ResNet-tests.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "import torch.nn as nn\n",
    "\n",
    "from tqdm import tqdm\n",
    "from functools import partial\n",
    "from dataclasses import dataclass\n",
    "from collections import OrderedDict\n",
    "import pickle\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "def getCNNdata(epochs, dtyp, epoch_start=0, epoch_end=1000):\n",
    "    data = {\"train\": [], \"val\": [], \"test\": []}\n",
    "\n",
    "    for epoch in range(epoch_start, epoch_end):\n",
    "        for phase in epochs[epoch].keys():\n",
    "            data[phase] = data[phase] + [epochs[epoch][phase][dtyp]]\n",
    "            #aucs[phase] = aucs[phase] + [cnn_epochs[epoch][phase][\"auc\"]]\n",
    "    return data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Implementing ResNet in PyTorch\n",
    "Today we are going to implement the famous ResNet from Kaiming He et al. (Microsoft Research). It won the 1st place on the ILSVRC 2015 classification task.\n",
    "\n",
    "The original paper can be read from [here ](https://arxiv.org/abs/1512.03385) and it is very easy to follow, additional material can be found in this [quora answer](https://www.quora.com/)\n",
    "![alt](./images/custom/rotated-resnet34.png)\n",
    "\n",
    "*Deeper neural networks are more difficult to train.* Why? One big problem of deeper network is the vanishing gradient. Basically, the model is not able to learn anymore.\n",
    "\n",
    "To solve this problem, the Authors proposed to use a reference to the previous layer to compute the output at a given layer. In ResNet, the output form the previous layer, called **residual**, is added to the output of the current layer. The following picture visualizes this operation\n",
    "\n",
    "![alt](./images/residual.png)\n",
    "\n",
    "We are going to make our implementation **as scalable as possible** using one think think unknown to mostly of the data scientiest: **object orienting programming**"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Basic Block\n",
    "\n",
    "Okay, the first thing is to think about what we need. Well, first of all we need a convolution layer and since PyTorch does not have the 'auto' padding in Conv2d, so we have to code ourself!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Conv2dAuto(nn.Conv2d):\n",
    "    def __init__(self, *args, **kwargs):\n",
    "        super().__init__(*args, **kwargs)\n",
    "        self.padding =  (self.kernel_size[0] // 2, self.kernel_size[1] // 2) # dynamic add padding based on the kernel_size\n",
    "        \n",
    "conv3x3 = partial(Conv2dAuto, kernel_size=3, bias=False)      \n",
    "        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Conv2dAuto(32, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n"
     ]
    }
   ],
   "source": [
    "conv = conv3x3(in_channels=32, out_channels=64)\n",
    "print(conv)\n",
    "del conv"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Residual Block\n",
    "To make clean code is mandatory to think about the main building block of each application, or of the network in our case. The residual block takes an input with `in_channels`, applies some blocks of convolutional layers to reduce it to `out_channels` and sum it up to the original input. If their sizes mismatch, then the input goes into an `identity`. We can abstract this process and create a interface that can be extedend."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "class ResidualBlock(nn.Module):\n",
    "    def __init__(self, in_channels, out_channels):\n",
    "        super().__init__()\n",
    "        self.in_channels, self.out_channels =  in_channels, out_channels\n",
    "        self.blocks = nn.Identity()\n",
    "        self.shortcut = nn.Identity()   \n",
    "    \n",
    "    def forward(self, x):\n",
    "        residual = x\n",
    "        if self.should_apply_shortcut: residual = self.shortcut(x)\n",
    "        x = self.blocks(x)\n",
    "        x += residual\n",
    "        return x\n",
    "    \n",
    "    @property\n",
    "    def should_apply_shortcut(self):\n",
    "        return self.in_channels != self.out_channels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "ResidualBlock(\n",
       "  (blocks): Identity()\n",
       "  (shortcut): Identity()\n",
       ")"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ResidualBlock(32, 64)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's test it with a dummy vector with one one, we should get a vector with two"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[[[2.]]]])"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dummy = torch.ones((1, 1, 1, 1))\n",
    "\n",
    "block = ResidualBlock(1, 64)\n",
    "block(dummy)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In ResNet each block has a expansion parameter in order to increase the `out_channels`. Also, the identity is defined as a Convolution followed by an Activation layer, this is referred as `shortcut`. Then, we can just extend `ResidualBlock` and defined the `shortcut` function."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "from collections import OrderedDict\n",
    "\n",
    "class ResNetResidualBlock(ResidualBlock):\n",
    "    def __init__(self, in_channels, out_channels, expansion=1, downsampling=1, conv=conv3x3, *args, **kwargs):\n",
    "        super().__init__(in_channels, out_channels)\n",
    "        self.expansion, self.downsampling, self.conv = expansion, downsampling, conv\n",
    "        self.shortcut = nn.Sequential(OrderedDict(\n",
    "        {\n",
    "            'conv' : nn.Conv2d(self.in_channels, self.expanded_channels, kernel_size=1,\n",
    "                      stride=self.downsampling, bias=False),\n",
    "            'bn' : nn.BatchNorm2d(self.expanded_channels)\n",
    "            \n",
    "        })) if self.should_apply_shortcut else None\n",
    "        \n",
    "        \n",
    "    @property\n",
    "    def expanded_channels(self):\n",
    "        return self.out_channels * self.expansion\n",
    "    \n",
    "    @property\n",
    "    def should_apply_shortcut(self):\n",
    "        return self.in_channels != self.expanded_channels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "ResNetResidualBlock(\n",
       "  (blocks): Identity()\n",
       "  (shortcut): Sequential(\n",
       "    (conv): Conv2d(32, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "    (bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "  )\n",
       ")"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ResNetResidualBlock(32, 64)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Basic Block\n",
    "A basic ResNet block is composed by two layers of `3x3` convs/batchnorm/relu. In the picture, the lines represnet the residual operation. The dotted line means that the shortcut was applied to match the input and the output dimension.\n",
    "![alt](./images/custom/Block.png)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's first create an handy function to stack one conv and batchnorm layer. Using `OrderedDict` to properly name each sublayer."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "from collections import OrderedDict\n",
    "def conv_bn(in_channels, out_channels, conv, *args, **kwargs):\n",
    "    return nn.Sequential(OrderedDict({'conv': conv(in_channels, out_channels, *args, **kwargs), \n",
    "                          'bn': nn.BatchNorm2d(out_channels) }))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Sequential(\n",
       "  (conv): Conv2d(3, 3, kernel_size=(3, 3), stride=(1, 1))\n",
       "  (bn): BatchNorm2d(3, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       ")"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "conv_bn(3, 3, nn.Conv2d, kernel_size=3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "class ResNetBasicBlock(ResNetResidualBlock):\n",
    "    expansion = 1\n",
    "    def __init__(self, in_channels, out_channels, activation=nn.ReLU, *args, **kwargs):\n",
    "        super().__init__(in_channels, out_channels, *args, **kwargs)\n",
    "        self.blocks = nn.Sequential(\n",
    "            conv_bn(self.in_channels, self.out_channels, conv=self.conv, bias=False, stride=self.downsampling),\n",
    "            activation(),\n",
    "            conv_bn(self.out_channels, self.expanded_channels, conv=self.conv, bias=False),\n",
    "        )\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ResNetBasicBlock(\n",
      "  (blocks): Sequential(\n",
      "    (0): Sequential(\n",
      "      (conv): Conv2dAuto(32, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "    )\n",
      "    (1): ReLU()\n",
      "    (2): Sequential(\n",
      "      (conv): Conv2dAuto(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "    )\n",
      "  )\n",
      "  (shortcut): Sequential(\n",
      "    (conv): Conv2d(32, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "    (bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "  )\n",
      ")\n"
     ]
    }
   ],
   "source": [
    "dummy = torch.ones((1, 32, 224, 224))\n",
    "\n",
    "block = ResNetBasicBlock(32, 64)\n",
    "block(dummy).shape\n",
    "print(block)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### BottleNeck\n",
    "To increase the network deepths but to decrese the number of parameters, the Authors defined a BottleNeck block that \n",
    "\"The three layers are 1x1, 3x3, and 1x1 convolutions, where the 1×1 layers are responsible for reducing and then increasing (restoring) dimensions, leaving the 3×3 layer a bottleneck with smaller input/output dimensions.\" We can extend the `ResNetResidualBlock` and create these blocks."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "class ResNetBottleNeckBlock(ResNetResidualBlock):\n",
    "    expansion = 4\n",
    "    def __init__(self, in_channels, out_channels, activation=nn.ReLU, *args, **kwargs):\n",
    "        super().__init__(in_channels, out_channels, expansion=4, *args, **kwargs)\n",
    "        self.blocks = nn.Sequential(\n",
    "           conv_bn(self.in_channels, self.out_channels, self.conv, kernel_size=1),\n",
    "             activation(),\n",
    "             conv_bn(self.out_channels, self.out_channels, self.conv, kernel_size=3, stride=self.downsampling),\n",
    "             activation(),\n",
    "             conv_bn(self.out_channels, self.expanded_channels, self.conv, kernel_size=1),\n",
    "        )\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ResNetBottleNeckBlock(\n",
      "  (blocks): Sequential(\n",
      "    (0): Sequential(\n",
      "      (conv): Conv2dAuto(32, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "    )\n",
      "    (1): ReLU()\n",
      "    (2): Sequential(\n",
      "      (conv): Conv2dAuto(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "    )\n",
      "    (3): ReLU()\n",
      "    (4): Sequential(\n",
      "      (conv): Conv2dAuto(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "    )\n",
      "  )\n",
      "  (shortcut): Sequential(\n",
      "    (conv): Conv2d(32, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "    (bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "  )\n",
      ")\n"
     ]
    }
   ],
   "source": [
    "dummy = torch.ones((1, 32, 10, 10))\n",
    "\n",
    "block = ResNetBottleNeckBlock(32, 64)\n",
    "block(dummy).shape\n",
    "print(block)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Layer\n",
    "A ResNet's layer is composed by blocks stacked one after the other. \n",
    "\n",
    "![alt](./images/custom/Layer.png)\n",
    "\n",
    "We can easily defined it by just stuck `n` blocks one after the other, just remember that the first convolution block has a stide of two since \"We perform downsampling directly by convolutional layers that have a stride of 2\"."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "class ResNetLayer(nn.Module):\n",
    "    def __init__(self, in_channels, out_channels, block=ResNetBasicBlock, n=1, *args, **kwargs):\n",
    "        super().__init__()\n",
    "        # 'We perform downsampling directly by convolutional layers that have a stride of 2.'\n",
    "        downsampling = 2 if in_channels != out_channels else 1\n",
    "        \n",
    "        self.blocks = nn.Sequential(\n",
    "            block(in_channels , out_channels, *args, **kwargs, downsampling=downsampling),\n",
    "            *[block(out_channels * block.expansion, \n",
    "                    out_channels, downsampling=1, *args, **kwargs) for _ in range(n - 1)]\n",
    "        )\n",
    "\n",
    "    def forward(self, x):\n",
    "        x = self.blocks(x)\n",
    "        return x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "ResNetLayer(\n",
       "  (blocks): Sequential(\n",
       "    (0): ResNetBasicBlock(\n",
       "      (blocks): Sequential(\n",
       "        (0): Sequential(\n",
       "          (conv): Conv2dAuto(64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n",
       "          (bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        )\n",
       "        (1): ReLU()\n",
       "        (2): Sequential(\n",
       "          (conv): Conv2dAuto(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "          (bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        )\n",
       "      )\n",
       "      (shortcut): Sequential(\n",
       "        (conv): Conv2d(64, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
       "        (bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      )\n",
       "    )\n",
       "    (1): ResNetBasicBlock(\n",
       "      (blocks): Sequential(\n",
       "        (0): Sequential(\n",
       "          (conv): Conv2dAuto(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "          (bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        )\n",
       "        (1): ReLU()\n",
       "        (2): Sequential(\n",
       "          (conv): Conv2dAuto(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "          (bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        )\n",
       "      )\n",
       "      (shortcut): None\n",
       "    )\n",
       "    (2): ResNetBasicBlock(\n",
       "      (blocks): Sequential(\n",
       "        (0): Sequential(\n",
       "          (conv): Conv2dAuto(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "          (bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        )\n",
       "        (1): ReLU()\n",
       "        (2): Sequential(\n",
       "          (conv): Conv2dAuto(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "          (bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        )\n",
       "      )\n",
       "      (shortcut): None\n",
       "    )\n",
       "  )\n",
       ")"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dummy = torch.ones((1, 32, 48, 48))\n",
    "\n",
    "layer = ResNetLayer(64, 128, block=ResNetBasicBlock, n=3)\n",
    "# layer(dummy).shape\n",
    "layer"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Encoder\n",
    "Similarly, the encoder is composed by multiple layer at increasing features size.\n",
    "\n",
    "![alt](./images/custom/rotated-Encoder.png)\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "class ResNetEncoder(nn.Module):\n",
    "    \"\"\"\n",
    "    ResNet encoder composed by increasing different layers with increasing features.\n",
    "    \"\"\"\n",
    "    def __init__(self, in_channels=3, blocks_sizes=[64, 128, 256, 512], deepths=[2,2,2,2], \n",
    "                 activation=nn.ReLU, block=ResNetBasicBlock, *args,**kwargs):\n",
    "        super().__init__()\n",
    "        \n",
    "        self.blocks_sizes = blocks_sizes\n",
    "        \n",
    "        self.gate = nn.Sequential(\n",
    "            nn.Conv2d(in_channels, self.blocks_sizes[0], kernel_size=7, stride=2, padding=3, bias=False),\n",
    "            nn.BatchNorm2d(self.blocks_sizes[0]),\n",
    "            activation(),\n",
    "            nn.MaxPool2d(kernel_size=3, stride=2, padding=1)\n",
    "        )\n",
    "        \n",
    "        self.in_out_block_sizes = list(zip(blocks_sizes, blocks_sizes[1:]))\n",
    "        self.blocks = nn.ModuleList([ \n",
    "            ResNetLayer(blocks_sizes[0], blocks_sizes[0], n=deepths[0], activation=activation, \n",
    "                        block=block,  *args, **kwargs),\n",
    "            *[ResNetLayer(in_channels * block.expansion, \n",
    "                          out_channels, n=n, activation=activation, \n",
    "                          block=block, *args, **kwargs) \n",
    "              for (in_channels, out_channels), n in zip(self.in_out_block_sizes, deepths[1:])]       \n",
    "        ])\n",
    "        \n",
    "        \n",
    "    def forward(self, x):\n",
    "        x = self.gate(x)\n",
    "        for block in self.blocks:\n",
    "            x = block(x)\n",
    "        return x"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Decoder\n",
    "The decoder is the last piece we need to create the full network. It is a fully connected layer that maps the features learned by the network to their respective classes. Easily, we can defined it as:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "class ResnetDecoder(nn.Module):\n",
    "    \"\"\"\n",
    "    This class represents the tail of ResNet. It performs a global pooling and maps the output to the\n",
    "    correct class by using a fully connected layer.\n",
    "    \"\"\"\n",
    "    def __init__(self, in_features, n_classes):\n",
    "        super().__init__()\n",
    "        self.avg = nn.AdaptiveAvgPool2d((1, 1))\n",
    "        self.decoder = nn.Linear(in_features, n_classes)\n",
    "\n",
    "    def forward(self, x):\n",
    "        x = self.avg(x)\n",
    "        x = x.view(x.size(0), -1)\n",
    "        x = self.decoder(x)\n",
    "        return x\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## ResNet\n",
    "\n",
    "Final, we can put all the pieces together and create the final model.\n",
    "\n",
    "![alt](./images/custom/rotated-resnet34.png)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "class ResNet(nn.Module):\n",
    "    \n",
    "    def __init__(self, in_channels, n_classes, *args, **kwargs):\n",
    "        super().__init__()\n",
    "        self.encoder = ResNetEncoder(in_channels, *args, **kwargs).to(device)\n",
    "        self.decoder = ResnetDecoder(self.encoder.blocks[-1].blocks[-1].expanded_channels, n_classes).to(device)\n",
    "        \n",
    "    def forward(self, x):\n",
    "        x = self.encoder(x)\n",
    "        x = self.decoder(x)\n",
    "        return x"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can now defined the five models proposed by the Authors, `resnet18,34,50,101,152`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "def resnet18(in_channels, n_classes):\n",
    "    return ResNet(in_channels, n_classes, block=ResNetBasicBlock, deepths=[2, 2, 2, 2])\n",
    "\n",
    "def resnet34(in_channels, n_classes):\n",
    "    return ResNet(in_channels, n_classes, block=ResNetBasicBlock, deepths=[3, 4, 6, 3])\n",
    "\n",
    "def resnet50(in_channels, n_classes):\n",
    "    return ResNet(in_channels, n_classes, block=ResNetBottleNeckBlock, deepths=[3, 4, 6, 3])\n",
    "\n",
    "def resnet101(in_channels, n_classes):\n",
    "    return ResNet(in_channels, n_classes, block=ResNetBottleNeckBlock, deepths=[3, 4, 23, 3])\n",
    "\n",
    "def resnet152(in_channels, n_classes):\n",
    "    return ResNet(in_channels, n_classes, block=ResNetBottleNeckBlock, deepths=[3, 8, 36, 3])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'device' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[0;32m/tmp/ipykernel_11989/4145029976.py\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[0;32mfrom\u001b[0m \u001b[0mtorchsummary\u001b[0m \u001b[0;32mimport\u001b[0m \u001b[0msummary\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      2\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 3\u001b[0;31m \u001b[0mmodel\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mresnet18\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m70\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m2\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      4\u001b[0m \u001b[0msummary\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mmodel\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcuda\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0;36m3\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m224\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m224\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/tmp/ipykernel_11989/172397865.py\u001b[0m in \u001b[0;36mresnet18\u001b[0;34m(in_channels, n_classes)\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[0;32mdef\u001b[0m \u001b[0mresnet18\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0min_channels\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mn_classes\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 2\u001b[0;31m     \u001b[0;32mreturn\u001b[0m \u001b[0mResNet\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0min_channels\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mn_classes\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mblock\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mResNetBasicBlock\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mdeepths\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m2\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m2\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m2\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m2\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      3\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      4\u001b[0m \u001b[0;32mdef\u001b[0m \u001b[0mresnet34\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0min_channels\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mn_classes\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      5\u001b[0m     \u001b[0;32mreturn\u001b[0m \u001b[0mResNet\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0min_channels\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mn_classes\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mblock\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mResNetBasicBlock\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mdeepths\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m3\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m4\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m6\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m3\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/tmp/ipykernel_11989/990910948.py\u001b[0m in \u001b[0;36m__init__\u001b[0;34m(self, in_channels, n_classes, *args, **kwargs)\u001b[0m\n\u001b[1;32m      3\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0m__init__\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0min_channels\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mn_classes\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m*\u001b[0m\u001b[0margs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m**\u001b[0m\u001b[0mkwargs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      4\u001b[0m         \u001b[0msuper\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m__init__\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 5\u001b[0;31m         \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mencoder\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mResNetEncoder\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0min_channels\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m*\u001b[0m\u001b[0margs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m**\u001b[0m\u001b[0mkwargs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mto\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdevice\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      6\u001b[0m         \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mdecoder\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mResnetDecoder\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mencoder\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mblocks\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m-\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mblocks\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m-\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mexpanded_channels\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mn_classes\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mto\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdevice\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      7\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mNameError\u001b[0m: name 'device' is not defined"
     ]
    }
   ],
   "source": [
    "from torchsummary import summary\n",
    "\n",
    "model = resnet18(70, 2)\n",
    "summary(model.cuda(), (3, 224, 224))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Downloading: \"https://download.pytorch.org/models/resnet152-394f9c45.pth\" to /home/hwixley/.cache/torch/hub/checkpoints/resnet152-394f9c45.pth\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "adedaf19d24e4d1eb04e3b83574957e7",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0.00/230M [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m/tmp/ipykernel_11989/79474960.py\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m      3\u001b[0m \u001b[0;31m# resnet101(False)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      4\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 5\u001b[0;31m \u001b[0msummary\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mmodels\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mresnet152\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;36m2\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcuda\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m20\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m75\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m~/miniconda3/envs/ai/lib/python3.9/site-packages/torchvision/models/resnet.py\u001b[0m in \u001b[0;36mresnet152\u001b[0;34m(pretrained, progress, **kwargs)\u001b[0m\n\u001b[1;32m    351\u001b[0m         \u001b[0mprogress\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0mbool\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m \u001b[0mIf\u001b[0m \u001b[0;32mTrue\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mdisplays\u001b[0m \u001b[0ma\u001b[0m \u001b[0mprogress\u001b[0m \u001b[0mbar\u001b[0m \u001b[0mof\u001b[0m \u001b[0mthe\u001b[0m \u001b[0mdownload\u001b[0m \u001b[0mto\u001b[0m \u001b[0mstderr\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    352\u001b[0m     \"\"\"\n\u001b[0;32m--> 353\u001b[0;31m     \u001b[0;32mreturn\u001b[0m \u001b[0m_resnet\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"resnet152\"\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mBottleneck\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0;36m3\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m8\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m36\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m3\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mpretrained\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mprogress\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m**\u001b[0m\u001b[0mkwargs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    354\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    355\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/miniconda3/envs/ai/lib/python3.9/site-packages/torchvision/models/resnet.py\u001b[0m in \u001b[0;36m_resnet\u001b[0;34m(arch, block, layers, pretrained, progress, **kwargs)\u001b[0m\n\u001b[1;32m    294\u001b[0m     \u001b[0mmodel\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mResNet\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mblock\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mlayers\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m**\u001b[0m\u001b[0mkwargs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    295\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0mpretrained\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 296\u001b[0;31m         \u001b[0mstate_dict\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mload_state_dict_from_url\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mmodel_urls\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0march\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mprogress\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mprogress\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    297\u001b[0m         \u001b[0mmodel\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mload_state_dict\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mstate_dict\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    298\u001b[0m     \u001b[0;32mreturn\u001b[0m \u001b[0mmodel\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/miniconda3/envs/ai/lib/python3.9/site-packages/torch/hub.py\u001b[0m in \u001b[0;36mload_state_dict_from_url\u001b[0;34m(url, model_dir, map_location, progress, check_hash, file_name)\u001b[0m\n\u001b[1;32m    584\u001b[0m             \u001b[0mr\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mHASH_REGEX\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0msearch\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mfilename\u001b[0m\u001b[0;34m)\u001b[0m  \u001b[0;31m# r is Optional[Match[str]]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    585\u001b[0m             \u001b[0mhash_prefix\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mr\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mgroup\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mif\u001b[0m \u001b[0mr\u001b[0m \u001b[0;32melse\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 586\u001b[0;31m         \u001b[0mdownload_url_to_file\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0murl\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mcached_file\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mhash_prefix\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mprogress\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mprogress\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    587\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    588\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0m_is_legacy_zip_format\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mcached_file\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/miniconda3/envs/ai/lib/python3.9/site-packages/torch/hub.py\u001b[0m in \u001b[0;36mdownload_url_to_file\u001b[0;34m(url, dst, hash_prefix, progress)\u001b[0m\n\u001b[1;32m    472\u001b[0m                   unit='B', unit_scale=True, unit_divisor=1024) as pbar:\n\u001b[1;32m    473\u001b[0m             \u001b[0;32mwhile\u001b[0m \u001b[0;32mTrue\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 474\u001b[0;31m                 \u001b[0mbuffer\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mu\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mread\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m8192\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    475\u001b[0m                 \u001b[0;32mif\u001b[0m \u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mbuffer\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;34m==\u001b[0m \u001b[0;36m0\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    476\u001b[0m                     \u001b[0;32mbreak\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/miniconda3/envs/ai/lib/python3.9/http/client.py\u001b[0m in \u001b[0;36mread\u001b[0;34m(self, amt)\u001b[0m\n\u001b[1;32m    460\u001b[0m             \u001b[0;31m# Amount is given, implement using readinto\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    461\u001b[0m             \u001b[0mb\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mbytearray\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mamt\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 462\u001b[0;31m             \u001b[0mn\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mreadinto\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mb\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    463\u001b[0m             \u001b[0;32mreturn\u001b[0m \u001b[0mmemoryview\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mb\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0mn\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtobytes\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    464\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/miniconda3/envs/ai/lib/python3.9/http/client.py\u001b[0m in \u001b[0;36mreadinto\u001b[0;34m(self, b)\u001b[0m\n\u001b[1;32m    504\u001b[0m         \u001b[0;31m# connection, and the user is reading more bytes than will be provided\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    505\u001b[0m         \u001b[0;31m# (for example, reading in 1k chunks)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 506\u001b[0;31m         \u001b[0mn\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mfp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mreadinto\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mb\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    507\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0mn\u001b[0m \u001b[0;32mand\u001b[0m \u001b[0mb\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    508\u001b[0m             \u001b[0;31m# Ideally, we would raise IncompleteRead if the content-length\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/miniconda3/envs/ai/lib/python3.9/socket.py\u001b[0m in \u001b[0;36mreadinto\u001b[0;34m(self, b)\u001b[0m\n\u001b[1;32m    702\u001b[0m         \u001b[0;32mwhile\u001b[0m \u001b[0;32mTrue\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    703\u001b[0m             \u001b[0;32mtry\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 704\u001b[0;31m                 \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_sock\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mrecv_into\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mb\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    705\u001b[0m             \u001b[0;32mexcept\u001b[0m \u001b[0mtimeout\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    706\u001b[0m                 \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_timeout_occurred\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;32mTrue\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/miniconda3/envs/ai/lib/python3.9/ssl.py\u001b[0m in \u001b[0;36mrecv_into\u001b[0;34m(self, buffer, nbytes, flags)\u001b[0m\n\u001b[1;32m   1239\u001b[0m                   \u001b[0;34m\"non-zero flags not allowed in calls to recv_into() on %s\"\u001b[0m \u001b[0;34m%\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1240\u001b[0m                   self.__class__)\n\u001b[0;32m-> 1241\u001b[0;31m             \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mread\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mnbytes\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mbuffer\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   1242\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1243\u001b[0m             \u001b[0;32mreturn\u001b[0m \u001b[0msuper\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mrecv_into\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mbuffer\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mnbytes\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mflags\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/miniconda3/envs/ai/lib/python3.9/ssl.py\u001b[0m in \u001b[0;36mread\u001b[0;34m(self, len, buffer)\u001b[0m\n\u001b[1;32m   1097\u001b[0m         \u001b[0;32mtry\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1098\u001b[0m             \u001b[0;32mif\u001b[0m \u001b[0mbuffer\u001b[0m \u001b[0;32mis\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1099\u001b[0;31m                 \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_sslobj\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mread\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mlen\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mbuffer\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   1100\u001b[0m             \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1101\u001b[0m                 \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_sslobj\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mread\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mlen\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "import torchvision.models as models\n",
    "\n",
    "# resnet101(False)\n",
    "\n",
    "summary(models.resnet152(1,2).cuda(), (1, 20, 75))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "58141634\n"
     ]
    }
   ],
   "source": [
    "count = 0\n",
    "for p in resnet152(1,2).parameters():\n",
    "    count += np.prod(p.shape)\n",
    "    \n",
    "print(count)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 87,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'resnet152_torchviz.pdf'"
      ]
     },
     "execution_count": 87,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from torchviz import make_dot\n",
    "\n",
    "model = resnet152(1,2)\n",
    "y = model(torch.zeros(512,1,20,75).to(device))\n",
    "\n",
    "make_dot(y, params=dict(list(model.named_parameters()))).render(\"resnet152_torchviz\", format=\"pdf\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import numpy as np\n",
    "from torch.utils.data import TensorDataset, DataLoader\n",
    "from tqdm import tqdm\n",
    "import torch.optim as optim\n",
    "from sklearn.metrics import roc_auc_score"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Device configuration\n",
    "device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "vers = \"16.4\"\n",
    "features = \"polar-channelled\"\n",
    "lag = \"0\"\n",
    "window_size = \"20\"\n",
    "model_save_dir = os.getcwd() + f\"/ml-models/db{vers}/{features}-lag{lag}-ws{window_size}/\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "def loadNpy(filename):\n",
    "    with open(os.getcwd() + \"/train-val-test/\"+ filename, \"rb\") as f: return np.load(f)\n",
    "\n",
    "db = f\"db{vers}/window-size-{window_size}/lag{lag}/{features}-features/\"\n",
    "xdtype = \"std-\"\n",
    "ydtype = \"exi-\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "DIR = os.getcwd() + \"/train-val-test/\" + db"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train, X_val, X_test = loadNpy(db + f\"{xdtype}X_train.npy\"), loadNpy(db + f\"{xdtype}X_val.npy\"), loadNpy(db + f\"{xdtype}X_test.npy\")\n",
    "y_train, y_val, y_test = loadNpy(db + f\"{ydtype}y_train.npy\"), loadNpy(db + f\"{ydtype}y_val.npy\"), loadNpy(db + f\"{ydtype}y_test.npy\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.2426140012845215\n"
     ]
    }
   ],
   "source": [
    "minAcc = np.mean(y_train)\n",
    "print(minAcc)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [],
   "source": [
    "cb = np.mean([0.200589, 0.242614])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.2216015\n",
      "0.7783985\n"
     ]
    }
   ],
   "source": [
    "print(cb)\n",
    "print(1-cb)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.2426140012845215\n"
     ]
    }
   ],
   "source": [
    "minAcc = np.mean(y_train)\n",
    "print(minAcc)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([0., 0., 0., ..., 0., 0., 1.])"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_train"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 251,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(array([], dtype=int64),)\n",
      "37674\n"
     ]
    }
   ],
   "source": [
    "print(np.where(np.logical_and(y_train != 1, y_train != 0)))\n",
    "print(len(y_train))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 252,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([], dtype=int64), array([], dtype=int64), array([], dtype=int64))"
      ]
     },
     "execution_count": 252,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.where(X_train == -np.inf)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "def npy_to_tensor(data, l=False):\n",
    "    if not l:\n",
    "        data = torch.from_numpy(data).float()\n",
    "    else:\n",
    "        data = torch.from_numpy(data).type(torch.LongTensor)\n",
    "    return data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train, X_val, X_test = npy_to_tensor(X_train), npy_to_tensor(X_val), npy_to_tensor(X_test)\n",
    "y_train, y_val, y_test = npy_to_tensor(y_train, 1), npy_to_tensor(y_val, 1), npy_to_tensor(y_test, 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "batch_size = 512"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_dataset = TensorDataset(X_train, y_train)\n",
    "\n",
    "val_dataset = TensorDataset(X_val, y_val)\n",
    "\n",
    "test_dataset = TensorDataset(X_test, y_test)\n",
    "\n",
    "# Data loader\n",
    "train_iterator = DataLoader(dataset=train_dataset,batch_size=batch_size,shuffle=False)\n",
    "\n",
    "val_iterator = DataLoader(dataset=val_dataset, batch_size=batch_size, shuffle=False)\n",
    "\n",
    "test_iterator = DataLoader(dataset=test_dataset, batch_size=batch_size,shuffle=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([37368, 20, 75])"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "loaders = {\n",
    "    'train': train_iterator,\n",
    "    'val': val_iterator,\n",
    "    'test': test_iterator\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "from libauc.losses import AUCMLoss, SOAPLoss, APLoss_SH\n",
    "from libauc.optimizers import PESG, SOAP_ADAM, SOAP_SGD\n",
    "from sklearn.metrics import roc_auc_score"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "imratio = minAcc\n",
    "SEED = 123\n",
    "BATCH_SIZE = 64\n",
    "lr =  1e-2\n",
    "weight_decay = 2e-4\n",
    "margin = 0.6\n",
    "beta = 0.99 # this refers to gamma for moving average in the paper\n",
    "posNum = 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<torch._C.Generator at 0x7f6450035b30>"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "rn = resnet18(1, 2)#.to(device)\n",
    "torch.manual_seed(SEED)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "AUCMLoss()"
      ]
     },
     "execution_count": 48,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "loss_func = AUCMLoss(imratio=imratio)\n",
    "loss_func"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "PESG (\n",
       "Parameter Group 0\n",
       "    a: tensor([0.], device='cuda:0', requires_grad=True)\n",
       "    alpha: tensor([0.], device='cuda:0', requires_grad=True)\n",
       "    b: tensor([0.], device='cuda:0', requires_grad=True)\n",
       "    clip_value: 1.0\n",
       "    gamma: 0.99\n",
       "    lr: 0.01\n",
       "    margin: 0.6\n",
       "    model_acc: [tensor([[[[0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          ...,\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          ...,\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          ...,\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          ...,\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          ...,\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          ...,\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "          [0., 0., 0.,  ..., 0., 0., 0.]]]], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "       device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "       device='cuda:0'), tensor([[[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]]], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "       device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "       device='cuda:0'), tensor([[[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]]], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "       device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "       device='cuda:0'), tensor([[[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]]], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "       device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "       device='cuda:0'), tensor([[[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]]], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "       device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "       device='cuda:0'), tensor([[[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]]], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0.], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0.], device='cuda:0'), tensor([[[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]]], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0.], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0.], device='cuda:0'), tensor([[[[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]]],\n",
       "\n",
       "\n",
       "        [[[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]]],\n",
       "\n",
       "\n",
       "        [[[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]]],\n",
       "\n",
       "\n",
       "        [[[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]]],\n",
       "\n",
       "\n",
       "        [[[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]]]], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0.], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0.], device='cuda:0'), tensor([[[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]]], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0.], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0.], device='cuda:0'), tensor([[[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]]], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0.], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0.], device='cuda:0'), tensor([[[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]]], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "       device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "       device='cuda:0'), tensor([[[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]]], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "       device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "       device='cuda:0'), tensor([[[[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]]],\n",
       "\n",
       "\n",
       "        [[[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]]],\n",
       "\n",
       "\n",
       "        [[[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]]],\n",
       "\n",
       "\n",
       "        [[[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]]],\n",
       "\n",
       "\n",
       "        [[[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]]]], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "       device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "       device='cuda:0'), tensor([[[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]]], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "       device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "       device='cuda:0'), tensor([[[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]]], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "       device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "       device='cuda:0'), tensor([[[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]]], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0.], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0.], device='cuda:0'), tensor([[[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]]], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0.], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0.], device='cuda:0'), tensor([[[[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]]],\n",
       "\n",
       "\n",
       "        [[[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]]],\n",
       "\n",
       "\n",
       "        [[[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]]],\n",
       "\n",
       "\n",
       "        [[[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]]],\n",
       "\n",
       "\n",
       "        [[[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]],\n",
       "\n",
       "         [[0.]]]], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0.], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0.], device='cuda:0'), tensor([[[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]]], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0.], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0.], device='cuda:0'), tensor([[[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]],\n",
       "\n",
       "\n",
       "        [[[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]],\n",
       "\n",
       "         [[0., 0., 0.],\n",
       "          [0., 0., 0.],\n",
       "          [0., 0., 0.]]]], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0.], device='cuda:0'), tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0.], device='cuda:0'), tensor([[0., 0., 0.,  ..., 0., 0., 0.],\n",
       "        [0., 0., 0.,  ..., 0., 0., 0.]], device='cuda:0'), tensor([0., 0.], device='cuda:0'), tensor([0.], device='cuda:0'), tensor([0.], device='cuda:0')]\n",
       "    model_ref: [tensor([[[[ 3.3737e-03, -1.7778e-03, -3.0353e-03,  ...,  3.4861e-03,\n",
       "            6.6034e-03, -2.1964e-03],\n",
       "          [-3.7917e-03,  7.6711e-03, -1.1925e-02,  ..., -1.4097e-02,\n",
       "            1.7938e-03,  1.8951e-02],\n",
       "          [ 4.9545e-03,  2.6920e-03, -7.7020e-04,  ..., -1.6896e-03,\n",
       "            9.1776e-03,  1.5810e-02],\n",
       "          ...,\n",
       "          [-1.1589e-02,  3.2547e-03, -6.3151e-03,  ..., -1.3250e-02,\n",
       "            1.7843e-03, -2.1338e-02],\n",
       "          [ 1.0524e-02, -3.8848e-03, -9.3435e-03,  ..., -1.0867e-02,\n",
       "            8.8054e-03,  1.5542e-02],\n",
       "          [ 6.2662e-03, -1.7549e-03,  9.8284e-04,  ...,  2.6621e-03,\n",
       "           -5.8504e-03,  8.7684e-03]]],\n",
       "\n",
       "\n",
       "        [[[ 1.6221e-02, -1.4779e-02,  1.1331e-02,  ...,  1.3139e-02,\n",
       "            1.0533e-02,  1.3881e-03],\n",
       "          [ 2.2473e-02, -8.0364e-03, -2.8084e-03,  ..., -6.5956e-03,\n",
       "           -7.9793e-03,  1.8383e-03],\n",
       "          [ 2.2935e-03,  5.1463e-03,  9.9376e-03,  ..., -1.0826e-02,\n",
       "           -4.4382e-04,  1.6236e-02],\n",
       "          ...,\n",
       "          [ 1.1674e-03,  4.4026e-03, -1.4465e-02,  ..., -5.4963e-03,\n",
       "            1.0042e-02,  8.2723e-03],\n",
       "          [-3.9481e-03,  4.8923e-03, -2.1681e-03,  ..., -1.6025e-02,\n",
       "           -1.0764e-02,  9.0315e-03],\n",
       "          [-7.2184e-03, -5.9508e-03, -7.1122e-03,  ..., -1.3729e-02,\n",
       "           -2.2150e-02, -1.3193e-02]]],\n",
       "\n",
       "\n",
       "        [[[-2.0915e-02,  9.6285e-03, -3.1861e-04,  ...,  7.6681e-03,\n",
