demo_nas.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "X4cRE8IbIrIV"
      },
      "source": [
        "Downloading PyTorch Vision Reference Scripts for Image Classification. These scripts are official reference implementations from PyTorch Vision that provide training and quantization utilities for image classification models."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "46CgrVgjg3E-",
        "outputId": "7fb20ebe-d7fd-43fa-dc9b-ebbedf31575e"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "--2025-05-22 07:47:03--  https://raw.githubusercontent.com/pytorch/vision/main/references/classification/presets.py\n",
            "Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 185.199.110.133, 185.199.111.133, 185.199.109.133, ...\n",
            "Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|185.199.110.133|:443... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 3885 (3.8K) [text/plain]\n",
            "Saving to: ‘presets.py.1’\n",
            "\n",
            "presets.py.1        100%[===================>]   3.79K  --.-KB/s    in 0s      \n",
            "\n",
            "2025-05-22 07:47:03 (24.3 MB/s) - ‘presets.py.1’ saved [3885/3885]\n",
            "\n",
            "--2025-05-22 07:47:04--  https://raw.githubusercontent.com/pytorch/vision/main/references/classification/sampler.py\n",
            "Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 185.199.109.133, 185.199.111.133, 185.199.110.133, ...\n",
            "Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|185.199.109.133|:443... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 2395 (2.3K) [text/plain]\n",
            "Saving to: ‘sampler.py.1’\n",
            "\n",
            "sampler.py.1        100%[===================>]   2.34K  --.-KB/s    in 0s      \n",
            "\n",
            "2025-05-22 07:47:04 (12.1 MB/s) - ‘sampler.py.1’ saved [2395/2395]\n",
            "\n",
            "--2025-05-22 07:47:04--  https://raw.githubusercontent.com/pytorch/vision/main/references/classification/train.py\n",
            "Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 185.199.110.133, 185.199.108.133, 185.199.111.133, ...\n",
            "Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|185.199.110.133|:443... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 23324 (23K) [text/plain]\n",
            "Saving to: ‘train.py.1’\n",
            "\n",
            "train.py.1          100%[===================>]  22.78K  --.-KB/s    in 0.007s  \n",
            "\n",
            "2025-05-22 07:47:04 (3.30 MB/s) - ‘train.py.1’ saved [23324/23324]\n",
            "\n",
            "--2025-05-22 07:47:04--  https://raw.githubusercontent.com/pytorch/vision/main/references/classification/train_quantization.py\n",
            "Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 185.199.108.133, 185.199.111.133, 185.199.109.133, ...\n",
            "Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|185.199.108.133|:443... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 11647 (11K) [text/plain]\n",
            "Saving to: ‘train_quantization.py.1’\n",
            "\n",
            "train_quantization. 100%[===================>]  11.37K  --.-KB/s    in 0.002s  \n",
            "\n",
            "2025-05-22 07:47:04 (6.26 MB/s) - ‘train_quantization.py.1’ saved [11647/11647]\n",
            "\n",
            "--2025-05-22 07:47:04--  https://raw.githubusercontent.com/pytorch/vision/main/references/classification/transformers.py\n",
            "Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 185.199.110.133, 185.199.111.133, 185.199.108.133, ...\n",
            "Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|185.199.110.133|:443... connected.\n",
            "HTTP request sent, awaiting response... 404 Not Found\n",
            "2025-05-22 07:47:05 ERROR 404: Not Found.\n",
            "\n",
            "--2025-05-22 07:47:05--  https://raw.githubusercontent.com/pytorch/vision/main/references/classification/utils.py\n",
            "Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 185.199.109.133, 185.199.108.133, 185.199.110.133, ...\n",
            "Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|185.199.109.133|:443... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 15791 (15K) [text/plain]\n",
            "Saving to: ‘utils.py.1’\n",
            "\n",
            "utils.py.1          100%[===================>]  15.42K  --.-KB/s    in 0.005s  \n",
            "\n",
            "2025-05-22 07:47:05 (3.21 MB/s) - ‘utils.py.1’ saved [15791/15791]\n",
            "\n"
          ]
        }
      ],
      "source": [
        "! wget https://raw.githubusercontent.com/pytorch/vision/main/references/classification/presets.py\n",
        "! wget https://raw.githubusercontent.com/pytorch/vision/main/references/classification/sampler.py\n",
        "! wget https://raw.githubusercontent.com/pytorch/vision/main/references/classification/train.py\n",
        "! wget https://raw.githubusercontent.com/pytorch/vision/main/references/classification/train_quantization.py\n",
        "! wget https://raw.githubusercontent.com/pytorch/vision/main/references/classification/transformers.py\n",
        "! wget https://raw.githubusercontent.com/pytorch/vision/main/references/classification/utils.py"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "HFASsisvIrIb"
      },
      "source": [
        "In this block, we build a “loss” function for our sequential policy gradient algorithm. When the right data is plugged in, the gradient of this loss is equal to the policy gradient."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "EaBokYCpg3FA"
      },
      "outputs": [],
      "source": [
        "import types\n",
        "from typing import List, Callable\n",
        "\n",
        "import torch\n",
        "from torch import nn, Tensor\n",
        "from torch.nn import functional as F\n",
        "from torchvision.models.resnet import BasicBlock\n",
        "\n",
        "\n",
        "def trp_criterion(trp_blocks: nn.ModuleList, shared_head: Callable, criterion: Callable, lambdas: List[float], hidden_state: Tensor, logits: Tensor, targets: Tensor, loss_normalization=False):\n",
        "    losses, rewards = criterion(logits, targets)\n",
        "    returns = torch.ones_like(rewards, dtype=torch.float32, device=rewards.device)\n",
        "    if loss_normalization:\n",
        "        coeff = torch.mean(losses).detach()\n",
        "\n",
        "    embeds = [hidden_state]\n",
        "    predictions = []\n",
        "    for k, w in enumerate(lambdas):\n",
        "        embeds.append(trp_blocks[k](embeds[-1]))\n",
        "        predictions.append(shared_head(embeds[-1]))\n",
        "        returns = returns + w * rewards\n",
        "        replica_losses, rewards = criterion(predictions[-1], targets, rewards)\n",
        "        losses = losses + replica_losses\n",
        "    loss = torch.mean(losses * returns)\n",
        "\n",
        "    if loss_normalization:\n",
        "        with torch.no_grad():\n",
        "            coeff = torch.exp(coeff) / torch.exp(loss.detach())\n",
        "        loss = coeff * loss\n",
        "\n",
        "    return loss"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "In this block, we build a TPBlock for the Task Replica Prediction (TRP) module; This implementation provides the backbone without the shared prediction head."
      ],
      "metadata": {
        "id": "_Ig0Jm2w8DPH"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "class TPBlock(nn.Module):\n",
        "    def __init__(self, depths: int, inplanes: int, planes: int):\n",
        "        super(TPBlock, self).__init__()\n",
        "\n",
        "        blocks = [BasicBlock(inplanes=inplanes, planes=planes) for _ in range(depths)]\n",
        "        self.blocks = nn.Sequential(*blocks)\n",
        "        for name, param in self.blocks.named_parameters():\n",
        "            if 'conv' in name:\n",
        "                nn.init.zeros_(param)  # Initialize weights\n",
        "            elif 'downsample' in name:\n",
        "                nn.init.zeros_(param)   # Initialize biases\n",
        "\n",
        "    def forward(self, x):\n",
        "        return self.blocks(x)"
      ],
      "metadata": {
        "id": "wkBlmJT96jZj"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "This implementation enables ResNet retraining in SPG mode.\n",
        "\n",
        "Components:\n",
        "-------------------------------------------------------------------------------\n",
        "1. gen_criterion()\n",
        "    - Purpose: compute per-sample losses and positional masks\n",
        "\n",
        "2. gen_shared_head()\n",
        "    - Purpose: Implements a shared prediction head that processes convolutional feature maps for prediction.\n",
        "\n",
        "3. gen_forward()\n",
        "    - Purpose: Extended forward pass supporting both traditional inference and SPG retraining."
      ],
      "metadata": {
        "id": "UGxQdKZaF2NT"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "class ResNetConfig:\n",
        "    @staticmethod\n",
        "    def gen_criterion(label_smoothing=0.0, top_k=1):\n",
        "        def func(input, target, mask=None):\n",
        "            \"\"\"\n",
        "            Args:\n",
        "                input (Tensor): Input tensor of shape [B, C].\n",
        "                target (Tensor): Target labels of shape [B] or [B, C].\n",
        "\n",
        "            Returns:\n",
        "                loss (Tensor): Scalar tensor representing the loss.\n",
        "                mask (Tensor): Boolean mask tensor of shape [B].\n",
        "            \"\"\"\n",
        "            label = torch.argmax(target, dim=1) if label_smoothing > 0.0 else target\n",
        "\n",
        "            unmasked_loss = F.cross_entropy(input, label, reduction=\"none\", label_smoothing=label_smoothing)\n",
        "            if mask is None:\n",
        "                mask = torch.ones_like(unmasked_loss, dtype=torch.float32, device=target.device)\n",
        "            losses = mask * unmasked_loss\n",
        "\n",
        "            with torch.no_grad():\n",
        "                topk_values, topk_indices = torch.topk(input, top_k, dim=-1)\n",
        "                mask = mask * torch.eq(topk_indices, label[:, None]).any(dim=-1).to(input.dtype)\n",
        "\n",
        "            return losses, mask\n",
        "        return func\n",
        "\n",
        "    @staticmethod\n",
        "    def gen_shared_head(self):\n",
        "        def func(x):\n",
        "            \"\"\"\n",
        "            Args:\n",
        "                x (Tensor): Hidden State tensor of shape [B, C, H, W].\n",
        "\n",
        "            Returns:\n",
        "                logits (Tensor): Logits tensor of shape [B, C].\n",
        "            \"\"\"\n",
        "            x = self.layer4(x)\n",
        "            x = self.avgpool(x)\n",
        "            x = torch.flatten(x, 1)\n",
        "            logits = self.fc(x)\n",
        "            return logits\n",
        "        return func\n",
        "\n",
        "    @staticmethod\n",
        "    def gen_forward(lambdas, loss_normalization=True, label_smoothing=0.0, top_k=1):\n",
        "        def func(self, x: Tensor, targets=None) -> Tensor:\n",
        "            x = self.conv1(x)\n",
        "            x = self.bn1(x)\n",
        "            x = self.relu(x)\n",
        "            x = self.maxpool(x)\n",
        "\n",
        "            x = self.layer1(x)\n",
        "            x = self.layer2(x)\n",
        "            hidden_state = self.layer3(x)\n",
        "            x = self.layer4(hidden_state)\n",
        "            x = self.avgpool(x)\n",
        "            x = torch.flatten(x, 1)\n",
        "            logits = self.fc(x)\n",
        "\n",
        "            if self.training:\n",
        "                shared_head = ResNetConfig.gen_shared_head(self)\n",
        "                criterion = ResNetConfig.gen_criterion(label_smoothing=label_smoothing, top_k=top_k)\n",
        "\n",
        "                loss = trp_criterion(self.trp_blocks, shared_head, criterion, lambdas, hidden_state, logits, targets, loss_normalization=loss_normalization)\n",
        "\n",
        "                return logits, loss\n",
        "\n",
        "            return logits\n",
        "\n",
        "        return func"
      ],
      "metadata": {
        "id": "kTZWkoLr8cfE"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "Applies TRP modules to the base ResNet (main backbone). The k-th TRP module corresponding to a deeper ResNet variant with an additional depth of 3 * sum(depths[:k+1])."
      ],
      "metadata": {
        "id": "cCn6vwItH1CW"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "def apply_trp(model, depths: List[int], planes: int, lambdas: List[float], **kwargs):\n",
        "    print(\"✅ Applying TRP to ResNet for Image Classification...\")\n",
        "    model.trp_blocks = torch.nn.ModuleList([TPBlock(depths=d, inplanes=planes, planes=planes) for d in depths])\n",
        "    model.forward = types.MethodType(ResNetConfig.gen_forward(lambdas), model)\n",
        "    return model"
      ],
      "metadata": {
        "id": "wXQF0oISH5Yp"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "The following is a training script for classification models, primarily based on the official TorchVision `train.py` reference implementation. We have made two modifications:\n",
        "\n",
        "Adding TRP Modules: We integrate TRP modules into the base model architecture before training begins:\n",
        "\n",
        "```python\n",
        "if args.apply_trp:\n",
        "    model = apply_trp(model, args.trp_depths,  args.trp_planes, args.trp_lambdas)\n",
        "```\n",
        "Removing TRP Modules: We remove the TRP components from the base model before saving the base model:\n",
        "```python\n",
        "if args.output_dir:\n",
        "    checkpoint = {\n",
        "        \"model\": model.state_dict() if not args.apply_trp else {k: v for k, v in model.state_dict().items() if not k.startswith(\"trp_blocks\")},\n",
        "        \"optimizer\": optimizer.state_dict(),\n",
        "        \"lr_scheduler\": lr_scheduler.state_dict(),\n",
        "        \"epoch\": epoch,\n",
        "        \"args\": args,\n",
        "    }\n",
        "    utils.save_on_master(checkpoint, os.path.join(args.output_dir, f\"model_{epoch}.pth\"))\n",
        "    utils.save_on_master(checkpoint, os.path.join(args.output_dir, \"checkpoint.pth\"))\n",
        "```"
      ],
      "metadata": {
        "id": "kDjSAv3PJr7P"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "hK4Y7Sqv4xUa"
      },
      "outputs": [],
      "source": [
        "import datetime\n",
        "import os\n",
        "import time\n",
        "import warnings\n",
        "\n",
        "import presets\n",
        "import torch\n",
        "import torch.utils.data\n",
        "import torchvision\n",
        "import utils\n",
        "from torch import nn\n",
        "from torchvision.transforms.functional import InterpolationMode\n",
        "\n",
        "\n",
        "def load_data(traindir, valdir):\n",
        "    # Data loading code\n",
        "    print(\"Loading data\")\n",
        "    interpolation = InterpolationMode(\"bilinear\")\n",
        "\n",
        "    print(\"Loading training data\")\n",
        "    st = time.time()\n",
        "    dataset = torchvision.datasets.ImageFolder(\n",
        "        traindir,\n",
        "        presets.ClassificationPresetTrain(crop_size=224, interpolation=interpolation, auto_augment_policy=None, random_erase_prob=0.0, ra_magnitude=9, augmix_severity=3),\n",
        "    )\n",
        "    print(\"Took\", time.time() - st)\n",
        "\n",
        "    print(\"Loading validation data\")\n",
        "    dataset_test = torchvision.datasets.ImageFolder(\n",
        "        valdir,\n",
        "        presets.ClassificationPresetEval(crop_size=224, resize_size=256, interpolation=interpolation)\n",
        "    )\n",
        "\n",
        "    print(\"Creating data loaders\")\n",
        "    train_sampler = torch.utils.data.RandomSampler(dataset)\n",
        "    test_sampler = torch.utils.data.SequentialSampler(dataset_test)\n",
        "\n",
        "    return dataset, dataset_test, train_sampler, test_sampler\n",
        "\n",
        "\n",
        "\n",
        "def train_one_epoch(model, optimizer, data_loader, device, epoch, args):\n",
        "    model.train()\n",
        "    metric_logger = utils.MetricLogger(delimiter=\"  \")\n",
        "    metric_logger.add_meter(\"lr\", utils.SmoothedValue(window_size=1, fmt=\"{value}\"))\n",
        "    metric_logger.add_meter(\"img/s\", utils.SmoothedValue(window_size=10, fmt=\"{value}\"))\n",
        "\n",
        "    header = f\"Epoch: [{epoch}]\"\n",
        "    for i, (image, target) in enumerate(metric_logger.log_every(data_loader, args.print_freq, header)):\n",
        "        start_time = time.time()\n",
        "        image, target = image.to(device), target.to(device)\n",
        "        with torch.amp.autocast(\"cuda\", enabled=False):\n",
        "            output, loss = model(image, target)\n",
        "\n",
        "        optimizer.zero_grad()\n",
        "        loss.backward()\n",
        "        optimizer.step()\n",
        "\n",
        "        acc1, acc5 = utils.accuracy(output, target, topk=(1, 5))\n",
        "        batch_size = image.shape[0]\n",
        "        metric_logger.update(loss=loss.item(), lr=optimizer.param_groups[0][\"lr\"])\n",
        "        metric_logger.meters[\"acc1\"].update(acc1.item(), n=batch_size)\n",
        "        metric_logger.meters[\"acc5\"].update(acc5.item(), n=batch_size)\n",
        "        metric_logger.meters[\"img/s\"].update(batch_size / (time.time() - start_time))\n",
        "\n",
        "\n",
        "def evaluate(model, criterion, data_loader, device, print_freq=100, log_suffix=\"\"):\n",
        "    model.eval()\n",
        "    metric_logger = utils.MetricLogger(delimiter=\"  \")\n",
        "    header = f\"Test: {log_suffix}\"\n",
        "\n",
        "    num_processed_samples = 0\n",
        "    with torch.inference_mode():\n",
        "        for image, target in metric_logger.log_every(data_loader, print_freq, header):\n",
        "            image = image.to(device, non_blocking=True)\n",
        "            target = target.to(device, non_blocking=True)\n",
        "            output = model(image)\n",
        "            loss = criterion(output, target)\n",
        "\n",
        "            acc1, acc5 = utils.accuracy(output, target, topk=(1, 5))\n",
        "            # FIXME need to take into account that the datasets\n",
        "            # could have been padded in distributed setup\n",
        "            batch_size = image.shape[0]\n",
        "            metric_logger.update(loss=loss.item())\n",
        "            metric_logger.meters[\"acc1\"].update(acc1.item(), n=batch_size)\n",
        "            metric_logger.meters[\"acc5\"].update(acc5.item(), n=batch_size)\n",
        "            num_processed_samples += batch_size\n",
        "    # gather the stats from all processes\n",
        "\n",
        "    num_processed_samples = utils.reduce_across_processes(num_processed_samples)\n",
        "    if (\n",
        "        hasattr(data_loader.dataset, \"__len__\")\n",
        "        and len(data_loader.dataset) != num_processed_samples\n",
        "        and torch.distributed.get_rank() == 0\n",
        "    ):\n",
        "        # See FIXME above\n",
        "        warnings.warn(\n",
        "            f\"It looks like the dataset has {len(data_loader.dataset)} samples, but {num_processed_samples} \"\n",
        "            \"samples were used for the validation, which might bias the results. \"\n",
        "            \"Try adjusting the batch size and / or the world size. \"\n",
        "            \"Setting the world size to 1 is always a safe bet.\"\n",
        "        )\n",
        "\n",
        "    metric_logger.synchronize_between_processes()\n",
        "\n",
        "    print(f\"{header} Acc@1 {metric_logger.acc1.global_avg:.3f} Acc@5 {metric_logger.acc5.global_avg:.3f}\")\n",
        "    return metric_logger.acc1.global_avg\n",
        "\n",
        "\n",
        "def main(args):\n",
        "    if args.output_dir:\n",
        "        utils.mkdir(args.output_dir)\n",
        "    print(args)\n",
        "\n",
        "    device = torch.device(args.device)\n",
        "\n",
        "    if args.use_deterministic_algorithms:\n",
        "        torch.backends.cudnn.benchmark = False\n",
        "        torch.use_deterministic_algorithms(True)\n",
        "    else:\n",
        "        torch.backends.cudnn.benchmark = True\n",
        "\n",
        "    train_dir = os.path.join(args.data_path, \"train\")\n",
        "    val_dir = os.path.join(args.data_path, \"val\")\n",
        "    dataset, dataset_test, train_sampler, test_sampler = load_data(train_dir, val_dir)\n",
        "\n",
        "    num_classes = len(dataset.classes)\n",
        "    data_loader = torch.utils.data.DataLoader(dataset, batch_size=args.batch_size, sampler=train_sampler, num_workers=16, pin_memory=True, collate_fn=None)\n",
        "    data_loader_test = torch.utils.data.DataLoader(dataset_test, batch_size=64, sampler=test_sampler, num_workers=16, pin_memory=True)\n",
        "\n",
        "    print(\"Creating model\")\n",
        "    model = torchvision.models.get_model(args.model, weights=args.weights, num_classes=num_classes)\n",
        "    if args.apply_trp:\n",
        "        model = apply_trp(model, args.trp_depths,  args.trp_planes, args.trp_lambdas)\n",
        "    model.to(device)\n",
        "\n",
        "    parameters = utils.set_weight_decay(model, args.weight_decay, norm_weight_decay=None, custom_keys_weight_decay=None)\n",
        "    optimizer = torch.optim.SGD(parameters, lr=args.lr, momentum=args.momentum, weight_decay=args.weight_decay, nesterov=False)\n",
        "\n",
        "    main_lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=args.lr_step_size, gamma=args.lr_gamma)\n",
