^$6>i0{WFh9y&Mg@}2;2S_zoGOE|rGCCo{M$cg1R8xPdV53=nk8a*(AZqys`^YN*2 z3c=z^%nbe6L8{wcr{03*-r*SDi}k-QT6geAv=>LfyR({NJ>YLJjt3ONVS#!cGR;>e zJ`f~uf=X5#Z$Xuizl^OZPM`9pMBI-+7EF%_9n{6RtKL;wdRG1)0Vb``P2@+l7-dLw zl_gX0&?0S2I1XAbMjZDVqmuVWP?#fTNQ}GjxrKIl=)w*_kuH6@L*rrCN=lb2d8|Ug{3{WG6A)P!lN6s~29^SNiH1d?W(RpY78;%_$-g)rd z$(%wo2bP0Opz0SwlQ)xhK@re5DN)>_Vie#^JNOXAfC`WkS~1@!1@hjrJh5Pt%|R@X zbbwLY!4J7+=szVuhA3+T-d42c6HI|kz|2J$Kux^23DSO{s%*PA2~i?<$#kPzffeW* zpT=DcpkgpxrtuP(q7``dYzsflnRurp+xYaxQ=zgM939qXe6EGPn$pvrA%22wKu+{%y z@$eW5d71u#O3KenU zvXYMFSk#Eb368DP1&9s6n9V?_4EoqtGT6dx7toHH-(PI)nAsiIF({0y>Oc^y@9@`5{Jr`;xxYZiGA0-S~M&IZ45%J zq(_1lk{ishSJgO~NV3IgX-jZ~NkVTQdLKEMRcTD)bZ>TzetX8fn0F}pccNkz$FNV6 zN8G{b0!U;m4tN{s4N+k9SR9N3l6=BFID3jKGDTdO%-Emq*F=Mf7dGjw`b6oRZ4b)t zZ8>q`#93S4iE|Ou%PlSDi#rYWIe3aaOTT*c*w`PjEd(73wvLvVy?k1sMms0&Y~@*j7l^vpY37Kfq@Os$5djvy}WF(tn3AMZp#4GWSe zNas%#p}zA7xT&w4CC)AjvqV4i_&X`RnEkpT$JqmUnwgK{^a@OPD~QZN6=I80v+;J$ zkQJwqgjH;q#_ivKapz z4AXsPed25#y$h%Vg;8C&Ae!VdKoJi)VIAwT)i#zOjaqT&SoW04sD&<`XP zDnIl_?c&6lTE-${cehM3wrIM5Rq#T-gu%Av4owIB&HK3*qxgT+3da90>5thcbjVb(&NBY0DOqze>;UTeboe)j+crZ-Dtf?`vbqSJ zEwrm$Lg|-h8NgXMb?*E%&ppSB{77n(l5hK`u5;()LCpom3 zt{vq`IARh3N5VJo+BRV>6_pg1Q53936P_NOD@>o9jy+BxGi`TZ=sUb4X`TKUav}`U zF%LN14<|o}VjSQ=Wh%zRBNg8?%usCqn5EX`@O4JI3hYh zPb$PkB%I)v5MoI`J6>UV=Y`veN2s3CnTsb0wvq{<)UjEDHEbURiglDn$-P*r}9xPQ(tRzUXQ5XhE zUjnlm0>!{Iq@x_{!kWs`akq-WN*2b@h##p?pnkn{Fvb?S@cR2>+Qt$zQRtaI1Fgnk zVpp)=KUL24Eg6y=#VxS0OBm-}YTC&6x2BvB?#}y-=xbDLg#b;Rc#>v75)`wjU~ud> z-JpDRcizdIfoFg$lG<4+RJ^p1P)2+Xc$`f@_xq*;OOsYZg)q9{L(u6bntXHjN+yfw zhE2WY1`CO15PvvkAW)F4d_7$NGq#SPAb#?^50#uZ6YR5X5?u9Fi#o7W5fd635wZ$F zyM&S!*`}$r@%_`Y!_Ty{Jk&~}R zyb!w8%?vGgx)M|7b=cX~U*>VbK|=8^7%;0B|5AWzU%Gw9P(sOuV%vCm z2NYqM)K{cyB!hu&L0~JUn1_%gMbtm3bEp!;S)Hf?ZvDP+ffmi#TQO#iL8~oO$1dtO zF}TNA94fMb?Db8b(K1K#!}_MCraBamfw9J;TZzH$Eg$6cl ztYBfk(!5L6anN6pydMRASopjBq-N{tP2Er zs1DPFL}8^P!3B7`RIuHopICa~ey&ITK-?fXaxk+%f;fKZr$-;~+BCj3N@fc{jR4s~4#{Mz+Ye7yU@>MDPc)qAu#4($snDUvI>c&;b-0To@y?c=T zZC+nJ*NleTnGG;5zZmM7(UwodhX~O4d5k5btas_3&tE5+|vk zA!QME;0D#nlu>B?Qfy6s=`P%fC~Nv~wbd5TwDb&d6vS^u0a85VfKIt&q!Batm-;F{ z;+!j0bgxunhTbX=7_eXbh;^E991V>+t#Hj!fY#(`P?DPGzY^paquaF~Sw4g^5vrCF zmQb)aU3{^fSI*{9*U;xbKnWo1+wv7ob;_e0j3PHGCT2A$B1nH6NPm)4)n-G|Y5@zA z54DwtPyF<4SlQhXVp7(8&zC~RhZ{e+LARF0%#TyRgIPtj6DBRVgm^`je+gPWgMb&- zkTy;{5$v4}c?56b5+4bQsPm^)l-c>Lm)?My?k(QRAASuIn8~z~{{H?{QW&Dp+5=)( z&j$au5XDd%*l3#YluGE+h=|7IqC${#S?YpgA%?;Az}BZrhW2JlL@B9sLo;lH4@|}Xo>JMET9Sxe75Oo{=H()5EPKaIc&15PLN0<5f_@UUdrEo9r}EV38`LEyIs&Cm(Px0x)fgs#YB5NMgi0zSCz zqmCvL-cOm_UPWVByI01QDUx!s08+p_Oxyv* zduJ=ZoMmvcCw?hFo(wz^TYO&<75kxu2VpLdNxIq~-D9vorNk4HUJ5OYhiw4{6LhBV z^FH@x($-~9#K`9))8FVtta^p3yX~79E6nzMbZ_tE&Ro6Gx9N`2^uwQHYv#2? zS1JZTK+%eoqdX<#z@I)v#h7wUGj}%KtStqoxKL^cJbCdx+9yvW`w0o z=2<5tuSxP=fhhb$`+w%M_lCV7h(wl$<;t6)_#+_LGKyadf{xF2{1|IgjFT~W3=N)u z`3-(&tczV-w88@d=h!-#fIa>BYEB%$i}eK4-vP^I@Bu!fI1(>K(GUuM2Ngvov|Xt( zbsl)qz>{|o!<@h~hnfJRh-eD)Z>%bA>`o2}BMIUkDU|?9Hs=ht<R|9cSqA2nax!pQRq<|i)I(^godoZRfD?Y|&@WL*}vP-;Z zAYeLfhAq;#$>VE^sk!7Iq)Hb`yfTSJjo4TD0Ce?1e!Qy=6A(P{2;)-x>R(+Hmi29k zww-v}4rH|zcgiF|vXh|V5T>n7l@Z-Pp22`rD;bCYU0}bOh(#kUry25P(SdUdeNp3> zxt`~3^)Rd08-^UkD%Wb!>x&}hCIJ^%1T0`*kvyyp71@U}Jb+x>=3yjKIm9VTDk^}t zZuInNY8&`a;v)RmMGy^vs}vQ2R-D7wvkFsLy}Iny{@;H_-6KL(A2~L;ufS95;&BZN z)B7P))Ia}3@)2TNDeZ1D-Ti6r&2pb2Ng8%KN&)kcOo(rdw9RYK&;sJqB)#M@+^0BP zKt*NU=ncbXb1z@}LI8y?)w9!a5;Fx#^0ByQKc~cvd?ZFm>hWjdx92Jr<2K8!!{dOLByzZ zYXn*-o~^CFr={<@r(6I?)=ebbWy6V?#%|~D*Lr;P2H;nE_9h;ofG~w;ej=v=>1lUw zHA`C8XTC)~B|5nRE@f)msY9t#0f!F>&=}@$sQuMCxi^DdrKR|^%YYE{8h)1zIzJ@W zKRd3G-J8{6f{G}gNGVPR@@H85Jl@t3LpJfK~|B#-%38czA82{R9 zNk3k2yle~Ly{_pq=v}tRFDqCSx|+_(4<2Z*B?t*LL@K*{-|sUz;TMQ~Nd5c&+>CaN&qz}N zlVVErB`COkm0U`>`MFy>P;LUr<*x`de*fI__S3kI$NYhv3Kf!fw>=O#10KNQ#zx&D zqW+k52s0ceN%HhG$Us7Ale`dEDR)b*OZhtBaeFx~&}ZkZ2(;u-T?U9~{!=Ue`p!`3 zSWUxPIZSfk(LE$T?TJSS1n&kv)#Q{=hJu4BsiYQqfftj!yk}cbm^gDh7h5H6LS!-Q z_<%u5W;pY>@Wfa@EQ8%9I8E{?b`T)}jxjhpHGsI*1;UTYKgJ(GuNaF%SRLk95~Bz~ zMd3sTo^T0Cf>$s@eilD)|Ci$!VQJS+(kGLS9yr@#w2=^O+I~L1Ez6etp1pzeeXM;Z z38}>OZOnB~JbuO^IxcS1)=o=;tcO(|?eMBmmCU0KDVoC8W_y3KP^HI7`0rRC(<+|# z!-nBSTD0y>2k00^9Ut86j*RU!m!9a~6&?s$#9fA_%Iwbx{5WDC_kH2%tN<*VV^~e$jGETxnxq_v- z;l(c?|3;!0p2l~xA&?D>Lgr78jjhoeE!NlWKTC#e7KgkfWS6$crOK7+Hx>uHuGH+= zR!+%$MrPk(_q$V#z{i4nXd?EUurA1s>vnH3(4`2ND})TXiw1}Qhs|#pSaHj2_rT}K zizO6!F(6};(Wi31awyN#DL_&fCp2q< z6%dd{XlW!fH2gH=jikTknYMLw(rxDRMLOx5$k4!YUd%f92ig`mg3E)#hD zFvwijs_j5Z!BtCTsuxs?l-;glZ>g~7CpVJGS=5uWF*g*12Qwe*Q~h)t1Q9e_Dsi9* z9a1#scU5uy0X~!$C)|>(y~HH_gq}IG4#P#-io`Jca%Kkb;OLLyDnv~HZy^>v@R~({ z>jUCuB|#+cApXZflckOm5YIQ+!TjryRo_!YJdk|Or>X=SL|hU*>mV)J{{Hn`CM&xq zHKEF17CAB#j>R`+YiDs?B<_@5@f2ek?z z>6E94vzh)3lFXV zERnH<`Yy!|6Xx)Zjd<^U`AQ!qJ$Cv= zfU*0NlEjtElFM;4g?AcWLJ5dRs9FD}OE3IQscZCw7zvPDaNa+&d);?PH%7{pW2+h? zK=KqG5R|2^8Uo0+Xv9j}tMd8DXjd*1B1wV=cuah7SIuBOzA-mfbM*baf8JxZOGV`)r&{#TJfDSy zwv{qx^)jY0*eCGUz^9k6Y-gPzymhT+^ecq!Vimb; zsFC2Q1uTEsZ%xKNhgW!RgCn8`TbWFO05o}2=@8_iwlOlcfD7A9A|Z~X9-cMDm0}(* z3!>Od**i4$PVEJslg^$UCb^vpPLw>a7CkFX)hh8|gh(leP$)%TQxnl0m@VF!ns7!0 zq{}YOCY+*5-tR6Lbji6;fc;-FGj;6lq{xCpPKv%;#x-%=did)#?D^OPWI13Fj~8W> z<0MdmEYQv~aubOR&2n)gs9>YS)f2SVaW0Qu=BU>MF zwOfmY`noVN(3l)NOsfs2jQbK; z%5Lb9SwqcFw!xoaY3^SbLfsb&1u-Mjhgu84W0PGTM|B5)S84uyKm$7}AZ6PF!k&fy=>H>N52!1C4tdrJRhws+;$9HOyC;y}=_jFGeplv}>1iF5etRXlotUD3h%F`r+ z5NT)lg67cCO{@KPd*VwM&3y>`h}qn=+rWJh^dZnwcl3ck1t=K;{p5fpyLD#Y%k#v_ zh~rL@dDQ&Ir_s0OW_=DQa#vHnn8S#3z{(~km_HZ@%t{NOu?L|8Pg}s!b~MNn6S!x` zLLZunK(eOoqj_eHQ1R95?Y%;djJ_vU?l#4UX_))>J2T&Tq}N&OJJUWjFf_z)SD;v; zMUGakR(DN)D|d=8)^xq|$PhCh%6H!lTWt`RmVn7D4$AC=z5?ETfZAghi#R~cC-cV5 zn}#?0q3_2f^B>`Ud#tKj6lh*d+{b~al3-}0)b~O2DX9Ch;&YJXMj&61e~+Bvq<)i| z1I?X9S1v;_+dxTP4fc%LqGK~=7;K}}?*4-~L!#MvxWyTCeH5?yj^1-U2_I!ba2#e@ z2`EjQhWlQn!;2w~x*Ew%oQfv^r{tn^lEhwg!+ENPbteQAqj87ieHtw;Zq?;&UTs zdkef7ZFQ);sDvUT88IBmFyQg-1IjOL_0JX(r&+1nw&{H!9g)5c+Q!fqSMmBOsDxz{ zY^e7Yxvfzmw0{@YFNqILM1vFiMBf_Fu}~?7C;Z_{bF)naBzFZ$i%>C5P8igj;zVc` z6e?SM_bZjI7bHq}iX6@jqXtBa~#z%ubQnefrYN>f=@iHY&F zON7K6ovDW3k~@^rt|MrTQrLauP7LFyh7cx!%w#kQu90y1x4NYDQXLXgoV&TlvITDCp=`fRQZf9bv)U?PpO8bA=a^Snad6wCy(`XQhtG z|6#MNLvHuZZ&{=fP0 zQ@-@r@bcHUltQzP@F)zdM5zsCZ(s!oHSK%<;^U%F6zFc#<|RE%9gUK^?<#dQr_XP> z7ZU*ek<(;pV=k6Iau8Sl2F(@l6KI(YBOZAX zzhfC`bN(cv8dYl~Xp`F-H53#`RuTy0i7k%6M=TEP(IPc6n0EYk-UDHod}2mJt=eR? zCJ(kH6bLe=ahOH}+?v*+V6>f(s7X3GA%bl8^J|Yhy`zu1ka8B!^Z*ZfjD#EtCT2dM zqd)mwkvPZzY9yNtvl+?YG_0oTxsw91)8 z_y^|jlPx<6nhL7lmM0CvcG>X1uix15^HkC3GYb?gkZ`Pg?Th|!204-KO_EDLwlAR^)P5+A-&KMCy00sen{b@chFEgCIkj+-ZNmah2#mZn; zbj2DCtptF*evuVeZ5$#?Z=_W|z{4EJ;V`7^gK zF=s#AWe9f|Suj9h-+>6RTlYU`hV;qoElAMrny}?DDSs;)PIQ?wKE{g~k6y1r*Xq{G z-#z3_u0RtKlX{MHTVZ(N{V2>Qa~2s}!Ms|GWOl!zB?{{5@X==&lgw)`#zrOzhTJHB zQ~ZNKNQ2gXE1p{lV9=C(59Y@}v~$lk_k;Kv@>|Atl)9C&8N&U$x6h^<@7jnI1-cv4Bp}T*kLJq15 z0*~mYr8Lff;^Z|Rms++7M1=946IW9Uqj$V9eDxArBK^6Mz^Hc&7%5qppAA_GV zx!JSXuOT(^+;68KZ=t&a5DnvysCx}gPL+gzjlPt@K8epJgT_Fm3_iXDM0x=25uRfJ zghC`ToQmk@S34I34*X&fm$bECBApznC}@+@#rTUgGU+0JCGECx=U%-kM;D7x3$sl@90J z;G;2ySs@P^VC%vJgd`Gbh`lFMk`uye!qLl&dn13!T`0xIEl#|b)z1G)trqs|245w9 z^_`ON{?TO;_v!zkTCKh8J(9f;i%4a);0!3WG2d@A(ubwyA`+4MRUh0xkaVdKj4G*SE*wyBgq$y43>^GdTXfW zQ&dD3oe9x>aGGC1mk#6fI6~k8;ONhTxd! zYdk_U*#NJTI0^9b)G0-~K`>~fY}7ED@jQ~?1`94FzC zl13?w_!?vQA&hRE8H>mYdRpME0iElx^~GskAq&zq6+AG#h+e$e23o#CSbqyoU(RS4 zx>+xXfKf{@p2lKzAxdrp&z#V9K^wFyHLtcmMTL}N_cr>vN5NbeLsIy7Z6>aq73Aau7vzeEmXxo$d0{XRd_(8?Fh##V{5FBIF6CiNT_|6-7+9KzN?muBnM6?qfT?vd^FA?#H~}K&cC{;7dL~D>;eIAL0p7hK3;! z$q)E{Ku;-?1sGlST*lX57pcP%# z+zENId8$$0R*2D|)NoJb3rfb_9Zz}l$N*cOJn#2~0e#@`ztO9)xX67B_*gyM4@SNa zp6RIMab1xPKBXPmi89IzE0Z@U2oKPy^)9(UDd3PYNxTB_Rc#lj z?MM}okEl&FGQ>v|AlxeLgb1@HMig4r$CX)^cuykk*G}p)6?x|W0xr2yRL_^lrUQ$L z-py37lbRGKnfYXQ#Llep$bv7J+^ifXp8H#yubE@_!KF!~a=Jk5H6STDobBYu<2!gN zY;X+_hdz!RDqN5QcX~3!J>E+ zyxsGP;>GXwIKv~pmshQlWrWBl8Xdy5C0!I|w_=)p=Oj6tx1&d%mF zQo$LY@ZLV)B}`4nsFZYW&E$=Fl7_2^gcUM5QAVKQC}?n?bYbgD6SC-Ma0G5*g5Zc! z1~1Oj@)dUUi2Q`&y}`1Di#V0^`^5Q@Px?n+&T#pJ3#esZzbFoAb9VMlv`%aFeRj1B zz{>jq3sO&qA0Zf(|Ah)}oVoLVJm@m&S}O8q*zpVhPQui^Si|E>*$MQX(VBdM?vapp zb@8B2u>}vArf4sc84Kb6o91a^(kKd`dH zg63AE%06f{Qe z3fiy^Gr^T`$o1o=Na3r*`~>)kzxk!;VAKYh1%$sK2MFat+A5T#D9zKD1#+%ZLxF`K z|M6P6D{NavEXyD~=eszhn~`I~iwPbLGbIjn=8O#FusHaqn(IPi-zaPzM&G`qPA=lc z$hnJ%VPHB49CHMN1SDY7CA*z)jspJ{T>nZ?@B;|jFvJk4bmv{X#HMyIyt1P<9)wxQGe=}4{^c$UI z{FcTWW3Y2TdG1KO-pA9cFvdBp;(V|6n~$UOo|O_j!`r*dM+YANhY8Vr>yM4TJmAyt z&yT#(*}v6d?w%OpMd*D@=e+%cH88Sju4{%LpH*^K-W+YR4p1E3L?NsbyO=)#p*`X?V4Xr7xVX@lXp%Gi$B5dW0 zAw2wFJgoW9Ik)Y+W>JF1m=ula13VwJk2-(iampmjV*Vc85*(UCB2c2UG41%Dppljn zIhfrx@YHrLGInm6o^o3Q!u6~PTcak)7(y`#KsuO=jBLF$Z)Y=2e6{kat zF&!?_%6K`lVlg~iLCpTpk^ayi=r@$YaIBYW(9!#o2}^SZjc)M+HG2_%x_1(}(P4VY zgG^%B(MQ%~VswPlSMztk69fU0g}}x7QK5nftV&}A5*gBg`}Hh7mvUC)XDLfc;>O1&cLj3T-}weY}foOKsX*oEY6y$o_Qaa|C>>APzwk{Oy9 zU^H5Co$8DNo{7-g;Oo^t7bV(siKI@vO%4Y zY9sZ2B=Ka!i#>aX268)|6|d;7|Mp*R48p&0-@9r`^bhJk&iZQVe(M?g7CpJX(tTlK z*tKgbSMNKluK88}(yYn9)K6aSeaYA>sA2J-T@goRLXv-7)6m_U`fDVT155KQWn^tj zsZe3uU$~6tnqVBAjz|>+px+9frr7QYD6+$lDg|Zj;?j;h3z#%!tp7!RJ5oOVDBS!W zCt7pUtYRS*aSeDd?}sn=XTB)hd|e)K{@Ex?xAdE_iQx$Tm^XDhmf3@I)uZ} zB+q<0j+28G&}Lr)=lNA>E3UC_4RTaB%D5Xe)d(}?B;qi#@@G(ot62_cFUBFhH(Wl# z7f=bZZZ8gH+zi}0q0M8_<`^gqIJDM^n{4KZ<0kiJIh+6o%EGyo({xL58hkgpK9o4H zmP^@h`fs$o2{@H&`#vnqlWJGHC{ZbCP{xqZ$P$twWJrb(lBtrZN%LTpsR$+Wlp&#_ zG?00o5|uG&LP_B}ueI2(z2D#e{r%tX_&&$+Zf{s?J^Lo zg_dKA-aCV|nA zZdHChSQQtDIczUh^0APpMwcODlt2Rodyq_90FvHNwbrUy9S!HaQHyfTs2<$>9Vi60 zLC0AFFufa1Fjt^nSGOVFP+lJrjBh_((!s;`KuWd$$VF7|l-4@4z!vsLF<5S zyj*Dc^7AAUtLo6GLXHv%(cA}gby0^K??$cVIN49Q zImgDVQ8ZK;^S={sO`EsYrW{hkWQ_KuAFkLPw+l@ePE>x!lf}?vH8NXZi)fgm1vVe1 z?J_Il^rN9Z+yB!=rs3EAABWf3HhR!_S%?q}@TkfVvUTIRo1;M<&54(^c^!eLJpMHe zfB#UE6wbVMO%p@OdpV zb+s*2T9@Dk?@;N?Fav%#ssm5%fPg=V-}nB+TUrYEgR7a#2!q~uR0mY!HtSnpLOCl7 zg5(=Q!@|Z{cflZ;2>7?PnI}O;;o*4U&sB*F#0($1w7}6$>zfOh>YA4cW}}*6`lFXo zKlpQt^@Z4T+}$FgM0n}`cmyt4RaaLxs}GRM%Er$>pN!`l6SG}aRRYoZNiP#pwoSQ` ziK(ylUmBkTXC8xHLeoK+txN*aeNr6Tik}Dw2#kp_yl>?%q!D`7%LLt-13G?WLZNU; zfVED1zep00FjywPT@alb-wr@py_Ts0d@CO!ZO+QhhRg=%z}_D$f|>t!V7=dY)`IhA z0lky}n!nPW0Fh>6ar0sY9OEhJaV|fxQiJ-wjH0_DTknK2}e+!%@?n*?+21f%vIE$zw}W%jc{~`?>M#ge)1gHZU{G~Q zSmxptINcQxw>R0Par(%EE2-?nb8Lg?Zi4n_r;MJg(bJTec#vU8YsbgHoVkT9o_c-7 zZZ!yZLqlVNrWwTVh9)y-&8rPKVzI4R*z_T|iPKF3I*}M4ObwBy{K`aZk&fO10t~oN zF6|%a=cT*z0(EFx4jFe=ANm$+7unM~Dm3kV#m-y{H=X-qNjK0xv$O&b&pnXZnf=nn zOjzB7y~kOrRnVJ9)$>}B8=3}O6~Gyh&(1xZ33O5uj_5$K z<3Rf&3?EG~e#xa7r>9feF0}MFn)V-z$0{9wMkgJmpkgu8mM&Sc1#eJ{G4^!a@!*wT z%w?1U{7kgA$7wk!93*1wc~Rt*?`p%aofb+Ocq%;*>vup~nd5LX1~BFlbB|D~n#ASW z3BhW&WFx6&v&}Ucvs~-nV2rYa=2mBn04+hK@~nCD=6!LgM*I;PGbm|YcfSIo#AZ3v zgZ}PDo9N{tB0I2PS~TyXe?TYq@Jsc1B5w|7#UjG zz;e;k2ZpA!*e;j^K_u8H!945A7@cdT!EaG0%p-dw0Yx3*EBOSYqu*l6_&b%)Gv~|^ zZ146S^v|`_$FOY-hh5V7I&CLvc`BepiM8E?h1eCRCymcn1aIzhN70E;dqY-Ejg0+uk8b)M5@oY=)sT&+W9l%` zc2vi^ckjeaa;8)Kp))Xm-!4ZpSaT`<=zg91Ryy7*bu_?umBQwqu(5yqqexlHJrxax zx4r6JTa)e>Ef5K4UPsOeNG!X+T44-p$_U7>OQ_ItM03zK>~^2JPOY1g)`$XUdJFB` ziUF?I@|sKG&AXu?(-C}s^4#UiT@+?jcxPvBuVn#s0u}`yw=;pw{uzYaKY$onDK!H1#+yVW97D2o29oPEJm;Ns=Ek zW6qp#lGW!qF*FcvhCT6!F_El&USo63Kv(w$jUk4)$Z58V#8Ic2*jZghc|NP0K};R^LUN)^ zzyL(hw_!^<0v2Q{c`d5!ME`BGbbYOk&@6?M-3M)-@R=;z8FukQWo)!#N2Sir?MfPO zS(*)(yW{>zY^yK@~;?>#OB|FPDMo}uCHx=NDn@KEqPl9@n_7(J5xrJF&>g>h@OMtZuU5&y&*!dODYDD>_<_a8pQ2fQ^i^#R zi3&3mWz=J)iYb+1b&Gy}ojhwyzH_B~nMfg+jgHrnZ=~`uSk0 z1&&5>X5jUt*6c*nJW>5r)4V@d%vFbvy!u)`wIX82CT7fzbtlHEV>=afq_FS@+Xege z+*sF~H5S`dQ=_qbdi#>4OD)l(rnP4s8g4CP^g(0h9S{&;+x{Ikj`nI1pE^-+2yC5D&{uq)pv1uhzKFPy1KfnkC#Y$WC5_2j^I$12?^!u`ZAff zNqtv0gD2E|^ki}$S%dx9ri)YSV)3sxlLW?07VoY5jSeuTev9_vae(Q~uzT}>ANf!W z^fx=1aEbh`C=(EUVNzY#+GT;*QLFdq#NX~Wp{wZtSTj+Z>MAL^L2Q-({m?~`4^{C& zc*491E~p2iI9#9|(UsO?m!BK$247bu7;QQ*<{EWco_o#+Ic7S*pKqZo{!QVH(U2#x za=;mihH{DW2z1M(S>%y0m@;-vXCaEZgQF0IC_xtmQqqH$k&385)n1I?F9MSdRB(QF zz#lh%J(9fl38!^^x+eBf{`bvC5y*}*08HTJv)WLY0e5{1S9|U4 zA@am9rrTC%#S{aUy^YW{Li^r6-3+H#Z1GS(KUU6&Fn4{cWG1TYYablCgNpE2s16y3 z^GcA%#8IvRVOT9j*khaCL7uiAiN>g%`>P}I!pg^JC7K%d@SAc4PU6?EUvG%*zkGSV zH1kgDCLG`t#HO%G+hDvkA~o((sJcowS*f(W@flS#U#*v)1(m%k<>R==od|ZzAG;nS z_Xs&yF`Bg@E{JJWO%M`dZS}Y82}~RulIPEzBO_CR()?A1kM+RTodL5ujiKS-kPPs8 z2NLPr8J*oLHhg1SP3TE3oQ6qX|y^%mCGr7DIs-xSG{ z100M{IL^5+!+lvfR+Ri7=)^E=&;!^jmFbAy+EEcuj17GUt~Uz$qNBR=Xm$9-hn+4L zMg2HU(Xg~NHhV%gj-||s$;tw?4FNOsBVN!=G&zTKUlF1M3Lc|x7_2ZVnipFcCp&kEFIjiieaGl5ZtrdwK%%w_?lKu;SvnpsxFT4_ z1FIYdO<6mh7@kJPET9@l+3CbeTehaXc1PtnOnfYlq?MnMzS@AzBj+9soG?ec`566% z2ZxL)t3$m)2{I`p^O{)f6!Vge@J3+!gG0;WSs0i?uf6}?Yf5Y&ng3(5WrZ>UM8Lsf zsU>33=Y!#(Z=%5CGD86ep{Qf48kW@_0%ft~44@gR;!} z^P`Pe4(;p|l_9pzBf%E@{{hT)Ne8E^c~^Nhz$EMHgbftf!7X4J!(ncVK8dnG=d@Go z6dpL;M%W|D>n&hPxhr9GRNyjF!U9>JeoEd~zX(?3L7X7-Cwi-9zYM_Y*E** z*3;M=2ar?NK020)JdrSqXrvuc29MI~@3KD2j(Z5CW4$9sqm)g?uxys~6RDa1`2AlX zH2vD+{V?*!e)(l~^KJG&I5?!#uGf!kOoqE_M=BnrmzG$6cldYsepL{uQV2p|OKcqO z${FUiXA#!CYVE2S(kxEc`C=|~7TU~Rw>M8ja({CLp4M5b?o1B;!@VwYfwD(M+hjPW za=qpAH$spmxtskkpYE`;OGK)(gux?i7kc~79qTweF8;)lK~9g?JD|`}sY>fg;2Loy z9`lEy#i5_gAGUGqf7{Ogt6-9Tkag%^^AG9YFV7`rc-Y$h?meV>L!5!2Bm+)uF|d*4 zno$w3L6pVczTGRvzPe+$5zPqoC+UcGJc@4*j^zkx6@4)`VPz=5FGfgkGLGo~ z9<=Y!=F@J)19N7EBJ93Z^`1U)=xe%gqRj5X>oawv4!t{tK_JeZ^%=VnK$fG#BOXO+ z7J%W>kuMW2NLv_up_A`0GmrOB*lZMcY9-`en|}TH?2AYk;FJUGu}%l%m*XUCUTXtq zDu-i_tFG2Yl34J8v)O1eGPoNRz3!My%W>OsR>6)e33hwaj! zqHy5jo2Y)#MJ}km886%{+h@oz-I-pE;BFg}Yk};tpd!Kes(A@C;utL|@}9S`iPbEv z7jvJ#e7PqTpF3Tb9(?!QkcE@UP9Y)H35H7nAn$JS-w4-$%DvI80zB8# zRRe^%Qb-PKYbN%Yoy|Hyjp;0^iCuOHh zitOq%7LQnSud@o9{)7ZG0-%L56BPxqVB(_@Ax7zE+{d8|MFGAV(*Y&$5iswPTI?es z>>wAk9lwdgv_}1(E^--p*mP%)ur?iT@JUCHB_a%GSL6W#e@MdpZDYH>@l3$9=Z z7J0AQQ~7DQvX9>CtkM{s!j0a`HFv>vWaN#wQ~q}%thqSDKl;D0R6~CnAn%)Z7m0)J zh@&-Z);4p*S4BfVn)Qil3xwygmcY-BZNJ#Bm*V<|Ul{gv(?{W$qX8H{gGb6X7*GGv zSS;gaDb|x8b{>m7M7E+T@}3Q#YX0Ol_fZOM|J{B3XLB8}Y}kn@d|IX+?7LmSq;BjBWVGgWzKL zxZ*`L^9^6qN5T2^;B?oKL(Lj{a$pX}sRcXL*uQ1}_vJsAa(D?ADJUzShdN;;1K57* z?b}ixu};&KS$pBgezti(%VA+*f&A~9BSpU>a&u#Rdmtdyp;UNbkb;6P%uf%@&^ihC2l0mW|z*oav zI{DJ=e`W2};XTCu%YOzv|J^P9V?8KhV!v(U@1u_lF3p!N9X7ZK6ov&({G2exTND-l z({nXC#fm~XJSJR4+K!w{hglxGXYH0PhpK@O@?U44x{bIid?