       "            2.7468e-04,  1.9929e-02],\n",
       "          [ 1.3708e-02, -5.0087e-03, -2.7928e-03,  ...,  6.3745e-05,\n",
       "           -9.8955e-03,  7.0161e-03],\n",
       "          [-9.4053e-03, -4.6806e-03,  1.0322e-02,  ...,  4.9275e-03,\n",
       "           -1.4078e-04, -2.7466e-03],\n",
       "          ...,\n",
       "          [-1.4596e-02, -2.1595e-02, -2.5822e-03,  ..., -8.0156e-03,\n",
       "           -8.1830e-03, -1.1820e-02],\n",
       "          [-2.8774e-03, -6.0430e-03,  6.0024e-03,  ..., -5.9217e-03,\n",
       "           -2.5479e-03,  1.1517e-02],\n",
       "          [-1.7858e-04,  4.2640e-03, -7.6574e-03,  ..., -1.2743e-02,\n",
       "            4.5128e-03, -2.2801e-03]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[ 7.6472e-03, -1.3979e-02,  1.1708e-02,  ..., -1.3396e-03,\n",
       "           -1.7715e-03,  5.9816e-03],\n",
       "          [ 5.5806e-03,  1.2676e-03,  2.6668e-02,  ..., -4.7441e-03,\n",
       "           -1.3016e-03,  1.0919e-03],\n",
       "          [ 1.1326e-02,  7.0819e-04,  4.8289e-04,  ...,  5.6939e-03,\n",
       "            7.0444e-03, -2.0261e-02],\n",
       "          ...,\n",
       "          [-1.1485e-02,  9.9384e-03, -3.1150e-03,  ..., -2.1282e-02,\n",
       "            7.7486e-03, -3.4859e-04],\n",
       "          [-1.3344e-02, -1.8486e-02, -1.1954e-02,  ..., -1.3607e-02,\n",
       "           -9.8877e-03,  1.0558e-02],\n",
       "          [ 2.5752e-03,  4.6830e-03,  9.9545e-03,  ..., -1.4934e-02,\n",
       "            1.4341e-02,  8.5578e-03]]],\n",
       "\n",
       "\n",
       "        [[[-1.9900e-02,  2.9635e-03, -3.8722e-03,  ..., -1.9215e-02,\n",
       "           -2.1690e-03, -1.1294e-02],\n",
       "          [ 4.9332e-03, -1.2497e-02, -4.3750e-03,  ..., -1.0945e-02,\n",
       "           -1.5449e-02, -3.5938e-03],\n",
       "          [-9.3079e-03,  1.4853e-02, -4.8568e-03,  ...,  1.7496e-02,\n",
       "            1.9481e-02, -2.3231e-06],\n",
       "          ...,\n",
       "          [ 2.3828e-03,  5.4325e-03, -6.4696e-03,  ..., -1.1839e-02,\n",
       "           -3.1790e-03,  4.2792e-03],\n",
       "          [ 1.1632e-02, -8.3270e-03,  1.7867e-03,  ..., -7.5995e-03,\n",
       "           -5.7536e-03,  6.7334e-03],\n",
       "          [ 1.2297e-02, -1.9892e-02,  3.6926e-03,  ...,  1.1352e-02,\n",
       "            1.7259e-02, -2.5453e-03]]],\n",
       "\n",
       "\n",
       "        [[[-4.4020e-03,  9.1621e-04,  8.1683e-03,  ...,  4.1912e-03,\n",
       "            1.2332e-03, -2.4843e-03],\n",
       "          [-3.8722e-03,  1.4279e-02, -1.8631e-03,  ..., -1.2254e-03,\n",
       "            9.8276e-04,  3.7084e-03],\n",
       "          [ 5.8154e-03,  4.3642e-03,  5.0590e-03,  ..., -1.1715e-02,\n",
       "           -4.3574e-03,  5.3393e-03],\n",
       "          ...,\n",
       "          [-6.5733e-03,  6.4727e-04,  6.6174e-03,  ...,  2.0343e-03,\n",
       "            1.3662e-02,  3.1803e-03],\n",
       "          [ 9.2622e-03,  1.1980e-02,  7.1154e-03,  ..., -1.4828e-02,\n",
       "            6.9698e-03, -7.9586e-03],\n",
       "          [-1.9972e-02,  1.1755e-02,  8.5802e-03,  ...,  9.7661e-03,\n",
       "           -1.0339e-02,  1.1689e-03]]]], device='cuda:0'), tensor([ 3.1085e-03,  7.1937e-04,  8.6122e-04, -1.0235e-02,  1.4318e-02,\n",
       "         1.5588e-03, -5.7761e-03,  5.9770e-04, -3.1236e-03,  5.8120e-03,\n",
       "         6.2591e-03, -8.6754e-03, -2.1759e-02,  5.3580e-03, -1.7770e-02,\n",
       "        -1.3277e-02, -2.1009e-02,  4.6472e-03, -1.5506e-05,  3.7130e-03,\n",
       "         8.2847e-03,  2.0659e-03, -8.5065e-03,  8.7705e-03,  1.7318e-02,\n",
       "         9.2422e-03,  4.2329e-04,  2.2413e-02,  2.3908e-03,  7.8835e-03,\n",
       "        -1.1296e-02, -2.1887e-02,  1.7146e-02, -2.4474e-03, -4.5338e-03,\n",
       "        -1.3577e-02,  1.4822e-02,  3.6886e-03, -8.4001e-03, -9.9289e-03,\n",
       "        -1.1551e-02,  5.3720e-05, -2.5383e-03,  4.2191e-03, -1.6730e-02,\n",
       "         1.2703e-03,  8.2874e-03,  8.3155e-03,  6.6869e-03, -2.0692e-02,\n",
       "        -1.4182e-03,  1.3660e-02, -8.1996e-03, -5.6485e-03,  1.8631e-03,\n",
       "        -4.7857e-03, -6.2961e-03,  1.6569e-02, -1.1913e-02, -1.3204e-03,\n",
       "        -7.1513e-04,  1.4851e-02, -1.5676e-02,  6.6451e-03], device='cuda:0'), tensor([-0.0057,  0.0124, -0.0035,  0.0250, -0.0057, -0.0164,  0.0017, -0.0011,\n",
       "        -0.0002,  0.0081, -0.0016,  0.0094,  0.0056, -0.0190,  0.0064,  0.0038,\n",
       "        -0.0070, -0.0068, -0.0127,  0.0015,  0.0011,  0.0030,  0.0040,  0.0092,\n",
       "        -0.0189,  0.0087,  0.0008, -0.0042, -0.0016,  0.0041,  0.0088,  0.0181,\n",
       "        -0.0068,  0.0088,  0.0060,  0.0086,  0.0027, -0.0010,  0.0204,  0.0061,\n",
       "        -0.0164,  0.0187,  0.0056, -0.0153,  0.0073,  0.0126, -0.0161, -0.0075,\n",
       "        -0.0012,  0.0056,  0.0016,  0.0116, -0.0054,  0.0149, -0.0098, -0.0178,\n",
       "         0.0164, -0.0055,  0.0072,  0.0011, -0.0225,  0.0008, -0.0163, -0.0043],\n",
       "       device='cuda:0'), tensor([[[[ 0.0077,  0.0046,  0.0016],\n",
       "          [ 0.0263,  0.0065, -0.0067],\n",
       "          [ 0.0008,  0.0024,  0.0013]],\n",
       "\n",
       "         [[-0.0112, -0.0091, -0.0123],\n",
       "          [ 0.0021, -0.0045,  0.0013],\n",
       "          [-0.0081, -0.0033,  0.0024]],\n",
       "\n",
       "         [[-0.0033, -0.0160, -0.0061],\n",
       "          [ 0.0068, -0.0152, -0.0182],\n",
       "          [-0.0033,  0.0045,  0.0069]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-0.0256, -0.0096,  0.0159],\n",
       "          [-0.0033, -0.0058, -0.0275],\n",
       "          [-0.0067,  0.0009, -0.0025]],\n",
       "\n",
       "         [[ 0.0048, -0.0054, -0.0018],\n",
       "          [-0.0205,  0.0072,  0.0057],\n",
       "          [ 0.0223,  0.0058, -0.0068]],\n",
       "\n",
       "         [[-0.0016,  0.0040, -0.0058],\n",
       "          [ 0.0120,  0.0083,  0.0062],\n",
       "          [ 0.0018,  0.0019,  0.0087]]],\n",
       "\n",
       "\n",
       "        [[[-0.0043, -0.0051, -0.0071],\n",
       "          [-0.0257, -0.0030, -0.0143],\n",
       "          [ 0.0059, -0.0116,  0.0009]],\n",
       "\n",
       "         [[-0.0049,  0.0067,  0.0143],\n",
       "          [ 0.0062, -0.0168, -0.0039],\n",
       "          [ 0.0098,  0.0091,  0.0070]],\n",
       "\n",
       "         [[ 0.0036,  0.0113,  0.0056],\n",
       "          [ 0.0065,  0.0070,  0.0193],\n",
       "          [ 0.0032, -0.0054,  0.0016]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-0.0159,  0.0074,  0.0161],\n",
       "          [ 0.0025,  0.0210, -0.0065],\n",
       "          [-0.0068,  0.0047,  0.0012]],\n",
       "\n",
       "         [[-0.0089,  0.0117,  0.0165],\n",
       "          [-0.0083, -0.0147, -0.0019],\n",
       "          [ 0.0144,  0.0151,  0.0057]],\n",
       "\n",
       "         [[-0.0003,  0.0305,  0.0146],\n",
       "          [-0.0181,  0.0014,  0.0067],\n",
       "          [-0.0012, -0.0022, -0.0043]]],\n",
       "\n",
       "\n",
       "        [[[ 0.0130, -0.0045, -0.0139],\n",
       "          [-0.0145,  0.0074,  0.0015],\n",
       "          [-0.0108, -0.0226,  0.0013]],\n",
       "\n",
       "         [[-0.0095,  0.0238,  0.0028],\n",
       "          [-0.0158,  0.0156,  0.0236],\n",
       "          [ 0.0099, -0.0011, -0.0061]],\n",
       "\n",
       "         [[-0.0062,  0.0106, -0.0078],\n",
       "          [-0.0157,  0.0012,  0.0006],\n",
       "          [-0.0023, -0.0184, -0.0082]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 0.0161,  0.0076, -0.0025],\n",
       "          [ 0.0167, -0.0044,  0.0116],\n",
       "          [-0.0089, -0.0081,  0.0004]],\n",
       "\n",
       "         [[-0.0074, -0.0031,  0.0122],\n",
       "          [ 0.0007,  0.0040, -0.0118],\n",
       "          [ 0.0098,  0.0016,  0.0032]],\n",
       "\n",
       "         [[-0.0076,  0.0146, -0.0133],\n",
       "          [-0.0148,  0.0030, -0.0149],\n",
       "          [-0.0091,  0.0065,  0.0081]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[ 0.0062,  0.0017, -0.0106],\n",
       "          [-0.0076,  0.0108,  0.0071],\n",
       "          [ 0.0328, -0.0076, -0.0023]],\n",
       "\n",
       "         [[-0.0063, -0.0130,  0.0088],\n",
       "          [-0.0023,  0.0005,  0.0032],\n",
       "          [-0.0030,  0.0008, -0.0044]],\n",
       "\n",
       "         [[ 0.0074, -0.0010,  0.0088],\n",
       "          [-0.0047,  0.0184, -0.0039],\n",
       "          [ 0.0094, -0.0017, -0.0028]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 0.0008, -0.0182,  0.0141],\n",
       "          [-0.0103, -0.0042, -0.0060],\n",
       "          [ 0.0185,  0.0169,  0.0075]],\n",
       "\n",
       "         [[ 0.0141, -0.0179, -0.0045],\n",
       "          [ 0.0094, -0.0059,  0.0131],\n",
       "          [-0.0024,  0.0059, -0.0022]],\n",
       "\n",
       "         [[ 0.0015,  0.0082,  0.0016],\n",
       "          [-0.0120, -0.0122,  0.0024],\n",
       "          [ 0.0098, -0.0046,  0.0087]]],\n",
       "\n",
       "\n",
       "        [[[-0.0053,  0.0028,  0.0050],\n",
       "          [ 0.0013, -0.0070,  0.0159],\n",
       "          [-0.0039, -0.0059,  0.0066]],\n",
       "\n",
       "         [[ 0.0030, -0.0033, -0.0039],\n",
       "          [-0.0089,  0.0239,  0.0090],\n",
       "          [ 0.0009,  0.0117, -0.0010]],\n",
       "\n",
       "         [[ 0.0024, -0.0131, -0.0096],\n",
       "          [-0.0181, -0.0098, -0.0014],\n",
       "          [-0.0130, -0.0063, -0.0090]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 0.0162, -0.0048,  0.0062],\n",
       "          [-0.0033, -0.0059, -0.0057],\n",
       "          [ 0.0085, -0.0195, -0.0243]],\n",
       "\n",
       "         [[-0.0317,  0.0015,  0.0035],\n",
       "          [-0.0154,  0.0098,  0.0013],\n",
       "          [ 0.0085, -0.0136, -0.0158]],\n",
       "\n",
       "         [[ 0.0078,  0.0132, -0.0076],\n",
       "          [-0.0001, -0.0088,  0.0026],\n",
       "          [-0.0027,  0.0003,  0.0016]]],\n",
       "\n",
       "\n",
       "        [[[-0.0167,  0.0096, -0.0132],\n",
       "          [-0.0029,  0.0050,  0.0094],\n",
       "          [ 0.0014, -0.0023, -0.0134]],\n",
       "\n",
       "         [[-0.0151, -0.0011,  0.0112],\n",
       "          [-0.0025, -0.0025,  0.0159],\n",
       "          [-0.0067,  0.0018, -0.0009]],\n",
       "\n",
       "         [[-0.0040, -0.0063,  0.0093],\n",
       "          [ 0.0025,  0.0076, -0.0085],\n",
       "          [-0.0075, -0.0041,  0.0107]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 0.0037,  0.0154, -0.0104],\n",
       "          [-0.0079,  0.0103, -0.0069],\n",
       "          [-0.0045, -0.0065,  0.0075]],\n",
       "\n",
       "         [[ 0.0062,  0.0122,  0.0003],\n",
       "          [ 0.0111,  0.0048,  0.0015],\n",
       "          [-0.0055, -0.0028,  0.0070]],\n",
       "\n",
       "         [[-0.0025, -0.0016, -0.0039],\n",
       "          [ 0.0054,  0.0211,  0.0218],\n",
       "          [ 0.0084,  0.0165, -0.0020]]]], device='cuda:0'), tensor([-1.2293e-02, -2.0499e-03, -4.3227e-03, -1.2614e-03, -1.4811e-02,\n",
       "         3.2586e-03,  1.5540e-03,  7.4437e-03,  1.1823e-02,  5.5155e-03,\n",
       "         2.1895e-03,  4.0799e-03, -1.6788e-03,  2.1149e-02, -1.2540e-02,\n",
       "        -4.7169e-03, -4.0685e-03,  1.0179e-03, -9.3295e-03, -9.8133e-03,\n",
       "        -2.4779e-03, -1.3873e-03,  2.3118e-03, -7.7937e-05,  5.7585e-03,\n",
       "        -9.6065e-03,  3.7104e-04,  8.9726e-03,  1.1210e-03, -4.4098e-03,\n",
       "        -8.8620e-03, -1.9110e-02, -2.5284e-03, -1.0314e-02,  1.8142e-02,\n",
       "        -1.0381e-02, -3.2956e-03,  6.7021e-03,  1.7981e-02, -4.8179e-03,\n",
       "         7.5269e-03, -2.1155e-03, -1.2131e-04, -1.9096e-02, -8.2774e-04,\n",
       "        -1.8608e-02,  8.7881e-03,  2.1397e-03,  7.9738e-03, -6.6625e-03,\n",
       "         1.8103e-02,  1.2500e-02,  2.7033e-02, -1.9152e-03,  1.9053e-02,\n",
       "        -3.9653e-03, -9.4644e-03,  3.6529e-03, -1.3495e-02, -1.5897e-03,\n",
       "        -5.4029e-03, -1.5227e-02, -8.4187e-06, -1.3208e-02], device='cuda:0'), tensor([-0.0087, -0.0213, -0.0059, -0.0044, -0.0110, -0.0156, -0.0142,  0.0019,\n",
       "         0.0082, -0.0063, -0.0056,  0.0227,  0.0050,  0.0022, -0.0073, -0.0191,\n",
       "        -0.0046, -0.0248,  0.0058,  0.0057,  0.0143,  0.0080,  0.0012, -0.0027,\n",
       "        -0.0188,  0.0109, -0.0059, -0.0046,  0.0044,  0.0009, -0.0011,  0.0067,\n",
       "        -0.0005, -0.0072, -0.0002, -0.0106,  0.0077, -0.0026, -0.0141,  0.0016,\n",
       "         0.0059, -0.0092, -0.0195, -0.0034, -0.0029, -0.0078,  0.0069,  0.0066,\n",
       "         0.0029, -0.0007,  0.0019, -0.0081,  0.0059,  0.0022, -0.0112,  0.0081,\n",
       "         0.0070,  0.0011, -0.0072,  0.0035,  0.0055, -0.0302,  0.0120, -0.0095],\n",
       "       device='cuda:0'), tensor([[[[ 1.4798e-02, -1.7001e-04, -3.3223e-03],\n",
       "          [-6.3156e-03,  7.9668e-03,  1.2552e-02],\n",
       "          [-1.4792e-02, -6.2006e-03,  7.1026e-03]],\n",
       "\n",
       "         [[ 6.5028e-03, -1.1904e-02, -1.0551e-02],\n",
       "          [ 9.9240e-04, -3.4410e-02, -1.9050e-04],\n",
       "          [ 7.9436e-03, -1.6306e-02,  1.9413e-02]],\n",
       "\n",
       "         [[-7.7880e-03,  9.1691e-03,  1.3746e-02],\n",
       "          [-1.2612e-02, -4.4192e-03, -1.3267e-02],\n",
       "          [-5.4559e-03, -1.5074e-03, -4.3264e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-4.7373e-04, -2.3006e-02,  8.1104e-03],\n",
       "          [-1.6727e-02, -4.4459e-03, -9.3219e-03],\n",
       "          [ 1.1610e-02, -1.3504e-02, -1.0587e-03]],\n",
       "\n",
       "         [[-1.2514e-02, -1.1922e-02, -5.6045e-03],\n",
       "          [ 4.2082e-03,  2.5615e-03,  7.1037e-03],\n",
       "          [ 7.2576e-03,  8.3409e-03,  6.5655e-03]],\n",
       "\n",
       "         [[-6.6635e-03,  1.7628e-02,  8.8650e-04],\n",
       "          [ 1.2057e-02,  2.0783e-04, -6.2277e-03],\n",
       "          [-5.7826e-03,  1.3312e-03, -1.0251e-02]]],\n",
       "\n",
       "\n",
       "        [[[ 1.1539e-02,  3.9657e-03, -6.5841e-04],\n",
       "          [ 3.0940e-03,  1.1844e-03,  1.7546e-02],\n",
       "          [ 4.7088e-04,  9.7530e-04,  6.2809e-03]],\n",
       "\n",
       "         [[ 1.3498e-03,  4.1453e-03, -2.7869e-03],\n",
       "          [-3.6123e-03, -1.6281e-02,  2.6315e-03],\n",
       "          [ 8.3438e-03,  1.1304e-04,  9.6177e-03]],\n",
       "\n",
       "         [[ 1.9679e-03,  7.0541e-03, -2.0786e-03],\n",
       "          [-4.9596e-03,  7.8762e-04, -5.4482e-04],\n",
       "          [-4.8906e-03,  8.3783e-03, -2.5913e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 1.7953e-02, -3.0867e-03, -1.5157e-02],\n",
       "          [ 7.5094e-03, -3.1444e-04,  5.3588e-03],\n",
       "          [-1.9660e-02,  8.3413e-03,  7.6387e-03]],\n",
       "\n",
       "         [[-1.6849e-02,  1.5037e-02,  8.9101e-03],\n",
       "          [ 1.3159e-02, -1.0347e-02, -3.1073e-03],\n",
       "          [-3.7357e-03, -9.2414e-03, -2.5166e-03]],\n",
       "\n",
       "         [[-4.6225e-04, -1.0029e-02,  3.2652e-03],\n",
       "          [-7.5310e-03, -1.7105e-02,  3.0244e-03],\n",
       "          [-2.1480e-02,  1.6711e-03,  2.2441e-02]]],\n",
       "\n",
       "\n",
       "        [[[ 3.2080e-03,  1.2284e-02, -6.6921e-03],\n",
       "          [-4.7445e-03, -2.1123e-02,  2.4132e-02],\n",
       "          [ 1.7485e-02,  8.5367e-03,  1.2546e-02]],\n",
       "\n",
       "         [[-1.7538e-02,  1.7485e-02,  1.6229e-04],\n",
       "          [-1.3322e-02,  5.3201e-03,  6.7211e-03],\n",
       "          [-1.4116e-02, -2.0547e-03,  8.5008e-03]],\n",
       "\n",
       "         [[ 6.3600e-03, -1.2874e-03,  5.8354e-03],\n",
       "          [-7.9353e-03,  3.1158e-03, -4.8515e-03],\n",
       "          [ 6.2019e-03,  1.3574e-02, -1.0692e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-1.7462e-02, -5.4624e-03,  3.1784e-03],\n",
       "          [ 1.5328e-03, -1.2207e-02, -5.1438e-03],\n",
       "          [-1.0822e-02,  9.1120e-03,  1.6435e-03]],\n",
       "\n",
       "         [[ 6.4783e-03,  1.7022e-02, -3.1721e-03],\n",
       "          [-9.8947e-03,  8.4223e-04,  8.9735e-03],\n",
       "          [ 1.7671e-03,  1.4196e-02, -1.1796e-02]],\n",
       "\n",
       "         [[-2.8363e-03,  6.7727e-03, -8.0028e-03],\n",
       "          [-1.0188e-02,  1.0375e-02, -1.3132e-03],\n",
       "          [-1.5907e-02, -1.8239e-02, -7.9856e-03]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[ 3.6128e-03,  1.1085e-02,  3.3822e-03],\n",
       "          [ 5.1706e-03,  6.2765e-03,  1.3503e-02],\n",
       "          [ 9.4816e-03,  1.6451e-02,  7.0624e-03]],\n",
       "\n",
       "         [[ 2.5091e-03,  1.2945e-02,  1.8205e-03],\n",
       "          [-1.0575e-02, -6.4160e-03, -1.5844e-02],\n",
       "          [-2.5359e-03, -4.0965e-03,  1.5708e-03]],\n",
       "\n",
       "         [[ 1.4268e-02, -1.2550e-03, -9.3914e-03],\n",
       "          [-8.7042e-03, -1.4991e-02,  9.2410e-03],\n",
       "          [-2.3316e-02, -1.6218e-03,  2.2077e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 9.1040e-04,  1.7738e-02, -7.4504e-03],\n",
       "          [-9.2579e-04, -5.1411e-03,  1.7815e-02],\n",
       "          [ 8.8369e-03, -2.9241e-03,  1.2524e-03]],\n",
       "\n",
       "         [[-2.3229e-04, -9.2118e-03, -1.4517e-02],\n",
       "          [ 4.2910e-03,  2.8299e-03, -1.0780e-03],\n",
       "          [ 2.2770e-04,  7.8059e-03,  1.0691e-02]],\n",
       "\n",
       "         [[ 5.3945e-03,  1.8873e-02,  1.1991e-04],\n",
       "          [ 8.4767e-03, -1.6305e-02,  3.5733e-03],\n",
       "          [-1.4259e-02,  1.3637e-02,  5.7048e-03]]],\n",
       "\n",
       "\n",
       "        [[[-1.2936e-02,  5.8586e-03, -3.8017e-03],\n",
       "          [-9.5718e-03, -1.7432e-02,  6.1267e-03],\n",
       "          [ 1.4650e-02,  8.5596e-03,  4.9897e-03]],\n",
       "\n",
       "         [[ 2.8846e-02,  1.4591e-02,  8.7470e-03],\n",
       "          [-1.1951e-02,  7.6965e-03,  1.2092e-02],\n",
       "          [ 8.7397e-03, -3.3905e-04,  1.1133e-02]],\n",
       "\n",
       "         [[-9.1566e-03, -1.2469e-03, -2.6689e-03],\n",
       "          [ 1.2924e-02,  3.5219e-03, -1.8869e-02],\n",
       "          [ 5.5672e-04,  1.8345e-02, -2.9269e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 1.8035e-03,  2.1409e-03,  1.0374e-02],\n",
       "          [-6.5351e-03, -1.8581e-03,  2.3259e-03],\n",
       "          [ 4.5986e-03,  4.2338e-03, -1.6634e-02]],\n",
       "\n",
       "         [[-9.4712e-03, -1.3237e-03, -1.7652e-02],\n",
       "          [-2.3882e-03,  2.2319e-03,  2.6923e-02],\n",
       "          [-1.6618e-02, -9.5127e-03,  1.0036e-02]],\n",
       "\n",
       "         [[-9.2893e-03,  1.3564e-02,  4.3857e-04],\n",
       "          [-1.3009e-02, -1.1595e-02, -5.0655e-03],\n",
       "          [ 1.2122e-02, -2.2194e-02,  2.3345e-03]]],\n",
       "\n",
       "\n",
       "        [[[ 3.2171e-03, -1.3386e-02, -7.2030e-04],\n",
       "          [-2.1997e-02,  8.4906e-04,  1.1830e-04],\n",
       "          [ 4.2290e-03, -9.6787e-03, -2.6545e-03]],\n",
       "\n",
       "         [[ 8.7121e-03, -1.3289e-02,  9.1391e-03],\n",
       "          [ 5.0244e-05, -5.6907e-03, -2.9439e-03],\n",
       "          [ 1.4083e-02,  4.6833e-03,  4.0747e-03]],\n",
       "\n",
       "         [[-2.3644e-02,  1.6085e-02,  1.0234e-02],\n",
       "          [ 1.2278e-02, -7.9973e-03, -9.5163e-03],\n",
       "          [-3.0229e-04,  8.3094e-03, -6.2113e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-2.6051e-03,  1.7901e-02,  5.0461e-03],\n",
       "          [ 8.3253e-03,  2.6952e-04,  1.6165e-02],\n",
       "          [ 1.1484e-02,  9.0496e-03,  8.0211e-05]],\n",
       "\n",
       "         [[-1.4534e-02, -1.8820e-04,  2.3973e-03],\n",
       "          [ 2.0174e-03, -9.7758e-04, -1.2427e-02],\n",
       "          [-1.7332e-03,  3.4503e-03,  1.5771e-03]],\n",
       "\n",
       "         [[-6.2294e-03, -2.0505e-02, -9.9376e-03],\n",
       "          [-5.7839e-03, -8.1187e-03, -1.5563e-02],\n",
       "          [ 6.8501e-03,  6.4751e-03,  6.1009e-03]]]], device='cuda:0'), tensor([ 0.0063,  0.0003,  0.0109,  0.0002,  0.0060,  0.0057, -0.0077, -0.0193,\n",
       "         0.0068,  0.0122, -0.0004,  0.0067, -0.0029, -0.0156,  0.0121,  0.0033,\n",
       "         0.0202, -0.0018, -0.0039, -0.0013,  0.0083, -0.0050,  0.0013,  0.0036,\n",
       "        -0.0041,  0.0001,  0.0041,  0.0134,  0.0021,  0.0184,  0.0042, -0.0005,\n",
       "         0.0143,  0.0027,  0.0138, -0.0158, -0.0010, -0.0032,  0.0093, -0.0064,\n",
       "        -0.0008, -0.0157, -0.0067, -0.0007, -0.0009,  0.0078, -0.0086, -0.0018,\n",
       "         0.0207,  0.0063, -0.0094, -0.0013,  0.0074,  0.0036, -0.0082, -0.0141,\n",
       "        -0.0158, -0.0053, -0.0208, -0.0079, -0.0078,  0.0195,  0.0035, -0.0117],\n",
       "       device='cuda:0'), tensor([ 0.0079,  0.0122, -0.0036,  0.0032,  0.0176,  0.0107,  0.0085, -0.0125,\n",
       "        -0.0095,  0.0033, -0.0007, -0.0021, -0.0094, -0.0019,  0.0095,  0.0135,\n",
       "         0.0190,  0.0177,  0.0089, -0.0162,  0.0082,  0.0053, -0.0045,  0.0079,\n",
       "         0.0067,  0.0182, -0.0073,  0.0205,  0.0083,  0.0025,  0.0056, -0.0015,\n",
       "        -0.0131, -0.0051,  0.0066, -0.0011, -0.0045,  0.0060,  0.0070,  0.0034,\n",
       "        -0.0129, -0.0153, -0.0026,  0.0096,  0.0016, -0.0077, -0.0094, -0.0038,\n",
       "         0.0074,  0.0112,  0.0350,  0.0069,  0.0015, -0.0070, -0.0110, -0.0044,\n",
       "         0.0036, -0.0122,  0.0159,  0.0145,  0.0195, -0.0027, -0.0022,  0.0148],\n",
       "       device='cuda:0'), tensor([[[[-4.1717e-04, -6.6875e-03,  5.4811e-03],\n",
       "          [ 2.4229e-02,  3.5995e-03,  1.2266e-02],\n",
       "          [-8.9412e-03, -1.6453e-02,  2.1240e-03]],\n",
       "\n",
       "         [[-8.2537e-03,  1.0031e-02, -8.4126e-03],\n",
       "          [ 7.8425e-03,  5.6838e-03, -2.9452e-03],\n",
       "          [ 8.9925e-03, -3.2775e-03,  1.6640e-03]],\n",
       "\n",
       "         [[ 3.0628e-03,  5.5798e-03, -2.0818e-03],\n",
       "          [ 3.5748e-03, -1.0338e-02, -1.0999e-02],\n",
       "          [ 5.7182e-03, -1.9391e-02,  6.3782e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-2.8393e-03, -8.3873e-03,  1.0896e-02],\n",
       "          [ 1.0442e-02, -1.0941e-02, -2.0220e-03],\n",
       "          [-9.5419e-04,  1.5720e-02, -4.5236e-03]],\n",
       "\n",
       "         [[-5.6190e-03,  4.1400e-04, -1.1071e-02],\n",
       "          [-8.9550e-03,  1.1874e-02,  7.1988e-03],\n",
       "          [-9.8119e-03, -1.4371e-02,  1.3926e-02]],\n",
       "\n",
       "         [[ 4.6850e-03, -1.3168e-02,  7.4218e-03],\n",
       "          [-1.4978e-02, -4.1510e-03, -2.0574e-04],\n",
       "          [-1.0861e-03,  7.8069e-03,  9.4120e-03]]],\n",
       "\n",
       "\n",
       "        [[[-1.5906e-03,  1.1687e-02,  6.7421e-03],\n",
       "          [ 3.9990e-03,  1.3788e-03, -8.8431e-05],\n",
       "          [-6.4906e-03,  9.5122e-03, -1.0690e-02]],\n",
       "\n",
       "         [[ 4.4878e-03,  5.4553e-04, -9.4073e-03],\n",
       "          [ 4.0872e-03, -2.3014e-03, -4.8102e-05],\n",
       "          [ 2.1327e-02,  7.4837e-03, -2.3729e-02]],\n",
       "\n",
       "         [[ 1.0138e-02, -8.3515e-03, -1.2046e-03],\n",
       "          [-3.3311e-03, -5.6278e-03, -9.2112e-03],\n",
       "          [ 1.6417e-02, -5.5229e-03,  4.4437e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 2.8133e-02, -2.7707e-03,  5.7310e-03],\n",
       "          [ 1.4830e-02, -1.1496e-02,  1.8991e-04],\n",
       "          [-3.1173e-03,  4.9787e-03,  5.8878e-03]],\n",
       "\n",
       "         [[ 1.5797e-02,  3.6438e-03,  4.5833e-04],\n",
       "          [-6.2619e-05, -7.6985e-03,  3.7328e-03],\n",
       "          [ 2.3191e-03,  1.3446e-02,  9.2054e-04]],\n",
       "\n",
       "         [[-1.4365e-02, -1.2583e-02,  1.7950e-02],\n",
       "          [-3.0623e-03,  4.5130e-04,  1.3215e-02],\n",
       "          [ 4.8584e-03,  6.1326e-03,  8.6938e-03]]],\n",
       "\n",
       "\n",
       "        [[[-1.5994e-02, -9.7475e-03, -5.2052e-04],\n",
       "          [-6.1502e-03,  9.0504e-03, -1.8425e-02],\n",
       "          [-9.5655e-03,  6.2727e-04, -2.4188e-03]],\n",
       "\n",
       "         [[-1.0699e-02, -7.2792e-03,  7.6471e-03],\n",
       "          [ 1.2173e-02,  1.1458e-02,  3.8546e-03],\n",
       "          [-1.4403e-02,  6.3818e-03,  5.3926e-04]],\n",
       "\n",
       "         [[-3.0440e-03, -7.9073e-03,  6.7964e-04],\n",
       "          [-4.4093e-03,  6.5672e-04, -1.7767e-02],\n",
       "          [ 6.3554e-03,  9.6975e-04,  1.3006e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-4.9577e-03, -2.3920e-03, -7.0139e-03],\n",
       "          [-2.0285e-02, -2.9653e-03,  7.1170e-03],\n",
       "          [ 1.4187e-03, -1.2151e-02,  3.5401e-03]],\n",
       "\n",
       "         [[-1.3519e-03,  1.8931e-05, -4.2479e-03],\n",
       "          [ 9.2366e-03, -9.5163e-04, -5.1078e-03],\n",
       "          [ 4.8096e-03, -1.5067e-02,  5.7184e-03]],\n",
       "\n",
       "         [[ 7.9026e-03, -9.8558e-03,  2.9289e-03],\n",
       "          [ 3.4222e-05, -1.3210e-02, -1.4747e-02],\n",
       "          [ 9.9499e-03,  1.3552e-02,  4.2705e-03]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[-1.3168e-02, -2.1474e-02,  3.6122e-03],\n",
       "          [ 6.4053e-04,  4.9342e-03,  7.0860e-03],\n",
       "          [-4.4162e-03, -5.8127e-03,  2.6966e-03]],\n",
       "\n",
       "         [[ 1.7901e-02, -1.8591e-02,  2.4055e-03],\n",
       "          [ 1.9716e-02,  1.5815e-02, -1.4959e-02],\n",
       "          [ 2.1474e-03,  5.6102e-03,  6.9284e-03]],\n",
       "\n",
       "         [[-1.0368e-02,  6.1012e-03,  9.1656e-03],\n",
       "          [ 2.0886e-02, -7.4714e-03, -1.1218e-03],\n",
       "          [ 5.4279e-03, -2.5211e-02,  2.1491e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 4.2030e-04,  7.7497e-03,  1.9958e-02],\n",
       "          [-2.3116e-04,  5.6045e-03, -4.6096e-03],\n",
       "          [-2.4917e-03, -5.0878e-03, -4.8137e-03]],\n",
       "\n",
       "         [[ 1.8665e-02,  6.9744e-03, -9.4732e-03],\n",
       "          [-9.4507e-03, -8.8008e-03, -9.2287e-03],\n",
       "          [-4.0649e-04, -7.6379e-03, -1.0383e-02]],\n",
       "\n",
       "         [[ 1.1463e-02,  5.2975e-03, -4.5704e-03],\n",
       "          [ 1.7256e-02,  2.3186e-03, -9.7370e-03],\n",
       "          [-4.6490e-03,  9.3922e-04,  8.8575e-05]]],\n",
       "\n",
       "\n",
       "        [[[ 6.6930e-03,  7.7344e-03, -7.7383e-03],\n",
       "          [-3.0767e-03, -1.7976e-02, -1.3326e-03],\n",
       "          [-3.0996e-03, -2.0303e-03, -2.5033e-03]],\n",
       "\n",
       "         [[-4.2907e-03,  6.3116e-03,  2.6182e-03],\n",
       "          [ 6.7666e-03,  3.9787e-03, -4.4939e-03],\n",
       "          [-5.0196e-04,  7.1746e-03,  3.2224e-03]],\n",
       "\n",
       "         [[-6.4388e-03, -1.1917e-02, -9.8350e-03],\n",
       "          [ 1.4514e-02, -3.3754e-03, -4.8695e-03],\n",
       "          [ 1.8563e-03, -1.8329e-02,  3.8633e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-9.9939e-03, -2.0813e-02,  9.2376e-03],\n",
       "          [ 8.7691e-03, -1.2196e-02,  1.3362e-02],\n",
       "          [ 8.4742e-03,  3.6546e-03,  8.1097e-03]],\n",
       "\n",
       "         [[-7.3390e-03,  7.6063e-03,  2.3582e-03],\n",
       "          [ 2.2468e-02, -1.7813e-03, -6.2879e-03],\n",
       "          [ 1.7130e-03,  6.6010e-03,  1.5058e-02]],\n",
       "\n",
       "         [[ 2.6311e-02, -2.1660e-02, -1.7502e-02],\n",
       "          [ 8.7344e-03,  9.6503e-04, -2.0155e-03],\n",
       "          [-1.1130e-02, -1.9506e-04,  1.0180e-02]]],\n",
       "\n",
       "\n",
       "        [[[-1.7745e-03, -9.8100e-03, -1.3715e-03],\n",
       "          [ 8.0071e-03,  1.8946e-04,  1.3893e-02],\n",
       "          [-1.4551e-02, -8.9697e-03,  1.0233e-02]],\n",
       "\n",
       "         [[-3.8693e-03,  2.9208e-03,  6.4929e-03],\n",
       "          [-4.8191e-03,  6.4024e-03,  2.5137e-02],\n",
       "          [ 1.3409e-03,  1.6935e-02,  6.5522e-03]],\n",
       "\n",
       "         [[-1.0834e-02, -2.3143e-04,  7.3926e-03],\n",
       "          [-7.4325e-03, -1.0665e-02, -4.7935e-03],\n",
       "          [ 9.3729e-03, -8.3298e-03,  3.9058e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 1.4725e-03, -6.9829e-03,  1.7639e-02],\n",
       "          [-5.8119e-03, -7.6966e-03, -5.3893e-03],\n",
       "          [ 1.0106e-02,  7.8562e-03,  5.8573e-03]],\n",
       "\n",
       "         [[-6.7511e-03, -4.4640e-03, -1.3842e-02],\n",
       "          [-3.8667e-03, -1.7369e-03,  9.6319e-03],\n",
       "          [ 7.5092e-03,  4.3740e-03, -1.1731e-02]],\n",
       "\n",
       "         [[-1.6533e-02,  1.3298e-02,  6.7015e-03],\n",
       "          [ 8.0050e-03,  2.7415e-04,  6.8810e-03],\n",
       "          [ 2.3097e-02,  7.6417e-03, -3.5673e-03]]]], device='cuda:0'), tensor([ 0.0084, -0.0088, -0.0039,  0.0098, -0.0166,  0.0045,  0.0033,  0.0107,\n",
       "        -0.0006,  0.0014, -0.0014, -0.0089, -0.0095, -0.0002,  0.0099, -0.0136,\n",
       "        -0.0061,  0.0102,  0.0076,  0.0080, -0.0078,  0.0019,  0.0002, -0.0005,\n",
       "         0.0142,  0.0050,  0.0121,  0.0136,  0.0092, -0.0006,  0.0124, -0.0105,\n",
       "         0.0001, -0.0002,  0.0040,  0.0044, -0.0108, -0.0102,  0.0033,  0.0030,\n",
       "         0.0168,  0.0103, -0.0013,  0.0013, -0.0039,  0.0010,  0.0135, -0.0105,\n",
       "        -0.0300, -0.0014, -0.0141, -0.0078, -0.0053, -0.0010,  0.0006,  0.0169,\n",
       "         0.0009,  0.0012, -0.0080, -0.0054, -0.0003, -0.0118, -0.0083,  0.0092],\n",
       "       device='cuda:0'), tensor([ 0.0033, -0.0065,  0.0031,  0.0053, -0.0005, -0.0030, -0.0042,  0.0020,\n",
       "        -0.0080,  0.0019,  0.0078,  0.0041,  0.0112,  0.0002,  0.0228, -0.0072,\n",
       "         0.0213, -0.0053,  0.0027, -0.0101, -0.0080,  0.0076,  0.0041, -0.0150,\n",
       "        -0.0020, -0.0100, -0.0059,  0.0023,  0.0103, -0.0084,  0.0022,  0.0019,\n",
       "        -0.0004, -0.0024, -0.0006, -0.0054,  0.0071,  0.0057, -0.0181,  0.0027,\n",
       "         0.0032,  0.0310,  0.0134,  0.0018, -0.0051, -0.0053, -0.0054,  0.0052,\n",
       "         0.0016, -0.0028,  0.0014,  0.0065,  0.0088,  0.0077,  0.0097,  0.0116,\n",
       "         0.0104,  0.0006, -0.0092,  0.0156, -0.0084,  0.0050, -0.0150,  0.0005],\n",
       "       device='cuda:0'), tensor([[[[ 7.5004e-03,  1.3863e-02, -1.0043e-02],\n",
       "          [-2.6385e-03, -2.2256e-03,  2.3970e-05],\n",
       "          [-1.1905e-02,  1.7783e-02,  1.3506e-02]],\n",
       "\n",
       "         [[ 5.8009e-04,  1.6322e-02, -1.0279e-03],\n",
       "          [ 8.8839e-03,  1.1809e-02,  2.3635e-03],\n",
       "          [-2.0766e-02,  9.0399e-04, -8.4693e-03]],\n",
       "\n",
       "         [[-6.2629e-03, -1.7602e-03,  1.3221e-02],\n",
       "          [-2.4076e-04,  1.4094e-02,  1.6124e-03],\n",
       "          [-1.3316e-02,  9.7415e-03,  1.7730e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 8.9534e-03,  1.4287e-03,  7.5536e-03],\n",
       "          [ 2.4966e-03,  8.7636e-03,  6.3723e-03],\n",
       "          [ 1.4134e-02, -7.3153e-03, -7.9250e-03]],\n",
       "\n",
       "         [[ 2.1978e-03,  7.4625e-03, -1.9995e-03],\n",
       "          [-9.7641e-03, -3.8152e-03,  3.2382e-03],\n",
       "          [-1.6107e-03,  4.9068e-03, -7.7599e-03]],\n",
       "\n",
       "         [[ 9.8043e-03,  2.0076e-02, -3.4654e-03],\n",
       "          [ 5.4094e-03, -2.3739e-03, -1.2927e-02],\n",
       "          [-4.6557e-04, -8.6288e-03,  3.5788e-03]]],\n",
       "\n",
       "\n",
       "        [[[ 1.3337e-02,  7.0012e-03, -6.9250e-03],\n",
       "          [-1.1093e-02,  6.0205e-03,  5.1149e-03],\n",
       "          [-1.6751e-02,  1.8781e-03,  2.1964e-03]],\n",
       "\n",
       "         [[-1.9691e-03,  5.3365e-03,  1.9614e-03],\n",
       "          [ 9.0858e-03, -8.6989e-03,  3.5837e-03],\n",
       "          [-3.9361e-03,  4.6468e-04, -7.0963e-03]],\n",
       "\n",
       "         [[ 9.5139e-03, -6.0677e-03,  8.0386e-03],\n",
       "          [-5.1066e-03,  2.5343e-02, -3.9700e-03],\n",
       "          [ 1.1152e-02, -6.9552e-03, -1.8952e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-5.1115e-03,  5.4634e-03,  5.7244e-03],\n",
       "          [-1.7906e-03, -8.9430e-03, -1.4305e-02],\n",
       "          [ 1.8555e-04,  7.4986e-03,  1.7950e-03]],\n",
       "\n",
       "         [[ 4.4585e-03,  1.1292e-02,  8.7672e-03],\n",
       "          [-9.7478e-03,  1.0577e-03,  2.8735e-02],\n",
       "          [-7.0554e-03, -2.0525e-03, -8.4399e-03]],\n",
       "\n",
       "         [[ 3.2687e-03,  1.1193e-02, -1.6680e-04],\n",
       "          [-1.2097e-02, -4.3868e-03,  8.9101e-03],\n",
       "          [ 2.0514e-02,  1.1816e-02,  8.3183e-03]]],\n",
       "\n",
       "\n",
       "        [[[-1.3728e-03,  6.7879e-03,  2.5156e-03],\n",
       "          [ 3.9244e-03, -3.9381e-03, -7.4870e-03],\n",
       "          [ 1.2994e-02, -3.7947e-03, -1.4262e-02]],\n",
       "\n",
       "         [[-1.4454e-02, -1.5808e-03, -3.3983e-03],\n",
       "          [-7.5848e-03, -6.3181e-03, -8.2854e-04],\n",
       "          [ 2.0699e-03,  3.2478e-03, -5.5971e-03]],\n",
       "\n",
       "         [[-3.7339e-03, -4.1793e-03,  1.0023e-02],\n",
       "          [ 3.0070e-02,  1.9627e-04,  5.4433e-03],\n",
       "          [-5.1996e-03,  4.5735e-03,  7.1632e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 3.5083e-03,  6.1175e-03, -1.2158e-02],\n",
       "          [ 1.6451e-02, -1.6464e-03, -5.0761e-03],\n",
       "          [ 2.6692e-02,  7.7704e-03,  3.8261e-03]],\n",
       "\n",
       "         [[ 1.8847e-03, -3.4990e-04, -1.7367e-03],\n",
       "          [-1.6764e-02,  1.2649e-03, -8.3855e-04],\n",
       "          [ 1.5942e-04, -1.9513e-03, -1.8074e-03]],\n",
       "\n",
       "         [[-3.8418e-03,  1.4677e-02,  1.3516e-02],\n",
       "          [-9.2768e-03, -8.5521e-03,  4.3561e-03],\n",
       "          [-8.5309e-03,  1.0747e-02,  4.2174e-03]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[ 8.6637e-03, -2.9737e-03,  8.9351e-03],\n",
       "          [-1.2149e-02,  1.9879e-02,  1.5093e-02],\n",
       "          [-3.4436e-03,  1.1560e-04,  1.3449e-02]],\n",
       "\n",
       "         [[ 6.1651e-03,  4.3020e-03, -1.2677e-02],\n",
       "          [ 1.5556e-02, -7.2148e-03,  2.6883e-03],\n",
       "          [ 2.4716e-02, -1.4307e-02, -8.1050e-03]],\n",
       "\n",
       "         [[-2.3156e-02, -1.6758e-02,  1.2588e-02],\n",
       "          [ 1.9322e-02, -1.2937e-03, -1.5351e-02],\n",
       "          [ 1.5588e-02, -5.4649e-03,  1.4281e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-8.5238e-03,  1.3645e-02, -1.1239e-02],\n",
       "          [-1.4391e-02, -6.2803e-03, -1.5302e-03],\n",
       "          [ 1.5643e-02, -7.3484e-03,  9.9169e-03]],\n",
       "\n",
       "         [[-4.5725e-03, -2.5296e-03, -1.2130e-02],\n",
       "          [ 9.5993e-03,  2.4146e-03, -1.0247e-02],\n",
       "          [ 4.5675e-03, -1.0158e-02, -1.8025e-03]],\n",
       "\n",
       "         [[-8.8278e-03, -7.5294e-03,  4.1751e-04],\n",
       "          [ 8.5794e-03,  1.7746e-02,  1.8407e-03],\n",
       "          [ 1.3317e-02, -2.0941e-02, -6.1637e-03]]],\n",
       "\n",
       "\n",
       "        [[[ 5.4050e-03,  4.4770e-03, -1.4213e-02],\n",
       "          [ 5.6655e-03,  5.4753e-03, -8.5283e-03],\n",
       "          [-6.6320e-03,  8.1662e-03,  1.6499e-02]],\n",
       "\n",
       "         [[ 1.8065e-03,  1.8089e-02,  1.1437e-02],\n",
       "          [-9.4199e-03, -4.6848e-04,  1.7979e-03],\n",
       "          [ 2.7529e-03, -1.2601e-02,  5.5523e-03]],\n",
       "\n",
       "         [[-8.8522e-03, -1.4140e-02, -1.7371e-02],\n",
       "          [ 5.2729e-03,  4.9189e-03, -1.0502e-02],\n",
       "          [-5.2656e-03, -1.6649e-02, -1.4744e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 2.6569e-02, -4.5777e-03,  3.6962e-03],\n",
       "          [-1.7672e-03,  5.2468e-03, -6.2439e-03],\n",
       "          [ 9.5082e-03,  1.3431e-03, -1.4920e-03]],\n",
       "\n",
       "         [[ 7.9465e-03,  5.4817e-03, -1.0039e-02],\n",
       "          [-1.9167e-03, -1.2827e-02, -1.7200e-02],\n",
       "          [-1.9016e-03, -7.0490e-03, -2.0945e-02]],\n",
       "\n",
       "         [[ 9.9802e-03, -5.2186e-04,  4.9441e-03],\n",
       "          [-9.6009e-03,  9.6084e-03,  4.6506e-03],\n",
       "          [ 2.1009e-02, -1.3402e-03, -4.9068e-03]]],\n",
       "\n",
       "\n",
       "        [[[ 3.9467e-03,  7.2799e-03, -1.2177e-02],\n",
       "          [ 5.0367e-03, -1.5682e-03, -3.3540e-03],\n",
       "          [ 1.1736e-02,  2.1734e-02,  1.2291e-03]],\n",
       "\n",
       "         [[-1.4164e-02, -1.1173e-02,  6.8556e-03],\n",
       "          [ 1.1248e-02,  2.9706e-03,  7.0689e-03],\n",
       "          [ 5.1398e-03, -3.8329e-03,  2.3078e-03]],\n",
       "\n",
       "         [[-1.3389e-02,  1.4469e-02, -7.9036e-03],\n",
       "          [-3.0630e-03,  5.2572e-03, -1.1384e-02],\n",
       "          [ 8.7824e-03,  5.2123e-03, -5.0574e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 8.6818e-03, -1.2858e-02, -9.1461e-03],\n",
       "          [-3.8657e-04,  8.5156e-04,  3.4113e-03],\n",
       "          [ 7.2010e-04,  2.3085e-02, -1.0579e-02]],\n",
       "\n",
       "         [[-1.3164e-02, -1.4085e-02, -1.1490e-02],\n",
       "          [ 3.7738e-03, -7.9587e-03,  7.7951e-03],\n",
       "          [ 2.3997e-03, -5.2085e-03,  4.9852e-03]],\n",
       "\n",
       "         [[ 1.1638e-02,  1.5014e-03, -3.9139e-03],\n",
       "          [-8.3491e-04, -5.0568e-03, -1.5396e-02],\n",
       "          [-1.8021e-03,  5.5265e-03,  1.1322e-03]]]], device='cuda:0'), tensor([ 1.4404e-02,  1.7908e-02,  1.2008e-02,  5.8821e-03,  1.0131e-02,\n",
       "         5.1855e-06,  5.2550e-03,  6.4546e-03, -7.7546e-03, -1.5545e-03,\n",
       "         6.5948e-03, -1.8005e-02,  5.8034e-03,  9.4142e-03,  6.6228e-03,\n",
       "         3.0458e-03,  4.5082e-03, -2.0249e-04, -1.3162e-02,  5.4772e-03,\n",
       "         1.4845e-03,  2.9531e-03, -1.1819e-04,  7.3824e-03, -5.5040e-03,\n",
       "         8.5169e-03,  3.1596e-04, -1.1076e-02, -4.3436e-03, -2.4620e-02,\n",
       "        -3.7622e-03,  1.4704e-02, -2.2762e-03, -1.5557e-03, -2.7045e-03,\n",
       "        -1.3941e-02, -6.3422e-03,  1.3776e-02, -1.3751e-02, -3.7370e-03,\n",
       "        -1.4348e-02, -2.4285e-03, -1.0594e-03, -5.5900e-03,  7.5146e-03,\n",
       "         2.8386e-05, -6.2880e-03,  1.1056e-02,  1.5833e-03, -7.2379e-03,\n",
       "         3.9271e-04, -2.9423e-03, -6.0899e-03, -8.9543e-03, -3.2500e-03,\n",
       "        -6.0529e-03, -8.0546e-04,  1.7896e-03, -1.1506e-03, -4.9236e-03,\n",
       "        -9.0213e-04,  3.7105e-03,  4.7355e-04, -1.8721e-02], device='cuda:0'), tensor([-6.1161e-05,  5.0555e-04, -1.0098e-02,  3.0907e-03, -2.7902e-03,\n",
       "         6.0807e-03, -4.3169e-03, -1.8142e-02,  2.3844e-02,  2.7948e-03,\n",
       "        -1.5862e-02, -4.8702e-03,  1.2972e-02,  1.0922e-02, -7.8938e-03,\n",
       "        -1.5578e-03, -4.5118e-04,  1.1716e-02,  1.1854e-02,  4.3668e-03,\n",
       "        -6.3006e-03,  2.1716e-04,  6.3447e-04,  3.2957e-03,  5.8879e-03,\n",
       "        -5.8010e-03,  1.5030e-02,  5.7026e-03,  2.9745e-03,  8.7124e-04,\n",
       "        -1.1183e-02, -1.9983e-02,  1.5978e-02,  3.3194e-04,  3.1000e-03,\n",
       "         1.4006e-02, -3.7436e-03,  3.0682e-03,  3.1796e-03,  6.6733e-03,\n",
       "         3.7043e-03,  5.3575e-03,  6.7300e-03, -1.5944e-04,  1.5771e-02,\n",
       "        -7.8938e-04, -2.5761e-02,  7.4583e-03, -3.1186e-03,  1.2476e-02,\n",
       "         5.5955e-03,  9.5786e-03, -7.1958e-04, -8.7900e-03, -2.0805e-02,\n",
       "         7.2037e-03, -4.2551e-03,  1.2678e-03, -8.1036e-03,  1.6621e-02,\n",
       "         4.9559e-03,  3.8370e-03,  5.8517e-03, -1.5619e-02], device='cuda:0'), tensor([[[[ 6.3516e-03, -6.2649e-03,  1.9242e-02],\n",
       "          [ 9.1221e-03, -3.4681e-03,  1.0018e-03],\n",
       "          [-1.1682e-04,  1.1405e-03,  2.8489e-03]],\n",
       "\n",
       "         [[ 1.6618e-03, -5.1650e-03,  6.3518e-03],\n",
       "          [-1.4340e-02,  2.9588e-03, -2.0143e-02],\n",
       "          [ 6.6671e-03,  7.6001e-03, -4.2080e-03]],\n",
       "\n",
       "         [[ 6.5982e-03, -2.2086e-03, -2.9666e-03],\n",
       "          [-1.0488e-02,  1.2002e-02, -7.6460e-03],\n",
       "          [-7.5628e-04,  1.4080e-02, -3.0375e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 2.4025e-03,  1.6607e-02, -8.6191e-03],\n",
       "          [ 1.2282e-03, -1.5448e-02, -1.2796e-02],\n",
       "          [-1.9589e-02,  5.4476e-03, -1.1437e-02]],\n",
       "\n",
       "         [[-9.1087e-03,  8.9464e-03,  1.2845e-02],\n",
       "          [ 7.3185e-03, -1.7042e-03,  1.0392e-02],\n",
       "          [-2.6063e-02, -3.9921e-03, -7.7260e-03]],\n",
       "\n",
       "         [[-9.9561e-03, -1.0139e-02,  1.4801e-02],\n",
       "          [ 4.5671e-03, -4.4897e-03, -1.0127e-03],\n",
       "          [ 1.0705e-02,  7.4140e-03,  3.0760e-03]]],\n",
       "\n",
       "\n",
       "        [[[ 4.1811e-03,  9.3508e-03,  7.4723e-03],\n",
       "          [-1.0377e-02,  8.1715e-03,  1.2827e-02],\n",
       "          [ 2.8753e-03, -1.9387e-03, -4.0353e-03]],\n",
       "\n",
       "         [[ 9.2136e-06,  1.7096e-03, -1.0516e-02],\n",
       "          [ 2.6421e-03, -1.0909e-02, -1.7011e-03],\n",
       "          [ 2.5037e-02,  1.4613e-02,  2.4814e-03]],\n",
       "\n",
       "         [[ 9.8243e-03, -5.1183e-03,  1.1624e-02],\n",
       "          [ 2.3134e-03, -9.6112e-03,  1.5195e-02],\n",
       "          [-1.4252e-02, -1.9580e-03,  7.4322e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 1.5983e-03, -1.6195e-02, -7.8166e-03],\n",
       "          [ 2.2844e-02, -2.8006e-02, -1.1706e-02],\n",
       "          [ 1.1932e-03,  1.5157e-03,  5.1264e-03]],\n",
       "\n",
       "         [[ 4.1069e-03, -6.9308e-03,  1.6355e-02],\n",
       "          [-7.6016e-03,  6.6358e-03, -2.8968e-03],\n",
       "          [ 5.7508e-03,  4.3272e-03,  1.2899e-03]],\n",
       "\n",
       "         [[ 6.0578e-03, -9.5776e-03,  2.0590e-03],\n",
       "          [ 3.9612e-03, -2.0004e-03,  1.2344e-02],\n",
       "          [ 2.5011e-02, -8.6032e-03, -8.0055e-03]]],\n",
       "\n",
       "\n",
       "        [[[ 1.2044e-02,  7.6362e-03, -1.2858e-02],\n",
       "          [ 1.0450e-02,  3.8942e-03,  9.9943e-03],\n",
       "          [-3.4491e-03,  5.5530e-03,  1.0180e-02]],\n",
       "\n",
       "         [[-1.0476e-02,  2.3193e-02,  2.7684e-03],\n",
       "          [ 7.2601e-03,  6.1465e-03, -3.2939e-03],\n",
       "          [ 7.4261e-03,  1.3165e-03,  3.8641e-03]],\n",
       "\n",
       "         [[ 1.7209e-02,  2.2305e-04, -8.7457e-03],\n",
       "          [-9.5562e-05,  2.1047e-03, -1.0576e-02],\n",
       "          [ 2.3311e-03,  2.4255e-03, -1.4619e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 1.6200e-02,  8.1989e-03, -4.8894e-03],\n",
       "          [ 5.1412e-03, -9.4172e-03,  1.0444e-03],\n",
       "          [-8.2484e-03,  1.1852e-02, -1.2167e-02]],\n",
       "\n",
       "         [[ 1.2209e-02,  1.3023e-02, -5.3707e-03],\n",
       "          [ 3.2197e-03, -2.0341e-03,  8.4395e-04],\n",
       "          [-1.9270e-02,  9.5659e-03, -8.1921e-03]],\n",
       "\n",
       "         [[ 1.0957e-02,  3.9835e-03,  1.3847e-02],\n",
       "          [-2.5988e-03,  6.8033e-03, -2.8865e-02],\n",
       "          [-1.3108e-02,  4.0918e-03, -1.8347e-04]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[-6.0554e-03,  9.5809e-04,  1.4813e-02],\n",
       "          [ 1.0164e-02,  2.9590e-03,  9.3727e-03],\n",
       "          [-1.3850e-03, -7.2474e-03, -2.5890e-03]],\n",
       "\n",
       "         [[-4.3764e-03,  1.1936e-02,  2.6271e-02],\n",
       "          [-1.5907e-02, -2.8792e-03,  1.1065e-03],\n",
       "          [-1.4784e-03, -6.9825e-03,  1.5167e-02]],\n",
       "\n",
       "         [[-4.1353e-03,  1.6902e-02, -1.2490e-02],\n",
       "          [-1.5743e-03, -1.4468e-02,  4.9572e-03],\n",
       "          [-1.3791e-02, -2.6426e-03, -1.0134e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 1.6685e-02, -1.3503e-02, -2.6645e-03],\n",
       "          [-1.5671e-02, -4.8440e-03,  9.1537e-03],\n",
       "          [ 7.5493e-03, -5.8162e-03, -8.1105e-03]],\n",
       "\n",
       "         [[-8.9825e-03, -2.0712e-02,  1.3966e-03],\n",
       "          [ 3.0107e-03,  4.4293e-03, -7.2740e-03],\n",
       "          [-1.5403e-03,  1.9889e-03, -9.0799e-04]],\n",
       "\n",
       "         [[-1.1990e-02, -6.9964e-03,  6.6458e-03],\n",
       "          [-8.3943e-03,  9.2694e-03,  1.3587e-02],\n",
       "          [ 5.5427e-03, -8.4198e-03, -7.6345e-03]]],\n",
       "\n",
       "\n",
       "        [[[ 5.1184e-03,  9.6354e-03,  1.8719e-02],\n",
       "          [ 3.6448e-03, -7.9986e-03,  1.7461e-02],\n",
       "          [-2.5058e-03,  4.3180e-03, -1.8773e-02]],\n",
       "\n",
       "         [[ 1.7205e-02,  8.3745e-03,  1.8622e-03],\n",
       "          [ 5.6501e-03, -1.2271e-02,  2.6063e-02],\n",
       "          [ 4.6548e-03, -3.7491e-03, -1.1001e-02]],\n",
       "\n",
       "         [[ 4.9579e-03, -1.1293e-02,  1.8421e-03],\n",
       "          [ 1.4391e-03, -1.3953e-02, -7.9446e-04],\n",
       "          [-7.5533e-04, -1.9969e-02, -9.6295e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 7.5495e-03, -1.4091e-02, -9.5465e-05],\n",
       "          [-1.3096e-02,  1.1511e-02,  4.7867e-03],\n",
       "          [-1.3505e-02,  1.5540e-02, -5.3072e-03]],\n",
       "\n",
       "         [[-1.6351e-02, -9.2165e-05, -4.9559e-03],\n",
       "          [-4.6563e-03,  6.8162e-03, -9.1445e-03],\n",
       "          [-5.0690e-03,  1.1782e-02, -4.7198e-03]],\n",
       "\n",
       "         [[ 1.5388e-02,  8.7222e-03, -2.8207e-03],\n",
       "          [-7.1540e-04, -9.1164e-03, -6.0542e-03],\n",
       "          [ 1.2723e-02,  3.7835e-03, -1.1113e-03]]],\n",
       "\n",
       "\n",
       "        [[[ 2.2203e-03, -6.4469e-03, -2.1546e-03],\n",
       "          [-3.5580e-03, -1.2024e-02, -1.4213e-02],\n",
       "          [ 2.2904e-03, -1.2408e-02,  1.0651e-02]],\n",
       "\n",
       "         [[ 1.0910e-02, -8.3521e-03,  1.4848e-02],\n",
       "          [-4.0450e-03, -1.1306e-02, -4.8260e-03],\n",
       "          [ 1.1617e-02, -5.7414e-03,  4.1936e-03]],\n",
       "\n",
       "         [[-1.5790e-02,  9.3585e-03,  2.0779e-02],\n",
       "          [ 1.3005e-02,  2.7514e-03, -1.5703e-03],\n",
       "          [ 1.0973e-03,  2.5846e-03, -1.4680e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-1.5917e-02, -1.6105e-02,  2.0429e-02],\n",
       "          [ 8.5506e-03, -3.8850e-03,  1.4871e-02],\n",
       "          [-7.1727e-03, -1.5087e-03, -8.5387e-03]],\n",
       "\n",
       "         [[-5.2109e-03, -1.1672e-02, -8.3780e-03],\n",
       "          [-8.1213e-04, -5.4529e-03,  2.7431e-03],\n",
       "          [ 5.7127e-03, -2.0669e-03, -1.4536e-02]],\n",
       "\n",
       "         [[ 1.3004e-03,  8.2684e-03, -5.7259e-03],\n",
       "          [-1.3422e-02, -3.2226e-03,  1.1438e-02],\n",
       "          [ 9.9749e-03, -7.2601e-03,  4.7709e-03]]]], device='cuda:0'), tensor([-0.0016,  0.0125, -0.0142,  0.0111,  0.0115,  0.0106,  0.0077,  0.0126,\n",
       "         0.0023,  0.0185, -0.0166, -0.0055, -0.0154,  0.0022,  0.0047, -0.0122,\n",
       "        -0.0010,  0.0041, -0.0047, -0.0127,  0.0030, -0.0001,  0.0031,  0.0027,\n",
       "         0.0003, -0.0149, -0.0133, -0.0132, -0.0102,  0.0107, -0.0043,  0.0020,\n",
       "         0.0095, -0.0019, -0.0132,  0.0129, -0.0088,  0.0033,  0.0116, -0.0037,\n",
       "        -0.0051,  0.0034, -0.0071, -0.0067,  0.0095,  0.0231,  0.0130,  0.0058,\n",
       "        -0.0081,  0.0145,  0.0060, -0.0105, -0.0023,  0.0045,  0.0051,  0.0196,\n",
       "        -0.0005, -0.0169,  0.0034, -0.0059,  0.0039, -0.0052,  0.0090,  0.0112,\n",
       "        -0.0256,  0.0022, -0.0088,  0.0097,  0.0152,  0.0027, -0.0075,  0.0091,\n",
       "         0.0120, -0.0033, -0.0077,  0.0168, -0.0029, -0.0008, -0.0042, -0.0053,\n",
       "        -0.0004,  0.0125,  0.0031, -0.0052, -0.0067, -0.0109, -0.0112, -0.0038,\n",
       "         0.0015, -0.0062, -0.0202, -0.0115,  0.0094, -0.0028,  0.0140, -0.0005,\n",
       "         0.0012,  0.0075,  0.0003,  0.0183,  0.0028,  0.0042,  0.0040,  0.0008,\n",
       "         0.0001,  0.0059, -0.0069,  0.0056,  0.0154,  0.0141, -0.0068, -0.0007,\n",
       "        -0.0083, -0.0105,  0.0042,  0.0024,  0.0102,  0.0062,  0.0074,  0.0039,\n",
       "        -0.0126, -0.0022, -0.0039, -0.0053,  0.0011, -0.0206,  0.0048,  0.0010],\n",
       "       device='cuda:0'), tensor([ 5.6428e-04,  3.0633e-03, -1.1906e-02, -2.7888e-03, -1.1484e-02,\n",
       "         1.5394e-02,  2.4354e-02, -4.9852e-03, -7.5381e-03, -3.9738e-03,\n",
       "         7.9320e-03, -6.9115e-03, -2.2571e-02, -1.6129e-02, -1.1964e-02,\n",
       "         1.6132e-03,  3.8496e-03, -1.4399e-02, -3.2866e-03,  1.0535e-02,\n",
       "         3.6746e-03,  4.2909e-03, -8.5898e-03,  5.4691e-03, -7.1584e-03,\n",
       "        -1.5158e-02,  2.2193e-03, -1.1633e-02, -4.1754e-03, -1.1296e-02,\n",
       "        -2.4666e-03,  1.4325e-02, -1.2415e-03,  1.3226e-02,  1.0978e-02,\n",
       "        -4.0710e-03,  1.5989e-02, -1.5579e-02,  5.9856e-03,  4.7488e-03,\n",
       "         4.1160e-03,  2.7570e-05, -6.2831e-03, -9.0356e-05,  1.0681e-02,\n",
       "         4.3200e-03, -5.7771e-03,  3.9421e-03,  1.1794e-03,  1.5922e-02,\n",
       "        -5.9411e-03, -9.6869e-03, -6.5258e-03, -2.8179e-03, -5.6685e-03,\n",
       "         2.1140e-02,  1.2592e-02,  1.1659e-03,  2.6744e-03, -4.9359e-03,\n",
       "        -5.0337e-03, -1.3636e-02, -1.2432e-02,  1.1954e-02,  6.8715e-03,\n",
       "        -2.8778e-02,  1.7896e-02,  9.8750e-03, -2.7859e-03,  4.0021e-03,\n",
       "         1.2144e-02,  4.2341e-03,  1.8990e-02,  5.8757e-03, -1.1842e-02,\n",
       "        -2.5816e-03, -2.1205e-03,  2.1467e-03,  1.4778e-02,  7.9542e-03,\n",
       "         1.1724e-02, -2.5660e-02, -3.5713e-03,  1.4590e-02, -8.4546e-03,\n",
       "        -6.7314e-03, -2.5435e-03, -1.7255e-03,  1.2656e-03, -4.4119e-03,\n",
       "        -5.8503e-03, -1.2111e-02, -1.1151e-02, -1.7010e-03, -7.5212e-03,\n",
       "        -1.3506e-03,  2.3325e-03,  1.8065e-02, -1.1875e-02,  2.0060e-02,\n",