        "    warmup_lr_scheduler = torch.optim.lr_scheduler.ConstantLR(optimizer, factor=args.lr_warmup_decay, total_iters=args.lr_warmup_epochs)\n",
        "    lr_scheduler = torch.optim.lr_scheduler.SequentialLR(optimizer, schedulers=[warmup_lr_scheduler, main_lr_scheduler], milestones=[args.lr_warmup_epochs])\n",
        "\n",
        "\n",
        "    print(\"Start training\")\n",
        "    start_time = time.time()\n",
        "    for epoch in range(args.epochs):\n",
        "        train_one_epoch(model, optimizer, data_loader, device, epoch, args)\n",
        "        lr_scheduler.step()\n",
        "        evaluate(model, nn.CrossEntropyLoss(), data_loader_test, device=device)\n",
        "        if args.output_dir:\n",
        "            checkpoint = {\n",
        "                \"model\": model.state_dict() if not args.apply_trp else {k: v for k, v in model.state_dict().items() if not k.startswith(\"trp_blocks\")},  # NOTE: remove TRP heads\n",
        "                \"optimizer\": optimizer.state_dict(),\n",
        "                \"lr_scheduler\": lr_scheduler.state_dict(),\n",
        "                \"epoch\": epoch,\n",
        "                \"args\": args,\n",
        "            }\n",
        "            utils.save_on_master(checkpoint, os.path.join(args.output_dir, f\"model_{epoch}.pth\"))\n",
        "            utils.save_on_master(checkpoint, os.path.join(args.output_dir, \"checkpoint.pth\"))\n",
        "\n",
        "    total_time = time.time() - start_time\n",
        "    total_time_str = str(datetime.timedelta(seconds=int(total_time)))\n",
        "    print(f\"Training time {total_time_str}\")\n"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Prepare the [ImageNet](http://image-net.org/) dataset manually and place it in `/path/to/imagenet`. For image classification examples, pass the argument `--data-path=/path/to/imagenet` to the training script. The extracted dataset directory should follow this structure:\n",
        "```setup\n",
        "/path/to/imagenet/:\n",
        "    train/:\n",
        "        n01440764:\n",
        "            n01440764_18.JPEG ...\n",
        "        n01443537:\n",
        "            n01443537_2.JPEG ...\n",
        "    val/:\n",
        "        n01440764:\n",
        "            ILSVRC2012_val_00000293.JPEG ...\n",
        "        n01443537:\n",
        "            ILSVRC2012_val_00000236.JPEG ...\n",
        "```\n",
        "\n",
        "Now you can apply the SPG algorithm in model retraining.\n",
        "\n",
        "**Implementation Note:**\n",
        "\n",
        "- This demonstration runs on Google Colab using a single GPU configuration\n",
        "- Performance Improvement: Enhances ResNet18 validation accuracy (ACC@1) from XX.X% to YY.Y%\n",
        "- For optimal results:\n",
        "  - Refer to our README.md for complete setup instructions\n",
        "  - Recommended hardware: 4× RTX A6000 GPUs"
      ],
      "metadata": {
        "id": "SV8s5k49KwgS"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "UDZxDNfT4xUb",
        "outputId": "bcf86aa0-eb77-4815-e0fa-05997f1e1f1b"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "namespace(data_path='/home/cs/Documents/datasets/imagenet', model='resnet18', device='cuda', batch_size=256, epochs=10, opt='sgd', lr=0.0004, momentum=0.9, weight_decay=0.0001, lr_warmup_epochs=1, lr_warmup_decay=0.0, lr_step_size=2, lr_gamma=0.5, print_freq=100, output_dir='resnet18', use_deterministic_algorithms=False, weights='ResNet18_Weights.IMAGENET1K_V1', apply_trp=True, trp_depths=[1, 1, 1], trp_planes=256, trp_lambdas=[0.4, 0.2, 0.1])\n",
            "Loading data\n",
            "Loading training data\n",
            "Took 2.6400649547576904\n",
            "Loading validation data\n",
            "Creating data loaders\n",
            "Creating model\n",
            "✅ Applying TRP to ResNet for Image Classification...\n",
            "Start training\n",
            "Epoch: [0]  [   0/5005]  eta: 5:44:15  lr: 0.0  img/s: 492.1456954856547  loss: 0.7194 (0.7194)  acc1: 69.1406 (69.1406)  acc5: 86.3281 (86.3281)  meanQV: 1.4840 (1.4840)  stdQV: 0.3240 (0.3240)  time: 4.1269  data: 3.6067  max mem: 8962\n",
            "Epoch: [0]  [ 100/5005]  eta: 0:46:06  lr: 0.0  img/s: 487.81080383655046  loss: 0.7315 (0.7366)  acc1: 68.7500 (68.9434)  acc5: 87.1094 (87.2022)  meanQV: 1.4813 (1.4826)  stdQV: 0.3240 (0.3238)  time: 0.5212  data: 0.0004  max mem: 8962\n",
            "Epoch: [0]  [ 200/5005]  eta: 0:43:26  lr: 0.0  img/s: 494.6728449606332  loss: 0.7197 (0.7298)  acc1: 68.7500 (69.0318)  acc5: 87.5000 (87.2785)  meanQV: 1.4813 (1.4832)  stdQV: 0.3251 (0.3236)  time: 0.5211  data: 0.0004  max mem: 8962\n",
            "Epoch: [0]  [ 300/5005]  eta: 0:41:58  lr: 0.0  img/s: 487.94536242744385  loss: 0.7055 (0.7270)  acc1: 70.7031 (69.1757)  acc5: 86.7188 (87.3053)  meanQV: 1.4949 (1.4842)  stdQV: 0.3192 (0.3232)  time: 0.5198  data: 0.0004  max mem: 8962\n",
            "Epoch: [0]  [ 400/5005]  eta: 0:40:48  lr: 0.0  img/s: 493.8419675273725  loss: 0.7652 (0.7331)  acc1: 67.9688 (68.9867)  acc5: 87.1094 (87.2409)  meanQV: 1.4758 (1.4829)  stdQV: 0.3262 (0.3237)  time: 0.5219  data: 0.0004  max mem: 8962\n",
            "Epoch: [0]  [ 500/5005]  eta: 0:39:45  lr: 0.0  img/s: 490.99324013193245  loss: 0.7112 (0.7341)  acc1: 69.1406 (68.9067)  acc5: 87.1094 (87.2653)  meanQV: 1.4840 (1.4823)  stdQV: 0.3228 (0.3239)  time: 0.5201  data: 0.0003  max mem: 8962\n",
            "Epoch: [0]  [ 600/5005]  eta: 0:38:46  lr: 0.0  img/s: 493.63241471415074  loss: 0.7323 (0.7309)  acc1: 69.5312 (69.0171)  acc5: 87.8906 (87.3297)  meanQV: 1.4867 (1.4831)  stdQV: 0.3216 (0.3236)  time: 0.5214  data: 0.0004  max mem: 8962\n",
            "Epoch: [0]  [ 700/5005]  eta: 0:37:49  lr: 0.0  img/s: 487.50804150353844  loss: 0.7489 (0.7310)  acc1: 68.7500 (69.0442)  acc5: 86.3281 (87.3167)  meanQV: 1.4813 (1.4833)  stdQV: 0.3240 (0.3235)  time: 0.5206  data: 0.0004  max mem: 8962\n",
            "Epoch: [0]  [ 800/5005]  eta: 0:36:53  lr: 0.0  img/s: 491.46023226850616  loss: 0.7270 (0.7316)  acc1: 69.1406 (69.0567)  acc5: 87.8906 (87.3166)  meanQV: 1.4840 (1.4834)  stdQV: 0.3216 (0.3235)  time: 0.5220  data: 0.0004  max mem: 8962\n",
            "Epoch: [0]  [ 900/5005]  eta: 0:35:58  lr: 0.0  img/s: 490.80023055787785  loss: 0.7573 (0.7338)  acc1: 68.3594 (68.9898)  acc5: 86.7188 (87.3010)  meanQV: 1.4785 (1.4829)  stdQV: 0.3262 (0.3237)  time: 0.5208  data: 0.0004  max mem: 8962\n",
            "Epoch: [0]  [1000/5005]  eta: 0:35:03  lr: 0.0  img/s: 493.60609129495185  loss: 0.7472 (0.7342)  acc1: 68.3594 (69.0087)  acc5: 87.1094 (87.2924)  meanQV: 1.4785 (1.4831)  stdQV: 0.3262 (0.3237)  time: 0.5212  data: 0.0004  max mem: 8962\n",
            "Epoch: [0]  [1100/5005]  eta: 0:34:09  lr: 0.0  img/s: 493.3883500186788  loss: 0.7273 (0.7337)  acc1: 68.3594 (69.0157)  acc5: 86.3281 (87.2832)  meanQV: 1.4785 (1.4831)  stdQV: 0.3251 (0.3236)  time: 0.5200  data: 0.0003  max mem: 8962\n",
            "Epoch: [0]  [1200/5005]  eta: 0:33:15  lr: 0.0  img/s: 488.9129516588334  loss: 0.7296 (0.7333)  acc1: 67.9688 (69.0434)  acc5: 87.1094 (87.2827)  meanQV: 1.4758 (1.4833)  stdQV: 0.3251 (0.3235)  time: 0.5207  data: 0.0003  max mem: 8962\n",
            "Epoch: [0]  [1300/5005]  eta: 0:32:22  lr: 0.0  img/s: 496.11964440830207  loss: 0.7214 (0.7334)  acc1: 69.1406 (69.0587)  acc5: 86.7188 (87.2835)  meanQV: 1.4840 (1.4834)  stdQV: 0.3240 (0.3235)  time: 0.5209  data: 0.0004  max mem: 8962\n",
            "Epoch: [0]  [1400/5005]  eta: 0:31:29  lr: 0.0  img/s: 489.82739879256343  loss: 0.7335 (0.7330)  acc1: 68.3594 (69.0832)  acc5: 86.7188 (87.2881)  meanQV: 1.4785 (1.4836)  stdQV: 0.3240 (0.3234)  time: 0.5208  data: 0.0004  max mem: 8962\n",
            "Epoch: [0]  [1500/5005]  eta: 0:30:35  lr: 0.0  img/s: 488.2193641437147  loss: 0.7221 (0.7324)  acc1: 69.1406 (69.0979)  acc5: 87.8906 (87.3184)  meanQV: 1.4840 (1.4837)  stdQV: 0.3228 (0.3234)  time: 0.5221  data: 0.0004  max mem: 8962\n",
            "Epoch: [0]  [1600/5005]  eta: 0:29:43  lr: 0.0  img/s: 491.9973130670899  loss: 0.7626 (0.7324)  acc1: 68.3594 (69.0891)  acc5: 86.7188 (87.3175)  meanQV: 1.4785 (1.4836)  stdQV: 0.3251 (0.3234)  time: 0.5204  data: 0.0004  max mem: 8962\n",
            "Epoch: [0]  [1700/5005]  eta: 0:28:50  lr: 0.0  img/s: 491.90219695845224  loss: 0.7566 (0.7321)  acc1: 70.3125 (69.0922)  acc5: 87.1094 (87.3330)  meanQV: 1.4922 (1.4836)  stdQV: 0.3192 (0.3234)  time: 0.5211  data: 0.0003  max mem: 8962\n",
            "Epoch: [0]  [1800/5005]  eta: 0:27:57  lr: 0.0  img/s: 492.04353203565745  loss: 0.7514 (0.7323)  acc1: 67.9688 (69.0897)  acc5: 86.3281 (87.3206)  meanQV: 1.4758 (1.4836)  stdQV: 0.3262 (0.3234)  time: 0.5207  data: 0.0004  max mem: 8962\n",
            "Epoch: [0]  [1900/5005]  eta: 0:27:04  lr: 0.0  img/s: 489.4212747065615  loss: 0.6998 (0.7323)  acc1: 68.7500 (69.0662)  acc5: 88.2812 (87.3253)  meanQV: 1.4813 (1.4835)  stdQV: 0.3240 (0.3235)  time: 0.5216  data: 0.0004  max mem: 8962\n",
            "Epoch: [0]  [2000/5005]  eta: 0:26:11  lr: 0.0  img/s: 495.7586395402095  loss: 0.7553 (0.7335)  acc1: 67.9688 (69.0319)  acc5: 87.5000 (87.2974)  meanQV: 1.4758 (1.4832)  stdQV: 0.3273 (0.3236)  time: 0.5215  data: 0.0003  max mem: 8962\n",
            "Epoch: [0]  [2100/5005]  eta: 0:25:19  lr: 0.0  img/s: 489.2644978319551  loss: 0.7095 (0.7331)  acc1: 68.7500 (69.0454)  acc5: 87.5000 (87.3048)  meanQV: 1.4813 (1.4833)  stdQV: 0.3240 (0.3235)  time: 0.5210  data: 0.0004  max mem: 8962\n",
            "Epoch: [0]  [2200/5005]  eta: 0:24:26  lr: 0.0  img/s: 496.86875392465674  loss: 0.7402 (0.7334)  acc1: 70.3125 (69.0735)  acc5: 86.7188 (87.3154)  meanQV: 1.4922 (1.4835)  stdQV: 0.3204 (0.3234)  time: 0.5210  data: 0.0004  max mem: 8962\n",
            "Epoch: [0]  [2300/5005]  eta: 0:23:34  lr: 0.0  img/s: 494.81235650908434  loss: 0.7301 (0.7332)  acc1: 68.7500 (69.0787)  acc5: 87.1094 (87.3180)  meanQV: 1.4813 (1.4836)  stdQV: 0.3251 (0.3234)  time: 0.5205  data: 0.0003  max mem: 8962\n",
            "Epoch: [0]  [2400/5005]  eta: 0:22:41  lr: 0.0  img/s: 492.8439314901557  loss: 0.7250 (0.7332)  acc1: 69.5312 (69.0799)  acc5: 86.7188 (87.3132)  meanQV: 1.4867 (1.4836)  stdQV: 0.3228 (0.3234)  time: 0.5205  data: 0.0003  max mem: 8962\n",
            "Epoch: [0]  [2500/5005]  eta: 0:21:49  lr: 0.0  img/s: 496.70487248638824  loss: 0.7133 (0.7329)  acc1: 67.9688 (69.0827)  acc5: 86.3281 (87.3009)  meanQV: 1.4758 (1.4836)  stdQV: 0.3262 (0.3234)  time: 0.5208  data: 0.0003  max mem: 8962\n",
            "Epoch: [0]  [2600/5005]  eta: 0:20:56  lr: 0.0  img/s: 491.3833130219057  loss: 0.7494 (0.7331)  acc1: 70.3125 (69.0795)  acc5: 86.7188 (87.2938)  meanQV: 1.4922 (1.4836)  stdQV: 0.3204 (0.3234)  time: 0.5203  data: 0.0003  max mem: 8962\n",
            "Epoch: [0]  [2700/5005]  eta: 0:20:04  lr: 0.0  img/s: 488.37014076537906  loss: 0.7300 (0.7330)  acc1: 68.7500 (69.0900)  acc5: 87.1094 (87.2955)  meanQV: 1.4813 (1.4836)  stdQV: 0.3240 (0.3234)  time: 0.5204  data: 0.0004  max mem: 8962\n",
            "Epoch: [0]  [2800/5005]  eta: 0:19:12  lr: 0.0  img/s: 491.6548870066586  loss: 0.7110 (0.7331)  acc1: 69.5312 (69.0993)  acc5: 88.2812 (87.3015)  meanQV: 1.4867 (1.4837)  stdQV: 0.3228 (0.3234)  time: 0.5216  data: 0.0005  max mem: 8962\n",
            "Epoch: [0]  [2900/5005]  eta: 0:18:19  lr: 0.0  img/s: 490.05632164917586  loss: 0.7100 (0.7334)  acc1: 69.5312 (69.0961)  acc5: 87.5000 (87.3092)  meanQV: 1.4867 (1.4837)  stdQV: 0.3228 (0.3234)  time: 0.5212  data: 0.0003  max mem: 8962\n",
            "Epoch: [0]  [3000/5005]  eta: 0:17:27  lr: 0.0  img/s: 491.02018376984734  loss: 0.7217 (0.7338)  acc1: 69.5312 (69.0922)  acc5: 88.2812 (87.3148)  meanQV: 1.4867 (1.4836)  stdQV: 0.3228 (0.3234)  time: 0.5212  data: 0.0003  max mem: 8962\n",
            "Epoch: [0]  [3100/5005]  eta: 0:16:35  lr: 0.0  img/s: 488.71979426958876  loss: 0.6803 (0.7335)  acc1: 68.3594 (69.0926)  acc5: 87.1094 (87.3073)  meanQV: 1.4785 (1.4836)  stdQV: 0.3251 (0.3234)  time: 0.5211  data: 0.0003  max mem: 8962\n",
            "Epoch: [0]  [3200/5005]  eta: 0:15:42  lr: 0.0  img/s: 492.2176638610237  loss: 0.7217 (0.7332)  acc1: 70.7031 (69.1058)  acc5: 87.5000 (87.3127)  meanQV: 1.4949 (1.4837)  stdQV: 0.3192 (0.3233)  time: 0.5208  data: 0.0004  max mem: 8962\n",
            "Epoch: [0]  [3300/5005]  eta: 0:14:50  lr: 0.0  img/s: 491.9402840653201  loss: 0.7205 (0.7333)  acc1: 69.1406 (69.1047)  acc5: 87.1094 (87.3212)  meanQV: 1.4840 (1.4837)  stdQV: 0.3216 (0.3233)  time: 0.5203  data: 0.0003  max mem: 8962\n",
            "Epoch: [0]  [3400/5005]  eta: 0:13:58  lr: 0.0  img/s: 487.95378866338956  loss: 0.7223 (0.7333)  acc1: 69.5312 (69.0997)  acc5: 86.7188 (87.3139)  meanQV: 1.4867 (1.4837)  stdQV: 0.3216 (0.3234)  time: 0.5214  data: 0.0004  max mem: 8962\n",
            "Epoch: [0]  [3500/5005]  eta: 0:13:05  lr: 0.0  img/s: 1043.7659108075188  loss: 0.7439 (0.7335)  acc1: 69.1406 (69.0932)  acc5: 86.3281 (87.3065)  meanQV: 1.4840 (1.4837)  stdQV: 0.3216 (0.3234)  time: 0.4635  data: 0.0003  max mem: 8962\n",
            "Epoch: [0]  [3600/5005]  eta: 0:12:07  lr: 0.0  img/s: 1048.0888024707265  loss: 0.7388 (0.7336)  acc1: 69.1406 (69.0959)  acc5: 86.7188 (87.3047)  meanQV: 1.4840 (1.4837)  stdQV: 0.3228 (0.3234)  time: 0.2795  data: 0.0011  max mem: 8962\n",
            "Epoch: [0]  [3700/5005]  eta: 0:11:15  lr: 0.0  img/s: 491.45843271282087  loss: 0.7387 (0.7335)  acc1: 69.5312 (69.0927)  acc5: 86.7188 (87.3025)  meanQV: 1.4867 (1.4836)  stdQV: 0.3228 (0.3234)  time: 0.5209  data: 0.0004  max mem: 8962\n",
            "Epoch: [0]  [3800/5005]  eta: 0:10:23  lr: 0.0  img/s: 496.6446795362795  loss: 0.7232 (0.7336)  acc1: 69.9219 (69.0989)  acc5: 87.8906 (87.3037)  meanQV: 1.4895 (1.4837)  stdQV: 0.3204 (0.3234)  time: 0.5210  data: 0.0004  max mem: 8962\n",
            "Epoch: [0]  [3900/5005]  eta: 0:09:32  lr: 0.0  img/s: 492.508007014193  loss: 0.7213 (0.7336)  acc1: 67.9688 (69.0933)  acc5: 87.1094 (87.2962)  meanQV: 1.4758 (1.4837)  stdQV: 0.3273 (0.3234)  time: 0.5213  data: 0.0003  max mem: 8962\n",
            "Epoch: [0]  [4000/5005]  eta: 0:08:40  lr: 0.0  img/s: 488.4174580719702  loss: 0.7132 (0.7336)  acc1: 68.7500 (69.0951)  acc5: 87.5000 (87.2945)  meanQV: 1.4813 (1.4837)  stdQV: 0.3240 (0.3234)  time: 0.5211  data: 0.0003  max mem: 8962\n",
            "Epoch: [0]  [4100/5005]  eta: 0:07:48  lr: 0.0  img/s: 495.24666769675684  loss: 0.7654 (0.7338)  acc1: 68.3594 (69.0866)  acc5: 87.1094 (87.2870)  meanQV: 1.4785 (1.4836)  stdQV: 0.3251 (0.3234)  time: 0.5212  data: 0.0004  max mem: 8962\n",
            "Epoch: [0]  [4200/5005]  eta: 0:06:56  lr: 0.0  img/s: 490.48949008347176  loss: 0.7638 (0.7341)  acc1: 66.7969 (69.0754)  acc5: 86.7188 (87.2836)  meanQV: 1.4676 (1.4835)  stdQV: 0.3262 (0.3234)  time: 0.5206  data: 0.0003  max mem: 8962\n",
            "Epoch: [0]  [4300/5005]  eta: 0:06:05  lr: 0.0  img/s: 487.7678139386137  loss: 0.7096 (0.7339)  acc1: 69.1406 (69.0847)  acc5: 87.1094 (87.2826)  meanQV: 1.4840 (1.4836)  stdQV: 0.3240 (0.3234)  time: 0.5218  data: 0.0004  max mem: 8962\n",
            "Epoch: [0]  [4400/5005]  eta: 0:05:13  lr: 0.0  img/s: 493.6394498964211  loss: 0.7463 (0.7340)  acc1: 68.3594 (69.0806)  acc5: 87.8906 (87.2854)  meanQV: 1.4785 (1.4836)  stdQV: 0.3262 (0.3234)  time: 0.5201  data: 0.0004  max mem: 8962\n",
            "Epoch: [0]  [4500/5005]  eta: 0:04:21  lr: 0.0  img/s: 485.9301278068194  loss: 0.7351 (0.7337)  acc1: 69.1406 (69.0835)  acc5: 87.1094 (87.2907)  meanQV: 1.4840 (1.4836)  stdQV: 0.3240 (0.3234)  time: 0.5212  data: 0.0003  max mem: 8962\n",
            "Epoch: [0]  [4600/5005]  eta: 0:03:29  lr: 0.0  img/s: 488.56480022495765  loss: 0.7084 (0.7340)  acc1: 68.3594 (69.0736)  acc5: 86.7188 (87.2877)  meanQV: 1.4785 (1.4835)  stdQV: 0.3262 (0.3234)  time: 0.5211  data: 0.0003  max mem: 8962\n",
            "Epoch: [0]  [4700/5005]  eta: 0:02:38  lr: 0.0  img/s: 494.1863106933297  loss: 0.7403 (0.7337)  acc1: 68.7500 (69.0823)  acc5: 87.1094 (87.2872)  meanQV: 1.4813 (1.4836)  stdQV: 0.3251 (0.3234)  time: 0.5210  data: 0.0004  max mem: 8962\n",
            "Epoch: [0]  [4800/5005]  eta: 0:01:46  lr: 0.0  img/s: 491.70689603851804  loss: 0.7220 (0.7338)  acc1: 71.0938 (69.0803)  acc5: 86.7188 (87.2814)  meanQV: 1.4977 (1.4836)  stdQV: 0.3167 (0.3234)  time: 0.5212  data: 0.0003  max mem: 8962\n",
            "Epoch: [0]  [4900/5005]  eta: 0:00:54  lr: 0.0  img/s: 493.2093111277595  loss: 0.7027 (0.7336)  acc1: 69.5312 (69.0827)  acc5: 86.3281 (87.2757)  meanQV: 1.4867 (1.4836)  stdQV: 0.3228 (0.3234)  time: 0.5207  data: 0.0004  max mem: 8962\n",
            "Epoch: [0]  [5000/5005]  eta: 0:00:02  lr: 0.0  img/s: 488.95235848734404  loss: 0.7240 (0.7339)  acc1: 69.5312 (69.0705)  acc5: 86.7188 (87.2706)  meanQV: 1.4867 (1.4835)  stdQV: 0.3216 (0.3234)  time: 0.5214  data: 0.0002  max mem: 8962\n",
            "Epoch: [0] Total time: 0:43:17\n"
          ]
        },
        {
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "/home/cs/anaconda3/envs/csenv/lib/python3.9/site-packages/torch/optim/lr_scheduler.py:243: UserWarning: The epoch parameter in `scheduler.step()` was not necessary and is being deprecated where possible. Please use `scheduler.step()` to step the scheduler. During the deprecation, if epoch is different from None, the closed form is used instead of the new chainable form, where available. Please open an issue if you are unable to replicate your use case: https://github.com/pytorch/pytorch/issues/new/choose.\n",
            "  warnings.warn(EPOCH_DEPRECATION_WARNING, UserWarning)\n"
          ]
        },
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Test:   [   0/6250]  eta: 2:26:50  loss: 0.8858 (0.8858)  acc1: 75.0000 (75.0000)  acc5: 100.0000 (100.0000)  time: 1.4097  data: 0.7689  max mem: 8962\n",
            "Test:   [ 100/6250]  eta: 0:02:03  loss: 0.2054 (0.6172)  acc1: 100.0000 (83.7871)  acc5: 100.0000 (95.6683)  time: 0.0066  data: 0.0006  max mem: 8962\n",
            "Test:   [ 200/6250]  eta: 0:01:22  loss: 0.6502 (0.6427)  acc1: 87.5000 (84.3284)  acc5: 87.5000 (95.0249)  time: 0.0064  data: 0.0007  max mem: 8962\n",
            "Test:   [ 300/6250]  eta: 0:01:08  loss: 1.1768 (0.8467)  acc1: 62.5000 (78.3638)  acc5: 87.5000 (93.2724)  time: 0.0075  data: 0.0010  max mem: 8962\n",
            "Test:   [ 400/6250]  eta: 0:00:59  loss: 1.1007 (0.9660)  acc1: 75.0000 (75.4676)  acc5: 87.5000 (92.2070)  time: 0.0074  data: 0.0017  max mem: 8962\n",
            "Test:   [ 500/6250]  eta: 0:00:55  loss: 1.1372 (1.0088)  acc1: 75.0000 (74.0519)  acc5: 87.5000 (91.8663)  time: 0.0080  data: 0.0021  max mem: 8962\n",
            "Test:   [ 600/6250]  eta: 0:00:51  loss: 0.5046 (0.9217)  acc1: 87.5000 (76.2479)  acc5: 100.0000 (92.6789)  time: 0.0071  data: 0.0009  max mem: 8962\n",
            "Test:   [ 700/6250]  eta: 0:00:49  loss: 0.5742 (0.9105)  acc1: 87.5000 (76.6762)  acc5: 100.0000 (92.7068)  time: 0.0052  data: 0.0006  max mem: 8962\n",
            "Test:   [ 800/6250]  eta: 0:00:47  loss: 0.8010 (0.9392)  acc1: 75.0000 (76.1392)  acc5: 87.5000 (92.3221)  time: 0.0077  data: 0.0010  max mem: 8962\n",
            "Test:   [ 900/6250]  eta: 0:00:46  loss: 0.4006 (0.8848)  acc1: 87.5000 (77.4695)  acc5: 100.0000 (92.8413)  time: 0.0066  data: 0.0006  max mem: 8962\n",
            "Test:   [1000/6250]  eta: 0:00:44  loss: 0.8992 (0.8725)  acc1: 75.0000 (77.6474)  acc5: 100.0000 (92.9570)  time: 0.0070  data: 0.0007  max mem: 8962\n",
            "Test:   [1100/6250]  eta: 0:00:43  loss: 1.0749 (0.9068)  acc1: 62.5000 (76.6803)  acc5: 100.0000 (92.8020)  time: 0.0079  data: 0.0019  max mem: 8962\n",
            "Test:   [1200/6250]  eta: 0:00:42  loss: 1.0901 (0.9146)  acc1: 75.0000 (76.3114)  acc5: 87.5000 (92.8809)  time: 0.0081  data: 0.0007  max mem: 8962\n",
            "Test:   [1300/6250]  eta: 0:00:41  loss: 0.6278 (0.9170)  acc1: 75.0000 (76.0857)  acc5: 100.0000 (93.0630)  time: 0.0062  data: 0.0006  max mem: 8962\n",
            "Test:   [1400/6250]  eta: 0:00:39  loss: 0.8946 (0.9150)  acc1: 75.0000 (76.1153)  acc5: 87.5000 (93.0496)  time: 0.0070  data: 0.0006  max mem: 8962\n",
            "Test:   [1500/6250]  eta: 0:00:38  loss: 0.7951 (0.9198)  acc1: 75.0000 (75.8328)  acc5: 100.0000 (93.0963)  time: 0.0068  data: 0.0006  max mem: 8962\n",
            "Test:   [1600/6250]  eta: 0:00:37  loss: 0.2566 (0.9163)  acc1: 87.5000 (75.6636)  acc5: 100.0000 (93.2386)  time: 0.0072  data: 0.0007  max mem: 8962\n",
            "Test:   [1700/6250]  eta: 0:00:36  loss: 1.0530 (0.9138)  acc1: 75.0000 (75.6393)  acc5: 100.0000 (93.3789)  time: 0.0087  data: 0.0016  max mem: 8962\n",
            "Test:   [1800/6250]  eta: 0:00:35  loss: 1.1882 (0.9209)  acc1: 62.5000 (75.3956)  acc5: 87.5000 (93.3509)  time: 0.0053  data: 0.0004  max mem: 8962\n",
            "Test:   [1900/6250]  eta: 0:00:34  loss: 1.0323 (0.9166)  acc1: 62.5000 (75.6707)  acc5: 100.0000 (93.4508)  time: 0.0059  data: 0.0006  max mem: 8962\n",
            "Test:   [2000/6250]  eta: 0:00:33  loss: 0.4827 (0.9227)  acc1: 75.0000 (75.5622)  acc5: 100.0000 (93.4283)  time: 0.0062  data: 0.0005  max mem: 8962\n",
            "Test:   [2100/6250]  eta: 0:00:32  loss: 0.4458 (0.9042)  acc1: 87.5000 (76.1185)  acc5: 100.0000 (93.5983)  time: 0.0059  data: 0.0005  max mem: 8962\n",
            "Test:   [2200/6250]  eta: 0:00:31  loss: 0.1983 (0.8953)  acc1: 87.5000 (76.3233)  acc5: 100.0000 (93.6620)  time: 0.0058  data: 0.0005  max mem: 8962\n",
            "Test:   [2300/6250]  eta: 0:00:30  loss: 0.6167 (0.8971)  acc1: 87.5000 (76.2929)  acc5: 100.0000 (93.6658)  time: 0.0067  data: 0.0010  max mem: 8962\n",
            "Test:   [2400/6250]  eta: 0:00:29  loss: 1.1606 (0.9060)  acc1: 62.5000 (76.1245)  acc5: 87.5000 (93.5652)  time: 0.0069  data: 0.0006  max mem: 8962\n",
            "Test:   [2500/6250]  eta: 0:00:29  loss: 0.9882 (0.9067)  acc1: 75.0000 (76.2095)  acc5: 100.0000 (93.5176)  time: 0.0063  data: 0.0005  max mem: 8962\n",
            "Test:   [2600/6250]  eta: 0:00:27  loss: 2.1575 (0.9289)  acc1: 50.0000 (75.7785)  acc5: 75.0000 (93.2045)  time: 0.0058  data: 0.0006  max mem: 8962\n",
            "Test:   [2700/6250]  eta: 0:00:27  loss: 0.6626 (0.9389)  acc1: 75.0000 (75.6248)  acc5: 100.0000 (93.0859)  time: 0.0069  data: 0.0006  max mem: 8962\n",