}1k+P19>P7^ec^4+_4 zYxEHwDLlJF(aG?KvM=l34*q|BV*j|kf36KI;dC`{k>M!VwZxDKWmQ$pOFR+|kkUsW z1#lYb9w%%Ih#AGLXGiWcg-u)l^34%d6UHJI^9!oeXwZq<(P<}asyBP11hjzVL@5@_ zYChP@Iv$IQuTDi+g4BWlBZ5`X4?6}tas;Ty^Or7BSdilS%PC{H2Ab}i|MqnAoEOLf zD0`YMq_G*rC8fyC7bLi$30Q5KeYolnSsvhn^+v&rGLxw6h-F;A@bXiCuxvwW{aC3Q zbmU~w$}+rGQeydyqV?&+Zy(YLEN zDMUd4#3VOX4@qnvfOgJG;7@gyuOHldJUBpbK+WYph7KRShKhzU5nsj7nnz;18As7G!&vhi@*=WTQlls8 zM7r|<+P^SwwDy>P%jRVn6OJ5v z+?p-2NhPN8^<^PItH;!bA$qv6{@8usP=UVe)QMxv7w1P5UdPsP6sq1{t;yh;3{^$I zIhK{h^@t;YW;%RpV&FjO>{JhB_X|@|QMpygKAKl=aJ-4$qx)%x~ ze+Uf~faHKb+ZnqR<#yX-)@!hjEBg*|cyXQTK%Fs-P=V?<6LVm?*M&#{+* z-G`xN7F9S#Y2UjoH_&TLD3;S-Dn;BCGLloC^Ope)!9>-0ys}5s0psr6xbXzVMHTUe zhQhvie-DkMC^Q13PyuB6|Fh5t;SZ|8{^!sLVH2pT>Jc9G*88vka{v;0!n@;sW6DLt zr@V$kXlzHi8$;qw@lX#S1CuNtI$5mG^O!#2g2E;}y#N8I8up8sAjn~q zLVB}FYgj%=e>Nd7VS@N9;9*P{=HFJS>(S%|NYN6;@*TG~FUuS8;T#L;!!hMnm)VU* z59T!V(8d=j7~XS*w6qT5YvMb)qBalTF1}yAU?+LzWC0x_G~L*WrfOGEM-fbfBBsOd z!r#y4<&A}wgm5~Nt$?my#&Aa65>QjCIJR<~|KR2=TPy%uRa)aMPyI=*;8DVn!EIHta`)2RbZQHufAeS4tHzMNkV>TNV*v`a?D@H3#Yl zRfq;)Lq|;C$m`^~>B2Tr3UP4C6+n)bFaR50{rjd}9m6@bozO_6v8lsiXhkKw0Csl? z1IzQ?Lk7qCEMYIjlk*g^!C<7A@+))WoPmCu*AEdV11q0mgW)`4cSkhwlmB`Nb2rkC)0qCB_m4w&^78 z1y-lVF>p;x0b>>O$!w=!dln)lI?VsrWvd`^2o1RMpfSkuHm+r1G@Dz`Vu0NNaIMopY?%X>#kI|qt zATGZ$ivo<=MAHj$KbvMsSl6=RA_#26?V6NsMzc3)&N}#^8*;AcKvxP!p)kZlR)Gxa z>Q|pQ?LGimsDdpQ2Tn`sx;}(#F^)K|_icQaYbs+iJOYV-P|<(nUClZ5e8N=btl6{g zf-%*Axq>#v@@_;#*2=ADhTuq!7P}8Y&0W>MTesei6cBe0Tiea+XWc| zaYI`!z9C3)Nv z9%4HfoI$Fn@@PhkRDlZGJBU{z0Ol+Ql?%Rc)6J+MfSC2UX$x0|1)$5`8GtfjIC*VQ zxlYhNjHgB5M&;lYK6&b;0R5S!295Dd-^IS%J+NSrP=5SvZ_N#P3$V;bWGfnbdF3@H z`h2cQQiMG1t^kQhNk(Xs-oBLExEs}0HJU87EQi7^`;2;^>S30ZF+!>9R&xEs2b_{% zGCKn}XT!f}17T_X5HO z&QO2>{yxcvM4IF~2>u+V!xT~Eu>$uh`pG*!qs?jS`YCeFH@*y zq4aZLtwoGqwB>j$_K*AV(}glc1ro&bmoMKR z5$vsA$vG8dL2ii?(lKA63Q3BIp%BXr9v(TnVq}ybGJS_}17e>?NNp)sMSZgA-}1i3 z`CE1cZ(zRiSw;Ld(Q}Yz-(+Qu!#C_2{Y{-G zsYJRR6MKo_7vg)P-ewdz;{g(^4vL&WL!!!^d`CY2!VLf}nx0{ZXfjk!D^qLWG9Xrt z#L;4uV#QEs^k)J4wxkXMXh)3rXhbrz0K>m}r?D^cAh0L=XH(E&6a@C3@?42Y*_#P?fz^)%y;A-6H=P3V|0}2_pUN!$6QJ>uG1eO{{-v7!D0I(CvMQY zZJQ=DTTN=Q-Iw1_PcPcXITb3K3Itii2|*?N`61tGiJtCxS=cZ=XuxHfUXN8sfwXis zMm!92w5%&n6R_f(3MYDtWNZlLpjRu^b?a{10lH{AI@+Qo5E@#bWUho4b<6 zKy$?O2hnB09ONaz8{{pYKYJEX^?B%|@DaQUXDtwk#X^iVGQ-9;HA3w^vEZm~xQ&rd zi}uz-7;)e)xm84CH9!fpL8QNa)8Y6ktGZl$SBo zk(2eA3UL#Ww8B^jFkpyDfCg!#0hJaoLn#4)M5FTli@B%_P_#+i1K*c?*Wj;<#lCby zN_P%yZ#y*XRS0{?gC9r(HH$p$&xB5-VcyrThY~>t-+fenIqS6^s}Ou(dXi;_B>uhe z!f;4E*_ z#`Ch>Fs8Dgo54%7gZ<7ghKtd6Sh60BmN@758D!H4T1+gxPW8)LADW(6pw@EdVZs@XuDmk)dtQ&dXDP1=CmK$0M=}5Nwa+aayr& zKqXM#mh`xRAZqG)=pV;6e*6J=1&QZILOt`j&A)`#*b$P>BeL(58v-Q;v#zs}s0&a5 zr*6x#KaTa^>HBcsQhYbp9@HXDRD%w9HxxckIJRm0@HHO_(lSBD9@PQ>P5j$p*1*7()R&QiKXRc$00MO7y{)wNp1nC7Ii24YU9I( zI@oLd%>$O8ozj$!s(AVEUve1XkhX{Yejb}pIR&7kd<1KV-L?r?C=$UC>{n#NEgOdd z{K7fFIACf1cCzv2ceY+i%orKo?_G=wWPiG?(?t;o3=z>RQ&#{SKk^JN?Ped8vmB68 zkR%iH9W?(JFw!n zZxh2f62o6`NMyuiHA-Fli$cFBLLNaNzG{VUo~iv7{FN7*G8rnNMWr3+8Gzj=4BSz9 zoUnmYTz?Y+u-^FJL}UNB_WxROHgvfG^3YyG==eX96&{2LqJ_jz%p1(6a~F5X? zPQ;HgkdDE{#Ye}bD+6au$`~?RthZl8@*uQa-+=m%+j!9|2Uk>K5Xu*gfb^&y_s~Vq zH?K<&-H`qlV{{pCQKjg3NO(XlbTcqJH&8G`LTiK`1fDI|it(#w2YViqwg1n0y<#Lr zr;)5t4jlzt-_sG%v^u1_$slMSq3+Kajgzb+M%Z4DoTmF39D*5$V@`T<3q*oRB>}kr zNhptM&mMWjR}`|3LZEk90NU3lEkigv>krPZ1g(^fm{ouXk(9JQM!am%e?|PiSpUAWJIhxy>fzh07W>Y zNFc-y8nYfGY1R7dkM_WRhPd45Z~NKyPw+#ulg&}Ogm@){iWWJ?P70_njv0|yWu1=w z;b(|!Cne(|$IpigRNWrjsW+{DaqidVeVI~+G848}3m#pCxPy9Whjj!#*dChGup0Qx zY(ck5#lBV-Cv4}epOm5wtjR`qSWV=#6b>#?zMhb=68F6M0ccd_d?$qu8(+#DTKfG$ zSHkvyGp?K*3MTK97H~{IH+OL9_&;XObc9sv0B}ao6qji;K$cafKQexqabX5e;@fFr z7&g0&_7LJ{O~6kXLp;uetDtKj{?lZVJbKwbA^^^*I!Y=k{ilqIKMzS+9_$k(Pl69S^RGr?pkA5-q^}quXf{nmmI)z8Z6O);B!)A;9hlS7n zA`G)o3WPH>68QR;ya)%YG3)rqv;A^dpu~Eke=|A5#SoB$U>)(aA)Z5|yyD;1k$oLA z7U5WYqWveljATE_|NqFB8h`)0B6e7F!p{lhi>9dkzq-pt5S&gQwHc*ARRBATkO9N2 zuO&vZ;GqZTf=71|Ff`f_*w9C zLJ3W|FEbQ{xSRfi77e7wNGCD;2JCf^;#w@;cwz``sny@=;bL%Qsr>7&F`wQ)b5d%p z^ig{Etlus~KGS@g`4;7M0=4g+-QSGwa(sag466k{&I~{FPR^@(wc6{|A7_3)S0;5R zNNLV@H=`h_Ycqf3ogN5W!L0p5s2Yu>x+ma+SkTMpbzAnY(5RhZvB)%&-mCWAX0KAN zbN8iogeEU8O)!K7c_;gE=6)EdX}_KwtrjfImOB)tRI?g?C6qY)cK^JO;jD#qUyV}# z(j@`_N48@5Hw!wv37}I%_qb6bsW9WA$|3fjG)nzL=d@<%@hFV>^3&x8noS5;;4mUv z{r)Ecr|ED1CKNl%!AkjGjtI*Bm!taMMF+P(vH?-ppgYK!oqTc8wi`mo-ESEJ(fK)1 z5z~PLGlghckICa$Axc;P`}WxX{Wm4suvCy0;vfdt$SluvcSou_LWz&($$s@_4&#d; zIc-KP@c*)g|G9YJf6$7tKgi?t+a70@4Q)KJk(j;=(~J zGlR!&*-(8i=v`v@fU`ugkijA89sQ-k=0(L|%#L zSr!1!(8eiIAQCL-`Sb4*0E{XuW5t{A1108Bfd`;-!fas1+cBhM2dF%^-(MA}{O3b) z$Vv1li9#$@GX&*4@PJqdD4FWtO}@^W7v@M7iJ?bvBb5bEff6$m8iWV{YqcnW#fMo(^zPMd-@uLJbo7i1tQ|_x*P$-q z`Ev#nPj$)P@`9$zP$=Mp@XEW%X3GHfJ{}|I@%mF)l2gDz;b+Zu2#0S~8D}j(5j%6% z45RBx!`D4asPn4%g_&&Q`Y%IBcgQl1c9Jk)kW}$9U#eYkeW&78>Z4?FX7iqaWg4{$ zOa2r*f89wbx!J0CQ07?j^P`LI@@kl*G;e(0j@RKEPD_ zZa;*Y>j$4UE*sdfc9BuTR^>f-;UCtAZvXpRKR-wB*_6Cw#K*CI_MiUTBxUTN9qmBc zZ_H$Tno8=|_-NEW-YOrFUuK>5Ku){UTUL#UqnIHdSz zz1rjKJ9arwIw=)Vs9Fi=gVV=;{GD$Yd z9+$CA`XCa^Qo$gXMFbuRwVDE1C-VqKbM|-ZqgZ6|y;&R-lsoydkD#yuCXQ4ADEuhA zcsHdxAFs^*NyrQ^*AkWw5LzgLVrW;C-XVG%e>O3LBZ~|DPADk=6d#JjAR6tEmQ#U% zss5MBc*A()dXcEtgu(Ov)>?G%Fr(E3;W|Aww2@y5mWe@)caY6?3U|a%FDDQ;JAufd zFmEN0?)*>8Lf|{DVU+PfSW%4q1|+OCzqKoJdyw1>8ONw01#69dsWhzxc$jchP(_2N zq25>ZO9GQ1^gMt_f_h)!3d2bs1Ph;%5tD@GkWpY1)ljRf1usHPh=8Q!HQUwb*Xneg z;l@n~u((B1tKE#2S-bjCH*@Rd`Jid!-5z4%FgJLs@|LK=0fDj5Xc*9J`*3Hv?@cgCPpssoDRNWbfXjU`WG^AhN(>ltt-k}sigo~mYlCQ4ZKErGjz?a?21OCAW%D3>rEGK zL>P>1QHH`b2k*}Sok1#TAh^_kh{Pta zvDSxm&hk^iHjbkvW=xWv8ljz&uY(%8mnEB*EaH{6+YGp5C+cTE3bzN$H5Nw;uzs*) zLoV78S^25vuji-$^IKCcDNRTh4cadf><%sTWhian3#nQe-9#T#a*zFHgJExO)>N<)5D-`rdcgQ-g8k66?Cy8)?KOh1~9w>Jotqc70jrs7j+p)%MRI&P3D zt0?TDfEh#6_Z~jFzZs3P!Vsxw&4bL%3?_HqVV%|{@c0rosdbilELj4hRX|$020G$w zB~bwP;#k52Jg)n0{mwh_6c*CNA zriYJ8J7HEV+5y?l!>~B;h+5ITGR=Z3{U=HVe4g6Yw)~jXxdX6K;ZLtj0M;PYjwG zIQ7UND47=cowbO#oO};!5wvCxQM$>IK^_nQ^>+ikAy%Qzu_nDLLV(kLBBJcXg+Mb_ zMe~!mzcU~T&c|6d!Q(SUnGUx>7oOE?OmQ@WtlG)f*O z#wzp-^oAVZg^b+W!E@)&i{c!lW+W%U7b-?ioxOb58Z>vHlbi^%;PF`q;ntK4x7};N z!ASs9RuYAYNX`h*^YZaABN$h}AJe_+La8w2538-gqvl*kCn@p9G1llqr3!^&tSO4~ z*TQ3_WaL|+J75;#&@v@ansu&yhKC9mu3@DQ@P&R>hm013C>=7k<=+DnULVzue$rz2 zd~vuq38o6Hxhjb0`X#r#A@lZw8`ClJu$?MFLy;PvnMrx_Oc0YQ1K}0RQHBXNnF*h? zVJgeC{H$;pPQLbqrDwREyW5N;t*ZAvIs*YL^y-DmeQUCzRV}~^MjX7nk5VWOy444~sYTM|>P@6<18 zsYVH!x$Dx(D9t={9C}WtSXT=z{ZN|1laZHS3GWblG7t^Y!m#V%E+k)c^ypf1qw|k6 z3i#{qnjOI1(x-1_3PEz;UYimPdvdvgE57Q0fphy3yoVRunPl+x*N`ahq$Y~j#z0t? zOp7e%QVk7vm69mT;N%YVYy|!>a(N^`fX7?8YE=P^(`-DY0N6El5M*S{eAy$#pG56V zpfnyOSWJD~5Mjj(VsCXi`T7}?!|C}A+A@Gd`wz8-UWQ$?YuC?60 zAe`9u9kkK?jAu!uI0p#LW2+bsJbA$Ph|U*WTHdGv6@+MM`{W$-nCTQA&!vEkesS#N zKeYqXQNLBA>jsO|g`4rm66lPt3q%Djrpk`WEyVs@srR5b{Tm6FB#|MCy$DcqUTgEZ zT1ga+Zo-9Yr{W)eJ6}ZwuK#gUn16tZH6&xp##Uvr;bX-r=_or#p= z`r4p9edVNIEqT60UtQl*H@n7lyYX1ypl=z8`(SNlG2~fm?k7~YuR*E|8FPgTwIX@a z17%wEC($Ht+yjxJt_^`}58i@J6~UjipxZ{u_p-tMhQS(r%lo*<{r#Pv(^?lqy9Y6u z`}%%-J}BUiO8+`N)56|ghESJ^$Pe`1NpV3y5QYc5RU^wz5CmV4m)aFQrd(4JBG8V| zbb+!c%@t@tZG?5z>Az{9$_4)}0M21Veh|$D3()4o zc&sGaobEsDUqcOZdICFuWK29VNXZZS3!y1kcDzeCTC@zDC({QkRQ(bBgLQD+U+DGr!i-co6EHzG#1B0N)34x zfZ@1IFO#_R4^dO7v#cc=ZLPA;Tb1eu6EnQpAoJ2gKv-{?_$$7o;+CPzSSVesLh<>Z{wlXnjb=wJp&W1$OZSTJxdHOEu zIHJ;d?cF(|Po^x8t#&SW_UxIIQ*8`-e#}F#1EZP`CoCExk4hNCw9E=0;9D|I$twrV zHK1yn=G>zP@nKeO54sLNzjxfwJrJ!3Ca1d*=oB%2!IhV!o;j*#R0v7mGkD9L5$X%O zpxD!1_|bNXge(TIRGvEyDcLFOh_~oFNcIBI;`qRK9mu*QOWSww1bjmO#Cx~4KY6?x zG6tbM<2h%SFxy`J$Bfm2u4AbmygMJSa4Q#|oYf_@KdsI` z2V!=*IkyF^T66ngk8O<*d(b&2Ux(hbn@{n8ugh;Mxuc2|S4F7@=T`HHsjWJoKW8uR zI>VFw^POCiFz3_q+VjT?PJNB|}ML8589e0amOb|}s(6$a}S(*XXNO8v5CuPAL_5kh4OGT+9O zL-Da!z#jz}<5mK>amt$BJrD#XlawXb+nA?Q+y)jdACJ1$O{b+>Gd~Du`#te1%g*R) zICC;E{}9h$?p*{tQf+;R{mN%8Q_pfiB?c=*ruLWI$Z;1>^X%I$Y$`eD2(nsL)Gy=% z?ks@lv;gz$8*_^Kpk>(;-<;1?_zA6f8zl@M9yv5~-n^SHz0D2+0^G{9fbR^edx$n% zqpVR>cwuqFnTDgSyo+=7P)VR;(<;y3*))15+>WS5<>GkSXyVwBO5uu@wzazyfoN#C zlA{%Q;RlB;y>io3AYKwkWpG6Ou#az|4`9L+31m6C4JXkjxgxkuKu~ax>psDv*^iOo zym!h_MdV$8G-0O0O+1}SnUMD;e6HPIjvHmG16nA+HBcl``B4;U6(FSKJ^Jmn{CX|3 zqIl}pKp%s=%MS^B6{Z{2-dlWgmBGVmrSNcO-YW>zVjC?GF!OT1>OSKRuN1rGFVKH)#uKko3qx_j`zSLEtezav#Mfg{Mpe&y@3#F!ANNLuI z+P(|=CX>`!g*&(;0Dz{!4{g@+X(G89!nVuuUT%$9$Y5#w)r+~;@VV_T&-H)SABcP@ zHkp=B=KF(CLxky$vK1XYHvoM2!C8$}33*@s`LaJ_P+BPWAv(1~qJDxv{gfk+;ux^!@?~nktU5Dnh`~o;RkZs&7K@NI@kUK^AuNUltj>WpWH% zI=&^OC)vMyHW4wf&`wkv1oIZr*`zcQ)dl17TUX{fbkY9J|&I^bQ09F zp&30$SP}`g=0z9Oj^}gPE1)B2bn-M0;1x5#o$V1->bMn&wMS4~RDdKj)-e;E|FT>T z?jN8Pk<_|Xf~{bt+F6Du{J<$sq1`4r^Gh2EW7O&xPW5kBC5dQ2IJU{f^F|_m(QIOq4sT;GzUAd zaYf)Jip2Bx=C_Z&U)pXF{(I9onG(1{)m^)G6|XrN?j+T+h02uJ2f8Rqv{f#(1)AYBLdSeWn$HY?@Lv_h*qG`wk#m^lW>YKmpvha_Q$ zufkwcDo1Gu8k98i6*R#C7CPJn(DX)0lqw`DKA{GNe_C|DKhoLqzE56(KHir@zFEzvF_RtbB0x1YrLAEx>mMIC-5AnzvgZQ<3mEOaZ^~O(l zh>D$Z`nM-7)TI<{J&)ZBX2Rp6L`rt#6yw2Y*F9Xc;QuGd7KiKB#uX^b% zU;p>~qU(q){jr}UuS6H=d5JN^hnNp3$Hxg}}p3IER($@6>*2q~d4=8xCZzDqjO1 z4h;2&N!q2sSB3d;o(|`c$&T%m#o+8xWu^3b8~e_bsvCffODU20ph8Ysa#w&o!VVEk z0@2Az><$?>=<*OTN{M}Nyk1(Vk|^NQ`#yo(mS0lu0?xXu{Kxpo62OluCyggtjta@= zR(1CaP{%F5S8iK)fhzCH^2TYxkk#4amTw@6%tM9Ca~fw?Io%bzV2VuC$@<@S=D466 z$&0zhBzfyNwX1J+2*4px;Kj+`^Ux)+bGdW`1<38)<}(K}liVMFNAN!0#cTiuA`Cg= zRwl)2#$1gy4UgC7MROxWqW}Ze5&c_QjZF6+M0}yZ)nlajn7^&@G}Xs6blfOn)YN;&q%RD z?>2WKqTX{yg&~U5mHg;=I3>4KH2bzu)p?-b2~R*@p$Ed$9idYI=7Va59Il-p4M024)0Q_S&L^+n}d=# zLIP514@2ZrtA3~jaa^pzzJ!qOqNX69rBLQ?*#5Z20-~+0GQ44LQV#6)seAGAzjx~H!W z=Ud^1Qt%Vs`Os2*4x;yq4$kNAqZw@#D~vw$5UnY@+%elha@YLVfYeXO9Ylq3K0-X3 zM_MRzGy-~Wfo2$cUBj)=S@h21@j+CCN!L2+myl~j2zIO19_qD;Y)}_a_PKNCa&7hG zTqB+g{-)N`cfyUj`S_IGXXrT(H12$Gh6mjpH7I+`*mk0Tgc#Jw_<+=-Zb1pL_n#m5 zEZ2N>X#6CxnKmJwAGwx-?W(%`nZtA`bU-rIdZ?(HGCsWD{?N8p$=PqmU~7v5TjJJu zjppBLV2LeHUdwFcw(o0nM+`GMPGN0}L&RG^;;FSefX*s%MZ20IfS3fxVzlVP%dmYt zucLR|@l>6?Ul_LnAP~7)E$U3i@-8?8n^1W!^R3)40I2siUG!HA4+!JCo}Rwq2+iF}f~Ax!LN!{Ke$ zm%VK(Ib`=#ewK4W31`64{)tIIZ*_N@{Mv3CAieJee0faqYw&PZJ1e-+eWF+otesXT zswuhVavs7S<2G205HgkP>ALm{sqR6umhZ}0shhlPZ#@iL%!+`_-glFY^E1n=r8-X) z0DV)%3B4pxhR>L?I|N1IK}*Tg0dQ-f#1hm}<*_O-OIA@b_5(usqyTC@r)I1+?d^wy z7s^`f!MUgAqTs}9)|%xCX1~i$%o5g%NAJx2SURs45V6_wq5`YddceH?O&4cT|v!+&0DYMXJczg3M2N1B}m zESv&*bq?mY+g8GrpY+r(b5}FKc=*dQ_fI^8RBX3k5GwpQM5vrh1$caq-5}DrhiBew)?jdSNkoh?|@q3=kL1!*tcy zNLv`R{WP+q3-0J9P|pSyfF7=x^Tqvz3(%acGV6If$|&KgZ0JSVi3kRmUPe-Bm=B8= zTT)pXwCl%}{Oi|db{+*UYFfF0CMB3<^dq$YSiSe!Ab&6teVoZ{K6SU?yVGyy<&K|R zbX_xIp@>LH_veO&8`r9G6WNZKhTa*?9MgT}**75crrJL+FJ+xsdJI*25;Df^L+)P| zs`tw`!kSeeXdpY(g4LaxK0tsBQw7Z#7pu^(wKR4~4$0?7cFXp|ic>-C^oU{EN5D0vANXYPE{X+Y={AJbV7! zALCP{+>~792Q>?5z^SJruy(udU95=1!5Dk@^);JFNX#8;^?#GpHtxapPP*QvG`k^tws2d%llGjAr4WoG777OiN zeiCS=Jsj4!`GUwl;N$-ErAVdue zNBh7wp!YpQS)`$x_lSiVv+l|F*Q5MT zHtOTY-)P#EG*AgF_1IoDtPCpFVs4y_P{x+>aqrYMIAU3uubmuSNTbcV& zU@PF3p64!L;*aX=yH7lg{?{Wd+mR7+_N~UGIuSrvc>!oXA(wc9WB&jqy7G>ScN}CS ze+*lnbiqA+9HYfcaY3%#ozV#X0zS5etK`GyI#7EHuPo2!eX?;X zJaVP%C$Lk8Uw1EkV)Cb@HX46RJn|_E=jhI8_SthWbV`x*AR`hM{7gYbB#||k6*4Lh z?UOAoJ@dI1UD0{H;!7(XH#WE^8`Jh0putT(upv_sAa9I(is+O0~TpVEEEEN*n|FT!TtNC+tcNsaMv6Ma1K#><{-_2LK z-$kk#KM3pi{R=bhKILSeFI0m%0SUC^qhmkr``n!-M(2Cdignl_gL0pP?3QDP5aoqH z;PNzy&}qv}*6Qv;m9Yx`7c-d?P9jiIbKDQy*Pc1ND6wgvxSZ#B3HBP8h{b!4ZJfHs zQ(?5Q0&&AF3TTFA8@8Jh3V{O?Y{IAMaOeAn09BQl+up6G^;Wd|mi)P(feYrK;E+JI zws9TSLJuYLsV{FZB z#Ax<%*2Ls|F3a9CZjK*D1t6>)HzmXnpjz;Fxog+1#m)?Y4NDFwwNJHgRqIG9vB*vi zS~a-lCGtv#?vCFoh{JpxZN>bM<=sTk?D@Q_19+@BGao?MEQfv&^RXuvsy$?WM^wQ= zq`~>D0dIuz$8r&*tuCIeW$4Qt^J7dnG7ax`G~E$`=t)R7IRf)!0)OH*M6C`ZS*`hF z%p@lXJu_gACXc4UM~Ih2e*Qfw771{(mcKM}U{Y8Ns?}bm+esskp)o>}()A%2K)`X{ z%LLU2{x|QM5$cy_r~WNpeG?G4m>(_3k(xH%tyy=ob+=jJ6p1Nw*JMo+bNE*knb2Rh0N_jO^Ft5g)U5~XKP>Yb_;OYNv0{eGYK1S;bm{>=JzTZs3_?Uzz>6-VG>iW+!ju6kC7*79i$>Z?HVY$(p|I` z2G6}zI_cU+wfB4>3tgAJaQ`N|bm@!VFu{^qhuwZ3Z_a;g=y?GqdnY{5%1lFk3vHR{wmWIHffdk;OOA2-oL zgS9zvuVToA>%x2L0j~<76S0>Gn0m8l_jg@CIk)EY;vA0tq!=*@l*tB+FnB}*XDD#=~7DMtB+{jU#Aq_${`QZuXUOPl1#y}nG@b-HTZXer-G!!7a|5f z6rjvy)NtuohXaoESffJuqQIT-?LjtyUu`4{gtB>Ja?WoIT!84t`MMno)_W{Nb zn;**H3wHc){7S$#!>=y1nV~m>nj-b7=zg`kvuOs32%QN)eR(C{v27|DI*c%cPTYOF zfmJ2NndRmcy(APBqm2r!suLC}*1df3LU)uDI4v4QKBddzS8KTN&M=i*u3^}kSr}!J zMXYby0?V9^HQ6|5R)qA)VHRAMTa4JIsYiLu9>6B50E{j=kS;#B|Dn4xW8Kp!$ezVX zuK?wje7yQeL`InDfmZHw%zCi{RRGA?J3e~_Fh)nvJSkmSV(s0~P(L}>K0j=o7E6Wf zL4(e2ucd_OJqp9)@-h104ZVc6eQ}-ZiNz~mnCP~&_8{`TxiA#NS65QzS=Y%WqpBpO zRTeC8X~pnjV?}XAd4;z9g$LE*srtK*AVO!J7ej|)Bf4#GK(|-d=dbf zk(bLH{MiLG*mQzeFJXyTEn$9>51XO>lEmYm!ejR3*t>IrDhPRb*De1_tgnYD0@0HT z6p4kn)D&Q0-2ZYeO1!zh7^ENhy07Dy4VuxNTyXP;yBbtBPM8ivO|Vm+!RKFEfdi_)#CdYjWZs2q}&L?^-{!|Yz(z9?6`1pXMB^= zXE^G627*t(EWEYdSH=fW!|cm8H8rk|HApK9;B)s&%SM$^rg!6eD}e(U0amowgf2R} z{XaZ?c_5VU`aaQQ8JbGARMxVUWe_UaLy{<2vky@zR8+EyP!zIfUrGy7DTS1zu@s4- zXt7jEkwo}iPv`Ufoj=a$n3;Fpd7tOGm+QLj`zA1GFNo4XM6oKjdwpY7zn)a527z7c z3TbdL$c7Ry=Vcp|(az!a-D_$6r1IrMmot~MH)KJ?<^nfIQ720$4> za4ZpI{lU!TEpl-X#)?&A#w zFY0Lc^(aKCINW1yE(o%R@vpzH0Vm-athT=0~2(jkvS$A8%XsyS0*2m57Dz%9@{t0xB;MVbwoi2!kFCy#x1dm%zXm7m~2dzU}2u zg$Tmdxu=FoCqJ;Oe%tLgS#r5i*vEDQaW!0Y!SU}eljpRitD}T*_a2gIpme*JGEkAF zg8cc`YhSHg3g}6~4%np*Kz(?lXdKF*=4A5*gukXJfB*USQT69z_Td zDvR_S7z%oCo(gUMue2Em6W=(P-CARoDM+(w-ha8~kh1zM5UywwghGv@M^~JC@*#R(%rb=MIO^N|{r&5P z_iJ#{rw|yde^dH#NgCPxDS*RJSM1C)dueX%nl)?Q}@s@5|_y*q-S#6WJQMF~A5O=rvg^no0P$J05Au2;?k z8FwfQA_}f;&E>E*3q_hH=b@z=e>Xq3VL$IlNeGlsEZxi}Z>^|jM><~&G+#S}bG-tg zIF7~9i@tl295AxX7Z(RrAZpENTW4GZP4po6xXOEvM8?m(pnV-e>7(!0-6B0Pv9WaU zo|UgjME(W}K!6Zw%U_U=N7=0(K5yY1SuON_zb;Q>X7+sVd|<#}8!Gt{Jofk?l5Zi= zD3SRAqp0Lng;sPN-Ae2R&Q5IEd;NFW+?t^;IHT0=wL>es$c_V>-ZHMJhtNcI^{v`< z_$$Kh1KB(1&T&dB)b=Z)c{Fh-!hyr~dOUMGPa)?{93p09UI)H7v>(z(SAMS@Rn24@) zwpao_~!iMio_n2W`f@x7G3wz)J5uZLb+rz`@F%y3+|DsKcR2jCUpd52#{v`Fc zV#7bqXIUTSnLIgNu6zOV4kF=|d-NA=B40aLoJBhYI{|YBKZ8R6d{Kssd5$(L;YKv3 zPX#+~-L@onelo@G;L}586A;fmn^2FH90wZrrX!k8Qbq_1V7RWj+eDII$pj@M#Lp@e zu_}~Om$~Z99Bhsf(fts06%zw^m=Lf)^_&R-@aZ5-(sD%`GrqC<71x}j$kBuI>(vx> z0G~=f@LC0OTRD*NJF*=#1iU~f1?oo%5w?qn%DksfhHdy%6^PV44ii17BqkScJY{UC+ck3t|w}!>*`y3X8vVu!>0 z`@Pld<QRD&$DvuDqCW0+H~gUM~r2#NSj zLD*z&^J-a7jt2zvLp0)vg7LdKlq%bJ1Ol$ehG4lC!fDG@v%Y@yN+6=ZZTmgHAzrrl zjs>bLbwh2J${}8Y&~G95XNVM*NzQ7Wn)@qs?b{!t!m&B_of(aC?_w1Cc1 z2mh_Zp@YzT?AMn)DPrRHK}HU>xrYj|L9M{riT_L!