       "        -7.8202e-03,  7.2494e-03,  6.9194e-03,  2.0845e-03,  5.5078e-03,\n",
       "         2.4470e-02,  1.3000e-03,  9.8640e-03, -5.6430e-03, -4.9570e-03,\n",
       "        -1.8434e-02, -5.0036e-03,  1.6587e-02, -5.6335e-04, -6.8262e-03,\n",
       "        -4.6112e-03, -1.9828e-02, -1.1871e-02, -5.4757e-03,  2.7513e-03,\n",
       "         1.0249e-02,  1.2801e-02, -2.6826e-04, -1.2005e-02, -9.0966e-03,\n",
       "         1.3017e-02, -1.6120e-02, -4.0990e-03], device='cuda:0'), tensor([[[[-4.1252e-03,  2.4253e-05, -1.4248e-03],\n",
       "          [ 1.0799e-02,  1.2250e-02,  2.0162e-02],\n",
       "          [-3.1661e-03, -1.2143e-02, -4.0336e-03]],\n",
       "\n",
       "         [[-2.0558e-02,  6.0123e-04,  1.1382e-02],\n",
       "          [ 1.9415e-03, -7.3678e-04, -1.0699e-02],\n",
       "          [ 5.6719e-03, -6.9449e-03, -1.2553e-02]],\n",
       "\n",
       "         [[ 2.0333e-02,  4.3775e-03,  1.0882e-02],\n",
       "          [-6.6703e-03,  1.9879e-03, -1.8472e-02],\n",
       "          [-1.8047e-02,  1.6338e-02,  5.6919e-04]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 1.4190e-03,  2.9945e-04, -1.9850e-03],\n",
       "          [ 2.0232e-03,  2.5840e-03, -8.5166e-03],\n",
       "          [ 1.3651e-02,  2.3409e-03, -4.2411e-03]],\n",
       "\n",
       "         [[-1.8069e-03, -3.4563e-04, -5.4500e-03],\n",
       "          [ 1.2815e-02, -1.1244e-02, -7.2066e-03],\n",
       "          [ 2.1919e-03,  3.0762e-03,  5.8704e-03]],\n",
       "\n",
       "         [[ 6.0849e-03,  9.8136e-03, -1.2758e-03],\n",
       "          [-1.6859e-02,  1.5810e-03,  2.4163e-03],\n",
       "          [ 1.5737e-03,  1.5503e-03,  1.4852e-02]]],\n",
       "\n",
       "\n",
       "        [[[-1.7921e-03,  6.4102e-03, -5.4103e-03],\n",
       "          [-2.4618e-03,  8.4853e-03,  1.9651e-02],\n",
       "          [-7.7665e-03,  2.5046e-03,  2.7426e-03]],\n",
       "\n",
       "         [[-4.3519e-03, -6.1342e-03,  2.0386e-02],\n",
       "          [ 3.3506e-03,  2.0482e-02,  3.5003e-03],\n",
       "          [-8.7203e-03,  8.0313e-03, -9.0419e-03]],\n",
       "\n",
       "         [[ 1.6636e-02, -2.4725e-03,  5.2132e-03],\n",
       "          [-1.9367e-02, -5.3317e-04, -1.1806e-02],\n",
       "          [-1.8701e-02, -1.0206e-02, -4.7511e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 1.9637e-03,  1.0751e-03,  1.4058e-03],\n",
       "          [ 1.0411e-03, -1.8240e-03,  6.8097e-03],\n",
       "          [-5.4831e-03, -6.7150e-03, -1.1987e-02]],\n",
       "\n",
       "         [[-2.0885e-03, -7.0884e-03,  2.2209e-03],\n",
       "          [ 6.5629e-03, -1.7976e-02, -3.3875e-04],\n",
       "          [-2.7980e-03, -2.5988e-02,  1.8978e-03]],\n",
       "\n",
       "         [[ 5.6271e-04,  1.2276e-02,  1.2304e-02],\n",
       "          [ 7.1426e-04,  3.3942e-03,  2.3995e-03],\n",
       "          [ 1.1754e-02,  1.4782e-02, -2.1329e-02]]],\n",
       "\n",
       "\n",
       "        [[[ 1.1378e-02, -1.8878e-02, -1.3544e-02],\n",
       "          [ 1.0389e-03, -7.1238e-03,  6.3507e-03],\n",
       "          [-6.8731e-03,  4.7405e-03, -3.0174e-03]],\n",
       "\n",
       "         [[-1.7584e-02, -2.1511e-03, -1.5041e-02],\n",
       "          [ 5.5860e-03, -3.7848e-03,  4.7908e-03],\n",
       "          [-8.5711e-03, -3.7173e-03,  9.0243e-03]],\n",
       "\n",
       "         [[-1.8050e-02,  1.2908e-02,  1.0175e-03],\n",
       "          [ 5.7978e-03,  1.3259e-02, -5.9808e-03],\n",
       "          [-1.1949e-02, -4.3889e-03,  4.6619e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-7.3764e-03, -2.9003e-04,  2.8502e-03],\n",
       "          [ 9.0200e-03, -6.1819e-03, -3.9543e-04],\n",
       "          [-8.5464e-03,  6.8680e-03,  1.3636e-02]],\n",
       "\n",
       "         [[-3.4428e-03, -1.2975e-02, -7.4932e-03],\n",
       "          [ 6.3957e-03, -1.0538e-02, -1.8228e-03],\n",
       "          [ 1.5782e-03, -5.3300e-03, -1.1206e-02]],\n",
       "\n",
       "         [[-6.5734e-03, -1.7218e-03,  4.7690e-03],\n",
       "          [ 6.2852e-03,  1.3482e-03,  1.1190e-03],\n",
       "          [-6.1939e-03, -2.2148e-02,  4.3028e-03]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[-8.5661e-03,  1.2165e-02, -1.6893e-02],\n",
       "          [-1.1807e-02, -8.9514e-03,  1.3484e-03],\n",
       "          [-1.9146e-02,  1.3974e-02, -1.3564e-02]],\n",
       "\n",
       "         [[ 4.3898e-03,  3.2316e-03, -1.3062e-02],\n",
       "          [ 5.6218e-03, -1.0857e-02, -9.1520e-03],\n",
       "          [ 5.4399e-03,  1.7107e-02,  6.5275e-03]],\n",
       "\n",
       "         [[ 5.2352e-03,  2.5323e-03, -8.2779e-04],\n",
       "          [ 5.9370e-03, -3.5103e-03, -4.5766e-03],\n",
       "          [ 1.0646e-02,  7.6041e-03,  7.0712e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 9.9450e-03, -2.5316e-03,  4.3903e-03],\n",
       "          [-1.6897e-02,  2.0833e-02, -1.3279e-02],\n",
       "          [-1.2570e-02,  5.8789e-03, -1.6885e-02]],\n",
       "\n",
       "         [[ 1.0560e-02,  6.8498e-03, -1.2931e-02],\n",
       "          [ 6.3123e-03, -1.1663e-02,  1.0265e-02],\n",
       "          [ 1.5791e-02, -5.2310e-03, -5.7547e-03]],\n",
       "\n",
       "         [[-1.0033e-02,  6.5958e-03, -1.4146e-02],\n",
       "          [ 2.0051e-02, -3.1694e-03,  9.5979e-03],\n",
       "          [ 4.9164e-04, -1.3560e-02,  2.8020e-03]]],\n",
       "\n",
       "\n",
       "        [[[ 1.2331e-02, -5.9871e-03,  3.0381e-03],\n",
       "          [ 1.5804e-02, -1.0105e-05,  7.5226e-03],\n",
       "          [-6.5382e-04, -6.1522e-04, -1.1664e-02]],\n",
       "\n",
       "         [[-1.9256e-02,  1.4946e-02,  3.4830e-03],\n",
       "          [-4.3930e-03,  2.6691e-04,  1.8448e-02],\n",
       "          [-6.5954e-03,  1.0826e-02,  5.7041e-03]],\n",
       "\n",
       "         [[ 7.8357e-03,  7.1879e-03, -8.8197e-03],\n",
       "          [-1.0787e-02,  6.9694e-03,  4.5258e-03],\n",
       "          [ 1.0265e-03,  7.8497e-03,  1.1883e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-3.4632e-03,  1.0110e-02,  5.0590e-03],\n",
       "          [ 5.5753e-03,  8.5249e-03, -1.9276e-02],\n",
       "          [-8.6465e-03,  5.2523e-03,  1.1581e-03]],\n",
       "\n",
       "         [[-1.2000e-02,  3.8461e-03, -3.4668e-03],\n",
       "          [ 2.3408e-02,  9.7442e-03,  1.1847e-02],\n",
       "          [-9.0750e-03, -1.1308e-02, -3.1508e-03]],\n",
       "\n",
       "         [[-3.6556e-03,  3.3597e-02,  3.9437e-03],\n",
       "          [-1.4407e-02, -1.5460e-02,  2.2904e-03],\n",
       "          [-3.6552e-03,  7.9574e-03,  6.6014e-03]]],\n",
       "\n",
       "\n",
       "        [[[-8.4532e-03,  5.3171e-03,  7.6745e-03],\n",
       "          [-9.0666e-03, -5.3365e-03,  8.7512e-03],\n",
       "          [-2.6742e-02, -9.6750e-03, -2.1992e-02]],\n",
       "\n",
       "         [[ 1.5401e-02,  1.6457e-03,  6.4128e-04],\n",
       "          [-4.6372e-03, -9.3046e-03,  7.7031e-04],\n",
       "          [ 3.6720e-04, -1.3178e-02,  1.3191e-02]],\n",
       "\n",
       "         [[-1.4907e-02,  1.8866e-02,  9.1537e-03],\n",
       "          [-8.9112e-03, -8.3237e-03, -1.2040e-02],\n",
       "          [ 2.1835e-03, -4.7723e-03, -1.4731e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 1.9642e-03,  7.6673e-03,  2.0389e-02],\n",
       "          [-1.0778e-03,  1.1322e-02,  5.1523e-04],\n",
       "          [ 6.0259e-03,  2.9915e-03,  2.5613e-03]],\n",
       "\n",
       "         [[-4.1679e-03,  7.4794e-03, -1.0460e-03],\n",
       "          [-2.1807e-02, -1.1977e-02, -4.2814e-03],\n",
       "          [-3.1936e-02, -7.8720e-03, -8.3262e-03]],\n",
       "\n",
       "         [[ 8.2609e-03, -5.5045e-03, -3.1492e-04],\n",
       "          [ 2.1242e-03,  3.0830e-04,  8.0049e-03],\n",
       "          [ 9.0059e-03, -4.7721e-04,  1.1109e-02]]]], device='cuda:0'), tensor([ 0.0019,  0.0023, -0.0135,  0.0070,  0.0148, -0.0066, -0.0064,  0.0036,\n",
       "        -0.0171,  0.0075,  0.0035,  0.0056, -0.0091, -0.0251, -0.0048, -0.0112,\n",
       "        -0.0027, -0.0023,  0.0093, -0.0084,  0.0073,  0.0057,  0.0057,  0.0065,\n",
       "         0.0161,  0.0015, -0.0081, -0.0045,  0.0123,  0.0175,  0.0058,  0.0046,\n",
       "        -0.0019, -0.0078, -0.0047,  0.0015,  0.0013,  0.0067, -0.0057, -0.0031,\n",
       "        -0.0008,  0.0066,  0.0144, -0.0022,  0.0021, -0.0139, -0.0034,  0.0017,\n",
       "        -0.0090,  0.0100,  0.0064,  0.0168,  0.0047,  0.0016, -0.0135,  0.0092,\n",
       "        -0.0186, -0.0018,  0.0009, -0.0201, -0.0097, -0.0028, -0.0061,  0.0048,\n",
       "        -0.0136,  0.0024, -0.0059,  0.0042,  0.0033, -0.0135, -0.0068,  0.0033,\n",
       "         0.0027, -0.0083,  0.0002, -0.0090, -0.0113, -0.0101, -0.0033, -0.0093,\n",
       "         0.0015, -0.0082, -0.0017,  0.0024, -0.0075, -0.0017,  0.0015,  0.0021,\n",
       "        -0.0011, -0.0040,  0.0028,  0.0020,  0.0069, -0.0051, -0.0046,  0.0074,\n",
       "        -0.0024, -0.0043, -0.0080, -0.0030, -0.0203, -0.0147,  0.0035, -0.0095,\n",
       "         0.0121,  0.0021, -0.0010, -0.0027,  0.0066, -0.0131, -0.0102, -0.0084,\n",
       "         0.0018,  0.0100,  0.0007, -0.0018, -0.0070,  0.0042,  0.0069, -0.0026,\n",
       "        -0.0057, -0.0073, -0.0133, -0.0011,  0.0176,  0.0048, -0.0210,  0.0050],\n",
       "       device='cuda:0'), tensor([-3.0386e-03,  1.0537e-02, -9.7432e-03, -6.4404e-03,  1.3102e-02,\n",
       "        -2.4992e-03, -1.9481e-02, -1.1776e-03, -1.8864e-02, -2.2639e-03,\n",
       "        -7.5794e-03, -8.1340e-03, -5.2170e-03,  6.5458e-03, -8.1191e-03,\n",
       "        -1.2442e-02,  1.5398e-03,  7.0895e-03,  2.3749e-02,  4.1176e-03,\n",
       "        -1.3212e-02, -7.1984e-03, -8.3844e-03, -1.3550e-04,  9.6934e-03,\n",
       "         1.1292e-02, -1.3160e-02,  1.4884e-03,  2.7925e-03,  3.0087e-03,\n",
       "        -2.1215e-02, -2.8788e-02, -1.0831e-03,  1.5967e-02, -4.5761e-03,\n",
       "        -7.4390e-03, -7.9543e-03,  9.4765e-03,  4.3695e-03, -2.8960e-03,\n",
       "        -7.2659e-03, -1.1381e-03,  5.1181e-03, -1.2499e-02, -1.2424e-02,\n",
       "        -5.3157e-03, -6.2314e-03,  1.1799e-02,  5.1106e-03,  2.6454e-03,\n",
       "        -5.4160e-03,  1.9580e-02,  3.7625e-03,  1.4611e-02, -1.0433e-02,\n",
       "         2.6720e-02,  7.0312e-03,  3.5068e-03,  1.5057e-02, -5.5245e-03,\n",
       "        -3.9472e-03,  1.3841e-03, -9.6688e-03,  8.2182e-03,  1.0931e-02,\n",
       "         1.0088e-02,  6.0885e-05,  7.4896e-03, -8.5632e-03,  3.8203e-03,\n",
       "         7.6306e-03,  5.0281e-03, -2.9817e-03, -1.5602e-02,  1.8934e-04,\n",
       "        -1.0937e-02,  4.2063e-03,  2.1584e-03,  1.0492e-02,  8.7281e-03,\n",
       "         3.6175e-03,  3.5243e-03, -7.8068e-03, -2.9407e-03,  3.1701e-02,\n",
       "         4.0486e-03,  2.3642e-03,  4.3035e-03, -8.6441e-03,  2.4216e-03,\n",
       "         3.2895e-04, -2.2595e-03, -1.0392e-03,  1.6587e-02,  4.4612e-03,\n",
       "         8.2489e-03, -2.3213e-04, -1.6781e-03, -1.3681e-02,  1.8936e-03,\n",
       "        -4.1402e-03,  1.3495e-02, -1.0110e-02,  1.7159e-02,  5.8419e-03,\n",
       "        -1.0236e-02,  7.9709e-03,  2.2637e-04,  1.0072e-02,  5.3190e-03,\n",
       "        -1.7313e-02,  2.0027e-02,  2.8326e-03,  4.2183e-03, -8.5814e-03,\n",
       "        -6.9413e-03,  4.8058e-03, -5.8785e-03,  1.4059e-02,  1.0791e-03,\n",
       "         1.6800e-03, -1.5742e-02,  4.0849e-03,  6.2027e-03,  1.5246e-03,\n",
       "        -2.9764e-03,  3.9343e-03, -1.3709e-02], device='cuda:0'), tensor([[[[-0.0087]],\n",
       "\n",
       "         [[-0.0318]],\n",
       "\n",
       "         [[-0.0060]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-0.0017]],\n",
       "\n",
       "         [[-0.0070]],\n",
       "\n",
       "         [[-0.0120]]],\n",
       "\n",
       "\n",
       "        [[[-0.0095]],\n",
       "\n",
       "         [[ 0.0015]],\n",
       "\n",
       "         [[ 0.0044]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-0.0024]],\n",
       "\n",
       "         [[-0.0002]],\n",
       "\n",
       "         [[ 0.0134]]],\n",
       "\n",
       "\n",
       "        [[[-0.0076]],\n",
       "\n",
       "         [[ 0.0044]],\n",
       "\n",
       "         [[ 0.0030]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 0.0068]],\n",
       "\n",
       "         [[ 0.0052]],\n",
       "\n",
       "         [[-0.0195]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[ 0.0144]],\n",
       "\n",
       "         [[ 0.0090]],\n",
       "\n",
       "         [[ 0.0090]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-0.0134]],\n",
       "\n",
       "         [[-0.0063]],\n",
       "\n",
       "         [[-0.0175]]],\n",
       "\n",
       "\n",
       "        [[[ 0.0193]],\n",
       "\n",
       "         [[ 0.0147]],\n",
       "\n",
       "         [[-0.0005]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 0.0108]],\n",
       "\n",
       "         [[-0.0093]],\n",
       "\n",
       "         [[ 0.0053]]],\n",
       "\n",
       "\n",
       "        [[[-0.0244]],\n",
       "\n",
       "         [[ 0.0079]],\n",
       "\n",
       "         [[ 0.0164]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 0.0032]],\n",
       "\n",
       "         [[-0.0015]],\n",
       "\n",
       "         [[ 0.0018]]]], device='cuda:0'), tensor([-1.1895e-02, -1.3182e-02,  5.7570e-03,  9.7358e-03, -9.8895e-03,\n",
       "        -1.1312e-02, -5.0024e-04, -3.7271e-04, -6.2812e-03, -8.1589e-03,\n",
       "        -7.9575e-03, -8.2362e-03,  4.1813e-03,  1.9182e-02,  1.2454e-02,\n",
       "        -6.9961e-03,  1.2657e-02, -2.4573e-02,  1.1265e-02,  5.4356e-03,\n",
       "        -1.6580e-02,  1.7650e-03, -4.7358e-03, -8.1760e-04, -6.0475e-03,\n",
       "        -1.4212e-02,  4.5998e-03, -8.7020e-03,  1.2897e-02, -3.0153e-03,\n",
       "         5.9400e-05,  2.0593e-03, -1.7009e-03,  5.6154e-04,  5.2240e-04,\n",
       "        -4.7969e-03, -2.7268e-03,  1.1287e-03,  1.1548e-03,  7.2209e-03,\n",
       "        -1.1729e-02, -1.0813e-03,  9.7356e-04, -7.1037e-03, -3.5461e-03,\n",
       "        -1.2339e-02,  1.8402e-03,  9.9209e-03, -1.5505e-02, -2.4365e-02,\n",
       "        -7.3313e-03,  3.1430e-03,  1.1183e-02, -1.1968e-02, -1.5265e-02,\n",
       "         1.8207e-03,  3.8616e-03, -1.8877e-03,  5.9296e-03, -5.1376e-03,\n",
       "         2.1804e-02,  2.9973e-05,  4.6183e-03, -5.6785e-03, -2.4120e-03,\n",
       "         1.7624e-02, -1.7874e-02, -2.1978e-03,  1.4353e-02, -1.3785e-02,\n",
       "         7.8385e-03, -7.9804e-04,  2.1151e-03,  1.1846e-03,  7.3858e-03,\n",
       "         2.9314e-03,  5.5418e-03, -1.4171e-04,  9.6863e-03, -9.5188e-03,\n",
       "        -7.4935e-03,  1.3748e-02, -1.2730e-02,  3.0836e-03,  2.2549e-03,\n",
       "        -1.9140e-02,  8.4956e-03,  2.7597e-03, -3.7464e-03, -1.2739e-02,\n",
       "        -1.6030e-02,  2.1804e-03,  1.6407e-02, -1.6532e-02, -1.0374e-02,\n",
       "         1.0586e-03, -4.2329e-03, -2.1754e-03, -4.8001e-03, -1.3941e-02,\n",
       "         7.6111e-03, -1.5964e-03,  6.2338e-03,  2.0175e-03, -1.3063e-03,\n",
       "        -1.5633e-03, -7.8066e-03,  1.5028e-02, -1.3452e-02, -1.1645e-02,\n",
       "         6.0278e-03,  6.3402e-03,  1.1361e-02,  1.7134e-03, -1.3501e-02,\n",
       "        -2.5177e-03, -2.0796e-02, -6.0804e-03,  6.8649e-03, -2.6063e-03,\n",
       "        -6.5591e-03, -5.0696e-03,  6.9648e-03,  1.2599e-02, -1.4352e-02,\n",
       "        -7.6801e-03,  7.9338e-03,  1.4844e-02], device='cuda:0'), tensor([ 7.4911e-03, -2.3554e-03, -1.4010e-02, -4.6686e-03,  1.4032e-03,\n",
       "         5.4957e-03, -1.0603e-02, -2.5280e-02,  3.4719e-03, -7.5237e-04,\n",
       "         3.8322e-03,  1.1247e-02, -1.7810e-03,  8.7882e-03,  1.5047e-02,\n",
       "         1.8396e-03,  1.6218e-02,  1.8586e-02, -2.0089e-04,  2.2746e-03,\n",
       "         1.1581e-03,  1.9972e-03,  2.1548e-03, -1.1129e-02, -1.0188e-02,\n",
       "         9.8964e-03, -8.8957e-03,  9.9871e-03,  4.0207e-04,  5.8431e-03,\n",
       "        -1.0072e-02,  1.2123e-05,  2.6665e-02,  7.0996e-03, -1.4029e-03,\n",
       "         1.7801e-02,  2.9324e-03,  1.3290e-03, -8.6332e-03, -2.9419e-03,\n",
       "        -3.6932e-03, -1.4909e-02, -1.4437e-03, -7.6738e-03, -3.5595e-03,\n",
       "         1.4006e-02,  3.9929e-03,  5.0148e-03, -2.3464e-04,  2.1625e-03,\n",
       "         8.6696e-03,  6.9189e-03, -8.6931e-03, -6.5701e-03, -4.4596e-03,\n",
       "         6.9380e-03, -1.0551e-02, -9.6864e-04, -3.6784e-03,  8.2486e-03,\n",
       "         1.4071e-02, -5.4097e-03,  2.6769e-04, -2.6861e-03, -9.9703e-03,\n",
       "         1.7784e-02,  1.2626e-02, -1.2297e-02, -1.1781e-02, -2.2416e-03,\n",
       "        -4.8473e-03,  2.2725e-03,  2.9671e-03, -2.0307e-02, -1.0629e-02,\n",
       "         1.0830e-02, -9.4673e-03,  2.3727e-04, -2.6862e-03, -1.4152e-02,\n",
       "         3.8545e-03,  1.0696e-02,  1.0994e-02, -4.6456e-03,  2.3845e-03,\n",
       "         3.3724e-03,  8.8150e-03,  1.2022e-02,  2.4335e-03,  1.2544e-02,\n",
       "        -1.6871e-02,  2.6903e-03,  1.3640e-02,  4.2095e-03,  1.4513e-02,\n",
       "        -5.9570e-03, -3.8011e-03,  8.0338e-03, -1.9146e-02, -5.0398e-03,\n",
       "        -5.4538e-03,  8.2766e-03,  4.5155e-03,  1.0598e-02, -4.6855e-03,\n",
       "         9.6730e-06,  1.8410e-03, -1.7226e-02,  4.6016e-03,  6.2469e-03,\n",
       "        -6.5490e-03,  4.1643e-03, -7.9273e-03, -3.8148e-04, -1.0270e-02,\n",
       "         9.2832e-03, -1.2627e-02,  2.7657e-02,  6.3719e-03, -3.0763e-03,\n",
       "         3.1395e-02, -4.3448e-03,  1.1253e-02,  2.0339e-03,  1.9198e-03,\n",
       "         2.8513e-03,  9.5423e-03,  5.0937e-03], device='cuda:0'), tensor([[[[ 1.6809e-02,  6.3374e-03, -1.6922e-02],\n",
       "          [ 6.2290e-03, -5.0984e-03, -7.0073e-03],\n",
       "          [-2.8778e-03, -1.6389e-02, -1.4487e-02]],\n",
       "\n",
       "         [[-1.3434e-02, -3.1501e-03,  1.5353e-03],\n",
       "          [-3.9773e-04,  2.4086e-04,  2.1643e-03],\n",
       "          [-5.2449e-03, -5.6115e-03,  1.2180e-03]],\n",
       "\n",
       "         [[ 2.5545e-02,  7.7650e-03,  1.1907e-02],\n",
       "          [ 5.5160e-03, -1.0188e-03,  5.7235e-03],\n",
       "          [ 6.6515e-04, -4.7770e-03, -8.9719e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-1.6407e-03, -7.0491e-03,  2.4251e-03],\n",
       "          [ 5.9667e-03, -7.0549e-03,  3.5715e-03],\n",
       "          [ 7.9216e-03,  6.2673e-04, -5.3584e-03]],\n",
       "\n",
       "         [[ 1.3255e-02, -4.0266e-03, -5.0744e-03],\n",
       "          [ 1.9749e-03, -5.4961e-03,  2.3474e-03],\n",
       "          [-1.6719e-02,  2.4423e-03, -4.4627e-03]],\n",
       "\n",
       "         [[-3.0157e-04,  2.3096e-03,  7.6441e-03],\n",
       "          [-5.8074e-03, -8.2392e-03,  2.3636e-03],\n",
       "          [ 5.1458e-03,  1.1449e-03, -2.8392e-03]]],\n",
       "\n",
       "\n",
       "        [[[-3.6789e-03,  1.1882e-03,  1.5730e-02],\n",
       "          [ 9.9352e-03, -6.5866e-03, -1.0403e-02],\n",
       "          [-3.5273e-04, -1.3896e-02, -2.4648e-03]],\n",
       "\n",
       "         [[-1.4566e-02,  9.9501e-03, -1.5199e-02],\n",
       "          [ 9.8063e-03, -1.6533e-03, -2.3488e-02],\n",
       "          [ 7.7215e-03, -1.1481e-02,  2.3130e-03]],\n",
       "\n",
       "         [[ 3.3054e-02,  5.6440e-03,  1.1912e-02],\n",
       "          [-7.0152e-03,  6.2932e-03,  6.2039e-03],\n",
       "          [ 2.5902e-03, -7.8423e-04,  9.8637e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-3.4272e-03,  8.7857e-04,  8.5797e-03],\n",
       "          [-4.8250e-03,  8.5758e-03, -8.0885e-03],\n",
       "          [ 2.4019e-03, -1.7736e-03,  3.4549e-03]],\n",
       "\n",
       "         [[-1.8514e-05,  1.6292e-04, -3.1313e-03],\n",
       "          [ 9.5076e-03,  9.0901e-04,  3.6418e-03],\n",
       "          [-2.3204e-03,  8.8067e-03,  6.7686e-03]],\n",
       "\n",
       "         [[ 3.4225e-03, -1.6727e-02, -1.5495e-03],\n",
       "          [-9.4315e-03, -2.9477e-02, -1.0698e-02],\n",
       "          [-5.0787e-03,  9.7877e-03,  1.0580e-02]]],\n",
       "\n",
       "\n",
       "        [[[-5.7579e-03,  6.3547e-03, -6.9851e-03],\n",
       "          [ 1.4315e-02,  4.8954e-03, -6.9685e-03],\n",
       "          [ 5.2777e-03,  1.0036e-03, -8.1785e-03]],\n",
       "\n",
       "         [[ 1.3358e-02,  1.1225e-03,  8.4465e-03],\n",
       "          [-4.9828e-03,  8.1863e-03, -1.1652e-03],\n",
       "          [ 3.5696e-03,  1.3777e-02,  7.1040e-04]],\n",
       "\n",
       "         [[-5.0527e-03, -2.6745e-03,  5.0086e-03],\n",
       "          [-1.1968e-02, -2.2063e-03,  1.1754e-02],\n",
       "          [ 5.2018e-03, -1.3443e-02,  5.7236e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-3.7591e-03, -7.2776e-06, -9.6959e-03],\n",
       "          [-2.1476e-03, -1.0189e-02,  1.4970e-02],\n",
       "          [-1.8254e-02, -3.2486e-03,  1.5304e-03]],\n",
       "\n",
       "         [[ 2.8758e-03,  8.0196e-03, -6.8998e-03],\n",
       "          [-1.2104e-02, -1.2985e-02, -9.8791e-03],\n",
       "          [-2.2451e-04, -7.8034e-04, -1.9856e-03]],\n",
       "\n",
       "         [[-8.6974e-03, -1.1546e-02,  1.3930e-02],\n",
       "          [ 1.9991e-03,  8.4401e-03,  1.3216e-02],\n",
       "          [ 1.3594e-02,  1.2092e-02, -4.5951e-03]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[ 1.1677e-02, -8.5384e-03,  1.0106e-03],\n",
       "          [ 5.6785e-03, -1.6552e-02, -4.6779e-03],\n",
       "          [-1.9262e-02,  1.4645e-02, -3.5079e-03]],\n",
       "\n",
       "         [[-6.6137e-03,  8.7092e-03, -1.3775e-02],\n",
       "          [-2.1478e-02, -2.4142e-03,  6.8187e-03],\n",
       "          [-7.7634e-03,  4.9752e-03,  1.5417e-03]],\n",
       "\n",
       "         [[ 6.5402e-03, -3.4653e-03, -1.4019e-02],\n",
       "          [ 1.1889e-02,  8.4931e-03,  1.0575e-02],\n",
       "          [-2.1802e-03,  1.2386e-02, -2.2439e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-7.1629e-03,  1.4654e-02, -8.0292e-03],\n",
       "          [-1.1460e-02, -6.5899e-04, -6.4074e-04],\n",
       "          [ 5.6051e-03, -8.7256e-03,  7.2637e-04]],\n",
       "\n",
       "         [[-5.4670e-03,  1.1738e-02, -1.0923e-02],\n",
       "          [-4.0185e-03,  1.0068e-03,  3.1061e-03],\n",
       "          [ 7.6951e-03, -2.8141e-03,  1.0804e-02]],\n",
       "\n",
       "         [[ 5.8461e-03, -5.3205e-03,  7.1767e-03],\n",
       "          [-5.8582e-03,  1.4854e-02, -1.1317e-02],\n",
       "          [-9.8838e-03, -2.4893e-02, -9.9650e-03]]],\n",
       "\n",
       "\n",
       "        [[[-2.5481e-02,  1.2211e-02, -4.4419e-03],\n",
       "          [-7.0035e-03,  1.5148e-02, -1.2556e-02],\n",
       "          [ 2.0138e-03,  3.2014e-03,  1.2761e-02]],\n",
       "\n",
       "         [[ 3.3479e-03, -2.7810e-02,  1.5635e-02],\n",
       "          [-7.2673e-04,  9.5699e-04, -6.7234e-03],\n",
       "          [-1.3277e-02,  1.0763e-02, -1.5554e-02]],\n",
       "\n",
       "         [[-8.5191e-03,  3.0616e-03,  1.2309e-02],\n",
       "          [-1.8236e-03,  3.0318e-03, -2.6051e-05],\n",
       "          [-6.1147e-03, -1.1917e-02, -4.6534e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-4.4334e-04,  4.3731e-03,  1.5777e-02],\n",
       "          [-6.9145e-04,  4.5085e-04,  9.9937e-03],\n",
       "          [ 7.9918e-03,  1.1387e-02, -1.5048e-02]],\n",
       "\n",
       "         [[ 2.1176e-02, -1.6353e-03, -2.1651e-02],\n",
       "          [-1.6850e-04, -8.7753e-03,  7.9331e-03],\n",
       "          [ 9.1103e-04, -4.1749e-03, -1.3418e-02]],\n",
       "\n",
       "         [[ 8.0241e-03,  2.3065e-03,  1.6066e-03],\n",
       "          [-6.2283e-03,  5.6802e-03,  1.5616e-02],\n",
       "          [ 5.5521e-03, -4.2764e-03,  1.1556e-02]]],\n",
       "\n",
       "\n",
       "        [[[ 7.3386e-03,  7.7852e-03, -1.7314e-02],\n",
       "          [-9.0163e-03, -2.3290e-02, -9.8780e-03],\n",
       "          [ 7.4756e-03, -5.9730e-03,  2.7121e-03]],\n",
       "\n",
       "         [[ 2.3331e-02,  3.7442e-03,  2.3796e-02],\n",
       "          [-2.8365e-03, -8.5227e-03, -1.2114e-02],\n",
       "          [-2.0178e-03,  2.6514e-03,  3.8259e-04]],\n",
       "\n",
       "         [[-2.8487e-02, -7.7779e-04, -6.9247e-03],\n",
       "          [ 4.8605e-03,  1.2650e-03,  4.3203e-04],\n",
       "          [-3.4872e-03,  8.4482e-03, -4.7668e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 8.5234e-03,  1.3942e-02, -3.6157e-03],\n",
       "          [ 3.6448e-03, -7.8761e-03, -1.6300e-03],\n",
       "          [ 8.5387e-04, -5.3321e-03,  7.1473e-04]],\n",
       "\n",
       "         [[-4.7462e-03,  1.4346e-02,  5.9856e-03],\n",
       "          [-1.0053e-02, -2.2991e-03,  3.6760e-04],\n",
       "          [ 9.9100e-03,  8.0818e-03,  8.5789e-03]],\n",
       "\n",
       "         [[-4.5313e-04, -2.1521e-02,  5.2564e-03],\n",
       "          [-1.0697e-02,  8.5215e-04, -1.7514e-02],\n",
       "          [ 1.2633e-03, -5.7826e-03, -5.3498e-03]]]], device='cuda:0'), tensor([-8.4561e-03, -1.8983e-03,  9.1095e-03, -1.1802e-02,  5.4721e-03,\n",
       "        -9.0581e-03, -1.8346e-02,  1.8783e-03, -2.7146e-03, -1.0293e-02,\n",
       "         9.2409e-05, -7.8104e-03, -9.6967e-03, -7.5854e-03, -1.1157e-02,\n",
       "        -3.2569e-02, -9.8361e-03,  2.0973e-02,  6.4402e-03, -4.7181e-03,\n",
       "         1.6043e-02,  8.6480e-03, -9.2302e-04, -1.8296e-02, -3.3736e-04,\n",
       "         1.8056e-02,  1.2548e-02,  4.4769e-04, -3.6712e-03, -4.4702e-03,\n",
       "        -1.1902e-02, -3.0442e-03,  3.9977e-03, -6.0679e-03, -9.4975e-04,\n",
       "         2.1715e-03, -8.5370e-03,  8.7954e-03,  1.6554e-02, -3.8987e-03,\n",
       "         1.5739e-03, -1.2870e-02, -1.0177e-02, -1.3418e-02, -6.5548e-03,\n",
       "        -5.2796e-03,  8.5988e-04,  4.9851e-03,  7.7002e-03, -2.2475e-03,\n",
       "         8.8825e-03, -8.5044e-03, -1.1623e-02,  8.7238e-03, -1.7580e-03,\n",
       "        -1.8258e-02,  7.9104e-04,  1.2022e-02,  9.2046e-03,  2.7373e-03,\n",
       "         1.0169e-02, -8.3360e-04, -1.1600e-02, -2.2233e-03,  1.0453e-03,\n",
       "         9.9088e-04,  1.1068e-02,  1.5488e-02, -1.1163e-02, -6.0760e-03,\n",
       "        -5.6502e-03,  7.3845e-03, -1.7928e-02, -5.7022e-03,  1.2205e-02,\n",
       "        -1.6608e-02,  6.8237e-03, -1.0793e-02,  2.6552e-03,  3.9206e-03,\n",
       "        -7.9108e-03, -1.4243e-02, -8.8964e-03,  1.8808e-02, -4.4353e-03,\n",
       "         3.1583e-03, -1.0036e-04,  4.7841e-03,  1.3848e-02, -1.0919e-02,\n",
       "        -7.2323e-03, -1.9889e-05, -1.6003e-02,  1.2823e-02, -9.4467e-03,\n",
       "         1.0417e-02, -4.9260e-03, -2.6777e-02, -4.8789e-03,  2.9529e-02,\n",
       "        -5.2341e-05,  2.7923e-04, -4.1434e-03,  1.7859e-02,  4.3024e-03,\n",
       "         7.2932e-03, -6.3080e-03, -1.9617e-03,  9.8588e-03, -7.1716e-03,\n",
       "         1.2659e-02, -4.3614e-03,  3.8554e-03, -1.4451e-02, -5.0931e-03,\n",
       "         2.7836e-03,  2.3949e-03,  4.2822e-03, -4.8705e-03, -9.4161e-03,\n",
       "        -2.0527e-03, -7.5252e-03,  8.1139e-03, -5.6571e-03, -6.9478e-03,\n",
       "        -4.8245e-03, -2.3153e-03,  1.3924e-02], device='cuda:0'), tensor([ 0.0140, -0.0129,  0.0106,  0.0123, -0.0123, -0.0165, -0.0044,  0.0051,\n",
       "         0.0066, -0.0039, -0.0143, -0.0041,  0.0071, -0.0123, -0.0170, -0.0006,\n",
       "         0.0137,  0.0070, -0.0174,  0.0036, -0.0124,  0.0184, -0.0020,  0.0248,\n",
       "        -0.0138,  0.0025,  0.0077,  0.0067, -0.0161, -0.0188,  0.0017,  0.0006,\n",
       "        -0.0061,  0.0092,  0.0008,  0.0040,  0.0046,  0.0009, -0.0022,  0.0198,\n",
       "         0.0063,  0.0008,  0.0006, -0.0070,  0.0032, -0.0053,  0.0083,  0.0033,\n",
       "        -0.0029, -0.0086,  0.0034, -0.0027, -0.0130, -0.0154, -0.0040, -0.0053,\n",
       "         0.0140,  0.0175, -0.0023, -0.0139,  0.0228,  0.0131, -0.0091, -0.0101,\n",
       "         0.0039,  0.0071,  0.0023, -0.0046,  0.0010, -0.0003,  0.0081,  0.0159,\n",
       "        -0.0145, -0.0211,  0.0059,  0.0049, -0.0034, -0.0044, -0.0108, -0.0010,\n",
       "         0.0041, -0.0062,  0.0136, -0.0209,  0.0157,  0.0117, -0.0055, -0.0039,\n",
       "         0.0002,  0.0090,  0.0021, -0.0104,  0.0033,  0.0043, -0.0047,  0.0171,\n",
       "        -0.0020,  0.0038,  0.0021, -0.0042, -0.0104,  0.0013,  0.0032,  0.0074,\n",
       "         0.0172,  0.0001,  0.0040, -0.0040, -0.0064, -0.0072,  0.0309, -0.0122,\n",
       "        -0.0142, -0.0104, -0.0086,  0.0107, -0.0214,  0.0039, -0.0046, -0.0158,\n",
       "        -0.0023, -0.0070,  0.0139, -0.0094,  0.0161,  0.0049,  0.0005, -0.0021],\n",
       "       device='cuda:0'), tensor([[[[-2.6487e-02,  1.2374e-02,  1.4853e-03],\n",
       "          [-3.4284e-02, -1.0568e-02,  1.0670e-02],\n",
       "          [-2.0637e-04,  5.3608e-03, -9.3355e-03]],\n",
       "\n",
       "         [[-4.8567e-03, -6.8341e-04,  1.1652e-02],\n",
       "          [-5.5633e-04,  1.2353e-03,  1.1836e-02],\n",
       "          [ 5.8188e-03, -2.7736e-03,  1.0451e-02]],\n",
       "\n",
       "         [[-9.0617e-03,  1.0210e-02,  8.1501e-03],\n",
       "          [ 6.3748e-04,  5.9352e-04,  2.5182e-03],\n",
       "          [-9.5302e-03,  9.8641e-03, -1.7205e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-1.1462e-02,  1.3312e-02,  3.8263e-03],\n",
       "          [-2.7055e-02,  6.6971e-03,  1.0011e-02],\n",
       "          [ 4.8276e-03,  7.6279e-03, -3.4309e-03]],\n",
       "\n",
       "         [[ 1.3465e-03, -4.8427e-03, -1.5976e-02],\n",
       "          [-3.0603e-04,  5.4521e-03, -1.8648e-02],\n",
       "          [-3.3185e-03,  1.2582e-02,  2.6007e-03]],\n",
       "\n",
       "         [[ 1.3905e-02,  2.0792e-03,  9.7754e-04],\n",
       "          [-2.1372e-02, -2.8062e-03, -3.1899e-03],\n",
       "          [-1.4065e-03, -7.7102e-03,  9.0545e-04]]],\n",
       "\n",
       "\n",
       "        [[[-4.3255e-04, -1.3425e-02,  1.1368e-02],\n",
       "          [ 2.3928e-02,  2.6985e-03,  5.0235e-03],\n",
       "          [-8.7602e-03,  5.5005e-03, -4.6793e-03]],\n",
       "\n",
       "         [[ 1.4632e-02, -3.9223e-04, -3.0333e-03],\n",
       "          [-9.4736e-03,  1.2735e-02,  1.5671e-03],\n",
       "          [ 2.6017e-03,  8.3144e-03, -5.2111e-03]],\n",
       "\n",
       "         [[-1.2699e-03, -6.7064e-03, -2.1758e-02],\n",
       "          [ 1.4407e-02,  2.3530e-06,  7.0398e-03],\n",
       "          [-5.1596e-03,  1.5882e-03,  9.9260e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-6.9790e-03,  2.0122e-03,  2.1480e-02],\n",
       "          [-1.5579e-02, -3.9523e-03,  5.4524e-03],\n",
       "          [ 1.1143e-02, -5.0951e-03, -5.0594e-03]],\n",
       "\n",
       "         [[ 4.3435e-03, -1.8543e-04,  1.0944e-02],\n",
       "          [ 7.5935e-03,  1.8015e-03, -1.1093e-02],\n",
       "          [-1.2439e-02,  3.9855e-03,  2.8946e-03]],\n",
       "\n",
       "         [[ 1.0318e-02, -1.8701e-02,  5.3276e-03],\n",
       "          [ 1.5689e-02,  5.1307e-03, -5.0214e-03],\n",
       "          [ 1.7428e-02,  1.5118e-02,  2.4461e-03]]],\n",
       "\n",
       "\n",
       "        [[[ 6.5329e-03,  8.7591e-03, -1.1089e-02],\n",
       "          [-7.7638e-03,  6.9195e-03,  1.1113e-02],\n",
       "          [-1.6276e-03, -5.6781e-03, -4.5645e-04]],\n",
       "\n",
       "         [[-9.5013e-03,  4.4020e-03,  1.1920e-02],\n",
       "          [-8.1184e-03, -1.8169e-02, -4.7026e-03],\n",
       "          [-2.3529e-03, -1.8633e-02,  1.9288e-02]],\n",
       "\n",
       "         [[-3.6039e-03,  6.9887e-03,  1.1894e-02],\n",
       "          [ 9.2873e-03,  5.3753e-03, -6.7320e-03],\n",
       "          [-7.0204e-03,  6.0418e-03,  8.1827e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 1.5367e-03,  1.0043e-02, -1.6642e-02],\n",
       "          [-9.9763e-03,  1.4814e-02,  7.9823e-03],\n",
       "          [-1.0012e-03,  7.9694e-04,  4.7251e-03]],\n",
       "\n",
       "         [[ 1.1852e-02, -1.0731e-02, -1.0989e-02],\n",
       "          [-4.1708e-03,  1.7719e-04,  8.3402e-03],\n",
       "          [-1.9635e-02, -1.2715e-02, -2.7041e-03]],\n",
       "\n",
       "         [[ 2.3944e-03,  4.3405e-03, -1.0807e-02],\n",
       "          [-2.2725e-02,  8.7152e-03,  8.9277e-03],\n",
       "          [-5.1153e-03,  4.3266e-03, -1.1306e-02]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[ 1.1063e-02, -2.7454e-03,  5.1185e-03],\n",
       "          [-6.0207e-03,  1.0322e-02,  2.5557e-03],\n",
       "          [-3.1054e-03,  1.7228e-02,  4.8856e-03]],\n",
       "\n",
       "         [[ 4.6359e-03,  1.2441e-02, -1.4253e-02],\n",
       "          [ 8.3944e-03,  2.3999e-03,  2.2046e-03],\n",
       "          [ 1.6853e-02, -5.1494e-03,  4.8868e-03]],\n",
       "\n",
       "         [[-9.0502e-05,  4.2261e-04, -2.2995e-03],\n",
       "          [-3.5192e-03,  2.8136e-03,  2.2733e-03],\n",
       "          [ 1.6329e-02,  6.4181e-04, -5.9982e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 7.2897e-03,  2.7521e-03,  4.1608e-03],\n",
       "          [ 7.4164e-03,  6.6352e-03, -1.7487e-02],\n",
       "          [ 5.7455e-03, -3.7119e-03,  1.6922e-03]],\n",
       "\n",
       "         [[ 8.1939e-03,  5.8793e-03,  1.2421e-02],\n",
       "          [-7.4225e-03,  2.5214e-03, -2.3607e-04],\n",
       "          [-1.3984e-03, -9.0080e-03, -5.0099e-03]],\n",
       "\n",
       "         [[ 1.0398e-02,  9.7547e-03, -2.5793e-03],\n",
       "          [-2.6586e-02, -3.9003e-03,  1.6685e-02],\n",
       "          [-9.2724e-03,  2.6775e-02, -8.3582e-03]]],\n",
       "\n",
       "\n",
       "        [[[-2.0645e-03,  6.2843e-03,  2.9762e-03],\n",
       "          [ 9.0882e-03,  7.6540e-05,  7.5502e-03],\n",
       "          [-6.1529e-03, -3.4236e-03, -5.5487e-03]],\n",
       "\n",
       "         [[-1.8514e-02,  1.6964e-02, -4.1767e-04],\n",
       "          [-5.8369e-03,  6.6886e-03, -3.1139e-03],\n",
       "          [-3.7092e-03, -4.4990e-03, -8.5055e-03]],\n",
       "\n",
       "         [[-7.1063e-03, -1.5491e-02,  1.6106e-03],\n",
       "          [ 1.7615e-02,  1.3082e-02, -7.8513e-03],\n",
       "          [ 2.4282e-03, -1.4037e-02, -3.8865e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-1.5474e-03, -1.4868e-03,  5.8097e-03],\n",
       "          [ 3.8577e-03,  4.0156e-03, -1.4458e-02],\n",
       "          [ 1.2821e-02, -6.0912e-03,  1.2284e-02]],\n",
       "\n",
       "         [[-1.8152e-03,  4.0270e-03, -4.3676e-03],\n",
       "          [ 1.4220e-03, -6.3132e-03,  6.3524e-03],\n",
       "          [ 8.3615e-03, -1.0610e-02,  2.4149e-03]],\n",
       "\n",
       "         [[-6.7783e-03, -1.0806e-02,  7.4091e-03],\n",
       "          [ 1.1879e-02,  6.2809e-03,  6.4883e-03],\n",
       "          [ 1.1408e-02,  6.5917e-03, -1.5945e-03]]],\n",
       "\n",
       "\n",
       "        [[[ 1.1873e-02, -6.0944e-03, -4.7541e-04],\n",
       "          [ 1.6443e-02,  1.1924e-02,  4.8401e-03],\n",
       "          [ 6.4984e-03,  1.3116e-03, -8.4544e-03]],\n",
       "\n",
       "         [[-4.0768e-03,  1.2999e-02,  1.4036e-02],\n",
       "          [-1.2475e-02,  1.7671e-03, -2.2596e-03],\n",
       "          [ 1.2004e-02, -6.5432e-03, -1.0389e-03]],\n",
       "\n",
       "         [[ 4.3409e-03, -1.2929e-03, -9.4530e-03],\n",
       "          [-3.6243e-04,  1.0608e-02,  9.0951e-03],\n",
       "          [-9.3388e-03, -1.1983e-02,  2.0144e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-7.7408e-03,  1.0950e-03, -9.0011e-03],\n",
       "          [-9.7053e-03, -1.8705e-02,  2.1015e-03],\n",
       "          [-2.0070e-03,  4.8298e-03,  1.4862e-02]],\n",
       "\n",
       "         [[-4.5314e-03,  1.6779e-02, -2.1882e-02],\n",
       "          [-1.6036e-02, -7.5207e-03, -4.8938e-04],\n",
       "          [-4.8860e-03,  6.6362e-03, -1.3265e-02]],\n",
       "\n",
       "         [[-3.3660e-03, -6.1209e-03, -6.2445e-03],\n",
       "          [-1.1972e-03,  1.7113e-02,  9.1023e-04],\n",
       "          [-1.8800e-02,  3.3764e-03,  1.7308e-02]]]], device='cuda:0'), tensor([ 7.3159e-03,  1.8054e-02, -1.1743e-02, -7.3294e-04, -2.8947e-02,\n",
       "        -7.9185e-03,  8.5283e-03, -5.5450e-03,  5.0838e-03, -2.3426e-02,\n",
       "        -8.1167e-03, -5.5914e-03,  1.1145e-02, -1.1318e-02, -7.4845e-04,\n",
       "         1.5206e-02, -1.5235e-02, -3.9978e-03, -6.9188e-03,  2.8604e-03,\n",
       "        -8.1019e-04, -6.1963e-05,  2.4875e-03, -3.8755e-03, -9.2465e-03,\n",
       "         8.9106e-06,  3.9145e-03,  1.2997e-03,  1.6362e-03, -1.7391e-02,\n",
       "        -7.2788e-03, -2.9480e-03,  3.7767e-03, -2.8688e-03, -1.7242e-02,\n",
       "        -3.4992e-03, -9.8465e-03,  1.2309e-02,  2.8014e-03, -8.1852e-03,\n",
       "        -3.0596e-04, -1.5056e-02, -8.3713e-03, -2.0595e-02,  4.8517e-04,\n",
       "         1.4979e-03,  4.1511e-03, -1.7145e-02,  1.9608e-02,  2.2064e-03,\n",
       "         1.3460e-03, -1.6360e-02, -1.7116e-03,  7.2288e-03, -9.3008e-03,\n",
       "        -8.4024e-03, -6.1220e-03, -1.2727e-02,  4.0544e-03, -1.1932e-02,\n",
       "        -6.0245e-03, -6.7731e-03,  1.4718e-02,  1.0484e-02,  8.4845e-03,\n",
       "        -1.1892e-02, -6.7385e-03,  1.9497e-02,  3.8311e-03, -2.3804e-03,\n",
       "         7.4346e-03,  4.0284e-03, -1.1131e-02, -5.1584e-03, -4.3278e-03,\n",
       "         4.5516e-03, -1.3223e-02, -1.6049e-02, -1.8854e-02, -1.2888e-02,\n",
       "         1.7943e-03, -9.5190e-03,  6.4457e-04,  5.5135e-04,  1.5786e-03,\n",
       "        -5.9913e-03, -1.3870e-02,  1.7171e-02,  3.7952e-03,  1.9731e-02,\n",
       "        -2.4416e-02,  7.1693e-03, -6.7855e-03, -4.2733e-03, -6.5833e-03,\n",
       "        -1.6011e-02,  3.5673e-03,  5.4684e-03, -9.9811e-04,  3.1899e-03,\n",
       "         1.4183e-03, -2.8873e-03,  2.8439e-02,  5.7169e-03,  4.4740e-03,\n",
       "        -2.1034e-03, -2.9599e-03, -1.4996e-02,  1.0464e-02,  8.7631e-03,\n",
       "        -1.5028e-03,  1.4498e-03, -6.1023e-03,  2.2462e-04, -1.3573e-02,\n",
       "        -6.3004e-03, -2.5213e-03,  2.1366e-02,  2.5758e-03, -2.2450e-04,\n",
       "         2.2657e-03,  6.7844e-03, -2.1984e-03, -4.6841e-03, -6.0028e-03,\n",
       "        -1.5686e-02,  6.1441e-03,  3.2061e-03], device='cuda:0'), tensor([ 9.4565e-04, -5.6617e-04,  1.5541e-02,  1.2032e-02, -1.9474e-02,\n",
       "        -1.5304e-03,  1.4761e-02, -1.5342e-02, -2.1226e-03,  7.9845e-04,\n",
       "         2.8471e-02,  1.1664e-02,  1.9262e-02,  1.6950e-02,  1.1225e-02,\n",
       "         6.3674e-05, -9.6110e-03, -1.5752e-02,  4.8937e-03,  8.2237e-03,\n",
       "         6.0948e-03,  6.7038e-03, -7.2348e-03, -9.8474e-03, -8.3235e-03,\n",
       "         1.7639e-02, -2.3560e-03,  5.8355e-03,  1.8498e-03, -3.9505e-03,\n",
       "         2.7319e-03, -1.4517e-02,  8.8682e-03,  7.2091e-03,  2.1793e-03,\n",
       "         2.2374e-03, -8.9236e-03, -5.5884e-03, -2.6757e-03,  1.0206e-02,\n",
       "        -1.4453e-02,  1.0240e-02, -4.8542e-03, -6.3858e-05,  1.5423e-02,\n",
       "         4.7576e-03, -1.9063e-03,  3.2215e-03, -9.5679e-03,  1.2273e-02,\n",
       "        -1.5121e-02, -1.8482e-02,  4.4986e-03,  1.2039e-02, -6.9552e-03,\n",
       "        -3.0450e-03, -6.9290e-03,  9.9987e-03, -2.3314e-03, -1.0494e-02,\n",
       "        -3.3925e-03, -1.9284e-02, -7.7524e-04, -9.7399e-03, -1.3523e-02,\n",
       "         4.6952e-03, -3.9081e-03,  6.0470e-03,  1.7893e-02,  7.8619e-03,\n",
       "        -1.2129e-02,  1.0393e-02, -1.8609e-02,  4.0297e-03, -1.8226e-02,\n",
       "        -3.6343e-03, -8.4441e-03,  5.2208e-03,  6.6114e-03, -6.5427e-03,\n",
       "        -1.0884e-03,  9.5492e-03,  1.0178e-02, -9.4209e-03, -6.2232e-04,\n",
       "        -8.9852e-05, -2.9288e-03, -1.1285e-02, -6.3153e-03, -1.3057e-03,\n",
       "        -1.2106e-02, -8.9756e-03,  9.7568e-03, -8.2520e-03,  1.0506e-02,\n",
       "         4.3007e-03,  5.4488e-03, -1.5934e-04, -7.0417e-03,  1.8932e-02,\n",
       "        -1.2252e-03, -7.5994e-04,  1.2235e-03,  2.5766e-03, -6.5477e-03,\n",
       "        -6.9214e-03, -1.7215e-02,  1.8315e-02, -5.6758e-04,  3.9703e-03,\n",
       "        -1.2634e-02,  4.1874e-03, -1.1835e-03, -5.9096e-03,  1.0435e-02,\n",
       "         2.2912e-03,  1.0089e-02,  1.6233e-02,  7.1053e-03,  2.3876e-04,\n",
       "         2.2465e-02,  1.2526e-03,  5.7900e-04, -1.1804e-02, -2.8777e-02,\n",
       "        -1.4459e-02,  9.6522e-03,  1.6470e-02], device='cuda:0'), tensor([[[[ 8.2269e-03, -1.0911e-02,  1.7998e-02],\n",
       "          [-1.0220e-02, -1.3003e-02,  6.1485e-03],\n",
       "          [ 4.6859e-03,  6.0181e-03, -6.9551e-03]],\n",
       "\n",
       "         [[-3.2319e-03,  2.1271e-03,  1.5372e-03],\n",
       "          [ 4.7902e-03,  7.7713e-03, -5.3833e-03],\n",
       "          [ 3.1715e-03, -6.8458e-04,  5.4871e-03]],\n",
       "\n",
       "         [[-8.0984e-03,  1.2771e-02,  1.2615e-03],\n",
       "          [-7.0425e-03,  4.0224e-04,  9.7609e-03],\n",
       "          [-2.2000e-02,  9.0775e-03, -6.3761e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-7.4306e-03, -8.0782e-03, -1.4328e-03],\n",
       "          [-3.5980e-03,  8.8702e-03,  1.2557e-03],\n",
       "          [-1.5241e-02,  7.4981e-03,  2.3517e-02]],\n",
       "\n",
       "         [[ 3.1072e-03, -1.6649e-03,  4.8493e-04],\n",
       "          [-1.4346e-02, -1.1925e-02, -1.9753e-03],\n",
       "          [ 3.1033e-03, -2.2511e-03, -1.9073e-02]],\n",
       "\n",
       "         [[-1.5366e-03, -5.7063e-04, -1.5839e-02],\n",
       "          [ 3.2266e-03, -8.3088e-03,  1.0219e-02],\n",
       "          [ 1.7993e-03, -3.0310e-03, -7.0256e-03]]],\n",
       "\n",
       "\n",
       "        [[[ 1.9291e-02,  5.8221e-03, -2.1168e-03],\n",
       "          [ 8.1518e-03, -4.3434e-04,  2.7738e-03],\n",
       "          [-1.1603e-03,  1.0858e-02, -7.4974e-03]],\n",
       "\n",
       "         [[ 5.4235e-03,  1.1097e-02, -4.2996e-03],\n",
       "          [ 9.0473e-03,  1.8438e-03, -1.5473e-02],\n",
       "          [-8.4477e-03, -4.5656e-03, -1.4805e-02]],\n",
       "\n",
       "         [[ 8.6207e-03,  9.2267e-03, -9.4352e-05],\n",
       "          [ 7.6556e-03, -1.4169e-03,  4.2392e-03],\n",
       "          [ 5.0116e-03, -4.8205e-03,  9.1712e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 1.4041e-02,  7.6731e-03, -1.6237e-03],\n",
       "          [-9.3168e-03, -2.2373e-02, -1.4741e-03],\n",
       "          [ 1.2013e-02, -1.1882e-02, -5.7048e-04]],\n",
       "\n",
       "         [[-1.1925e-02,  2.7593e-03,  6.0370e-03],\n",
       "          [ 2.0076e-03, -1.5425e-02, -1.4957e-02],\n",
       "          [ 1.6963e-03, -8.4430e-03,  2.3435e-03]],\n",
       "\n",
       "         [[-2.0305e-03,  2.1500e-02, -1.8376e-02],\n",
       "          [-8.1319e-03,  4.1232e-03,  3.9923e-03],\n",
       "          [-1.7107e-02, -1.0995e-02, -7.1275e-03]]],\n",
       "\n",
       "\n",
       "        [[[-1.5422e-02,  2.9853e-02,  6.4460e-03],\n",
       "          [ 5.5707e-03,  9.5682e-03,  5.8451e-03],\n",
       "          [ 6.1367e-03,  8.5991e-03, -1.1608e-02]],\n",
       "\n",
       "         [[ 9.2546e-03, -4.5291e-03,  1.6585e-03],\n",
       "          [ 1.1704e-02, -1.6580e-02, -1.1800e-02],\n",
       "          [-1.3363e-02,  5.8974e-03, -1.0070e-02]],\n",
       "\n",
       "         [[-3.9717e-03,  1.6830e-02, -9.0924e-04],\n",
       "          [ 1.7336e-03,  1.0118e-02, -1.5835e-02],\n",
       "          [-8.0096e-03, -7.8460e-03,  4.3933e-04]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 2.2704e-02,  2.4820e-03,  5.3157e-03],\n",
       "          [ 1.8866e-02,  3.6398e-03, -6.6061e-03],\n",
       "          [ 1.0226e-02, -8.7905e-03,  1.2495e-03]],\n",
       "\n",
       "         [[-8.0914e-03,  1.3771e-02, -3.7346e-03],\n",
       "          [-7.1451e-04,  1.6349e-02, -4.0153e-03],\n",
       "          [-4.2601e-04,  1.8820e-02, -5.2134e-03]],\n",
       "\n",
       "         [[-4.6245e-03,  8.6744e-03,  6.1165e-04],\n",
       "          [-1.2436e-02,  5.0216e-03,  6.7324e-03],\n",
       "          [ 1.4470e-02, -1.0236e-02, -1.0288e-03]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[ 4.4043e-03,  1.5042e-02, -1.1285e-02],\n",
       "          [ 1.2109e-02,  1.2423e-03, -2.3901e-03],\n",
       "          [-7.0266e-03,  1.0217e-02, -6.4349e-03]],\n",
       "\n",
       "         [[ 3.9758e-03, -1.2222e-02,  1.9334e-02],\n",
       "          [-1.5131e-03, -9.5143e-03, -1.3808e-02],\n",
       "          [ 4.0790e-03, -2.2294e-02,  1.9070e-02]],\n",
       "\n",
       "         [[-1.3806e-02,  1.1382e-02, -8.3251e-03],\n",
       "          [ 1.8742e-02, -1.3174e-02, -7.6689e-03],\n",
       "          [-1.3580e-02, -1.8077e-02,  1.2516e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 8.7430e-03,  2.8414e-03,  1.9684e-04],\n",
       "          [-7.0392e-03,  1.2379e-02,  3.1686e-03],\n",
       "          [ 5.0785e-03,  1.6441e-02,  2.1048e-02]],\n",
       "\n",
       "         [[ 8.3176e-03,  5.0487e-03,  1.4453e-02],\n",
       "          [-3.1964e-03, -1.9496e-03, -3.4543e-03],\n",
       "          [ 1.6137e-03,  8.5544e-03, -2.4333e-02]],\n",
       "\n",
       "         [[-7.0994e-03,  1.4007e-02, -4.6164e-03],\n",
       "          [-3.3924e-03,  1.3544e-02, -4.3296e-04],\n",
       "          [-1.4132e-02,  6.8349e-03, -1.1542e-02]]],\n",
       "\n",
       "\n",
       "        [[[-4.2998e-03,  3.9947e-03, -1.2417e-02],\n",
       "          [-1.1411e-02,  1.1342e-02,  7.7628e-03],\n",
       "          [ 5.4628e-04, -1.2710e-02,  9.2912e-04]],\n",
       "\n",
       "         [[-5.8187e-03,  1.3852e-02,  2.4198e-03],\n",
       "          [ 5.5005e-03, -2.2841e-03,  4.6877e-03],\n",
       "          [ 1.5635e-02,  1.8159e-03,  6.4867e-03]],\n",
       "\n",
       "         [[-1.4497e-02,  7.5289e-03, -1.1705e-02],\n",
       "          [ 1.7006e-03, -1.7136e-02,  6.8234e-03],\n",
       "          [ 1.2737e-03, -5.7320e-03,  1.0197e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 6.8707e-03,  1.3593e-02,  1.2206e-02],\n",
       "          [ 7.8577e-04,  1.0300e-02, -3.9987e-04],\n",
       "          [ 7.4049e-03,  8.6831e-03,  9.3592e-04]],\n",
       "\n",
       "         [[ 1.8538e-02, -3.1514e-03,  3.7160e-03],\n",
       "          [-9.6614e-03,  1.3702e-02, -1.8240e-03],\n",
       "          [-5.8125e-04, -1.4728e-02, -1.7707e-03]],\n",
       "\n",
       "         [[ 4.3250e-04,  2.7091e-03, -8.3458e-03],\n",
       "          [ 7.3594e-03,  1.1965e-02,  2.7375e-03],\n",
       "          [-1.6860e-02,  6.3431e-03, -1.9403e-02]]],\n",
       "\n",
       "\n",
       "        [[[ 4.8669e-03,  4.4956e-03,  6.7448e-04],\n",
       "          [-2.6704e-02, -5.8250e-04, -9.8249e-03],\n",
       "          [ 2.0741e-02,  4.7623e-03,  1.0033e-02]],\n",
       "\n",
       "         [[-1.2915e-02, -2.9690e-03, -2.7111e-04],\n",
       "          [ 8.0626e-03,  2.9953e-03, -1.7028e-03],\n",
       "          [-7.3534e-03, -5.0811e-03,  5.9948e-03]],\n",
       "\n",
       "         [[ 2.1775e-03,  1.4596e-02,  6.9896e-04],\n",
       "          [ 9.3966e-03,  6.3760e-03, -1.5157e-04],\n",
       "          [-7.5585e-04,  5.4806e-03, -1.0166e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-1.4730e-03,  1.4034e-02,  1.4700e-03],\n",
       "          [-1.0708e-02, -3.0343e-04, -6.1710e-03],\n",
       "          [ 2.0876e-02,  1.8911e-03, -5.1318e-03]],\n",
       "\n",
       "         [[-5.4467e-03, -1.3412e-02, -2.2881e-03],\n",
       "          [ 7.1934e-03,  1.4704e-02,  1.7195e-02],\n",
       "          [ 9.5656e-03,  1.2369e-02,  6.9133e-03]],\n",
       "\n",
       "         [[-2.2635e-04, -2.9383e-03, -1.2284e-03],\n",
       "          [-2.0430e-02,  1.0508e-02, -9.4942e-03],\n",
       "          [-2.1485e-02,  1.2181e-03, -1.5204e-02]]]], device='cuda:0'), tensor([-1.8585e-03, -4.3343e-03,  3.2631e-02, -1.1699e-02,  8.1737e-03,\n",
       "        -9.7197e-03,  1.3633e-02,  4.2201e-03, -7.1064e-03,  5.4687e-03,\n",
       "        -1.9368e-02,  3.0316e-03,  2.7406e-03, -9.3627e-04,  7.5178e-03,\n",
       "        -1.7163e-02,  1.7604e-02,  3.8370e-03, -7.1957e-03, -5.0017e-03,\n",
       "         8.6944e-03, -9.8844e-03,  1.1115e-02, -1.6036e-02, -5.1930e-03,\n",
       "         1.0203e-04,  1.3365e-02,  5.1468e-03,  8.5297e-03,  1.0533e-02,\n",
       "        -3.6269e-03,  2.8927e-03,  7.1858e-03, -6.6099e-03,  9.1883e-03,\n",
       "        -2.8466e-03, -6.8521e-04,  7.6596e-03,  6.1509e-03,  1.1471e-02,\n",
       "        -5.4933e-03, -1.4769e-03, -9.7059e-03,  8.7362e-03, -1.5475e-02,\n",
       "         1.6853e-02, -1.1413e-02, -1.0143e-02,  1.2971e-02, -7.8012e-03,\n",
       "         1.7233e-02,  4.5978e-03,  8.5367e-03,  1.5326e-02,  6.3290e-03,\n",
       "        -3.4864e-03,  5.3977e-03,  1.5405e-02, -1.1294e-02, -1.2407e-02,\n",
       "         6.4530e-03,  3.4566e-03, -2.9299e-03, -9.4394e-03, -1.0926e-02,\n",
       "         2.3435e-03,  9.8808e-03, -7.6553e-03,  5.5330e-03, -9.0084e-04,\n",
       "        -3.3800e-03, -8.5943e-03,  1.6931e-02, -7.6701e-03, -8.4853e-03,\n",
       "        -3.8078e-03, -1.3314e-03, -2.0906e-03,  6.2180e-03, -5.8336e-03,\n",
       "        -1.4896e-04,  5.1812e-03, -9.5530e-03, -3.5969e-03, -7.1016e-04,\n",
       "         2.3095e-02, -5.7157e-03,  5.9598e-03, -4.3262e-03, -7.8972e-03,\n",
       "        -4.7556e-03,  1.5269e-02,  1.2058e-02, -5.9058e-04,  4.6657e-04,\n",
       "         6.5216e-03, -1.7441e-03,  4.6822e-03,  2.7896e-02,  1.4494e-02,\n",
       "        -1.8592e-03, -2.7952e-03, -3.1282e-03,  1.2424e-02,  5.3594e-03,\n",
       "         9.9051e-03, -8.3684e-03,  1.7181e-02, -2.3660e-03,  5.7109e-03,\n",
       "         3.3598e-04,  3.6014e-05, -2.3045e-03, -2.2235e-03,  1.6262e-03,\n",
       "        -2.0942e-02,  9.6018e-03, -2.1048e-02, -6.0141e-03, -5.6825e-03,\n",
       "        -7.5447e-03, -1.9699e-02, -4.1626e-03,  6.6602e-04,  1.1916e-02,\n",
       "         6.9341e-03,  1.5607e-02,  7.2930e-03,  5.4982e-03,  4.5959e-03,\n",
       "        -1.1676e-03,  2.3161e-03,  2.7463e-03,  4.2520e-03, -1.3270e-02,\n",
       "         7.8885e-03, -2.0514e-02, -8.0670e-03, -2.3216e-02, -8.0703e-03,\n",
       "         8.2348e-04,  1.0074e-02,  2.5621e-02, -1.0094e-03,  9.0900e-03,\n",
       "         4.7179e-03, -1.0090e-02, -3.4155e-03, -6.8119e-03,  4.7521e-03,\n",
       "         1.9425e-02, -3.3413e-04, -6.2733e-03, -3.1455e-03, -4.3330e-03,\n",
       "        -1.2715e-02, -1.0268e-02, -1.4120e-02,  2.4234e-03, -1.2378e-02,\n",
       "        -1.2408e-02, -9.7705e-03, -1.5807e-02, -3.8453e-03,  1.1401e-02,\n",
       "         9.4934e-03, -7.1613e-03, -4.2412e-04, -1.1634e-02, -1.7496e-02,\n",
       "        -3.5682e-03,  1.8486e-02,  6.0099e-03,  8.2029e-03, -1.4064e-02,\n",
       "        -1.4684e-03, -4.6680e-03,  9.9620e-03,  1.1462e-02, -2.7483e-03,\n",
       "         1.6180e-03, -1.2439e-03, -2.0798e-02, -3.6341e-03, -3.2802e-03,\n",