            "Test:   [2800/6250]  eta: 0:00:26  loss: 1.7137 (0.9616)  acc1: 50.0000 (75.1205)  acc5: 87.5000 (92.8106)  time: 0.0062  data: 0.0006  max mem: 8962\n",
            "Test:   [2900/6250]  eta: 0:00:25  loss: 1.7213 (0.9804)  acc1: 50.0000 (74.7630)  acc5: 87.5000 (92.5758)  time: 0.0076  data: 0.0012  max mem: 8962\n",
            "Test:   [3000/6250]  eta: 0:00:24  loss: 2.1883 (1.0015)  acc1: 50.0000 (74.4585)  acc5: 75.0000 (92.2567)  time: 0.0066  data: 0.0013  max mem: 8962\n",
            "Test:   [3100/6250]  eta: 0:00:23  loss: 1.9025 (1.0238)  acc1: 50.0000 (73.9600)  acc5: 87.5000 (91.9824)  time: 0.0069  data: 0.0007  max mem: 8962\n",
            "Test:   [3200/6250]  eta: 0:00:22  loss: 0.7818 (1.0439)  acc1: 75.0000 (73.5786)  acc5: 100.0000 (91.7252)  time: 0.0066  data: 0.0009  max mem: 8962\n",
            "Test:   [3300/6250]  eta: 0:00:21  loss: 1.3260 (1.0571)  acc1: 50.0000 (73.2316)  acc5: 87.5000 (91.5972)  time: 0.0062  data: 0.0010  max mem: 8962\n",
            "Test:   [3400/6250]  eta: 0:00:21  loss: 2.2889 (1.0708)  acc1: 50.0000 (72.9638)  acc5: 75.0000 (91.4069)  time: 0.0054  data: 0.0004  max mem: 8962\n",
            "Test:   [3500/6250]  eta: 0:00:20  loss: 1.8139 (1.0788)  acc1: 62.5000 (72.8399)  acc5: 87.5000 (91.2704)  time: 0.0056  data: 0.0005  max mem: 8962\n",
            "Test:   [3600/6250]  eta: 0:00:19  loss: 0.5227 (1.0757)  acc1: 75.0000 (72.9797)  acc5: 100.0000 (91.2837)  time: 0.0061  data: 0.0004  max mem: 8962\n",
            "Test:   [3700/6250]  eta: 0:00:18  loss: 1.5864 (1.0885)  acc1: 50.0000 (72.7101)  acc5: 87.5000 (91.0970)  time: 0.0110  data: 0.0061  max mem: 8962\n",
            "Test:   [3800/6250]  eta: 0:00:18  loss: 0.7844 (1.0958)  acc1: 87.5000 (72.5993)  acc5: 87.5000 (90.9826)  time: 0.0059  data: 0.0005  max mem: 8962\n",
            "Test:   [3900/6250]  eta: 0:00:17  loss: 2.1642 (1.1114)  acc1: 37.5000 (72.2699)  acc5: 75.0000 (90.7684)  time: 0.0070  data: 0.0013  max mem: 8962\n",
            "Test:   [4000/6250]  eta: 0:00:16  loss: 1.3791 (1.1249)  acc1: 37.5000 (71.9789)  acc5: 87.5000 (90.6086)  time: 0.0061  data: 0.0009  max mem: 8962\n",
            "Test:   [4100/6250]  eta: 0:00:15  loss: 1.8430 (1.1361)  acc1: 50.0000 (71.8026)  acc5: 87.5000 (90.4566)  time: 0.0058  data: 0.0007  max mem: 8962\n",
            "Test:   [4200/6250]  eta: 0:00:14  loss: 0.5273 (1.1408)  acc1: 87.5000 (71.6556)  acc5: 100.0000 (90.4428)  time: 0.0054  data: 0.0004  max mem: 8962\n",
            "Test:   [4300/6250]  eta: 0:00:14  loss: 0.5790 (1.1499)  acc1: 75.0000 (71.5124)  acc5: 87.5000 (90.3075)  time: 0.0062  data: 0.0009  max mem: 8962\n",
            "Test:   [4400/6250]  eta: 0:00:13  loss: 0.8274 (1.1557)  acc1: 75.0000 (71.3730)  acc5: 100.0000 (90.2721)  time: 0.0063  data: 0.0009  max mem: 8962\n",
            "Test:   [4500/6250]  eta: 0:00:12  loss: 1.3455 (1.1605)  acc1: 75.0000 (71.3008)  acc5: 87.5000 (90.2327)  time: 0.0059  data: 0.0007  max mem: 8962\n",
            "Test:   [4600/6250]  eta: 0:00:11  loss: 1.6417 (1.1701)  acc1: 62.5000 (71.1476)  acc5: 87.5000 (90.0619)  time: 0.0066  data: 0.0008  max mem: 8962\n",
            "Test:   [4700/6250]  eta: 0:00:11  loss: 1.4379 (1.1789)  acc1: 62.5000 (70.8971)  acc5: 87.5000 (89.9197)  time: 0.0057  data: 0.0004  max mem: 8962\n",
            "Test:   [4800/6250]  eta: 0:00:10  loss: 1.3181 (1.1859)  acc1: 62.5000 (70.7899)  acc5: 87.5000 (89.8198)  time: 0.0061  data: 0.0006  max mem: 8962\n",
            "Test:   [4900/6250]  eta: 0:00:09  loss: 0.5202 (1.1891)  acc1: 87.5000 (70.7432)  acc5: 100.0000 (89.7648)  time: 0.0052  data: 0.0005  max mem: 8962\n",
            "Test:   [5000/6250]  eta: 0:00:08  loss: 2.0605 (1.1999)  acc1: 62.5000 (70.5934)  acc5: 75.0000 (89.6296)  time: 0.0062  data: 0.0012  max mem: 8962\n",
            "Test:   [5100/6250]  eta: 0:00:08  loss: 1.3460 (1.2044)  acc1: 62.5000 (70.5229)  acc5: 87.5000 (89.5976)  time: 0.0080  data: 0.0008  max mem: 8962\n",
            "Test:   [5200/6250]  eta: 0:00:07  loss: 1.1308 (1.2106)  acc1: 75.0000 (70.4264)  acc5: 87.5000 (89.5236)  time: 0.0055  data: 0.0005  max mem: 8962\n",
            "Test:   [5300/6250]  eta: 0:00:06  loss: 1.5583 (1.2231)  acc1: 62.5000 (70.1354)  acc5: 87.5000 (89.3157)  time: 0.0069  data: 0.0016  max mem: 8962\n",
            "Test:   [5400/6250]  eta: 0:00:06  loss: 0.8486 (1.2267)  acc1: 75.0000 (70.0542)  acc5: 87.5000 (89.2728)  time: 0.0064  data: 0.0012  max mem: 8962\n",
            "Test:   [5500/6250]  eta: 0:00:05  loss: 1.0672 (1.2307)  acc1: 62.5000 (69.9509)  acc5: 87.5000 (89.2179)  time: 0.0071  data: 0.0005  max mem: 8962\n",
            "Test:   [5600/6250]  eta: 0:00:04  loss: 1.0070 (1.2359)  acc1: 62.5000 (69.8648)  acc5: 87.5000 (89.1202)  time: 0.0096  data: 0.0041  max mem: 8962\n",
            "Test:   [5700/6250]  eta: 0:00:03  loss: 2.0962 (1.2487)  acc1: 37.5000 (69.5843)  acc5: 87.5000 (89.0019)  time: 0.0070  data: 0.0006  max mem: 8962\n",
            "Test:   [5800/6250]  eta: 0:00:03  loss: 0.7288 (1.2454)  acc1: 75.0000 (69.6582)  acc5: 100.0000 (89.0471)  time: 0.0061  data: 0.0005  max mem: 8962\n",
            "Test:   [5900/6250]  eta: 0:00:02  loss: 1.1190 (1.2431)  acc1: 75.0000 (69.6873)  acc5: 100.0000 (89.0781)  time: 0.0067  data: 0.0006  max mem: 8962\n",
            "Test:   [6000/6250]  eta: 0:00:01  loss: 0.6253 (1.2358)  acc1: 87.5000 (69.8425)  acc5: 87.5000 (89.1622)  time: 0.0068  data: 0.0006  max mem: 8962\n",
            "Test:   [6100/6250]  eta: 0:00:01  loss: 1.1478 (1.2438)  acc1: 62.5000 (69.6423)  acc5: 100.0000 (89.0776)  time: 0.0069  data: 0.0006  max mem: 8962\n",
            "Test:   [6200/6250]  eta: 0:00:00  loss: 0.2636 (1.2384)  acc1: 87.5000 (69.7650)  acc5: 100.0000 (89.1469)  time: 0.0071  data: 0.0024  max mem: 8962\n",
            "Test:  Total time: 0:00:44\n",
            "Test:  Acc@1 69.802 Acc@5 89.166\n",
            "Epoch: [1]  [   0/5005]  eta: 5:42:56  lr: 0.0004  img/s: 480.9318095087941  loss: 0.7240 (0.7240)  acc1: 67.1875 (67.1875)  acc5: 89.4531 (89.4531)  meanQV: 1.4703 (1.4703)  stdQV: 0.3293 (0.3293)  time: 4.1113  data: 3.5789  max mem: 8962\n",
            "Epoch: [1]  [ 100/5005]  eta: 0:45:51  lr: 0.0004  img/s: 493.27547238807784  loss: 0.7663 (0.7544)  acc1: 68.3594 (68.9047)  acc5: 85.9375 (87.0166)  meanQV: 1.4785 (1.4823)  stdQV: 0.3262 (0.3240)  time: 0.5211  data: 0.0004  max mem: 8962\n",
            "Epoch: [1]  [ 200/5005]  eta: 0:43:21  lr: 0.0004  img/s: 494.15947332208856  loss: 0.7673 (0.7576)  acc1: 68.7500 (69.0512)  acc5: 86.3281 (86.9928)  meanQV: 1.4813 (1.4833)  stdQV: 0.3240 (0.3234)  time: 0.5218  data: 0.0003  max mem: 8962\n",
            "Epoch: [1]  [ 300/5005]  eta: 0:41:55  lr: 0.0004  img/s: 490.8006792413158  loss: 0.7602 (0.7692)  acc1: 69.9219 (68.8824)  acc5: 88.2812 (86.9770)  meanQV: 1.4895 (1.4821)  stdQV: 0.3204 (0.3239)  time: 0.5209  data: 0.0004  max mem: 8962\n",
            "Epoch: [1]  [ 400/5005]  eta: 0:40:46  lr: 0.0004  img/s: 488.3024018976519  loss: 0.7641 (0.7740)  acc1: 67.9688 (68.7734)  acc5: 87.5000 (86.9837)  meanQV: 1.4758 (1.4814)  stdQV: 0.3262 (0.3243)  time: 0.5211  data: 0.0003  max mem: 8962\n",
            "Epoch: [1]  [ 500/5005]  eta: 0:39:44  lr: 0.0004  img/s: 491.2133664120496  loss: 0.7673 (0.7762)  acc1: 69.9219 (68.8451)  acc5: 87.5000 (87.0439)  meanQV: 1.4895 (1.4819)  stdQV: 0.3216 (0.3241)  time: 0.5217  data: 0.0003  max mem: 8962\n",
            "Epoch: [1]  [ 600/5005]  eta: 0:38:45  lr: 0.0004  img/s: 489.9574830025097  loss: 0.7744 (0.7785)  acc1: 68.7500 (68.8845)  acc5: 87.5000 (87.0379)  meanQV: 1.4813 (1.4821)  stdQV: 0.3240 (0.3240)  time: 0.5210  data: 0.0004  max mem: 8962\n",
            "Epoch: [1]  [ 700/5005]  eta: 0:37:48  lr: 0.0004  img/s: 490.1766813359385  loss: 0.7570 (0.7804)  acc1: 69.9219 (68.9227)  acc5: 87.1094 (87.0347)  meanQV: 1.4895 (1.4824)  stdQV: 0.3216 (0.3238)  time: 0.5210  data: 0.0003  max mem: 8962\n",
            "Epoch: [1]  [ 800/5005]  eta: 0:36:53  lr: 0.0004  img/s: 488.54479381319  loss: 0.7706 (0.7828)  acc1: 69.1406 (68.9124)  acc5: 86.3281 (87.0435)  meanQV: 1.4840 (1.4823)  stdQV: 0.3226 (0.3239)  time: 0.5213  data: 0.0003  max mem: 8962\n",
            "Epoch: [1]  [ 900/5005]  eta: 0:35:58  lr: 0.0004  img/s: 486.8855385165102  loss: 0.7941 (0.7848)  acc1: 68.3594 (68.9390)  acc5: 87.1094 (87.0465)  meanQV: 1.4785 (1.4825)  stdQV: 0.3251 (0.3238)  time: 0.5219  data: 0.0003  max mem: 8962\n",
            "Epoch: [1]  [1000/5005]  eta: 0:35:03  lr: 0.0004  img/s: 492.5010040446165  loss: 0.7671 (0.7867)  acc1: 69.5312 (68.9233)  acc5: 87.1094 (87.0430)  meanQV: 1.4867 (1.4824)  stdQV: 0.3228 (0.3238)  time: 0.5216  data: 0.0003  max mem: 8962\n",
            "Epoch: [1]  [1100/5005]  eta: 0:34:10  lr: 0.0004  img/s: 491.8478935923928  loss: 0.7918 (0.7885)  acc1: 68.3594 (68.8866)  acc5: 87.1094 (86.9948)  meanQV: 1.4785 (1.4821)  stdQV: 0.3251 (0.3240)  time: 0.5219  data: 0.0003  max mem: 8962\n",
            "Epoch: [1]  [1200/5005]  eta: 0:33:16  lr: 0.0004  img/s: 492.04127725035846  loss: 0.8054 (0.7904)  acc1: 68.3594 (68.8863)  acc5: 87.1094 (87.0059)  meanQV: 1.4785 (1.4821)  stdQV: 0.3240 (0.3240)  time: 0.5212  data: 0.0003  max mem: 8962\n",
            "Epoch: [1]  [1300/5005]  eta: 0:32:22  lr: 0.0004  img/s: 493.6503434775133  loss: 0.8003 (0.7921)  acc1: 68.7500 (68.8953)  acc5: 86.7188 (87.0238)  meanQV: 1.4813 (1.4822)  stdQV: 0.3240 (0.3239)  time: 0.5208  data: 0.0004  max mem: 8962\n",
            "Epoch: [1]  [1400/5005]  eta: 0:31:29  lr: 0.0004  img/s: 493.7590786833788  loss: 0.8469 (0.7947)  acc1: 67.5781 (68.8757)  acc5: 86.3281 (87.0087)  meanQV: 1.4730 (1.4821)  stdQV: 0.3273 (0.3240)  time: 0.5214  data: 0.0003  max mem: 8962\n",
            "Epoch: [1]  [1500/5005]  eta: 0:30:36  lr: 0.0004  img/s: 493.85105292835783  loss: 0.8278 (0.7964)  acc1: 67.9688 (68.8734)  acc5: 87.1094 (87.0126)  meanQV: 1.4758 (1.4820)  stdQV: 0.3260 (0.3240)  time: 0.5209  data: 0.0003  max mem: 8962\n",
            "Epoch: [1]  [1600/5005]  eta: 0:29:43  lr: 0.0004  img/s: 489.0960134101013  loss: 0.8061 (0.7976)  acc1: 69.5312 (68.8713)  acc5: 86.7188 (87.0096)  meanQV: 1.4867 (1.4820)  stdQV: 0.3228 (0.3240)  time: 0.5216  data: 0.0003  max mem: 8962\n",
            "Epoch: [1]  [1700/5005]  eta: 0:28:50  lr: 0.0004  img/s: 487.0217960638814  loss: 0.8219 (0.7990)  acc1: 69.1406 (68.8761)  acc5: 86.3281 (87.0118)  meanQV: 1.4840 (1.4821)  stdQV: 0.3204 (0.3240)  time: 0.5216  data: 0.0003  max mem: 8962\n",
            "Epoch: [1]  [1800/5005]  eta: 0:27:57  lr: 0.0004  img/s: 487.4064667244674  loss: 0.8217 (0.8005)  acc1: 67.1875 (68.8660)  acc5: 85.9375 (86.9964)  meanQV: 1.4703 (1.4820)  stdQV: 0.3273 (0.3240)  time: 0.5210  data: 0.0003  max mem: 8962\n",
            "Epoch: [1]  [1900/5005]  eta: 0:27:05  lr: 0.0004  img/s: 495.115358261054  loss: 0.8361 (0.8026)  acc1: 67.9688 (68.8474)  acc5: 86.3281 (86.9830)  meanQV: 1.4746 (1.4818)  stdQV: 0.3270 (0.3240)  time: 0.5206  data: 0.0003  max mem: 8962\n",
            "Epoch: [1]  [2000/5005]  eta: 0:26:12  lr: 0.0004  img/s: 488.846174928818  loss: 0.8257 (0.8035)  acc1: 69.1406 (68.8619)  acc5: 87.1094 (86.9956)  meanQV: 1.4840 (1.4819)  stdQV: 0.3228 (0.3240)  time: 0.5216  data: 0.0003  max mem: 8962\n",
            "Epoch: [1]  [2100/5005]  eta: 0:25:19  lr: 0.0004  img/s: 491.2070743506753  loss: 0.8428 (0.8053)  acc1: 66.7969 (68.8420)  acc5: 87.1094 (86.9922)  meanQV: 1.4676 (1.4818)  stdQV: 0.3303 (0.3241)  time: 0.5210  data: 0.0003  max mem: 8962\n",
            "Epoch: [1]  [2200/5005]  eta: 0:24:27  lr: 0.0004  img/s: 488.1822949084529  loss: 0.8558 (0.8064)  acc1: 68.3594 (68.8570)  acc5: 87.5000 (87.0222)  meanQV: 1.4785 (1.4819)  stdQV: 0.3251 (0.3240)  time: 0.5213  data: 0.0004  max mem: 8962\n",
            "Epoch: [1]  [2300/5005]  eta: 0:23:34  lr: 0.0004  img/s: 491.70599535560405  loss: 0.8227 (0.8072)  acc1: 68.3594 (68.8542)  acc5: 87.1094 (87.0248)  meanQV: 1.4785 (1.4819)  stdQV: 0.3262 (0.3240)  time: 0.5208  data: 0.0004  max mem: 8962\n",
            "Epoch: [1]  [2400/5005]  eta: 0:22:42  lr: 0.0004  img/s: 488.04228171446755  loss: 0.8344 (0.8086)  acc1: 69.5312 (68.8348)  acc5: 87.5000 (87.0231)  meanQV: 1.4867 (1.4817)  stdQV: 0.3228 (0.3241)  time: 0.5212  data: 0.0003  max mem: 8962\n",
            "Epoch: [1]  [2500/5005]  eta: 0:21:49  lr: 0.0004  img/s: 493.9346543052109  loss: 0.8142 (0.8099)  acc1: 67.1875 (68.8248)  acc5: 86.3281 (87.0107)  meanQV: 1.4703 (1.4817)  stdQV: 0.3273 (0.3241)  time: 0.5206  data: 0.0003  max mem: 8962\n",
            "Epoch: [1]  [2600/5005]  eta: 0:20:57  lr: 0.0004  img/s: 493.71140181999846  loss: 0.8310 (0.8114)  acc1: 69.5312 (68.8209)  acc5: 87.8906 (87.0080)  meanQV: 1.4855 (1.4816)  stdQV: 0.3216 (0.3241)  time: 0.5220  data: 0.0003  max mem: 8962\n",
            "Epoch: [1]  [2700/5005]  eta: 0:20:04  lr: 0.0004  img/s: 490.4764950910866  loss: 0.8228 (0.8124)  acc1: 68.7500 (68.8167)  acc5: 87.5000 (87.0055)  meanQV: 1.4813 (1.4816)  stdQV: 0.3238 (0.3241)  time: 0.5214  data: 0.0004  max mem: 8962\n",
            "Epoch: [1]  [2800/5005]  eta: 0:19:12  lr: 0.0004  img/s: 489.5152105330343  loss: 0.8128 (0.8133)  acc1: 69.1406 (68.8236)  acc5: 87.1094 (87.0178)  meanQV: 1.4840 (1.4817)  stdQV: 0.3240 (0.3241)  time: 0.5212  data: 0.0004  max mem: 8962\n",
            "Epoch: [1]  [2900/5005]  eta: 0:18:20  lr: 0.0004  img/s: 490.35195733551564  loss: 0.8154 (0.8143)  acc1: 67.5781 (68.8168)  acc5: 86.7188 (87.0159)  meanQV: 1.4730 (1.4816)  stdQV: 0.3273 (0.3241)  time: 0.5220  data: 0.0003  max mem: 8962\n",
            "Epoch: [1]  [3000/5005]  eta: 0:17:27  lr: 0.0004  img/s: 496.8430038188373  loss: 0.8169 (0.8153)  acc1: 69.1406 (68.8008)  acc5: 88.2812 (87.0080)  meanQV: 1.4840 (1.4815)  stdQV: 0.3228 (0.3242)  time: 0.5210  data: 0.0004  max mem: 8962\n",
            "Epoch: [1]  [3100/5005]  eta: 0:16:35  lr: 0.0004  img/s: 488.5263449079129  loss: 0.8308 (0.8165)  acc1: 70.3125 (68.7841)  acc5: 88.2812 (87.0036)  meanQV: 1.4922 (1.4814)  stdQV: 0.3192 (0.3242)  time: 0.5215  data: 0.0004  max mem: 8962\n",
            "Epoch: [1]  [3200/5005]  eta: 0:15:43  lr: 0.0004  img/s: 491.9060279391008  loss: 0.8449 (0.8176)  acc1: 68.3594 (68.7768)  acc5: 86.3281 (86.9960)  meanQV: 1.4785 (1.4813)  stdQV: 0.3262 (0.3242)  time: 0.5214  data: 0.0004  max mem: 8962\n",
            "Epoch: [1]  [3300/5005]  eta: 0:14:50  lr: 0.0004  img/s: 493.2963213941315  loss: 0.8492 (0.8186)  acc1: 67.9688 (68.7659)  acc5: 87.1094 (86.9994)  meanQV: 1.4758 (1.4812)  stdQV: 0.3248 (0.3243)  time: 0.5203  data: 0.0004  max mem: 8962\n",
            "Epoch: [1]  [3400/5005]  eta: 0:13:58  lr: 0.0004  img/s: 489.8253877234111  loss: 0.8006 (0.8190)  acc1: 68.3594 (68.7695)  acc5: 86.7188 (87.0057)  meanQV: 1.4785 (1.4813)  stdQV: 0.3192 (0.3243)  time: 0.5208  data: 0.0004  max mem: 8962\n",
            "Epoch: [1]  [3500/5005]  eta: 0:13:06  lr: 0.0004  img/s: 487.77468294972033  loss: 0.8505 (0.8198)  acc1: 68.7500 (68.7579)  acc5: 85.9375 (87.0026)  meanQV: 1.4813 (1.4812)  stdQV: 0.3248 (0.3243)  time: 0.5211  data: 0.0004  max mem: 8962\n",
            "Epoch: [1]  [3600/5005]  eta: 0:12:13  lr: 0.0004  img/s: 493.46703224484054  loss: 0.8441 (0.8204)  acc1: 68.3594 (68.7684)  acc5: 87.1094 (86.9979)  meanQV: 1.4785 (1.4813)  stdQV: 0.3262 (0.3243)  time: 0.5203  data: 0.0004  max mem: 8962\n",
            "Epoch: [1]  [3700/5005]  eta: 0:11:21  lr: 0.0004  img/s: 489.55359849141115  loss: 0.8019 (0.8213)  acc1: 69.9219 (68.7655)  acc5: 86.7188 (86.9992)  meanQV: 1.4891 (1.4812)  stdQV: 0.3214 (0.3243)  time: 0.5221  data: 0.0004  max mem: 8962\n",
            "Epoch: [1]  [3800/5005]  eta: 0:10:29  lr: 0.0004  img/s: 493.67462881592127  loss: 0.8116 (0.8217)  acc1: 69.1406 (68.7733)  acc5: 86.7188 (87.0059)  meanQV: 1.4840 (1.4813)  stdQV: 0.3216 (0.3243)  time: 0.5207  data: 0.0004  max mem: 8962\n",
            "Epoch: [1]  [3900/5005]  eta: 0:09:37  lr: 0.0004  img/s: 491.10327860529725  loss: 0.8894 (0.8224)  acc1: 67.5781 (68.7723)  acc5: 85.9375 (87.0182)  meanQV: 1.4730 (1.4813)  stdQV: 0.3273 (0.3243)  time: 0.5204  data: 0.0003  max mem: 8962\n",
            "Epoch: [1]  [4000/5005]  eta: 0:08:42  lr: 0.0004  img/s: 658.0155167215349  loss: 0.8130 (0.8233)  acc1: 68.7500 (68.7736)  acc5: 86.7188 (87.0157)  meanQV: 1.4813 (1.4813)  stdQV: 0.3228 (0.3243)  time: 0.3829  data: 0.0022  max mem: 8962\n",
            "Epoch: [1]  [4100/5005]  eta: 0:07:48  lr: 0.0004  img/s: 495.21469028956733  loss: 0.8295 (0.8240)  acc1: 69.1406 (68.7658)  acc5: 87.5000 (87.0174)  meanQV: 1.4816 (1.4812)  stdQV: 0.3235 (0.3243)  time: 0.5208  data: 0.0003  max mem: 8962\n",
            "Epoch: [1]  [4200/5005]  eta: 0:06:57  lr: 0.0004  img/s: 492.8694949672994  loss: 0.8703 (0.8249)  acc1: 68.3594 (68.7643)  acc5: 87.1094 (87.0155)  meanQV: 1.4773 (1.4812)  stdQV: 0.3262 (0.3243)  time: 0.5213  data: 0.0005  max mem: 8962\n",
            "Epoch: [1]  [4300/5005]  eta: 0:06:05  lr: 0.0004  img/s: 490.85205892006695  loss: 0.8436 (0.8255)  acc1: 67.5781 (68.7648)  acc5: 87.1094 (87.0136)  meanQV: 1.4730 (1.4812)  stdQV: 0.3262 (0.3243)  time: 0.5211  data: 0.0003  max mem: 8962\n",
            "Epoch: [1]  [4400/5005]  eta: 0:05:13  lr: 0.0004  img/s: 493.98669227672667  loss: 0.8510 (0.8258)  acc1: 68.3594 (68.7773)  acc5: 86.7188 (87.0168)  meanQV: 1.4785 (1.4813)  stdQV: 0.3251 (0.3242)  time: 0.5209  data: 0.0004  max mem: 8962\n",
            "Epoch: [1]  [4500/5005]  eta: 0:04:21  lr: 0.0004  img/s: 493.84991723495295  loss: 0.8170 (0.8264)  acc1: 68.3594 (68.7790)  acc5: 88.2812 (87.0156)  meanQV: 1.4785 (1.4813)  stdQV: 0.3238 (0.3242)  time: 0.5213  data: 0.0003  max mem: 8962\n",
            "Epoch: [1]  [4600/5005]  eta: 0:03:30  lr: 0.0004  img/s: 488.5125647070813  loss: 0.8377 (0.8271)  acc1: 68.3594 (68.7688)  acc5: 87.5000 (87.0158)  meanQV: 1.4770 (1.4812)  stdQV: 0.3262 (0.3243)  time: 0.5200  data: 0.0004  max mem: 8962\n",
            "Epoch: [1]  [4700/5005]  eta: 0:02:38  lr: 0.0004  img/s: 491.9828855090406  loss: 0.8226 (0.8275)  acc1: 68.3594 (68.7709)  acc5: 87.8906 (87.0220)  meanQV: 1.4785 (1.4812)  stdQV: 0.3216 (0.3243)  time: 0.5212  data: 0.0004  max mem: 8962\n",
            "Epoch: [1]  [4800/5005]  eta: 0:01:46  lr: 0.0004  img/s: 493.26708795863277  loss: 0.8528 (0.8281)  acc1: 68.7500 (68.7686)  acc5: 86.3281 (87.0159)  meanQV: 1.4809 (1.4812)  stdQV: 0.3251 (0.3243)  time: 0.5216  data: 0.0003  max mem: 8962\n",
            "Epoch: [1]  [4900/5005]  eta: 0:00:54  lr: 0.0004  img/s: 491.2724747144307  loss: 0.8068 (0.8286)  acc1: 69.5312 (68.7636)  acc5: 87.1094 (87.0159)  meanQV: 1.4867 (1.4812)  stdQV: 0.3228 (0.3243)  time: 0.5217  data: 0.0003  max mem: 8962\n",
            "Epoch: [1]  [5000/5005]  eta: 0:00:02  lr: 0.0004  img/s: 492.8498131172718  loss: 0.8547 (0.8292)  acc1: 67.9688 (68.7634)  acc5: 86.3281 (87.0148)  meanQV: 1.4758 (1.4812)  stdQV: 0.3262 (0.3243)  time: 0.5210  data: 0.0002  max mem: 8962\n",
            "Epoch: [1] Total time: 0:43:16\n",
            "Test:   [   0/6250]  eta: 1:21:57  loss: 0.8673 (0.8673)  acc1: 62.5000 (62.5000)  acc5: 100.0000 (100.0000)  time: 0.7867  data: 0.7725  max mem: 8962\n",
            "Test:   [ 100/6250]  eta: 0:01:29  loss: 0.1868 (0.6206)  acc1: 100.0000 (83.5396)  acc5: 100.0000 (95.4208)  time: 0.0063  data: 0.0008  max mem: 8962\n",
            "Test:   [ 200/6250]  eta: 0:01:03  loss: 0.7102 (0.6552)  acc1: 75.0000 (83.7687)  acc5: 100.0000 (94.6517)  time: 0.0056  data: 0.0004  max mem: 8962\n",
            "Test:   [ 300/6250]  eta: 0:00:54  loss: 1.2776 (0.8690)  acc1: 62.5000 (78.6130)  acc5: 87.5000 (93.0233)  time: 0.0061  data: 0.0009  max mem: 8962\n",
            "Test:   [ 400/6250]  eta: 0:00:48  loss: 0.9489 (0.9890)  acc1: 75.0000 (75.8416)  acc5: 87.5000 (92.0823)  time: 0.0060  data: 0.0005  max mem: 8962\n",
            "Test:   [ 500/6250]  eta: 0:00:45  loss: 1.0595 (1.0403)  acc1: 75.0000 (74.4261)  acc5: 87.5000 (91.6168)  time: 0.0053  data: 0.0007  max mem: 8962\n",
            "Test:   [ 600/6250]  eta: 0:00:42  loss: 0.4378 (0.9454)  acc1: 87.5000 (76.4975)  acc5: 100.0000 (92.3877)  time: 0.0051  data: 0.0004  max mem: 8962\n",
            "Test:   [ 700/6250]  eta: 0:00:40  loss: 0.4026 (0.9303)  acc1: 87.5000 (76.7653)  acc5: 100.0000 (92.4394)  time: 0.0056  data: 0.0006  max mem: 8962\n",
            "Test:   [ 800/6250]  eta: 0:00:39  loss: 0.7303 (0.9549)  acc1: 75.0000 (76.2797)  acc5: 87.5000 (92.0880)  time: 0.0063  data: 0.0006  max mem: 8962\n",
            "Test:   [ 900/6250]  eta: 0:00:37  loss: 0.3813 (0.8983)  acc1: 87.5000 (77.5943)  acc5: 100.0000 (92.6471)  time: 0.0063  data: 0.0006  max mem: 8962\n",
            "Test:   [1000/6250]  eta: 0:00:36  loss: 0.8989 (0.8855)  acc1: 75.0000 (77.7473)  acc5: 100.0000 (92.7947)  time: 0.0056  data: 0.0007  max mem: 8962\n",
            "Test:   [1100/6250]  eta: 0:00:35  loss: 1.1069 (0.9229)  acc1: 75.0000 (76.8847)  acc5: 100.0000 (92.6203)  time: 0.0064  data: 0.0022  max mem: 8962\n",
            "Test:   [1200/6250]  eta: 0:00:34  loss: 1.0983 (0.9314)  acc1: 75.0000 (76.4467)  acc5: 87.5000 (92.6624)  time: 0.0055  data: 0.0005  max mem: 8962\n",
            "Test:   [1300/6250]  eta: 0:00:33  loss: 0.5355 (0.9346)  acc1: 75.0000 (76.1145)  acc5: 100.0000 (92.8132)  time: 0.0057  data: 0.0005  max mem: 8962\n",
            "Test:   [1400/6250]  eta: 0:00:32  loss: 0.7667 (0.9310)  acc1: 75.0000 (76.2313)  acc5: 100.0000 (92.9069)  time: 0.0052  data: 0.0004  max mem: 8962\n",
            "Test:   [1500/6250]  eta: 0:00:31  loss: 0.8760 (0.9357)  acc1: 62.5000 (75.9494)  acc5: 100.0000 (92.9880)  time: 0.0053  data: 0.0007  max mem: 8962\n",
            "Test:   [1600/6250]  eta: 0:00:30  loss: 0.0993 (0.9286)  acc1: 100.0000 (75.9135)  acc5: 100.0000 (93.1371)  time: 0.0053  data: 0.0005  max mem: 8962\n",
            "Test:   [1700/6250]  eta: 0:00:29  loss: 1.1072 (0.9257)  acc1: 62.5000 (75.8157)  acc5: 100.0000 (93.2760)  time: 0.0068  data: 0.0021  max mem: 8962\n",
            "Test:   [1800/6250]  eta: 0:00:28  loss: 1.1913 (0.9373)  acc1: 62.5000 (75.5622)  acc5: 87.5000 (93.2537)  time: 0.0054  data: 0.0006  max mem: 8962\n",
            "Test:   [1900/6250]  eta: 0:00:28  loss: 1.0258 (0.9308)  acc1: 62.5000 (75.8417)  acc5: 100.0000 (93.3653)  time: 0.0050  data: 0.0006  max mem: 8962\n",
