customUNet.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "customUNet.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "mount_file_id": "1oSKnfq-OXtsCW8zmo48tDpr0oWoqBIro",
      "authorship_tag": "ABX9TyOsX3EdBPhPthvTh7834kGC",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/Orasz/CNN4COVID19/blob/main/customUNet.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "toywndUptxNa",
        "outputId": "fa5982af-7961-481a-f9ac-d292a04693b5"
      },
      "source": [
        "!pip install git+https://github.com/albumentations-team/albumentations.git"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Collecting git+https://github.com/albumentations-team/albumentations.git\n",
            "  Cloning https://github.com/albumentations-team/albumentations.git to /tmp/pip-req-build-5ihtlvvs\n",
            "  Running command git clone -q https://github.com/albumentations-team/albumentations.git /tmp/pip-req-build-5ihtlvvs\n",
            "Requirement already satisfied (use --upgrade to upgrade): albumentations==0.5.2 from git+https://github.com/albumentations-team/albumentations.git in /usr/local/lib/python3.7/dist-packages\n",
            "Requirement already satisfied: numpy>=1.11.1 in /usr/local/lib/python3.7/dist-packages (from albumentations==0.5.2) (1.19.5)\n",
            "Requirement already satisfied: scipy in /usr/local/lib/python3.7/dist-packages (from albumentations==0.5.2) (1.4.1)\n",
            "Requirement already satisfied: scikit-image>=0.16.1 in /usr/local/lib/python3.7/dist-packages (from albumentations==0.5.2) (0.16.2)\n",
            "Requirement already satisfied: imgaug>=0.4.0 in /usr/local/lib/python3.7/dist-packages (from albumentations==0.5.2) (0.4.0)\n",
            "Requirement already satisfied: PyYAML in /usr/local/lib/python3.7/dist-packages (from albumentations==0.5.2) (3.13)\n",
            "Requirement already satisfied: opencv-python>=4.1.1 in /usr/local/lib/python3.7/dist-packages (from albumentations==0.5.2) (4.1.2.30)\n",
            "Requirement already satisfied: imageio>=2.3.0 in /usr/local/lib/python3.7/dist-packages (from scikit-image>=0.16.1->albumentations==0.5.2) (2.4.1)\n",
            "Requirement already satisfied: pillow>=4.3.0 in /usr/local/lib/python3.7/dist-packages (from scikit-image>=0.16.1->albumentations==0.5.2) (7.0.0)\n",
            "Requirement already satisfied: PyWavelets>=0.4.0 in /usr/local/lib/python3.7/dist-packages (from scikit-image>=0.16.1->albumentations==0.5.2) (1.1.1)\n",
            "Requirement already satisfied: matplotlib!=3.0.0,>=2.0.0 in /usr/local/lib/python3.7/dist-packages (from scikit-image>=0.16.1->albumentations==0.5.2) (3.2.2)\n",
            "Requirement already satisfied: networkx>=2.0 in /usr/local/lib/python3.7/dist-packages (from scikit-image>=0.16.1->albumentations==0.5.2) (2.5)\n",
            "Requirement already satisfied: six in /usr/local/lib/python3.7/dist-packages (from imgaug>=0.4.0->albumentations==0.5.2) (1.15.0)\n",
            "Requirement already satisfied: Shapely in /usr/local/lib/python3.7/dist-packages (from imgaug>=0.4.0->albumentations==0.5.2) (1.7.1)\n",
            "Requirement already satisfied: pyparsing!=2.0.4,!=2.1.2,!=2.1.6,>=2.0.1 in /usr/local/lib/python3.7/dist-packages (from matplotlib!=3.0.0,>=2.0.0->scikit-image>=0.16.1->albumentations==0.5.2) (2.4.7)\n",
            "Requirement already satisfied: python-dateutil>=2.1 in /usr/local/lib/python3.7/dist-packages (from matplotlib!=3.0.0,>=2.0.0->scikit-image>=0.16.1->albumentations==0.5.2) (2.8.1)\n",
            "Requirement already satisfied: kiwisolver>=1.0.1 in /usr/local/lib/python3.7/dist-packages (from matplotlib!=3.0.0,>=2.0.0->scikit-image>=0.16.1->albumentations==0.5.2) (1.3.1)\n",
            "Requirement already satisfied: cycler>=0.10 in /usr/local/lib/python3.7/dist-packages (from matplotlib!=3.0.0,>=2.0.0->scikit-image>=0.16.1->albumentations==0.5.2) (0.10.0)\n",
            "Requirement already satisfied: decorator>=4.3.0 in /usr/local/lib/python3.7/dist-packages (from networkx>=2.0->scikit-image>=0.16.1->albumentations==0.5.2) (4.4.2)\n",
            "Building wheels for collected packages: albumentations\n",
            "  Building wheel for albumentations (setup.py) ... \u001b[?25l\u001b[?25hdone\n",
            "  Created wheel for albumentations: filename=albumentations-0.5.2-cp37-none-any.whl size=86173 sha256=fdd0148e9c9cb5f680ec28e4f5bffb84e3b49a8be20288d8d3410fe3d32100fb\n",
            "  Stored in directory: /tmp/pip-ephem-wheel-cache-whpksdva/wheels/e2/85/3e/2a40fac5cc1f43ced656603bb2fca1327b30ec7de1b1b66517\n",
            "Successfully built albumentations\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kOZF3gqPJy8x"
      },
      "source": [
        "\r\n",
        "import os\r\n",
        "from PIL import Image\r\n",
        "from torch.utils.data import Dataset\r\n",
        "import numpy as np\r\n",
        "import albumentations as A\r\n",
        "from albumentations.pytorch import ToTensorV2\r\n",
        "from tqdm import tqdm\r\n",
        "import torch.optim as optim\r\n",
        "\r\n",
        "\r\n",
        "#DATASET\r\n",
        "\r\n",
        "\r\n",
        "class QaTaDataset(Dataset):\r\n",
        "    def __init__(self, image_dir, mask_dir, transform=None):\r\n",
        "        self.image_dir = image_dir\r\n",
        "        self.mask_dir = mask_dir\r\n",
        "        self.transform = transform\r\n",
        "        self.images = os.listdir(image_dir)\r\n",
        "        self.masks = os.listdir(mask_dir)\r\n",
        "\r\n",
        "    def __len__(self):\r\n",
        "        return len(self.images)\r\n",
        "\r\n",
        "    def __getitem__(self, index):\r\n",
        "        img_path = os.path.join(self.image_dir, self.images[index])\r\n",
        "        mask_path = os.path.join(self.mask_dir, self.masks[index])\r\n",
        "        image = np.array(Image.open(img_path).convert(\"RGB\"))\r\n",
        "        mask = np.array(Image.open(mask_path).convert(\"L\"), dtype=np.float32)\r\n",
        "        mask[mask == 255.0] = 1.0\r\n",
        "\r\n",
        "        if self.transform is not None:\r\n",
        "            augmentations = self.transform(image=image, mask=mask)\r\n",
        "            image = augmentations[\"image\"]\r\n",
        "            mask = augmentations[\"mask\"]\r\n",
        "\r\n",
        "        return image, mask"
      ],
      "execution_count": 8,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8beJ2i_jenTc"
      },
      "source": [
        "#UTILS\r\n",
        "import torch\r\n",
        "import torchvision\r\n",
        "from torch.utils.data import DataLoader\r\n",
        "\r\n",
        "def save_checkpoint(state, filename=\"my_checkpoint.pth.tar\"):\r\n",
        "    print(\"=> Saving checkpoint\")\r\n",
        "    torch.save(state, filename)\r\n",
        "\r\n",
        "def load_checkpoint(checkpoint, model):\r\n",
        "    print(\"=> Loading checkpoint\")\r\n",
        "    model.load_state_dict(checkpoint[\"state_dict\"])\r\n",
        "\r\n",
        "def get_loaders(\r\n",
        "    train_dir,\r\n",
