Files
pyannote-audio/tutorials/training_a_model.ipynb

883 lines
166 KiB
Plaintext
Raw Permalink Blame History

This file contains invisible Unicode characters
This file contains invisible Unicode characters that are indistinguishable to humans but may be processed differently by a computer. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
{
"cells": [
{
"cell_type": "code",
"execution_count": 17,
"metadata": {},
"outputs": [],
"source": [
"# preparing notebook for visualization purposes\n",
"# (only show outputs between t=180s and t=240s)\n",
"from pyannote.core import notebook, Segment\n",
"notebook.crop = Segment(210, 240)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Training, fine-tuning, and transfer learning with `pyannote.audio`\n",
"\n",
"In this tutorial, you will learn how to use `pyannote.audio` to \n",
"\n",
"* train a voice activity detection model from scratch, \n",
"* fine-tune a pretrained speaker segmentation model,\n",
"* perform transfer learning (from speaker segmentation to overlapped speech detection)\n",
"\n",
"## Data preparation\n",
"\n",
"This tutorial assumes that \n",
"* the [AMI corpus](https://groups.inf.ed.ac.uk/ami/corpus/) has already been [setup for use with `pyannote`](https://github.com/pyannote/AMI-diarization-setup/tree/main/pyannote)\n",
"* the `PYANNOTE_DATABASE_CONFIG` environment variable is set accordingly. \n",
"\n",
"See [`pyannote.database` documentation](https://github.com/pyannote/pyannote-database#pyannote-database) to learn how to prepare your own dataset for training with `pyannote.audio`."
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {},
"outputs": [],
"source": [
"from pyannote.database import get_protocol\n",
"ami = get_protocol('AMI.SpeakerDiarization.only_words')"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Training a voice activity detection model from scratch"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Voice activity detection (VAD) is the task of detecting speech regions in a given audio stream or recording.\n",
"\n",
"We initialize a VAD *task* that describes how the model will be trained:\n",
"\n",
"* `ami` indicates that we will use files available in `ami.train()`.\n",
"* `duration=2.` and `batch_size=128` indicates that the model will ingest batches of 128 two seconds long audio chunks."
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {},
"outputs": [],
"source": [
"from pyannote.audio.tasks import VoiceActivityDetection\n",
"vad_task = VoiceActivityDetection(ami, duration=2.0, batch_size=128)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"We initialize one *model* with the `PyanNet` architecture used [in that paper](https://arxiv.org/abs/2104.04045). \n",
"In particular, we increase the default stride of the initial `sincnet` feature extraction layer to `10`.\n",
"\n",
"The model is also provided with the task (`task=vad_task`) for which it is being trained:"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {},
"outputs": [],
"source": [
"from pyannote.audio.models.segmentation import PyanNet\n",
"vad_model = PyanNet(task=vad_task, sincnet={'stride': 10})"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Now that everything is ready, let's train with `pytorch-ligthning`!"
]
},
{
"cell_type": "code",
"execution_count": 7,
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"GPU available: True, used: True\n",
"TPU available: False, using: 0 TPU cores\n",
"IPU available: False, using: 0 IPUs\n",
"LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n",
"\n",
" | Name | Type | Params | In sizes | Out sizes \n",
"-----------------------------------------------------------------------------------------------------------------------\n",
"0 | sincnet | SincNet | 42.6 K | [128, 1, 32000] | [128, 60, 115] \n",
"1 | lstm | LSTM | 589 K | [128, 115, 60] | [[128, 115, 256], [[4, 128, 128], [4, 128, 128]]]\n",
"2 | linear | ModuleList | 49.4 K | ? | ? \n",
"3 | classifier | Linear | 129 | [128, 115, 128] | [128, 115, 1] \n",
"4 | activation | Sigmoid | 0 | [128, 115, 1] | [128, 115, 1] \n",
"5 | validation_metric | AUROC | 0 | ? | ? \n",
"-----------------------------------------------------------------------------------------------------------------------\n",
"681 K Trainable params\n",
"0 Non-trainable params\n",
"681 K Total params\n",
"2.728 Total estimated model params size (MB)\n"
]
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"HBox(children=(HTML(value='Validation sanity check'), FloatProgress(value=1.0, bar_style='info', layout=Layout…"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "0a76f04fd4ab4cad9dac09344b932e49",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"HBox(children=(HTML(value='Training'), FloatProgress(value=1.0, bar_style='info', layout=Layout(flex='2'), max…"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"HBox(children=(HTML(value='Validating'), FloatProgress(value=1.0, bar_style='info', layout=Layout(flex='2'), m…"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\n"
]
}
],
"source": [
"import pytorch_lightning as pl\n",
"trainer = pl.Trainer(gpus=1, max_epochs=1)\n",
"trainer.fit(vad_model)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"For the purpose of this tutorial, the model is trained for only 1 epoch. One can obviously expect better performance by training longer and on more data.\n",
"\n",
"See [`pytorch-lightning`](https://www.pytorchlightning.ai/) documentation to learn more about the [`Trainer` API](https://pytorch-lightning.readthedocs.io/en/latest/common/trainer.html), in particular."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Once trained, the model can be applied to a test file:"
]
},
{
"cell_type": "code",
"execution_count": 8,
"metadata": {},
"outputs": [],
"source": [
"test_file = next(ami.test())\n",
"# here we use a test file provided by the protocol, but it could be any audio file\n",
"# e.g. test_file = \"/path/to/test.wav\"."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Because the model was trained on 2s audio chunks and that test files are likely to be much longer than that, we wrap the `model` with an `Inference` instance: it will take care of sliding a 2s window over the whole file and aggregate the output of the model."
]
},
{
"cell_type": "code",
"execution_count": 18,
"metadata": {},
"outputs": [
{
"data": {
"image/png": "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",
"text/plain": [
"<pyannote.core.feature.SlidingWindowFeature at 0x150493554b50>"
]
},
"execution_count": 18,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"from pyannote.audio import Inference\n",
"vad = Inference(vad_model)\n",
"\n",
"vad_probability = vad(test_file)\n",
"vad_probability"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Perfect voice activity detection output should look like that:"
]
},
{
"cell_type": "code",
"execution_count": 19,
"metadata": {},
"outputs": [
{
"data": {
"image/png": "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",
"text/plain": [
"<Timeline(uri=IS1009a, segments=[<Segment(54.95, 60.85)>, <Segment(61.54, 80.83)>, <Segment(81.04, 82.05)>, <Segment(83.39, 86.94)>, <Segment(87.49, 88.59)>, <Segment(88.82, 89.69)>, <Segment(90.21, 90.52)>, <Segment(91.31, 107.4)>, <Segment(107.88, 118.17)>, <Segment(118.76, 119.05)>, <Segment(119.38, 119.91)>, <Segment(122.4, 132.39)>, <Segment(132.87, 146.77)>, <Segment(149.44, 162.32)>, <Segment(165.19, 166.99)>, <Segment(188.06, 188.34)>, <Segment(195.04, 195.27)>, <Segment(195.56, 195.82)>, <Segment(196.38, 202.4)>, <Segment(203.78, 218.2)>, <Segment(219.44, 222.12)>, <Segment(222.76, 227.27)>, <Segment(236.11, 238.38)>, <Segment(241.01, 242.5)>, <Segment(243.94, 245.45)>, <Segment(248.32, 251.04)>, <Segment(251.8, 252.7)>, <Segment(252.79, 254.69)>, <Segment(255.24, 257.74)>, <Segment(260.02, 272.52)>, <Segment(274.68, 275.95)>, <Segment(276.23, 279.86)>, <Segment(280.08, 285.51)>, <Segment(288.1, 297.75)>, <Segment(308.92, 310.65)>, <Segment(311.19, 318.45)>, <Segment(320.53, 327.5)>, <Segment(328.77, 333.36)>, <Segment(333.57, 335.49)>, <Segment(336.02, 336.43)>, <Segment(337.47, 339.63)>, <Segment(347.67, 349.78)>, <Segment(355.86, 364.82)>, <Segment(367.25, 375.93)>, <Segment(377.25, 386.09)>, <Segment(386.53, 387.02)>, <Segment(387.4, 393.64)>, <Segment(395.23, 398.51)>, <Segment(398.76, 399.29)>, <Segment(400.24, 400.93)>, <Segment(402.86, 412.67)>, <Segment(419.16, 427.58)>, <Segment(428.69, 434.99)>, <Segment(436.56, 449.23)>, <Segment(450.37, 495.68)>, <Segment(497.24, 502.27)>, <Segment(502.77, 503.46)>, <Segment(503.78, 506.79)>, <Segment(508.17, 551.13)>, <Segment(552.51, 565.64)>, <Segment(566.02, 581.45)>, <Segment(581.48, 637.63)>, <Segment(638.22, 641.25)>, <Segment(641.3, 661.77)>, <Segment(663.98, 665.07)>, <Segment(665.22, 683.95)>, <Segment(685.84, 703.98)>, <Segment(704.61, 711.13)>, <Segment(711.71, 715.68)>, <Segment(715.72, 744.62)>, <Segment(745.25, 750.08)>, <Segment(751.38, 758)>, <Segment(764.72, 775.95)>, <Segment(777.27, 787.09)>, <Segment(787.58, 793.95)>, <Segment(794.48, 794.68)>, <Segment(795.57, 795.88)>, <Segment(796.5, 797.81)>, <Segment(797.96, 803.68)>, <Segment(803.8, 805.72)>])>"