       "         1.4438e-02, -1.3926e-03, -1.0662e-02,  1.9084e-02, -4.7021e-03,\n",
       "         2.3325e-04, -4.5387e-03,  2.2790e-03,  1.1197e-02,  2.0763e-02,\n",
       "        -4.0673e-03, -1.0657e-02, -1.8388e-02,  1.2875e-02,  1.4317e-02,\n",
       "        -1.5678e-03, -2.6635e-02, -1.6386e-03, -4.5955e-03, -2.0861e-02,\n",
       "        -4.8405e-03,  3.7788e-03,  4.3083e-04,  2.2224e-03, -2.4708e-03,\n",
       "         5.9286e-03,  9.4782e-03,  1.6494e-02, -3.9332e-04,  1.6008e-02,\n",
       "        -3.4753e-03, -2.9358e-03, -4.9611e-03,  5.2435e-03,  1.2959e-02,\n",
       "         5.2878e-03, -2.2151e-03, -9.8334e-03,  5.8758e-03, -1.5789e-03,\n",
       "         5.5142e-03, -5.0238e-04,  1.3122e-02, -2.4977e-03, -4.5443e-03,\n",
       "        -4.6356e-03,  3.3954e-03, -1.4413e-02, -8.3288e-03, -2.8632e-03,\n",
       "        -4.2855e-03, -6.4041e-03, -5.1457e-03,  1.1082e-02,  1.4369e-02,\n",
       "        -4.8952e-03,  8.5306e-04,  1.8437e-03,  1.3530e-03,  1.4762e-02,\n",
       "         6.7255e-03,  6.3584e-03, -2.3087e-03,  1.5869e-02,  1.1455e-02,\n",
       "         9.5241e-03, -1.7940e-03,  9.6086e-04, -1.3480e-02, -9.6467e-03,\n",
       "         1.7240e-03], device='cuda:0'), tensor([ 9.3668e-03,  7.1832e-04, -3.7813e-03,  1.1992e-02,  1.9857e-03,\n",
       "        -1.2192e-02, -1.0523e-02,  5.9977e-05, -2.7649e-03, -9.8924e-03,\n",
       "        -1.2245e-02,  1.3594e-02,  9.4389e-03, -4.8068e-03, -3.7936e-03,\n",
       "        -8.1909e-03, -4.2658e-03, -4.0240e-03,  5.8910e-03, -5.5547e-03,\n",
       "         8.8220e-03, -2.8043e-03,  1.3121e-02,  1.4949e-02, -2.5135e-02,\n",
       "        -1.9608e-03,  2.2480e-02, -9.4046e-04,  8.2385e-03,  4.9946e-03,\n",
       "        -7.9748e-03, -8.0581e-03,  1.6747e-02, -1.3500e-02,  1.2678e-02,\n",
       "         2.6758e-03,  8.7302e-03,  1.6821e-02,  2.6777e-02, -1.0108e-02,\n",
       "         3.6839e-03, -7.0380e-03,  1.2016e-02, -5.4759e-03, -2.1501e-03,\n",
       "         3.2912e-03,  1.1737e-02, -9.3349e-03,  4.2352e-03,  2.5682e-03,\n",
       "         1.5665e-02, -5.9777e-03, -2.0893e-02,  1.7438e-03, -1.8337e-02,\n",
       "         1.2117e-02,  1.4361e-02, -4.0497e-03, -1.2575e-03,  1.6755e-02,\n",
       "         1.6751e-03,  4.7496e-04, -3.4129e-03,  3.1816e-04, -8.7827e-06,\n",
       "         1.0027e-02, -6.4966e-03,  3.5783e-03, -3.9743e-03,  1.8221e-02,\n",
       "         2.4856e-02, -5.4643e-03, -1.6475e-02, -2.2327e-02, -2.8297e-03,\n",
       "         1.9804e-02,  8.7575e-03,  1.8578e-02, -7.1923e-03,  2.6986e-03,\n",
       "        -3.7632e-03,  3.0376e-03, -9.6120e-03, -2.1691e-02, -6.9328e-03,\n",
       "        -2.6716e-03, -2.4628e-03, -6.1374e-04,  1.1060e-02,  1.0910e-02,\n",
       "        -7.0888e-03, -1.5548e-03, -7.9271e-03, -6.4624e-03, -2.3291e-03,\n",
       "        -2.2186e-02,  1.9785e-02, -2.5865e-03,  2.7068e-04,  5.7853e-03,\n",
       "         5.1569e-04, -1.5412e-02,  7.3182e-03,  8.4224e-04, -1.8990e-02,\n",
       "         2.7569e-02, -4.4653e-03, -4.8304e-03, -4.5554e-03, -2.7696e-03,\n",
       "         1.1161e-02,  6.0880e-03, -2.8062e-03, -1.3276e-02, -3.5481e-03,\n",
       "        -7.1995e-04, -4.1697e-03, -1.4913e-02, -1.2873e-03, -3.1861e-03,\n",
       "        -5.0211e-03,  6.0808e-03, -8.3085e-03, -1.2428e-02,  1.1520e-02,\n",
       "         1.3843e-03,  1.4219e-02,  1.1373e-02, -1.2440e-02,  9.5207e-03,\n",
       "         5.7395e-03,  9.1966e-03, -4.4490e-04,  1.1491e-02,  1.5070e-02,\n",
       "         9.3273e-03, -1.7880e-02, -6.4285e-03,  1.6959e-02,  1.1182e-02,\n",
       "        -5.5337e-04, -2.9196e-03,  4.9454e-04,  1.6944e-03,  1.9150e-03,\n",
       "         1.8846e-02, -8.6008e-03,  7.1450e-03,  6.3474e-03,  2.2024e-03,\n",
       "         1.0662e-02,  6.4932e-03,  9.7802e-03, -1.8751e-02, -1.5124e-02,\n",
       "        -1.0493e-02,  6.7471e-03,  2.9268e-03, -4.1635e-03, -7.9160e-03,\n",
       "        -6.9796e-03,  1.1138e-02, -1.0274e-02,  1.1133e-02,  4.9109e-03,\n",
       "        -1.4494e-03,  4.6995e-03, -1.7656e-02,  1.2188e-02, -1.2891e-02,\n",
       "        -8.8227e-03, -9.8803e-03,  4.7119e-03, -1.4251e-02,  2.8713e-03,\n",
       "        -1.8075e-03, -7.3574e-03,  4.6558e-03, -3.4261e-03, -1.7766e-02,\n",
       "        -3.4120e-03, -5.2363e-03,  1.8998e-02,  8.9159e-03,  2.3214e-02,\n",
       "         2.8377e-04, -1.4721e-02,  8.2202e-03,  8.1126e-03, -5.8517e-03,\n",
       "        -1.0487e-02,  3.6083e-03, -1.3664e-02,  9.1649e-03, -5.9532e-03,\n",
       "         1.0060e-02, -1.0507e-02,  9.0885e-03,  5.5611e-03,  1.2434e-02,\n",
       "        -6.8478e-03,  1.7614e-02, -1.8778e-02, -1.5515e-02,  1.2613e-02,\n",
       "         1.1079e-05,  9.9459e-03, -1.1307e-04, -2.0954e-02, -2.1240e-02,\n",
       "        -3.2908e-04,  4.6926e-03,  1.3857e-02, -3.8439e-03, -6.4686e-03,\n",
       "         1.6696e-02, -5.8175e-04, -1.5978e-02,  1.7665e-02,  6.7303e-03,\n",
       "        -2.9947e-03,  2.8762e-03,  1.1354e-02,  5.6433e-03,  9.1541e-04,\n",
       "         1.6940e-02,  2.6344e-02, -1.0701e-02, -1.4498e-02, -8.0113e-03,\n",
       "        -1.4754e-02, -3.8610e-03, -1.0916e-02, -1.2582e-02,  9.7506e-03,\n",
       "         5.9018e-03,  1.1596e-02,  1.5201e-02, -3.1268e-03, -1.5757e-03,\n",
       "        -2.4559e-04,  9.0305e-03, -6.4612e-03,  1.6019e-03, -4.3539e-03,\n",
       "         7.2330e-03,  3.8163e-03, -4.5202e-03, -4.4722e-03, -7.8138e-03,\n",
       "        -5.2125e-03,  3.4447e-03,  5.0246e-03,  7.7985e-03, -1.3654e-02,\n",
       "        -7.3533e-03], device='cuda:0'), tensor([[[[-3.5441e-03, -1.4868e-02,  5.3632e-03],\n",
       "          [-2.4169e-03,  8.5065e-03,  7.5044e-03],\n",
       "          [-8.0467e-03,  1.0044e-02, -1.3858e-02]],\n",
       "\n",
       "         [[-1.4126e-02,  1.1885e-02, -7.6236e-03],\n",
       "          [ 1.4936e-02,  3.8221e-03,  1.0819e-02],\n",
       "          [ 1.4930e-03,  1.0341e-02,  3.9865e-03]],\n",
       "\n",
       "         [[-1.5334e-02, -6.9405e-04,  7.2754e-04],\n",
       "          [ 1.0792e-02, -4.4818e-03,  6.5917e-03],\n",
       "          [-2.7598e-02,  1.7902e-02,  5.2722e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 4.9189e-03, -7.2674e-03,  3.3939e-03],\n",
       "          [-2.1069e-02, -2.2740e-03, -1.0565e-02],\n",
       "          [ 3.0642e-03, -1.3922e-02, -1.4962e-02]],\n",
       "\n",
       "         [[-1.4003e-02, -1.1256e-02, -7.4426e-04],\n",
       "          [-5.0777e-03,  2.2920e-02,  8.4414e-03],\n",
       "          [ 1.2045e-02, -5.0055e-03, -2.2057e-02]],\n",
       "\n",
       "         [[-1.8834e-02,  1.0870e-02, -9.9726e-03],\n",
       "          [-4.1520e-03, -9.2572e-03,  4.2798e-03],\n",
       "          [ 8.1785e-03,  6.6474e-03, -2.6133e-03]]],\n",
       "\n",
       "\n",
       "        [[[ 7.6582e-03, -1.2762e-03,  1.3606e-03],\n",
       "          [-5.9148e-04, -1.7642e-02,  4.6205e-03],\n",
       "          [ 5.0601e-03,  5.0307e-03,  8.1524e-04]],\n",
       "\n",
       "         [[-3.4594e-03, -3.7227e-04,  1.1063e-02],\n",
       "          [-1.8382e-02,  5.0739e-03, -6.4185e-03],\n",
       "          [ 1.0854e-02,  4.3438e-03, -2.0000e-03]],\n",
       "\n",
       "         [[ 3.1082e-03, -2.2086e-03, -4.9479e-03],\n",
       "          [ 1.2607e-02, -8.2764e-03,  3.4221e-02],\n",
       "          [-3.3062e-04, -6.9344e-03,  1.2231e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 1.0235e-02, -9.1081e-03, -9.8874e-03],\n",
       "          [ 7.7790e-05,  7.3378e-03, -4.5344e-03],\n",
       "          [ 1.1651e-02,  5.0930e-03, -2.7727e-03]],\n",
       "\n",
       "         [[ 1.4839e-02, -1.1761e-02, -5.1813e-03],\n",
       "          [ 1.2805e-02,  1.1022e-03,  5.0755e-03],\n",
       "          [ 8.8626e-03, -6.8543e-03, -1.1920e-02]],\n",
       "\n",
       "         [[-1.2001e-02, -9.2327e-03,  1.3954e-02],\n",
       "          [ 6.6521e-03, -1.2554e-02, -3.8279e-04],\n",
       "          [ 5.4531e-03,  2.4483e-03,  6.1010e-03]]],\n",
       "\n",
       "\n",
       "        [[[ 9.5447e-03,  2.3109e-03, -1.0479e-02],\n",
       "          [ 1.9227e-03,  5.7942e-03,  6.4735e-03],\n",
       "          [ 7.0494e-04,  2.6417e-03, -8.0240e-03]],\n",
       "\n",
       "         [[ 3.5167e-03, -1.8321e-02, -4.1987e-03],\n",
       "          [-6.9768e-04, -1.4218e-02, -4.1771e-03],\n",
       "          [ 7.0057e-03, -7.2206e-03,  1.4746e-02]],\n",
       "\n",
       "         [[ 6.4413e-03,  1.4939e-02,  1.1595e-02],\n",
       "          [-2.5003e-03, -4.8342e-03, -1.8564e-02],\n",
       "          [ 1.9011e-02,  1.0421e-02, -2.8093e-04]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 6.4217e-03,  1.1544e-02, -1.0867e-02],\n",
       "          [-4.9651e-03, -6.1086e-03,  9.7897e-03],\n",
       "          [-2.2560e-02,  1.6925e-02, -4.4134e-03]],\n",
       "\n",
       "         [[ 1.1629e-02,  5.3765e-04, -3.8082e-03],\n",
       "          [ 7.9408e-03,  2.9676e-03,  1.1974e-02],\n",
       "          [ 8.8400e-03, -1.4132e-02, -5.7103e-03]],\n",
       "\n",
       "         [[-3.2684e-02, -2.2683e-03,  4.3107e-03],\n",
       "          [ 4.8888e-03, -3.7213e-03,  4.0552e-03],\n",
       "          [ 1.8194e-03,  3.5394e-03, -2.6965e-02]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[-2.9818e-03, -1.6207e-03,  1.0798e-02],\n",
       "          [-9.5618e-04,  4.6183e-03,  4.4481e-03],\n",
       "          [-2.9532e-02, -1.1497e-02, -4.3151e-03]],\n",
       "\n",
       "         [[ 2.4613e-03,  5.5156e-03,  5.6352e-04],\n",
       "          [ 4.0661e-03,  1.0254e-02, -1.5508e-03],\n",
       "          [ 9.8981e-03,  1.3626e-02,  2.9827e-03]],\n",
       "\n",
       "         [[ 3.2787e-03, -1.2217e-02,  4.4697e-04],\n",
       "          [-1.2623e-02,  7.6263e-03, -1.0976e-02],\n",
       "          [ 1.2789e-02,  2.2202e-02,  1.9291e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-1.6060e-03,  8.0959e-03,  1.6620e-03],\n",
       "          [-6.3908e-03, -9.5019e-04,  3.9186e-03],\n",
       "          [ 5.0046e-04, -8.2713e-03, -1.0016e-02]],\n",
       "\n",
       "         [[ 9.2049e-04,  1.8632e-02, -7.9445e-03],\n",
       "          [-6.1734e-04, -6.9916e-03, -4.5413e-03],\n",
       "          [-9.7601e-03, -2.0896e-03, -4.7340e-03]],\n",
       "\n",
       "         [[-1.2643e-02, -1.5007e-02, -9.1224e-03],\n",
       "          [-6.4259e-03,  5.5001e-03, -1.3151e-04],\n",
       "          [ 8.2587e-03,  1.5158e-02, -3.4610e-03]]],\n",
       "\n",
       "\n",
       "        [[[ 1.3456e-02, -8.6210e-03, -7.4097e-04],\n",
       "          [ 6.7862e-03, -8.3188e-03, -1.9043e-02],\n",
       "          [-4.8128e-03, -4.6926e-04, -1.2967e-02]],\n",
       "\n",
       "         [[ 9.5841e-03,  2.4896e-02, -1.4719e-02],\n",
       "          [ 3.4251e-03,  4.6683e-03, -6.5988e-03],\n",
       "          [ 1.3604e-03, -5.3507e-03,  9.2116e-03]],\n",
       "\n",
       "         [[-3.7153e-03,  1.3774e-02,  3.5959e-03],\n",
       "          [-1.2439e-02, -7.9521e-03,  2.1337e-03],\n",
       "          [-7.4189e-03,  1.6446e-02,  1.0043e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-7.9358e-03, -1.0769e-02, -1.0757e-02],\n",
       "          [-1.2823e-02, -8.0387e-03, -2.6423e-03],\n",
       "          [ 4.4455e-03, -8.5735e-03,  5.7204e-03]],\n",
       "\n",
       "         [[-5.2608e-04, -1.4138e-02,  1.8339e-02],\n",
       "          [ 6.1028e-03, -1.9368e-02, -1.5806e-02],\n",
       "          [ 8.0598e-03,  1.5613e-02, -1.4966e-02]],\n",
       "\n",
       "         [[ 1.5141e-02,  2.5664e-02, -1.4644e-02],\n",
       "          [-1.9080e-02,  1.0292e-03,  1.2883e-02],\n",
       "          [ 1.7073e-02, -4.7381e-03,  5.0343e-03]]],\n",
       "\n",
       "\n",
       "        [[[ 2.4560e-03, -1.3053e-02, -1.3527e-02],\n",
       "          [ 8.5509e-04,  3.7092e-03, -1.6023e-02],\n",
       "          [-8.2924e-03,  3.0524e-03, -5.3970e-03]],\n",
       "\n",
       "         [[ 2.1443e-02, -1.2514e-03,  4.4805e-03],\n",
       "          [-1.1230e-03, -2.1718e-03, -3.0731e-02],\n",
       "          [ 6.8533e-03, -6.9868e-03,  9.8327e-04]],\n",
       "\n",
       "         [[ 1.9808e-02,  1.5730e-02, -1.6130e-03],\n",
       "          [-9.4940e-03,  1.1497e-02,  6.4641e-03],\n",
       "          [-4.6573e-03, -2.0961e-02, -1.1040e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 9.7069e-03, -3.5257e-03, -3.0925e-03],\n",
       "          [ 3.1125e-03, -1.5285e-03, -5.0853e-03],\n",
       "          [ 9.5365e-03,  7.3046e-03, -7.1882e-03]],\n",
       "\n",
       "         [[-4.5853e-03, -6.5040e-03,  8.5126e-03],\n",
       "          [ 3.9917e-04,  1.1617e-02,  1.0571e-02],\n",
       "          [ 1.1812e-02, -6.5485e-03, -2.9340e-03]],\n",
       "\n",
       "         [[-8.4762e-04,  1.0010e-02,  1.0884e-03],\n",
       "          [-5.2424e-03, -7.7940e-03, -8.7257e-03],\n",
       "          [ 4.8590e-03,  1.3673e-03,  6.8982e-04]]]], device='cuda:0'), tensor([-4.2450e-03, -9.1937e-03, -7.2385e-03, -1.8956e-02,  2.0634e-03,\n",
       "        -5.1229e-03, -1.1272e-02,  2.2110e-03,  9.2103e-03,  2.6048e-03,\n",
       "         2.0170e-03, -3.0570e-03, -9.4844e-03, -9.4416e-03,  2.2529e-03,\n",
       "         3.7075e-04,  1.4607e-02, -5.7578e-03,  8.6353e-03,  1.0749e-02,\n",
       "        -7.0239e-03,  4.7971e-03, -1.0002e-02, -2.2456e-02, -3.7985e-03,\n",
       "        -9.5646e-03, -8.3303e-03, -3.8885e-03, -1.0803e-02,  1.2717e-03,\n",
       "        -2.4929e-03,  1.9001e-02,  1.4901e-02,  1.3355e-03,  2.3429e-04,\n",
       "        -1.1744e-02, -3.3910e-03, -1.2955e-02,  7.4414e-03,  8.7791e-03,\n",
       "         8.5780e-03, -3.6369e-03,  1.8207e-02,  8.3889e-03,  1.3983e-02,\n",
       "         1.5575e-03, -3.6421e-03, -2.7282e-03, -6.0497e-05,  2.7792e-03,\n",
       "        -2.3264e-02, -9.0778e-03, -1.3778e-02, -6.6673e-03, -6.3253e-03,\n",
       "         5.8239e-03, -9.0809e-03, -5.4554e-03, -2.2865e-03, -2.7388e-04,\n",
       "         3.6914e-03, -2.1327e-02,  2.0580e-04,  2.4740e-03,  1.3387e-02,\n",
       "        -2.0813e-02,  1.3285e-03, -5.7007e-04, -6.9051e-03,  5.9740e-03,\n",
       "         4.2015e-03,  3.4413e-04,  3.9400e-03,  7.2206e-03, -1.7924e-02,\n",
       "        -1.0731e-02, -1.3445e-02, -6.2540e-03, -1.5443e-03, -5.2260e-03,\n",
       "         1.0182e-02,  1.2387e-03,  7.4507e-03,  2.5118e-03, -1.9492e-02,\n",
       "        -5.6680e-03,  2.1162e-03,  6.2606e-03,  8.2028e-03,  7.3801e-03,\n",
       "         1.4894e-04, -3.9624e-03, -1.4301e-04,  1.7798e-02,  5.1281e-03,\n",
       "         1.7411e-03, -1.9492e-03,  1.0500e-02, -2.2235e-03, -1.0254e-02,\n",
       "        -1.1167e-02, -4.9383e-03,  4.9211e-03, -5.9313e-03, -5.5160e-03,\n",
       "        -7.8050e-03,  1.1671e-02, -5.3812e-03, -5.0349e-03, -7.2109e-03,\n",
       "         4.1156e-03,  5.8032e-03,  4.1234e-03, -1.7233e-02, -1.4829e-02,\n",
       "        -1.5040e-03, -5.0660e-04,  5.2046e-03,  1.0588e-03, -9.6215e-03,\n",
       "         5.4500e-03, -7.7955e-04,  3.9994e-03,  2.1864e-02, -1.2606e-02,\n",
       "        -9.4638e-05, -2.4679e-03,  8.0036e-04, -4.4661e-03, -7.0676e-04,\n",
       "         1.4655e-02,  1.0991e-02,  4.3097e-03,  4.9390e-03,  4.6982e-04,\n",
       "        -6.7061e-03, -1.6278e-03, -2.4310e-03,  1.5630e-02, -5.2157e-03,\n",
       "         3.7680e-03,  5.8129e-03,  7.4657e-04,  1.6862e-02, -1.5148e-02,\n",
       "        -1.5924e-03,  1.8057e-02, -2.1019e-02, -9.4220e-04, -6.6161e-03,\n",
       "         6.3431e-03,  4.9373e-03, -9.5686e-03, -3.4112e-03,  5.2357e-03,\n",
       "         1.0748e-02,  3.2716e-03, -8.0974e-03,  1.7967e-02, -5.3152e-03,\n",
       "         1.5305e-02,  8.5881e-03,  1.1292e-02, -8.8376e-03,  4.3076e-04,\n",
       "        -6.0492e-03,  1.8073e-03,  1.1686e-02,  2.7237e-04,  3.3392e-03,\n",
       "        -7.3122e-04, -5.4210e-03,  1.0999e-02,  1.8922e-02,  8.6953e-03,\n",
       "         1.2930e-02, -2.5762e-03, -7.6038e-03, -9.6958e-03,  7.3285e-03,\n",
       "         8.3488e-03,  9.3472e-03, -1.2341e-02, -6.5701e-03, -4.6116e-03,\n",
       "        -2.4659e-03, -6.1469e-03,  8.2471e-03, -3.1474e-04,  7.4317e-03,\n",
       "        -1.0086e-02,  2.3770e-03,  1.0237e-03, -9.9464e-03, -1.5833e-03,\n",
       "         1.1508e-03, -4.4038e-03, -1.9553e-02, -8.3575e-03, -1.2482e-02,\n",
       "        -4.3895e-03,  2.4583e-03,  2.0652e-03, -1.5507e-02, -1.2468e-02,\n",
       "         1.2474e-02, -1.8571e-03, -1.0345e-02,  3.8134e-04, -4.6251e-04,\n",
       "         4.5836e-03,  8.4277e-03, -5.4155e-03,  7.9515e-04, -6.6065e-03,\n",
       "        -7.7276e-03,  1.6530e-02,  5.7710e-03,  3.1103e-03,  1.5969e-03,\n",
       "         3.1336e-03,  7.5977e-03,  7.4709e-03,  1.7012e-02,  4.4680e-03,\n",
       "         2.2608e-03, -6.5336e-03,  3.8886e-03,  2.6040e-03, -9.3745e-03,\n",
       "         3.7934e-04, -9.5665e-03, -1.7457e-02, -9.3809e-03,  2.2275e-03,\n",
       "         1.2100e-02,  1.7256e-02, -1.0182e-02,  1.9237e-02, -5.4665e-03,\n",
       "         9.0741e-04, -9.3696e-03,  4.0270e-04,  2.9967e-04,  8.1688e-03,\n",
       "         1.3042e-02,  1.0653e-03,  2.1506e-02,  8.2180e-03, -5.0334e-03,\n",
       "         1.3922e-02,  1.3572e-03, -1.1149e-02,  1.3020e-03,  1.9174e-02,\n",
       "        -5.9521e-03], device='cuda:0'), tensor([ 5.4403e-03, -1.8394e-02, -2.3756e-02,  6.1009e-04, -7.3069e-03,\n",
       "        -5.5266e-03,  1.2618e-02, -4.6111e-03,  7.4135e-03, -1.0123e-02,\n",
       "        -1.2162e-02, -1.0845e-02,  8.8240e-03,  1.3489e-02, -1.0028e-02,\n",
       "         5.4289e-03,  5.9170e-03,  7.0252e-03, -8.8289e-03,  4.5186e-03,\n",
       "        -5.6482e-03, -1.1392e-03,  2.8467e-03,  1.0881e-02, -1.0544e-02,\n",
       "         7.5516e-03,  1.4132e-02,  1.6825e-02, -1.5387e-03, -6.8012e-03,\n",
       "         5.7649e-03, -9.2816e-03, -1.5079e-03, -1.6157e-02,  8.1225e-03,\n",
       "        -1.5390e-03,  1.0562e-02,  9.4553e-03, -9.8457e-03,  5.9915e-04,\n",
       "        -5.1581e-04,  1.4598e-03,  8.5979e-03,  8.7074e-05, -2.2101e-03,\n",
       "        -2.7661e-03,  7.9505e-03, -2.2265e-02, -8.2530e-03, -2.6153e-03,\n",
       "         1.1139e-02,  8.0510e-03, -3.9279e-03,  7.9003e-03,  1.3550e-02,\n",
       "        -1.6082e-03,  2.7019e-03,  1.5509e-02, -1.9544e-03, -8.6115e-03,\n",
       "         2.7714e-03,  5.2526e-03, -3.5110e-03,  9.5460e-03,  4.3331e-03,\n",
       "         1.2000e-02, -1.1109e-03, -3.4422e-03,  1.0596e-02,  8.3919e-04,\n",
       "         1.6823e-02, -1.1356e-02,  1.4822e-02, -1.3585e-02,  1.1052e-02,\n",
       "         1.3731e-02,  1.5404e-04,  2.0046e-02, -1.6342e-03, -7.6574e-03,\n",
       "         2.0765e-03,  2.0223e-02, -3.7094e-04, -4.8927e-03,  2.2669e-02,\n",
       "        -1.6691e-03, -1.1158e-02,  1.9092e-05,  8.6084e-03,  4.5018e-03,\n",
       "         6.6155e-03, -8.1124e-04,  1.4763e-02, -2.3083e-02,  3.9656e-03,\n",
       "        -7.2495e-03, -2.2555e-03, -9.4276e-03, -7.8211e-03,  1.8597e-02,\n",
       "        -1.9776e-02, -3.9379e-03, -8.2980e-03, -1.9001e-03, -2.0591e-03,\n",
       "        -3.3581e-03, -2.2656e-02,  6.2465e-04,  5.7490e-03,  8.3724e-03,\n",
       "         5.9264e-03,  5.4953e-03, -1.4512e-02,  2.7750e-03,  1.6075e-03,\n",
       "         1.7838e-02,  2.1127e-02, -6.0057e-03, -1.3233e-02,  5.4904e-03,\n",
       "        -3.0219e-03,  2.5209e-02,  7.7546e-03,  1.1565e-02, -1.4382e-02,\n",
       "         1.1940e-02,  1.7268e-02, -3.5781e-04,  1.6149e-02, -1.1288e-02,\n",
       "        -1.0510e-03,  1.6774e-02,  1.8090e-02,  3.3926e-04, -7.1144e-03,\n",
       "         1.5474e-04,  5.3057e-03, -6.9029e-03,  7.3308e-03, -4.4422e-05,\n",
       "        -5.1431e-03,  3.4009e-03,  2.1143e-02, -7.6649e-04, -3.1558e-04,\n",
       "        -1.1137e-02,  1.1683e-02,  9.0477e-03, -6.3522e-03, -1.2237e-02,\n",
       "        -8.8466e-03, -6.0189e-03,  2.2395e-03, -1.2695e-02, -1.3104e-02,\n",
       "        -8.0423e-03, -8.5380e-03,  3.3060e-03,  6.1726e-03, -1.0182e-03,\n",
       "         2.8030e-03,  8.8366e-03,  2.4716e-03, -1.7054e-02, -1.3861e-02,\n",
       "         1.0093e-03,  2.9862e-03, -2.9696e-03, -1.0660e-02, -5.0001e-03,\n",
       "         9.2675e-03,  3.4351e-03, -6.4859e-03,  1.2982e-02,  1.8497e-02,\n",
       "        -2.3863e-03,  7.6864e-03,  1.4484e-02,  9.2315e-03, -1.2355e-03,\n",
       "         1.4949e-02, -1.3455e-02, -7.4322e-03,  9.1307e-03,  2.1380e-04,\n",
       "        -7.8830e-03, -8.4831e-03, -8.5451e-03,  4.9191e-03, -4.0293e-03,\n",
       "         1.0006e-02,  1.6788e-02,  3.2255e-04, -3.7558e-03, -1.3835e-02,\n",
       "         1.0134e-02, -1.3861e-02, -7.7682e-03,  1.4534e-02, -4.3061e-03,\n",
       "         6.8305e-03, -3.6273e-03,  6.0035e-04,  2.7633e-02, -2.5928e-04,\n",
       "         1.5943e-02, -1.1714e-02, -8.5465e-03,  1.2902e-02,  1.7502e-03,\n",
       "         9.5815e-03,  1.6232e-02,  7.4884e-05,  1.0653e-02,  7.2824e-03,\n",
       "         1.4719e-02, -1.4265e-02,  6.0340e-03,  2.8320e-03,  8.5738e-03,\n",
       "        -3.3153e-03, -1.0697e-02, -5.7090e-03,  3.5094e-02, -6.1551e-03,\n",
       "        -8.0981e-03,  1.3674e-02,  1.3693e-02,  6.4131e-03, -1.3235e-03,\n",
       "        -1.5484e-02,  1.6043e-02,  1.6566e-03,  7.3752e-03, -1.9931e-02,\n",
       "         1.4207e-02, -4.5637e-03, -1.3600e-02,  6.2387e-03, -7.7793e-03,\n",
       "         4.4174e-04,  1.1292e-02, -3.5331e-04,  1.1603e-03, -5.1970e-03,\n",
       "         1.0808e-02,  4.1544e-03,  7.0313e-04, -9.2469e-03,  1.2769e-02,\n",
       "         2.4862e-03,  1.4393e-02, -8.9038e-03,  9.3714e-03, -4.0425e-03,\n",
       "        -8.0237e-03], device='cuda:0'), tensor([[[[ 0.0040]],\n",
       "\n",
       "         [[ 0.0139]],\n",
       "\n",
       "         [[-0.0072]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-0.0021]],\n",
       "\n",
       "         [[-0.0200]],\n",
       "\n",
       "         [[ 0.0027]]],\n",
       "\n",
       "\n",
       "        [[[ 0.0016]],\n",
       "\n",
       "         [[-0.0011]],\n",
       "\n",
       "         [[-0.0040]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 0.0062]],\n",
       "\n",
       "         [[ 0.0084]],\n",
       "\n",
       "         [[ 0.0032]]],\n",
       "\n",
       "\n",
       "        [[[ 0.0136]],\n",
       "\n",
       "         [[ 0.0215]],\n",
       "\n",
       "         [[ 0.0011]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 0.0050]],\n",
       "\n",
       "         [[-0.0071]],\n",
       "\n",
       "         [[ 0.0040]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[-0.0048]],\n",
       "\n",
       "         [[ 0.0017]],\n",
       "\n",
       "         [[-0.0022]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-0.0028]],\n",
       "\n",
       "         [[-0.0131]],\n",
       "\n",
       "         [[-0.0009]]],\n",
       "\n",
       "\n",
       "        [[[-0.0166]],\n",
       "\n",
       "         [[ 0.0052]],\n",
       "\n",
       "         [[ 0.0084]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-0.0204]],\n",
       "\n",
       "         [[ 0.0049]],\n",
       "\n",
       "         [[-0.0040]]],\n",
       "\n",
       "\n",
       "        [[[ 0.0092]],\n",
       "\n",
       "         [[ 0.0089]],\n",
       "\n",
       "         [[-0.0010]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-0.0153]],\n",
       "\n",
       "         [[-0.0172]],\n",
       "\n",
       "         [[ 0.0116]]]], device='cuda:0'), tensor([ 1.2934e-02,  5.0456e-03,  1.5843e-02, -8.8897e-03,  4.0090e-03,\n",
       "        -1.0328e-02,  6.3615e-03,  7.6986e-03,  8.7161e-03, -1.4872e-02,\n",
       "         8.7340e-03, -6.3595e-03,  5.8657e-03,  3.1103e-03,  2.7591e-03,\n",
       "        -9.2494e-05, -6.8905e-03,  2.1831e-03, -6.0304e-04,  6.7550e-04,\n",
       "         1.4736e-02,  2.0670e-02,  1.1362e-03, -1.4794e-03,  5.6978e-03,\n",
       "         5.3467e-03,  3.5222e-03,  1.7302e-03,  5.6015e-05,  1.7052e-02,\n",
       "        -2.8796e-03, -2.6903e-03, -1.0451e-02,  2.5703e-03, -4.7960e-03,\n",
       "         9.5627e-03, -1.0901e-02, -1.9190e-03,  3.8240e-03, -4.6737e-03,\n",
       "        -2.3205e-03, -1.8712e-02, -1.0417e-02, -1.2650e-02, -1.0391e-03,\n",
       "         1.2321e-02, -1.4023e-03,  1.7866e-03,  8.6026e-03, -1.1907e-02,\n",
       "        -1.5687e-02, -7.5773e-03, -1.5337e-02,  1.2451e-02, -3.2448e-03,\n",
       "         1.1126e-02, -2.0036e-03,  1.4429e-02, -1.3130e-02, -3.3273e-03,\n",
       "         7.7319e-03, -1.2035e-02, -5.4456e-03, -1.8414e-03,  2.0013e-02,\n",
       "        -2.9877e-03,  4.8867e-03,  9.4435e-03,  2.9894e-03,  1.3811e-02,\n",
       "         1.8297e-02,  8.5093e-03,  1.1853e-02,  6.3875e-04,  4.1413e-03,\n",
       "         4.6608e-03,  5.1780e-03, -1.3710e-02, -1.5177e-02,  1.8444e-03,\n",
       "         2.3106e-02, -2.5995e-03, -3.7906e-03,  7.7090e-03,  1.9286e-02,\n",
       "         5.1999e-03,  4.6572e-03, -2.6189e-03, -2.2706e-03, -2.4808e-03,\n",
       "         6.3481e-03, -1.1800e-02,  1.2491e-02, -7.7125e-03, -1.9558e-02,\n",
       "        -9.8518e-03, -1.4315e-02, -1.2146e-02,  5.8507e-03, -3.2071e-03,\n",
       "        -2.2728e-03,  1.2871e-02, -7.9405e-03, -1.3410e-02, -8.2150e-03,\n",
       "        -5.3219e-03,  3.6569e-03,  8.4227e-03,  1.6785e-02, -1.5081e-02,\n",
       "        -2.0260e-02,  1.8644e-02,  3.1331e-03,  3.8246e-03,  4.8987e-03,\n",
       "         4.0886e-03, -2.0235e-03,  6.9588e-05, -4.6824e-03,  2.2720e-03,\n",
       "        -9.9020e-03, -2.6714e-03,  3.2012e-04,  3.3064e-02,  1.7740e-03,\n",
       "         1.1326e-02, -9.0942e-03,  3.7904e-03,  6.6143e-03, -3.5697e-03,\n",
       "         3.0799e-03,  1.9091e-02, -1.5458e-02, -5.5381e-03, -1.3802e-03,\n",
       "         6.0334e-04,  1.7234e-02,  1.5722e-02,  1.2386e-02, -2.8884e-03,\n",
       "        -9.9445e-03,  3.8372e-04, -6.9564e-04,  1.3977e-02, -1.8107e-02,\n",
       "        -1.5748e-03,  1.0750e-02, -1.1328e-02, -8.7262e-03,  1.5739e-02,\n",
       "         5.5335e-03, -1.9796e-02, -5.7420e-03,  6.2346e-03,  6.3755e-03,\n",
       "         4.3773e-03, -3.4526e-03,  8.4292e-03, -2.6263e-02, -5.2488e-03,\n",
       "        -8.6106e-03,  1.0409e-03,  1.1393e-03, -2.4787e-03, -6.0519e-03,\n",
       "        -4.9655e-03,  6.5692e-03, -1.2133e-02, -3.5346e-03,  5.4039e-03,\n",
       "         1.8640e-02, -5.8053e-04,  8.5910e-03,  1.5110e-02,  2.1107e-02,\n",
       "        -1.3725e-02,  2.2941e-03, -3.3456e-03, -1.7574e-02, -3.7121e-03,\n",
       "         1.1206e-02,  2.9114e-03, -5.8212e-03,  1.1987e-02, -1.2005e-02,\n",
       "         7.8149e-04, -5.1319e-03, -1.3199e-02, -2.3938e-03,  1.1094e-02,\n",
       "        -1.8783e-03,  2.1002e-02,  5.2794e-03, -9.2626e-03,  3.2414e-03,\n",
       "        -2.6681e-03, -4.9184e-03, -4.6474e-03,  9.2348e-03, -1.1846e-02,\n",
       "        -2.7420e-03,  1.0361e-02, -2.8204e-02, -2.6186e-03,  1.1238e-02,\n",
       "         1.0256e-02,  1.4306e-02,  5.6469e-03, -8.5612e-03, -5.5649e-03,\n",
       "        -4.1265e-03, -1.5012e-02,  4.8033e-03,  1.2709e-02,  5.6323e-03,\n",
       "         5.5233e-03, -2.8788e-02,  7.9216e-03, -5.7036e-03, -4.8944e-04,\n",
       "        -8.2148e-03,  4.5967e-03, -1.8910e-02,  7.3077e-04, -7.2480e-03,\n",
       "        -7.8002e-03,  1.4059e-02, -1.0829e-02,  4.3578e-03, -3.5260e-03,\n",
       "        -3.0938e-03, -1.4102e-02,  3.4685e-03,  7.6165e-03,  1.0600e-02,\n",
       "         2.5707e-04, -2.5752e-02,  5.1542e-05,  9.6157e-03, -9.0616e-03,\n",
       "         2.8338e-02, -7.4617e-03, -2.5634e-03, -7.0153e-03, -5.0529e-03,\n",
       "         2.3712e-03,  5.3361e-04,  1.3925e-02, -8.1720e-03, -1.2850e-02,\n",
       "        -1.3279e-02, -8.1129e-03, -5.8222e-03,  2.3063e-02,  7.6056e-03,\n",
       "         1.4317e-02], device='cuda:0'), tensor([ 2.2396e-03, -6.2940e-03, -5.7466e-03, -1.7854e-02,  1.6424e-02,\n",
       "         7.2913e-03,  1.3096e-02, -8.4428e-03,  1.7592e-02, -1.4616e-03,\n",
       "         1.0078e-02, -1.2605e-03,  1.3036e-02, -6.5764e-03, -2.7203e-03,\n",
       "         4.2317e-03, -2.7066e-03,  9.9165e-03,  2.2497e-04, -5.7876e-03,\n",
       "        -3.6791e-04, -1.2436e-02,  1.2869e-02, -1.1555e-02,  1.2525e-02,\n",
       "        -2.0116e-02,  1.6661e-02, -1.1886e-03,  2.5409e-04,  2.0533e-02,\n",
       "        -3.6321e-03, -3.6039e-03,  4.6984e-03,  1.2293e-02,  1.5725e-02,\n",
       "        -9.6266e-03,  1.0400e-03, -8.2936e-03, -3.8812e-03, -8.8255e-03,\n",
       "         3.8380e-03, -1.4103e-02, -1.9419e-02,  5.0710e-03,  1.1689e-02,\n",
       "        -4.6681e-03,  1.3644e-02, -1.5239e-02,  3.1710e-03, -5.8301e-03,\n",
       "        -4.4203e-03, -2.9542e-03,  4.1728e-03, -1.8509e-03,  3.8139e-03,\n",
       "         2.7740e-03,  2.0236e-03, -1.9958e-02, -1.1784e-03, -1.8010e-03,\n",
       "         1.0839e-02,  1.0273e-02, -7.1010e-03,  2.6064e-02,  6.7895e-03,\n",
       "         3.4646e-03, -1.5186e-02, -1.0126e-02,  1.3130e-02, -9.9045e-03,\n",
       "         1.8930e-02,  3.3344e-04, -4.5940e-03, -1.2926e-02,  3.4448e-03,\n",
       "        -1.7428e-02, -3.1283e-03, -6.9232e-03, -3.0631e-03,  1.2423e-02,\n",
       "        -2.6505e-03, -1.5784e-02, -9.6546e-03,  2.7728e-03,  1.0943e-02,\n",
       "         1.7310e-02,  1.2259e-02, -6.4552e-03, -7.4604e-03, -1.0719e-03,\n",
       "         2.0630e-03,  1.7585e-02,  1.6778e-03,  2.2657e-02, -6.5374e-03,\n",
       "         4.3873e-03, -4.7618e-03, -7.1072e-03, -1.0124e-03,  8.8255e-03,\n",
       "        -8.8551e-03, -4.8750e-03,  9.0569e-03, -7.4097e-03, -9.1231e-03,\n",
       "         1.6842e-03, -7.4323e-04, -1.3490e-02, -1.9494e-02, -4.5543e-03,\n",
       "        -1.1705e-02, -4.1153e-04,  1.1697e-03,  3.6123e-03,  1.0787e-02,\n",
       "        -6.9326e-04,  1.8919e-02, -1.5434e-02,  2.2503e-02,  1.4316e-03,\n",
       "        -4.2084e-04,  1.0430e-02, -1.3361e-02, -1.5448e-02,  9.5711e-03,\n",
       "         3.9663e-03,  5.1187e-03,  6.1317e-03, -1.3010e-02,  5.9819e-03,\n",
       "        -2.7503e-03,  2.3822e-03, -4.8909e-03,  3.7870e-03, -1.6641e-02,\n",
       "        -1.7851e-02, -6.4926e-03,  9.3735e-03,  3.7992e-03,  2.4545e-03,\n",
       "         1.1323e-02,  3.4993e-03,  8.9751e-03, -8.6449e-03, -4.7276e-03,\n",
       "         7.9086e-04, -8.9500e-03,  4.3873e-03,  8.0568e-03, -8.6843e-03,\n",
       "         6.1461e-03, -2.0629e-02, -1.1844e-02, -1.9857e-02,  5.5502e-03,\n",
       "        -1.2446e-02, -1.3635e-02,  2.5846e-04, -7.2247e-03, -3.8492e-03,\n",
       "         3.7762e-03,  1.9939e-02, -1.0069e-02, -1.1571e-02,  2.3047e-02,\n",
       "         4.2708e-03, -1.2975e-02,  1.3549e-02, -1.4053e-03, -6.2944e-03,\n",
       "        -7.2653e-03,  3.9559e-03,  1.3265e-02, -2.0381e-02, -3.2047e-03,\n",
       "        -8.4047e-03,  1.4154e-02, -3.2805e-03,  2.2575e-03, -8.7203e-03,\n",
       "        -2.5571e-04,  4.2600e-03, -7.6947e-04,  5.2050e-03, -2.3506e-03,\n",
       "        -1.4273e-02, -5.0834e-03, -6.7461e-03,  2.5530e-02,  4.8826e-03,\n",
       "        -5.2829e-04, -1.0344e-02,  1.1320e-02,  5.9331e-03, -1.2847e-03,\n",
       "        -1.1337e-02,  1.0713e-02,  2.9206e-03, -1.8986e-02,  1.7643e-03,\n",
       "        -1.1827e-02, -7.9340e-03,  4.0385e-03, -2.6272e-03,  4.8365e-03,\n",
       "         1.4424e-02,  2.2845e-02, -2.1508e-03, -1.0786e-02, -5.6044e-03,\n",
       "         9.0721e-04,  1.0131e-02,  9.1223e-03,  7.8609e-03,  7.8473e-03,\n",
       "        -1.0189e-02, -1.4983e-02, -6.0461e-03,  8.2878e-04,  6.4576e-04,\n",
       "        -1.1862e-03,  6.7669e-03,  1.8442e-02,  1.8513e-02, -2.2149e-03,\n",
       "         4.9296e-03, -1.0074e-02, -2.2685e-03, -1.0822e-03, -1.0270e-02,\n",
       "         8.3153e-04, -9.7124e-03, -1.5270e-02, -5.1779e-03, -3.2435e-04,\n",
       "         5.0561e-03, -3.0038e-03, -1.1062e-02,  3.0950e-03,  1.6284e-04,\n",
       "         1.1347e-02,  7.1054e-03, -9.4087e-03,  9.5167e-03, -5.1918e-04,\n",
       "         7.6005e-03, -7.1318e-03, -4.6998e-03, -2.4286e-05, -3.6766e-03,\n",
       "        -3.4200e-02,  1.5584e-02, -3.0471e-03,  6.4053e-03, -8.0180e-03,\n",
       "        -2.7101e-04], device='cuda:0'), tensor([[[[ 3.0123e-03, -2.5234e-02,  9.2000e-04],\n",
       "          [-7.5549e-03,  1.5643e-03, -4.1168e-03],\n",
       "          [ 5.2187e-03,  1.0928e-02, -2.9816e-03]],\n",
       "\n",
       "         [[-4.7467e-03,  9.9417e-03, -6.2624e-03],\n",
       "          [ 1.3400e-02, -1.1948e-02, -7.2841e-03],\n",
       "          [ 8.3693e-03, -8.9509e-03, -1.2686e-02]],\n",
       "\n",
       "         [[-1.0677e-02,  1.1211e-02,  4.4583e-03],\n",
       "          [ 2.0238e-02, -1.1785e-02,  1.2178e-02],\n",
       "          [-1.2285e-03,  7.0262e-03,  4.7424e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-2.7021e-03,  1.7562e-03, -2.6772e-04],\n",
       "          [ 1.1668e-02,  1.9051e-03,  1.5837e-02],\n",
       "          [-1.0335e-02, -8.4336e-03,  7.0301e-03]],\n",
       "\n",
       "         [[-1.7904e-03,  5.9601e-03,  9.8092e-04],\n",
       "          [ 1.4766e-02, -1.2591e-02, -8.5591e-04],\n",
       "          [ 5.1920e-03,  1.0117e-02,  6.9696e-03]],\n",
       "\n",
       "         [[ 7.0642e-04,  5.0848e-03,  1.8558e-02],\n",
       "          [ 2.7178e-03,  3.9708e-03,  6.7983e-04],\n",
       "          [-3.4286e-03,  1.6796e-02,  2.1774e-02]]],\n",
       "\n",
       "\n",
       "        [[[-6.7555e-03, -1.5837e-02, -6.1799e-03],\n",
       "          [ 5.2982e-03,  1.4574e-02,  1.2809e-03],\n",
       "          [ 5.5415e-03, -5.3353e-04,  8.6275e-03]],\n",
       "\n",
       "         [[ 5.1406e-03,  8.2356e-03,  5.0433e-03],\n",
       "          [ 1.9234e-03, -1.9898e-02,  4.0127e-03],\n",
       "          [-7.1201e-03, -1.0790e-02, -1.6611e-02]],\n",
       "\n",
       "         [[ 1.8607e-02, -9.2525e-03, -5.6760e-03],\n",
       "          [ 6.3170e-03, -1.0622e-02,  8.7029e-03],\n",
       "          [ 4.8361e-03, -1.4479e-02,  4.3054e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-4.7096e-03,  2.0511e-02,  4.5040e-03],\n",
       "          [ 4.1761e-03, -4.4676e-03, -9.3149e-03],\n",
       "          [-1.6911e-03, -4.2945e-03,  1.5250e-02]],\n",
       "\n",
       "         [[-5.9392e-03, -5.4363e-03, -2.8526e-03],\n",
       "          [ 1.2446e-02,  1.4137e-02,  6.5531e-03],\n",
       "          [ 1.6112e-02,  1.3771e-02,  1.1126e-02]],\n",
       "\n",
       "         [[-1.0497e-02, -4.3522e-03, -2.8136e-02],\n",
       "          [ 3.8639e-03,  1.2663e-02, -1.8109e-02],\n",
       "          [ 2.6310e-02,  4.5065e-03,  1.1008e-02]]],\n",
       "\n",
       "\n",
       "        [[[ 2.6925e-02,  1.3691e-03, -9.1915e-03],\n",
       "          [-2.8459e-03, -3.6448e-03,  7.3671e-03],\n",
       "          [-8.6399e-03, -1.1523e-02,  1.3301e-03]],\n",
       "\n",
       "         [[ 1.1737e-03, -7.4475e-03, -1.8718e-02],\n",
       "          [-6.2961e-03, -3.7994e-04, -6.5332e-03],\n",
       "          [-2.0171e-04,  5.7100e-03, -1.5843e-02]],\n",
       "\n",
       "         [[-2.2867e-02,  5.1842e-03, -5.0722e-03],\n",
       "          [ 1.2337e-02,  1.4095e-02, -8.1602e-03],\n",
       "          [-2.1818e-03, -2.9311e-03,  1.1649e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-8.5117e-03, -1.1323e-02, -6.1051e-03],\n",
       "          [-8.4817e-04, -1.3721e-02,  8.2523e-03],\n",
       "          [-1.6952e-02, -1.1917e-02,  4.9028e-03]],\n",
       "\n",
       "         [[ 4.2367e-03,  1.5016e-02, -1.9075e-03],\n",
       "          [-1.5440e-03,  9.6809e-03, -1.0523e-03],\n",
       "          [ 1.3524e-02,  6.5310e-03, -5.9484e-03]],\n",
       "\n",
       "         [[-1.0424e-02, -1.9821e-02, -1.3426e-03],\n",
       "          [-4.7206e-03,  5.0003e-03, -8.5686e-03],\n",
       "          [ 1.3259e-02,  2.5538e-04, -2.0718e-03]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[ 1.5102e-02,  5.2231e-03,  4.4726e-03],\n",
       "          [-4.1778e-03, -9.3977e-03,  1.6532e-02],\n",
       "          [ 1.0614e-02,  7.2887e-03, -7.4468e-04]],\n",
       "\n",
       "         [[ 1.0763e-02,  6.1519e-03,  2.1504e-03],\n",
       "          [-1.9016e-03, -1.3529e-02,  1.7401e-02],\n",
       "          [ 6.6513e-04, -4.1438e-03,  2.2791e-02]],\n",
       "\n",
       "         [[ 8.3459e-04,  1.4551e-03, -5.0235e-03],\n",
       "          [-1.6313e-02, -1.7349e-03, -6.0017e-03],\n",
       "          [ 2.8310e-02,  1.3124e-03,  1.3878e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-9.7541e-03, -4.7499e-04,  4.1265e-03],\n",
       "          [ 1.0314e-02, -1.3863e-02,  1.5547e-02],\n",
       "          [-1.0823e-02, -2.3740e-02,  1.9122e-03]],\n",
       "\n",
       "         [[ 1.3212e-02, -4.1071e-03,  7.1466e-04],\n",
       "          [-1.0114e-02, -6.5000e-04,  1.9901e-02],\n",
       "          [ 8.6845e-03,  1.0247e-03, -2.9065e-04]],\n",
       "\n",
       "         [[ 9.3138e-03, -7.8741e-03,  2.5700e-03],\n",
       "          [ 1.5666e-03, -1.0707e-02, -6.8502e-04],\n",
       "          [ 2.4634e-04,  4.4442e-03, -2.3324e-02]]],\n",
       "\n",
       "\n",
       "        [[[-3.3083e-03,  1.4809e-02, -4.0153e-03],\n",
       "          [-1.4494e-02,  6.1520e-03, -1.0007e-03],\n",
       "          [ 1.2931e-02,  1.4701e-02,  7.8527e-03]],\n",
       "\n",
       "         [[ 2.1526e-02, -1.0305e-02,  4.1674e-03],\n",
       "          [-6.0330e-03,  8.1899e-03,  4.3665e-03],\n",
       "          [ 1.5772e-02,  2.4657e-04,  1.2546e-02]],\n",
       "\n",
       "         [[ 6.0730e-03, -7.0982e-03,  1.3163e-02],\n",
       "          [-1.5387e-02, -1.3502e-02, -6.8844e-03],\n",
       "          [-1.5068e-03,  2.9349e-03, -1.0234e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 7.7004e-05, -6.7412e-03,  1.8030e-02],\n",
       "          [-4.5774e-03, -1.2977e-02, -1.5477e-02],\n",
       "          [-2.9118e-03,  1.1530e-02,  2.0418e-02]],\n",
       "\n",
       "         [[-4.1328e-03, -3.6895e-03,  1.0616e-02],\n",
       "          [-5.8816e-03, -3.7106e-04, -4.4664e-03],\n",
       "          [-1.0426e-03,  1.3894e-03, -8.0046e-03]],\n",
       "\n",
       "         [[ 1.9536e-03, -1.1942e-02,  6.1748e-03],\n",
       "          [-2.1629e-03, -1.3851e-02,  1.4718e-03],\n",
       "          [ 1.1857e-02, -8.6440e-03,  1.2382e-02]]],\n",
       "\n",
       "\n",
       "        [[[-1.8960e-03,  2.4358e-02, -1.3733e-03],\n",
       "          [ 3.9417e-03,  4.8698e-03, -4.2662e-04],\n",
       "          [-4.6554e-03, -1.4273e-03, -5.5081e-03]],\n",
       "\n",
       "         [[ 5.8478e-03,  3.4685e-03,  2.7932e-03],\n",
       "          [-8.1369e-03, -9.3362e-03, -2.1123e-02],\n",
       "          [-3.0692e-03,  1.3881e-02, -5.5980e-03]],\n",
       "\n",
       "         [[ 7.3119e-03,  4.9663e-03,  5.8342e-03],\n",
       "          [ 9.0557e-03,  1.1302e-03, -4.4464e-03],\n",
       "          [ 1.4879e-02, -3.0173e-03, -9.3586e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 9.5988e-04,  4.9743e-04,  1.1327e-02],\n",
       "          [-3.2678e-03, -1.5548e-02, -1.2995e-03],\n",
       "          [-6.6941e-04,  1.8067e-02, -2.4167e-03]],\n",
       "\n",
       "         [[-7.6773e-03,  4.8775e-03, -2.4894e-03],\n",
       "          [ 3.0335e-03,  1.0739e-02,  1.2289e-02],\n",
       "          [ 1.5578e-02, -8.5115e-04,  1.2910e-02]],\n",
       "\n",
       "         [[ 7.5223e-03,  9.1693e-03, -1.5618e-02],\n",
       "          [-8.6244e-03,  3.9044e-04,  1.0690e-02],\n",
       "          [-2.5091e-03, -5.3830e-03,  5.8109e-03]]]], device='cuda:0'), tensor([-5.0299e-03, -1.5078e-02,  5.9565e-03,  1.1086e-02,  1.9399e-02,\n",
       "         7.6841e-03, -2.1089e-03, -5.1605e-03,  2.8099e-03,  1.2991e-02,\n",
       "        -7.2508e-04,  1.1185e-02, -4.8883e-04, -1.2721e-03,  8.2750e-03,\n",
       "        -6.2299e-03,  3.5889e-03, -1.7875e-02,  1.0026e-02,  8.1399e-03,\n",
       "         8.6637e-03, -3.7919e-03,  7.6515e-03,  3.0663e-03, -1.1761e-02,\n",
       "        -1.1849e-02, -3.8191e-03, -8.8251e-03,  2.4866e-03, -1.1315e-03,\n",
       "        -1.1899e-02, -1.0729e-02,  9.9312e-03, -1.5218e-02,  5.8916e-03,\n",
       "         2.7004e-03, -3.5824e-03,  4.6135e-03, -2.8064e-04, -8.0738e-03,\n",
       "         2.7559e-03,  8.0586e-03,  3.8931e-03,  1.5370e-02,  1.3896e-02,\n",
       "        -4.7697e-03, -2.0838e-04, -1.0247e-03, -9.0462e-03,  2.6313e-03,\n",
       "        -8.8148e-03,  3.9222e-03, -2.8532e-04,  1.1231e-02, -9.6936e-03,\n",
       "         1.9756e-02,  6.0232e-03, -2.0994e-02,  7.9172e-03,  4.7162e-04,\n",
       "        -8.5825e-03,  1.4947e-02, -1.3088e-02,  6.1257e-03,  1.0043e-02,\n",
       "        -1.2982e-03,  9.8027e-03,  4.1053e-03, -3.0754e-02,  7.4761e-03,\n",
       "        -1.6398e-02,  7.0743e-03,  6.5036e-03,  1.3737e-03, -1.7843e-02,\n",
       "         1.5517e-03,  1.0871e-02, -2.5765e-03, -2.1509e-02,  1.0182e-02,\n",
       "         1.3073e-02,  5.7569e-03, -3.4568e-05, -3.5943e-03, -1.5029e-03,\n",
       "         1.2168e-02, -2.3475e-03,  7.3320e-03, -1.9984e-02, -1.1748e-04,\n",
       "        -1.1976e-02,  2.4565e-03,  1.1387e-03,  1.2890e-02, -1.3694e-02,\n",
       "        -8.0248e-03, -6.9158e-03,  1.7430e-02, -1.0047e-02,  2.3584e-03,\n",
       "        -1.0699e-02,  6.6019e-03, -1.5061e-02,  4.2904e-03,  7.0625e-03,\n",
       "         1.5670e-02, -5.1434e-03,  9.9760e-03, -8.7747e-03,  1.6456e-03,\n",
       "        -1.3526e-02,  1.8836e-03, -2.2241e-03,  5.2336e-03,  1.3253e-02,\n",
       "         1.1401e-02,  7.5444e-03,  5.0621e-03, -4.6573e-03, -1.1528e-02,\n",
       "        -8.6413e-03, -6.2492e-03, -1.0206e-03, -1.2585e-03,  1.2986e-02,\n",
       "         1.9927e-02,  2.1005e-03, -3.6122e-03,  1.1343e-02, -6.3736e-03,\n",
       "        -8.9146e-05,  8.1330e-03,  1.2189e-03,  9.5553e-03,  9.4167e-03,\n",
       "        -1.6729e-02, -1.1140e-02,  4.2042e-03, -1.4117e-02,  3.7337e-03,\n",
       "        -1.2667e-03, -9.1241e-04, -1.6249e-02, -1.3762e-02, -2.9963e-03,\n",
       "         9.8174e-03, -2.8013e-03, -2.3205e-03,  6.0153e-03, -3.1564e-03,\n",
       "        -6.6101e-03,  1.4699e-02,  8.7644e-03, -1.7432e-02, -1.1227e-02,\n",
       "        -2.8472e-03, -1.8101e-02,  4.7689e-03, -1.3092e-02, -6.2563e-03,\n",
       "         6.5055e-03,  5.6215e-03, -1.4453e-02,  5.3425e-03,  1.3762e-02,\n",
       "        -4.2004e-03,  1.3035e-02,  5.8314e-03, -7.7772e-03, -1.0864e-02,\n",
       "         7.3680e-04, -2.2556e-03,  3.0314e-02, -1.1047e-02, -1.3772e-02,\n",
       "        -1.2759e-02, -1.2140e-02, -2.1867e-02,  1.1314e-02,  7.5617e-03,\n",
       "        -1.1769e-02,  1.7436e-02,  8.1046e-03, -4.8381e-03, -3.6135e-03,\n",
       "        -1.2451e-02,  9.9477e-03, -5.8730e-03, -8.8990e-03, -2.2504e-03,\n",
       "         2.6601e-03, -4.0485e-03,  1.5030e-02,  8.2472e-03, -4.8700e-03,\n",
       "         7.4260e-03,  3.1468e-03, -5.0705e-03, -6.8812e-03,  1.3771e-02,\n",
       "        -2.0179e-02, -1.8145e-02,  1.5253e-02, -5.3661e-03, -1.6100e-02,\n",
       "        -1.6710e-02, -8.4392e-03, -2.2922e-03, -1.5418e-02,  1.4584e-03,\n",
       "         1.4809e-02, -3.3259e-03,  4.3988e-03, -1.6483e-02,  6.5833e-03,\n",
       "        -4.2227e-03, -2.9590e-03,  1.5384e-02, -1.7285e-02, -1.3295e-02,\n",
       "         9.8343e-03,  9.5263e-03,  8.0227e-03,  1.2051e-02, -2.3011e-03,\n",
       "        -3.8223e-03, -1.2402e-02, -8.3020e-04,  2.2534e-02,  4.4566e-03,\n",
       "         1.1318e-02,  8.4073e-03,  7.1683e-04,  1.0017e-02, -7.6172e-03,\n",
       "         3.6854e-03, -6.7543e-03, -1.1059e-03, -5.9272e-04,  7.8050e-03,\n",
       "         1.3528e-03, -3.3196e-03, -9.6484e-03,  3.4728e-03,  5.3818e-03,\n",
       "         5.8593e-03,  3.4133e-04, -1.2310e-03, -3.8843e-03, -1.7769e-02,\n",
       "        -7.5848e-03, -1.1362e-02, -1.4026e-02, -1.2525e-02,  1.0596e-03,\n",
       "         2.3421e-02], device='cuda:0'), tensor([ 5.1183e-03, -2.3720e-02,  5.5677e-04,  1.9920e-02,  2.2464e-02,\n",
       "        -3.5456e-03,  6.3483e-03, -7.1453e-03,  2.2338e-02, -1.2057e-03,\n",
       "         1.8903e-02, -4.5957e-03, -8.8814e-03,  1.2741e-02,  2.3739e-03,\n",
       "         4.9565e-03, -2.9801e-03,  1.7478e-03, -2.1423e-02,  9.6669e-03,\n",
       "        -1.6062e-02, -5.2372e-03,  4.7235e-03, -1.5456e-02,  5.3238e-03,\n",
       "        -6.5491e-03,  8.9824e-03, -1.4955e-02,  7.1970e-03, -6.8741e-03,\n",
       "        -7.2768e-03,  3.6243e-03, -8.6751e-04,  1.7687e-02, -1.1002e-02,\n",
       "         1.5746e-02, -7.7388e-03, -4.9474e-03,  5.5722e-03, -6.4240e-03,\n",
       "        -4.5410e-03,  2.1564e-02, -3.3077e-04, -1.3262e-02, -6.9629e-04,\n",
       "        -1.2256e-02,  1.3758e-02, -1.5641e-02, -1.8174e-02,  1.1719e-02,\n",
       "        -4.2009e-03, -4.9243e-03,  1.2561e-02, -2.3005e-03, -3.1585e-03,\n",
       "        -3.8187e-04,  2.1920e-03, -1.2458e-02, -8.0238e-03,  7.0448e-04,\n",
       "        -5.3432e-03, -7.5478e-03, -6.3961e-04, -6.6905e-03,  4.6244e-03,\n",
       "         1.6575e-02, -9.2744e-03,  1.0144e-02,  4.7900e-03,  1.2390e-02,\n",
       "         7.0338e-03, -7.5529e-03,  2.6488e-03,  7.2831e-03, -5.1881e-03,\n",
       "         5.4263e-03, -1.1734e-02,  1.3118e-02, -5.4559e-03,  1.4374e-02,\n",
       "         1.6251e-02, -3.0435e-03,  1.4114e-02, -3.2982e-03, -4.8838e-04,\n",
       "        -1.1988e-02, -1.5181e-02, -1.1838e-02,  1.7212e-02, -5.7074e-03,\n",
       "        -7.8691e-04,  9.5832e-03, -4.2667e-05,  2.0212e-02,  7.3350e-04,\n",
       "        -3.3992e-03,  1.1298e-02, -8.9212e-03, -1.9257e-02, -9.8545e-04,\n",
       "        -1.4140e-02, -1.0728e-02,  1.4621e-02,  1.8194e-02,  2.4791e-03,\n",
       "        -2.9003e-02,  5.7619e-03, -3.4607e-03, -7.7743e-03,  5.9766e-03,\n",
       "        -1.0777e-02,  7.8840e-03, -1.2132e-03, -8.5499e-04, -1.8039e-03,\n",
       "        -3.2263e-03, -1.7045e-03,  9.5241e-04, -1.7573e-03,  5.2389e-03,\n",
       "        -1.2104e-02, -3.6677e-06, -2.1938e-02, -3.9791e-03,  1.7271e-02,\n",
       "         1.2298e-03, -1.1400e-02,  3.5845e-03,  3.5178e-05, -1.5136e-03,\n",
       "        -4.1099e-03, -1.2142e-03, -1.1452e-03, -2.0038e-03,  6.8551e-03,\n",
       "         7.2758e-03,  3.2065e-03, -1.7393e-03,  1.2646e-03,  3.9742e-03,\n",
       "        -3.8988e-03,  8.0159e-03,  8.9781e-04, -8.4806e-03,  1.4165e-02,\n",
       "        -1.6247e-03,  3.4298e-03, -4.2861e-03,  5.5198e-03,  3.4238e-03,\n",
       "         7.8335e-04,  7.9958e-04, -8.4638e-03, -4.8361e-03, -5.6407e-04,\n",
       "         1.3452e-02,  1.6951e-03,  4.6702e-03, -7.5698e-04, -1.6004e-04,\n",
       "        -4.1872e-03,  8.1194e-04,  1.3433e-02,  1.1798e-05,  9.1343e-03,\n",
       "         1.4408e-02,  3.7476e-03, -8.3086e-03, -5.2674e-04, -1.1784e-02,\n",
       "        -9.8103e-03,  1.6820e-03,  2.2560e-02,  1.7973e-03,  4.7771e-03,\n",
       "         2.1622e-04, -1.2950e-03, -1.4740e-02,  3.3887e-03, -1.9824e-03,\n",
       "        -8.6363e-03, -3.2500e-03,  4.6776e-03, -1.3695e-02,  2.1923e-03,\n",
       "        -3.3849e-04,  2.2672e-02, -3.9224e-03,  9.4087e-03, -6.0159e-03,\n",
       "         5.2120e-03,  9.0962e-03,  5.4334e-03,  1.1165e-02,  5.1573e-03,\n",
       "         2.1915e-03,  9.9874e-04,  1.1488e-02,  8.9934e-03,  2.0662e-02,\n",
       "        -1.9200e-03, -8.3166e-03, -1.0056e-02, -2.1120e-02,  5.4251e-03,\n",
       "         4.2212e-03, -5.0153e-03, -7.8350e-03, -5.3790e-03, -6.7873e-03,\n",
       "        -8.5998e-03, -3.4217e-03,  7.8029e-03,  1.0959e-02, -2.0421e-02,\n",
       "        -1.1677e-02,  1.1523e-02,  6.4666e-03, -3.9643e-03,  4.0370e-03,\n",
       "        -1.9086e-02,  1.0912e-02,  8.4250e-03, -2.6194e-03,  2.2222e-02,\n",
       "        -4.0603e-03, -5.9509e-04,  1.9967e-02,  1.5562e-03,  6.2953e-03,\n",
       "        -2.1303e-02, -8.5210e-03, -1.6507e-02, -7.6721e-06,  5.4559e-04,\n",
       "        -1.6409e-02, -2.3537e-03,  1.0315e-03, -7.0550e-03,  1.2152e-02,\n",
       "         9.7150e-03, -6.8069e-04, -4.8767e-03,  1.1949e-02,  6.5279e-03,\n",
       "         2.2521e-02, -1.5085e-02,  1.1869e-02,  1.3732e-02,  9.2700e-03,\n",
       "         3.3036e-03, -4.5243e-03,  6.0261e-03,  1.2868e-02,  4.5919e-03,\n",
       "         1.6301e-02], device='cuda:0'), tensor([[[[ 7.1531e-03, -2.4488e-03,  9.9282e-03],\n",
       "          [-3.0461e-05,  2.0701e-02, -9.4715e-04],\n",
       "          [ 6.1937e-03, -1.4480e-02,  1.9793e-03]],\n",
       "\n",
       "         [[ 3.2101e-03, -1.2180e-02,  8.8084e-03],\n",
       "          [-5.2678e-03, -1.0361e-02, -7.2663e-03],\n",
       "          [ 1.1494e-02,  1.0520e-03, -1.1634e-02]],\n",
       "\n",
       "         [[-5.2255e-03, -2.0319e-03,  1.8415e-02],\n",
       "          [ 1.1768e-02, -1.2646e-02,  5.9655e-03],\n",
       "          [-3.9649e-03, -1.6248e-02, -3.6351e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 6.3992e-03, -2.4469e-02, -1.2846e-02],\n",
       "          [-7.6614e-03, -1.0484e-02,  9.0753e-03],\n",
       "          [ 2.0637e-02, -1.6411e-02,  4.3958e-04]],\n",
       "\n",
       "         [[-6.0984e-03,  6.2573e-04, -5.7092e-03],\n",
       "          [ 1.0584e-02,  7.8688e-03,  1.2927e-02],\n",
       "          [-1.2453e-02, -9.3420e-03, -3.0689e-04]],\n",
       "\n",
       "         [[-2.1482e-03, -2.1952e-04,  7.8008e-03],\n",
       "          [-5.3773e-03, -6.5774e-03,  1.5440e-02],\n",
       "          [-1.1828e-02, -9.4091e-03,  2.3369e-04]]],\n",
       "\n",
       "\n",
       "        [[[-6.3052e-03, -4.5195e-03,  1.3073e-02],\n",
       "          [ 3.7258e-04,  1.4356e-03, -2.3313e-02],\n",
       "          [-6.2084e-03,  4.8311e-03,  2.9896e-03]],\n",
       "\n",
       "         [[-9.5249e-04,  1.8161e-02, -1.3584e-02],\n",
       "          [ 1.3189e-02,  1.9982e-03,  4.3831e-03],\n",
       "          [-1.1403e-02,  1.4154e-04,  1.2983e-02]],\n",
       "\n",
       "         [[-3.6068e-03, -4.7936e-03, -6.3732e-03],\n",
       "          [ 8.5934e-03, -6.0746e-03, -5.9924e-03],\n",
       "          [ 3.4898e-03, -4.2943e-03,  3.3558e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-7.0716e-03, -6.6768e-03, -1.5551e-02],\n",
       "          [ 1.0415e-02,  5.9920e-04, -6.7892e-03],\n",
       "          [ 8.1356e-04, -5.5911e-03,  1.4806e-02]],\n",
       "\n",
       "         [[ 8.5220e-03, -5.7602e-03, -2.0399e-03],\n",
       "          [ 1.0112e-02, -2.5105e-03,  7.6600e-03],\n",
       "          [ 9.2978e-03,  5.5444e-04,  1.1661e-02]],\n",
       "\n",
       "         [[ 4.1622e-03, -5.2810e-03,  6.0788e-05],\n",
       "          [ 5.4125e-03,  4.8653e-03,  2.2892e-02],\n",
       "          [ 1.0413e-02,  1.5900e-02, -7.0557e-03]]],\n",
       "\n",
       "\n",
       "        [[[-1.5072e-02, -3.1091e-03,  8.0775e-03],\n",
       "          [ 8.7555e-03, -2.7658e-03,  1.7347e-02],\n",
       "          [ 9.8639e-03, -9.2210e-03, -9.9635e-03]],\n",
       "\n",
       "         [[-2.9507e-03, -7.2914e-03, -1.5322e-02],\n",
       "          [-6.7043e-03, -3.1816e-03,  1.5910e-03],\n",