        "    train_maskdir,\r\n",
        "    val_dir,\r\n",
        "    val_maskdir,\r\n",
        "    batch_size,\r\n",
        "    train_transform,\r\n",
        "    val_transform,\r\n",
        "    num_workers=4,\r\n",
        "    pin_memory=True,\r\n",
        "):\r\n",
        "    train_ds = QaTaDataset(\r\n",
        "        image_dir=train_dir,\r\n",
        "        mask_dir=train_maskdir,\r\n",
        "        transform=train_transform,\r\n",
        "    )\r\n",
        "\r\n",
        "    train_loader = DataLoader(\r\n",
        "        train_ds,\r\n",
        "        batch_size=batch_size,\r\n",
        "        num_workers=num_workers,\r\n",
        "        pin_memory=pin_memory,\r\n",
        "        shuffle=True,\r\n",
        "    )\r\n",
        "\r\n",
        "    val_ds = QaTaDataset(\r\n",
        "        image_dir=val_dir,\r\n",
        "        mask_dir=val_maskdir,\r\n",
        "        transform=val_transform,\r\n",
        "    )\r\n",
        "\r\n",
        "    val_loader = DataLoader(\r\n",
        "        val_ds,\r\n",
        "        batch_size=batch_size,\r\n",
        "        num_workers=num_workers,\r\n",
        "        pin_memory=pin_memory,\r\n",
        "        shuffle=False,\r\n",
        "    )\r\n",
        "\r\n",
        "    return train_loader, val_loader\r\n",
        "\r\n",
        "def check_accuracy(loader, model, device=\"cuda\"):\r\n",
        "    num_correct = 0\r\n",
        "    num_pixels = 0\r\n",
        "    dice_score = 0\r\n",
        "    model.eval()\r\n",
        "\r\n",
        "    with torch.no_grad():\r\n",
        "        for x, y in loader:\r\n",
        "            x = x.to(device)\r\n",
        "            y = y.to(device).unsqueeze(1)\r\n",
        "            preds = torch.sigmoid(model(x))\r\n",
        "            preds = (preds > 0.5).float()\r\n",
        "            num_correct += (preds == y).sum()\r\n",
        "            num_pixels += torch.numel(preds)\r\n",
        "            dice_score += (2 * (preds * y).sum()) / (\r\n",
        "                (preds + y).sum() + 1e-8\r\n",
        "            )\r\n",
        "\r\n",
        "    print(\r\n",
        "        f\"Got {num_correct}/{num_pixels} with acc {num_correct/num_pixels*100:.2f}\"\r\n",
        "    )\r\n",
        "    print(f\"Dice score: {dice_score/len(loader)}\")\r\n",
        "    model.train()\r\n",
        "\r\n",
        "def save_predictions_as_imgs(\r\n",
        "    loader, model, folder=\"saved_images/\", device=\"cuda\"\r\n",
        "):\r\n",
        "    model.eval()\r\n",
        "    for idx, (x, y) in enumerate(loader):\r\n",
        "        x = x.to(device=device)\r\n",
        "        with torch.no_grad():\r\n",
        "            preds = torch.sigmoid(model(x))\r\n",
        "            preds = (preds > 0.5).float()\r\n",
        "        torchvision.utils.save_image(\r\n",
        "            preds, f\"{folder}/pred_{idx}.png\"\r\n",
        "        )\r\n",
        "        torchvision.utils.save_image(y.unsqueeze(1), f\"{folder}{idx}.png\")\r\n",
        "\r\n",
        "    model.train()"
      ],
      "execution_count": 3,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FPmo0FNsdTAX"
      },
      "source": [
        "#MODEL\r\n",
        "import torch.nn as nn\r\n",
        "class DoubleConv(nn.Module):\r\n",
        "    def __init__(self, in_channels, out_channels):\r\n",
        "        super(DoubleConv, self).__init__()\r\n",
        "        self.conv = nn.Sequential(\r\n",
        "            nn.Conv2d(in_channels, out_channels, 3, 1, 1, bias=False),\r\n",
        "            nn.BatchNorm2d(out_channels),\r\n",
        "            nn.ReLU(inplace=True),\r\n",
        "            nn.Conv2d(out_channels, out_channels, 3, 1, 1, bias=False),\r\n",
        "            nn.BatchNorm2d(out_channels),\r\n",
        "            nn.ReLU(inplace=True),\r\n",
        "        )\r\n",
        "\r\n",
        "    def forward(self, x):\r\n",
        "        return self.conv(x)\r\n",
        "\r\n",
        "class UNET(nn.Module):\r\n",
        "    def __init__(\r\n",
        "            self, in_channels=3, out_channels=1, features=[64, 128, 256, 512],\r\n",
        "    ):\r\n",
        "        super(UNET, self).__init__()\r\n",
        "        self.ups = nn.ModuleList()\r\n",
        "        self.downs = nn.ModuleList()\r\n",
        "        self.pool = nn.MaxPool2d(kernel_size=2, stride=2)\r\n",
        "\r\n",
        "        # Down part of UNET\r\n",
        "        for feature in features:\r\n",
        "            self.downs.append(DoubleConv(in_channels, feature))\r\n",
        "            in_channels = feature\r\n",
        "\r\n",
        "        # Up part of UNET\r\n",
        "        for feature in reversed(features):\r\n",
        "            self.ups.append(\r\n",
        "                nn.ConvTranspose2d(\r\n",
        "                    feature*2, feature, kernel_size=2, stride=2,\r\n",
        "                )\r\n",
        "            )\r\n",
        "            self.ups.append(DoubleConv(feature*2, feature))\r\n",
        "\r\n",
        "        self.bottleneck = DoubleConv(features[-1], features[-1]*2)\r\n",
        "        self.final_conv = nn.Conv2d(features[0], out_channels, kernel_size=1)\r\n",
        "\r\n",
        "    def forward(self, x):\r\n",
        "        skip_connections = []\r\n",
        "\r\n",
        "        for down in self.downs:\r\n",
        "            x = down(x)\r\n",
        "            skip_connections.append(x)\r\n",
        "            x = self.pool(x)\r\n",
        "\r\n",
        "        x = self.bottleneck(x)\r\n",
        "        skip_connections = skip_connections[::-1]\r\n",
        "\r\n",
        "        for idx in range(0, len(self.ups), 2):\r\n",
        "            x = self.ups[idx](x)\r\n",
        "            skip_connection = skip_connections[idx//2]\r\n",
        "\r\n",
        "            if x.shape != skip_connection.shape:\r\n",
        "                x = TF.resize(x, size=skip_connection.shape[2:])\r\n",
        "\r\n",
        "            concat_skip = torch.cat((skip_connection, x), dim=1)\r\n",
        "            x = self.ups[idx+1](concat_skip)\r\n",
        "\r\n",
        "        return self.final_conv(x)\r\n"
      ],
      "execution_count": 4,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yVeJU1z9daJb"
      },
      "source": [
        "#TRAINING\r\n",
        "# Hyperparameters etc.\r\n",
        "LEARNING_RATE = 1e-5\r\n",
        "DEVICE = \"cuda\" if torch.cuda.is_available() else \"cpu\"\r\n",
        "BATCH_SIZE = 16\r\n",
        "NUM_EPOCHS = 50\r\n",
        "NUM_WORKERS = 2\r\n",
        "IMAGE_HEIGHT = 224  \r\n",
        "IMAGE_WIDTH = 224 \r\n",
        "PIN_MEMORY = True\r\n",
        "LOAD_MODEL = False\r\n",
        "TRAIN_IMG_DIR = \"/content/drive/MyDrive/UNET/train_img/\"\r\n",
        "TRAIN_MASK_DIR = \"/content/drive/MyDrive/UNET/train_masks/\"\r\n",
        "VAL_IMG_DIR = \"/content/drive/MyDrive/UNET/val_img/\"\r\n",
        "VAL_MASK_DIR = \"/content/drive/MyDrive/UNET/val_masks/\"\r\n",