            "Test:   [2000/6250]  eta: 0:00:27  loss: 0.5154 (0.9364)  acc1: 87.5000 (75.7871)  acc5: 100.0000 (93.3221)  time: 0.0055  data: 0.0006  max mem: 8962\n",
            "Test:   [2100/6250]  eta: 0:00:27  loss: 0.3852 (0.9175)  acc1: 87.5000 (76.3267)  acc5: 100.0000 (93.4793)  time: 0.0061  data: 0.0012  max mem: 8962\n",
            "Test:   [2200/6250]  eta: 0:00:26  loss: 0.2291 (0.9102)  acc1: 87.5000 (76.4652)  acc5: 100.0000 (93.5654)  time: 0.0063  data: 0.0007  max mem: 8962\n",
            "Test:   [2300/6250]  eta: 0:00:25  loss: 0.6943 (0.9109)  acc1: 87.5000 (76.4776)  acc5: 100.0000 (93.5843)  time: 0.0062  data: 0.0005  max mem: 8962\n",
            "Test:   [2400/6250]  eta: 0:00:25  loss: 0.8836 (0.9201)  acc1: 75.0000 (76.3432)  acc5: 87.5000 (93.4975)  time: 0.0056  data: 0.0011  max mem: 8962\n",
            "Test:   [2500/6250]  eta: 0:00:24  loss: 0.7644 (0.9205)  acc1: 87.5000 (76.4794)  acc5: 100.0000 (93.4626)  time: 0.0061  data: 0.0010  max mem: 8962\n",
            "Test:   [2600/6250]  eta: 0:00:23  loss: 2.2800 (0.9451)  acc1: 37.5000 (75.9948)  acc5: 75.0000 (93.1469)  time: 0.0055  data: 0.0006  max mem: 8962\n",
            "Test:   [2700/6250]  eta: 0:00:23  loss: 0.7858 (0.9555)  acc1: 75.0000 (75.7960)  acc5: 87.5000 (93.0118)  time: 0.0060  data: 0.0006  max mem: 8962\n",
            "Test:   [2800/6250]  eta: 0:00:22  loss: 1.6738 (0.9770)  acc1: 62.5000 (75.3481)  acc5: 75.0000 (92.7704)  time: 0.0052  data: 0.0004  max mem: 8962\n",
            "Test:   [2900/6250]  eta: 0:00:21  loss: 1.9858 (0.9960)  acc1: 37.5000 (74.9138)  acc5: 87.5000 (92.5241)  time: 0.0071  data: 0.0006  max mem: 8962\n",
            "Test:   [3000/6250]  eta: 0:00:21  loss: 2.4274 (1.0179)  acc1: 50.0000 (74.6126)  acc5: 75.0000 (92.1984)  time: 0.0060  data: 0.0005  max mem: 8962\n",
            "Test:   [3100/6250]  eta: 0:00:20  loss: 1.8322 (1.0420)  acc1: 50.0000 (74.1172)  acc5: 75.0000 (91.9461)  time: 0.0064  data: 0.0005  max mem: 8962\n",
            "Test:   [3200/6250]  eta: 0:00:19  loss: 0.7034 (1.0615)  acc1: 75.0000 (73.6996)  acc5: 100.0000 (91.7291)  time: 0.0049  data: 0.0004  max mem: 8962\n",
            "Test:   [3300/6250]  eta: 0:00:19  loss: 1.2400 (1.0739)  acc1: 62.5000 (73.3528)  acc5: 87.5000 (91.6238)  time: 0.0050  data: 0.0006  max mem: 8962\n",
            "Test:   [3400/6250]  eta: 0:00:18  loss: 2.3805 (1.0877)  acc1: 37.5000 (73.0851)  acc5: 75.0000 (91.4327)  time: 0.0060  data: 0.0007  max mem: 8962\n",
            "Test:   [3500/6250]  eta: 0:00:17  loss: 1.7449 (1.0943)  acc1: 50.0000 (72.9899)  acc5: 87.5000 (91.2953)  time: 0.0074  data: 0.0023  max mem: 8962\n",
            "Test:   [3600/6250]  eta: 0:00:16  loss: 0.4245 (1.0910)  acc1: 87.5000 (73.1325)  acc5: 100.0000 (91.2941)  time: 0.0059  data: 0.0004  max mem: 8962\n",
            "Test:   [3700/6250]  eta: 0:00:16  loss: 1.6126 (1.1052)  acc1: 50.0000 (72.8519)  acc5: 87.5000 (91.1105)  time: 0.0121  data: 0.0065  max mem: 8962\n",
            "Test:   [3800/6250]  eta: 0:00:15  loss: 0.7444 (1.1120)  acc1: 87.5000 (72.7605)  acc5: 87.5000 (90.9761)  time: 0.0057  data: 0.0005  max mem: 8962\n",
            "Test:   [3900/6250]  eta: 0:00:14  loss: 2.3107 (1.1276)  acc1: 50.0000 (72.4494)  acc5: 75.0000 (90.7652)  time: 0.0052  data: 0.0005  max mem: 8962\n",
            "Test:   [4000/6250]  eta: 0:00:14  loss: 1.4615 (1.1421)  acc1: 62.5000 (72.1788)  acc5: 87.5000 (90.5992)  time: 0.0056  data: 0.0004  max mem: 8962\n",
            "Test:   [4100/6250]  eta: 0:00:13  loss: 1.7447 (1.1522)  acc1: 62.5000 (72.0129)  acc5: 87.5000 (90.4840)  time: 0.0059  data: 0.0005  max mem: 8962\n",
            "Test:   [4200/6250]  eta: 0:00:12  loss: 0.4051 (1.1574)  acc1: 87.5000 (71.8609)  acc5: 100.0000 (90.4725)  time: 0.0061  data: 0.0005  max mem: 8962\n",
            "Test:   [4300/6250]  eta: 0:00:12  loss: 0.4819 (1.1662)  acc1: 75.0000 (71.7420)  acc5: 87.5000 (90.3191)  time: 0.0058  data: 0.0007  max mem: 8962\n",
            "Test:   [4400/6250]  eta: 0:00:11  loss: 0.8488 (1.1727)  acc1: 75.0000 (71.5633)  acc5: 100.0000 (90.2835)  time: 0.0052  data: 0.0005  max mem: 8962\n",
            "Test:   [4500/6250]  eta: 0:00:11  loss: 1.1410 (1.1786)  acc1: 62.5000 (71.4619)  acc5: 87.5000 (90.2383)  time: 0.0058  data: 0.0004  max mem: 8962\n",
            "Test:   [4600/6250]  eta: 0:00:10  loss: 1.5850 (1.1891)  acc1: 50.0000 (71.2834)  acc5: 87.5000 (90.0701)  time: 0.0064  data: 0.0010  max mem: 8962\n",
            "Test:   [4700/6250]  eta: 0:00:09  loss: 1.3859 (1.1988)  acc1: 62.5000 (70.9982)  acc5: 87.5000 (89.9463)  time: 0.0051  data: 0.0005  max mem: 8962\n",
            "Test:   [4800/6250]  eta: 0:00:09  loss: 1.2475 (1.2056)  acc1: 62.5000 (70.8811)  acc5: 87.5000 (89.8563)  time: 0.0054  data: 0.0005  max mem: 8962\n",
            "Test:   [4900/6250]  eta: 0:00:08  loss: 0.4116 (1.2104)  acc1: 87.5000 (70.8121)  acc5: 100.0000 (89.7750)  time: 0.0058  data: 0.0009  max mem: 8962\n",
            "Test:   [5000/6250]  eta: 0:00:07  loss: 2.0346 (1.2224)  acc1: 62.5000 (70.6259)  acc5: 75.0000 (89.6321)  time: 0.0059  data: 0.0005  max mem: 8962\n",
            "Test:   [5100/6250]  eta: 0:00:07  loss: 1.2888 (1.2271)  acc1: 62.5000 (70.5327)  acc5: 87.5000 (89.5952)  time: 0.0069  data: 0.0006  max mem: 8962\n",
            "Test:   [5200/6250]  eta: 0:00:06  loss: 1.0585 (1.2335)  acc1: 75.0000 (70.4216)  acc5: 87.5000 (89.5236)  time: 0.0053  data: 0.0005  max mem: 8962\n",
            "Test:   [5300/6250]  eta: 0:00:05  loss: 1.5391 (1.2470)  acc1: 62.5000 (70.1401)  acc5: 87.5000 (89.3440)  time: 0.0058  data: 0.0006  max mem: 8962\n",
            "Test:   [5400/6250]  eta: 0:00:05  loss: 0.9108 (1.2497)  acc1: 75.0000 (70.0889)  acc5: 87.5000 (89.2936)  time: 0.0057  data: 0.0010  max mem: 8962\n",
            "Test:   [5500/6250]  eta: 0:00:04  loss: 0.9651 (1.2539)  acc1: 75.0000 (69.9668)  acc5: 87.5000 (89.2383)  time: 0.0060  data: 0.0006  max mem: 8962\n",
            "Test:   [5600/6250]  eta: 0:00:04  loss: 0.8866 (1.2592)  acc1: 62.5000 (69.8915)  acc5: 87.5000 (89.1381)  time: 0.0056  data: 0.0005  max mem: 8962\n",
            "Test:   [5700/6250]  eta: 0:00:03  loss: 1.9771 (1.2739)  acc1: 37.5000 (69.5799)  acc5: 75.0000 (88.9954)  time: 0.0055  data: 0.0004  max mem: 8962\n",
            "Test:   [5800/6250]  eta: 0:00:02  loss: 0.5795 (1.2705)  acc1: 75.0000 (69.6410)  acc5: 100.0000 (89.0407)  time: 0.0060  data: 0.0006  max mem: 8962\n",
            "Test:   [5900/6250]  eta: 0:00:02  loss: 1.1301 (1.2675)  acc1: 75.0000 (69.6831)  acc5: 100.0000 (89.0824)  time: 0.0058  data: 0.0005  max mem: 8962\n",
            "Test:   [6000/6250]  eta: 0:00:01  loss: 0.5545 (1.2598)  acc1: 87.5000 (69.8384)  acc5: 87.5000 (89.1685)  time: 0.0059  data: 0.0004  max mem: 8962\n",
            "Test:   [6100/6250]  eta: 0:00:00  loss: 1.2558 (1.2676)  acc1: 62.5000 (69.6648)  acc5: 87.5000 (89.0797)  time: 0.0053  data: 0.0004  max mem: 8962\n",
            "Test:   [6200/6250]  eta: 0:00:00  loss: 0.1734 (1.2610)  acc1: 87.5000 (69.7831)  acc5: 100.0000 (89.1610)  time: 0.0060  data: 0.0007  max mem: 8962\n",
            "Test:  Total time: 0:00:38\n",
            "Test:  Acc@1 69.816 Acc@5 89.174\n",
            "Epoch: [2]  [   0/5005]  eta: 5:24:29  lr: 0.0004  img/s: 488.37947021283276  loss: 0.7970 (0.7970)  acc1: 72.2656 (72.2656)  acc5: 87.1094 (87.1094)  meanQV: 1.5059 (1.5059)  stdQV: 0.3140 (0.3140)  time: 3.8899  data: 3.3657  max mem: 8962\n",
            "Epoch: [2]  [ 100/5005]  eta: 0:45:25  lr: 0.0004  img/s: 493.19209398441154  loss: 0.8641 (0.8689)  acc1: 67.9688 (68.4483)  acc5: 87.1094 (86.6530)  meanQV: 1.4758 (1.4788)  stdQV: 0.3251 (0.3251)  time: 0.5211  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [ 200/5005]  eta: 0:43:07  lr: 0.0004  img/s: 491.1342779344896  loss: 0.8738 (0.8655)  acc1: 68.7500 (68.5595)  acc5: 87.5000 (86.8898)  meanQV: 1.4813 (1.4796)  stdQV: 0.3240 (0.3248)  time: 0.5215  data: 0.0004  max mem: 8962\n",
            "Epoch: [2]  [ 300/5005]  eta: 0:41:47  lr: 0.0004  img/s: 487.00059052814623  loss: 0.8533 (0.8685)  acc1: 68.3594 (68.4191)  acc5: 87.5000 (86.8005)  meanQV: 1.4785 (1.4786)  stdQV: 0.3251 (0.3252)  time: 0.5215  data: 0.0004  max mem: 8962\n",
            "Epoch: [2]  [ 400/5005]  eta: 0:40:40  lr: 0.0004  img/s: 488.72224116185464  loss: 0.8569 (0.8636)  acc1: 69.1406 (68.5932)  acc5: 87.1094 (86.8931)  meanQV: 1.4840 (1.4799)  stdQV: 0.3216 (0.3247)  time: 0.5220  data: 0.0004  max mem: 8962\n",
            "Epoch: [2]  [ 500/5005]  eta: 0:39:39  lr: 0.0004  img/s: 485.98511188265843  loss: 0.8240 (0.8635)  acc1: 69.5312 (68.5403)  acc5: 87.8906 (86.8723)  meanQV: 1.4867 (1.4795)  stdQV: 0.3215 (0.3249)  time: 0.5207  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [ 600/5005]  eta: 0:38:41  lr: 0.0004  img/s: 493.6265143933163  loss: 0.8603 (0.8641)  acc1: 68.7500 (68.5791)  acc5: 86.3281 (86.8962)  meanQV: 1.4801 (1.4798)  stdQV: 0.3251 (0.3248)  time: 0.5222  data: 0.0004  max mem: 8962\n",
            "Epoch: [2]  [ 700/5005]  eta: 0:37:44  lr: 0.0004  img/s: 495.0733539463698  loss: 0.8482 (0.8641)  acc1: 68.3594 (68.5656)  acc5: 87.1094 (86.9194)  meanQV: 1.4785 (1.4797)  stdQV: 0.3260 (0.3249)  time: 0.5209  data: 0.0004  max mem: 8962\n",
            "Epoch: [2]  [ 800/5005]  eta: 0:36:49  lr: 0.0004  img/s: 490.3255347956828  loss: 0.8578 (0.8649)  acc1: 69.1406 (68.5959)  acc5: 86.7188 (86.9270)  meanQV: 1.4840 (1.4799)  stdQV: 0.3240 (0.3248)  time: 0.5212  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [ 900/5005]  eta: 0:35:55  lr: 0.0004  img/s: 487.8480383248159  loss: 0.8300 (0.8626)  acc1: 69.1406 (68.6646)  acc5: 87.5000 (86.9767)  meanQV: 1.4840 (1.4804)  stdQV: 0.3228 (0.3246)  time: 0.5215  data: 0.0004  max mem: 8962\n",
            "Epoch: [2]  [1000/5005]  eta: 0:35:01  lr: 0.0004  img/s: 491.5558526149826  loss: 0.8513 (0.8634)  acc1: 69.1406 (68.6805)  acc5: 87.5000 (86.9775)  meanQV: 1.4840 (1.4805)  stdQV: 0.3226 (0.3245)  time: 0.5211  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [1100/5005]  eta: 0:34:07  lr: 0.0004  img/s: 490.84757118315395  loss: 0.8577 (0.8636)  acc1: 68.3594 (68.6847)  acc5: 86.7188 (86.9710)  meanQV: 1.4785 (1.4805)  stdQV: 0.3251 (0.3245)  time: 0.5209  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [1200/5005]  eta: 0:33:14  lr: 0.0004  img/s: 494.1517410611354  loss: 0.8542 (0.8649)  acc1: 68.7500 (68.6462)  acc5: 86.3281 (86.9607)  meanQV: 1.4813 (1.4802)  stdQV: 0.3240 (0.3246)  time: 0.5209  data: 0.0004  max mem: 8962\n",
            "Epoch: [2]  [1300/5005]  eta: 0:32:20  lr: 0.0004  img/s: 487.5868519700021  loss: 0.8400 (0.8649)  acc1: 69.1406 (68.6611)  acc5: 86.3281 (86.9457)  meanQV: 1.4840 (1.4803)  stdQV: 0.3240 (0.3246)  time: 0.5207  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [1400/5005]  eta: 0:31:27  lr: 0.0004  img/s: 493.3883500186788  loss: 0.8124 (0.8646)  acc1: 69.1406 (68.6887)  acc5: 87.8906 (86.9658)  meanQV: 1.4840 (1.4805)  stdQV: 0.3240 (0.3245)  time: 0.5210  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [1500/5005]  eta: 0:30:34  lr: 0.0004  img/s: 495.2411855622316  loss: 0.8914 (0.8663)  acc1: 67.5781 (68.6449)  acc5: 87.5000 (86.9530)  meanQV: 1.4719 (1.4802)  stdQV: 0.3280 (0.3246)  time: 0.5220  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [1600/5005]  eta: 0:29:41  lr: 0.0004  img/s: 492.28401072281827  loss: 0.8554 (0.8661)  acc1: 69.9219 (68.6358)  acc5: 87.5000 (86.9669)  meanQV: 1.4891 (1.4802)  stdQV: 0.3216 (0.3246)  time: 0.5215  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [1700/5005]  eta: 0:28:49  lr: 0.0004  img/s: 488.3163923205836  loss: 0.8374 (0.8661)  acc1: 69.5312 (68.6529)  acc5: 87.5000 (86.9794)  meanQV: 1.4867 (1.4803)  stdQV: 0.3228 (0.3246)  time: 0.5213  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [1800/5005]  eta: 0:27:56  lr: 0.0004  img/s: 493.57046837953203  loss: 0.8886 (0.8664)  acc1: 68.3594 (68.6639)  acc5: 86.7188 (86.9808)  meanQV: 1.4785 (1.4804)  stdQV: 0.3251 (0.3245)  time: 0.5211  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [1900/5005]  eta: 0:27:03  lr: 0.0004  img/s: 490.1453552810568  loss: 0.8548 (0.8668)  acc1: 68.3594 (68.6598)  acc5: 87.1094 (86.9768)  meanQV: 1.4785 (1.4803)  stdQV: 0.3249 (0.3245)  time: 0.5213  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [2000/5005]  eta: 0:26:11  lr: 0.0004  img/s: 488.1270344653269  loss: 0.9081 (0.8673)  acc1: 67.9688 (68.6596)  acc5: 85.9375 (86.9690)  meanQV: 1.4754 (1.4803)  stdQV: 0.3270 (0.3245)  time: 0.5214  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [2100/5005]  eta: 0:25:18  lr: 0.0004  img/s: 489.5038292141757  loss: 0.8766 (0.8678)  acc1: 68.7500 (68.6416)  acc5: 87.5000 (86.9612)  meanQV: 1.4813 (1.4802)  stdQV: 0.3240 (0.3246)  time: 0.5213  data: 0.0004  max mem: 8962\n",
            "Epoch: [2]  [2200/5005]  eta: 0:24:26  lr: 0.0004  img/s: 493.80040608063206  loss: 0.8591 (0.8679)  acc1: 68.3594 (68.6433)  acc5: 86.7188 (86.9536)  meanQV: 1.4785 (1.4802)  stdQV: 0.3259 (0.3246)  time: 0.5222  data: 0.0004  max mem: 8962\n",
            "Epoch: [2]  [2300/5005]  eta: 0:23:33  lr: 0.0004  img/s: 487.89902333117954  loss: 0.8695 (0.8681)  acc1: 67.5781 (68.6313)  acc5: 86.3281 (86.9520)  meanQV: 1.4703 (1.4801)  stdQV: 0.3262 (0.3246)  time: 0.5212  data: 0.0004  max mem: 8962\n",
            "Epoch: [2]  [2400/5005]  eta: 0:22:41  lr: 0.0004  img/s: 495.53373554749567  loss: 0.9173 (0.8689)  acc1: 67.5781 (68.6229)  acc5: 86.3281 (86.9381)  meanQV: 1.4730 (1.4801)  stdQV: 0.3270 (0.3247)  time: 0.5204  data: 0.0004  max mem: 8962\n",
            "Epoch: [2]  [2500/5005]  eta: 0:21:48  lr: 0.0004  img/s: 492.0162505212319  loss: 0.8747 (0.8695)  acc1: 67.9688 (68.6191)  acc5: 87.1094 (86.9432)  meanQV: 1.4758 (1.4800)  stdQV: 0.3262 (0.3247)  time: 0.5214  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [2600/5005]  eta: 0:20:56  lr: 0.0004  img/s: 489.42819036679475  loss: 0.8458 (0.8694)  acc1: 69.9219 (68.6381)  acc5: 86.7188 (86.9568)  meanQV: 1.4895 (1.4802)  stdQV: 0.3216 (0.3246)  time: 0.5213  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [2700/5005]  eta: 0:20:04  lr: 0.0004  img/s: 491.20055774275215  loss: 0.8641 (0.8696)  acc1: 67.1875 (68.6336)  acc5: 86.7188 (86.9582)  meanQV: 1.4703 (1.4801)  stdQV: 0.3277 (0.3246)  time: 0.5221  data: 0.0004  max mem: 8962\n",
            "Epoch: [2]  [2800/5005]  eta: 0:19:11  lr: 0.0004  img/s: 494.1210418244721  loss: 0.8495 (0.8697)  acc1: 69.5312 (68.6488)  acc5: 87.5000 (86.9604)  meanQV: 1.4867 (1.4802)  stdQV: 0.3228 (0.3246)  time: 0.5220  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [2900/5005]  eta: 0:18:19  lr: 0.0004  img/s: 491.0172647329166  loss: 0.8790 (0.8703)  acc1: 68.3594 (68.6341)  acc5: 86.7188 (86.9494)  meanQV: 1.4785 (1.4801)  stdQV: 0.3251 (0.3246)  time: 0.5208  data: 0.0004  max mem: 8962\n",
            "Epoch: [2]  [3000/5005]  eta: 0:17:27  lr: 0.0004  img/s: 485.94684164959335  loss: 0.8755 (0.8706)  acc1: 68.3594 (68.6326)  acc5: 87.5000 (86.9472)  meanQV: 1.4785 (1.4801)  stdQV: 0.3192 (0.3246)  time: 0.5216  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [3100/5005]  eta: 0:16:34  lr: 0.0004  img/s: 494.1287731247124  loss: 0.9319 (0.8717)  acc1: 66.4062 (68.6098)  acc5: 84.7656 (86.9271)  meanQV: 1.4645 (1.4800)  stdQV: 0.3303 (0.3247)  time: 0.5206  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [3200/5005]  eta: 0:15:42  lr: 0.0004  img/s: 489.1026970888204  loss: 0.8573 (0.8715)  acc1: 67.1875 (68.6226)  acc5: 87.1094 (86.9358)  meanQV: 1.4699 (1.4801)  stdQV: 0.3291 (0.3246)  time: 0.5214  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [3300/5005]  eta: 0:14:50  lr: 0.0004  img/s: 493.99282848070345  loss: 0.9075 (0.8719)  acc1: 68.3594 (68.6160)  acc5: 85.9375 (86.9377)  meanQV: 1.4770 (1.4800)  stdQV: 0.3257 (0.3247)  time: 0.5221  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [3400/5005]  eta: 0:13:58  lr: 0.0004  img/s: 491.0915987137075  loss: 0.8952 (0.8724)  acc1: 67.1875 (68.6177)  acc5: 87.5000 (86.9409)  meanQV: 1.4688 (1.4800)  stdQV: 0.3283 (0.3247)  time: 0.5217  data: 0.0004  max mem: 8962\n",
            "Epoch: [2]  [3500/5005]  eta: 0:13:05  lr: 0.0004  img/s: 491.85172372724526  loss: 0.8616 (0.8726)  acc1: 67.9688 (68.6169)  acc5: 87.5000 (86.9428)  meanQV: 1.4758 (1.4800)  stdQV: 0.3262 (0.3247)  time: 0.5212  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [3600/5005]  eta: 0:12:13  lr: 0.0004  img/s: 493.1658175040636  loss: 0.8484 (0.8728)  acc1: 68.7500 (68.6238)  acc5: 88.2812 (86.9481)  meanQV: 1.4813 (1.4801)  stdQV: 0.3237 (0.3246)  time: 0.5210  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [3700/5005]  eta: 0:11:21  lr: 0.0004  img/s: 488.61460107584765  loss: 0.8767 (0.8732)  acc1: 67.9688 (68.6150)  acc5: 86.7188 (86.9436)  meanQV: 1.4758 (1.4800)  stdQV: 0.3240 (0.3247)  time: 0.5216  data: 0.0004  max mem: 8962\n",
            "Epoch: [2]  [3800/5005]  eta: 0:10:29  lr: 0.0004  img/s: 494.11217386622536  loss: 0.8734 (0.8735)  acc1: 66.7969 (68.6138)  acc5: 86.3281 (86.9486)  meanQV: 1.4676 (1.4800)  stdQV: 0.3251 (0.3247)  time: 0.5209  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [3900/5005]  eta: 0:09:36  lr: 0.0004  img/s: 490.99728148911294  loss: 0.8477 (0.8733)  acc1: 69.1406 (68.6243)  acc5: 87.1094 (86.9573)  meanQV: 1.4840 (1.4801)  stdQV: 0.3240 (0.3246)  time: 0.5215  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [4000/5005]  eta: 0:08:44  lr: 0.0004  img/s: 491.4019783456549  loss: 0.8635 (0.8735)  acc1: 68.7500 (68.6289)  acc5: 87.5000 (86.9596)  meanQV: 1.4813 (1.4801)  stdQV: 0.3240 (0.3246)  time: 0.5211  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [4100/5005]  eta: 0:07:52  lr: 0.0004  img/s: 494.1733465144578  loss: 0.8801 (0.8738)  acc1: 68.3594 (68.6256)  acc5: 85.9375 (86.9581)  meanQV: 1.4773 (1.4801)  stdQV: 0.3228 (0.3246)  time: 0.5216  data: 0.0004  max mem: 8962\n",
            "Epoch: [2]  [4200/5005]  eta: 0:07:00  lr: 0.0004  img/s: 489.6234710977453  loss: 0.8273 (0.8740)  acc1: 70.3125 (68.6327)  acc5: 87.8906 (86.9577)  meanQV: 1.4922 (1.4801)  stdQV: 0.3190 (0.3246)  time: 0.5215  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [4300/5005]  eta: 0:06:08  lr: 0.0004  img/s: 494.2581946216013  loss: 0.8741 (0.8740)  acc1: 68.7500 (68.6423)  acc5: 87.5000 (86.9612)  meanQV: 1.4813 (1.4802)  stdQV: 0.3251 (0.3246)  time: 0.5204  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [4400/5005]  eta: 0:05:15  lr: 0.0004  img/s: 745.4293803117672  loss: 0.8636 (0.8744)  acc1: 68.7500 (68.6341)  acc5: 87.1094 (86.9601)  meanQV: 1.4813 (1.4801)  stdQV: 0.3240 (0.3246)  time: 0.4351  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [4500/5005]  eta: 0:04:21  lr: 0.0004  img/s: 486.88907097457  loss: 0.8677 (0.8746)  acc1: 67.9688 (68.6392)  acc5: 86.3281 (86.9591)  meanQV: 1.4758 (1.4802)  stdQV: 0.3262 (0.3246)  time: 0.3115  data: 0.0013  max mem: 8962\n",
            "Epoch: [2]  [4600/5005]  eta: 0:03:29  lr: 0.0004  img/s: 489.810640387527  loss: 0.8474 (0.8748)  acc1: 68.7500 (68.6407)  acc5: 86.7188 (86.9571)  meanQV: 1.4813 (1.4802)  stdQV: 0.3240 (0.3246)  time: 0.5214  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [4700/5005]  eta: 0:02:38  lr: 0.0004  img/s: 494.03101273102885  loss: 0.8843 (0.8749)  acc1: 67.1875 (68.6427)  acc5: 85.9375 (86.9571)  meanQV: 1.4691 (1.4802)  stdQV: 0.3270 (0.3246)  time: 0.5212  data: 0.0004  max mem: 8962\n",
            "Epoch: [2]  [4800/5005]  eta: 0:01:46  lr: 0.0004  img/s: 491.7528352428831  loss: 0.9494 (0.8752)  acc1: 68.3594 (68.6450)  acc5: 85.9375 (86.9551)  meanQV: 1.4785 (1.4802)  stdQV: 0.3251 (0.3246)  time: 0.5206  data: 0.0003  max mem: 8962\n",
            "Epoch: [2]  [4900/5005]  eta: 0:00:54  lr: 0.0004  img/s: 489.231282062264  loss: 0.8820 (0.8755)  acc1: 67.1875 (68.6437)  acc5: 85.9375 (86.9490)  meanQV: 1.4691 (1.4802)  stdQV: 0.3290 (0.3246)  time: 0.5212  data: 0.0004  max mem: 8962\n",
            "Epoch: [2]  [5000/5005]  eta: 0:00:02  lr: 0.0004  img/s: 489.3046302811125  loss: 0.8896 (0.8759)  acc1: 68.3594 (68.6439)  acc5: 86.7188 (86.9525)  meanQV: 1.4785 (1.4802)  stdQV: 0.3248 (0.3246)  time: 0.5207  data: 0.0002  max mem: 8962\n",
            "Epoch: [2] Total time: 0:43:15\n",
            "Test:   [   0/6250]  eta: 1:13:07  loss: 0.9734 (0.9734)  acc1: 62.5000 (62.5000)  acc5: 100.0000 (100.0000)  time: 0.7020  data: 0.6951  max mem: 8962\n",
            "Test:   [ 100/6250]  eta: 0:01:26  loss: 0.1843 (0.6218)  acc1: 87.5000 (83.2921)  acc5: 100.0000 (95.2970)  time: 0.0060  data: 0.0006  max mem: 8962\n",
            "Test:   [ 200/6250]  eta: 0:01:02  loss: 0.7251 (0.6534)  acc1: 87.5000 (84.0174)  acc5: 87.5000 (94.8383)  time: 0.0060  data: 0.0005  max mem: 8962\n",
            "Test:   [ 300/6250]  eta: 0:00:53  loss: 1.2929 (0.8644)  acc1: 62.5000 (78.9037)  acc5: 87.5000 (93.1894)  time: 0.0062  data: 0.0005  max mem: 8962\n",
            "Test:   [ 400/6250]  eta: 0:00:48  loss: 0.8992 (0.9909)  acc1: 75.0000 (76.0910)  acc5: 87.5000 (92.3005)  time: 0.0060  data: 0.0004  max mem: 8962\n",
            "Test:   [ 500/6250]  eta: 0:00:45  loss: 1.1745 (1.0423)  acc1: 75.0000 (74.8503)  acc5: 87.5000 (91.9661)  time: 0.0065  data: 0.0010  max mem: 8962\n",
            "Test:   [ 600/6250]  eta: 0:00:42  loss: 0.4571 (0.9501)  acc1: 87.5000 (76.8095)  acc5: 100.0000 (92.5957)  time: 0.0061  data: 0.0005  max mem: 8962\n",
            "Test:   [ 700/6250]  eta: 0:00:40  loss: 0.4039 (0.9331)  acc1: 87.5000 (77.1220)  acc5: 100.0000 (92.5820)  time: 0.0057  data: 0.0006  max mem: 8962\n",
            "Test:   [ 800/6250]  eta: 0:00:39  loss: 0.7346 (0.9629)  acc1: 75.0000 (76.5605)  acc5: 87.5000 (92.1660)  time: 0.0058  data: 0.0006  max mem: 8962\n",
            "Test:   [ 900/6250]  eta: 0:00:37  loss: 0.3460 (0.9073)  acc1: 87.5000 (77.8857)  acc5: 100.0000 (92.6609)  time: 0.0059  data: 0.0005  max mem: 8962\n",
            "Test:   [1000/6250]  eta: 0:00:36  loss: 0.8425 (0.8922)  acc1: 75.0000 (78.1094)  acc5: 100.0000 (92.8322)  time: 0.0057  data: 0.0008  max mem: 8962\n",
            "Test:   [1100/6250]  eta: 0:00:35  loss: 0.9796 (0.9278)  acc1: 75.0000 (77.2252)  acc5: 100.0000 (92.6544)  time: 0.0073  data: 0.0016  max mem: 8962\n",
            "Test:   [1200/6250]  eta: 0:00:34  loss: 1.1206 (0.9326)  acc1: 75.0000 (76.8110)  acc5: 87.5000 (92.7560)  time: 0.0060  data: 0.0011  max mem: 8962\n",