       "          [ 1.1098e-02, -3.7144e-03, -9.3957e-03]],\n",
       "\n",
       "         [[ 3.4292e-03, -2.0320e-03,  7.8768e-03],\n",
       "          [-8.3209e-04, -1.4298e-02,  1.5111e-02],\n",
       "          [-5.6490e-03,  7.9963e-03, -2.0756e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-3.4094e-03, -5.1381e-04, -1.0360e-02],\n",
       "          [ 2.4285e-03, -1.8549e-02, -1.9759e-02],\n",
       "          [-7.8918e-03,  7.3503e-03,  7.4863e-03]],\n",
       "\n",
       "         [[-1.8219e-02,  7.5446e-03,  6.1323e-03],\n",
       "          [ 1.5368e-02, -2.8751e-02,  1.5912e-03],\n",
       "          [-5.2113e-03,  1.6577e-03,  6.1272e-03]],\n",
       "\n",
       "         [[-5.2508e-03,  3.9616e-04,  7.0667e-03],\n",
       "          [ 7.5545e-05, -1.8054e-03, -1.7395e-02],\n",
       "          [-1.0147e-02, -6.8732e-03,  1.9241e-03]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[-3.1650e-04, -4.6107e-03, -1.1643e-03],\n",
       "          [ 9.1168e-03,  6.2601e-03,  1.5667e-02],\n",
       "          [ 6.1688e-03, -4.2181e-03,  2.2160e-02]],\n",
       "\n",
       "         [[ 5.7497e-03, -1.8671e-02, -1.0100e-02],\n",
       "          [ 8.7723e-03, -4.7453e-03, -1.3130e-03],\n",
       "          [ 3.8129e-03,  4.9498e-03, -6.3315e-03]],\n",
       "\n",
       "         [[ 9.9992e-03,  7.4563e-03,  7.2091e-03],\n",
       "          [-4.1883e-03,  6.9619e-03,  1.6085e-02],\n",
       "          [ 8.0450e-04, -2.9206e-03, -1.0489e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 3.8729e-03, -6.5826e-03,  3.6475e-03],\n",
       "          [-1.2600e-02, -1.1231e-02, -4.3439e-03],\n",
       "          [ 2.5155e-03,  8.7439e-03, -7.8386e-03]],\n",
       "\n",
       "         [[-1.1651e-02, -1.3173e-02,  5.0002e-03],\n",
       "          [ 8.9672e-03, -2.8560e-03, -4.3548e-03],\n",
       "          [-7.6084e-03,  4.7448e-04,  5.1754e-03]],\n",
       "\n",
       "         [[-5.1142e-03, -2.9548e-03, -4.4921e-03],\n",
       "          [ 3.8440e-03,  1.0168e-02,  2.2553e-02],\n",
       "          [ 9.5992e-03, -1.8301e-02,  4.5928e-03]]],\n",
       "\n",
       "\n",
       "        [[[ 2.1010e-03,  4.6984e-03, -1.8943e-03],\n",
       "          [ 4.7896e-03,  5.1050e-03, -3.4655e-03],\n",
       "          [-1.1085e-02,  1.4970e-03,  1.0908e-02]],\n",
       "\n",
       "         [[ 6.6647e-03,  1.0472e-02,  1.9038e-02],\n",
       "          [-1.0134e-02, -1.7224e-02,  3.7393e-03],\n",
       "          [-6.2603e-03,  1.3592e-03, -1.4694e-02]],\n",
       "\n",
       "         [[ 5.8605e-03,  1.3163e-02,  1.2245e-02],\n",
       "          [-1.4769e-03,  1.1433e-02, -5.1404e-03],\n",
       "          [ 9.3760e-03,  2.7872e-03,  1.5798e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-1.1365e-02, -2.2716e-03, -6.2295e-03],\n",
       "          [ 5.8670e-03,  1.5931e-02, -1.1931e-02],\n",
       "          [-7.9316e-03,  1.7366e-02,  1.3128e-02]],\n",
       "\n",
       "         [[ 1.2688e-02,  5.5417e-03, -5.6396e-03],\n",
       "          [ 4.6370e-03, -1.0569e-03,  3.5471e-03],\n",
       "          [-6.5820e-03, -2.8516e-03, -2.1886e-02]],\n",
       "\n",
       "         [[-2.7568e-04, -1.1628e-03, -1.8307e-02],\n",
       "          [ 1.0953e-02,  2.3128e-03,  5.3144e-03],\n",
       "          [ 1.0909e-02,  2.4309e-03,  8.4345e-03]]],\n",
       "\n",
       "\n",
       "        [[[-2.4825e-03,  2.8269e-03,  1.3931e-02],\n",
       "          [ 2.7454e-03,  9.3926e-03,  1.0179e-03],\n",
       "          [ 7.1082e-03, -1.8351e-03,  6.9186e-03]],\n",
       "\n",
       "         [[-7.3601e-03,  1.1977e-03,  7.4330e-03],\n",
       "          [-5.5778e-03, -4.4960e-03,  1.0343e-02],\n",
       "          [-2.3902e-02,  1.7687e-02,  1.5813e-03]],\n",
       "\n",
       "         [[ 1.1688e-02, -1.0980e-02,  7.9456e-03],\n",
       "          [-4.9483e-04,  2.3768e-02,  4.0901e-03],\n",
       "          [-2.5440e-03,  1.8517e-02, -2.1098e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 2.1630e-02, -1.0583e-02, -2.5429e-03],\n",
       "          [-9.6703e-03,  1.4476e-02,  5.2965e-03],\n",
       "          [ 6.5737e-04,  9.7105e-03, -2.0616e-03]],\n",
       "\n",
       "         [[ 1.0204e-02, -2.4865e-02,  1.9282e-02],\n",
       "          [-4.4581e-03,  4.9019e-03,  1.0629e-03],\n",
       "          [ 1.7531e-02, -5.6232e-03,  5.0565e-03]],\n",
       "\n",
       "         [[-2.5677e-03,  1.9188e-04, -1.3935e-02],\n",
       "          [ 1.0325e-03, -8.3156e-03, -8.0352e-03],\n",
       "          [-1.4288e-02, -6.4789e-03,  3.3993e-03]]]], device='cuda:0'), tensor([-1.7790e-03, -1.5918e-02, -1.3489e-02, -1.3475e-02,  1.1623e-03,\n",
       "         1.8034e-02,  8.2856e-03, -3.5485e-03,  4.1270e-03, -3.7224e-03,\n",
       "         5.3705e-04,  1.2465e-02, -2.5284e-03, -1.1680e-02,  1.0709e-02,\n",
       "        -4.9699e-04,  4.9720e-03, -1.7110e-03,  1.6772e-03,  1.2350e-02,\n",
       "        -8.6769e-03, -4.3485e-03,  7.4537e-03, -1.0323e-02,  6.4583e-03,\n",
       "         8.9915e-03,  4.2013e-03,  9.5413e-03,  8.3423e-03,  5.1029e-03,\n",
       "         6.0070e-03, -2.4697e-03,  9.1835e-03, -3.8917e-05, -1.1270e-02,\n",
       "        -8.2191e-03,  7.7223e-03, -1.6087e-03, -5.3615e-03, -2.1801e-03,\n",
       "        -1.2338e-02, -7.7279e-03,  2.7629e-03,  3.0015e-03,  4.0460e-03,\n",
       "        -2.0689e-03,  2.3151e-02,  2.5301e-03, -4.4902e-03,  8.4916e-04,\n",
       "        -1.1348e-02,  1.6099e-02,  1.3562e-02, -1.6166e-02,  5.9767e-03,\n",
       "        -1.4317e-02,  1.7863e-02,  1.1313e-02,  1.1875e-02, -1.6349e-02,\n",
       "        -1.0238e-02, -7.6012e-04,  1.9673e-02, -2.4874e-02,  1.5561e-02,\n",
       "         8.1078e-04,  1.8190e-02, -3.1117e-03,  1.1763e-02, -1.1562e-02,\n",
       "         1.1952e-02, -1.6991e-02, -3.3448e-03,  1.3244e-02,  1.3008e-03,\n",
       "        -1.0358e-02, -2.3925e-02, -8.7621e-03,  3.9664e-03,  1.1824e-02,\n",
       "         6.6893e-03, -1.4862e-02,  1.7072e-02, -2.2567e-03,  1.8777e-04,\n",
       "         1.3523e-02,  9.0996e-04,  4.4206e-03, -2.5611e-02,  1.6789e-03,\n",
       "         5.7695e-05,  2.2051e-02, -5.7284e-03,  9.3571e-04,  3.5465e-03,\n",
       "         1.7756e-02, -7.9174e-03, -1.1477e-02,  6.3905e-03, -2.3807e-03,\n",
       "         4.7914e-03, -2.0508e-02,  7.0843e-03,  4.7286e-03, -1.3119e-03,\n",
       "         1.3659e-02,  1.3815e-02,  1.2033e-02,  8.0811e-03,  5.1546e-03,\n",
       "        -2.0556e-03,  1.0626e-03, -8.5112e-03, -4.4468e-03, -1.3215e-02,\n",
       "         1.3208e-02,  3.4964e-03,  1.9828e-03, -6.4482e-03,  1.1867e-02,\n",
       "         3.6969e-03, -7.9072e-03, -1.2063e-02,  1.6123e-03, -1.4353e-02,\n",
       "         5.8232e-03, -4.3863e-03, -4.4357e-03,  9.0630e-03,  2.6613e-02,\n",
       "         1.5171e-02,  2.2973e-03,  1.6440e-02, -5.7372e-03, -2.2846e-02,\n",
       "         3.6595e-03,  7.0167e-03,  6.5806e-03, -7.4321e-03,  2.1253e-02,\n",
       "         2.3014e-03, -2.2282e-02, -5.0852e-03,  4.8693e-03, -1.9433e-03,\n",
       "        -2.3674e-02, -9.6488e-03,  1.8092e-03, -1.0685e-02, -4.3121e-03,\n",
       "        -2.1774e-03,  2.5476e-02, -7.9337e-03, -1.9186e-02,  2.5981e-03,\n",
       "        -1.5940e-02, -2.1421e-02,  8.2083e-03, -2.8073e-03, -8.0797e-03,\n",
       "        -1.2249e-03, -7.3617e-03,  4.2595e-04, -2.3721e-03, -1.0946e-02,\n",
       "        -7.4953e-03,  8.3647e-03,  1.9496e-02,  1.3135e-02, -2.0868e-02,\n",
       "        -1.8977e-03, -8.0043e-04, -1.3366e-02, -4.8041e-03,  1.4631e-03,\n",
       "         7.6823e-03, -7.7323e-03, -9.6452e-03, -4.9685e-03,  6.8486e-06,\n",
       "        -1.5048e-02,  1.1255e-02,  1.4304e-02, -3.1703e-03,  1.1827e-02,\n",
       "         3.9947e-03, -2.7078e-03,  6.1139e-03,  2.0795e-02, -1.5273e-02,\n",
       "         4.0562e-03, -1.2527e-02,  1.1470e-02, -9.6730e-04, -5.4158e-03,\n",
       "        -6.9055e-03, -2.2884e-03,  1.3230e-02,  3.9521e-03,  5.1716e-03,\n",
       "         1.7320e-02, -5.0968e-03,  8.7211e-05, -5.0488e-03, -2.1919e-02,\n",
       "        -1.7128e-03,  1.2297e-02, -1.9834e-03,  8.8049e-03, -3.9941e-04,\n",
       "         5.5506e-03, -1.6339e-04, -1.4675e-03, -2.0454e-02,  9.6640e-03,\n",
       "        -1.5583e-02,  4.9840e-03, -1.2988e-03, -1.5386e-02,  2.8934e-03,\n",
       "        -6.8884e-03, -1.1628e-03, -4.7614e-03,  9.8413e-03, -1.9538e-02,\n",
       "         3.7429e-03,  6.4633e-03,  4.8141e-03,  1.2320e-03,  1.5004e-02,\n",
       "        -3.4461e-03,  2.1557e-02,  1.9790e-02,  1.6154e-02,  1.8815e-03,\n",
       "        -1.2715e-02, -3.9160e-03,  1.5717e-03, -1.2400e-02, -8.3046e-03,\n",
       "        -2.2743e-03,  7.2352e-03, -3.3130e-03, -4.2868e-03, -2.5931e-03,\n",
       "        -3.8998e-03,  1.2107e-03,  1.7480e-05, -6.2750e-03, -1.2241e-02,\n",
       "        -1.1316e-02, -3.3847e-03,  1.2436e-02,  4.1550e-03,  1.0829e-02,\n",
       "        -7.3581e-03], device='cuda:0'), tensor([-5.0984e-03, -1.9850e-03,  6.6830e-03, -6.7313e-03,  2.2331e-02,\n",
       "         1.0340e-02,  1.2673e-02,  8.0988e-03, -3.7476e-03, -3.0213e-03,\n",
       "         1.6810e-02, -4.8681e-03,  6.7707e-03,  2.1948e-02, -2.7761e-03,\n",
       "         3.7814e-03, -1.1903e-03, -4.5651e-03,  1.2044e-02, -7.0649e-03,\n",
       "        -5.0289e-03,  7.2234e-03,  5.5952e-03, -9.7691e-03, -6.2095e-03,\n",
       "        -2.3774e-03,  1.7101e-02,  1.0451e-02, -4.5419e-03, -1.9992e-03,\n",
       "         1.7044e-02,  2.3880e-02,  5.1603e-03,  1.7937e-02,  8.1444e-03,\n",
       "         1.4801e-02,  8.3150e-03, -1.3522e-03,  2.7817e-03, -3.0488e-02,\n",
       "         6.5857e-03,  6.1817e-03, -1.9225e-02, -5.7529e-03,  1.2067e-03,\n",
       "        -1.3200e-02, -5.9655e-03, -1.0446e-02, -1.1322e-02,  4.4394e-03,\n",
       "         1.2864e-02,  5.7964e-04, -2.9491e-03,  2.1868e-03, -1.2039e-02,\n",
       "        -2.2926e-02, -1.5455e-02,  1.7623e-02, -5.7339e-03,  1.7028e-02,\n",
       "         1.2110e-02, -2.6532e-03, -5.5238e-03, -6.2720e-03,  2.1529e-03,\n",
       "         9.8253e-03, -1.8042e-02,  8.6430e-03,  1.0401e-02,  1.1362e-02,\n",
       "        -7.8515e-03, -1.0078e-02, -1.4657e-03,  1.3121e-03,  9.8663e-03,\n",
       "         1.3851e-02, -1.3007e-02,  1.8718e-03,  1.3097e-02,  2.2348e-03,\n",
       "         1.3266e-02,  5.4616e-03, -1.1095e-02, -3.5132e-03,  1.8653e-03,\n",
       "        -2.1495e-03,  4.3810e-03, -1.0418e-04,  1.1214e-02,  1.7335e-02,\n",
       "        -1.6645e-02, -4.6227e-03, -2.1696e-02,  2.1731e-03,  7.8247e-03,\n",
       "         6.9389e-03, -1.7420e-03, -1.1999e-03,  1.6490e-02,  1.1578e-02,\n",
       "        -1.3929e-02, -3.5187e-03, -3.3157e-03, -1.5219e-02,  7.9846e-03,\n",
       "         1.7268e-04,  2.2056e-02,  8.9609e-03, -1.0961e-02,  2.4154e-03,\n",
       "        -1.9797e-02,  2.8841e-03,  2.5505e-03,  1.4231e-02, -2.1178e-03,\n",
       "         1.3575e-02, -5.3137e-03,  1.5420e-02,  1.6994e-02, -1.4324e-03,\n",
       "        -5.8608e-04, -3.9316e-03,  2.2938e-02,  1.0363e-02,  1.0295e-02,\n",
       "         4.0030e-03,  1.2598e-02, -1.2716e-02,  7.8583e-03,  1.7159e-02,\n",
       "        -2.2010e-02,  9.6082e-03, -2.1915e-03,  5.5657e-03,  4.0887e-03,\n",
       "        -1.0803e-02, -3.5690e-03, -5.9549e-03,  1.3679e-02,  2.2722e-02,\n",
       "        -2.2983e-03, -9.0016e-03, -3.6884e-03, -5.0179e-03,  9.1919e-03,\n",
       "        -1.8395e-04, -1.1359e-02,  3.3617e-03, -3.6153e-03,  2.1524e-03,\n",
       "         1.2112e-02, -1.4785e-02,  1.1532e-02, -1.0283e-02, -6.3545e-03,\n",
       "        -5.3186e-03,  3.6964e-03, -1.2596e-02,  9.3571e-03, -9.8959e-04,\n",
       "        -1.8138e-02, -9.9965e-03, -7.6845e-03, -1.8439e-03, -7.9266e-03,\n",
       "         1.0619e-02, -1.4780e-02,  1.3625e-02, -4.1796e-03,  1.6837e-02,\n",
       "         1.2889e-02, -9.0594e-03, -2.4758e-02, -5.0613e-04, -8.5362e-04,\n",
       "         1.2168e-02, -1.0151e-02,  4.2070e-03,  1.2242e-02, -1.5212e-03,\n",
       "         1.9653e-03,  1.1655e-02,  5.0363e-03, -9.4347e-03,  7.7629e-03,\n",
       "         5.9542e-03,  1.7366e-02, -9.4629e-03,  2.8023e-03, -3.9559e-03,\n",
       "        -3.3829e-03,  4.9462e-03, -4.8096e-03, -1.2516e-02, -2.3326e-02,\n",
       "        -7.8583e-03, -8.2508e-03, -3.5378e-03, -8.6313e-03,  1.0299e-03,\n",
       "        -1.6666e-03,  8.6152e-03,  3.6862e-03,  5.8169e-03,  6.4640e-03,\n",
       "         5.6540e-03,  2.0746e-02, -1.1908e-02, -6.3926e-03, -2.0127e-03,\n",
       "        -6.7338e-03, -1.0632e-02, -5.2863e-03,  6.4934e-04, -8.3847e-03,\n",
       "         1.7475e-03, -3.2313e-04,  1.7507e-03,  2.3031e-03,  6.5633e-03,\n",
       "         1.4137e-03,  1.0554e-02,  3.2409e-03, -1.4862e-02,  1.5629e-03,\n",
       "         4.2854e-04,  1.4801e-03, -5.3407e-03,  1.6255e-02, -1.1445e-02,\n",
       "        -8.3218e-03, -1.1026e-02, -1.0121e-04,  6.2170e-03, -4.9521e-03,\n",
       "         2.5649e-02, -2.4352e-02,  2.3909e-02,  7.1705e-03, -3.1413e-03,\n",
       "        -2.5907e-02,  4.8740e-03, -6.3159e-03,  2.9530e-04,  8.2044e-03,\n",
       "         8.7742e-05, -1.1437e-02, -1.6152e-02, -3.3668e-03,  1.1797e-02,\n",
       "         7.5508e-03,  4.7729e-03, -8.5734e-03, -1.4269e-03,  4.3756e-03,\n",
       "         2.9048e-03], device='cuda:0'), tensor([[[[-2.2577e-02,  1.3752e-02,  9.4669e-03],\n",
       "          [ 3.6387e-03,  1.7359e-02, -5.4821e-03],\n",
       "          [ 8.5312e-03,  8.9017e-03, -6.1334e-03]],\n",
       "\n",
       "         [[ 5.9313e-03, -1.2578e-03, -7.4252e-03],\n",
       "          [-9.4587e-03,  3.4107e-03,  1.5687e-02],\n",
       "          [ 4.3449e-03,  6.7849e-03,  8.0932e-03]],\n",
       "\n",
       "         [[-1.9989e-03,  9.6008e-04,  2.3653e-03],\n",
       "          [ 8.1109e-04,  6.3938e-03, -1.1926e-02],\n",
       "          [-9.8224e-04, -5.9111e-03, -1.5283e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-3.2141e-03,  1.5431e-02,  3.7192e-03],\n",
       "          [ 2.3710e-03, -8.7899e-03,  6.4978e-03],\n",
       "          [-8.2871e-04, -4.1269e-03,  2.2275e-02]],\n",
       "\n",
       "         [[ 3.0385e-04, -4.2339e-03, -9.6661e-03],\n",
       "          [ 4.9041e-03,  9.8383e-03,  6.2071e-03],\n",
       "          [ 7.8748e-03, -1.2086e-02, -2.8931e-03]],\n",
       "\n",
       "         [[-8.7342e-03, -9.0896e-03,  7.6345e-03],\n",
       "          [-5.5363e-03,  5.4211e-03,  4.8957e-03],\n",
       "          [ 2.4807e-03,  1.6030e-02, -3.0804e-05]]],\n",
       "\n",
       "\n",
       "        [[[-1.1303e-02,  1.2298e-02,  5.2299e-03],\n",
       "          [ 1.4192e-02,  3.7610e-03,  5.3705e-03],\n",
       "          [-5.8305e-04,  8.4430e-03,  3.2458e-03]],\n",
       "\n",
       "         [[-1.4858e-02,  1.0162e-02,  1.8314e-02],\n",
       "          [ 6.6231e-03, -3.3236e-03,  2.8650e-03],\n",
       "          [-6.6212e-03, -1.2637e-02, -1.4166e-02]],\n",
       "\n",
       "         [[-8.6743e-03,  1.3664e-02, -1.4015e-03],\n",
       "          [ 6.6737e-03, -8.8442e-03, -1.3545e-02],\n",
       "          [-2.7413e-03,  2.6337e-02, -1.8139e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-6.8344e-04,  4.5615e-03,  9.7561e-03],\n",
       "          [ 1.0614e-02, -6.7605e-03,  4.6051e-03],\n",
       "          [ 8.5541e-03,  7.9647e-03, -6.4417e-03]],\n",
       "\n",
       "         [[ 2.1992e-03,  1.3127e-03, -7.7563e-03],\n",
       "          [-1.9661e-02,  1.0211e-02,  3.7399e-03],\n",
       "          [-1.3801e-02,  9.4038e-03, -2.6139e-03]],\n",
       "\n",
       "         [[-2.9893e-03,  8.6078e-03,  1.1759e-02],\n",
       "          [-4.5477e-03,  1.9337e-02, -4.2921e-03],\n",
       "          [ 1.0102e-02, -1.6870e-02, -5.2084e-03]]],\n",
       "\n",
       "\n",
       "        [[[ 2.1383e-03,  1.2176e-02,  1.1591e-02],\n",
       "          [ 3.7577e-03, -5.4832e-03,  1.0140e-02],\n",
       "          [-8.7920e-03,  1.0349e-02,  1.4100e-03]],\n",
       "\n",
       "         [[-8.0345e-03, -4.6266e-03, -9.5116e-03],\n",
       "          [-8.1259e-03,  4.6856e-03, -1.2670e-02],\n",
       "          [ 8.5676e-03,  3.3729e-03, -1.7488e-02]],\n",
       "\n",
       "         [[-1.5513e-02, -1.1282e-03, -9.7890e-03],\n",
       "          [ 4.4102e-03, -9.0547e-04,  1.6271e-02],\n",
       "          [ 2.0500e-03, -1.7788e-02,  2.8471e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 4.2100e-03,  1.3155e-03, -9.9031e-03],\n",
       "          [-6.1505e-03, -1.0601e-02,  8.0671e-03],\n",
       "          [ 3.4054e-03,  3.8037e-03,  1.1634e-02]],\n",
       "\n",
       "         [[ 2.1622e-03, -1.3644e-02,  3.5335e-05],\n",
       "          [-6.9934e-03,  6.1348e-03,  8.0901e-03],\n",
       "          [-2.0602e-03, -3.2925e-03, -9.2595e-03]],\n",
       "\n",
       "         [[-1.8061e-02, -1.5700e-02,  1.2516e-02],\n",
       "          [ 3.1726e-03,  7.3061e-03,  8.7200e-03],\n",
       "          [ 8.1189e-03,  2.7638e-03, -1.3824e-04]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[-1.4963e-02, -1.5370e-03, -1.3911e-03],\n",
       "          [-1.5820e-02,  2.0995e-02, -2.5291e-03],\n",
       "          [ 1.0740e-02, -2.5800e-02, -9.3360e-03]],\n",
       "\n",
       "         [[ 3.9284e-03,  1.2789e-02,  1.5819e-02],\n",
       "          [-1.6344e-02,  5.5785e-03,  1.6333e-02],\n",
       "          [-2.4533e-03, -6.3436e-03,  2.1084e-02]],\n",
       "\n",
       "         [[ 5.3413e-03,  2.5761e-03,  8.7371e-03],\n",
       "          [ 8.6458e-03,  7.3298e-03, -7.8313e-03],\n",
       "          [ 1.2461e-03,  3.4754e-03, -5.0456e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-9.6646e-04,  3.0999e-03,  7.7567e-03],\n",
       "          [-5.7740e-03, -1.2985e-02,  1.9549e-02],\n",
       "          [ 7.6221e-03,  4.6445e-03,  1.7451e-02]],\n",
       "\n",
       "         [[-5.7525e-03,  3.7729e-04, -4.9639e-03],\n",
       "          [ 1.0246e-03, -3.3730e-03, -2.1512e-02],\n",
       "          [-8.5586e-04, -6.6154e-03, -3.9092e-03]],\n",
       "\n",
       "         [[-1.9246e-03, -1.4564e-02, -1.9036e-02],\n",
       "          [-9.3561e-03, -1.4404e-02,  1.3393e-02],\n",
       "          [-4.9639e-03, -1.8366e-03, -4.1003e-03]]],\n",
       "\n",
       "\n",
       "        [[[-1.6255e-02,  4.7740e-03,  1.9306e-02],\n",
       "          [ 1.8253e-02, -1.1207e-02,  1.3150e-02],\n",
       "          [ 1.2778e-02,  8.3874e-03, -1.6788e-02]],\n",
       "\n",
       "         [[ 6.8362e-05,  5.5745e-03,  7.4256e-03],\n",
       "          [ 1.2441e-03,  6.3926e-03, -5.4849e-03],\n",
       "          [-8.8234e-04,  1.2208e-02, -2.6358e-04]],\n",
       "\n",
       "         [[ 1.5345e-02,  1.8099e-03, -1.0433e-02],\n",
       "          [ 2.9189e-03, -4.4570e-03, -1.0821e-03],\n",
       "          [ 5.1586e-03,  8.7981e-04,  6.9227e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-1.5569e-04, -7.8585e-03,  2.6815e-02],\n",
       "          [ 2.4596e-03, -6.5338e-03,  6.0459e-03],\n",
       "          [ 5.6815e-03, -3.3566e-03,  2.2462e-03]],\n",
       "\n",
       "         [[ 3.6743e-03,  1.1892e-02, -1.5929e-02],\n",
       "          [ 4.7280e-03, -5.0710e-04,  3.0199e-02],\n",
       "          [-1.1956e-02, -6.2837e-03,  4.9985e-03]],\n",
       "\n",
       "         [[ 6.7929e-03, -4.1725e-03,  5.7713e-03],\n",
       "          [-1.4068e-04,  1.5702e-02, -3.6710e-04],\n",
       "          [-2.2866e-03, -1.3088e-02, -4.8885e-03]]],\n",
       "\n",
       "\n",
       "        [[[-6.5935e-03,  2.5413e-02, -1.3487e-03],\n",
       "          [ 2.9170e-04, -1.3131e-02, -4.0525e-03],\n",
       "          [-8.0160e-03, -2.0305e-02, -7.8108e-03]],\n",
       "\n",
       "         [[ 8.3917e-03,  1.4972e-02,  6.5452e-03],\n",
       "          [-1.0956e-02, -3.9994e-03,  4.4024e-03],\n",
       "          [-2.0115e-02,  1.8780e-02,  6.1073e-04]],\n",
       "\n",
       "         [[ 6.4319e-03, -8.2514e-03,  1.8477e-02],\n",
       "          [ 8.5547e-03, -1.0917e-02,  1.5456e-02],\n",
       "          [-2.4350e-02,  4.9577e-03, -3.6737e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 8.3394e-03, -9.1252e-03,  1.9758e-03],\n",
       "          [-7.7243e-03, -5.7571e-03, -7.2245e-03],\n",
       "          [ 1.0005e-02, -6.8252e-04,  5.5158e-03]],\n",
       "\n",
       "         [[ 9.4698e-03, -6.8992e-03,  1.0032e-02],\n",
       "          [-2.3140e-02,  8.1288e-04, -7.4484e-03],\n",
       "          [-2.5843e-02, -4.4504e-03,  6.6935e-03]],\n",
       "\n",
       "         [[ 2.5080e-02, -8.4016e-04,  9.3788e-03],\n",
       "          [-1.0645e-02, -3.6199e-03, -5.7187e-03],\n",
       "          [ 2.0026e-02, -4.4207e-03, -3.6430e-04]]]], device='cuda:0'), tensor([-6.7759e-04,  1.2041e-02, -1.6590e-03, -1.0698e-02, -7.8410e-03,\n",
       "         4.9076e-03,  2.1759e-03,  1.5727e-02, -4.6532e-03,  7.0635e-04,\n",
       "        -3.0785e-04,  7.8696e-05,  3.8150e-03, -4.3376e-03, -3.3254e-03,\n",
       "         1.0871e-02, -1.3007e-02, -1.0805e-02, -1.1373e-03, -6.8339e-04,\n",
       "        -3.3606e-03, -1.8644e-03, -1.8350e-02, -4.4347e-03, -9.1250e-03,\n",
       "         8.9531e-03,  5.2840e-03,  2.3572e-02, -1.7844e-03, -6.8344e-03,\n",
       "         1.3913e-03,  9.2256e-03,  3.2792e-02,  6.0510e-03,  1.3257e-02,\n",
       "        -1.2161e-03, -9.5303e-03, -9.6818e-03,  1.2606e-02, -4.8458e-03,\n",
       "        -2.9614e-03,  8.9978e-03, -9.2398e-03,  1.4360e-02, -1.4414e-02,\n",
       "         2.1822e-03, -1.2910e-02, -1.3800e-02,  8.3765e-03, -8.7672e-03,\n",
       "         7.4281e-03, -6.3465e-03, -9.9576e-03,  1.0874e-03, -7.2254e-03,\n",
       "        -1.8199e-02, -1.9902e-02, -1.0826e-02, -1.1232e-02,  2.8994e-03,\n",
       "         1.5326e-03, -7.1506e-03,  6.9281e-03,  1.6130e-02,  3.4065e-03,\n",
       "        -4.8435e-03,  1.4476e-03,  7.9059e-03, -1.6404e-03,  1.0517e-02,\n",
       "        -1.0452e-03, -1.1126e-02,  6.8192e-03, -3.1650e-03, -8.0826e-03,\n",
       "        -1.3040e-02, -1.2721e-02,  1.1785e-02,  1.4061e-02,  2.1057e-02,\n",
       "        -4.4569e-03,  1.4884e-03, -3.9420e-03, -7.3112e-03, -2.4770e-03,\n",
       "         1.5542e-03, -1.2518e-02, -4.8114e-03, -4.0942e-03,  6.3059e-03,\n",
       "         2.3065e-02, -9.2746e-03, -1.2433e-03,  3.9928e-03, -6.2737e-03,\n",
       "         2.3164e-03, -1.3901e-03,  2.5015e-03,  6.9786e-04,  4.5581e-03,\n",
       "         1.3526e-02, -2.7132e-02,  4.7898e-03, -1.1388e-02, -1.2071e-04,\n",
       "        -1.9809e-02, -5.7392e-03, -2.5664e-03, -1.1284e-02, -1.4224e-02,\n",
       "        -2.1295e-03, -5.5623e-03, -5.6572e-03,  4.2700e-03, -6.6753e-03,\n",
       "         5.0691e-03, -1.9005e-02, -2.0679e-02,  1.4014e-02, -7.4674e-03,\n",
       "        -8.7574e-03,  1.4160e-02, -1.3117e-03,  2.1643e-02, -1.7267e-02,\n",
       "        -3.2952e-03, -3.5703e-03, -1.0084e-02, -6.7689e-03,  6.0037e-03,\n",
       "         1.3757e-02, -1.2157e-02,  8.9548e-03,  1.5282e-03, -1.5394e-02,\n",
       "         8.4084e-03,  6.6124e-03, -1.3357e-02, -4.7085e-03,  1.5089e-03,\n",
       "         6.0776e-03,  1.8225e-02, -1.7599e-02, -1.4113e-02, -3.6203e-03,\n",
       "        -2.1515e-03,  7.6493e-04, -1.7721e-02, -1.1621e-02, -2.0004e-03,\n",
       "        -1.1386e-02,  2.3822e-03,  3.7914e-03,  1.2162e-02,  4.8907e-03,\n",
       "        -1.1631e-02, -7.8541e-03,  7.7622e-03,  1.1851e-02, -6.6761e-03,\n",
       "        -8.6786e-03, -6.6293e-03,  1.4140e-02, -5.2167e-03,  4.0482e-04,\n",
       "        -7.8944e-03, -6.6092e-03, -7.8592e-03, -1.9045e-02,  5.2217e-05,\n",
       "        -1.2769e-02, -1.4265e-03, -4.5979e-04,  1.0137e-02,  1.0404e-02,\n",
       "         8.5976e-03,  8.1777e-04,  7.8252e-03,  1.8691e-02, -8.7886e-03,\n",
       "         1.1380e-02,  1.2251e-02, -5.6670e-04, -1.2171e-02,  5.2281e-03,\n",
       "         5.8637e-03, -2.2913e-03, -1.1067e-02, -5.4233e-04,  3.2188e-03,\n",
       "        -1.0593e-02,  1.6139e-03, -4.3166e-03,  2.0942e-03,  7.5406e-03,\n",
       "         6.3227e-03, -6.7015e-03,  7.5394e-03, -1.2230e-02,  1.9165e-02,\n",
       "        -4.4532e-03, -5.7069e-03,  8.2898e-03,  8.4253e-03, -4.2025e-03,\n",
       "         9.1160e-03,  2.0954e-02, -3.5228e-03,  8.9240e-03,  1.3213e-02,\n",
       "        -1.4677e-02, -4.5028e-03, -9.1975e-03, -3.8003e-03,  1.0251e-02,\n",
       "        -1.7963e-02, -4.8891e-03, -1.5046e-02,  1.1705e-02,  1.1389e-03,\n",
       "         1.7228e-02,  3.9309e-03, -1.6550e-02,  1.3195e-02,  3.4193e-03,\n",
       "         2.5520e-03, -2.3196e-02,  3.7158e-03, -4.6952e-03, -2.2382e-03,\n",
       "        -4.9378e-03,  4.1521e-03,  1.0710e-02,  9.5778e-03,  3.0023e-03,\n",
       "         2.3439e-03, -6.4624e-03,  8.3212e-03, -1.0569e-02, -7.1209e-03,\n",
       "         9.6899e-03,  9.9931e-04, -4.1145e-03,  3.1821e-03,  1.1769e-02,\n",
       "         6.6671e-03,  5.3877e-03,  2.4393e-02, -1.9245e-03,  9.7934e-03,\n",
       "         4.8610e-03,  4.7919e-03, -1.7836e-02, -1.2454e-02,  8.6647e-03,\n",
       "         3.2333e-03, -2.4879e-03,  1.5768e-03, -1.9166e-02, -4.2487e-03,\n",
       "         1.3226e-02,  1.0320e-02,  6.4239e-03,  1.1212e-02, -2.4476e-03,\n",
       "         2.4964e-03,  4.7252e-03, -3.7882e-03, -6.6334e-03,  3.9050e-03,\n",
       "        -1.0048e-02, -1.7966e-02,  1.4141e-02, -2.1963e-02,  1.9095e-02,\n",
       "        -8.0450e-03, -2.4596e-03, -6.5952e-03,  4.5005e-03, -1.2388e-03,\n",
       "         1.3272e-02, -1.8786e-03,  6.5778e-03,  6.2278e-03,  5.0462e-03,\n",
       "        -8.9649e-03,  1.7204e-02,  5.4864e-03, -1.1540e-03,  1.0217e-02,\n",
       "        -4.7130e-03, -8.6831e-03,  7.6332e-03,  1.6385e-02,  4.9615e-03,\n",
       "        -8.3629e-03,  7.4808e-03, -7.7128e-03, -6.3401e-03,  1.1465e-03,\n",
       "         4.7864e-03, -1.0019e-02, -2.0175e-02,  1.2768e-03, -8.8003e-03,\n",
       "        -5.5279e-04,  1.0266e-02,  3.3084e-03, -5.1200e-03, -1.1078e-02,\n",
       "         1.6071e-04, -1.2931e-02, -2.0965e-03,  1.2221e-03,  1.4521e-02,\n",
       "        -4.8268e-03, -3.8262e-03, -1.9457e-02, -1.2654e-02, -1.5494e-03,\n",
       "         8.1408e-03, -7.8999e-03,  7.5929e-03, -2.7453e-03, -9.6311e-03,\n",
       "         1.8432e-02,  1.8749e-02, -9.2564e-03,  1.7644e-03,  4.5663e-03,\n",
       "        -5.2489e-03,  1.2049e-02,  2.1367e-03,  1.4381e-03,  4.4024e-03,\n",
       "         7.7524e-03,  6.9057e-03, -5.3155e-04, -7.3851e-06, -1.1253e-03,\n",
       "        -6.4602e-03, -1.1517e-02, -2.9374e-03, -2.2767e-03, -5.8742e-03,\n",
       "        -1.0475e-02,  1.0117e-02, -8.8193e-03,  3.6906e-03, -9.5876e-03,\n",
       "         1.0213e-03,  2.4575e-02, -1.1154e-02,  1.9300e-02,  6.2849e-03,\n",
       "        -4.7895e-03, -6.0887e-03, -1.0844e-02, -6.3748e-03,  2.2022e-02,\n",
       "        -1.2604e-02, -7.2845e-03, -4.3954e-03,  1.0537e-02,  3.0036e-03,\n",
       "        -9.4097e-03, -1.2739e-02, -2.0651e-02, -9.0954e-03,  1.3203e-02,\n",
       "         3.3389e-03,  1.4698e-02, -8.3908e-03,  2.0110e-02,  1.0333e-02,\n",
       "         3.8185e-03, -9.0457e-03, -2.1010e-03,  1.1743e-03,  5.7077e-04,\n",
       "         7.8588e-03, -2.3100e-02, -1.2563e-02,  1.4774e-03, -2.3934e-03,\n",
       "         1.4216e-02,  5.2085e-03,  2.2960e-02,  5.1234e-03, -6.3353e-03,\n",
       "        -4.8556e-03, -3.6225e-03,  5.7142e-04,  9.3165e-03, -1.8796e-03,\n",
       "        -6.6452e-03, -1.9335e-02,  1.9464e-02, -1.9803e-02, -1.5167e-02,\n",
       "        -2.5812e-02, -5.8817e-03, -1.1167e-02, -3.7556e-03,  8.6543e-03,\n",
       "         9.9251e-03,  1.7472e-02,  1.1432e-03,  9.1463e-03, -2.1392e-03,\n",
       "        -1.7477e-02, -2.9718e-03,  5.4751e-03,  2.8269e-02,  2.7534e-03,\n",
       "         6.0553e-03,  7.6218e-03, -1.5813e-02, -1.2227e-02, -8.3536e-03,\n",
       "         5.8386e-03,  6.2199e-03, -2.6587e-02, -5.7095e-03,  4.9386e-03,\n",
       "        -1.7583e-03,  6.5020e-03,  9.5534e-03, -1.2912e-02, -1.4351e-02,\n",
       "         1.9823e-03,  5.9171e-03,  1.1244e-02, -4.4970e-03, -9.2002e-03,\n",
       "        -5.6647e-03, -1.5230e-02,  1.3336e-02, -6.2698e-03, -5.2448e-03,\n",
       "         9.9902e-04,  5.6441e-03, -6.6273e-03, -1.0443e-02,  3.3530e-03,\n",
       "         2.4265e-03,  5.6859e-03, -1.1305e-02, -4.9574e-03,  4.9728e-03,\n",
       "         5.8559e-03, -5.1765e-03, -1.1811e-03,  1.6218e-02, -6.6046e-03,\n",
       "         1.4992e-03, -1.2191e-02, -1.4619e-02,  6.2027e-04,  1.2354e-02,\n",
       "        -2.8846e-03,  6.1893e-03, -7.6799e-03, -9.3246e-03, -1.3029e-02,\n",
       "         7.1681e-03,  5.1660e-03, -1.5185e-02,  6.0130e-03,  5.0153e-04,\n",
       "         8.0003e-04, -1.0118e-02,  4.6792e-03,  1.5578e-02,  7.8111e-03,\n",
       "        -7.5567e-04, -3.9571e-03, -1.2185e-02, -1.8983e-03, -5.0355e-03,\n",
       "         1.2306e-03, -4.3995e-03, -8.0946e-04,  6.8891e-03, -6.2868e-03,\n",
       "        -1.6024e-02, -5.4730e-03,  1.2474e-03,  3.5039e-03, -6.4948e-03,\n",
       "        -3.1854e-03,  1.6234e-03, -7.4641e-03, -7.9200e-03,  7.5682e-03,\n",
       "         3.3396e-04, -7.7865e-03, -1.2124e-02, -5.0640e-03, -2.3520e-02,\n",
       "        -2.4356e-02,  1.0388e-02, -2.8703e-02,  5.5318e-03,  7.7647e-03,\n",
       "         8.9240e-03, -2.8150e-03,  1.6793e-02, -4.5372e-03,  4.8573e-03,\n",
       "        -1.8615e-02,  2.0666e-02], device='cuda:0'), tensor([-5.3980e-03, -7.6513e-03, -2.0980e-02, -1.2326e-02, -1.8911e-02,\n",
       "        -3.7952e-03,  9.6768e-03, -1.6028e-02, -2.7005e-03, -6.8861e-03,\n",
       "         8.5946e-03, -1.0484e-03,  3.6995e-03, -6.7388e-03,  8.0159e-03,\n",
       "         5.4557e-03, -1.1712e-02,  1.4343e-02, -9.3835e-03, -1.3396e-02,\n",
       "         3.7427e-03, -4.2903e-03, -4.7663e-04, -7.1286e-03,  2.9564e-03,\n",
       "         1.4185e-02,  1.6740e-02,  1.5325e-02, -3.3529e-03, -2.7502e-03,\n",
       "        -1.5115e-02,  9.3197e-03, -1.9406e-03,  2.5710e-03, -4.5776e-03,\n",
       "         2.9894e-03,  1.2146e-02,  1.1013e-02,  1.1692e-02,  4.0114e-03,\n",
       "        -7.4208e-03, -3.0938e-03, -6.8580e-03,  2.2026e-03,  2.8330e-03,\n",
       "        -1.0311e-02,  4.8472e-03, -1.9708e-02,  6.2773e-03, -5.7951e-03,\n",
       "         4.6021e-03,  1.9899e-02,  6.7271e-04, -1.1429e-03,  4.3696e-04,\n",
       "        -3.4915e-05, -2.8688e-02,  2.0770e-03, -2.0274e-02, -3.0607e-03,\n",
       "         1.1190e-02, -1.1923e-02, -1.5233e-03, -3.2035e-03, -4.7901e-03,\n",
       "         8.0135e-04,  6.7765e-03,  1.0137e-02, -5.3969e-03,  7.0996e-04,\n",
       "         1.6859e-02, -4.8915e-03, -1.4828e-03,  5.7709e-03, -7.4082e-03,\n",
       "         1.2510e-02,  1.8698e-02,  1.2912e-03,  4.0772e-03,  3.0765e-03,\n",
       "        -2.1886e-03,  9.2119e-03,  1.7420e-03, -3.0957e-03,  2.0312e-02,\n",
       "        -3.1380e-03,  1.1616e-03, -1.2802e-02, -3.2385e-03, -2.5638e-03,\n",
       "        -6.8381e-03,  3.2043e-03, -1.4070e-02, -4.9708e-03, -1.2106e-02,\n",
       "        -6.2910e-03, -5.7759e-03,  1.1000e-02, -1.4667e-02, -8.7754e-03,\n",
       "        -7.9639e-03,  1.9101e-02, -3.7570e-04, -4.9321e-03,  8.6374e-03,\n",
       "         1.5178e-02, -3.7387e-05,  1.0560e-02, -1.3513e-02, -5.4365e-03,\n",
       "         8.7435e-03, -7.1671e-03, -1.3557e-02,  1.8722e-02,  7.6180e-03,\n",
       "        -9.3350e-03,  6.3636e-03, -1.9027e-03, -1.4752e-03, -1.9061e-03,\n",
       "        -5.8707e-03,  1.6870e-02,  6.7704e-03,  1.2696e-02,  6.9097e-03,\n",
       "         7.0438e-03,  7.2195e-03, -9.2946e-04,  6.0455e-03, -5.4452e-03,\n",
       "        -5.2497e-03,  9.9053e-03, -2.9991e-03,  4.9627e-03,  1.1181e-02,\n",
       "        -2.3209e-03, -8.1398e-03,  9.5535e-03,  1.1361e-02, -2.0861e-02,\n",
       "         6.8888e-03,  1.7372e-04, -3.2807e-03, -3.3823e-03, -8.9327e-03,\n",
       "        -2.6685e-03,  7.6243e-04, -4.8310e-03,  5.6696e-03, -9.5953e-03,\n",
       "        -1.1238e-02,  4.1899e-03, -6.1378e-03,  4.1433e-03, -5.9796e-03,\n",
       "        -1.6088e-02, -5.5784e-03,  1.8736e-02,  2.2434e-03,  1.4680e-02,\n",
       "         1.4116e-02,  5.9149e-03,  3.3467e-03, -7.9750e-03,  9.2047e-03,\n",
       "         1.5782e-02, -4.7414e-03,  1.4819e-02,  1.0361e-02,  7.7388e-03,\n",
       "         9.5219e-03, -1.7233e-02,  6.7424e-03, -9.3087e-03,  1.0890e-02,\n",
       "         4.0571e-03,  1.1460e-02,  6.8272e-03, -8.3254e-03,  2.5235e-02,\n",
       "        -1.2939e-02,  1.5110e-02,  6.6216e-03,  2.4084e-03,  5.3580e-04,\n",
       "        -2.1319e-03,  4.8108e-03, -2.8116e-03, -2.1196e-02, -2.0488e-02,\n",
       "         4.0743e-03,  1.0813e-02,  1.1764e-02, -6.7408e-04, -4.0359e-03,\n",
       "         5.8072e-03, -2.7047e-03,  4.1309e-03,  3.1105e-03, -1.6287e-03,\n",
       "         1.0710e-02,  1.3353e-02,  3.1308e-04,  4.6809e-03,  9.1547e-03,\n",
       "        -1.6778e-03, -9.9488e-03,  2.5271e-03,  8.2455e-03,  6.3620e-03,\n",
       "         1.1475e-02, -8.9750e-03,  9.0956e-03,  7.9138e-03,  1.5999e-02,\n",
       "        -5.4557e-03, -9.4713e-03, -1.7441e-02,  1.0362e-03, -6.8310e-03,\n",
       "        -7.4623e-03,  9.7798e-03, -5.9151e-04, -5.5151e-03,  3.2153e-03,\n",
       "         1.9596e-03, -1.7008e-02,  8.0801e-03, -1.5166e-02,  1.9814e-03,\n",
       "         1.1639e-02, -1.6111e-02, -8.1803e-03,  4.1010e-03, -1.2344e-02,\n",
       "         2.9196e-02, -4.1740e-03, -1.8294e-02,  1.5525e-02,  2.8216e-03,\n",
       "        -3.4965e-03,  5.0585e-03, -5.3674e-03, -1.8164e-02,  1.0457e-02,\n",
       "        -3.1063e-03, -3.2726e-03, -3.3779e-03,  2.7592e-02, -1.0196e-02,\n",
       "        -6.8025e-03, -7.7347e-03, -9.9138e-04,  3.5392e-03, -7.4569e-03,\n",
       "        -3.7808e-03,  4.8628e-03,  6.3506e-03, -3.4560e-03, -7.4036e-03,\n",
       "        -5.8001e-03, -7.6715e-03, -1.6449e-02,  1.3783e-02,  4.4572e-03,\n",
       "         9.2683e-03, -6.8423e-04,  7.6642e-03, -8.3741e-03, -1.2681e-02,\n",
       "         1.4408e-02,  1.2876e-04,  1.0896e-02, -1.4172e-02, -6.2663e-03,\n",
       "        -9.7566e-03, -8.7884e-03, -8.7713e-03,  1.4942e-02, -8.7657e-04,\n",
       "        -8.0872e-05, -2.0837e-03, -5.5590e-03, -5.1167e-03, -8.5668e-03,\n",
       "         3.5958e-03,  1.8649e-02,  1.0451e-02,  2.7044e-03,  2.0587e-04,\n",
       "        -3.4288e-03,  1.9417e-02, -8.4643e-03,  7.1253e-03,  5.8255e-03,\n",
       "        -1.5291e-02, -1.6628e-02, -3.0238e-03, -3.3736e-03, -2.1431e-03,\n",
       "         1.2618e-05, -5.7458e-03, -1.4400e-02, -9.0705e-03,  1.5167e-03,\n",
       "        -9.6770e-04,  8.0493e-03,  8.9582e-03, -7.0918e-03,  3.3556e-04,\n",
       "         5.0714e-03,  2.6998e-03,  2.2703e-03,  9.0681e-03,  1.4512e-02,\n",
       "        -2.8572e-03, -1.6030e-02, -1.2071e-02, -8.9885e-04,  2.3622e-02,\n",
       "        -1.2238e-02,  8.8177e-03, -1.3291e-04,  7.6743e-03, -1.2186e-03,\n",
       "        -6.5526e-03, -2.3930e-03,  5.7917e-03,  4.9712e-03,  2.9386e-03,\n",
       "         1.0679e-02,  1.7064e-02,  3.7121e-04, -3.1694e-03, -1.2950e-02,\n",
       "        -6.8499e-03,  8.1893e-03,  2.0150e-02, -1.0335e-02, -9.2031e-04,\n",
       "         1.4182e-02,  1.0776e-02, -1.5044e-02, -5.2439e-03, -5.6384e-04,\n",
       "         1.5562e-02, -1.3849e-02, -1.0461e-02, -2.8461e-03,  6.5053e-03,\n",
       "        -1.0842e-02, -4.1922e-03, -5.2208e-03, -9.7541e-03, -4.7982e-03,\n",
       "         4.9302e-03, -6.3034e-03,  9.7018e-03,  1.0231e-02, -5.3996e-03,\n",
       "         9.6955e-03,  1.6602e-02, -9.3634e-03,  2.2523e-02, -8.4379e-05,\n",
       "        -1.7992e-02,  1.4142e-02, -9.4607e-03,  4.0665e-03,  8.8017e-03,\n",
       "        -2.0787e-03,  7.1993e-03, -2.3422e-03, -1.1275e-02, -7.8989e-03,\n",
       "        -1.2289e-03,  1.8845e-02,  2.0363e-02,  1.3421e-02, -1.6787e-02,\n",
       "        -2.0591e-03,  1.5686e-02,  7.5205e-03, -3.5575e-03,  6.6110e-03,\n",
       "        -6.1160e-03, -5.8116e-03, -5.5581e-03,  8.0915e-03,  2.0875e-02,\n",
       "         6.4949e-03, -1.5413e-02,  1.5092e-02,  7.2799e-03, -6.8699e-03,\n",
       "         1.4911e-02,  5.9543e-03, -8.6391e-03,  1.1834e-03,  1.0988e-02,\n",
       "         1.3096e-02,  1.0092e-02, -8.8880e-03,  3.8488e-03, -7.6128e-03,\n",
       "        -7.9492e-04,  8.0106e-03, -1.7019e-02, -9.8391e-04,  3.7834e-03,\n",
       "         1.6312e-02,  6.0616e-03,  1.3716e-02, -1.8266e-03, -7.8851e-03,\n",
       "        -4.8226e-03, -5.2578e-03,  3.0887e-03,  7.7626e-03,  4.5338e-04,\n",
       "        -1.1954e-02, -7.9427e-03,  2.0185e-03,  2.5242e-03, -8.4806e-04,\n",
       "         1.2781e-02, -6.8868e-03, -6.4505e-03,  7.7546e-04, -1.0859e-03,\n",
       "        -5.9062e-04, -1.4762e-02, -2.4270e-05, -1.1604e-02,  1.7533e-03,\n",
       "         2.0224e-03,  8.0163e-03, -6.7898e-03,  1.9396e-03,  4.9651e-04,\n",
       "         1.0868e-02,  8.4347e-04, -1.8096e-02, -7.8396e-03, -1.6428e-02,\n",
       "        -7.8279e-03,  1.4188e-03, -6.2406e-03, -6.4245e-03, -1.8026e-03,\n",
       "         3.4440e-03, -1.2507e-02,  2.3704e-03, -7.7503e-03,  4.2754e-03,\n",
       "         9.6248e-04,  1.5730e-03,  9.3547e-03,  5.5148e-03,  1.2335e-02,\n",
       "        -9.2058e-03,  1.4892e-03,  2.1763e-02, -2.1048e-02, -1.1968e-03,\n",
       "        -5.1344e-03, -1.2631e-02,  3.4289e-03,  8.5711e-03, -1.2874e-02,\n",
       "        -7.0930e-04, -2.7246e-03, -1.7470e-02,  2.1749e-03, -7.9223e-04,\n",
       "        -1.8985e-03,  1.7027e-02, -2.7061e-03,  1.4907e-02, -1.6314e-02,\n",
       "         5.6560e-03, -9.1085e-03, -8.5220e-03,  1.1542e-02, -1.6584e-02,\n",
       "        -1.6172e-03, -6.8793e-03, -8.1936e-03,  1.8645e-02,  1.5956e-03,\n",
       "        -2.1283e-02,  2.3845e-03,  1.2354e-02,  1.2103e-02,  1.1517e-02,\n",
       "         5.7113e-03, -9.9206e-03,  4.9003e-03,  1.9723e-03, -1.0523e-02,\n",
       "         1.0689e-02,  1.6035e-02, -1.2544e-02, -5.3200e-03, -5.4737e-03,\n",
       "        -1.7662e-02, -5.0117e-03,  1.2381e-02, -1.1001e-02,  1.4282e-03,\n",
       "        -8.2914e-05,  5.9325e-03], device='cuda:0'), tensor([[[[ 4.4295e-03, -6.5151e-03,  7.5104e-03],\n",
       "          [-1.1725e-02,  3.1764e-03,  1.8794e-02],\n",
       "          [-1.5614e-02,  1.6506e-02,  8.2925e-03]],\n",
       "\n",
       "         [[-4.8624e-03, -1.3988e-02, -1.8934e-02],\n",
       "          [ 5.5885e-04,  1.4360e-02,  7.2004e-03],\n",
       "          [ 4.2170e-03,  1.0993e-02,  1.1784e-03]],\n",
       "\n",
       "         [[-1.4352e-03, -7.6109e-05, -3.2485e-03],\n",
       "          [ 6.4714e-03,  1.9360e-03, -2.8498e-03],\n",
       "          [ 1.6766e-02,  4.7048e-03,  1.0809e-04]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 3.3884e-03, -1.6587e-02,  7.7284e-04],\n",
       "          [-5.8177e-03, -2.7340e-03, -7.5699e-03],\n",
       "          [ 6.1630e-03, -2.3738e-03, -8.1450e-03]],\n",
       "\n",
       "         [[-6.5498e-03,  1.0691e-03,  9.3645e-03],\n",
       "          [-7.4012e-03,  1.8013e-02,  7.1279e-03],\n",
       "          [-4.9954e-03,  4.2305e-03, -1.5559e-02]],\n",
       "\n",
       "         [[ 1.2729e-02,  8.2617e-03,  1.5381e-02],\n",
       "          [ 2.0964e-03,  9.6238e-03,  6.2249e-03],\n",
       "          [-2.0626e-02, -8.6898e-03,  9.1291e-03]]],\n",
       "\n",
       "\n",
       "        [[[ 2.8246e-04, -3.3594e-03,  2.1100e-03],\n",
       "          [ 2.2762e-03,  1.2740e-02, -1.1169e-02],\n",
       "          [-1.2906e-02, -2.4654e-02, -1.0452e-03]],\n",
       "\n",
       "         [[ 6.3314e-03,  8.9839e-03, -6.1378e-03],\n",
       "          [ 1.1965e-02,  4.3106e-03, -9.9024e-03],\n",
       "          [ 1.5969e-02,  6.9246e-03,  5.3019e-03]],\n",
       "\n",
       "         [[-1.1885e-03, -1.8793e-02, -3.5427e-03],\n",
       "          [ 1.1091e-02,  9.9343e-04,  9.9237e-03],\n",
       "          [ 1.7823e-02,  6.8512e-03, -4.7806e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-3.9942e-03,  7.4151e-03,  5.3855e-03],\n",
       "          [ 6.2403e-03,  6.3885e-03, -1.4072e-02],\n",
       "          [ 4.0740e-04, -4.1221e-03, -2.5290e-02]],\n",
       "\n",
       "         [[ 1.0637e-03, -1.3088e-03,  5.4853e-03],\n",
       "          [ 5.2570e-03,  1.0038e-03, -6.0174e-03],\n",
       "          [-6.1422e-03, -3.4143e-03, -4.8268e-04]],\n",
       "\n",
       "         [[-6.2482e-03,  2.1080e-02,  6.0440e-03],\n",
       "          [-4.9168e-03, -4.0515e-03, -1.2512e-02],\n",
       "          [-3.4571e-04,  1.0047e-03,  1.0595e-02]]],\n",
       "\n",
       "\n",
       "        [[[-6.4050e-03,  4.9163e-03,  1.9019e-02],\n",
       "          [-3.4193e-03, -1.6695e-02, -4.3738e-03],\n",
       "          [ 9.3723e-03,  6.3111e-03, -1.7520e-03]],\n",
       "\n",
       "         [[-2.0983e-02,  1.0186e-03,  8.0269e-03],\n",
       "          [ 6.0186e-03,  1.9981e-04,  6.1965e-04],\n",
       "          [-6.4024e-03, -5.8889e-03,  7.7151e-03]],\n",
       "\n",
       "         [[-1.5637e-02,  7.9541e-03,  2.0852e-03],\n",
       "          [-5.7842e-03,  3.1390e-05, -2.9658e-03],\n",
       "          [-3.7257e-03,  4.5105e-03,  9.3714e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 1.2656e-03, -5.4300e-03,  1.1289e-02],\n",
       "          [ 1.4868e-02, -2.3313e-03, -5.8615e-03],\n",
       "          [-3.8636e-03,  1.4135e-02, -9.3912e-03]],\n",
       "\n",
       "         [[ 3.3039e-02, -4.6232e-03, -7.7379e-03],\n",
       "          [-1.3523e-02,  1.0006e-03, -1.3935e-03],\n",
       "          [ 5.5235e-03,  1.2763e-02, -1.0591e-02]],\n",
       "\n",
       "         [[-2.8813e-03, -1.0272e-02,  1.0805e-02],\n",
       "          [ 8.3352e-03,  1.8758e-02, -5.9340e-03],\n",
       "          [-4.1662e-03,  8.5611e-03, -1.7681e-02]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[ 3.7610e-03,  4.8340e-04,  1.1207e-02],\n",
       "          [ 7.0778e-03,  9.8191e-05,  2.0276e-03],\n",
       "          [-2.2616e-02, -8.7358e-03,  1.5336e-02]],\n",
       "\n",
       "         [[-1.1822e-02, -1.3924e-02,  3.2319e-03],\n",
       "          [-2.0360e-02,  1.0877e-03, -2.2793e-03],\n",
       "          [ 3.5027e-03, -3.1763e-03, -7.3209e-03]],\n",
       "\n",
       "         [[ 5.1533e-03, -1.0895e-02,  8.9568e-04],\n",
       "          [ 2.7975e-04,  2.6181e-03,  1.2327e-02],\n",
       "          [ 8.1404e-05,  1.0016e-02, -6.1140e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 1.0741e-02, -1.8951e-04,  1.6737e-02],\n",
       "          [-7.4780e-03,  9.1829e-03, -7.4867e-03],\n",
       "          [-1.3578e-02,  4.6959e-03, -1.5083e-02]],\n",
       "\n",
       "         [[-1.0460e-02, -1.3338e-02,  1.4956e-03],\n",
       "          [ 1.6324e-03, -9.0038e-03, -6.2749e-04],\n",
       "          [ 5.1703e-03, -7.7685e-04,  9.8520e-03]],\n",
       "\n",
       "         [[ 2.6528e-03,  2.7373e-03,  1.2620e-02],\n",
       "          [ 7.3571e-03,  7.5492e-03,  6.0705e-03],\n",
       "          [-1.3991e-02, -1.5798e-02, -8.1872e-03]]],\n",
       "\n",
       "\n",
       "        [[[ 1.1710e-03,  5.8826e-03,  4.6253e-03],\n",
       "          [-4.2032e-03,  2.0518e-03, -2.1809e-03],\n",
       "          [-6.4489e-03, -7.0525e-03, -1.8223e-02]],\n",
       "\n",
       "         [[ 3.4029e-03,  1.8150e-03,  7.6292e-03],\n",
       "          [ 9.0889e-03, -8.5812e-03,  1.1801e-02],\n",
       "          [-3.0625e-03,  1.6383e-02,  9.2410e-04]],\n",
       "\n",
       "         [[ 3.9754e-04, -1.6603e-03, -1.5101e-02],\n",
       "          [ 3.9128e-03,  1.1401e-02,  1.8282e-02],\n",
       "          [-1.8870e-02,  1.1925e-02,  2.7582e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 1.1434e-02,  2.4115e-03, -3.6959e-03],\n",
       "          [-9.8995e-04,  4.0947e-03, -1.5199e-02],\n",
       "          [ 1.3051e-02,  1.4892e-02, -1.1879e-02]],\n",
       "\n",
       "         [[ 1.8435e-02,  2.0349e-03, -1.6736e-03],\n",
       "          [-3.4443e-04, -2.5728e-03, -2.8167e-03],\n",
       "          [-4.1465e-03, -6.0623e-03,  7.8266e-03]],\n",
       "\n",
       "         [[-5.1711e-03, -1.7736e-03, -1.0210e-02],\n",
       "          [ 1.4473e-02,  1.1553e-02, -6.9340e-03],\n",
       "          [ 3.7257e-03,  8.5874e-03,  2.8965e-03]]],\n",
       "\n",
       "\n",
       "        [[[-5.9405e-03,  4.7215e-03,  9.9547e-03],\n",
       "          [-1.1582e-02,  9.1829e-03, -2.3016e-03],\n",
       "          [ 2.7842e-03, -4.0308e-03, -5.8050e-03]],\n",
       "\n",
       "         [[-3.4118e-03,  1.6678e-02,  1.3749e-03],\n",
       "          [-1.1832e-03,  8.5889e-03, -9.8722e-03],\n",
       "          [-7.8328e-03, -3.2416e-03,  1.3458e-02]],\n",
       "\n",
       "         [[-1.1110e-02,  1.5564e-02, -8.4402e-03],\n",
       "          [-2.4897e-04,  3.4906e-03,  1.9843e-02],\n",
       "          [-1.8505e-03, -3.9559e-03,  1.3881e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-1.4367e-03, -4.0698e-03,  1.1277e-02],\n",
       "          [ 9.1511e-03, -2.2257e-03,  1.4435e-02],\n",
       "          [-2.9068e-04, -3.1700e-02,  1.1866e-02]],\n",
       "\n",
       "         [[-1.1363e-02,  2.8021e-03,  5.5565e-03],\n",
       "          [ 2.1851e-03, -1.1253e-02,  8.0416e-03],\n",
       "          [ 1.4267e-03, -1.7411e-03, -9.1473e-03]],\n",
       "\n",
       "         [[ 1.8848e-03,  3.9590e-03, -1.5315e-03],\n",
       "          [ 1.1930e-02, -5.2985e-03,  4.9775e-03],\n",
       "          [ 1.2890e-02,  2.7654e-03,  8.5185e-03]]]], device='cuda:0'), tensor([ 1.2604e-02, -4.8141e-03,  1.3262e-02,  8.8881e-03,  6.4478e-03,\n",
       "        -3.5175e-04, -1.4527e-02, -1.0369e-03,  1.0245e-02, -2.4382e-03,\n",
       "        -1.7072e-02, -1.5387e-02,  1.6773e-03,  9.3352e-03, -1.0448e-02,\n",
       "        -7.1215e-03, -8.4437e-03, -6.8374e-03,  8.9706e-03, -2.8525e-03,\n",
       "         1.8655e-03,  1.4995e-02, -3.7437e-03, -7.4757e-03,  8.2533e-03,\n",
       "        -9.0432e-03,  1.1362e-02, -3.4551e-03,  9.0263e-03, -1.4609e-02,\n",
       "         1.9091e-03, -4.9598e-03, -1.5864e-02,  2.5946e-03,  1.7064e-03,\n",
       "        -1.2361e-02,  8.6391e-03,  1.2194e-02, -1.3327e-02,  2.2313e-02,\n",
       "         1.5162e-02,  1.8710e-02, -1.0413e-02,  1.8126e-02, -8.7324e-04,\n",
       "        -2.6980e-03,  1.0315e-03,  1.5621e-02, -3.1206e-03,  6.3217e-04,\n",
       "        -6.0285e-03,  4.7749e-03, -2.4352e-03, -5.9160e-04,  7.3199e-03,\n",
       "         9.1596e-04,  1.0274e-02, -5.0277e-04,  2.4924e-02, -3.8184e-03,\n",
       "        -1.2299e-02, -4.2888e-03, -1.1779e-04,  4.3472e-04, -1.7660e-02,\n",
       "         1.6611e-03, -1.0168e-02, -1.2611e-03,  3.1735e-03,  1.2978e-02,\n",
       "         1.2730e-03, -1.8850e-02, -1.2581e-02,  2.0153e-04, -1.3821e-02,\n",
       "        -2.2035e-03, -3.7410e-03,  2.2652e-03,  9.0273e-03, -1.6902e-02,\n",
       "        -4.2163e-03,  2.4663e-03, -3.4076e-03,  2.8310e-02, -7.7768e-03,\n",
       "        -1.2038e-02, -1.6267e-03,  2.0896e-03, -2.6120e-03,  1.1736e-02,\n",
       "        -7.2716e-03, -1.8253e-02,  1.8220e-02,  5.2735e-03,  4.1615e-03,\n",
       "         1.1380e-03,  3.4593e-03, -1.4693e-03, -6.8945e-03, -1.2216e-03,\n",
       "         2.1273e-03, -9.4595e-03, -1.0133e-02, -2.2444e-02,  1.0139e-02,\n",
       "         1.0603e-02, -8.9199e-03,  1.3513e-03,  3.8436e-03,  1.5773e-02,\n",
       "         1.3622e-02,  1.8034e-03, -4.1698e-03,  4.0546e-03,  9.3153e-03,\n",
       "         1.6695e-02,  9.5052e-03, -1.1707e-02,  1.5398e-03, -1.2011e-03,\n",
       "         6.7068e-03,  1.2447e-02, -1.8643e-03, -9.7128e-03, -1.1344e-03,\n",
       "        -2.4548e-03,  1.3909e-02,  2.3526e-03, -8.1619e-03, -6.9874e-03,\n",
       "        -6.0707e-04, -2.5411e-02,  1.0024e-02,  2.8127e-03,  1.3809e-02,\n",
       "         1.0493e-02, -3.4124e-03,  1.0730e-02, -6.8643e-03,  5.4552e-03,\n",
       "         7.8049e-03, -8.7397e-03,  1.1888e-02, -2.4738e-02,  7.8894e-03,\n",
       "        -3.9526e-03,  1.7414e-03,  1.2180e-02, -1.3880e-03,  9.2457e-03,\n",
       "        -3.3357e-03,  9.0993e-03, -2.1610e-02,  1.8310e-02, -2.0115e-03,\n",
       "         7.3215e-03, -4.9766e-03,  9.4304e-03, -2.2752e-02, -2.9859e-03,\n",
       "        -1.2027e-02, -1.5390e-02,  9.7385e-03,  4.8621e-03,  3.2613e-03,\n",
       "        -5.3610e-03,  1.7581e-03,  1.1686e-02, -1.1776e-03,  8.1026e-03,\n",
       "        -1.2652e-02, -1.1975e-02,  6.5292e-03,  3.2865e-02, -1.6246e-02,\n",
       "         1.3199e-02,  1.3390e-03,  7.1352e-03,  1.1427e-02,  8.2980e-04,\n",
       "         1.6970e-02,  7.7600e-03,  8.7114e-03,  1.3777e-02, -8.0782e-03,\n",
       "        -3.8476e-03, -7.1066e-03, -3.8011e-03,  2.6256e-03, -1.9201e-03,\n",
       "         1.2824e-02,  7.1115e-03,  1.0955e-02, -1.4206e-02,  1.6001e-02,\n",
       "         3.6246e-03,  9.0582e-03,  3.2116e-03, -2.9008e-03,  1.2907e-02,\n",
       "         1.3007e-03, -1.4027e-05, -3.6410e-03,  2.2906e-02, -1.0605e-02,\n",
       "        -4.1451e-03, -4.4753e-04,  2.3172e-02,  1.2419e-02, -4.9444e-03,\n",
       "        -6.5480e-03,  5.9070e-03,  3.0840e-03,  3.3855e-03,  1.1182e-02,\n",
       "        -6.5780e-03,  3.2027e-03,  1.4056e-02,  2.7642e-03,  5.9167e-03,\n",
       "         2.6591e-02, -1.0697e-02, -1.6400e-03, -6.2066e-03, -1.1989e-02,\n",
       "         8.9553e-04, -8.4076e-03,  4.3041e-03, -3.8863e-03, -4.2411e-03,\n",
       "        -1.9719e-03,  1.2064e-02,  3.0248e-03,  2.3671e-04, -3.7903e-03,\n",
       "         1.1845e-03, -6.9125e-04, -8.3980e-03,  7.0648e-03, -1.3351e-02,\n",
       "        -5.9450e-05, -1.0914e-03, -1.1813e-03, -1.0899e-02,  1.2157e-02,\n",
       "        -1.0610e-02, -1.0267e-02, -3.9500e-03,  1.0179e-02,  3.2946e-02,\n",
       "         1.6978e-02, -1.4264e-02, -1.6790e-03, -5.0871e-03,  7.1525e-04,\n",
       "        -2.2056e-02, -9.0232e-03, -3.3626e-03, -1.0605e-02, -2.1213e-02,\n",
       "         4.8673e-03,  8.8190e-03,  2.0229e-02, -9.2550e-03,  2.0225e-02,\n",
       "        -1.3284e-02, -1.6753e-03,  8.4168e-04, -1.6946e-02,  4.8658e-03,\n",
       "         6.9330e-04,  1.0020e-02,  1.1136e-02,  1.3610e-03, -9.0562e-03,\n",
       "         1.1986e-02,  1.0540e-02, -2.3219e-03,  5.2810e-04,  1.1201e-02,\n",
       "        -9.0336e-04,  8.8807e-03, -1.7506e-02,  1.2647e-02, -7.4496e-04,\n",
       "         2.4403e-02,  3.6632e-03,  1.4433e-02, -5.7378e-03, -1.0291e-02,\n",
       "         3.1439e-03,  7.1786e-03, -9.6571e-03, -1.6330e-03,  1.9078e-03,\n",
       "         1.2226e-02, -1.3079e-02, -5.6803e-03,  1.8660e-03,  6.9072e-03,\n",
       "         1.4376e-02,  1.4840e-03, -4.9252e-03, -1.3803e-03, -2.2927e-02,\n",
       "         8.3900e-03, -1.1954e-02,  1.9151e-03, -1.0489e-02,  6.4211e-03,\n",
       "        -1.6682e-02, -1.2071e-04, -3.6782e-03,  3.5746e-03,  5.7151e-03,\n",
       "        -2.7958e-03, -4.3771e-03, -5.6395e-03,  1.4871e-02, -3.2965e-03,\n",
       "        -1.9727e-02,  4.3846e-03, -4.2133e-03,  1.4465e-02, -7.7636e-03,\n",
       "         4.1323e-03, -7.6027e-03,  1.2941e-03, -2.2656e-03,  1.5043e-03,\n",