+-%_wKs|GS$dxn_Y27xyl!nhnn}z z^-iAL1j~6AeGtNFVbJrw2jR^4djF6A+~2@BVO1{NwJx-OK%58VH9WAA4G{154-K7$ ztLz7y|MX|iD+rjTx3JPr&d+Ur)<9*s0j2fZd+JvnQdt7NTq~o}YXW{Oz}ma2Uw)6g{?qiY_R1gHvcS z3&Diy-@kvmZOh!Ao~KMgfJU46)|a7U?Qe#J^4MAxjLAm0i#Db$4rSKcMiB_2Dme8q}b zc*BO*uZ6*V&~^3nK7F&edt`R{$elZPM19`YSVPmCvVPV%v7)eJq(V|&qF`+7Q1!h8 zF`|eSSZ}XwZJpZd@DwfF&mY34?9oIoc)>J0ohDlIXBOtEM)ECL5{g{M1$Oy-5$ARj z6J4P1Ivn-TSg;hR?|s3o+Ej}A<`CL~W5TgF}}CTkYATJD*o7dU`D2|oy)KMU~Pn3oR!6)j{7EQdu@&s-0;;YKzRte|Lw z<6e0Qb2e4|Z9H^+L&JxUA1{roR z%;sh;%=XYWGK$u?ckubV|!_RpcnYU;}T}Q`-J+kl7eXC4Av%K2Z*H;;C zCYNLNBm5~aP40;1N;|oE0P=Z?4cIc)e>|la?Kg~$+k{YJ<)>F$SsmkLsRNN#slwF8 z^c;zOyK)TXYtd{d^>1e zV2&|LOCxCLufT?`(~KEcS63U(-3KWYsSs{mE-o%zkh%S%qw$C)Z#mZb7t@$19*@JO z#wiCVXxw)smQQv$3+>;dShy|!ISr|8^*!fA1AZH~|Ng7g%HCe&DZq0dh&DDsWU~(V z!KR(1yWAj<{)n@mSbDnD5-%ObLUjlK{^sxkdf%V3Gsn&&lv&Ehr%&qa^cGtxJj{JThnOlz0i{WTNc;5cr(Rgbcn?wW;iSR%+MXQR; z8m~P?xJo*58k?F@Wh@FfIJvmQ0i@FdwW@5_y{V$Br2yTzfHaY&%kg7;TmlpDM6Il> z*0il1vX01}fr%uuzt2NKwuCeuLu;=W2uW?GFWL&-TLZ17!alIKxVR*s8%Z4<9ZA3> z$`kF_Cl(bqZP%4oREWzan|!1_W+`IWkJ#cLQKMb};yZ$-3*6QS@Ie#c z0!Wq(nts4;0^$5&8Xowo|FG@!^lSYei-)e!Y}X(A0gK9Jtkr4 zKl#-ruVdunH^V4cq$cbye#Dm*0NnkELqr+XI`!A{+xj9~ z-BAGVD{47*XH0LDfCs@3>wpq&@6HPi2(D@|^<#@O>F`~YJ_STSbB5_n`% zB%k*D4KzaLZmKL6LQ6~QBb<*U+SGl-x%u~PfJ-T(6W zLXu((mt+h>t+P?1@#kR63T4z6g!|MX;%NdFwtW16Q9Ps+IuI1MzA3i4VvOcdJd<)Z zURCLF&=^OYK0Q8Jd-BhDl6*VWp8tjPU)gWaG<+7Iz7*h!3%OfIUg5w!h2}bN2A;qO zA}*fE;yqnmU5wL0*e*>-f{b_V;`5n@`zCvbidQX8TCUm@pgq`j$>+vA;xoy)xjDxC zZ@MKU8)D!-K90u5$4lWc(z~y_J#K6~50{>RqL{aHsLeRKT+sl4Rk=yLay{ z$?vVPj#`zrRWs}CM@(}MIB|lW%E~eVDP8rUfdw)cRY1-Yz(-Lm9Zu)->I#wf52@MN z*%1kRMxo91^`RmgkH!TA1kkE%w}ac;hJwLur{-smDIe$u1WX zia}eJaN2_4;NV-qs8`#s<@^2Z=0BRL6aa)PB4~H+-hGegh{YoN5}*>?9gMeojD{u) z#M!(Nw9YGg&K3Se1fgl~;DA5{7jwtGP7E|GQd>} zKRiDdH2Drsjq&>NLq{X+-X{a9{` z&TZ6y{AfO4?Y4x)8veba?CpafZkX?qU9y{krd2vxXCTEf-MGigfX4&=q3x!hmx{1W z+yP*c%Bl8lyU_LWe9hn>i2t65e%O<=qkBugUk5mwseX6?oJj+cVCB(wJ4{T@*Q|1Z z*~4KE1swJs0YsV={W}tt^9uTi5yBC$IH6qT&XF%xMOk}3&&Y|1u|oIu7&^joK>u*+ z8kVgqRSwzQJ6LWKY2Vo2qnxREgv+Z+)3R0Wm+JEof08qxxHkro ziig+AeWgzTk%vRp;RLlAlx#bS)=k%9Cs!Yy!#lcf?pp?6;cTTxn{a2g>GgsKYY=!N zM6JUpH|#e`w&kY$>8T&CMOkumDRx!!RD{r2e{r;L`*yI%MAw|$MUF)&wi0v}AHTU= zE?le%h?uCH5wW$j#CzE9s))#;aM5YH?KmZ-8Q)Xw@G%^pr&I!?@7SVBoYdYq#@aQWwU&1J-=65}P%88m}4@EtUCLRM%Vmfo18ALoX^Fhno zw{JPPxLg$Vh+uFzQ}MSo%xe9E!!XL5Qk|hOPn-OX*rlN?)W28b?v*QYt=XHOIZ+lb zUK|k-;r6=L{hRm0to&D3_AG`|a)t==l4H@SD=nK-P&5+W=NLTqXYFDJ2bIda8NBtZ z>itwi8vvsh>vCX|Jl;6@{xLU>2>cR?&z4}4;KkC5G9V@VHCL{8xa;P-IRIImX@(jP zw)Yb#{7}cZkn_ZWomS@;dft%UhEum!G$IGmqcmUGw#oXpZ&$%_?kUPts6tQbu-I5` z4l#99z(%zdKlda#9Uwi%XGuBLJU(x%HvO?SKc1<$!I>O{=jZi!5ZkMXzM|*BD>uOQ zVm=WwX*}%(1akcY1NHeUu9lT4qB|mWJG0=dd52dO&~MGJU%v*pKF!#mfU=bfxqe%a zAc0zFz}Axt71laA6#_&;(7YHqSUr3)-Tz>zL?@i|!#H*S%JTAhbQ4s**?;!jxvijN z(A3ynUtizNT;)fFX-a7J(Vu||s=?NI8LD=L>=kYA^D*ep9fGr7*r&F?nM*$3RK zkmhgjX#HQm%G*3L)7EC8e{OGSZRJJk^{hQC!*Dl56P8O$#AIYRF?PeraNk`Fieg7@ z+SFc+*y#+mgez`7t<&x>N$&^+aHkf!l7?7@={b~>;BI~U(QKH!;lR5GD-nWxj3>_r zcih}={{ehwiA5n7n;iWGY$@ zr7X!-Sy4GG)}r{_9h`bDU})f zG;Y;YZbpLd_8n6E6S}#Pc$|XW*T_bt>E1wU|5SSlxB&Nqj0h)+=;IW9jzp%K*Yo*X z^4j4BHnxjbYekq8h9S2^nojq>hJCJdx(jumbR;JNs$OKOz1~o+B3q#m68ALY&8D-p zq=0}Xo@pYzE>Dq7j2)%#3BALHoVp$!L)5LYewO+S#w3?3ra+3`yM$usXrQoph&&GY z`?HI&BTzDW_3*r~7Kd09JX(?A5SKCVl!+6F>xMf^mjMiEw%Y}jm4w4k5xG2EwBo(z z>r_;d4D1K7k_bB+5cvuDpGSi4UraAt;)Tqcj7|TTz2eM?EhkQ#AUcjIav?StQ4{gC)Q9_e{&k5Lyhzl&dmHKRF?@Of z(T_k8+QgDGv0i?tCF4nUJt7+q-!>jHUVQQ-n%u0wZ!mW-`@`)^Mg|5# z0j8wo{+^f7&`9fRJ9_l!Fyci-`e6V@6&uWs^PKJ+15O1m)xzrrIDkXL_aloKJCdJ9 z_)jGa`oN|=J-Y4sK*EaMVhhK%=0|Xdl{nUFk!v74RW8;65AecWy&wwK4)j9SJ|XWo z83gkhGFh3KV+bf*QGs7#U#&cIqG&^2(nAt$f!e-s;$R8#@%o+~p$HlY!q5+=9$A=> z|ND`Vr`wb+@$Y0icmMwVJWKy42<;F!etpY4qsaCW5{vE`8Gur|RCG6z{{fmjHaV%5 z!DeRSQ0*Ie)uTd|sg0Jd>XjP(1-p>lqsKvA}c6N5R*VXs})WIaiOTfoq9DDOkza80|((VhuWTfvKQZ0fm z;6^#XvW*>xAkwuoG=$LL$Pf^tHD?z5>w6?AHAi!EbN9!$Rk?zk;KX(WJbB)|x-eMc z;65$U$E$atW`PY^!-G?^T)N5%qCI)@?1>l^BSNF6zHma)s( z*;!t^9~G-3ny0mxIrFSZCP<3hSI_GF|?ua+YT5b0&Phb>SMiRX?~iL1~l?$g@vYAi7=_LwWt>#+$;eE0aA zDj8AGlUTHvZj^ZN82TGBd)nfud#=2c2Igfq{9LoNG+B%3bg&`aFyA@qBbNS09v_b1 z<2)ZStYFa3k7xCC=_oT7C%6uXp=@%qXpqVg)1|sEO#8nyP%YdP&}eJ{{o`SY29QVQ zH$8XKbo;-3YyDhMWHlYN*lz`Ce9%8K!l}!FQN_)EVObh}Q4SZN#4GyKJ7}_GpD94( z)39+>RjOF7#*J!7yC&Cfzw*KE-I$ocWFB-L*i%4o$NR5iJ2AR3{=_pqD`i?Syqd>* z%c;LkJ*QCY)iSecHQ>NLM^aLm-qsQ+1hrCHT3@UG$i=j#%Zh9)d(&co)*c$VZzIHD zaWZF!|ATPd*HkDo>yAvhyQi{-=-@3}be8_)w1gFmfT=ODO0o8NFtLZCYM=R*jf|W; zcP(&qbo4MJXVgo3Rx=cTr|->wo&#k>%;tm_XHb#xC;f*!Wm- zfbP_IzyJLC44%K+RE@>;5rojziVIe5LlJ|rdrN(!5NiE0Iqxrj1)d@BgQ%q6f<-to zC<-61qau;&e*L<_!vl?#Y9=T8(ECKkG+iYoxm!%~18d>FB?9Ze!KiGeVNw9 zpXJTYh3m@TL5plY&KFTokY&2#^O{m%eZ%i+w{VCJzqzg8Gf%d;#BtcZeVWB0QRk+n z0#a@zW%%yA@-h!LnaVKowqhfnLDa}DCX-lf?=g-U+Nt?B5HjvFySTP$fQ^J zVkysRmUbw_Xu8zTb7+blD)i>X=eh(o1_w8HcvjX*pjC&FKi=A+q?8z%iumI(4%rL; zz4%;Ru#6roaNzH8BwuMeB{%=?d&YkJU=?Erjx;ptc-LR%#KnU>6;j)WlGqp+!o9q# ztgQdz$47E|0M|YvF(;p~pMGX1p)zdeVZp|fdwTC4iFIE%0=h1}ve(*)5(qFmX)QN1 zXOoQ+#ofdh<{9BVVITJXiHXzZ!F~+oD<9JW1W+4hZ~sk%S6}SbBO&SFY~ZUTEQr z)4>?Ml7K9O+rz`->+34+W8Zp>ll$ms)ldRN(dUw3z7FV`w3Zj*u@(~C?M{jBm-F*8 zd~@~khdo--lZ*U-_h$m0K%Ag<{4X*@jI)Cy>{wq`42zgOT34 zPL)Ypx-<--is0sJ9V6Bs{cL$WeCwzz2>^VMi=*Xr=;*R=A=7nppOjB#LgMRW;mXBO zROQFB=-+ZV!#jr-E@aW6pg~>jRVB>J%X_1;^255zkDVx%u8C|6mbv`9)gz0y*Heoc z#TPMILVl%XXj)*Kx6DeN9*|H|;|KY3=;+bGUwf$kU6rN#)&Zn|!>A8?vz9_1_Sd{% z#!^jP4Rm#(7(2=3W*#|3b@n>qHj(^9g4@A!3!Gh!yNoyBg&RXkOVQ;(fpz0#Og+`B z>a`?I6yME)a=?D=_TLq3I2kN;%IfY%QU?(Ck5E=5ubJjyepX~Vw(%lD>kK1%XsPSqC!&j)M*o*PMUy#$HdIKSLu5lh2z)kHV^OqZN5HL z$FaV?zBEkuQKPs3Gv|7zHjm3KjcO-Kkd%~EA(t2J-Wnjc)5rf#9lo4HDgW~S-}FkK zd=`xy{`XmORjj@er>gm)drD7P=^Qeu_y*NP*X^~ar>E&AYzaQKw};P?l?x)dR8?>O z*SCGnfn+w?%gDTf^?wfx=MbQK=`CUfAb0DZJ&T>)Aiyx%aU;)QKG>aASl;61!-o&8 zXR}Z#)Q?ior7BG?q^Gg73tpQ`U|$FdaschIwzETV)~Y+f^ADUhPAkttUT$$@FINAv zIYNL9&=<;B7t3gBrW?Fkv~XeJ=UyxekUnIlKzPX9*nT%KSXrWyQb+u}1n9^bP}mIa zqGg+7>u%g0Jz*ynNjRQ@^Y|Fzo`r25W(`jeL+bO;AVz>q!6{kfB+R&T{Fm&>lPBXC z8j1q{=njDNFD9i(L2SO@_daXS)Le)H6O>riBSX zEaB5B3s6s+ZuI0!(qX2+oJA#Nw!i0Fw6gIzyRCEl{1wIugEb7^*;b5N8 zOrB*lT6GH^p`<<=14|JL!s0y?l7?^aDgFqW)s`2#7*WXjqyPOn!Y6g2flLYOd-w}c z$-=LTlLeVHfP83wBe>ACW#3e)k}Cx}oY5(xgjzAAQs_j%QpB?wk4y)VA8m65Bd6oY{sifn?wVYB<%7KPaQXZ+-uYV8?x5YABiXOc64#$OCNB4wnO(&wTT# zzk*1LnJ-{g0Vq{$VuHQYC%O!ftmxYCjgYc7!i=1BH2iaMrpPLlglhhca9UJYN8MGF z69tGu9KmYXjc^Mmimau110HkyLtc|mp^`^A6wU9;!&W9OpqnG@fXR!+e&zn(i>5j} z$ki3OHcBsKWwC-GV~P<}ogZ{EpxkB3K+!0S=gK>v6$XJLzx0>bJs1D#BLlhVwuyd>TN;MkMK zfv;cVfZmUl==RqRUw}? zMULo<*VBkts+GE=F^4jpiTZ>mlFTviFvv*io>eQ)HS-K9Ll~cp1B`wVvh$)liO!Y#( z3abnQp~^|eai9=$e&on<|AVO=V>{8vgL0a)ej&nkYKrBRmusGxwm+Es{?2jC9=j9C z+hea4!r{Ug&kE){iYnx?l+ixoPqIQFS3%P^>#LdjrmSx`v4(Twaro~HLRfP_u>nK_ zAWLHZbcU*NUbDsQHXy z_C2m%D5+^Qd3V88MF3vBhE5@+NrEFqAP-w5=9EeNRHDa_P<94QR` zBD}xf$x}sGO`-hOtwyjH|Nk{I#sa+B2jG1ryGoaGh>iXH`967iH$DkemVn7DhsOGK z22ZZgI;sABgGsnh$sKb&WKcf9In;Iv%8 z8SPKh&RPI)%b9me`n6(oddu5)yo#Q{=FC;<;pd@N3pb7@S)iI-c(PjlmDb^={ z2th@^NpZ&j$VleXt!wFcSROkcf{Ka?4`<^LGo)Lv{H&weW6GfO+nPY;z;+HDfSu*g zw5p};t@2t9oC31S;cBJ>LtIw&jF235R;SV%^qbk4Dh8GcE zW%eY9Mt$cf81RQU?6~2t8G2F%X)$0BZ~)bNZ@11il#fb%3S3k-&;`nl?7rZqt`Deex+6jk%7S|<7OCyB|!Z{c`k*3D~SSNl2xQLQE%H zHv8Ah#>OatP;O}fMh-^Oa@!5vyLXSCVSyn*0XnASt#D_8Q`duoF^@fD!}r|)fqsDx zVRS%#;xf4i{|Pr;L*9qX|8dtJfY+|ym1qC=C-eYoFECy)T&YER~0Gr58ew7Ol@2f-6rKkF4sjO29G$j-u!Sv3}p6KrB$wjz* ze)0#<;^Eg-a@R9e6I}GL5ygFiy>g-e1}NXFA*V#*-zkxSpwAVFjr^=qp%|d*==OH= ze~jeVm*_8~hgwBF)gB!}AbvFMcV8sE=b8vnc_E5d)T>fjpdc&D1%^CvA*1~jD6e)X zXw_t@kk!JV9~vLrJ{bL>YssruQ2Pz^_lt5f>oCez9IkDVf`Y=)?Y@IdPOykI!SmVc zOsbDO`UOZ-y0ZSCLkBeXLpc;X2k4r-_#NmyC{G$18g@duTe(6DZX0l&^9P&|aLz8V zIoNc9BJ-%%bn^E0R!a>vp?bkuI(=X3ip1~TEUBY7IXnHbKY##@2d5MlsZ?iOlV{2RR)EFyKgREd`yK{L7s(8{?oK{L5Ko1#`+L#nR1bj<9y9g?!*RNm8 zE3zfalnfwA02j-$B9_L?K~lrF6@=)Ll`9V%szJ|i#)gv!0RS2>@73a~;4TL5@Sn*J z9v|cZnCgk!Z075%Q*9Rx7lNx=3_2D!k8-B^e)ahCaGVW%okYCH=v1W8Va8K*N`UOw z$lGe0q=euhX7Dg_lozW0g*th^2+jyRgi`nbcv_ybsPiS}MeQEBhBp0kbUpQ^{{`gI z{WI}~#%XN-K1edI7Ph#0D$Wa4x74=ka%iFks9*@zD-f>ylkd#dKChjhi^UMl%I_Wk zA%6V$uyyZ>(yj@iM5`;0Va`)KowhEiDE*dBLjD z3`-U37szGmxqCx6W{;K1l!QK1`)g*C9UC!pvp4XX>84mBO)Q{oixY*#J&1@d&f}OOD~SKz1oO4qbCA7 zPh-RRco@)X9Pb}@xB7R;r+$9o{^AYQLx8p^Obc40J$Z0ZSBR$s8KcM1hSM@7VxpoUI%kzFI0;dL_TIf zkLtz2$?5F(1pPpNWcPveotl}c^u7fmKYzXvv?Y`BFSodLu&;Sg9-&Sn>YcjFL7V0x z1Kj_>m6;=i3|{Gs^*>46#JmZTZ*~7++WNgUo)dQKU#4s26RvS?ZpH_pZKG!ffOT1Z zM*n9#J@tUGoSpYB?DC$DlZD7MG&D4C{0e@?&#o*J1V`6tG@9sV!Npu$7V?PRup2Q? z5Glu_k}QMEszHY6Cmspw*bffV@C^jCy5)*&dk`|}VMqD!El@=G4@bDCm^^?-1uOzm08+hwCDu zvwIf-RGK5k0`U;9(E?e;wbWsEr4ljlWExrfYJ*Omw65mY~XCnv$IAR@$R=OWReQmNpzQE9h^s`z3k=3)gIg;Vw zw|&=qdEN06K4^vu;wgSgeufGRb%=Z)Tcls!C5yf{j*V~U*CK<)0`=Ixg;7S>PiC2S zclY)6@#IAmbiqJ9zwP6=>+&v60P2EthG`KqAK9j)!#&`B`}^)Lhp0*)<-F~Xaia9Q zSu!F^n6%8cknq4t0scONWgrifprFW=GX}=Js%)8A8~41htpWlWdg;wvUjGTp9HEFyHkOJIOj@(R!bGlPe>qW; zbt07;~5ce?_Kgoq%dY0d@ zgH+(~^u)dkwa}_ep&I%*9%hd~C%Qu~uUUiVP8k`DAnI=+IwJ=>JF_=-wF0?LvlLmVMg%LPDuTr?|rzkyTlWL|hmc%T|6QO1xTgo$4q2!2hfiNx_=rzaScM#aB9uW5>*DThQ5nRnw9+-;kM^P{jGgsgeliFIX2uU0U1hOI!0Qr*qa&t9SI>Q&_4 zyv89tNYckl^n7mvYAasR4hTm+(g#GXa`mgI7hY9@Xztvv4prO_$TbUQdZ8I{`NE9r zEiT>d=~+@Q5^)gmv7+?t1b)yctl(C^efxHXRd^jVyksxtmC3{#hGW$~APjBq zq{qV@48T}p>Vr6a4iTHXkdX@GGLMm8+er2!>4zvR%8kpb z-K**jXg%!za6B++UblWk^DGvP1jq*Ly6Dw)$33Q26fsyZma(6owtan9yBc2c%+n+> zNlWu^gf|EliQofY@Y?OEs0OwQi;PnP6qGlR^!3&RT9&_CX=x5_M#u?eN5sMAb*QAv zZb?LRhz8eA0NbYEJ(dUvthL;+eH*?#guvVoBy?jv!2w8EDr|qo*U7>H-=PmjN{B(R zMPb$Mm;dS`3_cT|3E3{(irR@=F;E1)X#v9zmoA8~*e`@SgiFTH8=~56B*!L9>lE#ym1yWA1llthk+Pz76v)^7K4ZPv| z$%H@cuPBhbAn10%a+C9qu3BJt==Qtq`%4he@N9jFd|4*)wOBvNvt?A9iZj zLgf=I8h?bUgT1|5?GCh|w)u0w>&ow+KTUSuU={Eprdx9Os=>UY8;Fkc4k-Ents8dw z^l6>xVXUC?khWEbFr<^b|A5DbZ?0QiN%IrEqRoRR9F?4$tn<1H(VTL=4&Fx41_=pJ z9LM&<9y2w@YI#-Nq!4&loJZVXV*mqLhCQxlwG+8~nK=2^d4#xEq*d{+MEpzcDPL#8 z?j+$79T@^}fn0TVcXuBX!9>N_`{9uC0_27uuq;eDpG{y8%ZaA!%^NX-Lz-yKv8Tdmd3UI#F6k%+&I!uJ`2kOI z^Al!^JV$GWTUIFEN$&E%Jx+B~>3 zQW_*RT>5`IW;;|af*2Qtktl4s95}AB-QVDg&YUKpXE_WO6%$oZOh`HF7xG=iT(&dk zrI%9RPdC6Ot8jB$L3u5F1D|d7!(7kEh=b^&ugp_10dy4(EiTIUc(Hn3iWt^x}zB(1<@%S1DJw5U z`%P`cZ37=Zgae1-63(bx(0OR6btM|atOjGMi!Sc{-@hwx$V=l=o+^Z$MXVKzOXUGg z>8@~(b<(eWksGDp_`nGx4wd!}@$)}ub1W6<_>28opkRxc1PeX5rL;v4JznkiE5DIf&@I6!p*yQ;E=H$}l zK%52P_em_HxVX6T=*iNZDE5$3xX?AEV`?95Bnz=P~_B{KG5;pDzIIfdw<&=m8aN(VWAbGjVZzHx=5C zn4o!l;FrtT?UkmBa1Y6DvmU&WQBmm;g#7VoRruxS{Cxfa2{bBR06llJ>t}cE+$t$zf{33NIAO@&BCxQV@UNt~vt?gZ|eW!Xqy)&7tJZY`bce4;L5qssQ3> z%Fc5ke@V$}jZ^#&e1)Q3`CJzSd!&}t>qiAD(R)MguCqR1M;L=}_4-(k(7AIy&&kMySUg%f)&s_Cc% zHA2UT{x+$;2SF$ZL9?EVqx6OJR9}&FZUzSr&+?%8j@o(b(QYkV3uh{?@B@>R-uocP zZesQlRta*D$!w*CL|?LZ;*=^b_uTUz2Js|3EUf!@g2`8h%l`v9o(Ee(hha3bqG|_v z1Bf>aZ8J612hmBmwVzp8J$DhKp4j8wnE4r@5eY;wVMM+<|1=|YAJBxNOdmWm_zM8t zkhXW4V3841;Scb>LRyX}TZ9FqLLpQC>#T+WovFTYqet`XAyg>dsjkk2Ejb@@%KrK9 zgIFu&Pbd9X$==LXD~1Zsn)7{KT^;q(cbsw}r$K#^SfEBD0%{2Cke+i<>I!!IuX>H@ z{-@gniK6|FiLxG0REpROkK=!EDDeqaU8ABTmb`IuRPI*l@aRr1)I3*nbTnOO2Efv6 zn;wt<+l<##UVfXTRitp$tpoUUz*$?%W#WapJepa&h@jj)Ahy7eqXICZn&V*%T?;E1nupRAUvz><_LT8%yV$uXk`FO!U>j8EPx?Iy{7mLB$dZlpcp) zMI2xbW>D=g*~_3}qp=k$y)mFrN2j6M-n;=I2yx}%weJwXchld0`4WpkPox}$3FE7S zOo^H!ggZbTO&vgI_?AzjXt=FH4?rQ^Dj>p-!Lz1*=mKUzdL*k~(&|jGNt7?*LrD$R zO6c$GslQY8CyJqd;DgHomApB1XD!%PN=Qnkc?0-8a#q-+KLSJlngzeNXMgXVB}8LE zV3paI_m265`&8qd!06cLPk|P3e)k*ddKa*ff*X&@gU@*S;1i~w;H&Nb9cdJ~d5`^$ zjf->p-it8V72Y46DAr+M0rOgNLcXIJI;8&k0vbCKh{yNuTPqbxcgaru`QwZ*hOp7w z^mcI6gP7!LUAwlE)Nj<+I>JBb7@6EGs;%9EVBrknnfjM6lQps+QgJ_o-mf~c2%=Zw z;BUo)3p8(lkw{p(V-@&jl<`8c<^r=s>f8K;F&eE$^rr=G0*MOZA6>|};~&i9xVQPK z$NrPb0M`S|lIAqK2Wk`OJ+BPUpj*!jNNNC^ET#=qlur|C1gp;uIZbLV7>cop2^B5` zq;gmL4$9ISqL)aJRo&~hf!GUFsw_gfnD;&Svq({+zf!6g_+L~uZum@7TFY#|?zjaM z7NK0j4y<>4K-J|y1i)RV!aRfo2c!TYyTgX)YnlC@K1of_tyXej=MdZf)>D?QhfIz% zn$^ka$y96rXzAqYD#2Ct35CrdjS)_)aVDih1dKz);O+iuSRDWFZyhFPgK=q;9AUf= zOVPVR_^#8CLLj4)56}YTFDq>DePAFlw`x(O5S02E8?K8cc^||UgpM2nPnmx;0aZRm zE1QV)fsw%+_A_xtX`R%zQ9oOL0!s*0r2U^*o0D=p)ee+o+Rv3aifs5))nIYn&Iy$G zVHX-iwI(N0ozrrGc!$u!5cV*JHzCWCl*%!aXKAs|Q(GrS(^}VT6m^5yju20qBNW2$ zsFh9rXAi(cO|0-TdoKD6fR)?uCZQHu3jfF(0YtGJC?d)qP3BcS2^tS3#t?KL(bf0X z20PwZDc~gy_y$C32!eJsg~LaVd~#oSvnh|6M37sn!?{uL6x>B!pU!ZratIy!gx+Wv zD7xVJr2UXK3Raj4!=xkYez4yH?4~{LL%6dP8EH>0TznC^dg2RxM z74XVg7791r*S`TK+NDhwOU857vP9?eBvjB42hWC7T&Po& zeagH6tD zCLnW_rbui3G%=f|fNhXX3S9CJEwm^V?4$_`b0OnK{@uXx_77lcNZZFIY%-_@9iNZAu@A z7sLR}1o@8~`R01?GEzs>xpl;lW)r0Xd=HOFIl_f}ObPvZLMG|kMO~hK+jnFw37Y$B zL*pXD++guwN~m?ufJof%p8)nrPrcY47@ME(TBL7UUw1ta@TvEpm5H&$Ph{Ew?JRo+(Hn=Y zYIf=e8~8qg!Xt(?aHd?fdN#nFnBo<>w-`br~-1^p+$ z+~J_1#0|}pUL4`Ah8ztoPtpAVfm-p5h%f0@5v@Hm0dKPXx5mFK96=teqdBtOX8i`H zwn1$^8-;50?5Bk1Uq4d&Ls0@VRyB@v0=wAtlp|-ZR|Ncs0;pVT(vdL%uI09%j9S?J;0G$?wGjL7V^JlYA%OlXzaD~rq=&~P)UQF~6_tgTt0w@(j^$8< zce|EpWkD7P0#chp2OJuYHY-a2qFT7k` zaA`zbOKTNq9~8EodH$|_Sj;u{B@EGp?-V~z(36`=oZ)WrMAM)~hS^H<+cUiR7vQs} z+ZJ%R--XD7zyNVwiot(=N^Khg=^POLK{KmWTIYb-h{s)w1BY+Yg?vA5E~=~lW<|wP z$=eO!DHJ$GKl1gWqLo9;8vPRceGr+Xen2URZ`kidk+vH1A|jh4f)mJom;Y@c^7BwH zwDO`nsj|K58K~$N!(BQ3=MRaDv5ou^m7WV{!pnq-l5Qp+qu6py1rRo|i9Pb%1tMpY zpHbk66Wq5F^RFohFbd7}&nO1Ze?(P467%I;HbQ7p>t2_~N*1+@Z zYS6wS8M@CK5Ot~&O@pju8>o`N4^R^j;Wq9UHD8qP1E>t(7fbHJ7WBeumTrII!i?U8 zwl*|zk6z)d5&3>z7>b)bQf{CuGo;s2WjKqR>&}PbD?$ zw{EF5)$l_L0iA=asyUm(h3DmhCoaCIN;jVXzH!?9G^8+P)~veGp4Aqv7F;+gLo&U} z+NH^maEd;zcRO%^2f18*oK_LeOX!)`u6Nu+EaV{g)NbyLI`yRb|NlSD*i(9_PDven zC&vnfDB$W!xx(GheugAf^F-?0Mh9cOw2;AK@ioZb?8;*6B14l9p%Nn7&hEKsg4;J) z?WMT8XzOaPguA<99w01zqMQJk&T@boI#|e_0I-*$0q25er;}R-hJ7NN_uIPdWc-r> za1^f5m6bDE?L{~m(%vUIoujl&e0^t5fv?mw+HC^Rk1v`n`24Rl1C?tJ^zWNy-7ycp zeqDjXIEGibh(slzZjf~#GFv*i%YNEi5ABH2B|v<7g|IOcEwLaKwsl%|;A%)Fw-?D2 z=NQ9WNPTwy{3ue`lR^nvv~&+N1o_d&_wXmoKy|vnR^#g5g7u)IOwIfB@tBy8*xx;cPRO7bq=M-E?9yuAT!C zVi<1*0EF_Haa@V=d!4wv?wt6&-p)s3<3{eH%|$oTAn3#3 zjRS`^0lp!{$^Vl8rD`6Ly|+ep%pUTwnMNRt+@};)Yn_Y+0-nB7b6L1B3g&mBgOeJ+ zSUW+%%N`{vsgp524Cp0+*CIA%eDL@4TNXi|ot^EOUCZ7CMF=UgI$uWxH}x2ibo_!2 zG-@d`y+C{9W{NgkXg&djeM*TR+5^#my0;r~Cy1&tQBO`MGajV_IRD zBgYM-^>R}SqJ(fpq|*Ds?ee77eLC<7x}Q7a+W^#DVAbUFqH(QBv{kvV2eDFs-$h6~ z3oiei!9BWj-n=n9UraC{#znJYq84bc4Gt#VApj08Gu@N9JSPg>9Z(9%4v^mfsW@T% z!