            "Test:   [1300/6250]  eta: 0:00:33  loss: 0.4536 (0.9371)  acc1: 87.5000 (76.5661)  acc5: 100.0000 (92.8997)  time: 0.0059  data: 0.0005  max mem: 8962\n",
            "Test:   [1400/6250]  eta: 0:00:32  loss: 0.7222 (0.9336)  acc1: 75.0000 (76.6149)  acc5: 87.5000 (92.9425)  time: 0.0059  data: 0.0006  max mem: 8962\n",
            "Test:   [1500/6250]  eta: 0:00:31  loss: 0.9426 (0.9376)  acc1: 62.5000 (76.3658)  acc5: 100.0000 (93.0213)  time: 0.0057  data: 0.0005  max mem: 8962\n",
            "Test:   [1600/6250]  eta: 0:00:30  loss: 0.1496 (0.9321)  acc1: 100.0000 (76.2336)  acc5: 100.0000 (93.1839)  time: 0.0053  data: 0.0005  max mem: 8962\n",
            "Test:   [1700/6250]  eta: 0:00:29  loss: 1.0775 (0.9295)  acc1: 75.0000 (76.0802)  acc5: 100.0000 (93.3128)  time: 0.0051  data: 0.0006  max mem: 8962\n",
            "Test:   [1800/6250]  eta: 0:00:29  loss: 1.1900 (0.9398)  acc1: 62.5000 (75.8537)  acc5: 87.5000 (93.2954)  time: 0.0047  data: 0.0004  max mem: 8962\n",
            "Test:   [1900/6250]  eta: 0:00:28  loss: 0.9377 (0.9324)  acc1: 75.0000 (76.1376)  acc5: 100.0000 (93.4179)  time: 0.0056  data: 0.0005  max mem: 8962\n",
            "Test:   [2000/6250]  eta: 0:00:27  loss: 0.4859 (0.9373)  acc1: 87.5000 (76.0557)  acc5: 100.0000 (93.4158)  time: 0.0057  data: 0.0005  max mem: 8962\n",
            "Test:   [2100/6250]  eta: 0:00:27  loss: 0.3588 (0.9185)  acc1: 87.5000 (76.5826)  acc5: 100.0000 (93.5745)  time: 0.0059  data: 0.0004  max mem: 8962\n",
            "Test:   [2200/6250]  eta: 0:00:26  loss: 0.1941 (0.9123)  acc1: 87.5000 (76.7208)  acc5: 100.0000 (93.6336)  time: 0.0066  data: 0.0011  max mem: 8962\n",
            "Test:   [2300/6250]  eta: 0:00:25  loss: 0.6498 (0.9137)  acc1: 87.5000 (76.6949)  acc5: 100.0000 (93.5952)  time: 0.0052  data: 0.0008  max mem: 8962\n",
            "Test:   [2400/6250]  eta: 0:00:25  loss: 0.9289 (0.9226)  acc1: 75.0000 (76.5723)  acc5: 87.5000 (93.5079)  time: 0.0067  data: 0.0004  max mem: 8962\n",
            "Test:   [2500/6250]  eta: 0:00:24  loss: 0.9120 (0.9237)  acc1: 87.5000 (76.6593)  acc5: 100.0000 (93.4726)  time: 0.0054  data: 0.0004  max mem: 8962\n",
            "Test:   [2600/6250]  eta: 0:00:23  loss: 2.2991 (0.9477)  acc1: 50.0000 (76.1774)  acc5: 62.5000 (93.1565)  time: 0.0057  data: 0.0005  max mem: 8962\n",
            "Test:   [2700/6250]  eta: 0:00:23  loss: 0.7023 (0.9584)  acc1: 75.0000 (76.0089)  acc5: 100.0000 (93.0442)  time: 0.0061  data: 0.0007  max mem: 8962\n",
            "Test:   [2800/6250]  eta: 0:00:22  loss: 1.6450 (0.9808)  acc1: 62.5000 (75.5400)  acc5: 87.5000 (92.7749)  time: 0.0056  data: 0.0007  max mem: 8962\n",
            "Test:   [2900/6250]  eta: 0:00:21  loss: 1.8880 (1.0003)  acc1: 50.0000 (75.1034)  acc5: 87.5000 (92.5414)  time: 0.0057  data: 0.0005  max mem: 8962\n",
            "Test:   [3000/6250]  eta: 0:00:21  loss: 2.2391 (1.0214)  acc1: 50.0000 (74.7959)  acc5: 75.0000 (92.2193)  time: 0.0051  data: 0.0004  max mem: 8962\n",
            "Test:   [3100/6250]  eta: 0:00:20  loss: 1.8099 (1.0473)  acc1: 50.0000 (74.2785)  acc5: 75.0000 (91.9502)  time: 0.0058  data: 0.0006  max mem: 8962\n",
            "Test:   [3200/6250]  eta: 0:00:19  loss: 0.7831 (1.0671)  acc1: 62.5000 (73.8558)  acc5: 100.0000 (91.7565)  time: 0.0058  data: 0.0009  max mem: 8962\n",
            "Test:   [3300/6250]  eta: 0:00:18  loss: 1.2720 (1.0789)  acc1: 62.5000 (73.5497)  acc5: 87.5000 (91.6540)  time: 0.0060  data: 0.0010  max mem: 8962\n",
            "Test:   [3400/6250]  eta: 0:00:18  loss: 2.3960 (1.0930)  acc1: 37.5000 (73.2946)  acc5: 75.0000 (91.4363)  time: 0.0070  data: 0.0010  max mem: 8962\n",
            "Test:   [3500/6250]  eta: 0:00:17  loss: 1.8152 (1.0998)  acc1: 50.0000 (73.1612)  acc5: 75.0000 (91.3203)  time: 0.0075  data: 0.0021  max mem: 8962\n",
            "Test:   [3600/6250]  eta: 0:00:17  loss: 0.4449 (1.0965)  acc1: 87.5000 (73.3026)  acc5: 100.0000 (91.3219)  time: 0.0058  data: 0.0004  max mem: 8962\n",
            "Test:   [3700/6250]  eta: 0:00:16  loss: 1.6940 (1.1099)  acc1: 62.5000 (73.0276)  acc5: 87.5000 (91.1477)  time: 0.0131  data: 0.0075  max mem: 8962\n",
            "Test:   [3800/6250]  eta: 0:00:15  loss: 0.7985 (1.1155)  acc1: 87.5000 (72.9644)  acc5: 87.5000 (91.0484)  time: 0.0054  data: 0.0006  max mem: 8962\n",
            "Test:   [3900/6250]  eta: 0:00:15  loss: 2.2583 (1.1305)  acc1: 50.0000 (72.6833)  acc5: 75.0000 (90.8293)  time: 0.0053  data: 0.0004  max mem: 8962\n",
            "Test:   [4000/6250]  eta: 0:00:14  loss: 1.4313 (1.1455)  acc1: 62.5000 (72.4256)  acc5: 87.5000 (90.6492)  time: 0.0056  data: 0.0005  max mem: 8962\n",
            "Test:   [4100/6250]  eta: 0:00:13  loss: 1.7358 (1.1560)  acc1: 62.5000 (72.2507)  acc5: 87.5000 (90.5389)  time: 0.0056  data: 0.0004  max mem: 8962\n",
            "Test:   [4200/6250]  eta: 0:00:13  loss: 0.4747 (1.1604)  acc1: 87.5000 (72.1168)  acc5: 100.0000 (90.5380)  time: 0.0064  data: 0.0013  max mem: 8962\n",
            "Test:   [4300/6250]  eta: 0:00:12  loss: 0.5343 (1.1704)  acc1: 75.0000 (71.9513)  acc5: 87.5000 (90.3743)  time: 0.0054  data: 0.0004  max mem: 8962\n",
            "Test:   [4400/6250]  eta: 0:00:11  loss: 0.8444 (1.1773)  acc1: 75.0000 (71.7422)  acc5: 100.0000 (90.3175)  time: 0.0050  data: 0.0003  max mem: 8962\n",
            "Test:   [4500/6250]  eta: 0:00:11  loss: 1.2144 (1.1832)  acc1: 62.5000 (71.6119)  acc5: 87.5000 (90.2744)  time: 0.0060  data: 0.0006  max mem: 8962\n",
            "Test:   [4600/6250]  eta: 0:00:10  loss: 1.7535 (1.1929)  acc1: 50.0000 (71.4600)  acc5: 87.5000 (90.1190)  time: 0.0061  data: 0.0005  max mem: 8962\n",
            "Test:   [4700/6250]  eta: 0:00:09  loss: 1.4817 (1.2036)  acc1: 62.5000 (71.1498)  acc5: 87.5000 (89.9702)  time: 0.0050  data: 0.0006  max mem: 8962\n",
            "Test:   [4800/6250]  eta: 0:00:09  loss: 1.3074 (1.2111)  acc1: 62.5000 (71.0034)  acc5: 87.5000 (89.8563)  time: 0.0054  data: 0.0005  max mem: 8962\n",
            "Test:   [4900/6250]  eta: 0:00:08  loss: 0.4546 (1.2157)  acc1: 75.0000 (70.9090)  acc5: 100.0000 (89.7852)  time: 0.0053  data: 0.0006  max mem: 8962\n",
            "Test:   [5000/6250]  eta: 0:00:07  loss: 2.0130 (1.2274)  acc1: 62.5000 (70.7359)  acc5: 75.0000 (89.6296)  time: 0.0059  data: 0.0004  max mem: 8962\n",
            "Test:   [5100/6250]  eta: 0:00:07  loss: 1.3178 (1.2327)  acc1: 62.5000 (70.6136)  acc5: 87.5000 (89.5903)  time: 0.0061  data: 0.0005  max mem: 8962\n",
            "Test:   [5200/6250]  eta: 0:00:06  loss: 1.0106 (1.2388)  acc1: 75.0000 (70.5177)  acc5: 87.5000 (89.5236)  time: 0.0052  data: 0.0004  max mem: 8962\n",
            "Test:   [5300/6250]  eta: 0:00:05  loss: 1.5488 (1.2526)  acc1: 62.5000 (70.2297)  acc5: 87.5000 (89.3298)  time: 0.0053  data: 0.0005  max mem: 8962\n",
            "Test:   [5400/6250]  eta: 0:00:05  loss: 0.8032 (1.2557)  acc1: 75.0000 (70.1768)  acc5: 87.5000 (89.2728)  time: 0.0058  data: 0.0004  max mem: 8962\n",
            "Test:   [5500/6250]  eta: 0:00:04  loss: 0.8721 (1.2595)  acc1: 75.0000 (70.0827)  acc5: 87.5000 (89.2201)  time: 0.0061  data: 0.0006  max mem: 8962\n",
            "Test:   [5600/6250]  eta: 0:00:04  loss: 1.0362 (1.2646)  acc1: 62.5000 (70.0009)  acc5: 87.5000 (89.1269)  time: 0.0053  data: 0.0004  max mem: 8962\n",
            "Test:   [5700/6250]  eta: 0:00:03  loss: 2.2192 (1.2800)  acc1: 37.5000 (69.7049)  acc5: 75.0000 (88.9756)  time: 0.0052  data: 0.0005  max mem: 8962\n",
            "Test:   [5800/6250]  eta: 0:00:02  loss: 0.5529 (1.2755)  acc1: 75.0000 (69.7897)  acc5: 100.0000 (89.0321)  time: 0.0052  data: 0.0005  max mem: 8962\n",
            "Test:   [5900/6250]  eta: 0:00:02  loss: 1.1504 (1.2730)  acc1: 75.0000 (69.8187)  acc5: 87.5000 (89.0612)  time: 0.0059  data: 0.0006  max mem: 8962\n",
            "Test:   [6000/6250]  eta: 0:00:01  loss: 0.5121 (1.2652)  acc1: 87.5000 (69.9779)  acc5: 87.5000 (89.1518)  time: 0.0060  data: 0.0005  max mem: 8962\n",
            "Test:   [6100/6250]  eta: 0:00:00  loss: 1.3071 (1.2734)  acc1: 62.5000 (69.7959)  acc5: 87.5000 (89.0407)  time: 0.0048  data: 0.0004  max mem: 8962\n",
            "Test:   [6200/6250]  eta: 0:00:00  loss: 0.2227 (1.2671)  acc1: 100.0000 (69.9161)  acc5: 100.0000 (89.1207)  time: 0.0062  data: 0.0005  max mem: 8962\n",
            "Test:  Total time: 0:00:38\n",
            "Test:  Acc@1 69.952 Acc@5 89.136\n",
            "Epoch: [3]  [   0/5005]  eta: 5:49:16  lr: 0.0002  img/s: 494.13332106139956  loss: 0.8680 (0.8680)  acc1: 68.3594 (68.3594)  acc5: 87.8906 (87.8906)  meanQV: 1.4785 (1.4785)  stdQV: 0.3262 (0.3262)  time: 4.1871  data: 3.6690  max mem: 8962\n",
            "Epoch: [3]  [ 100/5005]  eta: 0:45:34  lr: 0.0002  img/s: 493.44095322259034  loss: 0.8700 (0.8771)  acc1: 69.5312 (68.9124)  acc5: 87.1094 (87.1906)  meanQV: 1.4867 (1.4820)  stdQV: 0.3216 (0.3238)  time: 0.5208  data: 0.0004  max mem: 8962\n",
            "Epoch: [3]  [ 200/5005]  eta: 0:43:11  lr: 0.0002  img/s: 491.6118921765009  loss: 0.8557 (0.8772)  acc1: 69.1406 (69.1056)  acc5: 87.1094 (87.0666)  meanQV: 1.4840 (1.4834)  stdQV: 0.3228 (0.3233)  time: 0.5215  data: 0.0003  max mem: 8962\n",
            "Epoch: [3]  [ 300/5005]  eta: 0:41:50  lr: 0.0002  img/s: 488.24378408069873  loss: 0.8735 (0.8765)  acc1: 68.3594 (69.1173)  acc5: 87.1094 (87.2859)  meanQV: 1.4773 (1.4834)  stdQV: 0.3259 (0.3233)  time: 0.5215  data: 0.0003  max mem: 8962\n",
            "Epoch: [3]  [ 400/5005]  eta: 0:40:42  lr: 0.0002  img/s: 491.1491050840666  loss: 0.8561 (0.8749)  acc1: 69.1406 (69.1338)  acc5: 87.1094 (87.3091)  meanQV: 1.4816 (1.4835)  stdQV: 0.3235 (0.3232)  time: 0.5215  data: 0.0004  max mem: 8962\n",
            "Epoch: [3]  [ 500/5005]  eta: 0:39:40  lr: 0.0002  img/s: 494.98206480200105  loss: 0.8668 (0.8747)  acc1: 68.3594 (69.2046)  acc5: 87.5000 (87.3082)  meanQV: 1.4785 (1.4840)  stdQV: 0.3228 (0.3230)  time: 0.5210  data: 0.0003  max mem: 8962\n",
            "Epoch: [3]  [ 600/5005]  eta: 0:38:42  lr: 0.0002  img/s: 492.9600064641953  loss: 0.8776 (0.8775)  acc1: 69.1406 (69.1595)  acc5: 87.5000 (87.2979)  meanQV: 1.4840 (1.4837)  stdQV: 0.3226 (0.3231)  time: 0.5206  data: 0.0004  max mem: 8962\n",
            "Epoch: [3]  [ 700/5005]  eta: 0:37:46  lr: 0.0002  img/s: 488.8984818192125  loss: 0.8980 (0.8788)  acc1: 69.1406 (69.1512)  acc5: 86.3281 (87.2465)  meanQV: 1.4836 (1.4837)  stdQV: 0.3228 (0.3231)  time: 0.5213  data: 0.0003  max mem: 8962\n",
            "Epoch: [3]  [ 800/5005]  eta: 0:36:50  lr: 0.0002  img/s: 489.088661574488  loss: 0.8472 (0.8815)  acc1: 68.7500 (69.0777)  acc5: 86.7188 (87.1776)  meanQV: 1.4801 (1.4831)  stdQV: 0.3248 (0.3233)  time: 0.5208  data: 0.0003  max mem: 8962\n",
            "Epoch: [3]  [ 900/5005]  eta: 0:35:56  lr: 0.0002  img/s: 491.4127733969181  loss: 0.8806 (0.8807)  acc1: 68.3594 (69.0921)  acc5: 86.7188 (87.1657)  meanQV: 1.4762 (1.4832)  stdQV: 0.3262 (0.3233)  time: 0.5213  data: 0.0004  max mem: 8962\n",
            "Epoch: [3]  [1000/5005]  eta: 0:35:01  lr: 0.0002  img/s: 491.22662521193246  loss: 0.9323 (0.8815)  acc1: 67.9688 (69.0762)  acc5: 86.7188 (87.1488)  meanQV: 1.4758 (1.4831)  stdQV: 0.3273 (0.3233)  time: 0.5216  data: 0.0003  max mem: 8962\n",
            "Epoch: [3]  [1100/5005]  eta: 0:34:08  lr: 0.0002  img/s: 490.98717822090043  loss: 0.8727 (0.8828)  acc1: 69.1406 (69.0541)  acc5: 87.5000 (87.1292)  meanQV: 1.4828 (1.4830)  stdQV: 0.3237 (0.3234)  time: 0.5207  data: 0.0003  max mem: 8962\n",
            "Epoch: [3]  [1200/5005]  eta: 0:33:14  lr: 0.0002  img/s: 489.31912417248043  loss: 0.8732 (0.8839)  acc1: 68.7500 (69.0385)  acc5: 86.3281 (87.0990)  meanQV: 1.4801 (1.4829)  stdQV: 0.3248 (0.3234)  time: 0.5208  data: 0.0004  max mem: 8962\n",
            "Epoch: [3]  [1300/5005]  eta: 0:32:21  lr: 0.0002  img/s: 488.60570732215643  loss: 0.8841 (0.8833)  acc1: 69.1406 (69.0139)  acc5: 87.1094 (87.1034)  meanQV: 1.4840 (1.4827)  stdQV: 0.3228 (0.3235)  time: 0.5211  data: 0.0004  max mem: 8962\n",
            "Epoch: [3]  [1400/5005]  eta: 0:31:27  lr: 0.0002  img/s: 489.7130175490766  loss: 0.8099 (0.8833)  acc1: 69.9219 (69.0252)  acc5: 87.1094 (87.0971)  meanQV: 1.4895 (1.4828)  stdQV: 0.3188 (0.3235)  time: 0.5214  data: 0.0003  max mem: 8962\n",
            "Epoch: [3]  [1500/5005]  eta: 0:30:35  lr: 0.0002  img/s: 490.8552003847328  loss: 0.8780 (0.8834)  acc1: 69.1406 (69.0147)  acc5: 86.7188 (87.0748)  meanQV: 1.4840 (1.4827)  stdQV: 0.3228 (0.3235)  time: 0.5206  data: 0.0003  max mem: 8962\n",
            "Epoch: [3]  [1600/5005]  eta: 0:29:42  lr: 0.0002  img/s: 492.7211280428486  loss: 0.9144 (0.8839)  acc1: 69.1406 (69.0086)  acc5: 86.7188 (87.0847)  meanQV: 1.4828 (1.4827)  stdQV: 0.3237 (0.3235)  time: 0.5212  data: 0.0003  max mem: 8962\n",
            "Epoch: [3]  [1700/5005]  eta: 0:28:49  lr: 0.0002  img/s: 494.0639741221239  loss: 0.8737 (0.8834)  acc1: 69.5312 (69.0187)  acc5: 86.3281 (87.0809)  meanQV: 1.4867 (1.4827)  stdQV: 0.3228 (0.3235)  time: 0.5216  data: 0.0004  max mem: 8962\n",
            "Epoch: [3]  [1800/5005]  eta: 0:27:56  lr: 0.0002  img/s: 494.14514608859565  loss: 0.9103 (0.8836)  acc1: 67.5781 (69.0094)  acc5: 87.1094 (87.0679)  meanQV: 1.4730 (1.4827)  stdQV: 0.3283 (0.3235)  time: 0.5212  data: 0.0004  max mem: 8962\n",
            "Epoch: [3]  [1900/5005]  eta: 0:27:04  lr: 0.0002  img/s: 488.1301411462633  loss: 0.8608 (0.8841)  acc1: 69.1406 (68.9943)  acc5: 86.7188 (87.0798)  meanQV: 1.4840 (1.4826)  stdQV: 0.3228 (0.3236)  time: 0.5210  data: 0.0004  max mem: 8962\n",
            "Epoch: [3]  [2000/5005]  eta: 0:26:11  lr: 0.0002  img/s: 493.9826015587656  loss: 0.9005 (0.8840)  acc1: 69.1406 (69.0001)  acc5: 87.1094 (87.0899)  meanQV: 1.4840 (1.4826)  stdQV: 0.3216 (0.3236)  time: 0.5209  data: 0.0003  max mem: 8962\n",
            "Epoch: [3]  [2100/5005]  eta: 0:25:18  lr: 0.0002  img/s: 491.0731815518808  loss: 0.8610 (0.8835)  acc1: 70.7031 (69.0287)  acc5: 87.8906 (87.0908)  meanQV: 1.4949 (1.4828)  stdQV: 0.3180 (0.3235)  time: 0.5210  data: 0.0005  max mem: 8962\n",
            "Epoch: [3]  [2200/5005]  eta: 0:24:26  lr: 0.0002  img/s: 486.65824433670525  loss: 0.8879 (0.8840)  acc1: 67.1875 (69.0139)  acc5: 87.1094 (87.0943)  meanQV: 1.4703 (1.4827)  stdQV: 0.3262 (0.3235)  time: 0.5212  data: 0.0004  max mem: 8962\n",
            "Epoch: [3]  [2300/5005]  eta: 0:23:33  lr: 0.0002  img/s: 489.28055002159914  loss: 0.8677 (0.8844)  acc1: 69.1406 (69.0091)  acc5: 88.6719 (87.0997)  meanQV: 1.4840 (1.4826)  stdQV: 0.3192 (0.3235)  time: 0.5220  data: 0.0004  max mem: 8962\n",
            "Epoch: [3]  [2400/5005]  eta: 0:22:41  lr: 0.0002  img/s: 489.7322262906224  loss: 0.8478 (0.8846)  acc1: 69.5312 (68.9997)  acc5: 88.2812 (87.0944)  meanQV: 1.4867 (1.4826)  stdQV: 0.3226 (0.3236)  time: 0.5206  data: 0.0003  max mem: 8962\n",
            "Epoch: [3]  [2500/5005]  eta: 0:21:49  lr: 0.0002  img/s: 494.49583330838766  loss: 0.9224 (0.8849)  acc1: 68.7500 (68.9983)  acc5: 86.3281 (87.0841)  meanQV: 1.4813 (1.4826)  stdQV: 0.3240 (0.3236)  time: 0.5211  data: 0.0003  max mem: 8962\n",
            "Epoch: [3]  [2600/5005]  eta: 0:20:56  lr: 0.0002  img/s: 491.60829085668  loss: 0.8767 (0.8851)  acc1: 69.9219 (69.0107)  acc5: 87.8906 (87.0868)  meanQV: 1.4883 (1.4826)  stdQV: 0.3214 (0.3235)  time: 0.5205  data: 0.0004  max mem: 8962\n",
            "Epoch: [3]  [2700/5005]  eta: 0:20:04  lr: 0.0002  img/s: 494.5233905012099  loss: 0.8684 (0.8851)  acc1: 69.5312 (69.0087)  acc5: 87.5000 (87.0942)  meanQV: 1.4855 (1.4826)  stdQV: 0.3216 (0.3235)  time: 0.5203  data: 0.0004  max mem: 8962\n",
            "Epoch: [3]  [2800/5005]  eta: 0:19:11  lr: 0.0002  img/s: 491.0152438815262  loss: 0.9061 (0.8855)  acc1: 69.1406 (68.9994)  acc5: 86.7188 (87.0876)  meanQV: 1.4840 (1.4826)  stdQV: 0.3228 (0.3236)  time: 0.5209  data: 0.0003  max mem: 8962\n",
            "Epoch: [3]  [2900/5005]  eta: 0:18:19  lr: 0.0002  img/s: 490.7789190394875  loss: 0.8594 (0.8853)  acc1: 69.1406 (68.9987)  acc5: 87.5000 (87.0913)  meanQV: 1.4820 (1.4826)  stdQV: 0.3215 (0.3236)  time: 0.5220  data: 0.0003  max mem: 8962\n",
            "Epoch: [3]  [3000/5005]  eta: 0:17:27  lr: 0.0002  img/s: 490.72441483952787  loss: 0.8432 (0.8854)  acc1: 69.5312 (68.9989)  acc5: 87.8906 (87.0884)  meanQV: 1.4867 (1.4826)  stdQV: 0.3202 (0.3236)  time: 0.5214  data: 0.0004  max mem: 8962\n",
            "Epoch: [3]  [3100/5005]  eta: 0:16:34  lr: 0.0002  img/s: 491.1839299405131  loss: 0.8755 (0.8852)  acc1: 70.3125 (69.0101)  acc5: 87.5000 (87.0990)  meanQV: 1.4898 (1.4826)  stdQV: 0.3192 (0.3235)  time: 0.5210  data: 0.0004  max mem: 8962\n",
            "Epoch: [3]  [3200/5005]  eta: 0:15:42  lr: 0.0002  img/s: 491.87718437506726  loss: 0.8693 (0.8855)  acc1: 69.5312 (69.0059)  acc5: 86.3281 (87.0879)  meanQV: 1.4863 (1.4826)  stdQV: 0.3216 (0.3235)  time: 0.5207  data: 0.0004  max mem: 8962\n",
            "Epoch: [3]  [3300/5005]  eta: 0:14:50  lr: 0.0002  img/s: 489.23663194863695  loss: 0.8853 (0.8859)  acc1: 69.9219 (69.0043)  acc5: 85.9375 (87.0807)  meanQV: 1.4871 (1.4826)  stdQV: 0.3202 (0.3235)  time: 0.5208  data: 0.0004  max mem: 8962\n",
            "Epoch: [3]  [3400/5005]  eta: 0:13:58  lr: 0.0002  img/s: 487.7279331806508  loss: 0.8678 (0.8862)  acc1: 68.3594 (68.9936)  acc5: 86.3281 (87.0718)  meanQV: 1.4785 (1.4825)  stdQV: 0.3262 (0.3236)  time: 0.5219  data: 0.0004  max mem: 8962\n",
            "Epoch: [3]  [3500/5005]  eta: 0:13:05  lr: 0.0002  img/s: 492.66031437817423  loss: 0.8921 (0.8862)  acc1: 68.3594 (68.9935)  acc5: 86.3281 (87.0740)  meanQV: 1.4781 (1.4825)  stdQV: 0.3260 (0.3236)  time: 0.5205  data: 0.0003  max mem: 8962\n",
            "Epoch: [3]  [3600/5005]  eta: 0:12:13  lr: 0.0002  img/s: 491.45280918624405  loss: 0.8688 (0.8865)  acc1: 69.5312 (68.9917)  acc5: 87.1094 (87.0669)  meanQV: 1.4855 (1.4825)  stdQV: 0.3226 (0.3236)  time: 0.5213  data: 0.0003  max mem: 8962\n",
            "Epoch: [3]  [3700/5005]  eta: 0:11:21  lr: 0.0002  img/s: 491.6204455225456  loss: 0.8652 (0.8867)  acc1: 68.3594 (68.9841)  acc5: 86.3281 (87.0616)  meanQV: 1.4773 (1.4824)  stdQV: 0.3262 (0.3236)  time: 0.5209  data: 0.0003  max mem: 8962\n",
            "Epoch: [3]  [3800/5005]  eta: 0:10:29  lr: 0.0002  img/s: 487.71309034173106  loss: 0.8378 (0.8872)  acc1: 69.1406 (68.9696)  acc5: 86.7188 (87.0521)  meanQV: 1.4840 (1.4823)  stdQV: 0.3228 (0.3236)  time: 0.5206  data: 0.0004  max mem: 8962\n",
            "Epoch: [3]  [3900/5005]  eta: 0:09:36  lr: 0.0002  img/s: 493.498784340653  loss: 0.9257 (0.8877)  acc1: 68.3594 (68.9631)  acc5: 87.1094 (87.0480)  meanQV: 1.4785 (1.4823)  stdQV: 0.3240 (0.3237)  time: 0.5218  data: 0.0003  max mem: 8962\n",
            "Epoch: [3]  [4000/5005]  eta: 0:08:44  lr: 0.0002  img/s: 494.29778520681486  loss: 0.8728 (0.8877)  acc1: 68.3594 (68.9672)  acc5: 87.5000 (87.0530)  meanQV: 1.4785 (1.4823)  stdQV: 0.3240 (0.3236)  time: 0.5206  data: 0.0003  max mem: 8962\n",
            "Epoch: [3]  [4100/5005]  eta: 0:07:52  lr: 0.0002  img/s: 491.56395393020676  loss: 0.9223 (0.8879)  acc1: 68.7500 (68.9693)  acc5: 86.7188 (87.0528)  meanQV: 1.4789 (1.4823)  stdQV: 0.3249 (0.3236)  time: 0.5214  data: 0.0003  max mem: 8962\n",
            "Epoch: [3]  [4200/5005]  eta: 0:07:00  lr: 0.0002  img/s: 489.26494371188943  loss: 0.8813 (0.8883)  acc1: 67.5781 (68.9555)  acc5: 86.3281 (87.0487)  meanQV: 1.4730 (1.4822)  stdQV: 0.3273 (0.3237)  time: 0.5211  data: 0.0003  max mem: 8962\n",
            "Epoch: [3]  [4300/5005]  eta: 0:06:07  lr: 0.0002  img/s: 493.6955115781962  loss: 0.8671 (0.8883)  acc1: 70.7031 (68.9589)  acc5: 86.3281 (87.0572)  meanQV: 1.4949 (1.4823)  stdQV: 0.3192 (0.3237)  time: 0.5204  data: 0.0003  max mem: 8962\n",
            "Epoch: [3]  [4400/5005]  eta: 0:05:15  lr: 0.0002  img/s: 493.47927899772964  loss: 0.9207 (0.8885)  acc1: 68.3594 (68.9580)  acc5: 85.9375 (87.0487)  meanQV: 1.4785 (1.4823)  stdQV: 0.3262 (0.3237)  time: 0.5198  data: 0.0003  max mem: 8962\n",
            "Epoch: [3]  [4500/5005]  eta: 0:04:23  lr: 0.0002  img/s: 492.8457412066241  loss: 0.8825 (0.8885)  acc1: 69.5312 (68.9623)  acc5: 87.1094 (87.0524)  meanQV: 1.4867 (1.4823)  stdQV: 0.3228 (0.3237)  time: 0.5207  data: 0.0004  max mem: 8962\n",
            "Epoch: [3]  [4600/5005]  eta: 0:03:31  lr: 0.0002  img/s: 491.5558526149826  loss: 0.8631 (0.8888)  acc1: 68.7500 (68.9587)  acc5: 86.7188 (87.0511)  meanQV: 1.4813 (1.4823)  stdQV: 0.3240 (0.3237)  time: 0.5209  data: 0.0003  max mem: 8962\n",
            "Epoch: [3]  [4700/5005]  eta: 0:02:39  lr: 0.0002  img/s: 491.1044017010751  loss: 0.9060 (0.8890)  acc1: 67.9688 (68.9607)  acc5: 86.7188 (87.0516)  meanQV: 1.4754 (1.4823)  stdQV: 0.3262 (0.3237)  time: 0.5205  data: 0.0003  max mem: 8962\n",
            "Epoch: [3]  [4800/5005]  eta: 0:01:46  lr: 0.0002  img/s: 488.3679195209023  loss: 0.8826 (0.8891)  acc1: 70.3125 (68.9605)  acc5: 87.5000 (87.0509)  meanQV: 1.4922 (1.4823)  stdQV: 0.3204 (0.3237)  time: 0.5207  data: 0.0003  max mem: 8962\n",
            "Epoch: [3]  [4900/5005]  eta: 0:00:54  lr: 0.0002  img/s: 657.254661723217  loss: 0.8709 (0.8890)  acc1: 69.1406 (68.9624)  acc5: 87.1094 (87.0525)  meanQV: 1.4840 (1.4823)  stdQV: 0.3208 (0.3237)  time: 0.3795  data: 0.0027  max mem: 8962\n",
            "Epoch: [3]  [5000/5005]  eta: 0:00:02  lr: 0.0002  img/s: 489.617889485732  loss: 0.8860 (0.8894)  acc1: 67.5781 (68.9478)  acc5: 86.3281 (87.0502)  meanQV: 1.4730 (1.4822)  stdQV: 0.3259 (0.3237)  time: 0.5212  data: 0.0002  max mem: 8962\n",
            "Epoch: [3] Total time: 0:43:15\n",
            "Test:   [   0/6250]  eta: 1:13:37  loss: 0.8178 (0.8178)  acc1: 62.5000 (62.5000)  acc5: 100.0000 (100.0000)  time: 0.7068  data: 0.7000  max mem: 8962\n",
            "Test:   [ 100/6250]  eta: 0:01:28  loss: 0.1440 (0.6006)  acc1: 87.5000 (83.4158)  acc5: 100.0000 (95.4208)  time: 0.0070  data: 0.0009  max mem: 8962\n",
            "Test:   [ 200/6250]  eta: 0:01:06  loss: 0.6774 (0.6502)  acc1: 87.5000 (84.2040)  acc5: 100.0000 (94.8383)  time: 0.0073  data: 0.0006  max mem: 8962\n",
            "Test:   [ 300/6250]  eta: 0:00:56  loss: 1.2668 (0.8598)  acc1: 62.5000 (79.2774)  acc5: 87.5000 (93.1478)  time: 0.0066  data: 0.0007  max mem: 8962\n",
            "Test:   [ 400/6250]  eta: 0:00:51  loss: 0.8958 (0.9858)  acc1: 75.0000 (76.1845)  acc5: 100.0000 (92.2693)  time: 0.0062  data: 0.0008  max mem: 8962\n",