]
},
"execution_count": 19,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"expected_output = test_file[\"annotation\"].get_timeline().support()\n",
"expected_output"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Fine-tuning a pretrained speaker segmentation model\n",
"\n",
"Speaker diarization is the task of partitioning a given audio stream of recording into according to the speaker identity.\n",
"\n",
"[`pyannote/segmentation`](https://hf.co/pyannote/segmentation) is a model that was pretrained to perform speaker diarization, but only locally, on 5s-long audio chunks. "
]
},
{
"cell_type": "code",
"execution_count": 20,
"metadata": {},
"outputs": [],
"source": [
"from pyannote.audio import Model\n",
"pretrained = Model.from_pretrained(\"pyannote/segmentation\")"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Let's visualize how it performs on our test file:"
]
},
{
"cell_type": "code",
"execution_count": 23,
"metadata": {},
"outputs": [
{
"data": {
"image/png": "iVBORw0KGgoAAAANSUhEUgAABHwAAAEiCAYAAACLNMUPAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjMuMywgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/Il7ecAAAACXBIWXMAAAsTAAALEwEAmpwYAAB7t0lEQVR4nO3dd4Acdf3/8ednZuvdXr9LcumVAIHQAqFLB0EQAQUVCyoIKsJXv4rtq/4sIHbEAiqKClKkKCJK74qQ0AIkoaT3XK7fbZ/P74/Z29yFS3It2du71yMsMzs75b17n52dec/n8xljrUVEREREREREREYOp9ABiIiIiIiIiIjI0FLCR0RERERERERkhFHCR0RERERERERkhFHCR0RERERERERkhFHCR0RERERERERkhFHCR0RERERERERkhAkMdgW1tbV26tSpQxCKiIiIiIiIiIgALFy4sMFaWzfQ5Qed8Jk6dSoLFiwY7GpERERERERERCTHGLNyMMurSZeIiIiIiIiIyAijhI+IiIiIiIiIyAijhI+IiIiIiIiIyAijhI+IiIiIiIiIyAijhI+IiIiIiIiIyAijhI+IiIiIiIiIyAijhI+IiIiIiIiIyAgTKHQAMjiJTIIVrStoTjbjWQ8s2K5/1h9uy2C2jhvT+/Ru4z1H+7ds93n6HcN2pneXtVk86+Uf1lo8vO1vq5f3NdTvaWfvx2AIOAECToCgE9zh0DXudre3I9ZaGuINJDIJ0jZNL8Xg7cv0Zabcurddpvuy3V/f3mvbK5fGmPww4kaYXD65TzHJyJb20jQnmmlMNNKZ6Xzb971r3BiDYxxc4+IYJ//o+h65xiXoBIkEIkTcSH7oOm6h32Kv1revZ0tiC/FMnHQ27X8/jMHByb9Xxzj570zXdIPxPyNyn5H1sNi3fWZYto4XyPb27bvS5PLJTCmfstu3KyIi0lepbIoNHRvY0LGBtJfe7nx9OX7f9th8oPp6rtBXBpM/jsGQH3eMk3+9MlzJ9Mrp+WnSf0r4FKH17etpTDby97f+zj1v3UNbqq3QIcku0pUc6m8iqDXZSlu6uMvFHlV7cOcZdxY6DNlFrLUks0naUm20plppTDSyJb6FLYktbIlvYUPHBta2r2Vt+1o2dW4a8oOM7rqSQOWhcsaWjGVMyRjGlY5jYmwiE8omML50POXhcipCFQTd4C6Lo0sqm+IrT32F+1fcv8u3NRpdst8lfGr/TxU6DBEREQA60508tvoxnlj7BKtbV7O+Yz2b45sLHdawMbNyJr8+8dfUldQVOpSipIRPEfrkQ59kectyAk6AE6ecyPGTj6c6Up1PAnSvadL1r0uP2hY7qJWxs3n6Mr3Hera3bF/W02O05/p7u5oPW2vT7Ox9DfV76uvnlbVZ0tk0aZsmnU2T8TKkvZ0P054//85OfqOBKNMrphMLxQiYQN+z4n282N5b7afeajbl5+ullti25dL61dPyNRFioVjfgpEhkcgkeL3p9R7lzVpLebgcz3rEM3ESmQQpL0VFqIKaaA1jS8biGIfGRCPrO9azvn096zvW055uJ5lJksgmSGQS+aROa6qVRCaRn769K1aOcRhTMoYJsQnMr5/P+Nh4aiO1VEWqiAVjOFmP4KI3cZcux0kkcdY3QEccsllswCWz7yxSx80nW1NB1maxWLJeFg+PdDad334ikyCe9d9XPBOnOdnMps5NvLrlVR5e9XCv8eVrBgUi1EXr2Kt6L/ap3YdTp59K2A33+fO21tKaamVd+zrWdaxjQ8cGMl4GgGfWP8NTa5/iE/t+gv3r9icSiBByQz1q63SvqdNVc6d7zc78PhEnf9Wse82g/FW1Xn4ndpddmcTbkTElYwqyXRERke6WNS/jly/9ksdXP04im6AmUsPMqpkcOeFI6mP1jC8dz7jScTs9vuhLa4Ch+p0fqvVs2yKl+zEOkB9f3baaHzz3A6569ip+fMyPh2Tbo40ZbBWvefPm2QULFgxRONIXD696mM50J4ePP5yaaE2hwxGREWBZ8zLe/bd3D8m6ooEoETdCOBAm4kYoD5dTHvIf0UCUsBsmEohQFiqjLFhGWaiMmmgNNZEaqqPVVIYrt5uk7Hz+BdZ94Quk1671JxhDsL4et7ISHAevvZ3UihUQCFB5ztmM+d8v4MZK+/0ePOuxuXMza9rXsL5jvZ+0SrbSlmojkfUTRBs6NrB4y2La0m2cMPkEfnzMj9920OVZj8WNi1m4YSFr29eyrn0dazvWsr7dT4z1JuJG+MwBn+Ejcz7S77hFRERk+Fvdtpr3/f19GGM4ddqpvHPaOzlgzAFqutSLb/z7Gzy08iGePO/JUfn5GGMWWmvnDXR51fApQsdPPr7QIYjICDOudBy/PP6XPZoOArSl2nAdN5/ECTgBWpItNCQa2NThN7WqDFcyPuZfhRpXMm6XNXuKL1rEqo9/nMCYOib87BpKDzsMp7QU4/T88U+tWkXjjX+g6ZZb6Pj3f5h0/XWEp03r17Yc4zC2dCxjS8fucD7Petyw6AZ+9sLPeGb9Mxw2/jAAMl6Gf634FzcsuoE3m98EIBaMMT42ngmlEzh47MGMj43PP+pL6wm7YTzrEXbDhNxQv+IVERGR4nHt89cCcPu7bmdi2cQCRzO87V+3P3e9cRcrWlcwvWJ6ocMpOkr4iIgIJcESjpp4VKHD2K5MYyOrL76EQHU1U2++mUBt7XbnDU2ezLiv/x/lp53Kmks/y+pPXMi0O+/wawENMcc4fGTOR7ht6W38/IWfc+DYA/nbm3/jd6/8jrXta5lZOZP/d/j/4+iJR1Mb3X7MIiIiMjq0p9p5ZPUjvGfme5Ts6YN9a/cF4NWGV5XwGQAlfEREZNhrvvNOslu2MPnuu3aY7Omu5KCDmHTdr1hx3vvZ/LNrGff1/9slsYXcEBfueyHf+e93mHeTX+N239p9+eLBX+SYSceMyurHIiIi0rvnNjxHMpvk5KknFzqUotCVFFvfsb7AkRQnJXxERGRYs9bScuddRA86iMhee/Vr2ejcuVSddx5Nt95K5bnvIzJ79i6J8b2z30skEGF5y3IOHX8o88fN71MniiIiIjK6vNb4Go5xmFM7p9ChFIVIIEJ1pFoJnwHSZUcRERnW4gsXklqxgsqzzx7Q8nWfvRQnFqPh578Y4si2cozDu2e+m8sPupxD6w9VskdERER69dqW15heMZ1oIFroUIrGuNJxSvgMkBI+IiIyrDXfcSdOaSnlpwys6rNbWUnVB95P20MPkVy+fIijExEREem7pY1L2bN6z0KHUVTqS+vZ0L6h0GEUJSV8RERk2Mq2t9N6//2Un3oqTknJgNdTff75mGCQxt/fOHTBiYiIiPRDIpNgY+dGppRPKXQoRaW+tJ51Heuw1hY6lKKjhI+IiAxbrf+4DxuPU3nOwJpzdQnU1lLxnvfQ8te/ktm8eYiiExEREem7te1rAZhcNrnAkRSXupI64pk48Uy80KEUHSV8RERkWPI6Omj83e8Iz5pJZO7cQa+v5oKPYtNpGm+6eQiiExEREemfVa2rAJhUNqnAkRSXmkgNAFviWwocSfFRwkdERIYday0bvnslqVWrGPvVrw1JJ8ihqVMpO/FEmm65hWx7xxBEKSIiItJ3q9tWAzC5XDV8+qM6Ug3AloQSPv2lhI+IiAwryTfeYN0VV9By113UXPxJSg+dP2Trrvn4x/BaW2n+y1+GbJ0iIiIifbGqbRVloTIqwhWFDqWo1ERzNXyU8Om3QKEDEBGR0cdLpdj0ve+RWr2G2JFH4JSVE6ippvmOO2l78EFMMEjNRRdRd+mlQ7rd6H77UTJ/PltuuIGq970Xp7R0SNcvIiIisj1r2tao/54B6Krh05hoLHAkxUcJHxER2e0afvUrmv58CyYYpOPJJ/PTTSRC7Wc+Q9UH3k+gunqXbLvu8stY+f4P0PinP1F78cW7ZBsiIiIi21rVtop9avYpdBhFR334DJwSPiIigtfRQfOdd1F++rsIVFXtkm0k33qLlrvvJrOlkZa//pWKd59B/fe+R7apCa+1lUxDA+FZs3Ardm0155IDDiB2/PFs+e0NVJ577i57vyIiIiJd0l6a9e3rOWXqKYUOpegE3SBloTLV8BkA9eEjIiKk161j45VX0nzrrflp1lq8zs5Br9taS+PNN7P8rLPZ8rvf03LPPVSccTrjvvUtjDEEqqsJTZ1Kybx5uzzZ02XM5ZfhdXbS+Lvf75btiYiIyOi2sWMjGZvRHboGqCZSoxo+A6AaPiIiQnjWLEqPOorGP/6JkoMPJvHqqzTe/GfSq1YR3X9/Jv7yFwNqYuUlEqz/6tdo/cc/KD3qKMZf+V3cmhqMU9jrDeFZs4gdeyzNd99N3WcvxQSDBY1HRERERraGeAMAdSV1BY6kOFVHqlXDZwCU8BEREQDGfvELrPjg+aw8/0MARA88kPJ3vpPGP/yB1Rd9kil/uHGHnRxv/N7VxF98kfJ3nkJ80SukN6wntWIl2YYG6i6/nJpPXjQkt1cfKpVnn037ww/T/sQTlB1/fKHDERERkRGsK1nR1QGx9E9NtIY3m98sdBhFRwkfEREB/Fov0+/9O53PPUd4+nQie+0FQHT//VjzmUtZ89nLmPSrX2JCobct2/700zTeeCNOWRkbr/oebnU14ZkzKT3kYCrf+15KDztsd7+dnYodfRRuXS3Nd9yphI+IiIjsUkr4DI5q+AyMEj4iIpIXHDOGitNO6zGt7Nhjqf/Wt1j/1a+y7itfZfz3r+7RJMtay+af/Yzg+PFM/8e9pDdsIDRx4rBvJmUCASrf/W62/P5GMg0NBGprCx2SiIiIjFBd/c903XFK+qcmWkNLsoW0lyboDO9jzOFEnTaLiMhOVZ59FnWf+xyt997Lpquvxlqbfy3xyiskXnqZ6o9/DCcaJTxt2rBP9nSpOPNMyGZpuffeQociIiIiI1hjopGyUBlBtziOkYabrkRZU6KpwJEUFyV8RESkT2ou/ARVH/4QjX/4I4033JCf3njjHzDRKBVnnFHA6AYmPHMmkblzabnr7h5JLBEREZGh1JhoVO2eQehqCqc7dfWPmnSJiEifGGMY+6Uvkd3SyKYf/ggTCuNWVtD6j39Qc9FFuGVlhQ5xQCrPOosN3/wmrffd97bmbCIiIiJDoTHRqP57BqEm6ifLtiSU8OkPJXxERKTPjOMw/qoryba2svHKKwGIzJ1L7ScvKnBkA1d5ztm0/PWvbPjGN4nstRfh6dMLHZKIiIiMMFviW5heqWOMgepKlqlJV/+oSZeIiPSLCYWYdN2vmHDtz5jw058y5U9/3OHt2oc7Ewgw/oc/xIRCrPzg+aTXry90SCIiIjLCqIbP4HR9drpTV/8o4SMiIv1mXJfyE0+k/JSTccLhQoczaKGJE5hy05/wkknWffEKbDZb6JBERERkhMh4GZqTzUr4DEIsGCPoBNWkq5+U8BEREQHC06cz7v/+j87nnqPhuusKHY6IiIiMEM3JZixWnTYPgjGG6kg1jXHV8OkP9eEjIiKSU3Hmu+n4979puPbnONESaj52QaFDEhERkSLXdWep6qhq+AxGdaSapqT68OkPJXxERERyjDGM/+53sOk0m77/faL770/JgQcUOiwREREpYl39zqhJ1+BUR1XDp7/UpEtERKQbEwox/srv4tbUsPknP8F6XqFDEhERkSKmhM/QqA5Xq9PmflLCR0REZBtOSQl1l32Wzueeo+lPfyp0OCIiIlLEupp01UTVh89gVEf8hI+1ttChFA0lfERERHpR+d73EjvuODb98Eckli4tdDgiIiJSpBoTjQScAGXBskKHUtSqo9UksgnimXihQykaSviIiIj0whhD/Xe+jVNZwdr/+RzZtrb8a2rmJSIiIn3VmGikOlKNMabQoRS1riZxujV73ynhIyIish2B6mom/PBHpFauZMX7zqXl3n+w/v/+jyX7zmXFee+n87nnCh2iiIiIDHNbElt0