       "        -1.2953e-02,  2.0459e-04,  1.6762e-02,  1.2426e-03,  9.6988e-03,\n",
       "         1.9670e-03,  4.8527e-04, -8.4507e-03,  9.2655e-04, -1.2000e-02,\n",
       "         1.0591e-02, -6.5443e-03, -6.2307e-03,  1.7418e-05,  7.1799e-03,\n",
       "         1.1494e-02,  2.2391e-02,  4.1629e-03,  3.3356e-03, -9.3472e-04,\n",
       "        -1.0395e-02,  8.4937e-03, -1.3108e-03, -2.6641e-03,  3.4668e-04,\n",
       "        -8.0797e-03, -8.7791e-03, -3.1942e-04,  1.4530e-02, -9.7799e-03,\n",
       "         3.6499e-02, -9.5966e-03, -1.1461e-02,  7.7692e-03, -2.1857e-02,\n",
       "         1.5567e-02, -2.7686e-03,  6.6259e-03,  6.5922e-03, -2.8224e-03,\n",
       "         3.1515e-02, -2.0327e-02, -6.5935e-03, -1.1397e-02, -1.1603e-02,\n",
       "         1.4023e-02,  8.3442e-03,  3.4522e-03, -1.6165e-02, -9.7723e-03,\n",
       "        -7.7388e-03, -1.3658e-03, -1.3956e-02,  9.8019e-04, -1.0604e-02,\n",
       "        -1.0876e-02, -7.4724e-03, -1.3318e-02, -4.3488e-03, -1.8149e-02,\n",
       "         5.0577e-03,  1.1402e-02,  1.4671e-02,  7.1265e-03, -2.3932e-03,\n",
       "        -8.6341e-04, -4.7672e-03,  5.0975e-03, -6.6169e-03,  4.5026e-03,\n",
       "        -1.3366e-02, -1.4111e-02, -1.9242e-02, -4.5833e-03, -4.8288e-03,\n",
       "        -7.9064e-03, -9.0336e-03,  4.0426e-03, -6.6109e-03,  9.2299e-04,\n",
       "         1.1470e-02, -7.3758e-03, -6.7241e-03, -4.3696e-03, -4.3447e-03,\n",
       "        -2.1406e-02,  2.0812e-02,  1.1829e-02,  5.0548e-03, -2.6322e-03,\n",
       "         1.0954e-02, -1.6022e-02, -5.7273e-03, -8.5768e-03, -8.9098e-03,\n",
       "        -1.1078e-03,  1.9551e-03, -1.6632e-03, -1.8012e-03, -2.4717e-03,\n",
       "         1.3055e-02, -1.1516e-02, -7.0611e-03,  9.5208e-04,  1.1671e-02,\n",
       "         5.1347e-03, -1.1308e-02,  1.4886e-03, -4.6180e-03,  7.5004e-03,\n",
       "         1.2057e-02, -9.7009e-03,  1.7364e-02, -3.6672e-03, -1.8135e-03,\n",
       "         9.7222e-03, -1.0052e-02,  1.8546e-02,  6.8618e-03, -7.7363e-03,\n",
       "         2.1947e-02,  5.4350e-03, -8.6781e-03, -6.8176e-03, -1.0640e-03,\n",
       "        -1.0383e-02,  1.4203e-02,  1.7907e-03, -1.7534e-02,  1.2486e-02,\n",
       "        -1.3474e-02, -1.2361e-03,  3.6817e-03,  5.1715e-03,  7.0774e-03,\n",
       "         3.4396e-03, -1.4958e-03, -6.1044e-03,  3.0111e-04, -1.2982e-02,\n",
       "        -5.5741e-03,  9.7460e-03,  2.9775e-03,  1.2963e-03, -6.9276e-03,\n",
       "        -1.5948e-02, -2.6274e-03,  1.5543e-03, -1.7635e-03,  6.2728e-03,\n",
       "         2.5535e-02, -4.0238e-03,  2.1222e-02, -5.2781e-03,  3.0184e-03,\n",
       "         2.5131e-03, -8.6241e-03, -7.8797e-03, -4.5308e-03,  3.3696e-03,\n",
       "        -9.3325e-03, -3.6802e-03, -2.3747e-03, -2.5038e-03,  3.3330e-02,\n",
       "        -3.9818e-04, -4.3260e-03, -8.2479e-03,  1.2289e-02,  2.6237e-03,\n",
       "        -9.7724e-03,  1.0824e-03,  8.2739e-04, -7.1239e-03, -2.3491e-03,\n",
       "         7.3658e-04, -1.6903e-02,  3.5101e-03, -3.9035e-03,  1.3293e-02,\n",
       "        -1.0281e-02, -8.6166e-03], device='cuda:0'), tensor([-6.8418e-03, -1.6127e-02,  7.6426e-03,  3.3076e-03, -1.9131e-03,\n",
       "         4.6171e-03, -7.1016e-03,  5.3429e-03, -4.3288e-03, -1.7326e-03,\n",
       "         1.4901e-02,  5.9355e-03, -3.7831e-04,  1.4714e-02,  1.3154e-02,\n",
       "         1.1613e-02,  3.0846e-03, -6.5604e-03,  1.1446e-03,  1.0625e-02,\n",
       "         7.9399e-03, -1.2944e-02, -7.6250e-03,  9.4269e-03, -6.0627e-03,\n",
       "        -1.0250e-03, -6.6071e-03,  3.7468e-03, -8.9670e-03, -5.8810e-03,\n",
       "         1.4353e-02, -8.4263e-03, -2.9745e-03, -9.3886e-03,  1.7302e-02,\n",
       "        -2.6296e-04, -1.7872e-03,  4.8621e-03, -3.3110e-03, -1.6761e-02,\n",
       "        -5.1825e-03,  1.9986e-04, -2.2010e-02,  1.3837e-02,  1.6524e-02,\n",
       "        -6.0026e-03, -1.8548e-03, -5.2225e-03, -2.6421e-03, -3.9543e-04,\n",
       "        -7.7404e-03,  2.9761e-03, -4.1111e-03, -1.5681e-04, -5.0131e-03,\n",
       "        -1.2304e-02, -7.5502e-03, -2.6742e-02,  5.2642e-03,  2.2160e-02,\n",
       "        -1.1979e-03,  5.4417e-03, -1.3271e-02, -1.0772e-02,  1.9662e-03,\n",
       "        -1.9276e-02, -1.0583e-02,  1.2190e-02, -1.1026e-02,  4.3777e-03,\n",
       "        -1.3758e-03, -2.2631e-02,  1.4217e-02,  2.6429e-03, -5.6203e-03,\n",
       "        -1.6155e-02,  1.5693e-02, -2.4655e-02, -1.7639e-02,  3.1208e-03,\n",
       "        -1.3631e-02, -1.9845e-02, -1.8061e-02, -1.4496e-02, -8.9965e-03,\n",
       "         8.0189e-03, -2.9473e-02,  5.4363e-03,  3.2271e-03, -1.0307e-02,\n",
       "        -5.9192e-04,  7.0717e-03,  1.2094e-02, -2.0607e-03,  1.0261e-02,\n",
       "         1.4075e-02,  1.9775e-03, -1.6535e-02,  3.1963e-03, -1.8475e-02,\n",
       "        -4.5484e-03,  3.3900e-03,  1.3149e-03, -2.3454e-02, -3.8640e-03,\n",
       "         2.5743e-02, -1.8635e-02,  1.1383e-02,  9.8478e-03, -3.7937e-03,\n",
       "        -8.4237e-03, -5.0219e-03,  7.9837e-03, -1.0864e-02, -4.5372e-03,\n",
       "        -5.7268e-03,  1.0339e-02, -1.4952e-02, -3.5126e-03,  1.2650e-03,\n",
       "         1.0663e-02, -9.3575e-03, -4.1936e-04,  4.4732e-04, -3.8819e-03,\n",
       "         7.1295e-03, -4.8187e-03, -7.3870e-03, -4.0634e-04, -1.2477e-02,\n",
       "         3.4515e-03, -3.9573e-03, -2.6224e-02,  2.2995e-04,  4.6229e-03,\n",
       "        -1.0082e-02, -5.7216e-03,  3.9535e-03, -1.5301e-02,  4.9579e-04,\n",
       "         1.5289e-02, -4.9780e-03,  1.4482e-02, -1.3082e-02, -2.5878e-02,\n",
       "        -2.6503e-03, -1.5146e-02,  7.3612e-03,  8.0422e-03,  1.0561e-02,\n",
       "         2.2569e-02,  1.1554e-02,  3.2273e-03,  7.5206e-03, -3.8512e-03,\n",
       "         8.5044e-03, -5.3824e-03, -2.5676e-02,  8.7226e-03,  1.2984e-02,\n",
       "        -1.7973e-02, -7.8659e-03, -1.9885e-02, -3.3450e-03,  1.7078e-02,\n",
       "        -1.1816e-02,  2.0239e-02, -2.3674e-03, -3.0796e-03,  8.1652e-03,\n",
       "         4.8898e-03,  1.0776e-02, -1.0748e-04,  1.0400e-02, -1.4038e-02,\n",
       "         1.6099e-02, -5.0953e-03, -7.1830e-03, -7.8103e-03, -1.6141e-02,\n",
       "         2.8185e-03, -1.9952e-03, -2.4291e-03, -2.4296e-03,  3.4830e-03,\n",
       "        -8.9293e-04, -6.8464e-03, -1.2330e-02,  1.2861e-02,  8.5169e-03,\n",
       "        -1.1390e-03, -5.6008e-03, -1.1187e-03,  3.7727e-03,  3.4888e-03,\n",
       "         2.0280e-02,  2.0048e-03,  7.0927e-03,  1.3286e-02,  2.2429e-02,\n",
       "        -1.2622e-03,  4.0365e-03, -9.2950e-03, -3.5279e-03, -1.0737e-03,\n",
       "         1.4230e-02, -7.4160e-04, -3.9648e-03,  2.2777e-02, -2.3382e-03,\n",
       "         9.3487e-03,  1.6923e-03,  4.1596e-04,  2.2967e-02, -3.6549e-03,\n",
       "         1.0204e-02, -1.1555e-03,  1.4424e-03,  1.1906e-02,  6.0669e-03,\n",
       "        -9.4465e-03, -3.7034e-03, -1.4737e-02,  6.9029e-03, -1.6364e-02,\n",
       "        -1.2604e-02, -4.1405e-03, -7.7746e-03, -1.5980e-02, -2.1202e-03,\n",
       "        -2.5552e-04, -5.3355e-03, -4.3502e-03,  1.7798e-02, -5.8712e-03,\n",
       "         1.3959e-02, -9.9410e-03,  1.2171e-02, -2.1315e-02,  9.0466e-03,\n",
       "        -4.8023e-03, -2.9784e-03, -6.2053e-03, -9.8332e-03,  8.1490e-03,\n",
       "         1.5361e-02, -1.6078e-03,  7.1924e-04,  9.2385e-03, -6.0101e-03,\n",
       "         1.3320e-02, -4.9469e-03, -1.8179e-02,  7.7807e-03, -8.2463e-03,\n",
       "         2.1890e-02, -4.8953e-03,  5.1758e-03, -9.7149e-03, -6.6196e-03,\n",
       "         6.9607e-03, -3.3384e-03,  4.3791e-03, -1.7395e-02, -1.0760e-02,\n",
       "         1.3678e-02,  9.1055e-05, -1.0373e-02, -5.3644e-03, -5.9993e-03,\n",
       "        -1.6094e-02, -1.1014e-02,  6.6907e-03,  1.0274e-02,  7.9160e-03,\n",
       "        -3.8534e-03,  1.2004e-02, -4.7317e-03, -4.5308e-03, -1.0859e-02,\n",
       "         1.7395e-03, -6.9041e-03,  1.4182e-02, -8.1770e-03, -3.4459e-03,\n",
       "        -6.8759e-03,  4.3618e-03,  8.7316e-03, -1.0548e-02,  1.2549e-02,\n",
       "        -8.4216e-03,  1.4298e-03,  1.7556e-02,  7.6684e-04,  5.2109e-03,\n",
       "        -8.2857e-03, -6.4701e-03,  1.9540e-02, -1.2384e-02,  9.1479e-03,\n",
       "        -1.9003e-03,  1.7492e-02,  4.0085e-05, -9.0100e-03,  2.9599e-03,\n",
       "         1.3153e-02, -1.0836e-02,  5.1828e-03, -7.3971e-03, -9.2654e-03,\n",
       "         1.3967e-02, -8.5858e-03,  5.7195e-03,  8.6738e-04,  3.6174e-03,\n",
       "         3.1215e-03,  1.1249e-02,  4.2949e-03, -1.6230e-03, -9.1252e-03,\n",
       "         1.5527e-02, -9.4219e-03, -8.4164e-03,  5.9040e-03,  1.1699e-02,\n",
       "         8.9052e-03,  1.0806e-02, -6.2951e-03,  7.9186e-03, -1.5549e-02,\n",
       "         1.9174e-03, -2.1294e-05, -9.2804e-03,  8.0994e-03, -8.5501e-03,\n",
       "        -2.6857e-03, -1.9709e-02,  2.7310e-04, -6.9610e-03, -2.9931e-03,\n",
       "        -9.0015e-03, -4.3810e-03,  1.5753e-04,  1.3465e-02,  1.6031e-03,\n",
       "        -7.1728e-03, -1.6773e-02,  7.7553e-03,  2.4262e-03, -3.7122e-03,\n",
       "        -1.6011e-02, -1.2063e-02,  5.0408e-03, -4.2131e-03,  4.8963e-03,\n",
       "         8.0345e-03, -8.8698e-03,  1.7835e-03,  1.5589e-02, -9.9035e-04,\n",
       "        -1.5858e-02, -6.8513e-03,  1.8493e-02, -1.1662e-02, -1.1809e-03,\n",
       "         9.6503e-03, -1.0809e-02, -2.4395e-04,  1.3794e-02,  1.4814e-04,\n",
       "         3.5573e-03, -1.2352e-02,  3.4576e-05,  1.2992e-02, -1.1515e-03,\n",
       "         1.7896e-02, -1.4716e-02, -4.2620e-03,  1.7808e-02,  1.1597e-02,\n",
       "        -4.2781e-03,  4.4767e-03,  4.4688e-03, -9.6661e-04,  3.9528e-03,\n",
       "        -9.8764e-03,  5.6290e-03, -2.4101e-03, -1.1267e-02, -2.3302e-03,\n",
       "         1.5391e-02, -6.3792e-04, -2.5210e-03,  2.9276e-03,  6.8478e-04,\n",
       "        -3.1719e-04,  1.6994e-02,  1.4138e-03, -1.4003e-03,  1.1259e-02,\n",
       "        -6.9985e-03, -4.8140e-03,  1.0107e-02, -1.6749e-02,  2.7450e-02,\n",
       "        -1.0316e-02, -2.4978e-03, -4.6773e-04,  2.6340e-02,  1.9601e-02,\n",
       "        -4.1125e-04,  1.2190e-02,  1.6112e-02,  8.9787e-03,  3.6775e-03,\n",
       "        -1.3842e-02, -1.0003e-02,  2.9574e-03, -2.2813e-02, -2.5135e-04,\n",
       "        -7.3664e-03, -2.6284e-03, -3.3369e-03, -4.4595e-03,  1.3582e-02,\n",
       "         1.4202e-02, -3.3267e-03, -7.4621e-03,  1.0019e-02, -4.5839e-04,\n",
       "         1.0575e-02,  1.3176e-03,  3.5312e-03,  3.0277e-03,  4.4055e-03,\n",
       "        -1.1034e-03,  2.2574e-03,  1.9040e-03,  4.4999e-03,  9.0846e-03,\n",
       "         1.4091e-03, -5.1978e-03, -4.9504e-03,  3.4892e-03,  9.3682e-03,\n",
       "         9.0448e-03, -5.2618e-03, -5.1436e-03,  1.4058e-02, -2.0086e-03,\n",
       "         9.5423e-03, -5.5487e-03,  5.1120e-03, -1.0353e-02,  1.1892e-02,\n",
       "        -1.1457e-02, -6.3413e-03,  6.4377e-03, -1.0444e-02, -1.3075e-03,\n",
       "        -8.4837e-03, -8.6170e-03,  8.2895e-04,  2.6549e-03,  3.8225e-04,\n",
       "        -1.4851e-02,  6.3742e-03,  9.7116e-03,  1.6392e-02,  3.5908e-03,\n",
       "        -3.0259e-03, -9.9394e-03,  8.8482e-03,  1.6304e-02,  1.4792e-03,\n",
       "        -4.6263e-03, -6.0864e-03, -2.1905e-03,  1.6941e-02,  8.1597e-03,\n",
       "        -1.2712e-02,  1.1075e-02, -1.8249e-03,  1.8915e-02,  1.3547e-02,\n",
       "         7.8093e-04, -8.2603e-03, -6.4465e-03,  1.1496e-03, -1.9395e-05,\n",
       "         1.2112e-02,  5.4915e-03, -1.2348e-02,  5.7736e-03,  2.1907e-02,\n",
       "         1.9674e-02,  2.2471e-03,  1.0141e-02,  2.7233e-04,  4.5066e-03,\n",
       "         2.8093e-03, -8.7889e-03,  1.5947e-02,  1.2915e-03,  1.3980e-02,\n",
       "         1.6338e-02,  1.2205e-02,  1.1390e-02,  2.0411e-02,  3.4305e-03,\n",
       "        -3.9119e-03,  1.3166e-02], device='cuda:0'), tensor([[[[ 9.3711e-05]],\n",
       "\n",
       "         [[ 4.4122e-04]],\n",
       "\n",
       "         [[ 2.7816e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 4.5479e-03]],\n",
       "\n",
       "         [[ 4.4943e-04]],\n",
       "\n",
       "         [[ 2.0602e-02]]],\n",
       "\n",
       "\n",
       "        [[[-1.8591e-02]],\n",
       "\n",
       "         [[-5.0184e-03]],\n",
       "\n",
       "         [[ 4.2528e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-1.6222e-02]],\n",
       "\n",
       "         [[ 3.9802e-04]],\n",
       "\n",
       "         [[-3.3959e-03]]],\n",
       "\n",
       "\n",
       "        [[[-2.8163e-03]],\n",
       "\n",
       "         [[ 1.0552e-02]],\n",
       "\n",
       "         [[-1.1514e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 7.4174e-03]],\n",
       "\n",
       "         [[-5.1553e-03]],\n",
       "\n",
       "         [[ 2.0514e-02]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[-4.1484e-03]],\n",
       "\n",
       "         [[-6.5004e-03]],\n",
       "\n",
       "         [[ 1.5528e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-3.8314e-03]],\n",
       "\n",
       "         [[ 6.1812e-03]],\n",
       "\n",
       "         [[-1.3043e-02]]],\n",
       "\n",
       "\n",
       "        [[[-1.1295e-04]],\n",
       "\n",
       "         [[-8.2154e-03]],\n",
       "\n",
       "         [[-7.0833e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 1.3709e-02]],\n",
       "\n",
       "         [[ 1.4145e-02]],\n",
       "\n",
       "         [[-1.3801e-02]]],\n",
       "\n",
       "\n",
       "        [[[ 1.4895e-03]],\n",
       "\n",
       "         [[-3.3866e-03]],\n",
       "\n",
       "         [[-1.0652e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 9.0421e-03]],\n",
       "\n",
       "         [[-1.7180e-02]],\n",
       "\n",
       "         [[-7.4173e-03]]]], device='cuda:0'), tensor([-5.9069e-03,  6.3644e-03, -6.2488e-03,  1.1669e-03, -1.4515e-02,\n",
       "        -1.9358e-02,  4.3090e-03,  2.5970e-03,  2.7585e-03, -1.4506e-03,\n",
       "        -1.0755e-02, -1.8627e-04,  2.0793e-02,  1.2043e-02, -2.7669e-02,\n",
       "         5.9845e-04,  9.9822e-03,  7.9075e-03,  7.1549e-03,  5.6142e-04,\n",
       "         1.4262e-02, -3.6505e-03, -2.4845e-04, -1.3484e-02,  8.5068e-03,\n",
       "         1.0054e-02,  9.6207e-03,  7.6388e-03,  6.5056e-03, -9.0591e-03,\n",
       "        -9.1147e-03, -1.2878e-02,  4.1715e-03, -2.1025e-03, -1.7960e-02,\n",
       "         1.8512e-03,  1.1668e-02,  1.6091e-03,  3.6527e-03,  1.0722e-03,\n",
       "        -3.3796e-03, -1.1485e-03,  3.9873e-03,  3.7048e-03,  1.4062e-02,\n",
       "        -5.2163e-03,  1.2789e-02, -5.0332e-03,  7.2868e-03, -5.2788e-03,\n",
       "         6.1022e-03, -1.2870e-02,  3.4304e-04, -8.5479e-03,  4.7454e-03,\n",
       "        -1.8061e-03,  1.1229e-02, -9.6346e-03, -2.2983e-03, -1.0297e-02,\n",
       "        -2.1131e-03, -4.6100e-03,  5.5350e-03,  9.9253e-03,  7.4169e-03,\n",
       "        -5.3365e-03,  4.9250e-03, -4.9012e-03,  9.6906e-03,  3.5126e-03,\n",
       "         3.7300e-03, -7.1452e-03, -1.1479e-02, -6.0386e-03, -9.6645e-03,\n",
       "         2.5813e-02, -7.2034e-03,  3.5037e-03, -8.2810e-03,  6.4876e-03,\n",
       "        -1.1342e-02, -4.9883e-03,  5.0260e-03,  8.0726e-04,  1.1932e-03,\n",
       "        -2.7272e-03,  6.0661e-03,  7.8535e-03, -5.7170e-03,  2.3788e-03,\n",
       "         2.4855e-02,  9.3773e-03,  5.7183e-03,  8.7841e-03, -1.0945e-02,\n",
       "        -7.7361e-03,  5.8024e-03, -6.3612e-03, -6.9979e-03,  8.8231e-03,\n",
       "         8.9850e-03,  6.1528e-03, -1.2534e-02, -8.0012e-03, -1.1380e-02,\n",
       "        -1.1734e-02, -3.1462e-03,  2.1573e-02,  3.4188e-03, -5.3439e-03,\n",
       "         1.8212e-02,  8.0925e-03, -7.9138e-04, -1.0297e-02, -2.0211e-03,\n",
       "        -3.7052e-03, -9.6547e-05, -1.1537e-02,  2.3908e-03, -4.0816e-03,\n",
       "         6.5204e-03, -2.6338e-04,  2.9845e-03, -2.4690e-02,  1.4018e-02,\n",
       "         5.3937e-04, -1.2487e-02, -2.3079e-03,  2.4495e-03, -1.2682e-02,\n",
       "         5.6684e-04, -5.3452e-03, -1.3610e-02, -2.5724e-03,  7.0998e-04,\n",
       "        -6.4553e-03,  7.2791e-03, -1.6768e-02,  1.1614e-02, -8.0957e-03,\n",
       "        -1.5631e-02,  1.8406e-02,  1.9471e-02, -2.7193e-03,  8.3181e-04,\n",
       "        -1.0267e-03, -8.0000e-03,  6.8132e-04,  4.0028e-03,  3.4120e-03,\n",
       "         8.0353e-03,  3.9945e-04,  1.1153e-02, -3.3035e-04, -1.6467e-03,\n",
       "         1.7391e-02, -2.8405e-03,  7.9789e-04,  5.5038e-03, -1.8596e-02,\n",
       "         1.4254e-02, -1.9492e-02, -4.5948e-03,  8.8929e-03,  9.7634e-04,\n",
       "        -1.7544e-02,  1.4654e-02, -1.5128e-03, -2.4132e-04,  1.0971e-02,\n",
       "         9.2977e-03,  1.9484e-02,  5.8552e-03, -1.0509e-02,  5.3520e-03,\n",
       "         9.5572e-03,  1.8554e-02,  1.6011e-03,  2.3675e-02, -2.4990e-02,\n",
       "         3.4707e-03,  3.7746e-03, -1.5182e-02, -2.0212e-03, -1.1629e-02,\n",
       "         1.3140e-02, -1.7255e-02,  2.4336e-02,  1.2972e-02,  7.3353e-03,\n",
       "         3.8337e-04,  2.3385e-03, -9.4299e-03,  1.0752e-03,  2.9670e-03,\n",
       "        -1.6769e-02,  7.1166e-03, -1.4645e-02, -1.9415e-03,  8.8224e-04,\n",
       "        -8.2889e-03,  3.4068e-04, -3.0000e-03, -1.3276e-03,  1.4859e-02,\n",
       "         6.3667e-03, -4.6125e-03,  8.4244e-03, -1.2526e-03, -7.1836e-03,\n",
       "        -2.2973e-02,  3.6630e-03,  2.6399e-03,  1.4345e-02, -1.5786e-02,\n",
       "        -2.4430e-03,  8.1830e-03,  1.1462e-03,  3.4112e-03, -3.5572e-03,\n",
       "         3.2363e-02, -6.0736e-03,  3.8608e-03,  1.3092e-02,  1.4215e-03,\n",
       "         1.7584e-02,  5.1762e-03,  1.4145e-02, -7.7036e-03,  6.8633e-04,\n",
       "         1.1676e-02, -3.1995e-03, -2.6810e-04, -5.5617e-03, -1.3087e-02,\n",
       "        -2.1753e-03,  8.5593e-03,  4.6719e-03,  1.0310e-03, -3.4248e-03,\n",
       "        -7.7149e-03, -7.6671e-03,  1.3350e-02, -1.6682e-02,  2.0022e-02,\n",
       "         6.2980e-04,  7.2938e-03,  2.0577e-02, -5.1478e-03, -3.3814e-03,\n",
       "         1.2412e-02,  1.6193e-02,  7.7287e-04, -6.2214e-03,  1.3721e-02,\n",
       "         1.1791e-02, -9.3817e-03,  4.1857e-03,  4.3195e-03, -6.8169e-03,\n",
       "        -1.4936e-02,  9.9830e-04, -7.7709e-03,  1.6072e-02, -3.3382e-03,\n",
       "         6.5383e-03, -9.9611e-03, -2.6938e-03, -5.5907e-05, -1.8207e-02,\n",
       "        -1.7234e-02,  6.8752e-03, -8.0415e-04,  1.4717e-02, -1.7839e-02,\n",
       "        -5.4717e-03, -5.8538e-03, -5.6156e-03,  8.8326e-03,  5.1945e-03,\n",
       "         8.1909e-04,  1.0172e-02,  1.4245e-02,  1.5974e-02, -1.3981e-03,\n",
       "        -1.1451e-03, -1.8038e-02, -1.0529e-02, -2.4361e-03, -1.0464e-03,\n",
       "        -7.4469e-03,  6.3574e-03,  2.0668e-03, -2.1316e-03,  9.5682e-04,\n",
       "         1.1115e-02,  1.0705e-02,  1.4125e-02, -1.3563e-02,  4.4516e-03,\n",
       "         1.1660e-03, -6.8986e-03,  1.0377e-02, -5.2767e-03, -9.1208e-03,\n",
       "        -4.7941e-03,  8.4527e-04, -1.3397e-02, -2.2313e-02,  5.9197e-03,\n",
       "         3.6981e-05,  2.3516e-02,  1.4997e-02,  1.8509e-02, -3.9025e-03,\n",
       "        -1.9090e-03,  6.8346e-03, -6.1801e-03, -5.7383e-03,  7.8930e-03,\n",
       "        -1.5081e-02, -7.4106e-03,  1.3484e-03,  7.4655e-03, -7.5562e-03,\n",
       "         1.2550e-02, -1.2648e-02,  3.7106e-03, -9.0020e-03, -1.4656e-02,\n",
       "         1.4327e-03, -2.7049e-02, -1.2760e-02, -4.8020e-03, -1.1265e-02,\n",
       "        -1.0922e-02,  9.0018e-03,  1.4061e-02,  8.3719e-03, -8.2921e-03,\n",
       "         9.5000e-03,  4.6326e-03, -1.8306e-03, -5.8548e-03, -2.0840e-03,\n",
       "         1.2886e-02,  3.4819e-03,  4.7102e-04, -5.7022e-05,  8.5511e-03,\n",
       "        -8.8318e-03, -8.6608e-03, -4.2198e-03,  6.5136e-04,  1.5883e-02,\n",
       "         2.4574e-03, -2.0797e-02,  1.7111e-04,  1.7594e-02, -5.8550e-03,\n",
       "        -2.0156e-02, -2.8931e-03,  9.1897e-03, -6.8082e-03, -2.5071e-03,\n",
       "        -6.4695e-03,  6.6398e-04, -3.4461e-03, -1.0335e-02, -2.3386e-03,\n",
       "         1.4201e-02,  1.1714e-02,  1.5458e-02, -3.1585e-03, -1.0923e-02,\n",
       "         3.4470e-03,  1.0274e-02, -2.1448e-02,  1.8030e-02, -1.8445e-02,\n",
       "         1.8054e-02, -2.4233e-03,  2.1807e-03,  9.6192e-03, -1.6917e-02,\n",
       "        -4.7861e-03, -1.2461e-02, -1.0759e-02, -4.6284e-04, -9.0889e-03,\n",
       "        -4.0036e-03,  1.3448e-02,  9.7244e-03,  7.0230e-04,  3.1340e-03,\n",
       "         5.6031e-04,  2.3013e-02, -1.1397e-02, -3.1153e-03,  1.1654e-02,\n",
       "        -1.8747e-02,  6.0612e-03,  1.0383e-02, -9.5469e-03, -6.6296e-03,\n",
       "         1.5686e-02,  8.3901e-03, -1.3255e-02, -5.7866e-03,  2.6942e-03,\n",
       "        -9.1433e-03, -3.7349e-03,  6.2285e-03,  2.3203e-03,  5.0963e-03,\n",
       "         1.5883e-02, -2.1491e-02,  4.7349e-03, -6.9553e-03,  1.0882e-02,\n",
       "         3.2866e-03,  3.7121e-03,  2.5851e-03, -2.3524e-03,  2.4672e-03,\n",
       "         1.3662e-02, -1.1378e-02, -6.9984e-03, -8.8519e-03, -1.2309e-02,\n",
       "        -1.3042e-02,  1.9942e-02,  1.1697e-02, -7.6043e-03, -4.7070e-03,\n",
       "        -8.3274e-03, -8.4514e-03,  1.7803e-02, -6.7521e-03,  1.1749e-02,\n",
       "        -4.9290e-03,  1.3334e-02, -4.3597e-03, -6.0072e-03, -8.6945e-03,\n",
       "         4.1591e-03, -1.7551e-03,  4.7086e-03,  1.0116e-02, -1.6765e-02,\n",
       "        -2.0109e-03, -1.6515e-02, -5.4048e-03,  4.7607e-03, -1.8864e-02,\n",
       "         1.7687e-02, -1.0301e-02, -2.1254e-02, -1.7375e-03, -1.8495e-03,\n",
       "         4.8026e-03,  3.3421e-03,  2.5822e-02,  8.9419e-04, -1.2029e-02,\n",
       "        -6.8218e-03,  5.6582e-03,  1.5373e-03, -7.9254e-03, -9.1070e-03,\n",
       "         1.6303e-03,  1.6834e-02, -7.3836e-03, -5.6434e-03,  1.9111e-04,\n",
       "         6.0652e-03,  4.0752e-03, -1.1935e-02,  8.3427e-03, -8.1419e-03,\n",
       "         1.6393e-02, -7.2426e-03, -1.3288e-02,  5.3437e-03,  2.0282e-02,\n",
       "         8.4710e-03, -2.5417e-03, -2.7990e-03, -1.7211e-02,  9.4847e-03,\n",
       "        -9.4028e-04,  1.5681e-02,  2.0122e-03,  1.8632e-02,  6.4797e-04,\n",
       "         5.6031e-03, -9.4799e-03, -7.5453e-03, -1.8634e-03, -9.8710e-03,\n",
       "        -2.0605e-02, -2.4154e-03,  5.8834e-03, -1.2878e-02, -8.6739e-03,\n",
       "         4.0921e-03,  3.0821e-03, -2.4973e-05,  1.7673e-02,  7.8779e-03,\n",
       "        -3.9334e-03, -4.1510e-04], device='cuda:0'), tensor([-4.8159e-03,  4.8597e-03,  1.1170e-03,  6.1231e-03,  5.8392e-03,\n",
       "         1.4493e-03,  1.5153e-02, -5.1514e-03, -1.5361e-04,  7.8972e-03,\n",
       "         9.6273e-03, -1.2505e-02,  1.0562e-02,  1.7822e-03,  6.1972e-04,\n",
       "         5.6427e-03,  8.0403e-03, -7.1212e-03,  1.1812e-02,  8.1344e-03,\n",
       "        -9.3135e-03, -1.1092e-02, -5.2446e-03,  1.6343e-02,  1.0544e-02,\n",
       "         1.1248e-02,  6.7589e-03, -9.4410e-03,  1.7352e-03, -4.0385e-03,\n",
       "        -1.0641e-02,  1.2169e-02,  2.6462e-02,  3.1670e-03, -1.3504e-03,\n",
       "        -1.7429e-02,  9.8725e-03, -6.6688e-03,  6.0352e-03,  7.7010e-03,\n",
       "         1.1224e-02, -2.3797e-03, -5.0029e-03,  3.3894e-03, -1.1032e-04,\n",
       "         5.0027e-03,  6.1890e-03, -2.0093e-03,  5.3224e-03,  8.1052e-04,\n",
       "         4.3190e-05, -6.4848e-03,  8.8397e-03, -3.7057e-03,  3.0249e-03,\n",
       "         5.4943e-04, -1.9002e-02, -3.1118e-03, -5.7705e-03,  6.2906e-03,\n",
       "        -8.2252e-03, -1.1308e-02, -3.0111e-03, -5.3070e-03,  7.3775e-03,\n",
       "         2.0067e-02,  2.4495e-02,  1.5493e-02,  7.2306e-03,  3.1919e-03,\n",
       "        -7.6167e-03,  1.3635e-02, -1.0527e-02,  2.4177e-03, -8.3092e-03,\n",
       "        -1.2913e-02,  1.4605e-02, -6.8641e-03,  4.5351e-03,  6.0089e-03,\n",
       "         6.8713e-03,  1.1613e-03,  4.0928e-03, -4.6414e-03, -1.1474e-03,\n",
       "        -2.0532e-02,  1.5574e-02, -3.8895e-03,  2.3726e-03,  4.0773e-03,\n",
       "        -1.1986e-03, -8.9988e-03, -1.3669e-02,  8.4122e-03, -7.1754e-03,\n",
       "        -4.9204e-03,  2.9480e-03,  3.0718e-03,  1.3284e-02,  7.4150e-03,\n",
       "         1.6352e-02,  4.6484e-03,  1.1131e-02,  9.1190e-03,  1.5627e-03,\n",
       "         2.4656e-04, -5.3499e-03,  5.8226e-03,  9.5796e-03,  2.6477e-02,\n",
       "        -6.2176e-03, -1.5643e-03,  2.8792e-03, -1.9090e-03, -1.9381e-03,\n",
       "        -6.8903e-03, -5.4209e-03,  9.5587e-03, -9.9232e-04, -2.6330e-03,\n",
       "         5.3414e-03,  1.4644e-02, -1.7020e-04, -6.6907e-03, -3.1464e-04,\n",
       "         6.2673e-03, -3.1213e-03, -9.8524e-03,  2.6195e-03, -8.7531e-03,\n",
       "         9.3705e-03, -7.6460e-04, -2.3387e-02, -1.4252e-02,  1.6270e-02,\n",
       "        -8.3742e-03,  4.6346e-03,  9.8021e-03,  6.3568e-03,  1.2834e-03,\n",
       "         1.8178e-03,  1.6114e-02, -1.4934e-02, -4.8609e-03,  1.2710e-04,\n",
       "         2.8721e-03, -1.3482e-02,  5.6329e-03, -7.0953e-03,  7.5702e-03,\n",
       "        -9.4101e-03,  1.2621e-02,  6.3489e-03,  6.8782e-03,  6.2761e-03,\n",
       "        -4.7920e-03,  1.0383e-02, -6.2663e-04, -3.3287e-02, -4.4015e-03,\n",
       "         3.8723e-03, -1.4514e-02, -6.9145e-03, -5.6047e-03, -8.2862e-03,\n",
       "         7.3027e-03,  5.5842e-03, -7.0428e-03,  1.6231e-02,  1.8350e-02,\n",
       "         8.6977e-03, -4.3044e-03,  5.6629e-03, -2.0111e-02, -3.8979e-03,\n",
       "        -5.3698e-03, -8.9642e-03,  1.1287e-02, -9.1614e-03,  6.9802e-03,\n",
       "        -1.4603e-02, -1.2043e-02, -1.5655e-02,  7.2415e-03, -1.8618e-02,\n",
       "        -1.2819e-02, -1.6604e-02,  2.0429e-02, -8.5207e-03, -1.5945e-03,\n",
       "         1.2094e-02, -2.0372e-02,  1.5497e-03,  1.7022e-02,  3.8462e-04,\n",
       "        -3.9767e-03,  1.1832e-02,  8.9236e-03,  5.4834e-03,  6.0768e-03,\n",
       "         5.1001e-03,  1.1128e-02, -1.5854e-03, -6.5011e-03, -4.6727e-03,\n",
       "         2.6581e-03, -6.0700e-03, -2.0124e-02,  1.5657e-04,  2.4397e-04,\n",
       "        -1.0370e-03,  5.7924e-03,  1.3080e-02, -2.5277e-02, -3.4080e-04,\n",
       "        -1.1348e-04, -5.7069e-04, -1.4759e-02, -8.6270e-03,  1.4786e-02,\n",
       "        -7.2494e-03, -7.2872e-03,  1.2229e-02,  7.6841e-03,  8.9514e-03,\n",
       "         7.7187e-03,  2.5591e-03, -2.3710e-03,  7.0021e-04,  3.3682e-03,\n",
       "         1.1493e-02,  5.3640e-03,  1.2751e-03, -2.5423e-03,  4.7854e-03,\n",
       "         3.9880e-03,  1.3244e-02, -1.5155e-02,  4.9905e-04, -6.2915e-03,\n",
       "        -1.0382e-02,  2.0252e-03,  1.3111e-02,  3.0372e-03, -3.7866e-03,\n",
       "        -2.5775e-03,  5.5125e-03, -2.5680e-03,  1.6562e-02,  1.6160e-03,\n",
       "        -2.3769e-02, -1.5026e-02, -1.0701e-02,  6.5383e-03, -2.5398e-04,\n",
       "        -1.5448e-02, -8.9729e-04,  7.8748e-03, -1.1283e-02, -5.3352e-03,\n",
       "         6.3728e-03, -5.7387e-04,  1.0935e-02,  4.7723e-03,  4.6966e-03,\n",
       "         1.6885e-02, -1.5447e-02, -9.5787e-03,  1.3467e-02,  6.6576e-03,\n",
       "         2.3497e-03,  8.9901e-03,  8.5968e-03, -4.7430e-03, -1.0071e-02,\n",
       "         7.0626e-03,  2.9779e-02,  5.6172e-03, -1.0809e-02, -1.1715e-02,\n",
       "        -4.2318e-03, -8.1185e-03, -3.8893e-03,  1.2943e-02,  9.2490e-03,\n",
       "         1.2626e-02, -3.6812e-03, -2.4527e-03, -2.3322e-03, -1.3052e-02,\n",
       "         1.4280e-03,  9.6985e-03,  3.7648e-03,  1.9750e-03, -7.6264e-04,\n",
       "         1.0370e-02,  6.5414e-03,  2.0170e-02, -1.6729e-03, -6.4254e-03,\n",
       "         2.4957e-03,  1.8426e-02, -4.4788e-03,  7.0778e-03,  8.4652e-03,\n",
       "         9.8388e-04, -6.3773e-03,  1.2211e-02, -1.9240e-03,  6.6854e-03,\n",
       "         6.0893e-03, -6.7769e-03, -4.5700e-03,  1.9890e-02,  3.1522e-03,\n",
       "         4.3650e-03, -1.0010e-02, -2.8674e-03,  1.1991e-02,  5.4798e-03,\n",
       "         7.3023e-03, -4.5534e-03, -8.1835e-05,  1.2804e-02,  2.4595e-03,\n",
       "        -6.8992e-03, -7.0965e-03, -9.0561e-03,  1.0187e-02,  1.1537e-02,\n",
       "        -2.7121e-03,  4.4606e-03,  4.4255e-03, -1.0159e-02, -5.8892e-03,\n",
       "         4.5422e-03, -2.8374e-03,  3.9592e-03, -9.0762e-03,  1.1963e-04,\n",
       "        -1.2150e-02, -6.0283e-04, -3.2825e-03,  1.1032e-02, -8.7154e-04,\n",
       "        -2.5034e-03,  2.2999e-03, -4.0857e-04,  7.3994e-03, -9.8087e-03,\n",
       "         1.1717e-02, -1.3850e-02,  1.5548e-02, -9.8278e-03,  1.5670e-02,\n",
       "         1.5265e-02, -5.4925e-04, -1.0518e-02,  1.4564e-02, -1.2278e-02,\n",
       "        -5.2853e-03,  1.0699e-02,  9.0222e-03, -8.1938e-03, -2.5910e-03,\n",
       "        -1.0290e-02, -7.6794e-03, -1.4404e-02, -1.4034e-02, -1.2312e-02,\n",
       "        -1.0655e-04,  1.7912e-02,  4.8561e-03,  1.3090e-02, -1.1695e-02,\n",
       "         2.9631e-02, -1.0332e-03, -9.3915e-03,  6.7776e-03,  7.8918e-03,\n",
       "         6.2087e-03,  8.1801e-03, -3.4574e-03,  6.8269e-03, -2.9116e-03,\n",
       "         1.2374e-02, -1.5613e-02, -2.6834e-03,  1.1241e-02, -1.7284e-03,\n",
       "        -1.2754e-02, -6.1486e-03,  7.4757e-03, -8.7622e-04, -1.4031e-02,\n",
       "        -1.5449e-03,  9.7402e-04, -8.5993e-03,  2.0705e-02,  1.6256e-02,\n",
       "        -2.1639e-03, -4.7635e-03,  5.0402e-03, -1.9277e-03,  1.5245e-03,\n",
       "         1.4535e-02, -1.4911e-02, -4.3082e-03, -1.1350e-02, -1.1738e-02,\n",
       "         8.3944e-04,  1.9731e-03, -5.8819e-04,  4.8101e-03,  1.1747e-03,\n",
       "         3.6155e-03,  2.9724e-03,  1.2825e-02,  1.7730e-02,  1.1352e-02,\n",
       "         3.4846e-03,  4.6073e-03, -3.0512e-02, -1.4994e-02,  2.1276e-04,\n",
       "         6.6377e-03, -1.7025e-03, -9.7700e-03,  3.5210e-03, -1.1629e-02,\n",
       "        -1.6597e-02,  1.1928e-02, -2.4176e-04, -7.6437e-03,  7.4294e-03,\n",
       "        -7.4122e-03, -4.4624e-03,  3.7773e-03, -3.2075e-03, -1.2873e-02,\n",
       "         8.8988e-03, -6.7787e-03,  8.4091e-03, -8.1995e-03, -3.0851e-04,\n",
       "        -8.1522e-03, -2.3370e-03, -1.4322e-03,  6.8027e-03, -3.5249e-03,\n",
       "        -4.3642e-03,  1.1990e-02, -1.0706e-02, -8.0642e-03,  1.3325e-02,\n",
       "        -8.3081e-03,  4.8739e-03,  9.8032e-03,  5.7276e-03,  5.5676e-03,\n",
       "         2.4524e-03, -9.0494e-03, -1.3118e-02,  1.1247e-02,  7.7989e-03,\n",
       "        -3.3020e-03,  1.1666e-02,  4.4267e-04,  4.6183e-03, -1.5710e-02,\n",
       "        -3.1645e-03, -4.1885e-03,  5.5050e-03, -1.0489e-02, -4.2214e-03,\n",
       "         4.9738e-03, -8.3674e-03, -5.5471e-03,  6.8396e-05, -1.0636e-02,\n",
       "         1.1185e-02, -2.4601e-03, -1.5225e-03,  4.2450e-03,  3.8676e-03,\n",
       "         2.2096e-03,  1.1302e-02,  2.1583e-02,  1.2749e-02,  4.4861e-03,\n",
       "         1.7685e-03, -3.4971e-03,  2.2318e-02,  6.3555e-03,  5.4252e-03,\n",
       "         2.2825e-02,  3.7709e-03, -4.2574e-03,  7.1849e-03, -4.8086e-04,\n",
       "         9.3464e-03,  3.7707e-03,  1.8198e-02,  7.4887e-03, -1.0940e-02,\n",
       "        -8.7115e-03, -2.5427e-04, -3.1893e-03, -1.0037e-02, -9.4188e-04,\n",
       "         9.3448e-03,  8.6234e-04], device='cuda:0'), tensor([[[[-1.4799e-03,  2.4256e-02, -8.3507e-03],\n",
       "          [ 1.0510e-02, -1.4556e-02, -8.8185e-03],\n",
       "          [-4.0148e-03,  3.7756e-03, -1.4728e-03]],\n",
       "\n",
       "         [[ 9.5768e-03,  1.6253e-02,  1.2285e-02],\n",
       "          [ 1.2409e-02, -7.3635e-03, -1.3443e-02],\n",
       "          [-2.1825e-02, -5.9356e-03, -1.1579e-02]],\n",
       "\n",
       "         [[ 1.1463e-02,  9.9034e-03, -2.8043e-03],\n",
       "          [-1.0982e-02, -1.2171e-02, -6.7756e-03],\n",
       "          [-2.0318e-02, -1.4594e-02, -5.1145e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 1.9299e-02, -3.0302e-03, -3.1302e-04],\n",
       "          [-4.2062e-03, -1.4688e-03,  1.6999e-02],\n",
       "          [ 1.3236e-02, -6.7943e-03,  3.9680e-03]],\n",
       "\n",
       "         [[-2.2338e-02, -5.9635e-03, -9.1418e-03],\n",
       "          [ 5.1952e-03, -7.2958e-03,  4.6193e-03],\n",
       "          [-3.4543e-03, -6.1199e-03,  1.1556e-03]],\n",
       "\n",
       "         [[ 6.1723e-03, -3.1090e-04,  9.9133e-03],\n",
       "          [-8.2781e-03, -7.8145e-03,  2.6532e-02],\n",
       "          [ 1.8608e-02, -4.2505e-03,  1.5937e-03]]],\n",
       "\n",
       "\n",
       "        [[[-7.9603e-03, -1.0275e-02, -8.1043e-03],\n",
       "          [ 3.6620e-03, -7.9582e-03, -1.0879e-02],\n",
       "          [-8.6984e-03, -2.4546e-03, -6.7376e-03]],\n",
       "\n",
       "         [[-6.0690e-03, -1.4571e-03, -8.1954e-03],\n",
       "          [ 7.5746e-04,  6.5336e-03, -4.4286e-03],\n",
       "          [-4.9511e-03, -2.6797e-02,  2.7144e-04]],\n",
       "\n",
       "         [[-9.2219e-03,  2.6931e-02, -3.8868e-03],\n",
       "          [ 5.2155e-03, -7.4768e-04,  1.3708e-02],\n",
       "          [ 4.2290e-03, -1.2462e-03, -3.7507e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-4.9809e-04, -9.0852e-04, -6.7392e-03],\n",
       "          [ 1.2561e-02, -1.2642e-03,  1.0187e-02],\n",
       "          [ 6.7277e-03,  2.3478e-03, -1.2391e-03]],\n",
       "\n",
       "         [[-4.0946e-03, -2.7727e-03,  5.0070e-03],\n",
       "          [-1.7966e-02, -1.1608e-03, -5.4950e-03],\n",
       "          [ 9.4457e-03, -8.7043e-03,  1.6607e-02]],\n",
       "\n",
       "         [[-2.4063e-03, -1.1667e-03,  1.6827e-02],\n",
       "          [ 3.5510e-04,  4.8668e-04, -6.5619e-03],\n",
       "          [-3.8877e-03,  6.3146e-03, -4.9294e-03]]],\n",
       "\n",
       "\n",
       "        [[[-5.5923e-03,  6.5371e-03,  4.9372e-03],\n",
       "          [ 5.1399e-03,  1.3467e-02,  6.0032e-03],\n",
       "          [-3.0763e-03,  1.7039e-02,  4.5592e-03]],\n",
       "\n",
       "         [[ 5.9397e-03,  1.0959e-02, -1.5542e-02],\n",
       "          [ 4.9499e-03, -6.0960e-04, -4.3157e-03],\n",
       "          [ 8.1105e-04,  6.2761e-03, -5.1625e-03]],\n",
       "\n",
       "         [[ 1.3351e-03, -1.5446e-02, -5.9708e-03],\n",
       "          [-1.0166e-02,  6.6154e-03, -9.3133e-03],\n",
       "          [ 4.6864e-03,  2.1056e-03, -1.1301e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-1.4082e-03,  1.5356e-02, -4.3829e-03],\n",
       "          [-6.5899e-03, -1.3469e-02, -2.3649e-03],\n",
       "          [-1.9540e-02, -5.1705e-03,  4.8407e-03]],\n",
       "\n",
       "         [[-4.1321e-04,  4.3594e-04, -1.7505e-02],\n",
       "          [ 5.9240e-03,  2.5682e-03,  2.2097e-02],\n",
       "          [-2.2712e-02, -7.7848e-03, -1.7546e-03]],\n",
       "\n",
       "         [[ 7.5355e-03, -9.3129e-03,  1.0410e-02],\n",
       "          [-5.2076e-03,  1.7100e-02, -3.3991e-03],\n",
       "          [ 7.8020e-04,  5.8486e-03, -7.5831e-03]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[ 9.1042e-03, -3.5960e-04,  6.0801e-03],\n",
       "          [-2.5886e-03, -1.1920e-02,  7.7576e-03],\n",
       "          [ 3.2124e-03,  7.9184e-03, -1.8710e-02]],\n",
       "\n",
       "         [[-1.1839e-02,  6.6793e-03,  1.0057e-02],\n",
       "          [ 8.2512e-03, -1.0609e-02, -6.4029e-03],\n",
       "          [ 4.7900e-04,  8.7305e-03,  1.3705e-02]],\n",
       "\n",
       "         [[ 1.7617e-03,  3.2043e-03,  1.3128e-04],\n",
       "          [ 7.2745e-03, -2.2193e-03,  1.8041e-02],\n",
       "          [-4.6735e-03, -8.9473e-04,  6.2935e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 1.2098e-03,  1.3731e-02,  3.5517e-03],\n",
       "          [ 8.5862e-03,  1.8719e-03,  3.1926e-03],\n",
       "          [-3.6079e-03, -3.4135e-03,  2.1686e-03]],\n",
       "\n",
       "         [[-5.6672e-03, -7.2165e-03, -3.6010e-03],\n",
       "          [ 1.8740e-02,  7.3840e-03, -5.8288e-03],\n",
       "          [-5.2565e-03, -1.5145e-02,  3.1148e-03]],\n",
       "\n",
       "         [[-9.1640e-05, -5.7415e-03,  4.8104e-03],\n",
       "          [-1.1605e-02,  1.5546e-03, -1.3430e-02],\n",
       "          [ 1.2688e-02,  6.2142e-03,  6.0474e-04]]],\n",
       "\n",
       "\n",
       "        [[[ 2.8671e-03, -1.6776e-03,  1.7007e-02],\n",
       "          [-1.4730e-02,  7.6433e-03, -1.7768e-03],\n",
       "          [ 2.7183e-03,  1.4987e-02, -1.2986e-03]],\n",
       "\n",
       "         [[-4.0643e-03, -1.8221e-02,  4.4767e-03],\n",
       "          [ 2.0149e-02, -1.4130e-02,  5.1630e-03],\n",
       "          [ 3.4095e-03,  2.9777e-04,  3.5641e-03]],\n",
       "\n",
       "         [[-4.1834e-03, -6.6311e-04,  1.1938e-03],\n",
       "          [-5.0252e-03,  3.9201e-03,  3.1081e-03],\n",
       "          [-2.6508e-03, -7.1305e-03, -5.9618e-04]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-1.5322e-03, -9.8282e-03,  4.8225e-04],\n",
       "          [-5.4432e-03, -2.0967e-02, -2.9882e-02],\n",
       "          [ 2.2678e-04,  1.9917e-02, -4.6655e-03]],\n",
       "\n",
       "         [[ 1.2508e-02, -7.6128e-03, -4.6181e-03],\n",
       "          [ 3.6604e-03,  1.4514e-02, -9.6119e-03],\n",
       "          [-7.1996e-03, -4.1334e-04,  7.8978e-03]],\n",
       "\n",
       "         [[-5.1460e-04,  2.8918e-03, -4.9797e-03],\n",
       "          [ 6.2585e-03,  8.9842e-03,  1.1606e-02],\n",
       "          [ 2.6917e-02, -4.4505e-03, -5.2738e-03]]],\n",
       "\n",
       "\n",
       "        [[[ 1.1304e-02,  8.1093e-03,  7.1993e-03],\n",
       "          [ 7.8615e-03,  1.6543e-02,  1.0767e-02],\n",
       "          [ 2.4879e-02,  3.6625e-04,  2.2539e-02]],\n",
       "\n",
       "         [[-1.6821e-02,  8.3043e-03, -1.3685e-03],\n",
       "          [-1.7652e-03, -3.4325e-03, -1.3816e-02],\n",
       "          [-3.3305e-03, -3.2514e-03, -2.4317e-03]],\n",
       "\n",
       "         [[ 1.5599e-02, -4.0402e-03,  1.1276e-03],\n",
       "          [-1.6642e-02,  1.0404e-02,  6.8046e-03],\n",
       "          [ 1.3059e-02, -1.2933e-02,  1.0835e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-2.2585e-03, -3.8351e-03,  8.8015e-04],\n",
       "          [ 9.9842e-03,  6.0468e-03, -6.1393e-03],\n",
       "          [ 9.3756e-03,  2.3193e-02, -1.3350e-02]],\n",
       "\n",
       "         [[-1.0891e-02,  5.2731e-03,  4.3871e-03],\n",
       "          [-1.9138e-03,  3.2336e-04,  1.1801e-02],\n",
       "          [-1.8068e-02,  9.0378e-03, -2.2327e-02]],\n",
       "\n",
       "         [[-1.4359e-02,  1.2110e-02,  2.7252e-03],\n",
       "          [ 7.6865e-03,  2.5614e-03, -2.2058e-02],\n",
       "          [-6.3699e-03,  5.7810e-03,  8.7121e-03]]]], device='cuda:0'), tensor([-1.7258e-02, -7.5341e-03, -1.3577e-02, -2.6304e-03, -3.5324e-04,\n",
       "         4.0321e-03,  4.9863e-03,  4.5790e-03,  7.0491e-03, -3.1710e-04,\n",
       "        -1.4214e-02, -7.7035e-03,  6.4813e-03,  1.1655e-02,  9.8990e-03,\n",
       "        -8.8156e-03, -5.7913e-03, -5.6432e-03, -1.0487e-02, -2.5190e-03,\n",
       "         2.4272e-03,  1.9679e-02,  1.2650e-02,  5.1857e-03, -4.8110e-03,\n",
       "        -1.2679e-02,  1.3762e-03, -1.2598e-02, -1.1091e-03, -1.8247e-03,\n",
       "        -2.5453e-02, -6.1265e-03,  1.4640e-02,  6.5161e-04,  3.4268e-03,\n",
       "         5.9899e-03,  7.6044e-04, -5.6179e-03,  4.1553e-04,  5.5622e-04,\n",
       "         9.8300e-05,  4.5079e-03, -1.8502e-03, -6.6144e-03,  2.3232e-03,\n",
       "        -3.9004e-03,  1.4912e-02, -1.3270e-02, -3.8225e-03,  1.8668e-03,\n",
       "        -3.9813e-03, -3.0045e-03, -8.6797e-03, -2.2111e-02,  1.1226e-03,\n",
       "        -7.2115e-03,  1.2947e-02, -1.0276e-05,  1.5151e-02, -1.6530e-02,\n",
       "        -5.4210e-03, -3.2878e-03, -5.9600e-03, -8.1538e-03,  1.5000e-02,\n",
       "         5.7085e-03, -1.9487e-02, -3.5278e-03, -7.1491e-03,  9.8184e-03,\n",
       "        -1.5444e-03,  7.8250e-03,  1.1045e-02, -1.0096e-02,  1.7354e-02,\n",
       "        -1.7580e-02, -1.8195e-02,  1.5791e-02,  4.3757e-03,  1.7733e-02,\n",
       "        -5.3030e-03, -4.0324e-03, -6.8989e-03,  4.6098e-03, -1.1668e-02,\n",
       "        -7.5984e-03,  5.2528e-03, -9.3423e-03, -6.5651e-03,  1.2041e-03,\n",
       "         3.5059e-03,  9.2983e-03,  6.9045e-03, -5.1129e-03,  8.9275e-03,\n",
       "        -5.4264e-03,  3.6976e-03, -1.2442e-02, -5.1492e-05, -4.9814e-03,\n",
       "         1.0248e-04, -1.0815e-02,  5.3313e-03, -2.6029e-03,  1.2400e-02,\n",
       "         1.6007e-03,  2.0066e-03, -2.1542e-04, -1.7204e-03, -4.3330e-03,\n",
       "        -5.3784e-03, -5.3987e-04, -5.4536e-03,  6.1983e-03,  4.3354e-03,\n",
       "         3.1662e-03, -8.0641e-03,  9.5105e-03, -4.8903e-03,  3.5531e-03,\n",
       "         1.9530e-02, -1.1145e-02,  5.0612e-03, -5.6448e-03, -5.8933e-03,\n",
       "        -6.6077e-03, -2.2136e-02,  1.0320e-02,  4.2448e-03,  7.9348e-04,\n",
       "        -7.7271e-04,  6.2596e-03, -2.8542e-03, -7.8791e-04,  1.8653e-02,\n",
       "        -9.2703e-03,  2.9521e-03, -6.6160e-03, -8.1069e-03,  5.9733e-03,\n",
       "        -6.6754e-03, -1.3054e-04, -9.9750e-03,  4.6627e-03, -1.4082e-02,\n",
       "        -1.1161e-02, -1.0260e-02,  1.6006e-02,  5.9317e-03,  1.2125e-02,\n",
       "        -1.9775e-03, -4.8755e-04,  1.2734e-03, -6.0543e-03,  2.4134e-03,\n",
       "        -1.2070e-03,  2.8932e-03,  1.2262e-03,  1.8755e-03, -1.2113e-02,\n",
       "        -2.5411e-03,  1.6823e-02,  1.4136e-02,  6.9231e-04, -1.8593e-02,\n",
       "        -1.8200e-03, -8.2333e-03,  5.6003e-03, -3.1906e-03, -1.3026e-02,\n",
       "        -1.3832e-02, -7.8784e-04,  7.4648e-03, -4.6581e-03, -2.1103e-02,\n",
       "         1.4616e-03,  1.3878e-04,  8.9167e-03,  8.2410e-03,  1.2619e-02,\n",
       "        -1.7458e-02, -1.8468e-02,  1.9288e-03, -1.3918e-02,  4.4320e-04,\n",
       "         5.1737e-03,  4.0038e-03, -5.5210e-03,  1.0083e-04, -2.5794e-03,\n",
       "        -6.7087e-04, -1.8533e-03, -1.4320e-02,  3.9233e-03, -9.4972e-03,\n",
       "        -3.0549e-02, -1.1640e-02, -8.5754e-04,  1.0591e-02,  4.7911e-03,\n",
       "        -3.8665e-03, -9.0136e-04, -1.3765e-02,  1.0890e-02,  1.2326e-02,\n",
       "        -2.4000e-03,  2.1503e-03, -1.7779e-03,  7.2753e-03, -4.0273e-03,\n",
       "        -1.9566e-02,  7.2065e-03, -1.3098e-02,  3.0668e-03, -1.6407e-02,\n",
       "        -2.2632e-04,  7.4331e-04, -6.2870e-03, -1.2224e-02,  1.2887e-02,\n",
       "         7.1281e-03,  6.8640e-03, -2.4890e-03, -4.5136e-03, -4.9761e-03,\n",
       "        -5.5128e-03, -3.9866e-03, -6.8520e-03, -2.7357e-03,  6.4714e-03,\n",
       "         1.1593e-02, -4.4975e-03,  1.4812e-03,  5.5913e-05, -9.3267e-03,\n",
       "         3.0521e-03, -8.4040e-04,  1.1000e-03, -2.0701e-03,  1.7274e-03,\n",
       "        -2.3681e-02,  9.3339e-04, -4.4270e-03,  1.6111e-03,  9.1644e-03,\n",
       "         2.6211e-04, -1.6402e-02, -1.4990e-02,  2.6383e-03,  8.0738e-03,\n",
       "         1.0839e-02, -4.2852e-04, -1.0257e-02, -1.5032e-02,  4.4595e-03,\n",
       "         8.2856e-03,  2.6447e-02,  9.9349e-03,  6.3583e-03, -1.5545e-02,\n",
       "         8.7207e-03,  8.9694e-03,  6.6162e-03,  1.2034e-03,  1.2123e-02,\n",
       "        -7.2143e-03, -8.0744e-03, -3.0838e-03,  7.3637e-03,  5.2681e-03,\n",
       "         1.1412e-02,  1.8601e-04, -2.2665e-02,  5.3894e-03, -1.3628e-02,\n",
       "        -6.2959e-04,  3.0365e-03, -6.1033e-03, -8.5153e-04,  7.7484e-03,\n",
       "        -3.9072e-03,  9.9920e-03, -1.9158e-02, -3.1567e-03,  3.5970e-03,\n",
       "         1.2587e-02, -1.6100e-03, -3.2973e-03,  1.0584e-02,  1.9759e-02,\n",
       "         9.5961e-03, -7.1729e-05,  6.0242e-03, -6.5439e-03, -3.0870e-03,\n",
       "         6.0453e-03,  1.2321e-02, -2.4358e-03,  7.5542e-03, -2.6123e-03,\n",
       "         4.5838e-03,  7.2846e-03, -1.0470e-02,  2.1064e-02,  6.6519e-05,\n",
       "         4.3446e-03,  3.1281e-03,  3.8298e-03,  3.8842e-03,  8.5259e-04,\n",
       "         7.7414e-03,  5.3919e-03,  2.9654e-03, -1.7151e-02,  3.4280e-03,\n",
       "         2.2282e-02, -2.1050e-03,  2.4450e-03, -8.7518e-03,  3.4268e-03,\n",
       "        -8.5385e-04,  6.0668e-03,  1.4340e-02, -2.3513e-03, -6.2465e-03,\n",
       "        -5.0943e-03, -5.6621e-03, -9.9029e-04,  2.2410e-02,  3.6158e-03,\n",
       "        -9.5588e-04, -1.7294e-02, -1.3036e-02, -1.4713e-02, -1.5983e-03,\n",
       "        -2.3862e-03, -2.1103e-04, -8.5995e-03,  3.4342e-03,  9.9680e-03,\n",
       "        -6.7287e-03,  6.2298e-03, -3.5206e-03, -1.4687e-02,  4.3223e-03,\n",
       "        -1.2202e-02,  1.1251e-02,  1.1579e-02, -5.0725e-03, -4.6419e-03,\n",
       "        -8.2764e-03, -1.5711e-03, -1.0073e-02,  3.2689e-03, -5.6013e-03,\n",
       "         5.8360e-03, -1.8085e-03,  5.2838e-03, -3.7277e-03, -4.2000e-03,\n",
       "         6.6053e-04, -4.7347e-03,  7.1789e-03,  6.2731e-03, -2.9798e-03,\n",
       "         1.2030e-02, -3.0542e-03, -7.5590e-03,  9.1702e-03,  3.8209e-03,\n",
       "         9.5165e-03, -2.0354e-02,  1.2016e-02, -3.7205e-03,  2.7688e-02,\n",
       "         1.1196e-03,  3.8833e-03,  8.5391e-03,  2.0543e-02, -2.1503e-03,\n",
       "         2.4221e-02,  7.5140e-03, -5.7985e-03, -1.2626e-02,  1.9996e-02,\n",
       "         1.1606e-02, -4.7217e-03, -6.6421e-03,  1.0564e-02,  1.5876e-02,\n",
       "        -1.3372e-02, -1.0586e-03,  3.7371e-03,  1.4381e-03, -4.4513e-04,\n",
       "         1.8615e-03, -7.2909e-03,  3.1958e-03,  1.9988e-03,  6.6937e-03,\n",
       "         7.4297e-03, -6.1480e-03,  1.3161e-02,  3.5612e-03, -1.3562e-02,\n",
       "        -6.4286e-03, -2.9358e-03, -1.2982e-02, -6.5381e-03, -6.5919e-03,\n",
       "         6.6146e-03, -1.0269e-02,  1.3011e-02,  3.7192e-03, -1.2713e-02,\n",
       "         9.5773e-03,  1.0772e-02,  1.4435e-02,  7.8298e-03, -1.7894e-02,\n",
       "         5.5479e-03,  1.3942e-02,  1.1665e-02,  3.8173e-03, -2.6681e-03,\n",
       "        -8.3287e-03,  1.7031e-02,  1.1512e-02, -2.5806e-02, -2.7487e-03,\n",
       "         8.5363e-03, -3.1411e-03,  1.2726e-02, -7.1399e-03,  1.8644e-03,\n",
       "         2.1517e-03, -1.6571e-02, -1.0610e-02, -1.2905e-02, -9.1976e-03,\n",
       "         7.7370e-03, -2.9731e-03, -5.3695e-03,  6.6420e-04, -8.1057e-03,\n",
       "        -1.8478e-02, -4.3882e-03, -1.9144e-02,  1.9150e-02, -3.0218e-03,\n",
       "         1.8339e-03, -2.6454e-02, -3.6032e-03,  5.9801e-04,  1.4644e-02,\n",
       "        -4.8114e-03, -1.4812e-02, -9.2795e-03, -2.2076e-02, -1.6530e-02,\n",
       "         7.0713e-03, -3.0461e-03,  4.2006e-03, -2.8249e-03, -1.5336e-02,\n",
       "         1.0887e-02,  2.5097e-02, -7.1952e-03, -1.3070e-02, -1.1205e-02,\n",
       "         3.8567e-03,  3.2189e-03,  1.0658e-02, -1.1652e-02,  1.0858e-02,\n",
       "         9.0320e-03, -1.5036e-02,  9.8405e-04,  1.8064e-03,  8.6756e-03,\n",
       "        -1.2532e-02, -2.0940e-03, -1.5045e-02,  7.6907e-03, -8.6759e-04,\n",
       "        -5.6449e-03,  1.4848e-02,  1.3976e-02, -7.9783e-03, -1.4772e-02,\n",
       "         9.1314e-03, -7.1990e-03,  4.5170e-03,  7.9581e-03, -1.3864e-03,\n",
       "        -1.4690e-02, -8.7319e-03,  2.2573e-03, -1.4200e-02,  1.1405e-02,\n",
       "        -1.2142e-02, -4.2934e-03, -3.2088e-03,  6.1530e-03, -1.5814e-03,\n",
       "         3.8328e-03, -6.9089e-03,  5.3247e-03,  8.5576e-03, -1.4444e-03,\n",
       "         2.0066e-02,  1.1639e-02], device='cuda:0'), tensor([-1.0849e-02,  2.8582e-03,  1.4744e-02,  1.3720e-02, -9.9376e-03,\n",
       "         9.5732e-03, -9.8857e-03, -1.2619e-03, -5.2179e-03,  9.1609e-03,\n",
       "         5.5592e-03,  1.1763e-02, -4.2512e-03,  1.4029e-02,  1.8409e-02,\n",
       "         1.0487e-02, -3.7720e-04,  2.8468e-03,  5.8797e-03,  3.3938e-03,\n",
       "        -1.1929e-03, -6.5961e-04, -1.3383e-03, -1.4048e-02,  1.6502e-02,\n",
       "         5.2191e-03,  2.0527e-03, -1.0438e-02, -4.1184e-03,  7.8561e-04,\n",
       "         1.4767e-02,  3.5560e-02, -2.1550e-03, -1.7272e-02,  3.3323e-02,\n",
       "         5.4419e-03,  4.4875e-03,  7.6945e-03,  5.6694e-03,  1.0804e-02,\n",
       "        -6.3658e-04, -1.4596e-02, -2.6196e-03, -7.6293e-03, -1.7715e-02,\n",
       "        -1.2833e-02,  1.8766e-03, -2.1124e-03, -4.1042e-03, -1.2897e-02,\n",
       "        -1.0135e-02, -1.0547e-02,  2.0538e-02,  1.6603e-03,  1.8979e-02,\n",
       "        -1.1521e-02, -5.0738e-04, -1.1693e-02,  9.2120e-04,  8.0596e-04,\n",
       "        -6.2382e-03,  9.4212e-03,  5.4291e-03, -1.3237e-02, -1.0643e-02,\n",
       "        -1.6814e-02,  3.5725e-03, -2.8009e-03,  8.8448e-03, -1.5653e-02,\n",
       "        -1.3391e-04, -1.1262e-03, -1.1277e-03, -5.4583e-04,  6.4641e-04,\n",
       "        -4.8966e-03, -1.8300e-02, -1.4302e-02,  2.7901e-03, -2.8756e-02,\n",
       "         5.0988e-03,  1.1836e-02, -9.2808e-03, -4.2671e-03, -1.4689e-02,\n",
       "         1.5243e-03,  8.3667e-03,  5.7401e-03, -5.5354e-03,  5.8350e-03,\n",
       "         1.2707e-02,  5.4360e-03, -9.9252e-04,  1.0305e-02,  1.3039e-02,\n",
       "        -1.9814e-02, -3.4649e-04,  6.0582e-03, -3.0341e-03,  2.0297e-02,\n",
       "         6.1038e-03, -1.3706e-02,  8.9059e-03, -1.5544e-02, -2.9420e-03,\n",
       "         5.4760e-03, -9.3517e-03,  9.8176e-03,  9.7258e-03, -4.8086e-04,\n",
       "         7.6353e-03,  9.1441e-03,  1.7908e-02, -3.3870e-03, -1.7937e-03,\n",
       "        -1.8493e-03, -8.2087e-03, -4.5603e-03, -2.4291e-02,  2.8784e-03,\n",
       "        -8.6964e-03, -7.0814e-03, -8.8984e-03, -3.3163e-03, -6.0721e-03,\n",
       "        -5.5879e-03, -4.5032e-03, -2.2750e-02,  1.1959e-02,  3.4619e-03,\n",
       "        -9.8801e-04, -3.3903e-03,  7.3601e-04, -1.6378e-02, -8.7690e-03,\n",
       "         4.9092e-03,  1.5252e-02, -6.9164e-03, -1.4054e-02, -1.4963e-02,\n",
       "         2.4067e-02, -2.1451e-04,  1.3963e-02,  3.4999e-03,  4.2755e-03,\n",
       "         2.3813e-04,  2.3977e-03, -1.6457e-02, -8.6197e-03,  7.6981e-04,\n",
       "        -1.4030e-02, -5.6098e-03, -1.4091e-02, -1.2501e-02,  1.3674e-02,\n",
       "         4.0020e-03, -8.9662e-03,  4.4328e-03, -7.8184e-03, -6.8739e-03,\n",
       "         1.6402e-02,  6.2145e-03, -2.3194e-02, -9.0331e-03,  2.4205e-02,\n",
       "         3.9486e-03,  1.0436e-02, -8.6292e-03, -9.9542e-03,  1.0779e-02,\n",
       "         5.5543e-03,  2.0337e-03,  7.8181e-03, -8.5739e-03,  2.5421e-03,\n",
       "         4.9210e-03, -2.0721e-02,  1.7437e-03,  4.5375e-03,  4.3494e-03,\n",
       "        -3.2843e-02,  6.7468e-03,  3.5580e-03,  2.5270e-03,  1.0254e-02,\n",
       "        -2.1659e-02, -8.5265e-03,  1.6529e-02,  1.0723e-02,  1.1755e-02,\n",
       "         5.1126e-03,  1.3884e-02,  5.4530e-03, -1.7505e-03,  2.0649e-02,\n",
       "         2.4222e-03,  2.4493e-03, -5.7937e-03,  4.7520e-03,  5.2866e-03,\n",
       "        -1.3371e-02, -2.2938e-03,  8.5538e-03,  2.8750e-02,  5.4422e-03,\n",
       "         3.2040e-03,  3.5391e-02,  8.3084e-03, -2.7086e-03,  6.2297e-03,\n",
       "         3.5582e-04,  1.1223e-02, -5.0491e-03, -7.0438e-03, -5.6324e-04,\n",
       "         1.2392e-02, -5.7847e-03,  1.4814e-03, -4.3391e-03,  5.5179e-03,\n",
       "         1.9831e-02, -3.2505e-03, -1.9779e-02,  1.7589e-03,  1.2636e-02,\n",
       "         9.8085e-03, -1.1063e-02,  7.9857e-03, -2.0016e-02,  6.3357e-03,\n",
       "         2.3702e-02,  8.1008e-03,  8.4795e-03,  6.6032e-03,  1.3429e-02,\n",