}&UG7J8vLKnqHSMUIY+L^6Zc)0OofH_B=Apcw%&T2lFhyzMmjHZHTX72=(6NZ5w@ zZ!k;YVxRY*igMvr`;z+9#1uJ`uJWfRLTJ1v{ip zHh*&G!!0}n)c0*H_F9~q!Il!Uddw!0fwO}12 zCj?0Siy;Ce8q27R36d)eQO6!cBJ8rHnaC~>ul$%8u|yMyGxfSC;6|68;aaifapJdl zQHoaMnS;S~oG3$hf+~n|G_>T`xP)x@xshFIBaRc%>1e1yM5Ku3IY=E=L*VcT>2`Ne z51KGNAKG@N99lO*6kjt2Gzd{j2uNuno}((F(-RSg;M*c0#z(_DxS`59UO`0`oY+U? z>(D=9@*YHHD_SluG}^!q6%pW?{Rh&Jx1#He;Nn%A>abIl(8}bfV|sd}u@e7(0)^S# zfT2S|D^ z4<0-SbJsvWFgZ3CulqZsaMFl^dD8&u0D(JlXx=M1@EA?WUbZUYQ?_r?D;Sjq5kMZmiK?x`uKENiUAY{nNt|*LR+%LxqbcSofG&_ z@9+Hx=$Amc!@+otOFdXgE`>N0HOcEajGuN74Gpao%)tZ;7N9>gU5ioXMHHqS`NCh} z(S~5Qx37<2iuFsksN9f6pjOIjnjp;RPlGB{7 z$dgH}4W4gpEGzKV!?_}T&eFSr(RBzdKb}0<`VeiorvKq7`foaZzlFA5^0V`2yb-L8 z0naR{C1#qGerZHpiU)ITn0_BcdC1iv&f+|gstGSe2!;bY1iVd})pYPVk(DbAEWOYO zAue9y{t>wrDU6l@8<;m*OvL#~$p^yzx%%sAMe7CpIGUusxx9Y)@}=h=s_5e>JsuU4 z3tF8{m-mi?>Fwz?*av@0p5^OoVV=l-Mzz!}Ud&k2fntPxD08A`hfmNDGexGv8LUs4 zTln1I@NhJoM{~zv)|jh$g!DQdn0pltDPZnzb=>_S@W}X=#){(hF4#LAwtNu?_>k#F zZr(ag%q-u+nKMFJDD7=WJNo=r1<=P-Bu&!S;4~mys!?g0euKDa`32P{e4RIr|MFVZ z+ti2P!hJ_FPXT*9)H4@@zJq5+A0|YFOu4H~@p&7PWel$=FN3*=0Kin4YrZjCYu9}o zZus-&rd>i?G@4ESE7c>7i&mV5BP1h9$3!es_W=}L(r%S5+62N$V!^ z<&4urJeu65wEM+2Tp#JZA$EV@t)~QLbCb-MvW0xd^JnINg(}s4Ykjg)*V(C&sf2Zg zTl0qjVtN0p!OK%<)C0=uA$NuB&OCOoG<7Ge5OlYg+)SP3Z{E##4|7gB6^LF(BoEYp zs+{n=%(1jw4@ENK=(ItKTCNU(n`FyOW5sryawE+VMx*bMjw-rSVM|GPS zI*IDgvw+XiJF>H6OyJ!2&j0hOj>We~X~_J0lMFyTh!9j=JmGLpmz{kta&&O7FGpit zxJtCIn+)Gq?Y6@w!sW3!s1I_`QC-_f@<^mhz&1IPh%K6EooGwXzKqivR!M0&uhLPq zZ{K@;oaSXVzft@~6)|z5EV5zO+E zVXCoQ&t-7Kyh|)EUeGxB1Q3?eFa4GO_ejr1>S4Rbg~l|VBEGj~i~-3N_sTdWHN_*d zS5AbM6Zu0BFa@MWuKoVboU2(LH*EG07;lR3tz>=^<=B1lt=9+jQ!WOr)*ecmQnKjUjvnW3M)-@KKWarLax1-gy znA?O+92hZ-%%o|j<-?vC`UH;S2!TBeL#9x~TjC(QMrGQg^Sq9494%o3#Kk1eN`yuL zN!i`Kt;=BqVq}0;Lz%=E-mc)Un5dbQ;T&XEY;zsW+;+Er-#jr$iKG7+^ZoxOWHjdYwGGzKq z<8kBWsPJnD1R<3k)9WNPIueV}$w%MD`*j=YHKa2}QW{2)cd@7s;0`%zU57rkNUcHq zXN~pu!6DsIB?S+6^&5YB*=q~Nux^ghAmz}NDu4YjzilenhDR`+&gQ+zLrQu1JjUVS za`d15`4g zWkq04{d%f6Z1vd)yolaqG|t;!_Z(Bl@UX)j4Xyfk0wo$uKL=!)(^Vi&5XV@4x>J``rSc zpS$GwOt6sSSZs(nNth8FpRkze%zZkYh^*y+zkP7@c{I~kg?NlR@@qjB75iICWwBMs zsEgU6WiR#I&RB`(5c23bSJcQ>T%^N?yGCNR>$3PAJ9c<%Q|K>Me+-!6jfeEHEdBg< zC)fXAx|eZc%=U#_RZaZktghtV9?(v~`wAVw`oh}YC7ER6-92#MV&!@6^W=!Y&y~?S zzR!+oH^~@e<)JCp_q--q-Se zP1=1dG=d41FJ7ZRwMkXJ=qQ5V^l>lr8M8(a5Z`;{qM_4{c?XnvFTpzUR{WL1#^$|h zkc90#Cx=3>F>CaVLWdMC{}%@~h%yS@nO{ft^gn&}>;}9VvF@cee~Nv9Kea&3`s!Pj zq)BLa_#3`mn}H{Dg3DBYFTMAZoVwJqJ@olWE9QBmQZWBSKa#Og^)6=v^n{AS z6i=IdKh1QC-Ht!BeIp;qbzyH%uw7j@bolUu^K;T(u5SK&!`5%ZpWjYO>z&D45x#T7 zf>&7ng{e!=Hh$*~yhn}ao#PIYh9aP#L*q6MY1a7^Xo_(nG~SpJXx;vuc3Gj}Gfxa} zKJ>ue+Qvqt8)^34g)pk!CLH_o%PG9!YORa;G=o;0<3xQ8ZLaxP6^cik?`zWoYFv@0 z4{e^M-5}y3xged1=u}*M^r3$?p~c`yIm?l>oLRmtzEn&t_;2zZ`LL~9-||8!8eO); z1&jxIoA=4O!Gs1nMZ*m-|DgYlj02xh3H2j(eiN4mX=rK+ z>$s+;&;oONCc(Tf>I01D^;YSJ0_aomlCC{t+x|E&dAcY1!lyalD6wBkSttW>&r5f3 zRdE{>-;T_b!O221hn`7f-MLlneU^K~pv=DeytAR3|CTRiY}|~LL+S23F|Y=Kl=@FJ z$Ba4P@PXGJ?D@r6Z{LCxVYU72cS)D6@rJ`Pu7MN+v=`osIq`#ShJ%mh6ZvHGsBs@s z1+o|4J(T(R{rffI7$l#rNS1EE#y%X?qTTe6r);poZ;+sEfei!sU2vJ<>h1)~uB`J!W| z)prk#&8xp(0I=PN0weyiXyRGUQax9Tr{_H#s2M$5R-`D+7+87lo}OYCN0F z1D$u#Wn0xQ%tDZsOcrXtM|I?aM9sG#lcJQ8?5CRUdEfP5OgrnNtNq?U5it)pX7kiK^Ba(d{ZNbEpcgN5GMWxX6 z?LRL-3)Z62&zY_$#K#=n1QsNe1hM&fNLz(7CaFs3nWeU-Ib!&r7``#9Gg&CrX|bb$ zf|(WK@p7Yub8!5-4TrCg(LBF%@$P}0Ig{sqKh#x0;Rd4h#OF6&yqMDz;{#{-?w%W} zwf-{$w$!1h(23sKHhJ>;g?MoEZi_@Okt0=o6uJdZ%=Yt3{=>MbQFrz<`n%hAKg1G) zqcivFvv*_vk|8a2Uo@YK+O*F*1OSU(9%aX-@BCf@XXhX$#opdDzLxs6_vw4tD1E>7ZtvxTsR7=yjADNN$8Cog>v&-sh5o^h z#ve^9sCtL_xhC}X7nFefhp_DIY@N4KSBA}AmPV&5J-Oq>${A}-;5gk^Axkz zc5Uh8z9d84*jSlLYs7^aLhK!~>4F#EkJ5&ocNI+PpPrxc^~;3CN*2FnzyK}#@4F9T z#V*21hv%WXztU(9k&k1_^YexHB|6Y2_7z3`6$( z%m6d)?LOF!A?kLo)cu1ydHR#-K?_M$J=RBGq!lq#&}s9WA)l*rw*ES>;-env=aU|e zqIL`gT)1tj_oy}ycnL=p08jzlwEZE=;}7StNWT91wP)pt?@ga)&6zrW^+GnCWAz36 zyr^H$AN31udz>NQy7$>iB}m*(+_4qU@>X|6*B;%&7p;-dOucd)GiHqQw3Aa(%Y>mM zbJNoOb{w*=eFnTR)uz7pi z35WEbb5YtsX1ut3(Q2&SeqwS9B8`s7tARbEy6yH}W3%$D?k3L|ml@~akN)(*0B4$o z!6UzYuu!0DRr>YJ5vWBB1}L+^-6 z_YpJ}cxy&5665nr(89dc*g36|kx_*Kz)h-N+=-Np?0vBmdh=CPMOp&eiegO1>oThM zSg{;M_DA{{pC^a#PGk*Do6g%?RJ*Q?j*xLusf#Z-7I8zTHC);vT&OS3&mR1N{=B-? zWX)F(r;I|D6$oa+rB7poU#id$_o8=!Gc5)SM9RW>bFXB-VL$^&7ppyWnm@yA-{iR7 z>-3cZOvlbn;K2i(_^e8{K`6@l5+^!rcH%u3(v{nog0w0qO4aD*D{6{D5J11Ni!+tg z)V8B%!N=VgTiL(u^V1P3qhcv{+O7ZR*%(Mticw7>UA6=0B&xS>A2tatF+7*=Q6!l` z65_Y&wZ9XQS#may0-{Fa$B6UpQxNpy8>sQOq;PKT)4vuo1%h&&tj+DhlJn)l(W6JV z40QV2sxfWU7?{{IHp`>D01tJ}R6lsIYt$pS&D))Pxx{|UCi8hyZ4@Tm>>vSexRq@m z7uz>9PC2dUoxSfr!bYQxbOStG_?S14eVK`+EwdSYS%ED3o*t>=MtSV zUj~(HB+|CiCXB83`O2q$@gH>>p#eUzVaUQM1hmV(VZ1 z(L&p?A|N@OGkTKYKN3Rg4=5p@6wSdD60KayK;*?DAP2bxb@;8i$-Re8xEUrGDZOej z;}ETlhkrv}MB8UxvARzkH7`qiesu?Q$i5!SIEdKyW3NnK#dQy9y}7I3E;vXsBYMDV z%>R{u*d}YE{R!+_W)o%C^dPZ}$^;HK+vFr&w_leo8ky}u<^n80{;dHuOl|j%uVp5q z>k8od-(W-F|(vUjTnSUeak85Oz#Kd5@+hV zoe8}ck6=>ZiIiLh`W13PEOR{vbr9`wlHj1@GfhWPxh)5 zd`b0T_7ah$q^H!x>y%9N<6%%n9=V6&XnwtSI&9o}m-V`~n-%6&Tn9e84jtXO1(4^P zd>`@z8Rh)?`CNVlUG2SVSXg+?SXr;@7@5=;3;8h>I%-+PMWp{@aUmIGA_Vig*^m=6cwF z9+u9sEIyO4h%D*Vi_iyNUI@!df_O#29nG`Evr1byJBJ1sJ!h`Ow>sbb!D^`3hWVVA zdWa1e28_8gFIFr=AT&o?$Dkb}i5YE?F)&Kb^{=XyC8CL-3qmqcikSEKv0WXsNlFF~62 zJ@*o`u}8LX!d9MipH*%+>i-nxjxr>8=ecT)UCRV1yL%u7Lq^^*UG_?1=J_SM3v5)7 zLQ!(CXWD>v}Qj!T2k#LYGIS2rdjlFuI#;}{Kg1T7C zq2yKJaEZ&c-Qz zPm(Yioo;c+FL}deQJpswu$i6+K@HW0!AP6F7XT_ax1|3s!t}+aCBt)gCbo9z&@{74 z%|0p@w@kp7lvHRfr=p`j(#G=^sEEkn6Lzk16>Kom@9Xfa^z%fUA(}Nly$iR;A)mP= zvW`WJGDi>y|Ddq5Y>}@@hPkPTm!5HL&;E;9+*83nY~(6Ec9)BSpg0y&_2zI!nPH#CfOR|UkN|}cJo&rK^?$t4Z3mi?hf5`!1;8F@) zg7c;W=bzr3+1>hitm}t|JaNRNu>tR7TK4Y(^L%J;H2hn_Q*${9!kWl>6H&V@7xXPD zF0#GvC=Kso$gkJy@HRKI%7vbO#Ta&;LxsYQdJ`eM_23r)yE;ig8{uCp?;Frk6sF23 zv=YCSUYx%-osUn~#Ue_8%(-21MtC$lsmn{&ESC3=Uay0N*LP=!h7n}yJeKzrAQ(u? zU|EyfWB-v5nwz98YXx}gH=1uuW%vIcgD=tAYej0mShAf;qDL&9g9sPwcl^@4c zh$=qI>nqZ_m@Dv^e(R;0vkkbN>nCx6Ax%b4nix4WcErsj>5Dg0x|*b@!NI`<%+VV~wS53EUNq-U>&{}`jc=hEf0)U1)0t&qf^Xp5lHGY5 zl;agE#pnYr&$F7POf+|*y&68>X*+r?Dt0ujf?;gEhe@F7vU7vZ!r2h)#<2%_MVlr2 zLZo%-96wtrSncxV%ObcUHJ|ydqVJK)PYr-o7w%^Cl(~Mp%U&`XC3s`MFvH2WPtcoq zVG$8pDoO|`JI@ZqTh$ooQg(cX-XADJE#fDrvoxxPu@E?eEUBwS`=w|RnJ~4OWk4$+ z^4Oq`Md+)IOUTf^{@Bxm?{PWp%DdutIm&6}MSGA9A>L0#Gt?ewY^5LDGuk96oG2Pk zhyy`naYHSrv^jdv(1oN2tn}O8T5# zY}(0GqLAxJWhQ&1{G%u1Ti_-l7UJ&x)yOYnSB-ACbb192{bFw&_{OGnDuMc^s)-r(9#s8S_xbvT#&Ob1$IFy^D6zA z5QurJS#L)X8$r8yI)5z~Lbq<+3K@`Isok_?Y?Wy^BCyVy9jxYM4HtS7t3U&!7mV%Y0bW%53hgw>6EBo(ho|i3UzgpuWu6;7*T_u zls>bO7ttBJjGg1_+gn`$fImo1tGmAbRlyc0mw6fro8u;FnKYE|+sym)cznGV(+JU3 z3x5bfDPNl^>^n0fOdztG|JE#JAx$O{ToHBi3}h#`_f zKYV#+Ibd~3?UY->NU_QxQg`ROO2DSwpnh{EX#hrnFBl|l8P+INpEn0cn%vgi*JX3d zI1#(Egf;3GlKJ1?@1f` zde)Vg;#;tQGdp^-0jA_9`|Rgt*K~p%Ke;?|r?u8VX^~soN#bYD`PTK(>CvDyx*&%D4I2iz93 zKY5}1!)Yw6Pn<_?c14rIxqGVv7B+S>s-HoqN*jToRs~(m%M-A0%IA1_lhPO)4lS}A zyOkK5r+bS?kL+?O5-9k*XC~Cy-El0U z`h`<%!`1w5`Zu3AMLq4eUaqo#lwXo7Y{Hsz&aI1fP`Kf#5ISh>yv4h%HQU$U9vVe_ zs?!j=^!p7#X!G>Nh2>f7<>Q{9_;|LNIss~mGIX`GRW3nw$%f33F zwz0B09W%7~{aC*QY13~BhR-v_87-V&&L+?+KZ>t2e0$H>3brw?mcz(l^E7P}*W|DJ zDJxoO=|;A(ZwLI07Xy;K|3>Drf-0HW-eUbn_U_8zcr zj#e?5JygnnB7;D=u5NW7XrjKG4ca%ns;q2x&pVuO3L&faJaa6fq;cZ3OmJSuy?(hy zZJ)+b+RiZppMYsrjbuNNte9?hf+>AsG%b-FY`M1(#aCs6#^*da*}T(aNoFSvGX?eq z;Du3lZwrGa0vz8C!ac9ow+`rS?JwcMrFBVR!DKDE#e%3BJ&gc*WJ3K1Np*Z@v)?&l z>LkY^WG)6||1oxS_%mk-#09=9j81wv;1F(^tnDb`8`X^cNTTAkZwU|fH=dlPuh)(i z-J`pY*#@f-q!qzwp0o4~>#BOR3Ev8uM@e<W*WlgQWF zHJ?1y`KRlZbtZXposH%@4XIQVb3-oEWJfE<0?^Ztek(uw<;hvQ+$&)4OTtHD3`D0^ z5OnDCwoJQOzp7c|fXk5L(a)Am(i7P75<3*#u z22>Y?4LInh_MA=n6>Iad7pf33d{qRp7Tmu-xv9h&3WK-8L8=t0()#BQYUMy*RN1A7 ziQ=@JI>LiX25zl`-C-OTvK|XF4_SulFgF6n3^WH{-X1UD|ej?(v5r@0EUe%hs)( zg@B_|d0GPYAGQ$)1gt{xFV0d0l+>gxO-d=tC8u)>X$TbQIzEpTl<5^doMW{ajJ%4T zL@_3&)$Zj2yC`)40W%K49`9=;emjO}9~~Ly9p?Y=E{lb$ZB@9N;%kdy#~Cwb)G}xg zeT|ZaZG_rg287F_>a;yf{Ygwwir=h#U8ZcyZ2TWZFL>YF&uGD`E<0>PNX;#;V)DB} z-oe=AP)u4^lRwN2w|S~LcyNxv-NjzrIG)8cI4a?&bLpAq@5T=u?Ag902(U5svv#E; z_R>_1A@8CQi4L&@ywsOUpU68D5zw*2ZDmcgwD%82s)`Ez)SV_}e?pa=l@An#sZAXc ze<$Clr@vxx;?SdDUVHZJnbcb&Oqe;AgMn{DH^WClhxX8Wy{ri{+RD;GX4dDm8BK*ET%0L$#7^T$^o8bAmzU0P{yt~B|JAEwvTg_U z(2xDrv470^;_~zl7rAQ0gf4{>@roIu;iM_s$e zKHFU|+bLw-I{7!YYPZ;U_Zn!9jWyDU3QHJZkf3q_8RgLYE)Irg&}Uj%si%&&t^SUsty&D^f&H4eiNr`9!}D5kVroR8{h-`Xu_4HN zNX@+Yp2&AGJD|SoN3NAQDs9QAkBt}F)|;XdWlVs`TMslnSTET> zUmR+o&M0ue`_D{Ra`XE2{_I5UtFskwWkq0kXz16S+}APyVCDSdHuG+@CmWHwGpS3u zO!}=b=dxclWu3%t%MDEfw(oy4tH%8aik#Q4uFPH9pUs$&m;CE}6d{Pf^ypP$fgJ zE{I40n|td2#dU?-i?Nin>%bEnHU$?m4J zJaeixj2{s4BD#sPCgC~Yf@##bKM{WyTkXZYzJ16I8$Pn=?E=P!`P{85Q=>udhOOUR^EZVX&j3Z!i=c|~VoLb;{}p6>xz^FdHEx>7aI($G zSRvdM?J+_3J0wrMm8I#gIMsRGxyA?^odj(cn}(BVvxr)^fXa~RHl<5_ZS4XHV*yFg z$YSF{_~;z?ID;maeQ*o$pzNLE8#$NyPc44%Z<^N3h8E#3px*V`g_(s}AmW4on9Q6R z{^-D?&T1KNFH`>cemSl!qL(QtJg1&8DARHoR*Hw9u=t{g{1vYN$Sg|3|A8*3j2ut* zZT+xj-Ma6Qy~%PGiT>!r9n_Rdbk_yFe{i^;01(vrS}vTtLcwL`iG)#y)-}~@PYSxh z${7s#0CFSWOTWNxbo0H}rg{QIda>|XjB$V#lmq`N0Dm!;G4M}He|Z4p;%iesG*Uoj zLWO%OFr*LrPj_H`1i4~1RjTwqb zL_YA9ntj}mVU_w&1TXTMLA7ol9?uvgVXTR~V$23DHj6_cF0?Fu_bRQdip8Ey_TOa~ z{k{l3dgpzCs#8AreJZC2qb9o^xI&Vd-SU&6-wo$+C-Jqt_1S|-CM$BX|9}DFw#IEP zr8lse4#JVVC-5~+>OiRx6Y(2(BICGx@eUO#QbJjaxM`2fCbDjeibD1n0FN78c*beI zN}?jJoC9Opf-tqXtXvBPTb}fGag#0QU2cHJf@w38r8S10a13#dpPpfU2eo>VcX|*1 z)3^UHSgJ!OmB!EUvj6H3y>hifNG)K;r=2`s%i24Vsykz1r^3TOav*45)w_b*&%yS z?YUmvFlvSLDxxs2&FNjoXUUjz#T1Ajoc<{pV~NevP1rC-Eky0VUA%n$pYL);j(6QJ zF%DFQYFwzYWz>HYCJ=Rwf=ydyQs&#MOP-se66g${(Lc05((N-LJD`nRFRw9xC5xhK zdIkI&{F$AcF|@px$PH^MIqz3yp&t?(cWKSH0V2SaZTKxcOz_U>v@^;1a31;@C4{m7 zkVlJgk8p#O^aWHcLIIh&cL~o2yCjA_AfGWxT|zp7&+0`2-KQzI&Lz*r(o)#sp8qFw z|Nh<#Alb=lpPB#S@#7gyarf91ZZMcu=un~iT@{b9EGvIklesi}S1#jSc}}02)}Za% zwiO~A`%2j0Jnku*k2S13;$UDat%+2rKK~M4py|KI2u_84N??i*06in8R_e72#F%uV z#9=?bMQn_t2dxGyLm1J8gY_C8SnA_18oG`ekQctCDbR=>(u1{slDd!jwml`6jiQW5 zxWbh)$ntAz{WX^sMFGxsE4CAEZLjF(g8@0YeZ_@E#S_7h9f8P*e*(v@{d}fl zMPlUeCvDtX^=bjlGeMOoNg{xYRsQwI z4-`TnAt8l*>_J>;EL9bc<;QC|T`Mp@2|NlWkFdK(A3I)KL83no09>lG zYk%w2x@WXp{#VQLtzfH^wy=L}f;xUE49(>=*~SO$Qi)e88FvzRaQKr300)6ggZg>z zT4sC%^~mPvIMW{%;TFf}{~?Bi>Y^&Ik`%n08y}(Nqs~>-2G$o1UA;j1WTS!uf7Ekv zKZ;$L6m`xN(NmDCz9hF#aAo-jV<}0=h;#WUELIEdj4$NM@BuEwN30Ekdn}t_0v%R4DAKsfo>jvn{);U-v#2 zj|@g!?PV)gSi(nok9Zbv7M1ED^XZGqGRyq*+42z*s#8#-`HCO=J#PC#zX8{c^}mzg z3_#UgSkP!Eg)O1ez-_a6Z%k%F>F3uXEKwY~;)xlL;k`@#9S-*GQzj5vh(*T)MaCUv7#W8*r%obptP+q9YdfBC42B z!r4Kh2RTh~*s;*>7=Z?UO^1+*isOgwl}Ny?I5ZtiKJIPiacAIf1}%P;@o{!!~> z&wd_8UD6kD%hzo>5So_EbObjD4Tq#iO5D@3oM>B}jV!!Kh=ul0EG*W4{N6$VQPcL{ zhd4AsMs;#`cNfu8>^YrDqwfHdzmYxN>+5J5Rj@InM0ZQq3x~_CGX?Q_bN}UfT@UTj z-F3Jga^=Fa_z;<3nm^!?mZ)T8CT)2OULJ|IB?XT#VWQlrUf|^-19W|Z zU7VKZ*MD3FREn78rdF|+@ZV{e=znF7&Yd@Do1-OqiJb8`MwBRQ3RsU~2osSwA)`uv z-hHm!%^#}K`_#A;zX8$oD#(trjvU&Q~vkN z4CBn3f?Rkrg}evEXRlEKgAN=#Sbf<(|JgyMimMcGxPPFRE`Vm-qi$4@Iidov} z8^0HsNgs6hT@i$F_NLZ##?bWcu~e+RvmWa@hWz)FWS_)8y(_+-?=fU*0Z{0&wm)kt zkI&e3LOuTLKRLHkDn;~3+^|+E%X*&$I%9QDJa}7IcP!#Yh=J3`BrbqQ?j*9hXxyvP z95Y2@NLL7G&WdBR)ZdM@vCrIX&tC{X!hv(=`N2Tu)I*?_=_uItd|djjwpM$6@%Km!rwsU> z+R%_TYqn)9MJM;$;QxCTu7z~BDQ=TyTUK!2y=$0!>4u4MHMbMzSioxS=PY@?H_cu> z4otg_kym|%mH$e}Xf$|2yqC10^Xh*eA#ilcNX7?yCc(4gW&w&`IZg_3mB4F0WSD%o zjJFQ+K-`(9(js_u$ANx^-ZEVao^T=ZG<}NWy0v?sytigw@JVS@A9V7^0zzj~90h=i zIWb8@wpqxKQyB>la;BB8xx~|}WG&J<=h!Z#2dHbT0kRWQBlkP`@O{?rV>e}**6lrd zhs}%qGci?YnkSM!);@Yf$Kt+S!^|9(kMpXhW`tyLpmtW5WOICD-UT_oPjn1{_ga;L zza@5PV)e>SM97NyQZ}Aqto-R;aO@8iBwQ{%(qX%W&`t-@_EWvV7NhF@yUfp6u+e$9 zxQMq#-w#v66*Ux*lJ#A!=HI`2Rc(ZoY9G0gUAN~@@^G7jwy=sY_R;5VxYuOjRsUMq(LfZ$ZkjwV2%|DbHW(s;b{>;>}qD_2s3@Bce#;bh=_vO zbNq835;fX$J}1Zi1_aPRITCgLYCg7K~R;5L+=stYSN`gCCuNdul_17@~_#q zuU=v9i}Zfk_}}RZ9e7DMyP<#eNQ38C>^yxyLLQNWmfgPYvQ|mA7TtK@Lbly{DP-AV zwo?ZnA))Wk`&q@}d*5&@gG}GV-8WIZqgxfqLB3;Qj2&Jmj{f=uPaJ9q(a`yRohBSY z;U=VZw?2K^g_K>g4j(@J%-uv=&55-|PRjE4#XYS4XH^o14)H?AT!thjU$5EQXuGY< zs5V|Hb${ZrQ57K6||2$4*Z0sCugLRr$ zoaZ&#uZw?}G9M^sN7ZcDr!Ta`IlbbnqZX3!2C6Zco8dZMR*COp&Gr3LZ5O0m@yMs^z;L|htJ&_D2mQ5KyQ2E{yg2it=a_1?j zYwtGn`pp~zI_pVF!%Bd?ws1UsJGre_Wm0H~tMZ>(Ht?l?5uLiF4Qzil{0qVdH`QHb zDy~KmpBis_W9WjTm>21OeD^5sxc$h_*p)6AqF-PR?UM>*WP1)P=@d&SSYqF>Io*YH zXRp0h;5312Mn6SWo#-+_YwQm;euV>v0wNN$56~u+4OjLT#UM87HMf97vIA?K-R|ml zFaztfJACTP$fM09xUMk8hx+u@Z^}L&y}n)kbav-RyyLf@E~d%7?C?{G0@^ut0g2NQ zi#GNOAG*){5%+r10|-_6!He~K1L&M&fN+3pbKfnrvXVr2!*vEekl*<-Z}kK3Ba^gq zIjhdSFwvClD0gr1X60zJCB5V%?z*1C!I-+(FY_gAQm+l47-a6-+{V9Oz5RA z-L89l$!_a;?fu*=xI5Zo!=$_#UoS5&$H%tq+x@W_r&V%FeC*pE;m82q04UY(KNV;EqnT~02L+di8vQpkka#p{qsU%N!kcs1!%dEd2JEny z0Sxr!Ga|N;`yWG(r~u6j&dZOB@jWzDsIH4>-@`utZT>klL>2xlm}`;nt(+TM(#e%! z0-KZ;Pv|?E*5CP^QCA6CBPsH+V-LfLr2Ek_347fsK#QCf8+(BaKiIy>Mm9ad^mh<5 zJ6s3O9(8i8) zy|JDWJTvU*RHukqf~jntLKyewYXZf$GsBY#y7QaN&sC zdIqfMJ~>GOFP8^QP3dRJfQ{)#{#Y@Q)7(al*&?>$X7O=_sj%Xht41_i9vy=S`OgZw_@1=sxDOifKw z*h_@JH>*_^T^Pz0MxJ*7CTwhYK1<)fKS+xh-ixGfH}ayX8IBz|&y*vlPc;?ew6_Aw zr_gqDZ1!WLy1C+?Wg-op@>UFV5KNzFLBRlf4w_F)y>Q-Eo|@b5e|iI@XMEr30`C`i4&3 z3!g%8>7nv=i=8J%BM~4$!00Z7z-lYCh}IX;KPb6E5yYfeZBwFHO!F|_TizbU!#0%y zNI;J;0X`%f?nul~J7>F9jk83=J<{|fQ$L9$X7VEkRZXvL*MD>nN(&nf#I?WfvEmcW zqxY-tG%b&lG(qGG2G4nO?}`Y9O4_(2S1(F_x>Ds=RKy={|HY8x znB(iP%09L1bc{62@w^}Zammgs*XrV*^gwvLfeY`gnBATUC-*F1E-T;1%c^(AjNKoM*V=0oT$zVwrCz4YUF4>+_ZrYtBG82{DH>e%rb|!}-ep z2~`wBigKT`H&nM;AMTgFEg_*??^pUMPNkfS*-9%Zntdl2#7iH(x+6aN#Keh2TukZ_ zZLO}(R$6T|QMG%R2l6WICp&oqdlll60K?C$JX!U`@@R#@NXv$DyPJZ`m)^O3n}r23 zh3$!5|DMtY*=4mDRqqVWv2}ki%iGDOT>xxMvL|`6a%$^x_KDFZAMNc2AT=%}V3HtQp?D1Dd zNmDakGp>DJrv1hJz!)BndjAjD0Il8n%0%1kOxpi`#On0wrrA6TFhZ(t4bKAlc*}Wj zz>d?plJwZ5 z+3l+49sP7nX+{a%Fd3|)}_)4IZIkB(c+bGv8RrMZ=22;gQ_y#Mz%`y?B`cK$^9A_M?f zIC@nUocQce;Y_D;V8qt(uvWbXs!Jl|+@0DoCJJ;GC2H1v(_i~SLqkvNNPuO$YHgrg z31_5I%ej-T*%sxSK7C@1i*-zoSKvHh#Vv~p=Z`SP@LiW*l2Vqv@Xfn-gVv|LqH`px zx+vuqfZ^Iwla&Wn*OYFVKk_f)*tNfAaL{Rd$-CP8>&q^_&DY$R62Bi)It|tt8G7|) z27OCsV6N$#Eb`}EsBFepvRy@YJ6lT5TGQ1!#)hZ=*Ee;4+7E4Go+hOCWu+drrQD*b zxoLCHA3TmTy(wy{zsi)EURMFJQYNR>5^|hLv38^7R3)hxUW*PMAlPg@Na@P!%qoMLHXc2FgJ!n$JR3n-;f2 zdHwmT>wj?uJ^OE8jMZZgsEKZiM=+NSlAj5(+twuv7e=gm3g88LPtH-g_Yc(?E@yCd z@%uZm6Dq^MqbE9dKUYFw?