        "\r\n",
        "def train_fn(loader, model, optimizer, loss_fn, scaler):\r\n",
        "    loop = tqdm(loader)\r\n",
        "\r\n",
        "    for batch_idx, (data, targets) in enumerate(loop):\r\n",
        "        data = data.to(device=DEVICE)\r\n",
        "        targets = targets.float().unsqueeze(1).to(device=DEVICE)\r\n",
        "\r\n",
        "        # forward\r\n",
        "        with torch.cuda.amp.autocast():\r\n",
        "            predictions = model(data)\r\n",
        "            loss = loss_fn(predictions, targets)\r\n",
        "\r\n",
        "        # backward\r\n",
        "        optimizer.zero_grad()\r\n",
        "        scaler.scale(loss).backward()\r\n",
        "        scaler.step(optimizer)\r\n",
        "        scaler.update()\r\n",
        "\r\n",
        "        # update tqdm loop\r\n",
        "        loop.set_postfix(loss=loss.item())\r\n",
        "\r\n",
        "\r\n",
        "def main():\r\n",
        "    train_transform = A.Compose(\r\n",
        "        [\r\n",
        "            A.Resize(height=IMAGE_HEIGHT, width=IMAGE_WIDTH),\r\n",
        "            #A.centerCrop(224),\r\n",
        "            A.Rotate(limit=35, p=1.0),\r\n",
        "            A.HorizontalFlip(p=0.5),\r\n",
        "            A.VerticalFlip(p=0.1),\r\n",
        "            A.Normalize(\r\n",
        "                mean=[0.0, 0.0, 0.0],\r\n",
        "                std=[1.0, 1.0, 1.0],\r\n",
        "                max_pixel_value=255.0,\r\n",
        "            ),\r\n",
        "            ToTensorV2(),\r\n",
        "        ],\r\n",
        "    )\r\n",
        "\r\n",
        "    val_transforms = A.Compose(\r\n",
        "        [\r\n",
        "            A.Resize(height=IMAGE_HEIGHT, width=IMAGE_WIDTH),\r\n",
        "            #A.centerCrop(224),\r\n",
        "            A.Normalize(\r\n",
        "                mean=[0.0, 0.0, 0.0],\r\n",
        "                std=[1.0, 1.0, 1.0],\r\n",
        "                max_pixel_value=255.0,\r\n",
        "            ),\r\n",
        "            ToTensorV2(),\r\n",
        "        ],\r\n",
        "    )\r\n",
        "\r\n",
        "    model = UNET(in_channels=3, out_channels=1).to(DEVICE)\r\n",
        "    loss_fn = nn.BCEWithLogitsLoss()\r\n",
        "    optimizer = optim.Adam(model.parameters(), lr=LEARNING_RATE)\r\n",
        "\r\n",
        "    train_loader, val_loader = get_loaders(\r\n",
        "        TRAIN_IMG_DIR,\r\n",
        "        TRAIN_MASK_DIR,\r\n",
        "        VAL_IMG_DIR,\r\n",
        "        VAL_MASK_DIR,\r\n",
        "        BATCH_SIZE,\r\n",
        "        train_transform,\r\n",
        "        val_transforms,\r\n",
        "        NUM_WORKERS,\r\n",
        "        PIN_MEMORY,\r\n",
        "    )\r\n",
        "\r\n",
        "    if LOAD_MODEL:\r\n",
        "        load_checkpoint(torch.load(\"my_checkpoint.pth.tar\"), model)\r\n",
        "\r\n",
        "\r\n",
        "    check_accuracy(val_loader, model, device=DEVICE)\r\n",
        "    scaler = torch.cuda.amp.GradScaler()\r\n",
        "\r\n",
        "    for epoch in range(NUM_EPOCHS):\r\n",
        "        train_fn(train_loader, model, optimizer, loss_fn, scaler)\r\n",
        "        check_accuracy(val_loader, model, device=DEVICE)\r\n",
        "\r\n",
        "        # print some examples to a folder\r\n",
        "        save_predictions_as_imgs(\r\n",
        "            val_loader, model, folder=\"/content/drive/MyDrive/UNET/immagini_salvate_unet_10epoche/\", device=DEVICE\r\n",
        "        )\r\n",
        "\r\n"
      ],
      "execution_count": 29,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "id": "hMRdDgdlwgGu",
        "outputId": "98c89d3e-198f-471a-c5e0-69277f7e7399"
      },
      "source": [
        "main()"
      ],
      "execution_count": 30,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\n",
            "\n",
            "\n",
            "\n",
            " 11%|█         | 16/148 [00:06<00:49,  2.68it/s, loss=0.343]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 11%|█▏        | 17/148 [00:06<00:48,  2.70it/s, loss=0.343]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 11%|█▏        | 17/148 [00:06<00:48,  2.70it/s, loss=0.352]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 12%|█▏        | 18/148 [00:06<00:48,  2.68it/s, loss=0.352]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 12%|█▏        | 18/148 [00:07<00:48,  2.68it/s, loss=0.292]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 13%|█▎        | 19/148 [00:07<00:48,  2.67it/s, loss=0.292]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 13%|█▎        | 19/148 [00:07<00:48,  2.67it/s, loss=0.334]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 14%|█▎        | 20/148 [00:07<00:47,  2.69it/s, loss=0.334]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 14%|█▎        | 20/148 [00:08<00:47,  2.69it/s, loss=0.324]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 14%|█▍        | 21/148 [00:08<00:47,  2.67it/s, loss=0.324]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 14%|█▍        | 21/148 [00:08<00:47,  2.67it/s, loss=0.337]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 15%|█▍        | 22/148 [00:08<00:47,  2.67it/s, loss=0.337]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 15%|█▍        | 22/148 [00:08<00:47,  2.67it/s, loss=0.341]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 16%|█▌        | 23/148 [00:08<00:46,  2.68it/s, loss=0.341]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 16%|█▌        | 23/148 [00:09<00:46,  2.68it/s, loss=0.304]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 16%|█▌        | 24/148 [00:09<00:46,  2.67it/s, loss=0.304]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 16%|█▌        | 24/148 [00:09<00:46,  2.67it/s, loss=0.311]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 17%|█▋        | 25/148 [00:09<00:46,  2.64it/s, loss=0.311]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 17%|█▋        | 25/148 [00:09<00:46,  2.64it/s, loss=0.332]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 18%|█▊        | 26/148 [00:09<00:46,  2.65it/s, loss=0.332]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 18%|█▊        | 26/148 [00:10<00:46,  2.65it/s, loss=0.391]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 18%|█▊        | 27/148 [00:10<00:45,  2.63it/s, loss=0.391]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 18%|█▊        | 27/148 [00:10<00:45,  2.63it/s, loss=0.343]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 19%|█▉        | 28/148 [00:10<00:45,  2.65it/s, loss=0.343]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 19%|█▉        | 28/148 [00:11<00:45,  2.65it/s, loss=0.342]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 20%|█▉        | 29/148 [00:11<00:44,  2.64it/s, loss=0.342]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 20%|█▉        | 29/148 [00:11<00:44,  2.64it/s, loss=0.363]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 20%|██        | 30/148 [00:11<00:44,  2.64it/s, loss=0.363]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 20%|██        | 