       "        -9.8942e-03, -1.2977e-03, -4.3502e-04,  4.5727e-03,  1.5690e-03,\n",
       "        -1.7848e-02, -4.4055e-03, -6.9321e-03,  7.2674e-03, -9.9501e-03,\n",
       "        -1.0265e-02, -5.8439e-03, -4.5349e-04, -7.5935e-03, -5.2584e-03,\n",
       "        -4.9743e-03,  6.1453e-03,  9.0052e-03, -2.5426e-03, -6.0893e-03,\n",
       "        -7.3753e-03, -3.7859e-03, -9.9653e-04,  6.8251e-03, -4.6958e-03,\n",
       "         5.5066e-03, -3.6986e-03,  3.1961e-03,  1.3392e-02, -1.1483e-02,\n",
       "        -1.6550e-02,  9.4067e-03,  4.5146e-03, -1.6918e-02, -2.1523e-02,\n",
       "         8.0660e-03,  9.5124e-03,  1.4327e-03, -1.5959e-02, -9.0216e-03,\n",
       "         1.8491e-02,  9.0351e-03,  3.1116e-03,  1.2704e-02,  4.0561e-03,\n",
       "        -1.3334e-02, -9.6391e-03, -6.7211e-03,  1.2055e-02,  7.6454e-03,\n",
       "        -1.1073e-02,  5.4013e-03,  7.6636e-03,  1.2912e-02, -5.7377e-03,\n",
       "         2.1616e-03, -2.2566e-02, -8.2967e-03, -1.1366e-02, -1.4477e-02,\n",
       "        -2.9957e-03,  6.5215e-03, -1.0905e-02,  1.1524e-02, -1.3931e-02,\n",
       "         3.2216e-03, -1.1384e-02,  9.2983e-03,  7.9520e-03,  1.1367e-02,\n",
       "         1.1677e-02,  2.3901e-03,  1.3924e-02, -3.7299e-03,  9.7934e-03,\n",
       "         8.1546e-04,  7.6495e-03,  1.1928e-02, -1.1860e-02,  3.8910e-03,\n",
       "         8.7887e-03, -1.1752e-02,  3.3223e-03, -9.9588e-03,  1.0851e-02,\n",
       "         2.4667e-02,  1.5540e-02,  9.7085e-04,  8.4655e-03,  3.3163e-03,\n",
       "        -7.8607e-03,  1.3078e-02, -8.0986e-03, -2.6221e-03,  4.1713e-03,\n",
       "        -3.4344e-03, -1.7286e-02, -8.9719e-04, -9.0363e-04, -2.5254e-02,\n",
       "         3.4670e-03, -1.2474e-02,  8.2840e-03,  4.1366e-03, -1.0472e-02,\n",
       "        -1.6353e-02,  6.2683e-03, -8.2141e-03, -3.8941e-03, -1.8081e-02,\n",
       "        -1.8497e-03, -2.2849e-03,  4.3613e-03,  2.0512e-03, -6.4837e-03,\n",
       "         1.3688e-02,  9.2005e-03, -1.6089e-02, -1.2930e-03,  1.5932e-03,\n",
       "        -4.0401e-03, -1.3631e-02, -1.6461e-02, -6.0281e-03,  4.1122e-03,\n",
       "        -3.7947e-03,  2.7153e-03, -6.2818e-03,  3.1784e-04,  1.0498e-02,\n",
       "        -7.5791e-04, -2.0148e-02, -6.3061e-03,  1.0038e-02,  3.5681e-03,\n",
       "        -4.7013e-03, -6.0504e-03,  1.3386e-02, -2.1209e-02, -1.0828e-02,\n",
       "         7.9302e-03,  1.0125e-02,  1.9776e-02,  1.9016e-02, -2.0617e-02,\n",
       "         1.2713e-02,  1.0212e-02,  2.5481e-03,  5.4486e-04,  1.5811e-02,\n",
       "         2.3705e-02, -4.5113e-03,  2.5411e-03, -2.4598e-03, -3.3070e-03,\n",
       "        -3.1131e-03,  1.6390e-02,  1.4433e-02,  1.1406e-02, -3.3474e-03,\n",
       "         1.1888e-02, -3.7580e-03,  3.6125e-03, -4.8251e-04,  8.5458e-03,\n",
       "         6.6967e-03,  1.4092e-02, -1.6749e-02, -9.1424e-03,  4.4249e-03,\n",
       "        -8.8700e-03, -1.3478e-02,  8.0661e-03, -3.7252e-03, -6.2146e-03,\n",
       "        -6.1490e-03,  1.2021e-02, -5.0485e-03,  2.1848e-02, -6.4656e-03,\n",
       "         5.3987e-03, -3.5905e-03,  9.0792e-03,  1.0471e-02, -7.2700e-03,\n",
       "         6.4709e-03,  9.1374e-03, -1.0221e-02, -2.3257e-02,  2.1945e-03,\n",
       "         5.5287e-03, -3.5902e-03,  6.9206e-03, -2.3565e-03, -6.0602e-03,\n",
       "         7.9215e-03, -7.9345e-03, -1.1697e-02, -5.3003e-03,  2.5113e-03,\n",
       "        -1.0349e-02, -2.0366e-02,  3.0130e-03,  1.2776e-02,  1.1910e-04,\n",
       "         5.6960e-03,  1.2197e-02, -2.1118e-02,  2.1466e-03, -4.5468e-03,\n",
       "         9.4469e-03, -1.0090e-02,  1.2559e-02, -6.0092e-03,  1.9605e-03,\n",
       "         7.4258e-03, -1.3493e-02,  1.7304e-02,  3.6492e-04,  1.9991e-02,\n",
       "        -6.0948e-05,  3.7634e-04, -8.1885e-03,  1.1347e-03, -9.4440e-03,\n",
       "         1.3100e-02,  1.6716e-02, -8.6736e-03,  4.8550e-03,  2.4904e-03,\n",
       "        -1.7374e-02, -1.5961e-02,  1.1283e-02,  5.7156e-03,  1.2468e-02,\n",
       "        -3.7054e-03,  5.9416e-03,  7.0166e-03, -6.7195e-03,  1.2974e-03,\n",
       "        -5.9513e-03,  1.3247e-02,  1.3928e-02, -6.6631e-03,  4.6550e-03,\n",
       "        -3.4092e-03, -1.2255e-02,  8.9777e-03,  1.1767e-02,  8.4902e-03,\n",
       "         7.9759e-03,  1.6863e-02,  1.2321e-02,  7.6479e-03, -4.1667e-04,\n",
       "        -8.8474e-03,  5.7922e-03,  1.0054e-02,  7.4631e-03, -1.5918e-03,\n",
       "        -4.8722e-03, -7.2649e-03, -7.5143e-03,  1.7391e-03,  9.0892e-03,\n",
       "        -4.5741e-03, -8.1303e-04,  4.6770e-03, -1.5965e-02, -1.0646e-03,\n",
       "        -7.1175e-03, -3.8096e-03, -6.2636e-04,  1.0543e-02,  1.1458e-02,\n",
       "        -6.4599e-03,  9.8427e-05], device='cuda:0'), tensor([[[[ 5.9825e-03, -9.8642e-03, -2.3776e-02],\n",
       "          [-6.3865e-03,  2.0559e-03,  1.4169e-03],\n",
       "          [ 6.3997e-03,  1.8996e-02,  1.9468e-02]],\n",
       "\n",
       "         [[ 5.1254e-03,  5.9665e-03,  2.1855e-03],\n",
       "          [ 5.7432e-03,  6.4460e-03,  1.3053e-02],\n",
       "          [-4.9044e-03,  4.4820e-03,  7.2956e-03]],\n",
       "\n",
       "         [[ 1.1331e-02,  8.3455e-03,  9.1286e-03],\n",
       "          [ 1.3284e-02,  1.6665e-02, -1.7463e-04],\n",
       "          [ 7.0520e-03,  1.3622e-02, -1.8442e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[ 1.5525e-02, -5.2053e-03, -3.8082e-03],\n",
       "          [ 9.3295e-03,  1.5399e-03, -1.1195e-02],\n",
       "          [ 4.2811e-03, -1.0922e-02,  3.9375e-03]],\n",
       "\n",
       "         [[ 9.6316e-03, -5.9918e-03, -4.7335e-03],\n",
       "          [ 4.7130e-03, -7.0769e-03, -2.4894e-02],\n",
       "          [-5.3774e-03, -2.1604e-02, -8.3792e-03]],\n",
       "\n",
       "         [[ 4.8408e-03,  1.0889e-02,  5.7786e-03],\n",
       "          [-1.3517e-02,  4.4374e-03,  1.3567e-02],\n",
       "          [-7.3681e-03,  7.0929e-03,  1.4853e-02]]],\n",
       "\n",
       "\n",
       "        [[[-1.5125e-02, -9.7769e-04, -1.0969e-02],\n",
       "          [-5.1386e-03,  1.6284e-02, -3.3451e-03],\n",
       "          [-1.2112e-02,  8.8760e-03, -8.8337e-03]],\n",
       "\n",
       "         [[-1.0909e-03,  1.6318e-02, -3.0882e-03],\n",
       "          [ 1.3196e-02, -1.0816e-02,  2.0871e-03],\n",
       "          [ 8.3527e-03, -1.0491e-02, -7.8158e-03]],\n",
       "\n",
       "         [[-1.4933e-03, -2.9346e-03,  8.2724e-03],\n",
       "          [ 1.0161e-02, -7.2294e-03, -2.6764e-03],\n",
       "          [-2.2160e-02, -6.9931e-03, -2.1924e-02]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-7.0873e-03, -6.2516e-03,  7.3452e-03],\n",
       "          [ 6.8954e-03, -1.3112e-02, -1.9203e-02],\n",
       "          [ 1.2183e-02, -9.2066e-03, -7.6577e-03]],\n",
       "\n",
       "         [[ 1.1445e-02,  8.9282e-03, -1.3259e-02],\n",
       "          [ 1.4204e-02,  7.1233e-03,  7.5728e-03],\n",
       "          [-4.5792e-03,  5.5767e-03,  3.1953e-03]],\n",
       "\n",
       "         [[ 2.8939e-03,  2.7172e-02, -5.6823e-03],\n",
       "          [ 5.0553e-03, -1.3458e-02, -3.6936e-03],\n",
       "          [-1.0200e-02, -3.2477e-03,  1.4212e-03]]],\n",
       "\n",
       "\n",
       "        [[[-1.3732e-02,  1.2464e-02,  5.8371e-03],\n",
       "          [-1.9521e-02, -3.8052e-03, -8.9782e-03],\n",
       "          [-1.1602e-03,  1.6760e-03, -1.4884e-02]],\n",
       "\n",
       "         [[-2.2343e-02,  4.6942e-03, -2.0399e-03],\n",
       "          [ 2.8259e-03, -1.1031e-02,  3.5784e-03],\n",
       "          [-1.5934e-02,  4.5758e-03, -1.4830e-02]],\n",
       "\n",
       "         [[-1.3525e-02,  4.2852e-03,  2.4567e-03],\n",
       "          [ 1.0132e-02, -2.1496e-03,  8.7318e-03],\n",
       "          [ 3.4017e-03,  5.4134e-03, -8.4404e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-4.5198e-03,  2.5200e-03,  1.8349e-03],\n",
       "          [ 1.7207e-02, -6.8995e-04, -7.2393e-04],\n",
       "          [ 5.8919e-03,  1.4039e-02,  1.2007e-02]],\n",
       "\n",
       "         [[ 4.0742e-04, -8.1797e-03,  3.9395e-03],\n",
       "          [-3.6602e-05,  3.6766e-03, -3.4873e-03],\n",
       "          [ 8.4502e-03,  8.2280e-04,  2.1140e-02]],\n",
       "\n",
       "         [[-1.3078e-02, -7.2862e-03, -1.0220e-03],\n",
       "          [ 3.0819e-03, -1.0809e-02, -4.6925e-03],\n",
       "          [-4.3582e-03,  1.2625e-02, -8.6088e-03]]],\n",
       "\n",
       "\n",
       "        ...,\n",
       "\n",
       "\n",
       "        [[[-6.5604e-03,  1.8172e-03,  7.3033e-03],\n",
       "          [-8.2582e-03, -6.8492e-03,  1.3334e-02],\n",
       "          [ 1.9203e-03,  1.2832e-02, -6.5253e-03]],\n",
       "\n",
       "         [[ 7.3473e-03, -1.6289e-03, -6.4595e-03],\n",
       "          [ 9.8911e-03, -1.6122e-02,  1.6438e-03],\n",
       "          [-7.7270e-03,  1.8170e-02, -8.1877e-03]],\n",
       "\n",
       "         [[-2.8945e-03,  2.0994e-03, -1.6370e-02],\n",
       "          [ 1.1046e-03, -6.4471e-04,  1.1720e-02],\n",
       "          [ 3.2748e-03,  6.3203e-04,  3.8048e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-8.7739e-03, -1.5921e-02,  3.3439e-03],\n",
       "          [-1.0082e-02,  1.8639e-03, -6.3282e-03],\n",
       "          [-4.8569e-03, -4.8882e-03,  5.1394e-03]],\n",
       "\n",
       "         [[-1.7403e-02,  3.7116e-03,  3.8183e-03],\n",
       "          [-3.4689e-03,  5.4646e-03, -1.2816e-02],\n",
       "          [-3.7880e-04,  3.5798e-03, -1.5750e-02]],\n",
       "\n",
       "         [[ 2.3011e-03, -5.2772e-03, -2.2296e-03],\n",
       "          [-9.2149e-03, -2.2837e-02,  2.8750e-03],\n",
       "          [-2.3538e-02, -1.7044e-02,  4.5212e-03]]],\n",
       "\n",
       "\n",
       "        [[[-2.1224e-03,  5.9864e-03,  5.6353e-03],\n",
       "          [ 8.7035e-03, -3.1978e-03, -8.4231e-04],\n",
       "          [-2.8105e-03,  6.0736e-04,  1.1672e-02]],\n",
       "\n",
       "         [[ 1.6907e-03, -4.3150e-03, -3.5084e-03],\n",
       "          [ 3.6657e-03,  5.1332e-03,  6.5392e-03],\n",
       "          [ 9.7660e-03,  7.0095e-03, -1.5606e-02]],\n",
       "\n",
       "         [[-1.0276e-02,  4.9614e-03,  1.3180e-02],\n",
       "          [ 1.7860e-03, -1.4747e-02, -4.7579e-03],\n",
       "          [ 4.9425e-03, -1.5357e-02, -1.5861e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-1.9088e-03, -6.3721e-03,  2.9948e-03],\n",
       "          [-1.8572e-02,  3.7405e-03,  6.5173e-03],\n",
       "          [-7.3208e-03,  4.1004e-03,  1.3823e-02]],\n",
       "\n",
       "         [[ 3.2001e-03,  2.1208e-02, -7.5090e-03],\n",
       "          [ 2.4839e-04, -3.2776e-03,  4.6126e-03],\n",
       "          [ 6.7446e-03,  1.0896e-02, -1.0713e-02]],\n",
       "\n",
       "         [[ 1.6366e-02, -1.3400e-02, -1.7644e-02],\n",
       "          [-1.8112e-03, -6.2391e-03,  6.5879e-04],\n",
       "          [-3.3502e-03, -2.1631e-03, -1.8181e-03]]],\n",
       "\n",
       "\n",
       "        [[[-4.7593e-03, -6.4122e-03,  7.7653e-03],\n",
       "          [-3.0255e-04,  6.8777e-03,  6.8807e-03],\n",
       "          [-6.9442e-03, -2.3422e-03,  5.6281e-03]],\n",
       "\n",
       "         [[-1.0763e-02,  6.7248e-03,  3.0879e-02],\n",
       "          [ 2.9354e-03, -1.2999e-02, -1.5396e-02],\n",
       "          [ 3.4164e-03,  7.2224e-03,  1.0345e-02]],\n",
       "\n",
       "         [[ 1.1368e-02, -9.4554e-03, -3.4933e-03],\n",
       "          [ 3.2214e-03,  5.6008e-04,  1.9642e-02],\n",
       "          [-4.3519e-03, -1.2545e-02,  8.7331e-03]],\n",
       "\n",
       "         ...,\n",
       "\n",
       "         [[-7.1707e-03, -2.8401e-03, -1.4553e-03],\n",
       "          [ 9.5930e-03,  1.3115e-02, -5.9249e-04],\n",
       "          [-4.3822e-03, -3.5748e-03,  6.7548e-03]],\n",
       "\n",
       "         [[ 2.8164e-03, -4.5363e-03, -1.0356e-03],\n",
       "          [-8.1346e-03,  9.0248e-03,  9.5881e-03],\n",
       "          [ 9.2300e-03,  6.4628e-03,  1.8221e-02]],\n",
       "\n",
       "         [[-1.0848e-02, -3.7621e-03, -1.7564e-02],\n",
       "          [ 4.1330e-03,  2.0232e-02, -1.4707e-03],\n",
       "          [-3.1589e-03, -1.6111e-03, -6.4679e-03]]]], device='cuda:0'), tensor([ 2.4995e-03,  6.7636e-05, -4.9219e-03,  4.8189e-03,  3.4726e-03,\n",
       "         2.8408e-04, -8.8637e-04, -1.0238e-02, -7.2537e-03,  5.0447e-03,\n",
       "        -1.2509e-03, -9.0432e-03,  1.2597e-02,  6.5254e-03,  2.7249e-02,\n",
       "        -3.3446e-03,  1.3459e-02, -8.6749e-04,  1.6385e-02,  1.1186e-02,\n",
       "         2.2492e-02, -1.7346e-03,  1.1796e-02,  1.4210e-02,  1.7506e-02,\n",
       "        -1.2452e-02, -2.3698e-03,  8.6120e-03,  2.8843e-03, -1.7826e-03,\n",
       "        -6.8312e-03, -6.9676e-03,  1.8075e-02,  6.7090e-03,  2.0915e-02,\n",
       "         3.2441e-03, -8.1365e-03, -1.5230e-02,  5.7254e-03,  3.1614e-03,\n",
       "        -1.5037e-03,  8.6432e-03, -1.0019e-02, -6.6746e-03,  1.1339e-02,\n",
       "        -2.1987e-02, -1.5186e-03,  4.3134e-03, -1.2316e-02,  1.1499e-02,\n",
       "         1.4633e-02, -4.7991e-03, -2.5636e-03, -2.9403e-03,  5.6716e-03,\n",
       "         1.8397e-03, -4.7738e-03, -1.2790e-03,  4.6200e-03, -1.3867e-02,\n",
       "         9.6664e-03, -5.4662e-03,  1.8035e-03, -1.4188e-02,  4.5939e-03,\n",
       "        -2.9549e-03,  1.2658e-02, -5.0086e-03,  4.8498e-04,  9.5112e-03,\n",
       "        -5.6589e-03,  8.2412e-03,  9.3448e-03, -1.1853e-03,  7.1926e-03,\n",
       "         5.3856e-03,  7.2134e-03, -1.2225e-02, -5.1004e-03,  3.7203e-03,\n",
       "         9.8609e-03,  1.6939e-03, -7.3241e-03, -1.5204e-02,  1.1345e-02,\n",
       "         8.3185e-03,  1.2536e-02, -4.2982e-04,  9.6913e-03, -8.0781e-03,\n",
       "        -7.4620e-03, -2.2527e-02,  9.2697e-03,  1.5163e-02, -1.0626e-02,\n",
       "         6.4054e-03, -1.5143e-03, -1.1116e-03, -3.0729e-02, -8.8627e-03,\n",
       "        -5.8560e-03,  1.2732e-04, -8.3390e-03, -4.4281e-03, -7.0977e-03,\n",
       "         5.1851e-03, -5.7799e-03,  6.1632e-04, -1.7744e-02, -2.2469e-03,\n",
       "         6.9056e-03, -7.5949e-03,  3.9463e-03, -1.4191e-02,  5.0631e-05,\n",
       "        -1.0981e-02, -8.6482e-04,  4.8180e-03, -4.8287e-03, -2.3195e-03,\n",
       "        -5.4264e-03,  1.4775e-02,  1.9534e-02,  4.9340e-05,  8.6237e-03,\n",
       "        -2.9225e-03,  8.5766e-03,  3.7441e-03, -4.1840e-04,  1.1267e-02,\n",
       "         3.5298e-03,  6.7320e-03,  4.4928e-03, -1.0841e-02,  6.1196e-03,\n",
       "         3.2240e-03, -3.5874e-03,  1.3686e-03,  9.3452e-05, -1.1483e-02,\n",
       "        -7.6647e-04, -1.1647e-02,  1.0101e-02,  2.2709e-02, -1.0565e-03,\n",
       "        -1.4594e-02, -1.5422e-02, -7.8018e-03, -5.6072e-03,  2.2381e-02,\n",
       "        -1.0649e-02, -2.1384e-03,  1.9819e-02, -2.2394e-02,  1.0280e-02,\n",
       "         1.1796e-03,  1.2694e-02,  1.1005e-02,  8.4717e-03,  2.6225e-03,\n",
       "        -1.0399e-02,  1.2361e-02, -8.9435e-03, -1.3384e-02, -1.5686e-02,\n",
       "        -5.2162e-03, -1.7455e-02, -1.4316e-02, -1.8368e-04, -1.2427e-02,\n",
       "        -2.3138e-02,  2.7757e-03,  1.6096e-03, -1.0642e-02,  6.2456e-03,\n",
       "        -1.5500e-03, -1.6471e-03, -5.9960e-03, -1.5821e-02,  7.6526e-03,\n",
       "        -2.5538e-03, -1.4371e-02, -5.0870e-03, -4.1461e-03,  9.2111e-03,\n",
       "        -9.3484e-04,  1.6605e-02,  8.6401e-03,  1.3940e-02, -1.7617e-02,\n",
       "         1.3078e-03, -5.4079e-03, -2.6242e-03, -3.2006e-03, -1.9541e-03,\n",
       "         8.8790e-03,  1.2137e-02,  3.2004e-03,  2.1117e-03,  8.1683e-05,\n",
       "        -6.3656e-05, -7.6382e-03, -7.9282e-03, -7.4179e-03,  4.4099e-03,\n",
       "         7.1110e-03,  4.6418e-03, -1.9084e-02,  6.1679e-03,  2.1888e-02,\n",
       "        -1.3984e-02,  3.1445e-03,  2.2855e-02,  2.4411e-02, -3.3397e-03,\n",
       "        -6.4067e-03, -1.2846e-02, -1.0409e-03,  5.4699e-03, -1.5387e-02,\n",
       "        -7.2736e-03,  2.5832e-03, -6.4315e-03,  4.5589e-03, -1.2369e-02,\n",
       "        -2.0232e-03,  8.5735e-03,  1.0517e-03,  1.6504e-02, -2.0612e-03,\n",
       "         1.2162e-02, -3.1865e-03, -1.5211e-02,  1.4966e-02,  9.4513e-03,\n",
       "         4.8294e-03, -2.8224e-03,  7.3146e-03,  4.5752e-03,  5.6841e-03,\n",
       "        -1.4377e-03,  2.0361e-03, -6.0336e-03, -1.7236e-02, -8.7343e-03,\n",
       "         7.9150e-03,  2.2080e-03,  1.5888e-02,  8.7545e-04, -1.3551e-02,\n",
       "        -3.7096e-03,  4.5867e-04,  9.3941e-04, -7.0453e-03,  1.5564e-03,\n",
       "        -6.7041e-04, -5.2131e-04, -3.7310e-03, -2.4008e-03,  1.3090e-02,\n",
       "         4.8870e-04, -3.4057e-03,  1.9876e-02, -2.9472e-03,  5.5093e-03,\n",
       "        -1.5910e-03,  3.5577e-03, -8.9646e-03,  1.8639e-02,  1.7686e-02,\n",
       "        -5.4770e-03, -6.6964e-03,  3.9453e-03, -1.2983e-04,  5.9903e-03,\n",
       "         1.0788e-02,  7.4277e-03, -1.3635e-02,  5.1571e-03,  7.4569e-03,\n",
       "         5.1172e-03,  2.1193e-02,  9.0416e-03,  7.5350e-03,  1.0827e-02,\n",
       "        -1.1232e-02, -7.4105e-03,  1.3846e-02,  3.1626e-03,  1.5353e-03,\n",
       "         3.1827e-03, -9.0961e-03, -4.3135e-03,  4.4567e-03, -3.3852e-03,\n",
       "        -8.5693e-04,  1.3355e-02,  1.3411e-02, -1.2287e-02,  1.6760e-02,\n",
       "         7.9429e-03, -3.4098e-03,  2.6463e-03,  5.7468e-03,  2.5695e-03,\n",
       "         1.9168e-02,  1.0297e-02,  1.0531e-02,  7.5585e-03,  6.1191e-04,\n",
       "        -8.2110e-04,  4.9124e-03,  2.2666e-03, -2.7276e-03, -1.6628e-05,\n",
       "         7.8717e-03,  2.0697e-02, -7.0206e-03, -7.3266e-03,  2.7750e-03,\n",
       "        -1.0109e-02,  2.3429e-03,  1.0175e-02,  1.1615e-02,  1.6462e-04,\n",
       "        -8.9161e-04,  1.0257e-03, -3.6628e-03, -3.6628e-03,  4.4668e-03,\n",
       "         1.9970e-03, -1.3444e-03,  3.5862e-03, -3.9847e-03, -1.7877e-02,\n",
       "         7.5036e-03, -6.0321e-03, -4.4075e-03,  2.0788e-04,  6.8256e-03,\n",
       "         1.1922e-02, -2.0957e-02,  8.9242e-03,  2.0633e-02,  7.5434e-03,\n",
       "         1.4172e-03, -1.1073e-02, -6.0778e-03,  5.5386e-03,  9.8242e-03,\n",
       "         6.6961e-03, -6.8265e-03, -2.4160e-02,  1.5464e-02, -8.5245e-03,\n",
       "         1.2630e-02,  7.6228e-03, -4.9196e-03, -1.3053e-02,  5.3913e-03,\n",
       "         7.3699e-03,  1.2344e-02,  1.6224e-02,  1.1974e-02, -2.7290e-03,\n",
       "        -2.2613e-03, -1.0218e-04, -3.4441e-03,  2.8553e-03, -2.3268e-02,\n",
       "         1.9540e-02, -2.0711e-03, -4.0932e-03, -2.1863e-02,  2.2115e-03,\n",
       "        -1.0654e-02, -3.8006e-04,  1.0395e-02,  1.1393e-03, -7.1634e-03,\n",
       "         1.4993e-02, -5.4025e-03,  1.1120e-02,  2.0271e-02, -3.6493e-03,\n",
       "        -1.6301e-02,  3.1815e-03, -3.5938e-03, -5.8314e-03,  1.0081e-02,\n",
       "         1.2686e-02, -2.7166e-04, -1.3064e-02,  1.5880e-03, -1.0494e-02,\n",
       "        -1.0503e-02, -4.8298e-04,  6.3995e-03,  7.7188e-03, -5.4999e-03,\n",
       "         1.3121e-03,  5.7753e-03, -5.5792e-03,  7.4464e-03, -3.9686e-03,\n",
       "         9.2764e-03,  7.6184e-03, -2.3792e-03, -1.3910e-02,  3.3033e-03,\n",
       "        -2.7152e-03,  1.8647e-02,  6.3613e-03, -3.0999e-03,  3.1818e-03,\n",
       "         2.0386e-03,  1.6242e-02,  5.3737e-03,  2.4407e-02, -1.2260e-02,\n",
       "         9.7209e-03, -6.1941e-03, -2.2372e-03, -2.7818e-04,  1.4603e-03,\n",
       "        -9.9742e-03, -6.0537e-03,  1.9138e-02,  5.5890e-03, -3.3907e-03,\n",
       "        -2.9135e-03, -9.7312e-03, -7.8615e-03,  2.0535e-02,  5.1689e-03,\n",
       "        -2.6715e-02, -2.6179e-03, -3.5554e-03,  2.8237e-03,  2.2063e-02,\n",
       "        -1.0591e-04, -8.8408e-03, -1.5421e-02, -6.9703e-03, -1.5026e-02,\n",
       "         1.8405e-02,  7.7299e-03, -1.0066e-02, -5.0764e-03, -3.2793e-03,\n",
       "        -1.9353e-02,  1.3165e-02,  2.0988e-03, -8.5898e-03,  5.4623e-03,\n",
       "         5.5044e-03, -1.0826e-02,  2.2189e-02, -1.0094e-02, -4.1623e-03,\n",
       "        -7.5951e-03,  1.9893e-03, -4.8724e-03, -1.2937e-02, -1.7090e-04,\n",
       "         2.8248e-03, -9.8249e-03,  2.7266e-03,  2.7459e-04, -4.3454e-03,\n",
       "         2.6197e-02,  8.3717e-03,  9.1365e-03,  9.0226e-03,  2.0461e-03,\n",
       "         1.0300e-02, -1.1514e-02,  2.9053e-02, -1.5062e-02,  3.2117e-03,\n",
       "        -1.3874e-02,  7.6658e-03, -1.0043e-02,  1.2345e-02, -9.8778e-03,\n",
       "        -7.3896e-03, -1.9954e-02, -1.1480e-03,  5.5366e-03, -4.1853e-03,\n",
       "        -7.1099e-04,  6.3517e-03,  1.3942e-04,  1.1386e-02,  1.2005e-02,\n",
       "        -1.4739e-02, -4.0238e-03, -1.8009e-02,  4.2596e-03,  1.8350e-02,\n",
       "         1.8999e-02,  1.7724e-02,  8.1455e-03, -7.8040e-03,  2.5489e-02,\n",
       "        -1.0844e-02, -1.2998e-02, -1.5440e-02,  1.6150e-02,  2.7923e-02,\n",
       "         3.3283e-03, -8.8197e-03], device='cuda:0'), tensor([-7.0712e-03,  4.4925e-03,  9.0639e-03, -1.0613e-02,  5.2597e-03,\n",
       "        -1.0949e-02,  1.2076e-02, -1.4050e-02, -3.8972e-03, -1.3104e-02,\n",
       "         3.3335e-03,  8.3995e-03, -1.2423e-02,  6.4353e-04, -1.5311e-02,\n",
       "         1.3523e-03, -2.0258e-02, -1.0907e-02,  5.5997e-03,  1.7725e-02,\n",
       "        -2.7589e-03,  3.6315e-03,  1.4219e-02, -1.1538e-02, -1.0328e-02,\n",
       "         1.0905e-02,  2.9337e-03, -1.3610e-03, -1.7431e-02,  6.3307e-03,\n",
       "         1.6055e-03, -1.3649e-02,  3.5589e-03, -9.9202e-04,  1.3060e-02,\n",
       "        -2.1076e-02, -1.7888e-02,  2.2082e-02, -5.0712e-03,  1.4352e-03,\n",
       "         2.7096e-03, -1.7472e-02, -8.8698e-03,  1.2596e-02, -1.1126e-02,\n",
       "         4.4913e-03,  3.0401e-03, -3.6561e-03, -8.2229e-03, -3.6464e-03,\n",
       "         9.5406e-03,  3.9808e-03,  1.4234e-02,  4.6266e-03,  1.8569e-03,\n",
       "        -9.9957e-04, -1.2028e-02,  2.8046e-03,  3.2670e-02,  1.4857e-02,\n",
       "        -1.9506e-02,  1.8857e-02, -1.0551e-02,  3.4772e-03,  8.0917e-03,\n",
       "         4.8823e-03, -9.0020e-03, -1.1715e-02,  1.7691e-02,  5.1648e-04,\n",
       "         1.0071e-02, -1.5627e-02,  1.3496e-02, -1.8897e-03,  5.8174e-03,\n",
       "         5.7233e-03, -1.0019e-02, -1.5575e-03,  7.1959e-04,  2.1891e-02,\n",
       "         3.4293e-03, -6.3986e-03, -4.2965e-03, -2.1038e-03,  1.2120e-03,\n",
       "        -4.6094e-03, -1.5139e-02,  7.3755e-03, -5.9004e-03, -6.5849e-03,\n",
       "        -4.1341e-03, -1.0685e-02,  9.6614e-03, -1.5635e-02, -9.4783e-03,\n",
       "        -1.6335e-02, -1.6498e-03,  1.0630e-02,  2.0327e-03, -7.0966e-03,\n",
       "        -7.9724e-03, -3.3987e-03, -9.4839e-03,  3.0985e-03,  3.2897e-03,\n",
       "         8.0969e-03, -6.8932e-03,  2.0970e-03, -2.9993e-03, -4.4240e-03,\n",
       "         6.2214e-03,  6.8220e-03,  1.0401e-02, -8.0750e-03,  2.5348e-03,\n",
       "         6.5208e-03,  3.3188e-03,  9.1365e-03,  1.6917e-02, -1.4983e-02,\n",
       "        -1.0341e-02,  1.2212e-02,  2.5006e-03, -8.1008e-03,  7.1058e-04,\n",
       "         9.0623e-03, -8.3917e-03, -7.5519e-03, -7.5690e-03,  9.5975e-03,\n",
       "         7.3377e-03, -8.4224e-03, -1.6373e-02,  3.5391e-03, -7.3357e-03,\n",
       "         1.0432e-02,  1.0500e-03, -9.9952e-03, -4.0910e-03,  1.9441e-02,\n",
       "        -5.6051e-03,  2.0282e-02, -1.0555e-02,  3.7598e-03, -6.5803e-04,\n",
       "         1.1066e-02,  9.2714e-03,  6.2086e-04,  3.4710e-03, -1.8330e-03,\n",
       "         7.0665e-03,  8.2024e-03, -9.8011e-04,  8.2599e-03,  2.1643e-02,\n",
       "         4.3850e-03,  1.1693e-02,  3.4716e-03, -9.8208e-03, -6.6694e-03,\n",
       "         1.1430e-02, -2.2852e-03,  1.1630e-02,  8.1731e-04,  6.8877e-03,\n",
       "         1.0181e-02, -6.9477e-03,  1.4634e-02,  1.8282e-02,  4.1825e-03,\n",
       "        -1.7835e-02,  1.5080e-02,  9.7756e-03, -1.9901e-02, -2.2939e-03,\n",
       "        -1.5858e-03,  1.1840e-02, -6.5814e-03, -6.6656e-03,  9.1926e-03,\n",
       "        -1.3496e-02, -2.9849e-03, -1.0301e-02, -5.3973e-03,  7.1637e-03,\n",
       "         1.5140e-03,  2.2379e-02, -1.1260e-02, -4.8894e-03, -2.1724e-03,\n",
       "         4.1448e-03, -2.6584e-03, -4.1087e-03, -1.3957e-04,  1.1162e-02,\n",
       "         2.4991e-03, -3.6593e-03, -1.4276e-02,  4.0436e-03, -9.0857e-03,\n",
       "        -3.9735e-03,  2.0402e-02, -1.3702e-02,  5.9231e-04,  5.3250e-03,\n",
       "        -4.4842e-03,  2.3542e-02, -1.0171e-03,  2.0188e-02,  9.2509e-03,\n",
       "         2.1542e-03, -2.3431e-03, -6.5333e-03, -1.0175e-02, -1.8583e-02,\n",
       "        -5.6633e-03,  1.1897e-02,  1.0621e-02, -2.5174e-03,  8.3549e-05,\n",
       "        -1.1085e-03,  9.4035e-03, -6.4080e-03, -1.5803e-02, -8.4027e-03,\n",
       "         9.4018e-03, -2.0610e-02, -8.3794e-04,  3.1821e-03, -6.1969e-03,\n",
       "        -1.0606e-02, -7.0590e-03,  6.4736e-03, -1.0573e-02, -5.0050e-03,\n",
       "        -3.9343e-04, -6.7799e-03,  3.3275e-03, -1.1694e-02, -8.9371e-03,\n",
       "         1.5348e-03, -1.4834e-02,  7.7661e-03, -1.1057e-02, -5.0432e-03,\n",
       "         6.9137e-03,  5.3818e-03, -5.0290e-03, -4.2762e-03, -3.6813e-03,\n",
       "         7.3155e-03, -2.0047e-03, -6.9546e-03, -1.0701e-02, -3.0664e-03,\n",
       "         8.6154e-04, -6.6396e-03,  7.3848e-03,  1.2429e-02,  4.5872e-04,\n",
       "        -5.8436e-03,  7.7834e-03,  2.1491e-02,  1.7167e-03, -4.5821e-03,\n",
       "         9.2673e-04,  6.9060e-03,  1.5359e-02, -1.0044e-02, -6.6311e-03,\n",
       "         1.9440e-02,  7.1299e-03,  5.5763e-03,  2.5084e-03, -4.6947e-03,\n",
       "        -4.1462e-03, -6.9306e-03, -9.4900e-03, -1.2723e-02, -8.2040e-03,\n",
       "        -1.4336e-02, -3.9205e-03,  1.3590e-02,  2.2001e-03,  1.0531e-03,\n",
       "        -1.2640e-02, -3.5980e-03, -6.9636e-03, -1.1781e-02,  9.0677e-03,\n",
       "        -1.4279e-03,  1.3353e-02, -1.3459e-02,  1.8818e-04,  4.7165e-04,\n",
       "        -3.4203e-03, -2.1322e-03,  1.0824e-02,  1.7442e-02,  5.5902e-03,\n",
       "        -1.4575e-03,  2.0404e-02, -1.3539e-02,  5.1244e-03, -2.3588e-02,\n",
       "        -7.6564e-03, -1.4703e-05,  7.3704e-03,  1.8509e-03,  9.3888e-03,\n",
       "         2.9188e-03,  9.7819e-03, -1.5402e-02, -9.4739e-03,  1.8049e-02,\n",
       "         5.5744e-03, -4.8734e-03, -4.8927e-03, -1.0858e-02, -2.8326e-03,\n",
       "         9.8157e-03,  3.6750e-04,  3.3435e-03, -8.3083e-03, -1.8059e-03,\n",
       "        -8.2058e-03,  5.3926e-03,  1.6633e-02, -1.8132e-03, -5.0965e-03,\n",
       "         9.1376e-03,  1.2194e-02, -1.4541e-03,  9.3076e-03, -1.1895e-02,\n",
       "         2.9284e-03,  5.9344e-03, -5.2412e-03, -1.1842e-02,  1.7874e-03,\n",
       "         1.0240e-02,  2.6409e-03, -9.5957e-03, -1.5130e-02,  1.3678e-02,\n",
       "        -8.6237e-03, -6.6975e-03,  7.6115e-03, -6.2352e-03,  1.8511e-02,\n",
       "        -4.8858e-03, -6.7028e-03, -4.3477e-03, -2.8792e-03, -1.1016e-02,\n",
       "        -5.8947e-03, -1.7788e-02,  2.2999e-03, -9.8822e-03,  4.6290e-03,\n",
       "         1.6026e-02, -3.0357e-04,  1.6253e-03, -1.4353e-02, -2.4502e-02,\n",
       "         1.4686e-02, -4.2717e-03,  4.3903e-03,  1.0663e-02, -1.5471e-02,\n",
       "         3.8012e-03, -1.9474e-03, -7.5553e-03, -8.4806e-03, -9.6873e-03,\n",
       "         8.5463e-03,  9.7483e-03, -5.2491e-03,  1.8847e-02,  1.6873e-02,\n",
       "         1.2809e-02, -6.2797e-03, -2.3808e-02,  1.4749e-02, -4.1833e-03,\n",
       "        -1.6762e-02,  3.8221e-03, -1.3395e-03, -6.2646e-03,  7.1897e-03,\n",
       "         9.6621e-03, -3.6245e-03,  3.1735e-03,  7.0897e-03,  2.2526e-02,\n",
       "         2.7236e-03, -2.3878e-02, -6.8264e-04, -3.9944e-03, -2.3754e-03,\n",
       "         5.7223e-03,  5.1832e-03,  2.2398e-02,  1.4690e-02, -8.3528e-03,\n",
       "         2.7627e-03,  1.0361e-02,  5.3912e-04,  9.6905e-03,  2.4330e-02,\n",
       "         9.2200e-03,  2.0261e-02,  5.6526e-03, -1.4023e-02, -7.3169e-04,\n",
       "         5.2120e-04, -8.0866e-03, -1.5795e-02,  1.9199e-02,  6.4104e-03,\n",
       "         1.8358e-03, -1.2934e-02, -9.2095e-04,  1.0712e-02, -9.3379e-03,\n",
       "         6.9883e-04, -1.0149e-02,  1.0836e-03,  5.0871e-03,  7.4284e-03,\n",
       "         1.2419e-03,  1.7163e-03,  4.7596e-03,  3.9946e-03, -2.7214e-03,\n",
       "         1.3198e-03,  1.9491e-02,  4.2989e-03, -1.8044e-02, -5.3742e-03,\n",
       "         1.0179e-02,  3.1575e-03, -1.6595e-03, -1.3934e-02, -1.4361e-02,\n",
       "         1.4266e-03, -5.6741e-03,  2.8754e-03,  1.3568e-02, -8.3395e-03,\n",
       "         3.5569e-04, -2.1566e-03,  6.7897e-03, -1.8947e-02,  7.9670e-03,\n",
       "         1.7915e-02,  1.5730e-02, -1.6006e-03,  4.0837e-03, -4.5180e-03,\n",
       "        -1.0211e-02, -4.7099e-03, -3.0190e-03,  1.9660e-02,  2.5726e-03,\n",
       "        -3.4834e-03, -7.4241e-03,  2.4775e-04,  3.8942e-04, -7.5450e-03,\n",
       "         2.4633e-02,  1.0720e-03,  1.3190e-02,  8.2422e-03,  1.5329e-02,\n",
       "         6.7939e-03, -4.3089e-03, -1.1055e-02, -2.1171e-03,  7.8146e-03,\n",
       "        -1.3811e-02, -7.1968e-03, -2.6865e-03,  5.0846e-03, -1.2240e-02,\n",
       "        -9.4904e-03, -5.8880e-04, -7.1443e-03, -1.9542e-02,  2.0224e-03,\n",
       "        -1.5931e-02, -1.4418e-03, -1.8138e-02, -1.9770e-04,  3.7847e-03,\n",
       "         5.1495e-03, -7.2872e-04,  2.0510e-03,  2.7803e-03,  1.2248e-02,\n",
       "        -1.0355e-02, -1.0372e-02,  3.0169e-03,  4.5474e-03,  2.3548e-03,\n",
       "        -7.4182e-03,  2.3385e-03,  1.3674e-03,  5.6179e-03, -4.3820e-03,\n",
       "        -1.3094e-02, -4.7490e-03], device='cuda:0'), tensor([[-0.0026,  0.0097,  0.0006,  ...,  0.0031,  0.0006,  0.0077],\n",
       "        [-0.0029,  0.0168,  0.0022,  ..., -0.0022, -0.0060,  0.0072]],\n",
       "       device='cuda:0'), tensor([0.0085, 0.0006], device='cuda:0'), tensor([-0.0036], device='cuda:0'), tensor([-0.0131], device='cuda:0')]\n",
       "    p: 0.16326530612244897\n",
       "    weight_decay: 0.0002\n",
       ")"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "optimizer = PESG(rn, \n",
    "                 a=loss_func.a, \n",
    "                 b=loss_func.b, \n",
    "                 alpha=loss_func.alpha, \n",
    "                 imratio=imratio, \n",
    "                 lr=lr, \n",
    "                 gamma=beta, \n",
    "                 margin=margin, \n",
    "                 weight_decay=weight_decay) #optim.Adam(rn.parameters(), lr = 0.0001, weight_decay=5e-2)   \n",
    "optimizer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 98,
   "metadata": {},
   "outputs": [],
   "source": [
    "swa_model = torch.optim.swa_utils.AveragedModel(rn)\n",
    "scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max=300)\n",
    "swa_start = 160\n",
    "swa_scheduler = optim.swa_utils.SWALR(optimizer, swa_lr=0.05)\n",
    "sched_type = \"swa\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|█████████████████████████████████████████| 100/100 [29:08<00:00, 17.49s/it]\n"
     ]
    }
   ],
   "source": [
    "torch.cuda.empty_cache()\n",
    "\n",
    "rn = resnet152(1, 2)#.to(device)\n",
    "torch.manual_seed(123)\n",
    "\n",
    "loss_func = nn.CrossEntropyLoss()\n",
    "optimizer = optim.SGD(rn.parameters(), lr = 0.01, momentum=0.9, weight_decay=0.0001)\n",
    "scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer, factor = 0.1, patience=5)\n",
    "sched_type = \"\"\n",
    "\n",
    "num_epochs = 100\n",
    "nn_epochs = {}\n",
    "\n",
    "# Train the model\n",
    "total_step = len(train_iterator)\n",
    "for epoch in tqdm(range(num_epochs)):\n",
    "    nn_epochs[epoch] = {}\n",
    "    losses = []\n",
    "    #running_loss = 0\n",
    "    \n",
    "    for phase in [\"train\", \"val\", \"test\"]:\n",
    "        total_step = len(loaders[phase].dataset)\n",
    "        correct = 0\n",
    "        total = 0\n",
    "        pred_ys = []\n",
    "        labs = []\n",
    "        \n",
    "        if phase == \"train\":\n",
    "            rn.train(True)\n",
    "            \n",
    "            for i, (images, labels) in enumerate(loaders[phase]):\n",
    "                images = images[:,None,:,:].to(device)\n",
    "                labels = labels.to(device)\n",
    "                \n",
    "                outputs = rn(images)\n",
    "                pred_y = torch.max(outputs, 1)[1].data.squeeze()\n",
    "                \n",
    "                loss = loss_func(outputs, labels)\n",
    "\n",
    "                losses.append(loss.item())\n",
    "                optimizer.zero_grad()\n",
    "\n",
    "                loss.backward()\n",
    "                optimizer.step()\n",
    "\n",
    "                pred_ys = pred_ys + pred_y.tolist()\n",
    "                labs = labs + labels.tolist()\n",
    "                total += len(labels)\n",
    "                correct += (pred_y == labels).sum().item()\n",
    "        else:\n",
    "            rn.train(False)\n",
    "            \n",
    "            with torch.no_grad():\n",
    "                for i, (images, labels) in enumerate(loaders[phase]):\n",
    "                    images = images[:,None,:,:].to(device)\n",
    "                    labels = labels.to(device)\n",
    "                    \n",
    "                    outputs = rn(images)\n",
    "                    pred_y = torch.max(outputs, 1)[1].data.squeeze()\n",
    "                    \n",
    "                    loss = loss_func(outputs, labels)\n",
    "\n",
    "                    losses.append(loss.item())\n",
    "                    optimizer.zero_grad()\n",
    "                    \n",
    "                    pred_ys = pred_ys + pred_y.tolist() #test_output.flatten().tolist()\n",
    "                    labs = labs + labels.tolist()\n",
    "                    total += len(labels)\n",
    "                    correct += (pred_y == labels).sum().item()\n",
    "        \"\"\"\n",
    "        for i, (images, labels) in enumerate(loaders[phase]):  \n",
    "            # Move tensors to the configured device\n",
    "            #print(images.shape)\n",
    "            images = images[:,None,:,:].to(device)\n",
    "            labels = labels.to(device)\n",
    "\n",
    "            # Forward pass\n",
    "            #print(images.shape)\n",
    "            outputs = rn(images)\n",
    "            #print(torch.max(outputs, 1)[1])\n",
    "            pred_y = torch.max(outputs, 1)[1].data.squeeze()\n",
    "            ##print(outputs.shape)\n",
    "            #print(labels.shape)\n",
    "            #print(pred_y.shape)\n",
    "            #pred_y = torch.max(outputs, 1)[1].data.squeeze()\n",
    "            loss = loss_func(outputs, labels)\n",
    "            \n",
    "            losses.append(loss.item())\n",
    "            #running_loss += loss.item()\n",
    "            # Backward and optimize\n",
    "            optimizer.zero_grad()\n",
    "            \n",
    "            if phase == \"train\":\n",
    "                loss.backward()\n",
    "                optimizer.step()\n",
    "            #print(i)\n",
    "\n",
    "\n",
    "            #print(pred_y)\n",
    "            pred_ys = pred_ys + pred_y.tolist() #test_output.flatten().tolist()\n",
    "            labs = labs + labels.tolist()\n",
    "            total += len(labels)\n",
    "            correct += (pred_y == labels).sum().item()\n",
    "            \"\"\"\n",
    "\n",
    "       # print ('Epoch [{}/{}], Step [{}/{}], Loss: {:.4f}' \n",
    "        #       .format(epoch+1, num_epochs, i+1, total_step, loss.item()))\n",
    "\n",
    "        #print(f'Accuracy of the model on the {total} images: %.3f' % (correct/total))\n",
    "\n",
    "        cm = np.zeros((2,2))\n",
    "        for i, p in enumerate(pred_ys):\n",
    "            label = labs[i]\n",
    "            cm[label, p] += 1\n",
    "\n",
    "        #print(f\"Sensitivity = {cm[1,1]/(cm[1,1] + cm[1,0])}\")\n",
    "        #print(f\"Specificity = {cm[0,0]/(cm[0,0] + cm[0,1])}\")\n",
    "        #print(f\"AUC = {roc_auc_score(labs, pred_ys)}\")\n",
    "\n",
    "        nn_epochs[epoch][phase] = {\"loss\": loss.item(), \"acc\": correct/total, \n",
    "                                 \"sens\": cm[1,1]/(cm[1,1] + cm[1,0]), \"spec\": cm[0,0]/(cm[0,0] + cm[0,1]),\n",
    "                                \"auc\": roc_auc_score(labs, pred_ys)}\n",
    "\n",
    "    #print()\n",
    "    avg_loss = sum(losses)/len(losses)\n",
    "    scheduler.step(avg_loss)\n",
    "\n",
    "    if sched_type == \"swa\":\n",
    "        if epoch > swa_start:\n",
    "            swa_model.update_parameters(rn)\n",
    "            swa_scheduler.step()\n",
    "        else:\n",
    "            scheduler.step()\n",
    "    elif sched_type == \"12\":         \n",
    "        scheduler1.step()\n",
    "        scheduler2.step()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 260,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open(f\"{model_save_dir}ep{num_epochs}-bs{batch_size}-{xdtype}{ydtype}resnet18-sgd-lr0.01-lrplat.pickle\", \"ab\") as f:\n",
    "    pickle.dump(nn_epochs, f)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open(f\"{model_save_dir}ep1000-{xdtype}{ydtype}resnet152-sgd-lr0.01-lrplat.pickle\", \"rb\") as f:\n",
    "    nn_epochs = pickle.load(f)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5, 1.02, 'ResNet152 (polar channelled features, lag 0ms, ws 20)')"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, axs = plt.subplots(3)\n",
    "fig.tight_layout()\n",
    "\n",
    "losses = getCNNdata(nn_epochs, \"loss\", 0, num_epochs)\n",
    "aucs = getCNNdata(nn_epochs, \"auc\", 0, num_epochs)\n",
    "sens = getCNNdata(nn_epochs, \"sens\", 0, num_epochs)\n",
    "specs = getCNNdata(nn_epochs, \"spec\", 0, num_epochs)\n",
    "#print(aucs)\n",
    "lw = 1\n",
    "\n",
    "#val_aucs = [el*1.1 for el in aucs[\"val\"]]\n",
    "#test_aucs = [el*1.1 for el in aucs[\"test\"]]\n",
    "\n",
    "    \n",
    "axs[0].plot(np.arange(len(aucs[\"train\"])), aucs[\"train\"], label=\"Train\", lw=lw)\n",
    "axs[0].plot(np.arange(len(aucs[\"val\"])), aucs[\"val\"], label=\"Val\", lw=lw)\n",
    "axs[0].plot(np.arange(len(aucs[\"test\"])), aucs[\"test\"], label=\"Test\", lw=lw)\n",
    "#axs[0].set_xlabel(\"Epoch #\")\n",
    "axs[0].set_ylabel(\"AUC\")\n",
    "axs[0].grid(True)\n",
    "\n",
    "\n",
    "axs[1].plot(np.arange(len(sens[\"train\"])), sens[\"train\"], label=\"Train\", lw=lw)\n",
    "axs[1].plot(np.arange(len(sens[\"val\"])), sens[\"val\"], label=\"Val\", lw=lw)\n",
    "axs[1].plot(np.arange(len(sens[\"test\"])), sens[\"test\"], label=\"Test\", lw=lw)\n",
    "#axs[1].set_xlabel(\"Epoch #\")\n",
    "axs[1].set_ylabel(\"Sensitivity\")\n",
    "axs[1].grid(True)\n",
    "\n",
    "leg = axs[1].legend(bbox_to_anchor=(1, 1))\n",
    "for legobj in leg.legendHandles:\n",
    "    legobj.set_linewidth(2.0)\n",
    "\n",
    "#axs[1].legend(bbox_to_anchor=(1, 1))\n",
    "\n",
    "axs[2].plot(np.arange(len(specs[\"train\"])), specs[\"train\"], label=\"Train\", lw=lw)\n",
    "axs[2].plot(np.arange(len(specs[\"val\"])), specs[\"val\"], label=\"Val\", lw=lw)\n",
    "axs[2].plot(np.arange(len(specs[\"test\"])), specs[\"test\"], label=\"Test\", lw=lw)\n",
    "axs[2].set_xlabel(\"Epoch #\")\n",
    "axs[2].set_ylabel(\"Specificity\")\n",
    "axs[2].grid(True)\n",
    "#axs[1].legend(lw=1)\n",
    "    \n",
    "plt.suptitle(f\"ResNet152 (polar channelled features, lag 0ms, ws 20)\",y=1.02, fontsize=14)\n",
    "\n",
    "#fig.savefig(f\"{model_save_dir}-bs{batch_size}-{xdtype}{ydtype}resnet152-sgd-lr0.01-lrplat.pdf\", bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AUC:\n",
      "1.0\n",
      "0.9245297237327638\n",
      "0.9242265452478269\n",
      "Sens:\n",
      "1.0\n",
      "0.8641034139874048\n",
      "0.8627645502645502\n",
      "Spec:\n",
      "1.0\n",
      "0.9849560334781227\n",
      "0.9856885402311035\n",
      "66\n"
     ]
    }
   ],
   "source": [
    "opt = \"test\"\n",
    "idx = np.where(aucs[opt] == np.max(aucs[opt]))[0][0]\n",
    "print(\"AUC:\")\n",
    "print(aucs[\"train\"][idx])\n",
    "print(aucs[\"val\"][idx])\n",
    "print(aucs[\"test\"][idx])\n",
    "print(\"Sens:\")\n",
    "print(sens[\"train\"][idx])\n",
    "print(sens[\"val\"][idx])\n",
    "print(sens[\"test\"][idx])\n",
    "print(\"Spec:\")\n",
    "print(specs[\"train\"][idx])\n",
    "print(specs[\"val\"][idx])\n",
    "print(specs[\"test\"][idx])\n",
    "\n",
    "print(idx)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([37368, 20, 75])"
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "import onnx"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "rn.eval()\n",
    "dummy_input = torch.randn(1, 1, 20, 75).to(device)\n",
    "input_names = [ \"actual_input\" ]\n",
    "output_names = [ \"output\" ]\n",
    "model = rn"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "torch.onnx.export(model, \n",
    "                  dummy_input,\n",
    "                  \"polar-window20-lag0-resnet152-v2.onnx\",\n",
    "                  verbose=False,\n",
    "                  input_names=input_names,\n",
    "                  output_names=output_names,\n",
    "                  export_params=True,\n",
    "                  )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 193,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x7ffab91f11f0>]"
      ]
     },
     "execution_count": 193,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(np.arange(len(losses[\"train\"])), losses[\"train\"], label=\"Train\", lw=lw*4)\n",
    "plt.plot(np.arange(len(losses[\"val\"])), losses[\"val\"], label=\"Val\", lw=lw*1.5)\n",
    "plt.plot(np.arange(len(losses[\"test\"])), losses[\"test\"], label=\"Test\", lw=lw)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 238,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'train': [7.270035266876221,\n",
       "  0.515286386013031,\n",
       "  0.5183351039886475,\n",
       "  0.5408487319946289,\n",
       "  0.5167276859283447,\n",
       "  0.5146437883377075,\n",
       "  0.5169643759727478,\n",
       "  0.5154601335525513,\n",
       "  0.5274099111557007,\n",
       "  0.5202822685241699,\n",
       "  0.5189491510391235,\n",
       "  0.5256572365760803,\n",
       "  0.5200202465057373,\n",
       "  0.5174275636672974,\n",
       "  0.5167351961135864,\n",
       "  0.5166274309158325,\n",
       "  0.5167020559310913,\n",
       "  0.516844630241394,\n",
       "  0.5177151560783386,\n",
       "  0.5175105333328247,\n",
       "  0.5173516869544983,\n",
       "  0.5161396265029907,\n",
       "  0.5169404745101929,\n",
       "  0.5191710591316223,\n",
       "  0.5198994874954224,\n",
       "  0.519963264465332,\n",
       "  0.5196201205253601,\n",
       "  0.5189990997314453,\n",
       "  0.5184279680252075,\n",
       "  0.5180664658546448,\n",
       "  0.5178685784339905,\n",
       "  0.5177380442619324,\n",
       "  0.5176403522491455,\n",
       "  0.5175607204437256,\n",
       "  0.517490804195404,\n",
       "  0.5174334049224854,\n",
       "  0.5173697471618652,\n",
       "  0.5173293352127075,\n",
       "  0.517278254032135,\n",
       "  0.5172346234321594,\n",
       "  0.5159891843795776,\n",
       "  0.5156828761100769,\n",
       "  0.5155401229858398,\n",
       "  0.5154498219490051,\n",
       "  0.5153809785842896,\n",
       "  0.515328049659729,\n",
       "  0.5152803659439087,\n",
       "  0.5152902603149414,\n",
       "  0.5153022408485413,\n",
       "  0.5153120756149292,\n",
       "  0.5153184533119202,\n",
       "  0.515316367149353,\n",
       "  0.5153151154518127,\n",
       "  0.515313446521759,\n",
       "  0.5153128504753113,\n",
       "  0.515312671661377,\n",
       "  0.5153119564056396,\n",
       "  0.5153135061264038,\n",
       "  0.5153141021728516,\n",
       "  0.5153130292892456,\n",
       "  0.5153132081031799,\n",
       "  0.5153130292892456,\n",
       "  0.5153133869171143,\n",
       "  0.5153130292892456,\n",
       "  0.515313446521759,\n",
       "  0.5153136849403381,\n",
       "  0.5153135657310486,\n",
       "  0.5153138637542725,\n",
       "  0.5153133869171143,\n",
       "  0.5153133273124695,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038,\n",
       "  0.5153135061264038],\n",
       " 'val': [11.01222038269043,\n",
       "  0.47715458273887634,\n",
       "  0.48654910922050476,\n",
       "  0.4893379807472229,\n",
       "  0.5090436935424805,\n",
       "  0.4890928566455841,\n",
       "  0.4871407151222229,\n",
       "  0.4926392734050751,\n",
       "  0.4785277247428894,\n",
       "  0.487530916929245,\n",
       "  0.4864850640296936,\n",
       "  0.4950650632381439,\n",
       "  0.5032860636711121,\n",
       "  0.49847233295440674,\n",
       "  0.49585235118865967,\n",
       "  0.4951067566871643,\n",
       "  0.4934813380241394,\n",
       "  0.49294623732566833,\n",
       "  0.49798068404197693,\n",
       "  0.49688956141471863,\n",
       "  0.4983184039592743,\n",
       "  0.49538248777389526,\n",
       "  0.4888975918292999,\n",
       "  0.48722636699676514,\n",
       "  0.4885667562484741,\n",
       "  0.4901258945465088,\n",
       "  0.4909127950668335,\n",
       "  0.49075064063072205,\n",
       "  0.4899916350841522,\n",
       "  0.48927414417266846,\n",
       "  0.48884817957878113,\n",
       "  0.48860979080200195,\n",
       "  0.48851048946380615,\n",
       "  0.4884355664253235,\n",
       "  0.48840922117233276,\n",
       "  0.48837581276893616,\n",
       "  0.4883820414543152,\n",
       "  0.4883759617805481,\n",
       "  0.4883843660354614,\n",
       "  0.4884595572948456,\n",
       "  0.4886966943740845,\n",
       "  0.4888593554496765,\n",
       "  0.48894959688186646,\n",
       "  0.4890107214450836,\n",
       "  0.4890632927417755,\n",
       "  0.4891216456890106,\n",
       "  0.48915809392929077,\n",
       "  0.48927414417266846,\n",
       "  0.48932915925979614,\n",
       "  0.48936423659324646,\n",
       "  0.4893963932991028,\n",
       "  0.48941630125045776,\n",
       "  0.4894358515739441,\n",
       "  0.48943689465522766,\n",
       "  0.48944053053855896,\n",
       "  0.4894450902938843,\n",
       "  0.48944735527038574,\n",
       "  0.489446759223938,\n",
       "  0.48944729566574097,\n",
       "  0.48945170640945435,\n",
       "  0.48945099115371704,\n",
       "  0.48945140838623047,\n",
       "  0.4894520938396454,\n",
       "  0.48945215344429016,\n",
       "  0.4894508719444275,\n",
       "  0.48945218324661255,\n",
       "  0.4894520044326782,\n",
       "  0.48945152759552,\n",
       "  0.48945218324661255,\n",
       "  0.48945215344429016,\n",
       "  0.4894528090953827,\n",
       "  0.4894518256187439,\n",
       "  0.48945116996765137,\n",
       "  0.48945116996765137,\n",
       "  0.489451140165329,\n",
       "  0.48945099115371704,\n",
       "  0.48945191502571106,\n",
       "  0.48945242166519165,\n",
       "  0.4894523024559021,\n",
       "  0.4894523024559021,\n",
       "  0.4894523024559021,\n",
       "  0.4894523024559021,\n",
       "  0.4894523024559021,\n",
       "  0.4894523024559021,\n",
       "  0.4894523024559021,\n",
       "  0.4894523024559021,\n",
       "  0.4894523024559021,\n",
       "  0.4894523024559021,\n",
       "  0.4894523024559021,\n",
       "  0.4894523024559021,\n",
       "  0.4894523024559021,\n",
       "  0.4894523024559021,\n",
       "  0.4894523024559021,\n",
       "  0.4894523024559021,\n",
       "  0.4894523024559021,\n",
       "  0.4894523024559021,\n",
       "  0.4894523024559021,\n",
       "  0.4894523024559021,\n",
       "  0.4894523024559021,\n",
       "  0.4894523024559021],\n",
       " 'test': [13.745423316955566,\n",
       "  0.6060864925384521,\n",
       "  0.5939598679542542,\n",
       "  0.5951346755027771,\n",
       "  0.5962017178535461,\n",
       "  0.5942010879516602,\n",
       "  0.5933181047439575,\n",
       "  0.5918688774108887,\n",
       "  0.6097509860992432,\n",
       "  0.5913447737693787,\n",
       "  0.5918089747428894,\n",
       "  0.5914990305900574,\n",
       "  0.5936291813850403,\n",
       "  0.5923927426338196,\n",
       "  0.5919671654701233,\n",
       "  0.5918610095977783,\n",
       "  0.591401219367981,\n",
       "  0.5910488963127136,\n",
       "  0.5923630595207214,\n",
       "  0.5913559198379517,\n",
       "  0.5917371511459351,\n",
       "  0.5915464162826538,\n",
       "  0.5902040004730225,\n",
       "  0.5899884104728699,\n",
       "  0.5890851020812988,\n",
       "  0.5883572697639465,\n",
       "  0.5880500078201294,\n",
       "  0.5881310105323792,\n",
       "  0.5884732604026794,\n",
       "  0.5888490080833435,\n",
       "  0.5891156792640686,\n",
       "  0.5892918705940247,\n",
       "  0.5894098281860352,\n",
       "  0.5895031690597534,\n",
       "  0.5895687341690063,\n",
       "  0.5896384716033936,\n",
       "  0.5896890163421631,\n",
       "  0.5897292494773865,\n",
       "  0.5897806882858276,\n",
       "  0.5897985100746155,\n",
       "  0.5899157524108887,\n",
       "  0.5898067355155945,\n",
       "  0.5897901058197021,\n",
       "  0.5897966623306274,\n",
       "  0.5897970199584961,\n",
       "  0.5897948145866394,\n",
       "  0.5897983908653259,\n",
       "  0.5897893905639648,\n",
       "  0.5897617340087891,\n",
       "  0.5897383093833923,\n",
       "  0.5897214412689209,\n",
       "  0.5897172093391418,\n",
       "  0.5897058844566345,\n",
       "  0.5897093415260315,\n",
       "  0.5897094011306763,\n",
       "  0.589704692363739,\n",
       "  0.5897049307823181,\n",
       "  0.5897060632705688,\n",
       "  0.5897055864334106,\n",
       "  0.5897031426429749,\n",
       "  0.5897023677825928,\n",
       "  0.5897035598754883,\n",
       "  0.5897029638290405,\n",
       "  0.5897030830383301,\n",
       "  0.5897039175033569,\n",
       "  0.5897030830383301,\n",
       "  0.5897032618522644,\n",
       "  0.5897035002708435,\n",
       "  0.5897034406661987,\n",
       "  0.5897035598754883,\n",
       "  0.5897032022476196,\n",
       "  0.5897035002708435,\n",
       "  0.5897038578987122,\n",
       "  0.5897037386894226,\n",
       "  0.5897035598754883,\n",
       "  0.5897035598754883,\n",
       "  0.5897031426429749,\n",
       "  0.5897027850151062,\n",
       "  0.5897029638290405,\n",
       "  0.5897029638290405,\n",
       "  0.5897029638290405,\n",
       "  0.5897029638290405,\n",
       "  0.5897029638290405,\n",
       "  0.5897029638290405,\n",
       "  0.5897029638290405,\n",
       "  0.5897029638290405,\n",
       "  0.5897029638290405,\n",
       "  0.5897029638290405,\n",
       "  0.5897029638290405,\n",
       "  0.5897029638290405,\n",
       "  0.5897029638290405,\n",
       "  0.5897029638290405,\n",
       "  0.5897029638290405,\n",
       "  0.5897029638290405,\n",
       "  0.5897029638290405,\n",
       "  0.5897029638290405,\n",
       "  0.5897029638290405,\n",
       "  0.5897029638290405,\n",
       "  0.5897029638290405,\n",