{Sp#!fOd>1g>pdeHA=<`)si)d~egCifN1KLa8-RT7&0_&b%G!UMnXfB6+sE6XK(+lb*s8 zT2rcidHt_{D8hYf6+cyG4s5r{iBNR;DdH^IcG}t98K3OK+~GBbD@xN6`-dT zf-%yf&m0 zoxK)zpl>tZX!I9~m+JM$cKm0dp{lKXZcFYhf4qks~8{+@KK+VyGKhhv%6FczRX5?WTP zg){7uMX>n;rG1uV%CETnr6qJ)$KlIW6o#H6dohbKq$SkZZ%Vs!;dgVhTAv@^zL}#w zAf0bBedfSn!zKs++j|B#G=T`K#wE|Zfmc*QU6Ska_Hz)`v5Y$8i5k%#!`8Rf3T4%} z9{Via+xc{2N1l2dXlz+iUu6K&=iXGZiFZk60$>c5#PGszd=Dg0${RKhnRQMF>>-@1 zhxuYgQ~B<|gRYRvA(5MiS{Zd%yO))<(PjayiO|9y9WvAZW-&Ff>EN$#-wb3mg}(fX zUR=x>w;&x>h0nrbaI;>m&$*aYmt4Bwfig~v1^x)D))Yfcf9J9A&utpBt}#QI4zhJ& z%z*>)L@y1WVUW=66?^NV$yuQu{}wFM5h~dg*Eu(Px4N^zn@^NYIw8qqt==QuYE4H{ zf-*Zo<&Qgs_g!gkoua*!<@uWQF!e1jPP#*?bP$%gh+4zm&bC2gCnwLakP$&Gsrjkt zRU)dsVb~E{iy!%1Q88mINTt|SCe6$~&>cDLSfgKrCiS~3h@G1z*&cMsgZy45= zuYCSMsCLT(7*V%bm|)8O?h~OEj4XB<@)Kj>IY)%^reCTct5j=9du{KeyR9^~D7bTH z+k97`>oGAEuC|V{CSn3WQ`joR&xxM8#D{zi`qj`^@7N1ht0qb>8q)cOJ}DP_sHogj z>C(Jg1AV2B&kUJvC*3C`OEbedb%#b~ZhM@bQrHcXF`U7HwdG+^MXpxE!9vjXjjnbW zDA_phBts={l;e9YW#UI_yXnzSVV1T~!}WQ`^9H`wK3A38Zaw#TP;xCw@zo;(2jN<) zA@BK!v!<`AR%CM!O;q+TJv?|KsR~8TuUr26@7vXfBKRR*;}a*kQ-(TJ`u5u; zZ$5w4tqDy}v2kd%x8d6l7sPF4hPw3#hdK&m$|gMj!C%VJ6rBE)^2I*2_SpOT5@y!w zG4vC%{~mp*vQY*%;!gA47G_)fh+IDm3J58+3*Q=7gl^JIg_0Ce9+2YBffZ?5R?f8B zg{y1fBIx(0kJ>oDv%`!Nk`s(1AXaZ`>2&(DXMx4b)BpDE8^sY#^x4eQp@;empi2A8~5eC7zK9G8#@((&D$zYq(^FUtc(UipqG~GCQYN@S=c&edi)d82Gj(7l4~9#p zm2~i{jimpv<%LhU=c#DeC0Ox=$)B+mD&2cC8avGGuhBggt0hb=I0Oil{-P>vtAz}WwtVxkGC$Xw-8HhN2@$r)O$SeXm1p*Pu7wOsVE*aJ&>{*sdxoBE zeDF6I(h7yYp`(k8{&_}L4iSNEEZjKTGPTX@YnQD{S+si%l3?sLjZAJIG<9V7W~#)V zH6O0Yqi7K6BwFqLL$zl(?lJv0+>j3z{pC=0l?;Vuw>yt<%h95>0_eC0CraJJ9p?|r z87(__Rye6XIS$s)NOnxNKu0@f@VK775ISB9degQefjCnnFNAV7tq*5(9Hr z$m8{jTNdPGS;9`brey>Pl`q_$wcN$Ja{h-l$|$a>SRkh^+cNnP-ije^-#^^jK}od3 zLV*aoJs_4$L{mh3;dy_f-g3`yz7-?I8v@B{*)i@ZGcMnKKB`Z?+q}%p6uuWdMb1hg z@q=jegw46VbnCN^D>U!Eys){Q>~b&B317pZ!fkR!A`UQqq@gr6wge8+{=_&5n|WDE zSlO!F(CJW%)#u*=HOh~v(Cgm+EO%WtdEQWgc8Y(JiD9eXJ09cKQ^w%T8$LMpX|I*U zqcAC~6E|rX`n0l39~%44z&+Ts2y@fHEh1}zPb6Jl8z@pcfT(Ua^z}+{Goe6z(Bz)K z>EIPnl<8XUiK?S95+N9OC6|4DWnQ?7Lt)W93-ghprTl#9;>A0Mf=vmz7Ku<)j=OQ{ zq&j_XbQAN4Tf#-mg$WP!j76lWUE`5*9W#`Z&Ng`~O8MV5h&-xj64_5E; z)oeix01KA?w{2)u1C=*IOw7U}8mdED)@V%YB?rm^=Id93VXVKUf-VU(z<8Ns1HlmK zT^=wQN{Dd4io}eb5>-EWwHRBqsPGx4XDEuL?>i6}+{cGmR>!8vQSBEHTCGnSaZ~G_ z-d{ITQquoZs}vi6BX6B?Q*z?+xU$H#A^{Xv4hP^@kZUoLuH>8c>e?HE*-?^a7g5m( z%3k}!H(_ilfIAjM74k!ZS&gTE`f#$7kXG_kL_$QO?^oKtN&b%eJmEhSq%1Jf03$%xkflW!?A)3LpeqxS7Lo-D=nUOz zvk)NNJ(RwT%UaUldTgq8;n@m)RBiH&1Sm>@ZJ8|ILLtVkn0JZ1k;E*pIR#Bt?ZDM z)i0GX6@a5DrdAnsMhT603dZc}oj|YD@Z!o2++GPPWny>m+uQ4;`Y1wARV)9`wx(h} zn01Gb z#5U;M61|$(-F_3BBO&4WT=DH#8A*^*R)lCzL|>Ji z6L{mzADoj(+{^X(v#@P`x_}7PZSic`AIo#@)-=gqKPS;aLSmAY`P7BLSS8>q4!w!S zI^;gk#pKYv6F}zIfc~g{wM4l){e*w+VuF|U?-KG<_J5u3@ggzgwER7ZS8w04m4XS( z%l$0e7~h^fecBt_u;-5)2)(tHT837LMSo1ZU* z8;w2vI6S57F5biVsP3~Slqvt9=NC!DUnEfFR~JU&NHCbsQtK3T&CpEtg0)?6?NaE{ z=+;&SQ!o1jNzYMU z^ig&F24ACdBc*B%n?z3BQBCZKeN*N+mGBcY41G)dzOSR{@6w}#p@^lkK4sBqGm620 znWqwuJI^6uudhwa0v^-mx!KG4}0GK2DuX!6GcBF&FxAJP1w6iW7;#7d=(- z_9AwIl&4cVHLo4`R?;luzaj4dm%qjY0~Ec^O;y^#jk)~ICA=PG8{wQrO*sZeg7eY9c|VbF4;;lO9b0FQX31*K<*R|Lef}z&~Ay>*>X* zZwb_9z1m2kZN-f5KYrxY%wfjlY2qTp&|CzPVon-Lr3e2gUL_H&(MW}sfz`o7w?Y19; zWQZe5mfRX6ZcQS$`$Jw=KICLlv`eTdVF7(6U7md zkv}IT$Wl)w4ViCwycqOD9E#YcsN4+q+DCV^w_e7HHS&~E>#n_DuHU+KE2GC=IV(*`tN%@1t_;qMI6Stol(uu7cfy#Ky+ogAFEGWD**8M9&g4Uv7N)#lhVCUhttX) zdj9(seoQ7kiPD?QG+9ooc&px;(w0u0o`yL@P^FL$uFOGmosMmgiLd7lHPW2#Y3c%v zZ3bQTx)$0cCP&(Lhg!+mAJ~6Ionf=H&!nf9A_P}t7I8J|fU?zq@s_;9|9NGv|MUJR z6Ss~`-b2sPee{9DLVW~QK)0nibbm!h`)OYo7%Kis)E0B-k(w!*-v-x{@Z^xLA5XE;|Qsfsr6nZlal=-rj%;j zx_2+i#anaWkHPA4YTo?kYz}k81&OoOco6WNo0;qBo`+^*NBpgMg;hJiMz;N$gJz!h z%{C=efs+Owl1KuIy}aGSY_18ujqPxP14fSQ1t8K*+v4tT*vJ#Njk9D3PwDK?pHxlr zz`HIh#5yWk8LBeK~S3ON<5x0f1%OW6)x zDt|-Y{_kfhKTnjWDKpX%y`ZZ4TV43ImQ6r#85tQF#@T+?sy6kO4WeUX=z1w&8YuA% zzc2ooBKBwI#l^0RZ#BLg2RD&`bl3my0y~eH6TAEzr{{7+9vEmV-(q=!u5)OJ6c-v9r9TNE0!%u10&TiIO=qN31I zR)aR7tdfyRDxr)dqZEZul9k!O(NxIFJQXs_D&v2D@%jF4xBu<-y?wvmb531d*Y$pl z=ku|i>8g1dZX{l~Z%pv5N$o^1z})vQay!hb=D~He=`w6h&%BIh=mL9 z|6KXFewh)vzOEOBr(5wWDYb7MBYpG{I`q%B`0k@Ts7+?f;nF2_B`UJ10dfg<-CRn5 zjU|m!*Qges*2x-ljiDMyoM2H0^Wg2Ev)11DX>o*__rcDh0QHLkMz{&F zr~-kHAXiSqGDd&-FmClh{8Ji-a2n>YG(Zjaa4-E5n6s6#K!GhSTXR_KzK-GJ54n zsmhPT#zBQ=O4R7|!cSE_C_SCtjfPp{rkT+HXVvrKo~MhI#=L_&g$ zia}JbLsrTcF5NvoXEHOf?1(U)jO5z}gsh0S#kcyM?W-0ZGE7@rI}N@iT=u77I>?)t zSHkCw79SE)O`h*W5H*fK*z$e*>Moy|=}wsae0(3_vJzpAFhFLzr2QmKs7>H3jWjgI z1J%^lL$mLPhnd!R($Naq#fA_&@zKtKg!nM=-DcuM*-^))3*^lJvyntfjk8k|5J<@^ zwfb3Ioi%|tQAlMs#l>kMb5n=4cHg*T2to@9dK+tWh&Nagh!pIiki&^rBF)@@dRh{- zrtt!yYuhw(lEKFL$JqE5(fz;Mx7o@K^0O|=v7A2-aBzctyLQFI z-wC5vepm;iu$K_+c$UyszJUZLn&;r6!QLpw&rIYqc33NNO$U+CJb?RV;K-3l$zwy4 zuw{)8KecsRwtVW1!Q#jq_%Ahr15b)cuHy-t20ltO3)F(OHEkuUylU)nWE2+QTZu_> z5J~M+j1%Af3iN zD-Ly%vR^jBcuqKaP9QcsDH3c{U}W@OFL$+bbaWJ%dM|J7+A?>Pm_nQT)w5+_aB#4g zBSVf=0VktR?`S}*n8lrb3wtKK;FZUZ32hXRDh1(nu7p1y0XZImQ-y2St_d8K@yYMr z4GaD)RQv~h?;Jt6AA!FTL9cGjo3DKN)z68dNe8Vo4Q(-Ei*;mlF)rNODh-Gvzmc5y zjHhzcaF3Bs7(d;37ph(cwFn`38#xJN|ap z?tN86{}o}w{`mTij<{69W#^YAt56ZBCtqFLhX_<~Uop6t%6#gj)vH8i+r3?)Gt6u2 zd}dK!P8i`H;`D-U1PIy`)G3%XBBbswOLiyxTFy*vwFcWOrjtZczQE*A(NP*$mrrT@ zRRw1)OhpTO??4hM>Joeub#G?@+e6u62(+9q;RU-&j5=R9XW;TFA838($K=o?8nzts z93?`}oO0~D@oU9YvVw`MZQ8s3ZMB}y?SXGPeQOU7C57~(7zsp&CtjnBiSBZKRQ~wE zKut|Sd7Cm-4XAPO*&Yuwg~>L4nXi~#D?SmGpUTh2IeS&9I=%jr(yiwhBZ}%|rO62Y z!<}HX4&FM1fx|z4Z{I+|wJ@ zaK?MClMOvOHZX33f|ji?;==X*`7dl2mzSRoR=pB4xy3$=UuZ(fHU9lH$s22bwpUw; z6q93I#j6M#hHMz_jikpb2w`D-?$%=k*5G2gOTV+IZX_A&P{l7uzp*BkM!I6BLfwNQ z`iqP>eDfTG##vJmA{PW&Eq(WpK^dl_MjifcGR!fJCU=14lchBu%;eEQ0hm5KHf=*} z$?vW?2h1V%>we!q6VTr-Ol1+$oogI9rmZLDUUXS3w1iRnOy866;p%UjGZeP$-O;L7 z;sU-Jiq~)ObLTd64^|=r*&`5dC#Ds%)^UvD(pp}#hp&MDpnluAfiqp2K!m5~*rYu; zos)mwzuCPYbhGYGmkqQ+XwhpLKHp(cm8oZu)xn$S@U<_nuXlahhL9{nlg7wBs4PgO zH)}=U+n*LPKlE*5qvaHT28blNn5r(~9(aOTxe zLN`_}9DK}%gOK|~U3rfr?K>J}4oa=*Lw!YS!R>qNoRjF) zXq=k9>^Ps{MoJ1CSk5-M~PeboFP=Ze{VB(g8zDw%)ZA#KmEE-)ZIV zQ71yVTVzfrF|<0k?JqH`<8Yu`!FP?`o>{5f`A)I#uDcehK*AXhIe=g>~%d=34(D#lR=5n6>~e`FZhVE@qQ%@gD2k^9fypGT98BeB;Ln zjlW&QffK0<`gB2zAH_HSf;S+>KFl#)v3z-*_N(^o+L^}8D!!l{W^9shChv5%_NweXKP58xHkhfCu+2l7X{ zV~Xw__9MH91V6d@KJa!sqz6@piqBb$8nWHk>j zxxBCG;&mF&!Hf)d%Vc+=>Pr2{ovJ|LbLeXp*N$}YAS(w81QOKn7Hc0u7H3RNX%{*S zB4_&Ik5i&kBYEA*C(F+V{W(or9-U&`3#(z({%uDx+%ShUzu&`c0rwQsbsqDiJLr@E z@kPbPrj>W+Qp}3EG4S)FXVbi-=>w>Jh!Nwe3 z`-J=SAYz}~->cE4u5=$aJ}EU4G_cA`ln=t#ON~?&_y)J#furUw3dTgU?=J2wD|;nm zoPk07z7+^qtfDFm#16$CE1SAvSkGX4*89$1WiU+vKL_Z(gyEbnRly|4+Cay{Rc>5f zc+E*~O1-dSw>!;r&=1Y#((mnrFJbsE@tvN(tR&_u$ zBwfYl;VJd=Cr#>#ORcM<*~-T8Z2#UHR@K%#d?>B8_euULi_2%QQf`t>SwI;w?cOEM zBsJ=uGv$;A2c{26nLearM6TmA>>15w_pwvg1{#=|?H+kUO!7N6YHOn2l96U+A1yzw zn;ZKd%7@JF|Lzy^riP}1o|r@re>HNh9?*0-cBuDOddsV86{er?$Meu|=cv-0W+v#E)`v2YBi3YTmiW`b8kG>Af%e0NZE_V?LW*rx`T4` z)4nCp$c%>XGN;+-*h)Sfx8(YpM@IQw{bA1wWYfR}4`Xwu>zd=u?XZ6Yh#ECsR^O3Z zSo02YnymqA$Tu751lthj>`(x+bWFu$3%A_sp)kn5E4(5}mPXoTr#1Y3Y9ug&v{p^l zO_J*x!k(}vUw!_JczHbOAzp!;EFw3!-`jywe#s+R9ZvpntZe9;* z>4Wu&@ZneMyUnWBBfixyvQz^h2mF2qwGLRwp2dzy^y;S8vV6WovcmdLL{@(#U~~_# z6w$cRyx$5b^EvGjx32QZlm1+rVwi-WO(D!8YA+Y|Q-b{I$xj#HMN zEC|IpQLDbbn={lUSBe~jtF#BJ0uNJdONFH0PvG%aZtkFy2-wwhtbJ5C!-R{0 z=nPL2Yio!7+S9ab2Hx!Ed||mRx9d6zNDaahtQFk&5)a+n!(mTF)kJgVh6o{7DLqs9 z>)VGDI@c%-n|;!K`_7MylNUlouwet|82|+tI^%iXR#Qr(#1KazFVUQ|eLUQ5?2%s@ z>go;Z!l9hS8wGLDZvrzv|-=L^f!7#(oY`N}%rZpfeKk z3@J71)?I|aP5Y+{5uZ8}d{o5o5#s?4P?vvwZXcsTVqjW!L$5vjn?oLQd6WO)DV(m# zv#b__@cbfmr4nK(y}H$ZW_4C}IA-SXhe?`}jD4sd?a9=^J!|Nw>V?!e@RGu`)JVqr zMUA~QKlz?WxR%?E*R(s_H{UYm$rii4CV*KYkm`w;@Am$&`zAL!ay#t#WaSlP}=dHxmirDz@8>yXk|F-)>M3?*? z%3;3|?Hn;f-b;M+o81t!RBid4NVV9_CLSK9xX2S)|D?akR%gw+H%ZRLT9Aqd zJj#lyQx{NkW)1A=@~vkA*d}*~K-XYEE$Q9t-mQfK89AWWC^HYKX*+JN9wf1E<4=rf z7m2O3U*S{1uZ=v!5jY?`adz$x!mZ}y8`wrgl~RGcQlAH7EV8>{cmF`_I$hEa9SPu6 zu5U5$n$`IJJ(@nsEZ=k~-4|eCLHc{oU9z3w8v5hgwg+uHckJjqe5$lKTcs*zg@gnL zr}Eusef4TF^ZIwYP4f|F3xKhrbC+Zi>Lc|oYKuW4;)Pf&#D7FeQVvL;J@o4?w_s|V zLUFjUy8+(<_z|Ps%HG(%_hd!W7%b_N6K>hf{ooznZRCIq?%B00Y$ab!FGJ`^VUrKa zevp`9?tgsVWyYXGA>7F9#7Z-;GZIPAfb3~qx@A5n>7i-_LWbxlzG|l(g6+@7J6Yz5G=a zj3=%K>88mN>Y}ER34>QGU3!hX$>)u>M{1{Rw>k$zBFTC4QO2gZM* zpoPXb^&?_x++qYqn=WaZ)w_1Lwyi%kNc!E6r2NCJ#)ZjH$SoG)4uw7nsU~g zu*78=>fui@<2>)_3Q@uYv99n~>lr9T%}EBm^uE?zrv(}P2YPJIih((Cz%~=S-!XH# z7@t{hNO7fX;;&4*&;dCGl!q+!q|%HwZ5rwuVy_J38snVP?^~~ss0uAq|$2LA9`b3-bB{ zAzBCzA5To!*eK#^@>h}2r?=OB!knMLC|FJ7=r;nxJ%v2LREbFj19V>4Y;(xs1K*sa{Oi1{$7QJs?dqj}QpWB^Y$4f0$9EH-< zb?I-uTrPU53TIx(?V-KAWHGY7gjU0!9n$Z{#4mhn7j-G8xLW;~OV**csxn8N3M0k8 zTh+R<*d2+)52xsgr8PhQ?V{$XvnDa3f9B`Z`LjY4Qs3+N`EDDEbN`*r4kr~Ef4=}p z$-@^0iolDz@*Kl=8ln;|`@D=)`Gi(Pg#;m^B&-+-Xv7-|KX_2%aLy7U>x{OYLsroy z_?}kX>U0{}xHvHz#Pr3m98>M9UF9+Tp;NF`kzcw4+tQi=%iNfyHz9)~XlfVT7tSNqih_ zN59R+1JPPqJO>P8J!m3AzI!M7x4Bp#^TI`o7p8F8Y8L2e|9s05&!vkOUqLM0=EtDO zT}LwLEM@qoy4VZcGTu7*uo4x%^Q>it8#`>UKA3WIH!EkUd@(C3?fojGc=2J5FF+2A z%x^6}H`(647;ryBS?pu^G5U+TTR)n{lqKP|yabF@mf5Bju|hV(yC<%{)GXMFL+( z9)Az&WKlR+cY}`|Y0QRkSH8kL1ua@Yd;7fu8k7Cc8XILevLRU(*FGfM0+KLO&&mH; z(eJJ_sHQk|(af*Il^2)TvkeI(lj{Eg8hPGjUE*mC&yt}rs`G#Sv#nz{0|SHKRm#z- zC1$_nTuPGXniE;mBsvURD&`CKl>r*nWChs+>s&e+9Wv9lE!yhHY zgt21|NozTV>-r+A-FGu|_ayD?kczHugMIb>^A`=BlB87ge$hHzUx>^0CjtiV;p#`( zq9ZH*1i5GrUS6`jJ-)HuTS)_B=$IM)Yc25KX$$7GXnH-NcdXK14@MC1ae_aZR8w8W z#GNWo11|?wB<77QIfrFzh7te6UGTWSJvt68&Hp8@+ zvF}x<0a&9+mzFfY{ZW&>9E-}njgFaxaNUJnon;wf3fyt~H&Fdhq%5?ihiB}wKg?vV zF=KYIA3wIF#+33CruRl0zWe;~<1iUF<#F7qz$4|*4@vAhGAnZ#0RkI#+7(#>Gqvhqm&@CJ7Czh&)IEGC_I(#^bfzG zxBNEx$ccM7(8WXXDIDE&Be6?#Cj2@7m}+$3@NJ~qn}Y%a6CRD(^^A%r$lmXm-XN5L(Viu}hHkRr4GFsS$q$gOu-2@3m-m&; zOFO%>Ys2LPh%3hMUA=yNLd6}&ET^ixZQ_ARok}kR@$jCUk2I^xyC!PX^Opt{ovt3Y z8C=Hk`ou4HnT&njB;}5;2mp*z&g1=~Rc|A5C?w5siH)tQqYewy`Ao-c-}nDjO8qD= z=+}n$nleexJrH|U;hU?6bWmZjb4IAzJz$XY9;f9tz49{FI~-6IP1JwKYeK)oUR5;a z9WDueo3rZ2H~qe6CoOZ^sOyV5Ezv>qD@8t>590A-y`T2DqG+t_;%JzbmNxamQ+Dqx zGs!L27*+a15fpOuXU!&~OsAFXHW10=XepESon#K&9G}VBwk5xQaeriiN|`dP(UUsn zWvFqvT1PCkRWHR|(C}5u1Y8ArJ=W>EVN4^{%ju=zJ5h&FPFpVC4;$jL>J*lFoeAGw z=#5$N8z*Dg8wA>+a(RwfR3KTgfwW%$w!E2t?*Kb6$lN3~inz;YCYiQrQ zMugqlu+PWJn3f3kYfmq;Xgh|gP`vv0uH8a62P3uA6?~QYCO$dp=2mWg;t+sOm(FFH zC0oQ2B2;7n2*MSwUE1JM1y$9qe=<3WTI~XAPGVXVjV@i@%*$&oPIP$P1-P1z@R3-i zFj;rU=1)vOfAD;=Z+QVL%V^<)p41(dbmg>g9-mZqk6JOT*XGG@=YFQS6MQbmeHix< zuz*+4yuYE?8)PqbyI76xYaK^P-}2tG`nOxhMK?92%dCch9Vuo7e9jTAz}FSQq$Zmb zyrKyiXQ%iM8wadAfncm`(v?_fxO=J^E^!;?>fT_)!7=6Z1t^%5XnhcSDWHY|N!k1^ zgdrM3jPy;)m11~~BB~8XtOR~(tFfOGCB)k+v^5%LSN{Xp*~Y^Vs13oveMeR|TtaX& z{e}^&ngA8RYmmU0i9wdK65F`S4Pjvo$&Rl*vT{rK)l3a%sx*;DFQaM{bv%o^D28-` zJb#sip|`jOZ$*N*e26p ztD@=r-|u6z%%60hxkOXlc4rn@X|ok&`i)MU_TM&8BGj4dWZR*SVZL+ppEnc91>Q8| z0={7zolds8!zQ)mJ#gpc_G%-tazpDThMIE?DUVw-`ta>>OePjRg!Dj1`vdwY9nBZ( zzoq0xYlm_IkwZdmV%u+19!&$CuMH^)_{G-}-|v*3-p(@6wYq#Ua1u&SvefrdBNvQtLG8&fpd@%`5 zq?2O>Y?HP&6wf}5>iCqYAHpx$N*|e~U>$BkgxSFin&juHWt_ zvc;rF`+Xb|H76^IVw&6$0j18KhtgYXT*2rk8wGnmSXT^SCpI9h{`r6ZEm-;+3DDFt zLsQa@CB`zrtZA=IAJdALze<|M0DV#?8(;&>Ox$R;i}NJuhfLt*Jq8P+cclJ!@*L0s zeUNmo57|ZT{Qrl)8W%iS0nQEn76F1L?z@?=q-pnJ=g&X3=?}f|eZQDeFT&|W8;5v7 zB=$jb{nzMGvnG|UUJS5DcW$R}mReZx;<)EHE(HTGVh!O8MP!oDaNp-fm^x&K=oJL+ zClv1UU3s=vb4Cjh426fFZde2d0dECm7qGQh(O6_<^2bDz*sBfnO3bALP2WCx01~z( zD`EWqTAkW!8{JwrA<6TgDiU%YyO$Jh^O|3OXR93`9tT+05?EhcgxJ>?e?eL*;2JtQ zAw;EKOnWi|1j=e;`!Q&e1iC9)QtE*1LuPd40K|db&TAYBSrV_?ftTZ_gYu2REcMx$)bDYy*| z4L_hXlyhC<^*^_03MxKjCS%OW_l%$7E{_=i3NaNk77Y7%gbD&NpPLg#0my^od*2K+ zJ*~b6))8UoPdil+*i>hi>s{H7pS7R?{(#P>>G-OGFA5?!55}pIA@oDG>W~PS@lJ;I zDmroUAViwAkUwzaKAoA3Hy@O#yhNz`m@>i(1Qe7_v*(tl5j3~ zV5Sm=ynshVSI+7oVc8HkPJs7W$MnC9IH`zS4aol*m#82F4lX~(nNLy2FC9FA-}@lY z>hugv6R?d!AdhS!ro>?Hld02>-nrfJD+{Oc9`kUr_h)d2sdY| zHIFWG;@4gwzx+h^1cYJv!tkE^0st3ITh_X0l!2n`>%+oj?Jo>hW42chN!tD8?qJ-_ zqYb>u{FjavW2?`fsDAVU?{0=g7qt#R>H6n>bL5*x89lT|`w$tIEv5fn-GRXqvg)OI z3b7kHoOSy+To1=I639hom#Wb$-4(qx>%mq05 zYiVc{SpAXV|M=KQjc$0uGoYKQ(bMUzD$%NmFOOl41Kx`!38%8m;$US((Sm|f)EV$V zFB;1gjZcFSDbpV1K7HXYk1t4_#hsh(c0k;<;FM6pf zZr{ehn`He@iS=B_iDNP2OXMsb%|{}%|Ge<2`wt#mV{BIaV)unWSrZiWBgNbXQ)#ov z)J}DyZqn6pJTAyd{jfqz9hMt-as*p@Uz;ej&HE>d(E2k6X|0CT*Zy1kM+elggfHp+ zeJZV97egx{(leh%+{arcB17li-tp>fGR%o62%$X>Iqd@^ojHYxl@m~?)L(x|AfnHq zwG-8h?f$WQM-JDU#i4T&$>%=TGeGvJQ1HmH-wHT-UY#Zqz-wEXq?~5m0gS5l^NjRK zy=s*Q0^kf=)6&}N!KoEAo5X0VGozpbjnFy#Yu1ChNJxk8Rjr_^Zf@zV@7{qEDFPh3 zA{(W;J1ihPOhH*$T;MPY#tGCX0ly*E!erawpA)2+IGXMqs=2mb_)6s60_YQ-Z#EY@ zX{{24OR>;ketHB^7}WNV8c$+c*&KNOw66<9TP>7zX@-5O7CA(B@mi-N%C23NKgs#G zyzc54G|xRomECT0y`5I&u%d87O2Ka}ElG{p`w%9QG(#4Hjv3sI^OyLKr_spX&BJ)J7DCTy~Wa@$i! zj~@LT*4p-OrC8fCq_^khbF6Ns*bi$raytV83%ci??IxAjpKH+2{<8SH@hX$aSJT>a zA?Q^YlsT3|H$);v&V;U-*{etD`Xb{KL9$jSmbY8S|E+pSfPl_NYBHgL6<^(kcWk=F zW9g=PYCDGi(?z$FXojVC5|9Nwp(Hf59JUUv7db)p^fab?o>=`-r2y%zx4}s&z?C?q ziGY%2gNCHtIp&;@&~tu_DxSI}5Z^X6{svApB)-tEj7epD7g1T1dmHA?i`1P&?@8hP zNq(U_fs5u!RXT`4ILw8@fnw2|4dHXGdfw|#|u$Vbh`O@0o zaa4I#bFYY0mHHMgezKhxgP_~Yo)>#MmoFLqAq-OG1;lLi?b%d|@f~O+!InQTf@E9j z?=8fq;9$M|JCRBLN%g?)wZL}dIVO=QyA=zD#ru)*@*%z?;8e} zwlbNU#{s->1!;h{{DbS)H~w?;MwBX$2%F43C-sJlUg)JRmH(K{$e6mmN?MHUtgC5- zb)QzR`Nau@!r@^Xg4Y<|v~BJL9+sGm`s&oYbjOeJlOpf$Msvlf3P z@m}u@ctGiz--i+y&f(2AdryDzv=9w)|h>Q>FrNzLV=!x zx|N)?O4#md*{y4Y|Ldgc82X>osxb`OW6cRqal*j8=z!`mnTZd;8|d$ZKSp#UiOSg*VR z9KVM*{N{qLx`z@?q&gkxt~hB}wpBNz{k=}7cKL1Z1P z>{fbm6-P+r5L=5|7?v@~A4Z(hj=2k;B06ZK{2e{L1YCaa3h5n@~=`^#XmFUu`$5 zdO)6?o_I+iy9lhEMc(zCLj54mw%<}C>7<=?{LK^l9pWn%2ENcSjH~B?v{!5KouGE6 z9lMdphg9j^iT(Zq@q8hrO*v<^wll|oqU~0`EB8Sr78Vw$tJ*^$#ZA@bOfvK=Iq)+O zukRJJUBF1O{m1na+Bz(lV$#!j(+zT-TngNWdh-g_|Lb;Fd-*mq;a1Odd2TU|{7t@25+y~V_QTFWV(<|AY z4A?3{bdgz;#w+zPpGR{i+AVg7DB|%ooNfqPprt*HV1~GcZ(NqmYY?PmGoHGv9yxLZ zdW%#9IXHE*8Wt@O?d+*OeWOTd{F#v`*8$J9gl|>6c zT~lu^Sw8s!N2&)GXV8y3^c!XMW7qzQC9Osh#-Rh4LQ)A_7-C z3}C~>a=f2GgVwP_8lPP5PaKPgiCT3}iY|phMYy;Cgv8)bR#@iZTPN@3DI-0TK?8l2 zdR|Ud_1}>Fy1s(lTvVIig~9@rYsF`2hH+HR`UT6$P1+sV%#}Y1oNk;XLVjHlLo7>aGee*f6bG_6m0(D|v{ z8W=a_Sb1Vx9M?-2$YiDD0LSL0cSaw%#OS{|lSCV)`vM5S) zmtNiW=f&j&BZxQXHT&9?<-XoH()x{EEybCJ-@qX|mg3{U#4AZmlYLIspSY{BHLdiM z<#t0QhH`NW7lD}lO1++Pt&Y}l(^yX%ztuENU>fV8+PK>m1dTi7{hl_=vq*;fyRBOA zVFMKvl@~X{I;p7a9=8+uAIB#fVX5NK(oHJLX#;&-wHEkyJg@2AsZ(Z@Dqr5Sh6X7@ zCCaX4{qNtgmIz6@n_gEaBjYG{XKJ_MepQhE?