S/Yh0JXwUT8+faeEj4iIyA6Uzj+Eyb/7HdnWVtb97//SfPdfKTvpRJKvv87KD32Y1gcfLHSIIiIiMow1xht1S/Yh0JU0a0ro1ux9pduyi4iI7ETp/EOY+dCDpFavxi0rI1hfT7alhZUf+jAN1/6c8hNPLHSIIiIiMkw1JhqZXjm90GEUPdXw6T/V8BEREekDJxolssceBOvrAXArKqg8+yySr79OauXKAkcnIiIiw5G1li2JLbol+xCoilQBSvj0hxI+IiIiA1R2wgkAtD3yaIEjERERkeGoLd1GMpukNlpb6FCKXiQQoSxYxsaOjYUOpWgo4SMiIjJAwQkTCM2cQceTTxQ6FBERERmGVretBmBi2cQCRzIyjIuNY0PHhkKHUTSU8BERERmE2NHvoPO5BXgdHYUORURERIaZ1a1+wmdy2eQCRzIyjC8dz/qO9YUOo2go4SMiIjIIsaOPwqbTdPz3v4UORURERIaZVW2rANXwGSr1pfWs61hX6DCKhhI+IiIig1By4IE4JSW0P6p+fERERKSnla0rGRMdQzQQLXQoI0J9rJ62VBvtqfZCh1IUlPAREREZBBMKETvuOFofeBCbShU6HBERERlGFjcuZlbVrEKHMWJMiE0AtvaNJDumhI+MWp0LF7Lx+z+g5d5/YDOZQocjIkWs4ozT8VpaaLrjjkKHIiIiIsNEe6qdN5veZL+6/QodyoixZ/WegJ9Ik50LFDoAkd3Bi8cx4TA2kaD1/gdovvMO4gsW5l/f9IMfUHLIIdRecjHh6dMLGKmIFKPSo46i9PDD2XjV93DCESrPPqvQIYmIiEiBPbfhOSxWCZ8hNKlsEmXBMl5teJWzZul4a2eU8JERo/OFF9jy2xtIvPIKgdpaQjOm40SiJJYuIfHyIkwggAVIpwlNncqYK66g6tz30f7007Te+w/aH3uMtgceoO6yy6g852zc8vJCvyURKRLGGCb89Cesvfxy1n/1q8RffJHqCz6qBLKIiMgolfbSXP/y9YwrHce8cfMKHc6I4RiHvWv3ZuHGhVhrMcYUOqRhzVhrB7WCefPm2QULFgxROCID03jzzWy88ioC1dWUHHoo2cZGksuWYRMJQlOnUjL/EMhmsZ5H2bHHEj3ooLftHNKbNrHh/75O++OPQyBA7MgjqTjjdCJz5hCor8cJhQr07kSkWNh0mo0/+AFNt9wK6TQlhx5K1fvfT+yYd+CEw4UOT0RERHaDlza/xDXPX8NzG57jB0f/gFOmnVLokEaU25fezref+TZ/PvXP7Fu3b6HD2aWMMQuttQPOGCrhI0Wv9f4HWHvZZcSOPZbx378at6xswOuy1hJ/8UXaHnqI1r/fS2bTJv8FxyE4aSLhGTMJz5hBeNZMQlOmEKivJ1Bbi3HUHRb4n59NpbCJBF4iiU0m8BIJbDKJCQZxq6oJVFVilDwbfpJtsOwxwIAxPYeBEMTGQXk9RCpz02VHMg0NNN95F0233Upm3XoIBAjPmEFk772JzJnjD/ecjVNSUuhQRUREZAj9+uVfc+0L11IVruIzB3yG981+X6FDGnHaUm288653MrZkLL84/heMKx1X6JB2GSV8RqHUmrXgZcFxMQEXHAcTCPhJh22Hrjvsq7n1Wga3ndbbPJ5Hx7PPsubTnyEycyqTr7wMJwB4Gf/zsV5umPWHAE4A3KA/zI/nnruBHuPWGuKvvUFq7QbSazeQXLGK5LLlpFaugu6dPAcCBMeNIzBuLG5FJW5FRe5RjlNWhnED/t/JOP6JsmP8v4njALnnrv+6cR3/79rb0HXzf1McJ7+MTaexqaSfaEkm8VIpMuvWEX/pZb8z6q5tOrnlHMfftuvkpnV7Lbc9m8ngdXRgk0lsJpN7pLHJlL+dVArrZcEC2SxeRwfZtja89nbwvJ3+zZ2yMtzqKgKVVbjV1biVleA6/t/Zs2QaGnArK5nwg+/3sRTJYKU3LCZ43aE7nc8Gon7iJzYOW1YPpbV+2U62Yjq3QGcjZOL+dy5UigmXQbgcwmUQqfCHGEh3QqrDf9huZcY4W78rXeOdDbDlLUi2+2XXOP53NRACNwRuODce3jrNCfrfb7drnsDW8a7vPgbScT+WrmE2Bdm0P/Ry3/NAeOu6Te677Lhbx6NVUDUFqqZC5RQIlebfg/UsHf/5L50vvERi8WISr71Gdktj7r0aghMmEKip8b8H1VUEqqsJ5Mbd3Hg+KWT8ZYwx/nhXUi6/i889d1yM62LcgP9b4Lpv33/093fB2tx+uP9D6+X2xZ4Htuu55++bs1n/N8DL5r7/HtbmXvc8rM3iZfxpNuvh5ebzslms9fCynv+3S7Rj21rw2prw2lqx7W3Yjna8znZsRxwbj+N1xEk3dBI75XjGfOGH/Xv/MmC3L72dPy/+M+NKxzGmZEz+URYqI+JGiAT8h2McTO6fYxz/5xG/rBr8cr/t8655DSb3c7p1HT2WM/5Fme7zG7N1nY5x8vN1baP7fLL7dB0TWmzvz3PDrYOt07e3bH7dO1nn2+a3drvzWCzpbJpkNkkym2Rd+zoOG38YkUBk4G9e+uXG+68iFe9gUvlkysIVlIZKCQUiuG4g9wgScIMEAkFcJ4jrBgi4/jDohnEDARwnd/zt5I7Lu34fna7jkLfvA7rKRX6IJZlO0Ni+mU1t63ly5WPc+cotnDLmaD69x8cJZ7utP7dup6QENxbzzxMC2/Sw0n172xkv1H7J5n6n/d/nbuOe9Y/lrO35GXYf91fQY2i3rrj3YRdjev49co9/r/s3lz/2P6Rsmv3r9mdaxTTGlowlFooRC8YoCZYQdIIEnSABJ+CPu93Gne2Mu0ECJlCQz9lai2c9OjOd3LfsPla0ruBL87+khM9os/iQvaG1H383Y7udEOzEQItDr8vtni9JoDTLtBM3E4jsPNEwFKwHqbYAqXaXTKdLOvfIdLpkU07uYbDZwtb6cUs8TND6f5vcAA/A9Dgny79uzda/owETsDiuhdz5Ncbmzm/9Ydef1xgwQYsT9LfnuP6yxu16gHEteJBNOnhJg5d0yCZyw6TBSxmwhq4A3IglUxdl9j0v78ZPbHR7a10Dl157W+7PajFYP7+AJUyasaaJsaaRcaYpN97EGJqoNa0AtBGl2ZbRaGPECWMxRElSZuLEiFNmOokRp9QkAUjZAB1EiBMmi5uLwt+uk3sYLC4erZSw3I6nhVJcPBwsAbKESBMiQ8ik8+NB0gTJEiRDkAyBbuNBMgTM2/cTSRskQYg4YdIESeOSIUAmF1eINOHcNpzc9v2h/yglgdPHnae1kEk4JBqDJJqCpNoCZBMOmaRLNumQSTi578IulvtdGOot2fz/2D3vo69cC0FDtsyh+fCjOPLq6wod0ahx94WnsOeTK/PP89/AXPHorbjY7q8NwXyD0stXe1eV7F32jRncof4O7bLPYhfGvKuEf/x59j7xE4UOY9R45B1zqN+464/9Pfz9iTXgma3jxoLrgeMNt9ted/vy9PULuqP5huNvei8sW/9G5Ma7ptnu07pNh7fP3/Uuu++DusZ39prp7aPvvuxO1rPtMg7QWAFHPrtkUAkfddpchFpPPBTTsgXjWUwum2o8u3XYNd0jX2Oi1xoyQM9vuH37pF7n2/p0QF/9XhbaUb7IWLAmd/ppup+GQrYsQnLGGBaVTaEpVE/WBPGM6z+6ThuNg5e7uufYLI7N4tqMPySL0zVuM7npGf9ELj/dX8bJzevaTG7rYHJZk/zzXCbFYCGTxUllMFkv93fycl9su3U5zx83XVfP838vtk7r+jta6y/f9fe0uemuwboOJjck4OBFAtjY0PQXYnv5g/U2rccfrfu8fSlPvazTxMYyuw/xydCorazgg2e+C8+zZD1L1oLnWTxr8Xcr/tWsTg9WGFixk/X5RdRfFmvzxRWb9l83gfx8PZbrZT09X9/JAr2uo9sU6+F4/vfcYEk7ETzjbjP/jre57esBL0lFagNVyfVUptYR9BL5/YAhtz/O7yf877ud1n0NZusWrMUkMzidSdx4ChNP4aSybE2Hdu13/DjyBw1d7zy37/B/A7b+FhjrbZ2W/73Y5kDZdIujW+ov/zx/Qr3zabZbs0Brug3pes7WWo6514zpOW67XWHtuqpnTdfVwq4aYAbPDZF1w6QjZWRKysmUVJCKVZMprSJTWoGNRjBOoKuyI3PGVyC7j5l2GC1tyfxziyWFR8Z4eEAWi7dt7Q1yB952m+/fts97TO9aqmvh3HSzzZXkbuvvPt59rh6JJLazzYHayYHT9n9fe5t3e6sbaLyFOaHb7lZ7P/TcyvZ8pS/Rm53N2Id1dt/vOhgcayh3A8yavF8fIpChEjvuSFob19JOhhRZUmTxsLnjFf+4hVwtnK5jGKBnDZ3cvP5vM/kHFpyu53T9rprcMDePAesYrOP/RoWMQ8hxqSJMyA1A0MGGXP/YnK7jevwf77SHSWUxqax/XN+b7X2N+1pZY0ezbW8d+QONbtMMW2sXdf1+m67f+twM3efp2iHbbd6zoZeVv/2p7W0+22292zw32zz3sGStJYOX/32x+XLg78/9v3/uODX3eVibm9d0Jflsfn9s6f5bkptmuv0ymG6hmV4+eoNfEzX/tkyP17ZO3rrfMRgqbZCq0lp4dsm2a+wX1fARERERERERERlmBtuHz/CqgSYiIiIiIiIiIoOmhI+IiIiIiIiIyAijhI+IiIiIiIiIyAijhI+IiIiIiIiIyAijhI+IiIiIiIiIyAijhI+IiIiIiIiIyAijhI+IiIiIiIiIyAijhI+IiIiIiIiIyAhjrLWDW4Exm4EOoGFIIhIZnmpRGZeRT+VcRgOVcxkNVM5lNFA5l5GuFii11tYNdAWDTvgAGGMWWGvnDXpFIsOUyriMBirnMhqonMtooHIuo4HKuYx0Q1HG1aRLRERERERERGSEUcJHRERERERERGSEGaqEz6+HaD0iw5XKuIwGKucyGqicy2igci6jgcq5jHSDLuND0oePiIiIiIiIiIgMH2rSJSIiIiIiIiIywijhIyIiIiIiIiIywijhIyIiIiIiIiIywijhIyIiIiIiIiIywijhIyIiIiIiIiIywijhIyIiIiIiIiIywijhIyIiIiIiIiIywijhIyIiIiIiIiIywijhIyIiIiIiIiIywgQGu4La2lo7derUIQhFREREREREREQAFi5c2GCtrRvo8oNO+EydOpUFCxYMdjUiIiIiIiIiIpJjjFk5mOXVpEtEREREREREZIRRwkdEREREREREZIRRwkdEREREREREZIRRwkdEREREREREZIRRwkdEREREREREZIRRwkdEREREREREZIRRwkdEREREREREZIQJDHYFixsXc8jNhwxFLNJH+9ftz69P+nWhwxg1VrSs4Lx/nJd/bjA9hrknPV8zZofzd73uGAfHOLjGffvQ6fk8P+70Mm9uGHSCBN0gWxJbWNu2li2JLWD9bdqukW3GAazt+Xx7ur+vHtPp3/TeJm93Hf3YpsEQCUSIBqJEA1FKg6VUR6qpidZQG63lwDEHsl/dfriO23tcRebptU9z5xt3srFjI03JJjzrkbVZPOthrfWHWAwGYwz5f6aX5xjeOe2dfPbAzxb6bUlOe6qdv775V57d8CyvbXmNZDaZ/5taa/PD3mz7vdne96WXiX2ab6Dr3973eSDLDfQ9DfX6d/aeuvZLpcFSYsEY0yqmceSEIzl8/OEYY/Csh2McMk1NJF55leTrS8lsbiDb0oLX3o5Np7GZDDaTgUwGEwpholGcaBQnGsFEoriVFZSffDLhmTN3GIuIiMhA3fjKjTy+5nGak82kvTQBE8B1tp6jOMbJH1f2OIdxXEJuiIgbIeSGiGfitKfa6Uh30JnpzM/TfX3dn5cESjhl6ikcP+X4Qn8ERWnQCZ/qSDXnzj53KGLZrfp6gjscjY+NL3QIo0osFOOsWWcBvZebruRJ12s9Eiu9TOv+WtfJ+bZDz3pkva3Tuh5d09M23etyaS9NKpuiKlLFrKpZHBY9DNf0ntzoywlbb++zt/fRH9v7LIZimxZLPBPPPzrSHSxuXMyW+Bba0+0AzKiYwfeO/h57Vu/Zr7iHm9uX3s63n/k2ddE6ZlTOYELZBIJOEIPJ/+g6xq/E2T0x0DXelTgA//P08JhYNrFg70d6emrtU1zxxBW0plqZXDaZeePmURYswxjT44Cqt+/tQBO6ff1u9mX92/vu7mz9vcbQj33Gzpbta6wDfU+9LefhEU/H6ch00Jxs5o7X7+CmxTdxxPgjqCup4+HX/s6VT09i7IIVkMkAYKJR3IoK3LIYBIKYQMB/uC5eRwfeli148Tg2HsdLJPDa29ly/a+Z+KtfEjviiJ3GOVBeKkV6zVq81hbCe++NEwrtsm2JiMjw0pnpxLMek8smE3bDZGyGjJcha7P5Y8muY8yu85OUlyKbyZLMJklmk6S9NBE3QmmolFgoRl1J3dbzHS9LxmbIellS2RRxL07GZmjobOBfK/7FDSfdwCH1qmjSX2awiY958+bZBQsWDFE4IiK7RkuyhSfXPslPFv6EeDrOn079EzMqZxQ6rAGx1nL6X0+nMlzJ70/+PUE3WOiQZACaEk280fQG0yunUxutzU9f2riUD973QaZVTOObh3+TOTVzChilDLV0Ns2tS2/l5y/8nGQ2ydw30nz5Lx6V730v5aedRmSvPXErKvq1zszmzaz62MfJNDQw7a93Exw79m3zZJubiS96hczGDWRbWnFiMdyKcsJ77EF4+vT8fPEXX6T5zjsBg4lGsMkU6dWrSK1cRXr9evA8AMJ77smUP/4Bt7x8UJ+HiIjIjiQyCY669SjO2eMcrjjkikKHs9sZYxZaa+cNeHklfERkNFnfvp733fs+ZlXN4oaTbuhTE5Ph5q3mtzjzb2fyjcO+wTl7nFPocGSAHljxAJ9//PMAzK6azWnTT+OAMQfwtae/Rme6k9tPv71HIkhGls60f6X0nmsv54DfPM3U+/9JdMrUAa8v+dZbLD/nvQTq6qj+6EdwoiVkW5pJr11H54IFJJcsge0c84VnzaLu8suI7Lsvy894N15HB05lBTaewLguwSlTCE2e7D+mTMbr7GTD//sWY7/8Jao/8pEBxywiItIXFz1wEZvjm7n73XcXOpTdbrAJn0E36RIRKSb1sXo+tf+nuPK/V/LEmid4x6R3FDqkftsc3wzA1PKphQ1EBmV+/Xx+feKveb3pdR5Y8QA/XvhjAMJumOtPvF7JnhGuJFgCwMR0GQBHPfAuzt3/I1x+0OUEnf7X2gvPmMHkG37L+q9/nY3f+nZ+ugmHiR5wALWXfoaSAw8iOHEibmUFXns72eZmOhcupPnWW1nz6c/484dCTL3jL0Rmz97h9ppuvY3Wf92vhI+IiOxy+43Zj+tfup60lx7Qb+RopoSPiIw65+xxDjcvvplfvPgLjp54dNHV8mlNtgJQHlZTimJWEa7gsPGHcdj4w/jInI+wqnUVL25+kf3r9mdy+eRChye7yeRMBa0hhyNmHM8fX/sjzclmvnPEdwa0Xyo58ECm//3vpNesAcCtqMCJxTDO22/K6sZiBMeNI7LnnlS997003XEHmXXrKD/jDCJ77LHTbcWOOYYtv/0tXjKJEw73O1YREZG+qi+tx2LZ1LmJCbEJhQ6nqCjhIyKjTtAJ8qG9PsR3/vsdXml4hX3r9i10SP3SmsolfEJK+Iwkk8snK9EzCgVbOympq+ea467hupeu4xcv/oLKcCWfO+hzA7qjoDGG0KRJ/VsmFKL6Ax/o1zKRvfaEbJbkm28SnaN+pkREZNcZVzoO8LtmUMKnf95+yUdEZBQ4ZdopAPx73b8LHEn/KeEjMnJktzTi1tQA8Mm5n+S82efxx9f+yPvufR9PrX0Kz3oFjrB34VyTr+SSpQWORERERrr60noA1nesL3AkxUc1fERkVKoIVzC9YjqLGhYVOpR+a022EjABooFooUMRkUHKNDYSHOdfuTTG8JX5X2HeuHn8ZOFPuOShS5hUNokp5VPY0LGBC/a5gNOnnz4smqGGJk/GRCIk33ij0KGIiMgI11XDZ0PHhgJHUnxUw0dERq25