       "  0.5897029638290405]}"
      ]
     },
     "execution_count": 238,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "losses"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 101,
   "metadata": {},
   "outputs": [
    {
     "ename": "KeyError",
     "evalue": "1",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyError\u001b[0m                                  Traceback (most recent call last)",
      "\u001b[0;32m/tmp/ipykernel_246677/2585576021.py\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0maccs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mgetCNNdata\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mnn_epochs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m\"acc\"\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m0\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mnum_epochs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m/tmp/ipykernel_246677/682399639.py\u001b[0m in \u001b[0;36mgetCNNdata\u001b[0;34m(epochs, dtyp, epoch_start, epoch_end)\u001b[0m\n\u001b[1;32m     13\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     14\u001b[0m     \u001b[0;32mfor\u001b[0m \u001b[0mepoch\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mrange\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mepoch_start\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mepoch_end\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 15\u001b[0;31m         \u001b[0;32mfor\u001b[0m \u001b[0mphase\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mepochs\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mepoch\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mkeys\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     16\u001b[0m             \u001b[0mdata\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mphase\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mdata\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mphase\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0mepochs\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mepoch\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mphase\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mdtyp\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     17\u001b[0m             \u001b[0;31m#aucs[phase] = aucs[phase] + [cnn_epochs[epoch][phase][\"auc\"]]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyError\u001b[0m: 1"
     ]
    }
   ],
   "source": [
    "accs = getCNNdata(nn_epochs, \"acc\", 0, num_epochs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 117,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x7f1f6411da00>,\n",
       " <matplotlib.lines.Line2D at 0x7f1f6411da30>,\n",
       " <matplotlib.lines.Line2D at 0x7f1f6411db80>,\n",
       " <matplotlib.lines.Line2D at 0x7f1f6411dca0>,\n",
       " <matplotlib.lines.Line2D at 0x7f1f6411ddc0>,\n",
       " <matplotlib.lines.Line2D at 0x7f1f6411dee0>,\n",
       " <matplotlib.lines.Line2D at 0x7f1f64129040>,\n",
       " <matplotlib.lines.Line2D at 0x7f1f64129160>,\n",
       " <matplotlib.lines.Line2D at 0x7f1f64129280>,\n",
       " <matplotlib.lines.Line2D at 0x7f1f641293a0>,\n",
       " <matplotlib.lines.Line2D at 0x7f1f6417b3a0>,\n",
       " <matplotlib.lines.Line2D at 0x7f1f641294c0>,\n",
       " <matplotlib.lines.Line2D at 0x7f1f641296d0>,\n",
       " <matplotlib.lines.Line2D at 0x7f1f641297f0>,\n",
       " <matplotlib.lines.Line2D at 0x7f1f64129910>,\n",
       " <matplotlib.lines.Line2D at 0x7f1f64129a30>,\n",
       " <matplotlib.lines.Line2D at 0x7f1f64129b50>,\n",
       " <matplotlib.lines.Line2D at 0x7f1f64129c70>,\n",
       " <matplotlib.lines.Line2D at 0x7f1f64129d90>,\n",
       " <matplotlib.lines.Line2D at 0x7f1f64129eb0>]"
      ]
     },
     "execution_count": 117,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXoAAAD4CAYAAADiry33AAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjQuMiwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8rg+JYAAAACXBIWXMAAAsTAAALEwEAmpwYAACENElEQVR4nOz9eZQkyX3fCX7M3cM97iPvo/Koyjq7q+/qC2gSDTZAgCBISiQwEqGhdjniYsBZiiJFrXY0q32j1Vs+aebNDIkZUcvhUhyNRqSwQwAjkOCJG+gGutF3d3XXmXXkfWfch1+2f7jHlRlZlVWVWZmV5Z/3LM3czdzDIzLi+/vZz8zNhZSSgICAgICDi7LXFxAQEBAQsLsEQh8QEBBwwAmEPiAgIOCAEwh9QEBAwAEnEPqAgICAA4621xfQiZ6eHjk+Pr7XlxEQEBBwz/D666+vSCl7O9XtS6EfHx/ntdde2+vLCAgICLhnEEJc36ouCN0EBAQEHHACoQ8ICAg44ARCHxAQEHDACYQ+ICAg4IATCH1AQEDAAScQ+oCAgIADTiD0AQEBAQecfTmPPuD2kFJiLy9jXb+Oef06TqFI6qc+idbTs9eXFhAQsIcEQn+PIaXEWVvDvH4d85on6PVkXb+OWy63tV/+H/9Huv7u36X77/1nqMnkHl11QEDAXhII/T5ESomTzWJeu9Ym4ua165hTU7jFYrOxqhI6NIw+Nkb0zBn0sTEvjY8hLYuVf/U7rP7P/zPrX/gC3b/09+j6T/9TlEhk795cQEDAlkgpEULs+HnFfnzC1JkzZ+T9sASCk802PXJfxOvbbj7fbKgohIaHmyLuC7k+OkpoeBgRCt3wdarnzrH825+n+J3voPb20PPLv0zmU59C6Pouv8OAgICbIS2L4ksvkf/qn2EtzDP+7//9bZ1HCPG6lPJMx7pA6HcXJ5/HvF4X8BYP/dp1nFyu2VAIQkND6GNjhMZGW0R9HP3Q8I6Icvn111n6rd+i8trrhA4dovfv/wrJT34Soap3fO6AgIDtI12XyptvkvvqVyn8xV/iZLMoqRTJj32MgX/6/7it33sg9DuMa5o461mcbBZnfd3Ls1mc7DrOehZ7fQ1ratobEF1fbztWGxzc7JmPjREaGUG5Cx62lJLSiy+y9Fu/Re39cxjHjtL7D/4B8Rde2JUuY0BAQJPqxYvk//Sr5P/sz7Dm5hDhMIkf+zDJT36S+HPP3ZFDFwj9FkgpkZUKzvo6dl2s6wK+UcRbyhsHPFsR0ShaOk3o0KE2IW+IeTi86+9rO0jXpfDXf83yb38e89o1wo88TN+v/zqxZ57Z60sLCDhQWLOz5P7sz8l/9avULl4EVSX2gQ+Q+uRPEn/hI6jx2I68zn0h9FJK3EKhozDbjXJuk3hL09zynEoyiZpOeymTRmuUM8396QxqJt3YVgzjTt/+XUXaNrn/+B9Z/p1/jT0/T/TZZ+j79V8n8vDDe31pAQH3LPb6OoW//EtyX/0zKq+/DkDk0UdJfvKTJH/i42jd3Tv+mveN0J9/+BGwrM2VioKaSrULdIs4a237/XIqhdDun0lJbq1G9gtfYOV3/2ec9XUSH/0Ivb/6qxjHju31pQUE3BO45TKFb3yT/Fe/SvGll8C20ScmSP3UJ0n+5E+ij4zs6uvfF0IPsP6FLyCM8CbvW0kkEEpwE/B2cIol1v7d/8raH/wvuKUSqZ/+aXr+/q+gHzq015cWELDvkJZF6fvfJ/enX6XwjW8gKxW0gQGSP/kJUj/1UxgnTty1sa/7RugDdg57fZ3V/+/vs/6Hf4h0XTKf/jQ9v/w5tN6OTyoLCLhvkK5L5a23yP3pn26aMZP6qU8SeeKJPXEsA6EPuG2sxUVW/vX/h+yXvoTQNLp+4Rfo/qW/h5pK7fWlBQTcVbaeMfNTxJ/74J7flxIIfcAdY16/zvL/9K/I/9mfocTjdP+9v0fX3/0FlGh0ry8tIGDXsGZnyf35n5P/092dMbMTBEIfsGNUL1xg+bd+m+K3v43a00PP5z5H5j/59J57M/cSUkpwHKRtI20b/NxLDtgWsl5v2eA066Rtee0dp73OspFO/Xwt53b8Oturp34e1yHUP4BxdALj6FFCIyPBjXM+ezFjZicIhD5gxym/8SbLv/VblF99ldDwMD2/8iukfvqn7muxcEslb/XQpSXspWXspSUvLTfL1vIy8gb3YewamobQNISqerPJhMDJZhvVQtfRjxzBmJjAOHYUfWICY+Io+ujIfTH7zC2XKXzzW+T/9E/3ZMbMThAIfcCuIKWk9NL3Wf6t36L63nvoExP0/oNfJfHRjx6ou2zdSqVdrOtC3rLPXlrCLZU2HSvCYbT+PkK9fWh9fWi9vSjxOEJTffENeeIb0kBVve2QJ8gNcfYTqrahLoTQ1PY6rV7fWqd2/H84xSLmlSvULl2mNjlJ7fIlzMuTWHNzzesPhdAPH8Y4ehT9qCf+xrGj6CMjN11jaT8hHcczwnPz2AvzWPPzWPMLWPPz2PPz1K5e3dMZMzvBHQu9EOLjwOcBFfh9KeW/7NDmeeC3gRCwIqX8kBBiBPh3wADgAr8npfz8zV4vEPp7Cyklhb/6a5Y//3nMq1cJnz5N76//GrEPfGBf/1DcWm2TWNe3W71yt1DYdKwwDE+4ffHW+noJ1bcb+/o8Ud/Hn0En3FKJ2pUr1C5PYk5ebhgCa2am2SgUwhgf98T/6FHPABydQB8bu+sGoL7aq72w4Au4J951IbcW5rEXl8Bx2o5T4nFCgwP+siTjJH/8o3s2Y2YnuCOhF0KowEXgo8AM8Crw81LK91vapIHvAx+XUk4JIfqklEtCiEFgUEr5hhAiAbwO/I3WYzsRCP29ibRtcl/5E5Z/519hz80Tffpp+n7914g8+uj2z+G6SNNEWtbmvNO+DblrmuDnG49x8wVfzD0Rb1tUzkeEQm1C3Uy+kNcFPJm85wT8TnHLZWpXrnrif/kytcuT1C5f9gxAXUc0DX18DOPoMS8M5BsCfWzstsdx3EoFa35hgyc+h90Q8gVkpdJ+UChEaGDAS0ODaAODhAYHG8IeGhxETSTu8BPZX9yp0D8L/DMp5cf87X8CIKX8Fy1t/gtgSEr5T29yrq8A/0pK+bUbtQuE/t7GNU2yX/j/sfK7v4uztkb4kYcRoZAvuhbSMv18s3hv9LruGEVB6DoiFEKJx1sEu6+DoPeiptP3nYDfKW6lQu3KFczJyYb41yYvY01NtxuAsbF28ffHAJxsdmtPfG6+bSyhjtrbQ2hwyBPywUG0wQFve9DbVru771nP/Ha5U6H/FJ6n/kv+9i8AT0spf6WlzW/jhWweBBLA56WU/27DecaB7wKnpZR5NiCE+CzwWYDR0dEnrl+/vt33F7BPcUsl1v63/43id7/nxY59wd0yD4UQeggR0v28WafoOvj5zc/Rsu8+Hhzea9xqFfPq1TbxNy9PYk5NgetueZySSBAaGEAbGiTUwRPX+vvvykqv9xo3EvrtDKd3cm82WgcNeAJ4AYgAPxBCvCylvOhfQBz4EvBrnUQeQEr5e8DvgefRb+O6AvY5SixGz+c+R8/nPrfXlxKwByjhMOFTpwifOtW2363VMK9do3bpMtbMNGqmq+GJa4ODqPH4Hl3xwWU7Qj8DtM4tOgTMdWizIqUsASUhxHeBR4CLQogQnsj/oZTyyztwzQEBAfcwimEQPnGC8IkTe30p9w3bCWK9ChwTQhwWQujA3wb+ZEObrwA/IoTQhBBR4GngnPCCnf8GOCel/B928sIDAgICArbHTT16KaUthPgV4K/wplf+gZTyPSHE5/z635VSnhNC/CXwDt40yt+XUp4VQjwH/ALwrhDiLf+U/5WU8s93480EBAQEBGwmuGEqICAg4ABwo8HY+2v+UUBAQMB9SCD0AQEBAQecQOgDAgICDjiB0AcEBAQccAKhDwgICDjgBEIfEBAQcMAJhD4gICDggBMIfUBAQMABJxD6gICAgANOIPQBAQEBB5xA6AMCAgIOOIHQBwQEBBxwAqEPCAgIOOAEQh8QEBBwwAmEPiAgIOCAEwh9QEBAwAEnEPqAgICAA04g9AEBAQEHnEDoAwICAg44gdAHBAQEHHC0vb6AneQ3zk+R1FTGI4afdIYNHU0Re31pAQEBAR2puS7nilXeKZTJ2g6/Ota/469xYITekZLX82WuVmrUXNnYrwkYCett4l8vj4Z1ImrQqQkICLg7tIr6O4UK7xTKnCtVsaSnWYNGiF8Z7UMRO+ucbkvohRAfBz4PqMDvSyn/ZYc2zwO/DYSAFSnlh7Z77E6gCsG3nzqJKyULNYtrFZNrlZqXqibXyjVez5fI227bcYNGqCn+YYPxaL2skwodGDsYEBBwl7mZqKc1lYcTEf7zkV4eTkR5JBFhNKwjdljkAYSU8sYNhFCBi8BHgRngVeDnpZTvt7RJA98HPi6lnBJC9Ekpl7ZzbCfOnDkjX3vttdt/V1sgpWTddrhW9sW/UuNqpcZ13ygsmXZb+0wjDLS5R9Cna7vyDwkI2C3ytsP5YoVzpSqT5RqHowbPZxKMR3ZHXO4ntivqDyeiuybqQojXpZRnOtVtx2V9Crgspbzin+wLwM8ArWL9GeDLUsopACnl0i0ce9cQQtAV0uhKaTyeim2qL9kO1xsGwOS63yN4LV/mK0tZWvsCEUVpiP5YROdwMC4QsE8wXZfJco1zpSrnfGE/V6wwW7MabQxFNEKco2Gd57sS/GgmwXOZOOmgJ3tD9pOnvl228x8dBqZbtmeApze0OQ6EhBDfBhLA56WU/26bx+4bYprKA/EID8Qjm+pM12WmanHVF//rFZOrlRqXy1W+uZZvGxdQBaQ0lZSmkrxJntiwndJUoqoSeFgBN0VKyWzN4lyxwvlSlXOlKu8XK0yWaw3R0QRMRMM8mYrxf4pHOBkLcyoe4ZAR4mrF5DvrBb6zlufLi+v8u7lVFOCxZJQPdSV4PpPgsWSM0H3stNRcl/OlKm/n7x1R78R2hL7TFW+M92jAE8ALQAT4gRDi5W0e672IEJ8FPgswOjq6jcu6u+iKwpGowZGosanOlZL5mtUwANerJuuWTd52yNkOedthvmaR98sV98bhsrqhSG7TMGxsEwsMxYEjZ9meZ+575+dLVc6XKm1jTsNGiJOxCB/pTnLKF/SjUQNd6TzhoP59/sXhHixX8ka+xLfXCnx3vcBvX1vkf7i2SFxVeC4T50NdST6USXD4AId5Doqod2I7Qj8DjLRsHwLmOrRZkVKWgJIQ4rvAI9s8FgAp5e8BvwdejH5bV79PUIRgOKwzHNb5YObm7Wuu2xD9uiHI227bdiO3vHzJrDX2lx33hudXgKSmkg6pjIZ1xvwZRvUw03jEIKmpO/Pm7wFsV7JoWlRcl6iiEFMVoqq6Lz3VmutyuVxrC7mcL1WZawm7JDWFU7EIf7Mvw6l4hFOxMCdj4TuaPBBSBE+n4zydjvN/Z5CsZfPiepHvrhf41lqBv1zJA94Mtue7EnzoHg7zWK7kerXGpVKVS+Ual8pVLhSrB0bUO7GdwVgNb0D1BWAWb0D1M1LK91ranAL+FfAxQAd+CPxt4PzNju3Ebg3GHhQsV3YwFM4mQ7Fq2Uz7Yw5rltN2joymMtYi/GNhnVG/PGiEUO+RL7SUkhXLZrZqMVczmatZzFbb80XTwunwNdeFIKrWhb+ZYqra3K8obW3a6jYe4xsRQxE3FQRXSqarphdyaYh6lclKtXGtISE4FjU41RJyORULM2SE7qrgSCm5VjH59nqB764VeHG9QMFxUYBHk1E+lEnwfFeCx/dZmKdkO1wqe+HVS+W6sFe5VjEbgg7ezLtjUWNXB0rvBjcajL2p0Psn+ATe1EkV+AMp5W8KIT4HIKX8Xb/N/w34RcDFm0b521sde7PXC4R+5ynYDlO+6NdnGU1VTK5Va8xUTeyWr4EuBCMtwj8W1hsGYTSsE7tLvQEpPYM2V7OY8UW7VcDnaiZzVQtzw3fYUARDRoghQ2c4HGLY0BkKh4ipKmXHpew4lByXsuN2yB3Kruu38/aXHKft87kZCmwyDK3GYaFmcb5UpdTSMxsJ641wy6lYmJPxMBOR8L4SzjqWK3kzX/Lj+wXeyJdxgbiq8MFM3Bf+5F0J89QN/UXfO79crnKp5Hnprb0gVcDhiMGxaJijUYNjsXCjnDggvds7Fvq7TSD0dxfblczVzMb4QnOswctzdntvoFfXGPNDQaMRnbGwN9toLGLQfwvTTsuO2xDr2ZrZ9Mr97bma1SaG4P1gB/QQw2G9o5gPGTrdIXXHBcZ0b2wY6tvl1v0d27t06yqnYhFOxcOcikU4EQvf02KTs2xezBb5zlqBb68VmKqaABwKh3g+k+RDXV6YJ3MHYR5HSqYqJpdavPO6p976/YypiifkUU/Ij8UMjkbDjEf0LccqDgqB0AfcEVnL5pov/FOVll5BtcZc1dow7VQw0hB+T/xTmspCzWK24Y17Yr6+wYAA9OlaQ7zrQj4UDnHIz/v0eyesdL9yrVLzBnXXCnyvJczzSCLqxfe7EjyxRZin7Lhc8QX8YqnKZT+GfqVca+u59ekaR6NhjrV458eiXtjxXgu57BSB0AfsGvVpp9crtUZvoGEMqmabR57WVIaMpje+0SsfMEIYB9zrut+wXcmbhTLfXsvz3bUibxRKONLzvD+YjvNkKsaKaXOx7In6TNVsTMtTgLGI7odYPO/8eDTMRNS4JweBd5tA6AP2BCklq5ZDzrYZMLwYecD9Tc6yeSlbbEzjvFYxiSiCox1i50eiRmD4b4E7vTM2IOC2EELQo2v06MHXLMAjFdL4RG+aT/SmAVi3bFKauuOLeAW0E/wCAwIC9ow7GaAN2D5BvyggICDggBMIfUBAQMABJxD6gICAgANOIPQBAQEBB5xA6AMCAgIOOIHQBwQEBBxwAqEPCAgIOOAEk1gDAgIC9hhrqUzl3RWcfI3M3zy24+cPhD4gICDgLiOlxJotUnlvlcrZFezlCgD6eBLpuAh1Z4MtgdAHBAQE3AWkKzGn8lTOeuLuZGuggHE4RfwDQ0Qe6EZNbX5U6U4QCH1AQEDALiEdl9qVHJWzK1TeX8UtWKAKwscyJF8YJfxAN2ostOvXcaCEfv6/+SFCU1ATOkpCR/WTktRRE6HGtohs/+EYAQEBAbeCtFyql9apvLdK9dwqbtlGhBTCJ7uIPNhN+GQXSvjuSu+BEXrpSsKnunALFk7BxJwp4OZNpNXhQdqaQI3rTYOQ1FHjId8gtBiIuI5Q959BkI5E1mzcqoNbtZFVB7fm51Ubt+YgqzbSkYQGYuiH4mi9UcQ+fCxdQMBBwK3ZVC+sUzm7QvX8OtJ0EGGVyKluIqe7MY5lUPS9W6b7wAi9UARf/vp/QySVom98gr7Hj9A7dpju/hFEVeIWTJyCieMbAjdv4hRN7NUK5rUcbtnucFJQYqHNPYR67yCpo8a9HsN2/olSSqTlImstAl31BFu2CHRDwDu1q9lIs4Px2ogivMmz/sNOha4SGo6jj8TRDyXQDyVQM0bQswkIuE3cskXl3Jon7pfWwZYo8RDRR3uJnO7BOJJCaPtjBvuBEXrHtjj8+BmWrl7h3Pe+ydt//WcAKKpK1/AIfeNHPANw+Ah9jx/BiMbajpe2i1O0PIPgGwEnb7YYCBN7sYRTsMDd/LAWYahNIxDXkbbbUcA7HbvpXLqCCGsoYRXF0BBhlVDaQBgqir+/Xi8Mv13Ya1evR1NAgr1cxpwpYs4UMGeKFF+aA8e7BiWmERpOoI8k0A95BkBN6Dvw3wgIOJg4BdObKfPeCrXJHLgSNaUTf2qQyOke9PHkvuw5H8gnTEnXJbu0wPK1Kyxdu8LS1UmWrl2hlF1vtEn1D7SL//gE8UzXNs4tcSt20yD4RqDVILhFC6EpLcKrNUTa26du2KehGGpDuHczXCRtF2uh1BT/6QL2Upn689vUlIF+KE6oRfzvdjwxIGA/Ya9XvZky761gXs+DBK07TOR0D5HTPYQOxW+rZyylJL+8xOKVSyxcuczi5EWsWo3P/L//+9u6zvvuCVNCUcgMDJEZGOL4M8819pey6w3Rr+eXXvl+oz6aSvvif4S+wxP0jR8h3T+IaHmcmVAEaiyEGgsRGmjvFdwLCE1phG5gEADXdLBmm16/OVOg8t5q4xitJ+KJ/yHf+x+KIUL757GA0pW4RQsnX/OMbb6lN1Y3xnkTHBcR0VAinvFV/LJobKuNuka7+vY+6YIH3B2s5bI3U+bsKtZsEYDQQIzkC6NETveg9UdvWdyLa6sNQffyS1QKeQAUVaN3bJyBieNI123TnJ3gQHr0t0KtXGL52lWWrjUNwOrsNK7jAKBHIvSOHfY8f98AdB8aQdV2f0rUXuKWrbaQT31wGwAFQv0x9JEEId/rD/VHd/wmj4aAF0xPxDuIt5M3cYtmo0fSihJvjqUoCR2hKd6YR8X2emUV2xv7qNiNsYytECGlo5FQIn4vLRJqGAqxsd5Qg7GQfY6UEmu+1BB3e6kMgD6SIHK6m8iDPWg9kW2fr5zPsXjlMguTF1n0Rb24vgaAEArdI6P0HznKwMRxBo4cpWd0HE2/s7DpHT8cXAjxceDzgAr8vpTyX26ofx74CnDV3/VlKeU/9+t+HfglvJ/iu8AvSimrN3q9vX44uG1ZrE5f94T/2iRLV6+wfP0qVs27bEXV6B4ZbQ/9jB1Gj0T37JrvBk6u1ib85kwRWfEHsTUFfSjmib4f9tG6Ix3jldKVuCXLF+pac3C8VcD9cFhHAY+FvIHwZMusqfqMqaTh7YuHbskLl5bbJvxuxTcIVRu33NwvNxgIt+INkHe6zgaCNgPQdl2iNReI1u36RutH6O9rttvQZsO22OpcQnj/G1UgVOEZaUUgNOHlquLXKV4osbWdX6ZepwgvVNlyvnrdprKigEJHwyddCa5E2i7SkeC4SFsi/dzbrtfV27lQb+Pva9a1n6PRrsM5nGwNZ70GwruBKfJgN+HTPWjbuIGpWip6Yu4L+sKVS+SXlxr1maFDDEwcY+DIUfqPHKNv/AihcPim571V7kjohRAqcBH4KDADvAr8vJTy/ZY2zwP/SEr5yQ3HDgMvAg9IKStCiP8d+HMp5b+90WvutdB3wnUdsgvzzdCP7/3Xu14IQWZgkN7xCfrGDhPv6iYcTxBJJAjHk0QSCYxYDEXZPyGPO0VKSXUuR2lymdpUHmehglh3EY73I3aETVHJUxVlYkYSQ4mh2RqiKm8o4JvF+/YF/G4gXenNkGoxAp16DfV90h8MRzY/B+9nuOFzqZfrv9F6detvtrWNBNdxNiXpuLiug3QcXMdFul5SFA1NDaEoGopQUFAQUsA2JnXdEYKmMVEEuE3h3XHqxkrzDY62YVv1jVREI3Kii/ADXajxrT1rs1ph6eqk761fYvHKJdbn5xr1qb5++n0vvf/IMfqPTGya+LFb3GmM/ingspTyin+yLwA/A7x/w6PaXyMihLCAKDB3k/b7EkVR6Ro6RNfQIU5+8EOAJ3TFtdWG1790bZKFyxe4+IPvdT6JEISjMcKJBJF40s8ThBPJZp5IEI4lvDq/nWbcvWmQllmjkstRzuco57NU8nmvnKuXs1Ty9focdq3W/hYRJEPd9ERH6I2P0KUPkKGHSqXAirVCxSlScYrYmk0oHSHSnyZxqI/M4WHSY2PE0pl7LswhFNEI6+w00nWpFPKUclnK2Szl3LpXzmUpZdcp57ON7XIui3Q3q7SiakTTaWKpNNFMmlg6QySRpJzPkVtaILe4SGFtpc2AhIwwmb4h0r0DJHsGSPf0kejuI5HpIZHqQlE0zyN2WrxjV4Lteoav7k3X6zaUvdzz4GntKWhN8d2uSDePaz8Hqrij75Jtmixfv9oIvyxMXmJtdgYpvc843t3DwJGjPPCjLzAwcYz+I0eJJJK3/Xq7yXa+mcPAdMv2DPB0h3bPCiHexhPyfySlfE9KOSuE+O+AKaAC/LWU8q87vYgQ4rPAZwFGR0dv4S3sHUIIEt09JLp7mHii+ZHUymUq+RyVYp5qoUC1WKBSLFApFKgW835eoJRdZ3VmikqhgFWtbPk6aii0wSAkvN7CRiMRrxuIJOFYHEVVsU2Tcj7XFOdcu1BXWvNcrhGe6nQN0WSaSDJJNJWma+gQkVSaaDJFNJki4ufRlFcOGeG2H1ndKK7OTLE6M83qrJ+/+Qq1F0uNdkYsRvfwKN2HRug+NEr38Ahdh0ZJdPfccwZgK1zX8YxmLtsu2rks5ex6m3CX87mO4q1qGtFUhmgqTaKrm/7DR4ml00RTXoql0kTTGWKpDEYsdtPPzjZN8itL5BYXyC7Ok1taILu4yNriLFfee73doAtBvKubdN8Aqf6BZt4/SGpogEgiec/9rxzbZmX6eiP0sjB5idXp642xukgyxcDEMY4/80HfUz+6rVl6+4XthG4+DXxMSvlL/vYvAE9JKf9+S5sk4Eopi0KITwCfl1IeE0JkgC8BfwvIAn8MfFFK+e9v9Jr7MXSz2zi2RbVYpFLwjEOlxSBUCnk/9wxFo12x0PgidkLTDWyz1rFO1TRfnJviHU0miSRSRFPppnD74h0KR3blxyulpJzL+gag3QhU8rlGOz0SoWt4pGkERkbpHh4l2dO74zMUbhXbsqgUclTyeU/AC57hrBTyXt7Yl/f35RteYStqKLRBpNMNMa+LeCztbRvRm4v3TlH/H2UXF/wewIJvCObJLS40BhnrhMIR0n39pPoHG4Yg3e8Zg2Rv3x1NZJBS4lgWZrWCXath1apY1aqX12qY1Yq/z6uza9XGtlfnH1OrYlebx9RKRRzbG28yYjH6jxzz4+rH6J84SqK7d98brzuN0T8L/DMp5cf87X8CIKX8Fzc45hpwBvgw8HEp5d/z9/9d4Bkp5X9xo9e8H4X+dpBSYlYqzV5CId/ScyhgVitE4glPyJO+eKc88dYjtz497G5Tzuea4j8zxZpvAFrvh9AMg+7hkYbn333IMwSpvv7bGg/xPtOy38vJUynkWsr59n2+kJuVLXpjQniffyJJJJkikkg2DGfUF/GGqKfT98T/pBOWWSO/tNgwBHUDkFtaJLe4gG2ZjbZCKCR6ekj1DZDqGyDZ04vrOljVSkOc66JtVave/hZxtqq1jkZyS4QgZIQJGQahcNgr1/OW/eFYnL7xIwxMHCfVP3BP/h/uVOg1vMHYF4BZvMHYz0gp32tpMwAsSimlEOIp4IvAGF58/w+AJ/FCN/8WeE1K+T/d6DUDoQ+4EZVigbXW8M/MFKuz0xRXVxpt1FCIrqFDjfBP96FRjFi84Xl7Qt0Uc8/r9oTcdTosh+GfM5JMEU2kiCSTDeGuC3lrOZJMEo7HD9Tg++0gXZdSdp2s3xPYaAzKuSxAiwAb7WIcbhFkI0woHGkXbcPw94UJhY2mgIe9pIX0e1K0b4edmF75CeC38aZX/oGU8jeFEJ8DkFL+rhDiV4BfBmw8Qf+HUsrv+8f+v/BCNzbwJvBLUsrO8QSfQOgDbodaucTa7AwrM9dZnZlmzTcArVPdWjFiMV+c68Lt5dEWsW6Ieodxh4A7x7EtFEXd8/DbXiOlZLJS40q5xo/3pG7rHHcs9HebQOgDdhKzWmFtZhqzWvXGIZIpwvEEqnYgbwwPuEcwXZdXsiW+tprna6s5rlZMkprC+x98CO021su575ZACAhoRQ9HGDh6fK8vIyCAFdPmm2t5/nolx7fXChQdF0MRfDAd5z8f6eMj3cnbEvmbEQj9HSKlxJKSiuNScSVlx8WSkqNRAzXo5gcE3NdIKTlXqvK1Fc9rfz1fRgL9usbf6Mvw0Z4kz2XixNTdHcs50EIvpaTmSiquS9lxqbiuJ8i+KFccl3J9n5+3tmvUObJ57MZzuW7HG/qGjBB/e7CLvz3QxWhkd54DGRAQsP+oOi4vZYteSGYlx2zNAuCRRITfGB/gx3uSnI5HUO6iI3ighP4jr14ga9ttwnyrIxACiKgKEUUhogoiikLU306HVIbUkF+nEPXzRlu/bEnJnyxl+a1ri/zWtUU+lEnwmaFuPtaTxLjPB50CAg4iCzWLr/ux9u+uFam4LlFV4UOZBL8xnuSF7iT9xt4thHighP5U3FsoqFWIo2qrGCtEFNEu0i11UUXBUO7stuk6Pz/YzUzV5Avza/yH+VU++941ukIqn+7v4ueHujgZ2/5KeAEBAfsLV0reKVT42mqOr63meafg3UtxKOz15H+8O8mz6TjhHV7R9XYJZt3cBRwp+e5agT+aX+MvV3JYUnImGeUzQ938TG+amHZ/z7W+G7hS4kp2ZaAr4P6g5Dh8b63IX6/m+PpqniXTRgHOpGJ8tDvJR7qTnIzt3RTcYHrlPmLFtPniwhp/OL/KpXKNmKrwN/syfGaoi8cS9+adkfuJquNypVLjUrnKpVKNy+Uql8pVJss1aq6kK6TRp2v06yF6DS/v0zX69BB9eoh+wyvHVSX4X+wg9UkLpisx/bzmj29FVYWE5vWo99tnPl01+dqK57V/P1uk5kqSmsKHu5J8tDvJh7uSdOv7IzASCP0+RErJa/kyfzi3yleWslRcl1OxMJ8Z7ObnBjJ0hfbHl2e/krVsLpXrgl7lUtkT9amK2VhlVwAjYZ1j0TBHYwZxVWHZtFkyLRZrXr5s2pgdfgMRRfEMgtE0BBuNQ78eolvX9vXsKillYzZY2XEo+5MJyo7ria4rqclm2RNh1xdiT5xrbmt9y3GuxJLNtuaG89SPs/xJETdDARKaSkJTSKgqSU0lrnrbreWEpvr1Xjnut60fdyfhV0dK3siX+Wtf3M+XvEX+JiIGH+nxxP3pVJzQPuwZBkK/zynYDl9ZyvKHc6u8WSijC8EnelP8ncFuPpiJ39XR+f2ElJLZmuV55SVf1P3yitVcpsBQBEciBsdiYY5GDY5HwxyLhTkcMYjeJEYqpSRrOyyaFss1m0XTYsn085pXXvL35ezNC8gpQE+jR+AbBCNEr95iEPztrabQOVI2xLfSIsSN1NjndNjXqV1z+3YmJLSiCdAVBUMIQopAVwSGUNAVb7tebiThjXPpiiAkBIbSUi8Eur9t+PWaEJQdl4LjUrAdCrZD3nEo2m572fHqqtswGCEhGqLvGQLFNwR+UpW2clJTKTku31jN8821PGuWgybg6VScH+/xQjIT0Z1/UMhOEwj9PcT7xQp/NL/KFxfWydoOo2Gdzwx28bcGuxg07uxRY/sV03W5WjF9Qa82PPXL5Rplp7mAVUpTORb1BP1Y1Bf1WJiRsH5XvOqK47JsNsV/0bR9Y+CXTYulms2yZXWcchtXFfr0EKqgTYy34+22ogkv3BFVVC+vJ0Vp24502BdTVX8WmSDcEGGlIbyeCCu+KIt952SYrkvBdin6wp/3y3nboeC4FO1OZYeCbzjqZauD7nWFVH6sK8lHe5I8n0mQusd61YHQ34NUHZe/WMnxR/OrfG+9iAL8WHeSvzPYxUe6U/uy63gzCrazKXZ+qVTjWrXWJozDRqgRbjkW9UT9WMygJ6TtuxhuJ1wpWbVslk2bxVqrYfDKrpS+8KqbhLiTYG/cpwdTdO+I+v01DQPgOAjgwXhkT8NwlcoMxdIFenteuK3jA6G/x7leqfEf5tf4wvwaC6ZFr67xtwa6+MxgN0ei++NmLCkledthrmYx76e5mslCzWKqanKpVGPBtBrtNQGHI55HXvfOj8XCHI0YwSykgPsG2y6xtPwXzM9/mWz2FTQtwY8890MU5dZ774HQHxBsV/LNtTx/NL/K11bzOBKeScX4O0Pd/GRv+qbx6NtFSsma5TBfMzcJebNstYVZwBsM7dU1hg2dozEvdl4X9LGwcU/2SgIC7hQpXdazrzA//yWWl/8KxykTiYwxOPhzDA78TcLhods6byD0B5DFmsX/vrDGH82vNla9+9n+Lv7OYBcPJaLbPo8jJSum7Qv4BiGvekK+YFqb4siqgAE9xKARYtDQ/dxLQ0aIwbBOv64FYYaAAJ9y+TrzC19mYeH/oFqdRVXj9Pf/JIMDP0sq9cQdhyUDob+LOE4NxykRCt2dh1xLKflBtsQfza/y1eUsVVfycDzCZ4a6+aneNBXXbQh3m5BXPY980bSwN3wFdCHahHvQ0BkKtwq5Tu8+n1YYELAfsO0CS0t/wdz8l8jlXgMEXV3PMTjws/T2/jiqunOzeQKhv0WklLhuFcvKYtk5LGsd28o1tm0r21KXxfZzy8rhut6t0OHwIXp6PkxP94dJp59BVXc/lp6zbL7sT9M8W+z8eLuIojAcDnneeNgT7YYX7ot6d0i9JwY9AwL2I1I6rK+/zPz8l1ha/itct0o0eoTBgZ9jYOBnCIcHd+V171uhl1LiOCUsK4dtbxDnjWJt5bDsbKOt65pbnlcInVAoTSiUIqSl0UIpb1vzcqHoZLOvsrb2Iq5bRVEidHV9kJ6eH6On+3kMo/+O39vNeKdQ5jtrBbpCWltIJakFIh4QsBuUy1eZn/8S8wv/B7XaApqWoL//pxgc+DmSyUd2/Xd3Xwi9lJJ33v0clrWGZTU9bSk7P/8TQFEinliH0mhau1hrbWVP0OttFSWyrX+a41RZz77Mysq3WF35JtXaHACJxIP0dP8YPT0/RiJxGiGCOHZAwL2IZeVZWvoz5ue/RC7/JqDQ3f0jDA78HD09H7krPfk694XQA7z+xs8jhNYQbK3F6w6FUmgtYq1pqR2Nj90MKSWl0kVWVr7Fyuo3yeXeBFx0vYfu7ufp6f4xuro+iKbF79o1BQQE3DpSOqytvcj8/JdZXvlrXNckFjvG4MDPMjDwM3elx96J+0bo7yVMc43Vte+yuvItVte+g20XECJEJv00PT0fprv7w0SjY3t9mfcsUkpMc4VKdYpqZYZKZQpX2oSNQcLhQQw/17TEXl/qXUVKF9NcxTSXEUJFVWNoWgxVjd3W3O37iVLpMvPzX2Zh4T9SMxfRtDQD/T/F4ODPkkg8tOch0UDo9zmua5HLvcHK6jdZWfk25fJlAKLRCX9A98dIpR5HUfbuwQX7EdsuUa16Il7x82plhkp1mkplGtetbjhCAdrn+mtaoiH6YWMIIzzoG4Mhf/8AirI/bkq7GbZdpFZbbE9m+7ZpLm8ZzhRC90U/6hkANYaqxVvKfq7Gm2UthqbGm8f47T3DcW8tIdAJy8qyuPhV5he+TD7/NkKodHd9iMHBn6On58M7+t1wXZd8Pk86nb6t4wOhv8col6+zuvotVla+xXr2FaS00LQEXV0/Sk/Pj9Hd9aPoetdeX+au47o2tdqCJ+B1Qa9MN0Tdstba2qtqnEhkhEj4EJHIKOHIISLhEa8cHkYIDdNcolqdo1qbp+bn1eo8teo81dr8pnMChELdhMNDLT2BobaegWH0IcTu3c3ruhamudIi2AvUzCUvry1Sqy1Rqy3iOMVNx6pqHMMYIGz0oxt9GMYAhtGPofcicXHsErZTxLFLOE4J2ym1l50Stl3CcYqN8kZjuRWKYvhGwjMMrb2HVkPSzONoWqKxzzMacTQtfld7G65rs7b2PeYXvszy8teR0iQeO8Hg4M/RP/AzGHrPjr2WlJKFhQXeffddzp49C8Cv/dqvodzG/SeB0N/D2HaRtfWXvAHd1W9hmiuAQir1aGNANxY7vufdxttBSoltZz3xrkxRqcy0hFqmqdbm2rxPIVTCxjCRyIgv4qNEIr6ohw/tyL0LjlOlVptvMQYt5doC1eocjlNqO0YIFUPv93oDvhHY2DPodG3e+8+3CPZWXvgKbFiDUggNQ+/DMPrRjX4Mo98X83p5AF3vQ9Nid/R5bKQ+9dgzCMUWQ9BuKDwjUWwpl7D99o5Tbtu/8b11wuttxH2jEW/0HNr2NXoaCa9+kxHxjtlq8kOxeIH5+S+xsPgnmOYyoVAX/f0/xdDgzxGPP7Cjv7HV1VXOnj3Lu+++y8rKCoqicPToUU6fPs2DDz6IehsPC79joRdCfBz4PKACvy+l/Jcb6p8HvgJc9Xd9WUr5z/26NPD7wGm8/+h/JqX8wY1eLxD6zkjpUiicbQzoFgqeBxA2huj2p25mMs/e1UHmra7TdU1ct4brmth2ruGJV+uiXvXEfKMXGgp1EYmM+l75SEPEI5FRDGNgz8MBnjgXNhiDjT2DBaRsn56rKOFGD0BK2xf3pQ7hJQiFMr7X3fTAdaOPcN0bN/oJhboOxGwtKV1P+H1D4PUuWvJGudRW5xkNv943IJ0+y040exlN8besLMXi+wih0d39PEODP0d39/M72pMoFAoNcZ+b82bgjY2N8dBDD/HAAw8QjW7/jvZO3JHQC69PehH4KDADvAr8vJTy/ZY2zwP/SEr5yQ7H/6/A96SUvy+E0IGolDJ7o9cMhH571GqLrKx+m5WVb7K29hKuW0FRwt6c/e4Pk8k8ixCaL7i1FvHdzrbZtu34uXTNRnmrYzaKXCuKEvbDK75X3hB1T9B32gPdC6R0Ma01zwBU56nW5hqhoWp1HkUJNbxxT8hbBF3vu6tT8g4Srmu19Rxsu9BiHEobDEd7LlDp7fsYA/0/ja5379g1VSoVzp07x7vvvsvVq54fPDg4yEMPPcSDDz5IKpXasde6kdBvxz16Crgspbzin+wLwM8A79/wKK9tEvhR4P8MID0F2FoFAm4Jw+hneOhvMTz0t3CcGtnsK/6A7rdYWfnGbZ9XCA1FMfyk+6l12/DjpsaGuo3bfls11gi16HrPPRlmuhWEUDD0Hgy9h2Ty4b2+nPsGRQmhKGlCofSeXodpmly8eJF3332XS5cu4bouXV1dfOhDH+Khhx6ip2fnYvzbZTtCPwxMt2zPAE93aPesEOJtYA7Pu38POAIsA/+LEOIR4HXgH0gpSxsPFkJ8FvgswOjo6C29iZ3Gu6PWwXEcbNtupNbtG9W5rks8HiedTpPJZIjFYrc1uHIrqKpBd/eP0t39oxw/9l9TKl0in38LhOINivniKxS9Ud5KqHdzYDEg4CDiOA6Tk5OcPXuW8+fPY5omiUSCp59+mtOnTzM0NLSnDs52hL7T1W2M97wBjEkpi0KITwD/ETjmn/9x4O9LKV8RQnwe+C+B/+emE0r5e8DvgRe62fY7aOHFF1/EsqybCvN2BHwnUVWVdDrdljKZTKMci8V29EsghCAeP048fnzHzhkQENCO67pMT0/z7rvv8t5771GpVAiHw5w+fZqHHnqIsbGxXXfwtst2hH4GGGnZPoTntTeQUuZbyn8uhPjXQoge/9gZKeUrfvUX8YR+V/jOd76DZVmoqoqmaY18Y1nTNHRd37Juq+O2205RFAqFAtlslmw2y/r6eqM8Pz9PuVxuu25N0zaJf2uKRqMHPtwREHAvsHE6ZD6fR9M0Tp48yUMPPcTExASatv/uH9jOFb0KHBNCHAZmgb8NfKa1gRBiAFiUUkohxFN4d6as+tvTQogTUsoLwAtsI7Z/u/zjf/yPUVV1X1jRcDhMb29vx7pardYQ/o1penqaarV99oCu6x0NQN0whMPhwBAEBOwinaZDTkxM8JGPfIQTJ05gGPt7AP2mQi+ltIUQvwL8Fd70yj+QUr4nhPicX/+7wKeAXxZC2EAF+NuyOZ3n7wN/6M+4uQL84i68DwBCoXvjzlHDMOjv76e/v/OaGNVqdZMBqPcKrl+/Tq1W23S+TgYglUoRDocJh8Poun5bc3MDAu5X6tMhz549y+zsLOBNh3zmmWc4deoUsdi9M0MsuGHqHqRSqXQMC9W3LcvqeFwoFCIcDmMYRiNvLXfat7F+L7ultm1jmiaWZd0w36pOVVWSyWQjpVIpksnkXRksD7g36DQdcmBggIceeojTp0/v6HTIneZOp1cG7DMikQiRSITBwc0PMJBSUi6XyWaz5PN5qtUq1WqVWq1GrVbbVM7lco19WxmIVlRVvalx2LhP07Q7Euh67rrbu/W+TigUIhQKoes6oVAIx3E4d+4cjuO0tVMUpaMBaC1Ho9F70hg4jkOlUqFSqVAulxupUvEeTKOqaiPcWS9vtW+7bfbic3Jd179r192UttrfWpfNZjl79iyXL1/GcZzGdMjTp09vGYLdSayaw5U3l8gtV3jqp47s+PkDoT9gCCGIxWLEYjEGB4dQbuEB3I7jbGkQNpZb962trbXtuxVaxbguyLquE4lE2rY7tdkqrw+0dxKcuiHM5/Pkcjny+XxbeXZ2tqMx2Ko3sNEY7OZYiW3bDZFuFeyNAt5a3jjeczcQQmzbYCiK0ibE2xHlTmkniMfjPPnkkzz00EN3ZTqkdCVzl7Ocf3mBydeXsGoO6f4oT/zEOKq2s8YyEPoDQilXY3mqwMp0gaXrBZanC5SyJiMnMxw908+Rx3oxIjf+d6uqSjQavaNbsV3XxbKsNoNg23ZHQd5KjHeTdkPY+ZFu3rMDSpuMQL08PT3Ne++9t0lgNE3b0hjUtyMR76E1pmneVKQ3lk1z63sN68ax/v/r6upq295Yrl+H4zi4rtu4b6TT9u222c5xruuiKErju1BPQoi27e3U3c4x9bpwOMzw8PBd+T7mlsucf3mBCy8vUFitEgqrHH2ij5PPDjA4kUbcgnO2XYIY/T2GlJJS1mR5Ks/yVKGRSjlfBASk+6L0jiaIJnSuvrNMfqWKognGHuzm2JP9jD/cQ0gPBmbvBNd1b2gM8vk8hUKhozEAbnivhmEYDVHuJNKd6vbjlL6AJmbF5vIbS5z/wTzzl3Mg4NCJDCefHeTIY7078nsMVq+8R5FSUlirtgn68lSBSsGLpQsB6YEYfaMJev3UMxJHD2tt51i6VuDSq4tcen2Rcs5EM1QOP9zD8Sf7GXmga8e7iQEerutSLBY3GQAhxJaCHYlEgtlRBwTXlcycX+P8Dxa4+tYytuWS7o9y8tkBjj81QKJrZxcfDIT+HkBKSX6lwtL19vBLreR5fkIRdA3G6B1L0DuSoG8sQfdwnJCxfVFwXcn8pSwXX1tk8o0laiUbI6ox8Vgvx57sZ+h45pZi+gEBAZtZXyhx/gcLXHhlgVK2hhHVOHamnxPPDtA/nty12P99I/T/4Z+/glAE4VjIT5qXx71tY8N+I6qhqHffm5WuJLtUbvfUp4uYFU/UFVXQPRxveOm9Iwm6D8XQQjvn6Tm2y/S5NS69tsjVt1awag7RpM7RJ/o49mQ//Yd37wsZEHDQqJYsLr26yPmXF1i6lkcogtEHuzj5zCDjD3fv6G93K+6L6ZVSSvrGk1SLFtWixepskVrZolqyke7WxsyIap4BiGodDEKIcFwjHG03FnpY3bYIuo7L+mK7qK9MF7Fq3qwOVVPoPhTn2JP9jRBM11Bs18MpqqYw/lAP4w/1YJkO199d5dJri7z3vTne+dYMie4wx870c+zJfrqHd3YtnvsFx3IprFcprFUprDaTUAXxtEEsbRDPeCmWNgjHQsHnfA/hOC7T761x/gfzXH13BdeWdA/H+OCnjnLsyX5iqf1zt+yB8ug7IV2JWbWplmyqJctLRcszAkWrsb9Wryt5++redScURWDUewt1wxBvGouQoZFdKLE87Ym6bXkDcpqu0HOoGU/vHU2QGYyi7kGvYivMis2Vt5e59Ooi0+fWka4kMxDl2JP9HDvTT7r/zh6OcJCwLYfiWo38aqUh4vnVKsU1Ly/lam3L/wkBsYyBdCTlvMnGn56qKcQyRpsRaBiDdJhY2iCa0oPw2h6zMlPg/A8WuPjDBSoFi0gixLEn+zn5zCA9I/E9M9b3TehmJ3Ecl1rJbjEITSNQLzeMQ7FpLOqiHjJUekbi9I0m6R2N0zuaJD0Qvad+pJWiyeQbnujPXc6ChN7RhC/6fcQze/skq93GNh0Kvmg3PfKKt71WpZzb+BQpQbzLINEdJtEdIdEVJtkd9rfDxNNGI1ToOi7lvElxvUYpW2vm2RrF9Wqj7NobHiGoCGIp3TMAaYNYB2MQTxuoof3jPBwEynmTS68ucu4H86zOFFFUwfjDPZx8ZoDR09074qy5rmS1ZNKbuL2eQCD0dxHbdKhVbKIJfVfmw+4VxfUql19f4tKriyxdLwAweDTF8Sf7mXi8j0ji7j28eaewao7vhfse+VrTKy+sVhqzm+ooqiDeIt7J7jCJLl/Uuz2R3UlDLqWkWrQoZmuU1j3hL/mGoNVA1MOArUQSoRZjECae1omlww3jEM8YbbOzAjbjWC7X3l3h/MsLTJ1dxXUlfWMJTjwzyPEn+wnHb39tLdtxubJS4uxsjrOzec7O5Tg3lycR1vj+P3nhts4ZCH3AjpJdKnP5tUUuvrrE+nwJoQhGTmY49mQ/hx+9+Y1Zu4XrSsyKTa3s9cRqFRuzbFOr2NRKNuV8rU3Mq8V2IVc1xfO+u4yGeDe98si+DZuYFbvFGFQbBqBpGGqb3iuAFlKIJHQiiRCRpE4koRNNhPx9/v6ETjShE06E9lWIcbeoT0c+//I8l15bpFayiaV0jj89wMlnBukauvWFzGq2w6XFoifqcznem8tzbj5P1e/9h0MKpwaTnB5KcXo4yX9yZuS2wj/3jdB//uuX6E8anB5Ocaw/jqHdv/ORSzWbxXyVse4Y6i6Jk5SStbkSF19d5PJri+RXqqiawthp78assYe6b+lGECklVtXrETXEumxvLd5lP1W8dlZ1s2fbihpSWrzwulceaZQPWi+sFdtyKGVNStmqHx6qUcmbVAoWlYJJudAsu05nTTCiWsMARDcYg0hCJ5pslo2Idk99lsX1Ghd/uMD5H8yzvlBGDSkceaSHk88OcuhU17YNfMV0OLeQ570WT/3iYgHL/0wThsYDQ0lOD3ui/uBQiiM9MbQdMKL3hdCbtsuTv/l1chXPcwmpguP9iYaVfHA4xamBJJEDdkeolJKFfJVz83nen8tzbr7A+/N5rq2WkBISYY2nD3fxzJFunjnSzanB5K4Iv5SSxWt5Lr26yOXXlyjnTEKGyuFHehg+kcGx3IZw18W62hDxpoDf7Ouoh1X0qIYRDWFENG/WVMTb1uvlmNao0yPeNFojqhEytj9b6n5FSolZdXwj4Im/ZwSahsAzDF65WrI2P28Ob7wi3OgRdO4lRBI64biGdL3pvl6SuPWyI3EsF9dp1jm2i+vnjuP6bev7/bLj+sfJ9vP6+x2n5TX816uPrQ1OpDjxzABHz/TftGear1q8P5fn7KznpZ+dzTG5XKQ+yS8TDXF6OMWDvgadHkox2rV743T3hdCD13WfWis3ukde/CvHetkTf0XARG/c//A9q/rAUJJk+N5Yx960XS4vFT1Rn8838my52S0f6YrwwGCSU4NJBpJh3p7J8vKVNa6ueI/pTYY1njrczbMT3TxzpItTA8kd/+K5rmTuUpZLr/o3ZpWbM5g0XcGIaOh1oa6LcsSb5qq3ibcv4A3R1vZl6OR+xnVcqiW7pVdgUslvNgh1Q9FpPOFOUDSBqimoqoKqCRRN8bb9/YqqoIYEqqr4df7+ejvV2zZiGhOP9W05q2y1WPM0ZS7He76nfn21+aS4/qTB6aEUDw6nOO1ry2Dq7j4Q6L4R+k5IKZnPVf34mNelem8uz0K+uarfeHeUB+viP5Ti9HCKrtjeDi5myybvb/DSLy81u4CGpnBiINEQ9QeGkpwcSJDYwmjN5yq8cmWNl6+s8oMrq40vaSoS4qnDXTzre/wnBxI7KqaO7VJYrXpiHdGC2SB3gbJps5SvsZivsliosZSvslTwt/NVlvI1Voo1RrqiPDqS5rHRDI+OpDnSE9t1Q2qZTlvvoFqyvNUuNQU11BTeVlH2BFv49UpD3BVV7LiQSilZzNca8fSzs3nen8sxl2vqxUhXpKETDw554ZfbnSmzk9zXQr8Vy4Ua77V6/nM5ptcqjfqhVNi3zk3vvz9p7PgXy3Ul19fKLaEXz0ufb/li9SYMT8wHk5waTPDgUJLx7juL681lK7xydZUfTK7y8pU1ptY84U9HQ41Qz7MT3Rzv21nhD7h9KqbDUqHKYr7WzOvi7Qv5Ur5Gobb5HhBDU+hPhulPGvQlw3RFda6tlnhrKttonwxrPOIL/2MjaR4dSZPZY4dnN7Ecl6srJc4vFDg/n+e9uTzvzeVYKXrTZoWAIz0xL54+lOLB4SQPDqZIRfdnBCAQ+m2SK1tN8Z/zwj5XVkqNuHFPXG+Lt50eTnEoE9m2+JdNm/MLhTZRP79QoGz6d8kqgoneWIuoe+lueAuz2QovT67y8pVVXr662jB6mWiIpxuhnm6O9cUD4d9hqpbDcsPjrjWEeylfZbFF0PPVzQKuawr9SYP+RJi+pEFfIkx/MkxfwmgT9mRY6/g9dV3JlZUib0xleWs6y5tTWS4s5Btx5vHuaMPjf3QkzanBJPo9tgie60pm1itcWCxwcbHA+YUCFxcKXFkpNnrImiI41p/g9FCy4didGkwSM+6dKaiB0N8BpZrNufl8M/Qzl+fSYgHb/yUkw1pT/P2Bl8M9MZYLtYZ3/v58nnNzea6uNo1GwtA4NeQJel3Uj/XHCd+FNTG2w8x6mZfroZ7JVWaznvB3xXSeOdIc3D3Wt3d3At4MKSXFmk22bLFWMimbDlJKHClxJbhS4rpe2XGl92ALCY70yo7boZ1f57oSR7JlO1fKRnJc/HN75WzFbBP2+gSCVnRV8YW7LthhelvEuy7mqcjOL5tQqtm8O5vzhX+dN6eyLBW8B8romsLpoWRD/B8bTTOc3r6zs5tIKVku1LiwWODCgifqXl6kYjXHBg5lIpzoT3B8IOHl/Qkm+mL3/Cy9+0bov/T6DMOZCA8OJbeMVe8EVcvh4mKhMX3qvdkc5xYKmLa/1IEiGoYA2gdI6/mt9AT2A9NrZc/b98W/LvzdMd0X/S6enehmond3hN91JbmKxXrZ9FKppVy2yPr71sqmV/b3WVtMFbwbKMLrpQkhvLIQKEKQjIToa/HCWz3w+v50dP+se1Mf56oL/1vTWd6ZyVHzv+89cYPHRtMN4X/4UJr4LnvCuYrVIuTNfL1lYkJPXOd4f4ITdUEfSHCsL76r2rCX3BdCb9ou3/2n3yWP5BoOpViI8GCMvsMpThzu2nXxtxxvRsx7c3kuLRUYSkU4NZjk5GDinpnVs12k9LrCP7iyysuT3uBufUyhJ67z9JHuxuDuRO/mBdEsx2W9bDY87bowt5azZdPf9gQ9W7G2nHqpKYJ0VKcrFiId1clEQ2SiOpmYV05HdbqiOlFD9cRW8QRXEaAI4YuxJ8qK2H6dd55mnWgR84Me3rIclwsLBc/jn87y1lSWK/7MLkXA8f5EQ/gfHclwtC9+W9N6q5bD5aUiFxYKbZ566xhW3NA43h9vE/Tj/Ql64ns/QHo3uS+EXjqS+S+cozxXRMuaaC2e3Dou13FZCyvI7jDJ4QSHjnZx8nj3gbXudxMpJdNrFX5wZYWXr6zxg8nVxqymnrjBiYE4hartCXbJ6jhYWCccUjyRjupkYk2Rbgh2TCddF3K/TdzoHH8OuLtkyyZvTTdj/W9NZxthqbih8fChVNssn9axJ9txubZa4sJC0Rf0PBcXi1xfLTXGC3RV4WifJ+iepx7nxECSobs8jXE3ka687RvN7guhb0W6EidXw14qsz5dYO16Dmu5TCxvEWl5slsJyYIGlWSIUF+MnrEUh092k+iP3VN39e03pJRcX62Hela5ulomHQltEOmQ73F7+7r88n4Zowi4c6SUXFstN8I9b05lOTefb4Q1D2UinBxIMJutMrlUxHS8H6ciYLw71gy7+MI+3h3dkTtI9wNSStyChTVfxJwvYfkJKRn4jY5afVPuWOiFEB8HPg+owO9LKf/lhvrnga8AV/1dX5ZS/vOWehV4DZiVUn7yZq+3W4OxUkrcosXy9Syzl9cpzBRR1qqkKw5dsinsJpJsRMXJGMSH4gwcyRAfjqN1RxD32IyDgID9RNVyONsY6M1ycbHAsD84Whf0o337Z1LCTiAdF2upgjVfbAi6NV/CLTXHE9SUQWgwRmgoRvKjY3d/rRtfpC8CHwVmgFeBn5dSvt/S5nngH20l4kKIfwicAZJ7KfQ3YnG5yOS5FZau5jAXSxh5k0FbMERT2F2gEtNQeyOkR5NEBuOE+qJovRGUA7a0QkBAwK3jlKxNgm4tlaEeStYEof4YoYGYJ+yDMfTBGMoOzM2/0ydMPQVcllJe8U/2BeBngPdveFTzxQ8BPwn8JvAPt3XFe0B/b5z+3jj8aHPfUr7K2evrTF1aIzdTwF2p0FuyGC/V0K7lqdK0uk7cG/w1+mNofRHUaAihqwhDRegqiq40yiKkbP8JVa7Ecl0sR2LZbqNsOy6W42LaEtvtXA6pgqcOdwXjEAGbkI4Lys7fWXq/IF2JvdLupZvzJdx88xkFSiJEaDBO/HgG3Rd1rSeKUO/+Z74doR8Gplu2Z4CnO7R7VgjxNjCH592/5+//beAfA4kbvYgQ4rPAZwFGR0e3cVm7T18yzI89NAgPDTb2LeWrvDub4z9OZVm4lqU8XyJTdRgvWoxdqjJ+aR2dG/8jXaAmvFQVkipQQVJGUgHKUlLCpeRKSv6+CtJv0yy37q8CnVYR0VWFHz3ew8dPD/LRU/23dVefdCVICf58cH+iuL8frywlONKbGeO3l65XFqpAiYdQoqFg7GOXkY7ELZo4+XqqbSq7BRO3bCNCCmpSR0kaqCkdNWmgJvVmOaWjJnTEAYmL3y5u1caaKzXj6QslrIUy+NNLUQShvgjhiXTDSw8NxlDj++eu4u0Ifadf5sZ4zxvAmJSyKIT4BPAfgWNCiE8CS1LK1/3wzpZIKX8P+D3wQjfbuK49oS8Z5oVkmBdO9Tf2LearvDuT4+3ZHH80k2V5vkjYgbgiiAlBTChEFUFUCqJCEAHC0ssNCWEJMRe6JOiuJORAyJWEpES5hU/CVQUypEBIAV3FViBfsShcKOOeu8w7XCYe0ojrKtGQiiJbhLtFmBvC7eIJ/E79NxRQYjpqPISS8ERETYRQ4i1lf78IVppsQ0qJW7Ybgu12FHITt2hu/n8JvM81qaN1R1APp1DjIdyq0zjenCrg5FaaIYbWY2Mh1JRvBJJ1I1A3DjpqyjgQ/y/pSpy1qj842vTUnWyt0UaJaoQGY8SfGWyKel9034/dbUfoZ4CRlu1DeF57AyllvqX850KIfy2E6AE+CPy0L/5hICmE+PdSyv/0zi99/9CfDNP/QJiPPNB/88a3iLRdpOngmg6y5iBNF7fmIE0vNco1B9d0W8oO0nJJdUVAEeSrNvOFKpP5KvmSiYukO24w3BXlUFeUSNhfP1yAP1Hc++Eq+Pv9fYpfL4S3X6GlXD8Ov63XTjoubsnCKZi4BS93iib2YgmnaG0WFwBNoPoGQNnKIMS9srjHB+7cmi/gOdP7jPK1RrlVzDt9TkpUQ00aKEnd8yKTLZ65X1bi2+tJNYxJrtZ83ZzpG5UaznoN83oet7x5eqzQlZYeQcvrp7xrU1MGalzfsbBFowdpuUjL8XMvuabj/24212217ZYsrIUy0l+OBAFaTwR9NEHo6cFmLD2p35MGbTuDsRreYOwLwCzeYOxnWkIzCCEGgEUppRRCPAV8Ec/Dly1tnucGA7at7KclEA4aUkrOzRf4i7Pz/Pm780wulxACnhjN8BMPDfLx0wMMpyN39XpkxfZErWB5YYd62TcIXm55sxQ6fF2FofoGIeQZgnhn4yAM1RMH1w9DtYag6mWnfV9bfqM6v77eK+rUxrVdCmULUbUxqg6yYOHkzaa4bHxPSf/a62KZaIpnvbwXnqS0nKZhqvcoci09i1wNp9DBMLX0LBqhobiOdDsIdqtY+87ORoG+rZ6mwBsjCymIkNooK2ENrT+KPhj3Yun90XtugsVOTK/8BF6sXQX+QEr5m0KIzwFIKX9XCPErwC8DNlAB/qGU8vsbzvE8gdDvOy4tFviLswv8+bvznF/wngX7yEianzg9wE+cHmCs+9YfnbZbSEc2egYNA9BiEFoNhbzJ06buJg7gILH9cgHJCi55VVALq7gxDTWpE+mKkOiJ0N0fZ6A3xkAqTOgejY9LV+KWrRsbg5yJ9Bdqa4pvuwBvua1vp62KElK8tpqC0FXYhaWN9wv33Q1TAbfHtZUSf3F2gb84O887MzkAHhhMeqL/0CBH++J7fIXbR1oOTsFqMQgm0vRnmqiiGVpSNoSdvEVpGmGpTXUtx+erNtezZa6ulbmyWmZytcTllRIzuYov7h5jPTEmeuMc7Ysz0esZzvlclflchbmsl8/nqhQ2rE4pBPTGDQbTEYZSYQZTEYbSYQZayn2J8K49KvJuIG33roqv60pqtkvNdrzc8sqWIzFCCoamYGhqo6yr258ht9cEQh9wy0yvlfmr9xb4i7MLvH59HYBjfXF+4qFBfuL0ACcHEvfMD+BOkFIyl6tyeanYSJPLRSaXiqyWmlPpDE1hojfORF+co76oH+2LM94T3faqiIWqxUKuylyuyny20sjrRmE+V20saV1HVQT9Cc8YDKbCDKUjDCTDDKU9YzCYDtMTM/Z87R0pJbYrsR2J6U8Nth2J5XhCW7XcjgLs7XOpWS1l2/Hrb9zebGlfbRH0W8UTfwUjpDbLLcbA0NRN9eFGuVO7lvKG4yK6ymDq9kKngdAH3BELuaov+vP88OoarvTWKa+L/kPDqXte9C3H5fpquSHkraLeKq6pSMgT8RYxP9oXZzgd2XUxlVKSr9jM5SqbegPzfnkuV22solpHVxX6U4bXC0iFGUh5RkFVRENwzRbh9ZJXtl3vnozWcv1eDcuu3+PhdhRwa8P5dorbEd7W+jYR9oVXU0SLIblFw2JtbZBu9X33xHVe+6cfva3P5b4R+t/5nd8BIBwOYxjGLSdd11GUezMmerdYKdb46/cW+Yuz83x/chXHlQynI354Z4DHRjJ77j3eiLJpM7lU4vJyoUXMS1xbKbUtLT2YCvuhlnhb3hPf37MupJSslUzmc1XmshUW8tWmQchWmctVWMxXtxQgVRFoikBXFTRVEFIVP3llTVXQVYHWsq9e79W1lzVFENL8Nn5ZUwS6pqApXtsbCXY41L7vXgqlgPecA3PLnsdmAyKE4KcfGbqt17pvhP4rX/kKlUqFWq22KVnW5oc7dELXdQzDuG1jUU/3g8FYL5l87dwif3l2gRcvrWA6Lv1Jg48/6MX0nxzvuuP4sZSep1iuOZQth4ppUzYdSjWHiuWVy6ZDuWb79U5zn9+24pcX87XGOvrgidpYd7Qh4nUvfaIvvuvrqe8lritZLZlIJCFFaYhvSFXu6Xj/vYLjOOTzebLZ7KYE8Iu/+Iu3dd77RuhvhOM4mKbZ0QjcKFWr1bZt0zRv/mIChoeGmZiY4MiRIxw6dAhNO7jCAZCvWnzz3BJ/cXaeb19Ypma79MR1PvrAAM8c6cJyZFOkzaZg14W5ZNqNcsVyKNX8bcvBcbf/HRUCoiGViK4RM1QiIZWorhLVNbrjelvIZaw7ds89Fm+nsCwLTQuWd94NHMehUCh0FPL19XXy+TwbdTeZTJJOp0l3pfnZv/Gzt/W6gdDvIK7rNgxGtVplJj/DpfVLTOYnuVa8xlRlikVzESQoroImNTSpEQvFSEQSZOIZkpEkES3STKFI+/ZNkqHu/EPKd5JSzebbF5b587PzfOv80qYBRKAhwBG9KcRRvb5PI3aDumhIJWo090dCKjHDKxvavdW13y1c16VYLLK+vs7a2hrr6+uNtLa2RrlcJhqNMjQ0xPDwMMPDwwwNDRGP3zszq/YK13W39Miz2Sz5fB7XbR8nSSQSnpCn02QymUY5nU5T1ap8d+67fGPqG6xX1/niT3/xtq7rThc1CwDyZp5L65eaKXuJy+uXKViFRpvB2CDH+o7x46kfRwhBsVpkObvMan6VbClLtpJlJbuCVCXo4CgOpjSpObUbvPJmBIKIFiGshRviH9Wim4xHykjx8fGP80D3Azv9cdyQmKHxkw8P8pMPD1K1HK6vlomEmsIdCan7Oo5/r2BZFtlstqOQZ7NZbLs5XVMIQSqVIpPJcPLkSVKpFOvr68zNzTE5OdnwMFOpVEP0h4eHGRwcJBwO79Vb3BNc193SI89ms+RyuS2FfGRkpE3E0+k0qVRqU49+Kj/FN6a+wTfOf4O3l98GYDw5zkfGPoLlWoSUnV2IMPDoN2A5FldyV7iUbRf1hdJCo00ilOBY5piX0sc43nWco+mjJPQbrtvG2toak5OTTE5OcvXqVWo1T+AHhwYZOTzC0NgQmb4MpjSp2JVGqtpVKnaFsl1u299atylZFdZr61iuxYPdD/Kp45/iE4c/QTQU3dXPL2DnkFJSLpc7Cvn6+jqFQqGtva7rZDKZRurq6mqUO4lNnVqtxvz8PHNzc8zOzjI7O9uIFwP09PS0ef0DAwP3dChSSkmlUtnU27mRkMfj8U0CXvfOk8kkodCNhVlKycX1i564T32Di+sXATjVdYqPjH2EF0Zf4EjqyB31RoPQTQeklMyX5rm0fomL6xcbgn4tdw1bep6QpmgcSR1pCPqxzDGOZ47TH+2/4/CA4zgNb2pycpKZmRmklOi6zvj4OBMTE0xMTNDd3X3br5U38/zZlT/jjy/+MZfWLxHVovzkkZ/k08c/zanuU3d0/QE7g+M4jdhtpzDLxjGhRCLRUcgzmQyx2Obn894upVKpIfz1vFTynwmrKAwMDLSFfXp6evbVBIT6gOdWRrLuZNWJxWKbQiqtHvnNhLwTrnR5Z/mdhrhPF6YRCB7re4yPjH2EHxv9MYbjwzv1lgOhz9VyDSGve+mXs5cpWsVGm6HYUJuXfixzjPHU+I53obaiUqlw7dq1hvCvr3s3KaVSqYboHz58mGj01j1yKSVvL7/NH1/8Y/7q2l9Rc2qc7j7Np45/ip84/BOBl38LuK6Lbdsdk+M4W9a1pkKh0BCeXC7XNjCnqmpHEa+n2xGcnUBKSS6X2yT+dUOk6zqDg4NtYZ90Or2r4yW1Wm2TcayXN3rliqJs2dvJZDLo+s4sKWy5Fq8tvMY3pr7BN6e+yXJlGU3ReHrwaT4y+hGeH3menkjPjrzWRu4boTcdk6u5q56H7ov6xfWLLJWXGm0SesILt2SONzz0o+mjxPX9NQi1VZhnaGioIfy3M5snV8vx1Stf5YsXv8jl7GVioRifPPJJPnX8U5zsOrkbb6UN13W9VRJd966l+uttV4hvJNobu/S3QzQa3SQ09e14PL6vPOMb4bouq6urbSGfhYUFHMcbfL/Twd76gPJWXnm5XG5rH4lEthTzZDK5a59r1a7y/bnv842pb/Dt6W+TN/NEtAjPDT/HC6Mv8KOHfvSmYd2d4L4Qesu1ePbfP0NNeh5GSAk1wy4tXvpOhF3uNrsR5unk5T/U8xCfPv5pPjb+sU1evuM4jZlGlUqFarXaSDfbrtVqDdHdaxRFQVVVNE3bdtrJ9qqq3jNCfjvYts3S0lKb17+8vLzlYG9fXx+lUqmjkN9oQLmTVx6J3L1VVwtmge/OeDNlXpx9kYpdIaEn+PDIh3lh9AWeHXqWiHb3rgfuE6GXjsN//9knSK+ZHJE9HB56gPjJBzBOniR86hShQ4fuOYHfimq1ytWrV7cM8xw5coQjR460hXmklFiW1VGUV0urfG/1e7yYf5FFZxEDg5PyJCeqJ4iVY1Sr1ZveP6AoCuFwuJEikUijrOt6Q+C2m4QQt9R+u0lV762lZ3cL1zQRodBd+U3UajUWFhYaXv/c3FzjO7uR1gHljUKeTqf39P+3WlnlW9Pf4utTX+eV+VewXZueSA8vjL7AC6MvcGbgzF0L9XbivhH68iuvUD13nuq5c1TPn8O8chV8L1KJxzFOniB88hThUycxTp7EOHYMZYdic3vJ2toaV65cYXJykitXrjTCPL29vTiO0xD1m3nUIT1EIVHgUvQSk+okjnA4pB7iA7EP8HTqaZLR5CYRbxXzg2JIDwpSSgpmgeWp8yxcfoelqQssL15lNTtHtpajT0lycvhRTj/64xx67qOod3EOfX2wd3l5mXg83hD2aDS6r75Hc8U5vjH1Db5+/eu8ufQmEsmh+KHGTJmHex9GEfujh3ZfCD0AlXWIZBqbbrVK7dIlqufOUTt/3jMCFy4g67E9TcM4csQXfs8AhE+eRE2nd+aN7AGtYZ65uTl0Xd/S027dNgyjzVvK1XL86eSf8scX/5gruSskQglvxs6JT3M8c3wP3+H9TcWusF5dZ626xlp1jdXKKmvVNdar66yWl1lZm2W1uEjWzLOuVLC3eBZlWGpURTMski7CkWqC46mjPHDsAzx4+sMczhzZUw91r7iSvcLXp77O169/nXNr5wA4ljnGR0Y9cT+eOb6vjFGd+0PopYT/9jBoERh+HIaf8NLQYxBONpu5LtbUFNW68J8/R+3ceeyl5oCtNjhI+OTJhud/0EI/t4KUkjeW3uCPL/4xX7v2NUzX5JHeR/jU8U/xsfGP3fU45EHDcizWa75wV9ZYq/l5tZnWq+usVj1Br9iVjufRbUGq5JIsQaosSdZUusNddKeH6O0bp2/kBH1HTtPTNUxXuAtd1VmvrnNh6T3Ovvdtzk29zuXqNFPxKrbmfc9DrmBc6+fk4COcGHyIE10nOJE5QSac6XgN9yo1p8al9UsNz/1a/hoAD/c+3BD30eTo3l7kNrg/hN424bV/A7Ove2ntil8hoOe4L/y+Aeg/DVp7yMZeXaV6/nzT8+8Q+gmfPIlx6lTDCOhHjx6I0M92yVaz/Mnkn/DFS1/kau4qiVCCT058kk8f/zTHMsf2+vJ2HCkllmtRc2rUnBpVu4rpmFSdDbldbbRpJLu2eZ+/P2/mPW+8ukrBLHR8bU1odIW7yITTpN0IqYogkTWJLxaITq+QWCySKktSJehK9JOeOInh90iNEyfQR0cRtxHPriwtcP6lP+W997/DhZVzXE1WmeoVrCeaTk5fpJdjXcc5kTnB8YyXj6fG0ZT9dxNV2SqzWF70UmlD7pfXa954gSpUzgyc4SOjH+HDIx+mP7bzz4DuhFutYq+sYi8vIatVYs8+e1vnuT+EHvjlr/8yI4kRfmT4R3gydZTw4nsw+6Yv/q9BadlrqOow8HDT6x9+ArqOwIbZENsK/UxM+AbgpBf/P3lif4R+KlmorEF6DJSdHcCSUvL64uuel3/9a1iuxaO9jza8/LC2N7fMO65DzsyRrWXJVrNeXk/VLEWreHMxtmvU3GadvK0Hk3poioahGptS0kh6Im5k6Ip00R3upivcRdoxiM6uE7uyiHbxKub5i9QuXULWB8JDIYyjRwmfOOH1Nk+cxDhxHC2zOx62dF1q589T/N6LzL3yLc4tnmWqy+H6UIjpsQhTsQo2niOkKzoT6QlP+LuaBiAdTu/OtUlJ0SpuEu3F8iIL5YVGuZMhzRgZ+mP99Ef9FOvnUPwQHxj6wI5dr3QcnLU17JUVLy37+coyTtv2Cm7LHc5qdzfHX3rxtl7zvhD6ql3lN77zG/xw/odUnSqGanBm4Aw/MvwjPDf8HGOJUcjNND3+2Tdg7k2wvLv9CKdg6PH2sE9iYNPrbCv009+PEo16SymClwtaQj/C39da77dh474N56jX44Jj+qkGtpeEU/XKrg0CUFREJAmRFEQziGgGNAOhKiAUUBSE4uUoAiEUUNWWsl9fL9eP8Y/PqSZ/Hb7MV/XzzKg54tLgx52TfFKe5ojoBUVFiUYxjhxGP3p026JkuRa5Wo5sNct6bZ1cLdfMq+ttIl7fVzALWwpzSAmR0BPtoqttFuGwFkZXdC9XdcLqhrxl/43OY6gG6hYGVrou1sxMswd5/gK18+ex5uYabdRMpiHm4ZMnME6ewjhyGLFHN0wBOMUS5R++QvF736P04ktUZqeZ64aZUz3Mne7nWr/gCsusVFcbx/RF+xqiXzcAY8mxG3r/UkpytVxDwBdKCx098bLdPo9eIOiOdLcJeGs+EB2gN9p7246IlBK3UGgRbl+0Nwi3vbKCs7bWiAa0osTjaD09aD09qL09aD29jW2ttwett5fwqdu7a/2+EPo6NafG6wuv873Z7/Hi7IuNeNtIYoTnhp/jueHneHLgSS+27DqwfKFF/F+HxfdA+qstJofbhX/w0bZ4fyutoZ/axUu4pn+LtcQbP6h/zlICsnk3ZGt9Sxvp2mBWwK6AVfHKVgXsqpc7tidp9X+fUEAL+8kA1UAKFWolqBWQZtl/LZBq2BvL0CJIxQAlhJQuOC74NxjhukjXAbde9uq2KkspeX9U8PVHBa+cENia4MS05KNvuTxxSVILQSECpb4E1bF+SsNpyr0JiimdQlSQo+IJuy/erXctbySieQu2ZYxMW54Op0kbXsoYGVJhry5tpIlokbs+xuIUi5jXrmNev4Z5/TrW9euY165Tu3QJt94rVBT08XGvV3jSF/UTJ9H6evf9mJB5/TrFF1+k9L0XKf3wh15PNxTCfOo0889MMDMRZzKU5eL6Ra7krmC73uCvoRoN7388OU7BLGzyyjcu9KcIhd5Ib5snPhAbaBPy3kgvIfXWDKGUElkuY69ncVaWO3jgvqD727LDNGMRCm0WbV+41Ua5F627G2UX5/rfV0K/kenCNC/OvsiLsy+2e/v9ZxrCP5Yca/6orArMv9Mu/utX/bMJ6D3RHu/ve3BTvH9bWFXITUP2Oqxfh+yUV85OeakeZqqjGpAehcyYl6dHvbBM2t+O9TS9/07UijD3Bkz/EGZe9fLKmldnJL33MvIUHHoKDp2BSPqW3k7dOOC6rFbW+NMrf8IXJ/8PporTNzwuUpMkypAwFVJKzBPpRC9dmWG6+8fo7hsjHck0BDxtpPcsNNQJt1LBnJrCvOqJeWtyVlba2moDA+hjYxgTE814+tGju/rjv1u4pknljTcpvfQixRdfonbOm62idncT++AHMJ57lqWHhpl0F7m4fpELaxe4sH6BteoamqK1eeED0QH6on1tot4d6b7pGIBbq+FkszjZnJ9ncXL+dq6+3VqXw83mkJ0eSiQEalfXFsLd2/TAe3pQksl9YZTva6FvZStv/1D8EM8NP8ePHPqRprffSnnNC/XMvu6J5cxrUPZ/xKoBgx3i/Y7lC/kGAa+LenGh/TWUEKRHWgR8FDLjTVGP9W0aQ7gjpPQGrKdfaYr/0vsg/e5mzwlP+Ovi33P8ll9fSsmrC6/yzso7JPUkmXBTsFN6ikTRxr06RW1yEnNyktrlSWqTkzirza6/iEYxjhzBmJhAn5jAODqBMTHhzYK6SzfPuLUa1vS0J+DXrvleupfsxcW2tmpvD/rYmJ/G0cf9fHgQpTwLi2c9g5wagdQhiA/s7P91n2AvL1N86SVKL75E6aWXcPwbpMIPPEDsR36E+HMfJPLoo5RkjWgo2jYXXVqWJ8h1Uc7lcNY3i/RG4ZbV6pbXIwwDNZ1GTaW8vK2cQk1nGsKt9vSgdXUh7rEVOgOh34KZwkzT21/4IRW7gq7oPDnwZGdvv46Unohvivf73fFQzC+3fLZC9X7YDa+8xRtPj0JicO9/8LWC936mX4UZX/wr/h2M4RQMn/GF/0nP6w+nduUy7PV1T/gnr1CbvIzpG4BWURW6jt4wAEcwJo5iHJ3wZpvcRhxbWhbmzEwjxFK7dq0RarHm55thNbz4eUPMx7085Au7Go95n9nie7BwFhbf9fLl817YbSNKCJJDTeFvpJZtY3+tw3SrSNel+v45Si9+j+KLL1J5622wbZRYjMjjj4PjtAm3W9w6bIemtYt0J+Gu55lmWbkP1tS/Y6EXQnwc+DygAr8vpfyXG+qfB74C1GMcX5ZS/nMhxAjw74ABwAV+T0r5+Zu93l6sR19zary++HpD+K/mvLdyU2+/jmPDih/vXzjr3bjVGmpJDIG6sx6ClJK16lrbgFVrXjALnOw6yeN9j/N4/+OMJ8dvrYspJaxe9j3+H3oGYOl9PAMmoPckjDzpefwjT0H3sV01Vk6hgHnlSsPzrxsBa3a22UjT0MfHMI543r/XCziKPj6O0DSsubk2j9y85sfPZ2fBaT4JS0kk0MfHNwm6PjaGmvINnOt6Yb2Fdz1PfeGsl+dawlXRHhg47U3pHXgI+h/0jH5uxmvXls9Afq45RlQnnPaEP72FMYj3b2tmleM6VB3v+QU1p0bGyOzJyqVOsUj55Zc90X/zLUTY87a1dBqlo4CnG563soNLMR807kjohRAqcBH4KDADvAr8vJTy/ZY2zwP/SEr5yQ3HDgKDUso3hBAJ4HXgb7Qe24n98CjBmcIML82+xIuzL/LKwisNb//MQDO2f8vCeQvcTMTrg1aW2x5fbI13RkIR3lt5j2wtC3jTyh7re8xL/Y/xQNcDtzx4RTXvGbN6nH/mVah65yec9jz9Q095BmD4zJaD1zuJWy5Tu3q1LfxjXr6MOT3dnPmg+LOJWuKxIhrdIOTjjbKaybT/b2sFWHy/6aEvnvW267O2hAo9x3xBPw39D3l5vP/GYyctSCmpmkUquetU1q9RyU1Rzc9QKcxTKS1SLq9QraxSsatUFEFFKFSFoKKqVPQYFT1CRTOoqBoVRaEioCpdKq5Fxal2fJJZX6SPkeQIo4lRRpOjjCRGGEuOMZIYIRaK3fb/ZFdwXW9ygln2esxW2S+X7mBfGcySN3tNDXlTr1W9pWy0lFvqNb1D2w77NGMb5/X3a4aXMuO39fHcqdA/C/wzKeXH/O1/AiCl/BctbZ6ng9B3ONdXgH8lpfzajdrtB6FvxXRMXl9sxvbr3v5wfNjz9oc9b3+73tFOiHh/rDnrYCA2wEB0gP5YP13hrvZ4p5RczV/lzcU3eWPpDd5cepPpgudxGqrBQz0P8VjfYzze/ziP9D5y68upuq7n9c/8sCH87uIFVu0xZsyHmOVZKqKX8eEsRycqZPqjEO3yejwRP492QSi6bUHc9qXVap73PnmZ2uXLSNNseOmhsTG03g4zW6T0xlAaHrov7I0BebyQVV3I68LeexJCkQ2nkqxUVriev85UYYrr+etMF6bJ1rJUrErDu25Nt0pEaISFQkRCxHWJODZhu+aVXZeIlH5KELYeRKudgNJhnEI/brhEKT3DXHKSC9FzzKjt4w3daoTRUJKRUIJRLcFYSzmhhGib9iXrU7rqeiKb+zbVt+yTrjcBopP4btxXD43eCqGol/RoSznmlyONsqPqWE4Ny655uWNiOjUs18RyTGzXxHIsLNfCdC0sx8aSNpZbzx1M6XplJJYASwhsBJYAUwiseoJG2duP306QECr/5v9yQz94S+5U6D8FfFxK+Uv+9i8AT0spf6WlzfPAl/A8/jk80X9vw3nGge8Cp6WU+Q6v81ngswCjo6NPXL9+fZtvr8lbX58ingnTfzhJPLN7D9CeLc7y4syLm7z9J/qf4Lnh53hm6Bls194k4oulZnknRPx2WS4v8+bSm7y55In/hbULONJBIDieOd4Q/sf6HmMgtvlego1IKVmbLzF7YZ3ZC1lmL65RK3vhh3RknbBcZ6E6Dih0a1c5Fn6Ro+GXSGktwqIavvhnWgxBprNRaN0O3UHs1ap4oai6h75w1out13J+AwFdh1vCLn6eOtQwSlJKVqurTOWnmCpMMZWfagj7VH6qba63JjQOJQ7RFe4iEvKe8xtWw23P+d24HVEjm54FXG8T1sKdvw+uQ2F6moX3Z5m/kmNh1mVlPYyUCuDSFZqjT7tAwell0TqOLb3PMKEuEAtfwI1cYD1+hdnYMlMhjamQxtKGgcmM4zBi2YzaNqOWxahle8m2STXmj7fcB9Iob9yneP/DUKwpxnrME+EtRLljfYuQ21qYnLTI2hXW/Rvo1mvrjXsyNuW1LKV6z2wHUYVKSNG8JFpzlZBQCAmVEAohoaAjiGYhVTH4r/+LL9zW692p0H8a+NgGoX9KSvn3W9okAVdKWRRCfAL4vJTyWEt9HPgO8JtSyi/f7IJvx6N3LJff/43vYpvelyya1OkbT9Lvp77xBEZ05282qXv79dj+ldyVTW06TR+r5wOxnRNxx7ZZm51m8cplFq9eJr+8xPDJBzny2Bm6RzoMKvuUrTLvrLzT8PrfXn674V0OxYZ4rP8xHu/zhH8iPYFAkFuqMHtxnZkL68xeWKdS8AxXsifM8PEMwycyHDqRIZY2ACiu15h8dZZLry2yOOWdu6/P4ejhAkeH5kgoS97spsp6M5XXvCmgzg2WSA5FW4Q/fWPDYBZb4unvej2R+iyjUMyLn7fG0/seACOOlJJsLdvmmdcFfbow3TbnXxUqw/FhRpOjjCXHGE34eXKUwdjgriwT4Douq7Ml5iezLEzmmJ/MUVz3wjSartB/OMngRJqBIykGjiS930GtCK6F48DKTJn5qyXmrxaZv1KkUvTmu4djGoNHEgxOJOkaD1NJrTFbnWWqMMNUYZrp4gxThSkWSottN6ol9SSjiVFGkiONz2AkMcJocpSMkbklB8yVLgWz0LhJrpFvIdjr1XXy5iY/skFUizZnf4XTjfssknoSXdUJKaFGHlJDXt6a1NCWda3HaULb8oa5Oo5tM3PuLOe//32uvvEKpewqqq7zq//2j1FuY0bZroduOhxzDTgjpVwRQoSArwJ/JaX8H7ZzwbcbunnnnV/FMWOYhUEKC70sX0mzNq2Dfzdpuj/qi74n/j2H4qihnR08nCvO8fri60S16I6K+EYc22Z1ZsoTdV/YV65fw7b8B6+EI8QzXazPewOVyd4+Dj/2JEceP8PIgw8T0o0tz227NhfWLzSE/43FN6jlXIZzxxgrPsBI/gShihemiqZ0Dp30RH34eIZkz83nhOdXK1x+fYnLry2xPOXd/j1wJMnRJ/o5+kRfwzgAXvfeKrcI/7on/m3b653rXbvzBaRG28Mu/achc5icVeB6/npDwBuCXrjediu9IhSGYkONWHZdyMeSYwzFh3Z9xcdaxWbxSs7z1idzLF7NY9W8HlQsbTA4kWJgIsXgRIruQ3FUdfvfPSkl2cUy85M55i9nmb+cI7fsGWYtpNB/xDMag0dTDBxOoUc0ak6NmcJMo1czXZhulOdL87h1YwrEQ/G2cYDB+CBlq9wu4hvufG49vhVd0cmEMw3hzhiZpnhvzH1hN9Stv/e7heO4FFarZJdyLM++z8zFH7B6/SKFhXVcSyJUSWK4SGq8SGzQ5Sd/9s3bep07FXoNbzD2BWAWbzD2M62hGSHEALAopZRCiKeALwJjfvX/CqxJKX9tuxd8O0LvujZf+vzP4SqraPEiRtJET1jokSghZRynOkxlrZ/1qS7yi71YpR4URaXnUNzz+g97BiDdF0Uo+2tU37EtVqY9UV+66gn78tQ1HH9gUY9E6Ds8Qf/ho/Qf8VJmYAihKBRWV7j65mtcefNVrr/7FnathqYbjJ5+mCOPP8nhx54k2dO76TVLuRqzF9eZPe957fkVb2qga5gsZa5xMfIWs6mLlKNZTvee9sI9fY/zaN+jpIztT7vMLpUbor86WwQBQ0fTHH2ij4nH+4gmb3PROCm9AdRW4VcNCpkxpqycJ+gFT8jrYp5rhGy82+kHY4MdPfND8UO3Poh9m0gpKaxWmZ/MNbz11bkiSC8C0n0ozuCRFANHUwxOpNtCllK6mOYKleo01coMleqMn09Trc4TiYyQTp0hnT5DMvkIqtrZSJdyNeYv55if9IR/ZbqAbHn9oaNpBo964h9LtQup6ZjMFmfbxH+qMMV0fprZ4iyOP8NIE1rjzubtCvde3O28FVbNIbdcIbdUZH1likJ+kkr1KpYzhWPPUs3mKc4JinNRpKughm2SIybpkRg946Mk0xOkuo7R3XecRPL2lkHeiemVnwB+G2965R9IKX9TCPE5ACnl7wohfgX4ZcAGKsA/lFJ+XwjxHPA94F2gbpb/Kynln9/o9W5P6B1+97O/QKXQ3m3TYxrhlESLlwnFCuhJEyNpoSclujGAVRygsNhLZX0AMz8I9jC9I11t4r/xy7ubOLbFytR1Fn1BX7wyycrUVRz/kWp6JEr/4Qn6fEHvP3yUzMCgtx7NTbBNk5n332XyjVe5+uar5Ja8GHnP6Dijp58gljlOudjF3MUc6wtl//U0ho+nG6GYrsEYQhGsVdd4a+mtRpz//ZX3saV3jUfTRxuzex7vf5yh2BBCCKSU2NLGcixMx8R0zUZuORa5xQrzb5dZOVujuiJBSCKjEuOEiTZRxtZr3rGu6Q2KOd7xjbJ/vnqb1nNbrtVY9reVgdgAY4mmR173Ng8lDu2Z97cyVWThii+skznKuXovTfXDL5633jeeQKgFqtW6iE+3iPkM1eoMrtse9tL1XsLhQxhGP+XyFUqliwAIoZFInCadeoJ0+gyp1BPoenfHazSrNotX8sxNZpm/nGXxSh7b8n7eyd4IQxOphvCn+7d+kIjlWqxWVomFYsRD8X0j2p2QUlItWuSWK2SXyuRW5z0xr1zDcqdR9Dn0xCKh+BJCsamu6+SvJchdS1Je9sZAIukwh05PcPzpZzn88Icwdni55/vqhqlqsUhuaYHs4oKfz5Pzy/mVZW9tFh+hgpFwCSXKXg8gaaEnbDQjjmt3Y5WHMPODqIyS6T5B3+gg/YeT9I4m0MN3Hmu1LYuVqWuN0MvilcusTF3Hdfw1QaIx+o9M0Ff31A9PkO7fnqjfjGrZ4sLL57j4/ZdZuPI2ZmkKz00Mk+g5zsiDT3D6+WcZPjGIso0eTsWucHblbEP43156uxG7jmpRHOlgOub2VoOU0FUeZGL1MY6uPk6q2osjHGZTF5jsfpOrXe9iapVGXFRXdEJqqC1vjbfW26SMlCfovrCPJEb2fDmFasli4UrTW1+61hTNRHeYwaMa3eNlUoM5tMgK1dqsJ+yVaarVWRynfRBR09JEIocIhw8RiRwiEh4hHB4mEvHyjV67ZWXJ5d4km3uNbPY18vl3kP5zl6PRiRbhP0MkMtpRjOvGaX4yy9wlzzhVi15vMxwPMdgi/L2jiVsKJd1NXFdSXK+SX6543vnKKvncFSrVa9juNGpkHj2+iJ5YRNVbZkhJDeEMYa4NUpiOsHq1RGnV++4PTBzn6JPPMHHmaboPdf78mueRUFyCxO0tj3xfCf2NcGybwupKm/jnFj1jkF2cw6y037moRWz0RL0HYKGGVRQljZR9GOHDJFPH6Rk4xeD44ZvGQm3T9ES91VOfbhH1WKwt9NJ/+Cip/oEd83KsmsP8ZJbZC+vMXMiyfD2PlKBqCgMTKQYOGyCnWZs9y7W3XqdSyCOEwtCJkxx5/KmbDuhuxHEdLmcv88bSG0zlpwgpITRFaxPeTmLcKtS6qqMJjcq8ZOHdMrNvFyitmSiaYPRUF0fP9HP4kZ42o+u6Npa1jmmtYpmrmOYqpuXlrltFUQwvCR1F9XPFQFFa8437NtYbCHFrz1t1XZdyrsz6/BrZJYeVmRrzkzmyS2uEYqsY8VXShwok+/IYyVVEaImaNYtt59rOo6oxIuFDhCMjfn6oZXsYTbvF6bEb/29OjULhXbK518llXyObe71xDbreSzp1hlT6CdKpM8Tjp1A6DC434vyXvTj/3GSO/BZx/v7DKVRV4DgS13ZxbInruDhtZb/OcXFtuTm3vfauI9vzRhu3ef7G6/jnlWVQl3CVORxmCEUXPM88sUgo0v7ZK7IfQx8jFjtMKnMUwzjE6mSF6Xcuc+XN16kW8qiaxujpR5g48wwTTzxFvKtzr8j/Unizvq6/BNdehOvf92YT/frZ2/rfBUK/TSrFQsMAZBcXyC7OsT5/ndzSAsW1fPuKBorr9wAs9LiLqkcxot0kug7TM/wAdiVEYbFAfnGJ7MIU+eVZbzVIIBSOkRkc99LQOF1Dh4ll/Dndwh869kVE+H+amtLSxm/Xpjf1pYyF18Weu+SJ++LVPK4jURRB/5Ekw8e9UEz/kSRaqH2E33UdFi5f4uqbrzL5xqssX/NmEiV6ejny2JMceeLJmw7o7hRSSmy7gGWtUqutsDw7zfyVKVYXZnHlOlqkQKyrgh4vgZLFtrMdzyOEhqKEcV2z4bHe+rWAayk4NQW7piEtA9cycE0dpxbCrqrYNRW7omBXBXatuYK0Y8lm8BJQdImRsNCTVe87lLDQ4xbhFCR6+kmkhjeJeSQygqal72qIQ0qXUuky2dxrvvC/RrXqDfCrapRU8jFS6TOkU0+QTD6KpnW+yaoR57/sefz1OP+uIDwHRlUFqm6jx9YJxVcIxVYIRVdQI8uo4WUUfRkllN9waAojNEY8foRkeoJY7AjR6GEikVFUNUIpu86VN17l8qs/YOrdt7EtEyMW48hjTzJx5hkOP/o4emSL+2lcBxbegWsveeJ+/fvNmw3TozD2HIx/EB75+dt6hkQg9DuAY9sUVpbJLi2QXZhnbX6StflrZBcWKK4WcMzOMwNUwybSbWOkBKG4gRZOgEjhminsSgq7msSpJbGrKZxaDNjZbq0Q0Dua4NBJb1bMwETqlsNOhTV/QPeN15h69y2sWrUxoFufyZPs6dv2+RynhmU1vW3P815p874tc61RlrLD6oKAoiRxzSTVfBSzFEdaSRKZAXqHDtE/Nko40ouu96Dr3Wiat8Kg6zhUSwUqhXUqhSyVYpZqKUe1WKBazFMtFamVS9RKJWrlCrVSBbNcpVauYVVNbhZ5UkJeUnWJZkhUQ6KFHbSwi2rYaGEbnAhOJYlVCFHNS8rrVRyrfdmDcCxOsrefZG8fqb4+v9xPqtcrG9G7v3RBnWp1zvf4Xyebe41i8TwgEUIlHn+AdPrJhudv6D0dz2FWbRau5BqzrlRNQdUUFFV4uSZQVQXFF2zFr1c1gaI2c0V1sJwlLHsW05qjZs1SqzYHnmtm+01gQoQIh4daDGg9tDVKNDpOKJTedK1rczNcfvVlJl97hblL50FKkr19TJx5mqNnnmH45IOonRZAcyyYf9v31l+CqZeh5huWriMw9kEYf87L0yN38i/x31sg9LuKN1Dj9QaWZy+zcPU8elyQ6FdRo2Ucdw3bWcV2VrAdryzl5tvRQUVTMqhqF5rSg6Z2oypdqEoXmtKNqnajCq+siKTn1rcsa+/deNiy+Jam0De2M/cPSOniOGVqlTzT77/NtbfeYuqd9ymseIuepQczDJ4apP9kN+lDYVxZwXEqOE7J98jXPCE3V3GczotWKYrhCXOom5Deja53o4e8POSXNS0DdgTHDGFVTGrlEtVSiYUry8xfWmR5egW7VkEoJtGERA87QM0X7yJm5cZ3nwpFQw1FUBQDRBjX1XFdHSG8bS830CMxoskE8UySRHeaVF+aVF+KRFeEeCZMPG1se+qulJJKIU9+aZHc8hL55WaeX14it7yIXWv/vrQagmRvX8MAeIahHyN695YvsO0CudwbZP1QTz7/Fq7rXW8kMk46faYxuycSubVlQ6R0qZlL3gBzZbox8Fwfo6jV5pFtawMphI2BLUNbhtGPt6rLDV7TdZm7dIHJ117m8muvsD43A0Df4QmOnvHi7b1jhze/D7vmLW7YEPZXmktk9BxvF/bk4LY/g+1y/wj91/+Zl2sR7267eh6Keg/kaN3XKdeMHb8NvxNSShyniGmuUKstY1ormLVlTHMZ01ylZi77Hq63LV0TRYLiShTHy1WpYqgpDCWFriYwRIyQiBISEUKE0YSOqkSwu0cxe0ZxVHCcckN8G7lb8fd3qHOada67eeVFKaGW1clPxclPxSkuRMEVqIZNcrRCZtyh67CCEY81RFtVu1BkEuw40o4iLQOnFsIxFayq3fSmS0WqDc+6nt9cqBGCkBFBCAPbDoHUUbQIyZ4UsXQC19VxbB2rpmFWVWplFYSBEGEQBqChRzTimTCJjEE8YxDvCnt5JtzYDul3Z4lk6GwI8itLnhFYWuxoCIxYrK0H0GoIkr19hGO7tyKm65oUCu81BnhzudexLM8hCIW6GsKfSp8hET+FbeepVGe9WUOVmQ1iPrcp1Kbrfc0BZz+kVR9wNoxBlNu4j8Eya0y9+xaXX32FK2/8kHIui6KqjDz4MBNnnmbiiac3T0O2qt5aT/UY+8yrzRVK+x6EsQ94oZixD0J8+z3e2+X+Efr/dgKqOXA7d/VvjvAGQ7TwDfKNRiKyuY0W9q7BrnpfBtt/vJ9d8fPW/S31Vku9n6RdQ3Ra3vYWkEApqpJPaBQSGvm4RiGuIRUVVY2iqhE/b00t+5QIqhrz98U6tPe27SrMvH+Ja2+9y7W336CSzyGEQmZwCKtW27ZQG9EoRjSOEYsRjsYwYrHGthGNEY7FMGJxjEZdjHDMq9fDkcasJMd2mT63xuXXlrjy9jJW1UENKQ3RTmwh4kbk3lqHvGEIOvUGlrzcqrV/h/RIFD0aRdNCqKEQqhZCDWl+HkLVtJb97dvaFvvVUGjzMZqXm/Yipeo5SuV3KRbfoWbNIFSJUL3FOqUrkI7AdQSqkiGk9qNr/YS0XjSlh5DahSoyKEoS6XihVMeycGzLyy0Lx7GbZdvC9suubWPb1ob2tle2vXIln8O2TPRIlMOPneHomac5/NiZ9p6RWfLWc7r+khdnn33Nv2tbeHdS1731sQ94d2TfZe4foa/j2P4j+Krbz61y8zF9DSG+2TH+o/5udIt+A9FiKPzH/dV7ETuyP4zUQliYWG4Z0y1im2voa3OElq8TWppEXTiP4q8vL5UQDJxG1J+TO/S49/SsHXqQuOs6LE5e5sobP2R1Zho9GvUEukWwjVi8KeR+nR4O78j00Y04lotVczBi2r6er70bSCmpFvLkF+fIzU+TX5onv7yEKVUc22mKX5t42o39dZF0/Xrbtti9kdQ7Y6PRqRsyRdOaRm2jgdJChONxxh99gpEHTqNqfo+gmveF/UVP2Ofe8J/FrMLgI763/hyMPnPLT2TbDe4/ob/buE67cbBr/rKjLUKshu5KWOiGNB6Y8ob3pZ19A+begvrt/aGY9wUefhyGHvPyzOG9v+6DjJTeoF392cBW2c9by+WmM9KxruXZwp3q6mnjOveNp6Od8ZeXPuM9DOcm/28pJa6zlYFoGodWw9D0tO22dq7joKjqlr0DTfNEulW0mwKttW0rqnpnRryShakfNGPs8297ayEpmucINYT9aTDubBrrbhAIfcDW1JcZbgj/G94zc+trl0cyvug/4X3Zhx+HxM1XtLzvcCwor3qptNIsd9pXK7SL8kYB3g5KqBk2rK/kWM/rYcTWfW1tI97xy+f9x2O+5RkKgFivL/xPeE8SG3r8rjxT4K5TXIal97xnCiy9D/NveauXIr314YfPNOPrI095q2TucwKhD7g1HMv78jc8/zf958n6gpQYavf6hx7zDMJBQUpvGlx51VsorSHSfl5a3bzdsk7OJsJp7+Ht0W4vGckNAhzuIMp+rnUSc7+HuFM4lrc88+xrMOM/VGb1kl8pvJDeoTNNz7/31I4/LW3XqBU9g7b0vi/q78HSOSgtN9tEu72F7cY+4MXZh8/c2fLXe0Qg9AF3jln2bvZoDfusTTbru440Pf7hJ2DgYW+N8LuN63oD4Y7l57Y3huJa3mDaRm+7TcRbRH2rAX3V8EW7y3tMYF28N+6rC3uk694RxVYq697/eOY13wC85i0OB16Ib+ixptc/fGZXpgveEo4Fq5MtXvo5r7x+rdkmFPUeENP/gDcrpu+UtzT1XZgRczcIhD5gd6ise93+RtjnTcj7z28VqvdDGnrMe8SedNtF12kV4xuUXf+Y7ZZvNQwSyfhiXRftrhbvuy7a3c1tPXZ/jllICWtX/MdIvuZ5/QvvNg1icrjF63/SG+vZDUMvpfds3aX3W7z092HlYnNShFCh+2hTyPse8MQ9Pb6rzzTeawKhD7h7FBbb4/2zbzQ9wTqq7t8+qm2j7Kd6WdGaz9i81XIo2i7ikcy96W3vF6yqJ/Yzrza9/qz/ZDiheiJ76MmmAeg+emtCW1lvCvmiH3JZOtceJksON4W87qX3HL8nQy93SiD0AXtHfV34ulgr6v3pEd8vFJeboj/zqtfLq9/2H055Yb261z/8hNdbsiqwfGGzl16Yb543nGoJt7SI+j6Y1rhfuJHQB+5MwO4ixMGctRHQmXgvnPgJL4E3ZrJyscXrfx2+9981H+EYH4DSUnNbNaD3OBz+kC/ofkoOBQ7CHRAIfUBAwO6hKNB30kuP/4K3r1b0pjPOvObNiEmNNL30riNBOG0XCD7RgICAu4sR96Yxjj+311dy33Bwh6ADAgICAoBA6AMCAgIOPIHQBwQEBBxwAqEPCAgIOOAEQh8QEBBwwAmEPiAgIOCAEwh9QEBAwAEnEPqAgICAA86+XOtGCLEMXL/Nw3uAlR28nHuZ4LNoJ/g82gk+jyYH4bMYk1L2dqrYl0J/JwghXttqYZ/7jeCzaCf4PNoJPo8mB/2zCEI3AQEBAQecQOgDAgICDjgHUeh/b68vYB8RfBbtBJ9HO8Hn0eRAfxYHLkYfEBAQENDOQfToAwICAgJaCIQ+ICAg4IBzYIReCPFxIcQFIcRlIcR/udfXs5cIIUaEEN8SQpwTQrwnhPgHe31Ne40QQhVCvCmE+OpeX8teI4RICyG+KIQ4739Hnt3ra9pLhBC/7v9Ozgoh/oMQ4sA9WfxACL0QQgV+B/gJ4AHg54UQD+ztVe0pNvAbUspTwDPA//U+/zwA/gFwbq8vYp/weeAvpZQngUe4jz8XIcQw8KvAmf9/O3fvGkUURnH4d3AVTESwNSskAbGxiYWIAQtjp2hroUV6BStB/wYRO5uojQGLmMJC0MI+iFEQtZEoyWrENH5gE8VjcUfYQkFh4F3uvk81c6tTzBzuxzC29wNbgNOxqdpXRdEDB4HXtldsbwJ3gFPBmcLYXre93Fx/pbzIY7Gp4kjqAseBuegs0STtBI4ANwBsb9r+FBoqXgfYLqkDjADvg/O0rpaiHwPW+u57DHGx9ZM0DkwBS8FRIl0DLgI/g3MMgklgA7jVbGXNSRqNDhXF9jvgCrAKrAOfbT+MTdW+Wopefxgb+u9GJe0A7gIXbH+JzhNB0gngo+0n0VkGRAc4AFy3PQV8A4b2TEvSLsrqfwLYDYxKOhObqn21FH0P2NN336XC5df/kLSVUvLzthej8wSaBk5KekvZ0jsq6XZspFA9oGf79wpvgVL8w+oY8Mb2hu3vwCJwODhT62op+sfAXkkTkrZRDlPuBWcKI0mUPdhXtq9G54lk+5Ltru1xynPxyHZ1M7Z/ZfsDsCZpXzM0A7wMjBRtFTgkaaR5b2ao8HC6Ex2gDbZ/SDoHPKCcmt+0/SI4VqRp4CzwXNKzZuyy7ftxkdIAOQ/MN5OiFWA2OE8Y20uSFoBlytdqT6nwdwj5C4SUUqpcLVs3KaWU/iKLPqWUKpdFn1JKlcuiTymlymXRp5RS5bLoU0qpcln0KaVUuV8qdsVCcy2MeAAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "sample = X_train[6,:,13:33]\n",
    "\n",
    "plt.plot(np.arange(10), sample)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 111,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor(1)"
      ]
     },
     "execution_count": 111,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_train[6]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 109,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([0, 0, 0,  ..., 0, 0, 1])"
      ]
     },
     "execution_count": 109,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_train"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 110,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([    6,    13,    49, ..., 37668, 37669, 37673]),)"
      ]
     },
     "execution_count": 110,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.where(y_train == 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "num_epochs = 10\n",
    "nn_epochs = {}\n",
    "\n",
    "# Train the model\n",
    "total_step = len(val_iterator)\n",
    "for epoch in range(num_epochs):\n",
    "    for i, (images, labels) in enumerate(val_iterator):  \n",
    "        # Move tensors to the configured device\n",
    "        #print(images.shape)\n",
    "        images = images.reshape(-1, 70, 70, 70).to(device)\n",
    "        labels = labels.to(device)\n",
    "        \n",
    "        # Forward pass\n",
    "        #print(images.shape)\n",
    "        outputs = rn(images)\n",
    "        pred_y = torch.max(outputs, 1)[1].data.squeeze()\n",
    "        loss = loss_func(pred_y, labels)\n",
    "        \n",
    "        # Backward and optimize\n",
    "        optimizer.zero_grad()\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "        #print(i)\n",
    "\n",
    "    print ('Epoch [{}/{}], Step [{}/{}], Loss: {:.4f}' \n",
    "           .format(epoch+1, num_epochs, i+1, total_step, loss.item()))\n",
    "    nn_epochs[epoch] = {\"loss\": loss.item(), #\"train\": eval_rn(rn, \"train\"), \n",
    "                 \"val\": eval_rn(rn, \"val\"), \"test\": eval_rn(rn, \"test\")}\n",
    "    rn.train()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.9.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}