            "Test:   [ 500/6250]  eta: 0:00:48  loss: 1.1632 (1.0374)  acc1: 75.0000 (74.9002)  acc5: 87.5000 (91.9411)  time: 0.0061  data: 0.0007  max mem: 8962\n",
            "Test:   [ 600/6250]  eta: 0:00:45  loss: 0.3993 (0.9455)  acc1: 87.5000 (76.9343)  acc5: 100.0000 (92.6165)  time: 0.0064  data: 0.0006  max mem: 8962\n",
            "Test:   [ 700/6250]  eta: 0:00:43  loss: 0.4326 (0.9321)  acc1: 87.5000 (77.1576)  acc5: 100.0000 (92.6712)  time: 0.0067  data: 0.0006  max mem: 8962\n",
            "Test:   [ 800/6250]  eta: 0:00:42  loss: 0.8489 (0.9613)  acc1: 75.0000 (76.5918)  acc5: 87.5000 (92.2441)  time: 0.0066  data: 0.0008  max mem: 8962\n",
            "Test:   [ 900/6250]  eta: 0:00:41  loss: 0.3832 (0.9033)  acc1: 87.5000 (77.9550)  acc5: 100.0000 (92.7303)  time: 0.0068  data: 0.0005  max mem: 8962\n",
            "Test:   [1000/6250]  eta: 0:00:39  loss: 0.9345 (0.8886)  acc1: 75.0000 (78.1094)  acc5: 100.0000 (92.9196)  time: 0.0061  data: 0.0009  max mem: 8962\n",
            "Test:   [1100/6250]  eta: 0:00:38  loss: 1.0317 (0.9263)  acc1: 75.0000 (77.0777)  acc5: 100.0000 (92.7679)  time: 0.0071  data: 0.0008  max mem: 8962\n",
            "Test:   [1200/6250]  eta: 0:00:37  loss: 1.0881 (0.9341)  acc1: 75.0000 (76.5820)  acc5: 87.5000 (92.8393)  time: 0.0069  data: 0.0007  max mem: 8962\n",
            "Test:   [1300/6250]  eta: 0:00:36  loss: 0.4206 (0.9358)  acc1: 75.0000 (76.3259)  acc5: 100.0000 (93.0150)  time: 0.0074  data: 0.0009  max mem: 8962\n",
            "Test:   [1400/6250]  eta: 0:00:35  loss: 0.8187 (0.9322)  acc1: 75.0000 (76.4632)  acc5: 87.5000 (93.0496)  time: 0.0061  data: 0.0009  max mem: 8962\n",
            "Test:   [1500/6250]  eta: 0:00:34  loss: 0.8115 (0.9372)  acc1: 75.0000 (76.2408)  acc5: 100.0000 (93.1379)  time: 0.0061  data: 0.0006  max mem: 8962\n",
            "Test:   [1600/6250]  eta: 0:00:33  loss: 0.1571 (0.9323)  acc1: 100.0000 (76.1165)  acc5: 100.0000 (93.2933)  time: 0.0064  data: 0.0006  max mem: 8962\n",
            "Test:   [1700/6250]  eta: 0:00:32  loss: 0.9674 (0.9303)  acc1: 75.0000 (76.0068)  acc5: 100.0000 (93.3936)  time: 0.0071  data: 0.0007  max mem: 8962\n",
            "Test:   [1800/6250]  eta: 0:00:32  loss: 1.2235 (0.9404)  acc1: 62.5000 (75.7843)  acc5: 87.5000 (93.3717)  time: 0.0083  data: 0.0008  max mem: 8962\n",
            "Test:   [1900/6250]  eta: 0:00:31  loss: 1.0283 (0.9336)  acc1: 62.5000 (76.0652)  acc5: 100.0000 (93.4903)  time: 0.0054  data: 0.0006  max mem: 8962\n",
            "Test:   [2000/6250]  eta: 0:00:30  loss: 0.4590 (0.9386)  acc1: 87.5000 (76.0057)  acc5: 100.0000 (93.4533)  time: 0.0056  data: 0.0009  max mem: 8962\n",
            "Test:   [2100/6250]  eta: 0:00:29  loss: 0.3773 (0.9199)  acc1: 87.5000 (76.5469)  acc5: 100.0000 (93.6042)  time: 0.0062  data: 0.0007  max mem: 8962\n",
            "Test:   [2200/6250]  eta: 0:00:28  loss: 0.2202 (0.9123)  acc1: 87.5000 (76.7208)  acc5: 100.0000 (93.6733)  time: 0.0054  data: 0.0006  max mem: 8962\n",
            "Test:   [2300/6250]  eta: 0:00:27  loss: 0.6553 (0.9135)  acc1: 87.5000 (76.7058)  acc5: 100.0000 (93.6821)  time: 0.0067  data: 0.0020  max mem: 8962\n",
            "Test:   [2400/6250]  eta: 0:00:27  loss: 0.8205 (0.9230)  acc1: 75.0000 (76.5775)  acc5: 87.5000 (93.5964)  time: 0.0078  data: 0.0017  max mem: 8962\n",
            "Test:   [2500/6250]  eta: 0:00:26  loss: 0.7730 (0.9230)  acc1: 87.5000 (76.6693)  acc5: 100.0000 (93.5726)  time: 0.0055  data: 0.0006  max mem: 8962\n",
            "Test:   [2600/6250]  eta: 0:00:25  loss: 2.1682 (0.9463)  acc1: 50.0000 (76.2351)  acc5: 75.0000 (93.2526)  time: 0.0053  data: 0.0004  max mem: 8962\n",
            "Test:   [2700/6250]  eta: 0:00:24  loss: 0.8635 (0.9574)  acc1: 75.0000 (76.0459)  acc5: 87.5000 (93.1137)  time: 0.0055  data: 0.0005  max mem: 8962\n",
            "Test:   [2800/6250]  eta: 0:00:23  loss: 1.4083 (0.9782)  acc1: 62.5000 (75.6025)  acc5: 87.5000 (92.8865)  time: 0.0055  data: 0.0004  max mem: 8962\n",
            "Test:   [2900/6250]  eta: 0:00:22  loss: 1.8681 (0.9967)  acc1: 37.5000 (75.2025)  acc5: 87.5000 (92.6706)  time: 0.0063  data: 0.0009  max mem: 8962\n",
            "Test:   [3000/6250]  eta: 0:00:22  loss: 2.3759 (1.0186)  acc1: 50.0000 (74.8542)  acc5: 75.0000 (92.3484)  time: 0.0055  data: 0.0006  max mem: 8962\n",
            "Test:   [3100/6250]  eta: 0:00:21  loss: 1.6813 (1.0453)  acc1: 50.0000 (74.2986)  acc5: 75.0000 (92.0872)  time: 0.0058  data: 0.0007  max mem: 8962\n",
            "Test:   [3200/6250]  eta: 0:00:20  loss: 0.7106 (1.0641)  acc1: 62.5000 (73.9261)  acc5: 100.0000 (91.8658)  time: 0.0050  data: 0.0004  max mem: 8962\n",
            "Test:   [3300/6250]  eta: 0:00:19  loss: 1.2310 (1.0768)  acc1: 62.5000 (73.5800)  acc5: 87.5000 (91.7487)  time: 0.0060  data: 0.0005  max mem: 8962\n",
            "Test:   [3400/6250]  eta: 0:00:19  loss: 2.4905 (1.0913)  acc1: 50.0000 (73.3203)  acc5: 75.0000 (91.5466)  time: 0.0059  data: 0.0005  max mem: 8962\n",
            "Test:   [3500/6250]  eta: 0:00:18  loss: 1.7458 (1.0984)  acc1: 50.0000 (73.2112)  acc5: 87.5000 (91.4310)  time: 0.0060  data: 0.0007  max mem: 8962\n",
            "Test:   [3600/6250]  eta: 0:00:17  loss: 0.4052 (1.0948)  acc1: 87.5000 (73.3477)  acc5: 100.0000 (91.4295)  time: 0.0060  data: 0.0008  max mem: 8962\n",
            "Test:   [3700/6250]  eta: 0:00:16  loss: 1.7100 (1.1095)  acc1: 62.5000 (73.0444)  acc5: 87.5000 (91.2456)  time: 0.0104  data: 0.0052  max mem: 8962\n",
            "Test:   [3800/6250]  eta: 0:00:16  loss: 0.7023 (1.1154)  acc1: 87.5000 (72.9742)  acc5: 87.5000 (91.1471)  time: 0.0053  data: 0.0005  max mem: 8962\n",
            "Test:   [3900/6250]  eta: 0:00:15  loss: 2.4000 (1.1317)  acc1: 37.5000 (72.6609)  acc5: 75.0000 (90.9382)  time: 0.0053  data: 0.0005  max mem: 8962\n",
            "Test:   [4000/6250]  eta: 0:00:14  loss: 1.5344 (1.1465)  acc1: 62.5000 (72.3882)  acc5: 87.5000 (90.7523)  time: 0.0056  data: 0.0006  max mem: 8962\n",
            "Test:   [4100/6250]  eta: 0:00:14  loss: 1.4738 (1.1567)  acc1: 62.5000 (72.2171)  acc5: 87.5000 (90.6364)  time: 0.0052  data: 0.0004  max mem: 8962\n",
            "Test:   [4200/6250]  eta: 0:00:13  loss: 0.3974 (1.1608)  acc1: 87.5000 (72.0930)  acc5: 100.0000 (90.6362)  time: 0.0068  data: 0.0005  max mem: 8962\n",
            "Test:   [4300/6250]  eta: 0:00:12  loss: 0.4784 (1.1701)  acc1: 75.0000 (71.9542)  acc5: 100.0000 (90.4935)  time: 0.0052  data: 0.0006  max mem: 8962\n",
            "Test:   [4400/6250]  eta: 0:00:12  loss: 0.9728 (1.1769)  acc1: 75.0000 (71.7763)  acc5: 100.0000 (90.4368)  time: 0.0052  data: 0.0005  max mem: 8962\n",
            "Test:   [4500/6250]  eta: 0:00:11  loss: 1.1849 (1.1823)  acc1: 75.0000 (71.6785)  acc5: 87.5000 (90.3855)  time: 0.0058  data: 0.0004  max mem: 8962\n",
            "Test:   [4600/6250]  eta: 0:00:10  loss: 1.6483 (1.1923)  acc1: 50.0000 (71.5116)  acc5: 87.5000 (90.2413)  time: 0.0056  data: 0.0007  max mem: 8962\n",
            "Test:   [4700/6250]  eta: 0:00:10  loss: 1.3664 (1.2023)  acc1: 62.5000 (71.2189)  acc5: 87.5000 (90.1032)  time: 0.0055  data: 0.0007  max mem: 8962\n",
            "Test:   [4800/6250]  eta: 0:00:09  loss: 1.3209 (1.2095)  acc1: 62.5000 (71.0841)  acc5: 87.5000 (89.9969)  time: 0.0051  data: 0.0008  max mem: 8962\n",
            "Test:   [4900/6250]  eta: 0:00:08  loss: 0.4727 (1.2146)  acc1: 87.5000 (71.0161)  acc5: 100.0000 (89.9077)  time: 0.0049  data: 0.0004  max mem: 8962\n",
            "Test:   [5000/6250]  eta: 0:00:08  loss: 2.0785 (1.2263)  acc1: 62.5000 (70.8483)  acc5: 75.0000 (89.7520)  time: 0.0054  data: 0.0005  max mem: 8962\n",
            "Test:   [5100/6250]  eta: 0:00:07  loss: 1.2702 (1.2315)  acc1: 62.5000 (70.7410)  acc5: 87.5000 (89.7153)  time: 0.0059  data: 0.0007  max mem: 8962\n",
            "Test:   [5200/6250]  eta: 0:00:06  loss: 1.0106 (1.2380)  acc1: 75.0000 (70.6138)  acc5: 87.5000 (89.6510)  time: 0.0047  data: 0.0004  max mem: 8962\n",
            "Test:   [5300/6250]  eta: 0:00:06  loss: 1.4912 (1.2517)  acc1: 50.0000 (70.3122)  acc5: 87.5000 (89.4619)  time: 0.0056  data: 0.0005  max mem: 8962\n",
            "Test:   [5400/6250]  eta: 0:00:05  loss: 0.9010 (1.2548)  acc1: 75.0000 (70.2671)  acc5: 100.0000 (89.4071)  time: 0.0057  data: 0.0006  max mem: 8962\n",
            "Test:   [5500/6250]  eta: 0:00:04  loss: 0.9291 (1.2585)  acc1: 75.0000 (70.1622)  acc5: 87.5000 (89.3656)  time: 0.0056  data: 0.0005  max mem: 8962\n",
            "Test:   [5600/6250]  eta: 0:00:04  loss: 0.9468 (1.2638)  acc1: 62.5000 (70.0857)  acc5: 87.5000 (89.2586)  time: 0.0060  data: 0.0016  max mem: 8962\n",
            "Test:   [5700/6250]  eta: 0:00:03  loss: 2.1460 (1.2784)  acc1: 37.5000 (69.7926)  acc5: 75.0000 (89.1203)  time: 0.0050  data: 0.0006  max mem: 8962\n",
            "Test:   [5800/6250]  eta: 0:00:02  loss: 0.6562 (1.2747)  acc1: 75.0000 (69.8522)  acc5: 100.0000 (89.1657)  time: 0.0053  data: 0.0008  max mem: 8962\n",
            "Test:   [5900/6250]  eta: 0:00:02  loss: 1.0713 (1.2722)  acc1: 75.0000 (69.8759)  acc5: 100.0000 (89.2095)  time: 0.0053  data: 0.0007  max mem: 8962\n",
            "Test:   [6000/6250]  eta: 0:00:01  loss: 0.5802 (1.2647)  acc1: 87.5000 (70.0342)  acc5: 87.5000 (89.2955)  time: 0.0057  data: 0.0006  max mem: 8962\n",
            "Test:   [6100/6250]  eta: 0:00:00  loss: 1.2877 (1.2731)  acc1: 62.5000 (69.8328)  acc5: 87.5000 (89.2046)  time: 0.0052  data: 0.0008  max mem: 8962\n",
            "Test:   [6200/6250]  eta: 0:00:00  loss: 0.2487 (1.2671)  acc1: 87.5000 (69.9544)  acc5: 100.0000 (89.2739)  time: 0.0061  data: 0.0008  max mem: 8962\n",
            "Test:  Total time: 0:00:39\n",
            "Test:  Acc@1 69.994 Acc@5 89.292\n",
            "Epoch: [4]  [   0/5005]  eta: 6:27:39  lr: 0.0002  img/s: 465.53450010860746  loss: 0.8434 (0.8434)  acc1: 71.0938 (71.0938)  acc5: 87.5000 (87.5000)  meanQV: 1.4977 (1.4977)  stdQV: 0.3180 (0.3180)  time: 4.6472  data: 4.0972  max mem: 8962\n",
            "Epoch: [4]  [ 100/5005]  eta: 0:45:55  lr: 0.0002  img/s: 490.74347871791  loss: 0.9550 (0.8984)  acc1: 67.1875 (69.1097)  acc5: 86.7188 (87.1519)  meanQV: 1.4691 (1.4833)  stdQV: 0.3260 (0.3231)  time: 0.5208  data: 0.0004  max mem: 8962\n",
            "Epoch: [4]  [ 200/5005]  eta: 0:43:22  lr: 0.0002  img/s: 493.9701025625455  loss: 0.8835 (0.9029)  acc1: 68.7500 (68.8783)  acc5: 87.5000 (87.0375)  meanQV: 1.4809 (1.4816)  stdQV: 0.3240 (0.3238)  time: 0.5217  data: 0.0003  max mem: 8962\n",
            "Epoch: [4]  [ 300/5005]  eta: 0:41:56  lr: 0.0002  img/s: 491.63170037870236  loss: 0.9149 (0.9015)  acc1: 67.9688 (68.7227)  acc5: 87.8906 (87.0367)  meanQV: 1.4758 (1.4805)  stdQV: 0.3260 (0.3243)  time: 0.5222  data: 0.0005  max mem: 8962\n",
            "Epoch: [4]  [ 400/5005]  eta: 0:40:46  lr: 0.0002  img/s: 489.97201559708594  loss: 0.8910 (0.8983)  acc1: 68.7500 (68.8055)  acc5: 86.3281 (87.0305)  meanQV: 1.4797 (1.4812)  stdQV: 0.3246 (0.3240)  time: 0.5209  data: 0.0003  max mem: 8962\n",
            "Epoch: [4]  [ 500/5005]  eta: 0:39:44  lr: 0.0002  img/s: 494.2242973791529  loss: 0.8698 (0.8962)  acc1: 69.9219 (68.8599)  acc5: 86.7188 (87.0556)  meanQV: 1.4895 (1.4815)  stdQV: 0.3214 (0.3239)  time: 0.5205  data: 0.0003  max mem: 8962\n",
            "Epoch: [4]  [ 600/5005]  eta: 0:38:45  lr: 0.0002  img/s: 491.0853097714258  loss: 0.8708 (0.8966)  acc1: 70.3125 (68.8962)  acc5: 87.1094 (87.0457)  meanQV: 1.4922 (1.4818)  stdQV: 0.3188 (0.3238)  time: 0.5203  data: 0.0003  max mem: 8962\n",
            "Epoch: [4]  [ 700/5005]  eta: 0:37:48  lr: 0.0002  img/s: 493.5375728016555  loss: 0.9078 (0.8985)  acc1: 68.7500 (68.8353)  acc5: 87.1094 (87.0018)  meanQV: 1.4813 (1.4814)  stdQV: 0.3248 (0.3240)  time: 0.5213  data: 0.0003  max mem: 8962\n",
            "Epoch: [4]  [ 800/5005]  eta: 0:36:52  lr: 0.0002  img/s: 490.2532232540004  loss: 0.9188 (0.8979)  acc1: 67.9688 (68.8904)  acc5: 86.3281 (86.9953)  meanQV: 1.4746 (1.4818)  stdQV: 0.3270 (0.3238)  time: 0.5209  data: 0.0003  max mem: 8962\n",
            "Epoch: [4]  [ 900/5005]  eta: 0:35:57  lr: 0.0002  img/s: 493.18552460181996  loss: 0.9238 (0.8994)  acc1: 67.9688 (68.8575)  acc5: 86.3281 (86.9680)  meanQV: 1.4758 (1.4815)  stdQV: 0.3262 (0.3239)  time: 0.5208  data: 0.0004  max mem: 8962\n",
            "Epoch: [4]  [1000/5005]  eta: 0:35:03  lr: 0.0002  img/s: 490.8677666455156  loss: 0.8737 (0.8983)  acc1: 69.9219 (68.8835)  acc5: 87.1094 (86.9892)  meanQV: 1.4895 (1.4817)  stdQV: 0.3192 (0.3238)  time: 0.5206  data: 0.0004  max mem: 8962\n",
            "Epoch: [4]  [1100/5005]  eta: 0:34:09  lr: 0.0002  img/s: 494.22497982810273  loss: 0.8912 (0.8984)  acc1: 69.1406 (68.9065)  acc5: 86.7188 (87.0005)  meanQV: 1.4840 (1.4819)  stdQV: 0.3204 (0.3237)  time: 0.5215  data: 0.0004  max mem: 8962\n",
            "Epoch: [4]  [1200/5005]  eta: 0:33:16  lr: 0.0002  img/s: 491.8136503316876  loss: 0.9000 (0.8996)  acc1: 68.7500 (68.8476)  acc5: 87.5000 (86.9653)  meanQV: 1.4801 (1.4814)  stdQV: 0.3248 (0.3239)  time: 0.5211  data: 0.0003  max mem: 8962\n",
            "Epoch: [4]  [1300/5005]  eta: 0:32:22  lr: 0.0002  img/s: 494.33123736827326  loss: 0.8564 (0.8992)  acc1: 70.3125 (68.8515)  acc5: 87.5000 (86.9686)  meanQV: 1.4918 (1.4815)  stdQV: 0.3192 (0.3239)  time: 0.5202  data: 0.0003  max mem: 8962\n",
            "Epoch: [4]  [1400/5005]  eta: 0:31:29  lr: 0.0002  img/s: 493.6809842522038  loss: 0.8581 (0.8993)  acc1: 69.5312 (68.8392)  acc5: 87.5000 (86.9747)  meanQV: 1.4867 (1.4814)  stdQV: 0.3216 (0.3239)  time: 0.5203  data: 0.0004  max mem: 8962\n",
            "Epoch: [4]  [1500/5005]  eta: 0:30:36  lr: 0.0002  img/s: 492.017828708216  loss: 0.8798 (0.8990)  acc1: 68.7500 (68.8564)  acc5: 86.7188 (86.9821)  meanQV: 1.4801 (1.4815)  stdQV: 0.3216 (0.3239)  time: 0.5223  data: 0.0004  max mem: 8962\n",
            "Epoch: [4]  [1600/5005]  eta: 0:29:43  lr: 0.0002  img/s: 494.6657802925495  loss: 0.8522 (0.8990)  acc1: 69.9219 (68.8637)  acc5: 87.1094 (86.9842)  meanQV: 1.4887 (1.4816)  stdQV: 0.3213 (0.3239)  time: 0.5216  data: 0.0003  max mem: 8962\n",
            "Epoch: [4]  [1700/5005]  eta: 0:28:50  lr: 0.0002  img/s: 499.96709108899364  loss: 0.8898 (0.8983)  acc1: 69.5312 (68.8800)  acc5: 86.7188 (86.9978)  meanQV: 1.4867 (1.4817)  stdQV: 0.3215 (0.3238)  time: 0.5219  data: 0.0004  max mem: 8962\n",
            "Epoch: [4]  [1800/5005]  eta: 0:27:57  lr: 0.0002  img/s: 491.00491534530653  loss: 0.9142 (0.8985)  acc1: 68.3594 (68.8823)  acc5: 86.7188 (86.9988)  meanQV: 1.4781 (1.4817)  stdQV: 0.3251 (0.3238)  time: 0.5209  data: 0.0003  max mem: 8962\n",
            "Epoch: [4]  [1900/5005]  eta: 0:27:05  lr: 0.0002  img/s: 489.01270554630025  loss: 0.8833 (0.8986)  acc1: 68.7500 (68.8708)  acc5: 87.1094 (86.9945)  meanQV: 1.4801 (1.4816)  stdQV: 0.3248 (0.3238)  time: 0.5220  data: 0.0003  max mem: 8962\n",
            "Epoch: [4]  [2000/5005]  eta: 0:26:12  lr: 0.0002  img/s: 485.6457289761854  loss: 0.8689 (0.8982)  acc1: 69.1406 (68.8986)  acc5: 87.5000 (87.0073)  meanQV: 1.4824 (1.4818)  stdQV: 0.3235 (0.3238)  time: 0.5208  data: 0.0004  max mem: 8962\n",
            "Epoch: [4]  [2100/5005]  eta: 0:25:19  lr: 0.0002  img/s: 490.798211492559  loss: 0.8893 (0.8982)  acc1: 68.7500 (68.8917)  acc5: 87.5000 (87.0095)  meanQV: 1.4813 (1.4817)  stdQV: 0.3251 (0.3238)  time: 0.5210  data: 0.0003  max mem: 8962\n",
            "Epoch: [4]  [2200/5005]  eta: 0:24:27  lr: 0.0002  img/s: 493.2516794802064  loss: 0.8902 (0.8984)  acc1: 68.3594 (68.8870)  acc5: 87.1094 (87.0112)  meanQV: 1.4785 (1.4817)  stdQV: 0.3248 (0.3238)  time: 0.5206  data: 0.0003  max mem: 8962\n",
            "Epoch: [4]  [2300/5005]  eta: 0:23:34  lr: 0.0002  img/s: 487.6654664365519  loss: 0.8884 (0.8983)  acc1: 68.7500 (68.8970)  acc5: 86.3281 (87.0163)  meanQV: 1.4813 (1.4818)  stdQV: 0.3240 (0.3238)  time: 0.5224  data: 0.0002  max mem: 8962\n",
            "Epoch: [4]  [2400/5005]  eta: 0:22:42  lr: 0.0002  img/s: 487.97818203963027  loss: 0.9129 (0.8987)  acc1: 68.3594 (68.9003)  acc5: 87.5000 (87.0210)  meanQV: 1.4785 (1.4818)  stdQV: 0.3251 (0.3238)  time: 0.5218  data: 0.0003  max mem: 8962\n",
            "Epoch: [4]  [2500/5005]  eta: 0:21:49  lr: 0.0002  img/s: 493.4107957036287  loss: 0.8948 (0.8988)  acc1: 68.7500 (68.9018)  acc5: 86.3281 (87.0080)  meanQV: 1.4809 (1.4818)  stdQV: 0.3240 (0.3238)  time: 0.5208  data: 0.0003  max mem: 8962\n",
            "Epoch: [4]  [2600/5005]  eta: 0:20:57  lr: 0.0002  img/s: 491.60288897587867  loss: 0.8879 (0.8991)  acc1: 68.7500 (68.9072)  acc5: 86.3281 (87.0161)  meanQV: 1.4813 (1.4818)  stdQV: 0.3226 (0.3237)  time: 0.5208  data: 0.0003  max mem: 8962\n",
            "Epoch: [4]  [2700/5005]  eta: 0:20:04  lr: 0.0002  img/s: 491.1488804236057  loss: 0.8520 (0.8993)  acc1: 69.9219 (68.9277)  acc5: 87.8906 (87.0213)  meanQV: 1.4895 (1.4820)  stdQV: 0.3204 (0.3237)  time: 0.5217  data: 0.0004  max mem: 8962\n",
            "Epoch: [4]  [2800/5005]  eta: 0:19:12  lr: 0.0002  img/s: 493.3804151641137  loss: 0.8855 (0.8992)  acc1: 69.1406 (68.9271)  acc5: 87.1094 (87.0320)  meanQV: 1.4828 (1.4820)  stdQV: 0.3228 (0.3237)  time: 0.5209  data: 0.0004  max mem: 8962\n",
            "Epoch: [4]  [2900/5005]  eta: 0:18:20  lr: 0.0002  img/s: 494.2067818344083  loss: 0.9216 (0.8993)  acc1: 68.7500 (68.9377)  acc5: 86.3281 (87.0301)  meanQV: 1.4813 (1.4821)  stdQV: 0.3251 (0.3237)  time: 0.5215  data: 0.0003  max mem: 8962\n",
            "Epoch: [4]  [3000/5005]  eta: 0:17:27  lr: 0.0002  img/s: 491.1448365704542  loss: 0.9209 (0.8991)  acc1: 68.3594 (68.9410)  acc5: 86.7188 (87.0258)  meanQV: 1.4785 (1.4821)  stdQV: 0.3249 (0.3237)  time: 0.5223  data: 0.0004  max mem: 8962\n",
            "Epoch: [4]  [3100/5005]  eta: 0:16:35  lr: 0.0002  img/s: 492.0568356895722  loss: 0.8835 (0.8994)  acc1: 68.7500 (68.9353)  acc5: 87.1094 (87.0167)  meanQV: 1.4813 (1.4820)  stdQV: 0.3226 (0.3237)  time: 0.5203  data: 0.0004  max mem: 8962\n",
            "Epoch: [4]  [3200/5005]  eta: 0:15:43  lr: 0.0002  img/s: 491.44133759107467  loss: 0.8832 (0.8996)  acc1: 69.9219 (68.9327)  acc5: 87.1094 (87.0172)  meanQV: 1.4883 (1.4820)  stdQV: 0.3214 (0.3237)  time: 0.5212  data: 0.0005  max mem: 8962\n",
            "Epoch: [4]  [3300/5005]  eta: 0:14:50  lr: 0.0002  img/s: 487.73613035951706  loss: 0.8594 (0.8997)  acc1: 69.1406 (68.9283)  acc5: 87.8906 (87.0199)  meanQV: 1.4840 (1.4820)  stdQV: 0.3228 (0.3237)  time: 0.5213  data: 0.0004  max mem: 8962\n",
            "Epoch: [4]  [3400/5005]  eta: 0:13:58  lr: 0.0002  img/s: 491.204377802288  loss: 0.9090 (0.8999)  acc1: 68.7500 (68.9294)  acc5: 86.3281 (87.0150)  meanQV: 1.4789 (1.4820)  stdQV: 0.3246 (0.3237)  time: 0.5213  data: 0.0004  max mem: 8962\n",
            "Epoch: [4]  [3500/5005]  eta: 0:13:06  lr: 0.0002  img/s: 495.01811999100084  loss: 0.9384 (0.9003)  acc1: 68.3594 (68.9144)  acc5: 86.3281 (87.0097)  meanQV: 1.4785 (1.4819)  stdQV: 0.3262 (0.3237)  time: 0.5219  data: 0.0003  max mem: 8962\n",
            "Epoch: [4]  [3600/5005]  eta: 0:12:13  lr: 0.0002  img/s: 488.85240665181243  loss: 0.8764 (0.9000)  acc1: 69.9219 (68.9205)  acc5: 87.8906 (87.0174)  meanQV: 1.4895 (1.4819)  stdQV: 0.3204 (0.3237)  time: 0.5217  data: 0.0004  max mem: 8962\n",
            "Epoch: [4]  [3700/5005]  eta: 0:11:21  lr: 0.0002  img/s: 488.62905411616515  loss: 0.8644 (0.8999)  acc1: 69.9219 (68.9198)  acc5: 86.7188 (87.0206)  meanQV: 1.4895 (1.4819)  stdQV: 0.3204 (0.3237)  time: 0.5218  data: 0.0004  max mem: 8962\n",
            "Epoch: [4]  [3800/5005]  eta: 0:10:29  lr: 0.0002  img/s: 488.65351505542134  loss: 0.8901 (0.8997)  acc1: 68.7500 (68.9341)  acc5: 87.5000 (87.0288)  meanQV: 1.4789 (1.4820)  stdQV: 0.3248 (0.3237)  time: 0.5216  data: 0.0003  max mem: 8962\n",
            "Epoch: [4]  [3900/5005]  eta: 0:09:37  lr: 0.0002  img/s: 493.65601740808734  loss: 0.8595 (0.8997)  acc1: 68.3594 (68.9418)  acc5: 87.1094 (87.0309)  meanQV: 1.4785 (1.4821)  stdQV: 0.3262 (0.3237)  time: 0.5212  data: 0.0004  max mem: 8962\n",
            "Epoch: [4]  [4000/5005]  eta: 0:08:44  lr: 0.0002  img/s: 492.87718715199964  loss: 0.8968 (0.8997)  acc1: 69.9219 (68.9440)  acc5: 87.5000 (87.0334)  meanQV: 1.4895 (1.4821)  stdQV: 0.3204 (0.3237)  time: 0.5208  data: 0.0003  max mem: 8962\n",
            "Epoch: [4]  [4100/5005]  eta: 0:07:52  lr: 0.0002  img/s: 488.15255480193474  loss: 0.8810 (0.8995)  acc1: 68.3594 (68.9461)  acc5: 87.5000 (87.0394)  meanQV: 1.4785 (1.4821)  stdQV: 0.3251 (0.3237)  time: 0.5210  data: 0.0004  max mem: 8962\n",
            "Epoch: [4]  [4200/5005]  eta: 0:07:00  lr: 0.0002  img/s: 490.5603024844127  loss: 0.9360 (0.8999)  acc1: 68.7500 (68.9386)  acc5: 85.9375 (87.0355)  meanQV: 1.4813 (1.4820)  stdQV: 0.3240 (0.3237)  time: 0.5213  data: 0.0004  max mem: 8962\n",
            "Epoch: [4]  [4300/5005]  eta: 0:06:08  lr: 0.0002  img/s: 493.6328685908266  loss: 0.8927 (0.9000)  acc1: 67.9688 (68.9293)  acc5: 87.1094 (87.0393)  meanQV: 1.4758 (1.4820)  stdQV: 0.3273 (0.3237)  time: 0.5209  data: 0.0003  max mem: 8962\n",
            "Epoch: [4]  [4400/5005]  eta: 0:05:15  lr: 0.0002  img/s: 500.0339136176978  loss: 0.8938 (0.9000)  acc1: 67.9688 (68.9281)  acc5: 87.5000 (87.0415)  meanQV: 1.4746 (1.4820)  stdQV: 0.3237 (0.3237)  time: 0.5203  data: 0.0003  max mem: 8962\n",
            "Epoch: [4]  [4500/5005]  eta: 0:04:23  lr: 0.0002  img/s: 491.36240054840806  loss: 0.8792 (0.9000)  acc1: 69.5312 (68.9270)  acc5: 87.1094 (87.0380)  meanQV: 1.4832 (1.4820)  stdQV: 0.3213 (0.3237)  time: 0.5218  data: 0.0003  max mem: 8962\n",
            "Epoch: [4]  [4600/5005]  eta: 0:03:31  lr: 0.0002  img/s: 494.30324647494075  loss: 0.9115 (0.9000)  acc1: 69.1406 (68.9302)  acc5: 86.7188 (87.0371)  meanQV: 1.4828 (1.4820)  stdQV: 0.3238 (0.3237)  time: 0.5208  data: 0.0003  max mem: 8962\n",
            "Epoch: [4]  [4700/5005]  eta: 0:02:39  lr: 0.0002  img/s: 494.0173748211402  loss: 0.8860 (0.9000)  acc1: 69.9219 (68.9322)  acc5: 87.1094 (87.0404)  meanQV: 1.4895 (1.4820)  stdQV: 0.3202 (0.3237)  time: 0.5212  data: 0.0004  max mem: 8962\n",
            "Epoch: [4]  [4800/5005]  eta: 0:01:47  lr: 0.0002  img/s: 493.9932830203957  loss: 0.9224 (0.9002)  acc1: 68.7500 (68.9312)  acc5: 87.1094 (87.0430)  meanQV: 1.4801 (1.4820)  stdQV: 0.3248 (0.3237)  time: 0.5215  data: 0.0004  max mem: 8962\n",