30/148 [00:11<00:44,  2.64it/s, loss=0.326]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 21%|██        | 31/148 [00:11<00:43,  2.66it/s, loss=0.326]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 21%|██        | 31/148 [00:12<00:43,  2.66it/s, loss=0.307]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 22%|██▏       | 32/148 [00:12<00:43,  2.66it/s, loss=0.307]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 22%|██▏       | 32/148 [00:12<00:43,  2.66it/s, loss=0.328]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 22%|██▏       | 33/148 [00:12<00:43,  2.63it/s, loss=0.328]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 22%|██▏       | 33/148 [00:12<00:43,  2.63it/s, loss=0.308]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 23%|██▎       | 34/148 [00:12<00:43,  2.64it/s, loss=0.308]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 23%|██▎       | 34/148 [00:13<00:43,  2.64it/s, loss=0.349]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 24%|██▎       | 35/148 [00:13<00:42,  2.64it/s, loss=0.349]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 24%|██▎       | 35/148 [00:13<00:42,  2.64it/s, loss=0.377]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 24%|██▍       | 36/148 [00:13<00:43,  2.60it/s, loss=0.377]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 24%|██▍       | 36/148 [00:14<00:43,  2.60it/s, loss=0.401]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 25%|██▌       | 37/148 [00:14<00:42,  2.62it/s, loss=0.401]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 25%|██▌       | 37/148 [00:14<00:42,  2.62it/s, loss=0.316]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 26%|██▌       | 38/148 [00:14<00:41,  2.62it/s, loss=0.316]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 26%|██▌       | 38/148 [00:14<00:41,  2.62it/s, loss=0.36] \u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 26%|██▋       | 39/148 [00:14<00:41,  2.62it/s, loss=0.36]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 26%|██▋       | 39/148 [00:15<00:41,  2.62it/s, loss=0.341]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 27%|██▋       | 40/148 [00:15<00:41,  2.62it/s, loss=0.341]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 27%|██▋       | 40/148 [00:15<00:41,  2.62it/s, loss=0.388]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 28%|██▊       | 41/148 [00:15<00:40,  2.61it/s, loss=0.388]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 28%|██▊       | 41/148 [00:16<00:40,  2.61it/s, loss=0.358]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 28%|██▊       | 42/148 [00:16<00:40,  2.63it/s, loss=0.358]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 28%|██▊       | 42/148 [00:16<00:40,  2.63it/s, loss=0.306]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 29%|██▉       | 43/148 [00:16<00:39,  2.65it/s, loss=0.306]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 29%|██▉       | 43/148 [00:16<00:39,  2.65it/s, loss=0.291]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 30%|██▉       | 44/148 [00:16<00:39,  2.63it/s, loss=0.291]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 30%|██▉       | 44/148 [00:17<00:39,  2.63it/s, loss=0.305]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 30%|███       | 45/148 [00:17<00:38,  2.66it/s, loss=0.305]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 30%|███       | 45/148 [00:17<00:38,  2.66it/s, loss=0.31] \u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 31%|███       | 46/148 [00:17<00:38,  2.66it/s, loss=0.31]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 31%|███       | 46/148 [00:17<00:38,  2.66it/s, loss=0.349]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 32%|███▏      | 47/148 [00:17<00:38,  2.65it/s, loss=0.349]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 32%|███▏      | 47/148 [00:18<00:38,  2.65it/s, loss=0.328]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 32%|███▏      | 48/148 [00:18<00:37,  2.66it/s, loss=0.328]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 32%|███▏      | 48/148 [00:18<00:37,  2.66it/s, loss=0.293]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 33%|███▎      | 49/148 [00:18<00:37,  2.65it/s, loss=0.293]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 33%|███▎      | 49/148 [00:19<00:37,  2.65it/s, loss=0.33] \u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 34%|███▍      | 50/148 [00:19<00:36,  2.65it/s, loss=0.33]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 34%|███▍      | 50/148 [00:19<00:36,  2.65it/s, loss=0.304]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 34%|███▍      | 51/148 [00:19<00:36,  2.67it/s, loss=0.304]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 34%|███▍      | 51/148 [00:19<00:36,  2.67it/s, loss=0.414]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 35%|███▌      | 52/148 [00:19<00:36,  2.66it/s, loss=0.414]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 35%|███▌      | 52/148 [00:20<00:36,  2.66it/s, loss=0.279]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 36%|███▌      | 53/148 [00:20<00:35,  2.66it/s, loss=0.279]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 36%|███▌      | 53/148 [00:20<00:35,  2.66it/s, loss=0.388]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 36%|███▋      | 54/148 [00:20<00:35,  2.67it/s, loss=0.388]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 36%|███▋      | 54/148 [00:20<00:35,  2.67it/s, loss=0.35] \u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 37%|███▋      | 55/148 [00:20<00:34,  2.66it/s, loss=0.35]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 37%|███▋      | 55/148 [00:21<00:34,  2.66it/s, loss=0.336]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 38%|███▊      | 56/148 [00:21<00:34,  2.66it/s, loss=0.336]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 38%|███▊      | 56/148 [00:21<00:34,  2.66it/s, loss=0.294]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 39%|███▊      | 57/148 [00:21<00:34,  2.65it/s, loss=0.294]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 39%|███▊      | 57/148 [00:22<00:34,  2.65it/s, loss=0.314]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 39%|███▉      | 58/148 [00:22<00:34,  2.65it/s, loss=0.314]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 39%|███▉      | 58/148 [00:22<00:34,  2.65it/s, loss=0.389]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 40%|███▉      | 59/148 [00:22<00:33,  2.64it/s, loss=0.389]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 40%|███▉      | 59/148 [00:22<00:33,  2.64it/s, loss=0.31] \u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 41%|████      | 60/148 [00:22<00:33,  2.66it/s, loss=0.31]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 41%|████      | 60/148 [00:23<00:33,  2.66it/s, loss=0.392]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 41%|████      | 61/148 [00:23<00:32,  2.66it/s, loss=0.392]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 41%|████      | 61/148 [00:23<00:32,  2.66it/s, loss=0.344]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 42%|████▏     | 62/148 [00:23<00:32,  2.67it/s, loss=0.344]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 