_g zNZ`82cd}ZyrqA6|UQ?1@%3g?w)hE5A;ZG5Kwc{;Hxk{@huq~UV4on79gK=YN8GV;| zM5)ksqwd`NntvT!It7M)|Mrd8FdIGRgO%PH6RV1fA_f_4>)NGD7a=&ixUuWUmd}1% zqIV44rv$T~i+l!0ymsMp{;hd(u;#@tYT(!T#r{aI zAzkaU%qbHhXR#BkRt$M`ewj|M*qjp~3_#`f$Tjw^7~h+CkE0Gi&M++t%@ePv+lZmV{I?vWJHAL7?I z${OoQ?<2VEz=_@RhVVoFPcH3U(MLNqG4L7al!2fFxPRVvLf2F=6oc8enPZJLJ%u=J_%p3u4jbsMnNR1F z`Xo8g4P+v}YfoN-gRZkg3LnA5d4`&v!oB8f*8Vo{P5m4P(QVN`2&Ez?-{QFmVPe`0 zfd56aQ{Q~;OHMLVV(yH*o|fHWCmu&f(QM3{=BA5_+F)+$AeuZ;H_)y1CsZM(Rd|** zX=mOq!^M-z9CRV+V+6xgWFQAy52dAvQ)+x{Nj(R}!8L0c6&_DC zeo56p6?GN#l_|g*`QCaz7JYJF8Uv>e^g5Uui~r%apBC~20mc9N)f({n+HGo&vZe=9 ztEW!wWD#_(0%3%TdL ztpY%M1kVukx_VRG48EtLJzKa%sLLgMob1KVB$bb6nk;%W@kBam6@hXHrR82fI?T8i z@(Tw03OE}ZW;lRl6n#2iiFPL*`xjBO_rTo1n|P; zBrwv&3=MGTs0E7Gr*rb<-@khEbsbWlNq^D@e?pVW-&NoMC!A11mG%_`ki`{Q;yEacLA*OefNnGJ6S~>hh z0IBb1`a2XreDLdg{wA5jBDB?eUs6GJDsqtG!*~I9g^T7I6JC?hH!(<6#}n&BIIto3 zS>dQSv;hzDjsuP#l%Be7l~*K>6ho^(%kw>`J2|(rC;jkGeDaM3yA6L|m?{KPvM0QE zv})nW^gP1lq3p7ba~4EJMTx1-KfB*Y(Qs_xoAZorfmh-NYE!pYg_F9&j!9OREg(-f zDCg~=o+AmA3%{dl?QW3Kt~Yk*C|oT)%|sA}4N}sALF|&dr)a^_A}d zC}yj9YE5UDZ5S2Sjre^r{ufhf5u&42*E!OWfcx$EMhoZjDPqYyy*o(*V6L>RxWDJ` zP}UWbLz=((`1#$EbL%SXm`U5?T|VUl<|V%`oNcv=8deE4D`OhwbV{xy9&3h@y62kqUM#qZBM( zqNq!Vxru{5Xo-vOb0L_mEIcYMxOeT-CHHow9Hg$_DKRSYT$z92*O2w}EfHnhkzr0+AfPV*?a z*V=%ZB}vA&lH8=1qUa8Ok~Aj7WYEVEjQdeN`mBjw5>|P{D_3vb#PBtAo~F-5GR^r& zM#ANpUAu-!Qr7WBZGJ=DY*W9EO72-c^o79hUnJ@z1P_%n7K4Z8 zdI;f2Zrc#1Jqx${b@#@vz*6CVH8yL9BF?4RTH{LP1_>&>KIM~=*K_v;2S=yQ_L8p! z=`nuj=?)qMw#}*vyY6n;jX;W1!M$E1eh{K#bTc@=9;PJ%39avbb_Siz)tQvDd5NC;k(XS}%5KhV%nrZnVl%{0oYb1P2Ibo*`FHc; zxjxf)aoibyUYy8@nHN=~m+1XVmY{I^I*O?!<`Le-NA`k_cH$g9s`XH2VNR)Ti@v=Y zD#>gVa|Op&GX~X2yN#WVP0EaYm|f7%*XbV{Y+SE?)ieS>Nj2KwxD>m*tp)PEJsXGS zUCfqkd^wrx%2fJIy3zDepK%xHVnerdx%sxm1H9V+^;@_X@ z`7r6E&ZJZGZ@N1kxSm_39v(GUzdUk@5gNPB^~)x^OqgU-E!($mHy$#w+qj<>)3Y?v z2l|r#BP~_=&(%{$6Lc5$)ux7@*2S-(+1uvLM2d7ib_Xg4h=JK;KVDko^XK0x56VOs zxA<CBP2CRR3S_=5Zz`NuBzM?RpG=y&NyYcT|{L_y=Q?iwS&<4K26@39kGQag1qVq^ary$^n&xLz@qLBN>*vmuXp!AKp?b)o%=6~`t<3`~0a=m|^5viBo?PE8oBt{%V)R$x{#WfP;`IZ^N=DgG< z6`>yaY8J@y2S zk(1`^Lyf5wp=B#EzhPzHmA@eZkj~G89J)9nYGFz#>brJ?FW1!6EZ(O^?HyS@iG~!8 zv)kYN=$0fi{SD8~(H34Yb;^mT=-7d{+ey z=C1?OQ^}X@A2&r*^0MbI@#mzvbwi}dE%2hyds2z(%3(Pf&+ZBSNyH2Tb=}ymt#I)- z)((+`4oH;VDwn##&kx^F?d7$@uc;81_18CGn>2{zRyZ81P4b#6ySN>akCIidY5$#@ z#YlfDRsPXgw!N{7s8k=H(_09gZB1}~%t_R=JllCGHj&e}=HkgRE~ra!I;s!OOxUv& z2xfgg1D4vcFq6(29c#Gbds>-UWKxw61ETl#;k}D#g^aFto_nm(N+qS=4aCtR=FBcR zW?`?@BC`CbQ#Z_bFKkVwIXZgx8l!e?%+?-#M_OJU;pd##8_#CH-@jUickMa&%AUa| zLySUJuKdvFAEb4H`ybh5L}P)^LKGcw=>AgOi=y+px*l5CUCr`j=cNq4d%K0_GR>jY zuN@hRorNo5?&$;?iA^pNDKUtrwUMeg^YhWE(z1T?U^vu?A|~}LV*mP!f_j-B$d$6{ zW*iY9xqkP$hd1o)Q#k39Ax1}K?Cj>;N2{Dp2}0$G4LZ=3`@41NVh&5l2y;K@h28Q3 zp2P{V<54dm;7A{)b}dea8!)|H_rMuvF7Ih*@J8;TNTTS=&)3Fc5T4W6jP$Wf4+VMo zjKLM?10C<=`q$i4!ZRdJ;WOo66|XKnnBH!>KQ})GzVy2S(|z3aF>CspbQ<;jIrU2B z%lQHdf=40LD8u2PrGU6^?h+8U)>k3sT|3$FWQP&_v_U}a=&PZLAU@GpTn~r~Uqsb)P$ak)2rI zBofz`|5>#n0~D9QDs=~zexl;?65{~-lEq&dosInN#_npS0bTE-U)<9!G0>CAl3QFS zYf}Q_f~}v^#1_MP$xPE1S%x;|)mL&)3WT>!lY1n!LJBK4zNfL?TPPD;7Q7w4a%uckf0EIx1ALDn^n+tEosW0ATyWJ>EkdYi72!(uldt##0m3%`X zz+U{kuuJWdLPaq=W(^AwMAfD}w0^i1hlm7>OQEBnEj$IZm z`-5qU0ultdi(s)=`N0?c2FxcKDI?eWfW%f1Chc@zi%~C zdJN)&gE3-~+jZ=yRYN`mp#$YPR`vCTcmYQLiz z)MM)B71jZuLFqu>u0g2NRO>Zp9tsIT*rML$&;D_*uhJOUDff^5h9$m)2UXvf#>Sdkk-pBMx z$d;e8dmWsSa9MaTP06Bl))y=?k`eMPv>LkJj1pA;o$H3sHyj~uIZj7Kf5H*oJjy8i z=WJah&v&dg^BCxY;ob5wN;|wTTERxH4QZ@DqMVf^lvWZcQp^@Qm3%vP zQ>Et=Ao+G`jy?~PJHI4x+b}(8O*7@?jWyw{dn1f@lQSU{JV6Z+qEvty#q%d?`z>ZJ$#&gYJC$C* z?U;U@&nu{mGPq+!ei+t1U=#U|w(Gcp7rU#6i*#xc;Q#0EeBXO&f2$A#o$kL9o^ZnA zlG!>UwCCJWs^5~IaZB{FPn~+v<@T{s?zPMp%BOG-xnmbh{z;KhxBC8*1$d+^R~U&H z8dMdg^^OQiY6JJ^`VO!hK7IbXX9OlzM7?VK`rj=jtSL!`ziBkAMBx)>iyTHR_*wr8 zVT*yK)%Qj$L4Z%939eBe$C3}UAuY-J`4jX=$vVKtPMvPLF(N|PPDwq67wo83TqINS z=*qo&_ZnQTfGP@-(7|PbLHSjLn~6OQX-EWEjNh{Nh_Z%{oNn4l3r9^`Ep(a*?vVse zAQUE6PW9p%@eQMrK02tZZK$8Aq*?Nh^$$PWk}DYNdewl03iq4V_n~o6G_CBf#|MTQ8QX_ijwBC z$!G%k2(u>G|C#VJogthJAGsPf02xN-%)NPb&K(@MKpIReiS8e3*1N3`ZW1(l^+{io zEhJL4vA@ThG;bT z{RLlq;%N#itu&`8EHHss6tm`q{uXljIU*8KepsF9?AgiUe?x&7hwP$j5>OippgaJT z@n9w*fbMAopFV_M!94~U;s?d5aDiQ*>T zn5|pg_2H8z^Tbbr8F_HVh}d>vYeos+MwDZ!mUr7w_XOwBhCvJTim4PTSsSAFz;-^K zo84@Tb&&{sVU0?1UaH~-DyG_Ru0#8r=uEpX%^ zuR1OxfkbsDAA%yHlX&Ntj$FacQirz7AtUlo8N7bb$s4|(`L0BBG->D1O=#T?P=(Dr zw=}&WD?9N^*YJkNZUP}+7zP9}{OV0bV-OOO%9N2C7f^jz^<69Us(|%rT|7`4h-zBv z2GIJ=qmK|bQvPxPhNYNBm`n{o4EKz6hjK~6{EW-0hS&&15$@^5S0%yx{1y8 z8W(!m9`u>@i(GFE9j|O#-J+P zvMV2O6m^lUC0xfFVRAAl`73}ZMRNCmKo-w zaqv!_rD#UE%+*CoTWTj@8{u~c&5#duYoS02Dq)6QIGP86W#+b=CKxsUhy!;%Af-2Q zERjzi{Kgns8|T5Nt_oBU*W}y)9w_26Dh{zT?{>XE7!|k(_eQ0ZNz<5d%ZAdaEIom2c)Xio-gE!GrhNKYa(ugc&$U}Pm4dH zq!aGQuOlB8l_Vw#2Wjr-q#7%Yb#n*158fx!qep#|d_TY2^-+@3vnv2yTz< zTwf&eW6C~j4)hd>zd2szc+3IsJhQfLW|t{l_1iFJpOKcQsM3*zp=!4ruG+S_EHu)i zi7{DA3!h8eJ7qad-m4MQQ(JwKpNt6}1BiqdS(EkNT&IB|adrCYVK3c>HU;J`e5e?n zxLj%TrcEgxw*SA=LSE};CA`+vuT)N(NEO!5)0?`zYp2PKR8Je$5nD{p)oaPLNgNZB zmlk5%UbN2{VY6;*8vr;(5MUwf@v%jvUk>Um-(1{M%mJbzB%wgAmirqLg)JfI<81FG zN{3W$B?hY1;>6A(v>h8hAt>fKCN@tc)LOjE*dLB0qq1(ASc`t@FuaRnMM#05#| z`!OsEByS#UQqnQ@JDgF}7B|uPZ^7KKwXLiPeBOp)aAUC5IWy!mRvp_&!oo#j8jTk8 z&yI$$vD27i6Qi~BerPr)CPcu*gf)@vG_{+5nWV8Gfw6&^l2<~&P11RBbb98s);|B2 zr!!RWn0+UFX{g+wk-=V4-F&B__Mfg7FL!fhTAp1+j>0YBB8-tgVvs;pZ;|+=FOelr zo9A2(tna?eUpD7!uM1z#s8PrrcyR+}XlZeN>}GwnaI^v9$FaV)ku)znQDajNJTas@ ziGY2}$G|rrc1dYSHNd>SHwHlI`Vy7=pjYw}Zp@FW;cuUVOU~Lp+$DGLlxhLVC7s;O zRG9jf5TXEk8FGxzoPMT1UmOQ5HiUwY_o#j{GW4m{&#wHa9y;ad9d}No>9Oj|-<*ax zbD8n3v9YB0llOc}PRStN$|9KHJ-zzoylQl)vKK!;;H8A_yk(YtPh&AlC~g)xi2v>m ze0+AvhG=6F4=v|Rx$S$J#b#png^YmRO=~F9Inl;P)oRAg835H8#&J{i{Q;b#-`fvn z$XNINdd#CO4{VzM;5w3x5w1_Rpf9#eDCL4n9rq7Mt5wJ0RGlxRBs{m*)tT)fXT$Ms zJ}Ws#9;;uJFs;7Tr&pcxh@~$lCbS|2Nt_#)lVNUMckge+`0atV`#;eQ zEuKplN87h!Zu(PkDqI|SMcH<(39g;80n?Fd@g!~_aC5nOMN3n&BOT|<-0cp7d(I?R z-E_T6o2oBfAZE$;NZb*3OkDb@>G6gV8qA9MDx)@aYM7&(m^R?l_tw^!^HHABwC5XS z!&sw%bN@MdM+g88FI$fo9sb%fx_Wd;yWMVK2Dst*x%%1jz&cZhr*^>r>$79ivnNS5 z$w#LTIr!z0&>Ps~r#D;~SW{>_`r^uuVOq{(MvvAbu{zFmy`T+)j)}&R zGD<{RG>>@R&@lM?`_H|c@Iwcmz!6-c=5vL;pn=3k40AEY)(;k>#PS^=9nu) zAjR5K0KsxSzt>(0-D7xSxO2`xdQG!?kA$~#ZM&f~e^Hn-34*|p;K8jwx~)nZayB*; zH*#qsdP}ABXx{ze_3I(Qj*I!+&HX^pG({WwPxyzQtt*m(Jd3}IYWA%RW4~wR9jnYc zcXt5S2Ib;Gk8fxzx}`GF+{5`CWPK+DhlF@gB!;OkIdqy> z^yRw&+pO8zuUr-YEm+PwJU{#cYJ!sjnEUGgjfv=o@66^xsAR;HeMmWLZ?ena{X5Co z=+bvX;*jy@*ZdqQr8eC)9+ySs-Ua_}`ipH+c&4peV_OpOgdX8YcP8+v^A{wod90vW zFQXz>egC96CvAbeGH|37a`OgtbJ2-A?cmy0H1l<1BrI3GXp6F(q@j}OUMn`)3->C+ zt+o$#2Ea&9bXrm(4qwIo&?nVyHh&fRqzkZy%;UHZFmnxh3 zK3%$DmWLSOrF8??cMBOkl|#xXUt;=X>y6wSH_TK&<|aUYa=$%%N=xxcu02JpQ`qf+ zRP5s*P6(eFb6^yHZsxA-N3EASW{$|$ofBkY#dqeN)2?Fp)bb?bnCLx1CBn-)Tpf>O z*D5YiK(eo%kL^9py;H%!^V2Q6anp2Ed$iM0&@=9B;7~@79!-L%X*(_PBZ1q{0HQz* zlDS#@|LB?{c{RG}|BszPe~DckJ<&$Tyg-4IwMd11X>TWi=@C+H;PvZ_Ey}IS0 zrwQ5y+NSYA6SO^lyb}?Ii-ZI^!1V2)8p&osT5G$z5de{qD2n|siHZqUxAAbqE9)Kc zZdN{YT}19p?Kq%L|D|JeXdzLC)#Z+KEjFTJS9RjxnGlhpgp6p&UXht3J0hEgQC27v z85POO*6-YU|JU_@ulIf4>nXnTzCYu9&N(07OV19%x7ubT0kvB3?Z~h%wd3mv~yL&yw+ky%9PcP2FU-J;1#$z&HyZFRE?@=P#MCvv+*YAa|5w2yy#TU3 zp!H`a)KamvPk=7mpqa>88%&g93v#nC$L)y5wg6I#6eNFv|v$dDng&A?!fN zIxYyXn|%AT76Q7sK!WD^b#SGXScCkfS`yhrxvulg`8uXEo0;mBwVMTCGTb&V|f0Af>uPQjE%_$u2U#fZ%)xE}Q#21=bi!VNc8BDFkAUFI~n`K($5o93xY zXscFvL(W1N2AWBDYZ5Ptu(tghy>hWU=*x4uoFa+kP116SA|~=aPyh?jTTF!d}YG!s?bS z#M!MOu_1t!*cF4%c=SYD#gb5z>_d!iKpr0e%2PVv&tn5id)-oL_+JMpGVj`T7blVn z_QD8J50pKMQRcOJwB6<21N7h)PP!=&D;bv`54pRU_l?QJ;>|i|>XCL@H4X_`I|)Mh z&0%@Iji3RwWtgJyWAYLP--}!a(1dZIxklJc?IT^|p{iHpro%cEMuLL?+UQ4{^1#aE zb^g>-qvG7!*+w82Mp&_o6Qu+N8Z^V6&JTy?kwWZ+@&I<@#li*=@dp$(3h>rKmD`ul zD~_;zgOkbL6(cOcf`xbvq;q(bLv$OCYRWYu`5KJMr+=^_12N*Tg*(i2_BF+bVXmmf z9lQm@rDBMO&Ebqi*6^V1Qf4+^&aTK8B}En?q+LfXFnoGcQUUzxgd?kZuw;EEVIxX1 zfxQgA8H2;`LbRAPRoQb7=ui(`c8pU`MeqzN+GU|X%XQ1p78b0Ggib7cPZISyo@!u} ze&}0&9MCgmun!J!_O=_-0iuCUjoI;cAc}iA2bid;}|cAQS%ZoRAj5d zjy)9bMgDw(feQdY)$EQN>iHfq`-c7U=(v@eK^#5kieS=O-4R48#% z>w{IOP4M4RRIyX>prpfxm#jwyaV!Z% z?gO>G+i%gD;Btk}_M)63q-`LZRp3~`j_;w*nKvwV=A5v(qjVOCYycoH4#Wk1^tc!T z?Z(TWK=Zc$iZpvQj3|+XnQ%~UK*EO#*RJEbswT_CdDK~M<787kK7@maG*HQJ)r=H9 zL}#Az*9Agh<@)cqL)8GJ5 zg{}7{IoDPrgVQiE@y_iWaO?m2)y%Oy;tbhx0V!2}?V9bRnK)LXR<*(p?oS=WMjM#r z$WV7ICM(WIYYPxDdB;uF2!@CJdlS{)Qr)H?DoGr*&Ae79mh97bROvd{ky(g+6R>LQ zASMfQKlG}=->U%n{w`WS!aRUh+q#44&y|YLp{P`WvETU1^aLN|xoJ1zcnkd7%c7>v z?3X$k3^B0eum^=P79??_-)9iIL4AJ3Uv{gn%;Ue56PsW~cX_P4Jvs8GqcsrUm=EXL z&|DM2vCfm-tjwvvd!rc$|K6aW8}szX^=Nn5afJT7tY-3|eF+4n8sIyWZ+a@*Gj;C< z1}fpZ%o1*(m~N&~y@T@b2xbVVRBzCf3C83cZH6NBF^*l*9cvcmRJ>*~fG9ASTUoj7=`kRaMA-AGTbAj;_PN~ zFbdG#uKB0Vbx#vit#5N)#N=~!l*Vp9hBgQvK?SBw`=6SeO~PU6cTs8;Xa^{1;rT1)xFCpg|Zk` zA7#!+>AqM+hlddb3l{CdPu{^FTb+ zlAf*p2vEYTA5GDcnf&UWQfPw^Er!Z^;cZxQ&_c6;1`5SEevqWk1z*x4aBKV)S2*pu zEQ;}1t}b8jiBJl93hSfaO!HAT7yk4YCq_ap@feB|WE%v&4r*d%CKNes`6~z24uaX^ z_K6eeZYx+=m7Q<}YmOMX+33Y7uJVZjn^Q>stqC7ODTn(B8LpO69AbxIyUM5lH;*cV zuWURp*#Yo8T-M}TEzWC6#(0buM|;aYv3-r^mY0t`{b6Zk*<-&5&|r!ih(|5lcRQOI zq)>e&H{V04KsTGE&-MCu7JUMlYa2Rv>B+90-Q=pcdvF zW>SXV*^B>>6;L`5g~ar159V9wDgFbBl9t4~=gux)0@P#1y@qgD2;z%r%KMM#o% z1Mrh$Dy*9AOCU*Rq+25jnBNw94fz^=9^)?}n*T?W0_Q6Z#;CW47~v|F!%0ef#F<=yw~U|5Fl zd#TEg{3?`tBj#aoaXWz3iMXm|450+Ryze2V$LiCUw6=iR0ke0BxY3#Ktw%=hozOxJ znPumSO*bgZYPRla$5$1@yAEq{dAl|h$-Aq{U`UCy_oRK41R78pd zzBaECLT`)R-9~Ww6Ol!x7oxFI6#Mw1k%q~R`pauHMr#^z>8;GGpHB{h?GL82G$jYz z%KDd?AWPJW3!={C%F==i4pQMYpyA43k-^o7+|cugmo>OuBObWFWCIZ>$*n*E736!nUX+@ zBnV|H_u6p|KgY6Z=S)rN17 zxDcm)1Y9uG*FjqY$4s&!XZU)mFX9uL*e~+$2w=w92EGDAJ)<)yQr@R)F~vI{-x=rl zNapHvy4w9{UIBs4gw_{JDDZyUc7=dWP|}Aphv|oJqV6UT?r{W$jflpB{*UOa@Sf)F z9@8V0FJGZwsiqeet*XP2#h)R=no#QyoSWR?$7hE6nf*!0mIm>g7sh1Ny<5qfIs>IB5H@^!_xd}*v(GN<||Bf!y=jy4|} z*xm6r3M8R_RTJbPTNQ9dat-#&@6 zSfB&Uud0Sp7rVZhKb589UIHdeLFEjPRd?&xoF6d z0yLMXu7_r8dhFBb=SqHKzu{7PhuHn1H%;E8gGF8oQxgb;5^thm3`{_>2nlU?l6Oot zc#`2UF#GRwf~ZeR*FTd4a6pcCa1N?K!t`<|$IXN*Rgj+aDcIs6Kn-+ml(uP8(;Y4# z;BlZVY9ZogcUooJ$oQ#-@lWmTjKn+-%KUVGJsf;v9yh48ezXi?wJa!iy`5Z8gan$i zh#kaz)4_J<{VM=HBfj)i7@($!rnGoQo9KEJ)3(P0^!Rmx7K5M>n+)hLOJz)B2OC=; zIu&$8DnmFo8w{_CW%03y5d)?7@N5-W$bldTpS_CAx_89*iA~DRA9?aC3Z&B6gaaNM zQNH0L+4LX11vso*P_8Qj?-M=ung%@a-+!X8{#xsaMihcmBCriIUMA5Z{>2_hi8hOV z%rfLd_i1K8fu-pb7^X7hzbyyu8<(c0|9KIdc>-?Ys@Gy)?fRZ1B7#t&#HdXGHNP{| z;nS%?f#5bpTu9-@c!TD9VKXYf7@)~;?%73*C^lKHA!C1CN|83U(*mn5DI2K| zk{1f2!Ad33y0|Qm5IL2tWZ{5z2%k0KgQAfLm)5=$6!duD9a_3ThSenqL^utk>@nVv zcam~loDgIf*4=ybAeiJW_}{(305=FjC_=n|^g$zf=VxDEUl@Y^2YC&x%bJDLq+=xe zg%MZP^43xuR4gW*mIMohxGq`0h?)Z0%MBPFhoEdsao?Vgz*<>dy$+)B)t^teX9Hh0 z!ZSc*48nq2T6A>)>WStiiOh$2pKa5s2hT$AI2M8%0$ zj;7cn+Qrm7a=(i}F4zR1ANare0zYV@==I_sk1gYdl1(Yu-mkGUi?}g5Ipw-|GaW>l z54Qk668U!LW65e=k1%^p|&Z<9NZ|JLf6AnPN9&$@{!00j^W*76>~^7Tuzc zL*T64IuP+YEjCCmlLuy`=%LAe+?9vpKKB%djZlL7A5Gze5ea@FT5(X0<6Epfr+A&!=$H(v#^gNvt%rcRTHlZh_y!!RgOs?Y`> z!bC{}3iX4Kzp0}VpN)Ae*@z6O4t9#aQsQom0XQ)idOqTe!sZ(aSZ;(d?2U{;7e;^j z^v6E-ENikX8q<3XP;L$rYCDks22a(;AMSw(@&NsH(uoW;7*42R7^8f>1A&v?#KeS- ziAi-~cKjGd&^CBA*dRe(4WM&1j1bikaMgjDdpbO~d~xaPT9ObD;CKL=a|1lat3jE{ zhm6Ys*9mTLCz27jYr4ATrH|M~qlbO~Smqv1%mbwFGndOiG$!W$lW<-&x3$$EyScz1 zh_tMK06ca#E;|RRYz|zLZ-}5ZXr0MIT0o88dtOSbA3eH|*ca7?DJ3_i3Vb-bi(m2Y zZgammtg-aT9U}TY;8~NUwRkbK*!)VECHpRYIhgdp>NPkiZ*W~|u|^hE;yp0cAK>oO zLL%k_DJI$2OL8PCYF^YsTqvBmVCukuI-e7D0@-E@dci^R0mB6&3L5H)_90S3x zZvnH1j)6fLfc6`}rfR5T>(ip1w`LhhyI*IPJV~P58vOW3t=6H1B8rqWH2OA#S2~ z5e*i;pj$Wz#B&6Wy0>Q5Ev<=sl7hE2(fbmkD}$noTkr)h&3rT|^&C6log>%;>@k~8 zFH6q?%>XfNA;T&#DUNm@N=1DB2OwfGRhh?a0CLvSfDI64`z7jmRZy+hVFF)_uKj=z4`H5w z2$e=Wx?5AVXjxcR_i>}G!Q_Rz#CJ5CED|IRJaTsucrlj37ZrN>n7Rhoe4fe@DO>G!zo5f93{}F(tmg=|yZJ{q4kB zsY;X;P%5L2kjX|x!-W23I4%ej4pjPS_)bdWf8s>1?%y35c`X-#%;wLMMYB0%*YZQ6TmfgQlgItk!rE?JTfjI7BWOVEWFh9ZQLJ z%Q)x`gaPF7j7#0HxD~*ztoVxHJk6| zpZK!4yB4E(2#3ko0q3&{p^@~FqH8H&wgbQ5Gt*Kg89ZQhf$xz&Q;!%%{u(4tKVuV^ z>XC*Oe{c$i4XUu1!gjU;S1(7~M@;~k-2bZ?YFMBOoM>eIScgH0!KP9PLDQos{83he zMh?3Q?6QhYH57dJ`t|Fj&8s-+opivM@EpBC^I(X+0D~C{j`e7(wWk8Qks{v5msm&P zld8r-8lQU-nR6G2IyBBGbw*j}@xrcsU@;{wF(7sOp@qVv6t|U^z(go&t&jcAH@hBn6b=+*mARmH z$^tv_f71G84{xH?QluQr1r(WeIOU|SLyD$RA|_Zl)lBEmXXf`W;2@JX$#@>0iYbBw zc}9{k7cXdpp{N1sh`aa|Fhy)&AB-ai;SeR##w?kzAI>8!h{P{zZ%4c`ITikLNvy(oYXJGt^|y) zwD}nd3M_)tiFP?}O2+1z(Gr_>DO{tGV}QvJtNLy#qHa*W0N^mc<=bg)8`L82k)4cC zxic$YkOdP0kMd8tvBFc2%p?W{jW@QknK;R7z+Y#6FrXW4fH|Z1$`d>_( z8c;|P`pxPif)ea$og{?`B+`9O31f*tmS1)@VA)8N3L>&Ycy_|Q%{-lJ0u36k4Q}?l zMu;p193V}vM6mxnLe_>92rn~|_6ad);k*~KMnG_|%0K7?$KLCW7<|v)qsz_wBm?#x zyAKV>@JqSA?=WL^eiD>Lg68#C&k3sRB+vj%{YYP^jmrSyScbh^9xPSKnTxeO55Nm4 z5fV~PRoiIb-xIGwF)JUyQ^YPuj%z;&iw5X_7%$LmxCP8X5@LSU`^!MEri|e_DSH8T z49i10fUWHntn0wl-|9mHI+=crl~%QGOAVAu(|YPOVt-x=%3KAE~GyU6j~;{ z0WB!Z1F$`sEVI{rRAUDT5SIalsd~(7F%uyL*lr)P%tz7X zDQSULMF#LFY51`$Aab^3`W;;X;qxHP0jC1T%u~lFycKXV?hy`If&vql0icnbtIa>~ znVW>CgKJB6Bj6O_pX7&~kwEe1Ww#jM)xkz& zZ>!p&=58i0e9`E$AZ6-}>9EACET3}`$Q)iE_{cV=L{ww=>PlV!y->WqH?;)17vIp& z1yV?SEb?xm0Q|^#!o>Mz(YDrH5>t-x6bEJ{E!Lezf$C&BvGZol`FM=)hRh`ewd z4BV%NMIqA9wrty$#`w9zb7h$lK~@#tI9YoDKINtob{UmMyopdoP)Tp+ zEtXHufUi%>nIDU{h_&gv03VXYzwY)ts6SCEtXHBuHnMvokohGp1JR6j8}|K|)kZ}< zHz9B>pZ9jm!9sD9=qrEL-T6tBooJF;X8v@;E!=)tfQJq9LwOp6dCHVEjHP6y?2Vx}V zz#k0RM1>p>eYkAc%e_H;Ld&-XL4i*0VM0H?GBT+~(Z}n`uh@wHKS-JLcA;_{w|9ff zkG7jZK^oD=+Mu6a5409`^fyr-NPu#KsscR|Jsv{|6C9j1#e%IIj7ZX3Q1(HJpiNq0 zY=OE#Gg(hLNc!w#pH-P$jsObRn^^Vur6S~eTuiibRzWS{wOFC+0=D$_XYE{GfPrlg zD?XKbA|?ui?>Or)_PKZ}jjo`BNqqhK^^Y(c^e%j%>F@>V^Y3rREFnv5*HJ<-Ok@$p zDsz~N0E#Tp<+H(Gy+ExC6Rqd4fa)!;-Kz^~`xe9$#cOtW66DMvG(7|OU+HY*PcF&T zr-q{o;*Sfwq&6Yqe=D8+_ln^B9D;&jYV_3_Qvf8`h>ZA_QF7P7utC&)^4voV&`%e` z`SJK_t%+~o`7VWLcy|aMWSJS#V70S3<<_l!2;}JH@nJ1j*B2H>euWP;nGB!>sDBwA z<@*GAA?=4kba)Llg7sr<4&?iy61flAN}y_h7S#6kI;#Eu!5w!20-%_HQ_WX2vg8j@ zGp?1EE=pi(4G4x)9I(Rga7gbhBY%ejNHjt?87K*LdMrG{0h7FURELc~%~rm4eCs4; zwB6U-+#C!{PRFq``)C`^VrEh8+8bag;h8{!T#VQ;g;1q8e)uoM8W{i3`q02EawcNO z$L_4(6sG92${V_Z(0j)hxx&uH#UH=Al4}9H{qv8D$)hg*CbH6W0}JLI9KfNXpSVzR zfI6YfdROcx4Pw{a`TA?9v)g-gfK92?-d56f%)rQtuYB}Oc-7Ox*dw@GOl%^8eF~+y zwSm77x}mAee={Pz0kq{1vR>vO^xfDRQ4jBFuA_SRrYYWU;NfKz152Xmpk%@7*G%k= zQH!@YlTmK-*85SrXGXiGZim?0ixEu_G|VH#WH$ijJ1OINd}>OYtS!zMHc*g`Ds}j4 zf4?`3avYgzI_Z;l6{|iKvYbO3%{VI>py>L*2>X!mogn)`s-$?CoHdjbO3k-u>z?;0 zAaTY%Sv~m_Bg+qHu0wE;qi#b_&bN=R?