            "Epoch: [4]  [4900/5005]  eta: 0:00:54  lr: 0.0002  img/s: 494.8479308631688  loss: 0.8930 (0.9002)  acc1: 67.9688 (68.9332)  acc5: 86.7188 (87.0443)  meanQV: 1.4758 (1.4820)  stdQV: 0.3262 (0.3237)  time: 0.5202  data: 0.0004  max mem: 8962\n",
            "Epoch: [4]  [5000/5005]  eta: 0:00:02  lr: 0.0002  img/s: 492.44182560982995  loss: 0.8594 (0.8999)  acc1: 70.3125 (68.9486)  acc5: 87.5000 (87.0523)  meanQV: 1.4922 (1.4821)  stdQV: 0.3192 (0.3236)  time: 0.5203  data: 0.0002  max mem: 8962\n",
            "Epoch: [4] Total time: 0:43:33\n",
            "Test:   [   0/6250]  eta: 1:16:02  loss: 0.7909 (0.7909)  acc1: 75.0000 (75.0000)  acc5: 100.0000 (100.0000)  time: 0.7299  data: 0.7206  max mem: 8962\n",
            "Test:   [ 100/6250]  eta: 0:01:29  loss: 0.1878 (0.6247)  acc1: 87.5000 (83.4158)  acc5: 100.0000 (95.2970)  time: 0.0065  data: 0.0006  max mem: 8962\n",
            "Test:   [ 200/6250]  eta: 0:01:04  loss: 0.6666 (0.6644)  acc1: 87.5000 (84.3284)  acc5: 100.0000 (94.5274)  time: 0.0077  data: 0.0020  max mem: 8962\n",
            "Test:   [ 300/6250]  eta: 0:00:54  loss: 1.2709 (0.8699)  acc1: 62.5000 (78.9037)  acc5: 87.5000 (93.0648)  time: 0.0064  data: 0.0008  max mem: 8962\n",
            "Test:   [ 400/6250]  eta: 0:00:49  loss: 0.8989 (0.9948)  acc1: 75.0000 (76.1222)  acc5: 100.0000 (92.1135)  time: 0.0065  data: 0.0005  max mem: 8962\n",
            "Test:   [ 500/6250]  eta: 0:00:45  loss: 1.1087 (1.0409)  acc1: 75.0000 (74.9251)  acc5: 87.5000 (91.7914)  time: 0.0068  data: 0.0025  max mem: 8962\n",
            "Test:   [ 600/6250]  eta: 0:00:43  loss: 0.4547 (0.9505)  acc1: 87.5000 (76.8927)  acc5: 100.0000 (92.4293)  time: 0.0059  data: 0.0005  max mem: 8962\n",
            "Test:   [ 700/6250]  eta: 0:00:40  loss: 0.3772 (0.9356)  acc1: 87.5000 (77.1755)  acc5: 100.0000 (92.4215)  time: 0.0055  data: 0.0004  max mem: 8962\n",
            "Test:   [ 800/6250]  eta: 0:00:40  loss: 0.8068 (0.9664)  acc1: 75.0000 (76.5762)  acc5: 87.5000 (92.0568)  time: 0.0060  data: 0.0004  max mem: 8962\n",
            "Test:   [ 900/6250]  eta: 0:00:38  loss: 0.3764 (0.9114)  acc1: 87.5000 (77.8996)  acc5: 100.0000 (92.5361)  time: 0.0062  data: 0.0004  max mem: 8962\n",
            "Test:   [1000/6250]  eta: 0:00:37  loss: 0.9310 (0.8980)  acc1: 75.0000 (78.1094)  acc5: 100.0000 (92.7572)  time: 0.0063  data: 0.0006  max mem: 8962\n",
            "Test:   [1100/6250]  eta: 0:00:37  loss: 1.1327 (0.9363)  acc1: 75.0000 (77.1571)  acc5: 100.0000 (92.5636)  time: 0.0073  data: 0.0006  max mem: 8962\n",
            "Test:   [1200/6250]  eta: 0:00:36  loss: 1.0381 (0.9410)  acc1: 75.0000 (76.7902)  acc5: 87.5000 (92.6728)  time: 0.0076  data: 0.0006  max mem: 8962\n",
            "Test:   [1300/6250]  eta: 0:00:35  loss: 0.4042 (0.9409)  acc1: 87.5000 (76.5853)  acc5: 100.0000 (92.8997)  time: 0.0066  data: 0.0007  max mem: 8962\n",
            "Test:   [1400/6250]  eta: 0:00:34  loss: 0.7884 (0.9356)  acc1: 75.0000 (76.6774)  acc5: 87.5000 (92.9961)  time: 0.0063  data: 0.0005  max mem: 8962\n",
            "Test:   [1500/6250]  eta: 0:00:33  loss: 0.9547 (0.9392)  acc1: 75.0000 (76.4074)  acc5: 100.0000 (93.1046)  time: 0.0058  data: 0.0005  max mem: 8962\n",
            "Test:   [1600/6250]  eta: 0:00:32  loss: 0.1298 (0.9348)  acc1: 87.5000 (76.2492)  acc5: 100.0000 (93.2464)  time: 0.0063  data: 0.0006  max mem: 8962\n",
            "Test:   [1700/6250]  eta: 0:00:32  loss: 1.0550 (0.9324)  acc1: 75.0000 (76.1023)  acc5: 100.0000 (93.3715)  time: 0.0092  data: 0.0041  max mem: 8962\n",
            "Test:   [1800/6250]  eta: 0:00:31  loss: 1.2747 (0.9422)  acc1: 62.5000 (75.8537)  acc5: 87.5000 (93.3648)  time: 0.0064  data: 0.0004  max mem: 8962\n",
            "Test:   [1900/6250]  eta: 0:00:30  loss: 0.9273 (0.9352)  acc1: 75.0000 (76.1244)  acc5: 100.0000 (93.4705)  time: 0.0065  data: 0.0005  max mem: 8962\n",
            "Test:   [2000/6250]  eta: 0:00:30  loss: 0.4722 (0.9411)  acc1: 87.5000 (76.0120)  acc5: 100.0000 (93.4658)  time: 0.0064  data: 0.0006  max mem: 8962\n",
            "Test:   [2100/6250]  eta: 0:00:29  loss: 0.3972 (0.9220)  acc1: 87.5000 (76.5469)  acc5: 100.0000 (93.6161)  time: 0.0064  data: 0.0005  max mem: 8962\n",
            "Test:   [2200/6250]  eta: 0:00:28  loss: 0.1838 (0.9145)  acc1: 87.5000 (76.6924)  acc5: 100.0000 (93.6733)  time: 0.0052  data: 0.0004  max mem: 8962\n",
            "Test:   [2300/6250]  eta: 0:00:27  loss: 0.7064 (0.9173)  acc1: 87.5000 (76.6515)  acc5: 100.0000 (93.6441)  time: 0.0061  data: 0.0019  max mem: 8962\n",
            "Test:   [2400/6250]  eta: 0:00:26  loss: 0.9312 (0.9272)  acc1: 75.0000 (76.5618)  acc5: 87.5000 (93.5392)  time: 0.0057  data: 0.0006  max mem: 8962\n",
            "Test:   [2500/6250]  eta: 0:00:26  loss: 0.9157 (0.9291)  acc1: 75.0000 (76.6144)  acc5: 100.0000 (93.4976)  time: 0.0060  data: 0.0006  max mem: 8962\n",
            "Test:   [2600/6250]  eta: 0:00:25  loss: 2.2484 (0.9531)  acc1: 50.0000 (76.1246)  acc5: 62.5000 (93.1949)  time: 0.0059  data: 0.0005  max mem: 8962\n",
            "Test:   [2700/6250]  eta: 0:00:24  loss: 0.8119 (0.9641)  acc1: 75.0000 (75.9210)  acc5: 87.5000 (93.0766)  time: 0.0057  data: 0.0004  max mem: 8962\n",
            "Test:   [2800/6250]  eta: 0:00:23  loss: 1.5186 (0.9860)  acc1: 62.5000 (75.4418)  acc5: 75.0000 (92.8240)  time: 0.0058  data: 0.0004  max mem: 8962\n",
            "Test:   [2900/6250]  eta: 0:00:23  loss: 1.7899 (1.0040)  acc1: 37.5000 (75.0603)  acc5: 87.5000 (92.6060)  time: 0.0078  data: 0.0007  max mem: 8962\n",
            "Test:   [3000/6250]  eta: 0:00:22  loss: 2.3188 (1.0252)  acc1: 50.0000 (74.7418)  acc5: 75.0000 (92.2817)  time: 0.0062  data: 0.0006  max mem: 8962\n",
            "Test:   [3100/6250]  eta: 0:00:21  loss: 1.8096 (1.0517)  acc1: 50.0000 (74.2341)  acc5: 75.0000 (92.0106)  time: 0.0056  data: 0.0004  max mem: 8962\n",
            "Test:   [3200/6250]  eta: 0:00:20  loss: 0.6736 (1.0715)  acc1: 75.0000 (73.8402)  acc5: 100.0000 (91.8072)  time: 0.0057  data: 0.0004  max mem: 8962\n",
            "Test:   [3300/6250]  eta: 0:00:19  loss: 1.2702 (1.0843)  acc1: 62.5000 (73.4853)  acc5: 87.5000 (91.6919)  time: 0.0060  data: 0.0009  max mem: 8962\n",
            "Test:   [3400/6250]  eta: 0:00:19  loss: 2.4354 (1.0978)  acc1: 37.5000 (73.2468)  acc5: 75.0000 (91.5062)  time: 0.0059  data: 0.0005  max mem: 8962\n",
            "Test:   [3500/6250]  eta: 0:00:18  loss: 1.7555 (1.1037)  acc1: 50.0000 (73.1720)  acc5: 75.0000 (91.3846)  time: 0.0058  data: 0.0006  max mem: 8962\n",
            "Test:   [3600/6250]  eta: 0:00:17  loss: 0.5104 (1.0999)  acc1: 87.5000 (73.3164)  acc5: 100.0000 (91.3948)  time: 0.0054  data: 0.0005  max mem: 8962\n",
            "Test:   [3700/6250]  eta: 0:00:17  loss: 1.7609 (1.1145)  acc1: 62.5000 (73.0276)  acc5: 87.5000 (91.2017)  time: 0.0116  data: 0.0071  max mem: 8962\n",
            "Test:   [3800/6250]  eta: 0:00:16  loss: 0.6589 (1.1207)  acc1: 87.5000 (72.9446)  acc5: 87.5000 (91.1076)  time: 0.0069  data: 0.0008  max mem: 8962\n",
            "Test:   [3900/6250]  eta: 0:00:15  loss: 2.3705 (1.1368)  acc1: 50.0000 (72.6224)  acc5: 75.0000 (90.8837)  time: 0.0063  data: 0.0010  max mem: 8962\n",
            "Test:   [4000/6250]  eta: 0:00:14  loss: 1.3760 (1.1507)  acc1: 62.5000 (72.3663)  acc5: 87.5000 (90.7179)  time: 0.0052  data: 0.0004  max mem: 8962\n",
            "Test:   [4100/6250]  eta: 0:00:14  loss: 1.3587 (1.1609)  acc1: 62.5000 (72.1836)  acc5: 87.5000 (90.6059)  time: 0.0056  data: 0.0005  max mem: 8962\n",
            "Test:   [4200/6250]  eta: 0:00:13  loss: 0.4211 (1.1651)  acc1: 87.5000 (72.0543)  acc5: 100.0000 (90.5975)  time: 0.0050  data: 0.0004  max mem: 8962\n",
            "Test:   [4300/6250]  eta: 0:00:12  loss: 0.4845 (1.1744)  acc1: 75.0000 (71.9339)  acc5: 87.5000 (90.4354)  time: 0.0056  data: 0.0006  max mem: 8962\n",
            "Test:   [4400/6250]  eta: 0:00:12  loss: 0.9234 (1.1813)  acc1: 75.0000 (71.7422)  acc5: 100.0000 (90.3885)  time: 0.0054  data: 0.0005  max mem: 8962\n",
            "Test:   [4500/6250]  eta: 0:00:11  loss: 1.2129 (1.1870)  acc1: 62.5000 (71.6230)  acc5: 87.5000 (90.3299)  time: 0.0060  data: 0.0008  max mem: 8962\n",
            "Test:   [4600/6250]  eta: 0:00:10  loss: 1.6169 (1.1973)  acc1: 50.0000 (71.4519)  acc5: 87.5000 (90.1652)  time: 0.0059  data: 0.0006  max mem: 8962\n",
            "Test:   [4700/6250]  eta: 0:00:10  loss: 1.4572 (1.2073)  acc1: 62.5000 (71.1471)  acc5: 87.5000 (90.0181)  time: 0.0054  data: 0.0005  max mem: 8962\n",
            "Test:   [4800/6250]  eta: 0:00:09  loss: 1.1656 (1.2140)  acc1: 62.5000 (71.0347)  acc5: 87.5000 (89.9188)  time: 0.0057  data: 0.0005  max mem: 8962\n",
            "Test:   [4900/6250]  eta: 0:00:08  loss: 0.4505 (1.2183)  acc1: 87.5000 (70.9779)  acc5: 100.0000 (89.8312)  time: 0.0050  data: 0.0005  max mem: 8962\n",
            "Test:   [5000/6250]  eta: 0:00:08  loss: 2.1536 (1.2300)  acc1: 62.5000 (70.8008)  acc5: 75.0000 (89.6821)  time: 0.0061  data: 0.0004  max mem: 8962\n",
            "Test:   [5100/6250]  eta: 0:00:07  loss: 1.3107 (1.2352)  acc1: 62.5000 (70.6920)  acc5: 100.0000 (89.6442)  time: 0.0064  data: 0.0008  max mem: 8962\n",
            "Test:   [5200/6250]  eta: 0:00:06  loss: 1.0626 (1.2412)  acc1: 75.0000 (70.5850)  acc5: 87.5000 (89.5741)  time: 0.0053  data: 0.0005  max mem: 8962\n",
            "Test:   [5300/6250]  eta: 0:00:06  loss: 1.4655 (1.2552)  acc1: 62.5000 (70.2650)  acc5: 87.5000 (89.3864)  time: 0.0063  data: 0.0005  max mem: 8962\n",
            "Test:   [5400/6250]  eta: 0:00:05  loss: 0.8139 (1.2579)  acc1: 75.0000 (70.2162)  acc5: 87.5000 (89.3469)  time: 0.0053  data: 0.0004  max mem: 8962\n",
            "Test:   [5500/6250]  eta: 0:00:04  loss: 0.9340 (1.2610)  acc1: 75.0000 (70.1213)  acc5: 87.5000 (89.3065)  time: 0.0055  data: 0.0004  max mem: 8962\n",
            "Test:   [5600/6250]  eta: 0:00:04  loss: 1.0789 (1.2661)  acc1: 62.5000 (70.0433)  acc5: 87.5000 (89.2140)  time: 0.0064  data: 0.0013  max mem: 8962\n",
            "Test:   [5700/6250]  eta: 0:00:03  loss: 2.0526 (1.2814)  acc1: 37.5000 (69.7422)  acc5: 75.0000 (89.0567)  time: 0.0050  data: 0.0004  max mem: 8962\n",
            "Test:   [5800/6250]  eta: 0:00:02  loss: 0.5955 (1.2775)  acc1: 75.0000 (69.8220)  acc5: 100.0000 (89.0989)  time: 0.0068  data: 0.0007  max mem: 8962\n",
            "Test:   [5900/6250]  eta: 0:00:02  loss: 1.1023 (1.2751)  acc1: 62.5000 (69.8483)  acc5: 100.0000 (89.1353)  time: 0.0056  data: 0.0006  max mem: 8962\n",
            "Test:   [6000/6250]  eta: 0:00:01  loss: 0.5431 (1.2677)  acc1: 87.5000 (69.9988)  acc5: 87.5000 (89.2289)  time: 0.0058  data: 0.0004  max mem: 8962\n",
            "Test:   [6100/6250]  eta: 0:00:00  loss: 1.3317 (1.2758)  acc1: 62.5000 (69.8164)  acc5: 87.5000 (89.1370)  time: 0.0054  data: 0.0004  max mem: 8962\n",
            "Test:   [6200/6250]  eta: 0:00:00  loss: 0.2130 (1.2696)  acc1: 87.5000 (69.9444)  acc5: 100.0000 (89.2114)  time: 0.0065  data: 0.0005  max mem: 8962\n",
            "Test:  Total time: 0:00:40\n",
            "Test:  Acc@1 69.982 Acc@5 89.216\n",
            "Epoch: [5]  [   0/5005]  eta: 6:26:06  lr: 0.0001  img/s: 472.08087977640673  loss: 0.9772 (0.9772)  acc1: 70.3125 (70.3125)  acc5: 87.8906 (87.8906)  meanQV: 1.4922 (1.4922)  stdQV: 0.3204 (0.3204)  time: 4.6287  data: 4.0864  max mem: 8962\n",
            "Epoch: [5]  [ 100/5005]  eta: 0:45:55  lr: 0.0001  img/s: 493.412156113808  loss: 0.8671 (0.8977)  acc1: 69.1406 (69.1445)  acc5: 87.1094 (87.3028)  meanQV: 1.4840 (1.4833)  stdQV: 0.3228 (0.3230)  time: 0.5215  data: 0.0004  max mem: 8962\n",
            "Epoch: [5]  [ 200/5005]  eta: 0:42:34  lr: 0.0001  img/s: 633.4831033504211  loss: 0.8743 (0.9023)  acc1: 69.5312 (69.1465)  acc5: 87.1094 (87.1774)  meanQV: 1.4867 (1.4834)  stdQV: 0.3216 (0.3230)  time: 0.4189  data: 0.0004  max mem: 8962\n",
            "Epoch: [5]  [ 300/5005]  eta: 0:37:51  lr: 0.0001  img/s: 491.4145726182402  loss: 0.8887 (0.9009)  acc1: 69.5312 (69.1938)  acc5: 87.8906 (87.1872)  meanQV: 1.4867 (1.4838)  stdQV: 0.3204 (0.3229)  time: 0.3729  data: 0.0032  max mem: 8962\n",
            "Epoch: [5]  [ 400/5005]  eta: 0:37:47  lr: 0.0001  img/s: 488.7140108034497  loss: 0.9087 (0.9030)  acc1: 68.3594 (69.0539)  acc5: 87.5000 (87.1396)  meanQV: 1.4781 (1.4828)  stdQV: 0.3251 (0.3233)  time: 0.5207  data: 0.0003  max mem: 8962\n",
            "Epoch: [5]  [ 500/5005]  eta: 0:37:23  lr: 0.0001  img/s: 489.01871879991273  loss: 0.8969 (0.9043)  acc1: 67.5781 (68.9605)  acc5: 87.1094 (87.0493)  meanQV: 1.4730 (1.4821)  stdQV: 0.3249 (0.3237)  time: 0.5216  data: 0.0003  max mem: 8962\n",
            "Epoch: [5]  [ 600/5005]  eta: 0:36:50  lr: 0.0001  img/s: 494.24204166440353  loss: 0.9185 (0.9048)  acc1: 68.3594 (68.9502)  acc5: 87.1094 (87.0873)  meanQV: 1.4785 (1.4821)  stdQV: 0.3262 (0.3237)  time: 0.5209  data: 0.0004  max mem: 8962\n",
            "Epoch: [5]  [ 700/5005]  eta: 0:36:12  lr: 0.0001  img/s: 490.8462248780818  loss: 0.9053 (0.9051)  acc1: 68.3594 (68.9840)  acc5: 85.9375 (87.0927)  meanQV: 1.4781 (1.4823)  stdQV: 0.3240 (0.3236)  time: 0.5222  data: 0.0003  max mem: 8962\n",
            "Epoch: [5]  [ 800/5005]  eta: 0:35:30  lr: 0.0001  img/s: 493.49674301159627  loss: 0.8652 (0.9059)  acc1: 70.3125 (68.9992)  acc5: 87.5000 (87.0879)  meanQV: 1.4918 (1.4824)  stdQV: 0.3202 (0.3235)  time: 0.5209  data: 0.0004  max mem: 8962\n",
            "Epoch: [5]  [ 900/5005]  eta: 0:34:46  lr: 0.0001  img/s: 492.24203409939537  loss: 0.8904 (0.9059)  acc1: 69.1406 (68.9659)  acc5: 87.5000 (87.0920)  meanQV: 1.4840 (1.4822)  stdQV: 0.3224 (0.3236)  time: 0.5210  data: 0.0003  max mem: 8962\n",
            "Epoch: [5]  [1000/5005]  eta: 0:34:00  lr: 0.0001  img/s: 493.5423367172554  loss: 0.9307 (0.9063)  acc1: 69.1406 (68.9541)  acc5: 85.9375 (87.0731)  meanQV: 1.4840 (1.4821)  stdQV: 0.3228 (0.3236)  time: 0.5217  data: 0.0004  max mem: 8962\n",
            "Epoch: [5]  [1100/5005]  eta: 0:33:14  lr: 0.0001  img/s: 489.2419819520163  loss: 0.8769 (0.9064)  acc1: 69.9219 (68.9540)  acc5: 87.1094 (87.0718)  meanQV: 1.4891 (1.4821)  stdQV: 0.3212 (0.3236)  time: 0.5218  data: 0.0003  max mem: 8962\n",
            "Epoch: [5]  [1200/5005]  eta: 0:32:26  lr: 0.0001  img/s: 499.6392909558434  loss: 0.8655 (0.9056)  acc1: 67.9688 (68.9608)  acc5: 86.7188 (87.0716)  meanQV: 1.4746 (1.4821)  stdQV: 0.3257 (0.3236)  time: 0.5207  data: 0.0003  max mem: 8962\n",
            "Epoch: [5]  [1300/5005]  eta: 0:31:37  lr: 0.0001  img/s: 493.8163030755964  loss: 0.9314 (0.9051)  acc1: 67.5781 (68.9860)  acc5: 86.7188 (87.0968)  meanQV: 1.4727 (1.4823)  stdQV: 0.3280 (0.3235)  time: 0.5206  data: 0.0004  max mem: 8962\n",
            "Epoch: [5]  [1400/5005]  eta: 0:30:49  lr: 0.0001  img/s: 494.9975815812936  loss: 0.8642 (0.9041)  acc1: 69.5312 (69.0129)  acc5: 87.8906 (87.0929)  meanQV: 1.4844 (1.4825)  stdQV: 0.3216 (0.3234)  time: 0.5214  data: 0.0003  max mem: 8962\n",
            "Epoch: [5]  [1500/5005]  eta: 0:29:59  lr: 0.0001  img/s: 488.4738953923197  loss: 0.9535 (0.9047)  acc1: 67.9688 (68.9798)  acc5: 86.3281 (87.0698)  meanQV: 1.4758 (1.4823)  stdQV: 0.3262 (0.3235)  time: 0.5208  data: 0.0004  max mem: 8962\n",
            "Epoch: [5]  [1600/5005]  eta: 0:29:09  lr: 0.0001  img/s: 489.3487835801782  loss: 0.8809 (0.9042)  acc1: 69.1406 (69.0108)  acc5: 87.1094 (87.0811)  meanQV: 1.4828 (1.4825)  stdQV: 0.3237 (0.3235)  time: 0.5214  data: 0.0004  max mem: 8962\n",
            "Epoch: [5]  [1700/5005]  eta: 0:28:19  lr: 0.0001  img/s: 489.42685183665805  loss: 0.8938 (0.9043)  acc1: 68.3594 (68.9845)  acc5: 85.9375 (87.0756)  meanQV: 1.4785 (1.4823)  stdQV: 0.3262 (0.3235)  time: 0.5206  data: 0.0003  max mem: 8962\n",
            "Epoch: [5]  [1800/5005]  eta: 0:27:29  lr: 0.0001  img/s: 489.34900659686224  loss: 0.8771 (0.9038)  acc1: 69.9219 (69.0081)  acc5: 89.0625 (87.1057)  meanQV: 1.4867 (1.4825)  stdQV: 0.3192 (0.3235)  time: 0.5219  data: 0.0004  max mem: 8962\n",
            "Epoch: [5]  [1900/5005]  eta: 0:26:39  lr: 0.0001  img/s: 488.80233479720977  loss: 0.9229 (0.9041)  acc1: 67.9688 (69.0192)  acc5: 86.7188 (87.1049)  meanQV: 1.4746 (1.4825)  stdQV: 0.3236 (0.3234)  time: 0.5216  data: 0.0004  max mem: 8962\n",
            "Epoch: [5]  [2000/5005]  eta: 0:25:48  lr: 0.0001  img/s: 492.5014558430955  loss: 0.9266 (0.9045)  acc1: 67.9688 (69.0048)  acc5: 86.7188 (87.0990)  meanQV: 1.4758 (1.4824)  stdQV: 0.3262 (0.3235)  time: 0.5205  data: 0.0004  max mem: 8962\n"
          ]
        },
        {
          "ename": "KeyboardInterrupt",
          "evalue": "",
          "output_type": "error",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
            "Cell \u001b[0;32mIn[8], line 27\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[38;5;28;01mfrom\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[38;5;21;01mtypes\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[38;5;28;01mimport\u001b[39;00m SimpleNamespace\n\u001b[1;32m      3\u001b[0m args \u001b[38;5;241m=\u001b[39m SimpleNamespace(\n\u001b[1;32m      4\u001b[0m     data_path\u001b[38;5;241m=\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124m/home/cs/Documents/datasets/imagenet\u001b[39m\u001b[38;5;124m\"\u001b[39m,\n\u001b[1;32m      5\u001b[0m     model\u001b[38;5;241m=\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mresnet18\u001b[39m\u001b[38;5;124m\"\u001b[39m,\n\u001b[0;32m   (...)\u001b[0m\n\u001b[1;32m     24\u001b[0m     trp_lambdas\u001b[38;5;241m=\u001b[39m[\u001b[38;5;241m0.4\u001b[39m, \u001b[38;5;241m0.2\u001b[39m, \u001b[38;5;241m0.1\u001b[39m],\n\u001b[1;32m     25\u001b[0m )\n\u001b[0;32m---> 27\u001b[0m \u001b[43mmain\u001b[49m\u001b[43m(\u001b[49m\u001b[43margs\u001b[49m\u001b[43m)\u001b[49m\n",
            "Cell \u001b[0;32mIn[7], line 151\u001b[0m, in \u001b[0;36mmain\u001b[0;34m(args)\u001b[0m\n\u001b[1;32m    149\u001b[0m start_time \u001b[38;5;241m=\u001b[39m time\u001b[38;5;241m.\u001b[39mtime()\n\u001b[1;32m    150\u001b[0m \u001b[38;5;28;01mfor\u001b[39;00m epoch \u001b[38;5;129;01min\u001b[39;00m \u001b[38;5;28mrange\u001b[39m(args\u001b[38;5;241m.\u001b[39mepochs):\n\u001b[0;32m--> 151\u001b[0m     \u001b[43mtrain_one_epoch\u001b[49m\u001b[43m(\u001b[49m\u001b[43mmodel\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43moptimizer\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mdata_loader\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mdevice\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mepoch\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43margs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    152\u001b[0m     lr_scheduler\u001b[38;5;241m.\u001b[39mstep()\n\u001b[1;32m    153\u001b[0m     evaluate(model, nn\u001b[38;5;241m.\u001b[39mCrossEntropyLoss(), data_loader_test, device\u001b[38;5;241m=\u001b[39mdevice)\n",
            "Cell \u001b[0;32mIn[7], line 61\u001b[0m, in \u001b[0;36mtrain_one_epoch\u001b[0;34m(model, optimizer, data_loader, device, epoch, args)\u001b[0m\n\u001b[1;32m     59\u001b[0m acc1, acc5 \u001b[38;5;241m=\u001b[39m utils\u001b[38;5;241m.\u001b[39maccuracy(output, target, topk\u001b[38;5;241m=\u001b[39m(\u001b[38;5;241m1\u001b[39m, \u001b[38;5;241m5\u001b[39m))\n\u001b[1;32m     60\u001b[0m batch_size \u001b[38;5;241m=\u001b[39m image\u001b[38;5;241m.\u001b[39mshape[\u001b[38;5;241m0\u001b[39m]\n\u001b[0;32m---> 61\u001b[0m metric_logger\u001b[38;5;241m.\u001b[39mupdate(loss\u001b[38;5;241m=\u001b[39m\u001b[43mloss\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mitem\u001b[49m\u001b[43m(\u001b[49m\u001b[43m)\u001b[49m, lr\u001b[38;5;241m=\u001b[39moptimizer\u001b[38;5;241m.\u001b[39mparam_groups[\u001b[38;5;241m0\u001b[39m][\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mlr\u001b[39m\u001b[38;5;124m\"\u001b[39m])\n\u001b[1;32m     62\u001b[0m metric_logger\u001b[38;5;241m.\u001b[39mmeters[\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124macc1\u001b[39m\u001b[38;5;124m\"\u001b[39m]\u001b[38;5;241m.\u001b[39mupdate(acc1\u001b[38;5;241m.\u001b[39mitem(), n\u001b[38;5;241m=\u001b[39mbatch_size)\n\u001b[1;32m     63\u001b[0m metric_logger\u001b[38;5;241m.\u001b[39mmeters[\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124macc5\u001b[39m\u001b[38;5;124m\"\u001b[39m]\u001b[38;5;241m.\u001b[39mupdate(acc5\u001b[38;5;241m.\u001b[39mitem(), n\u001b[38;5;241m=\u001b[39mbatch_size)\n",
            "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
          ]
        }
      ],
      "source": [
        "from types import SimpleNamespace\n",
        "\n",
        "args = SimpleNamespace(\n",
        "    data_path=\"/home/cs/Documents/datasets/imagenet\",  # Replace with your /path/to/imagenet\n",
        "    model=\"resnet18\",\n",
        "    device=\"cuda\",\n",
        "    batch_size=256,\n",
        "    epochs=10,\n",
        "    lr=0.0004,\n",
        "    momentum=0.9,\n",
        "    weight_decay=1e-4,\n",
        "    lr_warmup_epochs=1,\n",
        "    lr_warmup_decay=0.0,\n",
        "    lr_step_size=2,\n",
        "    lr_gamma=0.5,\n",
        "    print_freq=100,\n",
        "    output_dir=\"resnet18\",\n",
        "    use_deterministic_algorithms=False,\n",
        "    weights=\"ResNet18_Weights.IMAGENET1K_V1\",\n",
        "    apply_trp=True,\n",
        "    trp_depths=[1, 1, 1],\n",
        "    trp_planes=256,\n",
        "    trp_lambdas=[0.4, 0.2, 0.1],\n",
        ")\n",
        "\n",
        "main(args)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "a3KD3WXU3l-O"
      },
      "source": [
        "# Fine-tuning a language model"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "JAscNNUD3l-P"
      },
      "source": [
        "In this notebook, we'll see how to fine-tune one of the [🤗 Transformers](https://github.com/huggingface/transformers) model on a language modeling tasks. We will cover two types of language modeling tasks which are:\n",
        "\n",
        "- Causal language modeling: the model has to predict the next token in the sentence (so the labels are the same as the inputs shifted to the right). To make sure the model does not cheat, it gets an attention mask that will prevent it to access the tokens after token i when trying to predict the token i+1 in the sentence.\n",
        "\n",
        "![Widget inference representing the causal language modeling task](https://github.com/huggingface/notebooks/blob/main/examples/images/causal_language_modeling.png?raw=1)\n",
        "\n",