dXOLM+GTaqU8XD4sTvpEZHCyjY24VVX558YYTp56MveceQ9XHXUVY0vGsqZtDa5x+epTX+WmxTcVMNqtjOsSHDeO9AYdfIuIyK4VDUSpDFeqhs8AqIaPiIxak8om0ZhoJJ1NE3SLp8f/1lSrmnOJjBBeRwdOrPRt00NuiHdNfxfvmv4ufz7rcclDl/Crl37F2bPOzt/lq5ACY8eS2bix0GGIiMgoUButZUt8S6HDKDqq4SMio1YsGAOgLd1W4Ej6pzWphI/ISGCtxevsxCl9e8JnW45xuGS/S2hLtXHf8vt2Q3Q7Fxg7RgkfERHZLWqiNTQkGgodRtFRwkdERq2yUBkAbakiS/ikWikLlxU6DBEZJJtIgOfhlPStts5+dfuxZ/We3PjqjaS99C6ObueCY8eS3rwZ6w3PjqVFRGTkUA2fgVHCR0RGra6ET3uqvcCR9E9HuiNfO0lEipfX2QnQ54SPMYZP7fcpVrau5G9v/m1XhtYngTFjIZ0m29RU6FBERGSEq4nU0JhoxFpb6FCKihI+IjJqFWuTrrSXJuSECh2GiAxSPuHThyZdXY6ZdAz71e3Hr178FYlMYleF1ieBcWMBSK9TJ5oiIrJr1UZriWfidGY6Cx1KUVHCR0RGrWJt0lVsnUyLSO+8jg6g7zV8wK/lc/mBl7Mpvonblt62q0Lrk/C0aQCklr1V0DhERGTkq4nWAKhZVz8p4SMio1axNulKeSmCjhI+IsVua5OuvtfwAZg3bh4HjjmQO16/o6BV20NTpkAwSPLNNwsWg4iIjA61kVoAGuLquLk/lPARkVErFvKbdLWmWgscSf+kvbQSPiIjQL6GT2n/b7F+5swzWdG6gpc2vzTUYfWZCQYJT51K8vU3ChaDiIiMDvkaPgnV8OkPJXxEZNQqDfhX1dvTRVbDJ5si5KoPH5Fi53UMrIYPwElTTyIaiPK3twrbeXN41izV8BERkV2uK+GjGj79o4SPiIxaruMSC8aKqkmXtdbvtFkJH5GiN5BOm7uUBks5YfIJ/HP5P+lMF64Dy/CsmaTXrs3XVhIREdkVqsJVOMZRHz79pISPiIxqsVCsqJp0ZbwMgJp0iYwAg2nSBXDWrLPoSHfw0KqHhjKsfgnPmgVA8i113CwiIruO67hUhatUw6eflPARkVEtGogW/NbG/ZH20gC6LbvICLC10+aBJXwOGnsQ9aX1PLjywaEMq1/yCZ/XX3/baxs7NvL02qd1NVZERIZETbRGffj0kxI+IjKqRQNR4pl4ocPos1Q2BaDbsouMAF5nJwQCmNDAErjGGI6ZdAzPrHumYPux4MSJuBUVdD63oMf0bGsrP/zP97j4oYu55617ChKbiIiMLLXRWl1E6CclfERkVIu4keJK+Hi5hI+adIkUPa+jA6ekBGPMgNdxzKRjSGQTPLPumSGMrO+M61J61FG0P/kkNpMh09DAuiu+xOuHzOfcLz1MRQecPPXkgsQmIiIjS220Vk26+kkJHxEZ1aLB4qrhk2/SpU6bRYqe19k5oA6buzt47MHEgjEeWf3IEEXVf+Wnnkq2sZHlZ53NWyedTMs//kFo5gxK29McH9yH8bHxBYtNRERGjppoDVviW7DWFjqUoqGEj4iMaiWBkqLqwyffpEs1fESKXlcNn8EIukGOnXQsD658sGB36yo77ljqPvc53MpKyk4+mel/v4f6b30LgFjaLUhMIiIy8tRGakl5KdrSbYUOpWgo4SMio1qx9eGjGj4iI4fX2TnohA/AOXucQ0e6o6C1fGovupApf/wD46+6kvC0aTilMQBiKR1qiojI0KiJ1gCoWVc/6FdYREa1okv4ZP2Ej2r4iBS/oWjSBbBf3X5E3AivbXltCKIaGm7Mf1+lyYH3TyQiItJdbbQWQB0394MSPiIyqkXcCIls8TTp0m3ZRUaOoWjSBeA6LrOqZrG0cekQRDU0nLIyAKJJ9bMgIiJDoyvhoxo+faeEj4iMal2dNnvWK3QofaLbsouMHEPVpAtgdvVsljQuGTYdWZrc+1LCR0REhooSPv2nhI+IjGrRQBSgaDpu7qrhoyZdIsXP6+gYkiZdAPvW7ktrqpU3mt8YkvUNVsZ4JIIQUcJHRESGSHmonIATUMKnH5TwEZFRrSvhUyz9+HTV8FGnzSLFbyhr+Bw+/nAA7n7j7mFRyyeeidMZhlAiW+hQRERkhDDGUButVcKnH5TwEZFRLeJGAIqmHx/V8BEZGWw2i43Hh6yGz7jScRw87mBuWnwTVz17FVmvsImWfMInniloHCIiMrLURmrVaXM/KOEjIqNaNJir4ZMukho+nmr4iIwEXtzf5wxVDR+A6064jg/v/WFuWXILn3/887SmWods3f0Vz8SJhyCYSBcsBhERGXlUw6d/lPARkVGtJOCfbBVLky7dll1kZPA6OoGhTfiE3BBfOPgLXHHwFTy6+lHO/OuZPLHmiQGta7DNwvwaPoZAZ2pQ6xEREemuJlqjhE8/KOEjIqNaVx8+nZnOAkfSN/nbsquGj0hR8zraAYasSVd35+99Pn8+7c9URar49MOf5iP//AivN73ep2VT2RT/+/j/csztxwzqNu/xTJw1tRB8YxXpdesGvB4REZHuaqO1NCWbCt50uVgo4SMio1pVuAqApmRTgSPpm/xt2VXDR6SoZTb7VycDtTW7ZP1zaubw59P+zBUHX8HK1pWcd+95/OC5H9CcaN7hcr948Rfcv+J+GhONXP7o5bS8fj/cfTGJ2z/C7c/9lO888x2+88x3WN22eofrSWQS3DvfAWNovvvuIXxnIiIymtVGa/GsR2OisdChFIVAoQMQESmk6mg1AI3x4vjRUKfNIiNDZtNGAAJjx+6ybYTdMOfvfT6nTDuFa56/hpsW38Rdb9zFBftcwPl7nU9J8O3NyR5e9TBHTzyaj+/zcT5+/wWc8vTnmJS1rHQdOl97nqAJkLFZnljzBLe96zaqIlU9ln+r+S3ufONOHln1CFvKDYHf/YjaeSfusvcoIiKjy5iSMQBs7NxIXUldgaMZ/lTDR0RGtYpQBY5xiuYqQWemE4MhEogUOhQRGYTMpk0ABMbsuoRPl9poLd8+4tvcefqdHDzuYK594VpOu/s0bnrtJt5oeiPfX097qp2VrSuZWzqRA5++jj9sauWd2RA1k4/gtGmn8LvmNAvbQtxy6p/Z3LmZa56/psd27lt2H+f8/RxuWXILMypn8NX5X2XmvBMwxuzy9ygiIqPD+Nh4ANa1q7lwX6iGj4iMaq7jUhmuLJqET3uqndJgKY5Rvl6kmKU3bsQpieJGw7ttmzOrZvKz437Gi5te5CcLf8LVz10NwPSK6ZyzxzmEHL9vsL1fuRc2rWDu9GOYe9J3oHqav4KaQ+HOjzNn05uct+d53Lz4Zs6edTb71u3Lqw2v8tWnvsp+Y/bjR+/4ETXRXdNUTURERreuhM/6jvUFjqQ4KOEjIqNedaS6eBI+aT/hIyLFLbNmBQG3Fa49ED77Ajjubtv2/mP258ZTbmRl60oWbFzA3W/ezfef+z4AteFK9lu5CN75Qzj4Ez0XnPMeeOx78MQP+NiH7uL2pbfzgfs+wKnTTuW5Dc9RHa3mmmOvoSJcsdvei4iIjC7loXJiwRhr29cWOpSioISPiIx6NZGa4kn4pNopC5UVOgwRGaTM2tUEoh40r4Rlj8HM43fr9o0xTK2YytSKqZyzxzksbVzK0qalHN6whvIlL8P0Y9++kOPC0V+Auy+ibvWz3PKuW7h96e3cu+xegk6QXxz/CyV7RERkl6uP1bO+XTV8+kJtAkRk1KuOVLMlvqXQYfRJW7qNWDBW6DBEZBCstSTXrCdUlvEnvHTr1hczSdi8FDa+BunEbotpdvVszphxBrUbXoNoFVRP733Gfc6Gmllw10Xs8eKdfO3Qr/Gf9/+Hx973GHPr5u62eEVEZPSaXDaZlW0rCx1GUVANHxEZ9aZUTOH+lffTke4Y9s2lOlIdVER0BV2kmGU2b8brSBKudmD/8+HFm2DLGxCIwLoXIJNL9DgBGLMX1O/v16ypmrLrg1uzACYeDNvraNkNwIf/Brd/GB7/Huz7XkztTFyz+5qkiYjI6DazciaPrX6MZDZJ2N19feEVI9XwEZFR74C6A/Csx0ubXyp0KDvVnm6nLKgmXSLFLPXmmwCEJ42Bfc7yJybbwHow72Nw1m/g7Bvg8M9CbCy8cif85lh45S7I3VFrl0i0wuYlfsJnRyomwHk3g3HgxZt3XTwiIiK9mFk1k6zNsrxleaFDGfZUw0dERr25dXNxjMOLm17k8PGHFzqcHWpLtRELqUmXSDFLLF4CQHj6VL/vnkuf95tQbVurZt9z/GHDG3DnJ+COC2DRX+C0H0N5/dAHtnYhYGHivJ3PWzYOZp7gN0c77mu7tdNpEREZ3fao3AOAJY1L2LN6zwJHM7ypho+IjHqxUIxZlbN4YdMLhQ5lpzrSHerDR6TIdf73GUJlGQLT9vUn1MzYfhMqgNpZ8ImH4cRvw1uPwC8OgT+cDtcdBX84A576CXjZwQe26A4Ilu68hk+X/T8Ibetg2aOD37aIiEgfTa2YSk2khqfWPlXoUIY9JXxERPBvU/zy5pfJeJlCh7JdaS9NIptQwkekiNlUis6FCykZk4S6flyVdANwxGfhkn/DrJMg1eHXsok3wkPfhPu/OrjAmlbCK3fA3PdBuI/NRme/0+/gecHvB7dtERGRfnCMwzGTjuHJNU/SkmwpdDjDmpp0iYgA88bO47alt/HCphc4eFwfr27vZm2pNgA16RIpYi333IPX0UnZxATUze7/CmpmwDk39Jz2ry/DM7+Eyklw2Kf7v850Au6+2O+T5+j/7ftygTAcfCE88X14+md+Qmpb2TRg/ISViIgUreY77yT51jKic/clOncugfp6zI5qp3aTXreOtocepvVf/yK5ZAkWCI4bR3j2HkRm7+kP99yTwLhxfV7neXuex91v3s23n/k2Vx11FUEnOIh3N3Lp11dEBHjHpHdQFizjliW3DNuET1eTs2kV0wociYgMRHrdOjb96MdEJpZROqUdavcYmhWf9B1oXgX3fwWe/LFf86e0FkrroKwe6veD8Qe8vZ8ga/1Omh/8Bqz6t99RdMXE/m37mC9Bw+vw4P/Bkn/4CalQqb9uL+03QTv2a7DfuUPzXmWn/v7W37lt6W3UReuoK6kjFowRC8WIBWPURGuojdZSE6mhJlpDNBAtdLgiUiQSi5fQfNttNKbTAJiSEtxYjGB9PYHx9QTrxxOsr8etrCDb1kZy6eukli0j+eabZJuaAAjvtRcV73kPJhAgtWYNiUWv0PbPf+W34ZSXE545E7eqivC0qUTmziU6Z46fCHJdrLX5hNCe1Xvy2QM+y0+f/ylNiSYuPeBS9qvbr88Jo9FiSBI+Npv1f9hdVx+wiBSlaCDKB/f+INe9dB1XP3s1p884nRmVM4bNrR496/HXN/5Keah80Akpm8ngdXZik0m8ZAqvrZWOZ/5LaMpkyo47bogiFhmdrLX+dysex+voxOvoILNhPZ0Ln6f5zjuxiU7GH7EB9jufNe0ea5q2EHQNdbEItWUhHGOwFiwWz+bWl1t3wDG4jiHgODiGrcdcjgvn/B6e/wNseBk6GqBjMzStgNb1kE3684XLoaQGgiX+XcESzZBsBTcE7/w+7HsOzyzbwo8feJ23NrcTcA1B1yHkOgRdh5Kwy4y6GLPHljG9rhSAseURZp95PcGJ8/w+gJY/4a/bGDCun9QqH7+7/wyjmmMcIoEIS5uW8uyGZ2lPt+NZr9d5S4Ol1ERqqI5UUxGuoCJcQXmoPD9eEfKHpcFSIoEIYTdMNBAl7Ibzzx2jHiJERoNxF57JmPcfS3JjnPiSN0mvXU+2vYP0hk0kX3uN9ocfwaZS+fmdigqCY8ZQduKJhGfOpPSIwwlPmQTpDkh1gpcBN0g2nia5bCXJt1aQeP1NUstXkF61ivYnnoBccgnHwQQCWM+j9JBDqDjz3ZSdfDIf3/fjVEWq+NGCH/Ghf36IKeVTmDd2HntV78WUiilMjE1kbOnYUV37x9hB3t5zTmmJ/cvkKZjcaizgH4UAroMNOBDIDYOOPy3o+AcCjn+gYv2jFn+Z3NA6Btg6D/3OI+3OxJMF2/Npz5e3mWC3Gdnusjb/uXZ/PVs/jgN+/ZeBhSr9tumFZ9l02SU9J3Yvr11l1Nk6zRreVqZ7XFXt/txss95tp/WXxd/+dl4bzHqL3k72dxZ4y+1gk5PMTwvhELIODuBiMLkPt8efrcf49v94ttuH2Fsk9m3/93lA3GRJGY/J2SiTOoKYtOfvS23ufdnc/sKz+YfxLGS9ns/THibT+4F/Zv4c9v3DHduNX4bW63/4Gfa31/n7DsfB5vYn1ji5acafZoCs9QtC1/6Frb+b1uR+L7vGc/uR/LK5ee0OLshs+0p+HzKI772F3G/YDsr9Dte//RctvcScm930Jej8z+/b5zVe7jfd8zDW5h94uaHNfZc863+XMhayFpOxufHtbNNYonVpxh3YzJraqXzS+zJvdQ6udoXrmPzPSSToUlMaorIkRNA1GGNwDATJMimzkhnp15mcXkbMayNskyTcUjqcMtaHpvBs8BBWZypY3xxnU1uSCZVRjt6jDmstqaxHOmtJZzxaE2ne2NTO5rZkjzhCAYdx5RGqSoKEAg7GGFzjJ6cAPjh/Mu/cdxfcVUx69eo3P4+576Gtx9O5/UbWgayBjGPJGEg7kDGWtNPtYSwZY8nmvmBd+wIL/vroOX3bcQP5fY3pPrHr17H7eLeZbH4bQ3/8vqvOCIb6sKivcQ7Fdne4jm1OPobqffZ9PX3f/pT3fYyTz/7cACOS/lp+/fuZtv6+7b5uLWSTDulUABtwcEvAOi6ecXCsR9BL4G73R3IrD4esCZD2giSaQySbAqTjAaxnsFlLag1k243fG3F5EC/o4gUcWtwsrU6GDscjY/x9U9e+C8AxBoP/2+jQ7Ripa9Dj/Cy3H8sfU+X2U93m27o/7PntzT/r7WCFt33Fei7TY5o/9b23L1xore3D7TN7N+gaPo7rUbpXyv/Ac3tra/03Yj2DzYDNGsiCzYJNG2wC/8CV3AFa16fXbfmte36Gfo+6rX6uv8ffrve/zs6n9XZev93lbI/XU5mOPkQpQ6Uz3kHEbes5sausekCGHmXY2p7zvG1ab8P8eG7n0ZcyuZ0jgz6d7AxgvSPGTt7fHGAvIGUMGQNZvNyBr+mxW+rSa+Jmu3+bgYVnLLhYwhaiXjsmAE7YYpPkz/vp9jCuhWBuugM4FhyDcSy4YIJgghbjAK6BgMWtsSyfPqYPEcpQiRPEuCE/iZCxOJ6/UzE2k0swdEuYOJA/Numx/7Bv25+YXvczfia492OP3ZDN3WHh3wU7ncGssuu75HQb755c63rugo0aCBhswMVzc+Oug5e74OWFHGzQJRsJ0F5bRVO4hteDs3ktdgTzqmJ8dGIF02tLSWc9NrUl2dKewmLzB6T+5kz++DPrWTKe7Tb0crWBIJ7KsqUjRXNniqxn8axfQyhhHZaaaSwJTsUL+DWG/HxVrgZRxlLiuJRHAsyYVcv+kyp570GTiIa2f5v1xo4Uyxs6cB3DqsZOXlnbwsbWBE2dadIZj6y1ZDyPZMYvW2lvJFwxKB4dxKmqSvQ4DslfFOh6ePjJyW2PVbrGu2YxBg//ukGP1W37m2h72ZPs7Aez2/PuJ0BDkG+WnKI6pOulDOxM84aGXRKK9O6+2gtY1bg/tV4jLlkcm8XFw8HDtRkcPBzXwy3xcMh0ey2LZw1xEyFhwsSJECdMBpcAWYJk8o8AGYI27T93MgSqMwSrMwTI5pJFBru/pWJjK7ENbYTaUgQzKcjAmDTUAVj/9Cxl/GHGkP+t9DD5ceh5zNQ13uu03HTTy7TtJXZyq+h9+m78cg66hs+8efPsggULhigcERERERERERExxgyqho8a3YqIiIiIiIiIjDBK+IiIiIiIiIiIjDBK+IiIiIiIiIiIjDBK+IiIiIiIiIiIjDBK+IiIiIiIiIiIjDBK+IiIiIiIiIiIjDBK+IiIiIiIiIiIjDBK+IiIiIiIiIiIjDDGWju4FRjTBiwdmnBEhq1aoKHQQYjsYirnMhqonMtooHIuo4HKuYwGs621ZQNdODAEASy11s4bgvWIDFvGmAUq5zLSqZzLaKByLqOByrmMBirnMhoYYxYMZnk16RIRERERERERGWGU8BERERERERERGWGGIuHz6yFYh8hwp3Iuo4HKuYwGKucyGqicy2igci6jwaDK+aA7bRYRERERERERkeFFTbpEREREREREREYYJXxEREREREREREYYJXxEREREREREREYYJXxEREREREREREYYJXxEREREREREREYYJXxEREREREREREYYJXxEREREREREREYYJXxEREREREREREYYJXxEREREREREREaYwGBXUFtba6dOnToEoYiIiIiIiIiICMDChQsbrLV1A11+0AmfqVOnsmDBgsGuRkREREREREREcowxKwezvJp0iYiIiIiIiIiMMEr4iIiIiIiIiIiMMEr4iIiIiIiIiIiMMEr4iIiIiIiIiIiMMEr4iIiIiIiIiIiMMEr4iIiIiIiIiIiMMEr4iIiIiIiIiIiMMIHBrqAx0ch3nvkOnvXI2iye9YYiLpFhpSRQwpfnf7nQYYwaqWyKG1+9EWstHh5Y8PCw1mKxGAzGGBwcMODg+M+Nn8N2jIPBEHAClARKKAmWUBIooTRYyrSKadREawr8Dncday3xTJxUNkXGZkhn02RtNv9ZZm0Wz/OHWZsl62Xz4wCRQISoGyUaiPrjgShhN4wxZre+j3Q2zYaODazvWE9DvIHGRCNtqTYyNkPW839rMjaDtTZfFhzH8Yem56M0WEosGKM0WEp1pJoxJWOojdZSEizZre9JREREpDdZL8tbLW+xoWMDHekO4pl4/niu67i261jX4A+zNkvGy+TPwV3jEnSCVIQrqC+tZ0JsAmNKxuA6bqHfnhTQoBM+HekO7l9xP65xcY2LMWa3nxiI7GrlofJChzCqpLIprn3h2rdN7/qx60r8DFR9aT2nTDuFi+deXPQn/avbVvPiphdZ0riE15te5/Wm12lMNA7pNgwmn/wZUzKGw8YfxpHjj+SgsQf1OIjY3LmZhRsXYrH5iwD5hJKXJZlNkvJSJDIJktlkfpjMJunMdBJPx+lId7A5vplNnZt6/Rs7xsn/3nQldLq2t+1jZ2UkFoxRG63NJ4DGlY5jz+o92bN6T6aUT8knEEV2Fy8eJ/7SyyReWURi8RLSGzaQbWoCazGhULdHkEB1DcEJEwhOnEDpYYcRmjSp0OGLiEg/tKfaeXT1ozy19ikWblzIxs6NQ76NgAkwtnQsE2IT2K9uPz66z0d1XjPKGGsHftIEMG/ePLtgwYIhCkdExK+lkvEyfgI5dxWjt0SytTZ/Yt+VBOr+PO2liWfidKY76cx00pps5Y3mN3h+4/M8svoRppRP4Xcn/44xJWMK8C4H77eLfss1z18DQMgJMbNqJntU7cHU8qlEAhECJkDQDeaTJF01YVzHzSdN8uOOi7WWRCZBIpvIX1mKZ+IkMon8cHnrcl7Y9AIZL8PMypl8Zv/PMKFsAo+seoQbFt1Aykv1KfaIGyEcCBN2woQD4R41sWqiNUyITaC+tJ7xsfHUldRRE6mhLFTWrySMtZaszdKZ6aQj1UFbuo3GRCObO/2EUkO8gU2dm9gc38zmzs1s6NxAxssAfq2+2dWz2at6Lw4edzBHTzyakBvq/x9Jik7aS7Nw40IWbFhAR7qDVDblJyuzKX9/kukkkUmQ9tJkbAbP84iFYlRHqvOPmmiNP4zUMLt6NhXhiu1uz1pLNpmg8Re/ovHmm7GdnQAExtcTmjgJt7ISXAebSmNTKWw6jU0myTQ0kF6/HjIZnJISptz0JyJ7772bPiURERmM25bcxg8X/JBENkF1pJoDxxzIOya9g+kV04mFYkRdv6Z12A33uLAF5Me7jt8CTsCv8eNlSWVTtKRaWNu+lnXt61jXvo617WtZ07aGV7a8wuSyyfz+lN9TG60t8CcgfWWMWWitnTfg5ZXwEZHR6L/r/8ulj1zK/HHzufb4t9cmGu4aE42c8JcTOGLCEVx6wKVMr5hOwBl0pc0+6Uh38Njqx/jVS79iZevK/PRTpp7CR/f5KFE3ijGGgAn0SCiF3BBhN0zICQ3LmqDpbJq3Wt5i8ZbFLG5czJLGJSxpXEI8E6ciXMFp007j3TPfzV7Vew3L+GXnktkkj69+nFVtq/LNPruaQcYzcdZ3rOfpdU/TlmrDMQ4lgRJCbihfdqOBKCWBEiKBCEEnmC/bralWGhONNMYbaUo29WjebjAcPO5gvnvkdxlXOu5tMV3xxBVM/Mu/OeHBBspPPZXyM04nut9+BKqqdvp+bCZDauVKVp7/ISJz92Xy9dcP+jNKLFlCavVqguPqCU2ehPU8Gn/3O1r+dg+4LhWnn07dpZ/BBIOD3paIyGj0rxX/4guPf4EjJxzJJ+d+krl1c3dLreKFGxdy0QMXccKUE7j66Kt3+fZkaCjhIyIyQN/6z7e4f8X9PHXeU0V3Av/Emif49MOf5vcn/5554wb8GzAoaS/NE2ueIJ1Ns1fNXkwpn7J7trt+PYnFi/HicUoOPJBgff0u21bWy/LM+mf465t/5ZFVj5DyUuxZvSeXHXgZR044cpdtV3aNW5bcwpX/vbLX18JumIpwBYfWH8pxk4/jsPrDBtTk07MerUk/AbShcwMvbHqBG1+5kX1q9+GGk2/ocVC/pm0N77zrnfz6mgzjD3kHU667bkDvq+FXv2LzNT9j2t/+SmT27J3Ob60Fz8O4Pft12HLjjWz63jYnAa4Lnkfs+OPAQvvDD1Nx9lmM/+53BxSriMho1pJs4Yy/nsGE2AT+cMofCLq7N3n+04U/5YZXbuCBsx+gPrbrjp9k6Aw24bN7LgeLiAxD0yum56/MF1tHzq83vQ7AHtV7FCyGoBPk+MnH77btZdvb2XjVVbTcdTd0u1gRmTOH4ORJBKqqcCsrcWtriey5F5HZe+CUlg5qm67jcsSEIzhiwhG0JFv41/J/cdPim7jkoUu49IBLuWjuRYN9W7IbnTrtVKaUT+GAMQcA5GviRNzIkHVq6RiHykgllZFKpldO5/DxhzO+dDxf//fX+dNrf+Ijcz6Sn/fR1Y/iZi2VnZCaPfCEadUHPsCW3/yWLb/+DRN+9MP8dJtK0Xz3X2l/9FGy7W0Y45DtaCe9ajVeMkl0n32Izp1LYEwdNuux+ac/JXbC8dR+8pNkNm4ktXIV2aZGKt79bsKzZgGw8erv03jjjdR89KP5aSIi0je/fPGXNCebue6E63Z7sgfg7Flnc8MrN3D/ivv56D4f3e3bl91PCR8RGbWmVUwDYHnL8qJL+CxtXMqE2IRR0/Ge19HByve/n+Rby6i+4ALKTz4JEwrR/uRTdDz5JMnFS+hsaiLb2ppPBplQiPLTTqPu0s8QHD9+0DFUhCs4d89zOWvWWXz1qa/yixd/wQmTT2B65fRBr1t2j4pwBYePP3y3b/fMmWfy2OrHuOb5azh8/OHMqvITJU+tfYqSpD9PSygz4PW7FRVUnneen4j55EXgebQ/9hhNt91OZv16QlOnEqirw3oewboxRPfZF6e0lPjzz9N0223YRAKA6LyDGP+97+HGYrDvvr1uq+bCT9B444203v8AdUr4iIj0WTKb5O437+Zd09/FXjV7FSSGSeWT2KdmH/654p9K+IwSSviIyKiVT/i0Li9Ys6iBWtu+lsllkwsdxm7TcN11JN98i0m//jWxo7Y2pYrstRe1F12Yf26zWTKbN5N47TXan3ySlr/+jbZHHmHyb35NdO7cIYkl6Ab50vwv8ejqR7nx1Rv51hHfGpL1yshljOEbh3+D9/ztPXz1qa9y86k3c+OrN/Lvdf/m/XXHAg/RGEgMahs1F3yUljvvZPkZ785PKznkEOq/9S1Kjzxiu81WrbV4bW1kW1sJTpiw0+atgepqonPn0v7449R95tODillEZDT5z7r/EM/EOW3aaQWN45Rpp/DDBT9kecvy/LGwjFy656yIjFp10ToAGuNDexvz3aE93U5ZqKzQYewWXjxO0+1/oezkk3ske3pjXJfguHGUHXcc9d/4BtP/9lfc8nJWffQCOp59dshiqo5U845J7+DZDUO3ThnZqiPVfP3Qr7O4cTHH3H4MP3vhZwCcP/FMADa5nYNaf6CujonX/YqaCz/B+B/+kJmPPMyUP/6B2FFH7jCJY4zBLS8nNHFin/syix3zDhKLFpFpaBhUzCIio8mCDQsIOSEOHndwQeM4ddqpRNwI175QfDctkf5TwkdERq2gGyTiRmhPtxc6lH7rSHdQGhxc/zTFovW++/BaWqj6wPv7vWxo8mSm3HQTgfH1rL7wItqfeGLI4tqrei/Wtq+lNdU6ZOuUke34KcdzwZwLaE21Uhmu5I7T76A6HQZgnWkZ9PpLDjiAMZ//PBXvOm1ImjFuT+wd7wCg/cmndtk2RERGmqVNS5lROaMgffd0V1dSx4VzL+TBlQ9y+9LbCxqL7HpK+IjIqBYLxWhLtRU6jH4bLQkfay2NN99MeNYsSg4e2BWx4NgxTPnTnwjPmMHqT3+G+KJFQxLbntV7An5/SiJ99bl5n+OO0+/gjtPvYHb1bLw2P2G42jYVOLK+C++1F05FBfGXXix0KCIiRcFay+tNr+ePHQrt4/t8nKMmHMVVz17F8xufL3Q4sgsp4SMio1osWHwJH896dKQ7iIVihQ5ll4u/+CLJ1xZT9cEP9rm5SW8CVVVM/t0NBGprWfv5/yXbPvhaXbOr/dtfL2lcMuh1yegyu3o2Y0vHAvgdjQMrvE2FDKlfjDEEx48ns2FjoUMRESkKWxJbaEw0skdV4e6u2p3ruHzv6O8xITaBzz32OTZ0bCh0SLKLKOEjIqNaWais6Jp0dab9vj5iwZGf8Gm6+c84sRgVp79r0OtyKyuZ8MMfkF67ljUXX4LXObg+U2qjtdRF65TwkUHJtvoJ53VOS/67XQyCY8eS3qiEj4hIX6xpWwPA5PLhc8ON8lA51xx7DfFMnMsevYyGuPplG4mU8BGRUS0WjNGeKq6ET0e6A4CSYEmBI9m1UqtX03r//VS85z04pUPTfK3koIMY//2r6Xz+eVZ/6tNk2wZXu2t29WwlfGRQvNYWbNAlFYDlLcsLHU6fBcaOJaOEj4hIn6xuWw3AxLKJBY6kpxmVM7j66Kt5s+lNTr7jZK55/hoSmcHdNVKGFyV8RGRUi4VitKWLq0lXV8JnpNfw2XjlVTjBIDWf+PiQrrfitNMY/72r6Hz2Wd48/gTWf/0bJJcN7ER7z+o9Wda8jHQ2PaQxyuhgraXzxRdxa2rBGP697t+FDqnPguPGkm1sxEsmCx2KiMiwt6Z9DQbDhNiEQofyNsdMOoa73n0XJ009id8u+i0XP3QxqWyq0GHJEFHCR0RGtbJQWdHV8OlqgjaSO21u/ec/aX/0UWo//WmCY8cO+forzjiDqbffTuyYd9Byzz2seP/7B5T0mVo+lYzNsK5j3ZDHKCNfYtEi4gsWUveJC5lTM4cn1z5Z6JD6LDB2HACZTcXT95CISKGsaVvDmJIxhN1woUPp1ZTyKVx11FVceeSVLNy4kP/3n/+HtbbQYckQUMJHREa1WDBWdH34dMU7Umv4JN96i/Vf/RqR/eZS/eEP7bLtRPeZw4Tvf5/p9/wN47qsvvhiMk39u1NSV1v8Va2rdkWIMsIl31oGQOzIIzhq4lG8tPklWpKDvz377hCoqwMgs3lzgSMRERn+1rWvG5a1e7Z1+ozTuWS/S7jnrXu4afFNhQ5HhoASPiIyqsVCMeKZOGmveJrkdHXsOhJr+GTb21nzmUsx0SgTr7kGEwzu8m2GJk9m0i9/QWbjRtZ8+jP9aqIyqWwSAKvalPCR/kutWgmuS3D8eI6acBSe9Xhq7VOFDqtP3KoqALLNzYUNRESkCGzq3MTYkqGvsbwrXLLfJRwz6Rh+uvCnvNn0ZqHDkUFSwkdERrWyYBkAHamOAkfSdyO1SZe1lvVf/gqpVauY8OMfExw3brdtO7r//oy/+mrizz/Pxiuv6vNyNZEaooFo/u4bIv2RXrmKYH09JhRiTs0cxpaM5e/L/l7osPokUFUJQLafteJEREYbay2bOjdRV1JX6FD6xBjDNw/7JrFQjK889RX1U1jklPARkVGtLOQnfIqp4+Z4Jg5AJBApcCRDx0ul2PCtb9H24IOM+d//pXT+Ibs9hvJTTqb64x+j+bbb6HzuuT4tY4xhctlk1fCRAUmtXk1ost8s0HVczpp1Fk+vfZr/efR/hv1V1XwNHyV8RER2qDXVSiKbYEzJmEKH0mc10Rq+fujXWdy4mBteuaHQ4cggKOEjIqNaLOT3g1NMHTd33Tkh4hZ/wsd6Hq3/up/l7zmL5ltupfqCC6j+6EcKFk/dZz5DYHw9G77zXWwm06dlJpdPVh8+MiCZTZsIdOuU/GP7fIxP7PsJnt3wLB+47wPD+jbtJhrFhMP97vdKRGS02dTpd25fLE26uhw/5XhOnnoyv3n5NyzavIiM17fjIhlelPARkVGtq0lXW6p4avgks34fM+HA8LzTQ2/SGzey4btXsvrTn6Hj3/6tp73OTlZf9EnWXn45ABN/9UvGXvFFjDEFi9OJRhl7xZdILl3Kph/+qE/LTCybyNr2tWS97C6OTkYar70dp2xr5+uRQITLDryMO06/g7Ab5otPfHHY3hrXGINbVUW2qbnQoYiIDGtdCZ9iquHT5Svzv0JdSR0fuO8DvOdv78GzXqFDkn4KFDoAEZFC6qrhU0xNupLZJI5xCJjhvQtPvvUWDddfT7ZhC50LFmA9D7eiglWPPsrYL32J1n/+k/hLLzH2/75G1XnnYVy30CEDUH7ySXSefz6NN95IaOpUqs47d4fzTy6bTNpLs7FzI+Nj43dTlFLsbDaL19GBGyt722v1sXq+e+R3+fTDn+YrT32Fk6acxKyqWUyrmFaASLfPT/ioho+IyI4Uc8KnOlLNz4/7OV988oscOOZA4pn4iOtDcqQb3mcLIiK7WFcNn2Jr0hV2wwWtCbMzHc/8l7WXXYbNZglNnUrFe95DzYWfwK2sYuUHP8jGK6/ERKNM+MlPKD/5pEKH+zZjv3QFqdWr2PDtbxOcNJHYEUdsd96uO3WtbluthI/0mdfhdxTvlL094QNw9MSjuWjuRfz65V9z/4r7CZgAlx14GR+Z85Fh8913Kyt0ly4RkZ3Y2LkRKM6ED8DMqpncdcZdhQ5DBkgJHxEZ1fJ9+KSLJ+GTyCQIuaFCh9Gr+Msv0/inm2i97z5CU6Yw6frrCE2a1GOeqbfeQvuTTxKdO3e33omrP0wgwIQf/ZiVH/gAay+7nKm33kJ45sxe551c5ne6u6ptFfPr5+/OMKWIee3+Psft1qRrW5cecCnnzj6XzfHN3LDoBn608Ec8sPIBvnroV5lTM2d3hbpdgaoqEq8tLnQYIiLD2qbOTVSFq4btsZuMbOrDR0RGtXyTriLqwyflpQg7w6//ns4FC1jx/g/Q/sQTVJ17LlNvv/1tyR7w+8kpP+mkYZvs6eLGSpl03a8wkQjLzzqbDd/6Fp0LFtD64IO0PfIoNuv32TO2dCwhJ8Tq1tUFjliKSbbNT/g4se0nfMC/IjynZg4/esePuOLgK9jQsYFP3P+JYdGhs1upJl0iIjuzqXNT0dbukeKnGj4iMqoFnSARN1JUTbqS2eSwvErUdMutuLEYMx56EHc7zVSKTXD8eCb/9jc0/OIXNN16G01/viX/Ws0lFzPmsstwjMOEsgmsblPCR/rOa/eTzE4vffj0xhjD+Xufz3GTj+O8e8/jK09+hZtOvQnXKVzfV25VFdnWVmwmgwnokFJEpDdK+EghqYaPiIx6sVCsqJp0dfXhM5zYTIa2Rx6h7JRTRkyyp0tkzz2ZeO21zHjgASb+6pdMveMOovvvT8tdd+dr+Uwum8yqNt2aXfou2+YnfHbUpKs342Pj+fL8L/PKllf485I/74rQ+sytqgJryba2FjQOEZHhbGPnRiV8pGCU8BGRUS8WjBVVk65kNjnsbsmeWr0aG48T3X//Qoeyy4QmTqDs2GOJ7jOHqg+dT2bjRuIvvQT4HTevbluNtbbAUUqx8Np33Gnzjpwy9RSOnng0175wLWvb1w51aH3mVlUCqFmXiMh2pLIpGhONjC0dW+hQZJRSwkdERr3yUHlR1fBJZpPDroZP6q23AAjPnFHgSHaP2FFHgevS/tjjgJ/wiWfibElsKXBkUiy2NunqXw0f8Jt3fW3+1zAYvv2fbxcs0RioqgKU8BER2Z7N8c0AjC1RwkcKQwkfERn1YqFYUfXhk8qmhl0fPsm3lgEQmja9wJHsHm55OSUHHkj7k08CMLk8d6euVjXrkr7JN+kaQMIHoD5Wz2cP/CxPr3uafyz/x1CG1mduLuGTUcJHRKRXmzo3AcV7S3Ypfkr4iMioFwvGaE0VTx8UiUxi2NXwSb71JoH6etxYaaFD2W1KDp1PcskSss3NTCrz70amjpulr7LNzZhQCBONDngd580+j7m1c/n+s9+nKbH7ky6uaviIiOzQxs6NgBI+UjhK+IjIqFcWKiuqJl3DsdPm1FvLCE8fHbV7upTOnw/W0rlwIeNj43GNy7KWZYUOS4pEtqGBQG0txpgBr8N1XL5x+DdoSjZxz1v3DGF0fdx+V8Jni5oyioj0ZmOHn/BRky4pFCV8RGTUiwWLq0nXcLstu/U8ksuWjZr+e7pE5s7FhMN0PvssQSfIQWMP4l/L/0XWyxY6NCkCmYYtuLW1g17PHlV7MDE2kRc2vTAEUfWPEw7j1tSQXrdut29bRKQYbOrcRNgNUx4qL3QoMkop4SMio14sFCORTZD20oUOpU+GWw2f9Lr12Hic0PTRlfBxQiGiBxxAx3+fBeDc2eeyrmMd9y2/r8CRSTHI5Gr4DIUDxhzAC5teKEjnzcGJE0itWbPbtysiUgxWtq5kQmzCoGpzigyGEj4iMuqVhfzbIhdLLZ+klyTkDJ8aPolXXwUgPGtWgSPZ/UrnH0Jy6VKyzc2cMOUE5tTM4afP/5TOdGehQ5NhbigTPkdMOILGRCOPr3l8SNbXH6EJE0mvVQ0fEZFtWWt5afNLzK2bW+hQZBRTwkdERr1Y0L9LTrEkfFLZFJFApNBh5HU89SROWRnRffcpdCi7Xckhh/j9+CxYgGMcrjjkCjZ1buLq567eaW2LrJfFs95uilSGE5vJkG1sHLKEz0lTT2JibCK/eulXu72WT3DCBNLr12OzasooItLd8tblNCeb2a9uv0KHIqOYEj4iMurFQn7Cpy3dVuBIds5aO+z68Gl/+mlKDzsMEwwWOpTdLrLvvphIJN+s64AxB3Dhvhdy1xt38d3/fne7NX0a4g188qFP8rtXfrc7w5VhItPYCNbi1tYMyfqCTpAL517Ia1te45/L/zkk6+zztidPgnSa9Nq1u3W7IiLD2br2dXztqa8RdsMcPv7wQocjo1ig0AGIiBRaWbB4mnRlvAye9YZNk67Mli1k1q0n+qEPFzqUgnBCIUoOPIDOZ5/NT7v0gEuJZ+LctPgmHlr5EMdOPpYDxxzI9MrppLNpnljzBH9e8mcyXoZ3TX9XAaOXQsmsXw9AcFz94FbU2QipDqicxBkzzuDON+7km//5JvWxeg4Yc8AQRLp9nvVwjENkr70BSLz2GqHJk3fpNkVEioG1lq89/TWWNC7hW0d8i/Gx8YUOSUYx1fARkVGvqw+fYqjh05nxa4yUBEsKHIkv8dprAET23rvAkRROySF+Pz6ZpiYAjDFcccgV3HjKjRw49kDuW3YfX3nqK5x373l86J8f4oZXbuCQcYdw5xl3cubMMwsbvBREarXfyXFo0sSBr+Qf/wvfnwbX7AeNywg4AX56zE8ZUzKGSx66hBc3vTg0wW6jPdXOx+7/GMfdfhwvbnqR8B6zMMEgiVde2SXbExEpNm80v8FzG57jfw76H13YkYJTDR8RGfUqw5UANCYaCxtIH7Sl/KRUV79DhRZ/+WUAInvvVeBICqfkkPkAdD7zDOXvfGd++kFjD+KgsQeRyqZY3baaZS3LcI3L/mP2pzpSXahwZRhIr1kNQHDiABM+L/4ZnvsN7PkuWHIvLP0XHPYp6krquOGkG/jY/R/j4/d/nC8c/AXOnX3ukN4d5k+L/8RzG54D4LeLfsvPj/854dmzib+8aMi2ISJSzBZt9veHR088usCRiKiGj4gItSW1GAybOjcVOpSd6kh3AFtrJe1u8VdepenWW1l21lls+tGPabr1VqIHHIBbVph4hoPo3H1xKypof+yxXl8PuSFmVM7gxCknctzk45TsEVJr1uDW1uJEo/1b0Fp45jr466dg2jvg7Bugdja8+VB+lrGlY/nTqX9ifv18vvvf73LVs1fl9xv9lfWy3LfsPk6961RuX3o77al2/vTanzh20rGcO/tcnt3wLOlsmpJ584i/+CJeMjmg7YiIjCSvbnmVsmAZk8omFToUEdXwEREJOkFqo7Vs7NhY6FB2qquGT2mwdLduN9vewaarr6b5L38BwIRCbHltMQCVl1++W2MZbkwgQOyYd9D22ON4qRROaHj0ryTDV3rVakL9rd2TTcOdH4fX/gZ7nALn/B6CEZh0CCy9z08G5WryVEeq+fnxP+cHz/2AmxbfxOOrH+emU2+irqSuz5t7cs2T/P7V3+dr81z97NX8c/k/aUu18cn9PsmGjg3ctvQ2FjUsYtah82m88UbiL7xA6aGH9u99iYiMMK80vMLetXvjGNWtkMJTKRQRAcaWjGVj5/BP+OzuGj7WWtoefpjlZ55J8x13UP2xjzHtnr8x+6UXmXLLn5l0/XVUnHXWbollOCs//Qy8lhbaH3640KFIEUguW0Zo+vS+zdyxBZ78MfzsAD/Zc8L/g/ffCqFcP15j50DnFmjvWUPRMU6+L6mmZBMf/ddHuW/ZfTu9bXtnupOvPfU1PvXwp3ir+S0uP/By/n7m36mJ1vD8pue57MDLmFMzJ3+b4UUNiyg9+GCc0lKa77iz35+FiMhIkswmeaP5Dfap2afQoYgAquEjIgL4zSBWtKwodBg71dWx9O7ow8fr6GDNpZfS8e//EJoyhSk3/YmSgw7Kv15ywK69C1AxKT3sUIKTJtHw699QdtJJGNctdEgyTGWamsg2NBCeMWPnMy/5B9z1SUi1wbSj4bQfwR4n95xn7Bx/uOlVKBv7tlUcNPYgrj/xer7+9Ne54skruOb5a9i3bl9OmXoK+4/Zn9pobX7eDR0buPCBC1nZupKL5l7ExXMvJugGAfjHWf8gkUnkk8210VrGlY7j1YZXceZ8hMr3vY/GP/6R6o98hOi+OtERkdHp9cbXyXgZ5tTOKXQoIoASPiIigF/D55n1z2CtHdIOToda163jY6Fdm/DpfP551n3py6TXrGHs/32NqnPPxQT0k7E9xnUZ8z+Xs/Zzn2fzz65lzP9cXuiQZJhKvfUWAOFZM3c846pn4C8X+Amdd/8Cxm7nTnhjc8mVdS/CjON6neWAMQfwtzP/xn3L7+ORVY/w7PpnuX/F/QBMLZ/Kh/b+EDMqZ/D5xz5PMpvktyf9lkPGHQwrn/ZrDoVKCU48mGBJdY+mY3Nr57Jw40I861F7ycW03ncf6778JabdeSdOONzvz0ZEpNi9ssW/Y6Fq+MhwoaN3ERFgWsU0OtIdrO9Yz/jY+EKHs13t6VzCZwhr+NhMhuSbb5JasYLOZ58juWwZnf/9L8Hx45n8+99TOv+QIdvWSFb2zndS+Z9n2HL99TglJdR84uM7rOmTfPNNTDBIaMqU3RilFFrnCy8AEN5jD3j6Z4CF8gn+3bbWvQBV06Butn8nroqJcP6dULKDjr5LqqF6BqxduMPtOsbhXdPfxbumv4u0l+blzS/zSsMr3L/ifr79zLcBGF86nutPvJ7ZWQN/OB1WPLl1BcaFsnHQtsGP9z3XccykY3hg5QMsaljEfnX7Uf+db7P6wotY/5WvMv4H38c46jlAREaXFza+wJiSMYwrHVfoUESAIUz4ePE4LX/7G8nX36D0yCMoO673q0wiIsPRPrX+lZhXGl4Z3gmfVDsBEyDs9v/qubUWr6MDJxLJ19Zp+cc/2PyTn5Jes8afyRic8nJqP/Upqi+4ADe2ezuHLmbGGMb939fItrWx+Sc/oeXv91B+6qmEZ84kUFcHmQyZzZtJvvkmnS+8QOd/n6Xs5JOY+JOfFDp0GSLWWrLNzTihENZabCKBzXpkm5sI1NSQbWmh6aabiR50EMGV98CD/7d14WiV32yr4Q1Y9hhMPRLec92Okz1dJs7zl+lW+2ZHgk6Qg8YexEFjD+LDe3+YZzc8y6KGRZw68VjGP/YDePk2CJXBqT+EKUdAvAmWPQoNr0PFJL+p2R0XcMwnHyfshrnrjbvYr24/YkcdRd3//A+bf/ITgpMmMmaUd+guIqOLtZaFmxZy4JgDh3VtcRldhiTh0/L3e9n882tJr1yFCYVo+vOfqf3UJdReeqkKu4gUhT2q9iDgBFjUsIiTpp5U6HC2qz3dTiwU2+G+1UsmSa9ZQ3rDBryWFv/EM5Vmy/XXk1qxAoJBQpMnYxyH5BtvEN57L+q/+13cqipiRx0Jrqsr8wNkQiEm/OTHtJ5wAo1/+iMNP7v27TM5DuFZs6j5+Mep/tgFuz/IUczr6KDjmWeIHXfcoI5PvGSSbFMTqRUrSL7xJsk33vAfb76J19a2w2VNNMr4Oa3wzy/4d9s6Lpf0qZnp33UL/Dty5frO6ZMZx/tJmlfuhH3P6dd7McYwv34+8+vn+/0FvXwbHPZpOOIyiI3ZOuPUI7aO73MW/OY4yp78Ce+Z+R7ueOMOzt/rfGZVzaLmogtJr1nNluuuBwt1l1+mY0ERGRVeb3qdTZ2bOHjcwYUORSTP7OxuDTuz37hx9tbKKtyqKib86IeUHHII67/5TVruuJPyU99J9Uc/SnTu3CEKV0Rk17n4oYtZ2riU+8++n5A7PG+tfcUTV/Dy5pe576z7SLz6Gu2PPEJm8ybSmzaRXrMWr7OTzIYN/pX+bYSmT6fiPWfitbaRXL4MMlmi+82l5qKLhq6TYWth02JY/gQ0LoNUB1RPgymH+32NRMqHZjtFItveTmrFSrJNjZhAALeqitC0aerfpEAar72Sjb/4E8HacqJzZhOcOhW3oganogaMIduwAa+tBZtMYJMJvGQKr72NbHMz2dZ2su0dZNuT2FSmx3rdiCFckSFcniRYVwbhKiipxERLMYEwTiRAtrkFY1LEom8QZDMc9zU47DPgDsG1Ny8LvznOrx1UMx2S7WA9iFRAtBIilVBS4z93g+AEIRD2p5XUQGkttK6Fv3wUjvwfOOGbO9/mP6+A/17HlmO/zFkb/klpsJRfn/hrJpZNxGYybPh//4/mv9xB+WmnMe4bX8ctH13ffRldGuINlIfKh+2xg+we333mu9z5xp088t5HqIxUFjocGSGMMQuttfMGvPxgEz77T51qH/ziF6m96KJ8EwHreTT88lds+c1vsMkk0YMOovK951B+4ok4pWoeIIXlJZNkNjeQ2bSJzMYNZFtacCsqcCsqcMrKcSvKMZEITjiMiUQwoZCuTu5mmYbNrLv8EjJbmglPmUB4z70I7zWX4NSZhCZNxCkp6fc6vUTCP2lrasoPE4sXE1/4HMEJE6n73OdZyEoufOBC3jHxHVw490L2rtmboNOPq+y7mGc9TrnzFI7aUsv5D6ZJvPIKOA6Bmho/kTBlCk5JlOCkyYSmTCZYX49bUeEvm0gQ2XvvXXf3qNb18Pwf4MWboXmVPy1cDqEYtK3bOl9srN9HSdlYv8lIshU6G/1mKMEoBCI9h8Gof3KaSUAmCcbZOt0N+Set+WHYH4bLtp7khkrBCWw9yXWD4AyTO2gl2yDV2eudlWTXsIvuouXaL9O+LEm8MUgm7oLdZv9uLI5rMY7FuOAEPdyQhxv2cEPWH48Y3FiUUIVDqDZEYPwkTO0sv8y3rIamldCyxv8bpzv8chsug3AF1Mzwkz0TB3zs1ruWtfDA1yAdh3DM32a8GRLN/rBzCyRawEtvfx3VM+DCh/3mZV2fmbW0JjJsbkuyuS1JPJ2hpjTMnmPChO+5BF69m5fmfZBLWl8gkU1y7KRjOWriURw05iAif/4HDT+7Fre6mqoPvJ+yE04gPHOmahDuYsP95gMj0ecf+zzPb3qes2adxf51+zOzcia1JbXD6hhCdq3lLcs5+56zOW36aXz7iG8XOpwhYT0Pr62NbEsLicVLSK1YQe0nLyp0WKNOwRM+c+fsZf/y2x/2+pqNJ+DJZ7EPPw2btvgH9HU1MH4MVJRBSRRTEoWAC4EABAPguv58PR70Mq2vr3d7bUf68jnsdJYhWEef/h47mWeQf9M+r2O4fB7ZLLR1QGs7Nh6HdCb3SG8dT6WguQ1aWqEjvvN1bquinL3++9/+LycDkmptYO3J83HDHqnWAOnOnlfAjYv/3SZXhHoUE5Ob0PWdt/5Tr7ctWaI1aRLNQQg4BOZN58Vah8e95SQdD881xAKlVARKqQjEiDoRQiZAwAQIOAGMMdiufYsxgMEa659oQe41Px7TFYfFv/LuAXhY6/lNrqzFWg88i8VunY4/DWtpzrTRvGkNH3jcw40FiR42DnfuZLzyKv8dd30Y1ssNbX5o8LZ+WDYfCMbzMF4ax0vhZJM4XgqTTeWep7pNT/d4DeOQDlWQDpYTSm4hktiMxbCpdj7Lxp3CyvKD2RIchzFQmm2hvuUlqjtXUBFfSVnnGqKpLQSynWQCpSRDlRjAzSYJeAncbBI3m8D1kriZBI5Nk3XCeG4YrIebTeLYHZy07oTFYJ0AnhPEmgDW8R+eCebHrQlgjcHmC5vF2Kz/WdosWA9j/UKVitQQj9YTL5lAMjrGXy+O//e3FifrvycnG8fNJgik2yhtX0n1loU07XEutef9YsDvRfrnvkXr+fbfX2U8DUwyG6i2zVRlWqlMtWCA5mgNncEYWSeIZwJknBCeEyIeKCfhVpAIlmHdCK4bwHUNQcfgOg5B1+A6hqDrEHBM7jUH1zGEHA/HCWAcQybrkc5aMp4/jKeydKayJDNZoOeuzPMsqYxHMuPlhtn8eMazOAYcx+AYQ01piH0mVDC+Mko44JArev633FqsBS83tABeBieTJJRqIpJuJpJqAi/DstgBLG8zNLSnaO5MsaU9xeb2JKnM23egkaDDwZMruCx5PfMa/so6N8gNtfU8EHVoNv78JQSYvznG6Y/GmfhWBwCZSIB4TSnpWIRMeZRsSRgbDmJDwa1DxwHHAcc/frO5cWty0w2Aye93c/9tfZ7PJ219fXts91l7jmC79ts9FrDbTLLbrMhicscutsfLtmcY3Y5vehyLb++4x9rcPr5bbL0sE+hIUvPYi1R891vscfgpva9Lhtx/LjmPzjeW8kZlks4wJEKQCBqcUIiIGyEUDOO6QQK5h+MGMI6DcQM4rovBwTUOjvGHXeWwK3Fn8uW8+/St4375d/JlrOdy3Uue2Wa0ly9H17Lb/d70MZnYbQXbX6IPr+SPWcD/BTdbj6d6fBvt1uOgbgPT47vSbST3vTHbrCb//bUelm7HZIDFy+1Le05vzrTzcmI5WetxQdUJlNkgpNLYTBY8D7IeeB7Gdfz9U9ffzHq5UPI7bP+YL+vljuW6PkeT3x+CAc/z153N+vOmc9vpmt/k5gd/v7m9v3XX4XIqDZ0JbEccOhPQGYeOTogne+6TohFm//vfONFoL38x2VUKnvCZM7HMvvqJHV+psRbim0N0bAyTbA2Qag2QSTp4KQfr6QqEDB3/iqz/cHJD44DjWgLRLIGoRyCS9ccjHoGSLG7YI5srj9m0IZtysFmDzRo8z2CzkDBhJv1lWaHf3qjRHk/ykW/9Aouh1CSYmNrE1I6N1Hc2UN7RQTiVImA8TFciBXLjW3/YjfFf83DwrCEdDJAJB0iHAiTCITqCUZqiMdYExzCpYyPzX3oVd0P27Vf7h6FgTZrM0SGCQY9y00E5cbp+tbt/KjY/vu1zf9zLfWIpAiRtkBRBkgRJESBlc0O2DpO263mQAFkqTDsVdNBoy1lhx3Kvdxir7K6ordI9gedzyRIiTYgMIdKETW5ImhhxKkwHFXQQNUkCZAmQJUiWABmCJpMb75qe8Yema1oGF4uLh8HDxfPLEQ5ZnNxUJ/f5wRjTzAQaGGcaCZrsDt9J0gboIMIaW8cz3t5kZp/Opz70/l3wmUlvFqxo5LbnVudyqJastXi5ZIjnWX9oyY9nPEvW84eZrEfWs6Sz/rS05z/PZC3p/Gu5YW65rNf7MZbrGAKOoSTkUhIKEA44+SLeVdIdYwgHHUKuQzjgEgo4hAIO4YCfVMrHbS3rmhMs2dBKIt1rZrvPjIFx5RHGlIWpKg1RXRKirjxMXSxMXZk/jIZcNrYmeGZZI8+taGTZ5nbmZl7lMPdVZpk1jDGNtIfaWRlJsSZseTMYZHkoiNtm2Gs1zFpnqW2Big5LZQeUJiGc6pajkUFbVwUN5x/PuZ/+eaFDGTUWffhwyldsINkeIJMx2IzB0TmO7GpOrjaq4x/35ln/OK9bnqx3Xb87Tq72asjihHK1Wrs/D3qEyjM4VRD5dsOufEfSi4InfPads5e9+fofDHwFmYz/6KqJkc1dmfa2Xpmm64DJ2rc9TI/Xva3Z0d4eO63e2ocd805nGYJ19KEart3ZSoaiKm9f1rE7Po+dzeA42LISKCuD0K6pOuu4QfY9bPh25DvSeJ5lVWNn/uTL5obZfO0Xfz7/gsjWq9ZdV7C32uY6bG8XebpNN+k4gbWv4yQ6MOkEpFOYTCq/r/FPELNkvBSpbIp8taFu+yHTtR/C23qJHbpdiTa5K9TGvypnHIzj+lesHBfHcXBwwPi1A3BcjDE4xvVrjESiJGYcjA1vvS276b6JHlO2Ttv2ql9vy5leluvt+dYrjD1fc40hEnQJBxzCuaG1kLX+iW/X37HrJLVrV9/13ObH6THvtn8n2Hp1u7cL492vfG/7E7fTdeTn29okoquiaG/vvcfn4Hm4iUYcMvlaVQaDDUSxwQjWjWByzclM7mpbeSTAmPIIMjJ5np9UymT9fVfA8ZM1jjP0J4JZz9KezJBMZ/0Uaa42gH+h13Sr8GzyFaAds/V1gIBjCLj9S71Ya2nqTNMaT5PpSpTlkmKZTBKTaCGbSkI2RTrTRjzdSsam8bwMWTJkbRbPy2JTKWw8gUklMZ4FL+sPbbdxr1vtxa6HH0X+uNF25Ye7v5az9Xyoa2ex9Tu+zZvaWsth69JbX+5RYeLtOwOz7RrzO5Ht71hNt+fWdJu2bSpsJ9u2wQDhyXty4D7HUl1Zte07k13kzddeoKN5s1/7EwDr1zRPpTBe1u9nK1fLI//Ien6TmWzK/520Wf/7gJervGLztfS6/WLla+3ljoD8qfl57NblYGstEej+VehaU26k93NB08v07Z037vBsst/L2K3XebapnWTpdj5nus6H/B2a7fryd6/NYrrV78l/v7at5dT9iT/B69ovGv+YzK9F5eCQO27LvwYYlzK3lJDTrW++UBCCIbxArvWK64Jxtu7DstavuYiTr4mDk0vQuI7/cBz/Ep3N7ftyf3M8m5vH7XahYNu/Zc/h2/Z928xvtjM9/8l0zecY5h52MrJ7FTzhM2/ePLtgwYJBrUNERERERERERLYabMJHNWhFREREREREREYYJXxEREREREREREYYJXxEREREREREREYYJXxEREREREREREYYJXxEREREREREREYYJXxEREREREREREYYJXxEREREREREREYYJXxEREREREREREYYY60d3AqM2Qx0AA1DEpHI8FSLyriMfCrnMhqonMtooHIuo4HKuYx0tUCptbZuoCsYdMIHwBizwFo7b9ArEhmmVMZlNFA5l9FA5VxGA5VzGQ1UzmWkG4oyriZdIiIiIiIiIiIjjBI+IiIiIiIiIiIjzFAlfH49ROsRGa5UxmU0UDmX0UDlXEYDlXMZDVTOZaQbdBkfkj58RERERERERERk+FCTLhERERERERGREWanCR9jzCRjzKPGmNeMMa8aYy7LTX9v7rlnjJm3zTJfNsa8aYxZaow5eVcFLzJU+lvOjTFTjTFxY8yLucd1hYtepG92UM5/YIxZYox52RhztzGmstsy2p9L0ehvGde+XIrRDsr5t3Nl/EVjzAPGmPG56cYY87PcvvxlY8yBhX0HIjs3gHJ+jDGmpdv+/OuFfQciO7e9ct7t9c8bY6wxpjb3vN/785026TLG1AP11trnjTFlwELgTMACHnA98L/W2gW5+fcGbgEOAcYDDwF7WGuz/Xv7IrvPAMr5VOBea+0+hYlYpP92UM4nAo9YazPGmKsBrLVXaH8uxWYAZXwq2pdLkdlBOV9jrW3NzfNZYG9r7cXGmFOBS4FTgfnANdba+YWJXqRvBlDOj8E/Vn9XgUIW6bftlXNr7WvGmEnAb4E9gYOstQ0D2Z/vtIaPtXa9tfb53HgbsBiYYK1dbK1d2ssi7wZutdYmrbXLgTfxTxZEhq0BlHORorODcv6AtTaTm+0Z/JNj0P5ciswAyrhI0dlBOW/tNlsp/kUr8Pflf7S+Z4DK3EmGyLA1gHIuUnS2V85zL/8E+CI9y3i/9+f96sMndyXsAOC/O5htArC62/M13YIWGfb6WM4BphljXjDGPG6MOWrXRyYydHZQzj8G/DM3rv25FK0+lnHQvlyK2Lbl3BjzXWPMauCDQFeTFu3Lpaj1sZwDHGaMeckY809jzJzdH6nIwHUv58aYdwNrrbUvbTNbv/fnfU74GGNiwJ3A5dtkVkVGjH6U8/XAZGvtAcDngD8bY8p3R4wig7W9cm6M+SqQAW4uVGwiQ6EfZVz7cilavZVza+1XrbWT8Mv4ZwoZn8hQ6Ec5fx6YYq3dD7gW+GsBwhUZkO7lHP845Sv0TGYOWJ8SPsaYYC6Am621d+1k9rXApG7PJ+amiQxr/SnnuSYuW3LjC4G3gD12fZQig7O9cm6M+SjwLuCDdmvnbtqfS9HpTxnXvlyKVR+OWW4Gzs6Na18uRak/5dxa22qtbc+N3wcEuzq6FRnOeinnM4BpwEvGmBX4++znjTHjGMD+vC936TLADcBia+2P+xDzPcB5xpiwMWYaMAt4tg/LiRRMf8u5MabOGOPmxqfjl/NluzZKkcHZXjk3xpyC30b4DGttZ7dFtD+XotLfMq59uRSjHZTzWd1mezewJDd+D/Dh3N1dDgVarLXrd1vAIgPQ33JujBmXWwZjzCH457lbdl/EIv3XWzm31i6y1o6x1k611k7Fb7Z1oLV2AwPYnwf6EMcRwIeARcaYF3PTvgKE8avL1QH/MMa8aK092Vr7qjHmduA1/OpIn9YdXaQI9KucA0cD3zLGpPHv4nWxtbZx94ct0i/bK+c/wy/rD+aOlZ6x1l6s/bkUoX6VcbQvl+K0vXL+cWPMbPyyvBK4OPfaffh3dHkT6AQu2K3RigxMf8v5OcAlxpgMEAfO61ZjWWS46rWc52qp9abf+/Od3pZdRERERERERESKS7/u0iUiIiIiIiIiIsOfEj4iIiIiIiIiIiOMEj4iIiIiIiIiIiOMEj4iIiIiIiIiIiOMEj4iIiIiIiIiIiOMEj4iIiJStIwxNcaYF3OPDcaYtbnxdmPMLwsdn4iIiEih6LbsIiIiMiIYY74JtFtrf1joWEREREQKTTV8REREZMQxxhxjjLk3N/5NY8wfjDFPGmNWGmPOMsZ83xizyBjzL2NMMDffQcaYx40xC40x9xtj6gv7LkREREQGTgkfERERGQ1mAMcBZwA3AY9aa/cF4sBpuaTPtcA51tqDgN8B3y1UsCIiIiKDFSh0ACIiIiK7wT+ttWljzCLABf6Vm74ImArMBvYBHjTGkJtnfQHiFBERERkSSviIiIjIaJAEsNZ6xpi03dqJoYd/PGSAV621hxUqQBEREZGhpCZdIiIiIrAUqDPGHAZgjAkaY+YUOCYRERGRAVPCR0REREY9a20KOAe42hjzEvAicHhBgxIREREZBN2WXURERERERERkhFENHxERERERERGREUYJHxERERERERGREUYJHxERERERERGREUYJHxERERERERGREUYJHxERERERERGREUYJHxERERERERGREUYJHxERERERERGREUYJHxERERERERGREeb/A0KZQ28YdpupAAAAAElFTkSuQmCC",
"text/plain": [
"<pyannote.core.feature.SlidingWindowFeature at 0x150451ad1670>"
]
},
"execution_count": 23,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"spk_probability = Inference(pretrained, step=2.5)(test_file)\n",
"spk_probability"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"A perfect output would look like that:"
]
},
{
"cell_type": "code",
"execution_count": 22,
"metadata": {},
"outputs": [
{
"data": {
"image/png": "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",
"text/plain": [
"<pyannote.core.feature.SlidingWindowFeature at 0x150451793a30>"
]
},
"execution_count": 22,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"test_file[\"annotation\"].discretize(notebook.crop, resolution=0.010)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"We are going to fine-tune this pretrained model on the AMI dataset:"
]
},
{
"cell_type": "code",
"execution_count": 24,
"metadata": {},
"outputs": [],
"source": [
"from pyannote.audio.tasks import Segmentation\n",
"seg_task = Segmentation(ami, duration=5.0, max_num_speakers=4)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"To check that fine-tuning was actually helpful, we need to evaluate the performance of the pretrained model, and compute the average local diarization error rate on a 5s window sliding over the whole test set. To do so, we need to create a helper function:"
]
},
{
"cell_type": "code",
"execution_count": 15,
"metadata": {},
"outputs": [],
"source": [
"def test(model, protocol, subset=\"test\"):\n",
" from pyannote.audio.utils.signal import binarize\n",
" from pyannote.audio.utils.metric import DiscreteDiarizationErrorRate\n",
" from pyannote.audio.pipelines.utils import get_devices\n",
"\n",
" (device,) = get_devices(needs=1)\n",
" metric = DiscreteDiarizationErrorRate()\n",
" files = list(getattr(protocol, subset)())\n",
"\n",
" inference = Inference(model, device=device)\n",
"\n",
" for file in files:\n",
" reference = file[\"annotation\"]\n",
" hypothesis = binarize(inference(file))\n",
" uem = file[\"annotated\"]\n",
" _ = metric(reference, hypothesis, uem=uem)\n",
" \n",
" return abs(metric)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"We can then evaluate the model and see its local DER:"
]
},
{
"cell_type": "code",
"execution_count": 16,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Local DER (pretrained) = 18.3%\n"
]
}
],
"source": [
"der_pretrained = test(model=pretrained, protocol=ami, subset=\"test\")\n",
"print(f\"Local DER (pretrained) = {der_pretrained * 100:.1f}%\")"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Next, we prepare the model for fine-tuning, simply by overriding its `task` attribute..."
]
},
{
"cell_type": "code",
"execution_count": 29,
"metadata": {},
"outputs": [],
"source": [
"from copy import deepcopy\n",
"finetuned = deepcopy(pretrained)\n",
"finetuned.task = seg_task"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"... and we train it (for just one epoch)"
]
},
{
"cell_type": "code",
"execution_count": 30,
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"GPU available: True, used: True\n",
"TPU available: False, using: 0 TPU cores\n",
"IPU available: False, using: 0 IPUs\n",
"LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n",
"\n",
" | Name | Type | Params | In sizes | Out sizes \n",
"-------------------------------------------------------------------------------------------------------------------\n",
"0 | sincnet | SincNet | 42.6 K | [32, 1, 80000] | [32, 60, 293] \n",
"1 | lstm | LSTM | 1.4 M | [32, 293, 60] | [[32, 293, 256], [[8, 32, 128], [8, 32, 128]]]\n",
"2 | linear | ModuleList | 49.4 K | ? | ? \n",
"3 | classifier | Linear | 516 | [32, 293, 128] | [32, 293, 4] \n",
"4 | activation | Sigmoid | 0 | [32, 293, 4] | [32, 293, 4] \n",
"5 | validation_metric | AUROC | 0 | ? | ? \n",
"-------------------------------------------------------------------------------------------------------------------\n",
"1.5 M Trainable params\n",
"0 Non-trainable params\n",