+0!cjC?i1zfFgKw328GcSi&G?;&6-qE2>7 zJ%Z5z`*!mFu>}n?%0IW6&g1zh$fDO)=ZW9HD{zx#sLVmHNi3XHbPC4H{tcI889BLD z=RG)!FfUcP7^_@C&Y;@acr(AuNZ zhJyoY`N`ILi=kZ@L!B#i7W@~Id9cM6o{$eF{7Ivq^zE5itD?qdFOQ@2`L$ohPlx~ z(K%inIj9u4fW2;{&ifxH!@a+E49PtE|-A!Y~^SG>E zXGBk0IrGpqZW4wmrf0k0@AaXtbaj+j0dLB~aeRd7NnanXcTS(&E3kW#yjk#y7OoM> zc-d*(N2pTUAa73k_sSgy>OOezhO!ezC&Ka8gCfZm@iF1N%jEP=Yil-~@tx29xAw`Z zU=m;yN=1m>_pU6aGKw8W&q2^ki#REJgXl2}*zj2kxHZH`k8cZLOHswMTX`3vy|C^u zmVw3oW}CBT&w8C<;O2ovnp$}%^um)yt}S$A5;ZW3)H8s}fS@;+>?>!sfk}B6BOx4e zN!Z6uF9^Vt7z}=XS2cFeQ%T{p5Vd^y{(a9|_k9Bcn3b)x$s0YDJ+MwrM-JFg!v1g; z=c`hq6B7vHC$F>T3{BgiW1H5%2io|Z6^Cfm0g)2@{?f^MIt%UQAr$XWQGz{Ji)a4d zuSB{kW5l2+x$Qy?^&Oh`-4=g+eAW|PGZr}-J7ekfkd?Ue8fw?4+TJps0)_c zHLbUX9*p-c&rsEWaylNMDPlXy-I5q+nbu?BXY8g2#mZTcRJ(u;V=J!`E-qPqW)nA# z!5|(Ev8Z)(384`kh{sQ9Cb(SzyJut! zvDF*MNvNe@KG9Ya#e@LO(RZI;UfG7Yqun}2&4RYTucD$tLU6$hhDx=l0UZCXTwiVe zQN+L&B<6Zu@&*)zw*bZj^!^&2n;7tJzJ(-v($UW5R}K;pU?kKUxA-nzpkf8_BL0bm zp@;TgfbA$WgJ%~A%sDt(Ti5NpfC6+=;B=&(0*HTV?G6?ABOJTRK3nh_;JvYY5^<^d zJ66lwz;Nof@n7@;_;)NEzYQ)x?@yd~G;DxUwA&rC4x|xJZUANL2TVIS;DH!}=QV7& z1PhQ#$y>Vh&Z%zRcwo=r_i_OWDx$&2-00X); z6n38qw*YN`{9bMVtb2op>5MM}4`=CBfBeV{3IqJ!DsV8xU@7xMeuB7~xNd3JBR2XadoBe6a1s<3g^K zBo_TcR>+kRH)0Xod9tAq(JImW1_na|eO`Z4sB-I=KBTNZ!H>)P3F3r{&z_@>`FA6) zO)Z1CD5+k3XrfKGLWCCZ7bT?yibjb#%u?A7gmj)_2YH#8TgAr4{sw>Z+R73up}&U> zVo38?SR8UNESjR14`EBPUI@+LSIsYem?&m3_d`)fsva=jOWntskxve!U06AQCCKO%y+hTO zfo}<&z|^98^??qZ#0mabtfIR;hI4YRa^Xhy8$hS+OuwZyv z^Ft6eTXbZbG)HB70(MzBix9VbxSj+9hEr&kH=$h}C~eMvJv}{5)<}AebmD@D{*5OD z6l&HgXP?R0S)5@!@4{DdO%y$s++aU8mFMzPugL>**ULM+9SsW@_B4El=;>Ut|G&Kh z_OLLgUu79LX}9fLH%QhPqilt^1++6Pg5V8%I;#F=oQe3cTA&E@KJ|N4S*RLNB_Ys8 zvtZ^+FOZzv@%=I!_k7x!6wot?UT9)B7CBa}Ec7Tr6a`bDCv!tFi$)tNJC>seU4~%yLgv$F- zRgc=#Ehv^kQHN@TIg+eo3Jud-+kIwjct%Z8UDXlbX-IQ>`FqXyY}Wj);T`>{!LJj8 zJqEtVI%vd+OJULLO%!O%d_bBcLr%f#$Y3Af(JCP2=;V~NmMzh^?oLIu0o5)9T*iE! zzUOA^C_l_TxIkdtf~}TKfP4YIQtga!l+VEv!p(y`;R~xe%B9Hi3m9LZemu&&CF*Ao z?ITuLmuU!9XK`4{VBFzwpPmAs_ZR5za3))83jwo{`Y*q6(^MIQh_TW0xDPmLSpt@G zjLT6eL|+YZz+P9Dav!{w6!$BX2mi+6zmMBCIdfjbhh@WN961oFFug|y z_#2dul4CX~IT-i>qeVCi`tM^ITR2BiRlgNiqIg177vFB*0a*z_TU z8PBdNFe>(;H?8wHK49NI4gu(-7w;9{Z$ts_2Z!#z?wS|;I5}zQ*LZx$&0n%JDW+I}` zqwo=koSg%r@#<05B$7w4u+!Ql-1K9U>o;!X1Yux050yT>5e;JZVL--NxxXuyT(>aG zuKB7gRKXlWWB&Q-EPfB0Z7W1Hl3_)+x#TjD1?bgo$3s;hOJMAZLRcmn(ByXX_-Vsm zvG;Vkn5=Zo#HJ)6C#0JanL2F-@SC0*mqqi=Mb*l1V7bhO>`w{A3a-@SJZ@01B+@4l zSqxwGn*6p2P1Dzbfm`GNE#P}2>TYc%0^@uC{RO1ltK@GU54iXKoA**P%ijFUWP(+W zPy(Z)%RfNFu^zH>*xDC-DHj7{DAS*nuqEMTp%a_w$%C{)vj;K_>c@l)xeDT4V!Xn3 z7>0DjB@Zu2VCOE_rgseld@p?6Fa`#EAb#DO=6Yr>kV#I!bHuu!$(X28JmhmeGbv(c z2tZvj*;{=!hd=Um7@%L)s(qlGb>?DF!)LUMle*87bp=SQgt7r}Qp>jqpregDk7*XE zSYGS{^o~`K3DVn8hI7Cj1Cu?sty`7+(KJpstApN<3`HppY+Cuv0?VE5VPh%?MCdp%Kl?a3aZcCN8 z>%=kh!8LH|)c|P;aGdqpxA@$Esz0W;w!rl@nURC`8XSQ@^BITN=)7b%o1-fSQtkiB zVXqDq1t#sv#F@Ut#kTTEqe-xK8nHKrPb2uRZsQ36B2JpMLNTO0slu=Gju^FJqH&rb z)9Ww>Z9t(^m}5{2kp)#HH>Lp+z-DkaWES$Yhmera90o(EP9QeDfGY@RXuU{9Cldes zxv0UO=DiHLJxq6Sv&-78Pz5OGEl7XEgzhzfr2`R|yAo%Q8!O%Co;3I>9|yvw0@a@* znxD*?1=yNlFj-Muee3j&_{d0q)Zo>Xl{b;&SdJmKg{^=$dmIaB9t&wbeFIA7s(0SF zPuah}{~Mt7^!!Z1!-qbAeo!Ns;{KP;lG+*r%Sk5LHP>BEgQtnDHI6V8YZ`{a2!+mF z&`CCxl$_a~+DJVm2{w$YwtMH85E0m+z`rl_t9~pF~bXk zo$tqOr)ZXiF-G5{^jbu#&2wS;&5yC?B>rzOPc=BZ%x|7QCwqB>hLWRPLMOBJ`^Z## zbiPgdW}nE5E;En89W*{}v>-9ZXvcTImtZtn?KN(H?(uF%Ty%FHqE6?RX3RbTf25G+3|x`6;{kKrIs}pcd)`M z7FuSE0O?WQ6PfMxUQv7Pja%7)EZOhT%%Tn8u{0%g!?U+dvjg^o9Yw_W57_WwNH-Dr z4<5XM?zL~1yOf1Ydz{U9 z4_^J>@a_!_Fxa@~x(I41!pR=<82<;=LTiF}LQoSVjPyqp_0hu< za7id(;o!ERn66^6tPI1giqvjf+p?fd@dY9(_Sx;KcIE;)f*X+#lBmurCHbMVy|YU( zzNAkyC>F0706p|MpiI**EozL(xr5(VzwFoZt{BvAfWJU1yc3qo|NN4dPn{ZDH+LXZ zD|Gj!`)4?F9?~&r>9n0#>}E=_*_M-8_9WjX%_i-XO{U&hph2MCs76*nk=jE_{nwm7 z&kcG0jhvM)@47%IHvE2kxMb*Rp9yk3t9VrGlziXttGIasqIjq}5U#YPVd07FiJ9%N zbwu02&dgj7Q&1SNI%dk1g#C$6PEDOez*m{sZv6o4%}Hb*yNw$jP8v*lBtmH#ZsIW} zc@dCkE3-|quYqmWGbdY=5u=c~O}rd*f7k455T{*F)>mLP&MU_ORfRiEj(#BJ6qLNp z!fACvM$hsrl0VkcGzhS70?1Ut{Ja!w0YQLSs@|sft$M`iNO!S3W+N9og@aYAiGv=H zjkMGKqMwWHGTKiW$7iEb{fq)3O{7gtE&^<*1d+Zt{NDv|szCQy%-hzqCj8mDEzCAp zDXe0uYw+G{ru}p=xI&uBs#!c+WaS#}#l_h>vV(ss9(|i^Y=H(N@GU(i!^|Jw`u+P6 z)3NT)mlDJ0y>9A!->><^$jUacqswfgMt1Gxok)z67^c_CYGHUg2gch4&Vh~xez7|& zU%!2u>$Y;?#qoDf6X=0l6h+c|{9T@{Ey%U5JrR&sO#-|A~%DMb~dR_Tt}c;1sJ zI@xZ`$o(FFp6ZwyJ^t_;rTq(tLpY5%ec3ey!;j)opZ|&DEQq;3w16qd3qbq*S6&c8 zw2VkWxWwS7=&h95nilyQyone1P|ZV90Z#WvYi^E?adM5UY_YxH-H53{j$Y>A)2Dmm zp+S+kfbX@ibIzjEpoON2nt~$a(4#dJ@c~?k8!d-Gun3~s#n}%9-=vAx^7$)?deHO| zk6;s{Ib@>W7ZR@5fgqY&eEg@~9~`KmFh2B^ZYb?snPJ)lEUCZhw*L#D1=j$?ZQZoV z3>)myj?EeNtEE@wPu2)#msTN248dry+bI215~?5_As@9qaiW+1j1d7Zr!fp|J|Di! z-&SE~>`wY8@<=gM1xGdPa@4`?A8o=5ktgH(w0s<}(bA@-|;er7CEdA^^+s@^Y}T2EKnGg}>;h;+QN+zYjjb zL(GFL3;p*)ZIb8ikI48~im#=yrOFKqyN__K>m&CeVtV>AX@T$uLCvwU7MBwWD&`=# zlG(erVZ;Rmoq>G${;Sjc7|+P@G+dhKR}qGT`8IZ$d$0`+{v}S{=YKoUgmDf@QLmc{ zzq9keemro8=V|yd}O#)ToP^vda~dg$;+cOLIn8UuPOvuDqnAIq3e ze}<2#qY@hfLk-G(y#^(rJPX|3W#7APeLB&)dYSEkhUxHo&DF0`l#`1oU-%)Y|zE$H-V(;~{_I3*YXB2F;M*;DX{uIUuw>JD%w6Ce$LVgeqDk z|J)^sHd6!ls@B$%+#_N6$5DFCf%0G(y1@@6e9c86Yn`Lfzjfa#WsxAV%y}NRhBdbL znG~n;rd>S|VRos{Q5uynjNiO7C0UdRLmrTI}6q|%p! z?4uT!|5;VQ1@WBy#Nq{f-p%>PZP_gE36#i-?2psd<=OWzER1L68T@Sok2Tw~Sw5eU z7E}aD`!|9_qo*gfzkm@k`s-H{iq?i!=k0#1soHT1PY)OTBrr$wE<`?oCXuNsJacR; zdyxGjyw)1YcCyfI`g5w;#*D$QQ*=9a*iTG3LlElN6#20>dzjJk1%#-lF(bv{vz>Mh zR^>9#prO6(=f`&T)2An+F7uP;FLplXp#Pq^k4-qA^{9r%yQy)*iy#smisgEuEAR`A z>IcpbyS$dn9Cu=m;RHHWncce&uGh-Ce!&E7bYB(q2@2xHWH~DYhHEU>!ra_%wCllj z+(KLh&! zP0?VV3Lh{zL!+1YfX*yAVEoTo&~{A^!DtlAphvHW$$ZgBA z-B+Owr4-b}pyvU%Z{Ke0xR28B zb zFFR01xd!YiiOoZ`H8q#;sTaf2!SbUP4hLPkXLVK8)25REPdG8NH2rVw7K(c!>|2ng z__DCBsqw^nbZM@=H-s?3S!zFR5<=vjQtKY%+YUOwJHau>Vzz3UVu;uX^qT--o zjH`z6CHr;*py3PY{0w5oJCsv&RLY?0dt*_5<56IxG?O<+wZ~rSQ5VleT*A;rsg?k zOSbG%`0ReGz#MrnIFh*&oQyS%OX}a%KikMg{Janz$!+HMX24t#IVO-MjkfH@iyr>!U|d>2t77EQGp&DZvBDS+AP9DIBGa$+ZbznJasA(`m3% z`gClN+L}pd&7uxS0Z2q{9m@X^)UuX`R-aljGM)C(D&a1Dht(|~s zc051lsC_tSntT1&u=>?hcR3aO9?owleK&93oI5AI2*jj$x%8vXiA&*4FI-{4Rp_#iWZRMrBT}6hVG0UkWl9N(3UP?#*o$o10}mWH5X>Yv=F1j26FSzS zR3Ltpz1yPGvvOJwM6OJAyFD9(#tx3Pqy#;+=ff6J=#DE#@j~HpEGp1!C5S@Q$h}Jc zz*c+xa?i3zeyK7>VWUU@J?Cd~aOwM@Rp4!z%?{)J`i-oUK~-t4bhg`UKrQ`{Kr!E` zc(QBbDg7>xV`=5WH65VQRh<7upE(ffO>=^>szZ3|1=MMZFVOdrUTS*zyvXV^CV@DY zs~3aGvg*q?O$Q>$>UM5H4V@S6;-!bj}}P@*Jl#ywvsdn2ts5^SyQJRu}xytjzk1(>a8Mf)y7>#U~=M&U(M& zw;^yYZ1rC0mJ}BkvxYsAJb%VL3D`#NZB1ZO7SprP*gCqoz4*ZWG@l|q#@7Jdi)MmC z&q$eQ&yG2$e?O$jOH&o#`ibW0)CzJgHb8M$>^Of&#~X6xsv#%2QTRU19mpf%Tyt}9E=wk_WO^|dQsfMP|d z?TUAv?$ZmAqKBog^x}%7kc!atz9HVFAETn{#HTRkyJM^Y#1yrw7t`!tEd4M)Bq2@Z z*(}~r@zlgWjA=ubFPmmFZCPol8&plPVPWT;6ZHb9n`u*=aSs2+jrY8)N9chcqgYHk zYutGRHmSpCblTrQ2=f-|ASpCDucz6G2}vous5-19lKQ(*$qnx0J%T)BOQ(S%XzItg z6EXB2=vqK*=Y5QS>#XTO$5LBc8_@HsLum^-Z@fz@+Ktf63^G{gzJw0%9(q7SD_!-*8hPYDfVXeSQQJhDIYl!)FS~Xb^wr=Gr&`a z?)(5q_Vx;U+l4iu7s0^6T)!IYSpE7u#t(?~0w0Ade_5^f{ngsC$seim-K(I1M(I8V z3$nUL(S6eAF1Eqw#&`67n?fJ#hN@kgE$4jSwW(mrMbOGaL9lp-X(wH1Jzhv|Zmt`W zc5YGA+~hlvTF!YwbjQLwq)dm$OdHYE@B1`k$G?AnG@6)iH99eJUMr=YW5>aaL?QMc zF#SW<{*$A#F}3W4hv+U0+UB=)`oRg~xj~jCwyVWMSKasW%3V+zDgY7F%C(RUN=HK2 za~9a4oRG&T0PhdvX~xjEh%jHkQoclnD%{*PUytJi;|$v8L==jjKYu=;wgOC|DP+-k zurjVeT<$p>cctd%e1DLLokBC<#dF)l)i!S1*`T}--+3a44%dv7JHy|;T{1F?J(J#3 zXdlxv9Lm*Du~cqcLFywIT29D0ypX$ug1g_N={~M-6?H=c6gB5LuV?aSTzF=XH3R{h zPMOB$N4~djf4Z)KAE4Va@(E-UV!TPuI==NzNC*>E+@Ge|sKHI*qT(26M?#qW7H}qP z6istPXF3cW0%|=nv<#tsIeaP*D8@ZCxfk$g<`fuU{k662`{g9|N$4|16%yFZHmc+M ziBjuxKD%JV5}pL)XyeP5FY}J3?8`i85rDt9y#99v__`w;2Hc=?>hDfLtDe;O1{}3W zHIw<~-ad_?;a!&ZWbma(Wo!TWk(oV-ylwnqwH{5HcIV<;`EqL;`CU!rrHT4o?+(aY zy6#0w6W0K66vUc0;1Y?np^aK$&ms7U3Ov_XjH+PuoQHM6+-h%cD5~^HR85VD_uzcrcMxKZt$aS|gQN_c z$_e#Cls}0;PTQIjkFU#nCDc&A%WH4bW5|Nvg6;POO+_xQ;;3_Vv6+E;61Yn8&O@5W zyNMJOHkLz05-bb zoj1e+PvzPm7=y8d?bA4RY#1%vut3{xY3n0Bh&YYEAQVH{Mb0EHo~j5UIm%_Got^O|O9RqUxm@=%U6gBD127Sak6bElg7$8UW` zgj&QxC<6JwY)AU#Urb#YIBsH`@&*r%>8<~!3aDe*g!m7*Pb%+yD$8bbFGlx!-SkxL zQ?CPvi7*SDhIWiqb%e#5qX70Ws`S{v9geZ|-i5LKbT;=s;qKNd=Ua z3bn1El2JY2eEs@$cf6Bp#5Lo%$B(u>nax}mB5&%O*btQ82y1lECJ%z#8{o`jedP-_ zct`rnNJz8|&FA3Z-(db71wLTcbAyL%L*Kw8;AA)_(#B1mVyxvetmVhM@@^xDpL*PL zleAb6D!aP445BMAxV}@inz~`(XDtR|m#`>K1{7(?pZ7i!N|EAAxHcIeGoD{CLLQ1N z5aZ}>=7iby7o88AJfnU0BfDX8XPE?zm{VB z9}*H$=w_s<>i1z->*@|1a#hA--@d7(-;4@YaPZ4k*+W%;bn*qEb_(E!JBB)XpXm3A zqfh8b-yA$`FuL3%6ZAN7W9wV`mY>JpD%o3F^8Q5;BM-2L4ALRb?&^0rkmtFx;rZrA z(*_T7kN>nvzp4MX=q1Q@{hv=P$gLoNqsiz%>o!S-a5et4OJhZSPIh)rYX>w1kE#9r z`0kxXhnD>68(FY1ZIR6RYHJN268a-{JwR)*?x;&1%Rl2~om1PVxbac|4Kzrqb7)ae zyjDd0cQ77kkbx$f{)4EfJ;I(3*=$}nHIdl%3QPwy!A@iW?oJO}5wU=mpeFH!v^bI; zj#<3pEsI9V+Aww-fNZiD9mZn1ScB-M$2^ZLt#T<9J&pFx6bmOOCxX^$o-5=2i73FZ z%Z9jUxK-R(Q)*tjnU}K{Iknw_e1)FlDLv^&xyGUk^d3lJMLYsR;>hu8d@aQ=QUt%s zowuckGiL}0nz?sHN3wS;;rlBuoM=SKV7hx>N&o0pNzJ(ZVJITsKzc!X^vqBwBi@+T zJ0p9<${(#Kz3CeTTgFYGWCyhW8`vO3Y|aYPuO1rn{_S%3?<8T3TA? zfzV?r!<_Yotcho$$wEUY*I=5p5PG>+26@KP5|cXxueq$%b=d9?x8DOS#n9<~tR_%(wRC23=JPK2M-X-;rbxWlw@aO(w- zO0|7dsU1LIWNO#YOi4?dLS1^XEq)?cy7$uL>&CN??vr@td6%>&K;&p>A?>XXcr0}V zOw-5B0hg&LDMf9|r_5(&zoJWE_8>^z6<8iTa@&D&S8v^prb&eE zAX}qqLnQO3euLx^_R&*N+$&!#5aE1~(NnvO)9g=wlE;rNc{2s@Rdg(sV1G&7jkqcg zL~FmbPcboQ<;^XpdWkVb;1)?RSOX!BhGG2H&6~|(aTZf=QtYHs8V)q!lltS!nC2AK z%-IQrWEqzku&3o8KWu548p$6ntZ6SqIfP0EbXR0b49l9fTM4UCT`BZt#r)DdS1%xO zqjkSt_nUL$SM+~Oe1?W*5;6;K z$Z4fM%afpo0O1j~2x4G3Zk`TFigCWky;tQ=g}B9j^6f_^i1N296n{_+_U9bpmBivo zK|VU{z6^n7qY{0?^fDM#EG{hU=V_q9Fb#P6Fq*ii<)XbaYn$G@`HI3VXjX;%VB7(Z z!qzHk6gmS$hzXKg1CyN=vrW`Y%6E^MA3RKZQ({tlyYurUo}JE)$4}xjUHbFMA!a!X zi2u+>V~=1!96ABtvG?I|trTZJ>LLBu8wxHz7*M9X-(%(nAPCakeMoc7pC5Le| zKJ#Y=1y8KDrG2m>k%_5~z9kEz8Kwl2CnbY-M5{pra~XNh;W-#~I+l(sc^(3;@*AbnThzRgCF6G$S7?F?T4rxrcPf{tuaP*}QXDN_ zMGYILOdSoS>X6~vK07oZaHk8%0-wjzh7e18(JAhhT{>sk| zt&WcSI1fqPNh$bU93EcM|R}IHbh%=L|Hjzxbwpmx(pK#&BSR^&)h-T5;#)jrK?&_^{bbQwFJEFD{r)D-fNfiJKF|%- zu+v)VVSE@@e=R*^NT?eDGzs4u?;k%5;u-lOvE!iuM^u}w;5@p5G8PXIc-@g*RI}7E zcA3YFNpF|Psd!2%s=WE$H53PXcfY5gaBqR@A1^0Ki;V-+%=;fQ*|<%93;2*sZ29bW zdJ*~iz0~9l7v4aJQ=@WM2=J?DNyTx4^g~ZpIC>D^>~@Nr{k@EI-o$Du37poj6qQjVMUcY%mn0XF z7zj@B2cCh1J~fi#wKVY`!f>G1I*qaP$S4fSrdU4 zP4ESU5FanV+TfiLFlGYBiYZ7B#Z^(sB{(5ir{zb15AtMgD~PxZ4xP^{M?Ci&&jkkX zIDsH5OO^la@*n?=Gp@hvATuy^{zWGl4=TXJ2qE5=h>{obR$Jw?qFGU%Qvy>H6Jk_L zW1TsNeZC?FT29IFi;JJ^5UmEKMdij4LhLsTh9@EasRsBySk!y<9eU!trE!>;grvW< z=oaorT?CE6|Fw6f@l>vF``6yh!!D^bQE8x2g+`jJB$b4O+6^k9RAd*SamPZFs7MkK z4YDPn$S$N=M1!OBe%R@?RItwG*Pg>;Z!|rfYx^320E(snh4`dnsKME?|EG&z~Sl(&xd^oe^Mm%9rvBxojhR2 z*O|*!Xjnb|yCn8#UQ{7NTk!e@bsyyRcVUf+5bApz8z>UN;wG1{(9pjqq9ewS`Ig3` zmnO`iqww{9=Zki%0Hri(j#71J3QycM@tsAxmHQ>pSp`?hi-HYxrw*zZ zop^QqvxC%yrKScL*6-8ntSuadC;Td(RZzpzt4fufqTg}E^O3DYMAOH0r?{)RgB|v- zikKN@XlC09&>+nk`d<0qV>?5GVDPR97kWYl^-fYIV)#yGrhG7a!V`lO;`a@TRUoF4 z)Z`zv0X{lq@$kBfMmUn>PB;`vYc@ihlbwJ;B&p!Z+p<&?R`<4XgvQ%&%DtxYRe*=o zy-UB6RM~g(Mxn{_f?NrBJa68-LQoH{tsQUNxFJxXXc^Kvj34e)BnX^~ox!#8dZEG} zBx+~IC(wM$g=Ecg!t#jN>`q_V8xxrh;Rj;T+E`3H7S{a*pIl1c`mx?O)@|f85SWjD zq-yr~_bJ1GyPU;)9rg81Wh%;FUnllg#z?;OYgWj?u;qUfV;8_nPq8i7%4#X7Iu&Lh z*o_0XWO(2`n?&{nRi1*ORIhDo6{l|9|Lme!b>51-gGqv;mL(W|58O`mTb zp;SO97NBK9Oa{vn>DK4-QWuxR_KGeaZ??019Ze;qCU#Pvv;?$%#sHt^(SA~)varlh zkdsRhQ}_DHvQf<(-TB4yTy6k4Jb>p|W$AiUxGunw`=xF1=hXPUHu(^KWyYiGi+mP7 zs9dP4+b!oW2&p@RMvnAaZDEmm^3T}UFG={N>qkNhcb_{tD&MiNOU%>tZ1~7~0xlTr zT?Vq30>ojdR8W`ioKY~Yyu3Ul&6p^+$Mv!}v=GCxHr?~ykxn=9_fZx%Uy%GSL-cDF z{=|-+?_dD_i&c+M@H+cJmlwINW5PuYIW|75`W5PrO9*-{&FW&WyI?^vE)a_NHp|IY zy^PB!u;bI^o>>r_Z~d&IAS46Bls4v^KWf%~8kde~r?LK*lv@XL_L4)EUx}?GLKEk_#=&|r@HP3 z^^(2*YP8@)uAdG~o37N%@DX)bp!x9--t${S1^0}Z&GhXxcZAwz=+L1aY4g?xyJcq^ zE2>%UkI=-QA|05u%lcSzA<(H|gzaeFl3L0XHy5b&4Xj#{Ki?Zk@Esvtx!>WS)o8&~ zyi$q1fx=9nk@_gGkU4S+!7KR837kmLM9Cm+H|agO7Wu|go7iRKe|D#oCZ=qW2ynfG zbINm^oUm{XLg;*leRmD@KB$crRTP{Z^yJ+{G~t@`n50g)1TPbEMS-o=jU7FOM#Rq* zHrXm~fM9;U?|iC(Hh~{c^%Q1|@ucEgC-&2QJ6;fdGbax{yl39kUqZ7c`x|O^5x03g zA>k7OD9ddz$bL^olDdv}t4ZyY4{`580g!M|<0dDUaTM3sCGg9*U5#1$Y&R50Wb%Su*_X_xr}? z<|q6Pscl$bsf(tgV=_5C@<(sDi_~vnkq?FEP~4=~(em}}{(%Jj)mNP{xiDIP*f3>f zuYD8Nedu;?O6CjQv(0zDxuzifuo1LB*YJW zKHI#c>FBFbr14;pPbwNoN$94v4op0)Zkl>$4T>iB04rHJG&RF^dkue`|DfJ;n&IO$ zfIpU<9+^j)RLDeU8!em5b%-a3isnpOAk3>hIPxnQ_q>9_!UW8rELN#ao0ebU$J$hr z74oY!y;oJTk}+qwJ#Q_TE!(~WO#q7C7QqxUXKl8%TT>=h@Kv*He$fg%wC{MMz*Vo4j# zIkURJXM-p>(z?#VywZHY&DA{`5*iSuP~$f3M~%_2 z%daOa979OAPN`wB96i)Ro*SgF0ZMuwyknoHA__rPE4j{|{zPw(-ay-9mzQYRk!2KU zj2xZ4x%82O?5Sgy@2Z z%k@Qhkump7YZ;jiLiw-`bcDQdn120~ZIut$_KVo!vEfWM-GJ)LKTrD)sda+~FijA?v)F!O`v;z%D*}fMyt$W*itgC}JpSpzr`#Fxi z9#YP)i)H(^Sh{*HHSbaL>C;<6YkNw!M3gF~k)7Q-TjjW{-uk9c3tM=y0+6z>dkDnn zJ+!8a;9B-669nSMQc9c#d(yuNb|_@%JT%@#is!YNk&^l&v1EYV7cZ4cr!%q%?$*yJ z+DUVKpK~Fdw<|`B)JWPf+fTHSS5s3fRy_av*$>K~@o6O8#heEoY0zx%N7%<52qx9` zGGw!mFGYJ)%>nh5m8FU155x)%u%dRRi8ynpJruhW6(0B44(CgjVafGEej^7sg1~A7 zXlZrP#>Y(Nc8W#1UX1*sn#jiqp^=HR3@r`z(sOs%juoNLLVWj`3q%;bW~j+1J+p`; zI$h$j|>LgSryGeNyk_xvwb`1vug$I zsP_B(z>PF4>m{}~Mk=Lu&A3Z+AFk-Gpup375XADGJ z3|mFjz(bVCKAJoDfx_@2o2a24J-#oph@h3b=lQl-lP`@X=O(s7lrNAdH9~@6N{wY_ z4sB3eXBsxMqlW3!XWl8$5TkCaLmV=R?DSnUr((e$D$$qULCd5jT#MXUxPO1==(}^S zwY`p?zuX*~m*gX>d)VNU6#1|)Y=SMI=e9>OTg1Fib`r#%(?y5TQ5+=zSBoOv?(I4a zkcmO?lg15rc{{f_Oj6TSC-fWNMT07!XdW|h>hI&U{!oaf zEOdqa{%H9$8Yv&<{GjDQuIl`REtVyqVNc6q z);d<#)}kbg96VpU3lgD<;^H*kxe%A0#BMh`Te^M;!XN5F^2pj=-FF>MnZ!N&5q|b-M#kWFM(qw|nbvFyt_KM2N z_gglZJI(6wHi+*%WPIL6!Ybh>x6$j+xJb=GuEQet85M;*DUYWHX8xM#W#+hcT`~&) zZKBoVUDRksMzon}S!l4Anm!g+%Dd=hMQ{OR&}6$m5BiLFV;r^oRY{eX!vdlD5xfra zx~X=)wf=5>h3oB?=TXUxCp@I!2Zdx`A`L0nM5WQhNJZtvYJH|Ss5o!mZf?7Nm@?5I zQ~9D>4DPyv@~6&Q{G#aCv17x2Nft}a=^_`vv3EioH)P+7^DGtl0}3=Ap7wBXjmrLZ z2UdBk_Ya$q!1}95IZ@~haCUOCqr8~xtTz!sq-XD}{*El-lapp4RMSBN{`oFj?N5-b zHQ{yeom17doT`+PEcZ(@GZJ;q0*!0*wTV3;_B;d_$DOU53d+W44!!^2K@qia0!~Za zXUK7F;q0DU$WOvv(zR~lhUW$qMJmkRu6@QA?dzzf%YF(d6uWfKl5+qWk6`NLz1%tT z^RNm{8{K)*t+ClSaczuFyX!R&ls@@uK_{dHSH3oiCMF?UpB1^SidaQaw|!T4e}BP5 zo?qpcUqLkMzw=wMJdxmAz=L^DtD@i#qGyzYQX?vR!*Q~O8Fyfo%hO@spJZh@;_ENu zp0M{0pUVyePIEm$D!BfWR~xD2OzK+^x@hTEl2bdyo=A2@F9S>k zylUr>ElyDW&MutFJSv6KF?