        "- Masked language modeling: the model has to predict some tokens that are masked in the input. It still has access to the whole sentence, so it can use the tokens before and after the tokens masked to predict their value.\n",
        "\n",
        "![Widget inference representing the masked language modeling task](https://github.com/huggingface/notebooks/blob/main/examples/images/masked_language_modeling.png?raw=1)\n",
        "\n",
        "We will see how to easily load and preprocess the dataset for each one of those tasks, and how to use the `Trainer` API to fine-tune a model on it.\n",
        "\n",
        "A script version of this notebook you can directly run on a distributed environment or on TPU is available in our [examples folder](https://github.com/huggingface/transformers/tree/master/examples)."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "1r_n9OWV3l-Q"
      },
      "source": [
        "## Preparing the dataset"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "kswRMhPc3l-Q"
      },
      "source": [
        "For each of those tasks, we will use the [Wikitext 2]() dataset as an example. You can load it very easily with the 🤗 Datasets library."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "n2ZRs1cL3l-R"
      },
      "outputs": [],
      "source": [
        "from datasets import load_dataset\n",
        "datasets = load_dataset('wikitext', 'wikitext-2-raw-v1')"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "f1-9jepM3l-W"
      },
      "source": [
        "You can replace the dataset above with any dataset hosted on [the hub](https://huggingface.co/datasets) or use your own files. Just uncomment the following cell and replace the paths with values that will lead to your files:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "uxSaGa_l3l-W"
      },
      "outputs": [],
      "source": [
        "# datasets = load_dataset(\"text\", data_files={\"train\": path_to_train.txt, \"validation\": path_to_validation.txt}"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "jY1SwIrY3l-a"
      },
      "source": [
        "You can also load datasets from a csv or a JSON file, see the [full documentation](https://huggingface.co/docs/datasets/loading_datasets.html#from-local-files) for more information."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "u3EtYfeHIrIz"
      },
      "source": [
        "To access an actual element, you need to select a split first, then give an index:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "X6HrpprwIrIz"
      },
      "outputs": [],
      "source": [
        "datasets[\"train\"][10]"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "WHUmphG3IrI3"
      },
      "source": [
        "To get a sense of what the data looks like, the following function will show some examples picked randomly in the dataset."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "ur5sNUcZ3l-g"
      },
      "outputs": [],
      "source": [
        "from datasets import ClassLabel\n",
        "import random\n",
        "import pandas as pd\n",
        "from IPython.display import display, HTML\n",
        "\n",
        "def show_random_elements(dataset, num_examples=10):\n",
        "    assert num_examples <= len(dataset), \"Can't pick more elements than there are in the dataset.\"\n",
        "    picks = []\n",
        "    for _ in range(num_examples):\n",
        "        pick = random.randint(0, len(dataset)-1)\n",
        "        while pick in picks:\n",
        "            pick = random.randint(0, len(dataset)-1)\n",
        "        picks.append(pick)\n",
        "\n",
        "    df = pd.DataFrame(dataset[picks])\n",
        "    for column, typ in dataset.features.items():\n",
        "        if isinstance(typ, ClassLabel):\n",
        "            df[column] = df[column].transform(lambda i: typ.names[i])\n",
        "    display(HTML(df.to_html()))"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "1Uk8NROQ3l-k"
      },
      "outputs": [],
      "source": [
        "show_random_elements(datasets[\"train\"])"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "CKerdF353l-o"
      },
      "source": [
        "As we can see, some of the texts are a full paragraph of a Wikipedia article while others are just titles or empty lines."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "JEA1ju653l-p"
      },
      "source": [
        "## Causal Language modeling"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "v5GTGKZS3l-q"
      },
      "source": [
        "For causal language modeling (CLM) we are going to take all the texts in our dataset and concatenate them after they are tokenized. Then we will split them in examples of a certain sequence length. This way the model will receive chunks of contiguous text that may look like:\n",
        "```\n",
        "part of text 1\n",
        "```\n",
        "or\n",
        "```\n",
        "end of text 1 [BOS_TOKEN] beginning of text 2\n",
        "```\n",
        "depending on whether they span over several of the original texts in the dataset or not. The labels will be the same as the inputs, shifted to the left.\n",
        "\n",
        "We will use the [`distilgpt2`](https://huggingface.co/distilgpt2) model for this example. You can pick any of the checkpoints listed [here](https://huggingface.co/models?filter=causal-lm) instead:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "-WGBCO343l-q"
      },
      "outputs": [],
      "source": [
        "model_checkpoint = \"distilgpt2\""
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "5io6fY_d3l-u"
      },
      "source": [
        "To tokenize all our texts with the same vocabulary that was used when training the model, we have to download a pretrained tokenizer. This is all done by the `AutoTokenizer` class:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "iAYlS40Z3l-v"
      },
      "outputs": [],
      "source": [
        "from transformers import AutoTokenizer\n",
        "\n",
        "tokenizer = AutoTokenizer.from_pretrained(model_checkpoint, use_fast=True)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "rpOiBrJ13l-y"
      },
      "source": [
        "We can now call the tokenizer on all our texts. This is very simple, using the [`map`](https://huggingface.co/docs/datasets/package_reference/main_classes.html#datasets.Dataset.map) method from the Datasets library. First we define a function that call the tokenizer on our texts:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "lS2m25YM3l-z"
      },
      "outputs": [],
      "source": [
        "def tokenize_function(examples):\n",
        "    return tokenizer(examples[\"text\"])"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "M9xVAa3s3l-2"
      },
      "source": [
        "Then we apply it to all the splits in our `datasets` object, using `batched=True` and 4 processes to speed up the preprocessing. We won't need the `text` column afterward, so we discard it."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "NVAO0H8u3l-3"
      },
      "outputs": [],
      "source": [
        "tokenized_datasets = datasets.map(tokenize_function, batched=True, num_proc=4, remove_columns=[\"text\"])"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8qik3J_C3l-7"
      },
      "source": [
        "If we now look at an element of our datasets, we will see the text have been replaced by the `input_ids` the model will need:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "nYv_mcKk3l-7"
      },
      "outputs": [],
      "source": [
        "tokenized_datasets[\"train\"][1]"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "obvgcXda3l--"
      },
      "source": [
        "Now for the harder part: we need to concatenate all our texts together then split the result in small chunks of a certain `block_size`. To do this, we will use the `map` method again, with the option `batched=True`. This option actually lets us change the number of examples in the datasets by returning a different number of examples than we got. This way, we can create our new samples from a batch of examples.\n",
        "\n",
        "First, we grab the maximum length our model was pretrained with. This might be a big too big to fit in your GPU RAM, so here we take a bit less at just 128."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "DVHs5aCA3l-_"
      },
      "outputs": [],
      "source": [
        "# block_size = tokenizer.model_max_length\n",
        "block_size = 128"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "RpNfGiMw3l_A"
      },
      "source": [
        "Then we write the preprocessing function that will group our texts:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "iaAJy5Hu3l_B"
      },
      "outputs": [],
      "source": [
        "def group_texts(examples):\n",
        "    # Concatenate all texts.\n",
        "    concatenated_examples = {k: sum(examples[k], []) for k in examples.keys()}\n",
        "    total_length = len(concatenated_examples[list(examples.keys())[0]])\n",
        "    # We drop the small remainder, we could add padding if the model supported it instead of this drop, you can\n",
        "        # customize this part to your needs.\n",
        "    total_length = (total_length // block_size) * block_size\n",
        "    # Split by chunks of max_len.\n",
        "    result = {\n",
        "        k: [t[i : i + block_size] for i in range(0, total_length, block_size)]\n",
        "        for k, t in concatenated_examples.items()\n",
        "    }\n",
        "    result[\"labels\"] = result[\"input_ids\"].copy()\n",
        "    return result"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "LGJWXtNv3l_C"
      },
      "source": [
        "First note that we duplicate the inputs for our labels. This is because the model of the 🤗 Transformers library apply the shifting to the right, so we don't need to do it manually.\n",
        "\n",
        "Also note that by default, the `map` method will send a batch of 1,000 examples to be treated by the preprocessing function. So here, we will drop the remainder to make the concatenated tokenized texts a multiple of `block_size` every 1,000 examples. You can adjust this behavior by passing a higher batch size (which will also be processed slower). You can also speed-up the preprocessing by using multiprocessing:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "gXUSfBrq3l_C"
      },
      "outputs": [],
      "source": [
        "lm_datasets = tokenized_datasets.map(\n",
        "    group_texts,\n",
        "    batched=True,\n",
        "    batch_size=1000,\n",
        "    num_proc=4,\n",
        ")"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "6n84V8Gc3l_G"
      },
      "source": [
        "And we can check our datasets have changed: now the samples contain chunks of `block_size` contiguous tokens, potentially spanning over several of our original texts."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "hTeGCLl_3l_G"
      },
      "outputs": [],
      "source": [
        "tokenizer.decode(lm_datasets[\"train\"][1][\"input_ids\"])"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "iEmeQ7Xm3l_H"
      },
      "source": [
        "Now that the data has been cleaned, we're ready to instantiate our `Trainer`. We will a model:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "sPqQA3TT3l_I"
      },
      "outputs": [],
      "source": [
        "from transformers import AutoModelForCausalLM\n",
        "model = AutoModelForCausalLM.from_pretrained(model_checkpoint)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "VyPQTOF_3l_J"
      },
      "source": [
        "And some `TrainingArguments`:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "jElf8LJ33l_K"
      },
      "outputs": [],
      "source": [
        "from transformers import Trainer, TrainingArguments"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "YbSwEhQ63l_L"
      },
      "outputs": [],
      "source": [
        "model_name = model_checkpoint.split(\"/\")[-1]\n",
        "training_args = TrainingArguments(\n",
        "    f\"{model_name}-finetuned-wikitext2\",\n",
        "    eval_strategy = \"epoch\",\n",
        "    learning_rate=2e-5,\n",
        "    weight_decay=0.01,\n",
        "    push_to_hub=True,\n",
        ")"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "dsx-8ebng3FI"
      },
      "source": [
        "The last argument to setup everything so we can push the model to the [Hub](https://huggingface.co/models) regularly during training. Remove it if you didn't follow the installation steps at the top of the notebook. If you want to save your model locally in a name that is different than the name of the repository it will be pushed, or if you want to push your model under an organization and not your name space, use the `hub_model_id` argument to set the repo name (it needs to be the full name, including your namespace: for instance `\"sgugger/gpt-finetuned-wikitext2\"` or `\"huggingface/gpt-finetuned-wikitext2\"`)."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "sZRbT9ui3l_N"
      },
      "source": [
        "We pass along all of those to the `Trainer` class:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "OEuqwIra3l_N"
      },
      "outputs": [],
      "source": [
        "trainer = Trainer(\n",
        "    model=model,\n",
        "    args=training_args,\n",
        "    train_dataset=lm_datasets[\"train\"],\n",
        "    eval_dataset=lm_datasets[\"validation\"],\n",
        ")"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "6Vvz34Td3l_O"
      },
      "source": [
        "And we can train our model:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "NyZvu_MF3l_P"
      },
      "outputs": [],
      "source": [
        "trainer.train()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3APq-vUc3l_R"
      },
      "source": [
        "Once the training is completed, we can evaluate our model and get its perplexity on the validation set like this:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "diKZnB1I3l_R"
      },
      "outputs": [],
      "source": [
        "import math\n",
        "eval_results = trainer.evaluate()\n",
        "print(f\"Perplexity: {math.exp(eval_results['eval_loss']):.2f}\")"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "wY82caEX3l_i"
      },
      "source": [
        "You can now upload the result of the training to the Hub, just execute this instruction:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "KjTnsZXwg3FI"
      },
      "outputs": [],
      "source": [
        "trainer.push_to_hub()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lKIc9Jr8g3FI"
      },
      "source": [
        "You can now share this model with all your friends, family, favorite pets: they can all load it with the identifier `\"your-username/the-name-you-picked\"` so for instance:\n",
        "\n",
        "```python\n",
        "from transformers import AutoModelForCausalLM\n",
        "\n",
        "model = AutoModelForCausalLM.from_pretrained(\"sgugger/my-awesome-model\")\n",
        "```"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "q-EIELH43l_T"
      },
      "source": [
        "## Masked language modeling"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "LWk97-Ny3l_T"
      },
      "source": [
        "For masked language modeling (MLM) we are going to use the same preprocessing as before for our dataset with one additional step: we will randomly mask some tokens (by replacing them by `[MASK]`) and the labels will be adjusted to only include the masked tokens (we don't have to predict the non-masked tokens).\n",
        "\n",
        "We will use the [`distilroberta-base`](https://huggingface.co/distilroberta-base) model for this example. You can pick any of the checkpoints listed [here](https://huggingface.co/models?filter=masked-lm) instead:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "QRTpmyCc3l_T"
      },
      "outputs": [],
      "source": [
        "model_checkpoint = \"distilroberta-base\""
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "12F1ulgT3l_V"
      },
      "source": [
        "We can apply the same tokenization function as before, we just need to update our tokenizer to use the checkpoint we just picked:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "h8RCYcvr3l_V"
      },
      "outputs": [],
      "source": [
        "tokenizer = AutoTokenizer.from_pretrained(model_checkpoint, use_fast=True)\n",
        "tokenized_datasets = datasets.map(tokenize_function, batched=True, num_proc=4, remove_columns=[\"text\"])"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "MTuy8UUs3l_X"
      },
      "source": [
        "And like before, we group texts together and chunk them in samples of length `block_size`. You can skip that step if your dataset is composed of individual sentences."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "LVYPMwEs3l_X"
      },
      "outputs": [],
      "source": [
        "lm_datasets = tokenized_datasets.map(\n",
        "    group_texts,\n",
        "    batched=True,\n",
        "    batch_size=1000,\n",
        "    num_proc=4,\n",
        ")"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "nFJ49iHJ3l_Z"
      },
      "source": [
        "The rest is very similar to what we had, with two exceptions. First we use a model suitable for masked LM:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "PM10A9Za3l_Z"
      },
      "outputs": [],
      "source": [
        "from transformers import AutoModelForMaskedLM\n",
        "model = AutoModelForMaskedLM.from_pretrained(model_checkpoint)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "emlofW7Yg3FJ"
      },
      "source": [
        "We redefine our `TrainingArguments`:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "Hgd9wpjHg3FJ"
      },
      "outputs": [],
      "source": [
        "model_name = model_checkpoint.split(\"/\")[-1]\n",
        "training_args = TrainingArguments(\n",
        "    f\"{model_name}-finetuned-wikitext2\",\n",
        "    eval_strategy = \"epoch\",\n",
        "    learning_rate=2e-5,\n",
        "    weight_decay=0.01,\n",
        "    push_to_hub=True,\n",
        ")"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3tP4SZ3rg3FJ"
      },
      "source": [
        "Like before, the last argument to setup everything so we can push the model to the [Hub](https://huggingface.co/models) regularly during training. Remove it if you didn't follow the installation steps at the top of the notebook. If you want to save your model locally in a name that is different than the name of the repository it will be pushed, or if you want to push your model under an organization and not your name space, use the `hub_model_id` argument to set the repo name (it needs to be the full name, including your namespace: for instance `\"sgugger/bert-finetuned-wikitext2\"` or `\"huggingface/bert-finetuned-wikitext2\"`)."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "z6uuUnvz3l_b"
      },
      "source": [
        "Finally, we use a special `data_collator`. The `data_collator` is a function that is responsible of taking the samples and batching them in tensors. In the previous example, we had nothing special to do, so we just used the default for this argument. Here we want to do the random-masking. We could do it as a pre-processing step (like the tokenization) but then the tokens would always be masked the same way at each epoch. By doing this step inside the `data_collator`, we ensure this random masking is done in a new way each time we go over the data.\n",
        "\n",
        "To do this masking for us, the library provides a `DataCollatorForLanguageModeling`. We can adjust the probability of the masking:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "nRZ-5v_P3l_b"
      },
      "outputs": [],
      "source": [
        "from transformers import DataCollatorForLanguageModeling\n",
        "data_collator = DataCollatorForLanguageModeling(tokenizer=tokenizer, mlm_probability=0.15)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "bqHnWcYC3l_d"
      },
      "source": [
        "Then we just have to pass everything to `Trainer` and begin training:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "V-Y3gNqV3l_d"
      },
      "outputs": [],
      "source": [
        "trainer = Trainer(\n",
        "    model=model,\n",
        "    args=training_args,\n",
        "    train_dataset=lm_datasets[\"train\"],\n",
        "    eval_dataset=lm_datasets[\"validation\"],\n",
        "    data_collator=data_collator,\n",
        ")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "Y9TFqDG_3l_e"
      },
      "outputs": [],
      "source": [
        "trainer.train()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "KDBi0reX3l_g"
      },
      "source": [
        "Like before, we can evaluate our model on the validation set. The perplexity is much lower than for the CLM objective because for the MLM objective, we only have to make predictions for the masked tokens (which represent 15% of the total here) while having access to the rest of the tokens. It's thus an easier task for the model."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "4hSaANqj3l_g"
      },
      "outputs": [],
      "source": [
        "eval_results = trainer.evaluate()\n",
        "print(f\"Perplexity: {math.exp(eval_results['eval_loss']):.2f}\")"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "TDmITtEYg3FK"
      },
      "source": [
        "You can now upload the result of the training to the Hub, just execute this instruction:"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "-3CFIYUrg3FK"
      },
      "outputs": [],
      "source": [
        "trainer.push_to_hub()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Da10yeZNg3FK"
      },
      "source": [
        "You can now share this model with all your friends, family, favorite pets: they can all load it with the identifier `\"your-username/the-name-you-picked\"` so for instance:\n",
        "\n",
        "```python\n",
        "from transformers import AutoModelForMaskedLM\n",
        "\n",
        "model = AutoModelForMaskedLM.from_pretrained(\"sgugger/my-awesome-model\")\n",
        "```"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "RqqSgjeSg3FK"
      },
      "outputs": [],
      "source": []
    }
  ],
  "metadata": {
    "colab": {
      "provenance": []
    },
    "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": "3.9.21"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}