"1.5 M Total params\n",
"5.892 Total estimated model params size (MB)\n"
]
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"HBox(children=(HTML(value='Validation sanity check'), FloatProgress(value=1.0, bar_style='info', layout=Layout…"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "634877439d6744d88223cfd1313e5fc1",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"HBox(children=(HTML(value='Training'), FloatProgress(value=1.0, bar_style='info', layout=Layout(flex='2'), max…"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"HBox(children=(HTML(value='Validating'), FloatProgress(value=1.0, bar_style='info', layout=Layout(flex='2'), m…"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\n"
]
}
],
"source": [
"trainer = pl.Trainer(gpus=1, max_epochs=1)\n",
"trainer.fit(finetuned)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"We now evaluate the performance of the fine-tuned model..."
]
},
{
"cell_type": "code",
"execution_count": 19,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Local DER (finetuned) = 17.5%\n"
]
}
],
"source": [
"der_finetuned = test(model=finetuned, protocol=ami, subset=\"test\")\n",
"print(f\"Local DER (finetuned) = {der_finetuned * 100:.1f}%\")"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"... to confirm that it actually improved its performance on the AMI test set.\n",
"\n",
"\n",
"## Transfer learning\n",
"\n",
"What if you are only interested in detecting overlapped speech regions? \n",
"\n",
"Looking at the output of the `pyannote/segmentation` model, it seems that it would be a good starting point for training such a dedicated model:"
]
},
{
"cell_type": "code",
"execution_count": 37,
"metadata": {},
"outputs": [
{
"data": {
"image/png": "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",
"text/plain": [
"<pyannote.core.feature.SlidingWindowFeature at 0x150421dcef70>"
]
},
"execution_count": 37,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"Inference('pyannote/segmentation', step=2.5)(test_file)"
]
},
{
"cell_type": "code",
"execution_count": 38,
"metadata": {},
"outputs": [
{
"data": {
"image/png": "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",
"text/plain": [
"<pyannote.core.annotation.Annotation at 0x150493522fd0>"
]
},
"execution_count": 38,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"test_file[\"annotation\"]"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Let's update the model so that it specifically addresses overlapped speech detection. \n",
"This is achieved very simply by updating the `task` attribute..."
]
},
{
"cell_type": "code",
"execution_count": 39,
"metadata": {},
"outputs": [],
"source": [
"from pyannote.audio.tasks import OverlappedSpeechDetection\n",
"osd_task = OverlappedSpeechDetection(ami, duration=2.0)\n",
"\n",
"osd_model = Model.from_pretrained(\"pyannote/segmentation\")\n",
"osd_model.task = osd_task"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"... optionally freeezing a bunch of layers..."
]
},
{
"cell_type": "code",
"execution_count": 40,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"['sincnet',\n",
" 'sincnet.wav_norm1d',\n",
" 'sincnet.conv1d',\n",
" 'sincnet.conv1d.0',\n",
" 'sincnet.conv1d.0.filterbank',\n",
" 'sincnet.conv1d.1',\n",
" 'sincnet.conv1d.2',\n",
" 'sincnet.pool1d',\n",
" 'sincnet.pool1d.0',\n",
" 'sincnet.pool1d.1',\n",
" 'sincnet.pool1d.2',\n",
" 'sincnet.norm1d',\n",
" 'sincnet.norm1d.0',\n",
" 'sincnet.norm1d.1',\n",
" 'sincnet.norm1d.2',\n",
" 'lstm']"
]
},
"execution_count": 40,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"osd_model.freeze_up_to('lstm')"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"... and training it:"
]
},
{
"cell_type": "code",
"execution_count": 41,
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"GPU available: True, used: True\n",
"TPU available: False, using: 0 TPU cores\n",
"IPU available: False, using: 0 IPUs\n",
"LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n",
"\n",
" | Name | Type | Params | In sizes | Out sizes \n",
"-------------------------------------------------------------------------------------------------------------------\n",
"0 | sincnet | SincNet | 42.6 K | [32, 1, 32000] | [32, 60, 115] \n",
"1 | lstm | LSTM | 1.4 M | [32, 115, 60] | [[32, 115, 256], [[8, 32, 128], [8, 32, 128]]]\n",
"2 | linear | ModuleList | 49.4 K | ? | ? \n",
"3 | classifier | Linear | 129 | [32, 115, 128] | [32, 115, 1] \n",
"4 | activation | Sigmoid | 0 | [32, 115, 1] | [32, 115, 1] \n",
"5 | validation_metric | AUROC | 0 | ? | ? \n",
"-------------------------------------------------------------------------------------------------------------------\n",
"49.5 K Trainable params\n",
"1.4 M Non-trainable params\n",
"1.5 M Total params\n",
"5.890 Total estimated model params size (MB)\n"
]
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"HBox(children=(HTML(value='Validation sanity check'), FloatProgress(value=1.0, bar_style='info', layout=Layout…"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "17ca908b800f4d34841bc2c7d5ccdff3",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"HBox(children=(HTML(value='Training'), FloatProgress(value=1.0, bar_style='info', layout=Layout(flex='2'), max…"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"HBox(children=(HTML(value='Validating'), FloatProgress(value=1.0, bar_style='info', layout=Layout(flex='2'), m…"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\n"
]
}
],
"source": [
"trainer = pl.Trainer(gpus=1, max_epochs=1)\n",
"trainer.fit(osd_model)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Et voilà! A brand new overlapped speech detection model!"
]
},
{
"cell_type": "code",
"execution_count": 46,
"metadata": {},
"outputs": [
{
"data": {
"image/png": "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",
"text/plain": [
"<pyannote.core.feature.SlidingWindowFeature at 0x1504177e8370>"
]
},
"execution_count": 46,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"from pyannote.audio.utils.signal import binarize\n",
"binarize(Inference(osd_model)(test_file))"
]
},
{
"cell_type": "code",
"execution_count": 44,
"metadata": {},
"outputs": [
{
"data": {
"image/png": "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",
"text/plain": [
"<pyannote.core.annotation.Annotation at 0x150493522fd0>"
]
},
"execution_count": 44,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"test_file[\"annotation\"]"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Going further\n",
"\n",
"This tutorial only scratched the surface of the training API.\n",
"\n",
"Every task supports an `augmentation` parameter for training with data augmentation via [`torch-audiomentations`](https://github.com/asteroid-team/torch-audiomentations) library:\n",
"\n",
"```python\n",
"from torch_audiomentations import AddBackgroundNoise\n",
"augmentation = AddBackgroundNoise(\"/path/to/background/noise/directory\")\n",
"vad_task = VoiceActivityDetection(ami, augmentation=augmentation)\n",
"```\n",
"\n",
"We also benefit from all the nice things [`pytorch-lightning`](ttps://pytorch-lightning.readthedocs.io) has to offer (like multi-gpu training, for instance).\n",
"\n",
"```python\n",
"trainer = Trainer(gpus=4, strategy='ddp')\n",
"trainer.fit(model)\n",
"```\n",
"\n",
"Default optimizer (`Adam` with default parameters) is automatically set up for you. \n",
"Customizing optimizer (and scheduler) requires overriding [`model.configure_optimizers`](https://pytorch-lightning.readthedocs.io/en/stable/api/pytorch_lightning.core.lightning.html#pytorch_lightning.core.lightning.LightningModule.configure_optimizers) method:\n",
"\n",
"```python\n",
"from types import MethodType\n",
"from torch.optim import SGD\n",
"from torch.optim.lr_scheduler import ExponentialLR\n",
"def configure_optimizers(self):\n",
" return {\"optimizer\": SGD(self.parameters()),\n",
" \"lr_scheduler\": ExponentialLR(optimizer, 0.9)}\n",
"model.configure_optimizers = MethodType(configure_optimizers, model)\n",
"trainer.fit(model)\n",
"```"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.8.0"
}
},
"nbformat": 4,
"nbformat_minor": 4
}