-oi?l1)f!I4F~(mz^O^{1(+Pjhmh}cn2)F%d+NpgZZ?PMjJw&wA|x-x(Nc9DJJ!xt{p97trJtW<6s#2| zm$Do!J(fr0S076p72K8rw}r^B=T@%LR2#alZ)gW|!dmkDQFvnd!9JyM8NF6J49Pys zWuhP#m@Q}$88g`Z!w3ff4nE(fC(*=SG9Y7c-NtJo0N5%UX&zgE%N8uFJ3hL1k?_CY z%*8Z5>)x|x8tXwQ!mboTZ+}`i{l}17tFf-U;&j9B#EB8>SA0OjN>I60sQ{&zm?kK& zd<63d=x=lBk!MA`ME6VHs!$8%<{>@hPiPn-A7SeyV@`&Rz6D&+j zj(rVyg7UskbGE)0p$yyMXZ})rrg+;i7%jw)arQ2~0k~%jxF=)!>GHn5%X_IBMZ8`6 zY5ae_Z2n)p_8OvG*kX?pc#dTqf+2&7*7ifP2OeiNr?S}30>KX_kIKP(x6qC75lIO*r7GHck=+!Uc7ce#x^sG?}m6#d)|OYy&s zF|gXYbprJhANNg~v;WYcQmmsvRj}?$TaD{>JSr1n%-~sV%=QkssylTRwaed_o0fHS z2FXxOoos zx{#eb@hn^Z*@>~{4^cAQ?Idf8-up(x;{s|{1g*mS_c}jFZ^ecQYx>%ZO%hO~ih8Iq zI^SNly9NCU^(6Fi8#6=pIY#;B-b?MyepDVwrJxI?vI*{1&sd@R&3ooVEZU2#;}MH$ z4M3RPjqgg|AF1_o?>m1Jo1^Od^~n&Uo_6Hs5K65%jwHsvR?`r+8TC=@|N>{!evMpj2m@p$n#mI&9{?lAcE3ynB z5a9{Lo=E~#IPtZM4{~-(=!x(xrlx*AI(}0L@p^j6XUz&fN3A>bRicG<7cwN8GNqCP zl%mfcFZ<%8u8<8lKGYkns=B~as47ejm6INYWXQ&0Mv$MMUr@ILjj0J%s>TJ zk}4jsM&n!e0p?nkWbg^+v)^oh6zh)!nR@Rcf=#;fB|I$Tz+g83_5UNQeb6D)@u$YG z^_;YGV&dX+hO?tz|0+%e+NOyD8{8=}09JgRtr9L1AVJ=acV5`B*Qh&!smn{b-xJ_> zYE`1pFq9$~zoT_!nP|8rP=SJ5XOnD<^w#vYQK4-$bp`%(6A6#sGHtxRr$8q(>mg9QeA!_lRRm2)UPsc z*6sz?+ofOkOJY@{Q(M{{4}8b27oK-mBoo6~N87{OmVs zddYCCQF`whGn6f{n~#lNCyk2N$wBNTsjrE5N50f-Bo8ZyMW;B`+V@}sxJ;TPc{J{) zJ)*R&gGHdbXscQ7?0QhRmoEFWOV9+h(eJyqQa7v+2yx{XKfp0ik0l!-f(bEos_v2q z2w#ftj;u2%bTJPMGroW&=V}sJ>cfy@H@rM?05UPbF$JUpZZG?Dr*Dnc$t9r5iImMN zQY+t;&NHw(>GD@yLsvx%8Zt!Uy>mr>@1sZ4c%~&RpDiFc-ynH+tjc)Xr5QgF()t8< z-*w9RC*P3soP-lep}?<{{N=t@sHPf}LDlQm7rW7c{}w8+_acrlAIuzKU!yA&P%x`#~M6-{CJw4F*?#^V#Av>qoW3jSGVT$ z_qXrK2%xqgWm+B|I8il*Gm?EH{NXE~H_HqoM5|`iiq;wTbvkM!qDVElZ@#xW68m!@ z%_TC{|4~K8xTHcQrkFe#4Kx4{AxDV2NAOQPMT zTEiydXnCr=MzbD$5zc}Si-?cGh!=tAS^ z=@~+);$5vCUw6{Y?=-f)k2%ifM&3Rb)E#r<$6$3c0BX(5=K1XZE_)uXz_*SRVY(REHvQjdKDKH@rGA@bgWl^ zRjXi&!jgeDr}@oka#1s_Y3zVf-*2b7$>Lz|DV36tt*eev{c(wMpl)hLD9(&-m$oN# zGN&QfAC-CjQmJrsB;miZ{kzxaQ#fiJ!1@%JzFZ4`j&@Ftjs^gTKF%O~P`6+Ld>J+1$co z_W?U}xm@uhCkVYkg6l5Amd47v&Kck;AwIg|2$x9*2|+45V0TxKCay=EpJi_w8l0)1 zKYdm3OB3!5Do`LKW)b)ZKN=9Bz~(Ux8NZSeQDCnh2uFGy@}fHuk;5YPCu>%d1E=J9 zvzW7L?l#BSKX~L=sLw>_F>Y!l4Ad0_6-r7roUSqM<%e;QI>G@DU2zb@$@O32k)80~ zSO!psw4ch=l2F+xx7D-95!#QjwYQ}=Uf!!$fs>7|6A!R^hjY8;oqwMnFyVky3e7Bmf8AO;_GYV=v-UqnFRMy= z@e5#zcr_LZ98&MWG1V2PNTAj@Gyw}y6t~Mmx-f9c&?hPi>z?ljVRRC?F@40Y!~~yF zxz4uU78a*$Lu+3-S$P$MUF0e29afl|_vNTN=^feM88`_JsWY&xue@F*NoZ>IYunA+w_nTrLHgJuXlj_+hUu6b(F1I^3nB~Z0uU@% z3-mxv7SLw+wTc>KVE8-TLWO+0-=>cY@BT{U>qDq+jk$2)eWyi@umAcXA!_=DNahV( ze|RDYV}`PSStskYgKx#n_)u-RuyKGSMpL?P1rbn z9G=5$6&vz)gh0huJ^4ju;IBQ^LXQ6&J^O3)Dc9hREmp6-OHo0-Yu)eLp<~N92WqHA zF=RkoVIoECLW29d!(6E$P{F9Hft$B&3upf~K#bFQ@RnuUw@;x>+o>H*Kej0;iH9zh zJ#}N&DExc$>^U399X7jt@*xHHfL70;+csJHt7w=i%xa?&KYRMjIX0CDA{+&AciBZO zwsXKNG*Owui5tqSGrW+4{m!)TxI=e4MVibT;9BFI^ZGrZZzgBMIdEZpMYBr06#pbi znL|y5ur@{RD7VkeEsLaDT!(L9-~vwlD=%-&9?n?IW_)Afjb5Pq#wSsm3GKUT&Eci2 zy-*;@EY?Ec-z6^&a{PfzkX($t5;8DHV;3?kDDj%DKM=pg-rwK{C^(ds9 zG(&=TP@U}V1mczvKEKG^k=ac$Dqd|wgOcHzn#)F=?q@?YG4_rFiHV78dYk1FZb6+$ z2v$Rd|Ebsn7&0W`gD8LyEBXXIx7okXH0;^4#|i|-nA%5wo#W1gJ{(g3E$zjK=kI8+ z^`BE3&lRIpNQv^_varq$(l98!tZ?W;B0a>FMyqY)BA^r}5 z6FTvVC~cy7jAfUEYuE0Ly_`6%vj&FR22|;&Oq%3}ZT!r}#zy$Q$4Z6@;lFDnG_HzgBeE;0?i~!1+uC7 zAF(X{uRQJh8pIQ;qDtf+G9|2eP7XC?i)d*Q44!M4yyHg%syF+tqo1*JxFwxfp-${f zur>$5+F3k*jG`^lBG4RWbD}1=A5$hLls|V1DuUPF_uKw+wT>G>15)l6k>n@NZ~n4v zaLa@GwyIZ$=3l3_i8p>iXnh9+XGeXxGBs_=oI(liUxG~NuPK^Lvv+5bGC~LG^nfm_ z`>XK@iHa4MolV)|y?ggU0O1~^Kr2#wx^uas<0MziAVLk#S&_+!9*ha@KdVJQbCX6fo`}Kcs_P@sC|HkZp z4TjABXH#QILc5l$|F{DAt_oCD**d~JipgfcHV<0z?0Dg4Ey7PY@p53WVC7cB_EQ;Uj=XH!~wn(W}d)`N8@ z#>R{=I@=ivc0pyZ%AT6#q^_ow!7YET(j8j&=c}V3W)>akrw+LAxwiH!0Gn{?K>NF! z23jh0A6tG;=H$OyuS^@d>-3&IXO5ABgL>H6vss`X=Qv3v5_RFM=1_j*!Qh{JqKwSu z-~QbLdR*kZr&ZU4&owneEYGs%nrl{;KMAk-_*Og1O8vL#+jMl=YB_$~1-fwR{-=YK z){7qrABCJk%KcPulc+327EM}^Yq_-D@ENxr=uBTYZ{8UOi(}3EtG|9Y>679jAh1if z187oI`n3DO8*#^O-+$nH`0&6TiV6w~2recBrDqnl+}5dy|GB;~#GUXGvm_F_IY=a- znO0VSFxiL_M_@kK6UNO`UR59DhGImuDLcN$=ae>-hX;umzv#n<1;kTB*srO8?@jY* zcS>*@9;8oI!g1-^wyKTll|-G7iO4rh+ekQV9Jldz#vv3X+hJG$TGjdsL((aMk1D>x zCL3YBQRbz3z>Ip zu(27Tpy_y6Y4Bic#BTIGn|h`c!})2#t5zz-MGVr>h=edLJ^cx(jtbB+zgz8+I2gqW z3gqn~=)}~Z2ge|hy8{8Z4T6p|J2}|}fm0T@mg9t0L8?Vr>@3tnBKa_+d#G(*ym$7y z@2`?SVYqjGY1|kv1s{Gjg4~x0Fhl&RILrx;&%w7@9_ojgDpnhZr8QMc!QzHfsdeM~ z=Xuw#yU1x;zWnvlxiCRFkxP%GoYUrdjITLZ8=f;3q0R|fI-*bWqK`-fw2tMWSJA86 z4IMJXfV{PwZ12&B={;MY0wM&E)R2!u zY-)hyft7oJ!la0uO^rIX9g!b`oRV7Mzqb%+)mcOe_@JH@W}6H=of*xYrsXe!qzZ*) zH^)4DsIu}C(Ei~JEiG5{m_ zdT&+sIM`&Vd$-~bNg58g{cNYFf_PUJZ#jeCgLQ}zpNQ~S__P?F;&tfvSeu@*Q7EjuG{~SOWeTxefH_Ri4L7lrFTzt+r#D3<+LpfIJYd&N>&yA)i#3X3Cs_0)z z)vr=BgklJ;A=O>c=MB;d!HRf#c@1W5)>E|;f&0;fpoQU27D3vSVND?t6EfyDva%@S zyrJEloSluZH$7)OA|S)(W#NYxcnc#>1zf?!dx5Cn%lu(&Z7o(8=0oZf@%5c-9~vLU z6B7A1WmJm9%?~OPP{ZoLDh9L29|M|6j=I66WO5DqqG<>Fg`d_KW#u`E>BM2|Ahvdi z2t-0ClIW{d%cBv?nc3zmY>_c-tP?4M2$Y$2XGrBv{MaUtMk<%P$;$F%jF`c;-|zo8 z6kGSyA(5KpU>1_U)(#M-ntbP~MKs|qBXiKH6XD^Wy?gH>ZO4r4&g<8&gU7WDyQZqu zgZbC|w8 zyz<%4iV6$05yjUa_cNeR_J)H)dx4jA;vT8X*?+tKs2 zyB&dE^TkJM76P@?aAt;x@~YQfl+NpsrJ53@@H{oFm<`a12NkBm!o<;{oDo|oaDK?jfK zSC3Ysm0M;fczB2R;uIF%U<_7}!7TEs)U`|R0qINSyF zj|~>PFkqaARAt<_%(8KuKZBJ*i0ch`=+?NnTmi!;m5QEpDo?RWimTq|&z~oCI*R1V%w-u__}z?*LsBV7TwrOmnIXi{g}S*-4?!e8#lSY*yK`rtQV3v; zsmsaOpn0D^=j-v&$|H>Z?CtD=0U^4{dXx@Qval4%q^$R%H!Y#e4E+cgmFCywK2)&f zG|ZP@lTN+X7t<#u?HSH(U!NoWe7<4wDyGG)chTR98CRBzMO{4Svmq+z)LRME(R};% z&D-6*ooxBGFZ31;h<*K)+TG#o>$Db?h(-w@GSc*9O8V6ie_E`1S>p3ykZtf=^dR3@rMRbkwwAlOAT= zR@4MsA@wrd-mZ2Xw7m0E?l^fl73>$q|y>o z8ZRvyaG^D~au$h^O6?9`25?18-}#dO zxMcgpt-;;D$I(9Aj^!siZchGapUwkD#$Kf)IXEr-cU-KNtllOe*qGhZ6a3p$^LB1V z_4}mQaLzGFXxH)BPi3-)&5blMa&-KvHhtpHtDCXo-z4x&o7Lx%b_at1WO7tI+;Pgs zXLYRM?zp(P_FZ(nni90~7olmO(~d=LHMFNV>ttn<8Y-{r)CmlWP5RMEYZF;)$8Zjl zNR07@A-wHXmqbyuL?ZCh7@N)y=0h_+%6YDkLu>*h%3+Q{;+cl=YHHS09`+h={8`4P?;!$pY#(){ci)N z&4;fEhXyw7tb`(IpIGd|znKs?3p#u*`rnCcBvNqB*_3e!fNE3H7n4&5xA}sR)5(YC zf8h&ZkcfY{3MeN5x#Jv| z~+U%wvEj5MC7T=ikl^dQl}ULv6l&{+U=acWCcZeo*DX6NLz zNkJhh7#LF=V&zQZY&^)_r#rXIjYlB`-y@9Lw|pm~>I+5LkQ}7e}41hWixsJY2gJ5FeV+_0=}tqx9QL_17*I?8N!Vd%ZbiCtUu^0rn!$3JZJOk zSa9(_iR#}wB?zdjxx#{-=-dvICrCyd)w{i@<#0E3>-bN~EYl<_vMP}\n" + ] + } + ], "source": [ "from brainchip_utils.hardware_utils import get_akida_device\n", "\n", @@ -122,7 +222,7 @@ }, { "cell_type": "code", - "execution_count": null, + "execution_count": 4, "id": "cell-08", "metadata": {}, "outputs": [], @@ -153,10 +253,65 @@ }, { "cell_type": "code", - "execution_count": null, + "execution_count": 5, "id": "cell-10", "metadata": {}, - "outputs": [], + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Power measurement not available: only supported on AKD1500 (IpVersion.v1).\n", + " Repeat 1/10 done.\n", + " Repeat 2/10 done.\n", + " Repeat 3/10 done.\n", + " Repeat 4/10 done.\n", + " Repeat 5/10 done.\n", + " Repeat 6/10 done.\n", + " Repeat 7/10 done.\n", + " Repeat 8/10 done.\n", + " Repeat 9/10 done.\n", + " Repeat 10/10 done.\n", + "\n", + " Mean inference time: 405.299 ms (σ=7.030 ms)\n", + " Mean on-chip time: 396.036 ms (9900888 clocks)\n", + " Total inferences run: 1000\n", + "[Minimal] NPs=67 passes=3 latency@25MHz=396.036 ms projected@1000MHz=9.901 ms\n", + "Power measurement not available: only supported on AKD1500 (IpVersion.v1).\n", + " Repeat 1/10 done.\n", + " Repeat 2/10 done.\n", + " Repeat 3/10 done.\n", + " Repeat 4/10 done.\n", + " Repeat 5/10 done.\n", + " Repeat 6/10 done.\n", + " Repeat 7/10 done.\n", + " Repeat 8/10 done.\n", + " Repeat 9/10 done.\n", + " Repeat 10/10 done.\n", + "\n", + " Mean inference time: 404.907 ms (σ=7.040 ms)\n", + " Mean on-chip time: 395.646 ms (9891155 clocks)\n", + " Total inferences run: 1000\n", + "[AllNps] NPs=69 passes=3 latency@25MHz=395.646 ms projected@1000MHz=9.891 ms\n", + "Power measurement not available: only supported on AKD1500 (IpVersion.v1).\n", + " Repeat 1/10 done.\n", + " Repeat 2/10 done.\n", + " Repeat 3/10 done.\n", + " Repeat 4/10 done.\n", + " Repeat 5/10 done.\n", + " Repeat 6/10 done.\n", + " Repeat 7/10 done.\n", + " Repeat 8/10 done.\n", + " Repeat 9/10 done.\n", + " Repeat 10/10 done.\n", + "\n", + " Mean inference time: 257.312 ms (σ=5.065 ms)\n", + " Mean on-chip time: 254.591 ms (6364781 clocks)\n", + " Total inferences run: 1000\n", + "[HwPr] NPs=187 passes=8 latency@25MHz=254.591 ms projected@1000MHz=6.365 ms\n" + ] + } + ], "source": [ "from brainchip_utils.hardware_utils import full_model_benchmark, get_mapping_stats\n", "\n", @@ -184,7 +339,7 @@ "\n", "The per-layer benchmark (Minimal mapping) shows where the time goes, and the sparsity printout shows why. A layer's cost grows with the number of non-zero activations it receives, so on Akida, per-layer latency and input sparsity are linked: a layer with dense input and a large feature map is slow.\n", "\n", - "In this model, look at the last decoder blocks. They run at 64 × 64 and 128 × 128, so they receive far more activations than the encoder layers at 8 × 8 or 4 × 4, and their sparsity is likely to matter most. The per-layer results show whether it does.\n", + "In this model, the slowest layers are the encoder's first full convolutions (`conv_1` to `conv_3`), which apply 3 × 3 kernels across every channel of the 64 × 64 and 32 × 32 feature maps. The decoder's upsampling layers, at up to 128 × 128, come next, and the deep encoder layers at 8 × 8 and 4 × 4 cost little.\n", "\n", "The input layer (`conv_0`) is the exception to the sparsity rule: it runs on the dedicated input processor, whose cost doesn't depend on how many of its inputs are zero.\n", "\n", @@ -193,10 +348,99 @@ }, { "cell_type": "code", - "execution_count": null, + "execution_count": 6, "id": "cell-12", "metadata": {}, - "outputs": [], + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "\n", + "Layer Sparsity\n", + "----------------------------\n", + "conv_0 26.02%\n", + "conv_1 41.78%\n", + "conv_2 49.98%\n", + "conv_3 35.52%\n", + "dw_separable_4 11.18%\n", + "pw_separable_4 27.09%\n", + "dw_separable_5 9.95%\n", + "pw_separable_5 59.14%\n", + "dw_separable_6 13.68%\n", + "pw_separable_6 42.10%\n", + "dw_separable_7 13.74%\n", + "pw_separable_7 49.42%\n", + "dw_separable_8 19.40%\n", + "pw_separable_8 54.57%\n", + "dw_separable_9 20.78%\n", + "pw_separable_9 62.35%\n", + "dw_separable_10 25.13%\n", + "pw_separable_10 69.91%\n", + "dw_separable_11 32.66%\n", + "pw_separable_11 48.60%\n", + "dw_separable_12 24.16%\n", + "pw_separable_12 78.75%\n", + "dw_separable_13 39.64%\n", + "pw_separable_13 70.33%\n", + "dw_sepconv_t_0 66.37%\n", + "pw_sepconv_t_0 55.17%\n", + "dw_sepconv_t_1 54.85%\n", + "pw_sepconv_t_1 53.14%\n", + "dw_sepconv_t_2 46.27%\n", + "pw_sepconv_t_2 49.41%\n", + "dw_sepconv_t_3 41.03%\n", + "pw_sepconv_t_3 49.47%\n", + "dw_sepconv_t_4 41.18%\n", + "pw_sepconv_t_4 50.55%\n", + "head 0.00%\n", + "----------------------------\n", + "Mean 40.95%\n", + "\n", + "Layer Latency (ms) Clocks\n", + "-------------------------------------\n", + "conv_0 2.6448\n", + "conv_1 57.7118\n", + "conv_2 68.1212\n", + "conv_3 75.7012\n", + "dw_separable_4 5.3596\n", + "pw_separable_4 5.6954\n", + "dw_separable_5 4.3391\n", + "pw_separable_5 10.9067\n", + "dw_separable_6 1.7087\n", + "pw_separable_6 6.5842\n", + "dw_separable_7 1.6388\n", + "pw_separable_7 6.2570\n", + "dw_separable_8 1.4526\n", + "pw_separable_8 6.1165\n", + "dw_separable_9 1.3138\n", + "pw_separable_9 6.0867\n", + "dw_separable_10 1.1077\n", + "pw_separable_10 6.0284\n", + "dw_separable_11 0.9170\n", + "pw_separable_11 7.3058\n", + "dw_separable_12 1.2139\n", + "pw_separable_12 5.5492\n", + "dw_separable_13 0.6104\n", + "pw_separable_13 8.4229\n", + "dw_sepconv_t_0 1.3844\n", + "pw_sepconv_t_0 10.1121\n", + "dw_sepconv_t_1 3.2150\n", + "pw_sepconv_t_1 20.7983\n", + "dw_sepconv_t_2 6.7933\n", + "pw_sepconv_t_2 11.4291\n", + "dw_sepconv_t_3 13.8141\n", + "pw_sepconv_t_3 7.1611\n", + "dw_sepconv_t_4 9.3013\n", + "pw_sepconv_t_4 9.5852\n", + "head 3.6043\n", + "dequantizer 6.0524\n", + "-------------------------------------\n", + "Total 396.0441\n", + "Per-layer benchmark complete.\n" + ] + } + ], "source": [ "from brainchip_utils.hardware_utils import per_layer_benchmark\n", "from akida_models.sparsity import compute_sparsity\n", @@ -224,7 +468,16 @@ "name": "python3" }, "language_info": { - "name": "python" + "codemirror_mode": { + "name": "ipython", + "version": 3 + }, + "file_extension": ".py", + "mimetype": "text/x-python", + "name": "python", + "nbconvert_exporter": "python", + "pygments_lexer": "ipython3", + "version": "3.12.13" } }, "nbformat": 4, diff --git a/akida2/model_zoo/portrait128/portrait128_notebook_training.ipynb b/akida2/model_zoo/portrait128/portrait128_notebook_training.ipynb index 604b4a8..10025fd 100644 --- a/akida2/model_zoo/portrait128/portrait128_notebook_training.ipynb +++ b/akida2/model_zoo/portrait128/portrait128_notebook_training.ipynb @@ -472,10 +472,12 @@ "The input layer is the exception. It runs on the dedicated input processor, a\n", "conventional convolution whose cost doesn't depend on how many of its inputs are zero.\n", "\n", - "For this model, look at the decoder. Its last blocks run at 64 × 64 and 128 × 128, so\n", - "they process far more activations than the encoder layers at 8 × 8 or 4 × 4. Their\n", - "sparsity is therefore likely to have the most effect on latency; the per-layer\n", - "benchmark in `portrait128_notebook_benchmark.ipynb` measures it." + "For this model, the per-layer benchmark in `portrait128_notebook_benchmark.ipynb` shows\n", + "where the time goes. The slowest layers are the encoder's first full convolutions\n", + "(`conv_1` to `conv_3`), which apply 3 × 3 kernels across every channel of the 64 × 64 and\n", + "32 × 32 feature maps. The decoder's upsampling layers, which run at up to 128 × 128, come\n", + "next, and the deep encoder layers at 8 × 8 and 4 × 4 cost little. So sparsity in those\n", + "early convolutions has the most room to reduce latency." ] }, {