42%|████▏     | 62/148 [00:23<00:32,  2.67it/s, loss=0.374]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 43%|████▎     | 63/148 [00:23<00:32,  2.66it/s, loss=0.374]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 43%|████▎     | 63/148 [00:24<00:32,  2.66it/s, loss=0.371]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 43%|████▎     | 64/148 [00:24<00:31,  2.68it/s, loss=0.371]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 43%|████▎     | 64/148 [00:24<00:31,  2.68it/s, loss=0.308]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 44%|████▍     | 65/148 [00:24<00:31,  2.68it/s, loss=0.308]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 44%|████▍     | 65/148 [00:25<00:31,  2.68it/s, loss=0.347]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 45%|████▍     | 66/148 [00:25<00:30,  2.66it/s, loss=0.347]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 45%|████▍     | 66/148 [00:25<00:30,  2.66it/s, loss=0.327]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 45%|████▌     | 67/148 [00:25<00:30,  2.67it/s, loss=0.327]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 45%|████▌     | 67/148 [00:25<00:30,  2.67it/s, loss=0.297]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 46%|████▌     | 68/148 [00:25<00:29,  2.68it/s, loss=0.297]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 46%|████▌     | 68/148 [00:26<00:29,  2.68it/s, loss=0.395]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 47%|████▋     | 69/148 [00:26<00:29,  2.66it/s, loss=0.395]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 47%|████▋     | 69/148 [00:26<00:29,  2.66it/s, loss=0.315]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 47%|████▋     | 70/148 [00:26<00:29,  2.68it/s, loss=0.315]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 47%|████▋     | 70/148 [00:26<00:29,  2.68it/s, loss=0.362]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 48%|████▊     | 71/148 [00:26<00:28,  2.68it/s, loss=0.362]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 48%|████▊     | 71/148 [00:27<00:28,  2.68it/s, loss=0.308]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 49%|████▊     | 72/148 [00:27<00:28,  2.70it/s, loss=0.308]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 49%|████▊     | 72/148 [00:27<00:28,  2.70it/s, loss=0.34] \u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 49%|████▉     | 73/148 [00:27<00:27,  2.68it/s, loss=0.34]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 49%|████▉     | 73/148 [00:28<00:27,  2.68it/s, loss=0.364]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 50%|█████     | 74/148 [00:28<00:27,  2.67it/s, loss=0.364]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 50%|█████     | 74/148 [00:28<00:27,  2.67it/s, loss=0.353]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 51%|█████     | 75/148 [00:28<00:27,  2.67it/s, loss=0.353]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 51%|█████     | 75/148 [00:28<00:27,  2.67it/s, loss=0.343]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 51%|█████▏    | 76/148 [00:28<00:26,  2.68it/s, loss=0.343]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 51%|█████▏    | 76/148 [00:29<00:26,  2.68it/s, loss=0.279]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 52%|█████▏    | 77/148 [00:29<00:26,  2.67it/s, loss=0.279]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 52%|█████▏    | 77/148 [00:29<00:26,  2.67it/s, loss=0.354]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 53%|█████▎    | 78/148 [00:29<00:26,  2.68it/s, loss=0.354]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 53%|█████▎    | 78/148 [00:29<00:26,  2.68it/s, loss=0.36] \u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 53%|█████▎    | 79/148 [00:29<00:25,  2.68it/s, loss=0.36]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 53%|█████▎    | 79/148 [00:30<00:25,  2.68it/s, loss=0.337]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 54%|█████▍    | 80/148 [00:30<00:25,  2.69it/s, loss=0.337]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 54%|█████▍    | 80/148 [00:30<00:25,  2.69it/s, loss=0.307]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 55%|█████▍    | 81/148 [00:30<00:24,  2.70it/s, loss=0.307]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 55%|█████▍    | 81/148 [00:30<00:24,  2.70it/s, loss=0.289]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 55%|█████▌    | 82/148 [00:30<00:24,  2.70it/s, loss=0.289]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 55%|█████▌    | 82/148 [00:31<00:24,  2.70it/s, loss=0.322]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 56%|█████▌    | 83/148 [00:31<00:24,  2.71it/s, loss=0.322]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 56%|█████▌    | 83/148 [00:31<00:24,  2.71it/s, loss=0.346]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 57%|█████▋    | 84/148 [00:31<00:23,  2.72it/s, loss=0.346]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 57%|█████▋    | 84/148 [00:32<00:23,  2.72it/s, loss=0.345]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 57%|█████▋    | 85/148 [00:32<00:23,  2.71it/s, loss=0.345]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 57%|█████▋    | 85/148 [00:32<00:23,  2.71it/s, loss=0.337]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 58%|█████▊    | 86/148 [00:32<00:22,  2.71it/s, loss=0.337]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 58%|█████▊    | 86/148 [00:32<00:22,  2.71it/s, loss=0.311]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 59%|█████▉    | 87/148 [00:32<00:22,  2.72it/s, loss=0.311]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 59%|█████▉    | 87/148 [00:33<00:22,  2.72it/s, loss=0.334]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 59%|█████▉    | 88/148 [00:33<00:22,  2.72it/s, loss=0.334]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 59%|█████▉    | 88/148 [00:33<00:22,  2.72it/s, loss=0.303]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 60%|██████    | 89/148 [00:33<00:21,  2.71it/s, loss=0.303]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 60%|██████    | 89/148 [00:33<00:21,  2.71it/s, loss=0.345]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 61%|██████    | 90/148 [00:33<00:21,  2.69it/s, loss=0.345]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 61%|██████    | 90/148 [00:34<00:21,  2.69it/s, loss=0.361]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 61%|██████▏   | 91/148 [00:34<00:20,  2.72it/s, loss=0.361]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 61%|██████▏   | 91/148 [00:34<00:20,  2.72it/s, loss=0.346]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 62%|██████▏   | 92/148 [00:34<00:20,  2.71it/s, loss=0.346]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 62%|██████▏   | 92/148 [00:35<00:20,  2.71it/s, loss=0.345]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 63%|██████▎   | 93/148 [00:35<00:20,  2.70it/s, loss=0.345]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 63%|██████▎   | 93/148 [00:35<00:20,  2.70it/s, loss=0.304]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 64%|██████▎   | 94/148 [00:35<00:20,  2.69it/s, loss=0.304]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 64%|██████▎   | 94/148 [00:35<00:20,  2.69it/s, loss=0.41] \u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 64%|██████▍   | 95/148 [00:35<00:19,  2.67it/s, loss=0.41]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 64%|██████▍   | 95/148 [00:36<00:19,  2.67it/s, loss=0.306]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 65%|██████▍   | 96/148 [00:36<00:19,  2.66it/s, loss=0.306]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 65%|██████▍   | 96/148 [00:36<00:19,  2.66it/s, loss=0.372]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 66%|██████▌   | 97/148 [00:36<00:19,  2.67it/s, loss=0.372]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 66%|██████▌   | 97/148 [00:36<00:19,  2.67it/s, loss=0.36] \u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 66%|██████▌   | 98/148 [00:36<00:18,  2.66it/s, loss=0.36]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 66%|██████▌   | 98/148 [00:37<00:18,  2.66it/s, loss=0.302]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 67%|██████▋   | 99/148 [00:37<00:18,  2.66it/s, loss=0.302]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 67%|██████▋   | 99/148 [00:37<00:18,  2.66it/s, loss=0.365]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 68%|██████▊   | 100/148 [00:37<00:17,  2.68it/s, loss=0.365]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 68%|██████▊   | 100/148 [00:38<00:17,  2.68it/s, loss=0.358]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 68%|██████▊   | 101/148 [00:38<00:17,  2.67it/s, loss=0.358]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 68%|██████▊   | 101/148 [00:38<00:17,  2.67it/s, loss=0.331]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 69%|██████▉   | 102/148 [00:38<00:17,  2.68it/s, loss=0.331]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 69%|██████▉   | 102/148 [00:38<00:17,  2.68it/s, loss=0.426]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 70%|██████▉   | 103/148 [00:38<00:16,  2.68it/s, loss=0.426]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 70%|██████▉   | 103/148 [00:39<00:16,  2.68it/s, loss=0.339]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 70%|███████   | 104/148 [00:39<00:16,  2.67it/s, loss=0.339]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 70%|███████   | 104/148 [00:39<00:16,  2.67it/s, loss=0.39] \u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 71%|███████   | 105/148 [00:39<00:16,  2.67it/s, loss=0.39]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 71%|███████   | 105/148 [00:39<00:16,  2.67it/s, loss=0.343]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 72%|███████▏  | 106/148 [00:39<00:15,  2.67it/s, loss=0.343]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 72%|███████▏  | 106/148 [00:40<00:15,  2.67it/s, loss=0.36] \u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 72%|███████▏  | 107/148 [00:40<00:15,  2.66it/s, loss=0.36]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 72%|███████▏  | 107/148 [00:40<00:15,  2.66it/s, loss=0.385]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 73%|███████▎  | 108/148 [00:40<00:14,  2.68it/s, loss=0.385]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 73%|███████▎  | 108/148 [00:41<00:14,  2.68it/s, loss=0.311]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 74%|███████▎  | 109/148 [00:41<00:14,  2.68it/s, loss=0.311]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 74%|███████▎  | 109/148 [00:41<00:14,  2.68it/s, loss=0.29] \u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 74%|███████▍  | 110/148 [00:41<00:14,  2.69it/s, loss=0.29]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 74%|███████▍  | 110/148 [00:41<00:14,  2.69it/s, loss=0.303]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 75%|███████▌  | 111/148 [00:41<00:13,  2.69it/s, loss=0.303]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 75%|███████▌  | 111/148 [00:42<00:13,  2.69it/s, loss=0.379]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 76%|███████▌  | 112/148 [00:42<00:13,  2.69it/s, loss=0.379]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 76%|███████▌  | 112/148 [00:42<00:13,  2.69it/s, loss=0.329]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 76%|███████▋  | 113/148 [00:42<00:12,  2.71it/s, loss=0.329]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 76%|███████▋  | 113/148 [00:42<00:12,  2.71it/s, loss=0.346]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 77%|███████▋  | 114/148 [00:42<00:12,  2.69it/s, loss=0.346]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 77%|███████▋  | 114/148 [00:43<00:12,  2.69it/s, loss=0.32] \u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 78%|███████▊  | 115/148 [00:43<00:12,  2.70it/s, loss=0.32]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 78%|███████▊  | 115/148 [00:43<00:12,  2.70it/s, loss=0.324]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 78%|███████▊  | 116/148 [00:43<00:11,  2.70it/s, loss=0.324]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 78%|███████▊  | 116/148 [00:44<00:11,  2.70it/s, loss=0.304]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 79%|███████▉  | 117/148 [00:44<00:11,  2.69it/s, loss=0.304]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 79%|███████▉  | 117/148 [00:44<00:11,  2.69it/s, loss=0.342]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 80%|███████▉  | 118/148 [00:44<00:11,  2.69it/s, loss=0.342]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 80%|███████▉  | 118/148 [00:44<00:11,  2.69it/s, loss=0.316]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 80%|████████  | 119/148 [00:44<00:10,  2.70it/s, loss=0.316]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 80%|████████  | 119/148 [00:45<00:10,  2.70it/s, loss=0.366]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 81%|████████  | 120/148 [00:45<00:10,  2.72it/s, loss=0.366]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 81%|████████  | 120/148 [00:45<00:10,  2.72it/s, loss=0.342]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 82%|████████▏ | 121/148 [00:45<00:09,  2.70it/s, loss=0.342]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 82%|████████▏ | 121/148 [00:45<00:09,  2.70it/s, loss=0.324]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 82%|████████▏ | 122/148 [00:45<00:09,  2.72it/s, loss=0.324]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 82%|████████▏ | 122/148 [00:46<00:09,  2.72it/s, loss=0.327]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 83%|████████▎ | 123/148 [00:46<00:09,  2.73it/s, loss=0.327]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 83%|████████▎ | 123/148 [00:46<00:09,  2.73it/s, loss=0.408]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 84%|████████▍ | 124/148 [00:46<00:08,  2.73it/s, loss=0.408]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 84%|████████▍ | 124/148 [00:46<00:08,  2.73it/s, loss=0.335]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 84%|████████▍ | 125/148 [00:46<00:08,  2.73it/s, loss=0.335]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 84%|████████▍ | 125/148 [00:47<00:08,  2.73it/s, loss=0.297]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 85%|████████▌ | 126/148 [00:47<00:08,  2.72it/s, loss=0.297]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 85%|████████▌ | 126/148 [00:47<00:08,  2.72it/s, loss=0.332]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 86%|████████▌ | 127/148 [00:47<00:07,  2.71it/s, loss=0.332]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 86%|████████▌ | 127/148 [00:48<00:07,  2.71it/s, loss=0.348]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 86%|████████▋ | 128/148 [00:48<00:07,  2.70it/s, loss=0.348]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 86%|████████▋ | 128/148 [00:48<00:07,  2.70it/s, loss=0.323]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 87%|████████▋ | 129/148 [00:48<00:07,  2.68it/s, loss=0.323]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 87%|████████▋ | 129/148 [00:48<00:07,  2.68it/s, loss=0.37] \u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 88%|████████▊ | 130/148 [00:48<00:06,  2.68it/s, loss=0.37]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 88%|████████▊ | 130/148 [00:49<00:06,  2.68it/s, loss=0.323]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 89%|████████▊ | 131/148 [00:49<00:06,  2.66it/s, loss=0.323]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 89%|████████▊ | 131/148 [00:49<00:06,  2.66it/s, loss=0.329]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 89%|████████▉ | 132/148 [00:49<00:05,  2.67it/s, loss=0.329]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 89%|████████▉ | 132/148 [00:49<00:05,  2.67it/s, loss=0.304]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 90%|████████▉ | 133/148 [00:49<00:05,  2.69it/s, loss=0.304]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 90%|████████▉ | 133/148 [00:50<00:05,  2.69it/s, loss=0.319]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 91%|█████████ | 134/148 [00:50<00:05,  2.71it/s, loss=0.319]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 91%|█████████ | 134/148 [00:50<00:05,  2.71it/s, loss=0.391]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 91%|█████████ | 135/148 [00:50<00:04,  2.71it/s, loss=0.391]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 91%|█████████ | 135/148 [00:51<00:04,  2.71it/s, loss=0.313]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 92%|█████████▏| 136/148 [00:51<00:04,  2.70it/s, loss=0.313]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 92%|█████████▏| 136/148 [00:51<00:04,  2.70it/s, loss=0.304]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 93%|█████████▎| 137/148 [00:51<00:04,  2.70it/s, loss=0.304]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 93%|█████████▎| 137/148 [00:51<00:04,  2.70it/s, loss=0.387]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 93%|█████████▎| 138/148 [00:51<00:03,  2.68it/s, loss=0.387]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 93%|█████████▎| 138/148 [00:52<00:03,  2.68it/s, loss=0.331]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 94%|█████████▍| 139/148 [00:52<00:03,  2.66it/s, loss=0.331]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 94%|█████████▍| 139/148 [00:52<00:03,  2.66it/s, loss=0.36] \u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 95%|█████████▍| 140/148 [00:52<00:02,  2.68it/s, loss=0.36]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 95%|█████████▍| 140/148 [00:52<00:02,  2.68it/s, loss=0.395]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 95%|█████████▌| 141/148 [00:52<00:02,  2.66it/s, loss=0.395]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 95%|█████████▌| 141/148 [00:53<00:02,  2.66it/s, loss=0.387]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 96%|█████████▌| 142/148 [00:53<00:02,  2.68it/s, loss=0.387]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 96%|█████████▌| 142/148 [00:53<00:02,  2.68it/s, loss=0.334]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 97%|█████████▋| 143/148 [00:53<00:01,  2.69it/s, loss=0.334]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 97%|█████████▋| 143/148 [00:54<00:01,  2.69it/s, loss=0.349]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 97%|█████████▋| 144/148 [00:54<00:01,  2.69it/s, loss=0.349]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 97%|█████████▋| 144/148 [00:54<00:01,  2.69it/s, loss=0.378]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 98%|█████████▊| 145/148 [00:54<00:01,  2.72it/s, loss=0.378]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 98%|█████████▊| 145/148 [00:54<00:01,  2.72it/s, loss=0.33] \u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 99%|█████████▊| 146/148 [00:54<00:00,  2.69it/s, loss=0.33]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 99%|█████████▊| 146/148 [00:55<00:00,  2.69it/s, loss=0.402]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 99%|█████████▉| 147/148 [00:55<00:00,  2.71it/s, loss=0.402]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 99%|█████████▉| 147/148 [00:55<00:00,  2.71it/s, loss=0.359]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "100%|██████████| 148/148 [00:55<00:00,  2.67it/s, loss=0.359]\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Got 25261590/29654016 with acc 85.19\n",
            "Dice score: 0.184907466173172\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "\n",
            "\n",
            "\n",
            "\n",
            "  0%|          | 0/148 [00:00<?, ?it/s]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  0%|          | 0/148 [00:00<?, ?it/s, loss=0.333]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  1%|          | 1/148 [00:00<01:30,  1.63it/s, loss=0.333]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  1%|          | 1/148 [00:00<01:30,  1.63it/s, loss=0.338]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  1%|▏         | 2/148 [00:00<01:18,  1.86it/s, loss=0.338]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  1%|▏         | 2/148 [00:01<01:18,  1.86it/s, loss=0.346]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  2%|▏         | 3/148 [00:01<01:10,  2.04it/s, loss=0.346]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  2%|▏         | 3/148 [00:01<01:10,  2.04it/s, loss=0.29] \u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  3%|▎         | 4/148 [00:01<01:05,  2.20it/s, loss=0.29]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  3%|▎         | 4/148 [00:02<01:05,  2.20it/s, loss=0.305]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  3%|▎         | 5/148 [00:02<01:01,  2.34it/s, loss=0.305]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  3%|▎         | 5/148 [00:02<01:01,  2.34it/s, loss=0.312]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  4%|▍         | 6/148 [00:02<00:58,  2.44it/s, loss=0.312]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  4%|▍         | 6/148 [00:02<00:58,  2.44it/s, loss=0.29] \u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  5%|▍         | 7/148 [00:02<00:56,  2.51it/s, loss=0.29]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  5%|▍         | 7/148 [00:03<00:56,  2.51it/s, loss=0.319]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  5%|▌         | 8/148 [00:03<00:54,  2.55it/s, loss=0.319]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  5%|▌         | 8/148 [00:03<00:54,  2.55it/s, loss=0.398]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  6%|▌         | 9/148 [00:03<00:53,  2.61it/s, loss=0.398]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  6%|▌         | 9/148 [00:03<00:53,  2.61it/s, loss=0.347]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  7%|▋         | 10/148 [00:03<00:52,  2.64it/s, loss=0.347]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  7%|▋         | 10/148 [00:04<00:52,  2.64it/s, loss=0.302]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  7%|▋         | 11/148 [00:04<00:51,  2.64it/s, loss=0.302]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  7%|▋         | 11/148 [00:04<00:51,  2.64it/s, loss=0.291]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  8%|▊         | 12/148 [00:04<00:51,  2.67it/s, loss=0.291]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  8%|▊         | 12/148 [00:05<00:51,  2.67it/s, loss=0.375]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  9%|▉         | 13/148 [00:05<00:50,  2.69it/s, loss=0.375]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  9%|▉         | 13/148 [00:05<00:50,  2.69it/s, loss=0.335]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  9%|▉         | 14/148 [00:05<00:49,  2.68it/s, loss=0.335]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            "  9%|▉         | 14/148 [00:05<00:49,  2.68it/s, loss=0.422]\u001b[A\u001b[A\u001b[A\u001b[A\n",
            "\n",
            "\n",
            "\n",
            " 10%|█         | 15/148 [00:05<00:49,  2.67it/s, loss=0.422]\u001b[A\u001b[A\u001b[A\u001b[A"
          ],
          "name": "stderr"
        },
        {
          "output_type": "error",
          "ename": "KeyboardInterrupt",
          "evalue": "ignored",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
            "\u001b[0;32m<ipython-input-30-263240bbee7e>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mmain\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<ipython-input-29-6142d3a539a7>\u001b[0m in \u001b[0;36mmain\u001b[0;34m()\u001b[0m\n\u001b[1;32m     91\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     92\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[0mNUM_EPOCHS\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---> 93\u001b[0;31m         \u001b[0mtrain_fn\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtrain_loader\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mmodel\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0moptimizer\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mloss_fn\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mscaler\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     94\u001b[0m         \u001b[0mcheck_accuracy\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mval_loader\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mmodel\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mdevice\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     95\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m<ipython-input-29-6142d3a539a7>\u001b[0m in \u001b[0;36mtrain_fn\u001b[0;34m(loader, model, optimizer, loss_fn, scaler)\u001b[0m\n\u001b[1;32m     30\u001b[0m         \u001b[0moptimizer\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mzero_grad\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     31\u001b[0m         \u001b[0mscaler\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mscale\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mloss\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mbackward\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---> 32\u001b[0;31m         \u001b[0mscaler\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mstep\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0moptimizer\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     33\u001b[0m         \u001b[0mscaler\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mupdate\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     34\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.7/dist-packages/torch/cuda/amp/grad_scaler.py\u001b[0m in \u001b[0;36mstep\u001b[0;34m(self, optimizer, *args, **kwargs)\u001b[0m\n\u001b[1;32m    318\u001b[0m         \u001b[0;32massert\u001b[0m \u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0moptimizer_state\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m\"found_inf_per_device\"\u001b[0m\u001b[0;34m]\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\"No inf checks were recorded for this optimizer.\"\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    319\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 320\u001b[0;31m         \u001b[0;32mif\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0msum\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mv\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mitem\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mfor\u001b[0m \u001b[0mv\u001b[0m \u001b[0;32min\u001b[0m \u001b[0moptimizer_state\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m\"found_inf_per_device\"\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mvalues\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[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    321\u001b[0m             \u001b[0mretval\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0moptimizer\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mstep\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[0m\n\u001b[1;32m    322\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.7/dist-packages/torch/cuda/amp/grad_scaler.py\u001b[0m in \u001b[0;36m<genexpr>\u001b[0;34m(.0)\u001b[0m\n\u001b[1;32m    318\u001b[0m         \u001b[0;32massert\u001b[0m \u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0moptimizer_state\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m\"found_inf_per_device\"\u001b[0m\u001b[0;34m]\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\"No inf checks were recorded for this optimizer.\"\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    319\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 320\u001b[0;31m         \u001b[0;32mif\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0msum\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mv\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mitem\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mfor\u001b[0m \u001b[0mv\u001b[0m \u001b[0;32min\u001b[0m \u001b[0moptimizer_state\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m\"found_inf_per_device\"\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mvalues\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[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    321\u001b[0m             \u001b[0mretval\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0moptimizer\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mstep\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[0m\n\u001b[1;32m    322\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
          ]
        }
      ]
    }
  ]
}